On 9/26/2017 7:27 PM, Michał Winiarski wrote:
On Tue, Sep 26, 2017 at 02:49:48PM +0100, Chris Wilson wrote:
Quoting Sagar Arun Kamble (2017-09-26 14:24:43)
With this patch we disable GuC submission in i915_drm_suspend path.
This will destroy the client which will be setup back again. We also
reuse the complete sanitization done via intel_uc_runtime_suspend in
this path. Post i915_drm_resume, this state is recreated by
intel_uc_init_hw hence we need not have similar reuse for intel_uc_resume.
This also fixes issue where intel_uc_fini_hw was being called after GPU
reset happening in i915_gem_suspend during i915_driver_unload.
v2: Rebase w.r.t removal of GuC code restructuring. Added struct_mutex
protection for i915_guc_submission_disable.
v3: Rebase w.r.t updated GuC suspend function name.
v4: Added lockdep assert in i915_guc_submission_enable/disable.
Refined intel_uc_suspend to remove unnecessary locals and simplify
return. (Michal Winiarski)
Removed comment in guc_client_free about ignoring failure for
destroy_doorbell. (Oscar)
Rebase w.r.t i915_modparams change.
Cc: Michal Wajdeczko <michal.wajdeczko@xxxxxxxxx>
Cc: Michał Winiarski <michal.winiarski@xxxxxxxxx>
Reviewed-by: Michał Winiarski <michal.winiarski@xxxxxxxxx>
Signed-off-by: Sagar Arun Kamble <sagar.a.kamble@xxxxxxxxx>
---
drivers/gpu/drm/i915/i915_guc_submission.c | 15 ++++++++++++---
drivers/gpu/drm/i915/intel_uc.c | 11 +++++++----
2 files changed, 19 insertions(+), 7 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_guc_submission.c b/drivers/gpu/drm/i915/i915_guc_submission.c
index d1d6c0d..0c56765 100644
--- a/drivers/gpu/drm/i915/i915_guc_submission.c
+++ b/drivers/gpu/drm/i915/i915_guc_submission.c
@@ -885,9 +885,6 @@ static void guc_client_free(struct i915_guc_client *client)
* Be sure to drop any locks
*/
- /* FIXME: in many cases, by the time we get here the GuC has been
- * reset, so we cannot destroy the doorbell properly. Ignore the
- * error message for now */
destroy_doorbell(client);
guc_stage_desc_fini(client->guc, client);
i915_gem_object_unpin_map(client->vma->obj);
@@ -1154,6 +1151,12 @@ int i915_guc_submission_enable(struct drm_i915_private *dev_priv)
sizeof(struct guc_wq_item) *
I915_NUM_ENGINES > GUC_WQ_SIZE);
+ /*
+ * Assert that drm.struct_mutex is held.
Ahem.
+ * Needed for GuC client vma binding.
+ */
+ lockdep_assert_held(&dev_priv->drm.struct_mutex);
If you don't directly depend on struct_mutex, don't assert it. Otherwise
the person who removes that requirement will get very confused and upset.
-Chris
My bad - I suggested that.
Failed to notice *why* we're taking the mutex, and that we're already have
lockdep assert in the right place (__i915_gem_object_release_unless_active).
Sorry :)
-Michał
Will remove this. Thanks.
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
