On Thu, Feb 23, 2012 at 02:38:42PM -0800, Linus Torvalds wrote:
> On Thu, Feb 23, 2012 at 2:27 PM, Willy Tarreau <[email protected]> wrote:
> >
> > OK so indeed I will only be able to check that it boots :-/
>
> Well, we could do some trivial test-harness that just forces the issue
> with regular timer interrupts (and even without AES-NI). I think Peter
> talked about that when we were trying to hunt it down - but I think he
> was then able to reproduce the problem without anything special and we
> dropped it.
>
> Essentially, just doing something like
>
> if (irq_fpu_usable()) {
> kernel_fpu_begin();
> kernel_fpu_end();
> }
>
> in do_irq() and do_softirq() would stress-test things even without
> wireless, and even without AES-NI.
>
> You'd still need an x86-32 machine to test on, because x86-64 was
> immune to this issue.
>
> But yeah, the impact of this seems to be small enough that for older
> kernels (which are likely used on older systems for maintenance
> anyway) disabling AES-NI on x86-32 really might be the way to go.
I think that's already the case, 2.6.32 has the following depends for
CRYPTO_AES_NI_INTEL:
depends on (X86 || UML_X86) && 64BIT
It was this way until commit 0d258efb (crypto: aesni-intel - Ported
implementation to x86-32) which showed up in 2.6.38.
So we should be safe for 2.6.32 no changes needed, right?
greg k-h
--
To unsubscribe from this list: send the line "unsubscribe stable" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
