Re: [RFC PATCH] x86, entry: Switch stacks on a paranoid entry from userspace

From: Andy Lutomirski
Date: Mon Nov 17 2014 - 14:57:49 EST

On Mon, Nov 17, 2014 at 10:50 AM, Borislav Petkov <bp@xxxxxxxxx> wrote:
> On Fri, Nov 14, 2014 at 09:56:38PM +0000, Luck, Tony wrote:
>> ...
>> But I think that means we need more than one of these structures ...
>> we may not be done with one before a new machine check occurs. So
>> we'd have to make an NMI-safe allocator to grab one for use inside
>> do_machine_check()
> Well, I think we might do something with a lockless list as it is being
> done in ghes.c.
> It allocates entries from its own pool in the NMI handler and
> llist_add's them to a list.
> Then, in user context it does llist_del_all and then looks at each of
> the elements at leisure and stress-free :-)
> Pool alloc/free is NMI-safe too so we should be good. It looks pretty
> clean, I'd give it a try.

Would it be worth making a decision on task_work_add vs. stack switching first?

Stack switching pros: all this lockless allocation stuff is completely
unnecessary, and it's plausible that the stack switching code will be
added anyway.

task_work_add pros: conceptually simpler mce.c diff.

Tony, did the code survive your new stress test?


>> General testing note - one thing I did see was that if inject 1000
>> errors at 0.3s interval from my ssh'd login ... the serial console
>> keeps streaming messages for about 40 seconds after my test says it is
>> all done. This might be a factor in the other tests I've been running
>> against the stack-switching code (especially with extra debug) ... at
>> some point __log_buf must get full - what happens then?
> Start gets overwritten AFAICR.
> --
> Regards/Gruss,
> Boris.
> Sent from a fat crate under my desk. Formatting is fine.
> --

Andy Lutomirski
AMA Capital Management, LLC
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at
Please read the FAQ at