On 22/03/2018 02:29, Emilio G. Cota wrote:
> It is indeed hard to see a large impact on performance given the
> large size of our critical sections. But hey, rcu_read_unlock
> goes down from 0.24% to 0.08% of execution time when booting
> aarch64 linux!
I expect something a little better than 0.15% from virtio, especially
once I re-enable MemoryRegionCache in 2.13.
>> + int main(void) {
>> + syscall(__NR_membarrier, MEMBARRIER_CMD_QUERY, 0);
>> + syscall(__NR_membarrier, MEMBARRIER_CMD_SHARED, 0);
>> + }
>
> I think we should also check here that MEMBARRIER_CMD_SHARED is
> actually supported;
Checking run-time constraints at compile-time is a bit pointless. My
idea was to add a third mode where the choice is done at run-time, but
that wouldn't have made 2.12.
Paolo
