Merge tag 'drm-fixes-2025-03-07' of https://gitlab.freedesktop.org/drm/kernel

Pull drm fixes from Dave Airlie:
 "Fixes across the board, mostly xe and imagination with some amd and
  misc others.

  The xe fixes are mostly hmm related, though there are some others in
  there as well, nothing really stands out otherwise.

  The nouveau Kconfig to select FW_CACHE is in this, which we discussed
  a while back.

  nouveau:
   - rely on fw caching Kconfig fix

  imagination:
   - avoid deadlock on fence release
   - fix fence initialisation
   - fix timestamps firmware traces

  scheduler:
   - fix include guard

  bochs:
   - dpms fix

  i915:
   - bump max stream count to match pipes

  xe:
   - Remove double page flip on initial plane
   - Properly setup userptr pfn_flags_mask
   - Fix GT "for each engine" workarounds
   - Fix userptr races and missed validations
   - Userptr invalid page access fixes
   - Cleanup some style nits

  amdgpu:
   - Fix NULL check in DC code
   - SMU 14 fix

  amdkfd:
   - Fix NULL check in queue validation

  radeon:
   - RS400 HyperZ fix"

* tag 'drm-fixes-2025-03-07' of https://gitlab.freedesktop.org/drm/kernel: (22 commits)
  drm/bochs: Fix DPMS regression
  drm/xe/userptr: Unmap userptrs in the mmu notifier
  drm/xe/hmm: Don't dereference struct page pointers without notifier lock
  drm/xe/hmm: Style- and include fixes
  drm/xe: Add staging tree for VM binds
  drm/xe: Fix fault mode invalidation with unbind
  drm/xe/vm: Fix a misplaced #endif
  drm/xe/vm: Validate userptr during gpu vma prefetching
  drm/amd/pm: always allow ih interrupt from fw
  drm/radeon: Fix rs400_gpu_init for ATI mobility radeon Xpress 200M
  drm/amdkfd: Fix NULL Pointer Dereference in KFD queue
  drm/amd/display: Fix null check for pipe_ctx->plane_state in resource_build_scaling_params
  drm/xe: Fix GT "for each engine" workarounds
  drm/xe/userptr: properly setup pfn_flags_mask
  drm/i915/mst: update max stream count to match number of pipes
  drm/xe: Remove double pageflip
  drm/sched: Fix preprocessor guard
  drm/imagination: Fix timestamps in firmware traces
  drm/imagination: only init job done fences once
  drm/imagination: Hold drm_gem_gpuva lock for unmap
  ...

26 files changed, 456 insertions, 197 deletions

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_queue.c b/drivers/gpu/drm/amd/amdkfd/kfd_queue.c
index ecccd7adbab4..24396a2c77bd 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_queue.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_queue.c
@@ -266,8 +266,8 @@ int kfd_queue_acquire_buffers(struct kfd_process_device *pdd, struct queue_prope
 	/* EOP buffer is not required for all ASICs */
 	if (properties->eop_ring_buffer_address) {
 		if (properties->eop_ring_buffer_size != topo_dev->node_props.eop_buffer_size) {
-			pr_debug("queue eop bo size 0x%lx not equal to node eop buf size 0x%x\n",
-				properties->eop_buf_bo->tbo.base.size,
+			pr_debug("queue eop bo size 0x%x not equal to node eop buf size 0x%x\n",
+				properties->eop_ring_buffer_size,
 				topo_dev->node_props.eop_buffer_size);
 			err = -EINVAL;
 			goto out_err_unreserve;
diff --git a/drivers/gpu/drm/amd/display/dc/core/dc_resource.c b/drivers/gpu/drm/amd/display/dc/core/dc_resource.c
index 520a34a42827..a45037cb4cc0 100644
--- a/drivers/gpu/drm/amd/display/dc/core/dc_resource.c
+++ b/drivers/gpu/drm/amd/display/dc/core/dc_resource.c
@@ -1455,7 +1455,8 @@ bool resource_build_scaling_params(struct pipe_ctx *pipe_ctx)
 	DC_LOGGER_INIT(pipe_ctx->stream->ctx->logger);
 
 	/* Invalid input */
-	if (!plane_state->dst_rect.width ||
+	if (!plane_state ||
+			!plane_state->dst_rect.width ||
 			!plane_state->dst_rect.height ||
 			!plane_state->src_rect.width ||
 			!plane_state->src_rect.height) {
diff --git a/drivers/gpu/drm/amd/pm/swsmu/smu14/smu_v14_0.c b/drivers/gpu/drm/amd/pm/swsmu/smu14/smu_v14_0.c
index 9b2f4fe1578b..ddb6444406d2 100644
--- a/drivers/gpu/drm/amd/pm/swsmu/smu14/smu_v14_0.c
+++ b/drivers/gpu/drm/amd/pm/swsmu/smu14/smu_v14_0.c
@@ -1895,16 +1895,6 @@ static int smu_v14_0_allow_ih_interrupt(struct smu_context *smu)
 				    NULL);
 }
 
-static int smu_v14_0_process_pending_interrupt(struct smu_context *smu)
-{
-	int ret = 0;
-
-	if (smu_cmn_feature_is_enabled(smu, SMU_FEATURE_ACDC_BIT))
-		ret = smu_v14_0_allow_ih_interrupt(smu);
-
-	return ret;
-}
-
 int smu_v14_0_enable_thermal_alert(struct smu_context *smu)
 {
 	int ret = 0;
@@ -1916,7 +1906,7 @@ int smu_v14_0_enable_thermal_alert(struct smu_context *smu)
 	if (ret)
 		return ret;
 
-	return smu_v14_0_process_pending_interrupt(smu);
+	return smu_v14_0_allow_ih_interrupt(smu);
 }
 
 int smu_v14_0_disable_thermal_alert(struct smu_context *smu)
diff --git a/drivers/gpu/drm/i915/display/intel_dp_mst.c b/drivers/gpu/drm/i915/display/intel_dp_mst.c
index a65cf97ad12d..86d6185fda50 100644
--- a/drivers/gpu/drm/i915/display/intel_dp_mst.c
+++ b/drivers/gpu/drm/i915/display/intel_dp_mst.c
@@ -1867,7 +1867,8 @@ intel_dp_mst_encoder_init(struct intel_digital_port *dig_port, int conn_base_id)
 	/* create encoders */
 	mst_stream_encoders_create(dig_port);
 	ret = drm_dp_mst_topology_mgr_init(&intel_dp->mst_mgr, display->drm,
-					   &intel_dp->aux, 16, 3, conn_base_id);
+					   &intel_dp->aux, 16,
+					   INTEL_NUM_PIPES(display), conn_base_id);
 	if (ret) {
 		intel_dp->mst_mgr.cbs = NULL;
 		return ret;
diff --git a/drivers/gpu/drm/imagination/pvr_fw_meta.c b/drivers/gpu/drm/imagination/pvr_fw_meta.c
index c39beb70c317..6d13864851fc 100644
--- a/drivers/gpu/drm/imagination/pvr_fw_meta.c
+++ b/drivers/gpu/drm/imagination/pvr_fw_meta.c
@@ -527,8 +527,10 @@ pvr_meta_vm_map(struct pvr_device *pvr_dev, struct pvr_fw_object *fw_obj)
 static void
 pvr_meta_vm_unmap(struct pvr_device *pvr_dev, struct pvr_fw_object *fw_obj)
 {
-	pvr_vm_unmap(pvr_dev->kernel_vm_ctx, fw_obj->fw_mm_node.start,
-		     fw_obj->fw_mm_node.size);
+	struct pvr_gem_object *pvr_obj = fw_obj->gem;
+
+	pvr_vm_unmap_obj(pvr_dev->kernel_vm_ctx, pvr_obj,
+			 fw_obj->fw_mm_node.start, fw_obj->fw_mm_node.size);
 }
 
 static bool
diff --git a/drivers/gpu/drm/imagination/pvr_fw_trace.c b/drivers/gpu/drm/imagination/pvr_fw_trace.c
index 73707daa4e52..5dbb636d7d4f 100644
--- a/drivers/gpu/drm/imagination/pvr_fw_trace.c
+++ b/drivers/gpu/drm/imagination/pvr_fw_trace.c
@@ -333,8 +333,8 @@ static int fw_trace_seq_show(struct seq_file *s, void *v)
 	if (sf_id == ROGUE_FW_SF_LAST)
 		return -EINVAL;
 
-	timestamp = read_fw_trace(trace_seq_data, 1) |
-		((u64)read_fw_trace(trace_seq_data, 2) << 32);
+	timestamp = ((u64)read_fw_trace(trace_seq_data, 1) << 32) |
+		read_fw_trace(trace_seq_data, 2);
 	timestamp = (timestamp & ~ROGUE_FWT_TIMESTAMP_TIME_CLRMSK) >>
 		ROGUE_FWT_TIMESTAMP_TIME_SHIFT;
 
diff --git a/drivers/gpu/drm/imagination/pvr_queue.c b/drivers/gpu/drm/imagination/pvr_queue.c
index c4f08432882b..43411be930a2 100644
--- a/drivers/gpu/drm/imagination/pvr_queue.c
+++ b/drivers/gpu/drm/imagination/pvr_queue.c
@@ -109,12 +109,20 @@ pvr_queue_fence_get_driver_name(struct dma_fence *f)
 	return PVR_DRIVER_NAME;
 }
 
+static void pvr_queue_fence_release_work(struct work_struct *w)
+{
+	struct pvr_queue_fence *fence = container_of(w, struct pvr_queue_fence, release_work);
+
+	pvr_context_put(fence->queue->ctx);
+	dma_fence_free(&fence->base);
+}
+
 static void pvr_queue_fence_release(struct dma_fence *f)
 {
 	struct pvr_queue_fence *fence = container_of(f, struct pvr_queue_fence, base);
+	struct pvr_device *pvr_dev = fence->queue->ctx->pvr_dev;
 
-	pvr_context_put(fence->queue->ctx);
-	dma_fence_free(f);
+	queue_work(pvr_dev->sched_wq, &fence->release_work);
 }
 
 static const char *
@@ -268,6 +276,7 @@ pvr_queue_fence_init(struct dma_fence *f,
 
 	pvr_context_get(queue->ctx);
 	fence->queue = queue;
+	INIT_WORK(&fence->release_work, pvr_queue_fence_release_work);
 	dma_fence_init(&fence->base, fence_ops,
 		       &fence_ctx->lock, fence_ctx->id,
 		       atomic_inc_return(&fence_ctx->seqno));
@@ -304,8 +313,9 @@ pvr_queue_cccb_fence_init(struct dma_fence *fence, struct pvr_queue *queue)
 static void
 pvr_queue_job_fence_init(struct dma_fence *fence, struct pvr_queue *queue)
 {
-	pvr_queue_fence_init(fence, queue, &pvr_queue_job_fence_ops,
-			     &queue->job_fence_ctx);
+	if (!fence->ops)
+		pvr_queue_fence_init(fence, queue, &pvr_queue_job_fence_ops,
+				     &queue->job_fence_ctx);
 }
 
 /**
diff --git a/drivers/gpu/drm/imagination/pvr_queue.h b/drivers/gpu/drm/imagination/pvr_queue.h
index e06ced69302f..93fe9ac9f58c 100644
--- a/drivers/gpu/drm/imagination/pvr_queue.h
+++ b/drivers/gpu/drm/imagination/pvr_queue.h
@@ -5,6 +5,7 @@
 #define PVR_QUEUE_H
 
 #include <drm/gpu_scheduler.h>
+#include <linux/workqueue.h>
 
 #include "pvr_cccb.h"
 #include "pvr_device.h"
@@ -63,6 +64,9 @@ struct pvr_queue_fence {
 
 	/** @queue: Queue that created this fence. */
 	struct pvr_queue *queue;
+
+	/** @release_work: Fence release work structure. */
+	struct work_struct release_work;
 };
 
 /**
diff --git a/drivers/gpu/drm/imagination/pvr_vm.c b/drivers/gpu/drm/imagination/pvr_vm.c
index 363f885a7098..2896fa7501b1 100644
--- a/drivers/gpu/drm/imagination/pvr_vm.c
+++ b/drivers/gpu/drm/imagination/pvr_vm.c
@@ -293,8 +293,9 @@ err_bind_op_fini:
 
 static int
 pvr_vm_bind_op_unmap_init(struct pvr_vm_bind_op *bind_op,
-			  struct pvr_vm_context *vm_ctx, u64 device_addr,
-			  u64 size)
+			  struct pvr_vm_context *vm_ctx,
+			  struct pvr_gem_object *pvr_obj,
+			  u64 device_addr, u64 size)
 {
 	int err;
 
@@ -318,6 +319,7 @@ pvr_vm_bind_op_unmap_init(struct pvr_vm_bind_op *bind_op,
 		goto err_bind_op_fini;
 	}
 
+	bind_op->pvr_obj = pvr_obj;
 	bind_op->vm_ctx = vm_ctx;
 	bind_op->device_addr = device_addr;
 	bind_op->size = size;
@@ -598,20 +600,6 @@ err_free:
 }
 
 /**
- * pvr_vm_unmap_all() - Unmap all mappings associated with a VM context.
- * @vm_ctx: Target VM context.
- *
- * This function ensures that no mappings are left dangling by unmapping them
- * all in order of ascending device-virtual address.
- */
-void
-pvr_vm_unmap_all(struct pvr_vm_context *vm_ctx)
-{
-	WARN_ON(pvr_vm_unmap(vm_ctx, vm_ctx->gpuvm_mgr.mm_start,
-			     vm_ctx->gpuvm_mgr.mm_range));
-}
-
-/**
  * pvr_vm_context_release() - Teardown a VM context.
  * @ref_count: Pointer to reference counter of the VM context.
  *
@@ -703,11 +691,7 @@ pvr_vm_lock_extra(struct drm_gpuvm_exec *vm_exec)
 	struct pvr_vm_bind_op *bind_op = vm_exec->extra.priv;
 	struct pvr_gem_object *pvr_obj = bind_op->pvr_obj;
 
-	/* Unmap operations don't have an object to lock. */
-	if (!pvr_obj)
-		return 0;
-
-	/* Acquire lock on the GEM being mapped. */
+	/* Acquire lock on the GEM object being mapped/unmapped. */
 	return drm_exec_lock_obj(&vm_exec->exec, gem_from_pvr_gem(pvr_obj));
 }
 
@@ -772,8 +756,10 @@ err_cleanup:
 }
 
 /**
- * pvr_vm_unmap() - Unmap an already mapped section of device-virtual memory.
+ * pvr_vm_unmap_obj_locked() - Unmap an already mapped section of device-virtual
+ * memory.
  * @vm_ctx: Target VM context.
+ * @pvr_obj: Target PowerVR memory object.
  * @device_addr: Virtual device address at the start of the target mapping.
  * @size: Size of the target mapping.
  *
@@ -784,9 +770,13 @@ err_cleanup:
  *  * Any error encountered while performing internal operations required to
  *    destroy the mapping (returned from pvr_vm_gpuva_unmap or
  *    pvr_vm_gpuva_remap).
+ *
+ * The vm_ctx->lock must be held when calling this function.
  */
-int
-pvr_vm_unmap(struct pvr_vm_context *vm_ctx, u64 device_addr, u64 size)
+static int
+pvr_vm_unmap_obj_locked(struct pvr_vm_context *vm_ctx,
+			struct pvr_gem_object *pvr_obj,
+			u64 device_addr, u64 size)
 {
 	struct pvr_vm_bind_op bind_op = {0};
 	struct drm_gpuvm_exec vm_exec = {
@@ -799,11 +789,13 @@ pvr_vm_unmap(struct pvr_vm_context *vm_ctx, u64 device_addr, u64 size)
 		},
 	};
 
-	int err = pvr_vm_bind_op_unmap_init(&bind_op, vm_ctx, device_addr,
-					    size);
+	int err = pvr_vm_bind_op_unmap_init(&bind_op, vm_ctx, pvr_obj,
+					    device_addr, size);
 	if (err)
 		return err;
 
+	pvr_gem_object_get(pvr_obj);
+
 	err = drm_gpuvm_exec_lock(&vm_exec);
 	if (err)
 		goto err_cleanup;
@@ -818,6 +810,96 @@ err_cleanup:
 	return err;
 }
 
+/**
+ * pvr_vm_unmap_obj() - Unmap an already mapped section of device-virtual
+ * memory.
+ * @vm_ctx: Target VM context.
+ * @pvr_obj: Target PowerVR memory object.
+ * @device_addr: Virtual device address at the start of the target mapping.
+ * @size: Size of the target mapping.
+ *
+ * Return:
+ *  * 0 on success,
+ *  * Any error encountered by pvr_vm_unmap_obj_locked.
+ */
+int
+pvr_vm_unmap_obj(struct pvr_vm_context *vm_ctx, struct pvr_gem_object *pvr_obj,
+		 u64 device_addr, u64 size)
+{
+	int err;
+
+	mutex_lock(&vm_ctx->lock);
+	err = pvr_vm_unmap_obj_locked(vm_ctx, pvr_obj, device_addr, size);
+	mutex_unlock(&vm_ctx->lock);
+
+	return err;
+}
+
+/**
+ * pvr_vm_unmap() - Unmap an already mapped section of device-virtual memory.
+ * @vm_ctx: Target VM context.
+ * @device_addr: Virtual device address at the start of the target mapping.
+ * @size: Size of the target mapping.
+ *
+ * Return:
+ *  * 0 on success,
+ *  * Any error encountered by drm_gpuva_find,
+ *  * Any error encountered by pvr_vm_unmap_obj_locked.
+ */
+int
+pvr_vm_unmap(struct pvr_vm_context *vm_ctx, u64 device_addr, u64 size)
+{
+	struct pvr_gem_object *pvr_obj;
+	struct drm_gpuva *va;
+	int err;
+
+	mutex_lock(&vm_ctx->lock);
+
+	va = drm_gpuva_find(&vm_ctx->gpuvm_mgr, device_addr, size);
+	if (va) {
+		pvr_obj = gem_to_pvr_gem(va->gem.obj);
+		err = pvr_vm_unmap_obj_locked(vm_ctx, pvr_obj,
+					      va->va.addr, va->va.range);
+	} else {
+		err = -ENOENT;
+	}
+
+	mutex_unlock(&vm_ctx->lock);
+
+	return err;
+}
+
+/**
+ * pvr_vm_unmap_all() - Unmap all mappings associated with a VM context.
+ * @vm_ctx: Target VM context.
+ *
+ * This function ensures that no mappings are left dangling by unmapping them
+ * all in order of ascending device-virtual address.
+ */
+void
+pvr_vm_unmap_all(struct pvr_vm_context *vm_ctx)
+{
+	mutex_lock(&vm_ctx->lock);
+
+	for (;;) {
+		struct pvr_gem_object *pvr_obj;
+		struct drm_gpuva *va;
+
+		va = drm_gpuva_find_first(&vm_ctx->gpuvm_mgr,
+					  vm_ctx->gpuvm_mgr.mm_start,
+					  vm_ctx->gpuvm_mgr.mm_range);
+		if (!va)
+			break;
+
+		pvr_obj = gem_to_pvr_gem(va->gem.obj);
+
+		WARN_ON(pvr_vm_unmap_obj_locked(vm_ctx, pvr_obj,
+						va->va.addr, va->va.range));
+	}
+
+	mutex_unlock(&vm_ctx->lock);
+}
+
 /* Static data areas are determined by firmware. */
 static const struct drm_pvr_static_data_area static_data_areas[] = {
 	{
diff --git a/drivers/gpu/drm/imagination/pvr_vm.h b/drivers/gpu/drm/imagination/pvr_vm.h
index 79406243617c..b0528dffa7f1 100644
--- a/drivers/gpu/drm/imagination/pvr_vm.h
+++ b/drivers/gpu/drm/imagination/pvr_vm.h
@@ -38,6 +38,9 @@ struct pvr_vm_context *pvr_vm_create_context(struct pvr_device *pvr_dev,
 int pvr_vm_map(struct pvr_vm_context *vm_ctx,
 	       struct pvr_gem_object *pvr_obj, u64 pvr_obj_offset,
 	       u64 device_addr, u64 size);
+int pvr_vm_unmap_obj(struct pvr_vm_context *vm_ctx,
+		     struct pvr_gem_object *pvr_obj,
+		     u64 device_addr, u64 size);
 int pvr_vm_unmap(struct pvr_vm_context *vm_ctx, u64 device_addr, u64 size);
 void pvr_vm_unmap_all(struct pvr_vm_context *vm_ctx);
 
diff --git a/drivers/gpu/drm/nouveau/Kconfig b/drivers/gpu/drm/nouveau/Kconfig
index ce840300578d..1050a4617fc1 100644
--- a/drivers/gpu/drm/nouveau/Kconfig
+++ b/drivers/gpu/drm/nouveau/Kconfig
@@ -4,6 +4,7 @@ config DRM_NOUVEAU
 	depends on DRM && PCI && MMU
 	select IOMMU_API
 	select FW_LOADER
+	select FW_CACHE if PM_SLEEP
 	select DRM_CLIENT_SELECTION
 	select DRM_DISPLAY_DP_HELPER
 	select DRM_DISPLAY_HDMI_HELPER
diff --git a/drivers/gpu/drm/radeon/r300.c b/drivers/gpu/drm/radeon/r300.c
index 05c13102a8cb..d22889fbfa9c 100644
--- a/drivers/gpu/drm/radeon/r300.c
+++ b/drivers/gpu/drm/radeon/r300.c
@@ -359,7 +359,8 @@ int r300_mc_wait_for_idle(struct radeon_device *rdev)
 	return -1;
 }
 
-static void r300_gpu_init(struct radeon_device *rdev)
+/* rs400_gpu_init also calls this! */
+void r300_gpu_init(struct radeon_device *rdev)
 {
 	uint32_t gb_tile_config, tmp;
 
diff --git a/drivers/gpu/drm/radeon/radeon_asic.h b/drivers/gpu/drm/radeon/radeon_asic.h
index 1e00f6b99f94..8f5e07834fcc 100644
--- a/drivers/gpu/drm/radeon/radeon_asic.h
+++ b/drivers/gpu/drm/radeon/radeon_asic.h
@@ -165,6 +165,7 @@ void r200_set_safe_registers(struct radeon_device *rdev);
  */
 extern int r300_init(struct radeon_device *rdev);
 extern void r300_fini(struct radeon_device *rdev);
+extern void r300_gpu_init(struct radeon_device *rdev);
 extern int r300_suspend(struct radeon_device *rdev);
 extern int r300_resume(struct radeon_device *rdev);
 extern int r300_asic_reset(struct radeon_device *rdev, bool hard);
diff --git a/drivers/gpu/drm/radeon/rs400.c b/drivers/gpu/drm/radeon/rs400.c
index d6c18fd740ec..13cd0a688a65 100644
--- a/drivers/gpu/drm/radeon/rs400.c
+++ b/drivers/gpu/drm/radeon/rs400.c
@@ -256,8 +256,22 @@ int rs400_mc_wait_for_idle(struct radeon_device *rdev)
 
 static void rs400_gpu_init(struct radeon_device *rdev)
 {
-	/* FIXME: is this correct ? */
-	r420_pipes_init(rdev);
+	/* Earlier code was calling r420_pipes_init and then
+	 * rs400_mc_wait_for_idle(rdev). The problem is that
+	 * at least on my Mobility Radeon Xpress 200M RC410 card
+	 * that ends up in this code path ends up num_gb_pipes == 3
+	 * while the card seems to have only one pipe. With the
+	 * r420 pipe initialization method.
+	 *
+	 * Problems shown up as HyperZ glitches, see:
+	 * https://bugs.freedesktop.org/show_bug.cgi?id=110897
+	 *
+	 * Delegating initialization to r300 code seems to work
+	 * and results in proper pipe numbers. The rs400 cards
+	 * are said to be not r400, but r300 kind of cards.
+	 */
+	r300_gpu_init(rdev);
+
 	if (rs400_mc_wait_for_idle(rdev)) {
 		pr_warn("rs400: Failed to wait MC idle while programming pipes. Bad things might happen. %08x\n",
 			RREG32(RADEON_MC_STATUS));
diff --git a/drivers/gpu/drm/scheduler/gpu_scheduler_trace.h b/drivers/gpu/drm/scheduler/gpu_scheduler_trace.h
index c75302ca3427..f56e77e7f6d0 100644
--- a/drivers/gpu/drm/scheduler/gpu_scheduler_trace.h
+++ b/drivers/gpu/drm/scheduler/gpu_scheduler_trace.h
@@ -21,7 +21,7 @@
  *
  */
 
-#if !defined(_GPU_SCHED_TRACE_H) || defined(TRACE_HEADER_MULTI_READ)
+#if !defined(_GPU_SCHED_TRACE_H_) || defined(TRACE_HEADER_MULTI_READ)
 #define _GPU_SCHED_TRACE_H_
 
 #include <linux/stringify.h>
@@ -106,7 +106,7 @@ TRACE_EVENT(drm_sched_job_wait_dep,
 		      __entry->seqno)
 );
 
-#endif
+#endif /* _GPU_SCHED_TRACE_H_ */
 
 /* This part must be outside protection */
 #undef TRACE_INCLUDE_PATH
diff --git a/drivers/gpu/drm/tiny/bochs.c b/drivers/gpu/drm/tiny/bochs.c
index c67e1f906785..8706763af8fb 100644
--- a/drivers/gpu/drm/tiny/bochs.c
+++ b/drivers/gpu/drm/tiny/bochs.c
@@ -335,8 +335,6 @@ static void bochs_hw_setmode(struct bochs_device *bochs, struct drm_display_mode
 			 bochs->xres, bochs->yres, bochs->bpp,
 			 bochs->yres_virtual);
 
-	bochs_hw_blank(bochs, false);
-
 	bochs_dispi_write(bochs, VBE_DISPI_INDEX_ENABLE,      0);
 	bochs_dispi_write(bochs, VBE_DISPI_INDEX_BPP,         bochs->bpp);
 	bochs_dispi_write(bochs, VBE_DISPI_INDEX_XRES,        bochs->xres);
@@ -506,6 +504,9 @@ static int bochs_crtc_helper_atomic_check(struct drm_crtc *crtc,
 static void bochs_crtc_helper_atomic_enable(struct drm_crtc *crtc,
 					    struct drm_atomic_state *state)
 {
+	struct bochs_device *bochs = to_bochs_device(crtc->dev);
+
+	bochs_hw_blank(bochs, false);
 }
 
 static void bochs_crtc_helper_atomic_disable(struct drm_crtc *crtc,
diff --git a/drivers/gpu/drm/xe/display/xe_plane_initial.c b/drivers/gpu/drm/xe/display/xe_plane_initial.c
index 2eb9633f163a..2a2f250fa495 100644
--- a/drivers/gpu/drm/xe/display/xe_plane_initial.c
+++ b/drivers/gpu/drm/xe/display/xe_plane_initial.c
@@ -194,8 +194,6 @@ intel_find_initial_plane_obj(struct intel_crtc *crtc,
 		to_intel_plane(crtc->base.primary);
 	struct intel_plane_state *plane_state =
 		to_intel_plane_state(plane->base.state);
-	struct intel_crtc_state *crtc_state =
-		to_intel_crtc_state(crtc->base.state);
 	struct drm_framebuffer *fb;
 	struct i915_vma *vma;
 
@@ -241,14 +239,6 @@ intel_find_initial_plane_obj(struct intel_crtc *crtc,
 	atomic_or(plane->frontbuffer_bit, &to_intel_frontbuffer(fb)->bits);
 
 	plane_config->vma = vma;
-
-	/*
-	 * Flip to the newly created mapping ASAP, so we can re-use the
-	 * first part of GGTT for WOPCM, prevent flickering, and prevent
-	 * the lookup of sysmem scratch pages.
-	 */
-	plane->check_plane(crtc_state, plane_state);
-	plane->async_flip(NULL, plane, crtc_state, plane_state, true);
 	return;
 
 nofb:
diff --git a/drivers/gpu/drm/xe/xe_gt.c b/drivers/gpu/drm/xe/xe_gt.c
index 5d6fb79957b6..9f4f27d1ef4a 100644
--- a/drivers/gpu/drm/xe/xe_gt.c
+++ b/drivers/gpu/drm/xe/xe_gt.c
@@ -380,9 +380,7 @@ int xe_gt_init_early(struct xe_gt *gt)
 	if (err)
 		return err;
 
-	xe_wa_process_gt(gt);
 	xe_wa_process_oob(gt);
-	xe_tuning_process_gt(gt);
 
 	xe_force_wake_init_gt(gt, gt_to_fw(gt));
 	spin_lock_init(&gt->global_invl_lock);
@@ -474,6 +472,8 @@ static int all_fw_domain_init(struct xe_gt *gt)
 	}
 
 	xe_gt_mcr_set_implicit_defaults(gt);
+	xe_wa_process_gt(gt);
+	xe_tuning_process_gt(gt);
 	xe_reg_sr_apply_mmio(&gt->reg_sr, gt);
 
 	err = xe_gt_clock_init(gt);
diff --git a/drivers/gpu/drm/xe/xe_hmm.c b/drivers/gpu/drm/xe/xe_hmm.c
index 089834467880..392102515f3d 100644
--- a/drivers/gpu/drm/xe/xe_hmm.c
+++ b/drivers/gpu/drm/xe/xe_hmm.c
@@ -19,11 +19,10 @@ static u64 xe_npages_in_range(unsigned long start, unsigned long end)
 	return (end - start) >> PAGE_SHIFT;
 }
 
-/*
+/**
  * xe_mark_range_accessed() - mark a range is accessed, so core mm
  * have such information for memory eviction or write back to
  * hard disk
- *
  * @range: the range to mark
  * @write: if write to this range, we mark pages in this range
  * as dirty
@@ -43,15 +42,51 @@ static void xe_mark_range_accessed(struct hmm_range *range, bool write)
 	}
 }
 
-/*
+static int xe_alloc_sg(struct xe_device *xe, struct sg_table *st,
+		       struct hmm_range *range, struct rw_semaphore *notifier_sem)
+{
+	unsigned long i, npages, hmm_pfn;
+	unsigned long num_chunks = 0;
+	int ret;
+
+	/* HMM docs says this is needed. */
+	ret = down_read_interruptible(notifier_sem);
+	if (ret)
+		return ret;
+
+	if (mmu_interval_read_retry(range->notifier, range->notifier_seq)) {
+		up_read(notifier_sem);
+		return -EAGAIN;
+	}
+
+	npages = xe_npages_in_range(range->start, range->end);
+	for (i = 0; i < npages;) {
+		unsigned long len;
+
+		hmm_pfn = range->hmm_pfns[i];
+		xe_assert(xe, hmm_pfn & HMM_PFN_VALID);
+
+		len = 1UL << hmm_pfn_to_map_order(hmm_pfn);
+
+		/* If order > 0 the page may extend beyond range->start */
+		len -= (hmm_pfn & ~HMM_PFN_FLAGS) & (len - 1);
+		i += len;
+		num_chunks++;
+	}
+	up_read(notifier_sem);
+
+	return sg_alloc_table(st, num_chunks, GFP_KERNEL);
+}
+
+/**
  * xe_build_sg() - build a scatter gather table for all the physical pages/pfn
  * in a hmm_range. dma-map pages if necessary. dma-address is save in sg table
  * and will be used to program GPU page table later.
- *
  * @xe: the xe device who will access the dma-address in sg table
  * @range: the hmm range that we build the sg table from. range->hmm_pfns[]
  * has the pfn numbers of pages that back up this hmm address range.
  * @st: pointer to the sg table.
+ * @notifier_sem: The xe notifier lock.
  * @write: whether we write to this range. This decides dma map direction
  * for system pages. If write we map it bi-diretional; otherwise
  * DMA_TO_DEVICE
@@ -78,43 +113,84 @@ static void xe_mark_range_accessed(struct hmm_range *range, bool write)
  * Returns 0 if successful; -ENOMEM if fails to allocate memory
  */
 static int xe_build_sg(struct xe_device *xe, struct hmm_range *range,
-		       struct sg_table *st, bool write)
+		       struct sg_table *st,
+		       struct rw_semaphore *notifier_sem,
+		       bool write)
 {
+	unsigned long npages = xe_npages_in_range(range->start, range->end);
 	struct device *dev = xe->drm.dev;
-	struct page **pages;
-	u64 i, npages;
-	int ret;
+	struct scatterlist *sgl;
+	struct page *page;
+	unsigned long i, j;
 
-	npages = xe_npages_in_range(range->start, range->end);
-	pages = kvmalloc_array(npages, sizeof(*pages), GFP_KERNEL);
-	if (!pages)
-		return -ENOMEM;
+	lockdep_assert_held(notifier_sem);
 
-	for (i = 0; i < npages; i++) {
-		pages[i] = hmm_pfn_to_page(range->hmm_pfns[i]);
-		xe_assert(xe, !is_device_private_page(pages[i]));
+	i = 0;
+	for_each_sg(st->sgl, sgl, st->nents, j) {
+		unsigned long hmm_pfn, size;
+
+		hmm_pfn = range->hmm_pfns[i];
+		page = hmm_pfn_to_page(hmm_pfn);
+		xe_assert(xe, !is_device_private_page(page));
+
+		size = 1UL << hmm_pfn_to_map_order(hmm_pfn);
+		size -= page_to_pfn(page) & (size - 1);
+		i += size;
+
+		if (unlikely(j == st->nents - 1)) {
+			if (i > npages)
+				size -= (i - npages);
+			sg_mark_end(sgl);
+		}
+		sg_set_page(sgl, page, size << PAGE_SHIFT, 0);
 	}
+	xe_assert(xe, i == npages);
 
-	ret = sg_alloc_table_from_pages_segment(st, pages, npages, 0, npages << PAGE_SHIFT,
-						xe_sg_segment_size(dev), GFP_KERNEL);
-	if (ret)
-		goto free_pages;
+	return dma_map_sgtable(dev, st, write ? DMA_BIDIRECTIONAL : DMA_TO_DEVICE,
+			       DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_NO_KERNEL_MAPPING);
+}
+
+static void xe_hmm_userptr_set_mapped(struct xe_userptr_vma *uvma)
+{
+	struct xe_userptr *userptr = &uvma->userptr;
+	struct xe_vm *vm = xe_vma_vm(&uvma->vma);
+
+	lockdep_assert_held_write(&vm->lock);
+	lockdep_assert_held(&vm->userptr.notifier_lock);
+
+	mutex_lock(&userptr->unmap_mutex);
+	xe_assert(vm->xe, !userptr->mapped);
+	userptr->mapped = true;
+	mutex_unlock(&userptr->unmap_mutex);
+}
+
+void xe_hmm_userptr_unmap(struct xe_userptr_vma *uvma)
+{
+	struct xe_userptr *userptr = &uvma->userptr;
+	struct xe_vma *vma = &uvma->vma;
+	bool write = !xe_vma_read_only(vma);
+	struct xe_vm *vm = xe_vma_vm(vma);
+	struct xe_device *xe = vm->xe;
 
-	ret = dma_map_sgtable(dev, st, write ? DMA_BIDIRECTIONAL : DMA_TO_DEVICE,
-			      DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_NO_KERNEL_MAPPING);
-	if (ret) {
-		sg_free_table(st);
-		st = NULL;
+	if (!lockdep_is_held_type(&vm->userptr.notifier_lock, 0) &&
+	    !lockdep_is_held_type(&vm->lock, 0) &&
+	    !(vma->gpuva.flags & XE_VMA_DESTROYED)) {
+		/* Don't unmap in exec critical section. */
+		xe_vm_assert_held(vm);
+		/* Don't unmap while mapping the sg. */
+		lockdep_assert_held(&vm->lock);
 	}
 
-free_pages:
-	kvfree(pages);
-	return ret;
+	mutex_lock(&userptr->unmap_mutex);
+	if (userptr->sg && userptr->mapped)
+		dma_unmap_sgtable(xe->drm.dev, userptr->sg,
+				  write ? DMA_BIDIRECTIONAL : DMA_TO_DEVICE, 0);
+	userptr->mapped = false;
+	mutex_unlock(&userptr->unmap_mutex);
 }
 
-/*
+/**
  * xe_hmm_userptr_free_sg() - Free the scatter gather table of userptr
- *
  * @uvma: the userptr vma which hold the scatter gather table
  *
  * With function xe_userptr_populate_range, we allocate storage of
@@ -124,16 +200,9 @@ free_pages:
 void xe_hmm_userptr_free_sg(struct xe_userptr_vma *uvma)
 {
 	struct xe_userptr *userptr = &uvma->userptr;
-	struct xe_vma *vma = &uvma->vma;
-	bool write = !xe_vma_read_only(vma);
-	struct xe_vm *vm = xe_vma_vm(vma);
-	struct xe_device *xe = vm->xe;
-	struct device *dev = xe->drm.dev;
-
-	xe_assert(xe, userptr->sg);
-	dma_unmap_sgtable(dev, userptr->sg,
-			  write ? DMA_BIDIRECTIONAL : DMA_TO_DEVICE, 0);
 
+	xe_assert(xe_vma_vm(&uvma->vma)->xe, userptr->sg);
+	xe_hmm_userptr_unmap(uvma);
 	sg_free_table(userptr->sg);
 	userptr->sg = NULL;
 }
@@ -166,13 +235,20 @@ int xe_hmm_userptr_populate_range(struct xe_userptr_vma *uvma,
 {
 	unsigned long timeout =
 		jiffies + msecs_to_jiffies(HMM_RANGE_DEFAULT_TIMEOUT);
-	unsigned long *pfns, flags = HMM_PFN_REQ_FAULT;
+	unsigned long *pfns;
 	struct xe_userptr *userptr;
 	struct xe_vma *vma = &uvma->vma;
 	u64 userptr_start = xe_vma_userptr(vma);
 	u64 userptr_end = userptr_start + xe_vma_size(vma);
 	struct xe_vm *vm = xe_vma_vm(vma);
-	struct hmm_range hmm_range;
+	struct hmm_range hmm_range = {
+		.pfn_flags_mask = 0, /* ignore pfns */
+		.default_flags = HMM_PFN_REQ_FAULT,
+		.start = userptr_start,
+		.end = userptr_end,
+		.notifier = &uvma->userptr.notifier,
+		.dev_private_owner = vm->xe,
+	};
 	bool write = !xe_vma_read_only(vma);
 	unsigned long notifier_seq;
 	u64 npages;
@@ -199,19 +275,14 @@ int xe_hmm_userptr_populate_range(struct xe_userptr_vma *uvma,
 		return -ENOMEM;
 
 	if (write)
-		flags |= HMM_PFN_REQ_WRITE;
+		hmm_range.default_flags |= HMM_PFN_REQ_WRITE;
 
 	if (!mmget_not_zero(userptr->notifier.mm)) {
 		ret = -EFAULT;
 		goto free_pfns;
 	}
 
-	hmm_range.default_flags = flags;
 	hmm_range.hmm_pfns = pfns;
-	hmm_range.notifier = &userptr->notifier;
-	hmm_range.start = userptr_start;
-	hmm_range.end = userptr_end;
-	hmm_range.dev_private_owner = vm->xe;
 
 	while (true) {
 		hmm_range.notifier_seq = mmu_interval_read_begin(&userptr->notifier);
@@ -238,16 +309,37 @@ int xe_hmm_userptr_populate_range(struct xe_userptr_vma *uvma,
 	if (ret)
 		goto free_pfns;
 
-	ret = xe_build_sg(vm->xe, &hmm_range, &userptr->sgt, write);
+	ret = xe_alloc_sg(vm->xe, &userptr->sgt, &hmm_range, &vm->userptr.notifier_lock);
 	if (ret)
 		goto free_pfns;
 
+	ret = down_read_interruptible(&vm->userptr.notifier_lock);
+	if (ret)
+		goto free_st;
+
+	if (mmu_interval_read_retry(hmm_range.notifier, hmm_range.notifier_seq)) {
+		ret = -EAGAIN;
+		goto out_unlock;
+	}
+
+	ret = xe_build_sg(vm->xe, &hmm_range, &userptr->sgt,
+			  &vm->userptr.notifier_lock, write);
+	if (ret)
+		goto out_unlock;
+
 	xe_mark_range_accessed(&hmm_range, write);
 	userptr->sg = &userptr->sgt;
+	xe_hmm_userptr_set_mapped(uvma);
 	userptr->notifier_seq = hmm_range.notifier_seq;
+	up_read(&vm->userptr.notifier_lock);
+	kvfree(pfns);
+	return 0;
 
+out_unlock:
+	up_read(&vm->userptr.notifier_lock);
+free_st:
+	sg_free_table(&userptr->sgt);
 free_pfns:
 	kvfree(pfns);
 	return ret;
 }
-
diff --git a/drivers/gpu/drm/xe/xe_hmm.h b/drivers/gpu/drm/xe/xe_hmm.h
index 909dc2bdcd97..0ea98d8e7bbc 100644
--- a/drivers/gpu/drm/xe/xe_hmm.h
+++ b/drivers/gpu/drm/xe/xe_hmm.h
@@ -3,9 +3,16 @@
  * Copyright © 2024 Intel Corporation
  */
 
+#ifndef _XE_HMM_H_
+#define _XE_HMM_H_
+
 #include <linux/types.h>
 
 struct xe_userptr_vma;
 
 int xe_hmm_userptr_populate_range(struct xe_userptr_vma *uvma, bool is_mm_mmap_locked);
+
 void xe_hmm_userptr_free_sg(struct xe_userptr_vma *uvma);
+
+void xe_hmm_userptr_unmap(struct xe_userptr_vma *uvma);
+#endif
diff --git a/drivers/gpu/drm/xe/xe_pt.c b/drivers/gpu/drm/xe/xe_pt.c
index 1ddcc7e79a93..dc24baa84092 100644
--- a/drivers/gpu/drm/xe/xe_pt.c
+++ b/drivers/gpu/drm/xe/xe_pt.c
@@ -28,6 +28,8 @@ struct xe_pt_dir {
 	struct xe_pt pt;
 	/** @children: Array of page-table child nodes */
 	struct xe_ptw *children[XE_PDES];
+	/** @staging: Array of page-table staging nodes */
+	struct xe_ptw *staging[XE_PDES];
 };
 
 #if IS_ENABLED(CONFIG_DRM_XE_DEBUG_VM)
@@ -48,9 +50,10 @@ static struct xe_pt_dir *as_xe_pt_dir(struct xe_pt *pt)
 	return container_of(pt, struct xe_pt_dir, pt);
 }
 
-static struct xe_pt *xe_pt_entry(struct xe_pt_dir *pt_dir, unsigned int index)
+static struct xe_pt *
+xe_pt_entry_staging(struct xe_pt_dir *pt_dir, unsigned int index)
 {
-	return container_of(pt_dir->children[index], struct xe_pt, base);
+	return container_of(pt_dir->staging[index], struct xe_pt, base);
 }
 
 static u64 __xe_pt_empty_pte(struct xe_tile *tile, struct xe_vm *vm,
@@ -125,6 +128,7 @@ struct xe_pt *xe_pt_create(struct xe_vm *vm, struct xe_tile *tile,
 	}
 	pt->bo = bo;
 	pt->base.children = level ? as_xe_pt_dir(pt)->children : NULL;
+	pt->base.staging = level ? as_xe_pt_dir(pt)->staging : NULL;
 
 	if (vm->xef)
 		xe_drm_client_add_bo(vm->xef->client, pt->bo);
@@ -206,8 +210,8 @@ void xe_pt_destroy(struct xe_pt *pt, u32 flags, struct llist_head *deferred)
 		struct xe_pt_dir *pt_dir = as_xe_pt_dir(pt);
 
 		for (i = 0; i < XE_PDES; i++) {
-			if (xe_pt_entry(pt_dir, i))
-				xe_pt_destroy(xe_pt_entry(pt_dir, i), flags,
+			if (xe_pt_entry_staging(pt_dir, i))
+				xe_pt_destroy(xe_pt_entry_staging(pt_dir, i), flags,
 					      deferred);
 		}
 	}
@@ -376,8 +380,10 @@ xe_pt_insert_entry(struct xe_pt_stage_bind_walk *xe_walk, struct xe_pt *parent,
 		/* Continue building a non-connected subtree. */
 		struct iosys_map *map = &parent->bo->vmap;
 
-		if (unlikely(xe_child))
+		if (unlikely(xe_child)) {
 			parent->base.children[offset] = &xe_child->base;
+			parent->base.staging[offset] = &xe_child->base;
+		}
 
 		xe_pt_write(xe_walk->vm->xe, map, offset, pte);
 		parent->num_live++;
@@ -614,6 +620,7 @@ xe_pt_stage_bind(struct xe_tile *tile, struct xe_vma *vma,
 			.ops = &xe_pt_stage_bind_ops,
 			.shifts = xe_normal_pt_shifts,
 			.max_level = XE_PT_HIGHEST_LEVEL,
+			.staging = true,
 		},
 		.vm = xe_vma_vm(vma),
 		.tile = tile,
@@ -873,7 +880,7 @@ static void xe_pt_cancel_bind(struct xe_vma *vma,
 	}
 }
 
-static void xe_pt_commit_locks_assert(struct xe_vma *vma)
+static void xe_pt_commit_prepare_locks_assert(struct xe_vma *vma)
 {
 	struct xe_vm *vm = xe_vma_vm(vma);
 
@@ -885,6 +892,16 @@ static void xe_pt_commit_locks_assert(struct xe_vma *vma)
 	xe_vm_assert_held(vm);
 }
 
+static void xe_pt_commit_locks_assert(struct xe_vma *vma)
+{
+	struct xe_vm *vm = xe_vma_vm(vma);
+
+	xe_pt_commit_prepare_locks_assert(vma);
+
+	if (xe_vma_is_userptr(vma))
+		lockdep_assert_held_read(&vm->userptr.notifier_lock);
+}
+
 static void xe_pt_commit(struct xe_vma *vma,
 			 struct xe_vm_pgtable_update *entries,
 			 u32 num_entries, struct llist_head *deferred)
@@ -895,13 +912,17 @@ static void xe_pt_commit(struct xe_vma *vma,
 
 	for (i = 0; i < num_entries; i++) {
 		struct xe_pt *pt = entries[i].pt;
+		struct xe_pt_dir *pt_dir;
 
 		if (!pt->level)
 			continue;
 
+		pt_dir = as_xe_pt_dir(pt);
 		for (j = 0; j < entries[i].qwords; j++) {
 			struct xe_pt *oldpte = entries[i].pt_entries[j].pt;
+			int j_ = j + entries[i].ofs;
 
+			pt_dir->children[j_] = pt_dir->staging[j_];
 			xe_pt_destroy(oldpte, xe_vma_vm(vma)->flags, deferred);
 		}
 	}
@@ -913,7 +934,7 @@ static void xe_pt_abort_bind(struct xe_vma *vma,
 {
 	int i, j;
 
-	xe_pt_commit_locks_assert(vma);
+	xe_pt_commit_prepare_locks_assert(vma);
 
 	for (i = num_entries - 1; i >= 0; --i) {
 		struct xe_pt *pt = entries[i].pt;
@@ -928,10 +949,10 @@ static void xe_pt_abort_bind(struct xe_vma *vma,
 		pt_dir = as_xe_pt_dir(pt);
 		for (j = 0; j < entries[i].qwords; j++) {
 			u32 j_ = j + entries[i].ofs;
-			struct xe_pt *newpte = xe_pt_entry(pt_dir, j_);
+			struct xe_pt *newpte = xe_pt_entry_staging(pt_dir, j_);
 			struct xe_pt *oldpte = entries[i].pt_entries[j].pt;
 
-			pt_dir->children[j_] = oldpte ? &oldpte->base : 0;
+			pt_dir->staging[j_] = oldpte ? &oldpte->base : 0;
 			xe_pt_destroy(newpte, xe_vma_vm(vma)->flags, NULL);
 		}
 	}
@@ -943,7 +964,7 @@ static void xe_pt_commit_prepare_bind(struct xe_vma *vma,
 {
 	u32 i, j;
 
-	xe_pt_commit_locks_assert(vma);
+	xe_pt_commit_prepare_locks_assert(vma);
 
 	for (i = 0; i < num_entries; i++) {
 		struct xe_pt *pt = entries[i].pt;
@@ -961,10 +982,10 @@ static void xe_pt_commit_prepare_bind(struct xe_vma *vma,
 			struct xe_pt *newpte = entries[i].pt_entries[j].pt;
 			struct xe_pt *oldpte = NULL;
 
-			if (xe_pt_entry(pt_dir, j_))
-				oldpte = xe_pt_entry(pt_dir, j_);
+			if (xe_pt_entry_staging(pt_dir, j_))
+				oldpte = xe_pt_entry_staging(pt_dir, j_);
 
-			pt_dir->children[j_] = &newpte->base;
+			pt_dir->staging[j_] = &newpte->base;
 			entries[i].pt_entries[j].pt = oldpte;
 		}
 	}
@@ -1213,42 +1234,22 @@ static int vma_check_userptr(struct xe_vm *vm, struct xe_vma *vma,
 		return 0;
 
 	uvma = to_userptr_vma(vma);
-	notifier_seq = uvma->userptr.notifier_seq;
+	if (xe_pt_userptr_inject_eagain(uvma))
+		xe_vma_userptr_force_invalidate(uvma);
 
-	if (uvma->userptr.initial_bind && !xe_vm_in_fault_mode(vm))
-		return 0;
+	notifier_seq = uvma->userptr.notifier_seq;
 
 	if (!mmu_interval_read_retry(&uvma->userptr.notifier,
-				     notifier_seq) &&
-	    !xe_pt_userptr_inject_eagain(uvma))
+				     notifier_seq))
 		return 0;
 
-	if (xe_vm_in_fault_mode(vm)) {
+	if (xe_vm_in_fault_mode(vm))
 		return -EAGAIN;
-	} else {
-		spin_lock(&vm->userptr.invalidated_lock);
-		list_move_tail(&uvma->userptr.invalidate_link,
-			       &vm->userptr.invalidated);
-		spin_unlock(&vm->userptr.invalidated_lock);
-
-		if (xe_vm_in_preempt_fence_mode(vm)) {
-			struct dma_resv_iter cursor;
-			struct dma_fence *fence;
-			long err;
-
-			dma_resv_iter_begin(&cursor, xe_vm_resv(vm),
-					    DMA_RESV_USAGE_BOOKKEEP);
-			dma_resv_for_each_fence_unlocked(&cursor, fence)
-				dma_fence_enable_sw_signaling(fence);
-			dma_resv_iter_end(&cursor);
-
-			err = dma_resv_wait_timeout(xe_vm_resv(vm),
-						    DMA_RESV_USAGE_BOOKKEEP,
-						    false, MAX_SCHEDULE_TIMEOUT);
-			XE_WARN_ON(err <= 0);
-		}
-	}
 
+	/*
+	 * Just continue the operation since exec or rebind worker
+	 * will take care of rebinding.
+	 */
 	return 0;
 }
 
@@ -1514,6 +1515,7 @@ static unsigned int xe_pt_stage_unbind(struct xe_tile *tile, struct xe_vma *vma,
 			.ops = &xe_pt_stage_unbind_ops,
 			.shifts = xe_normal_pt_shifts,
 			.max_level = XE_PT_HIGHEST_LEVEL,
+			.staging = true,
 		},
 		.tile = tile,
 		.modified_start = xe_vma_start(vma),
@@ -1555,7 +1557,7 @@ static void xe_pt_abort_unbind(struct xe_vma *vma,
 {
 	int i, j;
 
-	xe_pt_commit_locks_assert(vma);
+	xe_pt_commit_prepare_locks_assert(vma);
 
 	for (i = num_entries - 1; i >= 0; --i) {
 		struct xe_vm_pgtable_update *entry = &entries[i];
@@ -1568,7 +1570,7 @@ static void xe_pt_abort_unbind(struct xe_vma *vma,
 			continue;
 
 		for (j = entry->ofs; j < entry->ofs + entry->qwords; j++)
-			pt_dir->children[j] =
+			pt_dir->staging[j] =
 				entries[i].pt_entries[j - entry->ofs].pt ?
 				&entries[i].pt_entries[j - entry->ofs].pt->base : NULL;
 	}
@@ -1581,7 +1583,7 @@ xe_pt_commit_prepare_unbind(struct xe_vma *vma,
 {
 	int i, j;
 
-	xe_pt_commit_locks_assert(vma);
+	xe_pt_commit_prepare_locks_assert(vma);
 
 	for (i = 0; i < num_entries; ++i) {
 		struct xe_vm_pgtable_update *entry = &entries[i];
@@ -1595,8 +1597,8 @@ xe_pt_commit_prepare_unbind(struct xe_vma *vma,
 		pt_dir = as_xe_pt_dir(pt);
 		for (j = entry->ofs; j < entry->ofs + entry->qwords; j++) {
 			entry->pt_entries[j - entry->ofs].pt =
-				xe_pt_entry(pt_dir, j);
-			pt_dir->children[j] = NULL;
+				xe_pt_entry_staging(pt_dir, j);
+			pt_dir->staging[j] = NULL;
 		}
 	}
 }
diff --git a/drivers/gpu/drm/xe/xe_pt_walk.c b/drivers/gpu/drm/xe/xe_pt_walk.c
index b8b3d2aea492..be602a763ff3 100644
--- a/drivers/gpu/drm/xe/xe_pt_walk.c
+++ b/drivers/gpu/drm/xe/xe_pt_walk.c
@@ -74,7 +74,8 @@ int xe_pt_walk_range(struct xe_ptw *parent, unsigned int level,
 		     u64 addr, u64 end, struct xe_pt_walk *walk)
 {
 	pgoff_t offset = xe_pt_offset(addr, level, walk);
-	struct xe_ptw **entries = parent->children ? parent->children : NULL;
+	struct xe_ptw **entries = walk->staging ? (parent->staging ?: NULL) :
+		(parent->children ?: NULL);
 	const struct xe_pt_walk_ops *ops = walk->ops;
 	enum page_walk_action action;
 	struct xe_ptw *child;
diff --git a/drivers/gpu/drm/xe/xe_pt_walk.h b/drivers/gpu/drm/xe/xe_pt_walk.h
index 5ecc4d2f0f65..5c02c244f7de 100644
--- a/drivers/gpu/drm/xe/xe_pt_walk.h
+++ b/drivers/gpu/drm/xe/xe_pt_walk.h
@@ -11,12 +11,14 @@
 /**
  * struct xe_ptw - base class for driver pagetable subclassing.
  * @children: Pointer to an array of children if any.
+ * @staging: Pointer to an array of staging if any.
  *
  * Drivers could subclass this, and if it's a page-directory, typically
  * embed an array of xe_ptw pointers.
  */
 struct xe_ptw {
 	struct xe_ptw **children;
+	struct xe_ptw **staging;
 };
 
 /**
@@ -41,6 +43,8 @@ struct xe_pt_walk {
 	 * as shared pagetables.
 	 */
 	bool shared_pt_mode;
+	/** @staging: Walk staging PT structure */
+	bool staging;
 };
 
 /**
diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
index 30259eba450b..ec6ec18ab3fa 100644
--- a/drivers/gpu/drm/xe/xe_vm.c
+++ b/drivers/gpu/drm/xe/xe_vm.c
@@ -579,51 +579,26 @@ out_unlock_outer:
 	trace_xe_vm_rebind_worker_exit(vm);
 }
 
-static bool vma_userptr_invalidate(struct mmu_interval_notifier *mni,
-				   const struct mmu_notifier_range *range,
-				   unsigned long cur_seq)
+static void __vma_userptr_invalidate(struct xe_vm *vm, struct xe_userptr_vma *uvma)
 {
-	struct xe_userptr *userptr = container_of(mni, typeof(*userptr), notifier);
-	struct xe_userptr_vma *uvma = container_of(userptr, typeof(*uvma), userptr);
+	struct xe_userptr *userptr = &uvma->userptr;
 	struct xe_vma *vma = &uvma->vma;
-	struct xe_vm *vm = xe_vma_vm(vma);
 	struct dma_resv_iter cursor;
 	struct dma_fence *fence;
 	long err;
 
-	xe_assert(vm->xe, xe_vma_is_userptr(vma));
-	trace_xe_vma_userptr_invalidate(vma);
-
-	if (!mmu_notifier_range_blockable(range))
-		return false;
-
-	vm_dbg(&xe_vma_vm(vma)->xe->drm,
-	       "NOTIFIER: addr=0x%016llx, range=0x%016llx",
-		xe_vma_start(vma), xe_vma_size(vma));
-
-	down_write(&vm->userptr.notifier_lock);
-	mmu_interval_set_seq(mni, cur_seq);
-
-	/* No need to stop gpu access if the userptr is not yet bound. */
-	if (!userptr->initial_bind) {
-		up_write(&vm->userptr.notifier_lock);
-		return true;
-	}
-
 	/*
 	 * Tell exec and rebind worker they need to repin and rebind this
 	 * userptr.
 	 */
 	if (!xe_vm_in_fault_mode(vm) &&
-	    !(vma->gpuva.flags & XE_VMA_DESTROYED) && vma->tile_present) {
+	    !(vma->gpuva.flags & XE_VMA_DESTROYED)) {
 		spin_lock(&vm->userptr.invalidated_lock);
 		list_move_tail(&userptr->invalidate_link,
 			       &vm->userptr.invalidated);
 		spin_unlock(&vm->userptr.invalidated_lock);
 	}
 
-	up_write(&vm->userptr.notifier_lock);
-
 	/*
 	 * Preempt fences turn into schedule disables, pipeline these.
 	 * Note that even in fault mode, we need to wait for binds and
@@ -641,11 +616,37 @@ static bool vma_userptr_invalidate(struct mmu_interval_notifier *mni,
 				    false, MAX_SCHEDULE_TIMEOUT);
 	XE_WARN_ON(err <= 0);
 
-	if (xe_vm_in_fault_mode(vm)) {
+	if (xe_vm_in_fault_mode(vm) && userptr->initial_bind) {
 		err = xe_vm_invalidate_vma(vma);
 		XE_WARN_ON(err);
 	}
 
+	xe_hmm_userptr_unmap(uvma);
+}
+
+static bool vma_userptr_invalidate(struct mmu_interval_notifier *mni,
+				   const struct mmu_notifier_range *range,
+				   unsigned long cur_seq)
+{
+	struct xe_userptr_vma *uvma = container_of(mni, typeof(*uvma), userptr.notifier);
+	struct xe_vma *vma = &uvma->vma;
+	struct xe_vm *vm = xe_vma_vm(vma);
+
+	xe_assert(vm->xe, xe_vma_is_userptr(vma));
+	trace_xe_vma_userptr_invalidate(vma);
+
+	if (!mmu_notifier_range_blockable(range))
+		return false;
+
+	vm_dbg(&xe_vma_vm(vma)->xe->drm,
+	       "NOTIFIER: addr=0x%016llx, range=0x%016llx",
+		xe_vma_start(vma), xe_vma_size(vma));
+
+	down_write(&vm->userptr.notifier_lock);
+	mmu_interval_set_seq(mni, cur_seq);
+
+	__vma_userptr_invalidate(vm, uvma);
+	up_write(&vm->userptr.notifier_lock);
 	trace_xe_vma_userptr_invalidate_complete(vma);
 
 	return true;
@@ -655,6 +656,34 @@ static const struct mmu_interval_notifier_ops vma_userptr_notifier_ops = {
 	.invalidate = vma_userptr_invalidate,
 };
 
+#if IS_ENABLED(CONFIG_DRM_XE_USERPTR_INVAL_INJECT)
+/**
+ * xe_vma_userptr_force_invalidate() - force invalidate a userptr
+ * @uvma: The userptr vma to invalidate
+ *
+ * Perform a forced userptr invalidation for testing purposes.
+ */
+void xe_vma_userptr_force_invalidate(struct xe_userptr_vma *uvma)
+{
+	struct xe_vm *vm = xe_vma_vm(&uvma->vma);
+
+	/* Protect against concurrent userptr pinning */
+	lockdep_assert_held(&vm->lock);
+	/* Protect against concurrent notifiers */
+	lockdep_assert_held(&vm->userptr.notifier_lock);
+	/*
+	 * Protect against concurrent instances of this function and
+	 * the critical exec sections
+	 */
+	xe_vm_assert_held(vm);
+
+	if (!mmu_interval_read_retry(&uvma->userptr.notifier,
+				     uvma->userptr.notifier_seq))
+		uvma->userptr.notifier_seq -= 2;
+	__vma_userptr_invalidate(vm, uvma);
+}
+#endif
+
 int xe_vm_userptr_pin(struct xe_vm *vm)
 {
 	struct xe_userptr_vma *uvma, *next;
@@ -1012,6 +1041,7 @@ static struct xe_vma *xe_vma_create(struct xe_vm *vm,
 			INIT_LIST_HEAD(&userptr->invalidate_link);
 			INIT_LIST_HEAD(&userptr->repin_link);
 			vma->gpuva.gem.offset = bo_offset_or_userptr;
+			mutex_init(&userptr->unmap_mutex);
 
 			err = mmu_interval_notifier_insert(&userptr->notifier,
 							   current->mm,
@@ -1053,6 +1083,7 @@ static void xe_vma_destroy_late(struct xe_vma *vma)
 		 * them anymore
 		 */
 		mmu_interval_notifier_remove(&userptr->notifier);
+		mutex_destroy(&userptr->unmap_mutex);
 		xe_vm_put(vm);
 	} else if (xe_vma_is_null(vma)) {
 		xe_vm_put(vm);
@@ -2286,8 +2317,17 @@ static int vm_bind_ioctl_ops_parse(struct xe_vm *vm, struct drm_gpuva_ops *ops,
 			break;
 		}
 		case DRM_GPUVA_OP_UNMAP:
+			xe_vma_ops_incr_pt_update_ops(vops, op->tile_mask);
+			break;
 		case DRM_GPUVA_OP_PREFETCH:
-			/* FIXME: Need to skip some prefetch ops */
+			vma = gpuva_to_vma(op->base.prefetch.va);
+
+			if (xe_vma_is_userptr(vma)) {
+				err = xe_vma_userptr_pin_pages(to_userptr_vma(vma));
+				if (err)
+					return err;
+			}
+
 			xe_vma_ops_incr_pt_update_ops(vops, op->tile_mask);
 			break;
 		default:
diff --git a/drivers/gpu/drm/xe/xe_vm.h b/drivers/gpu/drm/xe/xe_vm.h
index 23adb7442881..b882bfb31bd0 100644
--- a/drivers/gpu/drm/xe/xe_vm.h
+++ b/drivers/gpu/drm/xe/xe_vm.h
@@ -274,9 +274,17 @@ static inline void vm_dbg(const struct drm_device *dev,
 			  const char *format, ...)
 { /* noop */ }
 #endif
-#endif
 
 struct xe_vm_snapshot *xe_vm_snapshot_capture(struct xe_vm *vm);
 void xe_vm_snapshot_capture_delayed(struct xe_vm_snapshot *snap);
 void xe_vm_snapshot_print(struct xe_vm_snapshot *snap, struct drm_printer *p);
 void xe_vm_snapshot_free(struct xe_vm_snapshot *snap);
+
+#if IS_ENABLED(CONFIG_DRM_XE_USERPTR_INVAL_INJECT)
+void xe_vma_userptr_force_invalidate(struct xe_userptr_vma *uvma);
+#else
+static inline void xe_vma_userptr_force_invalidate(struct xe_userptr_vma *uvma)
+{
+}
+#endif
+#endif
diff --git a/drivers/gpu/drm/xe/xe_vm_types.h b/drivers/gpu/drm/xe/xe_vm_types.h
index 7f9a303e51d8..a4b4091cfd0d 100644
--- a/drivers/gpu/drm/xe/xe_vm_types.h
+++ b/drivers/gpu/drm/xe/xe_vm_types.h
@@ -59,12 +59,16 @@ struct xe_userptr {
 	struct sg_table *sg;
 	/** @notifier_seq: notifier sequence number */
 	unsigned long notifier_seq;
+	/** @unmap_mutex: Mutex protecting dma-unmapping */
+	struct mutex unmap_mutex;
 	/**
 	 * @initial_bind: user pointer has been bound at least once.
 	 * write: vm->userptr.notifier_lock in read mode and vm->resv held.
 	 * read: vm->userptr.notifier_lock in write mode or vm->resv held.
 	 */
 	bool initial_bind;
+	/** @mapped: Whether the @sgt sg-table is dma-mapped. Protected by @unmap_mutex. */
+	bool mapped;
 #if IS_ENABLED(CONFIG_DRM_XE_USERPTR_INVAL_INJECT)
 	u32 divisor;
 #endif
@@ -227,8 +231,8 @@ struct xe_vm {
 		 * up for revalidation. Protected from access with the
 		 * @invalidated_lock. Removing items from the list
 		 * additionally requires @lock in write mode, and adding
-		 * items to the list requires the @userptr.notifer_lock in
-		 * write mode.
+		 * items to the list requires either the @userptr.notifer_lock in
+		 * write mode, OR @lock in write mode.
 		 */
 		struct list_head invalidated;
 	} userptr;