Skip to contentSkip to conversation

More updates for MTL workarounds

PATCH In review [email protected] 3 participants 7 messages Feb 29, 2024

lore
Share & tools
lore
#0 lore raw

v2 cover More updates for MTL workarounds

From: John Harrison <[email protected]>

Upate another test to not fail when a MTL workaround is enabled. Also go back and add the w/a number to a previous one.

v2: Fix broken email address in from of first patch (review feedback from Kamil)

John Harrison (2):
tests/intel/gem_spin_batch: Skip compute engines on MTL due to w/a tests/intel/gem_exec_schedule: Add w/a tag to prior update for MTL
RCS/CCS w/a

 tests/intel/gem_exec_schedule.c |  1 +
 tests/intel/gem_spin_batch.c    | 16 ++++++++++++++++
 2 files changed, 17 insertions(+)
Signature · 1 line
2.43.0
#1 lore raw

PATCH v2 1/2 tests/intel/gem_spin_batch: Skip compute engines on MTL due to w/a

From: John Harrison <[email protected]>

There is a w/a on MTL that prevents concurrent execution of multiple processes on RCS and CCS. The 'spin-all-new' test is explicitly testing such execution across all engines. So either the entire test must be skipped on MTL or just the compute engine must be skipped. Do the latter as that at least leaves some test coverage in place.

---
 tests/intel/gem_spin_batch.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
tests/intel/gem_spin_batch.c +16 -0
index ec1458e35197..a0f35b23c718 100644
@@ -214,6 +214,19 @@ spin_on_all_engines(int fd, const intel_ctx_t *ctx,
 	igt_waitchildren();
 }
 
+/*
+ * Wa_14019159160:
+ * An RCS/CCS workaround on MTL means that contexts from different address spaces
+ * cannot run in parallel, they must be timesliced. However, a non-preemptible
+ * spinner cannot be timesliced. Thus the PARALLEL_SPIN_NEW_CTX test cannot be run
+ * across both RCS and CCS engines, one or other must be skipped for the test to
+ * not hit a heartbeat timeout and be killed.
+ */
+static bool skip_bad_engine(int fd, const struct intel_execution_engine2 *e)
+{
+	return IS_METEORLAKE(intel_get_drm_devid(fd)) && (e->class == I915_ENGINE_CLASS_COMPUTE);
+}
+
 static void spin_all(int i915, const intel_ctx_t *ctx, unsigned int flags)
 #define PARALLEL_SPIN_NEW_CTX BIT(0)
 {
@@ -230,6 +243,9 @@ static void spin_all(int i915, const intel_ctx_t *ctx, unsigned int flags)
 		if (gem_engine_can_block_ggtt_binder(i915, e))
 			continue;
 
+		if (skip_bad_engine(i915, e))
+			continue;
+
 		if (flags & PARALLEL_SPIN_NEW_CTX)
 			ctx = intel_ctx_create(i915, &cfg);
 		ahnd = get_reloc_ahnd(i915, ctx->id);
Signature · 1 line
2.43.0
Reviewed-by on v2 1/2
#2 lore raw

Re: [PATCH i-g-t v2 1/2] tests/intel/gem_spin_batch: Skip compute engines on MTL due to w/a

Quoted 38 lines · On 2/29/2024 11:39 AM, [email protected] wrote:

From: John Harrison <[email protected]>

There is a w/a on MTL that prevents concurrent execution of multiple processes on RCS and CCS. The 'spin-all-new' test is explicitly testing such execution across all engines. So either the entire test must be skipped on MTL or just the compute engine must be skipped. Do the latter as that at least leaves some test coverage in place.

---
  tests/intel/gem_spin_batch.c | 16 ++++++++++++++++
  1 file changed, 16 insertions(+)
tests/intel/gem_spin_batch.c +13 -0
index ec1458e35197..a0f35b23c718 100644
@@ -214,6 +214,19 @@ spin_on_all_engines(int fd, const intel_ctx_t *ctx,
  	igt_waitchildren();
  }
 
+/*
+ * Wa_14019159160:
+ * An RCS/CCS workaround on MTL means that contexts from different address spaces
+ * cannot run in parallel, they must be timesliced. However, a non-preemptible
+ * spinner cannot be timesliced. Thus the PARALLEL_SPIN_NEW_CTX test cannot be run
+ * across both RCS and CCS engines, one or other must be skipped for the test to
+ * not hit a heartbeat timeout and be killed.
+ */
+static bool skip_bad_engine(int fd, const struct intel_execution_engine2 *e)
+{
+	return IS_METEORLAKE(intel_get_drm_devid(fd)) && (e->class == I915_ENGINE_CLASS_COMPUTE);
+}
+
  static void spin_all(int i915, const intel_ctx_t *ctx, unsigned int flags)
  #define PARALLEL_SPIN_NEW_CTX BIT(0)
  {

tests/intel/gem_spin_batch.c static void spin_all(int i915, const intel_ctx_t *ctx, unsigned int flags) · line 248

Quoted 6 lines
tests/intel/gem_spin_batch.c +3 -0
@@ -230,6 +243,9 @@ static void spin_all(int i915, const intel_ctx_t *ctx, unsigned int flags)
  		if (gem_engine_can_block_ggtt_binder(i915, e))
  			continue;
 
+		if (skip_bad_engine(i915, e))
+			continue;
+
Quoted 3 lines
tests/intel/gem_spin_batch.c +0 -0
  		if (flags & PARALLEL_SPIN_NEW_CTX)
  			ctx = intel_ctx_create(i915, &cfg);
  		ahnd = get_reloc_ahnd(i915, ctx->id);
#3 lore raw

PATCH v2 2/2 tests/intel/gem_exec_schedule: Add w/a tag to prior update for MTL RCS/CCS w/a

In reply to [email protected]

From: John Harrison <[email protected]>

This test was recently updated to not fall foul of a particular workaround. The w/a tag itself was missed, however. So, add that in to make it easier to find in the future.

---
 tests/intel/gem_exec_schedule.c | 1 +
 1 file changed, 1 insertion(+)
tests/intel/gem_exec_schedule.c +1 -0
index 3fd91b2fae80..8612929cc403 100644
@@ -444,6 +444,7 @@
 IGT_TEST_DESCRIPTION("Check that we can control the order of execution");
 
 /*
+ * Wa_16019325821
  * The RCS/CCS workaround on MTL means that if spinners are sent to both RCS
  * and CCS engines concurrently then the system is effectively dead and will
  * require a reset of both engines to recover. Even pre-emptible spinners
Signature · 1 line
2.43.0
Reviewed-by on v2 2/2
#4 lore raw

Re: [PATCH i-g-t v2 2/2] tests/intel/gem_exec_schedule: Add w/a tag to prior update for MTL RCS/CCS w/a

Quoted 17 lines · On 2/29/2024 11:39 AM, [email protected] wrote:

From: John Harrison <[email protected]>

This test was recently updated to not fall foul of a particular workaround. The w/a tag itself was missed, however. So, add that in to make it easier to find in the future.

---
  tests/intel/gem_exec_schedule.c | 1 +
  1 file changed, 1 insertion(+)
tests/intel/gem_exec_schedule.c +0 -0
index 3fd91b2fae80..8612929cc403 100644

tests/intel/gem_exec_schedule.c line 447

Quoted 4 lines
tests/intel/gem_exec_schedule.c +1 -0
@@ -444,6 +444,7 @@
  IGT_TEST_DESCRIPTION("Check that we can control the order of execution");
 
  /*
+ * Wa_16019325821
Quoted 3 lines
tests/intel/gem_exec_schedule.c +0 -0
   * The RCS/CCS workaround on MTL means that if spinners are sent to both RCS
   * and CCS engines concurrently then the system is effectively dead and will
   * require a reset of both engines to recover. Even pre-emptible spinners
#5 lore raw

✓ CI.xeBAT: success for More updates for MTL workarounds (rev2)

In reply to [email protected]
== Series Details ==

== Series Details ==

Series: More updates for MTL workarounds (rev2)
URL : https://patchwork.freedesktop.org/series/130416/
State : success

== Summary ==

CI Bug Log - changes from XEIGT_7738_BAT -> XEIGTPW_10759_BAT
====================================================

Summary
-------

**SUCCESS**

No regressions found.

Participating hosts (4 -> 4)
------------------------------

No changes in participating hosts

Changes
-------

No changes found

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

* IGT: IGT_7738 -> IGTPW_10759
* Linux: xe-876-12db124b8cef9c32b7bc69a5abbaff016965cf99 -> xe-879-879fd254f581ab57486aae46f4a027e92642feb6

IGTPW_10759: 10759
IGT_7738: 7738
xe-876-12db124b8cef9c32b7bc69a5abbaff016965cf99: 12db124b8cef9c32b7bc69a5abbaff016965cf99 xe-879-879fd254f581ab57486aae46f4a027e92642feb6: 879fd254f581ab57486aae46f4a027e92642feb6

== Logs ==

For more details see: https://intel-gfx-ci.01.org/tree/intel-xe/IGTPW_10759/index.html

#6 lore raw

✗ Fi.CI.BAT: failure for More updates for MTL workarounds (rev2)

In reply to [email protected]
== Series Details ==

== Series Details ==

Series: More updates for MTL workarounds (rev2)
URL : https://patchwork.freedesktop.org/series/130416/
State : failure

== Summary ==

CI Bug Log - changes from CI_DRM_14369 -> IGTPW_10759
====================================================

Summary
-------

**FAILURE**

Serious unknown changes coming with IGTPW_10759 absolutely need to be verified manually.

If you think the reported changes have nothing to do with the changes
introduced in IGTPW_10759, please notify your bug team ([email protected]) to allow them to document this new failure mode, which will reduce false positives in CI.

External URL: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_10759/index.html

Participating hosts (42 -> 40)
------------------------------

Missing (2): fi-snb-2520m fi-bsw-n3050

Possible new issues
-------------------

Here are the unknown changes that may have been introduced in IGTPW_10759:

### IGT changes ###

#### Possible regressions ####

* igt@kms_pipe_crc_basic@read-crc-frame-sequence@pipe-a-edp-1:
- bat-arls-2: [PASS][1] -> [ABORT][2]
[1]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_14369/bat-arls-2/igt@kms_pipe_crc_basic@[email protected]
[2]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_10759/bat-arls-2/igt@kms_pipe_crc_basic@[email protected]

#### Suppressed ####

The following results come from untrusted machines, tests, or statuses. They do not affect the overall result.

* igt@kms_pipe_crc_basic@hang-read-crc@pipe-d-dp-2:
- {bat-rpls-3}: [PASS][3] -> [FAIL][4]
[3]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_14369/bat-rpls-3/igt@kms_pipe_crc_basic@[email protected]
[4]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_10759/bat-rpls-3/igt@kms_pipe_crc_basic@[email protected]

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

Here are the changes found in IGTPW_10759 that come from known issues:

### IGT changes ###

#### Issues hit ####

* igt@gem_lmem_swapping@basic:
- fi-apl-guc: NOTRUN -> [SKIP][5] ([fdo#109271] / [i915#4613]) +3 other tests skip
[5]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_10759/fi-apl-guc/igt@[email protected]

* igt@i915_pm_rpm@module-reload:
- fi-kbl-7567u: [PASS][6] -> [CRASH][7] ([i915#9947])
[6]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_14369/fi-kbl-7567u/igt@[email protected]
[7]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_10759/fi-kbl-7567u/igt@[email protected]

* igt@kms_hdmi_inject@inject-audio:
- fi-apl-guc: NOTRUN -> [SKIP][8] ([fdo#109271]) +17 other tests skip
[8]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_10759/fi-apl-guc/igt@[email protected]

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

[fdo#109271]: https://bugs.freedesktop.org/show_bug.cgi?id=109271
[i915#10262]: https://gitlab.freedesktop.org/drm/intel/issues/10262
[i915#4613]: https://gitlab.freedesktop.org/drm/intel/issues/4613
[i915#9947]: https://gitlab.freedesktop.org/drm/intel/issues/9947
[i915#9980]: https://gitlab.freedesktop.org/drm/intel/issues/9980

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

* CI: CI-20190529 -> None
* IGT: IGT_7738 -> IGTPW_10759

CI-20190529: 20190529
CI_DRM_14369: 879fd254f581ab57486aae46f4a027e92642feb6 @ git://anongit.freedesktop.org/gfx-ci/linux
IGTPW_10759: 10759
IGT_7738: 7738

Testlist changes
----------------

+igt@kms_big_joiner@2x-modeset
-igt@kms_big_joiner@force-joiner-basic
-igt@kms_big_joiner@force-joiner-invalid-modeset

== Logs ==

For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_10759/index.html