Re: [PATCH 05/11] drm/xe: Implement xe_pagefault_queue_work

[Francois Dugast <francois.dugast@intel.com>]

On Tue, Aug 05, 2025 at 11:22:36PM -0700, Matthew Brost wrote:
> Implement a worker that services page faults, using the same
> implementation as in xe_gt_pagefault.c.

Also the minor refactoring and cleanup along the way helps readability.

A few nits below.

> 
> Signed-off-by: Matthew Brost <matthew.brost@intel.com>
> ---
>  drivers/gpu/drm/xe/xe_pagefault.c | 240 +++++++++++++++++++++++++++++-
>  1 file changed, 239 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/xe/xe_pagefault.c b/drivers/gpu/drm/xe/xe_pagefault.c
> index 98be3203a9df..474412c21ec3 100644
> --- a/drivers/gpu/drm/xe/xe_pagefault.c
> +++ b/drivers/gpu/drm/xe/xe_pagefault.c
> @@ -5,12 +5,20 @@
>  
>  #include <linux/circ_buf.h>
>  
> +#include <drm/drm_exec.h>
>  #include <drm/drm_managed.h>
>  
> +#include "xe_bo.h"
>  #include "xe_device.h"
> +#include "xe_gt_printk.h"
>  #include "xe_gt_types.h"
> +#include "xe_gt_stats.h"

Move up to maintain alphabetically ordered.

> +#include "xe_hw_engine.h"
>  #include "xe_pagefault.h"
>  #include "xe_pagefault_types.h"
> +#include "xe_svm.h"
> +#include "xe_trace_bo.h"
> +#include "xe_vm.h"
>  
>  /**
>   * DOC: Xe page faults
> @@ -30,9 +38,239 @@ static int xe_pagefault_entry_size(void)
>  	return roundup_pow_of_two(sizeof(struct xe_pagefault));
>  }
>  
> +static int xe_pagefault_begin(struct drm_exec *exec, struct xe_vma *vma,
> +			      bool atomic, unsigned int id)

Please rename id to tile_id for clarity.

Francois

> +{
> +	struct xe_bo *bo = xe_vma_bo(vma);
> +	struct xe_vm *vm = xe_vma_vm(vma);
> +	int err;
> +
> +	err = xe_vm_lock_vma(exec, vma);
> +	if (err)
> +		return err;
> +
> +	if (atomic && IS_DGFX(vm->xe)) {
> +		if (xe_vma_is_userptr(vma)) {
> +			err = -EACCES;
> +			return err;
> +		}
> +
> +		/* Migrate to VRAM, move should invalidate the VMA first */
> +		err = xe_bo_migrate(bo, XE_PL_VRAM0 + id);
> +		if (err)
> +			return err;
> +	} else if (bo) {
> +		/* Create backing store if needed */
> +		err = xe_bo_validate(bo, vm, true);
> +		if (err)
> +			return err;
> +	}
> +
> +	return 0;
> +}
> +
> +static int xe_pagefault_handle_vma(struct xe_gt *gt, struct xe_vma *vma,
> +				   bool atomic)
> +{
> +	struct xe_vm *vm = xe_vma_vm(vma);
> +	struct xe_tile *tile = gt_to_tile(gt);
> +	struct drm_exec exec;
> +	struct dma_fence *fence;
> +	ktime_t end = 0;
> +	int err;
> +
> +	lockdep_assert_held_write(&vm->lock);
> +
> +	xe_gt_stats_incr(gt, XE_GT_STATS_ID_VMA_PAGEFAULT_COUNT, 1);
> +	xe_gt_stats_incr(gt, XE_GT_STATS_ID_VMA_PAGEFAULT_KB,
> +			 xe_vma_size(vma) / SZ_1K);
> +
> +	trace_xe_vma_pagefault(vma);
> +
> +	/* Check if VMA is valid, opportunistic check only */
> +	if (xe_vm_has_valid_gpu_mapping(tile, vma->tile_present,
> +					vma->tile_invalidated) && !atomic)
> +		return 0;
> +
> +retry_userptr:
> +	if (xe_vma_is_userptr(vma) &&
> +	    xe_vma_userptr_check_repin(to_userptr_vma(vma))) {
> +		struct xe_userptr_vma *uvma = to_userptr_vma(vma);
> +
> +		err = xe_vma_userptr_pin_pages(uvma);
> +		if (err)
> +			return err;
> +	}
> +
> +	/* Lock VM and BOs dma-resv */
> +	drm_exec_init(&exec, 0, 0);
> +	drm_exec_until_all_locked(&exec) {
> +		err = xe_pagefault_begin(&exec, vma, atomic, tile->id);
> +		drm_exec_retry_on_contention(&exec);
> +		if (xe_vm_validate_should_retry(&exec, err, &end))
> +			err = -EAGAIN;
> +		if (err)
> +			goto unlock_dma_resv;
> +
> +		/* Bind VMA only to the GT that has faulted */
> +		trace_xe_vma_pf_bind(vma);
> +		fence = xe_vma_rebind(vm, vma, BIT(tile->id));
> +		if (IS_ERR(fence)) {
> +			err = PTR_ERR(fence);
> +			if (xe_vm_validate_should_retry(&exec, err, &end))
> +				err = -EAGAIN;
> +			goto unlock_dma_resv;
> +		}
> +	}
> +
> +	dma_fence_wait(fence, false);
> +	dma_fence_put(fence);
> +
> +unlock_dma_resv:
> +	drm_exec_fini(&exec);
> +	if (err == -EAGAIN)
> +		goto retry_userptr;
> +
> +	return err;
> +}
> +
> +static bool
> +xe_pagefault_access_is_atomic(enum xe_pagefault_access_type access_type)
> +{
> +	return access_type == XE_PAGEFAULT_ACCESS_TYPE_ATOMIC;
> +}
> +
> +static struct xe_vm *xe_pagefault_asid_to_vm(struct xe_device *xe, u32 asid)
> +{
> +	struct xe_vm *vm;
> +
> +	down_read(&xe->usm.lock);
> +	vm = xa_load(&xe->usm.asid_to_vm, asid);
> +	if (vm && xe_vm_in_fault_mode(vm))
> +		xe_vm_get(vm);
> +	else
> +		vm = ERR_PTR(-EINVAL);
> +	up_read(&xe->usm.lock);
> +
> +	return vm;
> +}
> +
> +static int xe_pagefault_service(struct xe_pagefault *pf)
> +{
> +	struct xe_gt *gt = pf->gt;
> +	struct xe_device *xe = gt_to_xe(gt);
> +	struct xe_vm *vm;
> +	struct xe_vma *vma = NULL;
> +	int err;
> +	bool atomic;
> +
> +	/* Producer flagged this fault to be nacked */
> +	if (pf->consumer.fault_level == XE_PAGEFAULT_LEVEL_NACK)
> +		return -EFAULT;
> +
> +	vm = xe_pagefault_asid_to_vm(xe, pf->consumer.asid);
> +	if (IS_ERR(vm))
> +		return PTR_ERR(vm);
> +
> +	/*
> +	 * TODO: Change to read lock? Using write lock for simplicity.
> +	 */
> +	down_write(&vm->lock);
> +
> +	if (xe_vm_is_closed(vm)) {
> +		err = -ENOENT;
> +		goto unlock_vm;
> +	}
> +
> +	vma = xe_vm_find_vma_by_addr(vm, pf->consumer.page_addr);
> +	if (!vma) {
> +		err = -EINVAL;
> +		goto unlock_vm;
> +	}
> +
> +	atomic = xe_pagefault_access_is_atomic(pf->consumer.access_type);
> +
> +	if (xe_vma_is_cpu_addr_mirror(vma))
> +		err = xe_svm_handle_pagefault(vm, vma, gt,
> +					      pf->consumer.page_addr, atomic);
> +	else
> +		err = xe_pagefault_handle_vma(gt, vma, atomic);
> +
> +unlock_vm:
> +	if (!err)
> +		vm->usm.last_fault_vma = vma;
> +	up_write(&vm->lock);
> +	xe_vm_put(vm);
> +
> +	return err;
> +}
> +
> +static bool xe_pagefault_queue_pop(struct xe_pagefault_queue *pf_queue,
> +				   struct xe_pagefault *pf)
> +{
> +	bool found_fault = false;
> +
> +	spin_lock_irq(&pf_queue->lock);
> +	if (pf_queue->tail != pf_queue->head) {
> +		memcpy(pf, pf_queue->data + pf_queue->tail, sizeof(*pf));
> +		pf_queue->tail = (pf_queue->tail + xe_pagefault_entry_size()) %
> +			pf_queue->size;
> +		found_fault = true;
> +	}
> +	spin_unlock_irq(&pf_queue->lock);
> +
> +	return found_fault;
> +}
> +
> +static void xe_pagefault_print(struct xe_pagefault *pf)
> +{
> +	xe_gt_dbg(pf->gt, "\n\tASID: %d\n"
> +		  "\tFaulted Address: 0x%08x%08x\n"
> +		  "\tFaultType: %d\n"
> +		  "\tAccessType: %d\n"
> +		  "\tFaultLevel: %d\n"
> +		  "\tEngineClass: %d %s\n",
> +		  pf->consumer.asid,
> +		  upper_32_bits(pf->consumer.page_addr),
> +		  lower_32_bits(pf->consumer.page_addr),
> +		  pf->consumer.fault_type,
> +		  pf->consumer.access_type,
> +		  pf->consumer.fault_level,
> +		  pf->consumer.engine_class,
> +		  xe_hw_engine_class_to_str(pf->consumer.engine_class));
> +}
> +
>  static void xe_pagefault_queue_work(struct work_struct *w)
>  {
> -	/* TODO: Implement */
> +	struct xe_pagefault_queue *pf_queue =
> +		container_of(w, typeof(*pf_queue), worker);
> +	struct xe_pagefault pf;
> +	unsigned long threshold;
> +
> +#define USM_QUEUE_MAX_RUNTIME_MS      20
> +	threshold = jiffies + msecs_to_jiffies(USM_QUEUE_MAX_RUNTIME_MS);
> +
> +	while (xe_pagefault_queue_pop(pf_queue, &pf)) {
> +		int err;
> +
> +		if (!pf.gt)	/* Fault squashed during reset */
> +			continue;
> +
> +		err = xe_pagefault_service(&pf);
> +		if (err) {
> +			xe_pagefault_print(&pf);
> +			xe_gt_dbg(pf.gt, "Fault response: Unsuccessful %pe\n",
> +				  ERR_PTR(err));
> +		}
> +
> +		pf.producer.ops->ack_fault(&pf, err);
> +
> +		if (time_after(jiffies, threshold)) {
> +			queue_work(gt_to_xe(pf.gt)->usm.pf_wq, w);
> +			break;
> +		}
> +	}
> +#undef USM_QUEUE_MAX_RUNTIME_MS
>  }
>  
>  static int xe_pagefault_queue_init(struct xe_device *xe,
> -- 
> 2.34.1
>