[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Xen-devel] [PATCHv1 1/2] passthrough: use per-interrupt lock when injecting an interrupt



The use of the per-domain event_lock in hvm_dirq_assist() does not scale
with many VCPUs or interrupts.

Add a per-interrupt lock to reduce contention.  When a interrupt for a
passthrough device is being setup or teared down, we must take both
the event_lock and this new lock.

Signed-off-by: David Vrabel <david.vrabel@xxxxxxxxxx>
---
 xen/drivers/passthrough/io.c | 34 +++++++++++++++++++++++-----------
 xen/include/xen/hvm/irq.h    |  1 +
 2 files changed, 24 insertions(+), 11 deletions(-)

diff --git a/xen/drivers/passthrough/io.c b/xen/drivers/passthrough/io.c
index bda9374..7c86c20 100644
--- a/xen/drivers/passthrough/io.c
+++ b/xen/drivers/passthrough/io.c
@@ -106,7 +106,7 @@ static void pt_pirq_softirq_reset(struct hvm_pirq_dpci 
*pirq_dpci)
 {
     struct domain *d = pirq_dpci->dom;
 
-    ASSERT(spin_is_locked(&d->event_lock));
+    ASSERT(spin_is_locked(&pirq_dpci->lock));
 
     switch ( cmpxchg(&pirq_dpci->state, 1 << STATE_SCHED, 0) )
     {
@@ -209,7 +209,6 @@ int pt_irq_create_bind(
     if ( pirq < 0 || pirq >= d->nr_pirqs )
         return -EINVAL;
 
- restart:
     spin_lock(&d->event_lock);
 
     hvm_irq_dpci = domain_get_irq_dpci(d);
@@ -237,6 +236,8 @@ int pt_irq_create_bind(
     }
     pirq_dpci = pirq_dpci(info);
 
+    spin_lock(&pirq_dpci->lock);
+
     /*
      * A crude 'while' loop with us dropping the spinlock and giving
      * the softirq_dpci a chance to run.
@@ -245,11 +246,11 @@ int pt_irq_create_bind(
      * would have spun forever and would do the same thing (wait to flush out
      * outstanding hvm_dirq_assist calls.
      */
-    if ( pt_pirq_softirq_active(pirq_dpci) )
+    while ( pt_pirq_softirq_active(pirq_dpci) )
     {
-        spin_unlock(&d->event_lock);
+        spin_unlock(&pirq_dpci->lock);
         cpu_relax();
-        goto restart;
+        spin_lock(&pirq_dpci->lock);
     }
 
     switch ( pt_irq_bind->irq_type )
@@ -301,6 +302,7 @@ int pt_irq_create_bind(
                 pirq_dpci->dom = NULL;
                 pirq_dpci->flags = 0;
                 pirq_cleanup_check(info, d);
+                spin_unlock(&pirq_dpci->lock);
                 spin_unlock(&d->event_lock);
                 return rc;
             }
@@ -311,6 +313,7 @@ int pt_irq_create_bind(
 
             if ( (pirq_dpci->flags & mask) != mask )
             {
+                spin_unlock(&pirq_dpci->lock);
                 spin_unlock(&d->event_lock);
                 return -EBUSY;
             }
@@ -331,6 +334,7 @@ int pt_irq_create_bind(
         dest_mode = !!(pirq_dpci->gmsi.gflags & VMSI_DM_MASK);
         dest_vcpu_id = hvm_girq_dest_2_vcpu_id(d, dest, dest_mode);
         pirq_dpci->gmsi.dest_vcpu_id = dest_vcpu_id;
+        spin_unlock(&pirq_dpci->lock);
         spin_unlock(&d->event_lock);
         if ( dest_vcpu_id >= 0 )
             hvm_migrate_pirqs(d->vcpu[dest_vcpu_id]);
@@ -351,6 +355,7 @@ int pt_irq_create_bind(
 
         if ( !digl || !girq )
         {
+            spin_unlock(&pirq_dpci->lock);
             spin_unlock(&d->event_lock);
             xfree(girq);
             xfree(digl);
@@ -412,6 +417,7 @@ int pt_irq_create_bind(
                 hvm_irq_dpci->link_cnt[link]--;
                 pirq_dpci->flags = 0;
                 pirq_cleanup_check(info, d);
+                spin_unlock(&pirq_dpci->lock);
                 spin_unlock(&d->event_lock);
                 xfree(girq);
                 xfree(digl);
@@ -419,6 +425,7 @@ int pt_irq_create_bind(
             }
         }
 
+        spin_unlock(&pirq_dpci->lock);
         spin_unlock(&d->event_lock);
 
         if ( iommu_verbose )
@@ -430,6 +437,7 @@ int pt_irq_create_bind(
     }
 
     default:
+        spin_unlock(&pirq_dpci->lock);
         spin_unlock(&d->event_lock);
         return -EOPNOTSUPP;
     }
@@ -481,6 +489,8 @@ int pt_irq_destroy_bind(
     pirq = pirq_info(d, machine_gsi);
     pirq_dpci = pirq_dpci(pirq);
 
+    spin_lock(&pirq_dpci->lock);
+
     if ( pt_irq_bind->irq_type != PT_IRQ_TYPE_MSI )
     {
         unsigned int bus = pt_irq_bind->u.pci.bus;
@@ -549,6 +559,7 @@ int pt_irq_destroy_bind(
         pirq_cleanup_check(pirq, d);
     }
 
+    spin_unlock(&pirq_dpci->lock);
     spin_unlock(&d->event_lock);
 
     if ( what && iommu_verbose )
@@ -566,6 +577,7 @@ int pt_irq_destroy_bind(
 
 void pt_pirq_init(struct domain *d, struct hvm_pirq_dpci *dpci)
 {
+    spin_lock_init(&dpci->lock);
     INIT_LIST_HEAD(&dpci->digl_list);
     dpci->gmsi.dest_vcpu_id = -1;
 }
@@ -621,7 +633,7 @@ int hvm_do_IRQ_dpci(struct domain *d, struct pirq *pirq)
     return 1;
 }
 
-/* called with d->event_lock held */
+/* called with pirq_dhci->lock held */
 static void __msi_pirq_eoi(struct hvm_pirq_dpci *pirq_dpci)
 {
     irq_desc_t *desc;
@@ -675,7 +687,7 @@ static void hvm_dirq_assist(struct domain *d, struct 
hvm_pirq_dpci *pirq_dpci)
 {
     ASSERT(d->arch.hvm_domain.irq.dpci);
 
-    spin_lock(&d->event_lock);
+    spin_lock(&pirq_dpci->lock);
     if ( test_and_clear_bool(pirq_dpci->masked) )
     {
         struct pirq *pirq = dpci_pirq(pirq_dpci);
@@ -687,7 +699,7 @@ static void hvm_dirq_assist(struct domain *d, struct 
hvm_pirq_dpci *pirq_dpci)
 
             if ( pirq_dpci->flags & HVM_IRQ_DPCI_GUEST_MSI )
             {
-                spin_unlock(&d->event_lock);
+                spin_unlock(&pirq_dpci->lock);
                 return;
             }
         }
@@ -695,7 +707,7 @@ static void hvm_dirq_assist(struct domain *d, struct 
hvm_pirq_dpci *pirq_dpci)
         if ( pirq_dpci->flags & HVM_IRQ_DPCI_GUEST_MSI )
         {
             vmsi_deliver_pirq(d, pirq_dpci);
-            spin_unlock(&d->event_lock);
+            spin_unlock(&pirq_dpci->lock);
             return;
         }
 
@@ -709,7 +721,7 @@ static void hvm_dirq_assist(struct domain *d, struct 
hvm_pirq_dpci *pirq_dpci)
         {
             /* for translated MSI to INTx interrupt, eoi as early as possible 
*/
             __msi_pirq_eoi(pirq_dpci);
-            spin_unlock(&d->event_lock);
+            spin_unlock(&pirq_dpci->lock);
             return;
         }
 
@@ -723,7 +735,7 @@ static void hvm_dirq_assist(struct domain *d, struct 
hvm_pirq_dpci *pirq_dpci)
         ASSERT(pt_irq_need_timer(pirq_dpci->flags));
         set_timer(&pirq_dpci->timer, NOW() + PT_IRQ_TIME_OUT);
     }
-    spin_unlock(&d->event_lock);
+    spin_unlock(&pirq_dpci->lock);
 }
 
 static void __hvm_dpci_eoi(struct domain *d,
diff --git a/xen/include/xen/hvm/irq.h b/xen/include/xen/hvm/irq.h
index 4c9cb20..8b8e461 100644
--- a/xen/include/xen/hvm/irq.h
+++ b/xen/include/xen/hvm/irq.h
@@ -91,6 +91,7 @@ struct hvm_irq_dpci {
 
 /* Machine IRQ to guest device/intx mapping. */
 struct hvm_pirq_dpci {
+    spinlock_t lock;
     uint32_t flags;
     unsigned int state;
     bool_t masked;
-- 
2.1.4


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel


 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.