Re: Regarding patch "block/blk-mq: Don't complete locally if capacities are different"

[Qais Yousef]

On 08/05/24 22:54, MANISH PANDEY wrote:

> > > If no matching is required, it makes sense to set rq_affinity to 0. When
> > > matching is enabled, we need to rely on per-task iowait boost to help the
> > > requester to run at a bigger CPU, and naturally the completion will follow when
> > > rq_affinity=1. If the requester doesn't need the big perf, but the irq
> > > triggered on a bigger core, I struggle to understand why it is good for
> > > completion to run on bigger core without the requester also being on a similar
> > > bigger core to truly maximize perf.
> > 
> Not all the SoCs implements L3 as shared LLC. There are SoCs with L2 as LLC
> and not shared among all CPU clusters. So in this case, if we use rq=0, this
> would force to use a CPU, which doesn't shares L2 cache.
> Say in a system cpu[0-5] shares L2 as LLC and cpu[6-7] shares L2 as LLC,
> then any request from CPU[0-5] / CPU[6-7] would force to serve IRQ on CPUs
> which actually doesn't shares cache, would would result low performance.

For these systems rq_affinity=1 is what you want? the rq_affinity is not
supposed to be one size fits all. We wouldn't have different rq_affinity values
if one is supposed to work on all systems.