Re: [PATCH v2] drm/sched: Mark scheduler work queues with WQ_MEM_RECLAIM

[Matthew Brost <matthew.brost@intel.com>]

On Thu, Oct 24, 2024 at 01:44:41PM +0200, Nirmoy Das wrote:
> 
> On 10/24/2024 1:59 AM, Matthew Brost wrote:
> > drm_gpu_scheduler.submit_wq is used to submit jobs, jobs are in the path
> > of dma-fences, and dma-fences are in the path of reclaim. Mark scheduler
> > work queue with WQ_MEM_RECLAIM to ensure forward progress during
> > reclaim; without WQ_MEM_RECLAIM, work queues cannot make forward
> > progress during reclaim.
> >
> > v2:
> >  - Fixes tags (Philipp)
> >  - Reword commit message (Philipp)
> >
> > Cc: Luben Tuikov <ltuikov89@gmail.com>
> > Cc: Danilo Krummrich <dakr@kernel.org>
> > Cc: Philipp Stanner <pstanner@redhat.com>
> > Cc: stable@vger.kernel.org
> > Fixes: 34f50cc6441b ("drm/sched: Use drm sched lockdep map for submit_wq")
> > Fixes: a6149f039369 ("drm/sched: Convert drm scheduler to use a work queue rather than kthread")
> > Signed-off-by: Matthew Brost <matthew.brost@intel.com>
> 
> Acked-by: Nirmoy Das <nirmoy.das@intel.com>
> 
> Looks like Xe has a dependency on this now that xe->ordered_wq is allocated with  WQ_MEM_RECLAIM flag:
> 

Thanks for pointing this out.

I merged the Xe patches first not realizing this was going to break CI.
Hopefully I can merge this scheduler patch soon.

Matt

> https://intel-gfx-ci.01.org/tree/intel-xe/xe-pw-140135v2/bat-lnl-1/igt@xe_exec_fault_mode@twice-invalid-fault.html
> 
> > ---
> >  drivers/gpu/drm/scheduler/sched_main.c | 5 +++--
> >  1 file changed, 3 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/scheduler/sched_main.c b/drivers/gpu/drm/scheduler/sched_main.c
> > index 540231e6bac6..df0a5abb1400 100644
> > --- a/drivers/gpu/drm/scheduler/sched_main.c
> > +++ b/drivers/gpu/drm/scheduler/sched_main.c
> > @@ -1283,10 +1283,11 @@ int drm_sched_init(struct drm_gpu_scheduler *sched,
> >  		sched->own_submit_wq = false;
> >  	} else {
> >  #ifdef CONFIG_LOCKDEP
> > -		sched->submit_wq = alloc_ordered_workqueue_lockdep_map(name, 0,
> > +		sched->submit_wq = alloc_ordered_workqueue_lockdep_map(name,
> > +								       WQ_MEM_RECLAIM,
> >  								       &drm_sched_lockdep_map);
> >  #else
> > -		sched->submit_wq = alloc_ordered_workqueue(name, 0);
> > +		sched->submit_wq = alloc_ordered_workqueue(name, WQ_MEM_RECLAIM);
> >  #endif
> >  		if (!sched->submit_wq)
> >  			return -ENOMEM;