[Intel-gfx] [PATCH v2] drm/i915: Unwind i915_gem_init() failure

[Chris Wilson]

Since Michal introduced new errors other than -EIO during
i915_gem_init(), we need to actually unwind on the error path as we have
to abort the module load (and we expect to do so cleanly!).

As we now teardown key state and then mark the driver as wedged (on
EIO), we have to be careful to not allow ourselves to resume and
unwedge, thus attempting to use the uninitialised driver.

v2: Try not to free driver state for the suppressed EIO

References: 8620eb1dbbf2 ("drm/i915/uc: Don't use -EIO to report missing firmware")
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
Cc: Michal Wajdeczko <michal.wajdeczko at intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen at linux.intel.com>
Cc: Sagar Arun Kamble <sagar.a.kamble at intel.com>
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_gem.c | 82 +++++++++++++++++++++++++++++++++--------
 1 file changed, 67 insertions(+), 15 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index c7b5db78fbb4..ee243e1ef706 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -3235,7 +3235,12 @@ bool i915_gem_unset_wedged(struct drm_i915_private *i915)
 	if (!test_bit(I915_WEDGED, &i915->gpu_error.flags))
 		return true;
 
-	/* Before unwedging, make sure that all pending operations
+	/* Never successfully initialised, so can not unwedge? */
+	if (!i915->kernel_context)
+		return false;
+
+	/*
+	 * Before unwedging, make sure that all pending operations
 	 * are flushed and errored out - we may have requests waiting upon
 	 * third party fences. We marked all inflight requests as EIO, and
 	 * every execbuf since returned EIO, for consistency we want all
@@ -4853,7 +4858,8 @@ void i915_gem_resume(struct drm_i915_private *i915)
 	i915_gem_restore_gtt_mappings(i915);
 	i915_gem_restore_fences(i915);
 
-	/* As we didn't flush the kernel context before suspend, we cannot
+	/*
+	 * As we didn't flush the kernel context before suspend, we cannot
 	 * guarantee that the context image is complete. So let's just reset
 	 * it and start again.
 	 */
@@ -4874,8 +4880,10 @@ void i915_gem_resume(struct drm_i915_private *i915)
 	return;
 
 err_wedged:
-	DRM_ERROR("failed to re-initialize GPU, declaring wedged!\n");
-	i915_gem_set_wedged(i915);
+	if (!i915_terminally_wedged(&i915->gpu_error)) {
+		DRM_ERROR("failed to re-initialize GPU, declaring wedged!\n");
+		i915_gem_set_wedged(i915);
+	}
 	goto out_unlock;
 }
 
@@ -5158,22 +5166,28 @@ int i915_gem_init(struct drm_i915_private *dev_priv)
 	intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
 
 	ret = i915_gem_init_ggtt(dev_priv);
-	if (ret)
-		goto out_unlock;
+	if (ret) {
+		GEM_BUG_ON(ret == -EIO);
+		goto err_unlock;
+	}
 
 	ret = i915_gem_contexts_init(dev_priv);
-	if (ret)
-		goto out_unlock;
+	if (ret) {
+		GEM_BUG_ON(ret == -EIO);
+		goto err_ggtt;
+	}
 
 	ret = intel_engines_init(dev_priv);
-	if (ret)
-		goto out_unlock;
+	if (ret) {
+		GEM_BUG_ON(ret == -EIO);
+		goto err_context;
+	}
 
 	intel_init_gt_powersave(dev_priv);
 
 	ret = i915_gem_init_hw(dev_priv);
 	if (ret)
-		goto out_unlock;
+		goto err_pm;
 
 	/*
 	 * Despite its name intel_init_clock_gating applies both display
@@ -5187,9 +5201,48 @@ int i915_gem_init(struct drm_i915_private *dev_priv)
 	intel_init_clock_gating(dev_priv);
 
 	ret = __intel_engines_record_defaults(dev_priv);
-out_unlock:
+	if (ret)
+		goto err_init_hw;
+
+	if (i915_inject_load_failure()) {
+		ret = -EIO;
+		goto err_init_hw;
+	}
+
+	intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
+	mutex_unlock(&dev_priv->drm.struct_mutex);
+
+	return 0;
+
+	/*
+	 * Unwinding is complicated by that we want to handle -EIO to mean
+	 * disable GPU submission but keep KMS alive. We want to mark the
+	 * HW as irrevisibly wedged, but keep enough state around that the
+	 * driver doesn't explode during runtime.
+	 */
+err_init_hw:
+	i915_gem_wait_for_idle(dev_priv, I915_WAIT_LOCKED);
+	i915_gem_contexts_lost(dev_priv);
+	intel_uc_fini_hw(dev_priv);
+err_pm:
+	if (ret != -EIO) {
+		intel_cleanup_gt_powersave(dev_priv);
+		i915_gem_cleanup_engines(dev_priv);
+	}
+err_context:
+	if (ret != -EIO)
+		i915_gem_contexts_fini(dev_priv);
+err_ggtt:
+err_unlock:
+	intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
+	mutex_unlock(&dev_priv->drm.struct_mutex);
+
+	if (ret != -EIO)
+		i915_gem_cleanup_userptr(dev_priv);
+
 	if (ret == -EIO) {
-		/* Allow engine initialisation to fail by marking the GPU as
+		/*
+		 * Allow engine initialisation to fail by marking the GPU as
 		 * wedged. But we only want to do this where the GPU is angry,
 		 * for all other failure, such as an allocation failure, bail.
 		 */
@@ -5199,9 +5252,8 @@ int i915_gem_init(struct drm_i915_private *dev_priv)
 		}
 		ret = 0;
 	}
-	intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
-	mutex_unlock(&dev_priv->drm.struct_mutex);
 
+	i915_gem_drain_freed_objects(dev_priv);
 	return ret;
 }
 
-- 
2.15.1