Skip to contentSkip to conversation

drm/i915: Unlock the shared hwsp_gtt object after pinning

PATCH In review Mika Kuoppala 4 participants 4 messages Sep 3, 2020

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 drm/i915: Unlock the shared hwsp_gtt object after pinning

From: Thomas Hellström <[email protected]>

The hwsp_gtt object is used for sub-allocation and could therefore be shared by many contexts causing unnecessary contention during concurrent context pinning.
However since we're currently locking it only for pinning, it remains resident until we unpin it, and therefore it's safe to drop the lock early, allowing for concurrent thread access.

---
 drivers/gpu/drm/i915/gt/intel_context.c | 7 +++++++
 1 file changed, 7 insertions(+)
drivers/gpu/drm/i915/gt/intel_context.c +7 -0
index 61b05cd4c47a..65e956ba19e1 100644
@@ -271,6 +271,13 @@ int __intel_context_do_pin_ww(struct intel_context *ce,
 	i915_active_release(&ce->active);
 err_ctx_unpin:
 	intel_context_post_unpin(ce);
+
+	/*
+	 * Unlock the hwsp_ggtt object since it's shared. This is fine for now
+	 * since the lock has been used for pinning only, not fencing.
+	 */
+	i915_gem_ww_unlock_single(ce->timeline->hwsp_ggtt->obj);
+
 	return err;
 }
 
Signature · 6 lines
2.17.1

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#1 lore raw

Quoting Mika Kuoppala (2020-09-03 14:31:44)

Quoted 19 lines

From: Thomas Hellström <[email protected]>

The hwsp_gtt object is used for sub-allocation and could therefore be shared by many contexts causing unnecessary contention during concurrent context pinning.
However since we're currently locking it only for pinning, it remains resident until we unpin it, and therefore it's safe to drop the lock early, allowing for concurrent thread access.

---
 drivers/gpu/drm/i915/gt/intel_context.c | 7 +++++++
 1 file changed, 7 insertions(+)
drivers/gpu/drm/i915/gt/intel_context.c +0 -0
index 61b05cd4c47a..65e956ba19e1 100644

drivers/gpu/drm/i915/gt/intel_context.c int __intel_context_do_pin_ww(struct intel_context *ce, · line 279

Quoted 9 lines
drivers/gpu/drm/i915/gt/intel_context.c +6 -0
@@ -271,6 +271,13 @@ int __intel_context_do_pin_ww(struct intel_context *ce,
        i915_active_release(&ce->active);
 err_ctx_unpin:
        intel_context_post_unpin(ce);
+
+       /*
+        * Unlock the hwsp_ggtt object since it's shared. This is fine for now
+        * since the lock has been used for pinning only, not fencing.
+        */
+       i915_gem_ww_unlock_single(ce->timeline->hwsp_ggtt->obj);

The timeline is not special here, the same rules for locking/unlock can equally be applied to all the global state. You may even go as far as putting a local acquire context here, which would then have avoided the cross driver pollution.

Anyway, if it works for the timeline, there is no reason to keep the other globals locked. They are pinned and on completely different usage cycles to the user objects. [They can't be evicted without interacting with the HW to ensure the context has been flushed, so far there has been no way to do so without stalling for a freshly submitted request, hence we let them retire gracefully and kick the kernel timelines to flush completed contexts.]
-Chris
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#2 lore raw
Quoted 34 lines · Chris Wilson

Quoting Mika Kuoppala (2020-09-03 14:31:44)

Quoted 19 lines

From: Thomas Hellström <[email protected]>

The hwsp_gtt object is used for sub-allocation and could therefore be shared by many contexts causing unnecessary contention during concurrent context pinning.
However since we're currently locking it only for pinning, it remains resident until we unpin it, and therefore it's safe to drop the lock early, allowing for concurrent thread access.

---
  drivers/gpu/drm/i915/gt/intel_context.c | 7 +++++++
  1 file changed, 7 insertions(+)
drivers/gpu/drm/i915/gt/intel_context.c +0 -0
index 61b05cd4c47a..65e956ba19e1 100644

drivers/gpu/drm/i915/gt/intel_context.c int __intel_context_do_pin_ww(struct intel_context *ce, · line 279

Quoted 9 lines
drivers/gpu/drm/i915/gt/intel_context.c +6 -0
@@ -271,6 +271,13 @@ int __intel_context_do_pin_ww(struct intel_context *ce,
         i915_active_release(&ce->active);
  err_ctx_unpin:
         intel_context_post_unpin(ce);
+
+       /*
+        * Unlock the hwsp_ggtt object since it's shared. This is fine for now
+        * since the lock has been used for pinning only, not fencing.
+        */
+       i915_gem_ww_unlock_single(ce->timeline->hwsp_ggtt->obj);

The timeline is not special here, the same rules for locking/unlock can equally be applied to all the global state. You may even go as far as putting a local acquire context here, which would then have avoided the cross driver pollution.

Quoted 3 lines

Anyway, if it works for the timeline, there is no reason to keep the other globals locked. They are pinned and on completely different usage cycles to the user objects.

Agreed.  In principle everything we want to perma-pin we can unlock early, Keeping it under the same acquire context makes it possible to do this as part of a bigger ww transaction. Although initially this solution was discarded in favour of removing the suballocation to be able to ditch the perma-pinning in the future.

But is this acceptable to address the pi-isolated issue until we have something more thought-through in place? Or is there somethin specific you want me to change for a R-B?

Thanks,

Thomas

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#3 lore raw

[Intel-gfx] ✓ Fi.CI.BAT: success for drm/i915: Unlock the shared hwsp_gtt object after pinning

In reply to Mika Kuoppala
== Series Details ==

== Series Details ==

Series: drm/i915: Unlock the shared hwsp_gtt object after pinning
URL : https://patchwork.freedesktop.org/series/81290/
State : success

== Summary ==

CI Bug Log - changes from CI_DRM_8959 -> Patchwork_18439
====================================================

Summary
-------

**SUCCESS**

No regressions found.

External URL: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_18439/index.html

Known issues
------------

Here are the changes found in Patchwork_18439 that come from known issues:

### IGT changes ###

#### Issues hit ####

* igt@kms_cursor_legacy@basic-busy-flip-before-cursor-atomic:
- fi-icl-u2: [PASS][1] -> [DMESG-WARN][2] ([i915#1982])
[1]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_8959/fi-icl-u2/igt@[email protected]
[2]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_18439/fi-icl-u2/igt@[email protected]

#### Possible fixes ####

* igt@i915_selftest@live@gem_contexts:
- fi-tgl-u2: [INCOMPLETE][3] ([i915#2045]) -> [PASS][4]
[3]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_8959/fi-tgl-u2/igt@i915_selftest@live@gem_contexts.html
[4]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_18439/fi-tgl-u2/igt@i915_selftest@live@gem_contexts.html

* igt@i915_selftest@live@requests:
- fi-icl-y: [INCOMPLETE][5] ([i915#1958]) -> [PASS][6]
[5]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_8959/fi-icl-y/igt@i915_selftest@[email protected]
[6]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_18439/fi-icl-y/igt@i915_selftest@[email protected]

* igt@kms_chamelium@hdmi-crc-fast:
- fi-kbl-7500u: [DMESG-FAIL][7] ([i915#165]) -> [PASS][8]
[7]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_8959/fi-kbl-7500u/igt@[email protected]
[8]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_18439/fi-kbl-7500u/igt@[email protected]

{name}: This element is suppressed. This means it is ignored when computing
the status of the difference (SUCCESS, WARNING, or FAILURE).

[i915#165]: https://gitlab.freedesktop.org/drm/intel/issues/165
[i915#1958]: https://gitlab.freedesktop.org/drm/intel/issues/1958
[i915#1982]: https://gitlab.freedesktop.org/drm/intel/issues/1982
[i915#2045]: https://gitlab.freedesktop.org/drm/intel/issues/2045
[i915#2417]: https://gitlab.freedesktop.org/drm/intel/issues/2417

Participating hosts (39 -> 33)
------------------------------

Missing (6): fi-ilk-m540 fi-hsw-4200u fi-byt-squawks fi-bsw-cyan fi-ctg-p8600 fi-byt-clapper

Build changes
-------------

* Linux: CI_DRM_8959 -> Patchwork_18439

CI-20190529: 20190529
CI_DRM_8959: b570f98019dd32803271895707fb3ec98eb9636a @ git://anongit.freedesktop.org/gfx-ci/linux
IGT_5777: c240b5c00d58860e376b012cc3c883c17ae63f37 @ git://anongit.freedesktop.org/xorg/app/intel-gpu-tools
Patchwork_18439: d8b76f0b3574014ed467342e9b5aa029e449235e @ git://anongit.freedesktop.org/gfx-ci/linux

== Linux commits ==

d8b76f0b3574 drm/i915: Unlock the shared hwsp_gtt object after pinning

== Logs ==

For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_18439/index.html

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx