x86/pt: mask MSI vectors on unbind

When a MSI device with per-vector masking capabilities is detected or
added to Xen all the vectors are masked when initializing it. This
implies that the first time the interrupt is bound to a domain it's
masked.

This however only applies to the first time the interrupt is bound
because neither the unbind nor the pirq unmap will mask the vector
again. In order to fix this re-mask the interrupt when unbinding it
from a guest. This makes sure that pairs of bind/unbind will always
get the same masking state.

Note that no issues have been reported regarding this behavior because
QEMU always uses the newly introduced XEN_PT_GFLAGSSHIFT_UNMASKED when
binding interrupts, so it's always unmasked.

Signed-off-by: Roger Pau Monné <roger.pau@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>

diff --git a/xen/drivers/passthrough/io.c b/xen/drivers/passthrough/io.c
index 8f16e6c0a5e8afc2164a49f94bc4dd8f863ac3d1..bab3aa349a8e093a90f2adeb45bc368566f27818 100644 (file)
--- a/xen/drivers/passthrough/io.c
+++ b/xen/drivers/passthrough/io.c
@@ -645,7 +645,22 @@ int pt_irq_destroy_bind(
         }
         break;
     case PT_IRQ_TYPE_MSI:
+    {
+        unsigned long flags;
+        struct irq_desc *desc = domain_spin_lock_irq_desc(d, machine_gsi,
+                                                          &flags);
+
+        if ( !desc )
+            return -EINVAL;
+        /*
+         * Leave the MSI masked, so that the state when calling
+         * pt_irq_create_bind is consistent across bind/unbinds.
+         */
+        guest_mask_msi_irq(desc, true);
+        spin_unlock_irqrestore(&desc->lock, flags);
         break;
+    }
+
     default:
         return -EOPNOTSUPP;
     }