Re: [PATCH] x86: don't change affinity with interrupt unmasked

[Andrew Cooper <andrew.cooper3@citrix.com>]

On 20/03/15 16:40, Jan Beulich wrote:
> With ->startup unmasking the IRQ, setting the affinity afterwards
> without masking the IRQ again is invalid namely for MSI (which can't
> have their affinity updated atomically).
>
> Signed-off-by: Jan Beulich<jbeulich@suse.com>

Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>

> ---
> Changing the affinity of non-maskable MSI IRQs seems bogus too

Agreed.  Their affinity can clearly only be changed safely by a device 
driver which can guarantee that an interrupt will not be generated 
during the vulnerable period.

This further implies that Xen can't even safely set an affinity to start 
with...

> , but I
> can't immediately see what we can do about this (better than disabling
> affinity changes for them).

If we used interrupt remapping properly (which we don't), we could 
update the effective affinity without changing any device configuration, 
but this still doesn't provide a solution for the many systems out there 
without (functional) interrupt remapping.

~Andrew