Hi Simon.
This patch moves PAL code mapping area from xen identity mapping area
to EFI mapping area which overlaps with guest virtual address.
It means that PAL code area can't be pinned down always as
xen/ia64 vmm does now.
So PAL code should be pinned down after RR7 is switched
and right before firmware calls.
Without the followings patches, xen didn't boot up.
Initial boot. Not kexec boot.
[IA64] pin down correctly PAL CODE area.
So far PAL code is always pinned down by itr[IA64_TR_PALCODE].
However the rule was changed such that the area is pinned
down only by rid XEN_EFI_RR right before firmware call by set_one_rr_efi().
So pinning down PAL CODE area by efi_map_pal_code(), ia64_new_rr7(),
__vmx_switch_rr7() and ia64_reload_tr() is wrong.
So pin down PAL code area only by ia64_new_rr7_efi() and don't pin it
down by other functions.
Signed-off-by: Isaku Yamahata <yamahata@xxxxxxxxxxxxx>
diff -r e8056a7091a7 xen/arch/ia64/linux-xen/efi.c
--- a/xen/arch/ia64/linux-xen/efi.c Mon Jul 14 19:31:54 2008 +0900
+++ b/xen/arch/ia64/linux-xen/efi.c Mon Jul 14 19:33:20 2008 +0900
@@ -447,7 +447,11 @@
efi_map_pal_code (void)
{
#ifdef XEN
- u64 psr;
+ /*
+ * don't map PAL code.
+ * PAL code is mapped by set_one_rr_efi() right before
+ * firmware call.
+ */
(void)efi_get_pal_addr();
#else
void *pal_vaddr = efi_get_pal_addr ();
@@ -455,7 +459,6 @@
if (!pal_vaddr)
return;
-#endif
/*
* Cannot write to CRx with PSR.ic=1
@@ -466,6 +469,7 @@
IA64_GRANULE_SHIFT);
ia64_set_psr(psr); /* restore psr */
ia64_srlz_i();
+#endif
}
void __init
diff -r e8056a7091a7 xen/arch/ia64/linux-xen/mca_asm.S
--- a/xen/arch/ia64/linux-xen/mca_asm.S Mon Jul 14 19:31:54 2008 +0900
+++ b/xen/arch/ia64/linux-xen/mca_asm.S Mon Jul 14 19:33:20 2008 +0900
@@ -473,6 +473,7 @@
;;
srlz.d
;;
+#ifndef XEN
// 3. Reload ITR for PAL code.
GET_THIS_PADDR(r2, ia64_mca_pal_pte)
;;
@@ -491,6 +492,8 @@
;;
srlz.i
;;
+#endif
+
// 4. Reload DTR for stack.
#ifdef XEN
// Kernel registers are saved in a per_cpu cpu_kr_ia64_t
diff -r e8056a7091a7 xen/arch/ia64/vmx/vmx_entry.S
--- a/xen/arch/ia64/vmx/vmx_entry.S Mon Jul 14 19:31:54 2008 +0900
+++ b/xen/arch/ia64/vmx/vmx_entry.S Mon Jul 14 19:33:20 2008 +0900
@@ -598,7 +598,7 @@
/*
* in0: new rr7
* in1: virtual address of guest_vhpt
- * in2: virtual address of pal code segment
+ * in2: virtual addres of guest shared_info
* r8: will contain old rid value
*/
@@ -611,7 +611,7 @@
GLOBAL_ENTRY(__vmx_switch_rr7)
// not sure this unwind statement is correct...
.prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(1)
- alloc loc1 = ar.pfs, 4, 8, 0, 0
+ alloc loc1 = ar.pfs, 4, 7, 0, 0
1:{
mov r28 = in0 // copy procedure index
mov r8 = ip // save ip to compute branch
@@ -623,12 +623,9 @@
tpa loc2 = loc2 // get physical address of per cpu date
tpa r3 = r8 // get physical address of ip
dep loc5 = 0,in1,60,4 // get physical address of guest_vhpt
- dep loc6 = 0,in2,60,4 // get physical address of pal code
- dep loc7 = 0,in3,60,4 // get physical address of privregs
+ dep loc6 = 0,in2,60,4 // get physical address of privregs
;;
dep loc6 = 0,loc6,0,IA64_GRANULE_SHIFT
- // mask granule shift
- dep loc7 = 0,loc7,0,IA64_GRANULE_SHIFT
// mask granule shift
mov loc4 = psr // save psr
;;
@@ -725,46 +722,31 @@
;;
.vhpt_overlaps:
- // re-pin mappings for PAL code section
- mov r24=IA64_TR_PALCODE
- or loc6 = r25,loc6 // construct PA | page properties
- mov r23 = IA64_GRANULE_SHIFT<<2
- ;;
- ptr.i in2,r23
- ;;
- mov cr.itir=r23
- mov cr.ifa=in2
- ;;
- itr.i itr[r24]=loc6 // wire in new mapping...
- ;;
-
// r16, r19, r20 are used by
// ia64_switch_mode_phys()/ia64_switch_mode_virt()
// re-pin mappings for privregs
// r21 = (current physical addr) & (IA64_GRANULE_SIZE - 1)
// r17 = (guest_vhpt physical addr) & (IA64_GRANULE_SIZE - 1)
- // loc6 = (((pal phys addr) & (IA64_GRANULE_SIZE - 1) << 2)) |
PAGE_KERNEL
- // loc7 = (privregs physical addr) & (IA64_GRANULE_SIZE - 1)
- cmp.ne.unc p7,p0=r21,loc7 // check overlap with current stack
+ // loc6 = (privregs physical addr) & (IA64_GRANULE_SIZE - 1)
+ cmp.ne.unc p7,p0=r21,loc6 // check overlap with current stack
;;
-(p7) cmp.ne.unc p8,p0=r17,loc7 // check overlap with guest_vhpt
+(p7) cmp.ne.unc p8,p0=r17,loc6 // check overlap with guest_vhpt
;;
- // loc7 = (((privregs phys) & (IA64_GRANULE_SIZE - 1)) << 2) |
PAGE_KERNEL
- or loc7 = r25,loc7 // construct PA | page properties
+ // loc6 = (((privregs phys) & (IA64_GRANULE_SIZE - 1)) << 2) |
PAGE_KERNEL
+ or loc6 = r25,loc6 // construct PA | page properties
;;
- cmp.ne p9,p0=loc6,loc7
mov r22=IA64_TR_VPD
mov r24=IA64_TR_MAPPED_REGS
mov r23=IA64_GRANULE_SHIFT<<2
;;
-(p9) ptr.i in3,r23
-(p8) ptr.d in3,r23
+ ptr.i in2,r23
+(p8) ptr.d in2,r23
mov cr.itir=r23
- mov cr.ifa=in3
+ mov cr.ifa=in2
;;
-(p9) itr.i itr[r22]=loc7 // wire in new mapping...
+ itr.i itr[r22]=loc6 // wire in new mapping...
;;
-(p8) itr.d dtr[r24]=loc7 // wire in new mapping...
+(p8) itr.d dtr[r24]=loc6 // wire in new mapping...
;;
// done, switch back to virtual and return
diff -r e8056a7091a7 xen/arch/ia64/vmx/vmx_phy_mode.c
--- a/xen/arch/ia64/vmx/vmx_phy_mode.c Mon Jul 14 19:31:54 2008 +0900
+++ b/xen/arch/ia64/vmx/vmx_phy_mode.c Mon Jul 14 19:33:20 2008 +0900
@@ -172,7 +172,7 @@
ia64_dv_serialize_data();
vmx_switch_rr7(vrrtomrr(vcpu,VMX(vcpu, vrr[VRN7])),
(void *)vcpu->arch.vhpt.hash,
- pal_vaddr, vcpu->arch.privregs);
+ vcpu->arch.privregs);
ia64_set_pta(VMX(vcpu, mpta));
vmx_ia64_set_dcr(vcpu);
diff -r e8056a7091a7 xen/arch/ia64/vmx/vmx_vcpu.c
--- a/xen/arch/ia64/vmx/vmx_vcpu.c Mon Jul 14 19:31:54 2008 +0900
+++ b/xen/arch/ia64/vmx/vmx_vcpu.c Mon Jul 14 19:33:20 2008 +0900
@@ -197,12 +197,12 @@
}
void vmx_switch_rr7(unsigned long rid, void *guest_vhpt,
- void *pal_vaddr, void *shared_arch_info)
+ void *shared_arch_info)
{
__get_cpu_var(inserted_vhpt) = guest_vhpt;
__get_cpu_var(inserted_vpd) = shared_arch_info;
__get_cpu_var(inserted_mapped_regs) = shared_arch_info;
- __vmx_switch_rr7(rid, guest_vhpt, pal_vaddr, shared_arch_info);
+ __vmx_switch_rr7(rid, guest_vhpt, shared_arch_info);
}
IA64FAULT vmx_vcpu_set_rr(VCPU *vcpu, u64 reg, u64 val)
@@ -219,7 +219,7 @@
case VRN7:
if (likely(vcpu == current))
vmx_switch_rr7(vrrtomrr(vcpu,val), (void *)vcpu->arch.vhpt.hash,
- pal_vaddr, vcpu->arch.privregs);
+ vcpu->arch.privregs);
break;
case VRN4:
rrval = vrrtomrr(vcpu,val);
diff -r e8056a7091a7 xen/arch/ia64/xen/xenasm.S
--- a/xen/arch/ia64/xen/xenasm.S Mon Jul 14 19:31:54 2008 +0900
+++ b/xen/arch/ia64/xen/xenasm.S Mon Jul 14 19:33:20 2008 +0900
@@ -34,12 +34,13 @@
// unsigned long va_vhpt) /* in4 */
//Local usage:
// loc0=rp, loc1=ar.pfs, loc2=percpu_paddr, loc3=psr, loc4=ar.rse
-// loc5=pal_vaddr, loc6=xen_paddr, loc7=shared_archinfo_paddr,
+// loc5=shared_archinfo_paddr, loc6=xen_paddr,
// r16, r19, r20 are used by ia64_switch_mode_{phys, virt}()
+// loc5 is unused.
GLOBAL_ENTRY(ia64_new_rr7)
// FIXME? not sure this unwind statement is correct...
.prologue ASM_UNW_PRLG_RP|ASM_UNW_PRLG_PFS, ASM_UNW_PRLG_GRSAVE(1)
- alloc loc1 = ar.pfs, 5, 8, 0, 0
+ alloc loc1 = ar.pfs, 5, 7, 0, 0
movl loc2=PERCPU_ADDR
1: {
mov loc3 = psr // save psr
@@ -51,7 +52,7 @@
tpa in1=in1 // grab shared_info BEFORE changing rr7
adds r8 = 1f-1b,r8 // calculate return address for call
;;
- tpa loc7=in2 // grab arch_vcpu_info BEFORE chg rr7
+ tpa loc5=in2 // grab arch_vcpu_info BEFORE chg rr7
movl r17=PSR_BITS_TO_SET
mov loc4=ar.rsc // save RSE configuration
movl r16=PSR_BITS_TO_CLEAR
@@ -60,10 +61,7 @@
mov ar.rsc=0 // put RSE in enforced lazy, LE mode
or loc3=loc3,r17 // add in psr the bits to set
;;
- movl loc5=pal_vaddr // get pal_vaddr
- ;;
- ld8 loc5=[loc5] // read pal_vaddr
- ;;
+
andcm r16=loc3,r16 // removes bits to clear from psr
dep loc6=0,r8,0,KERNEL_TR_PAGE_SHIFT // Xen code paddr
br.call.sptk.many rp=ia64_switch_mode_phys
@@ -163,24 +161,12 @@
add r22=r22,in3
;;
ptr.d r22,r24
- or r23=loc7,r25 // construct PA | page properties
+ or r23=loc5,r25 // construct PA | page properties
mov cr.itir=r24
mov cr.ifa=r22
mov r21=IA64_TR_MAPPED_REGS
;;
itr.d dtr[r21]=r23 // wire in new mapping...
-
- // Purge/insert PAL TR
- mov r24=IA64_TR_PALCODE
- mov r23=IA64_GRANULE_SHIFT<<2
- dep r25=0,loc5,60,4 // convert pal vaddr to paddr
- ;;
- ptr.i loc5,r23
- or r25=r25,r26 // construct PA | page properties
- mov cr.itir=r23
- mov cr.ifa=loc5
- ;;
- itr.i itr[r24]=r25
// done, switch back to virtual and return
mov r16=loc3 // r16= original psr
@@ -361,6 +347,8 @@
mov cr.ifa=loc5
;;
itr.i itr[r24]=r25
+ ;;
+ srlz.i
// done, switch back to virtual and return
mov r16=loc3 // r16= original psr
diff -r e8056a7091a7 xen/include/asm-ia64/linux-xen/linux/efi.h
--- a/xen/include/asm-ia64/linux-xen/linux/efi.h Mon Jul 14 19:31:54
2008 +0900
+++ b/xen/include/asm-ia64/linux-xen/linux/efi.h Mon Jul 14 19:33:20
2008 +0900
@@ -25,6 +25,7 @@
#include <asm/system.h>
#ifdef XEN
+#include <asm/meminit.h> /* GRANULEROUNDDOWN */
extern void * pal_vaddr;
#endif
@@ -474,6 +475,10 @@
} while (0)
#define XEN_EFI_RR_RESTORE(rr6, rr7) do { \
+ ia64_ptr(0x1 /*I*/, \
+ GRANULEROUNDDOWN( \
+ (unsigned long)pal_vaddr), \
+ IA64_GRANULE_SHIFT); \
set_one_rr_efi(6UL << 61, rr6); \
set_one_rr_efi(7UL << 61, rr7); \
} while (0)
diff -r e8056a7091a7 xen/include/asm-ia64/vmx_vcpu.h
--- a/xen/include/asm-ia64/vmx_vcpu.h Mon Jul 14 19:31:54 2008 +0900
+++ b/xen/include/asm-ia64/vmx_vcpu.h Mon Jul 14 19:33:20 2008 +0900
@@ -104,9 +104,9 @@
extern int vmx_vcpu_pend_interrupt(VCPU * vcpu, uint8_t vector);
extern void vcpu_load_kernel_regs(VCPU * vcpu);
extern void __vmx_switch_rr7(unsigned long rid, void *guest_vhpt,
- void *pal_vaddr, void *shared_arch_info);
+ void *shared_arch_info);
extern void vmx_switch_rr7(unsigned long rid, void *guest_vhpt,
- void *pal_vaddr, void *shared_arch_info);
+ void *shared_arch_info);
extern void vmx_ia64_set_dcr(VCPU * v);
extern void inject_guest_interruption(struct vcpu *vcpu, u64 vec);
extern void vmx_asm_bsw0(void);
On Mon, Jul 14, 2008 at 07:21:52PM +1000, Simon Horman wrote:
> Map EFI regions into the same place they are maped into in Linux
>
> This is because of an unfortunate problem with the way that EFI interacts
> with Kexec. The call to map the EFI regions may only be made once. This
> means that after Kexec the EFI regions must be mapped into the same region
> that they were mapped into prior to Kexec.
>
> This is not usually a problem when kexecing from xen to xen or from linux
> to linux, as the mapping will be the same. However when kexecing from xen
> to linux or linux to xen, the mapping is different, and the problem
> manifests.
>
> So far Magnus Damm and I have come up with three different ideas for
> resolving this problem.
>
> 1. Leave the EFI in physical mode
> - This is nice and simple
> - There is a potential performance hit, but PAL calls are not
> made very often, so it shouldn't be a problem
> - I have patches to do this, some of which are in the
> series that accompany this patch.
> - The SGI people tell me that it won't work on SN because
> it allows the OS to provide EFI (or SAL?) code.
>
> 2. Always map EFI into the space that Linux uses
> - Not so simple
> - Requires Xen to jump through some hoops
> - But leaves Linux unmodified
> - But it will break if Linux ever changes its mapping
> - This patch series implements this change
>
> 3. Always map EFI to some agreed space
> - Similar to 2. but less likely to break in the future
> - But it requires Xen and Linux to agree on a space to be used
> - Reqires both Xen and Linux to be modified
>
> Cc: Isaku Yamahata <yamahata@xxxxxxxxxxxxx>
> Cc: Tristan Gingold <tgingold@xxxxxxx>
> Cc: Alex Williamson <alex.williamson@xxxxxx>
> Cc: Aron Griffis <aron@xxxxxx>
> Signed-off-by: Simon Horman <horms@xxxxxxxxxxxx>
>
> Index: xen-unstable.hg/xen/arch/ia64/linux-xen/efi.c
> ===================================================================
> --- xen-unstable.hg.orig/xen/arch/ia64/linux-xen/efi.c 2008-07-14
> 17:41:55.000000000 +1000
> +++ xen-unstable.hg/xen/arch/ia64/linux-xen/efi.c 2008-07-14
> 17:41:57.000000000 +1000
> @@ -612,6 +612,17 @@ efi_enter_virtual_mode (void)
>
> for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
> md = p;
> +#ifdef XEN
> + if (md->attribute & EFI_MEMORY_RUNTIME) {
> + if (md->attribute & EFI_MEMORY_WB)
> + md->virt_addr = __IA64_EFI_CACHED_OFFSET|
> + md->phys_addr;
> + else if (md->attribute & (EFI_MEMORY_UC|EFI_MEMORY_WC|
> + EFI_MEMORY_WT))
> + md->virt_addr = __IA64_EFI_UNCACHED_OFFSET|
> + md->phys_addr;
> + }
> +#else
> if (md->attribute & EFI_MEMORY_RUNTIME) {
> /*
> * Some descriptors have multiple bits set, so the
> order of
> @@ -644,6 +655,7 @@ efi_enter_virtual_mode (void)
> #endif
> }
> }
> +#endif
> }
>
> status = efi_call_phys(__va(runtime->set_virtual_address_map),
> Index: xen-unstable.hg/xen/include/asm-ia64/xenpage.h
> ===================================================================
> --- xen-unstable.hg.orig/xen/include/asm-ia64/xenpage.h 2008-07-14
> 17:41:55.000000000 +1000
> +++ xen-unstable.hg/xen/include/asm-ia64/xenpage.h 2008-07-14
> 17:41:57.000000000 +1000
> @@ -97,14 +97,7 @@ static inline u64 pa_clear_uc(u64 paddr)
> /* It is sometimes very useful to have unsigned long as result. */
> #define __va_ul(x) ({xen_va _v; _v.l = (long) (x); _v.f.reg = -1; _v.l;})
>
> -/* __va_efi() should just call __va() until the use of
> - * __IA64_EFI_CACHED_OFFSET is activated in efi_enter_virtual_mode()
> - */
> -#if 0
> #define __va_efi(x) ((unsigned long)(x) | __IA64_EFI_CACHED_OFFSET)
> -#else
> -#define __va_efi(x) __va(x)
> -#endif
>
> #endif
> #endif /* _ASM_IA64_XENPAGE_H */
>
--
yamahata
_______________________________________________
Xen-ia64-devel mailing list
Xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-ia64-devel
|
Home