drm/i915/perf: fix perf stream opening lock
authorLionel Landwerlin <lionel.g.landwerlin@intel.com>
Thu, 1 Mar 2018 11:06:13 +0000 (11:06 +0000)
committerRodrigo Vivi <rodrigo.vivi@intel.com>
Tue, 6 Mar 2018 00:08:28 +0000 (16:08 -0800)
We're seeing on CI that some contexts don't have the programmed OA
period timer that directs the OA unit on how often to write reports.

The issue is that we're not holding the drm lock from when we edit the
context images down to when we set the exclusive_stream variable. This
leaves a window for the deferred context allocation to call
i915_oa_init_reg_state() that will not program the expected OA timer
value, because we haven't set the exclusive_stream yet.

v2: Drop need_lock from gen8_configure_all_contexts() (Matt)

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Fixes: 701f8231a2f ("drm/i915/perf: prune OA configs")
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102254
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103715
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103755
Link: https://patchwork.freedesktop.org/patch/msgid/20180301110613.1737-1-lionel.g.landwerlin@intel.com
Cc: Jani Nikula <jani.nikula@linux.intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: intel-gfx@lists.freedesktop.org
Cc: <stable@vger.kernel.org> # v4.14+
(cherry picked from commit 41d3fdcd15d5ecf29cc73e8b79c2327ebb54b960)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
drivers/gpu/drm/i915/i915_perf.c

index 0be50e43507de0e15ef06245335940eed75d3f6b..f8fe5ffcdcfff8cb27667efdcec2914eb3b7b06b 100644 (file)
@@ -1303,9 +1303,8 @@ static void i915_oa_stream_destroy(struct i915_perf_stream *stream)
         */
        mutex_lock(&dev_priv->drm.struct_mutex);
        dev_priv->perf.oa.exclusive_stream = NULL;
-       mutex_unlock(&dev_priv->drm.struct_mutex);
-
        dev_priv->perf.oa.ops.disable_metric_set(dev_priv);
+       mutex_unlock(&dev_priv->drm.struct_mutex);
 
        free_oa_buffer(dev_priv);
 
@@ -1756,22 +1755,13 @@ static int gen8_switch_to_updated_kernel_context(struct drm_i915_private *dev_pr
  * Note: it's only the RCS/Render context that has any OA state.
  */
 static int gen8_configure_all_contexts(struct drm_i915_private *dev_priv,
-                                      const struct i915_oa_config *oa_config,
-                                      bool interruptible)
+                                      const struct i915_oa_config *oa_config)
 {
        struct i915_gem_context *ctx;
        int ret;
        unsigned int wait_flags = I915_WAIT_LOCKED;
 
-       if (interruptible) {
-               ret = i915_mutex_lock_interruptible(&dev_priv->drm);
-               if (ret)
-                       return ret;
-
-               wait_flags |= I915_WAIT_INTERRUPTIBLE;
-       } else {
-               mutex_lock(&dev_priv->drm.struct_mutex);
-       }
+       lockdep_assert_held(&dev_priv->drm.struct_mutex);
 
        /* Switch away from any user context. */
        ret = gen8_switch_to_updated_kernel_context(dev_priv, oa_config);
@@ -1819,8 +1809,6 @@ static int gen8_configure_all_contexts(struct drm_i915_private *dev_priv,
        }
 
  out:
-       mutex_unlock(&dev_priv->drm.struct_mutex);
-
        return ret;
 }
 
@@ -1863,7 +1851,7 @@ static int gen8_enable_metric_set(struct drm_i915_private *dev_priv,
         * to make sure all slices/subslices are ON before writing to NOA
         * registers.
         */
-       ret = gen8_configure_all_contexts(dev_priv, oa_config, true);
+       ret = gen8_configure_all_contexts(dev_priv, oa_config);
        if (ret)
                return ret;
 
@@ -1878,7 +1866,7 @@ static int gen8_enable_metric_set(struct drm_i915_private *dev_priv,
 static void gen8_disable_metric_set(struct drm_i915_private *dev_priv)
 {
        /* Reset all contexts' slices/subslices configurations. */
-       gen8_configure_all_contexts(dev_priv, NULL, false);
+       gen8_configure_all_contexts(dev_priv, NULL);
 
        I915_WRITE(GDT_CHICKEN_BITS, (I915_READ(GDT_CHICKEN_BITS) &
                                      ~GT_NOA_ENABLE));
@@ -1888,7 +1876,7 @@ static void gen8_disable_metric_set(struct drm_i915_private *dev_priv)
 static void gen10_disable_metric_set(struct drm_i915_private *dev_priv)
 {
        /* Reset all contexts' slices/subslices configurations. */
-       gen8_configure_all_contexts(dev_priv, NULL, false);
+       gen8_configure_all_contexts(dev_priv, NULL);
 
        /* Make sure we disable noa to save power. */
        I915_WRITE(RPM_CONFIG1,
@@ -2138,6 +2126,10 @@ static int i915_oa_stream_init(struct i915_perf_stream *stream,
        if (ret)
                goto err_oa_buf_alloc;
 
+       ret = i915_mutex_lock_interruptible(&dev_priv->drm);
+       if (ret)
+               goto err_lock;
+
        ret = dev_priv->perf.oa.ops.enable_metric_set(dev_priv,
                                                      stream->oa_config);
        if (ret)
@@ -2145,23 +2137,17 @@ static int i915_oa_stream_init(struct i915_perf_stream *stream,
 
        stream->ops = &i915_oa_stream_ops;
 
-       /* Lock device for exclusive_stream access late because
-        * enable_metric_set() might lock as well on gen8+.
-        */
-       ret = i915_mutex_lock_interruptible(&dev_priv->drm);
-       if (ret)
-               goto err_lock;
-
        dev_priv->perf.oa.exclusive_stream = stream;
 
        mutex_unlock(&dev_priv->drm.struct_mutex);
 
        return 0;
 
-err_lock:
+err_enable:
        dev_priv->perf.oa.ops.disable_metric_set(dev_priv);
+       mutex_unlock(&dev_priv->drm.struct_mutex);
 
-err_enable:
+err_lock:
        free_oa_buffer(dev_priv);
 
 err_oa_buf_alloc: