WARNING - OLD ARCHIVES

This is an archived copy of the Xen.org mailing list, which we have preserved to ensure that existing links to archives are not broken. The live archive, which contains the latest emails, can be found at http://lists.xen.org/
   
 
 
Xen 
 
Home Products Support Community News
 
   
 

xen-changelog

[Xen-changelog] [linux-2.6.18-xen] net sfc: Update sfc and sfc_resource

To: xen-changelog@xxxxxxxxxxxxxxxxxxx
Subject: [Xen-changelog] [linux-2.6.18-xen] net sfc: Update sfc and sfc_resource driver to latest release
From: "Xen patchbot-linux-2.6.18-xen" <patchbot-linux-2.6.18-xen@xxxxxxxxxxxxxxxxxxx>
Date: Tue, 31 Mar 2009 12:55:04 -0700
Delivery-date: Tue, 31 Mar 2009 12:55:45 -0700
Envelope-to: www-data@xxxxxxxxxxxxxxxxxxx
List-help: <mailto:xen-changelog-request@lists.xensource.com?subject=help>
List-id: BK change log <xen-changelog.lists.xensource.com>
List-post: <mailto:xen-changelog@lists.xensource.com>
List-subscribe: <http://lists.xensource.com/mailman/listinfo/xen-changelog>, <mailto:xen-changelog-request@lists.xensource.com?subject=subscribe>
List-unsubscribe: <http://lists.xensource.com/mailman/listinfo/xen-changelog>, <mailto:xen-changelog-request@lists.xensource.com?subject=unsubscribe>
Reply-to: xen-devel@xxxxxxxxxxxxxxxxxxx
Sender: xen-changelog-bounces@xxxxxxxxxxxxxxxxxxx
# HG changeset patch
# User Keir Fraser <keir.fraser@xxxxxxxxxx>
# Date 1238497150 -3600
# Node ID ad4d307bf9ced378d0b49d4559ae33ecbff3c1b7
# Parent  89d9c025b46c64922b8b082bf3ed69e0c03ad19e
net sfc: Update sfc and sfc_resource driver to latest release

...and update sfc_netfront, sfc_netback, sfc_netutil for any API changes

sfc_netback: Fix asymmetric use of SFC buffer table alloc and free
sfc_netback: Clean up if no SFC accel device found
sfc_netback: Gracefully handle case where page grant fails
sfc_netback: Disable net acceleration if the physical link goes down
sfc_netfront: Less verbose error messages, more verbose counters for
rx discard errors
sfc_netfront: Gracefully handle case where SFC netfront fails during
initialisation

Signed-off-by: Kieran Mansley <kmansley@xxxxxxxxxxxxxx>
---
 drivers/net/sfc/Kconfig                                          |   10 
 drivers/net/sfc/Makefile                                         |   50 
 drivers/net/sfc/Module.symvers                                   |    6 
 drivers/net/sfc/bitfield.h                                       |   23 
 drivers/net/sfc/debugfs.c                                        |  102 -
 drivers/net/sfc/driverlink.c                                     |   15 
 drivers/net/sfc/driverlink_api.h                                 |   39 
 drivers/net/sfc/efx.c                                            |  768 
+++-----
 drivers/net/sfc/efx.h                                            |   20 
 drivers/net/sfc/enum.h                                           |    4 
 drivers/net/sfc/ethtool.c                                        |   50 
 drivers/net/sfc/falcon.c                                         |  338 +--
 drivers/net/sfc/falcon.h                                         |   23 
 drivers/net/sfc/falcon_hwdefs.h                                  |   18 
 drivers/net/sfc/falcon_io.h                                      |    3 
 drivers/net/sfc/falcon_xmac.c                                    |  137 +
 drivers/net/sfc/i2c-direct.c                                     |    8 
 drivers/net/sfc/kernel_compat.c                                  |  576 ------
 drivers/net/sfc/kernel_compat.h                                  |  857 
----------
 drivers/net/sfc/mdio_10g.c                                       |    4 
 drivers/net/sfc/mtd.c                                            |    4 
 drivers/net/sfc/net_driver.h                                     |  167 +
 drivers/net/sfc/null_phy.c                                       |    6 
 drivers/net/sfc/pm8358_phy.c                                     |    4 
 drivers/net/sfc/rx.c                                             |  484 +++++
 drivers/net/sfc/rx.h                                             |   17 
 drivers/net/sfc/selftest.c                                       |   40 
 drivers/net/sfc/sfc_resource/ci/driver/efab/hardware.h           |   21 
 drivers/net/sfc/sfc_resource/ci/driver/efab/hardware/falcon.h    |   16 
 drivers/net/sfc/sfc_resource/ci/driver/resource/efx_vi.h         |    3 
 drivers/net/sfc/sfc_resource/ci/driver/resource/linux_efhw_nic.h |    6 
 drivers/net/sfc/sfc_resource/ci/efhw/common.h                    |    4 
 drivers/net/sfc/sfc_resource/ci/efhw/common_sysdep.h             |    2 
 drivers/net/sfc/sfc_resource/ci/efhw/efhw_types.h                |   43 
 drivers/net/sfc/sfc_resource/ci/efhw/eventq.h                    |    7 
 drivers/net/sfc/sfc_resource/ci/efhw/falcon.h                    |    5 
 drivers/net/sfc/sfc_resource/ci/efhw/hardware_sysdep.h           |   10 
 drivers/net/sfc/sfc_resource/ci/efhw/iopage.h                    |    8 
 drivers/net/sfc/sfc_resource/ci/efhw/iopage_types.h              |   79 
 drivers/net/sfc/sfc_resource/ci/efrm/driver_private.h            |   10 
 drivers/net/sfc/sfc_resource/ci/efrm/iobufset.h                  |    2 
 drivers/net/sfc/sfc_resource/ci/efrm/nic_table.h                 |   12 
 drivers/net/sfc/sfc_resource/ci/efrm/private.h                   |    8 
 drivers/net/sfc/sfc_resource/ci/efrm/resource_id.h               |    2 
 drivers/net/sfc/sfc_resource/ci/efrm/sysdep.h                    |    6 
 drivers/net/sfc/sfc_resource/ci/efrm/sysdep_linux.h              |   19 
 drivers/net/sfc/sfc_resource/ci/efrm/vi_resource.h               |    8 
 drivers/net/sfc/sfc_resource/ci/efrm/vi_resource_manager.h       |    6 
 drivers/net/sfc/sfc_resource/ci/efrm/vi_resource_private.h       |   15 
 drivers/net/sfc/sfc_resource/driver_object.c                     |   52 
 drivers/net/sfc/sfc_resource/driverlink_new.c                    |   64 
 drivers/net/sfc/sfc_resource/efx_vi_shm.c                        |   19 
 drivers/net/sfc/sfc_resource/eventq.c                            |   10 
 drivers/net/sfc/sfc_resource/falcon.c                            |  266 +--
 drivers/net/sfc/sfc_resource/falcon_mac.c                        |    4 
 drivers/net/sfc/sfc_resource/filter_resource.c                   |    2 
 drivers/net/sfc/sfc_resource/iobufset_resource.c                 |   32 
 drivers/net/sfc/sfc_resource/iopage.c                            |   10 
 drivers/net/sfc/sfc_resource/kernel_compat.c                     |  242 --
 drivers/net/sfc/sfc_resource/kernel_compat.h                     |  140 -
 drivers/net/sfc/sfc_resource/linux_resource_internal.h           |    3 
 drivers/net/sfc/sfc_resource/nic.c                               |   39 
 drivers/net/sfc/sfc_resource/resource_driver.c                   |   78 
 drivers/net/sfc/sfc_resource/resources.c                         |    4 
 drivers/net/sfc/sfc_resource/vi_resource_alloc.c                 |   20 
 drivers/net/sfc/sfc_resource/vi_resource_event.c                 |   32 
 drivers/net/sfc/sfc_resource/vi_resource_flush.c                 |    2 
 drivers/net/sfc/sfc_resource/vi_resource_manager.c               |   16 
 drivers/net/sfc/sfe4001.c                                        |   34 
 drivers/net/sfc/tenxpress.c                                      |    8 
 drivers/net/sfc/tx.c                                             |  137 -
 drivers/net/sfc/txc43128_phy.c                                   |    5 
 drivers/net/sfc/workarounds.h                                    |    4 
 drivers/net/sfc/xfp_phy.c                                        |   27 
 drivers/xen/sfc_netback/accel.c                                  |   48 
 drivers/xen/sfc_netback/accel_solarflare.c                       |   84 
 drivers/xen/sfc_netback/accel_xenbus.c                           |    3 
 drivers/xen/sfc_netback/ci/driver/resource/efx_vi.h              |    3 
 drivers/xen/sfc_netback/ci/efhw/common.h                         |    4 
 drivers/xen/sfc_netback/ci/efhw/common_sysdep.h                  |    8 
 drivers/xen/sfc_netback/ci/efhw/efhw_types.h                     |   43 
 drivers/xen/sfc_netback/ci/efhw/hardware_sysdep.h                |   10 
 drivers/xen/sfc_netback/ci/efhw/iopage_types.h                   |   79 
 drivers/xen/sfc_netback/ci/efrm/nic_table.h                      |   12 
 drivers/xen/sfc_netback/ci/efrm/sysdep_linux.h                   |   19 
 drivers/xen/sfc_netback/ci/tools/log.h                           |    7 
 drivers/xen/sfc_netback/ci/tools/platform/gcc_x86.h              |    9 
 drivers/xen/sfc_netfront/accel.h                                 |   18 
 drivers/xen/sfc_netfront/accel_debugfs.c                         |   16 
 drivers/xen/sfc_netfront/accel_msg.c                             |    2 
 drivers/xen/sfc_netfront/accel_netfront.c                        |   13 
 drivers/xen/sfc_netfront/accel_vi.c                              |   29 
 drivers/xen/sfc_netfront/ef_vi_falcon.h                          |    2 
 drivers/xen/sfc_netfront/etherfabric/ef_vi.h                     |   18 
 drivers/xen/sfc_netfront/falcon_event.c                          |    2 
 drivers/xen/sfc_netfront/falcon_vi.c                             |    8 
 drivers/xen/sfc_netfront/sysdep.h                                |   31 
 97 files changed, 2191 insertions(+), 3592 deletions(-)

diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/Kconfig
--- a/drivers/net/sfc/Kconfig   Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/Kconfig   Tue Mar 31 11:59:10 2009 +0100
@@ -2,6 +2,7 @@ config SFC
        tristate "Solarflare Solarstorm SFC4000 support"
        depends on PCI && INET
        select MII
+       select CRC32
        help
          This driver supports 10-gigabit Ethernet cards based on
          the Solarflare Communications Solarstorm SFC4000 controller.
@@ -28,8 +29,7 @@ config SFC_MTD
          new boot ROM to the NIC.
 
 config SFC_RESOURCE
-       depends on SFC && X86
-       tristate "Solarflare Solarstorm SFC4000 resource driver"
-       help
-         This module provides the SFC resource manager driver.
-
+        depends on SFC && X86
+        tristate "Solarflare Solarstorm SFC4000 resource driver"
+        help
+          This module provides the SFC resource manager driver.
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/Makefile
--- a/drivers/net/sfc/Makefile  Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/Makefile  Tue Mar 31 11:59:10 2009 +0100
@@ -1,43 +1,13 @@
+sfc-y                  += efx.o falcon.o tx.o rx.o mentormac.o falcon_gmac.o \
+                          falcon_xmac.o alaska.o i2c-direct.o selftest.o \
+                          driverlink.o ethtool.o xfp_phy.o mdio_10g.o \
+                          txc43128_phy.o tenxpress.o lm87_support.o boards.o \
+                          sfe4001.o pm8358_phy.o null_phy.o kernel_compat.o
+sfc-$(CONFIG_SFC_DEBUGFS) += debugfs.o
+obj-$(CONFIG_SFC)      += sfc.o
 
-# Final objects
-sfc_o = sfc.o
-sfc_mtd_o = sfc_mtd.o
-
-# Constituent objects
-sfc_elements_o :=
-sfc_elements_o += efx.o
-sfc_elements_o += falcon.o
-sfc_elements_o += tx.o
-sfc_elements_o += rx.o
-sfc_elements_o += mentormac.o
-sfc_elements_o += falcon_gmac.o
-sfc_elements_o += falcon_xmac.o
-sfc_elements_o += alaska.o
-sfc_elements_o += i2c-direct.o
-sfc_elements_o += selftest.o
-sfc_elements_o += driverlink.o
-ifeq ($(CONFIG_SFC_DEBUGFS),y)
-sfc_elements_o += debugfs.o
-endif
-sfc_elements_o += ethtool.o
-sfc_elements_o += xfp_phy.o
-sfc_elements_o += mdio_10g.o
-sfc_elements_o += txc43128_phy.o
-sfc_elements_o += tenxpress.o
-sfc_elements_o += lm87_support.o
-sfc_elements_o += boards.o
-sfc_elements_o += sfe4001.o
-sfc_elements_o += pm8358_phy.o
-sfc_elements_o += null_phy.o
-sfc_elements_o += phy.o
-sfc_elements_o += kernel_compat.o
-
-sfc_mtd_elements_o := mtd.o
-
-obj-$(CONFIG_SFC) += $(sfc_o)
-obj-$(CONFIG_SFC_MTD) += $(sfc_mtd_o)
-
-sfc-objs = $(sfc_elements_o)
-sfc_mtd-objs = $(sfc_mtd_elements_o)
+sfc_mtd-y = mtd.o
+obj-$(CONFIG_SFC_MTD)  += sfc_mtd.o
 
 obj-$(CONFIG_SFC_RESOURCE) += sfc_resource/
+
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/Module.symvers
--- /dev/null   Thu Jan 01 00:00:00 1970 +0000
+++ b/drivers/net/sfc/Module.symvers    Tue Mar 31 11:59:10 2009 +0100
@@ -0,0 +1,6 @@
+0x2e5e77fa     efx_dl_unregister_driver        drivers/net/sfc/sfc     
EXPORT_SYMBOL
+0x4ac7afe9     efx_dl_schedule_reset   drivers/net/sfc/sfc     EXPORT_SYMBOL
+0xbb52ca8e     efx_dl_register_driver_api_ver_1        drivers/net/sfc/sfc     
EXPORT_SYMBOL
+0x278552f6     efx_dl_register_callbacks       drivers/net/sfc/sfc     
EXPORT_SYMBOL
+0xc4414515     efx_dl_get_nic  drivers/net/sfc/sfc     EXPORT_SYMBOL
+0x42cae6c4     efx_dl_unregister_callbacks     drivers/net/sfc/sfc     
EXPORT_SYMBOL
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/bitfield.h
--- a/drivers/net/sfc/bitfield.h        Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/bitfield.h        Tue Mar 31 11:59:10 2009 +0100
@@ -54,11 +54,6 @@
 #define EFX_DWORD_2_WIDTH 32
 #define EFX_DWORD_3_LBN 96
 #define EFX_DWORD_3_WIDTH 32
-
-#define EFX_BYTE  1
-#define EFX_WORD  2
-#define EFX_DWORD 4
-#define EFX_OWORD 8
 
 /* Specified attribute (e.g. LBN) of the specified field */
 #define EFX_VAL(field, attribute) field ## _ ## attribute
@@ -505,7 +500,7 @@ typedef union efx_oword {
 #endif
 
 #define EFX_SET_OWORD_FIELD_VER(efx, oword, field, value) do { \
-       if (FALCON_REV(efx) == FALCON_REV_B0) {                    \
+       if (FALCON_REV(efx) >= FALCON_REV_B0) {                    \
                EFX_SET_OWORD_FIELD((oword), field##_B0, (value)); \
        } else { \
                EFX_SET_OWORD_FIELD((oword), field##_A1, (value)); \
@@ -513,7 +508,7 @@ typedef union efx_oword {
 } while (0)
 
 #define EFX_QWORD_FIELD_VER(efx, qword, field) \
-       (FALCON_REV(efx) == FALCON_REV_B0 ?     \
+       (FALCON_REV(efx) >= FALCON_REV_B0 ?     \
         EFX_QWORD_FIELD((qword), field##_B0) : \
         EFX_QWORD_FIELD((qword), field##_A1))
 
@@ -527,18 +522,4 @@ typedef union efx_oword {
                          ~((u64) 0) : ~((u32) 0))
 #define EFX_DMA_MASK(mask) ((mask) & EFX_DMA_MAX_MASK)
 
-/*
- * Determine if a DMA address is over the 4GB threshold
- *
- * Defined in a slightly tortuous way to avoid compiler warnings.
- */
-static inline int efx_is_over_4gb(dma_addr_t address)
-{
-       if (DMA_ADDR_T_WIDTH > 32)
-               return (((u64) address) >> 32) ? 1 : 0;
-       else
-               /* Can never be true */
-               return 0;
-}
-
 #endif /* EFX_BITFIELD_H */
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/debugfs.c
--- a/drivers/net/sfc/debugfs.c Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/debugfs.c Tue Mar 31 11:59:10 2009 +0100
@@ -27,9 +27,6 @@
 
 #include <linux/module.h>
 #include <linux/pci.h>
-/* For out-of-tree builds we always need procfs, if only for a compatibility
- * symlink.
- */
 #include <linux/proc_fs.h>
 #include <linux/dcache.h>
 #include <linux/seq_file.h>
@@ -38,13 +35,6 @@
 #include "debugfs.h"
 #include "falcon.h"
 
-/* EFX_USE_DEBUGFS is defined by kernel_compat.h so we can't decide whether to
- * include this earlier.
- */
-#ifdef EFX_USE_DEBUGFS
-#include <linux/debugfs.h>
-#endif
-
 #ifndef PRIu64
 #      if (BITS_PER_LONG == 64)
 #              define PRIu64 "lu"
@@ -53,8 +43,6 @@
 #      endif
 #endif
 
-#ifndef EFX_USE_DEBUGFS
-
 static void efx_debugfs_remove(struct proc_dir_entry *entry)
 {
        if (entry)
@@ -65,7 +53,6 @@ static void efx_debugfs_remove(struct pr
 #define debugfs_create_dir proc_mkdir
 #define debugfs_create_symlink proc_symlink
 
-#endif /* !EFX_USE_DEBUGFS */
 
 /* Parameter definition bound to a structure - each file has one of these */
 struct efx_debugfs_bound_param {
@@ -87,25 +74,6 @@ static struct dentry *efx_debug_cards;
 
 /* Sequential file interface to bound parameters */
 
-#if defined(EFX_USE_DEBUGFS)
-
-static int efx_debugfs_seq_show(struct seq_file *file, void *v)
-{
-       struct efx_debugfs_bound_param *binding =
-               (struct efx_debugfs_bound_param *)file->private;
-
-       return binding->param->reader(file,
-                                     binding->structure +
-                                     binding->param->offset);
-}
-
-static int efx_debugfs_open(struct inode *inode, struct file *file)
-{
-       return single_open(file, efx_debugfs_seq_show, inode->i_private);
-}
-
-#else /* EFX_NOT_UPSTREAM && !EFX_USE_DEBUGFS */
-
 static int efx_debugfs_seq_show(struct seq_file *file, void *v)
 {
        struct proc_dir_entry *entry = (struct proc_dir_entry *)file->private;
@@ -123,8 +91,6 @@ static int efx_debugfs_open(struct inode
 {
        return single_open(file, efx_debugfs_seq_show, PROC_I(inode)->pde);
 }
-
-#endif /* !EFX_NOT_UPSTREAM || EFX_USE_DEBUGFS */
 
 
 static struct file_operations efx_debugfs_file_ops = {
@@ -136,41 +102,10 @@ static struct file_operations efx_debugf
 };
 
 
-#if defined(EFX_USE_DEBUGFS)
-
-/**
- * efx_fini_debugfs_child - remove a named child of a debugfs directory
- * @dir:               Directory
- * @name:              Name of child
- *
- * This removes the named child from the directory, if it exists.
- */
-void efx_fini_debugfs_child(struct dentry *dir, const char *name)
-{
-       struct qstr child_name;
-       struct dentry *child;
-
-       child_name.len = strlen(name);
-       child_name.name = name;
-       child_name.hash = full_name_hash(child_name.name, child_name.len);
-       child = d_lookup(dir, &child_name);
-       if (child) {
-               /* If it's a "regular" file, free its parameter binding */
-               if (S_ISREG(child->d_inode->i_mode))
-                       kfree(child->d_inode->i_private);
-               debugfs_remove(child);
-               dput(child);
-       }
-}
-
-#else /* EFX_NOT_UPSTREAM && !EFX_USE_DEBUGFS */
-
 void efx_fini_debugfs_child(struct proc_dir_entry *dir, const char *name)
 {
        remove_proc_entry(name, dir);
 }
-
-#endif /* !EFX_NOT_UPSTREAM || EFX_USE_DEBUGFS */
 
 /*
  * Remove a debugfs directory.
@@ -283,22 +218,6 @@ static int efx_init_debugfs_files(struct
 
        while (param->name) {
                struct dentry *entry;
-#if defined(EFX_USE_DEBUGFS)
-               struct efx_debugfs_bound_param *binding;
-
-               binding = kmalloc(sizeof(*binding), GFP_KERNEL);
-               if (!binding)
-                       goto err;
-               binding->param = param;
-               binding->structure = structure;
-
-               entry = debugfs_create_file(param->name, S_IRUGO, parent,
-                                           binding, &efx_debugfs_file_ops);
-               if (!entry) {
-                       kfree(binding);
-                       goto err;
-               }
-#else
                entry = create_proc_entry(param->name, S_IRUGO, parent);
                if (!entry)
                        goto err;
@@ -314,7 +233,6 @@ static int efx_init_debugfs_files(struct
                entry->proc_fops = &efx_debugfs_file_ops;
                smp_wmb();
                entry->read_proc = (read_proc_t *) structure;
-#endif
 
                param++;
        }
@@ -392,7 +310,6 @@ static struct efx_debugfs_parameter efx_
 static struct efx_debugfs_parameter efx_debugfs_port_parameters[] = {
        EFX_NAMED_PARAMETER(enabled, struct efx_nic, port_enabled,
                            int, efx_debugfs_read_int),
-       EFX_INT_PARAMETER(struct efx_nic, net_dev_registered),
        EFX_INT_PARAMETER(struct efx_nic, rx_checksum_enabled),
        EFX_ATOMIC_PARAMETER(struct efx_nic, netif_stop_count),
        EFX_INT_PARAMETER(struct efx_nic, link_up),
@@ -668,6 +585,14 @@ static struct efx_debugfs_parameter efx_
        EFX_INT_PARAMETER(struct efx_channel, rx_alloc_level),
        EFX_INT_PARAMETER(struct efx_channel, rx_alloc_push_pages),
        EFX_INT_PARAMETER(struct efx_channel, rx_alloc_pop_pages),
+       EFX_UINT_PARAMETER(struct efx_channel, ssr.n_merges),
+       EFX_UINT_PARAMETER(struct efx_channel, ssr.n_bursts),
+       EFX_UINT_PARAMETER(struct efx_channel, ssr.n_slow_start),
+       EFX_UINT_PARAMETER(struct efx_channel, ssr.n_misorder),
+       EFX_UINT_PARAMETER(struct efx_channel, ssr.n_too_many),
+       EFX_UINT_PARAMETER(struct efx_channel, ssr.n_new_stream),
+       EFX_UINT_PARAMETER(struct efx_channel, ssr.n_drop_idle),
+       EFX_UINT_PARAMETER(struct efx_channel, ssr.n_drop_closed),
        {NULL},
 };
 
@@ -882,11 +807,7 @@ int efx_init_debugfs(void)
 int efx_init_debugfs(void)
 {
        /* Create top-level directory */
-#if defined(EFX_USE_DEBUGFS)
-       efx_debug_root = debugfs_create_dir("sfc", NULL);
-#else
        efx_debug_root = proc_mkdir("sfc", proc_root_driver);
-#endif
        if (!efx_debug_root)
                goto err;
 
@@ -895,11 +816,6 @@ int efx_init_debugfs(void)
        if (!efx_debug_cards)
                goto err;
 
-#if defined(EFX_USE_DEBUGFS)
-       /* Create compatibility sym-link */
-       if (!proc_symlink("sfc", proc_root_driver, "/sys/kernel/debug/sfc"))
-               goto err;
-#endif
        return 0;
 
  err:
@@ -914,9 +830,7 @@ int efx_init_debugfs(void)
  */
 void efx_fini_debugfs(void)
 {
-#if defined(EFX_USE_DEBUGFS)
        remove_proc_entry("sfc", proc_root_driver);
-#endif
        debugfs_remove(efx_debug_cards);
        efx_debug_cards = NULL;
        debugfs_remove(efx_debug_root);
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/driverlink.c
--- a/drivers/net/sfc/driverlink.c      Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/driverlink.c      Tue Mar 31 11:59:10 2009 +0100
@@ -140,20 +140,15 @@ void efx_dl_unregister_driver(struct efx
        printk(KERN_INFO "Efx driverlink unregistering %s driver\n",
                 driver->name);
 
-       /* Acquire lock.  We can't return failure, so have to use
-        * down() instead of down_interruptible()
-        */
+       /* Acquire lock.  We can't return failure */
        mutex_lock(&efx_driverlink_lock);
 
-       /* Remove all devices claimed by the driver */
        list_for_each_entry_safe(efx_handle, efx_handle_n,
                                 &driver->device_list, driver_node)
                efx_dl_del_device(&efx_handle->efx_dev);
 
-       /* Remove driver from driver list */
        list_del(&driver->node);
 
-       /* Release lock */
        mutex_unlock(&efx_driverlink_lock);
 }
 EXPORT_SYMBOL(efx_dl_unregister_driver);
@@ -252,22 +247,14 @@ int efx_dl_register_nic(struct efx_nic *
  * To avoid a branch point on the fast-path, the callbacks are always
  * implemented - they are never NULL.
  */
-#if defined(EFX_USE_FASTCALL)
 static enum efx_veto fastcall
-#else
-static enum efx_veto
-#endif
 efx_dummy_tx_packet_callback(struct efx_dl_device *efx_dev, struct sk_buff 
*skb)
 {
        /* Never veto the packet */
        return EFX_ALLOW_PACKET;
 }
 
-#if defined(EFX_USE_FASTCALL)
 static enum efx_veto fastcall
-#else
-static enum efx_veto
-#endif
 efx_dummy_rx_packet_callback(struct efx_dl_device *efx_dev,
                             const char *pkt_buf, int len)
 {
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/driverlink_api.h
--- a/drivers/net/sfc/driverlink_api.h  Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/driverlink_api.h  Tue Mar 31 11:59:10 2009 +0100
@@ -29,13 +29,8 @@
 #define EFX_DRIVERLINK_API_H
 
 #include <linux/list.h> /* for struct list_head */
-#if !defined(EFX_USE_FASTCALL)
-       #include <linux/version.h>
-       #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20)
-               #define EFX_USE_FASTCALL yes
-               #include <linux/linkage.h>
-       #endif
-#endif
+#include <linux/linkage.h>
+#define EFX_USE_FASTCALL yes
 
 /**
  * DOC: Efx driverlink API
@@ -327,25 +322,25 @@ enum efx_dl_falcon_resource_flags {
  *     The sfc driver will provide the appropriate lock semantics for
  *     the underlying hardware.
  * @buffer_table_min: First available buffer table entry
- * @buffer_table_max: Last available buffer table entry + 1
+ * @buffer_table_lim: Last available buffer table entry + 1
  * @evq_timer_min: First available event queue with timer
- * @evq_timer_max: Last available event queue with timer + 1
+ * @evq_timer_lim: Last available event queue with timer + 1
  * @evq_int_min: First available event queue with interrupt
- * @evq_int_max: Last available event queue with interrupt + 1
+ * @evq_int_lim: Last available event queue with interrupt + 1
  * @rxq_min: First available RX queue
- * @rxq_max: Last available RX queue + 1
+ * @rxq_lim: Last available RX queue + 1
  * @txq_min: First available TX queue
- * @txq_max: Last available TX queue + 1
+ * @txq_lim: Last available TX queue + 1
  * @flags: Hardware variation flags
  */
 struct efx_dl_falcon_resources {
        struct efx_dl_device_info hdr;
        spinlock_t *biu_lock;
-       unsigned buffer_table_min, buffer_table_max;
-       unsigned evq_timer_min, evq_timer_max;
-       unsigned evq_int_min, evq_int_max;
-       unsigned rxq_min, rxq_max;
-       unsigned txq_min, txq_max;
+       unsigned buffer_table_min, buffer_table_lim;
+       unsigned evq_timer_min, evq_timer_lim;
+       unsigned evq_int_min, evq_int_lim;
+       unsigned rxq_min, rxq_lim;
+       unsigned txq_min, txq_lim;
        enum efx_dl_falcon_resource_flags flags;
 };
 
@@ -426,13 +421,8 @@ struct efx_dl_callbacks {
         * may have multiple TX queues, running in parallel, please avoid
         * the need for locking if it all possible.
         */
-#if defined(EFX_USE_FASTCALL)
        enum efx_veto fastcall (*tx_packet) (struct efx_dl_device *efx_dev,
                                             struct sk_buff *skb);
-#else
-       enum efx_veto (*tx_packet) (struct efx_dl_device *efx_dev,
-                                   struct sk_buff *skb);
-#endif
 
        /*
         * rx_packet - Packet received.
@@ -457,13 +447,8 @@ struct efx_dl_callbacks {
         * allows for lockless operation between receive channels, so
         * please avoid the need for locking if at all possible.
         */
-#if defined(EFX_USE_FASTCALL)
        enum efx_veto fastcall (*rx_packet) (struct efx_dl_device *efx_dev,
                                             const char *pkt_hdr, int pkt_len);
-#else
-       enum efx_veto (*rx_packet) (struct efx_dl_device *efx_dev,
-                                   const char *pkt_hdr, int pkt_len);
-#endif
 
        /*
         * link_change - Link status change.
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/efx.c
--- a/drivers/net/sfc/efx.c     Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/efx.c     Tue Mar 31 11:59:10 2009 +0100
@@ -36,7 +36,6 @@
 #include <linux/in.h>
 #include <linux/crc32.h>
 #include <linux/ethtool.h>
-#include <asm/uaccess.h>
 #include "net_driver.h"
 #include "gmii.h"
 #include "driverlink.h"
@@ -93,13 +92,16 @@ const char *efx_phy_type_names[] = {
 
 const unsigned int efx_reset_type_max = RESET_TYPE_MAX;
 const char *efx_reset_type_names[] = {
-       [RESET_TYPE_INVISIBLE]    = "INVISIBLE",
-       [RESET_TYPE_ALL]          = "ALL",
-       [RESET_TYPE_WORLD]        = "WORLD",
-       [RESET_TYPE_DISABLE]      = "DISABLE",
-       [RESET_TYPE_MONITOR]      = "MONITOR",
-       [RESET_TYPE_INT_ERROR]    = "INT_ERROR",
-       [RESET_TYPE_RX_RECOVERY]  = "RX_RECOVERY",
+       [RESET_TYPE_INVISIBLE]     = "INVISIBLE",
+       [RESET_TYPE_ALL]           = "ALL",
+       [RESET_TYPE_WORLD]         = "WORLD",
+       [RESET_TYPE_DISABLE]       = "DISABLE",
+       [RESET_TYPE_MONITOR]       = "MONITOR",
+       [RESET_TYPE_INT_ERROR]     = "INT_ERROR",
+       [RESET_TYPE_RX_RECOVERY]   = "RX_RECOVERY",
+       [RESET_TYPE_RX_DESC_FETCH] = "RX_DESC_FETCH",
+       [RESET_TYPE_TX_DESC_FETCH] = "TX_DESC_FETCH",
+       [RESET_TYPE_TX_SKIP]       = "TX_SKIP",
 };
 
 const unsigned int efx_nic_state_max = STATE_MAX;
@@ -113,12 +115,28 @@ const char *efx_nic_state_names[] = {
 
 #define EFX_MAX_MTU (9 * 1024)
 
+/* RX slow fill workqueue. If memory allocation fails in the fast path,
+ * a work item is pushed onto this work queue to retry the allocation later,
+ * to avoid the NIC being starved of RX buffers. Since this is a per cpu
+ * workqueue, there is nothing to be gained in making it per NIC
+ */
+static struct workqueue_struct *refill_workqueue;
 
 /**************************************************************************
  *
  * Configurable values
  *
  *************************************************************************/
+
+/*
+ * Enable large receive offload (LRO) aka soft segment reassembly (SSR)
+ *
+ * This sets the default for new devices.  It can be controlled later
+ * using ethtool.
+ */
+static int lro = 1;
+module_param(lro, int, 0644);
+MODULE_PARM_DESC(lro, "Large receive offload acceleration");
 
 /*
  * Use separate channels for TX and RX events
@@ -150,7 +168,7 @@ static unsigned int monitor_reset = 1;
 /* This controls whether or not the driver will initialise devices
  * with invalid MAC addresses stored in the EEPROM or flash.  If true,
  * such devices will be initialised with a random locally-generated
- * MAC address.  This allows for loading the efx_mtd driver to
+ * MAC address.  This allows for loading the sfc_mtd driver to
  * reprogram the flash, even if the flash contents (including the MAC
  * address) have previously been erased.
  */
@@ -182,11 +200,7 @@ static unsigned int tx_irq_mod_usec = 15
  */
 static unsigned int allow_load_on_failure;
 
-/* Set to 1 to enable the use of Message-Signalled Interrupts (MSI).
- * MSI will not work on some motherboards due to limitations of the
- * chipset, so the default is off.
- *
- * This is the highest capability interrupt mode to use
+/* This is the first interrupt mode to try out of:
  * 0 => MSI-X
  * 1 => MSI
  * 2 => legacy
@@ -203,10 +217,8 @@ static unsigned int onload_offline_selft
  * i.e. the number of CPUs among which we may distribute simultaneous
  * interrupt handling.
  *
- * Cards without MSI-X will only target one CPU
- *
- * Default (0) means to use all CPUs in the system.  This parameter
- * can be set using "rss_cpus=xxx" when loading the module.
+ * Cards without MSI-X will only target one CPU via legacy or MSI interrupt.
+ * The default (0) means to assign an interrupt to each package (level II 
cache)
  */
 static unsigned int rss_cpus;
 module_param(rss_cpus, uint, 0444);
@@ -222,6 +234,13 @@ static void efx_fini_napi(struct efx_nic
 static void efx_fini_napi(struct efx_nic *efx);
 static void efx_fini_channels(struct efx_nic *efx);
 
+#define EFX_ASSERT_RESET_SERIALISED(efx)               \
+       do {                                            \
+               if ((efx->state == STATE_RUNNING) ||    \
+                   (efx->state == STATE_RESETTING))    \
+                       ASSERT_RTNL();                  \
+       } while (0)
+
 /**************************************************************************
  *
  * Event queue processing
@@ -253,6 +272,7 @@ static inline int efx_process_channel(st
                channel->rx_pkt = NULL;
        }
 
+       efx_flush_lro(channel);
        efx_rx_strategy(channel);
 
        /* Refill descriptor rings as necessary */
@@ -288,19 +308,11 @@ static inline void efx_channel_processed
  * NAPI guarantees serialisation of polls of the same device, which
  * provides the guarantee required by efx_process_channel().
  */
-#if !defined(EFX_HAVE_OLD_NAPI)
-static int efx_poll(struct napi_struct *napi, int budget)
-{
-       struct efx_channel *channel =
-               container_of(napi, struct efx_channel, napi_str);
-       struct net_device *napi_dev = channel->napi_dev;
-#else
 static int efx_poll(struct net_device *napi, int *budget_ret)
 {
        struct net_device *napi_dev = napi;
        struct efx_channel *channel = napi_dev->priv;
        int budget = min(napi_dev->quota, *budget_ret);
-#endif
        int unused;
        int rx_packets;
 
@@ -309,10 +321,8 @@ static int efx_poll(struct net_device *n
 
        unused = efx_process_channel(channel, budget);
        rx_packets = (budget - unused);
-#if defined(EFX_HAVE_OLD_NAPI)
        napi_dev->quota -= rx_packets;
        *budget_ret -= rx_packets;
-#endif
 
        if (rx_packets < budget) {
                /* There is no race here; although napi_disable() will
@@ -324,11 +334,7 @@ static int efx_poll(struct net_device *n
                efx_channel_processed(channel);
        }
 
-#if !defined(EFX_HAVE_OLD_NAPI)
-       return rx_packets;
-#else
        return (rx_packets >= budget);
-#endif
 }
 
 /* Process the eventq of the specified channel immediately on this CPU
@@ -387,8 +393,6 @@ static int efx_init_eventq(struct efx_ch
 {
        EFX_LOG(channel->efx, "chan %d init event queue\n", channel->channel);
 
-       ASSERT_RTNL();
-
        /* Initialise fields */
        channel->eventq_read_ptr = 0;
 
@@ -398,8 +402,6 @@ static void efx_fini_eventq(struct efx_c
 static void efx_fini_eventq(struct efx_channel *channel)
 {
        EFX_LOG(channel->efx, "chan %d fini event queue\n", channel->channel);
-
-       ASSERT_RTNL();
 
        falcon_fini_eventq(channel);
 }
@@ -429,7 +431,7 @@ static void efx_calc_rx_buffer_params(st
               EFX_MAX_FRAME_LEN(efx->net_dev->mtu) +
               efx->type->rx_buffer_padding);
 
-       /* Page-based allocation page-order */
+       /* Calculate page-order */
        for (order = 0; ((1u << order) * PAGE_SIZE) < len; ++order)
                ;
 
@@ -573,9 +575,7 @@ static void efx_stop_channel(struct efx_
        /* Wait for any NAPI processing to complete */
        napi_disable(&channel->napi_str);
 
-       /* Ensure that any worker threads have exited or will be
-        * no-ops.
-        */
+       /* Ensure that any worker threads have exited or will be no-ops */
        efx_for_each_channel_rx_queue(rx_queue, channel) {
                spin_lock_bh(&rx_queue->add_lock);
                spin_unlock_bh(&rx_queue->add_lock);
@@ -588,7 +588,8 @@ static void efx_fini_channels(struct efx
        struct efx_tx_queue *tx_queue;
        struct efx_rx_queue *rx_queue;
 
-       ASSERT_RTNL();
+       EFX_ASSERT_RESET_SERIALISED(efx);
+       BUG_ON(efx->port_enabled);
 
        efx_for_each_channel(channel, efx) {
                EFX_LOG(channel->efx, "shut down chan %d\n", channel->channel);
@@ -624,6 +625,11 @@ static void efx_remove_channel(struct ef
        channel->used_flags = 0;
 }
 
+void efx_schedule_slow_fill(struct efx_rx_queue *rx_queue, int delay)
+{
+       queue_delayed_work(refill_workqueue, &rx_queue->work, delay);
+}
+
 /**************************************************************************
  *
  * Port handling
@@ -636,12 +642,13 @@ static void efx_remove_channel(struct ef
  */
 static void efx_link_status_changed(struct efx_nic *efx)
 {
-       unsigned long flags __attribute__ ((unused));
        int carrier_ok;
 
-       /* Ensure no link status notifications get sent to the OS after the net
-        * device has been unregistered. */
-       if (!efx->net_dev_registered)
+       /* SFC Bug 5356: A net_dev notifier is registered, so we must ensure
+        * that no events are triggered between unregister_netdev() and the
+        * driver unloading. A more general condition is that NETDEV_CHANGE
+        * can only be generated between NETDEV_UP and NETDEV_DOWN */
+       if (!netif_running(efx->net_dev))
                return;
 
        carrier_ok = netif_carrier_ok(efx->net_dev) ? 1 : 0;
@@ -685,46 +692,50 @@ static void efx_link_status_changed(stru
                         (efx->loopback_mode ? " LOOPBACK]" : ""),
                         (efx->promiscuous ? " [PROMISC]" : ""));
        } else {
-               EFX_INFO(efx, "link down\n");
-       }
-
-}
-
-/* This call reinitialises the MAC to pick up new PHY settings
- * To call from a context that cannot sleep use reconfigure_work work item
- * For on_disabled=1 the caller must be serialised against efx_reset,
- * ideally by holding the rtnl lock.
- */
-void efx_reconfigure_port(struct efx_nic *efx, int on_disabled)
-{
-       mutex_lock(&efx->mac_lock);
-
-       EFX_LOG(efx, "reconfiguring MAC from PHY settings\n");
-
-       if (on_disabled)
-               ASSERT_RTNL();
-       else if (!efx->port_enabled)
-               goto out;
+               EFX_INFO(efx, "link down%s\n",
+                        efx->phy_powered ? "" : " [OFF]");
+       }
+
+}
+
+/* This call reinitialises the MAC to pick up new PHY settings. The
+ * caller must hold the mac_lock */
+static void __efx_reconfigure_port(struct efx_nic *efx)
+{
+       WARN_ON(!mutex_is_locked(&efx->mac_lock));
+
+       EFX_LOG(efx, "reconfiguring MAC from PHY settings on CPU %d\n",
+               raw_smp_processor_id());
 
        efx->mac_op->reconfigure(efx);
 
-out:
        /* Inform kernel of loss/gain of carrier */
        efx_link_status_changed(efx);
-
+}
+
+/* Reinitialise the MAC to pick up new PHY settings, even if the port is
+ * disabled. */
+void efx_reconfigure_port(struct efx_nic *efx)
+{
+       EFX_ASSERT_RESET_SERIALISED(efx);
+
+       mutex_lock(&efx->mac_lock);
+       __efx_reconfigure_port(efx);
        mutex_unlock(&efx->mac_lock);
 }
 
+/* Asynchronous efx_reconfigure_port work item. To speed up efx_flush_all()
+ * we don't efx_reconfigure_port() if the port is disabled. Care is taken
+ * in efx_stop_all() and efx_start_port() to prevent PHY events being lost */
 static void efx_reconfigure_work(struct work_struct *data)
 {
        struct efx_nic *efx = container_of(data, struct efx_nic,
                                           reconfigure_work);
 
-       EFX_LOG(efx, "MAC reconfigure executing on CPU %d\n",
-               raw_smp_processor_id());
-
-       /* Reinitialise MAC to activate new PHY parameters */
-       efx_reconfigure_port(efx, 0);
+       mutex_lock(&efx->mac_lock);
+       if (efx->port_enabled)
+               __efx_reconfigure_port(efx);
+       mutex_unlock(&efx->mac_lock);
 }
 
 static int efx_probe_port(struct efx_nic *efx)
@@ -789,47 +800,34 @@ static int efx_init_port(struct efx_nic 
        return 0;
 }
 
-/* Allow efx_reconfigure_port() to run, and propagate delayed changes
- * to the promiscuous flag to the MAC if needed */
+/* Allow efx_reconfigure_port() to be scheduled, and close the window
+ * between efx_stop_port and efx_flush_all whereby a previously scheduled
+ * efx_reconfigure_port() may have been cancelled */
 static void efx_start_port(struct efx_nic *efx)
 {
        EFX_LOG(efx, "start port\n");
-       ASSERT_RTNL();
-
        BUG_ON(efx->port_enabled);
 
        mutex_lock(&efx->mac_lock);
        efx->port_enabled = 1;
+       __efx_reconfigure_port(efx);
        mutex_unlock(&efx->mac_lock);
-
-       if (efx->net_dev_registered) {
-               int promiscuous;
-
-               netif_tx_lock_bh(efx->net_dev);
-               promiscuous = (efx->net_dev->flags & IFF_PROMISC) ? 1 : 0;
-               if (efx->promiscuous != promiscuous) {
-                       efx->promiscuous = promiscuous;
-                       queue_work(efx->workqueue, &efx->reconfigure_work);
-               }
-               netif_tx_unlock_bh(efx->net_dev);
-       }
-}
-
-/* Prevents efx_reconfigure_port() from executing, and prevents
+}
+
+/* Prevent efx_reconfigure_work and efx_monitor() from executing, and
  * efx_set_multicast_list() from scheduling efx_reconfigure_work.
  * efx_reconfigure_work can still be scheduled via NAPI processing
  * until efx_flush_all() is called */
 static void efx_stop_port(struct efx_nic *efx)
 {
        EFX_LOG(efx, "stop port\n");
-       ASSERT_RTNL();
 
        mutex_lock(&efx->mac_lock);
        efx->port_enabled = 0;
        mutex_unlock(&efx->mac_lock);
 
        /* Serialise against efx_set_multicast_list() */
-       if (efx->net_dev_registered) {
+       if (NET_DEV_REGISTERED(efx)) {
                netif_tx_lock_bh(efx->net_dev);
                netif_tx_unlock_bh(efx->net_dev);
        }
@@ -868,6 +866,7 @@ static int efx_init_io(struct efx_nic *e
 static int efx_init_io(struct efx_nic *efx)
 {
        struct pci_dev *pci_dev = efx->pci_dev;
+       dma_addr_t dma_mask = efx->type->max_dma_mask;
        int rc;
 
        EFX_LOG(efx, "initialising I/O\n");
@@ -886,20 +885,18 @@ static int efx_init_io(struct efx_nic *e
         * (e.g. x86_64 with iommu_sac_force set) will allow 40 bit
         * masks event though they reject 46 bit masks.
         */
-       efx->dma_mask = efx->type->max_dma_mask;
-       while (efx->dma_mask > 0x7fffffffUL) {
-               if (pci_dma_supported(pci_dev, efx->dma_mask) &&
-                   ((rc = pci_set_dma_mask(pci_dev, efx->dma_mask)) == 0))
+       while (dma_mask > 0x7fffffffUL) {
+               if (pci_dma_supported(pci_dev, dma_mask) &&
+                   ((rc = pci_set_dma_mask(pci_dev, dma_mask)) == 0))
                        break;
-               efx->dma_mask >>= 1;
+               dma_mask >>= 1;
        }
        if (rc) {
                EFX_ERR(efx, "could not find a suitable DMA mask\n");
                goto fail2;
        }
-       EFX_LOG(efx, "using DMA mask %llx\n",
-               (unsigned long long)efx->dma_mask);
-       rc = pci_set_consistent_dma_mask(pci_dev, efx->dma_mask);
+       EFX_LOG(efx, "using DMA mask %llx\n", (unsigned long long) dma_mask);
+       rc = pci_set_consistent_dma_mask(pci_dev, dma_mask);
        if (rc) {
                /* pci_set_consistent_dma_mask() is not *allowed* to
                 * fail with a mask that pci_set_dma_mask() accepted,
@@ -912,13 +909,7 @@ static int efx_init_io(struct efx_nic *e
        /* Get memory base address */
        efx->membase_phys = pci_resource_start(efx->pci_dev,
                                               efx->type->mem_bar);
-#if !defined(EFX_HAVE_MSIX_TABLE_RESERVED)
        rc = pci_request_region(pci_dev, efx->type->mem_bar, "sfc");
-#else
-       if (!request_mem_region(efx->membase_phys, efx->type->mem_map_size,
-                               "sfc"))
-               rc = -EIO;
-#endif
        if (rc) {
                EFX_ERR(efx, "request for memory BAR failed\n");
                rc = -EIO;
@@ -960,11 +951,7 @@ static void efx_fini_io(struct efx_nic *
        }
 
        if (efx->membase_phys) {
-#if !defined(EFX_HAVE_MSIX_TABLE_RESERVED)
                pci_release_region(efx->pci_dev, efx->type->mem_bar);
-#else
-               release_mem_region(efx->membase_phys, efx->type->mem_map_size);
-#endif
                efx->membase_phys = 0UL;
        }
 
@@ -972,33 +959,44 @@ static void efx_fini_io(struct efx_nic *
 }
 
 /* Probe the number and type of interrupts we are able to obtain. */
-static int efx_probe_interrupts(struct efx_nic *efx)
-{
+static void efx_probe_interrupts(struct efx_nic *efx)
+{
+       int max_channel = efx->type->phys_addr_channels - 1;
        struct msix_entry xentries[EFX_MAX_CHANNELS];
        int rc, i;
 
-       /* Select number of used RSS queues */
-       /* TODO: Can we react to CPU hotplug? */
-       if (rss_cpus == 0)
-               rss_cpus = num_online_cpus();
-
-       efx->rss_queues = 1;
        if (efx->interrupt_mode == EFX_INT_MODE_MSIX) {
-               unsigned int max_channel = efx->type->phys_addr_channels - 1;
-
                BUG_ON(!pci_find_capability(efx->pci_dev, PCI_CAP_ID_MSIX));
-               efx->rss_queues = min(max_channel + 1, rss_cpus);
+
+               if (rss_cpus == 0) {
+#ifdef topology_core_siblings
+                       cpumask_t core_mask;
+                       int cpu;
+
+                       cpus_clear(core_mask);
+                       efx->rss_queues = 0;
+                       for_each_online_cpu(cpu) {
+                               if (!cpu_isset(cpu, core_mask)) {
+                                       ++efx->rss_queues;
+                                       cpus_or(core_mask, core_mask,
+                                               topology_core_siblings(cpu));
+                               }
+                       }
+#else
+                       efx->rss_queues = num_online_cpus();
+#endif
+               } else {
+                       efx->rss_queues = rss_cpus;
+               }
+
+               /* Limit the number of rss queues appropriately */
+               efx->rss_queues = min(efx->rss_queues, max_channel + 1);
                efx->rss_queues = min(efx->rss_queues, EFX_MAX_CHANNELS);
-       }
-
-       /* Determine how many RSS queues we can use, and mark channels
-        * with the appropriate interrupt state */
-       if (efx->interrupt_mode == EFX_INT_MODE_MSIX) {
-               /* Build MSI request structure */
+
+               /* Request maximum number of MSI interrupts, and fill out
+                * the channel interrupt information the allowed allocation */
                for (i = 0; i < efx->rss_queues; i++)
                        xentries[i].entry = i;
-
-               /* Request maximum number of MSI interrupts */
                rc = pci_enable_msix(efx->pci_dev, xentries, efx->rss_queues);
                if (rc > 0) {
                        EFX_BUG_ON_PARANOID(rc >= efx->rss_queues);
@@ -1006,6 +1004,7 @@ static int efx_probe_interrupts(struct e
                        rc = pci_enable_msix(efx->pci_dev, xentries,
                                             efx->rss_queues);
                }
+
                if (rc == 0) {
                        for (i = 0; i < efx->rss_queues; i++) {
                                efx->channel[i].has_interrupt = 1;
@@ -1033,13 +1032,12 @@ static int efx_probe_interrupts(struct e
 
        /* Assume legacy interrupts */
        if (efx->interrupt_mode == EFX_INT_MODE_LEGACY) {
+               efx->rss_queues = 1;
                /* Every channel is interruptible */
                for (i = 0; i < EFX_MAX_CHANNELS; i++)
                        efx->channel[i].has_interrupt = 1;
                efx->legacy_irq = efx->pci_dev->irq;
        }
-
-       return 0;
 }
 
 static void efx_remove_interrupts(struct efx_nic *efx)
@@ -1059,7 +1057,7 @@ static void efx_remove_interrupts(struct
 /* Select number of used resources
  * Should be called after probe_interrupts()
  */
-static int efx_select_used(struct efx_nic *efx)
+static void efx_select_used(struct efx_nic *efx)
 {
        struct efx_tx_queue *tx_queue;
        struct efx_rx_queue *rx_queue;
@@ -1096,7 +1094,6 @@ static int efx_select_used(struct efx_ni
                        rx_queue++;
                }
        }
-       return 0;
 }
 
 static int efx_probe_nic(struct efx_nic *efx)
@@ -1112,29 +1109,22 @@ static int efx_probe_nic(struct efx_nic 
 
        /* Determine the number of channels and RX queues by trying to hook
         * in MSI-X interrupts. */
-       rc = efx_probe_interrupts(efx);
-       if (rc)
-               goto fail2;
+       efx_probe_interrupts(efx);
 
        /* Determine number of RX queues and TX queues */
-       rc = efx_select_used(efx);
-       if (rc)
-               goto fail3;
+       efx_select_used(efx);
 
        /* Register debugfs entries */
        rc = efx_init_debugfs_nic(efx);
        if (rc)
-               goto fail4;
+               goto fail2;
        /* Initialise the interrupt moderation settings */
        efx_init_irq_moderation(efx, tx_irq_mod_usec, rx_irq_mod_usec);
 
        return 0;
 
- fail4:
-       /* fall-thru */
- fail3:
+ fail2:
        efx_remove_interrupts(efx);
- fail2:
        falcon_remove_nic(efx);
  fail1:
        return rc;
@@ -1190,21 +1180,23 @@ static int efx_probe_all(struct efx_nic 
  fail3:
        efx_for_each_channel(channel, efx)
                efx_remove_channel(channel);
+       efx_remove_port(efx);
  fail2:
-       efx_remove_port(efx);
+       efx_remove_nic(efx);
  fail1:
        return rc;
 }
 
 /* Called after previous invocation(s) of efx_stop_all, restarts the
- * port, kernel transmit queue, NAPI processing and hardware interrupts.
+ * port, kernel transmit queue, NAPI processing and hardware interrupts,
+ * and ensures that the port is scheduled to be reconfigured.
  * This function is safe to call multiple times when the NIC is in any
  * state. */
 static void efx_start_all(struct efx_nic *efx)
 {
        struct efx_channel *channel;
 
-       ASSERT_RTNL();
+       EFX_ASSERT_RESET_SERIALISED(efx);
 
        /* Check that it is appropriate to restart the interface. All
         * of these flags are safe to read under just the rtnl lock */
@@ -1212,7 +1204,7 @@ static void efx_start_all(struct efx_nic
                return;
        if ((efx->state != STATE_RUNNING) && (efx->state != STATE_INIT))
                return;
-       if (efx->net_dev_registered && !netif_running(efx->net_dev))
+       if (NET_DEV_REGISTERED(efx) && !netif_running(efx->net_dev))
                return;
 
        /* Mark the port as enabled so port reconfigurations can start, then
@@ -1227,8 +1219,7 @@ static void efx_start_all(struct efx_nic
 
        /* Start hardware monitor if we're in RUNNING */
        if (efx->state == STATE_RUNNING)
-               queue_delayed_work(efx->workqueue, &efx->monitor_work,
-                                  efx_monitor_interval);
+               queue_work(efx->workqueue, &efx->monitor_work);
 }
 
 /* Flush all delayed work. Should only be called when no more delayed work
@@ -1236,24 +1227,6 @@ static void efx_start_all(struct efx_nic
  * since we're holding the rtnl_lock at this point. */
 static void efx_flush_all(struct efx_nic *efx)
 {
-#if defined(EFX_USE_CANCEL_DELAYED_WORK_SYNC)
-       struct efx_rx_queue *rx_queue;
-
-       /* Make sure the hardware monitor is stopped */
-       cancel_delayed_work_sync(&efx->monitor_work);
-
-       /* Ensure that all RX slow refills are complete. */
-       efx_for_each_rx_queue(rx_queue, efx) {
-               cancel_delayed_work_sync(&rx_queue->work);
-       }
-#endif
-
-#if defined(EFX_USE_CANCEL_WORK_SYNC)
-       /* Stop scheduled port reconfigurations */
-       cancel_work_sync(&efx->reconfigure_work);
-#endif
-
-#if !defined(EFX_USE_CANCEL_DELAYED_WORK_SYNC)
        /* Ensure that the hardware monitor and asynchronous port
         * reconfigurations are complete, which are the only two consumers
         * of efx->workqueue. Since the hardware monitor runs on a long period,
@@ -1270,9 +1243,8 @@ static void efx_flush_all(struct efx_nic
 
        /* efx_rx_work will disarm if !channel->enabled, so we can just
         * flush the refill workqueue twice as well. */
-       flush_workqueue(efx->refill_workqueue);
-       flush_workqueue(efx->refill_workqueue);
-#endif
+       flush_workqueue(refill_workqueue);
+       flush_workqueue(refill_workqueue);
 }
 
 /* Quiesce hardware and software without bringing the link down.
@@ -1284,7 +1256,7 @@ static void efx_stop_all(struct efx_nic 
 {
        struct efx_channel *channel;
 
-       ASSERT_RTNL();
+       EFX_ASSERT_RESET_SERIALISED(efx);
 
        /* port_enabled can be read safely under the rtnl lock */
        if (!efx->port_enabled)
@@ -1298,20 +1270,27 @@ static void efx_stop_all(struct efx_nic 
                if (channel->irq)
                        synchronize_irq(channel->irq);
 
-       /* Stop all synchronous port reconfigurations. */
-       efx_stop_port(efx);
-
        /* Stop all NAPI processing and synchronous rx refills */
        efx_for_each_channel(channel, efx)
                efx_stop_channel(channel);
 
+       /* Stop all asynchronous port reconfigurations. Since all
+        * event processing has already been stopped, there is no
+        * window to loose phy events */
+       efx_stop_port(efx);
+
        /* Flush reconfigure_work, refill_workqueue, monitor_work */
        efx_flush_all(efx);
+
+       /* Isolate the MAC from the TX and RX engines, so that queue
+        * flushes will complete in a timely fashion. */
+       falcon_deconfigure_mac_wrapper(efx);
+       falcon_drain_tx_fifo(efx);
 
        /* Stop the kernel transmit interface late, so the watchdog
         * timer isn't ticking over the flush */
        efx_stop_queue(efx);
-       if (efx->net_dev_registered) {
+       if (NET_DEV_REGISTERED(efx)) {
                netif_tx_lock_bh(efx->net_dev);
                netif_tx_unlock_bh(efx->net_dev);
        }
@@ -1360,16 +1339,15 @@ static int efx_run_selftests(struct efx_
        return rc;
 }
 
+/* A convinience function to safely flush all the queues */
 int efx_flush_queues(struct efx_nic *efx)
 {
        int rc;
 
-       ASSERT_RTNL();
+       EFX_ASSERT_RESET_SERIALISED(efx);
 
        efx_stop_all(efx);
 
-       /* We can't just flush the tx queues because the event queues
-        * may contain tx completions from that queue. Just flush everything */
        efx_fini_channels(efx);
        rc = efx_init_channels(efx);
        if (rc) {
@@ -1394,7 +1372,7 @@ void efx_init_irq_moderation(struct efx_
        struct efx_tx_queue *tx_queue;
        struct efx_rx_queue *rx_queue;
 
-       ASSERT_RTNL();
+       EFX_ASSERT_RESET_SERIALISED(efx);
 
        efx_for_each_tx_queue(tx_queue, efx)
                tx_queue->channel->irq_moderation = tx_usecs;
@@ -1413,20 +1391,13 @@ void efx_init_irq_moderation(struct efx_
  * efx_reconfigure_port via the mac_lock */
 static void efx_monitor(struct work_struct *data)
 {
-#if !defined(EFX_NEED_WORK_API_WRAPPERS)
-       struct efx_nic *efx = container_of(data, struct efx_nic,
-                                          monitor_work.work);
-#else
        struct efx_nic *efx = container_of(data, struct efx_nic,
                                           monitor_work);
-#endif
        int rc = 0;
 
        EFX_TRACE(efx, "hardware monitor executing on CPU %d\n",
                  raw_smp_processor_id());
 
-
-#if !defined(EFX_USE_CANCEL_DELAYED_WORK_SYNC)
        /* Without cancel_delayed_work_sync(), we have to make sure that
         * we don't rearm when port_enabled == 0 */
        mutex_lock(&efx->mac_lock);
@@ -1436,19 +1407,6 @@ static void efx_monitor(struct work_stru
        }
 
        rc = efx->mac_op->check_hw(efx);
-#else
-       /* If the mac_lock is already held then it is likely a port
-        * reconfiguration is already in place, which will likely do
-        * most of the work of check_hw() anyway. */
-       if (!mutex_trylock(&efx->mac_lock)) {
-               queue_delayed_work(efx->workqueue, &efx->monitor_work,
-                                  efx_monitor_interval);
-               return;
-       }
-
-       if (efx->port_enabled)
-               rc = efx->mac_op->check_hw(efx);
-#endif
        mutex_unlock(&efx->mac_lock);
 
        if (rc) {
@@ -1478,24 +1436,11 @@ static int efx_ioctl(struct net_device *
 static int efx_ioctl(struct net_device *net_dev, struct ifreq *ifr, int cmd)
 {
        struct efx_nic *efx = net_dev->priv;
-       int rc;
-
-       ASSERT_RTNL();
-
-       switch (cmd) {
-       case SIOCGMIIPHY:
-       case SIOCGMIIREG:
-               rc = generic_mii_ioctl(&efx->mii, if_mii(ifr), cmd, NULL);
-               break;
-       case SIOCSMIIREG:
-               rc = generic_mii_ioctl(&efx->mii, if_mii(ifr), cmd, NULL);
-               efx_reconfigure_port(efx, 0);
-               break;
-       default:
-               rc = -EOPNOTSUPP;
-       }
-
-       return rc;
+
+       if (!in_interrupt())
+           EFX_ASSERT_RESET_SERIALISED(efx);
+
+       return generic_mii_ioctl(&efx->mii, if_mii(ifr), cmd, NULL);
 }
 
 /**************************************************************************
@@ -1511,8 +1456,6 @@ static int efx_init_napi(struct efx_nic 
 {
        struct efx_channel *channel;
        int rc;
-
-       ASSERT_RTNL();
 
        /* Allocate the NAPI dev for the port */
        efx->net_dev = alloc_etherdev(0);
@@ -1523,29 +1466,15 @@ static int efx_init_napi(struct efx_nic 
        efx->net_dev->priv = efx;
        efx->mii.dev = efx->net_dev;
 
-       /* Set features based on module parameters and DMA mask.
-        * Enable DMA to ZONE_HIGHMEM if the NIC can access all memory
-        * directly.  This only has an effect on 32-bit systems and
-        * PAE on x86 limits memory to 64GB so 40 bits is plenty to
-        * address everything.  If the device can't address 40 bits
-        * then it's safest to turn NETIF_F_HIGHDMA off because this
-        * might be a PAE system with more than 4G of RAM and a 32-bit
-        * NIC.  The use of EFX_DMA_MASK is to eliminate compiler
-        * warnings on platforms where dma_addr_t is 32-bit.  We
-        * assume that in those cases we can access all memory
-        * directly if our DMA mask is all ones. */
-       efx->net_dev->features |= NETIF_F_IP_CSUM | NETIF_F_SG;
-       if (efx->dma_mask >= EFX_DMA_MASK(DMA_40BIT_MASK))
-               efx->net_dev->features |= NETIF_F_HIGHDMA;
+       efx->net_dev->features |= (NETIF_F_IP_CSUM | NETIF_F_SG |
+                                  NETIF_F_HIGHDMA);
+       efx->lro_enabled = lro;
 
        /* Copy MAC address */
        memcpy(&efx->net_dev->dev_addr, efx->mac_address, ETH_ALEN);
 
        /* Allocate the per channel devs */
        efx_for_each_channel(channel, efx) {
-#if !defined(EFX_HAVE_OLD_NAPI)
-               channel->napi_dev = efx->net_dev;
-#else
                channel->napi_dev = alloc_etherdev(0);
                if (!channel->napi_dev) {
                        rc = -ENOMEM;
@@ -1553,7 +1482,11 @@ static int efx_init_napi(struct efx_nic 
                }
                channel->napi_dev->priv = channel;
                atomic_set(&channel->napi_dev->refcnt, 1);
-#endif
+
+               /* Initialise LRO/SSR */
+               rc = efx_ssr_init(&channel->ssr, efx);
+               if (rc)
+                       goto err;
        }
 
        return 0;
@@ -1567,16 +1500,15 @@ static void efx_fini_napi(struct efx_nic
 {
        struct efx_channel *channel;
 
-       ASSERT_RTNL();
-
        efx_for_each_channel(channel, efx) {
+               /* Fini LRO/SSR */
+               efx_ssr_fini(&channel->ssr);
+
                /* Finish per channel NAPI */
-#if defined(EFX_HAVE_OLD_NAPI)
                if (channel->napi_dev) {
                        channel->napi_dev->priv = NULL;
                        free_netdev(channel->napi_dev);
                }
-#endif
                channel->napi_dev = NULL;
        }
 
@@ -1621,10 +1553,11 @@ static int efx_net_open(struct net_devic
 static int efx_net_open(struct net_device *net_dev)
 {
        struct efx_nic *efx = net_dev->priv;
-       ASSERT_RTNL();
+       EFX_ASSERT_RESET_SERIALISED(efx);
 
        EFX_LOG(efx, "opening device %s on CPU %d\n", net_dev->name,
                raw_smp_processor_id());
+
        efx_start_all(efx);
        return 0;
 }
@@ -1641,7 +1574,7 @@ static int efx_net_stop(struct net_devic
        EFX_LOG(efx, "closing %s on CPU %d\n", net_dev->name,
                raw_smp_processor_id());
 
-       /* Stop device and flush all the channels */
+       /* Stop the device and flush all the channels */
        efx_stop_all(efx);
        efx_fini_channels(efx);
        rc = efx_init_channels(efx);
@@ -1651,9 +1584,7 @@ static int efx_net_stop(struct net_devic
        return 0;
 }
 
-/* Context: process, dev_base_lock held, non-blocking.
- * Statistics are taken directly from the MAC.
- */
+/* Context: process, dev_base_lock held, non-blocking. */
 static struct net_device_stats *efx_net_stats(struct net_device *net_dev)
 {
        struct efx_nic *efx = net_dev->priv;
@@ -1662,23 +1593,27 @@ static struct net_device_stats *efx_net_
 
        if (!spin_trylock(&efx->stats_lock))
                return stats;
-       if (efx->state == STATE_RUNNING)
+       if (efx->state == STATE_RUNNING) {
                efx->mac_op->update_stats(efx);
+               falcon_update_nic_stats(efx);
+       }
        spin_unlock(&efx->stats_lock);
 
        stats->rx_packets = mac_stats->rx_packets;
        stats->tx_packets = mac_stats->tx_packets;
        stats->rx_bytes = mac_stats->rx_bytes;
        stats->tx_bytes = mac_stats->tx_bytes;
-       stats->tx_errors = mac_stats->tx_bad;
        stats->multicast = mac_stats->rx_multicast;
        stats->collisions = mac_stats->tx_collision;
-       stats->rx_length_errors = mac_stats->rx_gtjumbo;
-       stats->rx_over_errors = mac_stats->rx_overflow;
+       stats->rx_length_errors = (mac_stats->rx_gtjumbo +
+                                  mac_stats->rx_length_error);
+       stats->rx_over_errors = efx->n_rx_nodesc_drop_cnt;
        stats->rx_crc_errors = mac_stats->rx_bad;
        stats->rx_frame_errors = mac_stats->rx_align_error;
-       stats->rx_fifo_errors = 0;
+       stats->rx_fifo_errors = mac_stats->rx_overflow;
        stats->rx_missed_errors = mac_stats->rx_missed;
+       stats->tx_window_errors = mac_stats->tx_late_collision;
+
        stats->rx_errors = (stats->rx_length_errors +
                            stats->rx_over_errors +
                            stats->rx_crc_errors +
@@ -1686,11 +1621,8 @@ static struct net_device_stats *efx_net_
                            stats->rx_fifo_errors +
                            stats->rx_missed_errors +
                            mac_stats->rx_symbol_error);
-       stats->tx_aborted_errors = 0;
-       stats->tx_carrier_errors = 0;
-       stats->tx_fifo_errors = 0;
-       stats->tx_heartbeat_errors = 0;
-       stats->tx_window_errors = 0;
+       stats->tx_errors = (stats->tx_window_errors +
+                           mac_stats->tx_bad);
 
        return stats;
 }
@@ -1715,7 +1647,7 @@ static int efx_change_mtu(struct net_dev
        struct efx_nic *efx = net_dev->priv;
        int rc = 0;
 
-       ASSERT_RTNL();
+       EFX_ASSERT_RESET_SERIALISED(efx);
 
        if (new_mtu > EFX_MAX_MTU)
                return -EINVAL;
@@ -1738,15 +1670,11 @@ static int efx_change_mtu(struct net_dev
        if (rc)
                goto fail;
 
-       /* Reconfigure the MAC */
-       efx_reconfigure_port(efx, 1);
-
        /* Notify driverlink client of new MTU */
        EFX_DL_CALLBACK(efx, mtu_changed, new_mtu);
 
+ out:
        efx_start_all(efx);
-
- out:
        return rc;
 
  fail:
@@ -1760,7 +1688,7 @@ static int efx_set_mac_address(struct ne
        struct sockaddr *addr = data;
        char *new_addr = addr->sa_data;
 
-       ASSERT_RTNL();
+       EFX_ASSERT_RESET_SERIALISED(efx);
 
        if (!is_valid_ether_addr(new_addr)) {
                DECLARE_MAC_BUF(mac);
@@ -1772,7 +1700,7 @@ static int efx_set_mac_address(struct ne
        memcpy(net_dev->dev_addr, new_addr, net_dev->addr_len);
 
        /* Reconfigure the MAC */
-       efx_reconfigure_port(efx, 1);
+       efx_reconfigure_port(efx);
 
        return 0;
 }
@@ -1783,7 +1711,6 @@ static void efx_set_multicast_list(struc
        struct efx_nic *efx = net_dev->priv;
        struct dev_mc_list *mc_list = net_dev->mc_list;
        union efx_multicast_hash *mc_hash = &efx->multicast_hash;
-       unsigned long flags __attribute__ ((unused));
        int promiscuous;
        u32 crc;
        int bit;
@@ -1792,10 +1719,11 @@ static void efx_set_multicast_list(struc
        /* Set per-MAC promiscuity flag and reconfigure MAC if necessary */
        promiscuous = (net_dev->flags & IFF_PROMISC) ? 1 : 0;
        if (efx->promiscuous != promiscuous) {
-               if (efx->port_enabled) {
-                       efx->promiscuous = promiscuous;
+               efx->promiscuous = promiscuous;
+               /* Close the window between efx_stop_port() and efx_flush_all()
+                * by only queuing work when the port is enabled. */
+               if (efx->port_enabled)
                        queue_work(efx->workqueue, &efx->reconfigure_work);
-               }
        }
 
        /* Build multicast hash table */
@@ -1805,8 +1733,8 @@ static void efx_set_multicast_list(struc
                memset(mc_hash, 0x00, sizeof(*mc_hash));
                for (i = 0; i < net_dev->mc_count; i++) {
                        crc = ether_crc_le(ETH_ALEN, mc_list->dmi_addr);
-                       bit = (crc & ((1 << EFX_MCAST_HASH_BITS) - 1));
-                       set_bit_le(bit, (void *)mc_hash);
+                       bit = crc & (EFX_MCAST_HASH_ENTRIES - 1);
+                       set_bit_le(bit, mc_hash->byte);
                        mc_list = mc_list->next;
                }
        }
@@ -1835,6 +1763,21 @@ static struct notifier_block efx_netdev_
 static struct notifier_block efx_netdev_notifier = {
        .notifier_call = efx_netdev_event,
 };
+
+/* Prior to Linux 2.6.24, the bonding driver may call change_mtu()
+ * without holding the RTNL, unlike all other callers.  We try to
+ * mitigate the risk of a race with other reconfiguration using
+ * rtnl_trylock(), but we cannot eliminate it completely.
+ */
+static int efx_locked_change_mtu(struct net_device *net_dev, int new_mtu)
+{
+       int must_unlock = rtnl_trylock();
+       int rc = efx_change_mtu(net_dev, new_mtu);
+       if (must_unlock)
+               rtnl_unlock();
+       return rc;
+}
+#define efx_change_mtu efx_locked_change_mtu
 
 static int efx_register_netdev(struct efx_nic *efx)
 {
@@ -1861,8 +1804,6 @@ static int efx_register_netdev(struct ef
        /* Always start with carrier off; PHY events will detect the link */
        netif_carrier_off(efx->net_dev);
 
-       BUG_ON(efx->net_dev_registered);
-
        /* Clear MAC statistics */
        efx->mac_op->update_stats(efx);
        memset(&efx->mac_stats, 0, sizeof(efx->mac_stats));
@@ -1882,43 +1823,17 @@ static int efx_register_netdev(struct ef
                return rc;
        }
 
-       /* Allow link change notifications to be sent to the operating
-        * system.  The must happen after register_netdev so that
-        * there are no outstanding link changes if that call fails.
-        * It must happen before efx_reconfigure_port so that the
-        * initial state of the link is reported. */
-       mutex_lock(&efx->mac_lock);
-       efx->net_dev_registered = 1;
-       mutex_unlock(&efx->mac_lock);
-
-       /* Safety net: in case we don't get a PHY event */
-       rtnl_lock();
-       efx_reconfigure_port(efx, 1);
-       rtnl_unlock();
-
-       EFX_LOG(efx, "registered\n");
-
        return 0;
 }
 
 static void efx_unregister_netdev(struct efx_nic *efx)
 {
-       int was_registered = efx->net_dev_registered;
        struct efx_tx_queue *tx_queue;
 
        if (!efx->net_dev)
                return;
 
        BUG_ON(efx->net_dev->priv != efx);
-
-       /* SFC Bug 5356: Ensure that no more link status notifications get
-        * sent to the stack.  Bad things happen if there's an
-        * outstanding notification after the net device is freed, but
-        * they only get flushed out by unregister_netdev, not by
-        * free_netdev. */
-       mutex_lock(&efx->mac_lock);
-       efx->net_dev_registered = 0;
-       mutex_unlock(&efx->mac_lock);
 
        /* Free up any skbs still remaining. This has to happen before
         * we try to unregister the netdev as running their destructors
@@ -1926,7 +1841,7 @@ static void efx_unregister_netdev(struct
        efx_for_each_tx_queue(tx_queue, efx)
                efx_release_tx_buffers(tx_queue);
 
-       if (was_registered) {
+       if (NET_DEV_REGISTERED(efx)) {
                strlcpy(efx->name, pci_name(efx->pci_dev), sizeof(efx->name));
                efx_fini_debugfs_netdev(efx->net_dev);
                unregister_netdev(efx->net_dev);
@@ -1939,15 +1854,11 @@ static void efx_unregister_netdev(struct
  *
  **************************************************************************/
 
-/* This suspends the device (and acquires the suspend lock) without
- * flushing the descriptor queues.  It is included for the convenience
- * of the driverlink layer.
- */
+/* Serialise access to the driverlink callbacks, by quiescing event processing
+ * (without flushing the descriptor queues), and acquiring the rtnl_lock */
 void efx_suspend(struct efx_nic *efx)
 {
        EFX_LOG(efx, "suspending operations\n");
-
-       down(&efx->suspend_lock);
 
        rtnl_lock();
        efx_stop_all(efx);
@@ -1959,8 +1870,6 @@ void efx_resume(struct efx_nic *efx)
 
        efx_start_all(efx);
        rtnl_unlock();
-
-       up(&efx->suspend_lock);
 }
 
 /* The final hardware and software finalisation before reset.
@@ -1970,7 +1879,7 @@ static int efx_reset_down(struct efx_nic
 {
        int rc;
 
-       ASSERT_RTNL();
+       EFX_ASSERT_RESET_SERIALISED(efx);
 
        rc = efx->mac_op->get_settings(efx, ecmd);
        if (rc) {
@@ -1995,10 +1904,6 @@ static int efx_reset_up(struct efx_nic *
        rc = efx_init_channels(efx);
        if (rc)
                goto fail1;
-
-       /* In an INVISIBLE_RESET there might not be a link state transition,
-        * so we push the multicast list here. */
-       falcon_set_multicast_hash(efx);
 
        /* Restore MAC and PHY settings. */
        rc = efx->mac_op->set_settings(efx, ecmd);
@@ -2021,35 +1926,29 @@ static int efx_reset_up(struct efx_nic *
  *
  * This function will sleep.  You cannot reset from within an atomic
  * state; use efx_schedule_reset() instead.
+ *
+ * Grabs the dl_reset_lock, and to serialise with kernel interfaces the
+ * rtnl_lock.
  */
 static int efx_reset(struct efx_nic *efx)
 {
        struct ethtool_cmd ecmd;
-       unsigned long flags __attribute__ ((unused));
        enum reset_type method = efx->reset_pending;
        int rc;
 
+       /* Notify driverlink clients of imminent reset. */
        efx_dl_reset_lock();
-
-       rc = down_interruptible(&efx->suspend_lock);
-       if (rc) {
-               EFX_ERR(efx, "reset aborted by signal\n");
-               goto unlock_dl_lock;
-       }
-
-       /* We've got suspend_lock, which means we can only be in
-        * STATE_RUNNING or STATE_FINI. Don't clear
-        * efx->reset_pending, since this flag indicates that we
-        * should retry device initialisation.
-        */
+       efx_dl_reset_suspend(efx);
+
+       /* Serialise with kernel interfaces */
+       rtnl_lock();
+
+       /* If we're not RUNNING then don't reset. Leave the reset_pending
+        * flag set so that efx_pci_probe_main will be retried */
        if (efx->state != STATE_RUNNING) {
                EFX_INFO(efx, "scheduled reset quenched. NIC not RUNNING\n");
-               goto unlock_suspend_lock;
-       }
-
-       /* Notify driverlink clients of imminent reset. */
-       efx_dl_reset_suspend(efx);
-       rtnl_lock();
+               goto unlock_rtnl;
+       }
 
        efx->state = STATE_RESETTING;
        EFX_INFO(efx, "resetting (%s)\n", RESET_TYPE(method));
@@ -2104,15 +2003,16 @@ static int efx_reset(struct efx_nic *efx
                goto fail5;
 
        mutex_unlock(&efx->mac_lock);
-       efx_reconfigure_port(efx, 1);
        EFX_LOG(efx, "reset complete\n");
 
        efx->state = STATE_RUNNING;
        efx_start_all(efx);
 
+ unlock_rtnl:
        rtnl_unlock();
-
-       goto notify;
+       efx_dl_reset_resume(efx, 1);
+       efx_dl_reset_unlock();
+       return 0;
 
  fail5:
  fail4:
@@ -2122,22 +2022,13 @@ static int efx_reset(struct efx_nic *efx
        EFX_ERR(efx, "has been disabled\n");
        efx->state = STATE_DISABLED;
 
-       /* Remove the net_dev */
        mutex_unlock(&efx->mac_lock);
        rtnl_unlock();
+       /* Remove the net_dev */
        efx_unregister_netdev(efx);
        efx_fini_port(efx);
-
- notify:
-       /* Notify driverlink clients of completed reset */
-       efx_dl_reset_resume(efx, (rc == 0));
-
- unlock_suspend_lock:
-       up(&efx->suspend_lock);
-
- unlock_dl_lock:
+       efx_dl_reset_resume(efx, 0);
        efx_dl_reset_unlock();
-
        return rc;
 }
 
@@ -2170,6 +2061,7 @@ void efx_schedule_reset(struct efx_nic *
        case RESET_TYPE_RX_RECOVERY:
        case RESET_TYPE_RX_DESC_FETCH:
        case RESET_TYPE_TX_DESC_FETCH:
+       case RESET_TYPE_TX_SKIP:
                method = RESET_TYPE_INVISIBLE;
                break;
        default:
@@ -2185,11 +2077,7 @@ void efx_schedule_reset(struct efx_nic *
 
        efx->reset_pending = method;
 
-#if !defined(EFX_USE_CANCEL_DELAYED_WORK_SYNC)
        queue_work(efx->reset_workqueue, &efx->reset_work);
-#else
-       queue_work(efx->workqueue, &efx->reset_work);
-#endif
 }
 
 /**************************************************************************
@@ -2197,24 +2085,13 @@ void efx_schedule_reset(struct efx_nic *
  * List of NICs we support
  *
  **************************************************************************/
-
-enum efx_type_index {
-       EFX_TYPE_FALCON_A = 0,
-       EFX_TYPE_FALCON_B = 1,
-};
-
-static struct efx_nic_type *efx_nic_types[] = {
-       [EFX_TYPE_FALCON_A] = &falcon_a_nic_type,
-       [EFX_TYPE_FALCON_B] = &falcon_b_nic_type,
-};
-
 
 /* PCI device ID table */
 static struct pci_device_id efx_pci_table[] __devinitdata = {
-       {EFX_VENDID_SFC, FALCON_A_P_DEVID, PCI_ANY_ID, PCI_ANY_ID,
-        0, 0, EFX_TYPE_FALCON_A},
-       {EFX_VENDID_SFC, FALCON_B_P_DEVID, PCI_ANY_ID, PCI_ANY_ID,
-        0, 0, EFX_TYPE_FALCON_B},
+       {PCI_DEVICE(EFX_VENDID_SFC, FALCON_A_P_DEVID),
+        .driver_data = (unsigned long) &falcon_a_nic_type},
+       {PCI_DEVICE(EFX_VENDID_SFC, FALCON_B_P_DEVID),
+        .driver_data = (unsigned long) &falcon_b_nic_type},
        {0}                     /* end of list */
 };
 
@@ -2275,7 +2152,7 @@ static struct efx_board efx_dummy_board_
 /* This zeroes out and then fills in the invariants in a struct
  * efx_nic (including all sub-structures).
  */
-static int efx_init_struct(struct efx_nic *efx, enum efx_type_index type,
+static int efx_init_struct(struct efx_nic *efx, struct efx_nic_type *type,
                           struct pci_dev *pci_dev)
 {
        struct efx_channel *channel;
@@ -2288,7 +2165,6 @@ static int efx_init_struct(struct efx_ni
        spin_lock_init(&efx->biu_lock);
        spin_lock_init(&efx->phy_lock);
        mutex_init(&efx->spi_lock);
-       sema_init(&efx->suspend_lock, 1);
        INIT_WORK(&efx->reset_work, efx_reset_work);
        INIT_DELAYED_WORK(&efx->monitor_work, efx_monitor);
        efx->pci_dev = pci_dev;
@@ -2333,7 +2209,7 @@ static int efx_init_struct(struct efx_ni
                INIT_DELAYED_WORK(&rx_queue->work, efx_rx_work);
        }
 
-       efx->type = efx_nic_types[type];
+       efx->type = type;
 
        /* Sanity-check NIC type */
        EFX_BUG_ON_PARANOID(efx->type->txd_ring_mask &
@@ -2352,63 +2228,39 @@ static int efx_init_struct(struct efx_ni
        /* Higher numbered interrupt modes are less capable! */
        efx->interrupt_mode = max(efx->type->max_interrupt_mode,
                                  interrupt_mode);
-#if defined(EFX_NEED_DUMMY_MSIX)
-       if (efx->interrupt_mode == EFX_INT_MODE_MSIX)
-               efx->interrupt_mode = EFX_INT_MODE_MSI;
-#endif
-
-       /* Tasks that can fail are last */
-       efx->refill_workqueue = create_workqueue("sfc_refill");
-       if (!efx->refill_workqueue) {
-               rc = -ENOMEM;
-               goto fail1;
-       }
 
        efx->workqueue = create_singlethread_workqueue("sfc_work");
        if (!efx->workqueue) {
                rc = -ENOMEM;
-               goto fail2;
-       }
-
-#if !defined(EFX_USE_CANCEL_DELAYED_WORK_SYNC)
+               goto fail1;
+       }
+
        efx->reset_workqueue = create_singlethread_workqueue("sfc_reset");
        if (!efx->reset_workqueue) {
                rc = -ENOMEM;
-               goto fail3;
-       }
-#endif
+               goto fail2;
+       }
 
        return 0;
 
-#if !defined(EFX_USE_CANCEL_DELAYED_WORK_SYNC)
- fail3:
+ fail2:
        destroy_workqueue(efx->workqueue);
        efx->workqueue = NULL;
-#endif
-
- fail2:
-       destroy_workqueue(efx->refill_workqueue);
-       efx->refill_workqueue = NULL;
+
  fail1:
        return rc;
 }
 
 static void efx_fini_struct(struct efx_nic *efx)
 {
-#if !defined(EFX_USE_CANCEL_DELAYED_WORK_SYNC)
        if (efx->reset_workqueue) {
                destroy_workqueue(efx->reset_workqueue);
                efx->reset_workqueue = NULL;
        }
-#endif
        if (efx->workqueue) {
                destroy_workqueue(efx->workqueue);
                efx->workqueue = NULL;
        }
-       if (efx->refill_workqueue) {
-               destroy_workqueue(efx->refill_workqueue);
-               efx->refill_workqueue = NULL;
-       }
 }
 
 /**************************************************************************
@@ -2422,7 +2274,7 @@ static void efx_fini_struct(struct efx_n
  */
 static void efx_pci_remove_main(struct efx_nic *efx)
 {
-       ASSERT_RTNL();
+       EFX_ASSERT_RESET_SERIALISED(efx);
 
        /* Skip everything if we never obtained a valid membase */
        if (!efx->membase)
@@ -2456,43 +2308,29 @@ static void efx_pci_remove(struct pci_de
        /* Unregister driver from driverlink layer */
        efx_dl_unregister_nic(efx);
 
-       /* Mark the NIC as fini under both suspend_lock and
-        * rtnl_lock */
-       down(&efx->suspend_lock);
+       /* Mark the NIC as fini, then stop the interface */
        rtnl_lock();
        efx->state = STATE_FINI;
-       up(&efx->suspend_lock);
-
-       if (efx->membase) {
-               /* Stop the NIC. Since we're in STATE_FINI, this
-                * won't be reversed. */
-               if (efx->net_dev_registered)
-                       dev_close(efx->net_dev);
-
-               /* Release the rtnl lock. Any queued efx_resets()
-                * can now return early [we're in STATE_FINI]. */
-               rtnl_unlock();
-
-               efx_unregister_netdev(efx);
-               efx_fini_debugfs_channels(efx);
-
-               /* Wait for any scheduled resets to complete. No more will be
-                * scheduled from this point because efx_stop_all() has been
-                * called, we are no longer registered with driverlink, and
-                * the net_device's have been removed. */
-#if !defined(EFX_USE_CANCEL_DELAYED_WORK_SYNC)
-               flush_workqueue(efx->reset_workqueue);
-#else
-               flush_workqueue(efx->workqueue);
-#endif
-
-               /* Fini and remove all the software state */
-               rtnl_lock();
-               efx_pci_remove_main(efx);
-       }
-
+       dev_close(efx->net_dev);
+
+       /* Allow any queued efx_resets() to complete */
        rtnl_unlock();
 
+       if (efx->membase == NULL)
+               goto out;
+
+       efx_unregister_netdev(efx);
+       efx_fini_debugfs_channels(efx);
+
+       /* Wait for any scheduled resets to complete. No more will be
+        * scheduled from this point because efx_stop_all() has been
+        * called, we are no longer registered with driverlink, and
+        * the net_device's have been removed. */
+       flush_workqueue(efx->reset_workqueue);
+
+       efx_pci_remove_main(efx);
+
+out:
        efx_fini_io(efx);
        EFX_LOG(efx, "shutdown successful\n");
 
@@ -2593,7 +2431,7 @@ static int __devinit efx_pci_probe(struc
                                   const struct pci_device_id *entry)
 {
        struct efx_nic *efx;
-       enum efx_type_index type = entry->driver_data;
+       struct efx_nic_type *type = (struct efx_nic_type *) entry->driver_data;
        int i, rc;
 
        /* Allocate and initialise a struct efx_nic */
@@ -2614,50 +2452,27 @@ static int __devinit efx_pci_probe(struc
        if (rc)
                goto fail3;
 
-       /* From this point on we begin to expose the driver to the OS
-        * to varying degrees, so lets grab the suspend_lock and
-        * rtnl_lock to serialise against efx_reset() and
-        * friends. efx->state is not STATE_RUNNING yet, but we don't
-        * want these tasks to fail, just to block until we drop the
-        * lock
-        */
-       rc = down_interruptible(&efx->suspend_lock);
-       if (rc) {
-               EFX_ERR(efx, "suspend interrupted - aborting\n");
-               goto fail4;
-       }
-
-       rtnl_lock();
-
-       /* Probe, initialise and start everything. Run self-test */
+       /* No serialisation is required with the reset path because
+        * we're in STATE_INIT. */
        for (i = 0; i < 5; i++) {
                rc = efx_pci_probe_main(efx);
                if (rc == 0)
                        break;
 
-               /* Retry if a recoverably reset event has been scheduled */
-               if ((efx->reset_pending != RESET_TYPE_INVISIBLE) &&
-                   (efx->reset_pending != RESET_TYPE_ALL))
-                       goto fail5;
-
                /* Serialise against efx_reset(). No more resets will be
                 * scheduled since efx_stop_all() has been called, and we
                 * have not and never have been registered with either
                 * the rtnetlink or driverlink layers. */
-               rtnl_unlock();
-               up(&efx->suspend_lock);
-
-#if defined(EFX_USE_CANCEL_WORK_SYNC)
-               cancel_work_sync(&efx->reset_work);
-#else
                flush_workqueue(efx->reset_workqueue);
-#endif
-
-               down(&efx->suspend_lock);
-               rtnl_lock();
+
+               /* Retry if a recoverably reset event has been scheduled */
+               if ((efx->reset_pending != RESET_TYPE_INVISIBLE) &&
+                   (efx->reset_pending != RESET_TYPE_ALL))
+                       goto fail4;
 
                efx->reset_pending = RESET_TYPE_NONE;
-       };
+       }
+
        if (rc) {
                EFX_ERR(efx, "Could not reset NIC\n");
                goto fail5;
@@ -2671,15 +2486,13 @@ static int __devinit efx_pci_probe(struc
        /* Switch to the running state before we expose the device to
         * the OS.  This is to ensure that the initial gathering of
         * MAC stats succeeds. */
+       rtnl_lock();
        efx->state = STATE_RUNNING;
-
        rtnl_unlock();
 
        rc = efx_register_netdev(efx);
        if (rc)
                goto fail7;
-
-       up(&efx->suspend_lock);
 
        EFX_LOG(efx, "initialisation successful\n");
 
@@ -2691,18 +2504,12 @@ static int __devinit efx_pci_probe(struc
        return 0;
 
  fail8:
-       down(&efx->suspend_lock);
        efx_unregister_netdev(efx);
  fail7:
-       /* Re-acquire the rtnl lock around pci_remove_main() */
-       rtnl_lock();
        efx_fini_debugfs_channels(efx);
  fail6:
        efx_pci_remove_main(efx);
  fail5:
-       /* Drop the locks before fini */
-       rtnl_unlock();
-       up(&efx->suspend_lock);
  fail4:
        efx_fini_io(efx);
  fail3:
@@ -2749,6 +2556,12 @@ static int __init efx_init_module(void)
        if (rc)
                goto err_notifier;
 
+       refill_workqueue = create_workqueue("sfc_refill");
+       if (!refill_workqueue) {
+               rc = -ENOMEM;
+               goto err_refill;
+       }
+
        rc = pci_register_driver(&efx_pci_driver);
        if (rc < 0)
                goto err_pci;
@@ -2756,6 +2569,8 @@ static int __init efx_init_module(void)
        return 0;
 
  err_pci:
+       destroy_workqueue(refill_workqueue);
+ err_refill:
        unregister_netdevice_notifier(&efx_netdev_notifier);
  err_notifier:
        efx_fini_debugfs();
@@ -2768,6 +2583,7 @@ static void __exit efx_exit_module(void)
        printk(KERN_INFO "Solarflare NET driver unloading\n");
 
        pci_unregister_driver(&efx_pci_driver);
+       destroy_workqueue(refill_workqueue);
        unregister_netdevice_notifier(&efx_netdev_notifier);
        efx_fini_debugfs();
 
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/efx.h
--- a/drivers/net/sfc/efx.h     Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/efx.h     Tue Mar 31 11:59:10 2009 +0100
@@ -43,30 +43,19 @@ extern void efx_wake_queue(struct efx_ni
 extern void efx_wake_queue(struct efx_nic *efx);
 
 /* RX */
-#if defined(EFX_USE_FASTCALL)
 extern void fastcall efx_xmit_done(struct efx_tx_queue *tx_queue,
                                   unsigned int index);
-#else
-extern void efx_xmit_done(struct efx_tx_queue *tx_queue, unsigned int index);
-#endif
-#if defined(EFX_USE_FASTCALL)
 extern void fastcall efx_rx_packet(struct efx_rx_queue *rx_queue,
                                   unsigned int index, unsigned int len,
                                   int checksummed, int discard);
-#else
-extern void efx_rx_packet(struct efx_rx_queue *rx_queue, unsigned int index,
-                         unsigned int len, int checksummed, int discard);
-#endif
-extern void efx_fini_rx_buffer(struct efx_rx_queue *rx_queue,
-                              struct efx_rx_buffer *rx_buf);
+extern void efx_schedule_slow_fill(struct efx_rx_queue *rx_queue, int delay);
 
 /* Channels */
 extern void efx_process_channel_now(struct efx_channel *channel);
 extern int efx_flush_queues(struct efx_nic *efx);
 
 /* Ports */
-extern void efx_reconfigure_port(struct efx_nic *efx,
-                                int on_disabled);
+extern void efx_reconfigure_port(struct efx_nic *efx);
 
 /* Global */
 extern void efx_schedule_reset(struct efx_nic *efx, enum reset_type type);
@@ -91,13 +80,8 @@ static inline void efx_schedule_channel(
                  channel->channel, raw_smp_processor_id());
        channel->work_pending = 1;
 
-#if defined(EFX_HAVE_OLD_NAPI)
        if (!test_and_set_bit(__LINK_STATE_RX_SCHED, &channel->napi_dev->state))
                __netif_rx_schedule(channel->napi_dev);
-#else
-       netif_rx_schedule(channel->napi_dev, &channel->napi_str);
-#endif
 }
 
-
 #endif /* EFX_EFX_H */
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/enum.h
--- a/drivers/net/sfc/enum.h    Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/enum.h    Tue Mar 31 11:59:10 2009 +0100
@@ -98,6 +98,9 @@ enum efx_loopback_mode {
  * @RESET_TYPE_MONITOR: reset due to hardware monitor
  * @RESET_TYPE_INT_ERROR: reset due to internal error
  * @RESET_TYPE_RX_RECOVERY: reset to recover from RX datapath errors
+ * @RESET_TYPE_RX_DESC_FETCH: pcie error during rx descriptor fetch
+ * @RESET_TYPE_TX_DESC_FETCH: pcie error during tx descriptor fetch
+ * @RESET_TYPE_TX_SKIP: hardware completed empty tx descriptors
  */
 enum reset_type {
        RESET_TYPE_NONE = -1,
@@ -111,6 +114,7 @@ enum reset_type {
        RESET_TYPE_RX_RECOVERY,
        RESET_TYPE_RX_DESC_FETCH,
        RESET_TYPE_TX_DESC_FETCH,
+       RESET_TYPE_TX_SKIP,
        RESET_TYPE_MAX,
 };
 
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/ethtool.c
--- a/drivers/net/sfc/ethtool.c Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/ethtool.c Tue Mar 31 11:59:10 2009 +0100
@@ -28,7 +28,6 @@
 #include <linux/netdevice.h>
 #include <linux/ethtool.h>
 #include <linux/rtnetlink.h>
-#include <asm/uaccess.h>
 #include "net_driver.h"
 #include "selftest.h"
 #include "efx.h"
@@ -200,8 +199,15 @@ int efx_ethtool_get_settings(struct net_
                             struct ethtool_cmd *ecmd)
 {
        struct efx_nic *efx = net_dev->priv;
-
-       return efx->mac_op->get_settings(efx, ecmd);
+       int rc;
+
+       if (!in_interrupt())
+           mutex_lock(&efx->mac_lock);
+       rc = efx->mac_op->get_settings(efx, ecmd);
+       if (!in_interrupt())
+           mutex_unlock(&efx->mac_lock);
+
+       return rc;
 }
 
 /* This must be called with rtnl_lock held. */
@@ -211,14 +217,13 @@ int efx_ethtool_set_settings(struct net_
        struct efx_nic *efx = net_dev->priv;
        int rc;
 
+       mutex_lock(&efx->mac_lock);
        rc = efx->mac_op->set_settings(efx, ecmd);
-       if (rc)
-               return rc;
-
-       /* Push the settings to the MAC */
-       efx_reconfigure_port(efx, 0);
-
-       return 0;
+       mutex_unlock(&efx->mac_lock);
+       if (!rc)
+               efx_reconfigure_port(efx);
+
+       return rc;
 }
 
 static void efx_ethtool_get_drvinfo(struct net_device *net_dev,
@@ -418,7 +423,6 @@ static void efx_ethtool_get_stats(struct
                                  struct ethtool_stats *stats
                                  __attribute__ ((unused)), u64 *data)
 {
-       unsigned long flags __attribute__ ((unused));
        struct efx_nic *efx = net_dev->priv;
        struct efx_mac_stats *mac_stats = &efx->mac_stats;
        struct efx_ethtool_stat *stat;
@@ -429,7 +433,6 @@ static void efx_ethtool_get_stats(struct
 
        /* Update MAC and NIC statistics */
        net_dev->get_stats(net_dev);
-       falcon_update_nic_stats(efx);
 
        /* Fill detailed statistics buffer */
        for (i = 0; i < EFX_ETHTOOL_NUM_STATS; i++) {
@@ -461,7 +464,6 @@ static int efx_ethtool_set_tx_csum(struc
        if (rc)
                return rc;
 
-
        efx_flush_queues(efx);
 
        return 0;
@@ -668,14 +670,14 @@ static int efx_ethtool_set_pauseparam(st
        flow_control |= pause->autoneg ? EFX_FC_AUTO : 0;
 
        /* Try to push the pause parameters */
+       mutex_lock(&efx->mac_lock);
        rc = efx->mac_op->set_pause(efx, flow_control);
-       if (rc)
-               return rc;
-
-       /* Push the settings to the MAC */
-       efx_reconfigure_port(efx, 0);
-
-       return 0;
+       mutex_unlock(&efx->mac_lock);
+
+       if (!rc)
+               efx_reconfigure_port(efx);
+
+       return rc;
 }
 
 static void efx_ethtool_get_pauseparam(struct net_device *net_dev,
@@ -689,7 +691,6 @@ static void efx_ethtool_get_pauseparam(s
 }
 
 
-#if defined(EFX_USE_ETHTOOL_GET_PERM_ADDR)
 static int efx_ethtool_op_get_perm_addr(struct net_device *net_dev,
                                        struct ethtool_perm_addr *addr,
                                        u8 *data)
@@ -700,7 +701,6 @@ static int efx_ethtool_op_get_perm_addr(
 
        return 0;
 }
-#endif
 
 struct ethtool_ops efx_ethtool_ops = {
        .get_settings           = efx_ethtool_get_settings,
@@ -718,17 +718,11 @@ struct ethtool_ops efx_ethtool_ops = {
        .set_tx_csum            = efx_ethtool_set_tx_csum,
        .get_sg                 = ethtool_op_get_sg,
        .set_sg                 = ethtool_op_set_sg,
-#if defined(EFX_USE_ETHTOOL_FLAGS)
-       .get_flags              = ethtool_op_get_flags,
-       .set_flags              = ethtool_op_set_flags,
-#endif
        .self_test_count        = efx_ethtool_self_test_count,
        .self_test              = efx_ethtool_self_test,
        .get_strings            = efx_ethtool_get_strings,
        .phys_id                = efx_ethtool_phys_id,
        .get_stats_count        = efx_ethtool_get_stats_count,
        .get_ethtool_stats      = efx_ethtool_get_stats,
-#if defined(EFX_USE_ETHTOOL_GET_PERM_ADDR)
        .get_perm_addr          = efx_ethtool_op_get_perm_addr,
-#endif
 };
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/falcon.c
--- a/drivers/net/sfc/falcon.c  Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/falcon.c  Tue Mar 31 11:59:10 2009 +0100
@@ -25,8 +25,7 @@
  ****************************************************************************
  */
 
-#include <asm/io.h>
-#include <asm/bitops.h>
+#include <linux/bitops.h>
 #include <linux/delay.h>
 #include <linux/pci.h>
 #include <linux/module.h>
@@ -51,20 +50,29 @@
  * present in SFE400X evaluation boards
  */
 
+/**
+ * struct falcon_nic_data - Falcon NIC state
+ * @tx_dc_entries: Number of entries in each TX queue descriptor cache
+ * @rx_dc_entries: Number of entries in each RX queue descriptor cache
+ * @tx_dc_base: Base address in SRAM of TX queue descriptor caches
+ * @rx_dc_base: Base address in SRAM of RX queue descriptor caches
+ * @old_loopback_mode: Previous loopback mode used in deconfigure_mac_wrapper
+ * @external_sram_cfg: Size and number of banks of external SRAM
+ * @pci_dev2: The secondary PCI device if present
+ * @resources: Driverlink parameters
+ */
 struct falcon_nic_data {
-       /* Number of entries in each TX queue descriptor cache. */
        unsigned tx_dc_entries;
-       /* Number of entries in each RX queue descriptor cache. */
        unsigned rx_dc_entries;
-       /* Base address in SRAM of TX queue descriptor caches. */
        unsigned tx_dc_base;
-       /* Base address in SRAM of RX queue descriptor caches. */
        unsigned rx_dc_base;
 
-       /* Previous loopback mode used in deconfigure_mac_wrapper */
        enum efx_loopback_mode old_loopback_mode;
 
-       /* Driverlink parameters */
+       struct pci_dev *pci_dev2;
+
+       int external_sram_cfg;
+
        struct efx_dl_falcon_resources resources;
 };
 
@@ -150,21 +158,23 @@ MODULE_PARM_DESC(rx_xon_thresh_bytes, "R
 #endif
 
 /* TX DMA length mask (13-bit) */
-#define FALCON_TX_DMA_MASK (8192 - 1)
-
-/* Alignment of special buffers (4KB) */
-#define FALCON_BUF_ALIGN 4096
+#define FALCON_TX_DMA_MASK (4096 - 1)
+
+/* Size and alignment of special buffers (4KB) */
+#define FALCON_BUF_SIZE 4096
 
 /* Dummy SRAM size code */
 #define SRM_NB_BSZ_ONCHIP_ONLY (-1)
 
 /* Be nice if these (or equiv.) were in linux/pci_regs.h, but they're not. */
-#define PCI_EXP_DEVCAP_PWR_VAL_LBN     (18)
-/* This field takes up bits 26 and 27. */
-#define PCI_EXP_DEVCAP_PWR_SCL_LBN     (26)
-#define PCI_EXP_LNKSTA_LNK_WID         (0x3f0)
-#define PCI_EXP_LNKSTA_LNK_WID_LBN     (4)
-
+#define PCI_EXP_DEVCAP_PWR_VAL_LBN     18
+#define PCI_EXP_DEVCAP_PWR_SCL_LBN     26
+#define PCI_EXP_DEVCTL_PAYLOAD_LBN     5
+#define PCI_EXP_LNKSTA_LNK_WID         0x3f0
+#define PCI_EXP_LNKSTA_LNK_WID_LBN     4
+
+#define FALCON_IS_DUAL_FUNC(efx)               \
+       (FALCON_REV(efx) < FALCON_REV_B0)
 
 /**************************************************************************
  *
@@ -284,15 +294,23 @@ static struct efx_i2c_bit_operations fal
  *
  *************************************************************************/
 
-/* Adds the relevant entries to the full-mode buffer table. */
+/*
+ * Initialise a Falcon special buffer
+ *
+ * This will define a buffer (previously allocated via
+ * falcon_alloc_special_buffer()) in Falcon's buffer table, allowing
+ * it to be used for event queues, descriptor rings etc.
+ */
 static int
-falcon_pin_special_buffer_full(struct efx_nic *efx,
-                              struct efx_special_buffer *buffer)
+falcon_init_special_buffer(struct efx_nic *efx,
+                          struct efx_special_buffer *buffer)
 {
        efx_qword_t buf_desc;
        int index;
        dma_addr_t dma_addr;
        int i;
+
+       EFX_BUG_ON_PARANOID(!buffer->addr);
 
        /* Write buffer descriptors to NIC */
        for (i = 0; i < buffer->entries; i++) {
@@ -311,14 +329,17 @@ falcon_pin_special_buffer_full(struct ef
        return 0;
 }
 
-/* Clears the relevant entries from the buffer table */
+/* Unmaps a buffer from Falcon and clears the buffer table entries */
 static void
-falcon_clear_special_buffer_full(struct efx_nic *efx,
-                                struct efx_special_buffer *buffer)
+falcon_fini_special_buffer(struct efx_nic *efx,
+                          struct efx_special_buffer *buffer)
 {
        efx_oword_t buf_tbl_upd;
        unsigned int start = buffer->index;
        unsigned int end = (buffer->index + buffer->entries - 1);
+
+       if (!buffer->entries)
+               return;
 
        EFX_LOG(efx, "unmapping special buffers %d-%d\n",
                buffer->index, buffer->index + buffer->entries - 1);
@@ -337,11 +358,8 @@ falcon_clear_special_buffer_full(struct 
  * This allocates memory for a new buffer, clears it and allocates a
  * new buffer ID range.  It does not write into Falcon's buffer table.
  *
- * This call will allocate 4kB buffers, since Falcon can't use 8kB
- * buffers for event queues and descriptor rings.  It will always
- * allocate an even number of 4kB buffers, since when we're in
- * half-entry mode for the buffer table we can only deal with pairs of
- * buffers.
+ * This call will allocate 4KB buffers, since Falcon can't use 8KB
+ * buffers for event queues and descriptor rings.
  */
 static int falcon_alloc_special_buffer(struct efx_nic *efx,
                                       struct efx_special_buffer *buffer,
@@ -349,8 +367,7 @@ static int falcon_alloc_special_buffer(s
 {
        struct falcon_nic_data *nic_data = efx->nic_data;
 
-       /* Round size up to an 8kB boundary (i.e. pairs of 4kB buffers) */
-       len = (len + 8192 - 1) & ~(8192 - 1);
+       len = ALIGN(len, FALCON_BUF_SIZE);
 
        /* Allocate buffer as consistent PCI DMA space */
        buffer->addr = pci_alloc_consistent(efx->pci_dev, len,
@@ -358,8 +375,8 @@ static int falcon_alloc_special_buffer(s
        if (!buffer->addr)
                return -ENOMEM;
        buffer->len = len;
-       buffer->entries = len / 4096;
-       BUG_ON(buffer->dma_addr & (FALCON_BUF_ALIGN - 1));
+       buffer->entries = len / FALCON_BUF_SIZE;
+       BUG_ON(buffer->dma_addr & (FALCON_BUF_SIZE - 1));
 
        /* All zeros is a potentially valid event so memset to 0xff */
        memset(buffer->addr, 0xff, len);
@@ -375,34 +392,6 @@ static int falcon_alloc_special_buffer(s
                buffer->addr, virt_to_phys(buffer->addr));
 
        return 0;
-}
-
-/*
- * Initialise a Falcon special buffer
- *
- * This will define a buffer (previously allocated via
- * falcon_alloc_special_buffer()) in Falcon's buffer table, allowing
- * it to be used for event queues, descriptor rings etc.
- */
-static int falcon_init_special_buffer(struct efx_nic *efx,
-                                     struct efx_special_buffer *buffer)
-{
-       EFX_BUG_ON_PARANOID(!buffer->addr);
-
-       /* Write buffer descriptors to NIC */
-       return falcon_pin_special_buffer_full(efx, buffer);
-}
-
-/* Unmaps a buffer from Falcon and clears the buffer table
- * entries */
-static void falcon_fini_special_buffer(struct efx_nic *efx,
-                                      struct efx_special_buffer *buffer)
-{
-
-       if (!buffer->entries)
-               return;
-
-       falcon_clear_special_buffer_full(efx, buffer);
 }
 
 /* Release the buffer memory. */
@@ -487,11 +476,7 @@ static inline void falcon_notify_tx_desc
  * descriptor in the hardware TX descriptor ring (in host memory), and
  * write a doorbell.
  */
-#if defined(EFX_USE_FASTCALL)
 void fastcall falcon_push_buffers(struct efx_tx_queue *tx_queue)
-#else
-void falcon_push_buffers(struct efx_tx_queue *tx_queue)
-#endif
 {
 
        struct efx_tx_buffer *buffer;
@@ -603,8 +588,7 @@ static int falcon_flush_tx_queue(struct 
        falcon_write(efx, &tx_flush_descq, TX_FLUSH_DESCQ_REG_KER);
        msleep(FALCON_FLUSH_TIMEOUT);
 
-       /* If the NIC is resetting then don't bother checking */
-       if (EFX_WORKAROUND_7803(efx) || (efx->state == STATE_RESETTING))
+       if (EFX_WORKAROUND_7803(efx))
                return 0;
 
        /* Look for a flush completed event */
@@ -707,11 +691,7 @@ static inline void falcon_build_rx_desc(
 /* This writes to the RX_DESC_WPTR register for the specified receive
  * descriptor ring.
  */
-#if defined(EFX_USE_FASTCALL)
 void fastcall falcon_notify_rx_desc(struct efx_rx_queue *rx_queue)
-#else
-void falcon_notify_rx_desc(struct efx_rx_queue *rx_queue)
-#endif
 {
        efx_dword_t reg;
        unsigned write_ptr;
@@ -799,8 +779,7 @@ static int falcon_flush_rx_queue(struct 
        falcon_write(efx, &rx_flush_descq, RX_FLUSH_DESCQ_REG_KER);
        msleep(FALCON_FLUSH_TIMEOUT);
 
-       /* If the NIC is resetting then don't bother checking */
-       if (EFX_WORKAROUND_7803(efx) || (efx->state == STATE_RESETTING))
+       if (EFX_WORKAROUND_7803(efx))
                return 0;
 
        /* Look for a flush completed event */
@@ -863,8 +842,10 @@ void falcon_fini_rx(struct efx_rx_queue 
                        continue;
                break;
        }
-       if (rc)
+       if (rc) {
                EFX_ERR(efx, "failed to flush rx queue %d\n", rx_queue->queue);
+               efx_schedule_reset(efx, RESET_TYPE_INVISIBLE);
+       }
 
        /* Remove RX descriptor ring from card */
        EFX_ZERO_OWORD(rx_desc_ptr);
@@ -897,11 +878,7 @@ void falcon_remove_rx(struct efx_rx_queu
  * whereas channel->eventq_read_ptr contains the index of the "next to
  * read" event.
  */
-#if defined(EFX_USE_FASTCALL)
 void fastcall falcon_eventq_read_ack(struct efx_channel *channel)
-#else
-void falcon_eventq_read_ack(struct efx_channel *channel)
-#endif
 {
        efx_dword_t reg;
        struct efx_nic *efx = channel->efx;
@@ -947,10 +924,10 @@ static inline void falcon_handle_tx_even
                tx_ev_q_label = EFX_QWORD_FIELD(*event, TX_EV_Q_LABEL);
                tx_queue = &efx->tx_queue[tx_ev_q_label];
 
-               if (efx->net_dev_registered)
+               if (NET_DEV_REGISTERED(efx))
                        netif_tx_lock(efx->net_dev);
                falcon_notify_tx_desc(tx_queue);
-               if (efx->net_dev_registered)
+               if (NET_DEV_REGISTERED(efx))
                        netif_tx_unlock(efx->net_dev);
        } else if (EFX_QWORD_FIELD(*event, TX_EV_PKT_ERR) &&
                   EFX_WORKAROUND_10727(efx)) {
@@ -1290,11 +1267,7 @@ static void falcon_handle_driver_event(s
        }
 }
 
-#if defined(EFX_USE_FASTCALL)
 int fastcall falcon_process_eventq(struct efx_channel *channel, int *rx_quota)
-#else
-int falcon_process_eventq(struct efx_channel *channel, int *rx_quota)
-#endif
 {
        unsigned int read_ptr;
        efx_qword_t event, *p_event;
@@ -1555,6 +1528,7 @@ static inline void falcon_irq_ack_a1(str
  */
 static irqreturn_t falcon_fatal_interrupt(struct efx_nic *efx)
 {
+       struct falcon_nic_data *nic_data = efx->nic_data;
        efx_oword_t *int_ker = (efx_oword_t *) efx->irq_status.addr;
        efx_oword_t fatal_intr;
        int error, mem_perr;
@@ -1581,8 +1555,8 @@ static irqreturn_t falcon_fatal_interrup
 
        /* Disable DMA bus mastering on both devices */
        pci_disable_device(efx->pci_dev);
-       if (efx->type->is_dual_func)
-               pci_disable_device(efx->pci_dev2);
+       if (FALCON_IS_DUAL_FUNC(efx))
+               pci_disable_device(nic_data->pci_dev2);
 
        if (++n_int_errors < FALCON_MAX_INT_ERRORS) {
                EFX_ERR(efx, "SYSTEM ERROR - reset scheduled\n");
@@ -1603,13 +1577,9 @@ out:
  * interrupts are disabled, to allow for correct semantics of
  * efx_suspend() and efx_resume().
  */
-#if !defined(EFX_HAVE_IRQ_HANDLER_REGS)
-static irqreturn_t falcon_legacy_interrupt_b0(int irq, void *dev_id)
-#else
 static irqreturn_t falcon_legacy_interrupt_b0(int irq, void *dev_id,
                                              struct pt_regs *regs
                                              __attribute__ ((unused)))
-#endif
 {
        struct efx_nic *efx = (struct efx_nic *)dev_id;
        efx_oword_t *int_ker = (efx_oword_t *) efx->irq_status.addr;
@@ -1647,13 +1617,9 @@ static irqreturn_t falcon_legacy_interru
 }
 
 
-#if !defined(EFX_HAVE_IRQ_HANDLER_REGS)
-static irqreturn_t falcon_legacy_interrupt_a1(int irq, void *dev_id)
-#else
 static irqreturn_t falcon_legacy_interrupt_a1(int irq, void *dev_id,
                                              struct pt_regs *regs
                                              __attribute__ ((unused)))
-#endif
 {
        struct efx_nic *efx = (struct efx_nic *)dev_id;
        efx_oword_t *int_ker = (efx_oword_t *) efx->irq_status.addr;
@@ -1710,13 +1676,9 @@ static irqreturn_t falcon_legacy_interru
  * interrupts are disabled, to allow for correct semantics of
  * efx_suspend() and efx_resume().
  */
-#if !defined(EFX_HAVE_IRQ_HANDLER_REGS)
-static irqreturn_t falcon_msi_interrupt(int irq, void *dev_id)
-#else
 static irqreturn_t falcon_msi_interrupt(int irq, void *dev_id,
                                        struct pt_regs *regs
                                        __attribute__ ((unused)))
-#endif
 {
        struct efx_channel *channel = (struct efx_channel *)dev_id;
        struct efx_nic *efx = channel->efx;
@@ -1746,7 +1708,6 @@ static void falcon_setup_rss_indir_table
 {
        int i = 0;
        unsigned long offset;
-       unsigned long flags __attribute__ ((unused));
        efx_dword_t dword;
 
        if (FALCON_REV(efx) < FALCON_REV_B0)
@@ -1976,8 +1937,6 @@ void falcon_drain_tx_fifo(struct efx_nic
 void falcon_drain_tx_fifo(struct efx_nic *efx)
 {
        efx_oword_t temp;
-       efx_oword_t mcast_reg0;
-       efx_oword_t mcast_reg1;
        int count;
 
        if (FALCON_REV(efx) < FALCON_REV_B0)
@@ -1994,9 +1953,6 @@ void falcon_drain_tx_fifo(struct efx_nic
 
        EFX_SET_OWORD_FIELD(temp, TXFIFO_DRAIN_EN_B0, 1);
        falcon_write(efx, &temp, MAC0_CTRL_REG_KER);
-
-       falcon_read(efx, &mcast_reg0, MAC_MCAST_HASH_REG0_KER);
-       falcon_read(efx, &mcast_reg1, MAC_MCAST_HASH_REG1_KER);
 
        /* Reset the MAC and EM block. */
        falcon_read(efx, &temp, GLB_CTL_REG_KER);
@@ -2025,10 +1981,6 @@ void falcon_drain_tx_fifo(struct efx_nic
 
        spin_unlock(&efx->stats_lock);
 
-       /* Restore the multicast hash registers. */
-       falcon_write(efx, &mcast_reg0, MAC_MCAST_HASH_REG0_KER);
-       falcon_write(efx, &mcast_reg1, MAC_MCAST_HASH_REG1_KER);
-
        /* If we've reset the EM block and the link is up, then
         * we'll have to kick the XAUI link so the PHY can recover */
        if (efx->link_up && EFX_IS10G(efx) && EFX_WORKAROUND_5147(efx))
@@ -2053,6 +2005,12 @@ void falcon_deconfigure_mac_wrapper(stru
         * draining the TX fifo and resetting. */
        changing_loopback = (efx->loopback_mode != nic_data->old_loopback_mode);
        nic_data->old_loopback_mode = efx->loopback_mode;
+
+       if (EFX_WORKAROUND_11667(efx) && (efx->phy_type == PHY_TYPE_10XPRESS)) {
+               if (changing_loopback)
+                       return;
+       }
+
        if (changing_loopback || !efx->link_up)
                falcon_drain_tx_fifo(efx);
 }
@@ -2074,8 +2032,7 @@ void falcon_reconfigure_mac_wrapper(stru
        /* MAC_LINK_STATUS controls MAC backpressure but doesn't work
         * as advertised.  Disable to ensure packets are not
         * indefinitely held and TX queue can be flushed at any point
-        * while the link is down.
-        */
+        * while the link is down. */
        EFX_POPULATE_OWORD_5(reg,
                             MAC_XOFF_VAL, 0xffff /* max pause time */,
                             MAC_BCAD_ACPT, 1,
@@ -2091,12 +2048,12 @@ void falcon_reconfigure_mac_wrapper(stru
 
        falcon_write(efx, &reg, MAC0_CTRL_REG_KER);
 
-       /*
-        * Transmission of pause frames when RX crosses the threshold is
+       /* Restore the multicast hash registers. */
+       falcon_set_multicast_hash(efx);
+
+       /* Transmission of pause frames when RX crosses the threshold is
         * covered by RX_XOFF_MAC_EN and XM_TX_CFG_REG:XM_FCNTL.
-        *
-        * Action on receipt of pause frames is controller by XM_DIS_FCNTL
-        */
+        * Action on receipt of pause frames is controller by XM_DIS_FCNTL */
        tx_fc = (efx->flow_control & EFX_FC_TX) ? 1 : 0;
        falcon_read(efx, &reg, RX_CFG_REG_KER);
        EFX_SET_OWORD_FIELD_VER(efx, reg, RX_XOFF_MAC_EN, tx_fc);
@@ -2291,7 +2248,6 @@ static int falcon_mdio_read(struct net_d
        unsigned int phy_10g = phy_id & FALCON_PHY_ID_10G;
        efx_oword_t reg;
        int value = -1;
-       unsigned long flags __attribute__ ((unused));
 
        if (phy_addr == PHY_ADDR_INVALID)
                return -1;
@@ -2373,7 +2329,7 @@ static void falcon_init_mdio(struct mii_
        gmii->mdio_read = falcon_mdio_read;
        gmii->mdio_write = falcon_mdio_write;
        gmii->phy_id_mask = FALCON_PHY_ID_MASK;
-       gmii->reg_num_mask = ((1 << EFX_WIDTH(MD_DEV_ADR)) - 1);
+       gmii->reg_num_mask = ((1 << EFX_WIDTH(MD_PHY_ADR)) - 1);
 }
 
 static int falcon_probe_gmac_port(struct efx_nic *efx)
@@ -2493,17 +2449,16 @@ void falcon_remove_port(struct efx_nic *
 
 void falcon_set_multicast_hash(struct efx_nic *efx)
 {
-       union efx_multicast_hash falcon_mc_hash;
+       union efx_multicast_hash *mc_hash = &efx->multicast_hash;
 
        /* Broadcast packets go through the multicast hash filter.
         * ether_crc_le() of the broadcast address is 0xbe2612ff
-        * so we always add bit 0xff to the mask we are given.
+        * so we always add bit 0xff to the mask.
         */
-       memcpy(&falcon_mc_hash, &efx->multicast_hash, sizeof(falcon_mc_hash));
-       set_bit_le(0xff, (void *)&falcon_mc_hash);
-
-       falcon_write(efx, &falcon_mc_hash.oword[0], MAC_MCAST_HASH_REG0_KER);
-       falcon_write(efx, &falcon_mc_hash.oword[1], MAC_MCAST_HASH_REG1_KER);
+       set_bit_le(0xff, mc_hash->byte);
+
+       falcon_write(efx, &mc_hash->oword[0], MAC_MCAST_HASH_REG0_KER);
+       falcon_write(efx, &mc_hash->oword[1], MAC_MCAST_HASH_REG1_KER);
 }
 
 /**************************************************************************
@@ -2587,6 +2542,7 @@ out:
  * context and is allowed to sleep. */
 int falcon_reset_hw(struct efx_nic *efx, enum reset_type method)
 {
+       struct falcon_nic_data *nic_data = efx->nic_data;
        efx_oword_t glb_ctl_reg_ker;
        int rc;
 
@@ -2601,8 +2557,8 @@ int falcon_reset_hw(struct efx_nic *efx,
                                "function prior to hardware reset\n");
                        goto fail1;
                }
-               if (efx->type->is_dual_func) {
-                       rc = pci_save_state(efx->pci_dev2);
+               if (FALCON_IS_DUAL_FUNC(efx)) {
+                       rc = pci_save_state(nic_data->pci_dev2);
                        if (rc) {
                                EFX_ERR(efx, "failed to backup PCI state of "
                                        "secondary function prior to "
@@ -2635,8 +2591,8 @@ int falcon_reset_hw(struct efx_nic *efx,
 
        /* Restore PCI configuration if needed */
        if (method == RESET_TYPE_WORLD) {
-               if (efx->type->is_dual_func) {
-                       rc = pci_restore_state(efx->pci_dev2);
+               if (FALCON_IS_DUAL_FUNC(efx)) {
+                       rc = pci_restore_state(nic_data->pci_dev2);
                        if (rc) {
                                EFX_ERR(efx, "failed to restore PCI config for "
                                        "the secondary function\n");
@@ -2686,19 +2642,21 @@ fail6:
  */
 static int falcon_reset_sram(struct efx_nic *efx)
 {
+       struct falcon_nic_data *nic_data = efx->nic_data;
        efx_oword_t srm_cfg_reg_ker, gpio_cfg_reg_ker;
        int count, onchip, sram_cfg_val;
 
        /* Set the SRAM wake/sleep GPIO appropriately. */
-       onchip = (efx->external_sram_cfg == SRM_NB_BSZ_ONCHIP_ONLY);
+       onchip = (nic_data->external_sram_cfg == SRM_NB_BSZ_ONCHIP_ONLY);
        falcon_read(efx, &gpio_cfg_reg_ker, GPIO_CTL_REG_KER);
        EFX_SET_OWORD_FIELD(gpio_cfg_reg_ker, GPIO1_OEN, 1);
        EFX_SET_OWORD_FIELD(gpio_cfg_reg_ker, GPIO1_OUT, onchip ? 1 : 0);
        falcon_write(efx, &gpio_cfg_reg_ker, GPIO_CTL_REG_KER);
 
        /* Initiate SRAM reset */
-       sram_cfg_val = (efx->external_sram_cfg == SRM_NB_BSZ_ONCHIP_ONLY) ?
-               0 : efx->external_sram_cfg;
+       sram_cfg_val = nic_data->external_sram_cfg;
+       if (nic_data->external_sram_cfg == SRM_NB_BSZ_ONCHIP_ONLY)
+               sram_cfg_val = 0;
 
        EFX_POPULATE_OWORD_2(srm_cfg_reg_ker,
                             SRAM_OOB_BT_INIT_EN, 1,
@@ -2761,11 +2719,12 @@ static void falcon_spi_device_init(struc
 /* Extract non-volatile configuration */
 static int falcon_probe_nvconfig(struct efx_nic *efx)
 {
-       int rc;
+       struct falcon_nic_data *nic_data = efx->nic_data;
        struct falcon_nvconfig *nvconfig;
        struct efx_spi_device *spi;
        size_t offset, len;
        int magic_num, struct_ver, board_rev, onchip_sram;
+       int rc;
 
        nvconfig = kmalloc(sizeof(*nvconfig), GFP_KERNEL);
 
@@ -2831,18 +2790,18 @@ static int falcon_probe_nvconfig(struct 
         * automatically but might may been reset since boot.
         */
        if (onchip_sram) {
-               efx->external_sram_cfg = SRM_NB_BSZ_ONCHIP_ONLY;
+               nic_data->external_sram_cfg = SRM_NB_BSZ_ONCHIP_ONLY;
        } else {
-               efx->external_sram_cfg =
-                   EFX_OWORD_FIELD(nvconfig->srm_cfg_reg,
-                                   SRM_NUM_BANKS_AND_BANK_SIZE);
-               WARN_ON(efx->external_sram_cfg == SRM_NB_BSZ_RESERVED);
+               nic_data->external_sram_cfg =
+                       EFX_OWORD_FIELD(nvconfig->srm_cfg_reg,
+                                       SRM_NUM_BANKS_AND_BANK_SIZE);
+               WARN_ON(nic_data->external_sram_cfg == SRM_NB_BSZ_RESERVED);
                /* Replace invalid setting with the smallest defaults */
-               if (efx->external_sram_cfg == SRM_NB_BSZ_DEFAULT)
-                       efx->external_sram_cfg = SRM_NB_BSZ_1BANKS_2M;
+               if (nic_data->external_sram_cfg == SRM_NB_BSZ_DEFAULT)
+                       nic_data->external_sram_cfg = SRM_NB_BSZ_1BANKS_2M;
        }
        EFX_LOG(efx, "external_sram_cfg=%d (>=0 is external)\n",
-               efx->external_sram_cfg);
+               nic_data->external_sram_cfg);
 
  out:
        kfree(nvconfig);
@@ -2892,31 +2851,35 @@ static int falcon_dimension_resources(st
         */
        switch (FALCON_REV(efx)) {
        case FALCON_REV_A1:
-               res->rxq_min = res->txq_min = 16;
-               res->evq_int_min = res->evq_int_max = 4;
+               res->rxq_min = 16;
+               res->txq_min = 16;
+               res->evq_int_min = 4;
+               res->evq_int_lim = 5;
                res->evq_timer_min = 5;
-               res->evq_timer_max = 4096;
+               res->evq_timer_lim = 4096;
                internal_dcs_entries = 8192;
                break;
        case FALCON_REV_B0:
        default:
-               res->rxq_min = res->txq_min = res->evq_int_min = 0;
-               res->evq_int_max = 64;
+               res->rxq_min = 0;
+               res->txq_min = 0;
+               res->evq_int_min = 0;
+               res->evq_int_lim = 64;
                res->evq_timer_min = 64;
-               res->evq_timer_max = 4096;
+               res->evq_timer_lim = 4096;
                internal_dcs_entries = 4096;
                break;
        }
 
        buffer_entry_bytes = 8;
 
-       if (efx->external_sram_cfg == SRM_NB_BSZ_ONCHIP_ONLY) {
-               res->rxq_max = internal_dcs_entries / nic_data->rx_dc_entries;
-               res->txq_max = internal_dcs_entries / nic_data->tx_dc_entries;
+       if (nic_data->external_sram_cfg == SRM_NB_BSZ_ONCHIP_ONLY) {
+               res->rxq_lim = internal_dcs_entries / nic_data->rx_dc_entries;
+               res->txq_lim = internal_dcs_entries / nic_data->tx_dc_entries;
                /* Prog model says 8K entries for buffer table in internal
                 * mode.  But does this not depend on full/half mode?
                 */
-               res->buffer_table_max = 8192;
+               res->buffer_table_lim = 8192;
                nic_data->tx_dc_base = 0x130000;
                nic_data->rx_dc_base = 0x100000;
        } else {
@@ -2925,7 +2888,7 @@ static int falcon_dimension_resources(st
                /* Determine how much SRAM we have to play with.  We have
                 * to fit buffer table and descriptor caches in.
                 */
-               switch (efx->external_sram_cfg) {
+               switch (nic_data->external_sram_cfg) {
                case SRM_NB_BSZ_1BANKS_2M:
                default:
                        sram_bytes = 2 * 1024 * 1024;
@@ -2953,17 +2916,17 @@ static int falcon_dimension_resources(st
                max_vnics = sram_bytes / vnic_bytes;
                for (n_vnics = 1; n_vnics < res->evq_timer_min + max_vnics;)
                        n_vnics *= 2;
-               res->rxq_max = n_vnics;
-               res->txq_max = n_vnics;
+               res->rxq_lim = n_vnics;
+               res->txq_lim = n_vnics;
 
                dcs = n_vnics * nic_data->tx_dc_entries * 8;
                nic_data->tx_dc_base = sram_bytes - dcs;
                dcs = n_vnics * nic_data->rx_dc_entries * 8;
                nic_data->rx_dc_base = nic_data->tx_dc_base - dcs;
-               res->buffer_table_max = nic_data->rx_dc_base / 8;
-       }
-
-       if (efx->type->is_dual_func)
+               res->buffer_table_lim = nic_data->rx_dc_base / 8;
+       }
+
+       if (FALCON_IS_DUAL_FUNC(efx))
                res->flags |= EFX_DL_FALCON_DUAL_FUNC;
 
        if (EFX_INT_MODE_USE_MSI(efx))
@@ -2982,7 +2945,6 @@ static int falcon_probe_nic_variant(stru
        falcon_read(efx, &altera_build, ALTERA_BUILD_REG_KER);
        efx->is_asic = EFX_OWORD_FIELD(altera_build, VER_ALL) == 0;
 
-#if !defined(EFX_USE_PCI_DEV_REVISION)
        {
                int rc;
                rc = pci_read_config_byte(efx->pci_dev, PCI_CLASS_REVISION,
@@ -2990,7 +2952,7 @@ static int falcon_probe_nic_variant(stru
                if (rc)
                        return rc;
        }
-#endif
+
        switch (FALCON_REV(efx)) {
        case FALCON_REV_A0:
        case 0xff:
@@ -3189,24 +3151,28 @@ int falcon_probe_nic(struct efx_nic *efx
        efx->i2c.sda = 1;
        efx->i2c.scl = 1;
 
+       /* Allocate storage for hardware specific data */
+       nic_data = kzalloc(sizeof(*nic_data), GFP_KERNEL);
+       efx->nic_data = (void *) nic_data;
+
        /* Determine number of ports etc. */
        rc = falcon_probe_nic_variant(efx);
        if (rc)
                goto fail1;
 
        /* Probe secondary function if expected */
-       if (efx->type->is_dual_func) {
+       if (FALCON_IS_DUAL_FUNC(efx)) {
                struct pci_dev *dev = pci_dev_get(efx->pci_dev);
 
                while ((dev = pci_get_device(EFX_VENDID_SFC, FALCON_A_S_DEVID,
                                             dev))) {
                        if (dev->bus == efx->pci_dev->bus &&
                            dev->devfn == efx->pci_dev->devfn + 1) {
-                               efx->pci_dev2 = dev;
+                               nic_data->pci_dev2 = dev;
                                break;
                        }
                }
-               if (!efx->pci_dev2) {
+               if (!nic_data->pci_dev2) {
                        EFX_ERR(efx, "failed to find secondary function\n");
                        rc = -ENODEV;
                        goto fail2;
@@ -3244,11 +3210,6 @@ int falcon_probe_nic(struct efx_nic *efx
                efx->mii.phy_id = 2;
        }
 
-       /* Decide how many resources we can allocate, to ourselves
-        * and to driverlink clients */
-       nic_data = kzalloc(sizeof(*nic_data), GFP_KERNEL);
-       efx->nic_data = (void *) nic_data;
-
        rc = falcon_dimension_resources(efx);
        if (rc)
                goto fail6;
@@ -3256,21 +3217,21 @@ int falcon_probe_nic(struct efx_nic *efx
        return 0;
 
  fail6:
-       kfree(nic_data);
-       efx->nic_data = efx->dl_info = NULL;
+       efx->dl_info = NULL;
  fail5:
        falcon_remove_spi_devices(efx);
        falcon_free_buffer(efx, &efx->irq_status);
  fail4:
        /* fall-thru */
  fail3:
-       if (efx->pci_dev2) {
-               pci_dev_put(efx->pci_dev2);
-               efx->pci_dev2 = NULL;
+       if (nic_data->pci_dev2) {
+               pci_dev_put(nic_data->pci_dev2);
+               nic_data->pci_dev2 = NULL;
        }
  fail2:
        /* fall-thru */
  fail1:
+       kfree(efx->nic_data);
        return rc;
 }
 
@@ -3336,7 +3297,7 @@ static void falcon_init_ack_repl_timer(s
                                  &pcie_ctrl_stat_reg);
        pcie_devicectrl = (u16) EFX_EXTRACT_DWORD(pcie_ctrl_stat_reg, 0, 15);
        tlp_size = ((PCI_EXP_DEVCTL_PAYLOAD & pcie_devicectrl) >>
-                   ffs(PCI_EXP_DEVCTL_PAYLOAD));
+                   PCI_EXP_DEVCTL_PAYLOAD_LBN);
        EFX_WARN_ON_PARANOID(tlp_size > 3); /* => 1024 bytes */
        tlp_ack_factor = &tlp_ack_factor_lut[tlp_size & 0x3];
        tlp_size_decoded = tlp_ack_factor->tlp;
@@ -3450,6 +3411,7 @@ static void falcon_fini_pcie_core(struct
  */
 int falcon_init_nic(struct efx_nic *efx)
 {
+       struct falcon_nic_data *nic_data = efx->nic_data;
        struct falcon_nic_data *data;
        efx_oword_t temp;
        unsigned thresh;
@@ -3469,7 +3431,7 @@ int falcon_init_nic(struct efx_nic *efx)
        /* Use on-chip SRAM if needed.
         */
        falcon_read(efx, &temp, NIC_STAT_REG);
-       if (efx->external_sram_cfg == SRM_NB_BSZ_ONCHIP_ONLY)
+       if (nic_data->external_sram_cfg == SRM_NB_BSZ_ONCHIP_ONLY)
                EFX_SET_OWORD_FIELD(temp, ONCHIP_SRAM, 1);
        else
                EFX_SET_OWORD_FIELD(temp, ONCHIP_SRAM, 0);
@@ -3623,21 +3585,23 @@ void falcon_fini_nic(struct efx_nic *efx
 
 void falcon_remove_nic(struct efx_nic *efx)
 {
+       struct falcon_nic_data *nic_data = efx->nic_data;
+
+       falcon_remove_spi_devices(efx);
+       falcon_free_buffer(efx, &efx->irq_status);
+
+       /* Reset the NIC finally */
+       (void) falcon_reset_hw(efx, RESET_TYPE_ALL);
+
+       /* Release the second function after the reset */
+       if (nic_data->pci_dev2) {
+               pci_dev_put(nic_data->pci_dev2);
+               nic_data->pci_dev2 = NULL;
+       }
+
        /* Tear down the private nic state, and the driverlink nic params */
        kfree(efx->nic_data);
        efx->nic_data = efx->dl_info = NULL;
-
-       falcon_remove_spi_devices(efx);
-       falcon_free_buffer(efx, &efx->irq_status);
-
-       /* Reset the NIC finally */
-       (void) falcon_reset_hw(efx, RESET_TYPE_ALL);
-
-       /* Release the second function after the reset */
-       if (efx->pci_dev2) {
-               pci_dev_put(efx->pci_dev2);
-               efx->pci_dev2 = NULL;
-       }
 }
 
 void falcon_update_nic_stats(struct efx_nic *efx)
@@ -3657,7 +3621,6 @@ void falcon_update_nic_stats(struct efx_
  */
 
 struct efx_nic_type falcon_a_nic_type = {
-       .is_dual_func = 1,
        .mem_bar = 2,
        .mem_map_size = 0x20000,
        .txd_ptr_tbl_base = TX_DESC_PTR_TBL_KER_A1,
@@ -3679,7 +3642,6 @@ struct efx_nic_type falcon_a_nic_type = 
 };
 
 struct efx_nic_type falcon_b_nic_type = {
-       .is_dual_func = 0,
        .mem_bar = 2,
        /* Map everything up to and including the RSS indirection
         * table.  Don't map MSI-X table, MSI-X PBA since Linux
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/falcon.h
--- a/drivers/net/sfc/falcon.h  Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/falcon.h  Tue Mar 31 11:59:10 2009 +0100
@@ -28,8 +28,6 @@
 #ifndef EFX_FALCON_H
 #define EFX_FALCON_H
 
-#include <asm/io.h>
-#include <linux/spinlock.h>
 #include "net_driver.h"
 
 /*
@@ -42,11 +40,7 @@ enum falcon_revision {
        FALCON_REV_B0 = 2,
 };
 
-#if defined(EFX_USE_PCI_DEV_REVISION)
-#define FALCON_REV(efx) ((efx)->pci_dev->revision)
-#else
 #define FALCON_REV(efx) ((efx)->revision)
-#endif
 
 extern struct efx_nic_type falcon_a_nic_type;
 extern struct efx_nic_type falcon_b_nic_type;
@@ -63,46 +57,29 @@ extern int falcon_init_tx(struct efx_tx_
 extern int falcon_init_tx(struct efx_tx_queue *tx_queue);
 extern void falcon_fini_tx(struct efx_tx_queue *tx_queue);
 extern void falcon_remove_tx(struct efx_tx_queue *tx_queue);
-#if defined(EFX_USE_FASTCALL)
 extern void fastcall falcon_push_buffers(struct efx_tx_queue *tx_queue);
-#else
-extern void falcon_push_buffers(struct efx_tx_queue *tx_queue);
-#endif
 
 /* RX data path */
 extern int falcon_probe_rx(struct efx_rx_queue *rx_queue);
 extern int falcon_init_rx(struct efx_rx_queue *rx_queue);
 extern void falcon_fini_rx(struct efx_rx_queue *rx_queue);
 extern void falcon_remove_rx(struct efx_rx_queue *rx_queue);
-#if defined(EFX_USE_FASTCALL)
 extern void fastcall falcon_notify_rx_desc(struct efx_rx_queue *rx_queue);
-#else
-extern void falcon_notify_rx_desc(struct efx_rx_queue *rx_queue);
-#endif
 
 /* Event data path */
 extern int falcon_probe_eventq(struct efx_channel *channel);
 extern int falcon_init_eventq(struct efx_channel *channel);
 extern void falcon_fini_eventq(struct efx_channel *channel);
 extern void falcon_remove_eventq(struct efx_channel *channel);
-#if defined(EFX_USE_FASTCALL)
 extern int fastcall falcon_process_eventq(struct efx_channel *channel,
                                          int *rx_quota);
-#else
-extern int falcon_process_eventq(struct efx_channel *channel, int *rx_quota);
-#endif
-#if defined(EFX_USE_FASTCALL)
 extern void fastcall falcon_eventq_read_ack(struct efx_channel *channel);
-#else
-extern void falcon_eventq_read_ack(struct efx_channel *channel);
-#endif
 
 /* Ports */
 extern int falcon_probe_port(struct efx_nic *efx);
 extern void falcon_remove_port(struct efx_nic *efx);
 
 /* MAC/PHY */
-extern void falcon_check_xaui_link_up(struct efx_nic *efx);
 extern int falcon_xaui_link_ok(struct efx_nic *efx);
 extern int falcon_dma_stats(struct efx_nic *efx,
                            unsigned int done_offset);
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/falcon_hwdefs.h
--- a/drivers/net/sfc/falcon_hwdefs.h   Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/falcon_hwdefs.h   Tue Mar 31 11:59:10 2009 +0100
@@ -894,6 +894,15 @@
 #define XM_DIS_FCNTL_WIDTH 1
 
 /* XGMAC pause time count register */
+/* XGMAC management interrupt mask register */
+#define XM_MGT_INT_MSK_REG_MAC_B0 0x5
+#define XM_MSK_PRMBLE_ERR_LBN 2
+#define XM_MSK_PRMBLE_ERR_WIDTH 1
+#define XM_MSK_RMTFLT_LBN 1
+#define XM_MSK_RMTFLT_WIDTH 1
+#define XM_MSK_LCLFLT_LBN 0
+#define XM_MSK_LCLFLT_WIDTH 1
+
 #define XM_PAUSE_TIME_REG_MAC 0x9
 #define XM_TX_PAUSE_CNT_LBN 16
 #define XM_TX_PAUSE_CNT_WIDTH 16
@@ -922,6 +931,15 @@
 #define XX_PWRDNC_EN_LBN 14
 #define XX_PWRDNC_EN_WIDTH 1
 #define XX_PWRDNB_EN_LBN 13
+/* XGMAC management interrupt status register */
+#define XM_MGT_INT_REG_MAC_B0 0x0f
+#define XM_PRMBLE_ERR 2
+#define XM_PRMBLE_WIDTH 1
+#define XM_RMTFLT_LBN 1
+#define XM_RMTFLT_WIDTH 1
+#define XM_LCLFLT_LBN 0
+#define XM_LCLFLT_WIDTH 1
+
 #define XX_PWRDNB_EN_WIDTH 1
 #define XX_PWRDNA_EN_LBN 12
 #define XX_PWRDNA_EN_WIDTH 1
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/falcon_io.h
--- a/drivers/net/sfc/falcon_io.h       Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/falcon_io.h       Tue Mar 31 11:59:10 2009 +0100
@@ -28,8 +28,9 @@
 #ifndef EFX_FALCON_IO_H
 #define EFX_FALCON_IO_H
 
+#include <linux/io.h>
+#include <linux/spinlock.h>
 #include "net_driver.h"
-#include "falcon.h"
 
 /**************************************************************************
  *
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/falcon_xmac.c
--- a/drivers/net/sfc/falcon_xmac.c     Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/falcon_xmac.c     Tue Mar 31 11:59:10 2009 +0100
@@ -236,6 +236,43 @@ int falcon_reset_xaui(struct efx_nic *ef
        return rc;
 }
 
+static int falcon_xgmii_status(struct efx_nic *efx)
+{
+       efx_dword_t reg;
+
+       if (FALCON_REV(efx) < FALCON_REV_B0)
+               return 1;
+
+       /* The ISR latches, so clear it and re-read */
+       efx->mac_op->mac_readl(efx, &reg, XM_MGT_INT_REG_MAC_B0);
+       efx->mac_op->mac_readl(efx, &reg, XM_MGT_INT_REG_MAC_B0);
+
+       if (EFX_DWORD_FIELD(reg, XM_LCLFLT) ||
+           EFX_DWORD_FIELD(reg, XM_RMTFLT)) {
+               EFX_INFO(efx, "MGT_INT: "EFX_DWORD_FMT"\n", EFX_DWORD_VAL(reg));
+               return 0;
+       }
+
+       return 1;
+}
+
+static void falcon_mask_status_intr(struct efx_nic *efx, int enable)
+{
+       efx_dword_t reg;
+
+       if ((FALCON_REV(efx) < FALCON_REV_B0) || LOOPBACK_INTERNAL(efx))
+               return;
+
+       /* Flush the ISR */
+       if (enable)
+               efx->mac_op->mac_readl(efx, &reg, XM_MGT_INT_REG_MAC_B0);
+
+       EFX_POPULATE_DWORD_2(reg,
+                            XM_MSK_RMTFLT, !enable,
+                            XM_MSK_LCLFLT, !enable);
+       efx->mac_op->mac_writel(efx, &reg, XM_MGT_INT_MSK_REG_MAC_B0);
+}
+
 static int falcon_init_xmac(struct efx_nic *efx)
 {
        int rc;
@@ -257,6 +294,7 @@ static int falcon_init_xmac(struct efx_n
        if (rc)
                goto fail2;
 
+       falcon_mask_status_intr(efx, 1);
        return 0;
 
  fail2:
@@ -269,9 +307,7 @@ int falcon_xaui_link_ok(struct efx_nic *
 int falcon_xaui_link_ok(struct efx_nic *efx)
 {
        efx_dword_t reg;
-       int align_done;
-       int sync_status;
-       int link_ok = 0;
+       int align_done, sync_status, link_ok = 0;
 
        /* If we're in internal loopback, then the link is up.
         * The A1 FPGA/4G has RX and TX XAUI wired together, so the link is up.
@@ -292,13 +328,25 @@ int falcon_xaui_link_ok(struct efx_nic *
        EFX_SET_DWORD_FIELD(reg, XX_COMMA_DET, XX_COMMA_DET_RESET);
        EFX_SET_DWORD_FIELD(reg, XX_CHARERR, XX_CHARERR_RESET);
        EFX_SET_DWORD_FIELD(reg, XX_DISPERR, XX_DISPERR_RESET);
-
        efx->mac_op->mac_writel(efx, &reg, XX_CORE_STAT_REG_MAC);
 
+       /* If the link is up, then check the phy side of the xaui link
+        * (error conditions from the wire side propoagate back through
+        * the phy to the xaui side). */
+       if (efx->link_up && link_ok) {
+               int has_phyxs = efx->phy_op->mmds & (1 << MDIO_MMD_PHYXS);
+               if (has_phyxs)
+                       link_ok = mdio_clause45_phyxgxs_lane_sync(efx);
+       }
+
+       /* If the PHY and XAUI links are up, then check the mac's xgmii
+        * fault state */
+       if (efx->link_up && link_ok)
+               link_ok = falcon_xgmii_status(efx);
+
        return link_ok;
 }
 
-/* Do most of the heavy lifting of falcon_reconfigure_xmac */
 static void falcon_reconfigure_xmac_core(struct efx_nic *efx)
 {
        unsigned int max_frame_len;
@@ -367,7 +415,7 @@ static void falcon_reconfigure_xmac_core
        efx->mac_op->mac_writel(efx, &reg, XM_ADR_HI_REG_MAC);
 
        /* Handle B0 FPGA loopback where RAMBUS XGXS block not present */
-       if (FALCON_REV(efx) == FALCON_REV_B0 && !efx->is_asic) {
+       if (FALCON_REV(efx) >= FALCON_REV_B0 && !efx->is_asic) {
                int xgmii_loopback =
                        (efx->loopback_mode == LOOPBACK_XGMII) ? 1 : 0;
 
@@ -378,7 +426,6 @@ static void falcon_reconfigure_xmac_core
        }
 }
 
-/* Do most of the heavy lifting of falcon_reconfigure_xmac */
 static void falcon_reconfigure_xgxs_core(struct efx_nic *efx)
 {
        efx_dword_t reg;
@@ -387,7 +434,7 @@ static void falcon_reconfigure_xgxs_core
        int xgmii_loopback =
                (efx->loopback_mode == LOOPBACK_XGMII) ? 1 : 0;
 
-       if (FALCON_REV(efx) == FALCON_REV_B0 && !efx->is_asic)
+       if (FALCON_REV(efx) >= FALCON_REV_B0 && !efx->is_asic)
                /* RAMBUS XGXS block is not present */
                return;
 
@@ -442,52 +489,46 @@ static void falcon_reconfigure_xgxs_core
 }
 
 
-/* Sometimes the XAUI link between Falcon and XFP fails to come up. The state
- * of the link is checked during phy_reconfigure(). After XAIU is reset then
- * the MAC must be reconfigured.
- */
-#define MAX_XAUI_TRIES (5)     /* It's never been seen to take more than 2 */
-
-void falcon_check_xaui_link_up(struct efx_nic *efx)
+/* Try and bring the Falcon side of the Falcon-Phy XAUI link fails
+ * to come back up. Bash it until it comes back up */
+static int falcon_check_xaui_link_up(struct efx_nic *efx)
 {
        int max_tries, tries;
-       tries = EFX_WORKAROUND_5147(efx) ? MAX_XAUI_TRIES : 1;
+       tries = EFX_WORKAROUND_5147(efx) ? 5 : 1;
        max_tries = tries;
 
        if ((efx->loopback_mode == LOOPBACK_NETWORK) ||
            (efx->phy_type == PHY_TYPE_NONE) ||
            !efx->phy_powered)
-               return;
+               return 0;
 
        while (tries) {
                if (falcon_xaui_link_ok(efx))
-                       return;
+                       return 1;
 
                EFX_LOG(efx, "%s Clobbering XAUI (%d tries left).\n",
                        __func__, tries);
                (void) falcon_reset_xaui(efx);
-               /* Cannot use full reconfigure. Need to avoid recursion */
-
-               /* Give the poor thing time to sort itself out: if we retry
-                * too fast it will never train. */
                udelay(200);
-
-               falcon_reconfigure_xgxs_core(efx);
-
                tries--;
        }
 
-       EFX_ERR(efx, "Failed to bring XAUI link back up in %d tries!\n",
+       EFX_LOG(efx, "Failed to bring XAUI link back up in %d tries!\n",
                max_tries);
+       return 0;
 }
 
 static void falcon_reconfigure_xmac(struct efx_nic *efx)
 {
+       int xaui_link_ok;
+
+       falcon_mask_status_intr(efx, 0);
+
+       /* Deconfigure the mac wrapper, draining the tx fifo if necessary */
        falcon_deconfigure_mac_wrapper(efx);
 
-       /* In internal loopback modes disable transmit */
+       /* Reconfigure the PHY, disabling transmit in mac level loopback. */
        efx->tx_disabled = LOOPBACK_INTERNAL(efx);
-
        efx->phy_op->reconfigure(efx);
 
        falcon_reconfigure_xgxs_core(efx);
@@ -496,8 +537,11 @@ static void falcon_reconfigure_xmac(stru
        /* Reconfigure MAC wrapper */
        falcon_reconfigure_mac_wrapper(efx);
 
-       /* Ensure XAUI link is up - might repeat reconfigure_xmac_core */
-       falcon_check_xaui_link_up(efx);
+       /* Ensure XAUI link is up */
+       xaui_link_ok = falcon_check_xaui_link_up(efx);
+
+       if (xaui_link_ok && efx->link_up)
+               falcon_mask_status_intr(efx, 1);
 }
 
 static void falcon_fini_xmac(struct efx_nic *efx)
@@ -576,29 +620,28 @@ static void falcon_update_stats_xmac(str
 
 static int falcon_check_xmac(struct efx_nic *efx)
 {
-       unsigned link_ok, phyxs_ok = 1;
-       unsigned has_phyxs = efx->phy_op->mmds & (1 << MDIO_MMD_PHYXS);
-
-       /* Check the remote XAUI link status */
-       link_ok = falcon_xaui_link_ok(efx);
+       unsigned xaui_link_ok;
+       int rc;
 
        if ((efx->loopback_mode == LOOPBACK_NETWORK) ||
+           (efx->phy_type == PHY_TYPE_NONE) ||
            !efx->phy_powered)
                return 0;
 
-       if (link_ok && has_phyxs && !LOOPBACK_INTERNAL(efx)) {
-               /* Does the PHYXS think we have lane sync? */
-               phyxs_ok = mdio_clause45_phyxgxs_lane_sync(efx);
-       }
-
-       if (EFX_WORKAROUND_5147(efx) && (!link_ok || !phyxs_ok)) {
+       falcon_mask_status_intr(efx, 0);
+       xaui_link_ok = falcon_xaui_link_ok(efx);
+
+       if (EFX_WORKAROUND_5147(efx) && !xaui_link_ok)
                (void) falcon_reset_xaui(efx);
-               falcon_reconfigure_xgxs_core(efx);
-       }
 
        /* Call the PHY check_hw routine */
-       efx->phy_op->check_hw(efx);
-       return 0;
+       rc = efx->phy_op->check_hw(efx);
+
+       /* Unmask interrupt if everything was (and still is) ok */
+       if (xaui_link_ok && efx->link_up)
+               falcon_mask_status_intr(efx, 1);
+
+       return rc;
 }
 
 /* Simulate a PHY event */
@@ -659,12 +702,10 @@ static int falcon_xmac_set_pause(struct 
        reset = ((flow_control & EFX_FC_TX) &&
                 !(efx->flow_control & EFX_FC_TX));
        if (EFX_WORKAROUND_11482(efx) && reset) {
-               if (FALCON_REV(efx) == FALCON_REV_B0) {
+               if (FALCON_REV(efx) >= FALCON_REV_B0) {
                        /* Recover by resetting the EM block */
-                       mutex_lock(&efx->mac_lock);
                        if (efx->link_up)
                                falcon_drain_tx_fifo(efx);
-                       mutex_unlock(&efx->mac_lock);
                } else {
                        /* Schedule a reset to recover */
                        efx_schedule_reset(efx, RESET_TYPE_INVISIBLE);
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/i2c-direct.c
--- a/drivers/net/sfc/i2c-direct.c      Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/i2c-direct.c      Tue Mar 31 11:59:10 2009 +0100
@@ -25,13 +25,13 @@
  ****************************************************************************
  */
 
-#include <asm/io.h>
 #include <linux/delay.h>
 #include "net_driver.h"
 #include "i2c-direct.h"
 
-/* EEPROM access via I2C
- * data (SDA) and clock (SCL) line read/writes
+/*
+ * I2C data (SDA) and clock (SCL) line read/writes with appropriate
+ * delays.
  */
 
 static inline void setsda(struct efx_i2c_interface *i2c, int state)
@@ -79,7 +79,7 @@ static inline void i2c_release(struct ef
 {
        EFX_WARN_ON_PARANOID(!i2c->scl);
        EFX_WARN_ON_PARANOID(!i2c->sda);
-       /* Just in case */
+       /* Devices may time out if operations do not end */
        setscl(i2c, 1);
        setsda(i2c, 1);
        EFX_BUG_ON_PARANOID(getsda(i2c) != 1);
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/kernel_compat.c
--- a/drivers/net/sfc/kernel_compat.c   Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/kernel_compat.c   Tue Mar 31 11:59:10 2009 +0100
@@ -25,8 +25,6 @@
  ****************************************************************************
  */
 
-#define EFX_IN_KCOMPAT_C 1
-
 #include "net_driver.h"
 #include <linux/mii.h>
 #include <linux/ethtool.h>
@@ -47,514 +45,22 @@
 
 /**************************************************************************
  *
- * GMII-friendly versions of mii_ethtool_[gs]set
- *
- **************************************************************************
- *
- * Kernels prior to 2.6.12 don't support GMII PHYs via
- * mii_ethtool_gset and mii_ethtool_sset.  These are those functions
- * taken from a 2.6.12 kernel tree, with the tests for
- * mii->supports_gmii removed (since that field doesn't exist in older
- * kernels).
- *
- */
-
-#ifdef EFX_NEED_MII_ETHTOOL_FIX
-int efx_mii_ethtool_gset(struct mii_if_info *mii, struct ethtool_cmd *ecmd)
-{
-       struct net_device *dev = mii->dev;
-       u32 advert, bmcr, lpa, nego;
-       u32 advert2 = 0, bmcr2 = 0, lpa2 = 0;
-
-       ecmd->supported =
-           (SUPPORTED_10baseT_Half | SUPPORTED_10baseT_Full |
-            SUPPORTED_100baseT_Half | SUPPORTED_100baseT_Full |
-            SUPPORTED_Autoneg | SUPPORTED_TP | SUPPORTED_MII);
-       ecmd->supported |= SUPPORTED_1000baseT_Half | SUPPORTED_1000baseT_Full;
-
-       /* only supports twisted-pair */
-       ecmd->port = PORT_MII;
-
-       /* only supports internal transceiver */
-       ecmd->transceiver = XCVR_INTERNAL;
-
-       /* this isn't fully supported at higher layers */
-       ecmd->phy_address = mii->phy_id;
-
-       ecmd->advertising = ADVERTISED_TP | ADVERTISED_MII;
-       advert = mii->mdio_read(dev, mii->phy_id, MII_ADVERTISE);
-       advert2 = mii->mdio_read(dev, mii->phy_id, MII_CTRL1000);
-
-       if (advert & ADVERTISE_10HALF)
-               ecmd->advertising |= ADVERTISED_10baseT_Half;
-       if (advert & ADVERTISE_10FULL)
-               ecmd->advertising |= ADVERTISED_10baseT_Full;
-       if (advert & ADVERTISE_100HALF)
-               ecmd->advertising |= ADVERTISED_100baseT_Half;
-       if (advert & ADVERTISE_100FULL)
-               ecmd->advertising |= ADVERTISED_100baseT_Full;
-       if (advert2 & ADVERTISE_1000HALF)
-               ecmd->advertising |= ADVERTISED_1000baseT_Half;
-       if (advert2 & ADVERTISE_1000FULL)
-               ecmd->advertising |= ADVERTISED_1000baseT_Full;
-
-       bmcr = mii->mdio_read(dev, mii->phy_id, MII_BMCR);
-       lpa = mii->mdio_read(dev, mii->phy_id, MII_LPA);
-       bmcr2 = mii->mdio_read(dev, mii->phy_id, MII_CTRL1000);
-       lpa2 = mii->mdio_read(dev, mii->phy_id, MII_STAT1000);
-       if (bmcr & BMCR_ANENABLE) {
-               ecmd->advertising |= ADVERTISED_Autoneg;
-               ecmd->autoneg = AUTONEG_ENABLE;
-
-               nego = mii_nway_result(advert & lpa);
-               if ((bmcr2 & (ADVERTISE_1000HALF | ADVERTISE_1000FULL)) &
-                   (lpa2 >> 2))
-                       ecmd->speed = SPEED_1000;
-               else if (nego == LPA_100FULL || nego == LPA_100HALF)
-                       ecmd->speed = SPEED_100;
-               else
-                       ecmd->speed = SPEED_10;
-               if ((lpa2 & LPA_1000FULL) || nego == LPA_100FULL ||
-                   nego == LPA_10FULL) {
-                       ecmd->duplex = DUPLEX_FULL;
-                       mii->full_duplex = 1;
-               } else {
-                       ecmd->duplex = DUPLEX_HALF;
-                       mii->full_duplex = 0;
-               }
-       } else {
-               ecmd->autoneg = AUTONEG_DISABLE;
-
-               ecmd->speed = ((bmcr & BMCR_SPEED1000 &&
-                               (bmcr & BMCR_SPEED100) == 0) ? SPEED_1000 :
-                              (bmcr & BMCR_SPEED100) ? SPEED_100 : SPEED_10);
-               ecmd->duplex =
-                       (bmcr & BMCR_FULLDPLX) ? DUPLEX_FULL : DUPLEX_HALF;
-       }
-
-       /* ignore maxtxpkt, maxrxpkt for now */
-
-       return 0;
-}
-
-int efx_mii_ethtool_sset(struct mii_if_info *mii, struct ethtool_cmd *ecmd)
-{
-       struct net_device *dev = mii->dev;
-
-       if (ecmd->speed != SPEED_10 &&
-           ecmd->speed != SPEED_100 &&
-           ecmd->speed != SPEED_1000)
-               return -EINVAL;
-       if (ecmd->duplex != DUPLEX_HALF && ecmd->duplex != DUPLEX_FULL)
-               return -EINVAL;
-       if (ecmd->port != PORT_MII)
-               return -EINVAL;
-       if (ecmd->transceiver != XCVR_INTERNAL)
-               return -EINVAL;
-       if (ecmd->phy_address != mii->phy_id)
-               return -EINVAL;
-       if (ecmd->autoneg != AUTONEG_DISABLE && ecmd->autoneg != AUTONEG_ENABLE)
-               return -EINVAL;
-
-       /* ignore supported, maxtxpkt, maxrxpkt */
-
-       if (ecmd->autoneg == AUTONEG_ENABLE) {
-               u32 bmcr, advert, tmp;
-               u32 advert2 = 0, tmp2 = 0;
-
-               if ((ecmd->advertising & (ADVERTISED_10baseT_Half |
-                                         ADVERTISED_10baseT_Full |
-                                         ADVERTISED_100baseT_Half |
-                                         ADVERTISED_100baseT_Full |
-                                         ADVERTISED_1000baseT_Half |
-                                         ADVERTISED_1000baseT_Full)) == 0)
-                       return -EINVAL;
-
-               /* advertise only what has been requested */
-               advert = mii->mdio_read(dev, mii->phy_id, MII_ADVERTISE);
-               tmp = advert & ~(ADVERTISE_ALL | ADVERTISE_100BASE4);
-               advert2 = mii->mdio_read(dev, mii->phy_id, MII_CTRL1000);
-               tmp2 = advert2 & ~(ADVERTISE_1000HALF | ADVERTISE_1000FULL);
-               if (ecmd->advertising & ADVERTISED_10baseT_Half)
-                       tmp |= ADVERTISE_10HALF;
-               if (ecmd->advertising & ADVERTISED_10baseT_Full)
-                       tmp |= ADVERTISE_10FULL;
-               if (ecmd->advertising & ADVERTISED_100baseT_Half)
-                       tmp |= ADVERTISE_100HALF;
-               if (ecmd->advertising & ADVERTISED_100baseT_Full)
-                       tmp |= ADVERTISE_100FULL;
-               if (ecmd->advertising & ADVERTISED_1000baseT_Half)
-                       tmp2 |= ADVERTISE_1000HALF;
-               if (ecmd->advertising & ADVERTISED_1000baseT_Full)
-                       tmp2 |= ADVERTISE_1000FULL;
-               if (advert != tmp) {
-                       mii->mdio_write(dev, mii->phy_id, MII_ADVERTISE, tmp);
-                       mii->advertising = tmp;
-               }
-               if (advert2 != tmp2)
-                       mii->mdio_write(dev, mii->phy_id, MII_CTRL1000, tmp2);
-
-               /* turn on autonegotiation, and force a renegotiate */
-               bmcr = mii->mdio_read(dev, mii->phy_id, MII_BMCR);
-               bmcr |= (BMCR_ANENABLE | BMCR_ANRESTART);
-               mii->mdio_write(dev, mii->phy_id, MII_BMCR, bmcr);
-
-               mii->force_media = 0;
-       } else {
-               u32 bmcr, tmp;
-
-               /* turn off auto negotiation, set speed and duplexity */
-               bmcr = mii->mdio_read(dev, mii->phy_id, MII_BMCR);
-               tmp = bmcr & ~(BMCR_ANENABLE | BMCR_SPEED100 |
-                              BMCR_SPEED1000 | BMCR_FULLDPLX);
-               if (ecmd->speed == SPEED_1000)
-                       tmp |= BMCR_SPEED1000;
-               else if (ecmd->speed == SPEED_100)
-                       tmp |= BMCR_SPEED100;
-               if (ecmd->duplex == DUPLEX_FULL) {
-                       tmp |= BMCR_FULLDPLX;
-                       mii->full_duplex = 1;
-               } else {
-                       mii->full_duplex = 0;
-               }
-               if (bmcr != tmp)
-                       mii->mdio_write(dev, mii->phy_id, MII_BMCR, tmp);
-
-               mii->force_media = 1;
-       }
-       return 0;
-}
-#endif /* NEED_EFX_MII_ETHTOOL_GSET */
-
-/**************************************************************************
- *
- * unregister_netdevice_notifier : Has a race before 2.6.17
- *
- **************************************************************************
- *
- */
-
-#ifdef EFX_NEED_UNREGISTER_NETDEVICE_NOTIFIER_FIX
-/**
- * efx_unregister_netdevice_notifier - fixed unregister_netdevice_notifier
- * @nb:                notifier to unregister
- *
- * unregister_netdevice_notifier() does not wait for the notifier
- * to be unused before 2.6.17.  This wrapper fixes that.
- */
-int efx_unregister_netdevice_notifier(struct notifier_block *nb)
-{
-       int res;
-
-       res = unregister_netdevice_notifier(nb);
-       /* Ensure any outstanding calls complete. */
-       rtnl_lock();
-       rtnl_unlock();
-       return res;
-}
-#endif /* NEED_EFX_UNREGISTER_NETDEVICE_NOTIFIER */
-
-/**************************************************************************
- *
- * IOMMU-locking versions of pci_[un]map_single and
- * pci_{alloc,free}_consistent.  See SFC bug 4560.
- *
- **************************************************************************
- *
- */
-#ifdef EFX_NEED_IOMMU_LOCK
-
-/*
- * efx_use_iommu_lock - IOMMU lock use control
- *
- * If set to 1, the driver will attempt to mitigate the race condition
- * bug around IOMMU accesses in some 2.6 kernels.  If set to 2, the
- * driver will use the lock even if it thinks it doesn't need to.
- * Note that this is only a best-effort attempt; in particular, we
- * cannot do anything about other drivers touching the IOMMU.
- */
-static unsigned int efx_use_iommu_lock = 1;
-EXPORT_SYMBOL(efx_use_iommu_lock);
-
-/*
- * efx_iommu_lock - lock around IOMMU accesses
- *
- * This spinlock should be held while calling functions that access
- * the IOMMU if efx_use_iommu_lock is >= 2.  The efx_pci_*()
- * functions do this where possible.
- */
-static spinlock_t efx_iommu_lock = SPIN_LOCK_UNLOCKED;
-EXPORT_SYMBOL(efx_iommu_lock);
-
-/* Don't use the IOMMU lock if the device can access the whole of memory */
-#define EFX_DMA_CONSISTENT(_efx)                       \
-       (((_efx)->dma_mask >> PAGE_SHIFT) >= max_pfn)
-/**
- * efx_pci_map_single - map buffer for DMA, under IOMMU lock
- * @pci:               PCI device
- * @ptr:               Buffer
- * @size:              Buffer length
- * @direction:         DMA direction
- *
- * Wrapper for pci_map_single that uses efx_iommu_lock if necessary.
- */
-dma_addr_t efx_pci_map_single(struct pci_dev *pci, void *ptr, size_t size,
-                             int direction)
-{
-       struct efx_nic *efx = pci_get_drvdata(pci);
-       unsigned long flags __attribute__ ((unused));
-       dma_addr_t dma_addr;
-
-       if (unlikely((efx_use_iommu_lock &&
-                     (!EFX_NO_IOMMU) &&
-                     (!EFX_DMA_CONSISTENT(efx))) ||
-                    efx_use_iommu_lock >= 2)) {
-               spin_lock_irqsave(&efx_iommu_lock, flags);
-               dma_addr = pci_map_single(pci, ptr, size, direction);
-               spin_unlock_irqrestore(&efx_iommu_lock, flags);
-       } else {
-               dma_addr = pci_map_single(pci, ptr, size, direction);
-       }
-       return dma_addr;
-}
-
-/**
- * efx_pci_unmap_single - unmap buffer for DMA, under IOMMU lock
- * @pci:               PCI device
- * @dma_addr:          DMA address
- * @size:              Buffer length
- * @direction:         DMA direction
- *
- * Wrapper for pci_unmap_single that uses efx_iommu_lock if necessary.
- */
-void efx_pci_unmap_single(struct pci_dev *pci, dma_addr_t dma_addr,
-                         size_t size, int direction)
-{
-       struct efx_nic *efx = pci_get_drvdata(pci);
-       unsigned long flags __attribute__ ((unused));
-
-       if (unlikely((efx_use_iommu_lock &&
-                     (!EFX_NO_IOMMU) &&
-                     (!EFX_DMA_CONSISTENT(efx))) ||
-                    efx_use_iommu_lock >= 2)) {
-               spin_lock_irqsave(&efx_iommu_lock, flags);
-               pci_unmap_single(pci, dma_addr, size, direction);
-               spin_unlock_irqrestore(&efx_iommu_lock, flags);
-       } else {
-               pci_unmap_single(pci, dma_addr, size, direction);
-       }
-}
-
-/**
- * efx_pci_alloc_consistent - allocate DMA-consistent buffer, under IOMMU lock
- * @pci:               PCI device
- * @size:              Buffer length
- * @dma_addr:          DMA address
- *
- * Wrapper for pci_alloc_consistent that uses efx_iommu_lock if necessary.
- *
- * Bugs: Currently this can't use the spinlock because
- *     pci_alloc_consistent may block.
- */
-void *efx_pci_alloc_consistent(struct pci_dev *pci, size_t size,
-                              dma_addr_t *dma_addr)
-{
-       return pci_alloc_consistent(pci, size, dma_addr);
-}
-
-/**
- * efx_pci_free_consistent - free DMA-consistent buffer, under IOMMU lock
- * @pci:               PCI device
- * @size:              Buffer length
- * @ptr:               Buffer
- * @dma_addr:          DMA address
- *
- * Wrapper for pci_free_consistent that uses efx_iommu_lock if necessary.
- */
-void efx_pci_free_consistent(struct pci_dev *pci, size_t size, void *ptr,
-                            dma_addr_t dma_addr)
-{
-       struct efx_nic *efx = pci_get_drvdata(pci);
-       unsigned long flags __attribute__ ((unused));
-
-       if (unlikely((efx_use_iommu_lock &&
-                     (!EFX_NO_IOMMU) &&
-                     (!EFX_DMA_CONSISTENT(efx))) ||
-                    efx_use_iommu_lock >= 2)) {
-               spin_lock_irqsave(&efx_iommu_lock, flags);
-               pci_free_consistent(pci, size, ptr, dma_addr);
-               spin_unlock_irqrestore(&efx_iommu_lock, flags);
-       } else {
-               pci_free_consistent(pci, size, ptr, dma_addr);
-       }
-}
-
-module_param(efx_use_iommu_lock, uint, 0644);
-MODULE_PARM_DESC(efx_use_iommu_lock, "Enable lock for bug in free_iommu");
-
-#endif
-
-#ifdef EFX_NEED_COMPOUND_PAGE_FIX
-
-void efx_compound_page_destructor(struct page *page)
-{
-       /* Fake up page state to keep __free_pages happy */
-       set_page_count(page, 1);
-       page[1].mapping = NULL;
-
-       __free_pages(page, (unsigned long)page[1].index);
-}
-
-#endif /* NEED_COMPOUND_PAGE_FIX */
-
-/**************************************************************************
- *
- * print_hex_dump, taken from lib/hexdump.c.
- *
- **************************************************************************
- *
- */
-#ifdef EFX_NEED_HEX_DUMP
-
-#define hex_asc(x)     "0123456789abcdef"[x]
-#define isascii(c) (((unsigned char)(c))<=0x7f)
-
-static void hex_dump_to_buffer(const void *buf, size_t len, int rowsize,
-                              int groupsize, char *linebuf, size_t linebuflen,
-                              int ascii)
-{
-        const u8 *ptr = buf;
-        u8 ch;
-        int j, lx = 0;
-        int ascii_column;
-
-        if (rowsize != 16 && rowsize != 32)
-                rowsize = 16;
-
-        if (!len)
-                goto nil;
-        if (len > rowsize)              /* limit to one line at a time */
-                len = rowsize;
-        if ((len % groupsize) != 0)     /* no mixed size output */
-                groupsize = 1;
-
-        switch (groupsize) {
-        case 8: {
-                const u64 *ptr8 = buf;
-                int ngroups = len / groupsize;
-
-                for (j = 0; j < ngroups; j++)
-                        lx += scnprintf(linebuf + lx, linebuflen - lx,
-                               "%16.16llx ", (unsigned long long)*(ptr8 + j));
-                ascii_column = 17 * ngroups + 2;
-                break;
-        }
-
-        case 4: {
-                const u32 *ptr4 = buf;
-                int ngroups = len / groupsize;
-
-                for (j = 0; j < ngroups; j++)
-                        lx += scnprintf(linebuf + lx, linebuflen - lx,
-                               "%8.8x ", *(ptr4 + j));
-                ascii_column = 9 * ngroups + 2;
-                break;
-        }
-
-        case 2: {
-                const u16 *ptr2 = buf;
-                int ngroups = len / groupsize;
-
-                for (j = 0; j < ngroups; j++)
-                        lx += scnprintf(linebuf + lx, linebuflen - lx,
-                               "%4.4x ", *(ptr2 + j));
-                ascii_column = 5 * ngroups + 2;
-                break;
-        }
-
-        default:
-                for (j = 0; (j < rowsize) && (j < len) && (lx + 4) < 
linebuflen;
-                     j++) {
-                        ch = ptr[j];
-                        linebuf[lx++] = hex_asc(ch >> 4);
-                        linebuf[lx++] = hex_asc(ch & 0x0f);
-                        linebuf[lx++] = ' ';
-                }
-                ascii_column = 3 * rowsize + 2;
-                break;
-        }
-        if (!ascii)
-                goto nil;
-
-        while (lx < (linebuflen - 1) && lx < (ascii_column - 1))
-                linebuf[lx++] = ' ';
-       /* Removed is_print() check */
-        for (j = 0; (j < rowsize) && (j < len) && (lx + 2) < linebuflen; j++)
-                linebuf[lx++] = isascii(ptr[j]) ? ptr[j] : '.';
-nil:
-        linebuf[lx++] = '\0';
-}
-
-void print_hex_dump(const char *level, const char *prefix_str, int prefix_type,
-                   int rowsize, int groupsize,
-                   const void *buf, size_t len, int ascii)
-{
-        const u8 *ptr = buf;
-        int i, linelen, remaining = len;
-        char linebuf[200];
-
-        if (rowsize != 16 && rowsize != 32)
-                rowsize = 16;
-
-        for (i = 0; i < len; i += rowsize) {
-                linelen = min(remaining, rowsize);
-                remaining -= rowsize;
-                hex_dump_to_buffer(ptr + i, linelen, rowsize, groupsize,
-                                  linebuf, sizeof(linebuf), ascii);
-
-                switch (prefix_type) {
-                case DUMP_PREFIX_ADDRESS:
-                        printk("%s%s%*p: %s\n", level, prefix_str,
-                              (int)(2 * sizeof(void *)), ptr + i, linebuf);
-                        break;
-                case DUMP_PREFIX_OFFSET:
-                        printk("%s%s%.8x: %s\n", level, prefix_str, i, 
linebuf);
-                        break;
-                default:
-                        printk("%s%s%s\n", level, prefix_str, linebuf);
-                        break;
-                }
-        }
-}
-
-#endif /* EFX_NEED_HEX_DUMP */
-
-/**************************************************************************
- *
  * print_mac, from net/ethernet/eth.c in v2.6.24
  *
  **************************************************************************
  *
  */
-#ifdef EFX_NEED_PRINT_MAC
 char *print_mac(char *buf, const u8 *addr)
 {
         sprintf(buf, "%02x:%02x:%02x:%02x:%02x:%02x",
                 addr[0], addr[1], addr[2], addr[3], addr[4], addr[5]);
         return buf;
 }
-#endif /* EFX_NEED_PRINT_MAC */
 
 #ifdef EFX_NEED_CSUM_TCPUDP_NOFOLD
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,20)
-__wsum
-csum_tcpudp_nofold (__be32 saddr, __be32 daddr, unsigned short len,
-                   unsigned short proto, __wsum sum)
-#else
 __wsum
 csum_tcpudp_nofold (unsigned long saddr, unsigned long daddr,
                    unsigned short len, unsigned short proto, __wsum sum)
-#endif
 {
        unsigned long result;
 
@@ -570,85 +76,3 @@ csum_tcpudp_nofold (unsigned long saddr,
 
 }
 #endif /* EFX_NEED_CSUM_TCPUDP_NOFOLD */
-
-#ifdef EFX_NEED_RANDOM_ETHER_ADDR
-/* Generate random MAC address */
-void efx_random_ether_addr(uint8_t *addr) {
-        get_random_bytes (addr, ETH_ALEN);
-       addr [0] &= 0xfe;       /* clear multicast bit */
-       addr [0] |= 0x02;       /* set local assignment bit (IEEE802) */
-}
-#endif /* EFX_NEED_RANDOM_ETHER_ADDR */
-
-#ifdef EFX_NEED_MSECS_TO_JIFFIES
-/*
- * When we convert to jiffies then we interpret incoming values
- * the following way:
- *
- * - negative values mean 'infinite timeout' (MAX_JIFFY_OFFSET)
- *
- * - 'too large' values [that would result in larger than
- *   MAX_JIFFY_OFFSET values] mean 'infinite timeout' too.
- *
- * - all other values are converted to jiffies by either multiplying
- *   the input value by a factor or dividing it with a factor
- *
- * We must also be careful about 32-bit overflows.
- */
-#ifndef MSEC_PER_SEC
-#define MSEC_PER_SEC   1000L
-#endif
-unsigned long msecs_to_jiffies(const unsigned int m)
-{
-       /*
-        * Negative value, means infinite timeout:
-        */
-       if ((int)m < 0)
-               return MAX_JIFFY_OFFSET;
-
-#if HZ <= MSEC_PER_SEC && !(MSEC_PER_SEC % HZ)
-       /*
-        * HZ is equal to or smaller than 1000, and 1000 is a nice
-        * round multiple of HZ, divide with the factor between them,
-        * but round upwards:
-        */
-       return (m + (MSEC_PER_SEC / HZ) - 1) / (MSEC_PER_SEC / HZ);
-#elif HZ > MSEC_PER_SEC && !(HZ % MSEC_PER_SEC)
-       /*
-        * HZ is larger than 1000, and HZ is a nice round multiple of
-        * 1000 - simply multiply with the factor between them.
-        *
-        * But first make sure the multiplication result cannot
-        * overflow:
-        */
-       if (m > jiffies_to_msecs(MAX_JIFFY_OFFSET))
-               return MAX_JIFFY_OFFSET;
-
-       return m * (HZ / MSEC_PER_SEC);
-#else
-       /*
-        * Generic case - multiply, round and divide. But first
-        * check that if we are doing a net multiplication, that
-        * we wouldnt overflow:
-        */
-       if (HZ > MSEC_PER_SEC && m > jiffies_to_msecs(MAX_JIFFY_OFFSET))
-               return MAX_JIFFY_OFFSET;
-
-       return (m * HZ + MSEC_PER_SEC - 1) / MSEC_PER_SEC;
-#endif
-}
-#endif /* EFX_NEED_MSECS_TO_JIFFIES */
-
-#ifdef EFX_NEED_MSLEEP
-/**
- * msleep - sleep safely even with waitqueue interruptions
- * @msecs: Time in milliseconds to sleep for
- */
-void msleep(unsigned int msecs)
-{
-       unsigned long timeout = msecs_to_jiffies(msecs) + 1;
-
-       while (timeout)
-               timeout = schedule_timeout_uninterruptible(timeout);
-}
-#endif
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/kernel_compat.h
--- a/drivers/net/sfc/kernel_compat.h   Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/kernel_compat.h   Tue Mar 31 11:59:10 2009 +0100
@@ -37,406 +37,32 @@
 #include <linux/interrupt.h>
 #include <linux/skbuff.h>
 #include <linux/netdevice.h>
-
-#include "extraversion.h"
+#include <linux/rtnetlink.h>
 
 /*
  * Kernel backwards compatibility
  *
- * This file provides macros that enable the driver to be compiled on
- * any kernel from 2.6.9 onward (plus SLES 9 2.6.5), without requiring
- * explicit version tests scattered throughout the code.
+ * This file provides macros to facilitate backporting the driver.
  */
 
-/**************************************************************************
- *
- * Version/config/architecture tests to set feature flags
- *
- **************************************************************************
- *
- * NOTE: For simplicity, these initial version tests cover kernel.org
- * releases only.  Backported features in "enterprise" kernels are
- * handled further down.
- */
-
-#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,9) &&       \
-       !(defined(EFX_DIST_SUSE) &&                      \
-         LINUX_VERSION_CODE == KERNEL_VERSION(2,6,5) && \
-         EFX_DIST_KVER_LEVEL_1 == 7)
-       #error "This kernel version is now unsupported"
+#ifdef __ia64__
+       /* csum_tcpudp_nofold() is extern but not exported */
+       #define EFX_NEED_CSUM_TCPUDP_NOFOLD yes
 #endif
-
-#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,6)
-       #define EFX_NEED_RANDOM_ETHER_ADDR yes
-#endif
-
-#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,7)
-       #define EFX_NEED_I2C_CLASS_HWMON yes
-       #define EFX_NEED_IF_MII yes
-       #define EFX_NEED_MSLEEP yes
-       #define EFX_NEED_MSECS_TO_JIFFIES yes
-#endif
-
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,8)
-       #define EFX_USE_MTD_ERASE_FAIL_ADDR yes
-#else
-       #define EFX_NEED_MTD_ERASE_CALLBACK yes
-       #define EFX_NEED_DUMMY_PCI_DISABLE_MSI yes
-       #define EFX_NEED_DUMMY_MSIX yes
-#endif
-
-#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,9)
-       #define EFX_NEED_BYTEORDER_TYPES yes
-#endif
-
-#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,10)
-       #define EFX_NEED_MMIOWB yes
-       #define EFX_NEED_PCI_SAVE_RESTORE_WRAPPERS yes
-#endif
-
-#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,12)
-       #define EFX_NEED_DUMMY_SUPPORTS_GMII yes
-       #define EFX_NEED_MII_CONSTANTS yes
-       #define EFX_NEED_MII_ETHTOOL_FIX yes
-       #define EFX_HAVE_MSIX_TABLE_RESERVED yes
-#endif
-
-#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,14)
-       #define EFX_NEED_SCHEDULE_TIMEOUT_INTERRUPTIBLE yes
-       #define EFX_NEED_SCHEDULE_TIMEOUT_UNINTERRUPTIBLE yes
-       #define EFX_NEED_GFP_T yes
-       #define EFX_NEED_KZALLOC yes
-#endif
-
-#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
-       #define EFX_NEED_SETUP_TIMER yes
-       #ifdef CONFIG_HUGETLB_PAGE
-               #define EFX_USE_COMPOUND_PAGES yes
-       #endif
-#else
-       #define EFX_USE_COMPOUND_PAGES yes
-#endif
-
-#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,16)
-       #define EFX_NEED_MUTEX yes
-       #define EFX_NEED_SAFE_LISTS yes
-       #ifdef EFX_USE_COMPOUND_PAGES
-               #define EFX_NEED_COMPOUND_PAGE_FIX yes
-       #endif
-#endif
-
-#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17)
-       #define EFX_NEED_UNREGISTER_NETDEVICE_NOTIFIER_FIX yes
-       #define EFX_NEED_DEV_NOTICE yes
-#endif
-
-#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18)
-       #define EFX_NEED_IRQF_FLAGS yes
-       #define EFX_NEED_NETDEV_ALLOC_SKB yes
-       /* Fedora backported 2.6.18 netdevice.h changes */
-       #ifndef NETIF_F_GSO
-               #define EFX_NEED_NETIF_TX_LOCK yes
-       #endif
-#else
-       #define EFX_USE_MTD_WRITESIZE yes
-#endif
-
-#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,19)
-       #define EFX_NEED_IRQ_HANDLER_T yes
-       #define EFX_HAVE_IRQ_HANDLER_REGS yes
-#endif
-
-#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20)
-       #define EFX_NEED_WORK_API_WRAPPERS yes
-       #define EFX_USE_FASTCALL yes
-       #define EFX_NEED_CSUM_UNFOLDED yes
-#endif
-
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,21)
-       /*
-        * debugfs was introduced earlier, but only supports sym-links
-        * from 2.6.21
-        */
-       #ifdef CONFIG_DEBUG_FS
-               #define EFX_USE_DEBUGFS yes
-       #endif
-#endif
-
-#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,22)
-       #define EFX_NEED_SKB_HEADER_MACROS yes
-       #define EFX_NEED_HEX_DUMP yes
-#else
-       #define EFX_USE_CANCEL_WORK_SYNC yes
-#endif
-
-#if LINUX_VERSION_CODE == KERNEL_VERSION(2,6,22)
-       #define EFX_NEED_HEX_DUMP_CONST_FIX yes
-#endif
-
-#if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)) && \
-    (LINUX_VERSION_CODE <  KERNEL_VERSION(2,6,23))
-       #define EFX_USE_ETHTOOL_GET_PERM_ADDR yes
-#endif
-
-#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
-       #ifdef __ia64__
-               /* csum_tcpudp_nofold() is extern but not exported */
-               #define EFX_NEED_CSUM_TCPUDP_NOFOLD yes
-       #endif
-#else
-       #define EFX_USE_PCI_DEV_REVISION yes
-       #define EFX_USE_CANCEL_DELAYED_WORK_SYNC yes
-#endif
-
-#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,24)
-       #define EFX_HAVE_OLD_NAPI yes
-       #define EFX_NEED_GENERIC_LRO yes
-       #define EFX_NEED_PRINT_MAC yes
-#else
-       #define EFX_USE_ETHTOOL_FLAGS yes
-#endif
-
-/*
- * SFC Bug 4560: Some kernels leak IOMMU entries under heavy load.  Use a
- * spinlock to serialise access where possible to alleviate the
- * problem.
- *
- * NB. The following definition is duplicated in
- * the char driver.  Please keep in sync.
- */
-#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17) && \
-     defined(__x86_64__) && defined(CONFIG_SMP))
-       #define EFX_NEED_IOMMU_LOCK yes
-       #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
-               #if defined(CONFIG_GART_IOMMU)
-                       #define EFX_NO_IOMMU no_iommu
-               #else
-                       #define EFX_NO_IOMMU 1
-               #endif
-       #else
-               #define EFX_NO_IOMMU 0
-       #endif
-#endif
-
 #ifdef CONFIG_PPC64
        /* __raw_writel and friends are broken on ppc64 */
        #define EFX_NEED_RAW_READ_AND_WRITE_FIX yes
 #endif
 
-/**************************************************************************
- *
- * Exceptions for backported features
- *
- **************************************************************************
- */
+typedef irqreturn_t (*irq_handler_t)(int, void *, struct pt_regs *);
 
-/* RHEL4 */
-#if defined(EFX_DIST_RHEL) && LINUX_VERSION_CODE == KERNEL_VERSION(2,6,9)
-       #if EFX_DIST_KVER_LEVEL_1 >= 22
-               /* linux-2.6.9-mmiowb.patch */
-               #undef EFX_NEED_MMIOWB
-       #endif
-       #if EFX_DIST_KVER_LEVEL_1 >= 34
-               /* linux-2.6.9-net-mii-update.patch */
-               #undef EFX_NEED_DUMMY_SUPPORTS_GMII
-               #undef EFX_NEED_MII_CONSTANTS
-               #undef EFX_NEED_MII_ETHTOOL_FIX
-               /* linux-2.6.9-gfp_t-typedef.patch */
-               #undef EFX_NEED_GFP_T
-               /* linux-2.6.9-slab-update.patch */
-               #undef EFX_NEED_KZALLOC
-       #endif
-       #if EFX_DIST_KVER_LEVEL_1 >= 55
-               /* linux-2.6.18-sata-update.patch (ported from 2.6.18->2.6.9) */
-               #undef EFX_NEED_SCHEDULE_TIMEOUT_UNINTERRUPTIBLE
-               #undef EFX_NEED_IRQ_HANDLER_T
-       #endif
-#endif
+#define skb_mac_header(skb)    (skb)->mac.raw
+#define skb_network_header(skb) (skb)->nh.raw
+#define eth_hdr(skb)           ((struct ethhdr *)skb_mac_header(skb))
+#define tcp_hdr(skb)           (skb)->h.th
+#define ip_hdr(skb)            (skb)->nh.iph
+#define skb_tail_pointer(skb)   (skb)->tail
 
-/* RHEL5 */
-#if defined(EFX_DIST_RHEL) && LINUX_VERSION_CODE == KERNEL_VERSION(2,6,18)
-       #if EFX_DIST_KVER_LEVEL_1 >= 53
-               /* linux-2.6.18-sata-update.patch */
-               #undef EFX_NEED_SCHEDULE_TIMEOUT_UNINTERRUPTIBLE
-               #undef EFX_NEED_IRQ_HANDLER_T
-       #endif
-#endif
-
-#if defined(EFX_DIST_RHEL)
-       #if (LINUX_VERSION_CODE != KERNEL_VERSION(2,6,9)) && \
-            (LINUX_VERSION_CODE != KERNEL_VERSION(2,6,18))
-               #error "Unknown Red Hat Enterprise kernel version"
-       #endif
-#endif
-
-/* SLES9 */
-#if defined(EFX_DIST_SUSE) && LINUX_VERSION_CODE == KERNEL_VERSION(2,6,5) && \
-       EFX_DIST_KVER_LEVEL_1 == 7
-       #if EFX_DIST_KVER_LEVEL_2 >= 139
-               #undef EFX_NEED_MMIOWB
-       #endif
-       #if EFX_DIST_KVER_LEVEL_2 >= 191
-               #undef EFX_NEED_MSLEEP
-               #undef EFX_NEED_MSECS_TO_JIFFIES
-       #endif
-       #if EFX_DIST_KVER_LEVEL_2 >= 244
-               #undef EFX_NEED_BYTEORDER_TYPES
-       #endif
-       #if EFX_DIST_KVER_LEVEL_2 >= 252
-               #undef EFX_NEED_KZALLOC
-       #endif
-#endif
-
-/**************************************************************************
- *
- * Definitions of missing constants, types, functions and macros
- *
- **************************************************************************
- *
- */
-
-#ifndef DMA_40BIT_MASK
-       #define DMA_40BIT_MASK  0x000000ffffffffffULL
-#endif
-
-#ifndef spin_trylock_irqsave
-       #define spin_trylock_irqsave(lock, flags)       \
-       ({                                              \
-               local_irq_save(flags);                  \
-               spin_trylock(lock) ?                    \
-               1 : ({local_irq_restore(flags); 0;});   \
-       })
-#endif
-
-#ifndef raw_smp_processor_id
-       #define raw_smp_processor_id() (current_thread_info()->cpu)
-#endif
-
-#ifndef NETIF_F_LRO
-       #define NETIF_F_LRO 0
-#endif
-
-/* Cope with small changes in PCI constants between minor kernel revisions */
-#if PCI_X_STATUS != 4
-       #undef PCI_X_STATUS
-       #define PCI_X_STATUS 4
-       #undef PCI_X_STATUS_MAX_SPLIT
-       #define PCI_X_STATUS_MAX_SPLIT 0x03800000
-#endif
-
-#ifndef PCI_EXP_LNKSTA
-       #define PCI_EXP_LNKSTA          18          /* Link Status */
-#endif
-
-/* Used for struct pt_regs */
-#ifndef regs_return_value
-       #if defined(__x86_64__)
-               #define regs_return_value(regs) ((regs)->rax)
-       #elif defined(__i386__)
-               #define regs_return_value(regs) ((regs)->eax)
-       #elif defined(__ia64__)
-               #define regs_return_value(regs) ((regs)->r8)
-       #else
-               #error "Need definition for regs_return_value()"
-       #endif
-#endif
-
-#ifndef __GFP_COMP
-       #define __GFP_COMP 0
-#endif
-
-#ifndef __iomem
-       #define __iomem
-#endif
-
-#ifndef NET_IP_ALIGN
-       #define NET_IP_ALIGN 2
-#endif
-
-#ifndef PCI_CAP_ID_EXP
-#define PCI_CAP_ID_EXP         0x10    /* PCI Express */
-#endif
-
-#ifndef PCI_EXP_FLAGS
-#define PCI_EXP_FLAGS           2           /* Capabilities register */
-#define PCI_EXP_FLAGS_TYPE      0x00f0      /* Capability version */
-#define PCI_EXP_TYPE_ENDPOINT   0x0         /* Express Endpoint */
-#define PCI_EXP_TYPE_LEG_END    0x1         /* Legacy Endpoint */
-#define PCI_EXP_TYPE_ROOT_PORT  0x4         /* Root Port */
-#endif
-
-#ifndef PCI_EXP_DEVCAP
-#define PCI_EXP_DEVCAP          4           /* Device capabilities */
-#define PCI_EXP_DEVCAP_PAYLOAD  0x07        /* Max_Payload_Size */
-#define PCI_EXP_DEVCAP_PWR_VAL  0x3fc0000   /* Slot Power Limit Value */
-#define PCI_EXP_DEVCAP_PWR_SCL  0xc000000   /* Slot Power Limit Scale */
-#endif
-
-#ifndef PCI_EXP_DEVCTL
-#define PCI_EXP_DEVCTL          8           /* Device Control */
-#define PCI_EXP_DEVCTL_PAYLOAD  0x00e0      /* Max_Payload_Size */
-#define PCI_EXP_DEVCTL_READRQ   0x7000      /* Max_Read_Request_Size */
-#endif
-
-#ifndef PCI_EXP_LNKSTA
-#define PCI_EXP_LNKSTA         18          /* Link Status */
-#endif
-
-#ifndef NETDEV_TX_OK
-       #define NETDEV_TX_OK 0
-#endif
-
-#ifndef NETDEV_TX_BUSY
-       #define NETDEV_TX_BUSY 1
-#endif
-
-#ifndef __force
-       #define __force
-#endif
-
-#if ! defined(for_each_cpu_mask) && ! defined(CONFIG_SMP)
-       #define for_each_cpu_mask(cpu, mask)            \
-               for ((cpu) = 0; (cpu) < 1; (cpu)++, (void)mask)
-#endif
-
-/**************************************************************************/
-
-#ifdef EFX_NEED_IRQ_HANDLER_T
-       typedef irqreturn_t (*irq_handler_t)(int, void *, struct pt_regs *);
-#endif
-
-#ifdef EFX_NEED_I2C_CLASS_HWMON
-       #define I2C_CLASS_HWMON (1<<0)
-#endif
-
-#ifdef EFX_NEED_MII_CONSTANTS
-       #define MII_CTRL1000            0x09
-       #define MII_STAT1000            0x0a
-       #define BMCR_SPEED1000          0x0040
-       #define ADVERTISE_PAUSE_ASYM    0x0800
-       #define ADVERTISE_PAUSE_CAP     0x0400
-       #define LPA_PAUSE_ASYM          0x0800
-       #define LPA_PAUSE_CAP           0x0400
-       #define ADVERTISE_1000FULL      0x0200
-       #define ADVERTISE_1000HALF      0x0100
-       #define LPA_1000FULL            0x0800
-       #define LPA_1000HALF            0x0400
-#endif
-
-#ifdef EFX_NEED_DUMMY_SUPPORTS_GMII
-       #include <linux/mii.h>
-       /* Ugly; redirect nonexistent new field to an old unused field */
-       #undef supports_gmii
-       #define supports_gmii full_duplex
-#endif
-
-#ifdef EFX_NEED_SKB_HEADER_MACROS
-       #define skb_mac_header(skb)     (skb)->mac.raw
-       #define skb_network_header(skb) (skb)->nh.raw
-       #define tcp_hdr(skb)            (skb)->h.th
-       #define ip_hdr(skb)             (skb)->nh.iph
-       #define skb_tail_pointer(skb)   (skb)->tail
-#endif
 
 #ifdef EFX_NEED_RAW_READ_AND_WRITE_FIX
        #include <asm/io.h>
@@ -499,427 +125,60 @@
        #define __raw_readq efx_raw_readq
 #endif
 
-#ifdef EFX_NEED_SCHEDULE_TIMEOUT_INTERRUPTIBLE
-       static inline signed long
-       schedule_timeout_interruptible(signed long timeout)
-       {
-               set_current_state(TASK_INTERRUPTIBLE);
-               return schedule_timeout(timeout);
-       }
-#endif
+typedef u32 __wsum;
+#define csum_unfold(x) ((__force __wsum) x)
 
-#ifdef EFX_NEED_SCHEDULE_TIMEOUT_UNINTERRUPTIBLE
-       static inline signed long
-       schedule_timeout_uninterruptible(signed long timeout)
-       {
-               set_current_state(TASK_UNINTERRUPTIBLE);
-               return schedule_timeout(timeout);
-       }
-#endif
+#define DECLARE_MAC_BUF(var) char var[18] __attribute__((unused))
+extern char *print_mac(char *buf, const u8 *addr);
 
-#ifdef EFX_NEED_MMIOWB
-       #if defined(__i386__) || defined(__x86_64__)
-               #define mmiowb()
-       #elif defined(__ia64__)
-               #ifndef ia64_mfa
-                       #define ia64_mfa() asm volatile ("mf.a" ::: "memory")
-               #endif
-               #define mmiowb ia64_mfa
-       #else
-               #error "Need definition for mmiowb()"
-       #endif
-#endif
+/**
+ * queue_delayed_work in pre 2.6.20 can't rearm from inside
+ * the work member. So instead do a rather hacky sleep
+ */
+#define delayed_work work_struct
+#define INIT_DELAYED_WORK INIT_WORK
 
-#ifdef EFX_NEED_KZALLOC
-       static inline void *kzalloc(size_t size, int flags)
-       {
-               void *buf = kmalloc(size, flags);
-               if (buf)
-                       memset(buf, 0,size);
-               return buf;
-       }
-#endif
+static int inline efx_queue_delayed_work(struct workqueue_struct *wq,
+                                        struct work_struct *work,
+                                        unsigned long delay)
+{
+       if (unlikely(delay > 0))
+               schedule_timeout_uninterruptible(delay);
+       return queue_work(wq, work);
+}
+#define queue_delayed_work efx_queue_delayed_work
 
-#ifdef EFX_NEED_SETUP_TIMER
-       static inline void setup_timer(struct timer_list * timer,
-                                      void (*function)(unsigned long),
-                                      unsigned long data)
-       {
-               timer->function = function;
-               timer->data = data;
-               init_timer(timer);
-       }
-#endif
+/**
+ * The old and new work-function prototypes just differ
+ * in the type of the pointer returned, so it's safe
+ * to cast between the prototypes.
+ */
+typedef void (*efx_old_work_func_t)(void *p);
 
-#ifdef EFX_NEED_MUTEX
-       #define EFX_DEFINE_MUTEX(x) DECLARE_MUTEX(x)
-       #undef DEFINE_MUTEX
-       #define DEFINE_MUTEX EFX_DEFINE_MUTEX
+#undef INIT_WORK
+#define INIT_WORK(_work, _func)                                        \
+       do {                                                    \
+               INIT_LIST_HEAD(&(_work)->entry);                \
+               (_work)->pending = 0;                           \
+               PREPARE_WORK((_work),                           \
+                            (efx_old_work_func_t) (_func),     \
+                            (_work));                          \
+       } while (0)
 
-       #define efx_mutex semaphore
-       #undef mutex
-       #define mutex efx_mutex
+#define napi_str napi_dev[0]
 
-       #define efx_mutex_init(x) init_MUTEX(x)
-       #undef mutex_init
-       #define mutex_init efx_mutex_init
+static inline void netif_napi_add(struct net_device *dev,
+                                 struct net_device *dummy,
+                                 int (*poll) (struct net_device *, int *),
+                                 int weight)
+{
+       dev->weight = weight;
+       dev->poll = poll;
+}
 
-       #define efx_mutex_destroy(x) do { } while(0)
-       #undef mutex_destroy
-       #define mutex_destroy efx_mutex_destroy
+#define napi_enable netif_poll_enable
+#define napi_disable netif_poll_disable
 
-       #define efx_mutex_lock(x) down(x)
-       #undef mutex_lock
-       #define mutex_lock efx_mutex_lock
-
-       #define efx_mutex_lock_interruptible(x) down_interruptible(x)
-       #undef mutex_lock_interruptible
-       #define mutex_lock_interruptible efx_mutex_lock_interruptible
-
-       #define efx_mutex_unlock(x) up(x)
-       #undef mutex_unlock
-       #define mutex_unlock efx_mutex_unlock
-
-       #define efx_mutex_trylock(x) (!down_trylock(x))
-       #undef mutex_trylock
-       #define mutex_trylock efx_mutex_trylock
-
-       static inline int efx_mutex_is_locked(struct efx_mutex *m)
-       {
-               /* NB. This is quite inefficient, but it's the best we
-                * can do with the semaphore API. */
-               if ( down_trylock(m) )
-                       return 1;
-               /* Undo the effect of down_trylock. */
-               up(m);
-               return 0;
-       }
-       #undef mutex_is_locked
-       #define mutex_is_locked efx_mutex_is_locked
-#endif
-
-#ifndef NETIF_F_GSO
-       #define efx_gso_size tso_size
-       #undef gso_size
-       #define gso_size efx_gso_size
-       #define efx_gso_segs tso_segs
-       #undef gso_segs
-       #define gso_segs efx_gso_segs
-#endif
-
-#ifdef EFX_NEED_IRQF_FLAGS
-       #ifdef SA_PROBEIRQ
-               #define IRQF_PROBE_SHARED  SA_PROBEIRQ
-       #else
-               #define IRQF_PROBE_SHARED  0
-       #endif
-       #define IRQF_SHARED        SA_SHIRQ
-#endif
-
-#ifdef EFX_NEED_NETDEV_ALLOC_SKB
-       #ifndef NET_SKB_PAD
-               #define NET_SKB_PAD 16
-       #endif
-
-       static inline
-       struct sk_buff *netdev_alloc_skb(struct net_device *dev,
-                                        unsigned int length)
-       {
-               struct sk_buff *skb = alloc_skb(length + NET_SKB_PAD,
-                                               GFP_ATOMIC | __GFP_COLD);
-               if (likely(skb)) {
-                       skb_reserve(skb, NET_SKB_PAD);
-                       skb->dev = dev;
-               }
-               return skb;
-       }
-#endif
-
-#ifdef EFX_NEED_NETIF_TX_LOCK
-       static inline void netif_tx_lock(struct net_device *dev)
-       {
-               spin_lock(&dev->xmit_lock);
-               dev->xmit_lock_owner = smp_processor_id();
-       }
-       static inline void netif_tx_lock_bh(struct net_device *dev)
-       {
-               spin_lock_bh(&dev->xmit_lock);
-               dev->xmit_lock_owner = smp_processor_id();
-       }
-       static inline void netif_tx_unlock_bh(struct net_device *dev)
-       {
-               dev->xmit_lock_owner = -1;
-               spin_unlock_bh(&dev->xmit_lock);
-       }
-       static inline void netif_tx_unlock(struct net_device *dev)
-       {
-               dev->xmit_lock_owner = -1;
-               spin_unlock(&dev->xmit_lock);
-       }
-#endif
-
-#ifdef EFX_NEED_CSUM_UNFOLDED
-       typedef u32 __wsum;
-       #define csum_unfold(x) ((__force __wsum) x)
-#endif
-
-#ifdef EFX_NEED_HEX_DUMP
-       enum {
-               DUMP_PREFIX_NONE,
-               DUMP_PREFIX_ADDRESS,
-               DUMP_PREFIX_OFFSET
-       };
-#endif
-
-#ifdef EFX_NEED_PRINT_MAC
-       #define DECLARE_MAC_BUF(var) char var[18] __attribute__((unused))
-#endif
-
-#ifdef EFX_NEED_GFP_T
-       typedef unsigned int gfp_t;
-#endif
-
-#ifdef EFX_NEED_SAFE_LISTS
-       #define list_for_each_entry_safe_reverse(pos, n, head, member)       \
-               for (pos = list_entry((head)->prev, typeof(*pos), member),   \
-                    n = list_entry(pos->member.prev, typeof(*pos), member); \
-                    &pos->member != (head);                                 \
-                    pos = n,                                                \
-                    n = list_entry(n->member.prev, typeof(*n), member))
-#endif
-
-#ifdef EFX_NEED_DEV_NOTICE
-       #define dev_notice dev_warn
-#endif
-
-#ifdef EFX_NEED_IF_MII
-       #include <linux/mii.h>
-       static inline struct mii_ioctl_data *efx_if_mii ( struct ifreq *rq ) {
-               return ( struct mii_ioctl_data * ) &rq->ifr_ifru;
-       }
-       #undef if_mii
-       #define if_mii efx_if_mii
-#endif
-
-#ifdef EFX_NEED_MTD_ERASE_CALLBACK
-       #include <linux/mtd/mtd.h>
-       static inline void efx_mtd_erase_callback(struct erase_info *instr) {
-               if ( instr->callback )
-                       instr->callback ( instr );
-       }
-       #undef mtd_erase_callback
-       #define mtd_erase_callback efx_mtd_erase_callback
-#endif
-
-#ifdef EFX_NEED_DUMMY_PCI_DISABLE_MSI
-       #include <linux/pci.h>
-       static inline void dummy_pci_disable_msi ( struct pci_dev *dev ) {
-               /* Do nothing */
-       }
-       #undef pci_disable_msi
-       #define pci_disable_msi dummy_pci_disable_msi
-#endif
-
-#ifdef EFX_NEED_DUMMY_MSIX
-       struct msix_entry {
-               u16     vector; /* kernel uses to write allocated vector */
-               u16     entry;  /* driver uses to specify entry, OS writes */
-       };
-       static inline int pci_enable_msix(struct pci_dev* dev,
-                                         struct msix_entry *entries, int nvec)
-               {return -1;}
-       static inline void pci_disable_msix(struct pci_dev *dev) { /* Do 
nothing */}
-#endif
-
-#ifdef EFX_NEED_BYTEORDER_TYPES
-       typedef __u16 __be16;
-       typedef __u32 __be32;
-       typedef __u64 __be64;
-       typedef __u16 __le16;
-       typedef __u32 __le32;
-       typedef __u64 __le64;
-#endif
-
-/**************************************************************************
- *
- * Missing functions provided by kernel_compat.c
- *
- **************************************************************************
- *
- */
-#ifdef EFX_NEED_RANDOM_ETHER_ADDR
-       extern void efx_random_ether_addr(uint8_t *addr);
-       #ifndef EFX_IN_KCOMPAT_C
-               #undef random_ether_addr
-               #define random_ether_addr efx_random_ether_addr
-       #endif
-#endif
-
-#ifdef EFX_NEED_MII_ETHTOOL_FIX
-       extern int efx_mii_ethtool_gset(struct mii_if_info *mii,
-                                       struct ethtool_cmd *ecmd);
-       extern int efx_mii_ethtool_sset(struct mii_if_info *mii,
-                                       struct ethtool_cmd *ecmd);
-       #ifndef EFX_IN_KCOMPAT_C
-               #undef mii_ethtool_gset
-               #define mii_ethtool_gset efx_mii_ethtool_gset
-               #undef mii_ethtool_sset
-               #define mii_ethtool_sset efx_mii_ethtool_sset
-       #endif
-#endif
-
-#ifdef EFX_NEED_UNREGISTER_NETDEVICE_NOTIFIER_FIX
-       extern int efx_unregister_netdevice_notifier(struct notifier_block *nb);
-       #ifndef EFX_IN_KCOMPAT_C
-               #undef unregister_netdevice_notifier
-               #define unregister_netdevice_notifier \
-                               efx_unregister_netdevice_notifier
-       #endif
-#endif
-
-#ifdef EFX_NEED_IOMMU_LOCK
-       extern dma_addr_t efx_pci_map_single(struct pci_dev *pci, void *ptr,
-                                            size_t size, int direction);
-       extern void efx_pci_unmap_single(struct pci_dev *pci,
-                                        dma_addr_t dma_addr, size_t size,
-                                        int direction);
-       extern void * efx_pci_alloc_consistent(struct pci_dev *pci,
-                                              size_t size,
-                                              dma_addr_t *dma_addr);
-       extern void efx_pci_free_consistent(struct pci_dev *pci,
-                                           size_t size, void *ptr,
-                                           dma_addr_t dma_addr);
-       #ifndef EFX_IN_KCOMPAT_C
-               #undef pci_map_single
-               #undef pci_unmap_single
-               #undef pci_alloc_consistent
-               #undef pci_free_consistent
-               #define pci_map_single efx_pci_map_single
-               #define pci_unmap_single efx_pci_unmap_single
-               #define pci_alloc_consistent efx_pci_alloc_consistent
-               #define pci_free_consistent efx_pci_free_consistent
-       #endif
-#endif
-
-#ifdef EFX_NEED_PRINT_MAC
-       extern char *print_mac(char *buf, const u8 *addr);
-#endif
-
-#ifdef EFX_NEED_COMPOUND_PAGE_FIX
-       extern void efx_compound_page_destructor(struct page *page);
-#endif
-
-#ifdef EFX_NEED_HEX_DUMP
-       extern void
-       print_hex_dump(const char *level, const char *prefix_str,
-                      int prefix_type, int rowsize, int groupsize,
-                      const void *buf, size_t len, int ascii);
-#endif
-
-#ifdef EFX_NEED_MSECS_TO_JIFFIES
-       extern unsigned long msecs_to_jiffies(const unsigned int m);
-#endif
-
-#ifdef EFX_NEED_MSLEEP
-       extern void msleep(unsigned int msecs);
-#endif
-
-/**************************************************************************
- *
- * Wrappers to fix bugs and parameter changes
- *
- **************************************************************************
- *
- */
-
-#ifdef EFX_NEED_PCI_SAVE_RESTORE_WRAPPERS
-       #define pci_save_state(_dev)                                    \
-               pci_save_state(_dev, (_dev)->saved_config_space)
-
-       #define pci_restore_state(_dev)                                 \
-               pci_restore_state(_dev, (_dev)->saved_config_space)
-#endif
-
-#ifdef EFX_NEED_WORK_API_WRAPPERS
-       /**
-        * queue_delayed_work in pre 2.6.20 can't rearm from inside
-        * the work member. So instead do a rather hacky sleep
-        */
-       #define delayed_work work_struct
-       #define INIT_DELAYED_WORK INIT_WORK
-
-       static int inline efx_queue_delayed_work(struct workqueue_struct *wq,
-                                                struct work_struct *work,
-                                                unsigned long delay)
-       {
-               if (unlikely(delay > 0))
-                       schedule_timeout_uninterruptible(delay);
-               return queue_work(wq, work);
-       }
-       #define queue_delayed_work efx_queue_delayed_work
-
-       /**
-        * The old and new work-function prototypes just differ
-        * in the type of the pointer returned, so it's safe
-        * to cast between the prototypes.
-        */
-       typedef void (*efx_old_work_func_t)(void *p);
-
-       #undef INIT_WORK
-       #define INIT_WORK(_work, _func)                                 \
-               do {                                                    \
-                       INIT_LIST_HEAD(&(_work)->entry);                \
-                       (_work)->pending = 0;                           \
-                       PREPARE_WORK((_work),                           \
-                                    (efx_old_work_func_t) (_func),     \
-                                    (_work));                          \
-               } while (0)
-#endif
-
-#ifdef EFX_HAVE_OLD_NAPI
-       #define napi_str napi_dev[0]
-
-       static inline void netif_napi_add(struct net_device *dev,
-                                         struct net_device *dummy,
-                                         int (*poll) (struct net_device *,
-                                                      int *),
-                                         int weight)
-       {
-               dev->weight = weight;
-               dev->poll = poll;
-       }
-
-       #define napi_enable netif_poll_enable
-       #define napi_disable netif_poll_disable
-
-       #define netif_rx_complete(dev, dummy) netif_rx_complete(dev)
-#endif
-
-#ifdef EFX_NEED_COMPOUND_PAGE_FIX
-       static inline
-       struct page *efx_alloc_pages(gfp_t flags, unsigned int order)
-       {
-               struct page *p = alloc_pages(flags, order);
-               if ((flags & __GFP_COMP) && (p != NULL) && (order > 0))
-                       p[1].mapping = (void *)efx_compound_page_destructor;
-               return p;
-       }
-       #undef alloc_pages
-       #define alloc_pages efx_alloc_pages
-
-       static inline
-       void efx_free_pages(struct page *p, unsigned int order)
-       {
-               if ((order > 0) && (page_count(p) == 1))
-                       p[1].mapping = NULL;
-               __free_pages(p, order);
-       }
-       #define __free_pages efx_free_pages
-#endif
-
-#ifdef EFX_NEED_HEX_DUMP_CONST_FIX
-       #define print_hex_dump(v,s,t,r,g,b,l,a) \
-               print_hex_dump((v),(s),(t),(r),(g),(void*)(b),(l),(a))
-#endif
+#define netif_rx_complete(dev, dummy) netif_rx_complete(dev)
 
 #endif /* EFX_KERNEL_COMPAT_H */
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/mdio_10g.c
--- a/drivers/net/sfc/mdio_10g.c        Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/mdio_10g.c        Tue Mar 31 11:59:10 2009 +0100
@@ -106,8 +106,8 @@ static int mdio_clause45_check_mmd(struc
 }
 
 /* This ought to be ridiculous overkill. We expect it to fail rarely */
-#define MDIO45_RESET_TIME      HZ
-#define MDIO45_RESET_ITERS     (100)
+#define MDIO45_RESET_TIME      1000 /* ms */
+#define MDIO45_RESET_ITERS     100
 
 int mdio_clause45_wait_reset_mmds(struct efx_nic *efx,
                                  unsigned int mmd_mask)
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/mtd.c
--- a/drivers/net/sfc/mtd.c     Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/mtd.c     Tue Mar 31 11:59:10 2009 +0100
@@ -300,9 +300,7 @@ out:
                erase->state = MTD_ERASE_DONE;
        } else {
                erase->state = MTD_ERASE_FAILED;
-#if defined(EFX_USE_MTD_ERASE_FAIL_ADDR)
                erase->fail_addr = 0xffffffff;
-#endif
        }
        mtd_erase_callback(erase);
        return rc;
@@ -437,9 +435,7 @@ static __devinit int efx_mtd_register(st
 
        efx_mtd->mtd.size = spi->size;
        efx_mtd->mtd.erasesize = spi->erase_size;
-#if defined(EFX_USE_MTD_WRITESIZE)
        efx_mtd->mtd.writesize = 1;
-#endif
        if (snprintf(efx_mtd->name, sizeof(efx_mtd->name),
                     "%s %s", efx->name, type_name) >=
            sizeof(efx_mtd->name))
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/net_driver.h
--- a/drivers/net/sfc/net_driver.h      Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/net_driver.h      Tue Mar 31 11:59:10 2009 +0100
@@ -50,13 +50,8 @@
 #include "driverlink.h"
 #include "i2c-direct.h"
 
-       #ifndef EFX_USE_DEBUGFS
-               /* Sick, but we have no other use for dentry */
-               #define dentry proc_dir_entry
-       #endif
-
-#define EFX_MAX_LRO_DESCRIPTORS 8
-#define EFX_MAX_LRO_AGGR MAX_SKB_FRAGS
+/* Sick, but we have no other use for dentry */
+#define dentry proc_dir_entry
 
 /**************************************************************************
  *
@@ -66,7 +61,7 @@
 #ifndef EFX_DRIVER_NAME
 #define EFX_DRIVER_NAME        "sfc"
 #endif
-#define EFX_DRIVER_VERSION     "2.2.0101"
+#define EFX_DRIVER_VERSION     "2.2.0204"
 
 #ifdef EFX_ENABLE_DEBUG
 #define EFX_BUG_ON_PARANOID(x) BUG_ON(x)
@@ -76,11 +71,15 @@
 #define EFX_WARN_ON_PARANOID(x) do {} while (0)
 #endif
 
+#define NET_DEV_REGISTERED(efx)                                        \
+       ((efx)->net_dev &&                                      \
+        ((efx)->net_dev->reg_state == NETREG_REGISTERED))
+
 /* Include net device name in log messages if it has been registered.
  * Use efx->name not efx->net_dev->name so that races with (un)registration
  * are harmless.
  */
-#define NET_DEV_NAME(efx) ((efx)->net_dev_registered ? (efx)->name : "")
+#define NET_DEV_NAME(efx) (NET_DEV_REGISTERED(efx) ? (efx)->name : "")
 
 /* Un-rate-limited logging */
 #define EFX_ERR(efx, fmt, args...) \
@@ -336,6 +335,83 @@ struct efx_buffer {
 };
 
 
+
+/**
+ * struct efx_ssr_conn - Connection state for Soft Segment Reassembly (SSR) 
aka LRO
+ * @link: Link for hash table and free list.
+ * @active_link: Link for active_conns list
+ * @saddr: Source IP address
+ * @daddr: Destination IP address
+ * @source: Source TCP port number
+ * @dest: Destination TCP port number
+ * @n_in_order_pkts: Number of in-order packets we've seen with payload.
+ * @next_seq: Next in-order sequence number.
+ * @last_pkt_jiffies: Time we last saw a packet on this connection.
+ * @skb: The SKB we are currently holding.
+ *     If %NULL, then all following fields are undefined.
+ * @skb_tail: The tail of the frag_list of SKBs we're holding.
+ *     Only valid after at least one merge.
+ * @eh: The ethernet header of the skb we are holding.
+ * @iph: The IP header of the skb we are holding.
+ * @th: The TCP header of the skb we are holding.
+ * @th_last: The TCP header of the last packet merged.
+ */
+struct efx_ssr_conn {
+       struct list_head link;
+       struct list_head active_link;
+       unsigned saddr, daddr;
+       unsigned short source, dest;
+       unsigned n_in_order_pkts;
+       unsigned next_seq;
+       unsigned long last_pkt_jiffies;
+       struct sk_buff *skb;
+       struct sk_buff *skb_tail;
+       struct ethhdr *eh;
+       struct iphdr *iph;
+       struct tcphdr *th;
+       struct tcphdr *th_last;
+};
+
+/**
+ * struct efx_ssr_state - Port state for Soft Segment Reassembly (SSR) aka LRO
+ * @efx: The associated NIC.
+ * @conns_mask: Number of hash buckets - 1.
+ * @conns: Hash buckets for tracked connections.
+ * @conns_n: Length of linked list for each hash bucket.
+ * @active_conns: Connections that are holding a packet.
+ *     Connections are self-linked when not in this list.
+ * @free_conns: Free efx_ssr_conn instances.
+ * @last_purge_jiffies: The value of jiffies last time we purged idle
+ *     connections.
+ * @n_merges: Number of packets absorbed by SSR.
+ * @n_bursts: Number of bursts spotted by SSR.
+ * @n_slow_start: Number of packets not merged because connection may be in
+ *     slow-start.
+ * @n_misorder: Number of out-of-order packets seen in tracked streams.
+ * @n_too_many: Incremented when we're trying to track too many streams.
+ * @n_new_stream: Number of distinct streams we've tracked.
+ * @n_drop_idle: Number of streams discarded because they went idle.
+ * @n_drop_closed: Number of streams that have seen a FIN or RST.
+ */
+struct efx_ssr_state {
+       struct efx_nic *efx;
+       unsigned conns_mask;
+       struct list_head *conns;
+       unsigned *conns_n;
+       struct list_head active_conns;
+       struct list_head free_conns;
+       unsigned long last_purge_jiffies;
+       unsigned n_merges;
+       unsigned n_bursts;
+       unsigned n_slow_start;
+       unsigned n_misorder;
+       unsigned n_too_many;
+       unsigned n_new_stream;
+       unsigned n_drop_idle;
+       unsigned n_drop_closed;
+};
+
+
 /* Flags for channel->used_flags */
 #define EFX_USED_BY_RX 1
 #define EFX_USED_BY_TX 2
@@ -371,6 +447,7 @@ enum efx_rx_alloc_method {
  * @last_eventq_read_ptr: Last event queue read pointer value.
  * @eventq_magic: Event queue magic value for driver-generated test events
  * @debug_dir: debugfs directory
+ * @ssr: LRO/SSR state
  * @rx_alloc_level: Watermark based heuristic counter for pushing descriptors
  *     and diagnostic counters
  * @rx_alloc_push_pages: RX allocation method currently in use for pushing
@@ -395,9 +472,6 @@ struct efx_channel {
        unsigned int has_interrupt;
        unsigned int irq_moderation;
        struct net_device *napi_dev;
-#if !defined(EFX_HAVE_OLD_NAPI)
-       struct napi_struct napi_str;
-#endif
        struct work_struct reset_work;
        int work_pending;
        struct efx_special_buffer eventq;
@@ -409,6 +483,7 @@ struct efx_channel {
        struct dentry *debug_dir;
 #endif
 
+       struct efx_ssr_state ssr;
        int rx_alloc_level;
        int rx_alloc_push_pages;
        int rx_alloc_pop_pages;
@@ -533,10 +608,10 @@ enum phy_type {
 #define EFX_ISCLAUSE45(efx) ((efx)->phy_type != PHY_TYPE_1G_ALASKA)
 
 enum nic_state {
-       STATE_INIT = 0,      /* suspend_lock always held */
+       STATE_INIT = 0,
        STATE_RUNNING = 1,
        STATE_FINI = 2,
-       STATE_RESETTING = 3, /* suspend_lock always held */
+       STATE_RESETTING = 3, /* rtnl_lock always held */
        STATE_DISABLED = 4,
        STATE_MAX,
 };
@@ -548,10 +623,10 @@ enum nic_state {
  * This is the equivalent of NET_IP_ALIGN [which controls the alignment
  * of the skb->head for hardware DMA].
  */
-#ifdef __ia64__
-#define EFX_PAGE_IP_ALIGN 2
+#if defined(__i386__) || defined(__x86_64__)
+#define EFX_PAGE_IP_ALIGN 0
 #else
-#define EFX_PAGE_IP_ALIGN 0
+#define EFX_PAGE_IP_ALIGN NET_IP_ALIGN
 #endif
 
 /*
@@ -578,14 +653,14 @@ enum efx_fc_type {
  * @mac_writel: Write dword to MAC register
  * @mac_readl: Read dword from a MAC register
  * @init: Initialise MAC and PHY
- * @reconfigure: Reconfigure MAC and PHY (e.g. for new link parameters)
+ * @reconfigure: Reconfigure MAC and PHY. Serialised by the mac_lock
  * @update_stats: Update statistics
  * @fini: Shut down MAC and PHY
- * @check_hw: Check hardware
+ * @check_hw: Check hardware. Serialised by the mac_lock
  * @fake_phy_event: Simulate a PHY event on a port
- * @get_settings: Get ethtool settings
- * @set_settings: Set ethtool settings
- * @set_pause: Set pause parameters
+ * @get_settings: Get ethtool settings. Serialised by the mac_lock
+ * @set_settings: Set ethtool settings. Serialised by the mac_lock
+ * @set_pause: Set pause parameters. Serialised by the mac_lock
  */
 struct efx_mac_operations {
        void (*mac_writel) (struct efx_nic *efx,
@@ -711,8 +786,8 @@ struct efx_mac_stats {
 
 /* An Efx multicast filter hash */
 union efx_multicast_hash {
-       u8 byte[EFX_MCAST_HASH_ENTRIES / sizeof(u8)];
-       efx_oword_t oword[EFX_MCAST_HASH_ENTRIES / sizeof(efx_oword_t)];
+       u8 byte[EFX_MCAST_HASH_ENTRIES / 8];
+       efx_oword_t oword[EFX_MCAST_HASH_ENTRIES / sizeof(efx_oword_t) / 8];
 };
 
 /* Efx Error condition statistics */
@@ -732,12 +807,9 @@ struct efx_nic_errors {
  * struct efx_nic - an Efx NIC
  * @name: Device name (net device name or bus id before net device registered)
  * @pci_dev: The PCI device
- * @pci_dev2: The secondary PCI device if present
  * @type: Controller type attributes
- * @dma_mask: DMA mask
  * @legacy_irq: IRQ number
  * @workqueue: Workqueue for resets, port reconfigures and the HW monitor
- * @refill_workqueue: RX refill workqueue
  * @reset_work: Scheduled reset workitem
  * @monitor_work: Hardware monitor workitem
  * @membase_phys: Memory BAR value as physical address
@@ -746,15 +818,10 @@ struct efx_nic_errors {
  * @interrupt_mode: Interrupt mode
  * @is_asic: Is ASIC (else FPGA)
  * @is_10g: Is set to 10G (else 1G)
- * @external_sram_cfg: Size and number of banks of external SRAM
  * @i2c: I2C interface
  * @board_info: Board-level information
- * @state: Device state flag. Can only be manipulated when both
- *     suspend_lock and rtnl_lock are held.  Can be read when
- *     either is held.
+ * @state: Device state flag. Serialised by the rtnl_lock.
  * @reset_pending: Pending reset method (normally RESET_TYPE_NONE)
- * @suspend_lock: Device suspend lock.  This must not be acquired with
- *     rtnl_lock held.
  * @tx_queue: TX DMA queues
  * @rx_queue: RX DMA queues
  * @channel: Channels
@@ -774,23 +841,21 @@ struct efx_nic_errors {
  * @spi_lock: SPI bus lock
  * @n_rx_nodesc_drop_cnt: RX no descriptor drop count
  * @nic_data: Hardware dependant state
- * @mac_lock: MAC access lock. Protects efx->port_enabled/net_dev_registered
- *            and efx_reconfigure_port()
+ * @mac_lock: MAC access lock. Protects @port_enabled, efx_monitor() and
+ *     efx_reconfigure_port()
  * @port_enabled: Port enabled indicator.
- *     Serialises efx_stop_all and efx_start_all with kernel interfaces.
- *     Safe to read under the rtnl_lock, mac_lock, or netif_tx_lock, but
- *     all three must be held to modify it.
- * @net_dev_registered: Port is registered with operating system.
+ *     Serialises efx_stop_all(), efx_start_all() and efx_monitor() and
+ *     efx_reconfigure_work with kernel interfaces. Safe to read under any
+ *     one of the rtnl_lock, mac_lock, or netif_tx_lock, but all three must
+ *     be held to modify it.
  * @port_initialized: Port initialized?
  * @net_dev: Operating system network device. Consider holding the rtnl lock
  * @rx_checksum_enabled: RX checksumming enabled
  * @netif_stop_count: Port stop count
  * @netif_stop_lock: Port stop lock
- * @mac_stats: MAC statistics
- * @stats: Net device statistics.
- *     Hardware-specific code fills in @mac_stats, which provides a
- *     detailed breakdown.  Generic code aggregates these statistics
- *     into a standard &struct net_device_stats.
+ * @mac_stats: MAC statistics. These include all statistics the MACs
+ *     can provide.  Generic code converts these into a standard
+ *     &struct net_device_stats.
  * @stats_buffer: DMA buffer for statistics
  * @stats_lock: Statistics update lock
  * @mac_op: MAC interface
@@ -829,21 +894,14 @@ struct efx_nic {
 struct efx_nic {
        char name[IFNAMSIZ];
        struct pci_dev *pci_dev;
-       struct pci_dev *pci_dev2;
-#if !defined(EFX_USE_PCI_DEV_REVISION)
        u8 revision;
-#endif
        const struct efx_nic_type *type;
-       dma_addr_t dma_mask;
        int legacy_irq;
        struct workqueue_struct *workqueue;
-#if !defined(EFX_USE_CANCEL_DELAYED_WORK_SYNC)
        /* Since we can't use cancel_delayed_work_sync efx_reset() has to
         * flush efx->workqueue to serialise against efx_reconfigure_port
         * and efx_monitor. So it can't also run on workqueue */
        struct workqueue_struct *reset_workqueue;
-#endif
-       struct workqueue_struct *refill_workqueue;
        struct work_struct reset_work;
        struct delayed_work monitor_work;
        unsigned long membase_phys;
@@ -852,15 +910,12 @@ struct efx_nic {
        enum efx_int_mode interrupt_mode;
        unsigned int is_asic:1;
        unsigned int is_10g:1;
-       int external_sram_cfg;
 
        struct efx_i2c_interface i2c;
        struct efx_board board_info;
 
        enum nic_state state;
        enum reset_type reset_pending;
-
-       struct semaphore suspend_lock;
 
        struct efx_tx_queue tx_queue[EFX_MAX_TX_QUEUES];
        struct efx_rx_queue rx_queue[EFX_MAX_RX_QUEUES];
@@ -885,11 +940,11 @@ struct efx_nic {
 
        struct mutex mac_lock;
        int port_enabled;
-       int net_dev_registered;
 
        int port_initialized;
        struct net_device *net_dev;
        int rx_checksum_enabled;
+       int lro_enabled;
 
        atomic_t netif_stop_count;
        spinlock_t netif_stop_lock;
@@ -942,7 +997,6 @@ struct efx_nic {
 
 /**
  * struct efx_nic_type - Efx device type definition
- * @is_dual_func: Is dual-function (else single-function)
  * @mem_bar: Memory BAR number
  * @mem_map_size: Memory BAR mapped size
  * @txd_ptr_tbl_base: TX descriptor ring base address
@@ -965,7 +1019,6 @@ struct efx_nic {
  *     descriptors
  */
 struct efx_nic_type {
-       unsigned int is_dual_func;
        unsigned int mem_bar;
        unsigned int mem_map_size;
        unsigned int txd_ptr_tbl_base;
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/null_phy.c
--- a/drivers/net/sfc/null_phy.c        Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/null_phy.c        Tue Mar 31 11:59:10 2009 +0100
@@ -33,10 +33,8 @@ static int falcon_null_phy_check_hw(stru
        int link_ok = falcon_xaui_link_ok(efx);
 
        /* Generate PHY event that a PHY would have generated */
-       if (link_ok != efx->link_up) {
-               efx->link_up = link_ok;
+       if (link_ok != efx->link_up)
                efx->mac_op->fake_phy_event(efx);
-       }
 
        return 0;
 }
@@ -46,7 +44,7 @@ static void falcon_null_phy_reconfigure(
        /* CX4 is always 10000FD only */
        efx->link_options = GM_LPA_10000FULL;
 
-       falcon_null_phy_check_hw(efx);
+       efx->link_up = falcon_xaui_link_ok(efx);
 }
 
 struct efx_phy_operations falcon_null_phy_ops = {
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/pm8358_phy.c
--- a/drivers/net/sfc/pm8358_phy.c      Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/pm8358_phy.c      Tue Mar 31 11:59:10 2009 +0100
@@ -162,10 +162,8 @@ static int pm8358_phy_check_hw(struct ef
        int rc = 0;
        int link_up = pm8358_link_ok(efx);
        /* Simulate a PHY event if link state has changed */
-       if (link_up != efx->link_up) {
-               efx->link_up = link_up;
+       if (link_up != efx->link_up)
                efx->mac_op->fake_phy_event(efx);
-       }
 
        return rc;
 }
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/rx.c
--- a/drivers/net/sfc/rx.c      Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/rx.c      Tue Mar 31 11:59:10 2009 +0100
@@ -84,7 +84,7 @@ static int rx_alloc_method = RX_ALLOC_ME
 #define RX_ALLOC_LEVEL_LRO 0x2000
 #define RX_ALLOC_LEVEL_MAX 0x3000
 #define RX_ALLOC_FACTOR_LRO 1
-#define RX_ALLOC_FACTOR_SKB -2
+#define RX_ALLOC_FACTOR_SKB (-2)
 
 /* This is the percentage fill level below which new RX descriptors
  * will be added to the RX descriptor ring.
@@ -284,8 +284,8 @@ static inline void efx_free_rx_buffer(st
        }
 }
 
-inline void efx_fini_rx_buffer(struct efx_rx_queue *rx_queue,
-                              struct efx_rx_buffer *rx_buf)
+static inline void efx_fini_rx_buffer(struct efx_rx_queue *rx_queue,
+                                     struct efx_rx_buffer *rx_buf)
 {
        /* Unmap for DMA */
        efx_unmap_rx_buffer(rx_queue->efx, rx_buf);
@@ -364,7 +364,7 @@ static int __efx_fast_push_rx_descriptor
                  rx_queue->added_count - rx_queue->removed_count);
 
  out:
-       /* Send write pointer to card.  */
+       /* Send write pointer to card. */
        falcon_notify_rx_desc(rx_queue);
 
        /* If the fast fill is running inside from the refill tasklet, then
@@ -399,8 +399,7 @@ void efx_fast_push_rx_descriptors(struct
                 * that work is immediately pending to free some memory
                 * (e.g. an RX event or TX completion)
                 */
-               queue_delayed_work(rx_queue->efx->refill_workqueue,
-                                  &rx_queue->work, 0);
+               efx_schedule_slow_fill(rx_queue, 0);
        }
 }
 
@@ -409,11 +408,7 @@ void efx_rx_work(struct work_struct *dat
        struct efx_rx_queue *rx_queue;
        int rc;
 
-#if !defined(EFX_NEED_WORK_API_WRAPPERS)
-       rx_queue = container_of(data, struct efx_rx_queue, work.work);
-#else
        rx_queue = container_of(data, struct efx_rx_queue, work);
-#endif
 
        if (unlikely(!rx_queue->channel->enabled))
                return;
@@ -425,10 +420,8 @@ void efx_rx_work(struct work_struct *dat
        /* Push new RX descriptors, allowing at least 1 jiffy for
         * the kernel to free some more memory. */
        rc = __efx_fast_push_rx_descriptors(rx_queue, 1);
-       if (rc) {
-               queue_delayed_work(rx_queue->efx->refill_workqueue,
-                                  &rx_queue->work, 1);
-       }
+       if (rc)
+               efx_schedule_slow_fill(rx_queue, 1);
 }
 
 static inline void efx_rx_packet__check_len(struct efx_rx_queue *rx_queue,
@@ -513,14 +506,9 @@ static inline struct sk_buff *efx_rx_mk_
        return skb;
 }
 
-#if defined(EFX_USE_FASTCALL)
 void fastcall efx_rx_packet(struct efx_rx_queue *rx_queue,
                            unsigned int index, unsigned int len,
                            int checksummed, int discard)
-#else
-void efx_rx_packet(struct efx_rx_queue *rx_queue, unsigned int index,
-                  unsigned int len, int checksummed, int discard)
-#endif
 {
        struct efx_nic *efx = rx_queue->efx;
        struct efx_rx_buffer *rx_buf;
@@ -587,6 +575,7 @@ void __efx_rx_packet(struct efx_channel 
        struct efx_nic *efx = channel->efx;
        enum efx_veto veto;
        struct sk_buff *skb;
+       int lro = efx->lro_enabled;
 
        /* If we're in loopback test, then pass the packet directly to the
         * loopback layer, and free the rx_buf here
@@ -616,7 +605,14 @@ void __efx_rx_packet(struct efx_channel 
         * changed, then flush the LRO state.
         */
        if (unlikely(channel->rx_alloc_pop_pages != (rx_buf->page != NULL))) {
+               efx_flush_lro(channel);
                channel->rx_alloc_pop_pages = (rx_buf->page != NULL);
+       }
+       if (likely(checksummed && lro)) {
+               if (efx_ssr(&channel->ssr, rx_buf)) {
+                       channel->rx_alloc_level += RX_ALLOC_FACTOR_LRO;
+                       goto done;
+               }
        }
 
        /* Allow callback to veto the packet */
@@ -668,7 +664,7 @@ void efx_rx_strategy(struct efx_channel 
        enum efx_rx_alloc_method method = rx_alloc_method;
 
        /* Only makes sense to use page based allocation if LRO is enabled */
-       if (!(channel->efx->net_dev->features & NETIF_F_LRO)) {
+       if (!(channel->efx->lro_enabled)) {
                method = RX_ALLOC_METHOD_SKB;
        } else if (method == RX_ALLOC_METHOD_AUTO) {
                /* Constrain the rx_alloc_level */
@@ -725,8 +721,6 @@ int efx_init_rx_queue(struct efx_rx_queu
 
        EFX_LOG(rx_queue->efx, "initialising RX queue %d\n", rx_queue->queue);
 
-       ASSERT_RTNL();
-
        /* Initialise ptr fields */
        rx_queue->added_count = 0;
        rx_queue->notified_count = 0;
@@ -753,8 +747,6 @@ void efx_fini_rx_queue(struct efx_rx_que
        struct efx_rx_buffer *rx_buf;
 
        EFX_LOG(rx_queue->efx, "shutting down RX queue %d\n", rx_queue->queue);
-
-       ASSERT_RTNL();
 
        /* Flush RX queue and remove descriptor ring */
        falcon_fini_rx(rx_queue);
@@ -788,6 +780,12 @@ void efx_remove_rx_queue(struct efx_rx_q
        rx_queue->used = 0;
 }
 
+/* Flush LRO/SSR state for the given channel */
+void efx_flush_lro(struct efx_channel *channel)
+{
+       efx_ssr_end_of_burst(&channel->ssr);
+}
+
 
 module_param(rx_alloc_method, int, 0644);
 MODULE_PARM_DESC(rx_alloc_method, "Allocation method used for RX buffers");
@@ -796,3 +794,441 @@ MODULE_PARM_DESC(rx_refill_threshold,
 MODULE_PARM_DESC(rx_refill_threshold,
                 "RX descriptor ring fast/slow fill threshold (%)");
 
+
+
+/* Size of the LRO hash table.  Must be a power of 2.  A larger table
+ * means we can accelerate a larger number of streams.
+ */
+static unsigned lro_table_size = 128;
+module_param(lro_table_size, uint, 0644);
+MODULE_PARM_DESC(lro_table_size,
+                "Size of the LRO hash table.  Must be a power of 2");
+
+/* Maximum length of a hash chain.  If chains get too long then the lookup
+ * time increases and may exceed the benefit of LRO.
+ */
+static unsigned lro_chain_max = 20;
+module_param(lro_chain_max, uint, 0644);
+MODULE_PARM_DESC(lro_chain_max,
+                "Maximum length of chains in the LRO hash table");
+
+
+/* Maximum time (in jiffies) that a connection can be idle before it's LRO
+ * state is discarded.
+ */
+static unsigned lro_idle_jiffies = HZ / 10 + 1;        /* 100ms */
+module_param(lro_idle_jiffies, uint, 0644);
+MODULE_PARM_DESC(lro_idle_jiffies, "Time (in jiffies) after which an"
+                " idle connection's LRO state is discarded");
+
+
+/* Number of packets with payload that must arrive in-order before a
+ * connection is eligible for LRO.  The idea is we should avoid coalescing
+ * segments when the sender is in slow-start because reducing the ACK rate
+ * can damage performance.
+ */
+static unsigned lro_slow_start_packets = 20;
+module_param(lro_slow_start_packets, uint, 0644);
+MODULE_PARM_DESC(lro_slow_start_packets, "Number of packets that must "
+                "pass in-order before starting LRO.");
+
+
+int efx_ssr_init(struct efx_ssr_state *st, struct efx_nic *efx)
+{
+       unsigned i;
+       st->conns_mask = lro_table_size - 1;
+       if ((st->conns_mask + 1) & st->conns_mask) {
+               EFX_ERR(efx, "lro_table_size(=%u) must be a power of 2\n",
+                       lro_table_size);
+               return -EINVAL;
+       }
+       st->efx = efx;
+       st->conns = kmalloc((st->conns_mask + 1)
+                           * sizeof(st->conns[0]), GFP_KERNEL);
+       if (st->conns == NULL)
+               return -ENOMEM;
+       st->conns_n = kmalloc((st->conns_mask + 1)
+                             * sizeof(st->conns_n[0]), GFP_KERNEL);
+       if (st->conns_n == NULL) {
+               kfree(st->conns);
+               st->conns = NULL;
+               return -ENOMEM;
+       }
+       for (i = 0; i <= st->conns_mask; ++i) {
+               INIT_LIST_HEAD(&st->conns[i]);
+               st->conns_n[i] = 0;
+       }
+       INIT_LIST_HEAD(&st->active_conns);
+       INIT_LIST_HEAD(&st->free_conns);
+       return 0;
+}
+
+/* Drop the given connection, and add it to the free list */
+static inline void efx_ssr_drop(struct efx_ssr_state *st,
+                               struct efx_ssr_conn *c, unsigned conn_hash)
+{
+       EFX_BUG_ON_PARANOID(c->skb);
+       EFX_BUG_ON_PARANOID(st->conns_n[conn_hash] <= 0);
+       --st->conns_n[conn_hash];
+       list_del(&c->link);
+       list_add(&c->link, &st->free_conns);
+}
+
+void efx_ssr_fini(struct efx_ssr_state *st)
+{
+       struct efx_ssr_conn *c;
+       unsigned i;
+
+       /* Return cleanly if efx_ssr_init() has not been called. */
+       if (st->conns == NULL)
+               return;
+
+       EFX_BUG_ON_PARANOID(!list_empty(&st->active_conns));
+
+       for (i = 0; i <= st->conns_mask; ++i)
+               while (!list_empty(&st->conns[i])) {
+                       c = list_entry(st->conns[i].prev,
+                                      struct efx_ssr_conn, link);
+                       efx_ssr_drop(st, c, i);
+               }
+
+       while (!list_empty(&st->free_conns)) {
+               c = list_entry(st->free_conns.prev, struct efx_ssr_conn, link);
+               list_del(&c->link);
+               EFX_BUG_ON_PARANOID(c->skb);
+               kfree(c);
+       }
+
+       kfree(st->conns_n);
+       kfree(st->conns);
+       st->conns = NULL;
+}
+
+/* Calc IP checksum and deliver to the OS */
+static void efx_ssr_deliver(struct efx_ssr_state *st, struct efx_ssr_conn *c)
+{
+       struct efx_nic *efx = st->efx;
+       int veto, len;
+
+       EFX_BUG_ON_PARANOID(!c->skb);
+
+       ++st->n_bursts;
+
+       /* Finish off packet munging and recalculate IP header checksum. */
+       c->iph->tot_len = htons(c->iph->tot_len);
+       c->iph->check = 0;
+       c->iph->check = ip_fast_csum((u8 *) c->iph, c->iph->ihl);
+
+       len = c->skb->len + ((char *)c->iph - (char *)c->eh);
+       c->skb->truesize = len + sizeof(struct sk_buff);
+
+       c->th->window = c->th_last->window;
+       c->th->ack_seq = c->th_last->ack_seq;
+       if (c->th->doff == c->th_last->doff) {
+               /* Copy TCP options (take care to avoid going negative). */
+               len = ((c->th->doff - 5) & 0xf) << 2u;
+               memcpy(c->th + 1, c->th_last + 1, len);
+       }
+
+       /* Allow callback to veto the packet. */
+       veto = EFX_DL_CALLBACK(efx, rx_packet, (char *)c->eh, len);
+       if (unlikely(veto)) {
+               EFX_LOG(efx, "RX vetoed by driverlink %s driver\n",
+                       efx->dl_cb_dev.rx_packet->driver->name);
+               dev_kfree_skb_any(c->skb);
+       } else {
+               netif_receive_skb(c->skb);
+       }
+
+       c->skb = NULL;
+       list_del_init(&c->active_link);
+}
+
+/* Stop tracking connections that have gone idle in order to keep hash
+ * chains short.
+ */
+static void efx_ssr_purge_idle(struct efx_ssr_state *st, unsigned now)
+{
+       struct efx_ssr_conn *c;
+       unsigned i;
+
+       EFX_BUG_ON_PARANOID(!list_empty(&st->active_conns));
+
+       st->last_purge_jiffies = now;
+       for (i = 0; i <= st->conns_mask; ++i) {
+               if (list_empty(&st->conns[i]))
+                       continue;
+
+               c = list_entry(st->conns[i].prev, struct efx_ssr_conn, link);
+               if (now - c->last_pkt_jiffies > lro_idle_jiffies) {
+                       ++st->n_drop_idle;
+                       efx_ssr_drop(st, c, i);
+               }
+       }
+}
+
+/* Push held skbs down into network stack.
+ * Only called when active list is non-empty.
+ */
+void __efx_ssr_end_of_burst(struct efx_ssr_state *st)
+{
+       struct efx_ssr_conn *c;
+       unsigned j;
+
+       EFX_BUG_ON_PARANOID(list_empty(&st->active_conns));
+
+       do {
+               c = list_entry(st->active_conns.next, struct efx_ssr_conn,
+                              active_link);
+               EFX_BUG_ON_PARANOID(!c->skb);
+               efx_ssr_deliver(st, c);
+       } while (!list_empty(&st->active_conns));
+
+       j = jiffies;
+       if (unlikely(j != st->last_purge_jiffies))
+               efx_ssr_purge_idle(st, j);
+}
+
+/* Construct an skb Push held skbs down into network stack.
+ * Only called when active list is non-empty.
+ */
+static inline int
+efx_ssr_merge(struct efx_ssr_state *st, struct efx_ssr_conn *c,
+             struct tcphdr *th, int data_length)
+{
+       /* Increase lengths appropriately */
+       c->skb->len += data_length;
+       c->skb->data_len += data_length;
+
+       /*
+        * Keep track of max MSS seen and store in
+        * gso_size for kernel to use
+        */
+       if (data_length > skb_shinfo(c->skb)->gso_size)
+               skb_shinfo(c->skb)->gso_size = data_length;
+
+       /* Update the connection state flags */
+       c->iph->tot_len += data_length;
+       c->th->psh |= th->psh;
+       c->th_last = th;
+       ++st->n_merges;
+
+       /* Pass packet up now if another segment could overflow the IP
+        * length.
+        */
+       return (c->skb->len > 65536 - 9200);
+}
+
+static inline void
+efx_ssr_start(struct efx_ssr_state *st, struct efx_ssr_conn *c,
+             struct tcphdr *th, int data_length)
+{
+       /* Initialise gso_size appropriately */
+       skb_shinfo(c->skb)->gso_size = data_length;
+
+       /* Mangle header fields for later processing */
+       c->iph->tot_len = ntohs(c->iph->tot_len);
+
+       /* Move this connection the head of the active list */
+       list_del(&c->active_link);
+       list_add(&c->active_link, &st->active_conns);
+}
+
+static inline int
+efx_ssr_conn_page(struct efx_ssr_state *st, struct efx_ssr_conn *c,
+                 struct efx_rx_buffer *rx_buf, struct tcphdr *th,
+                 int hdr_length, int data_length)
+{
+       if (likely(c->skb)) {
+               struct skb_frag_struct *frag;
+               frag = skb_shinfo(c->skb)->frags;
+               frag += skb_shinfo(c->skb)->nr_frags;
+               frag->page = rx_buf->page;
+               frag->page_offset = RX_BUF_OFFSET(rx_buf) + hdr_length;
+               frag->size = data_length;
+               ++skb_shinfo(c->skb)->nr_frags;
+               rx_buf->page = NULL;
+
+               if (efx_ssr_merge(st, c, th, data_length) ||
+                   (skb_shinfo(c->skb)->nr_frags == MAX_SKB_FRAGS))
+                       efx_ssr_deliver(st, c);
+
+               return 1;
+       } else {
+               c->skb = efx_rx_mk_skb(rx_buf, st->efx, hdr_length);
+               if (unlikely(c->skb == NULL))
+                       return 0;
+
+               c->eh = eth_hdr(c->skb);
+               c->iph = (struct iphdr *)c->skb->data;
+               c->th = (struct tcphdr *)((u8 *) c->iph + c->iph->ihl * 4);
+               c->th_last = c->th;
+
+               efx_ssr_start(st, c, th, data_length);
+
+               return 1;
+       }
+}
+
+static inline void
+efx_ssr_conn_skb(struct efx_ssr_state *st, struct efx_ssr_conn *c,
+                struct efx_rx_buffer *rx_buf, struct ethhdr *eh,
+                struct iphdr *iph, struct tcphdr *th, int data_length)
+{
+       /* Transfer ownership of the rx_buf->skb to the LRO chain */
+       struct sk_buff *skb = rx_buf->skb;
+       rx_buf->skb = NULL;
+
+       if (likely(c->skb)) {
+               /* Remove the headers */
+               skb_pull(skb, skb->len - data_length);
+
+               /* Tack the new skb onto the head skb's frag_list. */
+               EFX_BUG_ON_PARANOID(skb->next);
+               if (!skb_shinfo(c->skb)->frag_list)
+                       skb_shinfo(c->skb)->frag_list = skb;
+               else
+                       c->skb_tail->next = skb;
+               c->skb_tail = skb;
+
+               if (efx_ssr_merge(st, c, th, data_length))
+                       efx_ssr_deliver(st, c);
+       } else {
+               c->skb = skb;
+               c->eh = eh;
+               c->iph = iph;
+               c->th = th;
+               c->th_last = th;
+
+               efx_ssr_start(st, c, th, data_length);
+       }
+}
+
+/* Process SKB and decide whether to dispatch it to the stack now or
+ * later.
+ */
+int efx_ssr(struct efx_ssr_state *st, struct efx_rx_buffer *rx_buf)
+{
+
+       int eh_proto, data_length, hdr_length, dont_merge;
+       struct efx_ssr_conn *c;
+       struct ethhdr *eh;
+       struct iphdr *iph;
+       struct tcphdr *th;
+       unsigned th_seq, conn_hash, pkt_length;
+
+       /* This does not handle VLAN code */
+       /* Find the IP header. The ethernet header is always at rx_buf->data */
+       eh = (struct ethhdr *)rx_buf->data;
+       if (rx_buf->page) {
+               eh_proto = eh->h_proto;
+               iph = (struct iphdr *)(eh + 1);
+       } else {
+               /* The skb head is at the IP header */
+               eh_proto = rx_buf->skb->protocol;
+               iph = (struct iphdr *)rx_buf->skb->data;
+       }
+
+       /* We're not interested if it isn't TCP over IPv4, or if fragged. */
+       if ((eh_proto - htons(ETH_P_IP)) |
+           (iph->protocol - IPPROTO_TCP) |
+           (iph->frag_off & htons(IP_MF | IP_OFFSET)))
+               return 0;
+
+       /* Get the TCP protocol */
+       th = (struct tcphdr *)((u8 *) iph + iph->ihl * 4);
+       hdr_length = (u8 *) th + th->doff * 4 - (u8 *) eh;
+       /* Cope with padding after IP header */
+       pkt_length = ntohs(iph->tot_len) + (u8 *)iph - (u8 *)eh;
+       rx_buf->len = min(pkt_length, rx_buf->len);
+       data_length = rx_buf->len - hdr_length;
+       th_seq = ntohl(th->seq);
+       dont_merge = ((data_length <= 0)
+                     | th->urg | th->syn | th->rst | th->fin);
+
+       /* Very cheap and crude hash. */
+       conn_hash = (th->source ^ th->dest) & st->conns_mask;
+
+       list_for_each_entry(c, &st->conns[conn_hash], link) {
+               if ((c->saddr - iph->saddr) | (c->daddr - iph->daddr) |
+                   (c->source - th->source) | (c->dest - th->dest))
+                       continue;
+
+               /* Re-insert at head of list to reduce lookup time. */
+               list_del(&c->link);
+               list_add(&c->link, &st->conns[conn_hash]);
+
+               if (unlikely(th_seq - c->next_seq)) {
+                       /* Out-of-order, so start counting again. */
+                       if (c->skb)
+                               efx_ssr_deliver(st, c);
+                       c->n_in_order_pkts = 0;
+                       c->next_seq = th_seq + data_length;
+                       ++st->n_misorder;
+                       return 0;
+               }
+               c->next_seq = th_seq + data_length;
+               c->last_pkt_jiffies = jiffies;
+
+               if (c->n_in_order_pkts < lro_slow_start_packets) {
+                       /* May be in slow-start, so don't merge. */
+                       ++st->n_slow_start;
+                       ++c->n_in_order_pkts;
+                       return 0;
+               }
+
+               if (unlikely(dont_merge)) {
+                       if (c->skb)
+                               efx_ssr_deliver(st, c);
+                       if (th->fin || th->rst) {
+                               ++st->n_drop_closed;
+                               efx_ssr_drop(st, c, conn_hash);
+                       }
+                       return 0;
+               }
+
+               if (rx_buf->page) {
+                       return efx_ssr_conn_page(st, c, rx_buf, th, hdr_length,
+                                                data_length);
+               } else {
+                       efx_ssr_conn_skb(st, c, rx_buf, eh, iph, th,
+                                        data_length);
+                       return 1;
+               }
+       }
+
+       /* We're not yet tracking this connection. */
+       if (dont_merge)
+               return 0;
+
+       if (st->conns_n[conn_hash] >= lro_chain_max) {
+               ++st->n_too_many;
+               return 0;
+       }
+
+       if (!list_empty(&st->free_conns)) {
+               c = list_entry(st->free_conns.next, struct efx_ssr_conn, link);
+               list_del(&c->link);
+       } else {
+               c = kmalloc(sizeof(*c), GFP_ATOMIC);
+               if (c == NULL)
+                       return 0;
+               c->skb = NULL;
+               INIT_LIST_HEAD(&c->active_link);
+       }
+
+       /* Create the connection tracking data */
+       ++st->conns_n[conn_hash];
+       list_add(&c->link, &st->conns[conn_hash]);
+       c->saddr = iph->saddr;
+       c->daddr = iph->daddr;
+       c->source = th->source;
+       c->dest = th->dest;
+       c->next_seq = th_seq + data_length;
+       c->n_in_order_pkts = 0;
+       EFX_BUG_ON_PARANOID(c->skb);
+       ++st->n_new_stream;
+       return 0;
+}
+
+
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/rx.h
--- a/drivers/net/sfc/rx.h      Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/rx.h      Tue Mar 31 11:59:10 2009 +0100
@@ -26,6 +26,7 @@
 #ifndef EFX_RX_H
 #define EFX_RX_H
 
+#include <linux/skbuff.h>
 #include "net_driver.h"
 
 
@@ -34,6 +35,7 @@ int efx_init_rx_queue(struct efx_rx_queu
 int efx_init_rx_queue(struct efx_rx_queue *rx_queue);
 void efx_fini_rx_queue(struct efx_rx_queue *rx_queue);
 
+void efx_flush_lro(struct efx_channel *channel);
 void efx_rx_strategy(struct efx_channel *channel);
 void efx_fast_push_rx_descriptors(struct efx_rx_queue *rx_queue);
 void efx_rx_work(struct work_struct *data);
@@ -41,4 +43,19 @@ void __efx_rx_packet(struct efx_channel 
                     struct efx_rx_buffer *rx_buf, int checksummed);
 
 
+
+extern int efx_ssr_init(struct efx_ssr_state *st, struct efx_nic *efx);
+extern void efx_ssr_fini(struct efx_ssr_state *st);
+
+extern void __efx_ssr_end_of_burst(struct efx_ssr_state *st);
+extern int efx_ssr(struct efx_ssr_state *st, struct efx_rx_buffer *rx_buf);
+
+
+static inline void efx_ssr_end_of_burst(struct efx_ssr_state *st)
+{
+       if (!list_empty(&st->active_conns))
+               __efx_ssr_end_of_burst(st);
+}
+
+
 #endif /* EFX_RX_H */
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/selftest.c
--- a/drivers/net/sfc/selftest.c        Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/selftest.c        Tue Mar 31 11:59:10 2009 +0100
@@ -35,7 +35,6 @@
 #include <linux/in.h>
 #include <linux/udp.h>
 #include <linux/rtnetlink.h>
-#include <asm/io.h>
 #include "net_driver.h"
 #include "ethtool.h"
 #include "efx.h"
@@ -245,8 +244,7 @@ static int efx_test_eventq_irq(struct ef
 
                if (channel->eventq_magic == magic)
                        goto eventq_ok;
-       }
-       while (++count < 2);
+       } while (++count < 2);
 
        EFX_ERR(channel->efx, "channel %d timed out in %ld jiffies waiting for"
                " event queue\n", channel->channel, jiffies - j_start);
@@ -471,9 +469,6 @@ static int efx_test_loopback(struct efx_
                             struct efx_tx_queue *tx_queue,
                             struct efx_loopback_self_tests *lb_tests)
 {
-#if !defined(EFX_HAVE_OLD_NAPI)
-       struct efx_channel *channel;
-#endif
        struct efx_selftest_state *state = efx->loopback_selftest;
        struct efx_loopback_payload *payload;
        struct sk_buff *skb;
@@ -513,17 +508,8 @@ static int efx_test_loopback(struct efx_
                udelay(10);
        }
 
-#if !defined(EFX_HAVE_OLD_NAPI)
-       /* NAPI polling is not enabled, so process channels synchronously */
-       schedule_timeout_uninterruptible(HZ / 50);
-       efx_for_each_channel_with_interrupt(channel, efx) {
-               if (channel->work_pending)
-                       efx_process_channel_now(channel);
-       }
-#else
        /* Allow time for processing */
        schedule_timeout_uninterruptible(HZ / 10);
-#endif
 
        if (state->flush)
                goto out3;
@@ -625,7 +611,7 @@ static int efx_test_loopbacks(struct efx
        struct ethtool_cmd ecmd, ecmd_loopback;
        struct efx_tx_queue *tx_queue;
        enum efx_loopback_mode old_mode, mode;
-       int old_powered, count, rc = 0;
+       int old_powered, count, rc = 0, link_up;
        int retry = EFX_WORKAROUND_8909(efx);
 
        /* Get current PHY settings */
@@ -663,7 +649,7 @@ static int efx_test_loopbacks(struct efx
                state->flush = 1;
                efx->phy_powered = 1;
                efx->loopback_mode = mode;
-               efx_reconfigure_port(efx, 0);
+               efx_reconfigure_port(efx);
 
                /* Wait for the PHY to signal the link is up */
                count = 0;
@@ -677,11 +663,21 @@ static int efx_test_loopbacks(struct efx
                        /* Wait for PHY events to be processed */
                        flush_workqueue(efx->workqueue);
                        rmb();
-               } while ((++count < 20) && !efx->link_up);
+
+                       /* efx->link_up can be 1 even if the XAUI link is down,
+                        * (bug5762). Usually, it's not worth bothering with the
+                        * difference, but for selftests, we need that extra
+                        * guarantee that the link is really, really, up.
+                        */
+                       link_up = efx->link_up;
+                       if (EFX_IS10G(efx) && !falcon_xaui_link_ok(efx))
+                               link_up = 0;
+
+               } while ((++count < 20) && !link_up);
 
                /* The link should now be up. If it isn't, there is no point
                 * in attempting a loopback test */
-               if (!efx->link_up) {
+               if (!link_up) {
                        EFX_ERR(efx, "loopback %s never came up\n",
                                LOOPBACK_MODE(efx));
                        rc = -EIO;
@@ -712,7 +708,7 @@ fail:
 
                        state->flush = 1;
                        efx->loopback_mode = first;
-                       efx_reconfigure_port(efx, 0);
+                       efx_reconfigure_port(efx);
 
                        retry = rc = 0;
                        --mode;
@@ -750,8 +746,6 @@ int efx_online_test(struct efx_nic *efx,
        struct efx_channel *channel;
        int rc = 0;
 
-       ASSERT_RTNL();
-
        EFX_LOG(efx, "performing online self-tests\n");
 
        rc |= efx_test_interrupts(efx, tests);
@@ -778,8 +772,6 @@ int efx_offline_test(struct efx_nic *efx
 {
        struct efx_selftest_state *state;
        int rc = 0;
-
-       ASSERT_RTNL();
 
        EFX_LOG(efx, "performing offline self-tests\n");
 
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/ci/driver/efab/hardware.h
--- a/drivers/net/sfc/sfc_resource/ci/driver/efab/hardware.h    Tue Mar 31 
11:49:12 2009 +0100
+++ b/drivers/net/sfc/sfc_resource/ci/driver/efab/hardware.h    Tue Mar 31 
11:59:10 2009 +0100
@@ -70,22 +70,23 @@
 #define efhw_nic_close_hardware(nic) \
        ((nic)->efhw_func->close_hardware(nic))
 
-#define efhw_nic_init_hardware(nic, ev_handlers, mac_addr) \
-       ((nic)->efhw_func->init_hardware((nic), (ev_handlers), (mac_addr)))
+#define efhw_nic_init_hardware(nic, ev_handlers, mac_addr, non_irq_evq) \
+       ((nic)->efhw_func->init_hardware((nic), (ev_handlers), (mac_addr), \
+                                        (non_irq_evq)))
 
 /*-------------- Interrupt support  ------------ */
 /** Handle interrupt.  Return 0 if not handled, 1 if handled. */
 #define efhw_nic_interrupt(nic) \
        ((nic)->efhw_func->interrupt(nic))
 
-#define efhw_nic_interrupt_enable(nic, index) \
-       ((nic)->efhw_func->interrupt_enable(nic, index))
-
-#define efhw_nic_interrupt_disable(nic, index) \
-       ((nic)->efhw_func->interrupt_disable(nic, index))
-
-#define efhw_nic_set_interrupt_moderation(nic, index, val) \
-       ((nic)->efhw_func->set_interrupt_moderation(nic, index, val))
+#define efhw_nic_interrupt_enable(nic) \
+       ((nic)->efhw_func->interrupt_enable(nic))
+
+#define efhw_nic_interrupt_disable(nic) \
+       ((nic)->efhw_func->interrupt_disable(nic))
+
+#define efhw_nic_set_interrupt_moderation(nic, val) \
+       ((nic)->efhw_func->set_interrupt_moderation(nic, val))
 
 /*-------------- Event support  ------------ */
 
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/ci/driver/efab/hardware/falcon.h
--- a/drivers/net/sfc/sfc_resource/ci/driver/efab/hardware/falcon.h     Tue Mar 
31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/sfc_resource/ci/driver/efab/hardware/falcon.h     Tue Mar 
31 11:59:10 2009 +0100
@@ -197,7 +197,7 @@ union __u64to32 {
 };
 
 static inline void
-falcon_write_ddd_d(efhw_ioaddr_t kva,
+falcon_write_ddd_d(volatile char __iomem *kva,
                   uint32_t d0, uint32_t d1, uint32_t d2, uint32_t d3)
 {
        writel(d0, kva + 0);
@@ -207,7 +207,7 @@ falcon_write_ddd_d(efhw_ioaddr_t kva,
        writel(d3, kva + 12);
 }
 
-static inline void falcon_write_q(efhw_ioaddr_t kva, uint64_t q)
+static inline void falcon_write_q(volatile char __iomem *kva, uint64_t q)
 {
        union __u64to32 u;
        u.u64 = q;
@@ -217,7 +217,7 @@ static inline void falcon_write_q(efhw_i
        writel(u.s.b, kva + 4);
 }
 
-static inline void falcon_read_q(efhw_ioaddr_t addr, uint64_t *q0)
+static inline void falcon_read_q(volatile char __iomem *addr, uint64_t *q0)
 {
        /* It is essential that we read dword0 first, so that
         * the shadow register is updated with the latest value
@@ -232,14 +232,14 @@ static inline void falcon_read_q(efhw_io
 }
 
 static inline void
-falcon_write_qq(efhw_ioaddr_t kva, uint64_t q0, uint64_t q1)
+falcon_write_qq(volatile char __iomem *kva, uint64_t q0, uint64_t q1)
 {
        writeq(q0, kva + 0);
        falcon_write_q(kva + 8, q1);
 }
 
 static inline void
-falcon_read_qq(efhw_ioaddr_t addr, uint64_t *q0, uint64_t *q1)
+falcon_read_qq(volatile char __iomem *addr, uint64_t *q0, uint64_t *q1)
 {
        falcon_read_q(addr, q0);
        *q1 = readq(addr + 8);
@@ -390,9 +390,6 @@ static inline int falcon_rx_dma_page_off
 /* Falcon nails down the event queue mappings */
 #define FALCON_EVQ_KERNEL0   (0)       /* hardwired for net driver */
 #define FALCON_EVQ_CHAR      (4)       /* char driver's event queue      */
-#define FALCON_EVQ_NONIRQ    (5)       /* char driver's non interrupting
-                                          queue. Subsequent queues are
-                                          available for user apps */
 
 /* reserved by the drivers */
 #define FALCON_EVQ_TBL_RESERVED           (8)
@@ -411,7 +408,8 @@ static inline int falcon_rx_dma_page_off
  *
  *---------------------------------------------------------------------------*/
 
-static inline void falcon_deadbeef(efhw_ioaddr_t efhw_kva, unsigned what)
+static inline void
+falcon_deadbeef(volatile char __iomem *efhw_kva, unsigned what)
 {
        writel(what, efhw_kva + 0x300);
        mmiowb();
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/ci/driver/resource/efx_vi.h
--- a/drivers/net/sfc/sfc_resource/ci/driver/resource/efx_vi.h  Tue Mar 31 
11:49:12 2009 +0100
+++ b/drivers/net/sfc/sfc_resource/ci/driver/resource/efx_vi.h  Tue Mar 31 
11:59:10 2009 +0100
@@ -190,8 +190,6 @@ efx_vi_filter_stop(struct efx_vi_state *
 /*! Constants for the type field in efx_vi_hw_resource */
 #define EFX_VI_HW_RESOURCE_TXDMAQ    0x0       /* PFN of TX DMA Q */
 #define EFX_VI_HW_RESOURCE_RXDMAQ    0x1       /* PFN of RX DMA Q */
-#define EFX_VI_HW_RESOURCE_TXBELL    0x2       /* PFN of TX Doorbell (EF1) */
-#define EFX_VI_HW_RESOURCE_RXBELL    0x3       /* PFN of RX Doorbell (EF1) */
 #define EFX_VI_HW_RESOURCE_EVQTIMER  0x4       /* Address of event q timer */
 
 /* Address of event q pointer (EF1) */
@@ -229,7 +227,6 @@ struct efx_vi_hw_resource {
  * Metadata concerning the list of hardware resource mappings
  */
 struct efx_vi_hw_resource_metadata {
-       int version;
        int evq_order;
        int evq_offs;
        int evq_capacity;
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/ci/driver/resource/linux_efhw_nic.h
--- a/drivers/net/sfc/sfc_resource/ci/driver/resource/linux_efhw_nic.h  Tue Mar 
31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/sfc_resource/ci/driver/resource/linux_efhw_nic.h  Tue Mar 
31 11:59:10 2009 +0100
@@ -38,12 +38,6 @@
 #ifndef __CI_DRIVER_RESOURCE_LINUX_RESOURCE__
 #define __CI_DRIVER_RESOURCE_LINUX_RESOURCE__
 
-#ifndef __linux__
-# error Silly
-#endif
-#ifndef __KERNEL__
-# error Silly
-#endif
 
 #include <ci/efhw/efhw_types.h>
 #include <linux/interrupt.h>
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/ci/efhw/common.h
--- a/drivers/net/sfc/sfc_resource/ci/efhw/common.h     Tue Mar 31 11:49:12 
2009 +0100
+++ b/drivers/net/sfc/sfc_resource/ci/efhw/common.h     Tue Mar 31 11:59:10 
2009 +0100
@@ -56,10 +56,6 @@ typedef union {
                uint32_t a;
                uint32_t b;
        } opaque;
-       struct {
-               uint32_t code;
-               uint32_t status;
-       } ev1002;
 } efhw_event_t;
 
 /* Flags for TX/RX queues */
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/ci/efhw/common_sysdep.h
--- a/drivers/net/sfc/sfc_resource/ci/efhw/common_sysdep.h      Tue Mar 31 
11:49:12 2009 +0100
+++ b/drivers/net/sfc/sfc_resource/ci/efhw/common_sysdep.h      Tue Mar 31 
11:59:10 2009 +0100
@@ -52,7 +52,7 @@
 
 /* Linux kernel also does not provide PRIx32... Sigh. */
 #define PRIx32 "x"
-
+ 
 #ifdef __ia64__
 # define PRIx64 "lx"
 #else
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/ci/efhw/efhw_types.h
--- a/drivers/net/sfc/sfc_resource/ci/efhw/efhw_types.h Tue Mar 31 11:49:12 
2009 +0100
+++ b/drivers/net/sfc/sfc_resource/ci/efhw/efhw_types.h Tue Mar 31 11:59:10 
2009 +0100
@@ -45,14 +45,6 @@
 
 /*--------------------------------------------------------------------
  *
- * hardware limits used in the types
- *
- *--------------------------------------------------------------------*/
-
-#define EFHW_KEVENTQ_MAX    8
-
-/*--------------------------------------------------------------------
- *
  * forward type declarations
  *
  *--------------------------------------------------------------------*/
@@ -72,7 +64,7 @@ struct efhw_buffer_table_allocation{
 
 struct eventq_resource_hardware {
        /*!iobuffer allocated for eventq - can be larger than eventq */
-       efhw_iopages_t iobuff;
+       struct efhw_iopages iobuff;
        unsigned iobuff_off;
        struct efhw_buffer_table_allocation buf_tbl_alloc;
        int capacity;           /*!< capacity of event queue */
@@ -85,7 +77,7 @@ struct eventq_resource_hardware {
  *--------------------------------------------------------------------*/
 
 struct efhw_keventq {
-       volatile int lock;
+       int lock;
        caddr_t evq_base;
        int32_t evq_ptr;
        uint32_t evq_mask;
@@ -115,7 +107,7 @@ struct efhw_func_ops {
        /*! initialise all hardware functional units */
        int (*init_hardware) (struct efhw_nic *nic,
                              struct efhw_ev_handler *,
-                             const uint8_t *mac_addr);
+                             const uint8_t *mac_addr, int non_irq_evq);
 
   /*-------------- Interrupt support  ------------ */
 
@@ -130,17 +122,17 @@ struct efhw_func_ops {
         */
        int (*interrupt) (struct efhw_nic *nic);
 
-       /*! Enable given interrupt mask for the given IRQ unit */
-       void (*interrupt_enable) (struct efhw_nic *nic, uint idx);
-
-       /*! Disable given interrupt mask for the given IRQ unit */
-       void (*interrupt_disable) (struct efhw_nic *nic, uint idx);
+       /*! Enable the interrupt */
+       void (*interrupt_enable) (struct efhw_nic *nic);
+
+       /*! Disable the interrupt */
+       void (*interrupt_disable) (struct efhw_nic *nic);
 
        /*! Set interrupt moderation strategy for the given IRQ unit
         ** val is in usec
         */
        void (*set_interrupt_moderation)(struct efhw_nic *nic,
-                                        uint idx, uint val);
+                                        uint val);
 
   /*-------------- Event support  ------------ */
 
@@ -255,8 +247,8 @@ struct efhw_device_type {
 
 /*! */
 struct efhw_nic {
-       /*! zero base index in efrm_nic_table.nic array */
-       volatile int index;
+       /*! zero base index in efrm_nic_tablep->nic array */
+       int index;
        int ifindex;            /*!< OS level nic index */
 #ifdef HAS_NET_NAMESPACE
        struct net *nd_net;
@@ -283,7 +275,7 @@ struct efhw_nic {
        /* hardware resources */
 
        /*! I/O address of the start of the bar */
-       efhw_ioaddr_t bar_ioaddr;
+       volatile char __iomem *bar_ioaddr;
 
        /*! Bar number of control aperture. */
        unsigned ctr_ap_bar;
@@ -312,14 +304,17 @@ struct efhw_nic {
        void (*irq_handler) (struct efhw_nic *, int unit);
 
        /*! event queues per driver */
-       struct efhw_keventq evq[EFHW_KEVENTQ_MAX];
+       struct efhw_keventq interrupting_evq;
 
 /* for marking when we are not using an IRQ unit
       - 0 is a valid offset to an IRQ unit on EF1! */
 #define EFHW_IRQ_UNIT_UNUSED  0xffff
-       /*! interrupt unit in use  */
-       unsigned int irq_unit[EFHW_KEVENTQ_MAX];
-       efhw_iopage_t irq_iobuff;       /*!<  Falcon SYSERR interrupt */
+       /*! interrupt unit in use for the interrupting event queue  */
+       unsigned int irq_unit;
+
+       struct efhw_keventq non_interrupting_evq;
+
+       struct efhw_iopage irq_iobuff;  /*!<  Falcon SYSERR interrupt */
 
        /* The new driverlink infrastructure. */
        struct efx_dl_device *net_driver_dev;
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/ci/efhw/eventq.h
--- a/drivers/net/sfc/sfc_resource/ci/efhw/eventq.h     Tue Mar 31 11:49:12 
2009 +0100
+++ b/drivers/net/sfc/sfc_resource/ci/efhw/eventq.h     Tue Mar 31 11:59:10 
2009 +0100
@@ -47,10 +47,9 @@ extern int efhw_keventq_poll(struct efhw
 
 /*! Callbacks for handling events. */
 struct efhw_ev_handler {
-       void (*wakeup_fn)(struct efhw_nic *nic, efhw_event_t *ev);
-       void (*timeout_fn)(struct efhw_nic *nic, efhw_event_t *ev);
-       void (*sw_fn)(struct efhw_nic *nic, efhw_event_t *ev);
-       void (*dmaq_flushed_fn) (struct efhw_nic *, int, int);
+       void (*wakeup_fn)(struct efhw_nic *nic, unsigned);
+       void (*timeout_fn)(struct efhw_nic *nic, unsigned);
+       void (*dmaq_flushed_fn) (struct efhw_nic *, unsigned, int);
 };
 
 extern int efhw_keventq_ctor(struct efhw_nic *, int instance,
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/ci/efhw/falcon.h
--- a/drivers/net/sfc/sfc_resource/ci/efhw/falcon.h     Tue Mar 31 11:49:12 
2009 +0100
+++ b/drivers/net/sfc/sfc_resource/ci/efhw/falcon.h     Tue Mar 31 11:59:10 
2009 +0100
@@ -71,11 +71,6 @@ falcon_handle_char_event(struct efhw_nic
 falcon_handle_char_event(struct efhw_nic *nic,
                         struct efhw_ev_handler *h, efhw_event_t *evp);
 
-/*! map event queue instance space (0,1,2,..) onto event queue
-  number. This function takes into account the allocation rules for
-  the underlying driver model */
-extern int falcon_idx_to_evq(struct efhw_nic *nic, uint idx);
-
 /*! Acknowledge to HW that processing is complete on a given event queue */
 extern void falcon_nic_evq_ack(struct efhw_nic *nic, uint evq, /* evq id */
                               uint rptr,       /* new read pointer update */
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/ci/efhw/hardware_sysdep.h
--- a/drivers/net/sfc/sfc_resource/ci/efhw/hardware_sysdep.h    Tue Mar 31 
11:49:12 2009 +0100
+++ b/drivers/net/sfc/sfc_resource/ci/efhw/hardware_sysdep.h    Tue Mar 31 
11:59:10 2009 +0100
@@ -50,6 +50,10 @@
 #error Unknown endianness
 #endif
 
+#ifndef __iomem
+#define __iomem
+#endif
+
 #ifndef mmiowb
        #if defined(__i386__) || defined(__x86_64__)
                #define mmiowb()
@@ -63,10 +67,8 @@
        #endif
 #endif
 
-typedef char *efhw_ioaddr_t;
-
 #ifndef readq
-static inline uint64_t __readq(void __iomem *addr)
+static inline uint64_t __readq(volatile void __iomem *addr)
 {
        return *(volatile uint64_t *)addr;
 }
@@ -74,7 +76,7 @@ static inline uint64_t __readq(void __io
 #endif
 
 #ifndef writeq
-static inline void __writeq(uint64_t v, void __iomem *addr)
+static inline void __writeq(uint64_t v, volatile void __iomem *addr)
 {
        *(volatile uint64_t *)addr = v;
 }
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/ci/efhw/iopage.h
--- a/drivers/net/sfc/sfc_resource/ci/efhw/iopage.h     Tue Mar 31 11:49:12 
2009 +0100
+++ b/drivers/net/sfc/sfc_resource/ci/efhw/iopage.h     Tue Mar 31 11:59:10 
2009 +0100
@@ -48,11 +48,11 @@
  *
  *--------------------------------------------------------------------*/
 
-extern int efhw_iopage_alloc(struct efhw_nic *, efhw_iopage_t *p);
-extern void efhw_iopage_free(struct efhw_nic *, efhw_iopage_t *p);
+extern int efhw_iopage_alloc(struct efhw_nic *, struct efhw_iopage *p);
+extern void efhw_iopage_free(struct efhw_nic *, struct efhw_iopage *p);
 
-extern int efhw_iopages_alloc(struct efhw_nic *, efhw_iopages_t *p,
+extern int efhw_iopages_alloc(struct efhw_nic *, struct efhw_iopages *p,
                              unsigned order);
-extern void efhw_iopages_free(struct efhw_nic *, efhw_iopages_t *p);
+extern void efhw_iopages_free(struct efhw_nic *, struct efhw_iopages *p);
 
 #endif /* __CI_DRIVER_RESOURCE_IOPAGE_H__ */
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/ci/efhw/iopage_types.h
--- a/drivers/net/sfc/sfc_resource/ci/efhw/iopage_types.h       Tue Mar 31 
11:49:12 2009 +0100
+++ b/drivers/net/sfc/sfc_resource/ci/efhw/iopage_types.h       Tue Mar 31 
11:59:10 2009 +0100
@@ -3,7 +3,8 @@
  *          resource management for Xen backend, OpenOnload, etc
  *           (including support for SFE4001 10GBT NIC)
  *
- * This file provides efhw_page_t and efhw_iopage_t for Linux kernel.
+ * This file provides struct efhw_page and struct efhw_iopage for Linux
+ * kernel.
  *
  * Copyright 2005-2007: Solarflare Communications Inc,
  *                      9501 Jeronimo Road, Suite 250,
@@ -38,77 +39,83 @@
 #ifndef __CI_EFHW_IOPAGE_LINUX_H__
 #define __CI_EFHW_IOPAGE_LINUX_H__
 
+#include <linux/version.h>
 #include <linux/gfp.h>
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,9)
 #include <linux/hardirq.h>
+#else
+#include <asm/hardirq.h>
+#endif
+#include <linux/errno.h>
 #include <ci/efhw/debug.h>
 
 /*--------------------------------------------------------------------
  *
- * efhw_page_t: A single page of memory.  Directly mapped in the driver,
- * and can be mapped to userlevel.
+ * struct efhw_page: A single page of memory.  Directly mapped in the
+ * driver, and can be mapped to userlevel.
  *
  *--------------------------------------------------------------------*/
 
-typedef struct {
+struct efhw_page {
        unsigned long kva;
-} efhw_page_t;
+};
 
-static inline int efhw_page_alloc(efhw_page_t *p)
+static inline int efhw_page_alloc(struct efhw_page *p)
 {
        p->kva = __get_free_page(in_interrupt()? GFP_ATOMIC : GFP_KERNEL);
        return p->kva ? 0 : -ENOMEM;
 }
 
-static inline int efhw_page_alloc_zeroed(efhw_page_t *p)
+static inline int efhw_page_alloc_zeroed(struct efhw_page *p)
 {
        p->kva = get_zeroed_page(in_interrupt()? GFP_ATOMIC : GFP_KERNEL);
        return p->kva ? 0 : -ENOMEM;
 }
 
-static inline void efhw_page_free(efhw_page_t *p)
+static inline void efhw_page_free(struct efhw_page *p)
 {
        free_page(p->kva);
        EFHW_DO_DEBUG(memset(p, 0, sizeof(*p)));
 }
 
-static inline char *efhw_page_ptr(efhw_page_t *p)
+static inline char *efhw_page_ptr(struct efhw_page *p)
 {
        return (char *)p->kva;
 }
 
-static inline unsigned efhw_page_pfn(efhw_page_t *p)
+static inline unsigned efhw_page_pfn(struct efhw_page *p)
 {
        return (unsigned)(__pa(p->kva) >> PAGE_SHIFT);
 }
 
-static inline void efhw_page_mark_invalid(efhw_page_t *p)
+static inline void efhw_page_mark_invalid(struct efhw_page *p)
 {
        p->kva = 0;
 }
 
-static inline int efhw_page_is_valid(efhw_page_t *p)
+static inline int efhw_page_is_valid(struct efhw_page *p)
 {
        return p->kva != 0;
 }
 
-static inline void efhw_page_init_from_va(efhw_page_t *p, void *va)
+static inline void efhw_page_init_from_va(struct efhw_page *p, void *va)
 {
        p->kva = (unsigned long)va;
 }
 
 /*--------------------------------------------------------------------
  *
- * efhw_iopage_t: A single page of memory.  Directly mapped in the driver,
+ * struct efhw_iopage: A single page of memory.  Directly mapped in the driver,
  * and can be mapped to userlevel.  Can also be accessed by the NIC.
  *
  *--------------------------------------------------------------------*/
 
-typedef struct {
-       efhw_page_t p;
+struct efhw_iopage {
+       struct efhw_page p;
        dma_addr_t dma_addr;
-} efhw_iopage_t;
+};
 
-static inline dma_addr_t efhw_iopage_dma_addr(efhw_iopage_t *p)
+static inline dma_addr_t efhw_iopage_dma_addr(struct efhw_iopage *p)
 {
        return p->dma_addr;
 }
@@ -120,9 +127,9 @@ static inline dma_addr_t efhw_iopage_dma
 
 /*--------------------------------------------------------------------
  *
- * efhw_iopages_t: A set of pages that are contiguous in physical memory.
- * Directly mapped in the driver, and can be mapped to userlevel.  Can also
- * be accessed by the NIC.
+ * struct efhw_iopages: A set of pages that are contiguous in physical
+ * memory.  Directly mapped in the driver, and can be mapped to userlevel.
+ * Can also be accessed by the NIC.
  *
  * NB. The O/S may be unwilling to allocate many, or even any of these.  So
  * only use this type where the NIC really needs a physically contiguous
@@ -130,44 +137,44 @@ static inline dma_addr_t efhw_iopage_dma
  *
  *--------------------------------------------------------------------*/
 
-typedef struct {
+struct efhw_iopages {
        caddr_t kva;
        unsigned order;
        dma_addr_t dma_addr;
-} efhw_iopages_t;
+};
 
-static inline caddr_t efhw_iopages_ptr(efhw_iopages_t *p)
+static inline caddr_t efhw_iopages_ptr(struct efhw_iopages *p)
 {
        return p->kva;
 }
 
-static inline unsigned efhw_iopages_pfn(efhw_iopages_t *p)
+static inline unsigned efhw_iopages_pfn(struct efhw_iopages *p)
 {
        return (unsigned)(__pa(p->kva) >> PAGE_SHIFT);
 }
 
-static inline dma_addr_t efhw_iopages_dma_addr(efhw_iopages_t *p)
+static inline dma_addr_t efhw_iopages_dma_addr(struct efhw_iopages *p)
 {
        return p->dma_addr;
 }
 
-static inline unsigned efhw_iopages_size(efhw_iopages_t *p)
+static inline unsigned efhw_iopages_size(struct efhw_iopages *p)
 {
        return 1u << (p->order + PAGE_SHIFT);
 }
 
-/* efhw_iopage_t <-> efhw_iopages_t conversions for handling physically
- * contiguous allocations in iobufsets for iSCSI.  This allows the
- * essential information about contiguous allocations from
- * efhw_iopages_alloc() to be saved away in the efhw_iopage_t array in an
- * iobufset.  (Changing the iobufset resource to use a union type would
+/* struct efhw_iopage <-> struct efhw_iopages conversions for handling
+ * physically contiguous allocations in iobufsets for iSCSI.  This allows
+ * the essential information about contiguous allocations from
+ * efhw_iopages_alloc() to be saved away in the struct efhw_iopage array in
+ * an iobufset.  (Changing the iobufset resource to use a union type would
  * involve a lot of code changes, and make the iobufset's metadata larger
  * which could be bad as it's supposed to fit into a single page on some
  * platforms.)
  */
 static inline void
-efhw_iopage_init_from_iopages(efhw_iopage_t *iopage,
-                           efhw_iopages_t *iopages, unsigned pageno)
+efhw_iopage_init_from_iopages(struct efhw_iopage *iopage,
+                             struct efhw_iopages *iopages, unsigned pageno)
 {
        iopage->p.kva = ((unsigned long)efhw_iopages_ptr(iopages))
            + (pageno * PAGE_SIZE);
@@ -176,8 +183,8 @@ efhw_iopage_init_from_iopages(efhw_iopag
 }
 
 static inline void
-efhw_iopages_init_from_iopage(efhw_iopages_t *iopages,
-                           efhw_iopage_t *iopage, unsigned order)
+efhw_iopages_init_from_iopage(struct efhw_iopages *iopages,
+                             struct efhw_iopage *iopage, unsigned order)
 {
        iopages->kva = (caddr_t) efhw_iopage_ptr(iopage);
        EFHW_ASSERT(iopages->kva);
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/ci/efrm/driver_private.h
--- a/drivers/net/sfc/sfc_resource/ci/efrm/driver_private.h     Tue Mar 31 
11:49:12 2009 +0100
+++ b/drivers/net/sfc/sfc_resource/ci/efrm/driver_private.h     Tue Mar 31 
11:59:10 2009 +0100
@@ -69,16 +69,16 @@ extern int efrm_driver_unregister_nic(st
  *--------------------------------------------------------------------*/
 
 struct vi_resource_dimensions {
-       unsigned evq_int_min, evq_int_max;
-       unsigned evq_timer_min, evq_timer_max;
-       unsigned rxq_min, rxq_max;
-       unsigned txq_min, txq_max;
+       unsigned evq_int_min, evq_int_lim;
+       unsigned evq_timer_min, evq_timer_lim;
+       unsigned rxq_min, rxq_lim;
+       unsigned txq_min, txq_lim;
 };
 
 /*! Initialise resources */
 extern int
 efrm_resources_init(const struct vi_resource_dimensions *,
-                   int buffer_table_min, int buffer_table_max);
+                   int buffer_table_min, int buffer_table_lim);
 
 /*! Tear down resources */
 extern void efrm_resources_fini(void);
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/ci/efrm/iobufset.h
--- a/drivers/net/sfc/sfc_resource/ci/efrm/iobufset.h   Tue Mar 31 11:49:12 
2009 +0100
+++ b/drivers/net/sfc/sfc_resource/ci/efrm/iobufset.h   Tue Mar 31 11:59:10 
2009 +0100
@@ -55,7 +55,7 @@ struct iobufset_resource {
        unsigned int n_bufs;
        unsigned int pages_per_contiguous_chunk;
        unsigned order;
-       efhw_iopage_t bufs[1];
+       struct efhw_iopage bufs[1];
        /*!< up to n_bufs can follow this, so this must be the last member */
 };
 
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/ci/efrm/nic_table.h
--- a/drivers/net/sfc/sfc_resource/ci/efrm/nic_table.h  Tue Mar 31 11:49:12 
2009 +0100
+++ b/drivers/net/sfc/sfc_resource/ci/efrm/nic_table.h  Tue Mar 31 11:59:10 
2009 +0100
@@ -62,21 +62,21 @@ struct efrm_nic_table {
 };
 
 /* Resource driver structures used by other drivers as well */
-extern struct efrm_nic_table efrm_nic_table;
+extern struct efrm_nic_table *efrm_nic_tablep;
 
 static inline void efrm_nic_table_hold(void)
 {
-       atomic_inc(&efrm_nic_table.ref_count);
+       atomic_inc(&efrm_nic_tablep->ref_count);
 }
 
 static inline void efrm_nic_table_rele(void)
 {
-       atomic_dec(&efrm_nic_table.ref_count);
+       atomic_dec(&efrm_nic_tablep->ref_count);
 }
 
 static inline int efrm_nic_table_held(void)
 {
-       return (atomic_read(&efrm_nic_table.ref_count) != 0);
+       return (atomic_read(&efrm_nic_tablep->ref_count) != 0);
 }
 
 /* Run code block _x multiple times with variable nic set to each
@@ -86,13 +86,13 @@ static inline int efrm_nic_table_held(vo
        for ((_nic_i) = (efrm_nic_table_hold(), 0);                     \
             (_nic_i) < EFHW_MAX_NR_DEVS || (efrm_nic_table_rele(), 0); \
             (_nic_i)++)                                                \
-               if (((_nic) = efrm_nic_table.nic[_nic_i]))
+               if (((_nic) = efrm_nic_tablep->nic[_nic_i]))
 
 #define EFRM_FOR_EACH_NIC_IN_SET(_set, _i, _nic)                       \
        for ((_i) = (efrm_nic_table_hold(), 0);                         \
             (_i) < EFHW_MAX_NR_DEVS || (efrm_nic_table_rele(), 0);     \
             ++(_i))                                                    \
-               if (((_nic) = efrm_nic_table.nic[_i]) &&                \
+               if (((_nic) = efrm_nic_tablep->nic[_i]) &&              \
                    efrm_nic_set_read((_set), (_i)))
 
 #endif /* __CI_EFRM_NIC_TABLE_H__ */
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/ci/efrm/private.h
--- a/drivers/net/sfc/sfc_resource/ci/efrm/private.h    Tue Mar 31 11:49:12 
2009 +0100
+++ b/drivers/net/sfc/sfc_resource/ci/efrm/private.h    Tue Mar 31 11:59:10 
2009 +0100
@@ -93,7 +93,15 @@ efrm_kfifo_id_ctor(struct kfifo **ids_ou
        unsigned int i;
        struct kfifo *ids;
        unsigned char *buffer;
+#ifndef TCP_CHIMNEY_SUPPORT
        unsigned int size = roundup_pow_of_two((limit - base) * sizeof(int));
+#else
+        /* ### TODO - Linux kfifos really are a power of two, sysdep_ci2linux
+                      does ci_fifo2's, which only actually hold 2^n - 1.
+                      We need to double buffer size, not add one, because
+                      ci_fifo2 can only be a power of two. */
+       unsigned int size = roundup_pow_of_two((limit - base) * 2 * 
sizeof(int));
+#endif
 
        EFRM_ASSERT(base <= limit);
        buffer = vmalloc(size);
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/ci/efrm/resource_id.h
--- a/drivers/net/sfc/sfc_resource/ci/efrm/resource_id.h        Tue Mar 31 
11:49:12 2009 +0100
+++ b/drivers/net/sfc/sfc_resource/ci/efrm/resource_id.h        Tue Mar 31 
11:59:10 2009 +0100
@@ -48,7 +48,7 @@
  * level.
  ***********************************************************************/
 
-typedef struct efrm_resource_handle_s {
+typedef struct {
        uint32_t handle;
 } efrm_resource_handle_t;
 
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/ci/efrm/sysdep.h
--- a/drivers/net/sfc/sfc_resource/ci/efrm/sysdep.h     Tue Mar 31 11:49:12 
2009 +0100
+++ b/drivers/net/sfc/sfc_resource/ci/efrm/sysdep.h     Tue Mar 31 11:59:10 
2009 +0100
@@ -41,14 +41,8 @@
 /* Spinlocks are defined in efhw/sysdep.h */
 #include <ci/efhw/sysdep.h>
 
-#if defined(__linux__) && defined(__KERNEL__)
 
 # include <ci/efrm/sysdep_linux.h>
 
-#else
-
-# include <ci/efrm/sysdep_ci2linux.h>
-
-#endif
 
 #endif /* __CI_EFRM_SYSDEP_H__ */
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/ci/efrm/sysdep_linux.h
--- a/drivers/net/sfc/sfc_resource/ci/efrm/sysdep_linux.h       Tue Mar 31 
11:49:12 2009 +0100
+++ b/drivers/net/sfc/sfc_resource/ci/efrm/sysdep_linux.h       Tue Mar 31 
11:59:10 2009 +0100
@@ -50,7 +50,11 @@
 #include <linux/workqueue.h>
 #include <linux/gfp.h>
 #include <linux/slab.h>
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,9)
 #include <linux/hardirq.h>
+#else
+#include <asm/hardirq.h>
+#endif
 #include <linux/kernel.h>
 #include <linux/if_ether.h>
 #include <linux/completion.h>
@@ -60,6 +64,21 @@
 /* get roundup_pow_of_two(), which was in kernel.h in early kernel versions */
 #include <linux/log2.h>
 #endif
+
+
+/********************************************************************
+ *
+ * Utility functions
+ *
+ ********************************************************************/
+
+#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,9))
+static inline unsigned long __attribute_const__ roundup_pow_of_two(unsigned 
long x)
+{
+        return (1UL << fls(x - 1));
+}
+#endif
+
 
 /********************************************************************
  *
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/ci/efrm/vi_resource.h
--- a/drivers/net/sfc/sfc_resource/ci/efrm/vi_resource.h        Tue Mar 31 
11:49:12 2009 +0100
+++ b/drivers/net/sfc/sfc_resource/ci/efrm/vi_resource.h        Tue Mar 31 
11:59:10 2009 +0100
@@ -168,4 +168,12 @@ uint32_t efrm_vi_rm_evq_bytes(struct vi_
 uint32_t efrm_vi_rm_evq_bytes(struct vi_resource *virs
                              /*,struct efhw_nic *nic */ );
 
+
+/* Fill [out_vi_data] with information required to allow a VI to be init'd.
+ * [out_vi_data] must ref at least VI_MAPPINGS_SIZE bytes.
+ */
+extern void efrm_vi_resource_mappings(struct vi_resource*, int nic_i,
+                                      void* out_vi_data);
+
+
 #endif /* __CI_EFRM_VI_RESOURCE_H__ */
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/ci/efrm/vi_resource_manager.h
--- a/drivers/net/sfc/sfc_resource/ci/efrm/vi_resource_manager.h        Tue Mar 
31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/sfc_resource/ci/efrm/vi_resource_manager.h        Tue Mar 
31 11:59:10 2009 +0100
@@ -67,11 +67,7 @@ struct vi_resource_evq_info {
        struct vi_resource *evq_virs;
 };
 
-#ifdef __ci_ul_driver__
-#define EFRM_VI_USE_WORKQUEUE 0
-#else
 #define EFRM_VI_USE_WORKQUEUE 1
-#endif
 
 /*! Global information for the VI resource manager. */
 struct vi_resource_manager {
@@ -115,7 +111,7 @@ struct vi_resource_nic_info {
 struct vi_resource_nic_info {
        struct eventq_resource_hardware evq_pages;
 #if defined(__CI_HARDWARE_CONFIG_FALCON__)
-       efhw_iopages_t dmaq_pages[EFRM_VI_RM_DMA_QUEUE_COUNT];
+       struct efhw_iopages dmaq_pages[EFRM_VI_RM_DMA_QUEUE_COUNT];
 #endif
 };
 
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/ci/efrm/vi_resource_private.h
--- a/drivers/net/sfc/sfc_resource/ci/efrm/vi_resource_private.h        Tue Mar 
31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/sfc_resource/ci/efrm/vi_resource_private.h        Tue Mar 
31 11:59:10 2009 +0100
@@ -70,14 +70,17 @@ efrm_eventq_base(struct vi_resource *vir
                                 hw->iobuff_off);
 }
 
-/*! Wakeup handler, see efhw_ev_handler_t for prototype */
-extern void efrm_handle_wakeup_event(struct efhw_nic *nic, efhw_event_t *ev);
+/*! Wakeup handler */
+extern void efrm_handle_wakeup_event(struct efhw_nic *nic, unsigned id);
 
-/*! Timeout handler, see efhw_ev_handler_t for prototype */
-extern void efrm_handle_timeout_event(struct efhw_nic *nic, efhw_event_t *ev);
+/*! Timeout handler */
+extern void efrm_handle_timeout_event(struct efhw_nic *nic, unsigned id);
 
-/*! DMA flush handler, see efhw_ev_handler_t for prototype */
-extern void efrm_handle_dmaq_flushed(struct efhw_nic *nic, int instance,
+/*! DMA flush handler */
+extern void efrm_handle_dmaq_flushed(struct efhw_nic *nic, unsigned id,
                                   int rx_flush);
 
+/*! SRAM update handler */
+extern void efrm_handle_sram_event(struct efhw_nic *nic);
+
 #endif /* __CI_EFRM_VI_RESOURCE_PRIVATE_H__ */
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/driver_object.c
--- a/drivers/net/sfc/sfc_resource/driver_object.c      Tue Mar 31 11:49:12 
2009 +0100
+++ b/drivers/net/sfc/sfc_resource/driver_object.c      Tue Mar 31 11:59:10 
2009 +0100
@@ -48,10 +48,10 @@
 */
 
 #define efrm_driver_lock(irqlock_state) \
-       spin_lock_irqsave(&efrm_nic_table.lock, irqlock_state)
+       spin_lock_irqsave(&efrm_nic_tablep->lock, irqlock_state)
 
 #define efrm_driver_unlock(irqlock_state)              \
-       spin_unlock_irqrestore(&efrm_nic_table.lock,    \
+       spin_unlock_irqrestore(&efrm_nic_tablep->lock,  \
                               irqlock_state);
 
 /* These routines are all methods on the architecturally singleton
@@ -63,8 +63,9 @@
 */
 
 /*! Exported driver state */
-struct efrm_nic_table efrm_nic_table;
-EXPORT_SYMBOL(efrm_nic_table);
+static struct efrm_nic_table efrm_nic_table;
+struct efrm_nic_table *efrm_nic_tablep;
+EXPORT_SYMBOL(efrm_nic_tablep);
 
 /* Internal table with resource managers.
  * We'd like to not export it, but we are still using efrm_rm_table
@@ -75,10 +76,8 @@ EXPORT_SYMBOL(efrm_rm_table);
 
 int efrm_driver_ctor(void)
 {
-       memset(&efrm_nic_table, 0, sizeof(efrm_nic_table));
-       memset(&efrm_rm_table, 0, sizeof(efrm_rm_table));
-
-       spin_lock_init(&efrm_nic_table.lock);
+        efrm_nic_tablep = &efrm_nic_table;
+       spin_lock_init(&efrm_nic_tablep->lock);
 
        EFRM_TRACE("%s: driver created", __FUNCTION__);
        return 0;
@@ -88,7 +87,10 @@ int efrm_driver_dtor(void)
 {
        EFRM_ASSERT(!efrm_nic_table_held());
 
-       spin_lock_destroy(&efrm_nic_table.lock);
+       spin_lock_destroy(&efrm_nic_tablep->lock);
+       memset(&efrm_nic_table, 0, sizeof(efrm_nic_table));
+       memset(&efrm_rm_table, 0, sizeof(efrm_rm_table));
+
        EFRM_TRACE("%s: driver deleted", __FUNCTION__);
        return 0;
 }
@@ -108,21 +110,21 @@ int efrm_driver_register_nic(struct efhw
                goto done;
        }
 
-       if (efrm_nic_table.nic_count == EFHW_MAX_NR_DEVS) {
+       if (efrm_nic_tablep->nic_count == EFHW_MAX_NR_DEVS) {
                EFRM_WARN("%s: filled up NIC table size %d", __FUNCTION__,
                          EFHW_MAX_NR_DEVS);
                rc = -E2BIG;
                goto done;
        }
 
-       EFRM_ASSERT(efrm_nic_table.nic[nic_index] == NULL);
-       efrm_nic_table.nic[nic_index] = nic;
+       EFRM_ASSERT(efrm_nic_tablep->nic[nic_index] == NULL);
+       efrm_nic_tablep->nic[nic_index] = nic;
        nic->index = nic_index;
 
-       if (efrm_nic_table.a_nic == NULL)
-               efrm_nic_table.a_nic = nic;
+       if (efrm_nic_tablep->a_nic == NULL)
+               efrm_nic_tablep->a_nic = nic;
 
-       efrm_nic_table.nic_count++;
+       efrm_nic_tablep->nic_count++;
        efrm_driver_unlock(lock_flags);
        return rc;
 
@@ -147,24 +149,24 @@ int efrm_driver_unregister_nic(struct ef
                goto done;
        }
 
-       EFRM_ASSERT(efrm_nic_table.nic[nic_index] == nic);
+       EFRM_ASSERT(efrm_nic_tablep->nic[nic_index] == nic);
 
        nic->index = -1;
-       efrm_nic_table.nic[nic_index] = NULL;
+       efrm_nic_tablep->nic[nic_index] = NULL;
 
-       --efrm_nic_table.nic_count;
+       --efrm_nic_tablep->nic_count;
 
-       if (efrm_nic_table.a_nic == nic) {
-               if (efrm_nic_table.nic_count == 0) {
-                       efrm_nic_table.a_nic = NULL;
+       if (efrm_nic_tablep->a_nic == nic) {
+               if (efrm_nic_tablep->nic_count == 0) {
+                       efrm_nic_tablep->a_nic = NULL;
                } else {
                        for (nic_index = 0; nic_index < EFHW_MAX_NR_DEVS;
                             nic_index++) {
-                               if (efrm_nic_table.nic[nic_index] != NULL)
-                                       efrm_nic_table.a_nic =
-                                           efrm_nic_table.nic[nic_index];
+                               if (efrm_nic_tablep->nic[nic_index] != NULL)
+                                       efrm_nic_tablep->a_nic =
+                                           efrm_nic_tablep->nic[nic_index];
                        }
-                       EFRM_ASSERT(efrm_nic_table.a_nic);
+                       EFRM_ASSERT(efrm_nic_tablep->a_nic);
                }
        }
 
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/driverlink_new.c
--- a/drivers/net/sfc/sfc_resource/driverlink_new.c     Tue Mar 31 11:49:12 
2009 +0100
+++ b/drivers/net/sfc/sfc_resource/driverlink_new.c     Tue Mar 31 11:59:10 
2009 +0100
@@ -66,21 +66,20 @@ init_vi_resource_dimensions(struct vi_re
                            const struct efx_dl_falcon_resources *res)
 {
        rd->evq_timer_min = res->evq_timer_min;
-       rd->evq_timer_max = res->evq_timer_max;
+       rd->evq_timer_lim = res->evq_timer_lim;
        rd->evq_int_min = res->evq_int_min;
-       rd->evq_int_max = res->evq_int_max;
+       rd->evq_int_lim = res->evq_int_lim;
        rd->rxq_min = res->rxq_min;
-       rd->rxq_max = res->rxq_max;
+       rd->rxq_lim = res->rxq_lim;
        rd->txq_min = res->txq_min;
-       rd->txq_max = res->txq_max;
+       rd->txq_lim = res->txq_lim;
        EFRM_TRACE
            ("Using evq_int(%d-%d) evq_timer(%d-%d) RXQ(%d-%d) TXQ(%d-%d)",
-            res->evq_int_min, res->evq_int_max, res->evq_timer_min,
-            res->evq_timer_max, res->rxq_min, res->rxq_max, res->txq_min,
-            res->txq_max);
-}
-
-#if defined(EFX_NOT_UPSTREAM)
+            res->evq_int_min, res->evq_int_lim, res->evq_timer_min,
+            res->evq_timer_lim, res->rxq_min, res->rxq_lim, res->txq_min,
+            res->txq_lim);
+}
+
 /* We have a module parameter that can tell us to only load the char driver
  * for 1 NIC (if there are multiple NICs in the system), and if so which one.
  * This tells us the PCI bus and slot of the NIC to load for, or -1 to just
@@ -98,7 +97,6 @@ MODULE_PARM_DESC(only_NIC,
 MODULE_PARM_DESC(only_NIC,
                 "Initialise sfc_resource driver for one NIC only, "
                 "with specified PCI bus and slot");
-#endif
 
 static int
 efrm_dl_probe(struct efx_dl_device *efrm_dev,
@@ -112,16 +110,14 @@ efrm_dl_probe(struct efx_dl_device *efrm
        struct pci_dev *dev;
        struct efhw_nic *nic;
        unsigned probe_flags = 0;
+       int non_irq_evq;
        int rc;
 
        efrm_dev->priv = NULL;
 
-       efx_dl_for_each_device_info_matching(dev_info, EFX_DL_FALCON_RESOURCES,
-                                            struct efx_dl_falcon_resources,
-                                            hdr, res) {
-               /* break out, leaving res pointing at the falcon resources */
-               break;
-       }
+       efx_dl_search_device_info(dev_info, EFX_DL_FALCON_RESOURCES,
+                                 struct efx_dl_falcon_resources,
+                                 hdr, res);
 
        if (res == NULL) {
                EFRM_ERR("%s: Unable to find falcon driverlink resources",
@@ -132,24 +128,24 @@ efrm_dl_probe(struct efx_dl_device *efrm
        if (res->flags & EFX_DL_FALCON_USE_MSI)
                probe_flags |= NIC_FLAG_TRY_MSI;
 
+#if defined(EFX_NOT_UPSTREAM)
+       if (only_NIC != -1 &&
+           (efrm_dev->pci_dev->bus->number !=
+            ((only_NIC >> 8) & 0xFFFF)
+            || PCI_SLOT(efrm_dev->pci_dev->devfn) !=
+            (only_NIC & 0xFF))) {
+         EFRM_NOTICE("Hiding char device %x:%x",
+                     efrm_dev->pci_dev->bus->number,
+                     PCI_SLOT(efrm_dev->pci_dev->devfn));
+         return -ENODEV;
+       }
+#endif
+       
        dev = efrm_dev->pci_dev;
        if (res->flags & EFX_DL_FALCON_DUAL_FUNC) {
                unsigned vendor = dev->vendor;
                EFRM_ASSERT(dev->bus != NULL);
                dev = NULL;
-
-#if defined(EFX_NOT_UPSTREAM)
-               if (only_NIC != -1 &&
-                   (efrm_dev->pci_dev->bus->number !=
-                    ((only_NIC >> 8) & 0xFFFF)
-                    || PCI_SLOT(efrm_dev->pci_dev->devfn) !=
-                    (only_NIC & 0xFF))) {
-                       EFRM_NOTICE("Hiding char device %x:%x",
-                                   efrm_dev->pci_dev->bus->number,
-                                   PCI_SLOT(efrm_dev->pci_dev->devfn));
-                       return -ENODEV;
-               }
-#endif
 
                while ((dev = pci_get_device(vendor, FALCON_S_DEVID, dev))
                       != NULL) {
@@ -174,10 +170,14 @@ efrm_dl_probe(struct efx_dl_device *efrm
 
        init_vi_resource_dimensions(&res_dim, res);
 
+       EFRM_ASSERT(res_dim.evq_timer_lim > res_dim.evq_timer_min);
+       res_dim.evq_timer_lim--;
+       non_irq_evq = res_dim.evq_timer_lim;
+
        rc = efrm_nic_add(dev, probe_flags, net_dev->dev_addr, &lnic,
                          res->biu_lock,
-                         res->buffer_table_min, res->buffer_table_max,
-                         &res_dim);
+                         res->buffer_table_min, res->buffer_table_lim,
+                         non_irq_evq, &res_dim);
        if (rc != 0)
                return rc;
 
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/sfc_resource/efx_vi_shm.c
--- a/drivers/net/sfc/sfc_resource/efx_vi_shm.c Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/sfc_resource/efx_vi_shm.c Tue Mar 31 11:59:10 2009 +0100
@@ -298,7 +298,7 @@ efx_vi_dma_map_pages(struct efx_vi_state
                /* TODO do we need to get_page() here ? */
 
                dma_addr = pci_map_page
-                   (linux_efhw_nic(efrm_nic_table.nic[efx_state->nic_index])->
+                   
(linux_efhw_nic(efrm_nic_tablep->nic[efx_state->nic_index])->
                     pci_dev, pages[i], 0, PAGE_SIZE, PCI_DMA_TODEVICE);
 
                efrm_buffer_table_set(&dm_state->bt_handle, i, dma_addr,
@@ -399,7 +399,7 @@ efx_vi_dma_unmap_pages(struct efx_vi_sta
 
        for (i = 0; i < dm_state->n_pages; ++i)
                pci_unmap_page(linux_efhw_nic
-                       (efrm_nic_table.nic[efx_state->nic_index])->pci_dev,
+                       (efrm_nic_tablep->nic[efx_state->nic_index])->pci_dev,
                        dm_state->dma_addrs[i], PAGE_SIZE, PCI_DMA_TODEVICE);
 
        kfree(dm_state->dma_addrs);
@@ -547,7 +547,7 @@ efx_vi_hw_resource_get_phys(struct efx_v
 {
        struct efx_vi_state *efx_state = vih;
        int i, ni = efx_state->nic_index;
-       struct linux_efhw_nic *lnic = linux_efhw_nic(efrm_nic_table.nic[ni]);
+       struct linux_efhw_nic *lnic = linux_efhw_nic(efrm_nic_tablep->nic[ni]);
        unsigned long phys = lnic->ctr_ap_pci_addr;
        struct efrm_resource *ep_res = &efx_state->vi_res->rs;
        unsigned ep_mmap_bytes;
@@ -555,11 +555,9 @@ efx_vi_hw_resource_get_phys(struct efx_v
        if (*length < EFX_VI_HW_RESOURCE_MAXSIZE)
                return -EINVAL;
 
-       mdata->version = 0;
-
-       mdata->nic_arch = efrm_nic_table.nic[ni]->devtype.arch;
-       mdata->nic_variant = efrm_nic_table.nic[ni]->devtype.variant;
-       mdata->nic_revision = efrm_nic_table.nic[ni]->devtype.revision;
+       mdata->nic_arch = efrm_nic_tablep->nic[ni]->devtype.arch;
+       mdata->nic_variant = efrm_nic_tablep->nic[ni]->devtype.variant;
+       mdata->nic_revision = efrm_nic_tablep->nic[ni]->devtype.revision;
 
        mdata->evq_order =
            efx_state->vi_res->nic_info[ni].evq_pages.iobuff.order;
@@ -634,9 +632,6 @@ efx_vi_hw_resource_get_phys(struct efx_v
                (unsigned long)efx_state->vi_res->nic_info[ni].
                        dmaq_pages[EFRM_VI_RM_DMA_QUEUE_RX].kva;
 
-       /* NB EFX_VI_HW_RESOURCE_TXBELL not used on Falcon */
-       /* NB EFX_VI_HW_RESOURCE_RXBELL not used on Falcon */
-
        i++;
        hw_res_array[i].type = EFX_VI_HW_RESOURCE_EVQTIMER;
        hw_res_array[i].mem_type = EFX_VI_HW_RESOURCE_PERIPHERAL;
@@ -648,7 +643,7 @@ efx_vi_hw_resource_get_phys(struct efx_v
        /* NB EFX_VI_HW_RESOURCE_EVQPTR not used on Falcon */
 
        i++;
-       switch (efrm_nic_table.nic[ni]->devtype.variant) {
+       switch (efrm_nic_tablep->nic[ni]->devtype.variant) {
        case 'A':
                hw_res_array[i].type = EFX_VI_HW_RESOURCE_EVQRPTR;
                hw_res_array[i].mem_type = EFX_VI_HW_RESOURCE_PERIPHERAL;
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/sfc_resource/eventq.c
--- a/drivers/net/sfc/sfc_resource/eventq.c     Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/sfc_resource/eventq.c     Tue Mar 31 11:59:10 2009 +0100
@@ -170,7 +170,7 @@ efhw_handle_rxdmaq_flushed(struct efhw_n
 efhw_handle_rxdmaq_flushed(struct efhw_nic *nic, struct efhw_ev_handler *h,
                           efhw_event_t *evp)
 {
-       int instance = (int)FALCON_EVENT_RX_FLUSH_Q_ID(evp);
+       unsigned instance = (unsigned)FALCON_EVENT_RX_FLUSH_Q_ID(evp);
        EFHW_TRACE("%s: instance=%d", __FUNCTION__, instance);
 
        if (!h->dmaq_flushed_fn) {
@@ -185,24 +185,28 @@ efhw_handle_wakeup_event(struct efhw_nic
 efhw_handle_wakeup_event(struct efhw_nic *nic, struct efhw_ev_handler *h,
                         efhw_event_t *evp)
 {
+       unsigned instance = (unsigned)FALCON_EVENT_WAKE_EVQ_ID(evp);
+
        if (!h->wakeup_fn) {
                EFHW_WARN("%s: no handler registered", __FUNCTION__);
                return;
        }
 
-       h->wakeup_fn(nic, evp);
+       h->wakeup_fn(nic, instance);
 }
 
 void
 efhw_handle_timeout_event(struct efhw_nic *nic, struct efhw_ev_handler *h,
                          efhw_event_t *evp)
 {
+       unsigned instance = (unsigned)FALCON_EVENT_WAKE_EVQ_ID(evp);
+
        if (!h->timeout_fn) {
                EFHW_WARN("%s: no handler registered", __FUNCTION__);
                return;
        }
 
-       h->timeout_fn(nic, evp);
+       h->timeout_fn(nic, instance);
 }
 
 /**********************************************************************
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/sfc_resource/falcon.c
--- a/drivers/net/sfc/sfc_resource/falcon.c     Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/sfc_resource/falcon.c     Tue Mar 31 11:59:10 2009 +0100
@@ -45,14 +45,11 @@
  *
  *---------------------------------------------------------------------------*/
 
-/* on for debug builds */
-#ifndef NDEBUG
-#  define FALCON_FULL_FILTER_CACHE 1   /* complete SW shadow of filter tbl */
-#  define FALCON_VERIFY_FILTERS    0
-#else /* Also adds duplicate filter check */
-#  define FALCON_FULL_FILTER_CACHE 1   /* keep this on for some security */
-#  define FALCON_VERIFY_FILTERS    0
-#endif
+/* Keep a software copy of the filter table and check for duplicates. */
+#define FALCON_FULL_FILTER_CACHE 1
+
+/* Read filters back from the hardware to detect corruption. */
+#define FALCON_VERIFY_FILTERS    0
 
 /* options */
 #define RX_FILTER_CTL_SRCH_LIMIT_TCP_FULL 8    /* default search limit */
@@ -73,11 +70,7 @@
  *
  *---------------------------------------------------------------------------*/
 
-#ifndef __KERNEL__
-#define _DEBUG_SYM_ extern
-#else
 #define _DEBUG_SYM_ static inline
-#endif
 
  /*----------------------------------------------------------------------------
   *
@@ -208,6 +201,7 @@ _falcon_nic_ipfilter_build(struct efhw_n
                EFHW_ASSERT(!rss_b0);
                break;
        case 'B':
+       case 'C':
                v4 |= scat_b0 << __DW4(SCATTER_EN_1_B0_LBN);
                v4 |= rss_b0 << __DW4(RSS_EN_1_B0_LBN);
                break;
@@ -407,7 +401,7 @@ falcon_dmaq_tx_q_init(struct efhw_nic *n
        uint index, desc_type;
        uint64_t val1, val2, val3;
        ulong offset;
-       efhw_ioaddr_t efhw_kva = EFHW_KVA(nic);
+       volatile char __iomem *efhw_kva = EFHW_KVA(nic);
 
        /* Q attributes */
        int iscsi_hdig_en = ((flags & EFHW_VI_ISCSI_TX_HDIG_EN) != 0);
@@ -471,6 +465,7 @@ falcon_dmaq_tx_q_init(struct efhw_nic *n
 
        switch (nic->devtype.variant) {
        case 'B':
+       case 'C':
                __DW3CHCK(TX_NON_IP_DROP_DIS_B0_LBN,
                          TX_NON_IP_DROP_DIS_B0_WIDTH);
                __DW3CHCK(TX_IP_CHKSM_DIS_B0_LBN, TX_IP_CHKSM_DIS_B0_WIDTH);
@@ -523,7 +518,7 @@ falcon_dmaq_rx_q_init(struct efhw_nic *n
        uint i, desc_type = 1;
        uint64_t val1, val2, val3;
        ulong offset;
-       efhw_ioaddr_t efhw_kva = EFHW_KVA(nic);
+       volatile char __iomem *efhw_kva = EFHW_KVA(nic);
 
        /* Q attributes */
 #if BUG5762_WORKAROUND
@@ -613,7 +608,7 @@ static void falcon_dmaq_tx_q_disable(str
        FALCON_LOCK_DECL;
        uint64_t val1, val2, val3;
        ulong offset;
-       efhw_ioaddr_t efhw_kva = EFHW_KVA(nic);
+       volatile char __iomem *efhw_kva = EFHW_KVA(nic);
 
        /* initialise the TX descriptor queue pointer table */
 
@@ -646,7 +641,7 @@ static void falcon_dmaq_rx_q_disable(str
        FALCON_LOCK_DECL;
        uint64_t val1, val2, val3;
        ulong offset;
-       efhw_ioaddr_t efhw_kva = EFHW_KVA(nic);
+       volatile char __iomem *efhw_kva = EFHW_KVA(nic);
 
        /* initialise the TX descriptor queue pointer table */
        offset = falcon_dma_rx_q_offset(nic, dmaq);
@@ -749,8 +744,8 @@ _falcon_nic_buffer_table_set32(struct ef
 {
        /* programming the half table needs to be done in pairs. */
        uint64_t entry, val, shift;
-       efhw_ioaddr_t efhw_kva = EFHW_KVA(nic);
-       efhw_ioaddr_t offset;
+       volatile char __iomem *efhw_kva = EFHW_KVA(nic);
+       volatile char __iomem *offset;
 
        EFHW_BUILD_ASSERT(BUF_ADR_HBUF_ODD_LBN == BUF_ADR_HBUF_EVEN_LBN + 32);
        EFHW_BUILD_ASSERT(BUF_OWNER_ID_HBUF_ODD_LBN ==
@@ -774,9 +769,9 @@ _falcon_nic_buffer_table_set32(struct ef
        val &= ~(((uint64_t) 0xffffffff) << shift);
        val |= (entry << shift);
 
-       EFHW_TRACE("%s[%x]: " ci_dma_addr_fmt ":%x:%" PRIx64 "->%x = %"
-                  PRIx64, __FUNCTION__, buffer_id, dma_addr, own_id, entry,
-                  (unsigned)(offset - efhw_kva), val);
+       EFHW_TRACE("%s[%x]: %lx:%x:%" PRIx64 "->%x = %"
+                  PRIx64, __FUNCTION__, buffer_id, (unsigned long) dma_addr,
+                  own_id, entry, (unsigned)(offset - efhw_kva), val);
 
        /* Falcon requires that access to this register is serialised */
        falcon_write_q(offset, val);
@@ -811,9 +806,9 @@ _falcon_nic_buffer_table_set64(struct ef
                               dma_addr_t dma_addr, uint bufsz,
                               uint region, int own_id, int buffer_id)
 {
-       efhw_ioaddr_t offset;
+       volatile char __iomem *offset;
        uint64_t entry;
-       efhw_ioaddr_t efhw_kva = EFHW_KVA(nic);
+       volatile char __iomem *efhw_kva = EFHW_KVA(nic);
 
        EFHW_ASSERT(region < FALCON_REGION_NUM);
 
@@ -826,9 +821,9 @@ _falcon_nic_buffer_table_set64(struct ef
        entry = falcon_nic_buffer_table_entry64_mk(dma_addr, bufsz, region,
                                                   own_id);
 
-       EFHW_TRACE("%s[%x]: " ci_dma_addr_fmt
-                  ":bufsz=%x:region=%x:ownid=%x",
-                  __FUNCTION__, buffer_id, dma_addr, bufsz, region, own_id);
+       EFHW_TRACE("%s[%x]: %lx:bufsz=%x:region=%x:ownid=%x",
+                  __FUNCTION__, buffer_id, (unsigned long) dma_addr, bufsz,
+                  region, own_id);
 
        EFHW_TRACE("%s: BUF[%x]:NIC[%x]->%" PRIx64,
                   __FUNCTION__, buffer_id,
@@ -870,7 +865,7 @@ static inline void _falcon_nic_buffer_ta
 static inline void _falcon_nic_buffer_table_commit(struct efhw_nic *nic)
 {
        /* MUST be called holding the FALCON_LOCK */
-       efhw_ioaddr_t efhw_kva = EFHW_KVA(nic);
+       volatile char __iomem *efhw_kva = EFHW_KVA(nic);
        uint64_t cmd;
 
        EFHW_BUILD_ASSERT(BUF_TBL_UPD_REG_KER_OFST == BUF_TBL_UPD_REG_OFST);
@@ -900,9 +895,9 @@ _falcon_nic_buffer_table_clear(struct ef
        uint64_t cmd;
        uint64_t start_id = buffer_id;
        uint64_t end_id = buffer_id + num - 1;
-       efhw_ioaddr_t efhw_kva = EFHW_KVA(nic);
-
-       efhw_ioaddr_t offset = (efhw_kva + BUF_TBL_UPD_REG_OFST);
+       volatile char __iomem *efhw_kva = EFHW_KVA(nic);
+
+       volatile char __iomem *offset = (efhw_kva + BUF_TBL_UPD_REG_OFST);
 
        EFHW_BUILD_ASSERT(BUF_TBL_UPD_REG_KER_OFST == BUF_TBL_UPD_REG_OFST);
 
@@ -962,12 +957,9 @@ static inline void falcon_nic_srm_upd_ev
         * updates */
 
        FALCON_LOCK_DECL;
-       efhw_ioaddr_t efhw_kva = EFHW_KVA(nic);
+       volatile char __iomem *efhw_kva = EFHW_KVA(nic);
 
        EFHW_BUILD_ASSERT(SRM_UPD_EVQ_REG_OFST == SRM_UPD_EVQ_REG_KER_OFST);
-
-       EFHW_ASSERT((evq == FALCON_EVQ_KERNEL0) || (evq == FALCON_EVQ_CHAR) ||
-                   (evq == FALCON_EVQ_NONIRQ));
 
        __DWCHCK(SRM_UPD_EVQ_ID_LBN, SRM_UPD_EVQ_ID_WIDTH);
        __RANGECHCK(evq, SRM_UPD_EVQ_ID_WIDTH);
@@ -991,7 +983,7 @@ falcon_nic_evq_ptr_tbl(struct efhw_nic *
        FALCON_LOCK_DECL;
        uint i, val;
        ulong offset;
-       efhw_ioaddr_t efhw_kva = EFHW_KVA(nic);
+       volatile char __iomem *efhw_kva = EFHW_KVA(nic);
 
        /* size must be one of the various options, otherwise we assert */
        for (i = 0; i < N_EVENTQ_SIZES; i++) {
@@ -1045,7 +1037,7 @@ falcon_nic_evq_ack(struct efhw_nic *nic,
 {
        uint val;
        ulong offset;
-       efhw_ioaddr_t efhw_kva = EFHW_KVA(nic);
+       volatile char __iomem *efhw_kva = EFHW_KVA(nic);
 
        EFHW_BUILD_ASSERT(FALCON_EVQ_CHAR == 4);
 
@@ -1078,38 +1070,13 @@ falcon_nic_evq_ack(struct efhw_nic *nic,
        mmiowb();
 }
 
-/*----------------------------------------------------------------------------
- *
- * Helper for evq mapping
- *
- * idx = 0 && char   => hw eventq[4]
- * idx = 0 && net    => hw eventq[0]
- *   0 < idx < 5     => hw eventq[idx]  (5 is non-interrupting)
- *
- *
- *---------------------------------------------------------------------------*/
-
-int falcon_idx_to_evq(struct efhw_nic *nic, uint idx)
-{
-       EFHW_BUILD_ASSERT(FALCON_EVQ_CHAR == 4);
-       EFHW_ASSERT(idx <= FALCON_EVQ_NONIRQ);
-       return (idx > 0) ? idx : FALCON_EVQ_CHAR;
-}
-
-static inline int falcon_evq_is_interrupting(struct efhw_nic *nic, uint idx)
-{
-       EFHW_BUILD_ASSERT(FALCON_EVQ_CHAR == 4);
-       EFHW_ASSERT(idx <= FALCON_EVQ_NONIRQ);
-
-       /* only the first CHAR driver event queue is interrupting */
-       return (idx == FALCON_EVQ_CHAR);
-}
+/*---------------------------------------------------------------------------*/
 
 static inline void
 falcon_drv_ev(struct efhw_nic *nic, uint64_t data, uint qid)
 {
        FALCON_LOCK_DECL;
-       efhw_ioaddr_t efhw_kva = EFHW_KVA(nic);
+       volatile char __iomem *efhw_kva = EFHW_KVA(nic);
 
        /* send an event from one driver to the other */
        EFHW_BUILD_ASSERT(DRV_EV_REG_KER_OFST == DRV_EV_REG_OFST);
@@ -1133,7 +1100,7 @@ falcon_timer_cmd(struct efhw_nic *nic,
        FALCON_LOCK_DECL;
        uint val;
        ulong offset;
-       efhw_ioaddr_t efhw_kva = EFHW_KVA(nic);
+       volatile char __iomem *efhw_kva = EFHW_KVA(nic);
 
        EFHW_BUILD_ASSERT(TIMER_VAL_LBN == 0);
 
@@ -1184,7 +1151,7 @@ void falcon_nic_pace(struct efhw_nic *ni
           Pacing only available on the virtual interfaces
         */
        FALCON_LOCK_DECL;
-       efhw_ioaddr_t efhw_kva = EFHW_KVA(nic);
+       volatile char __iomem *efhw_kva = EFHW_KVA(nic);
        ulong offset;
 
        if (pace > 20)
@@ -1200,6 +1167,7 @@ void falcon_nic_pace(struct efhw_nic *ni
                offset += (dmaq - TX_PACE_TBL_FIRST_QUEUE_A1) * 16;
                break;
        case 'B':
+       case 'C':
                /* Would be nice to assert this, but as dmaq is unsigned and
                 * TX_PACE_TBL_FIRST_QUEUE_B0 is 0, it makes no sense
                 * EFHW_ASSERT(dmaq >= TX_PACE_TBL_FIRST_QUEUE_B0);
@@ -1232,8 +1200,8 @@ static void falcon_nic_handle_fatal_int(
 static void falcon_nic_handle_fatal_int(struct efhw_nic *nic)
 {
        FALCON_LOCK_DECL;
-       efhw_ioaddr_t offset;
-       efhw_ioaddr_t efhw_kva = EFHW_KVA(nic);
+       volatile char __iomem *offset;
+       volatile char __iomem *efhw_kva = EFHW_KVA(nic);
        uint64_t val;
 
        offset = (efhw_kva + FATAL_INTR_REG_OFST);
@@ -1264,8 +1232,8 @@ static void falcon_nic_interrupt_hw_enab
 {
        FALCON_LOCK_DECL;
        uint val;
-       efhw_ioaddr_t offset;
-       efhw_ioaddr_t efhw_kva = EFHW_KVA(nic);
+       volatile char __iomem *offset;
+       volatile char __iomem *efhw_kva = EFHW_KVA(nic);
 
        EFHW_BUILD_ASSERT(DRV_INT_EN_CHAR_WIDTH == 1);
 
@@ -1288,8 +1256,8 @@ static void falcon_nic_interrupt_hw_disa
 static void falcon_nic_interrupt_hw_disable(struct efhw_nic *nic)
 {
        FALCON_LOCK_DECL;
-       efhw_ioaddr_t offset;
-       efhw_ioaddr_t efhw_kva = EFHW_KVA(nic);
+       volatile char __iomem *offset;
+       volatile char __iomem *efhw_kva = EFHW_KVA(nic);
 
        EFHW_BUILD_ASSERT(SRAM_PERR_INT_KER_WIDTH == 1);
        EFHW_BUILD_ASSERT(DRV_INT_EN_KER_LBN == 0);
@@ -1312,13 +1280,12 @@ static void falcon_nic_interrupt_hw_disa
        FALCON_LOCK_UNLOCK(nic);
 }
 
-#ifndef __ci_ul_driver__
 
 static void falcon_nic_irq_addr_set(struct efhw_nic *nic, dma_addr_t dma_addr)
 {
        FALCON_LOCK_DECL;
-       efhw_ioaddr_t offset;
-       efhw_ioaddr_t efhw_kva = EFHW_KVA(nic);
+       volatile char __iomem *offset;
+       volatile char __iomem *efhw_kva = EFHW_KVA(nic);
 
        offset = (efhw_kva + INT_ADR_REG_CHAR_OFST);
 
@@ -1332,7 +1299,6 @@ static void falcon_nic_irq_addr_set(stru
        FALCON_LOCK_UNLOCK(nic);
 }
 
-#endif
 
 
 /*--------------------------------------------------------------------
@@ -1345,7 +1311,7 @@ falcon_nic_set_rx_usr_buf_size(struct ef
 falcon_nic_set_rx_usr_buf_size(struct efhw_nic *nic, int usr_buf_bytes)
 {
        FALCON_LOCK_DECL;
-       efhw_ioaddr_t efhw_kva = EFHW_KVA(nic);
+       volatile char __iomem *efhw_kva = EFHW_KVA(nic);
        uint64_t val, val2, usr_buf_size = usr_buf_bytes / 32;
        int rubs_lbn, rubs_width, roec_lbn;
 
@@ -1361,6 +1327,7 @@ falcon_nic_set_rx_usr_buf_size(struct ef
                roec_lbn = RX_OWNERR_CTL_A1_LBN;
                break;
        case 'B':
+       case 'C':
                rubs_lbn = RX_USR_BUF_SIZE_B0_LBN;
                rubs_width = RX_USR_BUF_SIZE_B0_WIDTH;
                roec_lbn = RX_OWNERR_CTL_B0_LBN;
@@ -1392,7 +1359,7 @@ falcon_nic_rx_filter_ctl_get(struct efhw
                             uint32_t *tcp_wild,
                             uint32_t *udp_full, uint32_t *udp_wild)
 {
-       efhw_ioaddr_t efhw_kva = EFHW_KVA(nic);
+       volatile char __iomem *efhw_kva = EFHW_KVA(nic);
        FALCON_LOCK_DECL;
        uint64_t val;
 
@@ -1420,7 +1387,7 @@ falcon_nic_rx_filter_ctl_set(struct efhw
                             uint32_t udp_full, uint32_t udp_wild)
 {
        uint64_t val, val2;
-       efhw_ioaddr_t efhw_kva = EFHW_KVA(nic);
+       volatile char __iomem *efhw_kva = EFHW_KVA(nic);
        FALCON_LOCK_DECL;
 
        EFHW_ASSERT(tcp_full < nic->filter_tbl_size);
@@ -1480,7 +1447,7 @@ _DEBUG_SYM_ void falcon_nic_tx_cfg(struc
 _DEBUG_SYM_ void falcon_nic_tx_cfg(struct efhw_nic *nic, int unlocked)
 {
        FALCON_LOCK_DECL;
-       efhw_ioaddr_t efhw_kva = EFHW_KVA(nic);
+       volatile char __iomem *efhw_kva = EFHW_KVA(nic);
        uint64_t val1, val2;
 
        EFHW_BUILD_ASSERT(TX_CFG_REG_OFST == TX_CFG_REG_KER_OFST);
@@ -1516,7 +1483,7 @@ static void falcon_nic_pace_cfg(struct e
 static void falcon_nic_pace_cfg(struct efhw_nic *nic)
 {
        FALCON_LOCK_DECL;
-       efhw_ioaddr_t efhw_kva = EFHW_KVA(nic);
+       volatile char __iomem *efhw_kva = EFHW_KVA(nic);
        unsigned offset = 0;
        uint64_t val;
 
@@ -1527,6 +1494,7 @@ static void falcon_nic_pace_cfg(struct e
        switch (nic->devtype.variant) {
        case 'A':  offset = TX_PACE_REG_A1_OFST;  break;
        case 'B':  offset = TX_PACE_REG_B0_OFST;  break;
+       case 'C':  offset = TX_PACE_REG_B0_OFST;  break;
        default:   EFHW_ASSERT(0);                break;
        }
        falcon_write_qq(efhw_kva + offset, val, 0);
@@ -1697,15 +1665,6 @@ host_ipfilter_cache_check_not(uint nic, 
  *
  *---------------------------------------------------------------------------*/
 
-#ifdef __ci_ul_driver__
-
-static int falcon_nic_init_irq_channel(struct efhw_nic *nic, int enable)
-{
-       EFHW_ERR("%s: not implemented for ul driver", __FUNCTION__);
-       return -EOPNOTSUPP;
-}
-
-#else
 
 static int falcon_nic_init_irq_channel(struct efhw_nic *nic, int enable)
 {
@@ -1727,14 +1686,13 @@ static int falcon_nic_init_irq_channel(s
                falcon_nic_irq_addr_set(nic, 0);
        }
 
-       EFHW_TRACE("%s: " ci_dma_addr_fmt " %sable", __FUNCTION__,
-                  efhw_iopage_dma_addr(&nic->irq_iobuff), enable ?
-                       "en" : "dis");
+       EFHW_TRACE("%s: %lx %sable", __FUNCTION__,
+                  (unsigned long) efhw_iopage_dma_addr(&nic->irq_iobuff),
+                  enable ? "en" : "dis");
 
        return 0;
 }
 
-#endif
 
 static void falcon_nic_close_hardware(struct efhw_nic *nic)
 {
@@ -1747,14 +1705,10 @@ static void falcon_nic_close_hardware(st
        EFHW_NOTICE("%s:", __FUNCTION__);
 }
 
-#ifdef __ci_ul_driver__
-extern
-#else
 static
-#endif
 int falcon_nic_get_mac_config(struct efhw_nic *nic)
 {
-       efhw_ioaddr_t efhw_kva = nic->bar_ioaddr;
+       volatile char __iomem *efhw_kva = nic->bar_ioaddr;
        int is_mac_type_1g;
        uint32_t strap, altera;
        uint64_t rx_cfg, r;
@@ -1811,6 +1765,23 @@ int falcon_nic_get_mac_config(struct efh
                }
 #endif
                break;
+       case 'C':
+               /* Treat like B0 for now, but without the RX FIFO size check
+                * (don't need it, and RX_CFG_REG will likely change soon
+                * anyway).
+                */
+               is_mac_type_1g = (0 != (strap & 2));
+#if FALCON_MAC_SET_TYPE_BY_SPEED
+               /* Check the selected strap pins against the MAC speed -
+                * and adjust if necessary.
+                */
+               {
+                       int speed;
+                       speed = readl(efhw_kva + MAC0_CTRL_REG_OFST) & 0x3;
+                       is_mac_type_1g = (speed <= 2);
+               }
+#endif
+               break;
        default:
                EFHW_ASSERT(0);
                is_mac_type_1g = 0;
@@ -1834,7 +1805,7 @@ static int
 static int
 falcon_nic_init_hardware(struct efhw_nic *nic,
                         struct efhw_ev_handler *ev_handlers,
-                        const uint8_t *mac_addr)
+                        const uint8_t *mac_addr, int non_irq_evq)
 {
        int rc;
 
@@ -1868,7 +1839,7 @@ falcon_nic_init_hardware(struct efhw_nic
           IFDEF FALCON's can be removed from
           nic.c:efhw_nic_allocate_common_hardware_resources()
         */
-       nic->irq_unit[0] = INT_EN_REG_CHAR_OFST;
+       nic->irq_unit = INT_EN_REG_CHAR_OFST;
 
        /*****************************************************************
         * The rest of this function deals with initialization of the NICs
@@ -1877,7 +1848,7 @@ falcon_nic_init_hardware(struct efhw_nic
 
        /* char driver grabs SRM events onto the non interrupting
         * event queue */
-       falcon_nic_srm_upd_evq(nic, FALCON_EVQ_NONIRQ);
+       falcon_nic_srm_upd_evq(nic, non_irq_evq);
 
        /* RXDP tweaks */
 
@@ -1906,19 +1877,19 @@ falcon_nic_init_hardware(struct efhw_nic
                                     RX_FILTER_CTL_SRCH_LIMIT_UDP_WILD);
 
        if (!(nic->flags & NIC_FLAG_NO_INTERRUPT)) {
-               rc = efhw_keventq_ctor(nic, FALCON_EVQ_CHAR, &nic->evq[0],
-                                      ev_handlers);
+               rc = efhw_keventq_ctor(nic, FALCON_EVQ_CHAR,
+                                      &nic->interrupting_evq, ev_handlers);
                if (rc < 0) {
                        EFHW_ERR("%s: efhw_keventq_ctor() failed (%d) evq=%d",
                                 __FUNCTION__, rc, FALCON_EVQ_CHAR);
                        return rc;
                }
        }
-       rc = efhw_keventq_ctor(nic, FALCON_EVQ_NONIRQ,
-                              &nic->evq[FALCON_EVQ_NONIRQ], NULL);
+       rc = efhw_keventq_ctor(nic, non_irq_evq,
+                              &nic->non_interrupting_evq, NULL);
        if (rc < 0) {
                EFHW_ERR("%s: efhw_keventq_ctor() failed (%d) evq=%d",
-                        __FUNCTION__, rc, FALCON_EVQ_NONIRQ);
+                        __FUNCTION__, rc, non_irq_evq);
                return rc;
        }
 
@@ -1938,11 +1909,12 @@ falcon_nic_init_hardware(struct efhw_nic
  *--------------------------------------------------------------------*/
 
 static void
-falcon_nic_interrupt_enable(struct efhw_nic *nic, unsigned idx)
-{
-       int evq;
-
-       if (idx || (nic->flags & NIC_FLAG_NO_INTERRUPT))
+falcon_nic_interrupt_enable(struct efhw_nic *nic)
+{
+       struct efhw_keventq *q;
+       unsigned rdptr;
+
+       if (nic->flags & NIC_FLAG_NO_INTERRUPT)
                return;
 
        /* Enable driver interrupts */
@@ -1950,32 +1922,26 @@ falcon_nic_interrupt_enable(struct efhw_
        falcon_nic_interrupt_hw_enable(nic);
 
        /* An interrupting eventq must start of day ack its read pointer */
-       evq = falcon_idx_to_evq(nic, idx);
-
-       if (falcon_evq_is_interrupting(nic, evq)) {
-               struct efhw_keventq *q = &nic->evq[idx];
-               unsigned rdptr =
-                   EFHW_EVENT_OFFSET(q, q, 1) / sizeof(efhw_event_t);
-               falcon_nic_evq_ack(nic, evq, rdptr, false);
-               EFHW_NOTICE("%s: ACK evq[%d]:%x", __FUNCTION__, evq, rdptr);
-       }
-}
-
-static void falcon_nic_interrupt_disable(struct efhw_nic *nic, uint idx)
+       q = &nic->interrupting_evq;
+       rdptr = EFHW_EVENT_OFFSET(q, q, 1) / sizeof(efhw_event_t);
+       falcon_nic_evq_ack(nic, FALCON_EVQ_CHAR, rdptr, false);
+       EFHW_NOTICE("%s: ACK evq[%d]:%x", __FUNCTION__,
+                   FALCON_EVQ_CHAR, rdptr);
+}
+
+static void falcon_nic_interrupt_disable(struct efhw_nic *nic)
 {
        /* NB. No need to check for NIC_FLAG_NO_INTERRUPT, as
         ** falcon_nic_interrupt_hw_disable() will do it. */
-       if (idx)
-               return;
        falcon_nic_interrupt_hw_disable(nic);
 }
 
 static void
-falcon_nic_set_interrupt_moderation(struct efhw_nic *nic, uint idx,
+falcon_nic_set_interrupt_moderation(struct efhw_nic *nic,
                                    uint32_t val)
 {
-       falcon_timer_cmd(nic, falcon_idx_to_evq(nic, idx),
-                        TIMER_MODE_INT_HLDOFF, val / 5);
+       falcon_timer_cmd(nic, FALCON_EVQ_CHAR, TIMER_MODE_INT_HLDOFF,
+                        val / 5);
 }
 
 static inline void legacy_irq_ack(struct efhw_nic *nic)
@@ -1992,7 +1958,7 @@ static inline void legacy_irq_ack(struct
 
 static int falcon_nic_interrupt(struct efhw_nic *nic)
 {
-       volatile uint32_t *syserr_ptr =
+       uint32_t *syserr_ptr =
            (uint32_t *) efhw_iopage_ptr(&nic->irq_iobuff);
        int handled = 0;
        int done_ack = 0;
@@ -2102,7 +2068,7 @@ void
 void
 falcon_nic_ipfilter_ctor(struct efhw_nic *nic)
 {
-       if (nic->devtype.variant == 'B' && nic->fpga_version)
+       if (nic->devtype.variant >= 'B' && nic->fpga_version)
                nic->filter_tbl_size = 8 * 1024;
        else
                nic->filter_tbl_size = 16 * 1024;
@@ -2276,10 +2242,6 @@ static void falcon_nic_ipfilter_clear(st
 
 static inline void falcon_nic_buffer_table_lazy_commit(struct efhw_nic *nic)
 {
-#if defined(__ci_ul_driver__)
-       if (!(nic->options & NIC_OPT_EFTEST))
-               return;
-#endif
 
        /* Do nothing if operating in synchronous mode. */
        if (!nic->irq_handler)
@@ -2291,10 +2253,6 @@ static inline void falcon_nic_buffer_tab
        FALCON_LOCK_DECL;
        int count = 0, rc = 0;
 
-#if defined(__ci_ul_driver__)
-       if (!(nic->options & NIC_OPT_EFTEST))
-               return;
-#endif
 
        /* We can be called here early days */
        if (!nic->irq_handler)
@@ -2316,9 +2274,9 @@ static inline void falcon_nic_buffer_tab
                 * upcalls into the core driver */
                struct efhw_ev_handler handler;
                memset(&handler, 0, sizeof(handler));
-               nic->evq[FALCON_EVQ_NONIRQ].ev_handlers = &handler;
-               rc = efhw_keventq_poll(nic, &nic->evq[FALCON_EVQ_NONIRQ]);
-               nic->evq[FALCON_EVQ_NONIRQ].ev_handlers = NULL;
+               nic->non_interrupting_evq.ev_handlers = &handler;
+               rc = efhw_keventq_poll(nic, &nic->non_interrupting_evq);
+               nic->non_interrupting_evq.ev_handlers = NULL;
 
                if (rc < 0) {
                        EFHW_ERR("%s: poll ERROR (%d:%d) ***** ",
@@ -2353,10 +2311,6 @@ void falcon_nic_buffer_table_confirm(str
           an event or DMA queue */
        FALCON_LOCK_DECL;
 
-#if defined(__ci_ul_driver__)
-       if (!(nic->options & NIC_OPT_EFTEST))
-               return;
-#endif
 
        /* Do nothing if operating in synchronous mode. */
        if (!nic->irq_handler)
@@ -2463,7 +2417,7 @@ falcon_check_for_bogus_tx_dma_wptr(struc
        FALCON_LOCK_DECL;
        uint64_t val_low64, val_high64;
        uint64_t size, hwptr, swptr, val;
-       efhw_ioaddr_t efhw_kva = EFHW_KVA(nic);
+       volatile char __iomem *efhw_kva = EFHW_KVA(nic);
        ulong offset = falcon_dma_tx_q_offset(nic, dmaq);
 
        /* Falcon requires 128 bit atomic access for this register */
@@ -2500,7 +2454,7 @@ void falcon_clobber_tx_dma_ptrs(struct e
 {
        FALCON_LOCK_DECL;
        uint64_t val_low64, val_high64;
-       efhw_ioaddr_t efhw_kva = EFHW_KVA(nic);
+       volatile char __iomem *efhw_kva = EFHW_KVA(nic);
        ulong offset = falcon_dma_tx_q_offset(nic, dmaq);
 
        EFHW_WARN("Recovering stuck TXQ[%d]", dmaq);
@@ -2523,7 +2477,7 @@ __falcon_really_flush_tx_dma_channel(str
 __falcon_really_flush_tx_dma_channel(struct efhw_nic *nic, uint dmaq)
 {
        FALCON_LOCK_DECL;
-       efhw_ioaddr_t efhw_kva = EFHW_KVA(nic);
+       volatile char __iomem *efhw_kva = EFHW_KVA(nic);
        uint val;
 
        EFHW_BUILD_ASSERT(TX_FLUSH_DESCQ_REG_KER_OFST ==
@@ -2557,7 +2511,7 @@ __falcon_is_tx_dma_channel_flushed(struc
        FALCON_LOCK_DECL;
        uint64_t val_low64, val_high64;
        uint64_t enable, flush_pending;
-       efhw_ioaddr_t efhw_kva = EFHW_KVA(nic);
+       volatile char __iomem *efhw_kva = EFHW_KVA(nic);
        ulong offset = falcon_dma_tx_q_offset(nic, dmaq);
 
        /* Falcon requires 128 bit atomic access for this register */
@@ -2606,7 +2560,7 @@ __falcon_really_flush_rx_dma_channel(str
 __falcon_really_flush_rx_dma_channel(struct efhw_nic *nic, uint dmaq)
 {
        FALCON_LOCK_DECL;
-       efhw_ioaddr_t efhw_kva = EFHW_KVA(nic);
+       volatile char __iomem *efhw_kva = EFHW_KVA(nic);
        uint val;
 
        EFHW_BUILD_ASSERT(RX_FLUSH_DESCQ_REG_KER_OFST ==
@@ -2634,7 +2588,7 @@ __falcon_is_rx_dma_channel_flushed(struc
 {
        FALCON_LOCK_DECL;
        uint64_t val;
-       efhw_ioaddr_t efhw_kva = EFHW_KVA(nic);
+       volatile char __iomem *efhw_kva = EFHW_KVA(nic);
        ulong offset = falcon_dma_rx_q_offset(nic, dmaq);
 
        /* Falcon requires 128 bit atomic access for this register */
@@ -2678,17 +2632,13 @@ falcon_handle_char_event(struct efhw_nic
        case TX_DESCQ_FLS_DONE_EV_DECODE:
                EFHW_TRACE("TX[%d] flushed",
                           (int)FALCON_EVENT_TX_FLUSH_Q_ID(ev));
-#if !defined(__ci_ul_driver__)
                efhw_handle_txdmaq_flushed(nic, h, ev);
-#endif
                break;
 
        case RX_DESCQ_FLS_DONE_EV_DECODE:
                EFHW_TRACE("RX[%d] flushed",
                           (int)FALCON_EVENT_TX_FLUSH_Q_ID(ev));
-#if !defined(__ci_ul_driver__)
                efhw_handle_rxdmaq_flushed(nic, h, ev);
-#endif
                break;
 
        case SRM_UPD_DONE_EV_DECODE:
@@ -2698,16 +2648,16 @@ falcon_handle_char_event(struct efhw_nic
                break;
 
        case EVQ_INIT_DONE_EV_DECODE:
-               EFHW_TRACE("EVQ INIT");
+               EFHW_TRACE("%sEVQ INIT", "");
                break;
 
        case WAKE_UP_EV_DECODE:
-               EFHW_TRACE("WAKE UP");
+               EFHW_TRACE("%sWAKE UP", "");
                efhw_handle_wakeup_event(nic, h, ev);
                break;
 
        case TIMER_EV_DECODE:
-               EFHW_TRACE("TIMER");
+               EFHW_TRACE("%sTIMER", "");
                efhw_handle_timeout_event(nic, h, ev);
                break;
 
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/sfc_resource/falcon_mac.c
--- a/drivers/net/sfc/sfc_resource/falcon_mac.c Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/sfc_resource/falcon_mac.c Tue Mar 31 11:59:10 2009 +0100
@@ -65,7 +65,7 @@
 /*! Get MAC current address - i.e not necessarily the one in the EEPROM */
 static inline void mentormac_get_mac_addr(struct efhw_nic *nic)
 {
-       efhw_ioaddr_t mac_kva;
+       volatile char __iomem *mac_kva;
        uint val1, val2;
 
        MENTOR_MAC_ASSERT_VALID();
@@ -131,7 +131,7 @@ static inline void GDACT10mac_get_mac_ad
 static inline void GDACT10mac_get_mac_addr(struct efhw_nic *nic)
 {
        uint val1, val2;
-       efhw_ioaddr_t efhw_kva = EFHW_KVA(nic);
+       volatile char __iomem *efhw_kva = EFHW_KVA(nic);
        FALCON_LOCK_DECL;
 
        GDACT10_MAC_ASSERT_VALID();
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/filter_resource.c
--- a/drivers/net/sfc/sfc_resource/filter_resource.c    Tue Mar 31 11:49:12 
2009 +0100
+++ b/drivers/net/sfc/sfc_resource/filter_resource.c    Tue Mar 31 11:59:10 
2009 +0100
@@ -234,7 +234,7 @@ __efrm_filter_resource_set(struct filter
        EFRM_ASSERT(frs);
        EFRM_ASSERT(frs->pt);
 
-       if (efrm_nic_table.a_nic->devtype.variant >= 'B') {
+       if (efrm_nic_tablep->a_nic->devtype.variant >= 'B') {
                /* Scatter setting must match the setting for
                 * the corresponding RX queue */
                if (!(frs->pt->flags & EFHW_VI_JUMBO_EN))
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/iobufset_resource.c
--- a/drivers/net/sfc/sfc_resource/iobufset_resource.c  Tue Mar 31 11:49:12 
2009 +0100
+++ b/drivers/net/sfc/sfc_resource/iobufset_resource.c  Tue Mar 31 11:59:10 
2009 +0100
@@ -58,7 +58,7 @@ static inline size_t iobsrs_size(int no_
 static inline size_t iobsrs_size(int no_pages)
 {
        return offsetof(struct iobufset_resource, bufs) +
-           no_pages * sizeof(efhw_iopage_t);
+           no_pages * sizeof(struct efhw_iopage);
 }
 
 void efrm_iobufset_resource_free(struct iobufset_resource *rs)
@@ -74,21 +74,21 @@ void efrm_iobufset_resource_free(struct 
                efrm_buffer_table_free(&rs->buf_tbl_alloc);
 
        /* see comment on call to efhw_iopage_alloc in the alloc routine above
-          for discussion on use of efrm_nic_table.a_nic here */
-       EFRM_ASSERT(efrm_nic_table.a_nic);
+          for discussion on use of efrm_nic_tablep->a_nic here */
+       EFRM_ASSERT(efrm_nic_tablep->a_nic);
        if (rs->order == 0) {
                for (i = 0; i < rs->n_bufs; ++i)
-                       efhw_iopage_free(efrm_nic_table.a_nic, &rs->bufs[i]);
+                       efhw_iopage_free(efrm_nic_tablep->a_nic, &rs->bufs[i]);
        } else {
                /* it is important that this is executed in increasing page
                 * order because some implementations of
                 * efhw_iopages_init_from_iopage() assume this */
                for (i = 0; i < rs->n_bufs;
                     i += rs->pages_per_contiguous_chunk) {
-                       efhw_iopages_t iopages;
+                       struct efhw_iopages iopages;
                        efhw_iopages_init_from_iopage(&iopages, &rs->bufs[i],
                                                    rs->order);
-                       efhw_iopages_free(efrm_nic_table.a_nic, &iopages);
+                       efhw_iopages_free(efrm_nic_tablep->a_nic, &iopages);
                }
        }
 
@@ -126,7 +126,7 @@ efrm_iobufset_resource_alloc(int32_t n_p
        EFRM_RESOURCE_ASSERT_VALID(&vi_evq->rs, 0);
        EFRM_ASSERT(EFRM_RESOURCE_TYPE(vi_evq->rs.rs_handle) ==
                    EFRM_RESOURCE_VI);
-       EFRM_ASSERT(efrm_nic_table.a_nic);
+       EFRM_ASSERT(efrm_nic_tablep->a_nic);
 
        /* allocate the resource data structure. */
        object_size = iobsrs_size(n_pages);
@@ -186,11 +186,11 @@ efrm_iobufset_resource_alloc(int32_t n_p
                        /* due to bug2426 we have to specifiy a NIC when
                         * allocating a DMAable page, which is a bit messy.
                         * For now we assume that if the page is suitable
-                        * (e.g. DMAable) by one nic (efrm_nic_table.a_nic),
+                        * (e.g. DMAable) by one nic (efrm_nic_tablep->a_nic),
                         * it is suitable for all NICs.
                         * XXX I bet that breaks in Solaris.
                         */
-                       rc = efhw_iopage_alloc(efrm_nic_table.a_nic,
+                       rc = efhw_iopage_alloc(efrm_nic_tablep->a_nic,
                                             &iobrs->bufs[i]);
                        if (rc < 0) {
                                EFRM_ERR("%s: failed (rc %d) to allocate "
@@ -199,7 +199,7 @@ efrm_iobufset_resource_alloc(int32_t n_p
                        }
                }
        } else {
-               efhw_iopages_t iopages;
+               struct efhw_iopages iopages;
                unsigned j;
 
                /* make sure iobufs are in a known state in case we don't
@@ -209,7 +209,7 @@ efrm_iobufset_resource_alloc(int32_t n_p
 
                for (i = 0; i < iobrs->n_bufs;
                     i += iobrs->pages_per_contiguous_chunk) {
-                       rc = efhw_iopages_alloc(efrm_nic_table.a_nic,
+                       rc = efhw_iopages_alloc(efrm_nic_tablep->a_nic,
                                                &iopages, iobrs->order);
                        if (rc < 0) {
                                EFRM_ERR("%s: failed (rc %d) to allocate "
@@ -277,16 +277,16 @@ fail5:
        i = iobrs->n_bufs;
 fail4:
        /* see comment on call to efhw_iopage_alloc above for a discussion
-        * on use of efrm_nic_table.a_nic here */
+        * on use of efrm_nic_tablep->a_nic here */
        if (iobrs->order == 0) {
                while (i--) {
-                       efhw_iopage_t *page = &iobrs->bufs[i];
-                       efhw_iopage_free(efrm_nic_table.a_nic, page);
+                       struct efhw_iopage *page = &iobrs->bufs[i];
+                       efhw_iopage_free(efrm_nic_tablep->a_nic, page);
                }
        } else {
                unsigned int j;
                for (j = 0; j < i; j += iobrs->pages_per_contiguous_chunk) {
-                       efhw_iopages_t iopages;
+                       struct efhw_iopages iopages;
 
                        EFRM_ASSERT(j % iobrs->pages_per_contiguous_chunk
                                    == 0);
@@ -296,7 +296,7 @@ fail4:
                        efhw_iopages_init_from_iopage(&iopages,
                                                      &iobrs->bufs[j],
                                                      iobrs->order);
-                       efhw_iopages_free(efrm_nic_table.a_nic, &iopages);
+                       efhw_iopages_free(efrm_nic_tablep->a_nic, &iopages);
                }
        }
        efrm_vi_resource_release(iobrs->evq);
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/sfc_resource/iopage.c
--- a/drivers/net/sfc/sfc_resource/iopage.c     Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/sfc_resource/iopage.c     Tue Mar 31 11:59:10 2009 +0100
@@ -34,7 +34,7 @@
 #include "kernel_compat.h"
 #include <ci/efhw/common_sysdep.h> /* for dma_addr_t */
 
-int efhw_iopage_alloc(struct efhw_nic *nic, efhw_iopage_t *p)
+int efhw_iopage_alloc(struct efhw_nic *nic, struct efhw_iopage *p)
 {
        struct linux_efhw_nic *lnic = linux_efhw_nic(nic);
        dma_addr_t handle;
@@ -55,7 +55,7 @@ int efhw_iopage_alloc(struct efhw_nic *n
        return 0;
 }
 
-void efhw_iopage_free(struct efhw_nic *nic, efhw_iopage_t *p)
+void efhw_iopage_free(struct efhw_nic *nic, struct efhw_iopage *p)
 {
        struct linux_efhw_nic *lnic = linux_efhw_nic(nic);
        EFHW_ASSERT(efhw_page_is_valid(&p->p));
@@ -64,7 +64,9 @@ void efhw_iopage_free(struct efhw_nic *n
                                 efhw_iopage_ptr(p), p->dma_addr);
 }
 
-int efhw_iopages_alloc(struct efhw_nic *nic, efhw_iopages_t *p, unsigned order)
+int
+efhw_iopages_alloc(struct efhw_nic *nic, struct efhw_iopages *p,
+                  unsigned order)
 {
        unsigned bytes = 1u << (order + PAGE_SHIFT);
        struct linux_efhw_nic *lnic = linux_efhw_nic(nic);
@@ -91,7 +93,7 @@ int efhw_iopages_alloc(struct efhw_nic *
        return 0;
 }
 
-void efhw_iopages_free(struct efhw_nic *nic, efhw_iopages_t *p)
+void efhw_iopages_free(struct efhw_nic *nic, struct efhw_iopages *p)
 {
        unsigned bytes = 1u << (p->order + PAGE_SHIFT);
        struct linux_efhw_nic *lnic = linux_efhw_nic(nic);
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/kernel_compat.c
--- a/drivers/net/sfc/sfc_resource/kernel_compat.c      Tue Mar 31 11:49:12 
2009 +0100
+++ b/drivers/net/sfc/sfc_resource/kernel_compat.c      Tue Mar 31 11:59:10 
2009 +0100
@@ -48,22 +48,6 @@
 #define EFRM_IOPAGE_COUNTS_ENABLED 0
 
 
-
-/* I admit that it's a bit ugly going straight to the field, but it
- * seems easiest given that get_page followed by put_page on a page
- * with PG_reserved set will increment the ref count on 2.6.14 and
- * below, but not 2.6.15.  Also, RedHat have hidden put_page_testzero
- * in a header file which produces warnings when compiled.  This
- * doesn't agree with our use of -Werror.
- */
-#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,5)
-# define page_count_field(pg)  ((pg)->count)
-#else
-# define page_count_field(pg)  ((pg)->_count)
-#endif
-
-#define inc_page_count(page)   atomic_inc(&page_count_field(page))
-#define dec_page_count(page)   atomic_dec(&page_count_field(page))
 
 /* Bug 5531: set_page_count doesn't work if the new page count is an
  * expression. */
@@ -170,15 +154,6 @@
    *    Hon    - The PG_compound bit is honoured by munmap.
    *
    *                 OS      A       B       C       D
-   * 2.4.18                  NotDef  NU      resv    NotHon
-   * 2.4.29                  NotDef  NU      resv    NotHon
-   * 2.4.20-31.9     rhl9    NotDef  NU      resv    NotHon
-   *
-   * 2.4.21-4.EL     rhel3   Comp    NU      resv    Hon
-   * 2.4.21-15.EL    rhel3   Comp    NU      resv    Hon
-   * 2.4.21-32.EL    rhel3   Comp    NU      resv    Hon
-   * 2.4.21-40.EL    rhel3   Comp    NU      resv    Hon
-   *
    * 2.6.0                   Comp    NU      resv    NotHon
    *
    * 2.6.5-7.97      sles9   OptInv  NU      resv    NotHon
@@ -209,72 +184,39 @@
    * to one on all the sub-pages.  The SLES 9 range are affected, as
    * are kernels built without CONFIG_MMU defined.
    *
-   * Possible strategies for multi-page allocations:
-   *
-   * EFRM_MMAP_USE_COMPOUND
-   * 1. Allocate a compound page.  Reference counting should then work
-   *    on the whole allocation.  This is a good theory, but is broken
-   *    by bug/feature D (above).
-   *
-   * EFRM_MMAP_USE_SPLIT
-   * 2. Convert the multi-page allocation to many single page
-   *    allocations.  This involves incrementing the reference counts
-   *    and clearing PG_compound on all the pages (including the
-   *    first).  The references should be released _after_ calling
-   *    pci_free_consistent so that that call doesn't release the
-   *    memory.
-   *
-   * EFRM_MMAP_USE_INCREMENT
-   * 3. Increment the reference count on all the pages after
-   *    allocating and decrement them again before freeing.  This gets
-   *    round the zero reference count problem.  It doesn't handle the
-   *    case where someone else is holding a reference to one of our
-   *    pages when we free the pages, but we think VM_IO stops this
-   *    from happening.
+   * On all kernel versions, we just allocate a compound page.
+   * Reference counting should then work on the whole allocation but
+   * is broken by bug/feature D (above) on old kernels.
+   *
+   * EFRM_MMAP_USE_SPLIT 
+
+   *    On old kernels, we convert the multi-page allocation to many
+   *    single page allocations.  This involves incrementing the
+   *    reference counts and clearing PG_compound on all the pages
+   *    (including the first).  Given that the affected kernels are
+   *    inconsistent about the initial reference counts on high order
+   *    page allocations, we reinitialise the reference counts instead
+   *    of incrementing them.  The references are released _after_
+   *    calling pci_free_consistent so that that call doesn't release
+   *    the memory.
    */
 
-/* Should we use strategy 1?  This can be forced on us by the OS. */
-#if defined(PG_compound)
-#define EFRM_MMAP_USE_COMPOUND 1
-#else
-#define EFRM_MMAP_USE_COMPOUND 0
-#endif
-
-/* Should we use strategy 2?  This can be used even if strategy 1 is
- * used. */
+/* Should we split each multi-page allocation into single page
+ * allocations? */
 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,16)
 #define EFRM_MMAP_USE_SPLIT    1
 #else
 #define EFRM_MMAP_USE_SPLIT    0
 #endif
 
-/* Should we use strategy 3?  There's no point doing this if either
- * strategy 1 or strategy 2 is used. */
-#if !EFRM_MMAP_USE_COMPOUND && !EFRM_MMAP_USE_SPLIT
-#error "We shouldn't have to use this strategy."
-#define EFRM_MMAP_USE_INCREMENT 1
-#else
-#define EFRM_MMAP_USE_INCREMENT 0
-#endif
-
-#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17)
-#define EFRM_MMAP_RESET_REFCNT  1
-#else
-#define EFRM_MMAP_RESET_REFCNT  0
-#endif
-
 /* NB. 2.6.17 has renamed SetPageCompound to __SetPageCompound and
  * ClearPageCompound to __ClearPageCompound. */
-#if ((defined(PageCompound)        !=  defined(PG_compound)) ||        \
-     (defined(SetPageCompound)     !=  defined(PG_compound) && \
-      defined(__SetPageCompound)   !=  defined(PG_compound)) ||        \
-     (defined(ClearPageCompound)   !=  defined(PG_compound) && \
-      defined(__ClearPageCompound) !=  defined(PG_compound)) ||        \
-     (defined(__GFP_COMP)          && !defined(PG_compound)))
+#if ( ( !defined(PageCompound) ) ||                                    \
+      ( !defined(PG_compound) ) ||                                     \
+      ( !defined(SetPageCompound)   && !defined(__SetPageCompound) ) ||        
\
+      ( !defined(ClearPageCompound) && !defined(__ClearPageCompound) ) )
 #error Mismatch of defined page-flags.
 #endif
-
-extern int use_pci_alloc;      /* Use pci_alloc_consistent to alloc iopages */
 
 /****************************************************************************
  *
@@ -286,7 +228,7 @@ static inline void pci_mmap_pages_hack_a
 {
        unsigned pfn = __pa(kva) >> PAGE_SHIFT;
        struct page *start_pg = pfn_to_page(pfn);
-#if !defined(NDEBUG) || EFRM_MMAP_USE_SPLIT
+#if EFRM_MMAP_USE_SPLIT
        struct page *end_pg = start_pg + (1 << order);
        struct page *pg;
 #endif
@@ -308,45 +250,8 @@ static inline void pci_mmap_pages_hack_a
 
        /* Check the page count and PG_compound bit. */
 #ifndef NDEBUG
-#  if defined(PG_compound)
-       EFRM_ASSERT(PageCompound(start_pg) == EFRM_MMAP_USE_COMPOUND);
-#  endif
+       EFRM_ASSERT(PageCompound(start_pg) == 1);
        EFRM_ASSERT(page_count(start_pg) == 1);
-
-       {
-               /* Some kernels have the page count field hold (ref_count-1)
-                * rather than (ref_count).  This is so that decrementing the
-                * reference count to "zero" causes the internal value to change
-                * from 0 to -1 which sets the carry flag.  Other kernels store
-                * the real reference count value in the obvious way.  We handle
-                * this here by reading the reference count field of the first
-                * page, which is always 1. */
-               int pg_count_zero;
-               pg_count_zero = atomic_read(&page_count_field(start_pg)) - 1;
-               for (pg = start_pg + 1; pg < end_pg; pg++) {
-                       int pg_count;
-#  if defined(PG_compound)
-                       EFRM_ASSERT(PageCompound(pg) == EFRM_MMAP_USE_COMPOUND);
-#  endif
-
-                       /* Bug 5450: Some kernels initialise the page count
-                        * to one for pages other than the first and some
-                        * leave it at zero.  We allow either behaviour
-                        * here, but disallow anything strange.  Newer
-                        * kernels only define set_page_count in an
-                        * internal header file, so we have to make do with
-                        * incrementing and decrementing the reference
-                        * count.  Fortunately, those kernels don't set the
-                        * reference count to one on all the pages. */
-                       pg_count = atomic_read(&page_count_field(pg));
-#  if EFRM_MMAP_RESET_REFCNT
-                       if (pg_count != pg_count_zero)
-                               EFRM_ASSERT(pg_count == pg_count_zero + 1);
-#  else
-                       EFRM_ASSERT(pg_count == pg_count_zero);
-#  endif
-               }
-       }
 #endif
 
        /* Split the multi-page allocation if necessary. */
@@ -354,50 +259,33 @@ static inline void pci_mmap_pages_hack_a
        for (pg = start_pg; pg < end_pg; pg++) {
 
                /* This is no longer a compound page. */
-#  if EFRM_MMAP_USE_COMPOUND
                ClearPageCompound(pg);
                EFRM_ASSERT(PageCompound(pg) == 0);
-#  endif
 
 #  ifndef NDEBUG
                {
                        int pg_count = page_count(pg);
                        /* Bug 5450: The page count can be zero or one here. */
-                       if (pg == start_pg) {
+                       if (pg == start_pg)
                                EFRM_ASSERT(pg_count == 1);
-                       } else {
-#    if EFRM_MMAP_RESET_REFCNT
-                               if (pg_count != 0)
-                                       EFRM_ASSERT(pg_count == 1);
-#    else
-                               EFRM_ASSERT(pg_count == 0);
-#    endif
-                       }
+                       else if (pg_count != 0)
+                               EFRM_ASSERT(pg_count == 1);
                }
 #  endif
 
-               /* Get a reference which will be released after the pages have
-                * been passed back to pci_free_consistent. */
-#  if EFRM_MMAP_RESET_REFCNT
-               /* Bug 5450: Reset the reference count since the count might
-                * already be 1. */
+               /* Get a reference which will be released after the
+                * pages have been passed back to pci_free_consistent.
+                * Reset the page count instead of incrementing it
+                * because old kernels are inconsistent about
+                * initialising the reference count. */
                ci_set_page_count(pg, (pg == start_pg) ? 2 : 1);
-#  else
-               get_page(pg);
-#  endif
        }
 #endif
-
-       /* Fudge the reference count if necessary. */
-#if EFRM_MMAP_USE_INCREMENT
-       for (pg = start_pg; pg < end_pg; pg++)
-               inc_page_count(pg);
-#endif
 }
 
 static inline void pci_mmap_pages_hack_before_free(caddr_t kva, unsigned order)
 {
-#if EFRM_MMAP_USE_INCREMENT || !defined(NDEBUG)
+#if !defined(NDEBUG)
        /* Drop the references taken in pci_mmap_pages_hack_after_alloc */
        unsigned pfn = __pa(kva) >> PAGE_SHIFT;
        struct page *start_pg = pfn_to_page(pfn);
@@ -412,15 +300,9 @@ static inline void pci_mmap_pages_hack_b
        if (PageReserved(start_pg))
                return;
 
-#  if EFRM_MMAP_USE_INCREMENT
-       for (pg = start_pg; pg < end_pg; pg++)
-               dec_page_count(pg);
-#  endif
-
-#if !defined(NDEBUG)
        EFRM_ASSERT(page_count(start_pg) == 1+EFRM_MMAP_USE_SPLIT);
 
-#  if EFRM_MMAP_USE_COMPOUND && !EFRM_MMAP_USE_SPLIT
+#  if !EFRM_MMAP_USE_SPLIT
        for (pg = start_pg; pg < end_pg; pg++)
                EFRM_ASSERT(PageCompound(pg));
 #  else
@@ -432,7 +314,6 @@ static inline void pci_mmap_pages_hack_b
                EFRM_ASSERT(page_count(pg) == exp_pg_count);
        }
 #  endif
-#endif
 
 #endif
 }
@@ -483,52 +364,23 @@ void *efrm_dma_alloc_coherent(struct dev
 void *efrm_dma_alloc_coherent(struct device *dev, size_t size,
                              dma_addr_t *dma_addr, int flag)
 {
-       struct pci_dev *pci_dev;
        void *ptr;
        unsigned order;
-       EFRM_IOMMU_DECL;
 
        order = __ffs(size/PAGE_SIZE);
        EFRM_ASSERT(size == (PAGE_SIZE<<order));
+
+       /* Can't take a spinlock here since the allocation can
+        * block. */
+       ptr = dma_alloc_coherent(dev, size, dma_addr, flag);
+       if (ptr == NULL)
+               return ptr;
 
        /* NB. The caller may well set __GFP_COMP.  However we can't
         * rely on this working on older kernels.  2.6.9 only acts on
         * __GFP_COMP if CONFIG_HUGETLB_PAGE is defined.  If the flag
         * did have an effect then PG_compound will be set on the
         * pages. */
-
-       if (use_pci_alloc) {
-               /* Can't take a spinlock here since the allocation can
-                * block. */
-               ptr = dma_alloc_coherent(dev, size, dma_addr, flag);
-               if (ptr == NULL)
-                       return ptr;
-       } else {
-#ifdef CONFIG_SWIOTLB          /* BUG1340 */
-               if (swiotlb) {
-                       EFRM_ERR("%s: This kernel is using DMA bounce "
-                                "buffers.  Please upgrade kernel to "
-                                "linux2.6 or reduce the amount of RAM "
-                                "with mem=XXX.", __FUNCTION__);
-                       return NULL;
-               }
-#endif
-               ptr = (void *)__get_free_pages(flag, order);
-
-               if (ptr == NULL)
-                       return NULL;
-
-               EFRM_IOMMU_LOCK();
-               pci_dev = container_of(dev, struct pci_dev, dev);
-               *dma_addr = pci_map_single(pci_dev, ptr, size,
-                                          PCI_DMA_BIDIRECTIONAL);
-               EFRM_IOMMU_UNLOCK();
-               if (pci_dma_mapping_error(*dma_addr)) {
-                       free_pages((unsigned long)ptr, order);
-                       return NULL;
-               }
-       }
-
 #ifndef CONFIG_IA64
        pci_mmap_pages_hack_after_alloc(ptr, order);
 #endif
@@ -547,9 +399,7 @@ void efrm_dma_free_coherent(struct devic
 void efrm_dma_free_coherent(struct device *dev, size_t size,
                            void *ptr, dma_addr_t dma_addr)
 {
-       struct pci_dev *pci_dev;
        unsigned order;
-       EFRM_IOMMU_DECL;
 
        order = __ffs(size/PAGE_SIZE);
        EFRM_ASSERT(size == (PAGE_SIZE<<order));
@@ -564,19 +414,7 @@ void efrm_dma_free_coherent(struct devic
 #ifndef CONFIG_IA64
        pci_mmap_pages_hack_before_free(ptr, order);
 #endif
-       if (use_pci_alloc) {
-               EFRM_IOMMU_LOCK();
-               dma_free_coherent(dev, size, ptr, dma_addr);
-               EFRM_IOMMU_UNLOCK();
-       } else {
-               pci_dev = container_of(dev, struct pci_dev, dev);
-               EFRM_IOMMU_LOCK();
-               efrm_pci_unmap_single(pci_dev, dma_addr, size,
-                                     PCI_DMA_BIDIRECTIONAL);
-               EFRM_IOMMU_UNLOCK();
-
-               free_pages((unsigned long)ptr, order);
-       }
+       dma_free_coherent(dev, size, ptr, dma_addr);
 
 #ifndef CONFIG_IA64
        pci_mmap_pages_hack_after_free(ptr, order);
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/kernel_compat.h
--- a/drivers/net/sfc/sfc_resource/kernel_compat.h      Tue Mar 31 11:49:12 
2009 +0100
+++ b/drivers/net/sfc/sfc_resource/kernel_compat.h      Tue Mar 31 11:59:10 
2009 +0100
@@ -40,9 +40,12 @@
 #define DRIVER_LINUX_RESOURCE_KERNEL_COMPAT_H
 
 #include <linux/version.h>
+#include <linux/moduleparam.h>
+#include <linux/sched.h>
+#include <asm/io.h>
+#include <linux/pci.h>
 
 /********* wait_for_completion_timeout() ********************/
-#include <linux/sched.h>
 
 /* RHEL_RELEASE_CODE from linux/version.h is only defined for 2.6.9-55EL
  * UTS_RELEASE is unfortunately unusable
@@ -51,7 +54,7 @@
 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,11)) && \
        !defined(RHEL_RELEASE_CODE)
 
-static inline unsigned long fastcall __sched
+static inline unsigned long fastcall
 efrm_wait_for_completion_timeout(struct completion *x, unsigned long timeout)
 {
        might_sleep();
@@ -87,125 +90,24 @@ out:
 
 #endif
 
-/********* pci_map_*() ********************/
+/********* io mapping ********************/
 
-#include <linux/pci.h>
+#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,9)
 
-/* Bug 4560: Some kernels leak IOMMU entries under heavy load.  Use a
- * spinlock to serialise access where possible to alleviate the
- * problem.
- *
- * NB. This is duplicated in the net driver.  Please keep in sync. */
-#if ((LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)) && \
-     (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17)) && \
-      defined(__x86_64__) && defined(CONFIG_SMP))
+  #ifndef __iomem
+  #define __iomem
+  #endif
 
-#define EFRM_HAVE_IOMMU_LOCK 1
-
-#if ((LINUX_VERSION_CODE == KERNEL_VERSION(2,6,5)) &&  \
-      defined(CONFIG_SUSE_KERNEL))
-#define EFRM_NEED_ALTERNATE_MAX_PFN 1
-#endif
-
-#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
-#if defined(CONFIG_GART_IOMMU)
-#define EFRM_NO_IOMMU no_iommu
-#else
-#define EFRM_NO_IOMMU 1
-#endif
-#else
-#define EFRM_NO_IOMMU 0
-#endif
-
-/* Set to 0 if we should never use the lock.  Set to 1 if we should
- * automatically determine if we should use the lock.  Set to 2 if we
- * should always use the lock. */
-extern unsigned int efx_use_iommu_lock;
-/* Defined in the net driver. */
-extern spinlock_t efx_iommu_lock;
-/* Non-zero if there is a card which needs the lock. */
-extern int efrm_need_iommu_lock;
-
-/* The IRQ state is needed if the lock is being used.  The flag is
- * cached to ensure that every lock is followed by an unlock, even
- * if the global flag changes in the middle of the operation. */
-
-#define EFRM_IOMMU_DECL                                \
-       unsigned long efx_iommu_irq_state = 0;  \
-       int efx_iommu_using_lock;
-#define EFRM_IOMMU_LOCK()                                              \
-       do {                                                            \
-               efx_iommu_using_lock = (efx_use_iommu_lock &&           \
-                                       (efrm_need_iommu_lock ||        \
-                                        efx_use_iommu_lock >= 2));     \
-               if (efx_iommu_using_lock)                               \
-               spin_lock_irqsave(&efx_iommu_lock, efx_iommu_irq_state);\
-       } while (0)
-#define EFRM_IOMMU_UNLOCK()                                            \
-       do {                                                            \
-               if (efx_iommu_using_lock)                               \
-               spin_unlock_irqrestore(&efx_iommu_lock,                 \
-                                      efx_iommu_irq_state);            \
-       } while (0)
-
-#else /* defined(__x86_64__) && defined(CONFIG_SMP) */
-
-#define EFRM_HAVE_IOMMU_LOCK 0
-#define EFRM_IOMMU_DECL
-#define EFRM_IOMMU_LOCK()    do {} while (0)
-#define EFRM_IOMMU_UNLOCK()  do {} while (0)
+  static inline void efrm_iounmap(volatile void __iomem *addr)
+  {
+         iounmap((void __iomem *)addr);
+  }
+  #define iounmap(arg) efrm_iounmap(arg)
 
 #endif
 
-static inline dma_addr_t efrm_pci_map_single(struct pci_dev *hwdev, void *ptr,
-                                            size_t size, int direction)
-{
-       dma_addr_t dma_addr;
-       EFRM_IOMMU_DECL;
 
-       EFRM_IOMMU_LOCK();
-       dma_addr = pci_map_single(hwdev, ptr, size, direction);
-       EFRM_IOMMU_UNLOCK();
-
-       return dma_addr;
-}
-
-static inline void efrm_pci_unmap_single(struct pci_dev *hwdev,
-                                        dma_addr_t dma_addr, size_t size,
-                                        int direction)
-{
-       EFRM_IOMMU_DECL;
-
-       EFRM_IOMMU_LOCK();
-       pci_unmap_single(hwdev, dma_addr, size, direction);
-       EFRM_IOMMU_UNLOCK();
-}
-
-static inline dma_addr_t efrm_pci_map_page(struct pci_dev *hwdev,
-                                          struct page *page,
-                                          unsigned long offset, size_t size,
-                                          int direction)
-{
-       dma_addr_t dma_addr;
-       EFRM_IOMMU_DECL;
-
-       EFRM_IOMMU_LOCK();
-       dma_addr = pci_map_page(hwdev, page, offset, size, direction);
-       EFRM_IOMMU_UNLOCK();
-
-       return dma_addr;
-}
-
-static inline void efrm_pci_unmap_page(struct pci_dev *hwdev,
-                                      dma_addr_t dma_addr, size_t size,
-                                      int direction)
-{
-       EFRM_IOMMU_DECL;
-
-       EFRM_IOMMU_LOCK();
-       pci_unmap_page(hwdev, dma_addr, size, direction);
-       EFRM_IOMMU_UNLOCK();
-}
+/********* Memory allocation *************/
 
 #ifndef IN_KERNEL_COMPAT_C
 #  ifndef __GFP_COMP
@@ -215,6 +117,9 @@ static inline void efrm_pci_unmap_page(s
 #    define __GFP_ZERO 0
 #  endif
 #endif
+
+
+/********* pci_map_*() ********************/
 
 extern void *efrm_dma_alloc_coherent(struct device *dev, size_t size,
                                     dma_addr_t *dma_addr, int flag);
@@ -236,4 +141,11 @@ static inline void efrm_pci_free_consist
        efrm_dma_free_coherent(&hwdev->dev, size, ptr, dma_addr);
 }
 
+
+#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,8))
+static inline void efrm_pci_disable_msi(struct pci_dev *dev) {}
+#undef pci_disable_msi
+#define pci_disable_msi efrm_pci_disable_msi
+#endif
+
 #endif /* DRIVER_LINUX_RESOURCE_KERNEL_COMPAT_H */
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/linux_resource_internal.h
--- a/drivers/net/sfc/sfc_resource/linux_resource_internal.h    Tue Mar 31 
11:49:12 2009 +0100
+++ b/drivers/net/sfc/sfc_resource/linux_resource_internal.h    Tue Mar 31 
11:59:10 2009 +0100
@@ -65,7 +65,8 @@ extern int
 extern int
 efrm_nic_add(struct pci_dev *dev, unsigned int opts, const uint8_t *mac_addr,
             struct linux_efhw_nic **lnic_out, spinlock_t *reg_lock,
-            int bt_min, int bt_max, const struct vi_resource_dimensions *);
+            int bt_min, int bt_max, int non_irq_evq,
+            const struct vi_resource_dimensions *);
 extern void efrm_nic_del(struct linux_efhw_nic *);
 
 
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/sfc_resource/nic.c
--- a/drivers/net/sfc/sfc_resource/nic.c        Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/sfc_resource/nic.c        Tue Mar 31 11:59:10 2009 +0100
@@ -71,6 +71,17 @@ int efhw_device_type_init(struct efhw_de
                        return 0;
                }
                break;
+       case 0x7777:
+               dt->arch = EFHW_ARCH_FALCON;
+               dt->variant = 'C';
+               switch (class_revision) {
+               case 0:
+                       dt->revision = 0;
+                       break;
+               default:
+                       return 0;
+               }
+               break;
        default:
                return 0;
        }
@@ -92,8 +103,6 @@ void efhw_nic_init(struct efhw_nic *nic,
 void efhw_nic_init(struct efhw_nic *nic, unsigned flags, unsigned options,
                   struct efhw_device_type dev_type)
 {
-       int i;
-
        nic->devtype = dev_type;
        nic->flags = flags;
        nic->options = options;
@@ -102,8 +111,7 @@ void efhw_nic_init(struct efhw_nic *nic,
        nic->reg_lock = &nic->the_reg_lock;
        nic->mtu = 1500 + ETH_HLEN;
 
-       for (i = 0; i < EFHW_KEVENTQ_MAX; i++)
-               nic->irq_unit[i] = EFHW_IRQ_UNIT_UNUSED;
+       nic->irq_unit = EFHW_IRQ_UNIT_UNUSED;
 
        switch (nic->devtype.arch) {
        case EFHW_ARCH_FALCON:
@@ -118,6 +126,7 @@ void efhw_nic_init(struct efhw_nic *nic,
                        nic->ctr_ap_bar = FALCON_S_CTR_AP_BAR;
                        break;
                case 'B':
+               case 'C':
                        nic->flags |= NIC_FLAG_NO_INTERRUPT;
                        nic->ctr_ap_bar = FALCON_P_CTR_AP_BAR;
                        break;
@@ -135,18 +144,14 @@ void efhw_nic_init(struct efhw_nic *nic,
 
 void efhw_nic_close_interrupts(struct efhw_nic *nic)
 {
-       int i;
-
        EFHW_ASSERT(nic);
        if (!efhw_nic_have_hw(nic))
                return;
 
        EFHW_ASSERT(efhw_nic_have_hw(nic));
 
-       for (i = 0; i < EFHW_KEVENTQ_MAX; i++) {
-               if (nic->irq_unit[i] != EFHW_IRQ_UNIT_UNUSED)
-                       efhw_nic_interrupt_disable(nic, i);
-       }
+       if (nic->irq_unit != EFHW_IRQ_UNIT_UNUSED)
+               efhw_nic_interrupt_disable(nic);
 }
 
 void efhw_nic_dtor(struct efhw_nic *nic)
@@ -156,7 +161,6 @@ void efhw_nic_dtor(struct efhw_nic *nic)
        /* Check that we have functional units because the software only
         * driver doesn't initialise anything hardware related any more */
 
-#ifndef __ci_ul_driver__
        /* close interrupts is called first because the act of deregistering
           the driver could cause this driver to change from master to slave
           and hence the implicit interrupt mappings would be wrong */
@@ -168,19 +172,14 @@ void efhw_nic_dtor(struct efhw_nic *nic)
                efhw_nic_close_hardware(nic);
        }
        EFHW_TRACE("%s: functional units ... done", __FUNCTION__);
-#endif
 
        /* destroy event queues */
        EFHW_TRACE("%s: event queues ... ", __FUNCTION__);
 
-#ifndef __ci_ul_driver__
-       {
-               int i;
-               for (i = 0; i < EFHW_KEVENTQ_MAX; ++i)
-                       if (nic->evq[i].evq_mask)
-                               efhw_keventq_dtor(nic, &nic->evq[i]);
-       }
-#endif
+       if (nic->interrupting_evq.evq_mask)
+               efhw_keventq_dtor(nic, &nic->interrupting_evq);
+       if (nic->non_interrupting_evq.evq_mask)
+               efhw_keventq_dtor(nic, &nic->non_interrupting_evq);
 
        EFHW_TRACE("%s: event queues ... done", __FUNCTION__);
 
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/resource_driver.c
--- a/drivers/net/sfc/sfc_resource/resource_driver.c    Tue Mar 31 11:49:12 
2009 +0100
+++ b/drivers/net/sfc/sfc_resource/resource_driver.c    Tue Mar 31 11:59:10 
2009 +0100
@@ -45,15 +45,6 @@
 #include <ci/efrm/vi_resource_private.h>
 #include <ci/efrm/driver_private.h>
 
-#if EFRM_HAVE_IOMMU_LOCK
-#ifdef EFRM_NEED_ALTERNATE_MAX_PFN
-extern unsigned long blk_max_pfn;
-#define max_pfn blk_max_pfn
-#else
-#include <linux/bootmem.h>
-#endif
-#endif
-
 MODULE_AUTHOR("Solarflare Communications");
 MODULE_LICENSE("GPL");
 
@@ -63,11 +54,6 @@ static struct efhw_ev_handler ev_handler
        .dmaq_flushed_fn = efrm_handle_dmaq_flushed,
 };
 
-#if EFRM_HAVE_IOMMU_LOCK
-int efrm_need_iommu_lock;
-EXPORT_SYMBOL(efrm_need_iommu_lock);
-#endif
-
 const int max_hardware_init_repeats = 10;
 
 /*--------------------------------------------------------------------
@@ -76,7 +62,6 @@ const int max_hardware_init_repeats = 10
  *
  *--------------------------------------------------------------------*/
 /* See docs/notes/pci_alloc_consistent */
-int use_pci_alloc = 1;         /* Use pci_alloc_consistent to alloc iopages */
 static int do_irq = 1;         /* enable interrupts */
 
 #if defined(CONFIG_X86_XEN)
@@ -94,9 +79,6 @@ MODULE_PARM_DESC(irq_moderation, "IRQ mo
 MODULE_PARM_DESC(irq_moderation, "IRQ moderation in usec");
 module_param(nic_options, int, S_IRUGO);
 MODULE_PARM_DESC(nic_options, "Nic options -- see efhw_types.h");
-module_param(use_pci_alloc, int, S_IRUGO);
-MODULE_PARM_DESC(use_pci_alloc, "Use pci_alloc_consistent to alloc iopages "
-                "(autodetected by kernel version)");
 module_param(efx_vi_eventq_size, int, S_IRUGO);
 MODULE_PARM_DESC(efx_vi_eventq_size,
                 "Size of event queue allocated by efx_vi library");
@@ -176,7 +158,6 @@ static int efrm_nic_buffer_table_alloc(s
 {
        int capacity;
        int page_order;
-       int i;
        int rc;
 
        /* Choose queue size. */
@@ -189,16 +170,19 @@ static int efrm_nic_buffer_table_alloc(s
                } else if (capacity & nic->evq_sizes)
                        break;
        }
-       for (i = 0; i < EFHW_KEVENTQ_MAX; ++i) {
-               nic->evq[i].hw.capacity = capacity;
-               nic->evq[i].hw.buf_tbl_alloc.base = (unsigned)-1;
-       }
+
+       nic->interrupting_evq.hw.capacity = capacity;
+       nic->interrupting_evq.hw.buf_tbl_alloc.base = (unsigned)-1;
+
+       nic->non_interrupting_evq.hw.capacity = capacity;
+       nic->non_interrupting_evq.hw.buf_tbl_alloc.base = (unsigned)-1;
 
        /* allocate buffer table entries to map onto the iobuffer */
        page_order = get_order(capacity * sizeof(efhw_event_t));
        if (!(nic->flags & NIC_FLAG_NO_INTERRUPT)) {
                rc = efrm_buffer_table_alloc(page_order,
-                                            &nic->evq[0].hw.buf_tbl_alloc);
+                                            &nic->interrupting_evq
+                                            .hw.buf_tbl_alloc);
                if (rc < 0) {
                        EFRM_WARN
                            ("%s: failed (%d) to alloc %d buffer table entries",
@@ -207,7 +191,7 @@ static int efrm_nic_buffer_table_alloc(s
                }
        }
        rc = efrm_buffer_table_alloc(page_order,
-                                    &nic->evq[FALCON_EVQ_NONIRQ].hw.
+                                    &nic->non_interrupting_evq.hw.
                                     buf_tbl_alloc);
        if (rc < 0) {
                EFRM_WARN
@@ -223,16 +207,17 @@ static int efrm_nic_buffer_table_alloc(s
  */
 static void efrm_nic_buffer_table_free(struct efhw_nic *nic)
 {
-       int i;
-       for (i = 0; i <= FALCON_EVQ_NONIRQ; i++)
-               if (nic->evq[i].hw.buf_tbl_alloc.base != (unsigned)-1)
-                       efrm_buffer_table_free(&nic->evq[i].hw.buf_tbl_alloc);
-
+       if (nic->interrupting_evq.hw.buf_tbl_alloc.base != (unsigned)-1)
+               efrm_buffer_table_free(&nic->interrupting_evq.hw
+                                      .buf_tbl_alloc);
+       if (nic->non_interrupting_evq.hw.buf_tbl_alloc.base != (unsigned)-1)
+               efrm_buffer_table_free(&nic->non_interrupting_evq
+                                      .hw.buf_tbl_alloc);
 }
 
 static int iomap_bar(struct linux_efhw_nic *lnic, size_t len)
 {
-       efhw_ioaddr_t ioaddr;
+       volatile char __iomem *ioaddr;
 
        ioaddr = ioremap_nocache(lnic->ctr_ap_pci_addr, len);
        if (ioaddr == 0)
@@ -345,11 +330,9 @@ void linux_efrm_nic_dtor(struct linux_ef
 void linux_efrm_nic_dtor(struct linux_efhw_nic *lnic)
 {
        struct efhw_nic *nic = &lnic->nic;
-       efhw_ioaddr_t bar_ioaddr = nic->bar_ioaddr;
+       volatile char __iomem *bar_ioaddr = nic->bar_ioaddr;
 
        efhw_nic_dtor(nic);
-
-       efrm_nic_buffer_table_free(nic);
 
        /* Unmap the bar. */
        EFRM_ASSERT(bar_ioaddr);
@@ -369,7 +352,7 @@ static void efrm_tasklet(unsigned long p
 
        EFRM_ASSERT(!(nic->flags & NIC_FLAG_NO_INTERRUPT));
 
-       efhw_keventq_poll(nic, &nic->evq[0]);
+       efhw_keventq_poll(nic, &nic->interrupting_evq);
        EFRM_TRACE("tasklet complete");
 }
 
@@ -409,7 +392,7 @@ int
 int
 efrm_nic_add(struct pci_dev *dev, unsigned flags, const uint8_t *mac_addr,
             struct linux_efhw_nic **lnic_out, spinlock_t *reg_lock,
-            int bt_min, int bt_max,
+            int bt_min, int bt_lim, int non_irq_evq,
             const struct vi_resource_dimensions *res_dim)
 {
        struct linux_efhw_nic *lnic = NULL;
@@ -424,14 +407,14 @@ efrm_nic_add(struct pci_dev *dev, unsign
                   pci_name(dev) ? pci_name(dev) : "?", dev->irq);
 
        /* Ensure that we have room for the new adapter-structure. */
-       if (efrm_nic_table.nic_count == EFHW_MAX_NR_DEVS) {
+       if (efrm_nic_tablep->nic_count == EFHW_MAX_NR_DEVS) {
                EFRM_WARN("%s: WARNING: too many devices", __FUNCTION__);
                rc = -ENOMEM;
                goto failed;
        }
 
        if (n_nics_probed == 0) {
-               rc = efrm_resources_init(res_dim, bt_min, bt_max);
+               rc = efrm_resources_init(res_dim, bt_min, bt_lim);
                if (rc != 0)
                        goto failed;
                resources_init = 1;
@@ -467,7 +450,7 @@ efrm_nic_add(struct pci_dev *dev, unsign
        rc = efrm_driver_register_nic(nic, nic_index++);
        if (rc < 0) {
                EFRM_ERR("%s: cannot register nic %d with nic error code %d",
-                        __FUNCTION__, efrm_nic_table.nic_count, rc);
+                        __FUNCTION__, efrm_nic_tablep->nic_count, rc);
                goto failed;
        }
        registered_nic = 1;
@@ -484,7 +467,8 @@ efrm_nic_add(struct pci_dev *dev, unsign
           we want to make sure that we maximise our chances, so we
           loop a few times until all is good. */
        for (count = 0; count < max_hardware_init_repeats; count++) {
-               rc = efhw_nic_init_hardware(nic, &ev_handler, mac_addr);
+               rc = efhw_nic_init_hardware(nic, &ev_handler, mac_addr,
+                                           non_irq_evq);
                if (rc >= 0)
                        break;
 
@@ -509,18 +493,10 @@ efrm_nic_add(struct pci_dev *dev, unsign
                        EFRM_ERR("Interrupt initialisation failed (%d)", rc);
                        goto failed;
                }
-               efhw_nic_set_interrupt_moderation(nic, 0, irq_moderation);
-               efhw_nic_interrupt_enable(nic, 0);
+               efhw_nic_set_interrupt_moderation(nic, irq_moderation);
+               efhw_nic_interrupt_enable(nic);
        }
        EFRM_TRACE("interrupts are %sregistered", do_irq ? "" : "not ");
-
-#if EFRM_HAVE_IOMMU_LOCK
-       /* Bug 4560: We need the lock if there is memory which cannot be
-        * accessed by the card and there is an IOMMU to access it.  In that
-        * case, the kernel will use the IOMMU to access the high memory. */
-       if ((dev->dma_mask >> PAGE_SHIFT) < max_pfn && !EFRM_NO_IOMMU)
-               efrm_need_iommu_lock = 1;
-#endif
 
        *lnic_out = lnic;
        EFRM_ASSERT(rc == 0);
@@ -552,6 +528,8 @@ void efrm_nic_del(struct linux_efhw_nic 
        EFRM_TRACE("%s:", __FUNCTION__);
        EFRM_ASSERT(nic);
 
+       efrm_nic_buffer_table_free(nic);
+
        efrm_driver_unregister_nic(nic);
 
        /*
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/sfc_resource/resources.c
--- a/drivers/net/sfc/sfc_resource/resources.c  Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/sfc_resource/resources.c  Tue Mar 31 11:59:10 2009 +0100
@@ -40,11 +40,11 @@
 
 int
 efrm_resources_init(const struct vi_resource_dimensions *vi_res_dim,
-                   int buffer_table_min, int buffer_table_max)
+                   int buffer_table_min, int buffer_table_lim)
 {
        int i, rc;
 
-       rc = efrm_buffer_table_ctor(buffer_table_min, buffer_table_max);
+       rc = efrm_buffer_table_ctor(buffer_table_min, buffer_table_lim);
        if (rc != 0)
                return rc;
 
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/vi_resource_alloc.c
--- a/drivers/net/sfc/sfc_resource/vi_resource_alloc.c  Tue Mar 31 11:49:12 
2009 +0100
+++ b/drivers/net/sfc/sfc_resource/vi_resource_alloc.c  Tue Mar 31 11:59:10 
2009 +0100
@@ -82,13 +82,13 @@ static inline int efrm_vi_rm_alloc_id(ui
        int instance;
        int rc;
 
-       if (efrm_nic_table.a_nic == NULL)       /* ?? FIXME: surely not right */
+       if (efrm_nic_tablep->a_nic == NULL)     /* ?? FIXME: surely not right */
                return -ENODEV;
 
        spin_lock_irqsave(&efrm_vi_manager->rm.rm_lock, lock_flags);
 
        /* Falcon A1 RX phys addr wierdness. */
-       if (efrm_nic_table.a_nic->devtype.variant == 'A' &&
+       if (efrm_nic_tablep->a_nic->devtype.variant == 'A' &&
            (vi_flags & EFHW_VI_RX_PHYS_ADDR_EN)) {
                if (vi_flags & EFHW_VI_JUMBO_EN) {
                        /* Falcon-A cannot do phys + scatter. */
@@ -141,10 +141,10 @@ static void efrm_vi_rm_free_id(int insta
        irq_flags_t lock_flags;
        struct kfifo *instances;
 
-       if (efrm_nic_table.a_nic == NULL)       /* ?? FIXME: surely not right */
+       if (efrm_nic_tablep->a_nic == NULL)     /* ?? FIXME: surely not right */
                return;
 
-       if (efrm_nic_table.a_nic->devtype.variant == 'A' &&
+       if (efrm_nic_tablep->a_nic->devtype.variant == 'A' &&
            instance == FALCON_A1_ISCSI_DMAQ) {
                EFRM_ASSERT(efrm_vi_manager->iscsi_dmaq_instance_is_free ==
                            false);
@@ -361,7 +361,7 @@ efrm_vi_rm_init_evq(struct vi_resource *
 efrm_vi_rm_init_evq(struct vi_resource *virs, int nic_index)
 {
        int rc;
-       struct efhw_nic *nic = efrm_nic_table.nic[nic_index];
+       struct efhw_nic *nic = efrm_nic_tablep->nic[nic_index];
        int instance = EFRM_RESOURCE_INSTANCE(virs->rs.rs_handle);
        struct eventq_resource_hardware *evq_hw =
            &virs->nic_info[nic_index].evq_pages;
@@ -419,7 +419,7 @@ static inline void
 static inline void
 efrm_vi_rm_fini_evq(struct vi_resource *virs, int nic_index)
 {
-       struct efhw_nic *nic = efrm_nic_table.nic[nic_index];
+       struct efhw_nic *nic = efrm_nic_tablep->nic[nic_index];
        int instance = EFRM_RESOURCE_INSTANCE(virs->rs.rs_handle);
        struct vi_resource_nic_info *nic_info = &virs->nic_info[nic_index];
 
@@ -481,7 +481,7 @@ efrm_vi_rm_init_or_fini_dmaq(struct vi_r
                             int queue_type, int init, int nic_index)
 {
        int rc;
-       struct efhw_nic *nic = efrm_nic_table.nic[nic_index];
+       struct efhw_nic *nic = efrm_nic_tablep->nic[nic_index];
        int instance = EFRM_RESOURCE_INSTANCE(virs->rs.rs_handle);
        uint32_t buf_bytes;
        struct vi_resource *evq_virs;
@@ -490,7 +490,7 @@ efrm_vi_rm_init_or_fini_dmaq(struct vi_r
        struct vi_resource_nic_info *nic_info = &virs->nic_info[nic_index];
        int page_order;
        uint32_t num_pages;
-       efhw_iopages_t *iobuff;
+       struct efhw_iopages *iobuff;
 #endif
 
        if (!init)
@@ -553,6 +553,10 @@ destroy:
        /* Ignore disabled queues. */
        if (virs->dmaq_capacity[queue_type] == 0)
                return 0;
+
+       /* Ensure TX pacing turned off -- queue flush doesn't reset this. */
+       if (queue_type == EFRM_VI_RM_DMA_QUEUE_TX)
+               falcon_nic_pace(nic, instance, 0);
 
        /* No need to disable the queue here.  Nobody is using it anyway. */
 
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/vi_resource_event.c
--- a/drivers/net/sfc/sfc_resource/vi_resource_event.c  Tue Mar 31 11:49:12 
2009 +0100
+++ b/drivers/net/sfc/sfc_resource/vi_resource_event.c  Tue Mar 31 11:59:10 
2009 +0100
@@ -48,7 +48,7 @@ efrm_eventq_request_wakeup(struct vi_res
        struct efhw_nic *nic;
        int next_i;
        EFRM_ASSERT(efrm_nic_set_read(&virs->nic_set, nic_index));
-       nic = efrm_nic_table.nic[nic_index];
+       nic = efrm_nic_tablep->nic[nic_index];
        EFRM_ASSERT(nic);
        next_i = ((current_ptr / sizeof(efhw_event_t)) &
                  (virs->evq_capacity - 1));
@@ -61,7 +61,7 @@ EXPORT_SYMBOL(efrm_eventq_request_wakeup
 
 void efrm_eventq_reset(struct vi_resource *virs, int nic_index)
 {
-       struct efhw_nic *nic = efrm_nic_table.nic[nic_index];
+       struct efhw_nic *nic = efrm_nic_tablep->nic[nic_index];
        int instance = EFRM_RESOURCE_INSTANCE(virs->rs.rs_handle);
 
        EFRM_ASSERT(virs->evq_capacity != 0);
@@ -217,16 +217,18 @@ efrm_eventq_do_callback(struct efhw_nic 
        }
 }
 
-void efrm_handle_wakeup_event(struct efhw_nic *nic, efhw_event_t *ev)
-{
-       efrm_eventq_do_callback(nic,
-                               (unsigned int)FALCON_EVENT_WAKE_EVQ_ID(ev),
-                               false);
-}
-
-void efrm_handle_timeout_event(struct efhw_nic *nic, efhw_event_t *ev)
-{
-       efrm_eventq_do_callback(nic,
-                               (unsigned int)FALCON_EVENT_WAKE_EVQ_ID(ev),
-                               true);
-}
+void efrm_handle_wakeup_event(struct efhw_nic *nic, unsigned instance)
+{
+       efrm_eventq_do_callback(nic, instance, false);
+}
+
+void efrm_handle_timeout_event(struct efhw_nic *nic, unsigned instance)
+{
+       efrm_eventq_do_callback(nic, instance, true);
+}
+
+void efrm_handle_sram_event(struct efhw_nic *nic)
+{
+  if (nic->buf_commit_outstanding > 0)
+    nic->buf_commit_outstanding--;
+}
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/vi_resource_flush.c
--- a/drivers/net/sfc/sfc_resource/vi_resource_flush.c  Tue Mar 31 11:49:12 
2009 +0100
+++ b/drivers/net/sfc/sfc_resource/vi_resource_flush.c  Tue Mar 31 11:59:10 
2009 +0100
@@ -409,7 +409,7 @@ efrm_handle_tx_dmaq_flushed(struct efhw_
 }
 
 void
-efrm_handle_dmaq_flushed(struct efhw_nic *flush_nic, int instance,
+efrm_handle_dmaq_flushed(struct efhw_nic *flush_nic, unsigned instance,
                         int rx_flush)
 {
        irq_flags_t lock_flags;
diff -r 89d9c025b46c -r ad4d307bf9ce 
drivers/net/sfc/sfc_resource/vi_resource_manager.c
--- a/drivers/net/sfc/sfc_resource/vi_resource_manager.c        Tue Mar 31 
11:49:12 2009 +0100
+++ b/drivers/net/sfc/sfc_resource/vi_resource_manager.c        Tue Mar 31 
11:59:10 2009 +0100
@@ -76,7 +76,7 @@ efrm_create_or_destroy_vi_resource_manag
        struct list_head flush_pending;
        irq_flags_t lock_flags;
        int rc, i, n_evqs;
-       unsigned dmaq_min, dmaq_max;
+       unsigned dmaq_min, dmaq_lim;
 
        EFRM_ASSERT(rm_in_out);
 
@@ -85,11 +85,11 @@ efrm_create_or_destroy_vi_resource_manag
 
        EFRM_ASSERT(dims);
        EFRM_NOTICE("vi_resource_manager: evq_int=%u-%u evq_timer=%u-%u",
-                   dims->evq_int_min, dims->evq_int_max,
-                   dims->evq_timer_min, dims->evq_timer_max);
+                   dims->evq_int_min, dims->evq_int_lim,
+                   dims->evq_timer_min, dims->evq_timer_lim);
        EFRM_NOTICE("vi_resource_manager: rxq=%u-%u txq=%u-%u",
-                   dims->rxq_min, dims->rxq_max,
-                   dims->txq_min, dims->txq_max);
+                   dims->rxq_min, dims->rxq_lim,
+                   dims->txq_min, dims->txq_lim);
 
        efrm_vi_manager = kmalloc(sizeof(*efrm_vi_manager), GFP_KERNEL);
        if (efrm_vi_manager == NULL) {
@@ -102,12 +102,12 @@ efrm_create_or_destroy_vi_resource_manag
        efrm_vi_manager->iscsi_dmaq_instance_is_free = true;
 
        dmaq_min = max(dims->rxq_min, dims->txq_min);
-       dmaq_max = min(dims->rxq_max, dims->txq_max);
+       dmaq_lim = min(dims->rxq_lim, dims->txq_lim);
 
        efrm_vi_manager->with_timer_base =
            max(dmaq_min, dims->evq_timer_min);
        efrm_vi_manager->with_timer_limit =
-           min(dmaq_max, dims->evq_timer_max);
+           min(dmaq_lim, dims->evq_timer_lim);
        rc = efrm_kfifo_id_ctor(&efrm_vi_manager->instances_with_timer,
                                efrm_vi_manager->with_timer_base,
                                efrm_vi_manager->with_timer_limit,
@@ -118,7 +118,7 @@ efrm_create_or_destroy_vi_resource_manag
        efrm_vi_manager->with_interrupt_base =
            max(dmaq_min, dims->evq_int_min);
        efrm_vi_manager->with_interrupt_limit =
-           min(dmaq_max, dims->evq_int_max);
+           min(dmaq_lim, dims->evq_int_lim);
        efrm_vi_manager->with_interrupt_limit =
                max(efrm_vi_manager->with_interrupt_limit,
                    efrm_vi_manager->with_interrupt_base);
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/sfe4001.c
--- a/drivers/net/sfc/sfe4001.c Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/sfe4001.c Tue Mar 31 11:59:10 2009 +0100
@@ -130,18 +130,18 @@ void sfe4001_poweroff(struct efx_nic *ef
 
        /* Turn off all power rails */
        out = 0xff;
-       (void) efx_i2c_write(i2c, PCA9539, P0_OUT, &out, EFX_BYTE);
+       (void) efx_i2c_write(i2c, PCA9539, P0_OUT, &out, 1);
 
        /* Disable port 1 outputs on IO expander */
        cfg = 0xff;
-       (void) efx_i2c_write(i2c, PCA9539, P1_CONFIG, &cfg, EFX_BYTE);
+       (void) efx_i2c_write(i2c, PCA9539, P1_CONFIG, &cfg, 1);
 
        /* Disable port 0 outputs on IO expander */
        cfg = 0xff;
-       (void) efx_i2c_write(i2c, PCA9539, P0_CONFIG, &cfg, EFX_BYTE);
+       (void) efx_i2c_write(i2c, PCA9539, P0_CONFIG, &cfg, 1);
 
        /* Clear any over-temperature alert */
-       (void) efx_i2c_read(i2c, MAX6647, RSL, &in, EFX_BYTE);
+       (void) efx_i2c_read(i2c, MAX6647, RSL, &in, 1);
 }
 
 static int sfe4001_check_hw(struct efx_nic *efx)
@@ -163,7 +163,7 @@ static int sfe4001_check_hw(struct efx_n
        if (falcon_xaui_link_ok(efx))
                return 0;
 
-       rc = efx_i2c_read(i2c, PCA9539, P1_IN, &status, EFX_BYTE);
+       rc = efx_i2c_read(i2c, PCA9539, P1_IN, &status, 1);
        status &= ((1 << P1_AFE_PWD_LBN) | (1 << P1_DSP_PWD25_LBN));
 
        /* We know we can read from the IO expander because we did
@@ -223,12 +223,12 @@ int sfe4001_poweron(struct efx_nic *efx)
        /* Set DSP over-temperature alert threshold */
        EFX_INFO(efx, "DSP cut-out at %dC\n", xgphy_max_temperature);
        rc = efx_i2c_write(i2c, MAX6647, WLHO,
-                          &xgphy_max_temperature, EFX_BYTE);
+                          &xgphy_max_temperature, 1);
        if (rc)
                goto fail1;
 
        /* Read it back and verify */
-       rc = efx_i2c_read(i2c, MAX6647, RLHN, &in, EFX_BYTE);
+       rc = efx_i2c_read(i2c, MAX6647, RLHN, &in, 1);
        if (rc)
                goto fail1;
        if (in != xgphy_max_temperature) {
@@ -237,17 +237,17 @@ int sfe4001_poweron(struct efx_nic *efx)
        }
 
        /* Clear any previous over-temperature alert */
-       rc = efx_i2c_read(i2c, MAX6647, RSL, &in, EFX_BYTE);
+       rc = efx_i2c_read(i2c, MAX6647, RSL, &in, 1);
        if (rc)
                goto fail1;
 
        /* Enable port 0 and port 1 outputs on IO expander */
        cfg = 0x00;
-       rc = efx_i2c_write(i2c, PCA9539, P0_CONFIG, &cfg, EFX_BYTE);
+       rc = efx_i2c_write(i2c, PCA9539, P0_CONFIG, &cfg, 1);
        if (rc)
                goto fail1;
        cfg = 0xff & ~(1 << P1_SPARE_LBN);
-       rc = efx_i2c_write(i2c, PCA9539, P1_CONFIG, &cfg, EFX_BYTE);
+       rc = efx_i2c_write(i2c, PCA9539, P1_CONFIG, &cfg, 1);
        if (rc)
                goto fail2;
 
@@ -255,7 +255,7 @@ int sfe4001_poweron(struct efx_nic *efx)
        out = 0xff & ~((0 << P0_EN_1V2_LBN) | (0 << P0_EN_2V5_LBN) |
                       (0 << P0_EN_3V3X_LBN) | (0 << P0_EN_5V_LBN) |
                       (0 << P0_EN_1V0X_LBN));
-       rc = efx_i2c_write(i2c, PCA9539, P0_OUT, &out, EFX_BYTE);
+       rc = efx_i2c_write(i2c, PCA9539, P0_OUT, &out, 1);
        if (rc)
                goto fail3;
 
@@ -267,14 +267,14 @@ int sfe4001_poweron(struct efx_nic *efx)
                               (1 << P0_EN_3V3X_LBN) | (1 << P0_EN_5V_LBN) |
                               (1 << P0_X_TRST_LBN));
 
-               rc = efx_i2c_write(i2c, PCA9539, P0_OUT, &out, EFX_BYTE);
+               rc = efx_i2c_write(i2c, PCA9539, P0_OUT, &out, 1);
                if (rc)
                        goto fail3;
                msleep(10);
 
                /* Turn on 1V power rail */
                out &= ~(1 << P0_EN_1V0X_LBN);
-               rc = efx_i2c_write(i2c, PCA9539, P0_OUT, &out, EFX_BYTE);
+               rc = efx_i2c_write(i2c, PCA9539, P0_OUT, &out, 1);
                if (rc)
                        goto fail3;
 
@@ -283,7 +283,7 @@ int sfe4001_poweron(struct efx_nic *efx)
                schedule_timeout_uninterruptible(HZ);
 
                /* Check DSP is powered */
-               rc = efx_i2c_read(i2c, PCA9539, P1_IN, &in, EFX_BYTE);
+               rc = efx_i2c_read(i2c, PCA9539, P1_IN, &in, 1);
                if (rc)
                        goto fail3;
                if (in & (1 << P1_AFE_PWD_LBN))
@@ -302,14 +302,14 @@ fail3:
 fail3:
        /* Turn off all power rails */
        out = 0xff;
-       (void) efx_i2c_write(i2c, PCA9539, P0_OUT, &out, EFX_BYTE);
+       (void) efx_i2c_write(i2c, PCA9539, P0_OUT, &out, 1);
        /* Disable port 1 outputs on IO expander */
        out = 0xff;
-       (void) efx_i2c_write(i2c, PCA9539, P1_CONFIG, &out, EFX_BYTE);
+       (void) efx_i2c_write(i2c, PCA9539, P1_CONFIG, &out, 1);
 fail2:
        /* Disable port 0 outputs on IO expander */
        out = 0xff;
-       (void) efx_i2c_write(i2c, PCA9539, P1_CONFIG, &out, EFX_BYTE);
+       (void) efx_i2c_write(i2c, PCA9539, P0_CONFIG, &out, 1);
 fail1:
        return rc;
 }
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/tenxpress.c
--- a/drivers/net/sfc/tenxpress.c       Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/tenxpress.c       Tue Mar 31 11:59:10 2009 +0100
@@ -574,10 +574,8 @@ static int tenxpress_phy_check_hw(struct
 
        link_ok = phy_up && tenxpress_link_ok(efx, 1);
 
-       if (link_ok != efx->link_up) {
-               efx->link_up = link_ok;
+       if (link_ok != efx->link_up)
                efx->mac_op->fake_phy_event(efx);
-       }
 
        /* Nothing to check if we've already shut down the PHY */
        if (!phy_up)
@@ -652,7 +650,7 @@ static void tenxpress_reset_xaui(struct 
        soft_rst2 = mdio_clause45_read(efx, phy, MDIO_MMD_PCS,
                                       PCS_SOFT_RST2_REG);
 
-       /* Modify => put in reset */
+       /* Put in reset */
        test_select &= ~(1 << CLK312_EN_LBN);
        mdio_clause45_write(efx, phy, MDIO_MMD_PCS,
                            PCS_TEST_SELECT_REG, test_select);
@@ -666,7 +664,7 @@ static void tenxpress_reset_xaui(struct 
                            PCS_CLOCK_CTRL_REG, clk_ctrl);
        udelay(10);
 
-       /* Modify => remove reset */
+       /* Remove reset */
        clk_ctrl |= (1 << PLL312_RST_N_LBN);
        mdio_clause45_write(efx, phy, MDIO_MMD_PCS,
                            PCS_CLOCK_CTRL_REG, clk_ctrl);
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/tx.c
--- a/drivers/net/sfc/tx.c      Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/tx.c      Tue Mar 31 11:59:10 2009 +0100
@@ -61,8 +61,7 @@ void efx_stop_queue(struct efx_nic *efx)
        EFX_TRACE(efx, "stop TX queue\n");
 
        atomic_inc(&efx->netif_stop_count);
-       if (likely(efx->net_dev_registered))
-               netif_stop_queue(efx->net_dev);
+       netif_stop_queue(efx->net_dev);
 
        spin_unlock_bh(&efx->netif_stop_lock);
 }
@@ -77,12 +76,35 @@ inline void efx_wake_queue(struct efx_ni
        if (atomic_dec_and_lock(&efx->netif_stop_count,
                                &efx->netif_stop_lock)) {
                EFX_TRACE(efx, "waking TX queue\n");
-               if (likely(efx->net_dev_registered))
-                       netif_wake_queue(efx->net_dev);
+               netif_wake_queue(efx->net_dev);
                spin_unlock(&efx->netif_stop_lock);
        }
        local_bh_enable();
 }
+
+static inline void efx_dequeue_buffer(struct efx_tx_queue *tx_queue,
+                                     struct efx_tx_buffer *buffer)
+{
+       if (buffer->unmap_len) {
+               struct pci_dev *pci_dev = tx_queue->efx->pci_dev;
+               if (buffer->unmap_single)
+                       pci_unmap_single(pci_dev, buffer->unmap_addr,
+                                        buffer->unmap_len, PCI_DMA_TODEVICE);
+               else
+                       pci_unmap_page(pci_dev, buffer->unmap_addr,
+                                      buffer->unmap_len, PCI_DMA_TODEVICE);
+               buffer->unmap_len = 0;
+               buffer->unmap_single = 0;
+       }
+
+       if (buffer->skb) {
+               dev_kfree_skb_any((struct sk_buff *) buffer->skb);
+               buffer->skb = NULL;
+               EFX_TRACE(tx_queue->efx, "TX queue %d transmission id %x "
+                         "complete\n", tx_queue->queue, read_ptr);
+       }
+}
+
 
 /*
  * Add a socket buffer to a TX queue
@@ -239,17 +261,7 @@ static inline int efx_enqueue_skb(struct
                --tx_queue->insert_count;
                insert_ptr = tx_queue->insert_count & efx->type->txd_ring_mask;
                buffer = &tx_queue->buffer[insert_ptr];
-               if (buffer->unmap_len) {
-                       if (buffer->unmap_single)
-                               pci_unmap_single(pci_dev, buffer->unmap_addr,
-                                                buffer->unmap_len,
-                                                PCI_DMA_TODEVICE);
-                       else
-                               pci_unmap_page(pci_dev, buffer->unmap_addr,
-                                              buffer->unmap_len,
-                                              PCI_DMA_TODEVICE);
-               }
-               buffer->unmap_len = 0;
+               efx_dequeue_buffer(tx_queue, buffer);
                buffer->len = 0;
        }
 
@@ -269,56 +281,30 @@ static inline void efx_dequeue_buffers(s
 static inline void efx_dequeue_buffers(struct efx_tx_queue *tx_queue,
                                       unsigned int index)
 {
-       struct pci_dev *pci_dev = tx_queue->efx->pci_dev;
-       struct efx_tx_buffer *buffer;
+       struct efx_nic *efx = tx_queue->efx;
        unsigned int stop_index, read_ptr;
-
-       /* Calculate the stopping point.  Doing the check this way
-        * avoids wrongly completing every buffer in the ring if we
-        * get called twice with the same index.  (Hardware should
-        * never do this, since it can't complete that many buffers in
-        * one go.)
-        */
-       stop_index = (index + 1) & tx_queue->efx->type->txd_ring_mask;
-       read_ptr = tx_queue->read_count & tx_queue->efx->type->txd_ring_mask;
+       unsigned int mask = tx_queue->efx->type->txd_ring_mask;
+
+       stop_index = (index + 1) & mask;
+       read_ptr = tx_queue->read_count & mask;
 
        while (read_ptr != stop_index) {
-               buffer = &tx_queue->buffer[read_ptr];
+               struct efx_tx_buffer *buffer = &tx_queue->buffer[read_ptr];
                if (unlikely(buffer->len == 0)) {
                        EFX_ERR(tx_queue->efx, "TX queue %d spurious TX "
                                "completion id %x\n", tx_queue->queue,
                                read_ptr);
-                       atomic_inc(&tx_queue->efx->errors.spurious_tx);
-                       /* Don't reset */
-               } else {
-                       if (buffer->unmap_len) {
-                               if (buffer->unmap_single)
-                                       pci_unmap_single(pci_dev,
-                                                        buffer->unmap_addr,
-                                                        buffer->unmap_len,
-                                                        PCI_DMA_TODEVICE);
-                               else
-                                       pci_unmap_page(pci_dev,
-                                                      buffer->unmap_addr,
-                                                      buffer->unmap_len,
-                                                      PCI_DMA_TODEVICE);
-                               buffer->unmap_single = 0;
-                               buffer->unmap_len = 0;
-                       }
-                       if (buffer->skb) {
-                               dev_kfree_skb_any((struct sk_buff *)
-                                                 buffer->skb);
-                               buffer->skb = NULL;
-                               EFX_TRACE(tx_queue->efx, "TX queue %d "
-                                         "transmission id %x complete\n",
-                                         tx_queue->queue, read_ptr);
-                       }
-                       buffer->continuation = 1;
-                       buffer->len = 0;
+                       atomic_inc(&efx->errors.spurious_tx);
+                       efx_schedule_reset(efx, RESET_TYPE_TX_SKIP);
+                       return;
                }
+
+               efx_dequeue_buffer(tx_queue, buffer);
+               buffer->continuation = 1;
+               buffer->len = 0;
+
                ++tx_queue->read_count;
-               read_ptr = (tx_queue->read_count &
-                           tx_queue->efx->type->txd_ring_mask);
+               read_ptr = tx_queue->read_count & mask;
        }
 }
 
@@ -385,13 +371,8 @@ out:
        return rc;
 }
 
-#if defined(EFX_USE_FASTCALL)
 void fastcall efx_xmit_done(struct efx_tx_queue *tx_queue, unsigned int index)
-#else
-void efx_xmit_done(struct efx_tx_queue *tx_queue, unsigned int index)
-#endif
-{
-       unsigned long flags __attribute__ ((unused));
+{
        unsigned fill_level;
        struct efx_nic *efx = tx_queue->efx;
 
@@ -407,11 +388,7 @@ void efx_xmit_done(struct efx_tx_queue *
        if (unlikely(tx_queue->stopped)) {
                fill_level = tx_queue->insert_count - tx_queue->read_count;
                if (fill_level < EFX_NETDEV_TX_THRESHOLD(tx_queue)) {
-                       /* If the port is stopped and the net_dev isn't
-                        * registered, then the caller must be performing
-                        * flow control manually */
-                       if (unlikely(!efx->net_dev_registered))
-                               return;
+                       EFX_BUG_ON_PARANOID(!NET_DEV_REGISTERED(efx));
 
                        /* Do this under netif_tx_lock(), to avoid racing
                         * with efx_xmit(). */
@@ -464,8 +441,6 @@ int efx_init_tx_queue(struct efx_tx_queu
 {
        EFX_LOG(tx_queue->efx, "initialising TX queue %d\n", tx_queue->queue);
 
-       ASSERT_RTNL();
-
        /* Initialise fields */
        tx_queue->insert_count = 0;
        tx_queue->write_count = 0;
@@ -479,22 +454,26 @@ int efx_init_tx_queue(struct efx_tx_queu
 
 void efx_release_tx_buffers(struct efx_tx_queue *tx_queue)
 {
-       unsigned int last_index, mask;
-       if (tx_queue->buffer) {
-               /* Free any buffers left in the ring */
-               mask = tx_queue->efx->type->txd_ring_mask;
-               last_index = (tx_queue->insert_count - 1) & mask;
-               EFX_LOG(tx_queue->efx, "Will dequeue up to 0x%x from 0x%x\n",
-                       last_index, tx_queue->read_count & mask);
-               efx_dequeue_buffers(tx_queue, last_index);
+       struct efx_tx_buffer *buffer;
+
+       if (!tx_queue->buffer)
+               return;
+
+       /* Free any buffers left in the ring */
+       while (tx_queue->read_count != tx_queue->write_count) {
+               buffer = &tx_queue->buffer[tx_queue->read_count &
+                                          tx_queue->efx->type->txd_ring_mask];
+               efx_dequeue_buffer(tx_queue, buffer);
+               buffer->continuation = 1;
+               buffer->len = 0;
+
+               ++tx_queue->read_count;
        }
 }
 
 void efx_fini_tx_queue(struct efx_tx_queue *tx_queue)
 {
        EFX_LOG(tx_queue->efx, "shutting down TX queue %d\n", tx_queue->queue);
-
-       ASSERT_RTNL();
 
        /* Flush TX queue, remove descriptor ring */
        falcon_fini_tx(tx_queue);
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/txc43128_phy.c
--- a/drivers/net/sfc/txc43128_phy.c    Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/txc43128_phy.c    Tue Mar 31 11:59:10 2009 +0100
@@ -653,10 +653,9 @@ static int txc43128_phy_check_hw(struct 
        int link_up = txc43128_phy_read_link(efx);
 
        /* Simulate a PHY event if link state has changed */
-       if (link_up != efx->link_up) {
-               efx->link_up = link_up;
+       if (link_up != efx->link_up)
                efx->mac_op->fake_phy_event(efx);
-       } else if (EFX_WORKAROUND_10934(efx)) {
+       else if (EFX_WORKAROUND_10934(efx)) {
                if (link_up || (efx->loopback_mode != LOOPBACK_NONE))
                        data->bug10934_timer = jiffies;
                else {
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/workarounds.h
--- a/drivers/net/sfc/workarounds.h     Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/workarounds.h     Tue Mar 31 11:59:10 2009 +0100
@@ -34,7 +34,7 @@
 #define EFX_WORKAROUND_ALWAYS(efx) 1
 #define EFX_WORKAROUND_FALCON_A(efx) (FALCON_REV(efx) <= FALCON_REV_A1)
 #define EFX_WORKAROUND_FALCON_B0FPGA(efx) \
-       (FALCON_REV(efx) == FALCON_REV_B0 && !(efx)->is_asic)
+       (FALCON_REV(efx) >= FALCON_REV_B0 && !(efx)->is_asic)
 
 /* XAUI resets if link not detected */
 #define EFX_WORKAROUND_5147 EFX_WORKAROUND_ALWAYS
@@ -67,6 +67,8 @@
 #define EFX_WORKAROUND_11482 EFX_WORKAROUND_ALWAYS
 /* Flush events can take a very long time to appear */
 #define EFX_WORKAROUND_11557 EFX_WORKAROUND_ALWAYS
+/* 10Xpress is sensitive to unstable XAUI sync when going into loopback */
+#define EFX_WORKAROUND_11667 EFX_WORKAROUND_ALWAYS
 
 /* Spurious parity errors in TSORT buffers */
 #define EFX_WORKAROUND_5129 EFX_WORKAROUND_FALCON_A
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/net/sfc/xfp_phy.c
--- a/drivers/net/sfc/xfp_phy.c Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/net/sfc/xfp_phy.c Tue Mar 31 11:59:10 2009 +0100
@@ -114,11 +114,12 @@ static int xfp_phy_init(struct efx_nic *
        phy_data->tx_disabled = efx->tx_disabled;
 
        rc = xfp_reset_phy(efx);
-       if (rc < 0)
-               goto fail;
 
        EFX_INFO(efx, "XFP: PHY init %s.\n",
                 rc ? "failed" : "successful");
+       if (rc < 0)
+               goto fail;
+
        return 0;
 
  fail:
@@ -142,10 +143,8 @@ static int xfp_phy_check_hw(struct efx_n
        int rc = 0;
        int link_up = xfp_link_ok(efx);
        /* Simulate a PHY event if link state has changed */
-       if (link_up != efx->link_up) {
-               efx->link_up = link_up;
+       if (link_up != efx->link_up)
                efx->mac_op->fake_phy_event(efx);
-       }
 
        rc = efx->board_info.monitor(efx);
        if (rc) {
@@ -192,15 +191,13 @@ static void xfp_phy_fini(struct efx_nic 
 }
 
 struct efx_phy_operations falcon_xfp_phy_ops = {
-       .init            = xfp_phy_init,
-       .reconfigure     = xfp_phy_reconfigure,
-       .check_hw        = xfp_phy_check_hw,
-       .fini            = xfp_phy_fini,
-       .clear_interrupt = xfp_phy_clear_interrupt,
-       .reset_xaui      = efx_port_dummy_op_void,
-       .mmds            = XFP_REQUIRED_DEVS,
-       .loopbacks       = XFP_LOOPBACKS,
-       /* No loopback appears to be reliable enough for self-test
-        * operation. So don't do it. */
+       .init             = xfp_phy_init,
+       .reconfigure      = xfp_phy_reconfigure,
+       .check_hw         = xfp_phy_check_hw,
+       .fini             = xfp_phy_fini,
+       .clear_interrupt  = xfp_phy_clear_interrupt,
+       .reset_xaui       = efx_port_dummy_op_void,
+       .mmds             = XFP_REQUIRED_DEVS,
+       .loopbacks        = XFP_LOOPBACKS,
        .startup_loopback = LOOPBACK_PCS,
 };
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/xen/sfc_netback/accel.c
--- a/drivers/xen/sfc_netback/accel.c   Tue Mar 31 11:49:12 2009 +0100
+++ b/drivers/xen/sfc_netback/accel.c   Tue Mar 31 11:59:10 2009 +0100
@@ -38,7 +38,9 @@ static int netback_accel_netdev_event(st
        struct net_device *net_dev = (struct net_device *)ptr;
        struct netback_accel *bend;
 
-       if ((event == NETDEV_UP) || (event == NETDEV_DOWN)) {
+       if ((event == NETDEV_UP) || 
+           (event == NETDEV_DOWN) ||
+           (event == NETDEV_CHANGE)) {
                mutex_lock(&bend_list_mutex);
                bend = bend_list;
                while (bend != NULL) {
@@ -51,9 +53,16 @@ static int netback_accel_netdev_event(st
                        if (bend->shared_page == NULL)
                                goto next;
 
-                       if (bend->net_dev->ifindex == net_dev->ifindex)
-                               netback_accel_set_interface_state
-                                       (bend, event == NETDEV_UP);
+                       if (bend->net_dev->ifindex == net_dev->ifindex) {
+                               int ok;
+                               if (event == NETDEV_CHANGE)
+                                       ok = (netif_carrier_ok(net_dev) && 
+                                             (net_dev->flags & IFF_UP));
+                               else
+                                       ok = (netif_carrier_ok(net_dev) && 
+                                             (event == NETDEV_UP));
+                               netback_accel_set_interface_state(bend, ok);
+                       }
 
                next:
                        mutex_unlock(&bend->bend_mutex);
@@ -86,22 +95,31 @@ static int __init netback_accel_init(voi
 #endif
 
        rc = netback_accel_init_fwd();
+       if (rc != 0)
+               goto fail0;
 
-       if (rc == 0)
-               netback_accel_debugfs_init();
+       netback_accel_debugfs_init();
 
-       if (rc == 0)
-               rc = netback_accel_sf_init();
+       rc = netback_accel_sf_init();
+       if (rc != 0)
+               goto fail1;
 
-       if (rc == 0)
-               rc = register_netdevice_notifier
-                       (&netback_accel_netdev_notifier);
+       rc = register_netdevice_notifier
+               (&netback_accel_netdev_notifier);
+       if (rc != 0)
+               goto fail2;
 
-       /*
-        * What if no device was found, shouldn't we clean up stuff
-        * we've allocated for acceleration subsystem?
-        */
+       return 0;
 
+ fail2:
+       netback_accel_sf_shutdown();
+ fail1:
+       netback_accel_debugfs_fini();
+       netback_accel_shutdown_fwd();
+ fail0:
+#ifdef EFX_GCOV
+       gcov_provider_fini(THIS_MODULE);
+#endif
        return rc;
 }
 
diff -r 89d9c025b46c -r ad4d307bf9ce drivers/xen/sfc_netback/accel_solarflare.c
--- a/drivers/xen/sfc_netback/accel_solarflare.c        Tue Mar 31 11:49:12 
2009 +0100
+++ b/drivers/xen/sfc_netback/accel_solarflare.c        Tue Mar 31 11:59:10 
2009 +0100
@@ -170,25 +170,16 @@ static struct netback_accel_hooks accel_
  */
 static int efx_device_to_efab_nic_index(struct efx_dl_device *efx_dl_dev) 
 {
-       int i;
-
-       for (i = 0; i < EFHW_MAX_NR_DEVS; i++) {
-               struct efhw_nic *nic = efrm_nic_table.nic[i];
-
-               /*
-                * It's possible for the nic structure to have not
-                * been initialised if the resource driver failed its
-                * driverlink probe

_______________________________________________
Xen-changelog mailing list
Xen-changelog@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-changelog

<Prev in Thread] Current Thread [Next in Thread>
  • [Xen-changelog] [linux-2.6.18-xen] net sfc: Update sfc and sfc_resource driver to latest release, Xen patchbot-linux-2.6.18-xen <=