The engine registers really shouldn't be touched during GuC submission
as the GuC owns the registers. Don't call ring_is_idle and tie
intel_engine_is_idle strickly the engine pm.

Because intel_engine_is_idle tied to the engine pm, retire requests
before checking intel_engines_are_idle in gt_drop_caches, and lastly
increase the timeout in gt_drop_caches for the intel_engines_are_idle
check.

Signed-off-by: Matthew Brost <matthew.br...@intel.com>
---
 drivers/gpu/drm/i915/gt/intel_engine_cs.c | 13 +++++++++++++
 drivers/gpu/drm/i915/i915_debugfs.c       |  6 +++---
 drivers/gpu/drm/i915/i915_drv.h           |  2 +-
 3 files changed, 17 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/i915/gt/intel_engine_cs.c 
b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
index e34a61600c8c..591226b96201 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine_cs.c
+++ b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
@@ -1226,6 +1226,9 @@ static bool ring_is_idle(struct intel_engine_cs *engine)
 {
        bool idle = true;
 
+       /* GuC submission shouldn't access HEAD & TAIL via MMIO */
+       GEM_BUG_ON(intel_engine_uses_guc(engine));
+
        if (I915_SELFTEST_ONLY(!engine->mmio_base))
                return true;
 
@@ -1292,6 +1295,16 @@ bool intel_engine_is_idle(struct intel_engine_cs *engine)
        if (!i915_sched_engine_is_empty(engine->sched_engine))
                return false;
 
+       /*
+        * We shouldn't touch engine registers with GuC submission as the GuC
+        * owns the registers. Let's tie the idle to engine pm, at worst this
+        * function sometimes will falsely report non-idle when idle during the
+        * delay to retire requests or with virtual engines and a request
+        * running on another instance within the same class / submit mask.
+        */
+       if (intel_engine_uses_guc(engine))
+               return false;
+
        /* Ring stopped? */
        return ring_is_idle(engine);
 }
diff --git a/drivers/gpu/drm/i915/i915_debugfs.c 
b/drivers/gpu/drm/i915/i915_debugfs.c
index d540dd8029d0..2639961504b5 100644
--- a/drivers/gpu/drm/i915/i915_debugfs.c
+++ b/drivers/gpu/drm/i915/i915_debugfs.c
@@ -867,13 +867,13 @@ gt_drop_caches(struct intel_gt *gt, u64 val)
 {
        int ret;
 
+       if (val & DROP_RETIRE || val & DROP_RESET_ACTIVE)
+               intel_gt_retire_requests(gt);
+
        if (val & DROP_RESET_ACTIVE &&
            wait_for(intel_engines_are_idle(gt), I915_IDLE_ENGINES_TIMEOUT))
                intel_gt_set_wedged(gt);
 
-       if (val & DROP_RETIRE)
-               intel_gt_retire_requests(gt);
-
        if (val & (DROP_IDLE | DROP_ACTIVE)) {
                ret = intel_gt_wait_for_idle(gt, MAX_SCHEDULE_TIMEOUT);
                if (ret)
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 3cfa6effbb5f..aa359b8480cd 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -576,7 +576,7 @@ struct i915_gem_mm {
        u32 shrink_count;
 };
 
-#define I915_IDLE_ENGINES_TIMEOUT (200) /* in ms */
+#define I915_IDLE_ENGINES_TIMEOUT (500) /* in ms */
 
 unsigned long i915_fence_context_timeout(const struct drm_i915_private *i915,
                                         u64 context);
-- 
2.28.0

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Reply via email to