Re: [PATCH v3 1/5] drm/xe/gt: Use GT ordered workqueue for wedging

[Rodrigo Vivi <rodrigo.vivi@intel.com>]

On Mon, Sep 07, 2026 at 03:06:17PM +0200, Raag Jadav wrote:
> On Mon, Sep 07, 2026 at 02:02:59PM +0200, Michal Wajdeczko wrote:
> > On 9/7/2026 10:34 AM, Raag Jadav wrote:
> > > Currently, xe_gt_declare_wedged() stops GuC CT and initiates exec queue
> > > teardown synchronously. This can result in tdr timeout in cases where
> > > the device is declared wedged while jobs are still in-flight.
> > > 
> > > Introduce a worker for GT specific wedge handling and queue the teardown
> > > on GT ordered workqueue, so we don't disrupt the scheduler while jobs
> > > are still in-flight.
> > > 
> > > Fixes: c9474b726b93 ("drm/xe: Wedge the entire device")
> > > Signed-off-by: Raag Jadav <raag.jadav@intel.com>
> > > ---
> > > v2: Split fixes into separate patches (Rodrigo, Michal)
> > > ---
> > >  drivers/gpu/drm/xe/xe_gt.c       | 17 +++++++++++++++--
> > >  drivers/gpu/drm/xe/xe_gt_types.h |  6 ++++++
> > >  2 files changed, 21 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/drivers/gpu/drm/xe/xe_gt.c b/drivers/gpu/drm/xe/xe_gt.c
> > > index 478e047031f4..200e34331c19 100644
> > > --- a/drivers/gpu/drm/xe/xe_gt.c
> > > +++ b/drivers/gpu/drm/xe/xe_gt.c
> > > @@ -173,6 +173,7 @@ static void xe_gt_enable_comp_1wcoh(struct xe_gt *gt)
> > >  }
> > >  
> > >  static void gt_reset_worker(struct work_struct *w);
> > > +static void gt_wedge_worker(struct work_struct *w);
> > >  
> > >  static int emit_job_sync(struct xe_exec_queue *q, struct xe_bb *bb,
> > >  			 long timeout_jiffies, bool force_reset)
> > > @@ -704,6 +705,8 @@ static void xe_gt_fini(void *arg)
> > >  	struct xe_gt *gt = arg;
> > >  	int i;
> > >  
> > > +	disable_work_sync(&gt->wedge.worker);
> > > +
> > >  	if (disable_work_sync(&gt->reset.worker))
> > >  		/*
> > >  		 * If gt_reset_worker was halted from executing, take care of
> > > @@ -723,6 +726,7 @@ int xe_gt_init(struct xe_gt *gt)
> > >  	int i;
> > >  
> > >  	INIT_WORK(&gt->reset.worker, gt_reset_worker);
> > > +	INIT_WORK(&gt->wedge.worker, gt_wedge_worker);
> > 
> > hmm, shouldn't this be done in xe_gt_alloc() ?
> 
> The worker is touching hardware, which I'm assuming shouldn't be done
> before xe_device_probe()?
> 
> Raag
> 
> > as xe_gt_init_early() seems to be doing something else and much later
> > 
> > >  
> > >  	for (i = 0; i < XE_ENGINE_CLASS_MAX; ++i) {
> > >  		gt->ring_ops[i] = xe_ring_ops_get(gt, i);
> > > @@ -1005,6 +1009,14 @@ void xe_gt_reset_async(struct xe_gt *gt)
> > >  		xe_pm_runtime_put(xe);
> > >  }
> > >  
> > > +static void gt_wedge_worker(struct work_struct *w)
> > > +{
> > > +	struct xe_gt *gt = container_of(w, typeof(*gt), wedge.worker);
> > > +
> > > +	xe_uc_declare_wedged(&gt->uc);
> > > +	xe_tlb_inval_reset(&gt->tlb_inval);
> > > +}
> > > +
> > >  void xe_gt_suspend_prepare(struct xe_gt *gt)
> > >  {
> > >  	xe_uc_suspend_prepare(&gt->uc);
> > > @@ -1195,6 +1207,7 @@ void xe_gt_declare_wedged(struct xe_gt *gt)
> > >  {
> > >  	xe_gt_assert(gt, gt_to_xe(gt)->wedged.mode);
> > >  
> > > -	xe_uc_declare_wedged(&gt->uc);
> > > -	xe_tlb_inval_reset(&gt->tlb_inval);
> > > +	/* Prevent queueing uninitialized worker if hit during probe */

Perhaps this comment itself proves Michal's point?!

I believe it would be cleaner if this would be per device and not per_gt
and already INIT worker inside device_early_probe...

> > > +	if (gt->wedge.worker.func)
> > > +		queue_work(gt->ordered_wq, &gt->wedge.worker);

This can race with other wedge places now, like:
prepare_device_for_reset()

we should perhaps flush all workers there or whenever
we are in the sync path?! or perhaps move all to the
queue?

> > >  }
> > > diff --git a/drivers/gpu/drm/xe/xe_gt_types.h b/drivers/gpu/drm/xe/xe_gt_types.h
> > > index 628911346455..b11ac52b0ab7 100644
> > > --- a/drivers/gpu/drm/xe/xe_gt_types.h
> > > +++ b/drivers/gpu/drm/xe/xe_gt_types.h
> > > @@ -233,6 +233,12 @@ struct xe_gt {
> > >  		struct work_struct worker;
> > >  	} reset;
> > >  
> > > +	/** @wedge: state for GT wedge */
> > > +	struct {
> > > +		/** @wedge.worker: worker for GT wedge to be done async */
> > > +		struct work_struct worker;
> > > +	} wedge;
> > > +
> > >  	/** @tlb_inval: TLB invalidation state */
> > >  	struct xe_tlb_inval tlb_inval;
> > >  
> >