[PATCH 09/33] block: Protect against concurrent isolated cpuset change

[Jens Axboe]

On 12/31/25 7:02 AM, Frederic Weisbecker wrote:
> Le Mon, Dec 29, 2025 at 05:37:29PM -0700, Jens Axboe a ?crit :
>> On 12/24/25 6:44 AM, Frederic Weisbecker wrote:
>>> The block subsystem prevents running the workqueue to isolated CPUs,
>>> including those defined by cpuset isolated partitions. Since
>>> HK_TYPE_DOMAIN will soon contain both and be subject to runtime
>>> modifications, synchronize against housekeeping using the relevant lock.
>>>
>>> For full support of cpuset changes, the block subsystem may need to
>>> propagate changes to isolated cpumask through the workqueue in the
>>> future.
>>>
>>> Signed-off-by: Frederic Weisbecker <frederic at kernel.org>
>>> ---
>>>  block/blk-mq.c | 6 +++++-
>>>  1 file changed, 5 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/block/blk-mq.c b/block/blk-mq.c
>>> index 1978eef95dca..0037af1216f3 100644
>>> --- a/block/blk-mq.c
>>> +++ b/block/blk-mq.c
>>> @@ -4257,12 +4257,16 @@ static void blk_mq_map_swqueue(struct request_queue *q)
>>>  
>>>  		/*
>>>  		 * Rule out isolated CPUs from hctx->cpumask to avoid
>>> -		 * running block kworker on isolated CPUs
>>> +		 * running block kworker on isolated CPUs.
>>> +		 * FIXME: cpuset should propagate further changes to isolated CPUs
>>> +		 * here.
>>>  		 */
>>> +		rcu_read_lock();
>>>  		for_each_cpu(cpu, hctx->cpumask) {
>>>  			if (cpu_is_isolated(cpu))
>>>  				cpumask_clear_cpu(cpu, hctx->cpumask);
>>>  		}
>>> +		rcu_read_unlock();
>>
>> Want me to just take this one separately and get it out of your hair?
>> Doesn't seem to have any dependencies.
> 
> The patch could be applied alone but the rest of the patchset needs it,
> otherwise it may dereference freed memory. So I fear it needs to stay
> within the lot.
> 
> I appreciate the offer though. But an ack would help, even if I must admit
> this single patch (which doesn't change current behaviour) leaves a
> bitter taste because complete handling of cpuset isolated partition change
> will require more work.

That's fine too:

Acked-by: Jens Axboe <axboe at kernel.dk>

> Speaking of, is there a way that I missed to define/overwrite the above
> hctx->cpumask on runtime?

Only spot where it's set/manipulated right now is as part of setting up
the hctx <-> ctx queue mappings, when the queue is configured (or
reconfigured).

-- 
Jens Axboe