Re: [PATCH 32/33] drm/i915/guc: Implement GuC priority management

John Harrison <john.c.harrison@xxxxxxxxx> · Tue, 7 Jun 2022 15:40:14 -0700

On 6/7/2022 06:58, Tvrtko Ursulin wrote:

A blast from the past..

On 27/07/2021 01:23, Matthew Brost wrote:
Implement a simple static mapping algorithm of the i915 priority levels
(int, -1k to 1k exposed to user) to the 4 GuC levels. Mapping is as
follows:

i915 level < 0          -> GuC low level     (3)
i915 level == 0         -> GuC normal level  (2)
i915 level < INT_MAX    -> GuC high level    (1)
i915 level == INT_MAX   -> GuC highest level (0)

We believe this mapping should cover the UMD use cases (3 distinct user
levels + 1 kernel level).

I've noticed Vulkan has VK_EXT_global_priority and AFAICT Mesa 
implementation maps those priorities to i915 uapi:

    VK_QUEUE_GLOBAL_PRIORITY_LOW_KHR = -512
    VK_QUEUE_GLOBAL_PRIORITY_MEDIUM_KHR = 0
    VK_QUEUE_GLOBAL_PRIORITY_HIGH_KHR = +512
    VK_QUEUE_GLOBAL_PRIORITY_REALTIME_KHR = +1023

Wasn't aware of the REALTIME option. Does Mesa have that as well? Pretty 
sure it only had low/medium/high last I time looked but that was quite a 
while ago.

GuC as per this patch remaps to:

    VK_QUEUE_GLOBAL_PRIORITY_LOW_KHR = GUC_CLIENT_PRIORITY_NORMAL
    VK_QUEUE_GLOBAL_PRIORITY_MEDIUM_KHR = GUC_CLIENT_PRIORITY_KMD_NORMAL
    VK_QUEUE_GLOBAL_PRIORITY_HIGH_KHR = GUC_CLIENT_PRIORITY_HIGH
    VK_QUEUE_GLOBAL_PRIORITY_REALTIME_KHR = GUC_CLIENT_PRIORITY_HIGH

Last two being the same, so we lose one level. I don't think it is a 
problem according to VK spec, which is very loose, so considered fine 
for us and just unfortunate?
My recollection is that both Vulkan and Mesa describe the priority as 
merely a hint. The implementation is free to obey or ignore as it 
desires (or as it is capable). So while it would be good to support what 
we can, we don't actually fail any compatibility rules if we don't. And 
I think the idea was to reserve the highest priority for KMD only tasks 
such as the heartbeat.


Also, given drm/sched has three priorities - what is the plan to 
connect GuC's four into those three? I ask because I noticed the 
I guess the plan there is a straight 1:1 for low/medium/high. Matthew 
would be the person with more idea on the drm/scheduler, though.

fixed three levels does not really fit with some other drivers like 
msm for instance. There are up to four hardware priorities there and 
currently an odd mapping scheme is implemented. So I am getting a 
feeling drm/sched should rather be extended to allow more flexibility 
for drivers.

And finally a side question on this particular patch - shouldn't GuC 
map I915_PRIORITY_HEARTBEAT to GUC_CLIENT_PRIORITY_KMD_HIGH (highest)? 
I'd say yes - we certainly want heartbeats to preempt maximum user 
priority.
Hmm. Not sure what the point of the HEARTBEAT define is. The heartbeat 
pulse itself starts at minimum priority and then bumps upwards on each 
time period until it hits the max. So eventually, it will run at higher 
than maximum user priority. The HEARTBEAT prio itself seems to be the 
mid-point. So we don't want that to be super-high. It would even maybe 
make more sense for it to match NORMAL.

John.




Regards,

Tvrtko

In addition to static mapping, a simple counter system is attached to
each context tracking the number of requests inflight on the context at
each level. This is needed as the GuC levels are per context while in
the i915 levels are per request.

v2:
  (Daniele)
   - Add BUILD_BUG_ON to enforce ordering of priority levels
   - Add missing lockdep to guc_prio_fini
   - Check for return before setting context registered flag
   - Map DISPLAY priority or higher to highest guc prio
   - Update comment for guc_prio

Signed-off-by: Matthew Brost <matthew.brost@xxxxxxxxx>
Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@xxxxxxxxx>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@xxxxxxxxx>
---
  drivers/gpu/drm/i915/gt/intel_breadcrumbs.c   |   3 +
  drivers/gpu/drm/i915/gt/intel_context_types.h |   9 +-
  drivers/gpu/drm/i915/gt/intel_engine_user.c   |   4 +
  .../gpu/drm/i915/gt/uc/intel_guc_submission.c | 212 +++++++++++++++++-
  drivers/gpu/drm/i915/i915_request.c           |   5 +
  drivers/gpu/drm/i915/i915_request.h           |   9 +
  drivers/gpu/drm/i915/i915_scheduler.c         |   7 +
  drivers/gpu/drm/i915/i915_scheduler_types.h   |  12 +
  drivers/gpu/drm/i915/i915_trace.h             |  17 +-
  include/uapi/drm/i915_drm.h                   |   9 +
  10 files changed, 282 insertions(+), 5 deletions(-)

diff --git a/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c 
b/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c
index 2007dc6f6b99..209cf265bf74 100644
--- a/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c
+++ b/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c
@@ -245,6 +245,9 @@ static void signal_irq_work(struct irq_work *work)
              llist_entry(signal, typeof(*rq), signal_node);
          struct list_head cb_list;
  +        if (rq->engine->sched_engine->retire_inflight_request_prio)
+ rq->engine->sched_engine->retire_inflight_request_prio(rq);
+
          spin_lock(&rq->lock);
          list_replace(&rq->fence.cb_list, &cb_list);
          __dma_fence_signal__timestamp(&rq->fence, timestamp);
diff --git a/drivers/gpu/drm/i915/gt/intel_context_types.h 
b/drivers/gpu/drm/i915/gt/intel_context_types.h
index a5bc876face7..e54351a170e2 100644
--- a/drivers/gpu/drm/i915/gt/intel_context_types.h
+++ b/drivers/gpu/drm/i915/gt/intel_context_types.h
@@ -18,8 +18,9 @@
  #include "intel_engine_types.h"
  #include "intel_sseu.h"
  -#define CONTEXT_REDZONE POISON_INUSE
+#include "uc/intel_guc_fwif.h"
  +#define CONTEXT_REDZONE POISON_INUSE
  DECLARE_EWMA(runtime, 3, 8);
    struct i915_gem_context;
@@ -191,6 +192,12 @@ struct intel_context {
        /* GuC context blocked fence */
      struct i915_sw_fence guc_blocked;
+
+    /*
+     * GuC priority management
+     */
+    u8 guc_prio;
+    u32 guc_prio_count[GUC_CLIENT_PRIORITY_NUM];
  };
    #endif /* __INTEL_CONTEXT_TYPES__ */
diff --git a/drivers/gpu/drm/i915/gt/intel_engine_user.c 
b/drivers/gpu/drm/i915/gt/intel_engine_user.c
index 84142127ebd8..8f8bea08e734 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine_user.c
+++ b/drivers/gpu/drm/i915/gt/intel_engine_user.c
@@ -11,6 +11,7 @@
  #include "intel_engine.h"
  #include "intel_engine_user.h"
  #include "intel_gt.h"
+#include "uc/intel_guc_submission.h"
    struct intel_engine_cs *
  intel_engine_lookup_user(struct drm_i915_private *i915, u8 class, 
u8 instance)
@@ -115,6 +116,9 @@ static void set_scheduler_caps(struct 
drm_i915_private *i915)
              disabled |= (I915_SCHEDULER_CAP_ENABLED |
                       I915_SCHEDULER_CAP_PRIORITY);
  +        if (intel_uc_uses_guc_submission(&i915->gt.uc))
+            enabled |= I915_SCHEDULER_CAP_STATIC_PRIORITY_MAP;
+
          for (i = 0; i < ARRAY_SIZE(map); i++) {
              if (engine->flags & BIT(map[i].engine))
                  enabled |= BIT(map[i].sched);
diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c 
b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
index 3ff42d6e934f..b760cbf6ca0e 100644
--- a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
+++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
@@ -81,6 +81,7 @@ guc_create_virtual(struct intel_engine_cs 
**siblings, unsigned int count);
   */
  #define SCHED_STATE_NO_LOCK_ENABLED            BIT(0)
  #define SCHED_STATE_NO_LOCK_PENDING_ENABLE        BIT(1)
+#define SCHED_STATE_NO_LOCK_REGISTERED            BIT(2)
  static inline bool context_enabled(struct intel_context *ce)
  {
      return (atomic_read(&ce->guc_sched_state_no_lock) &
@@ -116,6 +117,24 @@ static inline void 
clr_context_pending_enable(struct intel_context *ce)
             &ce->guc_sched_state_no_lock);
  }
  +static inline bool context_registered(struct intel_context *ce)
+{
+    return (atomic_read(&ce->guc_sched_state_no_lock) &
+        SCHED_STATE_NO_LOCK_REGISTERED);
+}
+
+static inline void set_context_registered(struct intel_context *ce)
+{
+    atomic_or(SCHED_STATE_NO_LOCK_REGISTERED,
+          &ce->guc_sched_state_no_lock);
+}
+
+static inline void clr_context_registered(struct intel_context *ce)
+{
+    atomic_and((u32)~SCHED_STATE_NO_LOCK_REGISTERED,
+           &ce->guc_sched_state_no_lock);
+}
+
  /*
   * Below is a set of functions which control the GuC scheduling 
state which
   * require a lock, aside from the special case where the functions 
are called
@@ -1092,6 +1111,7 @@ static int steal_guc_id(struct intel_guc *guc)
            list_del_init(&ce->guc_id_link);
          guc_id = ce->guc_id;
+        clr_context_registered(ce);
          set_context_guc_id_invalid(ce);
          return guc_id;
      } else {
@@ -1201,10 +1221,15 @@ static int register_context(struct 
intel_context *ce, bool loop)
      struct intel_guc *guc = ce_to_guc(ce);
      u32 offset = intel_guc_ggtt_offset(guc, guc->lrc_desc_pool) +
          ce->guc_id * sizeof(struct guc_lrc_desc);
+    int ret;
        trace_intel_context_register(ce);
  -    return __guc_action_register_context(guc, ce->guc_id, offset, 
loop);
+    ret = __guc_action_register_context(guc, ce->guc_id, offset, loop);
+    if (likely(!ret))
+        set_context_registered(ce);
+
+    return ret;
  }
    static int __guc_action_deregister_context(struct intel_guc *guc,
@@ -1260,6 +1285,8 @@ static void guc_context_policy_init(struct 
intel_engine_cs *engine,
      desc->preemption_timeout = engine->props.preempt_timeout_ms * 
1000;
  }
  +static inline u8 map_i915_prio_to_guc_prio(int prio);
+
  static int guc_lrc_desc_pin(struct intel_context *ce, bool loop)
  {
      struct intel_engine_cs *engine = ce->engine;
@@ -1267,6 +1294,8 @@ static int guc_lrc_desc_pin(struct 
intel_context *ce, bool loop)
      struct intel_guc *guc = &engine->gt->uc.guc;
      u32 desc_idx = ce->guc_id;
      struct guc_lrc_desc *desc;
+    const struct i915_gem_context *ctx;
+    int prio = I915_CONTEXT_DEFAULT_PRIORITY;
      bool context_registered;
      intel_wakeref_t wakeref;
      int ret = 0;
@@ -1282,6 +1311,12 @@ static int guc_lrc_desc_pin(struct 
intel_context *ce, bool loop)
        context_registered = lrc_desc_registered(guc, desc_idx);
  +    rcu_read_lock();
+    ctx = rcu_dereference(ce->gem_context);
+    if (ctx)
+        prio = ctx->sched.priority;
+    rcu_read_unlock();
+
      reset_lrc_desc(guc, desc_idx);
      set_lrc_desc_registered(guc, desc_idx, ce);
  @@ -1290,7 +1325,8 @@ static int guc_lrc_desc_pin(struct 
intel_context *ce, bool loop)
      desc->engine_submit_mask = adjust_engine_mask(engine->class,
                                engine->mask);
      desc->hw_context_desc = ce->lrc.lrca;
-    desc->priority = GUC_CLIENT_PRIORITY_KMD_NORMAL;
+    ce->guc_prio = map_i915_prio_to_guc_prio(prio);
+    desc->priority = ce->guc_prio;
      desc->context_flags = CONTEXT_REGISTRATION_FLAG_KMD;
      guc_context_policy_init(engine, desc);
      init_sched_state(ce);
@@ -1693,11 +1729,17 @@ static inline void guc_lrc_desc_unpin(struct 
intel_context *ce)
      GEM_BUG_ON(ce != __get_context(guc, ce->guc_id));
      GEM_BUG_ON(context_enabled(ce));
  +    clr_context_registered(ce);
      deregister_context(ce, ce->guc_id, true);
  }
    static void __guc_context_destroy(struct intel_context *ce)
  {
+ GEM_BUG_ON(ce->guc_prio_count[GUC_CLIENT_PRIORITY_KMD_HIGH] ||
+           ce->guc_prio_count[GUC_CLIENT_PRIORITY_HIGH] ||
+ ce->guc_prio_count[GUC_CLIENT_PRIORITY_KMD_NORMAL] ||
+           ce->guc_prio_count[GUC_CLIENT_PRIORITY_NORMAL]);
+
      lrc_fini(ce);
      intel_context_fini(ce);
  @@ -1791,15 +1833,124 @@ static int guc_context_alloc(struct 
intel_context *ce)
      return lrc_alloc(ce, ce->engine);
  }
  +static void guc_context_set_prio(struct intel_guc *guc,
+                 struct intel_context *ce,
+                 u8 prio)
+{
+    u32 action[] = {
+        INTEL_GUC_ACTION_SET_CONTEXT_PRIORITY,
+        ce->guc_id,
+        prio,
+    };
+
+    GEM_BUG_ON(prio < GUC_CLIENT_PRIORITY_KMD_HIGH ||
+           prio > GUC_CLIENT_PRIORITY_NORMAL);
+
+    if (ce->guc_prio == prio || submission_disabled(guc) ||
+        !context_registered(ce))
+        return;
+
+    guc_submission_send_busy_loop(guc, action, ARRAY_SIZE(action), 
0, true);
+
+    ce->guc_prio = prio;
+    trace_intel_context_set_prio(ce);
+}
+
+static inline u8 map_i915_prio_to_guc_prio(int prio)
+{
+    if (prio == I915_PRIORITY_NORMAL)
+        return GUC_CLIENT_PRIORITY_KMD_NORMAL;
+    else if (prio < I915_PRIORITY_NORMAL)
+        return GUC_CLIENT_PRIORITY_NORMAL;
+    else if (prio < I915_PRIORITY_DISPLAY)
+        return GUC_CLIENT_PRIORITY_HIGH;
+    else
+        return GUC_CLIENT_PRIORITY_KMD_HIGH;
+}
+
+static inline void add_context_inflight_prio(struct intel_context *ce,
+                         u8 guc_prio)
+{
+    lockdep_assert_held(&ce->guc_active.lock);
+    GEM_BUG_ON(guc_prio >= ARRAY_SIZE(ce->guc_prio_count));
+
+    ++ce->guc_prio_count[guc_prio];
+
+    /* Overflow protection */
+    GEM_WARN_ON(!ce->guc_prio_count[guc_prio]);
+}
+
+static inline void sub_context_inflight_prio(struct intel_context *ce,
+                         u8 guc_prio)
+{
+    lockdep_assert_held(&ce->guc_active.lock);
+    GEM_BUG_ON(guc_prio >= ARRAY_SIZE(ce->guc_prio_count));
+
+    /* Underflow protection */
+    GEM_WARN_ON(!ce->guc_prio_count[guc_prio]);
+
+    --ce->guc_prio_count[guc_prio];
+}
+
+static inline void update_context_prio(struct intel_context *ce)
+{
+    struct intel_guc *guc = &ce->engine->gt->uc.guc;
+    int i;
+
+    BUILD_BUG_ON(GUC_CLIENT_PRIORITY_KMD_HIGH != 0);
+    BUILD_BUG_ON(GUC_CLIENT_PRIORITY_KMD_HIGH > 
GUC_CLIENT_PRIORITY_NORMAL);
+
+    lockdep_assert_held(&ce->guc_active.lock);
+
+    for (i = 0; i < ARRAY_SIZE(ce->guc_prio_count); ++i) {
+        if (ce->guc_prio_count[i]) {
+            guc_context_set_prio(guc, ce, i);
+            break;
+        }
+    }
+}
+
+static inline bool new_guc_prio_higher(u8 old_guc_prio, u8 
new_guc_prio)
+{
+    /* Lower value is higher priority */
+    return new_guc_prio < old_guc_prio;
+}
+
  static void add_to_context(struct i915_request *rq)
  {
      struct intel_context *ce = rq->context;
+    u8 new_guc_prio = map_i915_prio_to_guc_prio(rq_prio(rq));
+
+    GEM_BUG_ON(rq->guc_prio == GUC_PRIO_FINI);
        spin_lock(&ce->guc_active.lock);
      list_move_tail(&rq->sched.link, &ce->guc_active.requests);
+
+    if (rq->guc_prio == GUC_PRIO_INIT) {
+        rq->guc_prio = new_guc_prio;
+        add_context_inflight_prio(ce, rq->guc_prio);
+    } else if (new_guc_prio_higher(rq->guc_prio, new_guc_prio)) {
+        sub_context_inflight_prio(ce, rq->guc_prio);
+        rq->guc_prio = new_guc_prio;
+        add_context_inflight_prio(ce, rq->guc_prio);
+    }
+    update_context_prio(ce);
+
      spin_unlock(&ce->guc_active.lock);
  }
  +static void guc_prio_fini(struct i915_request *rq, struct 
intel_context *ce)
+{
+    lockdep_assert_held(&ce->guc_active.lock);
+
+    if (rq->guc_prio != GUC_PRIO_INIT &&
+        rq->guc_prio != GUC_PRIO_FINI) {
+        sub_context_inflight_prio(ce, rq->guc_prio);
+        update_context_prio(ce);
+    }
+    rq->guc_prio = GUC_PRIO_FINI;
+}
+
  static void remove_from_context(struct i915_request *rq)
  {
      struct intel_context *ce = rq->context;
@@ -1812,6 +1963,8 @@ static void remove_from_context(struct 
i915_request *rq)
      /* Prevent further __await_execution() registering a cb, then 
flush */
      set_bit(I915_FENCE_FLAG_ACTIVE, &rq->fence.flags);
  +    guc_prio_fini(rq, ce);
+
      spin_unlock_irq(&ce->guc_active.lock);
        atomic_dec(&ce->guc_id_ref);
@@ -2093,6 +2246,39 @@ static void guc_init_breadcrumbs(struct 
intel_engine_cs *engine)
      }
  }
  +static void guc_bump_inflight_request_prio(struct i915_request *rq,
+                       int prio)
+{
+    struct intel_context *ce = rq->context;
+    u8 new_guc_prio = map_i915_prio_to_guc_prio(prio);
+
+    /* Short circuit function */
+    if (prio < I915_PRIORITY_NORMAL ||
+        rq->guc_prio == GUC_PRIO_FINI ||
+        (rq->guc_prio != GUC_PRIO_INIT &&
+         !new_guc_prio_higher(rq->guc_prio, new_guc_prio)))
+        return;
+
+    spin_lock(&ce->guc_active.lock);
+    if (rq->guc_prio != GUC_PRIO_FINI) {
+        if (rq->guc_prio != GUC_PRIO_INIT)
+            sub_context_inflight_prio(ce, rq->guc_prio);
+        rq->guc_prio = new_guc_prio;
+        add_context_inflight_prio(ce, rq->guc_prio);
+        update_context_prio(ce);
+    }
+    spin_unlock(&ce->guc_active.lock);
+}
+
+static void guc_retire_inflight_request_prio(struct i915_request *rq)
+{
+    struct intel_context *ce = rq->context;
+
+    spin_lock(&ce->guc_active.lock);
+    guc_prio_fini(rq, ce);
+    spin_unlock(&ce->guc_active.lock);
+}
+
  static void sanitize_hwsp(struct intel_engine_cs *engine)
  {
      struct intel_timeline *tl;
@@ -2317,6 +2503,10 @@ int intel_guc_submission_setup(struct 
intel_engine_cs *engine)
          guc->sched_engine->disabled = guc_sched_engine_disabled;
          guc->sched_engine->private_data = guc;
          guc->sched_engine->destroy = guc_sched_engine_destroy;
+        guc->sched_engine->bump_inflight_request_prio =
+            guc_bump_inflight_request_prio;
+        guc->sched_engine->retire_inflight_request_prio =
+            guc_retire_inflight_request_prio;
          tasklet_setup(&guc->sched_engine->tasklet,
                    guc_submission_tasklet);
      }
@@ -2694,6 +2884,22 @@ void intel_guc_submission_print_info(struct 
intel_guc *guc,
      drm_printf(p, "\n");
  }
  +static inline void guc_log_context_priority(struct drm_printer *p,
+                        struct intel_context *ce)
+{
+    int i;
+
+    drm_printf(p, "\t\tPriority: %d\n",
+           ce->guc_prio);
+    drm_printf(p, "\t\tNumber Requests (lower index == higher 
priority)\n");
+    for (i = GUC_CLIENT_PRIORITY_KMD_HIGH;
+         i < GUC_CLIENT_PRIORITY_NUM; ++i) {
+        drm_printf(p, "\t\tNumber requests in priority band[%d]: %d\n",
+               i, ce->guc_prio_count[i]);
+    }
+    drm_printf(p, "\n");
+}
+
  void intel_guc_submission_print_context_info(struct intel_guc *guc,
                           struct drm_printer *p)
  {
@@ -2716,6 +2922,8 @@ void 
intel_guc_submission_print_context_info(struct intel_guc *guc,
          drm_printf(p, "\t\tSchedule State: 0x%x, 0x%x\n\n",
                 ce->guc_state.sched_state,
atomic_read(&ce->guc_sched_state_no_lock));
+
+        guc_log_context_priority(p, ce);
      }
  }
  diff --git a/drivers/gpu/drm/i915/i915_request.c 
b/drivers/gpu/drm/i915/i915_request.c
index 541a20371502..1f1d4a6a0eff 100644
--- a/drivers/gpu/drm/i915/i915_request.c
+++ b/drivers/gpu/drm/i915/i915_request.c
@@ -114,6 +114,9 @@ static void i915_fence_release(struct dma_fence 
*fence)
  {
      struct i915_request *rq = to_request(fence);
  +    GEM_BUG_ON(rq->guc_prio != GUC_PRIO_INIT &&
+           rq->guc_prio != GUC_PRIO_FINI);
+
      /*
       * The request is put onto a RCU freelist (i.e. the address
       * is immediately reused), mark the fences as being freed now.
@@ -924,6 +927,8 @@ __i915_request_create(struct intel_context *ce, 
gfp_t gfp)
        rq->rcustate = get_state_synchronize_rcu(); /* acts as 
smp_mb() */
  +    rq->guc_prio = GUC_PRIO_INIT;
+
      /* We bump the ref for the fence chain */
i915_sw_fence_reinit(&i915_request_get(rq)->submit);
i915_sw_fence_reinit(&i915_request_get(rq)->semaphore);
diff --git a/drivers/gpu/drm/i915/i915_request.h 
b/drivers/gpu/drm/i915/i915_request.h
index ac0e3326c067..e6a0e0ebc9aa 100644
--- a/drivers/gpu/drm/i915/i915_request.h
+++ b/drivers/gpu/drm/i915/i915_request.h
@@ -293,6 +293,15 @@ struct i915_request {
       */
      struct list_head guc_fence_link;
  +    /**
+     * Priority level while the request is inflight. Differs from i915
+     * scheduler priority. See comment above
+     * I915_SCHEDULER_CAP_STATIC_PRIORITY_MAP for details.
+     */
+#define    GUC_PRIO_INIT    0xff
+#define    GUC_PRIO_FINI    0xfe
+    u8 guc_prio;
+
      I915_SELFTEST_DECLARE(struct {
          struct list_head link;
          unsigned long delay;
diff --git a/drivers/gpu/drm/i915/i915_scheduler.c 
b/drivers/gpu/drm/i915/i915_scheduler.c
index 28dd887eb1be..17843c204356 100644
--- a/drivers/gpu/drm/i915/i915_scheduler.c
+++ b/drivers/gpu/drm/i915/i915_scheduler.c
@@ -241,6 +241,9 @@ static void __i915_schedule(struct 
i915_sched_node *node,
      /* Fifo and depth-first replacement ensure our deps execute 
before us */
      sched_engine = lock_sched_engine(node, sched_engine, &cache);
      list_for_each_entry_safe_reverse(dep, p, &dfs, dfs_link) {
+        struct i915_request *from = container_of(dep->signaler,
+                             struct i915_request,
+                             sched);
          INIT_LIST_HEAD(&dep->dfs_link);
            node = dep->signaler;
@@ -254,6 +257,10 @@ static void __i915_schedule(struct 
i915_sched_node *node,
GEM_BUG_ON(node_to_request(node)->engine->sched_engine !=
                 sched_engine);
  +        /* Must be called before changing the nodes priority */
+        if (sched_engine->bump_inflight_request_prio)
+            sched_engine->bump_inflight_request_prio(from, prio);
+
          WRITE_ONCE(node->attr.priority, prio);
            /*
diff --git a/drivers/gpu/drm/i915/i915_scheduler_types.h 
b/drivers/gpu/drm/i915/i915_scheduler_types.h
index eaef233e9080..b0a1b58c7893 100644
--- a/drivers/gpu/drm/i915/i915_scheduler_types.h
+++ b/drivers/gpu/drm/i915/i915_scheduler_types.h
@@ -179,6 +179,18 @@ struct i915_sched_engine {
      void    (*kick_backend)(const struct i915_request *rq,
                  int prio);
  +    /**
+     * @bump_inflight_request_prio: update priority of an inflight 
request
+     */
+    void    (*bump_inflight_request_prio)(struct i915_request *rq,
+                          int prio);
+
+    /**
+     * @retire_inflight_request_prio: indicate request is retired to
+     * priority tracking
+     */
+    void    (*retire_inflight_request_prio)(struct i915_request *rq);
+
      /**
       * @schedule: adjust priority of request
       *
diff --git a/drivers/gpu/drm/i915/i915_trace.h 
b/drivers/gpu/drm/i915/i915_trace.h
index 9613a7c19661..806ad688274b 100644
--- a/drivers/gpu/drm/i915/i915_trace.h
+++ b/drivers/gpu/drm/i915/i915_trace.h
@@ -904,6 +904,7 @@ DECLARE_EVENT_CLASS(intel_context,
                   __field(int, pin_count)
                   __field(u32, sched_state)
                   __field(u32, guc_sched_state_no_lock)
+                 __field(u8, guc_prio)
                   ),
                TP_fast_assign(
@@ -912,12 +913,19 @@ DECLARE_EVENT_CLASS(intel_context,
                 __entry->sched_state = ce->guc_state.sched_state;
                 __entry->guc_sched_state_no_lock =
atomic_read(&ce->guc_sched_state_no_lock);
+               __entry->guc_prio = ce->guc_prio;
                 ),
  -            TP_printk("guc_id=%d, pin_count=%d 
sched_state=0x%x,0x%x",
+            TP_printk("guc_id=%d, pin_count=%d 
sched_state=0x%x,0x%x, guc_prio=%u",
                    __entry->guc_id, __entry->pin_count,
                    __entry->sched_state,
-                  __entry->guc_sched_state_no_lock)
+                  __entry->guc_sched_state_no_lock,
+                  __entry->guc_prio)
+);
+
+DEFINE_EVENT(intel_context, intel_context_set_prio,
+         TP_PROTO(struct intel_context *ce),
+         TP_ARGS(ce)
  );
    DEFINE_EVENT(intel_context, intel_context_reset,
@@ -1017,6 +1025,11 @@ trace_i915_request_out(struct i915_request *rq)
  {
  }
  +static inline void
+trace_intel_context_set_prio(struct intel_context *ce)
+{
+}
+
  static inline void
  trace_intel_context_reset(struct intel_context *ce)
  {
diff --git a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h
index 975087553ea0..7f13d241417f 100644
--- a/include/uapi/drm/i915_drm.h
+++ b/include/uapi/drm/i915_drm.h
@@ -572,6 +572,15 @@ typedef struct drm_i915_irq_wait {
  #define   I915_SCHEDULER_CAP_PREEMPTION    (1ul << 2)
  #define   I915_SCHEDULER_CAP_SEMAPHORES    (1ul << 3)
  #define   I915_SCHEDULER_CAP_ENGINE_BUSY_STATS    (1ul << 4)
+/*
+ * Indicates the 2k user priority levels are statically mapped into 
3 buckets as
+ * follows:
+ *
+ * -1k to -1    Low priority
+ * 0        Normal priority
+ * 1 to 1k    Highest priority
+ */
+#define   I915_SCHEDULER_CAP_STATIC_PRIORITY_MAP    (1ul << 5)
    #define I915_PARAM_HUC_STATUS         42







