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: c6a4d46ec1d7 ("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

diff --git a/drivers/gpu/drm/xe/xe_device_types.h b/drivers/gpu/drm/xe/xe_device_types.h
index 646d04aec68cca8e84d5311e51801779fdde1a26..9e30dc7d6e5822b66d3e975715a02176a68d0e42 100644 (file)
--- a/drivers/gpu/drm/xe/xe_device_types.h
+++ b/drivers/gpu/drm/xe/xe_device_types.h
@@ -507,6 +507,12 @@ 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 {

diff --git a/drivers/gpu/drm/xe/xe_exec.c b/drivers/gpu/drm/xe/xe_exec.c
index 25a59b6934f6d1bebc6bd85d7f2a06780ca19c29..1dc0c54cac5f696c68e9b1acd0044913a071a849 100644 (file)
--- a/drivers/gpu/drm/xe/xe_exec.c
+++ b/drivers/gpu/drm/xe/xe_exec.c
@@ -238,6 +238,15 @@ retry:
                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)) {

diff --git a/drivers/gpu/drm/xe/xe_pm.c b/drivers/gpu/drm/xe/xe_pm.c
index ecfa7d77a2a3a59a34c650c64718668f2dc63865..6eea4190bbd2ca615f085711e90b3d3b28c44d7d 100644 (file)
--- a/drivers/gpu/drm/xe/xe_pm.c
+++ b/drivers/gpu/drm/xe/xe_pm.c
@@ -25,6 +25,7 @@
 #include "xe_pxp.h"
 #include "xe_sriov_vf_ccs.h"
 #include "xe_trace.h"
+#include "xe_vm.h"
 #include "xe_wa.h"
 
 /**
@@ -301,6 +302,19 @@ 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)
 {
@@ -310,6 +324,7 @@ 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)
@@ -326,6 +341,8 @@ 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;
@@ -352,6 +369,14 @@ 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;

diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
index 869f1db87d74cface469857bbb94908cdb9722da..346975bc10659a2814063d0ca4a66c32d0ecda1e 100644 (file)
--- a/drivers/gpu/drm/xe/xe_vm.c
+++ b/drivers/gpu/drm/xe/xe_vm.c
@@ -343,6 +343,9 @@ 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;
@@ -429,6 +432,33 @@ 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);
@@ -452,6 +482,11 @@ 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)
@@ -1470,6 +1505,7 @@ 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);
        }
 
        err = xe_svm_init(vm);
@@ -1649,8 +1685,12 @@ 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);
 

diff --git a/drivers/gpu/drm/xe/xe_vm.h b/drivers/gpu/drm/xe/xe_vm.h
index bec9bf89d937cc77e9045fa3cbc8ee958f4bdc63..f29a4baee05d552dfa30730c6b7ac39392bc97b0 100644 (file)
--- a/drivers/gpu/drm/xe/xe_vm.h
+++ b/drivers/gpu/drm/xe/xe_vm.h
@@ -271,6 +271,8 @@ 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

diff --git a/drivers/gpu/drm/xe/xe_vm_types.h b/drivers/gpu/drm/xe/xe_vm_types.h
index f6dd964be42fc40474417b00effdd82b59a0ed25..8e188b83ef303c624d0eab767708e2b282b95735 100644 (file)
--- a/drivers/gpu/drm/xe/xe_vm_types.h
+++ b/drivers/gpu/drm/xe/xe_vm_types.h
@@ -282,6 +282,11 @@ 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 */