OvmfPkg: End timer interrupt later to avoid stack overflow under load

RestoreTPL called while at TPL_HIGH_LEVEL unconditionally enables
interrupts even if called in interrupt handler. That opens a window while
interrupt is not completely handled but another interrupt could be
accepted.

If a VM starts on a heavily loaded host hundreds of periodic timer
interrupts might be queued while vCPU is descheduled (the behavior is
typical for a Xen host). The next time vCPU is scheduled again all of them
get delivered back to back causing OVMF to accept each one without
finishing a previous one and cleaning up the stack. That quickly results
in stack overflow and a triple fault.

Fix it by postponing sending EOI until we finished processing the current
tick giving interrupt handler opportunity to clean up the stack before
accepting the next tick.

Signed-off-by: Igor Druzhinin <igor.druzhinin@citrix.com>
Message-Id: <1592275782-9369-1-git-send-email-igor.druzhinin@citrix.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Ref: https://bugzilla.tianocore.org/show_bug.cgi?id=2815
Acked-by: Laszlo Ersek <lersek@redhat.com>
[lersek@redhat.com: add BZ ref; rewrap msg to silence PatchCheck.py]

1 files changed, 3 insertions, 2 deletions

diff --git a/OvmfPkg/XenTimerDxe/XenTimerDxe.c b/OvmfPkg/XenTimerDxe/XenTimerDxe.c
index 9f9e04766c..0bec59382b 100644
--- a/OvmfPkg/XenTimerDxe/XenTimerDxe.c
+++ b/OvmfPkg/XenTimerDxe/XenTimerDxe.c
@@ -61,8 +61,6 @@ TimerInterruptHandler (
 

   OriginalTPL = gBS->RaiseTPL (TPL_HIGH_LEVEL);

 

-  SendApicEoi();

-

   if (mTimerNotifyFunction != NULL) {

     //

     // @bug : This does not handle missed timer interrupts

@@ -71,6 +69,9 @@ TimerInterruptHandler (
   }

 

   gBS->RestoreTPL (OriginalTPL);

+

+  DisableInterrupts ();

+  SendApicEoi ();

 }

 

 /**