Re: [Xen-devel] PML causing race condition during guest bootstorm and host crash on Broadwell cpu.

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 13.02.17 at 17:56, <JBeulich@xxxxxxxx> wrote:
>>>> On 13.02.17 at 17:32, <anshul.makkar@xxxxxxxxxx> wrote:
>> On 09/02/17 16:22, Jan Beulich wrote:
>>> Mind giving the attached patch a try (which admittedly was only
>>> lightly tested so far - in particular I haven't seen the second of
>>> the debug messages being logged, yet)?
>> Patch looks promising. I couldn't do much thorough testing, but initial 
>> reboot cycles (around 20 reboots of 32 VMS) went fine.
> 
> Thanks. The most interesting part though is whether the 2nd of the
> two log messages ever showed up. I any event I'll submit the cleaned
> up patch, for more formal discussion of the approach to happen there.

Actually, no, while putting together the commit message I thought
of a second situation which likely would also need addressing: If
we bypass __context_switch() also on the context-switch-in path
(because of scheduling the same vCPU again) the CPU may also
have lost control of the VMCS. There's no context switch hook
though to place the reload invocation in, so I'll first have to think
about what the best model here would be (of course I'd be more
than happy about suggestions).

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel