Re: [PATCH 2/8] block/mq-deadline: serialize request dispatching

[Jens Axboe <axboe@kernel.dk>]

On 1/23/24 11:36 AM, Bart Van Assche wrote:
> On 1/23/24 09:34, Jens Axboe wrote:
>> +    struct {
>> +        spinlock_t lock;
>> +        spinlock_t zone_lock;
>> +    } ____cacheline_aligned_in_smp;
> 
> It is not clear to me why the ____cacheline_aligned_in_smp attribute
> is applied to the two spinlocks combined? Can this cause both spinlocks
> to end up in the same cache line? If the ____cacheline_aligned_in_smp
> attribute would be applied to each spinlock separately, could that
> improve performance even further? Otherwise this patch looks good to me,
> hence:

It is somewhat counterintuitive, but my testing shows that there's no
problem with them in the same cacheline. Hence I'm reluctant to move
them out of the struct and align both of them, as it'd just waste memory
for seemingly no runtime benefit.

> Reviewed-by: Bart Van Assche <bvanassche@acm.org>

Thanks!

-- 
Jens Axboe