Re: [PATCH] drm/xe: Use a dedicated workqueue for VM rebinds

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

On Mon, Oct 19, 2026 at 11:01:42PM +0000, Jia Yao wrote:
> VM rebind workers may block while handling HMM faults or waiting for VM
> fences. Running them on ordered_wq can stall user-fence notifications
> and VM TLB timeout work queued behind them, preventing forward progress.

I don't see how ' VM TLB timeout work' is related. Those are scheduled
on the GT ordered work queue. 

> 
> Move rebind workers to a dedicated ordered workqueue. This keeps rebinds
> serialized across VMs while allowing fence-related work to run
> independently.
> 

So this isn't a deadlock rather just a delay in user fences signaling?

> Signed-off-by: Jia Yao <jia.yao@intel.com>
> ---
>  drivers/gpu/drm/xe/xe_device.c       | 12 +++++++++++-
>  drivers/gpu/drm/xe/xe_device_types.h |  5 ++++-
>  drivers/gpu/drm/xe/xe_vm.c           |  2 +-
>  drivers/gpu/drm/xe/xe_vm.h           |  2 +-
>  4 files changed, 17 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/gpu/drm/xe/xe_device.c b/drivers/gpu/drm/xe/xe_device.c
> index db0b883322ef..f1dd4b7842a9 100644
> --- a/drivers/gpu/drm/xe/xe_device.c
> +++ b/drivers/gpu/drm/xe/xe_device.c
> @@ -385,6 +385,9 @@ static void xe_device_destroy(struct drm_device *dev, void *dummy)
>  	if (xe->preempt_fence_wq)
>  		destroy_workqueue(xe->preempt_fence_wq);
>  
> +	if (xe->rebind_wq)
> +		destroy_workqueue(xe->rebind_wq);
> +
>  	if (xe->ordered_wq)
>  		destroy_workqueue(xe->ordered_wq);

How about s/ordered_wq/userfence_wq to make this clear what is used for
and nothing else should be scheduled on this.

Matt

>  
> @@ -515,9 +518,16 @@ int xe_device_init_early(struct xe_device *xe)
>  	xe->preempt_fence_wq = alloc_ordered_workqueue("xe-preempt-fence-wq",
>  						       WQ_MEM_RECLAIM);
>  	xe->ordered_wq = alloc_ordered_workqueue("xe-ordered-wq", 0);
> +	/*
> +	 * Rebinding can block in HMM or wait for VM fences. Keep it off the
> +	 * queue serving user-fence notifications and VM TLB fence timeouts,
> +	 * which must be able to make progress while a rebind is waiting.
> +	 * Preserve serialization between rebind workers on different VMs.
> +	 */
> +	xe->rebind_wq = alloc_ordered_workqueue("xe-rebind-wq", 0);
>  	xe->unordered_wq = alloc_workqueue("xe-unordered-wq", WQ_PERCPU, 0);
>  	xe->destroy_wq = alloc_workqueue("xe-destroy-wq", WQ_PERCPU | WQ_MEM_RECLAIM, 0);
> -	if (!xe->ordered_wq || !xe->unordered_wq ||
> +	if (!xe->ordered_wq || !xe->rebind_wq || !xe->unordered_wq ||
>  	    !xe->preempt_fence_wq || !xe->destroy_wq) {
>  		/*
>  		 * Cleanup done in xe_device_destroy via
> diff --git a/drivers/gpu/drm/xe/xe_device_types.h b/drivers/gpu/drm/xe/xe_device_types.h
> index 5553415afeaf..705c487e3c6f 100644
> --- a/drivers/gpu/drm/xe/xe_device_types.h
> +++ b/drivers/gpu/drm/xe/xe_device_types.h
> @@ -387,9 +387,12 @@ struct xe_device {
>  	/** @preempt_fence_wq: used to serialize preempt fences */
>  	struct workqueue_struct *preempt_fence_wq;
>  
> -	/** @ordered_wq: used to serialize compute mode resume */
> +	/** @ordered_wq: ordered user-fence and VM TLB timeout work */
>  	struct workqueue_struct *ordered_wq;
>  
> +	/** @rebind_wq: serializes compute mode rebinds and resume */
> +	struct workqueue_struct *rebind_wq;
> +
>  	/** @unordered_wq: used to serialize unordered work */
>  	struct workqueue_struct *unordered_wq;
>  
> diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
> index e7f421a43516..88b4f6b19fdd 100644
> --- a/drivers/gpu/drm/xe/xe_vm.c
> +++ b/drivers/gpu/drm/xe/xe_vm.c
> @@ -501,7 +501,7 @@ static bool vm_suspend_rebind_worker(struct xe_vm *vm)
>   */
>  void xe_vm_resume_rebind_worker(struct xe_vm *vm)
>  {
> -	queue_work(vm->xe->ordered_wq, &vm->preempt.rebind_work);
> +	xe_vm_queue_rebind_worker(vm);
>  }
>  
>  static void preempt_rebind_work_func(struct work_struct *w)
> diff --git a/drivers/gpu/drm/xe/xe_vm.h b/drivers/gpu/drm/xe/xe_vm.h
> index 06138c3c4315..25c8760e2392 100644
> --- a/drivers/gpu/drm/xe/xe_vm.h
> +++ b/drivers/gpu/drm/xe/xe_vm.h
> @@ -259,7 +259,7 @@ int xe_vm_validate_protected(struct xe_vm *vm);
>  static inline void xe_vm_queue_rebind_worker(struct xe_vm *vm)
>  {
>  	xe_assert(vm->xe, xe_vm_in_preempt_fence_mode(vm));
> -	queue_work(vm->xe->ordered_wq, &vm->preempt.rebind_work);
> +	queue_work(vm->xe->rebind_wq, &vm->preempt.rebind_work);
>  }
>  
>  /**
> -- 
> 2.43.0
>