[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [PATCH 3/3] x86/vMSI-X: also snoop REP MOVS



>>> On 28.04.16 at 12:34, <andrew.cooper3@xxxxxxxxxx> wrote:
> On 28/04/16 10:50, Jan Beulich wrote:
>> ... as at least certain versions of Windows use such to update the
>> MSI-X table. However, to not overly complicate the logic for now
>> - only EFLAGS.DF=0 is being handled,
>> - only updates not crossing MSI-X table entry boundaries are handled.
>>
>> Signed-off-by: Jan Beulich <jbeulich@xxxxxxxx>
> 
> IMO, under those circumstances we should domain_crash() and be rather
> loud on the console.
> 
> Perhaps in a debug build only, but this will be a very unpleasant issue
> to debug for whomever finds an OS which falls into of those unhandled
> situations.

I disagree: At the time we snoop accesses, we don't know whether
they're targeting any MSI-X table entry. And we shouldn't crash the
guest just because it accessed _something else_ in a way the snoop
logic doesn't support. If any unsupported access gets done by a
guest, all that'll happen is that MSI-X again doesn't work inside the
guest, i.e. the same situation as the previous and this patches are
trying to deal with. (And yes, the debugging wasn't really pleasant,
but that's more because the original authors didn't design the whole
thing properly, and any other solution I could think of would have
caused issues with the qemu/Xen interface logic.)

Jan


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

 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.