Re: [PATCH 1/6] workqueue: Inherit NOIO and NOFS alloc flags
From: Haakon Bugge
Date: Tue May 14 2024 - 09:49:03 EST
> Hello,
>
> On Mon, May 13, 2024 at 02:53:41PM +0200, Håkon Bugge wrote:
> diff --git a/include/linux/workqueue.h b/include/linux/workqueue.h
> index 158784dd189ab..09ecc692ffcae 100644
> --- a/include/linux/workqueue.h
> +++ b/include/linux/workqueue.h
> @@ -398,6 +398,8 @@ enum wq_flags {
> __WQ_DRAINING = 1 << 16, /* internal: workqueue is draining */
> __WQ_ORDERED = 1 << 17, /* internal: workqueue is ordered */
> __WQ_LEGACY = 1 << 18, /* internal: create*_workqueue() */
> + __WQ_NOIO = 1 << 19, /* internal: execute work with NOIO */
> + __WQ_NOFS = 1 << 20, /* internal: execute work with NOFS */
>
> I don't quite understand how this is supposed to be used. The flags are
> marked internal but nothing actually sets them. Looking at later patches, I
> don't see any usages either. What am I missing?
Hi Tejun,
Thank you for so quickly looking into this. You are quite right. During re-basing, I missed a hunk from alloc_workqueue(), where I sample current->flags for PF_MALLOC_{NOIO,NOFS} bits and sets the corresponding bits in wq->flags. Fixed in v2.
> @@ -3184,6 +3189,12 @@ __acquires(&pool->lock)
>
> lockdep_copy_map(&lockdep_map, &work->lockdep_map);
> #endif
> + /* Set inherited alloc flags */
> + if (use_noio_allocs)
> + noio_flags = memalloc_noio_save();
> + if (use_nofs_allocs)
> + nofs_flags = memalloc_nofs_save();
> +
> /* ensure we're on the correct CPU */
> WARN_ON_ONCE(!(pool->flags & POOL_DISASSOCIATED) &&
> raw_smp_processor_id() != pool->cpu);
> @@ -3320,6 +3331,12 @@ __acquires(&pool->lock)
>
> /* must be the last step, see the function comment */
> pwq_dec_nr_in_flight(pwq, work_data);
> +
> + /* Restore alloc flags */
> + if (use_nofs_allocs)
> + memalloc_nofs_restore(nofs_flags);
> + if (use_noio_allocs)
> + memalloc_noio_restore(noio_flags);
>
> Also, this looks like something that the work function can do on entry and
> before exit, no?
It _can_ be done in the work functions, but that will be a code sprawl. Only in RDS, we have the following worker functions:
rds_ib_odp_mr_worker();
rds_ib_mr_pool_flush_worker()
rds_ib_odp_mr_worker()
rds_tcp_accept_worker()
rds_connect_worker()
rds_send_worker()
rds_recv_worker()
rds_shutdown_worker()
adding the ones from ib_cm, rdma_cm, mlx5_ib, and mlx5_core, I strongly prefer to have it in one place.
Thxs, Håkon