drm/xe/exec: Switch hw engine group execution mode upon job submission

If the job about to be submitted is a dma-fence job, update the current
execution mode of the hw engine group. This triggers an immediate suspend
of the exec queues running faulting long-running jobs.

If the job about to be submitted is a long-running job, kick a new worker
used to resume the exec queues running faulting long-running jobs once
the dma-fence jobs have completed.

v2: Kick the resume worker from exec IOCTL, switch to unordered workqueue,
    destroy it after use (Matt Brost)

v3: Do not resume if no exec queue was suspended (Matt Brost)

v4: Squash commits (Matt Brost)

v5: Do not kick the worker when xe_vm_in_preempt_fence_mode (Matt Brost)

Signed-off-by: Francois Dugast <francois.dugast@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240809155156.1955925-10-francois.dugast@intel.com

diff --git a/drivers/gpu/drm/xe/xe_exec.c b/drivers/gpu/drm/xe/xe_exec.c
index f36980a..484acfb 100644 (file)
--- a/drivers/gpu/drm/xe/xe_exec.c
+++ b/drivers/gpu/drm/xe/xe_exec.c
@@ -14,6 +14,7 @@
 #include "xe_bo.h"
 #include "xe_device.h"
 #include "xe_exec_queue.h"
+#include "xe_hw_engine_group.h"
 #include "xe_macros.h"
 #include "xe_ring_ops_types.h"
 #include "xe_sched_job.h"
@@ -124,6 +125,8 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
        bool write_locked, skip_retry = false;
        ktime_t end = 0;
        int err = 0;
+       struct xe_hw_engine_group *group;
+       enum xe_hw_engine_group_execution_mode mode, previous_mode;
 
        if (XE_IOCTL_DBG(xe, args->extensions) ||
            XE_IOCTL_DBG(xe, args->pad[0] || args->pad[1] || args->pad[2]) ||
@@ -182,6 +185,15 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
                }
        }
 
+       group = q->hwe->hw_engine_group;
+       mode = xe_hw_engine_group_find_exec_mode(q);
+
+       if (mode == EXEC_MODE_DMA_FENCE) {
+               err = xe_hw_engine_group_get_mode(group, mode, &previous_mode);
+               if (err)
+                       goto err_syncs;
+       }
+
 retry:
        if (!xe_vm_in_lr_mode(vm) && xe_vm_userptr_check_repin(vm)) {
                err = down_write_killable(&vm->lock);
@@ -199,7 +211,7 @@ retry:
                downgrade_write(&vm->lock);
                write_locked = false;
                if (err)
-                       goto err_unlock_list;
+                       goto err_hw_exec_mode;
        }
 
        if (!args->num_batch_buffer) {
@@ -312,6 +324,9 @@ retry:
                spin_unlock(&xe->ttm.lru_lock);
        }
 
+       if (mode == EXEC_MODE_LR)
+               xe_hw_engine_group_resume_faulting_lr_jobs(group);
+
 err_repin:
        if (!xe_vm_in_lr_mode(vm))
                up_read(&vm->userptr.notifier_lock);
@@ -324,6 +339,9 @@ err_unlock_list:
        up_read(&vm->lock);
        if (err == -EAGAIN && !skip_retry)
                goto retry;
+err_hw_exec_mode:
+       if (mode == EXEC_MODE_DMA_FENCE)
+               xe_hw_engine_group_put(group);
 err_syncs:
        while (num_syncs--)
                xe_sync_entry_cleanup(&syncs[num_syncs]);

diff --git a/drivers/gpu/drm/xe/xe_hw_engine_group.c b/drivers/gpu/drm/xe/xe_hw_engine_group.c
index e6c2351..8275052 100644 (file)
--- a/drivers/gpu/drm/xe/xe_hw_engine_group.c
+++ b/drivers/gpu/drm/xe/xe_hw_engine_group.c
@@ -17,9 +17,36 @@ hw_engine_group_free(struct drm_device *drm, void *arg)
 {
        struct xe_hw_engine_group *group = arg;
 
+       destroy_workqueue(group->resume_wq);
        kfree(group);
 }
 
+static void
+hw_engine_group_resume_lr_jobs_func(struct work_struct *w)
+{
+       struct xe_exec_queue *q;
+       struct xe_hw_engine_group *group = container_of(w, struct xe_hw_engine_group, resume_work);
+       int err;
+       enum xe_hw_engine_group_execution_mode previous_mode;
+
+       err = xe_hw_engine_group_get_mode(group, EXEC_MODE_LR, &previous_mode);
+       if (err)
+               return;
+
+       if (previous_mode == EXEC_MODE_LR)
+               goto put;
+
+       list_for_each_entry(q, &group->exec_queue_list, hw_engine_group_link) {
+               if (!xe_vm_in_fault_mode(q->vm))
+                       continue;
+
+               q->ops->resume(q);
+       }
+
+put:
+       xe_hw_engine_group_put(group);
+}
+
 static struct xe_hw_engine_group *
 hw_engine_group_alloc(struct xe_device *xe)
 {
@@ -30,7 +57,12 @@ hw_engine_group_alloc(struct xe_device *xe)
        if (!group)
                return ERR_PTR(-ENOMEM);
 
+       group->resume_wq = alloc_workqueue("xe-resume-lr-jobs-wq", 0, 0);
+       if (!group->resume_wq)
+               return ERR_PTR(-ENOMEM);
+
        init_rwsem(&group->mode_sem);
+       INIT_WORK(&group->resume_work, hw_engine_group_resume_lr_jobs_func);
        INIT_LIST_HEAD(&group->exec_queue_list);
 
        err = drmm_add_action_or_reset(&xe->drm, hw_engine_group_free, group);
@@ -134,7 +166,7 @@ int xe_hw_engine_group_add_exec_queue(struct xe_hw_engine_group *group, struct x
                if (err)
                        goto err_suspend;
 
-               queue_work(group->resume_wq, &group->resume_work);
+               xe_hw_engine_group_resume_faulting_lr_jobs(group);
        }
 
        list_add(&q->hw_engine_group_link, &group->exec_queue_list);
@@ -167,6 +199,16 @@ void xe_hw_engine_group_del_exec_queue(struct xe_hw_engine_group *group, struct
        up_write(&group->mode_sem);
 }
 
+/**
+ * xe_hw_engine_group_resume_faulting_lr_jobs() - Asynchronously resume the hw engine group's
+ * faulting LR jobs
+ * @group: The hw engine group
+ */
+void xe_hw_engine_group_resume_faulting_lr_jobs(struct xe_hw_engine_group *group)
+{
+       queue_work(group->resume_wq, &group->resume_work);
+}
+
 /**
  * xe_hw_engine_group_suspend_faulting_lr_jobs() - Suspend the faulting LR jobs of this group
  * @group: The hw engine group
@@ -177,6 +219,7 @@ static int xe_hw_engine_group_suspend_faulting_lr_jobs(struct xe_hw_engine_group
 {
        int err;
        struct xe_exec_queue *q;
+       bool need_resume = false;
 
        lockdep_assert_held_write(&group->mode_sem);
 
@@ -184,6 +227,7 @@ static int xe_hw_engine_group_suspend_faulting_lr_jobs(struct xe_hw_engine_group
                if (!xe_vm_in_fault_mode(q->vm))
                        continue;
 
+               need_resume = true;
                q->ops->suspend(q);
        }
 
@@ -196,6 +240,9 @@ static int xe_hw_engine_group_suspend_faulting_lr_jobs(struct xe_hw_engine_group
                        goto err_suspend;
        }
 
+       if (need_resume)
+               xe_hw_engine_group_resume_faulting_lr_jobs(group);
+
        return 0;
 
 err_suspend:
@@ -310,3 +357,16 @@ __releases(&group->mode_sem)
 {
        up_read(&group->mode_sem);
 }
+
+/**
+ * xe_hw_engine_group_find_exec_mode() - Find the execution mode for this exec queue
+ * @q: The exec_queue
+ */
+enum xe_hw_engine_group_execution_mode
+xe_hw_engine_group_find_exec_mode(struct xe_exec_queue *q)
+{
+       if (xe_vm_in_fault_mode(q->vm))
+               return EXEC_MODE_LR;
+       else
+               return EXEC_MODE_DMA_FENCE;
+}

diff --git a/drivers/gpu/drm/xe/xe_hw_engine_group.h b/drivers/gpu/drm/xe/xe_hw_engine_group.h
index e0deb7c..797ee81 100644 (file)
--- a/drivers/gpu/drm/xe/xe_hw_engine_group.h
+++ b/drivers/gpu/drm/xe/xe_hw_engine_group.h
@@ -22,4 +22,8 @@ int xe_hw_engine_group_get_mode(struct xe_hw_engine_group *group,
                                enum xe_hw_engine_group_execution_mode *previous_mode);
 void xe_hw_engine_group_put(struct xe_hw_engine_group *group);
 
+enum xe_hw_engine_group_execution_mode
+xe_hw_engine_group_find_exec_mode(struct xe_exec_queue *q);
+void xe_hw_engine_group_resume_faulting_lr_jobs(struct xe_hw_engine_group *group);
+
 #endif