drm/xe: Block exec and rebind worker while evicting for suspend / hibernate

When the xe pm_notifier evicts for suspend / hibernate, there might be
racing tasks trying to re-validate again. This can lead to suspend taking
excessive time or get stuck in a live-lock. This behaviour becomes
much worse with the fix that actually makes re-validation bring back
bos to VRAM rather than letting them remain in TT.

Prevent that by having exec and the rebind worker waiting for a completion
that is set to block by the pm_notifier before suspend and is signaled
by the pm_notifier after resume / wakeup.

It's probably still possible to craft malicious applications that block
suspending. More work is pending to fix that.

v3:
- Avoid wait_for_completion() in the kernel worker since it could
  potentially cause work item flushes from freezable processes to
  wait forever. Instead terminate the rebind workers if needed and
  re-launch at resume. (Matt Auld)
v4:
- Fix some bad naming and leftover debug printouts.
- Fix kerneldoc.
- Use drmm_mutex_init() for the xe->rebind_resume_lock (Matt Auld).
- Rework the interface of xe_vm_rebind_resume_worker (Matt Auld).

Link: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/4288
Fixes: c6a4d46 ("drm/xe: evict user memory in PM notifier")
Cc: Matthew Auld <matthew.auld@intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: <stable@vger.kernel.org> # v6.16+
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://lore.kernel.org/r/20250904160715.2613-4-thomas.hellstrom@linux.intel.com
(cherry picked from commit 5993345)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

Author: Thomas Hellström

drivers/gpu/drm/xe/xe_device_types.h

@@ -553,6 +553,12 @@ struct xe_device {
 
 	/** @pm_notifier: Our PM notifier to perform actions in response to various PM events. */
 	struct notifier_block pm_notifier;
+	/** @pm_block: Completion to block validating tasks on suspend / hibernate prepare */
+	struct completion pm_block;
+	/** @rebind_resume_list: List of wq items to kick on resume. */
+	struct list_head rebind_resume_list;
+	/** @rebind_resume_lock: Lock to protect the rebind_resume_list */
+	struct mutex rebind_resume_lock;
 
 	/** @pmt: Support the PMT driver callback interface */
 	struct {

drivers/gpu/drm/xe/xe_exec.c

@@ -237,6 +237,15 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
 		goto err_unlock_list;
 	}
 
+	/*
+	 * It's OK to block interruptible here with the vm lock held, since
+	 * on task freezing during suspend / hibernate, the call will
+	 * return -ERESTARTSYS and the IOCTL will be rerun.
+	 */
+	err = wait_for_completion_interruptible(&xe->pm_block);
+	if (err)
+		goto err_unlock_list;
+
 	vm_exec.vm = &vm->gpuvm;
 	vm_exec.flags = DRM_EXEC_INTERRUPTIBLE_WAIT;
 	if (xe_vm_in_lr_mode(vm)) {

drivers/gpu/drm/xe/xe_pm.c

@@ -24,6 +24,7 @@
 #include "xe_pcode.h"
 #include "xe_pxp.h"
 #include "xe_trace.h"
+#include "xe_vm.h"
 #include "xe_wa.h"
 
 /**
@@ -290,6 +291,19 @@ static u32 vram_threshold_value(struct xe_device *xe)
 	return DEFAULT_VRAM_THRESHOLD;
 }
 
+static void xe_pm_wake_rebind_workers(struct xe_device *xe)
+{
+	struct xe_vm *vm, *next;
+
+	mutex_lock(&xe->rebind_resume_lock);
+	list_for_each_entry_safe(vm, next, &xe->rebind_resume_list,
+				 preempt.pm_activate_link) {
+		list_del_init(&vm->preempt.pm_activate_link);
+		xe_vm_resume_rebind_worker(vm);
+	}
+	mutex_unlock(&xe->rebind_resume_lock);
+}
+
 static int xe_pm_notifier_callback(struct notifier_block *nb,
 				   unsigned long action, void *data)
 {
@@ -299,6 +313,7 @@ static int xe_pm_notifier_callback(struct notifier_block *nb,
 	switch (action) {
 	case PM_HIBERNATION_PREPARE:
 	case PM_SUSPEND_PREPARE:
+		reinit_completion(&xe->pm_block);
 		xe_pm_runtime_get(xe);
 		err = xe_bo_evict_all_user(xe);
 		if (err)
@@ -315,6 +330,8 @@ static int xe_pm_notifier_callback(struct notifier_block *nb,
 		break;
 	case PM_POST_HIBERNATION:
 	case PM_POST_SUSPEND:
+		complete_all(&xe->pm_block);
+		xe_pm_wake_rebind_workers(xe);
 		xe_bo_notifier_unprepare_all_pinned(xe);
 		xe_pm_runtime_put(xe);
 		break;
@@ -341,6 +358,14 @@ int xe_pm_init(struct xe_device *xe)
 	if (err)
 		return err;
 
+	err = drmm_mutex_init(&xe->drm, &xe->rebind_resume_lock);
+	if (err)
+		goto err_unregister;
+
+	init_completion(&xe->pm_block);
+	complete_all(&xe->pm_block);
+	INIT_LIST_HEAD(&xe->rebind_resume_list);
+
 	/* For now suspend/resume is only allowed with GuC */
 	if (!xe_device_uc_enabled(xe))
 		return 0;

drivers/gpu/drm/xe/xe_vm.c

@@ -393,6 +393,9 @@ static int xe_gpuvm_validate(struct drm_gpuvm_bo *vm_bo, struct drm_exec *exec)
 		list_move_tail(&gpuva_to_vma(gpuva)->combined_links.rebind,
 			       &vm->rebind_list);
 
+	if (!try_wait_for_completion(&vm->xe->pm_block))
+		return -EAGAIN;
+
 	ret = xe_bo_validate(gem_to_xe_bo(vm_bo->obj), vm, false);
 	if (ret)
 		return ret;
@@ -479,6 +482,33 @@ static int xe_preempt_work_begin(struct drm_exec *exec, struct xe_vm *vm,
 	return xe_vm_validate_rebind(vm, exec, vm->preempt.num_exec_queues);
 }
 
+static bool vm_suspend_rebind_worker(struct xe_vm *vm)
+{
+	struct xe_device *xe = vm->xe;
+	bool ret = false;
+
+	mutex_lock(&xe->rebind_resume_lock);
+	if (!try_wait_for_completion(&vm->xe->pm_block)) {
+		ret = true;
+		list_move_tail(&vm->preempt.pm_activate_link, &xe->rebind_resume_list);
+	}
+	mutex_unlock(&xe->rebind_resume_lock);
+
+	return ret;
+}
+
+/**
+ * xe_vm_resume_rebind_worker() - Resume the rebind worker.
+ * @vm: The vm whose preempt worker to resume.
+ *
+ * Resume a preempt worker that was previously suspended by
+ * vm_suspend_rebind_worker().
+ */
+void xe_vm_resume_rebind_worker(struct xe_vm *vm)
+{
+	queue_work(vm->xe->ordered_wq, &vm->preempt.rebind_work);
+}
+
 static void preempt_rebind_work_func(struct work_struct *w)
 {
 	struct xe_vm *vm = container_of(w, struct xe_vm, preempt.rebind_work);
@@ -502,6 +532,11 @@ static void preempt_rebind_work_func(struct work_struct *w)
 	}
 
 retry:
+	if (!try_wait_for_completion(&vm->xe->pm_block) && vm_suspend_rebind_worker(vm)) {
+		up_write(&vm->lock);
+		return;
+	}
+
 	if (xe_vm_userptr_check_repin(vm)) {
 		err = xe_vm_userptr_pin(vm);
 		if (err)
@@ -1714,6 +1749,7 @@ struct xe_vm *xe_vm_create(struct xe_device *xe, u32 flags, struct xe_file *xef)
 	if (flags & XE_VM_FLAG_LR_MODE) {
 		INIT_WORK(&vm->preempt.rebind_work, preempt_rebind_work_func);
 		xe_pm_runtime_get_noresume(xe);
+		INIT_LIST_HEAD(&vm->preempt.pm_activate_link);
 	}
 
 	if (flags & XE_VM_FLAG_FAULT_MODE) {
@@ -1895,8 +1931,12 @@ void xe_vm_close_and_put(struct xe_vm *vm)
 	xe_assert(xe, !vm->preempt.num_exec_queues);
 
 	xe_vm_close(vm);
-	if (xe_vm_in_preempt_fence_mode(vm))
+	if (xe_vm_in_preempt_fence_mode(vm)) {
+		mutex_lock(&xe->rebind_resume_lock);
+		list_del_init(&vm->preempt.pm_activate_link);
+		mutex_unlock(&xe->rebind_resume_lock);
 		flush_work(&vm->preempt.rebind_work);
+	}
 	if (xe_vm_in_fault_mode(vm))
 		xe_svm_close(vm);
 

drivers/gpu/drm/xe/xe_vm.h

@@ -273,6 +273,8 @@ struct dma_fence *xe_vm_bind_kernel_bo(struct xe_vm *vm, struct xe_bo *bo,
 				       struct xe_exec_queue *q, u64 addr,
 				       enum xe_cache_level cache_lvl);
 
+void xe_vm_resume_rebind_worker(struct xe_vm *vm);
+
 /**
  * xe_vm_resv() - Return's the vm's reservation object
  * @vm: The vm

drivers/gpu/drm/xe/xe_vm_types.h

@@ -293,6 +293,11 @@ struct xe_vm {
 		 * BOs
 		 */
 		struct work_struct rebind_work;
+		/**
+		 * @preempt.pm_activate_link: Link to list of rebind workers to be
+		 * kicked on resume.
+		 */
+		struct list_head pm_activate_link;
 	} preempt;
 
 	/** @um: unified memory state */