Xen devel - Jul 2009 - [PATCH V2] passthrough: deliver IRQs even if VCPU#0 is halted

Essentially nothing is changed from my previous patch.
Diffs are:
- split hvm_dirq_assist() for readability
- a little bit efficent for MP race condition

The attached patch is not so readable. 
I''ll show the following handmade diff for human.
====================================================================== void
hvm_dirq_assist(struct vcpu *v)
 {
     struct domain *d = v->domain;
     struct hvm_irq_dpci *hvm_irq_dpci = d->arch.hvm_domain.irq.dpci;
-    unsigned int irq;
+    unsigned int irq, i, dirq_mask_size;
+    unsigned long mask;
 
-    if ( !iommu_enabled || (v->vcpu_id != 0) || (hvm_irq_dpci == NULL) )
+    if ( !iommu_enabled || (hvm_irq_dpci == NULL) )
         return;
 
-    for ( irq = find_first_bit(hvm_irq_dpci->dirq_mask, d->nr_pirqs);
-          irq < d->nr_pirqs;
-          irq = find_next_bit(hvm_irq_dpci->dirq_mask, d->nr_pirqs, irq +
1) )
-    {
-        if ( !test_and_clear_bit(irq, hvm_irq_dpci->dirq_mask) )
-            continue;
-
-	__hvm_dirq_assist(d, hvm_irq_dpci, irq);
-    }
+    dirq_mask_size = BITS_TO_LONGS(d->nr_pirqs);
+    for (i = 0; i < dirq_mask_size; i++)
+        mask = xchg(&hvm_irq_dpci->dirq_mask[i], 0L);
+
+        while ( mask != 0 )
+        {
+            irq = find_first_set_bit(mask);
+            mask &= ~(1UL << irq);
+            irq += (i * BITS_PER_LONG);
+            if ( irq < d->nr_pirqs )
+                __hvm_dirq_assist(d, hvm_irq_dpci, irq);
+        }
+    }
 }
======================================================================
while d->nr_pirqs<=64, dirq_mask_size is 1 (d->nr_pirqs is normally
32).
Then the efficiency is not so worse than comparing v->vcpu_id != 0.

Thanks,
Kouya




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

On 10/07/2009 07:34, "Kouya Shimura" <kouya@jp.fujitsu.com>
wrote:
> Essentially nothing is changed from my previous patch.
> Diffs are:
> - split hvm_dirq_assist() for readability
> - a little bit efficent for MP race condition
Just target the i8259_target, and don''t bother with the MP efficiency
changes.

 -- Keir



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

Keir Fraser writes:
> Just target the i8259_target, and don''t bother with the MP
efficiency
> changes.
Sure. Attached. kdump works fine with this patch.
I was a little bit worried about i8259_target pointing the halted VCPU#0.

Thanks,
Kouya

Signed-off-by: Kouya Shimura <kouya@jp.fujitsu.com>



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

On 10/07/2009 10:27, "Kouya Shimura" <kouya@jp.fujitsu.com>
wrote:
> Keir Fraser writes:
>> Just target the i8259_target, and don''t bother with the MP
efficiency
>> changes.
> 
> Sure. Attached. kdump works fine with this patch.
> I was a little bit worried about i8259_target pointing the halted VCPU#0.
In your specific case it won''t when it matters (kdump kernel comes up,
reprograms IO-APIC which fixes i8259_target, then brings up PCI layer).

Really I think some of that dirq assist stuff should be put in a tasklet.
But let''s not worry about that now, and just patch it up this way.

 -- Keir



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