1 files changed, 99 insertions, 35 deletions

diff --git a/drivers/gpu/drm/xe/xe_exec_queue.c b/drivers/gpu/drm/xe/xe_exec_queue.c
index d098d2dd1b2d..fee22358cc09 100644
--- a/drivers/gpu/drm/xe/xe_exec_queue.c
+++ b/drivers/gpu/drm/xe/xe_exec_queue.c
@@ -8,6 +8,7 @@
 #include <linux/nospec.h>
 
 #include <drm/drm_device.h>
+#include <drm/drm_drv.h>
 #include <drm/drm_file.h>
 #include <uapi/drm/xe_drm.h>
 
@@ -16,6 +17,7 @@
 #include "xe_hw_engine_class_sysfs.h"
 #include "xe_hw_engine_group.h"
 #include "xe_hw_fence.h"
+#include "xe_irq.h"
 #include "xe_lrc.h"
 #include "xe_macros.h"
 #include "xe_migrate.h"
@@ -23,6 +25,7 @@
 #include "xe_ring_ops_types.h"
 #include "xe_trace.h"
 #include "xe_vm.h"
+#include "xe_pxp.h"
 
 enum xe_exec_queue_sched_prop {
 	XE_EXEC_QUEUE_JOB_TIMEOUT = 0,
@@ -36,6 +39,8 @@ static int exec_queue_user_extensions(struct xe_device *xe, struct xe_exec_queue
 
 static void __xe_exec_queue_free(struct xe_exec_queue *q)
 {
+	if (xe_exec_queue_uses_pxp(q))
+		xe_pxp_exec_queue_remove(gt_to_xe(q->gt)->pxp, q);
 	if (q->vm)
 		xe_vm_put(q->vm);
 
@@ -68,6 +73,7 @@ static struct xe_exec_queue *__xe_exec_queue_alloc(struct xe_device *xe,
 	q->gt = gt;
 	q->class = hwe->class;
 	q->width = width;
+	q->msix_vec = XE_IRQ_DEFAULT_MSIX;
 	q->logical_mask = logical_mask;
 	q->fence_irq = &gt->fence_irq[hwe->class];
 	q->ring_ops = gt->ring_ops[hwe->class];
@@ -75,6 +81,7 @@ static struct xe_exec_queue *__xe_exec_queue_alloc(struct xe_device *xe,
 	INIT_LIST_HEAD(&q->lr.link);
 	INIT_LIST_HEAD(&q->multi_gt_link);
 	INIT_LIST_HEAD(&q->hw_engine_group_link);
+	INIT_LIST_HEAD(&q->pxp.link);
 
 	q->sched_props.timeslice_us = hwe->eclass->sched_props.timeslice_us;
 	q->sched_props.preempt_timeout_us =
@@ -107,35 +114,37 @@ static struct xe_exec_queue *__xe_exec_queue_alloc(struct xe_device *xe,
 
 static int __xe_exec_queue_init(struct xe_exec_queue *q)
 {
-	struct xe_vm *vm = q->vm;
 	int i, err;
+	u32 flags = 0;
 
-	if (vm) {
-		err = xe_vm_lock(vm, true);
-		if (err)
-			return err;
+	/*
+	 * PXP workloads executing on RCS or CCS must run in isolation (i.e. no
+	 * other workload can use the EUs at the same time). On MTL this is done
+	 * by setting the RUNALONE bit in the LRC, while starting on Xe2 there
+	 * is a dedicated bit for it.
+	 */
+	if (xe_exec_queue_uses_pxp(q) &&
+	    (q->class == XE_ENGINE_CLASS_RENDER || q->class == XE_ENGINE_CLASS_COMPUTE)) {
+		if (GRAPHICS_VER(gt_to_xe(q->gt)) >= 20)
+			flags |= XE_LRC_CREATE_PXP;
+		else
+			flags |= XE_LRC_CREATE_RUNALONE;
 	}
 
 	for (i = 0; i < q->width; ++i) {
-		q->lrc[i] = xe_lrc_create(q->hwe, q->vm, SZ_16K);
+		q->lrc[i] = xe_lrc_create(q->hwe, q->vm, SZ_16K, q->msix_vec, flags);
 		if (IS_ERR(q->lrc[i])) {
 			err = PTR_ERR(q->lrc[i]);
-			goto err_unlock;
+			goto err_lrc;
 		}
 	}
 
-	if (vm)
-		xe_vm_unlock(vm);
-
 	err = q->ops->init(q);
 	if (err)
 		goto err_lrc;
 
 	return 0;
 
-err_unlock:
-	if (vm)
-		xe_vm_unlock(vm);
 err_lrc:
 	for (i = i - 1; i >= 0; --i)
 		xe_lrc_put(q->lrc[i]);
@@ -150,6 +159,9 @@ struct xe_exec_queue *xe_exec_queue_create(struct xe_device *xe, struct xe_vm *v
 	struct xe_exec_queue *q;
 	int err;
 
+	/* VMs for GSCCS queues (and only those) must have the XE_VM_FLAG_GSC flag */
+	xe_assert(xe, !vm || (!!(vm->flags & XE_VM_FLAG_GSC) == !!(hwe->engine_id == XE_HW_ENGINE_GSCCS0)));
+
 	q = __xe_exec_queue_alloc(xe, vm, logical_mask, width, hwe, flags,
 				  extensions);
 	if (IS_ERR(q))
@@ -159,12 +171,26 @@ struct xe_exec_queue *xe_exec_queue_create(struct xe_device *xe, struct xe_vm *v
 	if (err)
 		goto err_post_alloc;
 
+	/*
+	 * We can only add the queue to the PXP list after the init is complete,
+	 * because the PXP termination can call exec_queue_kill and that will
+	 * go bad if the queue is only half-initialized. This means that we
+	 * can't do it when we handle the PXP extension in __xe_exec_queue_alloc
+	 * and we need to do it here instead.
+	 */
+	if (xe_exec_queue_uses_pxp(q)) {
+		err = xe_pxp_exec_queue_add(xe->pxp, q);
+		if (err)
+			goto err_post_alloc;
+	}
+
 	return q;
 
 err_post_alloc:
 	__xe_exec_queue_free(q);
 	return ERR_PTR(err);
 }
+ALLOW_ERROR_INJECTION(xe_exec_queue_create, ERRNO);
 
 struct xe_exec_queue *xe_exec_queue_create_class(struct xe_device *xe, struct xe_gt *gt,
 						 struct xe_vm *vm,
@@ -240,12 +266,16 @@ struct xe_exec_queue *xe_exec_queue_create_bind(struct xe_device *xe,
 
 	return q;
 }
+ALLOW_ERROR_INJECTION(xe_exec_queue_create_bind, ERRNO);
 
 void xe_exec_queue_destroy(struct kref *ref)
 {
 	struct xe_exec_queue *q = container_of(ref, struct xe_exec_queue, refcount);
 	struct xe_exec_queue *eq, *next;
 
+	if (xe_exec_queue_uses_pxp(q))
+		xe_pxp_exec_queue_remove(gt_to_xe(q->gt)->pxp, q);
+
 	xe_exec_queue_last_fence_put_unlocked(q);
 	if (!(q->flags & EXEC_QUEUE_FLAG_BIND_ENGINE_CHILD)) {
 		list_for_each_entry_safe(eq, next, &q->multi_gt_list,
@@ -260,8 +290,17 @@ void xe_exec_queue_fini(struct xe_exec_queue *q)
 {
 	int i;
 
+	/*
+	 * Before releasing our ref to lrc and xef, accumulate our run ticks
+	 * and wakeup any waiters.
+	 */
+	xe_exec_queue_update_run_ticks(q);
+	if (q->xef && atomic_dec_and_test(&q->xef->exec_queue.pending_removal))
+		wake_up_var(&q->xef->exec_queue.pending_removal);
+
 	for (i = 0; i < q->width; ++i)
 		xe_lrc_put(q->lrc[i]);
+
 	__xe_exec_queue_free(q);
 }
 
@@ -392,6 +431,22 @@ static int exec_queue_set_timeslice(struct xe_device *xe, struct xe_exec_queue *
 	return 0;
 }
 
+static int
+exec_queue_set_pxp_type(struct xe_device *xe, struct xe_exec_queue *q, u64 value)
+{
+	if (value == DRM_XE_PXP_TYPE_NONE)
+		return 0;
+
+	/* we only support HWDRM sessions right now */
+	if (XE_IOCTL_DBG(xe, value != DRM_XE_PXP_TYPE_HWDRM))
+		return -EINVAL;
+
+	if (!xe_pxp_is_enabled(xe->pxp))
+		return -ENODEV;
+
+	return xe_pxp_exec_queue_set_type(xe->pxp, q, DRM_XE_PXP_TYPE_HWDRM);
+}
+
 typedef int (*xe_exec_queue_set_property_fn)(struct xe_device *xe,
 					     struct xe_exec_queue *q,
 					     u64 value);
@@ -399,6 +454,7 @@ typedef int (*xe_exec_queue_set_property_fn)(struct xe_device *xe,
 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_PXP_TYPE] = exec_queue_set_pxp_type,
 };
 
 static int exec_queue_user_ext_set_property(struct xe_device *xe,
@@ -410,7 +466,7 @@ static int exec_queue_user_ext_set_property(struct xe_device *xe,
 	int err;
 	u32 idx;
 
-	err = __copy_from_user(&ext, address, sizeof(ext));
+	err = copy_from_user(&ext, address, sizeof(ext));
 	if (XE_IOCTL_DBG(xe, err))
 		return -EFAULT;
 
@@ -418,7 +474,8 @@ static int exec_queue_user_ext_set_property(struct xe_device *xe,
 			 ARRAY_SIZE(exec_queue_set_property_funcs)) ||
 	    XE_IOCTL_DBG(xe, ext.pad) ||
 	    XE_IOCTL_DBG(xe, ext.property != DRM_XE_EXEC_QUEUE_SET_PROPERTY_PRIORITY &&
-			 ext.property != DRM_XE_EXEC_QUEUE_SET_PROPERTY_TIMESLICE))
+			 ext.property != DRM_XE_EXEC_QUEUE_SET_PROPERTY_TIMESLICE &&
+			 ext.property != DRM_XE_EXEC_QUEUE_SET_PROPERTY_PXP_TYPE))
 		return -EINVAL;
 
 	idx = array_index_nospec(ext.property, ARRAY_SIZE(exec_queue_set_property_funcs));
@@ -448,7 +505,7 @@ static int exec_queue_user_extensions(struct xe_device *xe, struct xe_exec_queue
 	if (XE_IOCTL_DBG(xe, ext_number >= MAX_USER_EXTENSIONS))
 		return -E2BIG;
 
-	err = __copy_from_user(&ext, address, sizeof(ext));
+	err = copy_from_user(&ext, address, sizeof(ext));
 	if (XE_IOCTL_DBG(xe, err))
 		return -EFAULT;
 
@@ -470,7 +527,7 @@ static int exec_queue_user_extensions(struct xe_device *xe, struct xe_exec_queue
 	return 0;
 }
 
-static u32 calc_validate_logical_mask(struct xe_device *xe, struct xe_gt *gt,
+static u32 calc_validate_logical_mask(struct xe_device *xe,
 				      struct drm_xe_engine_class_instance *eci,
 				      u16 width, u16 num_placements)
 {
@@ -532,15 +589,15 @@ int xe_exec_queue_create_ioctl(struct drm_device *dev, void *data,
 		u64_to_user_ptr(args->instances);
 	struct xe_hw_engine *hwe;
 	struct xe_vm *vm;
-	struct xe_gt *gt;
 	struct xe_tile *tile;
 	struct xe_exec_queue *q = NULL;
 	u32 logical_mask;
+	u32 flags = 0;
 	u32 id;
 	u32 len;
 	int err;
 
-	if (XE_IOCTL_DBG(xe, args->flags) ||
+	if (XE_IOCTL_DBG(xe, args->flags & ~DRM_XE_EXEC_QUEUE_LOW_LATENCY_HINT) ||
 	    XE_IOCTL_DBG(xe, args->reserved[0] || args->reserved[1]))
 		return -EINVAL;
 
@@ -548,15 +605,17 @@ int xe_exec_queue_create_ioctl(struct drm_device *dev, void *data,
 	if (XE_IOCTL_DBG(xe, !len || len > XE_HW_ENGINE_MAX_INSTANCE))
 		return -EINVAL;
 
-	err = __copy_from_user(eci, user_eci,
-			       sizeof(struct drm_xe_engine_class_instance) *
-			       len);
+	err = copy_from_user(eci, user_eci,
+			     sizeof(struct drm_xe_engine_class_instance) * len);
 	if (XE_IOCTL_DBG(xe, err))
 		return -EFAULT;
 
 	if (XE_IOCTL_DBG(xe, eci[0].gt_id >= xe->info.gt_count))
 		return -EINVAL;
 
+	if (args->flags & DRM_XE_EXEC_QUEUE_LOW_LATENCY_HINT)
+		flags |= EXEC_QUEUE_FLAG_LOW_LATENCY;
+
 	if (eci[0].engine_class == DRM_XE_ENGINE_CLASS_VM_BIND) {
 		if (XE_IOCTL_DBG(xe, args->width != 1) ||
 		    XE_IOCTL_DBG(xe, args->num_placements != 1) ||
@@ -565,8 +624,8 @@ int xe_exec_queue_create_ioctl(struct drm_device *dev, void *data,
 
 		for_each_tile(tile, xe, id) {
 			struct xe_exec_queue *new;
-			u32 flags = EXEC_QUEUE_FLAG_VM;
 
+			flags |= EXEC_QUEUE_FLAG_VM;
 			if (id)
 				flags |= EXEC_QUEUE_FLAG_BIND_ENGINE_CHILD;
 
@@ -585,8 +644,7 @@ int xe_exec_queue_create_ioctl(struct drm_device *dev, void *data,
 					      &q->multi_gt_link);
 		}
 	} else {
-		gt = xe_device_get_gt(xe, eci[0].gt_id);
-		logical_mask = calc_validate_logical_mask(xe, gt, eci,
+		logical_mask = calc_validate_logical_mask(xe, eci,
 							  args->width,
 							  args->num_placements);
 		if (XE_IOCTL_DBG(xe, !logical_mask))
@@ -613,7 +671,7 @@ int xe_exec_queue_create_ioctl(struct drm_device *dev, void *data,
 		}
 
 		q = xe_exec_queue_create(xe, vm, logical_mask,
-					 args->width, hwe, 0,
+					 args->width, hwe, flags,
 					 args->extensions);
 		up_read(&vm->lock);
 		xe_vm_put(vm);
@@ -756,20 +814,21 @@ bool xe_exec_queue_is_idle(struct xe_exec_queue *q)
  */
 void xe_exec_queue_update_run_ticks(struct xe_exec_queue *q)
 {
-	struct xe_file *xef;
+	struct xe_device *xe = gt_to_xe(q->gt);
 	struct xe_lrc *lrc;
-	u32 old_ts, new_ts;
+	u64 old_ts, new_ts;
+	int idx;
 
 	/*
-	 * Jobs that are run during driver load may use an exec_queue, but are
-	 * not associated with a user xe file, so avoid accumulating busyness
-	 * for kernel specific work.
+	 * Jobs that are executed by kernel doesn't have a corresponding xe_file
+	 * and thus are not accounted.
 	 */
-	if (!q->vm || !q->vm->xef)
+	if (!q->xef)
 		return;
 
-	xef = q->vm->xef;
-
+	/* Synchronize with unbind while holding the xe file open */
+	if (!drm_dev_enter(&xe->drm, &idx))
+		return;
 	/*
 	 * Only sample the first LRC. For parallel submission, all of them are
 	 * scheduled together and we compensate that below by multiplying by
@@ -780,7 +839,9 @@ void xe_exec_queue_update_run_ticks(struct xe_exec_queue *q)
 	 */
 	lrc = q->lrc[0];
 	new_ts = xe_lrc_update_timestamp(lrc, &old_ts);
-	xef->run_ticks[q->class] += (new_ts - old_ts) * q->width;
+	q->xef->run_ticks[q->class] += (new_ts - old_ts) * q->width;
+
+	drm_dev_exit(idx);
 }
 
 /**
@@ -820,7 +881,10 @@ int xe_exec_queue_destroy_ioctl(struct drm_device *dev, void *data,
 
 	mutex_lock(&xef->exec_queue.lock);
 	q = xa_erase(&xef->exec_queue.xa, args->exec_queue_id);
+	if (q)
+		atomic_inc(&xef->exec_queue.pending_removal);
 	mutex_unlock(&xef->exec_queue.lock);
+
 	if (XE_IOCTL_DBG(xe, !q))
 		return -ENOENT;