Hi Luca,
On 08/04/2022 11:41, Luca Fancellu wrote:
On 8 Apr 2022, at 08:16, Jan Beulich <jbeul...@suse.com> wrote:
See the code comment. The higher the rate of vCPU-s migrating across
pCPU-s, the less useful this attempted optimization actually is. With
credit2 the migration rate looks to be unduly high even on mostly idle
systems, and hence on large systems lock contention here isn't very
difficult to observe.
Signed-off-by: Jan Beulich <jbeul...@suse.com>
I’ve tested this patch on a Juno board, starting Dom0, creating/destroying few
guests,
doing some networking from guests of different cpu pool and everything worked.
FWIW, we don't support PIRQ on Arm (alloc_pirq_struct() will always
return NULL). So evtchn_mode_pirqs() is basically a NOP.
This patch has the advantage for Arm to avoid taking the lock for
nothing. I will comment on the x86 part separately.
Cheers,
--
Julien Grall
