drm/i915/gt: Include reset failures in the trace
authorChris Wilson <chris@chris-wilson.co.uk>
Fri, 4 Dec 2020 15:12:33 +0000 (15:12 +0000)
committerChris Wilson <chris@chris-wilson.co.uk>
Fri, 4 Dec 2020 15:41:33 +0000 (15:41 +0000)
The GT and engine reset failures are completely invisible when looking at
a trace for a bug, but are vital to understanding the incomplete flow.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20201204151234.19729-3-chris@chris-wilson.co.uk
drivers/gpu/drm/i915/gt/intel_reset.c

index 9fb4306..b3ccf78 100644 (file)
@@ -231,7 +231,7 @@ static int g4x_do_reset(struct intel_gt *gt,
                              GRDOM_MEDIA | GRDOM_RESET_ENABLE);
        ret =  wait_for_atomic(g4x_reset_complete(pdev), 50);
        if (ret) {
-               drm_dbg(&gt->i915->drm, "Wait for media reset failed\n");
+               GT_TRACE(gt, "Wait for media reset failed\n");
                goto out;
        }
 
@@ -239,7 +239,7 @@ static int g4x_do_reset(struct intel_gt *gt,
                              GRDOM_RENDER | GRDOM_RESET_ENABLE);
        ret =  wait_for_atomic(g4x_reset_complete(pdev), 50);
        if (ret) {
-               drm_dbg(&gt->i915->drm, "Wait for render reset failed\n");
+               GT_TRACE(gt, "Wait for render reset failed\n");
                goto out;
        }
 
@@ -265,7 +265,7 @@ static int ilk_do_reset(struct intel_gt *gt, intel_engine_mask_t engine_mask,
                                           5000, 0,
                                           NULL);
        if (ret) {
-               drm_dbg(&gt->i915->drm, "Wait for render reset failed\n");
+               GT_TRACE(gt, "Wait for render reset failed\n");
                goto out;
        }
 
@@ -276,7 +276,7 @@ static int ilk_do_reset(struct intel_gt *gt, intel_engine_mask_t engine_mask,
                                           5000, 0,
                                           NULL);
        if (ret) {
-               drm_dbg(&gt->i915->drm, "Wait for media reset failed\n");
+               GT_TRACE(gt, "Wait for media reset failed\n");
                goto out;
        }
 
@@ -305,9 +305,9 @@ static int gen6_hw_domain_reset(struct intel_gt *gt, u32 hw_domain_mask)
                                           500, 0,
                                           NULL);
        if (err)
-               drm_dbg(&gt->i915->drm,
-                       "Wait for 0x%08x engines reset failed\n",
-                       hw_domain_mask);
+               GT_TRACE(gt,
+                        "Wait for 0x%08x engines reset failed\n",
+                        hw_domain_mask);
 
        return err;
 }
@@ -407,8 +407,7 @@ static int gen11_lock_sfc(struct intel_engine_cs *engine, u32 *hw_mask)
                return 0;
 
        if (ret) {
-               drm_dbg(&engine->i915->drm,
-                       "Wait for SFC forced lock ack failed\n");
+               ENGINE_TRACE(engine, "Wait for SFC forced lock ack failed\n");
                return ret;
        }
 
@@ -1148,8 +1147,7 @@ int intel_engine_reset(struct intel_engine_cs *engine, const char *msg)
                ret = intel_guc_reset_engine(&engine->gt->uc.guc, engine);
        if (ret) {
                /* If we fail here, we expect to fallback to a global reset */
-               drm_dbg(&gt->i915->drm, "%sFailed to reset %s, ret=%d\n",
-                       uses_guc ? "GuC " : "", engine->name, ret);
+               ENGINE_TRACE(engine, "Failed to reset, err: %d\n", ret);
                goto out;
        }
 
@@ -1186,7 +1184,7 @@ static void intel_gt_reset_global(struct intel_gt *gt,
 
        kobject_uevent_env(kobj, KOBJ_CHANGE, error_event);
 
-       drm_dbg(&gt->i915->drm, "resetting chip, engines=%x\n", engine_mask);
+       GT_TRACE(gt, "resetting chip, engines=%x\n", engine_mask);
        kobject_uevent_env(kobj, KOBJ_CHANGE, reset_event);
 
        /* Use a watchdog to ensure that our reset completes */