[PATCH v6 3/6] drm/xe/xe_vm: Add per VM pagefault info

Lin, Shuicheng shuicheng.lin at intel.com
Fri Mar 7 23:19:29 UTC 2025


On Fri, Mar 7, 2025 2:41 PM Cavitt, Jonathan wrote
> Add additional information to each VM so they can report up to the first
> 50 seen pagefaults.  Only failed pagefaults are saved this way, as successful
> pagefaults should recover and not need to be reported to userspace.
> 
> v2:
> - Free vm after use (Shuicheng)
> - Compress pf copy logic (Shuicheng)
> - Update fault_unsuccessful before storing (Shuicheng)
> - Fix old struct name in comments (Shuicheng)
> - Keep first 50 pagefaults instead of last 50 (Jianxun)
> 
> Signed-off-by: Jonathan Cavitt <jonathan.cavitt at intel.com>
> Suggested-by: Matthew Brost <matthew.brost at intel.com>
> CC: Shuicheng Lin <shuicheng.lin at intel.com>
> CC: Zhang Jianxun <jianxun.zhang at intel.com>
> ---
>  drivers/gpu/drm/xe/xe_gt_pagefault.c | 17 +++++++++++
>  drivers/gpu/drm/xe/xe_vm.c           | 42 ++++++++++++++++++++++++++++
>  drivers/gpu/drm/xe/xe_vm.h           |  6 ++++
>  drivers/gpu/drm/xe/xe_vm_types.h     | 20 +++++++++++++
>  4 files changed, 85 insertions(+)
> 
> diff --git a/drivers/gpu/drm/xe/xe_gt_pagefault.c
> b/drivers/gpu/drm/xe/xe_gt_pagefault.c
> index c8a9058aa09f..964ca1efb698 100644
> --- a/drivers/gpu/drm/xe/xe_gt_pagefault.c
> +++ b/drivers/gpu/drm/xe/xe_gt_pagefault.c
> @@ -343,6 +343,22 @@ int xe_guc_pagefault_handler(struct xe_guc *guc,
> u32 *msg, u32 len)
>  	return full ? -ENOSPC : 0;
>  }
> 
> +static void save_pagefault_to_vm(struct xe_device *xe, struct
> +xe_pagefault *pf) {
> +	struct xe_vm *vm;
> +	struct xe_pagefault *store;
> +
> +	vm = asid_to_vm(xe, pf->asid);
> +	if (IS_ERR(vm))
> +		return;
> +
> +	spin_lock(&vm->pfs.lock);
> +	store = kmemdup(pf, sizeof(*pf), GFP_KERNEL);

I still prefer we could have NULL check for the store pointer.

Shuicheng

> +	xe_vm_add_pf_entry(vm, store);
> +	spin_unlock(&vm->pfs.lock);
> +	xe_vm_put(vm);
> +}
> +
>  #define USM_QUEUE_MAX_RUNTIME_MS	20
> 
>  static void pf_queue_work_func(struct work_struct *w) @@ -362,6 +378,7
> @@ static void pf_queue_work_func(struct work_struct *w)
>  		if (unlikely(ret)) {
>  			print_pagefault(xe, &pf);
>  			pf.fault_unsuccessful = 1;
> +			save_pagefault_to_vm(xe, &pf);
>  			drm_dbg(&xe->drm, "Fault response:
> Unsuccessful %d\n", ret);
>  		}
> 
> diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c index
> 22a26aff3a6e..eada3ecc2364 100644
> --- a/drivers/gpu/drm/xe/xe_vm.c
> +++ b/drivers/gpu/drm/xe/xe_vm.c
> @@ -778,6 +778,43 @@ int xe_vm_userptr_check_repin(struct xe_vm *vm)
>  		list_empty_careful(&vm->userptr.invalidated)) ? 0 : -
> EAGAIN;  }
> 
> +void xe_vm_add_pf_entry(struct xe_vm *vm, struct xe_pagefault *pf) {
> +	struct xe_vm_pf_entry *e = NULL;
> +
> +	e = kzalloc(sizeof(*e), GFP_KERNEL);
> +	xe_assert(vm->xe, e);
> +

The 2nd input parameter pf is missed to be assigned to e.

Shuicheng

> +	spin_lock(&vm->pfs.lock);
> +
> +	/**
> +	 * Limit the number of pfs in the pf list to prevent memory overuse.
> +	 */
> +	if (vm->pfs.len >= MAX_PFS) {
> +		kfree(e);
> +		spin_unlock(&vm->pfs.lock);
> +		return;
> +	}
> +
> +	list_add_tail(&e->list, &vm->pfs.list);
> +	vm->pfs.len++;
> +	spin_unlock(&vm->pfs.lock);
> +}
> +
> +void xe_vm_remove_pf_entries(struct xe_vm *vm) {
> +	struct xe_vm_pf_entry *e, *tmp;
> +
> +	spin_lock(&vm->pfs.lock);
> +	list_for_each_entry_safe(e, tmp, &vm->pfs.list, list) {
> +		list_del(&e->list);
> +		kfree(e->pf);
> +		kfree(e);
> +	}
> +	vm->pfs.len = 0;
> +	spin_unlock(&vm->pfs.lock);
> +}
> +
>  static int xe_vma_ops_alloc(struct xe_vma_ops *vops, bool array_of_binds)  {
>  	int i;
> @@ -1660,6 +1697,9 @@ struct xe_vm *xe_vm_create(struct xe_device *xe,
> u32 flags)
>  	init_rwsem(&vm->userptr.notifier_lock);
>  	spin_lock_init(&vm->userptr.invalidated_lock);
> 
> +	INIT_LIST_HEAD(&vm->pfs.list);
> +	spin_lock_init(&vm->pfs.lock);
> +
>  	ttm_lru_bulk_move_init(&vm->lru_bulk_move);
> 
>  	INIT_WORK(&vm->destroy_work, vm_destroy_work_func); @@ -
> 1930,6 +1970,8 @@ void xe_vm_close_and_put(struct xe_vm *vm)
>  	}
>  	up_write(&xe->usm.lock);
> 
> +	xe_vm_remove_pf_entries(vm);
> +
>  	for_each_tile(tile, xe, id)
>  		xe_range_fence_tree_fini(&vm->rftree[id]);
> 
> diff --git a/drivers/gpu/drm/xe/xe_vm.h b/drivers/gpu/drm/xe/xe_vm.h index
> 0ef811fc2bde..fe068e4448a6 100644
> --- a/drivers/gpu/drm/xe/xe_vm.h
> +++ b/drivers/gpu/drm/xe/xe_vm.h
> @@ -12,6 +12,8 @@
>  #include "xe_map.h"
>  #include "xe_vm_types.h"
> 
> +#define MAX_PFS	50
> +
>  struct drm_device;
>  struct drm_printer;
>  struct drm_file;
> @@ -257,6 +259,10 @@ int xe_vma_userptr_pin_pages(struct
> xe_userptr_vma *uvma);
> 
>  int xe_vma_userptr_check_repin(struct xe_userptr_vma *uvma);
> 
> +void xe_vm_add_pf_entry(struct xe_vm *vm, struct xe_pagefault *pf);
> +
> +void xe_vm_remove_pf_entries(struct xe_vm *vm);
> +
>  bool xe_vm_validate_should_retry(struct drm_exec *exec, int err, ktime_t
> *end);
> 
>  int xe_vm_lock_vma(struct drm_exec *exec, struct xe_vma *vma); diff --git
> a/drivers/gpu/drm/xe/xe_vm_types.h b/drivers/gpu/drm/xe/xe_vm_types.h
> index 84fa41b9fa20..3a2329f54fb1 100644
> --- a/drivers/gpu/drm/xe/xe_vm_types.h
> +++ b/drivers/gpu/drm/xe/xe_vm_types.h
> @@ -19,6 +19,7 @@
>  #include "xe_range_fence.h"
> 
>  struct xe_bo;
> +struct xe_pagefault;
>  struct xe_svm_range;
>  struct xe_sync_entry;
>  struct xe_user_fence;
> @@ -142,6 +143,13 @@ struct xe_userptr_vma {
> 
>  struct xe_device;
> 
> +struct xe_vm_pf_entry {
> +	/** @pf: observed pagefault */
> +	struct xe_pagefault *pf;
> +	/** @list: link into @xe_vm.pfs.list */
> +	struct list_head list;
> +};
> +
>  struct xe_vm {
>  	/** @gpuvm: base GPUVM used to track VMAs */
>  	struct drm_gpuvm gpuvm;
> @@ -305,6 +313,18 @@ struct xe_vm {
>  		bool capture_once;
>  	} error_capture;
> 
> +	/**
> +	 * @pfs: List of all pagefaults associated with this VM
> +	 */
> +	struct {
> +		/** @lock: lock protecting @pfs.list */
> +		spinlock_t lock;
> +		/** @list: list of xe_exec_queue_ban_entry entries */
> +		struct list_head list;
> +		/** @len: length of @pfs.list */
> +		unsigned int len;
> +	} pfs;
> +
>  	/**
>  	 * @tlb_flush_seqno: Required TLB flush seqno for the next exec.
>  	 * protected by the vm resv.
> --
> 2.43.0



More information about the dri-devel mailing list