linux-kernel - Re: [PATCH v2 2/5] mq-deadline: switch to use elevator lock

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <2ada440d-5121-6c3e-71db-1f8eb63864a7@huaweicloud.com>
Date: Thu, 31 Jul 2025 15:14:06 +0800
From: Yu Kuai <yukuai1@...weicloud.com>
To: Damien Le Moal <dlemoal@...nel.org>, Yu Kuai <yukuai1@...weicloud.com>,
 Hannes Reinecke <hare@...e.de>, jack@...e.cz, tj@...nel.org,
 josef@...icpanda.com, axboe@...nel.dk
Cc: cgroups@...r.kernel.org, linux-block@...r.kernel.org,
 linux-kernel@...r.kernel.org, yi.zhang@...wei.com, yangerkun@...wei.com,
 johnny.chenyi@...wei.com, "yukuai (C)" <yukuai3@...wei.com>
Subject: Re: [PATCH v2 2/5] mq-deadline: switch to use elevator lock

Hi,

在 2025/07/31 15:04, Damien Le Moal 写道:
> On 7/31/25 3:32 PM, Yu Kuai wrote:
>> Hi,
>>
>> 在 2025/07/31 14:22, Damien Le Moal 写道:
>>> On 7/31/25 3:20 PM, Hannes Reinecke wrote:
>>>> On 7/30/25 10:22, Yu Kuai wrote:
>>>>> From: Yu Kuai <yukuai3@...wei.com>
>>>>>
>>>>> Replace the internal spinlock 'dd->lock' with the new spinlock in
>>>>> elevator_queue, there are no functional changes.
>>>>>
>>>>> Signed-off-by: Yu Kuai <yukuai3@...wei.com>
>>>>> ---
>>>>>     block/mq-deadline.c | 58 +++++++++++++++++++++------------------------
>>>>>     1 file changed, 27 insertions(+), 31 deletions(-)
>>>>>
>>>>> diff --git a/block/mq-deadline.c b/block/mq-deadline.c
>>>>> index 9ab6c6256695..2054c023e855 100644
>>>>> --- a/block/mq-deadline.c
>>>>> +++ b/block/mq-deadline.c
>>>>> @@ -101,7 +101,7 @@ struct deadline_data {
>>>>>         u32 async_depth;
>>>>>         int prio_aging_expire;
>>>>>     -    spinlock_t lock;
>>>>> +    spinlock_t *lock;
>>>>>     };
>>>>>       /* Maps an I/O priority class to a deadline scheduler priority. */
>>>>> @@ -213,7 +213,7 @@ static void dd_merged_requests(struct request_queue *q,
>>>>> struct request *req,
>>>>>         const u8 ioprio_class = dd_rq_ioclass(next);
>>>>>         const enum dd_prio prio = ioprio_class_to_prio[ioprio_class];
>>>>>     -    lockdep_assert_held(&dd->lock);
>>>>> +    lockdep_assert_held(dd->lock);
>>>>>           dd->per_prio[prio].stats.merged++;
>>>>>     @@ -253,7 +253,7 @@ static u32 dd_queued(struct deadline_data *dd, enum
>>>>> dd_prio prio)
>>>>>     {
>>>>>         const struct io_stats_per_prio *stats = &dd->per_prio[prio].stats;
>>>>>     -    lockdep_assert_held(&dd->lock);
>>>>> +    lockdep_assert_held(dd->lock);
>>>>>           return stats->inserted - atomic_read(&stats->completed);
>>>>>     }
>>>>> @@ -323,7 +323,7 @@ static struct request *__dd_dispatch_request(struct
>>>>> deadline_data *dd,
>>>>>         enum dd_prio prio;
>>>>>         u8 ioprio_class;
>>>>>     -    lockdep_assert_held(&dd->lock);
>>>>> +    lockdep_assert_held(dd->lock);
>>>>>           if (!list_empty(&per_prio->dispatch)) {
>>>>>             rq = list_first_entry(&per_prio->dispatch, struct request,
>>>>> @@ -434,7 +434,7 @@ static struct request
>>>>> *dd_dispatch_prio_aged_requests(struct deadline_data *dd,
>>>>>         enum dd_prio prio;
>>>>>         int prio_cnt;
>>>>>     -    lockdep_assert_held(&dd->lock);
>>>>> +    lockdep_assert_held(dd->lock);
>>>>>           prio_cnt = !!dd_queued(dd, DD_RT_PRIO) + !!dd_queued(dd,
>>>>> DD_BE_PRIO) +
>>>>>                !!dd_queued(dd, DD_IDLE_PRIO);
>>>>> @@ -466,10 +466,9 @@ static struct request *dd_dispatch_request(struct
>>>>> blk_mq_hw_ctx *hctx)
>>>>>         struct request *rq;
>>>>>         enum dd_prio prio;
>>>>>     -    spin_lock(&dd->lock);
>>>>>         rq = dd_dispatch_prio_aged_requests(dd, now);
>>>>>         if (rq)
>>>>> -        goto unlock;
>>>>> +        return rq;
>>>>>           /*
>>>>>          * Next, dispatch requests in priority order. Ignore lower priority
>>>>> @@ -481,9 +480,6 @@ static struct request *dd_dispatch_request(struct
>>>>> blk_mq_hw_ctx *hctx)
>>>>>                 break;
>>>>>         }
>>>>>     -unlock:
>>>>> -    spin_unlock(&dd->lock);
>>>>> -
>>>>>         return rq;
>>>>>     }
>>>>>     @@ -538,9 +534,9 @@ static void dd_exit_sched(struct elevator_queue *e)
>>>>>             WARN_ON_ONCE(!list_empty(&per_prio->fifo_list[DD_READ]));
>>>>>             WARN_ON_ONCE(!list_empty(&per_prio->fifo_list[DD_WRITE]));
>>>>>     -        spin_lock(&dd->lock);
>>>>> +        spin_lock(dd->lock);
>>>>>             queued = dd_queued(dd, prio);
>>>>> -        spin_unlock(&dd->lock);
>>>>> +        spin_unlock(dd->lock);
>>>>>               WARN_ONCE(queued != 0,
>>>>>                   "statistics for priority %d: i %u m %u d %u c %u\n",
>>>>
>>>> Do you still need 'dd->lock'? Can't you just refer to the lock from the
>>>> elevator_queue structure directly?
>>>
>>> Indeed. Little inline helpers for locking/unlocking q->elevator->lock would be
>>> nice.
>>
>> How about the first patch to factor out inline helpers like dd_lock()
>> and dd_unlock(), still use dd->lock without any functional changes, and
>> then switch to use q->elevator->lock in the next patch? (same for bfq)
> 
> Patch one can introduce elv->lock and the helpers, then patch 2 use the helpers
> to replace dd->lock. Just don't say "no functional change" in the commit
> message and rather explain that things keep working the same way as before, but
> using a different lock. That will address Bart's comment too.
> And same for bfq in patch 3.
> 
Ok, this is what I did in the first RFC version:

https://lore.kernel.org/all/20250530080355.1138759-3-yukuai1@huaweicloud.com/

I somehow convince myself using dd->lock is better. :(
Will change this in the next version.

Thanks,
Kuai

>