From: Tetsu Yamamoto <tyamamot@redhat.com>
Date: Wed, 12 Mar 2008 11:27:15 -0400
Subject: [xen] ia64: ftp stress test fixes between HVM/Dom0
Message-id: 47D7F653.1030700@redhat.com
O-Subject: Re: [RHEL5.2 PATCH] Dom0 panic while we run ftp test tool between HVM and Dom0.
Bugzilla: 426015

This is an additional patch for BZ#426015.
https://bugzilla.redhat.com/show_bug.cgi?id=426015

The patch which was previously posted and committed in -67 kernel fixes
the problem caused by wrong ssm_i emulation, but we found that there is
another cause to panic while ftp test.

The vcpu emulation of psr.ic in vcpu_rfi() has also a problem
that the bit is always changed to be set.  This can set wrong psr.i
value on return from interruption.

This additional patch fixes the rfi emulation in the same way as in the
upstream, but does not include any other parts than psr.ic and psr.i to
minimize the modification.
- [IA64] vcpu_set_psr added and used by vcpu_rfi.
http://xenbits.xensource.com/xen-unstable.hg?rev/409f9849fa68

I've tested this patch with kernel -84.  Unfortunately, the problem
still occur with this patch, but the possibility of panic can be less.

The time the ftp test program can run for is:
  w/o this patch: 24 hours
  w/  this patch: 72 hours

So I believe this patch can make some improvement.

Please review and ACK.

Regards,

Tetsu Yamamoto

Tetsu Yamamoto wrote::
> These patches fix BZ#426015.
> https://bugzilla.redhat.com/show_bug.cgi?id=426015
>
> Dom0 panic while FTP stress test is caused by wrong ssm_i emulation
> which is not barriered.  This means that interruption may not be
> delivered correctly and kernel hit BUG_ON like a BUG_ON(irqs_disabled()).
>
> To fix it, the following upstream patches are backported:
> - [IA64] Fix vdso paravirtualization.
> http://xenbits.xensource.com/ext/ia64/linux-2.6.18-xen.hg?rev/8257a66108c0
> - [IA64] Cleanup vdso paravirtualization.
> http://xenbits.xensource.com/ext/ia64/linux-2.6.18-xen.hg?rev/b64323b3a963
> - [IA64] Fix xen_ssm_i()
> http://xenbits.xensource.com/ext/ia64/linux-2.6.18-xen.hg?rev/0f9032c33df4
> - [IA64] barrier in xen_rsm_i
> http://xenbits.xensource.com/ext/ia64/linux-2.6.18-xen.hg?rev/76c09a3354a8
>
> I've tested these patches with kernel-2.6.18-58.el5, and confirmed that
> Dom0 hang does not occur on FTP stress test.
>
> Please review and ACK.
>
> Regards,
>
> Tetsu Yamamoto
>

Acked-by: Bill Burns <bburns@redhat.com>

diff --git a/arch/ia64/xen/vcpu.c b/arch/ia64/xen/vcpu.c
index 71b120d..9b967c3 100644
--- a/arch/ia64/xen/vcpu.c
+++ b/arch/ia64/xen/vcpu.c
@@ -1335,21 +1335,21 @@ IA64FAULT vcpu_rfi(VCPU * vcpu)
 {
 	// TODO: Only allowed for current vcpu
 	PSR psr;
-	u64 int_enable, ifs;
+	u64 ifs, int_enable, psr_ic;
 	REGS *regs = vcpu_regs(vcpu);
 
 	psr.i64 = PSCB(vcpu, ipsr);
+	int_enable = psr.ia64_psr.i;
+	psr_ic = psr.ia64_psr.ic;
 	if (psr.ia64_psr.cpl < 3)
 		psr.ia64_psr.cpl = 2;
-	int_enable = psr.ia64_psr.i;
+
 	if (psr.ia64_psr.dfh) {
 		PSCB(vcpu, vpsr_dfh) = 1;
 	} else {
 		psr.ia64_psr.dfh = PSCB(vcpu, hpsr_dfh);
 		PSCB(vcpu, vpsr_dfh) = 0;
 	}
-	if (psr.ia64_psr.ic)
-		PSCB(vcpu, interrupt_collection_enabled) = 1;
 	if (psr.ia64_psr.dt && psr.ia64_psr.rt && psr.ia64_psr.it)
 		vcpu_set_metaphysical_mode(vcpu, FALSE);
 	else
@@ -1368,9 +1368,9 @@ IA64FAULT vcpu_rfi(VCPU * vcpu)
 
 	regs->cr_ipsr = psr.i64;
 	regs->cr_iip = PSCB(vcpu, iip);
-	PSCB(vcpu, interrupt_collection_enabled) = 1;
 	vcpu_bsw1(vcpu);
-	vcpu->vcpu_info->evtchn_upcall_mask = !int_enable;
+	PSCB(vcpu, interrupt_collection_enabled) = !!(psr_ic);
+	vcpu->vcpu_info->evtchn_upcall_mask = !(int_enable);
 	return IA64_NO_FAULT;
 }