Re: [Xen-devel] [PATCH] NMI: continue in case of PCI SERR erros

["Jan Beulich" <JBeulich@xxxxxxxxxx>]

>>> On 25.02.11 at 12:27, Stefano Stabellini <stefano.stabellini@xxxxxxxxxxxxx>
wrote:
> On Fri, 25 Feb 2011, Jan Beulich wrote:
>> >>> On 25.02.11 at 11:53, Stefano Stabellini 
>> >>> <stefano.stabellini@xxxxxxxxxxxxx>
>> wrote:
>> > Memory parity error is only valid for IBM PC-AT, newer machines use
>> > bit 7 (0x80) of 0x61 port for PCI SERR. While memory errors are
>> > usually reported via MCE.
>> > 
>> > Rename the memory parity error handler to pci serr handler and
>> > print a warning and continue instead of crashing.
>> > 
>> > Signed-off-by: Stefano Stabellini <stefano.stabellini@xxxxxxxxxxxxx>
>> > 
>> > ---
>> > 
>> > diff -r 598d1fc295b6 xen/arch/x86/traps.c
>> > --- a/xen/arch/x86/traps.c Thu Feb 24 09:33:19 2011 +0000
>> > +++ b/xen/arch/x86/traps.c Fri Feb 25 10:37:47 2011 +0000
>> > @@ -3075,23 +3075,12 @@ static void nmi_dom0_report(unsigned int
>> >      send_guest_trap(d, 0, TRAP_nmi);
>> >  }
>> >  
>> > -static void mem_parity_error(struct cpu_user_regs *regs)
>> > +static void pci_serr_error(struct cpu_user_regs *regs)
>> >  {
>> > -    switch ( opt_nmi[0] )
>> > -    {
>> > -    case 'd': /* 'dom0' */
>> > -        nmi_dom0_report(_XEN_NMIREASON_parity_error);
>> > -    case 'i': /* 'ignore' */
>> > -        break;
>> > -    default:  /* 'fatal' */
>> > -        console_force_unlock();
>> > -        printk("\n\nNMI - MEMORY ERROR\n");
>> > -        fatal_trap(TRAP_nmi, regs);
>> > -    }
>> > +    console_force_unlock();
>> > +    printk("\n\nNMI - PCI system error (SERR)\n");
>> 
>> Why do you remove the reporting of the event to Dom0? I'd rather
>> see this part stay and the public headers getting adjusted in a
>> compatible way.
>> 
> 
> Because it might not have anything to do with dom0: I wrote this patch

As you say: might.

> to fix http://bugzilla.xensource.com/bugzilla/show_bug.cgi?id=1744, that
> is due to Xen trying to mask an MSI-X of a device assigned to a guest on
> shutdown. The device seems to be in a state in which it refuses any
> interactions so it sends a PCI SERR that becomes an NMI and used to
> crash Xen.

I'm certainly agreeing to the part that avoids the crash.

Jan


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