Merge tag 'drm-xe-fixes-2024-02-22' of https://gitlab.freedesktop.org/drm/xe/kernel into drm-fixes

UAPI Changes:
- Remove support for persistent exec_queues
- Drop a reduntant sysfs newline printout

Cross-subsystem Changes:

Core Changes:

Driver Changes:
- A three-patch fix for a VM_BIND rebind optimization path
- Fix a modpost warning on an xe KUNIT module
Signed-off-by: Dave Airlie <airlied@redhat.com>

From: Thomas Hellstrom <thomas.hellstrom@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/ZdcsNrxdWMMM417v@fedora

drivers/gpu/drm/xe/tests/xe_mocs_test.c

@@ -21,4 +21,5 @@ kunit_test_suite(xe_mocs_test_suite);
 
 MODULE_AUTHOR("Intel Corporation");
 MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("xe_mocs kunit test");
 MODULE_IMPORT_NS(EXPORTED_FOR_KUNIT_TESTING);

drivers/gpu/drm/xe/xe_device.c

@@ -83,9 +83,6 @@ static int xe_file_open(struct drm_device *dev, struct drm_file *file)
 	return 0;
 }
 
-static void device_kill_persistent_exec_queues(struct xe_device *xe,
-					       struct xe_file *xef);
-
 static void xe_file_close(struct drm_device *dev, struct drm_file *file)
 {
 	struct xe_device *xe = to_xe_device(dev);
@@ -102,8 +99,6 @@ static void xe_file_close(struct drm_device *dev, struct drm_file *file)
 	mutex_unlock(&xef->exec_queue.lock);
 	xa_destroy(&xef->exec_queue.xa);
 	mutex_destroy(&xef->exec_queue.lock);
-	device_kill_persistent_exec_queues(xe, xef);
-
 	mutex_lock(&xef->vm.lock);
 	xa_for_each(&xef->vm.xa, idx, vm)
 		xe_vm_close_and_put(vm);
@@ -255,9 +250,6 @@ struct xe_device *xe_device_create(struct pci_dev *pdev,
 			xa_erase(&xe->usm.asid_to_vm, asid);
 	}
 
-	drmm_mutex_init(&xe->drm, &xe->persistent_engines.lock);
-	INIT_LIST_HEAD(&xe->persistent_engines.list);
-
 	spin_lock_init(&xe->pinned.lock);
 	INIT_LIST_HEAD(&xe->pinned.kernel_bo_present);
 	INIT_LIST_HEAD(&xe->pinned.external_vram);
@@ -570,37 +562,6 @@ void xe_device_shutdown(struct xe_device *xe)
 {
 }
 
-void xe_device_add_persistent_exec_queues(struct xe_device *xe, struct xe_exec_queue *q)
-{
-	mutex_lock(&xe->persistent_engines.lock);
-	list_add_tail(&q->persistent.link, &xe->persistent_engines.list);
-	mutex_unlock(&xe->persistent_engines.lock);
-}
-
-void xe_device_remove_persistent_exec_queues(struct xe_device *xe,
-					     struct xe_exec_queue *q)
-{
-	mutex_lock(&xe->persistent_engines.lock);
-	if (!list_empty(&q->persistent.link))
-		list_del(&q->persistent.link);
-	mutex_unlock(&xe->persistent_engines.lock);
-}
-
-static void device_kill_persistent_exec_queues(struct xe_device *xe,
-					       struct xe_file *xef)
-{
-	struct xe_exec_queue *q, *next;
-
-	mutex_lock(&xe->persistent_engines.lock);
-	list_for_each_entry_safe(q, next, &xe->persistent_engines.list,
-				 persistent.link)
-		if (q->persistent.xef == xef) {
-			xe_exec_queue_kill(q);
-			list_del_init(&q->persistent.link);
-		}
-	mutex_unlock(&xe->persistent_engines.lock);
-}
-
 void xe_device_wmb(struct xe_device *xe)
 {
 	struct xe_gt *gt = xe_root_mmio_gt(xe);

drivers/gpu/drm/xe/xe_device.h

@@ -42,10 +42,6 @@ int xe_device_probe(struct xe_device *xe);
 void xe_device_remove(struct xe_device *xe);
 void xe_device_shutdown(struct xe_device *xe);
 
-void xe_device_add_persistent_exec_queues(struct xe_device *xe, struct xe_exec_queue *q);
-void xe_device_remove_persistent_exec_queues(struct xe_device *xe,
-					     struct xe_exec_queue *q);
-
 void xe_device_wmb(struct xe_device *xe);
 
 static inline struct xe_file *to_xe_file(const struct drm_file *file)

drivers/gpu/drm/xe/xe_device_types.h

@@ -341,14 +341,6 @@ struct xe_device {
 		struct mutex lock;
 	} usm;
 
-	/** @persistent_engines: engines that are closed but still running */
-	struct {
-		/** @lock: protects persistent engines */
-		struct mutex lock;
-		/** @list: list of persistent engines */
-		struct list_head list;
-	} persistent_engines;
-
 	/** @pinned: pinned BO state */
 	struct {
 		/** @lock: protected pinned BO list state */

drivers/gpu/drm/xe/xe_exec_queue.c

@@ -60,7 +60,6 @@ static struct xe_exec_queue *__xe_exec_queue_create(struct xe_device *xe,
 	q->fence_irq = &gt->fence_irq[hwe->class];
 	q->ring_ops = gt->ring_ops[hwe->class];
 	q->ops = gt->exec_queue_ops;
-	INIT_LIST_HEAD(&q->persistent.link);
 	INIT_LIST_HEAD(&q->compute.link);
 	INIT_LIST_HEAD(&q->multi_gt_link);
 
@@ -326,23 +325,6 @@ static int exec_queue_set_preemption_timeout(struct xe_device *xe,
 	return q->ops->set_preempt_timeout(q, value);
 }
 
-static int exec_queue_set_persistence(struct xe_device *xe, struct xe_exec_queue *q,
-				      u64 value, bool create)
-{
-	if (XE_IOCTL_DBG(xe, !create))
-		return -EINVAL;
-
-	if (XE_IOCTL_DBG(xe, xe_vm_in_preempt_fence_mode(q->vm)))
-		return -EINVAL;
-
-	if (value)
-		q->flags |= EXEC_QUEUE_FLAG_PERSISTENT;
-	else
-		q->flags &= ~EXEC_QUEUE_FLAG_PERSISTENT;
-
-	return 0;
-}
-
 static int exec_queue_set_job_timeout(struct xe_device *xe, struct xe_exec_queue *q,
 				      u64 value, bool create)
 {
@@ -414,7 +396,6 @@ static const xe_exec_queue_set_property_fn exec_queue_set_property_funcs[] = {
 	[DRM_XE_EXEC_QUEUE_SET_PROPERTY_PRIORITY] = exec_queue_set_priority,
 	[DRM_XE_EXEC_QUEUE_SET_PROPERTY_TIMESLICE] = exec_queue_set_timeslice,
 	[DRM_XE_EXEC_QUEUE_SET_PROPERTY_PREEMPTION_TIMEOUT] = exec_queue_set_preemption_timeout,
-	[DRM_XE_EXEC_QUEUE_SET_PROPERTY_PERSISTENCE] = exec_queue_set_persistence,
 	[DRM_XE_EXEC_QUEUE_SET_PROPERTY_JOB_TIMEOUT] = exec_queue_set_job_timeout,
 	[DRM_XE_EXEC_QUEUE_SET_PROPERTY_ACC_TRIGGER] = exec_queue_set_acc_trigger,
 	[DRM_XE_EXEC_QUEUE_SET_PROPERTY_ACC_NOTIFY] = exec_queue_set_acc_notify,
@@ -441,6 +422,9 @@ static int exec_queue_user_ext_set_property(struct xe_device *xe,
 		return -EINVAL;
 
 	idx = array_index_nospec(ext.property, ARRAY_SIZE(exec_queue_set_property_funcs));
+	if (!exec_queue_set_property_funcs[idx])
+		return -EINVAL;
+
 	return exec_queue_set_property_funcs[idx](xe, q, ext.value,  create);
 }
 
@@ -704,9 +688,7 @@ int xe_exec_queue_create_ioctl(struct drm_device *dev, void *data,
 		}
 
 		q = xe_exec_queue_create(xe, vm, logical_mask,
-					 args->width, hwe,
-					 xe_vm_in_lr_mode(vm) ? 0 :
-					 EXEC_QUEUE_FLAG_PERSISTENT);
+					 args->width, hwe, 0);
 		up_read(&vm->lock);
 		xe_vm_put(vm);
 		if (IS_ERR(q))
@@ -728,8 +710,6 @@ int xe_exec_queue_create_ioctl(struct drm_device *dev, void *data,
 			goto kill_exec_queue;
 	}
 
-	q->persistent.xef = xef;
-
 	mutex_lock(&xef->exec_queue.lock);
 	err = xa_alloc(&xef->exec_queue.xa, &id, q, xa_limit_32b, GFP_KERNEL);
 	mutex_unlock(&xef->exec_queue.lock);
@@ -872,10 +852,7 @@ int xe_exec_queue_destroy_ioctl(struct drm_device *dev, void *data,
 	if (XE_IOCTL_DBG(xe, !q))
 		return -ENOENT;
 
-	if (!(q->flags & EXEC_QUEUE_FLAG_PERSISTENT))
 	xe_exec_queue_kill(q);
-	else
-		xe_device_add_persistent_exec_queues(xe, q);
 
 	trace_xe_exec_queue_close(q);
 	xe_exec_queue_put(q);

drivers/gpu/drm/xe/xe_exec_queue_types.h

@@ -105,16 +105,6 @@ struct xe_exec_queue {
 		struct xe_guc_exec_queue *guc;
 	};
 
-	/**
-	 * @persistent: persistent exec queue state
-	 */
-	struct {
-		/** @xef: file which this exec queue belongs to */
-		struct xe_file *xef;
-		/** @link: link in list of persistent exec queues */
-		struct list_head link;
-	} persistent;
-
 	union {
 		/**
 		 * @parallel: parallel submission state

drivers/gpu/drm/xe/xe_execlist.c

@@ -378,8 +378,6 @@ static void execlist_exec_queue_fini_async(struct work_struct *w)
 		list_del(&exl->active_link);
 	spin_unlock_irqrestore(&exl->port->lock, flags);
 
-	if (q->flags & EXEC_QUEUE_FLAG_PERSISTENT)
-		xe_device_remove_persistent_exec_queues(xe, q);
 	drm_sched_entity_fini(&exl->entity);
 	drm_sched_fini(&exl->sched);
 	kfree(exl);

drivers/gpu/drm/xe/xe_gt_idle.c

@@ -145,10 +145,10 @@ void xe_gt_idle_sysfs_init(struct xe_gt_idle *gtidle)
 	}
 
 	if (xe_gt_is_media_type(gt)) {
-		sprintf(gtidle->name, "gt%d-mc\n", gt->info.id);
+		sprintf(gtidle->name, "gt%d-mc", gt->info.id);
 		gtidle->idle_residency = xe_guc_pc_mc6_residency;
 	} else {
-		sprintf(gtidle->name, "gt%d-rc\n", gt->info.id);
+		sprintf(gtidle->name, "gt%d-rc", gt->info.id);
 		gtidle->idle_residency = xe_guc_pc_rc6_residency;
 	}
 

drivers/gpu/drm/xe/xe_guc_submit.c

@@ -1028,8 +1028,6 @@ static void __guc_exec_queue_fini_async(struct work_struct *w)
 
 	if (xe_exec_queue_is_lr(q))
 		cancel_work_sync(&ge->lr_tdr);
-	if (q->flags & EXEC_QUEUE_FLAG_PERSISTENT)
-		xe_device_remove_persistent_exec_queues(gt_to_xe(q->gt), q);
 	release_guc_id(guc, q);
 	xe_sched_entity_fini(&ge->entity);
 	xe_sched_fini(&ge->sched);

drivers/gpu/drm/xe/xe_pt.c

@@ -499,11 +499,13 @@ xe_pt_stage_bind_entry(struct xe_ptw *parent, pgoff_t offset,
 		 * this device *requires* 64K PTE size for VRAM, fail.
 		 */
 		if (level == 0 && !xe_parent->is_compact) {
-			if (xe_pt_is_pte_ps64K(addr, next, xe_walk))
+			if (xe_pt_is_pte_ps64K(addr, next, xe_walk)) {
+				xe_walk->vma->gpuva.flags |= XE_VMA_PTE_64K;
 				pte |= XE_PTE_PS64;
-			else if (XE_WARN_ON(xe_walk->needs_64K))
+			} else if (XE_WARN_ON(xe_walk->needs_64K)) {
 				return -EINVAL;
 			}
+		}
 
 		ret = xe_pt_insert_entry(xe_walk, xe_parent, offset, NULL, pte);
 		if (unlikely(ret))
@@ -545,13 +547,16 @@ xe_pt_stage_bind_entry(struct xe_ptw *parent, pgoff_t offset,
 		*child = &xe_child->base;
 
 		/*
-		 * Prefer the compact pagetable layout for L0 if possible.
+		 * Prefer the compact pagetable layout for L0 if possible. Only
+		 * possible if VMA covers entire 2MB region as compact 64k and
+		 * 4k pages cannot be mixed within a 2MB region.
 		 * TODO: Suballocate the pt bo to avoid wasting a lot of
 		 * memory.
 		 */
 		if (GRAPHICS_VERx100(tile_to_xe(xe_walk->tile)) >= 1250 && level == 1 &&
 		    covers && xe_pt_scan_64K(addr, next, xe_walk)) {
 			walk->shifts = xe_compact_pt_shifts;
+			xe_walk->vma->gpuva.flags |= XE_VMA_PTE_COMPACT;
 			flags |= XE_PDE_64K;
 			xe_child->is_compact = true;
 		}

drivers/gpu/drm/xe/xe_vm.c

@@ -2190,15 +2190,17 @@ static u64 xe_vma_max_pte_size(struct xe_vma *vma)
 {
 	if (vma->gpuva.flags & XE_VMA_PTE_1G)
 		return SZ_1G;
-	else if (vma->gpuva.flags & XE_VMA_PTE_2M)
+	else if (vma->gpuva.flags & (XE_VMA_PTE_2M | XE_VMA_PTE_COMPACT))
 		return SZ_2M;
+	else if (vma->gpuva.flags & XE_VMA_PTE_64K)
+		return SZ_64K;
 	else if (vma->gpuva.flags & XE_VMA_PTE_4K)
 		return SZ_4K;
 
 	return SZ_1G;	/* Uninitialized, used max size */
 }
 
-static u64 xe_vma_set_pte_size(struct xe_vma *vma, u64 size)
+static void xe_vma_set_pte_size(struct xe_vma *vma, u64 size)
 {
 	switch (size) {
 	case SZ_1G:
@@ -2207,9 +2209,13 @@ static u64 xe_vma_set_pte_size(struct xe_vma *vma, u64 size)
 	case SZ_2M:
 		vma->gpuva.flags |= XE_VMA_PTE_2M;
 		break;
+	case SZ_64K:
+		vma->gpuva.flags |= XE_VMA_PTE_64K;
+		break;
+	case SZ_4K:
+		vma->gpuva.flags |= XE_VMA_PTE_4K;
+		break;
 	}
-
-	return SZ_4K;
 }
 
 static int xe_vma_op_commit(struct xe_vm *vm, struct xe_vma_op *op)

drivers/gpu/drm/xe/xe_vm_types.h

@@ -29,6 +29,8 @@ struct xe_vm;
 #define XE_VMA_PTE_4K		(DRM_GPUVA_USERBITS << 5)
 #define XE_VMA_PTE_2M		(DRM_GPUVA_USERBITS << 6)
 #define XE_VMA_PTE_1G		(DRM_GPUVA_USERBITS << 7)
+#define XE_VMA_PTE_64K		(DRM_GPUVA_USERBITS << 8)
+#define XE_VMA_PTE_COMPACT	(DRM_GPUVA_USERBITS << 9)
 
 /** struct xe_userptr - User pointer */
 struct xe_userptr {

include/uapi/drm/xe_drm.h

@@ -1046,7 +1046,6 @@ struct drm_xe_exec_queue_create {
 #define   DRM_XE_EXEC_QUEUE_SET_PROPERTY_PRIORITY		0
 #define   DRM_XE_EXEC_QUEUE_SET_PROPERTY_TIMESLICE		1
 #define   DRM_XE_EXEC_QUEUE_SET_PROPERTY_PREEMPTION_TIMEOUT	2
-#define   DRM_XE_EXEC_QUEUE_SET_PROPERTY_PERSISTENCE		3
 #define   DRM_XE_EXEC_QUEUE_SET_PROPERTY_JOB_TIMEOUT		4
 #define   DRM_XE_EXEC_QUEUE_SET_PROPERTY_ACC_TRIGGER		5
 #define   DRM_XE_EXEC_QUEUE_SET_PROPERTY_ACC_NOTIFY		6