tests/perf_pmu: Compare against requested freq in frequency subtest
PATCH 1/1 tests/perf_pmu: Compare against requested freq in frequency subtest
With SLPC, even when we set the same min and max freq's, the requested and actual freq's can differ from the min/max freq set. For example "efficient freq" (when in effect) can override set min/max freq. In general FW is the final arbiter in determining freq and can override set values.
Therefore in igt@perf_pmu@frequency subtest, compare the requested freq reported by PMU not against the set freq's but against the requested freq reported in sysfs. Also add a delay after setting freq's to account for messaging delays in setting freq's in GuC.
v2: Introduce a 100 ms delay after setting freq
v3: Update commit message, code identical to v2
--- tests/i915/perf_pmu.c | 23 +++++++++++++++-------- 1 file changed, 15 insertions(+), 8 deletions(-)
tests/i915/perf_pmu.c +15 -8
@@ -1543,10 +1543,13 @@ test_interrupts_sync(int gem_fd)igt_assert_lte (target, busy); } +/* Wait for GuC SLPC freq changes to take effect */ +#define wait_freq_set () usleep(100000) +static void test_frequency (int gem_fd) { -uint32_t min_freq, max_freq, boost_freq; +uint32_t min_freq, max_freq, boost_freq, min_req, max_req;uint64_t val[2 ], start[2 ], slept;double min[2 ], max[2 ];igt_spin_t *spin;
@@ -1572,10 +1575,11 @@ test_frequency(int gem_fd)* Set GPU to min frequency and read PMU counters. */ igt_require (igt_sysfs_set_u32 (sysfs,"gt_min_freq_mhz" , min_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) == min_freq);igt_require (igt_sysfs_set_u32 (sysfs,"gt_max_freq_mhz" , min_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == min_freq);igt_require (igt_sysfs_set_u32 (sysfs,"gt_boost_freq_mhz" , min_freq)); +wait_freq_set (); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) == min_freq); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == min_freq);igt_require (igt_sysfs_get_u32 (sysfs,"gt_boost_freq_mhz" ) == min_freq);gem_quiescent_gpu (gem_fd);/* Idle to be sure the change takes effect */
@@ -1587,6 +1591,7 @@ test_frequency(int gem_fd) min[0 ] =1e9 *(val[0 ] - start[0 ]) / slept; min[1 ] =1e9 *(val[1 ] - start[1 ]) / slept; + min_req =igt_sysfs_get_u32 (sysfs,"gt_cur_freq_mhz" );igt_spin_free (gem_fd, spin);gem_quiescent_gpu (gem_fd);/* Don't leak busy bo into the next phase */
@@ -1597,11 +1602,11 @@ test_frequency(int gem_fd)* Set GPU to max frequency and read PMU counters. */ igt_require (igt_sysfs_set_u32 (sysfs,"gt_max_freq_mhz" , max_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == max_freq);igt_require (igt_sysfs_set_u32 (sysfs,"gt_boost_freq_mhz" , boost_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_boost_freq_mhz" ) == boost_freq); -igt_require (igt_sysfs_set_u32 (sysfs,"gt_min_freq_mhz" , max_freq)); +wait_freq_set (); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == max_freq); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_boost_freq_mhz" ) == boost_freq);igt_require (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) == max_freq);gem_quiescent_gpu (gem_fd);
@@ -1613,6 +1618,7 @@ test_frequency(int gem_fd) max[0 ] =1e9 *(val[0 ] - start[0 ]) / slept; max[1 ] =1e9 *(val[1 ] - start[1 ]) / slept; + max_req =igt_sysfs_get_u32 (sysfs,"gt_cur_freq_mhz" );igt_spin_free (gem_fd, spin);gem_quiescent_gpu (gem_fd);
@@ -1621,6 +1627,7 @@ test_frequency(int gem_fd)* Restore min/max. */ igt_sysfs_set_u32 (sysfs,"gt_min_freq_mhz" , min_freq); +wait_freq_set ();if (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) != min_freq)igt_warn ("Unable to restore min frequency to saved value [%u MHz], now %u MHz\n" , min_freq,igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ));
@@ -1633,12 +1640,12 @@ test_frequency(int gem_fd)igt_info ("Max frequency: requested %.1f, actual %.1f\n" , max[0 ], max[1 ]); -assert_within_epsilon (min[0 ], min_freq, tolerance); +assert_within_epsilon (min[0 ], min_req, tolerance);/* * On thermally throttled devices we cannot be sure maximum frequency * can be reached so use larger tolerance downards. */ -__assert_within_epsilon (max[0 ], max_freq, tolerance,0.15f ); +__assert_within_epsilon (max[0 ], max_req, tolerance,0.15f ); } static void
Signature · 1 line
2.38.1
Patchwork
CI[igt-dev] ✓ Fi.CI.BAT: success for tests/perf_pmu: Compare against requested freq in frequency subtest (rev3)
== Series Details ==
== Series Details ==
Series: tests/perf_pmu: Compare against requested freq in frequency subtest (rev3)
URL : https://patchwork.freedesktop.org/series/110574/
State : success
== Summary ==
CI Bug Log - changes from CI_DRM_12403 -> IGTPW_8132
====================================================
Summary
-------
**SUCCESS**
No regressions found.
External URL: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/index.html
Participating hosts (43 -> 29)
------------------------------
Additional (2): fi-kbl-soraka fi-hsw-4770
Missing (16): bat-dg1-7 fi-rkl-11600 fi-ilk-m540 fi-bdw-samus bat-dg1-6 bat-dg2-8 bat-adlm-1 bat-dg2-9 bat-adlp-6 bat-adlp-4 fi-ctg-p8600 bat-adln-1 bat-rplp-1 bat-rpls-2 bat-dg2-11 bat-jsl-1
Known issues
------------
Here are the changes found in IGTPW_8132 that come from known issues:
### IGT changes ###
#### Issues hit ####
* igt@gem_exec_gttfill@basic:
- fi-kbl-soraka: NOTRUN -> [SKIP][1] ([fdo#109271]) +9 similar issues
[1]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/fi-kbl-soraka/igt@[email protected]
- fi-pnv-d510: [PASS][2] -> [FAIL][3] ([i915#7229])
[2]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/fi-pnv-d510/igt@[email protected]
[3]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/fi-pnv-d510/igt@[email protected]
* igt@gem_huc_copy@huc-copy:
- fi-kbl-soraka: NOTRUN -> [SKIP][4] ([fdo#109271] / [i915#2190])
[4]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/fi-kbl-soraka/igt@[email protected]
* igt@gem_lmem_swapping@basic:
- fi-kbl-soraka: NOTRUN -> [SKIP][5] ([fdo#109271] / [i915#4613]) +3 similar issues
[5]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/fi-kbl-soraka/igt@[email protected]
* igt@i915_selftest@live@gem_contexts:
- fi-kbl-soraka: NOTRUN -> [INCOMPLETE][6] ([i915#7099])
[6]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/fi-kbl-soraka/igt@i915_selftest@live@gem_contexts.html
* igt@i915_selftest@live@gt_pm:
- fi-kbl-soraka: NOTRUN -> [DMESG-FAIL][7] ([i915#1886])
[7]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/fi-kbl-soraka/igt@i915_selftest@live@gt_pm.html
* igt@kms_addfb_basic@addfb25-y-tiled-small-legacy:
- fi-hsw-4770: NOTRUN -> [SKIP][8] ([fdo#109271]) +11 similar issues
[8]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/fi-hsw-4770/igt@[email protected]
* igt@kms_chamelium@common-hpd-after-suspend:
- fi-kbl-7567u: NOTRUN -> [SKIP][9] ([fdo#109271] / [fdo#111827])
[9]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/fi-kbl-7567u/igt@[email protected]
* igt@kms_chamelium@dp-crc-fast:
- fi-hsw-4770: NOTRUN -> [SKIP][10] ([fdo#109271] / [fdo#111827]) +8 similar issues
[10]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/fi-hsw-4770/igt@[email protected]
* igt@kms_chamelium@hdmi-hpd-fast:
- fi-kbl-soraka: NOTRUN -> [SKIP][11] ([fdo#109271] / [fdo#111827]) +7 similar issues
[11]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/fi-kbl-soraka/igt@[email protected]
* igt@kms_cursor_legacy@basic-busy-flip-before-cursor@atomic-transitions:
- fi-bsw-kefka: [PASS][12] -> [FAIL][13] ([i915#6298])
[12]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/fi-bsw-kefka/igt@kms_cursor_legacy@[email protected]
[13]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/fi-bsw-kefka/igt@kms_cursor_legacy@[email protected]
* igt@kms_pipe_crc_basic@suspend-read-crc:
- fi-kbl-7567u: NOTRUN -> [SKIP][14] ([fdo#109271])
[14]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/fi-kbl-7567u/igt@[email protected]
* igt@kms_psr@sprite_plane_onoff:
- fi-hsw-4770: NOTRUN -> [SKIP][15] ([fdo#109271] / [i915#1072]) +3 similar issues
[15]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/fi-hsw-4770/igt@kms_psr@sprite_plane_onoff.html
* igt@runner@aborted:
- fi-snb-2520m: NOTRUN -> [FAIL][16] ([i915#4312])
[16]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/fi-snb-2520m/igt@[email protected]
#### Possible fixes ####
* igt@i915_selftest@live@hangcheck:
- {fi-ehl-2}: [INCOMPLETE][17] ([i915#5153] / [i915#6106] / [i915#7351]) -> [PASS][18]
[17]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/fi-ehl-2/igt@i915_selftest@[email protected]
[18]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/fi-ehl-2/igt@i915_selftest@[email protected]
* igt@i915_suspend@basic-s3-without-i915:
- fi-kbl-7567u: [INCOMPLETE][19] ([i915#4817]) -> [PASS][20]
[19]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/fi-kbl-7567u/igt@[email protected]
[20]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/fi-kbl-7567u/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
[fdo#111827]: https://bugs.freedesktop.org/show_bug.cgi?id=111827
[i915#1072]: https://gitlab.freedesktop.org/drm/intel/issues/1072
[i915#1886]: https://gitlab.freedesktop.org/drm/intel/issues/1886
[i915#2190]: https://gitlab.freedesktop.org/drm/intel/issues/2190
[i915#4312]: https://gitlab.freedesktop.org/drm/intel/issues/4312
[i915#4613]: https://gitlab.freedesktop.org/drm/intel/issues/4613
[i915#4817]: https://gitlab.freedesktop.org/drm/intel/issues/4817
[i915#5153]: https://gitlab.freedesktop.org/drm/intel/issues/5153
[i915#6106]: https://gitlab.freedesktop.org/drm/intel/issues/6106
[i915#6298]: https://gitlab.freedesktop.org/drm/intel/issues/6298
[i915#7099]: https://gitlab.freedesktop.org/drm/intel/issues/7099
[i915#7229]: https://gitlab.freedesktop.org/drm/intel/issues/7229
[i915#7351]: https://gitlab.freedesktop.org/drm/intel/issues/7351
Build changes
-------------
* CI: CI-20190529 -> None
* IGT: IGT_7068 -> IGTPW_8132
CI-20190529: 20190529
CI_DRM_12403: 9c91ccfa558e7510add39958a56745ae7fa07753 @ git://anongit.freedesktop.org/gfx-ci/linux
IGTPW_8132: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/index.html
IGT_7068: 5c0ec905b6bbecfb8df8b8f3315d0470539e6ae3 @ https://gitlab.freedesktop.org/drm/igt-gpu-tools.git
== Logs ==
For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/index.html
Patchwork
CI[igt-dev] ✓ Fi.CI.IGT: success for tests/perf_pmu: Compare against requested freq in frequency subtest (rev3)
In reply to Ashutosh Dixit== Series Details ==
== Series Details ==
Series: tests/perf_pmu: Compare against requested freq in frequency subtest (rev3)
URL : https://patchwork.freedesktop.org/series/110574/
State : success
== Summary ==
CI Bug Log - changes from CI_DRM_12403_full -> IGTPW_8132_full
====================================================
Summary
-------
**SUCCESS**
No regressions found.
External URL: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/index.html
Participating hosts (11 -> 8)
------------------------------
Missing (3): pig-skl-6260u pig-kbl-iris pig-glk-j5005
Possible new issues
-------------------
Here are the unknown changes that may have been introduced in IGTPW_8132_full:
### IGT changes ###
#### Suppressed ####
The following results come from untrusted machines, tests, or statuses. They do not affect the overall result.
* igt@i915_pm_backlight@fade:
- {shard-dg1}: NOTRUN -> [SKIP][1]
[1]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-dg1-17/igt@[email protected]
* igt@i915_pm_backlight@fade-with-suspend:
- {shard-rkl}: NOTRUN -> [SKIP][2]
[2]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-rkl-5/igt@[email protected]
* igt@i915_pm_rpm@system-suspend-modeset:
- {shard-rkl}: [SKIP][3] ([fdo#109308]) -> [INCOMPLETE][4]
[3]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-rkl-4/igt@[email protected]
[4]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-rkl-3/igt@[email protected]
Known issues
------------
Here are the changes found in IGTPW_8132_full that come from known issues:
### IGT changes ###
#### Issues hit ####
* igt@api_intel_bb@crc32:
- shard-iclb: NOTRUN -> [SKIP][5] ([i915#6230])
[5]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb8/igt@[email protected]
* igt@feature_discovery@display-3x:
- shard-iclb: NOTRUN -> [SKIP][6] ([i915#1839])
[6]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb5/igt@[email protected]
* igt@gem_ccs@block-copy-inplace:
- shard-tglb: NOTRUN -> [SKIP][7] ([i915#3555] / [i915#5325])
[7]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb8/igt@[email protected]
* igt@gem_ctx_persistence@hostile:
- shard-snb: NOTRUN -> [SKIP][8] ([fdo#109271] / [i915#1099])
[8]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-snb2/igt@[email protected]
* igt@gem_ctx_persistence@many-contexts:
- shard-tglb: NOTRUN -> [FAIL][9] ([i915#2410])
[9]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb3/igt@[email protected]
* igt@gem_exec_balancer@parallel-bb-first:
- shard-iclb: [PASS][10] -> [SKIP][11] ([i915#4525])
[10]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-iclb2/igt@[email protected]
[11]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb3/igt@[email protected]
* igt@gem_exec_capture@capture-invisible@smem0:
- shard-iclb: NOTRUN -> [SKIP][12] ([i915#6334])
[12]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb5/igt@gem_exec_capture@[email protected]
- shard-tglb: NOTRUN -> [SKIP][13] ([i915#6334])
[13]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb6/igt@gem_exec_capture@[email protected]
* igt@gem_exec_fair@basic-deadline:
- shard-tglb: [PASS][14] -> [FAIL][15] ([i915#2846])
[14]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-tglb3/igt@[email protected]
[15]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb5/igt@[email protected]
* igt@gem_exec_fair@basic-pace@vcs0:
- shard-glk: NOTRUN -> [FAIL][16] ([i915#2842]) +4 similar issues
[16]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-glk6/igt@gem_exec_fair@[email protected]
* igt@gem_lmem_swapping@parallel-random:
- shard-tglb: NOTRUN -> [SKIP][17] ([i915#4613]) +1 similar issue
[17]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb7/igt@[email protected]
* igt@gem_lmem_swapping@parallel-random-engines:
- shard-iclb: NOTRUN -> [SKIP][18] ([i915#4613]) +1 similar issue
[18]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb7/igt@[email protected]
* igt@gem_lmem_swapping@smem-oom:
- shard-apl: NOTRUN -> [SKIP][19] ([fdo#109271] / [i915#4613]) +2 similar issues
[19]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-apl1/igt@[email protected]
* igt@gem_lmem_swapping@verify-random-ccs:
- shard-glk: NOTRUN -> [SKIP][20] ([fdo#109271] / [i915#4613]) +1 similar issue
[20]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-glk2/igt@[email protected]
* igt@gem_pxp@create-protected-buffer:
- shard-tglb: NOTRUN -> [SKIP][21] ([i915#4270])
[21]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb7/igt@[email protected]
- shard-iclb: NOTRUN -> [SKIP][22] ([i915#4270])
[22]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb7/igt@[email protected]
* igt@gem_render_copy@x-tiled-to-vebox-yf-tiled:
- shard-apl: NOTRUN -> [SKIP][23] ([fdo#109271]) +124 similar issues
[23]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-apl8/igt@[email protected]
* igt@gem_render_copy@yf-tiled-to-vebox-yf-tiled:
- shard-iclb: NOTRUN -> [SKIP][24] ([i915#768])
[24]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb2/igt@[email protected]
* igt@gem_userptr_blits@unsync-unmap-after-close:
- shard-tglb: NOTRUN -> [SKIP][25] ([i915#3297])
[25]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb6/igt@[email protected]
- shard-iclb: NOTRUN -> [SKIP][26] ([i915#3297])
[26]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb1/igt@[email protected]
* igt@gem_userptr_blits@vma-merge:
- shard-apl: NOTRUN -> [FAIL][27] ([i915#3318])
[27]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-apl8/igt@[email protected]
* igt@gen3_mixed_blits:
- shard-tglb: NOTRUN -> [SKIP][28] ([fdo#109289])
[28]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb8/igt@gen3_mixed_blits.html
* igt@gen9_exec_parse@allowed-all:
- shard-glk: [PASS][29] -> [DMESG-WARN][30] ([i915#5566] / [i915#716])
[29]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-glk9/igt@[email protected]
[30]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-glk5/igt@[email protected]
* igt@gen9_exec_parse@allowed-single:
- shard-apl: [PASS][31] -> [DMESG-WARN][32] ([i915#5566] / [i915#716])
[31]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-apl1/igt@[email protected]
[32]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-apl6/igt@[email protected]
* igt@gen9_exec_parse@cmd-crossing-page:
- shard-iclb: NOTRUN -> [SKIP][33] ([i915#2856]) +1 similar issue
[33]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb2/igt@[email protected]
* igt@gen9_exec_parse@shadow-peek:
- shard-tglb: NOTRUN -> [SKIP][34] ([i915#2527] / [i915#2856]) +1 similar issue
[34]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb6/igt@[email protected]
* igt@i915_pm_dc@dc9-dpms:
- shard-apl: [PASS][35] -> [SKIP][36] ([fdo#109271])
[35]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-apl3/igt@[email protected]
[36]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-apl3/igt@[email protected]
* igt@i915_pm_rpm@modeset-pc8-residency-stress:
- shard-iclb: NOTRUN -> [SKIP][37] ([fdo#109293] / [fdo#109506])
[37]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb5/igt@[email protected]
* igt@i915_pm_rpm@system-suspend-devices:
- shard-snb: NOTRUN -> [SKIP][38] ([fdo#109271]) +123 similar issues
[38]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-snb5/igt@[email protected]
* igt@i915_pm_rps@engine-order:
- shard-apl: NOTRUN -> [FAIL][39] ([i915#6537])
[39]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-apl7/igt@[email protected]
* igt@i915_query@test-query-geometry-subslices:
- shard-tglb: NOTRUN -> [SKIP][40] ([i915#5723])
[40]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb3/igt@[email protected]
* igt@kms_big_fb@4-tiled-max-hw-stride-64bpp-rotate-0:
- shard-tglb: NOTRUN -> [SKIP][41] ([i915#5286]) +2 similar issues
[41]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb5/igt@[email protected]
- shard-iclb: NOTRUN -> [SKIP][42] ([i915#5286]) +4 similar issues
[42]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb6/igt@[email protected]
* igt@kms_big_fb@y-tiled-64bpp-rotate-270:
- shard-iclb: NOTRUN -> [SKIP][43] ([fdo#110725] / [fdo#111614])
[43]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb2/igt@[email protected]
* igt@kms_big_fb@yf-tiled-max-hw-stride-32bpp-rotate-0-async-flip:
- shard-tglb: NOTRUN -> [SKIP][44] ([fdo#111615]) +4 similar issues
[44]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb3/igt@[email protected]
* igt@kms_big_fb@yf-tiled-max-hw-stride-64bpp-rotate-0-hflip-async-flip:
- shard-iclb: NOTRUN -> [SKIP][45] ([fdo#110723])
[45]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb8/igt@kms_big_fb@yf-tiled-max-hw-stride-64bpp-rotate-0-hflip-async-flip.html
* igt@kms_ccs@pipe-a-bad-pixel-format-4_tiled_dg2_mc_ccs:
- shard-tglb: NOTRUN -> [SKIP][46] ([i915#6095]) +5 similar issues
[46]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb3/igt@kms_ccs@pipe-a-bad-pixel-format-4_tiled_dg2_mc_ccs.html
* igt@kms_ccs@pipe-a-bad-rotation-90-y_tiled_gen12_mc_ccs:
- shard-iclb: NOTRUN -> [SKIP][47] ([fdo#109278] / [i915#3886]) +2 similar issues
[47]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb1/igt@kms_ccs@pipe-a-bad-rotation-90-y_tiled_gen12_mc_ccs.html
- shard-tglb: NOTRUN -> [SKIP][48] ([i915#3689] / [i915#3886])
[48]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb3/igt@kms_ccs@pipe-a-bad-rotation-90-y_tiled_gen12_mc_ccs.html
* igt@kms_ccs@pipe-a-crc-primary-rotation-180-4_tiled_dg2_rc_ccs:
- shard-tglb: NOTRUN -> [SKIP][49] ([i915#3689] / [i915#6095]) +1 similar issue
[49]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb3/igt@kms_ccs@pipe-a-crc-primary-rotation-180-4_tiled_dg2_rc_ccs.html
* igt@kms_ccs@pipe-a-crc-sprite-planes-basic-y_tiled_gen12_rc_ccs:
- shard-iclb: NOTRUN -> [SKIP][50] ([fdo#109278]) +12 similar issues
[50]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb3/igt@kms_ccs@pipe-a-crc-sprite-planes-basic-y_tiled_gen12_rc_ccs.html
* igt@kms_ccs@pipe-b-missing-ccs-buffer-y_tiled_gen12_rc_ccs_cc:
- shard-apl: NOTRUN -> [SKIP][51] ([fdo#109271] / [i915#3886]) +2 similar issues
[51]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-apl1/igt@kms_ccs@pipe-b-missing-ccs-buffer-y_tiled_gen12_rc_ccs_cc.html
* igt@kms_ccs@pipe-b-random-ccs-data-y_tiled_gen12_rc_ccs_cc:
- shard-glk: NOTRUN -> [SKIP][52] ([fdo#109271] / [i915#3886]) +2 similar issues
[52]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-glk6/igt@kms_ccs@pipe-b-random-ccs-data-y_tiled_gen12_rc_ccs_cc.html
* igt@kms_ccs@pipe-d-crc-sprite-planes-basic-yf_tiled_ccs:
- shard-tglb: NOTRUN -> [SKIP][53] ([fdo#111615] / [i915#3689]) +1 similar issue
[53]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb2/igt@kms_ccs@pipe-d-crc-sprite-planes-basic-yf_tiled_ccs.html
* igt@kms_ccs@pipe-d-random-ccs-data-y_tiled_gen12_mc_ccs:
- shard-tglb: NOTRUN -> [SKIP][54] ([i915#3689]) +3 similar issues
[54]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb1/igt@kms_ccs@pipe-d-random-ccs-data-y_tiled_gen12_mc_ccs.html
* igt@kms_chamelium@common-hpd-after-suspend:
- shard-apl: NOTRUN -> [SKIP][55] ([fdo#109271] / [fdo#111827]) +5 similar issues
[55]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-apl8/igt@[email protected]
* igt@kms_chamelium@dp-hpd-enable-disable-mode:
- shard-glk: NOTRUN -> [SKIP][56] ([fdo#109271] / [fdo#111827]) +6 similar issues
[56]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-glk8/igt@[email protected]
* igt@kms_chamelium@hdmi-hpd-storm-disable:
- shard-tglb: NOTRUN -> [SKIP][57] ([fdo#109284] / [fdo#111827]) +2 similar issues
[57]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb3/igt@[email protected]
- shard-iclb: NOTRUN -> [SKIP][58] ([fdo#109284] / [fdo#111827]) +4 similar issues
[58]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb7/igt@[email protected]
* igt@kms_color_chamelium@ctm-0-50:
- shard-snb: NOTRUN -> [SKIP][59] ([fdo#109271] / [fdo#111827]) +2 similar issues
[59]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-snb7/igt@[email protected]
* igt@kms_content_protection@content_type_change:
- shard-iclb: NOTRUN -> [SKIP][60] ([i915#7118])
[60]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb1/igt@kms_content_protection@content_type_change.html
* igt@kms_content_protection@dp-mst-type-0:
- shard-tglb: NOTRUN -> [SKIP][61] ([i915#3116] / [i915#3299])
[61]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb2/igt@[email protected]
- shard-iclb: NOTRUN -> [SKIP][62] ([i915#3116])
[62]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb3/igt@[email protected]
* igt@kms_content_protection@lic@pipe-a-dp-1:
- shard-apl: NOTRUN -> [TIMEOUT][63] ([i915#7173])
[63]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-apl2/igt@kms_content_protection@[email protected]
* igt@kms_content_protection@srm:
- shard-tglb: NOTRUN -> [SKIP][64] ([i915#7118])
[64]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb1/igt@[email protected]
* igt@kms_cursor_legacy@cursora-vs-flipb@atomic:
- shard-iclb: NOTRUN -> [SKIP][65] ([fdo#109274]) +10 similar issues
[65]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb8/igt@kms_cursor_legacy@[email protected]
* igt@kms_cursor_legacy@cursora-vs-flipb@varying-size:
- shard-tglb: NOTRUN -> [SKIP][66] ([fdo#109274] / [fdo#111825]) +5 similar issues
[66]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb5/igt@kms_cursor_legacy@[email protected]
* igt@kms_display_modes@extended-mode-basic:
- shard-tglb: NOTRUN -> [SKIP][67] ([fdo#109274])
[67]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb5/igt@[email protected]
* igt@kms_dsc@dsc-with-bpc:
- shard-iclb: NOTRUN -> [SKIP][68] ([i915#3840])
[68]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb3/igt@[email protected]
* igt@kms_dsc@dsc-with-bpc-formats:
- shard-glk: NOTRUN -> [SKIP][69] ([fdo#109271] / [i915#7205])
[69]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-glk8/igt@[email protected]
* igt@kms_flip@2x-plain-flip-fb-recreate:
- shard-tglb: NOTRUN -> [SKIP][70] ([fdo#109274] / [fdo#111825] / [i915#3637]) +6 similar issues
[70]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb7/igt@[email protected]
* igt@kms_flip@2x-plain-flip-fb-recreate@ac-hdmi-a1-hdmi-a2:
- shard-glk: NOTRUN -> [FAIL][71] ([i915#2122]) +1 similar issue
[71]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-glk3/igt@kms_flip@[email protected]
* igt@kms_flip_scaled_crc@flip-32bpp-ytile-to-32bpp-ytileccs-upscaling@pipe-a-valid-mode:
- shard-tglb: NOTRUN -> [SKIP][72] ([i915#2587] / [i915#2672]) +1 similar issue
[72]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb5/igt@kms_flip_scaled_crc@flip-32bpp-ytile-to-32bpp-ytileccs-upscaling@pipe-a-valid-mode.html
* igt@kms_flip_scaled_crc@flip-64bpp-4tile-to-32bpp-4tiledg2rcccs-downscaling@pipe-a-valid-mode:
- shard-iclb: NOTRUN -> [SKIP][73] ([i915#2587] / [i915#2672]) +5 similar issues
[73]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb6/igt@kms_flip_scaled_crc@flip-64bpp-4tile-to-32bpp-4tiledg2rcccs-downscaling@pipe-a-valid-mode.html
* igt@kms_flip_scaled_crc@flip-64bpp-4tile-to-32bpp-4tiledg2rcccs-upscaling@pipe-a-default-mode:
- shard-iclb: NOTRUN -> [SKIP][74] ([i915#2672]) +6 similar issues
[74]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb2/igt@kms_flip_scaled_crc@flip-64bpp-4tile-to-32bpp-4tiledg2rcccs-upscaling@pipe-a-default-mode.html
* igt@kms_flip_scaled_crc@flip-64bpp-yftile-to-16bpp-yftile-upscaling@pipe-a-valid-mode:
- shard-glk: NOTRUN -> [SKIP][75] ([fdo#109271]) +145 similar issues
[75]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-glk6/igt@kms_flip_scaled_crc@flip-64bpp-yftile-to-16bpp-yftile-upscaling@pipe-a-valid-mode.html
* igt@kms_frontbuffer_tracking@fbcpsr-1p-primscrn-cur-indfb-draw-blt:
- shard-tglb: NOTRUN -> [SKIP][76] ([i915#6497]) +5 similar issues
[76]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb5/igt@kms_frontbuffer_tracking@fbcpsr-1p-primscrn-cur-indfb-draw-blt.html
* igt@kms_frontbuffer_tracking@fbcpsr-2p-scndscrn-indfb-pgflip-blt:
- shard-iclb: NOTRUN -> [SKIP][77] ([fdo#109280]) +19 similar issues
[77]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb7/igt@kms_frontbuffer_tracking@fbcpsr-2p-scndscrn-indfb-pgflip-blt.html
* igt@kms_frontbuffer_tracking@fbcpsr-tiling-4:
- shard-iclb: NOTRUN -> [SKIP][78] ([i915#5438])
[78]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb5/igt@[email protected]
* igt@kms_frontbuffer_tracking@psr-2p-scndscrn-cur-indfb-draw-pwrite:
- shard-tglb: NOTRUN -> [SKIP][79] ([fdo#109280] / [fdo#111825]) +18 similar issues
[79]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb1/igt@kms_frontbuffer_tracking@psr-2p-scndscrn-cur-indfb-draw-pwrite.html
* igt@kms_hdr@bpc-switch-dpms:
- shard-tglb: NOTRUN -> [SKIP][80] ([i915#3555]) +3 similar issues
[80]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb6/igt@[email protected]
* igt@kms_hdr@static-toggle-suspend:
- shard-iclb: NOTRUN -> [SKIP][81] ([i915#3555]) +4 similar issues
[81]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb1/igt@[email protected]
* igt@kms_multipipe_modeset@basic-max-pipe-crc-check:
- shard-tglb: NOTRUN -> [SKIP][82] ([i915#1839])
[82]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb3/igt@[email protected]
* igt@kms_plane_alpha_blend@alpha-transparent-fb@pipe-b-hdmi-a-2:
- shard-glk: NOTRUN -> [FAIL][83] ([i915#4573]) +2 similar issues
[83]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-glk9/igt@kms_plane_alpha_blend@[email protected]
* igt@kms_plane_lowres@tiling-none@pipe-c-edp-1:
- shard-tglb: NOTRUN -> [SKIP][84] ([i915#3536]) +3 similar issues
[84]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb1/igt@kms_plane_lowres@[email protected]
* igt@kms_plane_lowres@tiling-yf:
- shard-tglb: NOTRUN -> [SKIP][85] ([fdo#112054] / [i915#5288])
[85]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb3/igt@[email protected]
* igt@kms_plane_lowres@tiling-yf@pipe-b-edp-1:
- shard-iclb: NOTRUN -> [SKIP][86] ([i915#3536]) +5 similar issues
[86]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb1/igt@kms_plane_lowres@[email protected]
* igt@kms_plane_scaling@plane-scaler-with-rotation-unity-scaling@pipe-a-edp-1:
- shard-iclb: NOTRUN -> [SKIP][87] ([i915#5176]) +2 similar issues
[87]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb3/igt@kms_plane_scaling@[email protected]
* igt@kms_plane_scaling@plane-scaler-with-rotation-unity-scaling@pipe-b-edp-1:
- shard-tglb: NOTRUN -> [SKIP][88] ([i915#5176]) +7 similar issues
[88]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb8/igt@kms_plane_scaling@[email protected]
* igt@kms_plane_scaling@planes-upscale-factor-0-25-downscale-factor-0-25@pipe-b-edp-1:
- shard-iclb: NOTRUN -> [SKIP][89] ([i915#5235]) +2 similar issues
[89]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb1/igt@kms_plane_scaling@planes-upscale-factor-0-25-downscale-factor-0-25@pipe-b-edp-1.html
* igt@kms_prime@d3hot:
- shard-tglb: NOTRUN -> [SKIP][90] ([i915#6524])
[90]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb8/igt@[email protected]
- shard-iclb: NOTRUN -> [SKIP][91] ([i915#6524])
[91]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb3/igt@[email protected]
* igt@kms_psr2_sf@overlay-plane-move-continuous-exceed-sf:
- shard-iclb: NOTRUN -> [SKIP][92] ([i915#658]) +1 similar issue
[92]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb1/igt@[email protected]
- shard-tglb: NOTRUN -> [SKIP][93] ([i915#2920])
[93]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb7/igt@[email protected]
* igt@kms_psr2_sf@overlay-plane-update-continuous-sf:
- shard-glk: NOTRUN -> [SKIP][94] ([fdo#109271] / [i915#658]) +3 similar issues
[94]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-glk7/igt@[email protected]
* igt@kms_psr2_sf@primary-plane-update-sf-dmg-area-big-fb:
- shard-apl: NOTRUN -> [SKIP][95] ([fdo#109271] / [i915#658]) +2 similar issues
[95]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-apl7/igt@[email protected]
* igt@kms_psr2_su@page_flip-nv12:
- shard-tglb: NOTRUN -> [SKIP][96] ([i915#7037])
[96]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb1/igt@kms_psr2_su@page_flip-nv12.html
- shard-iclb: NOTRUN -> [SKIP][97] ([fdo#109642] / [fdo#111068] / [i915#658])
[97]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb8/igt@kms_psr2_su@page_flip-nv12.html
* igt@kms_psr@psr2_cursor_blt:
- shard-iclb: [PASS][98] -> [SKIP][99] ([fdo#109441]) +1 similar issue
[98]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-iclb2/igt@kms_psr@psr2_cursor_blt.html
[99]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb7/igt@kms_psr@psr2_cursor_blt.html
* igt@kms_psr@psr2_sprite_mmap_cpu:
- shard-tglb: NOTRUN -> [FAIL][100] ([i915#132] / [i915#3467]) +1 similar issue
[100]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb6/igt@kms_psr@psr2_sprite_mmap_cpu.html
* igt@kms_psr@psr2_sprite_mmap_gtt:
- shard-iclb: NOTRUN -> [SKIP][101] ([fdo#109441]) +2 similar issues
[101]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb7/igt@kms_psr@psr2_sprite_mmap_gtt.html
* igt@kms_rotation_crc@primary-yf-tiled-reflect-x-0:
- shard-tglb: NOTRUN -> [SKIP][102] ([fdo#111615] / [i915#5289])
[102]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb2/igt@[email protected]
* igt@kms_writeback@writeback-check-output:
- shard-iclb: NOTRUN -> [SKIP][103] ([i915#2437])
[103]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb2/igt@[email protected]
* igt@perf_pmu@event-wait@rcs0:
- shard-iclb: NOTRUN -> [SKIP][104] ([fdo#112283])
[104]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb3/igt@perf_pmu@[email protected]
- shard-tglb: NOTRUN -> [SKIP][105] ([fdo#112283])
[105]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb8/igt@perf_pmu@[email protected]
* igt@prime_udl:
- shard-tglb: NOTRUN -> [SKIP][106] ([fdo#109291])
[106]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb5/igt@prime_udl.html
* igt@sysfs_clients@fair-0:
- shard-apl: NOTRUN -> [SKIP][107] ([fdo#109271] / [i915#2994])
[107]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-apl7/igt@[email protected]
- shard-tglb: NOTRUN -> [SKIP][108] ([i915#2994]) +2 similar issues
[108]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb3/igt@[email protected]
- shard-iclb: NOTRUN -> [SKIP][109] ([i915#2994])
[109]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb7/igt@[email protected]
* igt@sysfs_clients@sema-50:
- shard-glk: NOTRUN -> [SKIP][110] ([fdo#109271] / [i915#2994]) +1 similar issue
[110]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-glk5/igt@[email protected]
* igt@tools_test@sysfs_l3_parity:
- shard-iclb: NOTRUN -> [SKIP][111] ([fdo#109307])
[111]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb1/igt@tools_test@sysfs_l3_parity.html
#### Possible fixes ####
* igt@fbdev@unaligned-write:
- {shard-rkl}: [SKIP][112] ([i915#2582]) -> [PASS][113]
[112]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-rkl-4/igt@[email protected]
[113]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-rkl-6/igt@[email protected]
* igt@gem_create@create-clear@smem0:
- {shard-rkl}: [INCOMPLETE][114] ([i915#7397]) -> [PASS][115]
[114]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-rkl-5/igt@gem_create@[email protected]
[115]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-rkl-1/igt@gem_create@[email protected]
* igt@gem_ctx_exec@basic-nohangcheck:
- shard-tglb: [FAIL][116] ([i915#6268]) -> [PASS][117]
[116]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-tglb2/igt@[email protected]
[117]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb5/igt@[email protected]
* igt@gem_eio@kms:
- {shard-dg1}: [FAIL][118] ([i915#5784]) -> [PASS][119]
[118]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-dg1-13/igt@[email protected]
[119]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-dg1-17/igt@[email protected]
* igt@gem_exec_balancer@full-pulse:
- {shard-dg1}: [FAIL][120] ([i915#6032]) -> [PASS][121]
[120]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-dg1-17/igt@[email protected]
[121]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-dg1-17/igt@[email protected]
* igt@gem_exec_balancer@parallel-keep-in-fence:
- shard-iclb: [SKIP][122] ([i915#4525]) -> [PASS][123] +1 similar issue
[122]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-iclb7/igt@[email protected]
[123]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb1/igt@[email protected]
* igt@gem_exec_fair@basic-none@vecs0:
- shard-glk: [FAIL][124] ([i915#2842]) -> [PASS][125]
[124]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-glk6/igt@gem_exec_fair@[email protected]
[125]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-glk1/igt@gem_exec_fair@[email protected]
* igt@gem_exec_fair@basic-pace-solo@rcs0:
- shard-apl: [FAIL][126] ([i915#2842]) -> [PASS][127]
[126]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-apl1/igt@gem_exec_fair@[email protected]
[127]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-apl2/igt@gem_exec_fair@[email protected]
* igt@gem_exec_reloc@basic-gtt-cpu:
- {shard-rkl}: [SKIP][128] ([i915#3281]) -> [PASS][129] +5 similar issues
[128]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-rkl-1/igt@[email protected]
[129]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-rkl-5/igt@[email protected]
* igt@gem_pwrite@basic-self:
- {shard-rkl}: [SKIP][130] ([i915#3282]) -> [PASS][131]
[130]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-rkl-6/igt@[email protected]
[131]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-rkl-5/igt@[email protected]
* igt@gen9_exec_parse@bb-start-far:
- {shard-rkl}: [SKIP][132] ([i915#2527]) -> [PASS][133] +1 similar issue
[132]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-rkl-4/igt@[email protected]
[133]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-rkl-5/igt@[email protected]
* igt@i915_pm_dc@dc6-psr:
- shard-iclb: [FAIL][134] ([i915#3989] / [i915#454]) -> [PASS][135]
[134]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-iclb6/igt@[email protected]
[135]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb2/igt@[email protected]
* igt@i915_pm_rpm@drm-resources-equal:
- {shard-rkl}: [SKIP][136] ([fdo#109308]) -> [PASS][137]
[136]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-rkl-4/igt@[email protected]
[137]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-rkl-6/igt@[email protected]
* igt@i915_selftest@live@hangcheck:
- shard-tglb: [DMESG-WARN][138] ([i915#5591]) -> [PASS][139]
[138]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-tglb2/igt@i915_selftest@[email protected]
[139]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb1/igt@i915_selftest@[email protected]
* igt@kms_cursor_crc@cursor-suspend@pipe-c-dp-1:
- shard-apl: [DMESG-WARN][140] ([i915#180]) -> [PASS][141] +1 similar issue
[140]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-apl8/igt@kms_cursor_crc@[email protected]
[141]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-apl1/igt@kms_cursor_crc@[email protected]
* igt@kms_cursor_legacy@forked-bo@pipe-c:
- shard-iclb: [INCOMPLETE][142] ([i915#4391]) -> [PASS][143]
[142]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-iclb7/igt@kms_cursor_legacy@[email protected]
[143]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb3/igt@kms_cursor_legacy@[email protected]
* igt@kms_frontbuffer_tracking@fbc-badstride:
- {shard-rkl}: [SKIP][144] ([i915#1849] / [i915#4098]) -> [PASS][145] +5 similar issues
[144]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-rkl-4/igt@[email protected]
[145]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-rkl-6/igt@[email protected]
* igt@kms_plane@plane-panning-bottom-right-suspend@pipe-a-planes:
- {shard-rkl}: [SKIP][146] ([i915#3558]) -> [PASS][147] +1 similar issue
[146]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-rkl-1/igt@kms_plane@[email protected]
[147]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-rkl-6/igt@kms_plane@[email protected]
* igt@kms_plane@plane-panning-top-left@pipe-a-planes:
- {shard-rkl}: [SKIP][148] ([i915#1849] / [i915#3558]) -> [PASS][149] +1 similar issue
[148]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-rkl-2/igt@kms_plane@[email protected]
[149]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-rkl-6/igt@kms_plane@[email protected]
* igt@kms_psr@cursor_mmap_cpu:
- {shard-rkl}: [SKIP][150] ([i915#1072]) -> [PASS][151] +1 similar issue
[150]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-rkl-1/igt@kms_psr@cursor_mmap_cpu.html
[151]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-rkl-6/igt@kms_psr@cursor_mmap_cpu.html
* igt@kms_psr@psr2_no_drrs:
- shard-iclb: [SKIP][152] ([fdo#109441]) -> [PASS][153] +2 similar issues
[152]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-iclb6/igt@kms_psr@psr2_no_drrs.html
[153]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb2/igt@kms_psr@psr2_no_drrs.html
* igt@kms_psr_stress_test@flip-primary-invalidate-overlay:
- shard-tglb: [SKIP][154] ([i915#5519]) -> [PASS][155]
[154]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-tglb5/igt@[email protected]
[155]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb3/igt@[email protected]
* igt@kms_rotation_crc@primary-rotation-90:
- {shard-rkl}: [SKIP][156] ([i915#1845] / [i915#4098]) -> [PASS][157] +10 similar issues
[156]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-rkl-2/igt@[email protected]
[157]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-rkl-6/igt@[email protected]
* igt@kms_universal_plane@universal-plane-pageflip-windowed-pipe-b:
- {shard-rkl}: [SKIP][158] ([i915#4070] / [i915#4098]) -> [PASS][159]
[158]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-rkl-2/igt@kms_universal_plane@universal-plane-pageflip-windowed-pipe-b.html
[159]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-rkl-6/igt@kms_universal_plane@universal-plane-pageflip-windowed-pipe-b.html
* igt@kms_universal_plane@universal-plane-pipe-b-sanity:
- {shard-rkl}: [SKIP][160] ([i915#1845] / [i915#4070] / [i915#4098]) -> [PASS][161] +1 similar issue
[160]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-rkl-2/igt@[email protected]
[161]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-rkl-6/igt@[email protected]
* igt@perf@mi-rpc:
- {shard-rkl}: [SKIP][162] ([i915#2434]) -> [PASS][163]
[162]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-rkl-1/igt@[email protected]
[163]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-rkl-5/igt@[email protected]
* igt@prime_vgem@basic-write:
- {shard-rkl}: [SKIP][164] ([fdo#109295] / [i915#3291] / [i915#3708]) -> [PASS][165]
[164]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-rkl-2/igt@[email protected]
[165]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-rkl-5/igt@[email protected]
#### Warnings ####
* igt@gem_pwrite@basic-exhaustion:
- shard-apl: [INCOMPLETE][166] ([i915#7248]) -> [WARN][167] ([i915#2658])
[166]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-apl7/igt@[email protected]
[167]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-apl2/igt@[email protected]
- shard-tglb: [INCOMPLETE][168] ([i915#7248]) -> [WARN][169] ([i915#2658])
[168]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-tglb3/igt@[email protected]
[169]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-tglb8/igt@[email protected]
- shard-glk: [INCOMPLETE][170] ([i915#7248]) -> [WARN][171] ([i915#2658])
[170]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-glk5/igt@[email protected]
[171]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-glk5/igt@[email protected]
* igt@i915_pm_rc6_residency@rc6-idle@rcs0:
- shard-iclb: [FAIL][172] ([i915#2684]) -> [WARN][173] ([i915#2684])
[172]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-iclb6/igt@i915_pm_rc6_residency@[email protected]
[173]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb5/igt@i915_pm_rc6_residency@[email protected]
* igt@kms_psr2_sf@cursor-plane-move-continuous-sf:
- shard-iclb: [SKIP][174] ([i915#2920]) -> [SKIP][175] ([i915#658])
[174]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-iclb2/igt@[email protected]
[175]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb3/igt@[email protected]
* igt@kms_psr2_sf@overlay-plane-update-continuous-sf:
- shard-iclb: [SKIP][176] ([fdo#111068] / [i915#658]) -> [SKIP][177] ([i915#2920])
[176]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-iclb8/igt@[email protected]
[177]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-iclb2/igt@[email protected]
* igt@runner@aborted:
- shard-apl: ([FAIL][178], [FAIL][179], [FAIL][180]) ([i915#3002] / [i915#4312]) -> ([FAIL][181], [FAIL][182], [FAIL][183]) ([fdo#109271] / [i915#3002] / [i915#4312])
[178]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-apl8/igt@[email protected]
[179]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-apl7/igt@[email protected]
[180]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12403/shard-apl2/igt@[email protected]
[181]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-apl6/igt@[email protected]
[182]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-apl6/igt@[email protected]
[183]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/shard-apl1/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
[fdo#109274]: https://bugs.freedesktop.org/show_bug.cgi?id=109274
[fdo#109278]: https://bugs.freedesktop.org/show_bug.cgi?id=109278
[fdo#109280]: https://bugs.freedesktop.org/show_bug.cgi?id=109280
[fdo#109283]: https://bugs.freedesktop.org/show_bug.cgi?id=109283
[fdo#109284]: https://bugs.freedesktop.org/show_bug.cgi?id=109284
[fdo#109289]: https://bugs.freedesktop.org/show_bug.cgi?id=109289
[fdo#109291]: https://bugs.freedesktop.org/show_bug.cgi?id=109291
[fdo#109293]: https://bugs.freedesktop.org/show_bug.cgi?id=109293
[fdo#109295]: https://bugs.freedesktop.org/show_bug.cgi?id=109295
[fdo#109302]: https://bugs.freedesktop.org/show_bug.cgi?id=109302
[fdo#109307]: https://bugs.freedesktop.org/show_bug.cgi?id=109307
[fdo#109308]: https://bugs.freedesktop.org/show_bug.cgi?id=109308
[fdo#109309]: https://bugs.freedesktop.org/show_bug.cgi?id=109309
[fdo#109312]: https://bugs.freedesktop.org/show_bug.cgi?id=109312
[fdo#109441]: https://bugs.freedesktop.org/show_bug.cgi?id=109441
[fdo#109506]: https://bugs.freedesktop.org/show_bug.cgi?id=109506
[fdo#109642]: https://bugs.freedesktop.org/show_bug.cgi?id=109642
[fdo#110723]: https://bugs.freedesktop.org/show_bug.cgi?id=110723
[fdo#110725]: https://bugs.freedesktop.org/show_bug.cgi?id=110725
[fdo#111068]: https://bugs.freedesktop.org/show_bug.cgi?id=111068
[fdo#111614]: https://bugs.freedesktop.org/show_bug.cgi?id=111614
[fdo#111615]: https://bugs.freedesktop.org/show_bug.cgi?id=111615
[fdo#111656]: https://bugs.freedesktop.org/show_bug.cgi?id=111656
[fdo#111825]: https://bugs.freedesktop.org/show_bug.cgi?id=111825
[fdo#111827]: https://bugs.freedesktop.org/show_bug.cgi?id=111827
[fdo#112054]: https://bugs.freedesktop.org/show_bug.cgi?id=112054
[fdo#112283]: https://bugs.freedesktop.org/show_bug.cgi?id=112283
[i915#1072]: https://gitlab.freedesktop.org/drm/intel/issues/1072
[i915#1099]: https://gitlab.freedesktop.org/drm/intel/issues/1099
[i915#132]: https://gitlab.freedesktop.org/drm/intel/issues/132
[i915#1397]: https://gitlab.freedesktop.org/drm/intel/issues/1397
[i915#1755]: https://gitlab.freedesktop.org/drm/intel/issues/1755
[i915#180]: https://gitlab.freedesktop.org/drm/intel/issues/180
[i915#1825]: https://gitlab.freedesktop.org/drm/intel/issues/1825
[i915#1839]: https://gitlab.freedesktop.org/drm/intel/issues/1839
[i915#1845]: https://gitlab.freedesktop.org/drm/intel/issues/1845
[i915#1849]: https://gitlab.freedesktop.org/drm/intel/issues/1849
[i915#2122]: https://gitlab.freedesktop.org/drm/intel/issues/2122
[i915#2410]: https://gitlab.freedesktop.org/drm/intel/issues/2410
[i915#2434]: https://gitlab.freedesktop.org/drm/intel/issues/2434
[i915#2437]: https://gitlab.freedesktop.org/drm/intel/issues/2437
[i915#2527]: https://gitlab.freedesktop.org/drm/intel/issues/2527
[i915#2582]: https://gitlab.freedesktop.org/drm/intel/issues/2582
[i915#2587]: https://gitlab.freedesktop.org/drm/intel/issues/2587
[i915#2658]: https://gitlab.freedesktop.org/drm/intel/issues/2658
[i915#2672]: https://gitlab.freedesktop.org/drm/intel/issues/2672
[i915#2684]: https://gitlab.freedesktop.org/drm/intel/issues/2684
[i915#2705]: https://gitlab.freedesktop.org/drm/intel/issues/2705
[i915#280]: https://gitlab.freedesktop.org/drm/intel/issues/280
[i915#2842]: https://gitlab.freedesktop.org/drm/intel/issues/2842
[i915#2846]: https://gitlab.freedesktop.org/drm/intel/issues/2846
[i915#2856]: https://gitlab.freedesktop.org/drm/intel/issues/2856
[i915#2920]: https://gitlab.freedesktop.org/drm/intel/issues/2920
[i915#2994]: https://gitlab.freedesktop.org/drm/intel/issues/2994
[i915#3002]: https://gitlab.freedesktop.org/drm/intel/issues/3002
[i915#3116]: https://gitlab.freedesktop.org/drm/intel/issues/3116
[i915#3281]: https://gitlab.freedesktop.org/drm/intel/issues/3281
[i915#3282]: https://gitlab.freedesktop.org/drm/intel/issues/3282
[i915#3291]: https://gitlab.freedesktop.org/drm/intel/issues/3291
[i915#3297]: https://gitlab.freedesktop.org/drm/intel/issues/3297
[i915#3299]: https://gitlab.freedesktop.org/drm/intel/issues/3299
[i915#3318]: https://gitlab.freedesktop.org/drm/intel/issues/3318
[i915#3359]: https://gitlab.freedesktop.org/drm/intel/issues/3359
[i915#3361]: https://gitlab.freedesktop.org/drm/intel/issues/3361
[i915#3458]: https://gitlab.freedesktop.org/drm/intel/issues/3458
[i915#3467]: https://gitlab.freedesktop.org/drm/intel/issues/3467
[i915#3469]: https://gitlab.freedesktop.org/drm/intel/issues/3469
[i915#3536]: https://gitlab.freedesktop.org/drm/intel/issues/3536
[i915#3539]: https://gitlab.freedesktop.org/drm/intel/issues/3539
[i915#3546]: https://gitlab.freedesktop.org/drm/intel/issues/3546
[i915#3555]: https://gitlab.freedesktop.org/drm/intel/issues/3555
[i915#3558]: https://gitlab.freedesktop.org/drm/intel/issues/3558
[i915#3637]: https://gitlab.freedesktop.org/drm/intel/issues/3637
[i915#3638]: https://gitlab.freedesktop.org/drm/intel/issues/3638
[i915#3639]: https://gitlab.freedesktop.org/drm/intel/issues/3639
[i915#3689]: https://gitlab.freedesktop.org/drm/intel/issues/3689
[i915#3708]: https://gitlab.freedesktop.org/drm/intel/issues/3708
[i915#3734]: https://gitlab.freedesktop.org/drm/intel/issues/3734
[i915#3810]: https://gitlab.freedesktop.org/drm/intel/issues/3810
[i915#3826]: https://gitlab.freedesktop.org/drm/intel/issues/3826
[i915#3840]: https://gitlab.freedesktop.org/drm/intel/issues/3840
[i915#3886]: https://gitlab.freedesktop.org/drm/intel/issues/3886
[i915#3955]: https://gitlab.freedesktop.org/drm/intel/issues/3955
[i915#3989]: https://gitlab.freedesktop.org/drm/intel/issues/3989
[i915#4070]: https://gitlab.freedesktop.org/drm/intel/issues/4070
[i915#4077]: https://gitlab.freedesktop.org/drm/intel/issues/4077
[i915#4078]: https://gitlab.freedesktop.org/drm/intel/issues/4078
[i915#4079]: https://gitlab.freedesktop.org/drm/intel/issues/4079
[i915#4083]: https://gitlab.freedesktop.org/drm/intel/issues/4083
[i915#4098]: https://gitlab.freedesktop.org/drm/intel/issues/4098
[i915#4103]: https://gitlab.freedesktop.org/drm/intel/issues/4103
[i915#4212]: https://gitlab.freedesktop.org/drm/intel/issues/4212
[i915#4215]: https://gitlab.freedesktop.org/drm/intel/issues/4215
[i915#426]: https://gitlab.freedesktop.org/drm/intel/issues/426
[i915#4270]: https://gitlab.freedesktop.org/drm/intel/issues/4270
[i915#4312]: https://gitlab.freedesktop.org/drm/intel/issues/4312
[i915#433]: https://gitlab.freedesktop.org/drm/intel/issues/433
[i915#4349]: https://gitlab.freedesktop.org/drm/intel/issues/4349
[i915#4387]: https://gitlab.freedesktop.org/drm/intel/issues/4387
[i915#4391]: https://gitlab.freedesktop.org/drm/intel/issues/4391
[i915#4525]: https://gitlab.freedesktop.org/drm/intel/issues/4525
[i915#4538]: https://gitlab.freedesktop.org/drm/intel/issues/4538
[i915#454]: https://gitlab.freedesktop.org/drm/intel/issues/454
[i915#4573]: https://gitlab.freedesktop.org/drm/intel/issues/4573
[i915#4613]: https://gitlab.freedesktop.org/drm/intel/issues/4613
[i915#4812]: https://gitlab.freedesktop.org/drm/intel/issues/4812
[i915#4833]: https://gitlab.freedesktop.org/drm/intel/issues/4833
[i915#4852]: https://gitlab.freedesktop.org/drm/intel/issues/4852
[i915#4859]: https://gitlab.freedesktop.org/drm/intel/issues/4859
[i915#4860]: https://gitlab.freedesktop.org/drm/intel/issues/4860
[i915#4880]: https://gitlab.freedesktop.org/drm/intel/issues/4880
[i915#4885]: https://gitlab.freedesktop.org/drm/intel/issues/4885
[i915#4991]: https://gitlab.freedesktop.org/drm/intel/issues/4991
[i915#5176]: https://gitlab.freedesktop.org/drm/intel/issues/5176
[i915#5235]: https://gitlab.freedesktop.org/drm/intel/issues/5235
[i915#5286]: https://gitlab.freedesktop.org/drm/intel/issues/5286
[i915#5288]: https://gitlab.freedesktop.org/drm/intel/issues/5288
[i915#5289]: https://gitlab.freedesktop.org/drm/intel/issues/5289
[i915#5325]: https://gitlab.freedesktop.org/drm/intel/issues/5325
[i915#5327]: https://gitlab.freedesktop.org/drm/intel/issues/5327
[i915#533]: https://gitlab.freedesktop.org/drm/intel/issues/533
[i915#5438]: https://gitlab.freedesktop.org/drm/intel/issues/5438
[i915#5461]: https://gitlab.freedesktop.org/drm/intel/issues/5461
[i915#5519]: https://gitlab.freedesktop.org/drm/intel/issues/5519
[i915#5563]: https://gitlab.freedesktop.org/drm/intel/issues/5563
[i915#5566]: https://gitlab.freedesktop.org/drm/intel/issues/5566
[i915#5591]: https://gitlab.freedesktop.org/drm/intel/issues/5591
[i915#5723]: https://gitlab.freedesktop.org/drm/intel/issues/5723
[i915#5784]: https://gitlab.freedesktop.org/drm/intel/issues/5784
[i915#6032]: https://gitlab.freedesktop.org/drm/intel/issues/6032
[i915#6095]: https://gitlab.freedesktop.org/drm/intel/issues/6095
[i915#6230]: https://gitlab.freedesktop.org/drm/intel/issues/6230
[i915#6248]: https://gitlab.freedesktop.org/drm/intel/issues/6248
[i915#6252]: https://gitlab.freedesktop.org/drm/intel/issues/6252
[i915#6268]: https://gitlab.freedesktop.org/drm/intel/issues/6268
[i915#6334]: https://gitlab.freedesktop.org/drm/intel/issues/6334
[i915#6433]: https://gitlab.freedesktop.org/drm/intel/issues/6433
[i915#6497]: https://gitlab.freedesktop.org/drm/intel/issues/6497
[i915#6524]: https://gitlab.freedesktop.org/drm/intel/issues/6524
[i915#6537]: https://gitlab.freedesktop.org/drm/intel/issues/6537
[i915#658]: https://gitlab.freedesktop.org/drm/intel/issues/658
[i915#6621]: https://gitlab.freedesktop.org/drm/intel/issues/6621
[i915#6768]: https://gitlab.freedesktop.org/drm/intel/issues/6768
[i915#6946]: https://gitlab.freedesktop.org/drm/intel/issues/6946
[i915#7037]: https://gitlab.freedesktop.org/drm/intel/issues/7037
[i915#7116]: https://gitlab.freedesktop.org/drm/intel/issues/7116
[i915#7118]: https://gitlab.freedesktop.org/drm/intel/issues/7118
[i915#716]: https://gitlab.freedesktop.org/drm/intel/issues/716
[i915#7173]: https://gitlab.freedesktop.org/drm/intel/issues/7173
[i915#7205]: https://gitlab.freedesktop.org/drm/intel/issues/7205
[i915#7248]: https://gitlab.freedesktop.org/drm/intel/issues/7248
[i915#7397]: https://gitlab.freedesktop.org/drm/intel/issues/7397
[i915#768]: https://gitlab.freedesktop.org/drm/intel/issues/768
Build changes
-------------
* CI: CI-20190529 -> None
* IGT: IGT_7068 -> IGTPW_8132
* Piglit: piglit_4509 -> None
CI-20190529: 20190529
CI_DRM_12403: 9c91ccfa558e7510add39958a56745ae7fa07753 @ git://anongit.freedesktop.org/gfx-ci/linux
IGTPW_8132: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/index.html
IGT_7068: 5c0ec905b6bbecfb8df8b8f3315d0470539e6ae3 @ https://gitlab.freedesktop.org/drm/igt-gpu-tools.git
piglit_4509: fdc5a4ca11124ab8413c7988896eec4c97336694 @ git://anongit.freedesktop.org/piglit
== Logs ==
For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8132/index.html
Quoted 26 lines · Ashutosh Dixit
With SLPC, even when we set the same min and max freq's, the requested and actual freq's can differ from the min/max freq set. For example "efficient freq" (when in effect) can override set min/max freq. In general FW is the final arbiter in determining freq and can override set values.
Therefore in igt@perf_pmu@frequency subtest, compare the requested freq reported by PMU not against the set freq's but against the requested freq reported in sysfs. Also add a delay after setting freq's to account for messaging delays in setting freq's in GuC.
v2: Introduce a 100 ms delay after setting freq
v3: Update commit message, code identical to v2Cc Vinay Belgaumkar <[email protected]>Cc Tvrtko Ursulin <[email protected]>Signed-off-by Ashutosh Dixit <[email protected]>--- tests/i915/perf_pmu.c | 23 +++++++++++++++-------- 1 file changed, 15 insertions(+), 8 deletions(-)tests/i915/perf_pmu.c +0
-0
tests/i915/perf_pmu.c test_interrupts_sync(int gem_fd) · line 1547
Quoted 5 lines
tests/i915/perf_pmu.c +2
-0@@ -1543,10 +1543,13 @@ test_interrupts_sync(int gem_fd)igt_assert_lte (target, busy); } +/* Wait for GuC SLPC freq changes to take effect */ +#define wait_freq_set () usleep(100000)
A future task of maybe adding a drop_caches flag to wait for this?
Alternatively, if we have a sysfs or debugfs file which reflects the value once GuC has processed it, keep reading it until it is updated? Even if to something other than it was before the write, to handle the inability to set the exact requested frequency sometimes?
Quoted 24 lines
tests/i915/perf_pmu.c +5
-3+static void test_frequency (int gem_fd) { -uint32_t min_freq, max_freq, boost_freq; +uint32_t min_freq, max_freq, boost_freq, min_req, max_req;uint64_t val[2 ], start[2 ], slept;double min[2 ], max[2 ];igt_spin_t *spin;@@ -1572,10 +1575,11 @@ test_frequency(int gem_fd)* Set GPU to min frequency and read PMU counters. */ igt_require (igt_sysfs_set_u32 (sysfs,"gt_min_freq_mhz" , min_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) == min_freq);igt_require (igt_sysfs_set_u32 (sysfs,"gt_max_freq_mhz" , min_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == min_freq);igt_require (igt_sysfs_set_u32 (sysfs,"gt_boost_freq_mhz" , min_freq)); +wait_freq_set (); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) == min_freq); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == min_freq);igt_require (igt_sysfs_get_u32 (sysfs,"gt_boost_freq_mhz" ) == min_freq);gem_quiescent_gpu (gem_fd);/* Idle to be sure the change takes effect */
tests/i915/perf_pmu.c test_frequency(int gem_fd) · line 1594
Quoted 4 lines
tests/i915/perf_pmu.c +1
-0@@ -1587,6 +1591,7 @@ test_frequency(int gem_fd) min[0 ] =1e9 *(val[0 ] - start[0 ]) / slept; min[1 ] =1e9 *(val[1 ] - start[1 ]) / slept; + min_req =igt_sysfs_get_u32 (sysfs,"gt_cur_freq_mhz" );
This really leaves a bad taste for me still, sorry. First of all, it may make it more palatable if readback was immediately after wait_freq_set(), or as part of it. Otherwise it seems like even more confused test and sysfs ABI.
Did we close on my question of whether we can make readback of gt_max_freq_mhz actually represent the real max? Correct me please if I got confused - with min freq checking here we write max=300 and get 350? If so can we make gt_max_freq_mhz return 350 after the write of 300? Or return an error?
Quoted 18 lines
tests/i915/perf_pmu.c +3
-3igt_spin_free (gem_fd, spin);gem_quiescent_gpu (gem_fd);/* Don't leak busy bo into the next phase */ @@ -1597,11 +1602,11 @@ test_frequency(int gem_fd)* Set GPU to max frequency and read PMU counters. */ igt_require (igt_sysfs_set_u32 (sysfs,"gt_max_freq_mhz" , max_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == max_freq);igt_require (igt_sysfs_set_u32 (sysfs,"gt_boost_freq_mhz" , boost_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_boost_freq_mhz" ) == boost_freq); -igt_require (igt_sysfs_set_u32 (sysfs,"gt_min_freq_mhz" , max_freq)); +wait_freq_set (); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == max_freq); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_boost_freq_mhz" ) == boost_freq);igt_require (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) == max_freq);gem_quiescent_gpu (gem_fd);
tests/i915/perf_pmu.c test_frequency(int gem_fd) · line 1621
Quoted 4 lines
tests/i915/perf_pmu.c +1
-0@@ -1613,6 +1618,7 @@ test_frequency(int gem_fd) max[0 ] =1e9 *(val[0 ] - start[0 ]) / slept; max[1 ] =1e9 *(val[1 ] - start[1 ]) / slept; + max_req =igt_sysfs_get_u32 (sysfs,"gt_cur_freq_mhz" );
For the max freq use case we had 15% downward tolerance to account for thermally challenged.
a) Is that not enough with SLPC? Like there is a new "failure" mode?
b) Is the extra 10% needed with the sysfs readback added? Aka do we need both?
Quoted 13 lines
tests/i915/perf_pmu.c +1
-0igt_spin_free (gem_fd, spin);gem_quiescent_gpu (gem_fd);@@ -1621,6 +1627,7 @@ test_frequency(int gem_fd)* Restore min/max. */ igt_sysfs_set_u32 (sysfs,"gt_min_freq_mhz" , min_freq); +wait_freq_set ();if (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) != min_freq)igt_warn ("Unable to restore min frequency to saved value [%u MHz], now %u MHz\n" , min_freq,igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ));@@ -1633,12 +1640,12 @@ test_frequency(int gem_fd) igt_info("Max frequency: requested %.1f, actual %.1f\n",
tests/i915/perf_pmu.c test_frequency(int gem_fd) · line 1648
Quoted 10 lines
tests/i915/perf_pmu.c +2
-2max[0], max[1]); -assert_within_epsilon (min[0 ], min_freq, tolerance); +assert_within_epsilon (min[0 ], min_req, tolerance);/* * On thermally throttled devices we cannot be sure maximum frequency * can be reached so use larger tolerance downards. */ -__assert_within_epsilon (max[0 ], max_freq, tolerance,0.15f ); +__assert_within_epsilon (max[0 ], max_req, tolerance,0.15f );
This 0.15f is what I was talking about above.
Regards,
Tvrtko
Quoted 3 lines
tests/i915/perf_pmu.c +0
-0} static void
Quoted 1 line · Tvrtko Ursulin
Hi Tvrtko,
I am only offering an overall clarification below first, not answering all of your points for now.
Quoted 73 lines · Ashutosh Dixit
Quoted 26 lines
With SLPC, even when we set the same min and max freq's, the requested and actual freq's can differ from the min/max freq set. For example "efficient freq" (when in effect) can override set min/max freq. In general FW is the final arbiter in determining freq and can override set values.
Therefore in igt@perf_pmu@frequency subtest, compare the requested freq reported by PMU not against the set freq's but against the requested freq reported in sysfs. Also add a delay after setting freq's to account for messaging delays in setting freq's in GuC.
v2: Introduce a 100 ms delay after setting freq
v3: Update commit message, code identical to v2Cc Vinay Belgaumkar <[email protected]>Cc Tvrtko Ursulin <[email protected]>Signed-off-by Ashutosh Dixit <[email protected]>--- tests/i915/perf_pmu.c | 23 +++++++++++++++-------- 1 file changed, 15 insertions(+), 8 deletions(-)tests/i915/perf_pmu.c +0
-0tests/i915/perf_pmu.c test_interrupts_sync(int gem_fd) · line 1546
Quoted 4 lines
tests/i915/perf_pmu.c +1
-0@@ -1543,10 +1543,13 @@ test_interrupts_sync(int gem_fd)igt_assert_lte (target, busy); } +/* Wait for GuC SLPC freq changes to take effect */ +#define wait_freq_set () usleep(100000)A future task of maybe adding a
drop_cachesflag to wait for this?Alternatively, if we have a sysfs or debugfs file which reflects the value once GuC has processed it, keep reading it until it is updated? Even if to something other than it was before the write, to handle the inability to set the exact requested frequency sometimes?
Quoted 25 lines
tests/i915/perf_pmu.c +5
-3+static void test_frequency (int gem_fd) { -uint32_t min_freq, max_freq, boost_freq; +uint32_t min_freq, max_freq, boost_freq, min_req, max_req;uint64_t val[2 ], start[2 ], slept;double min[2 ], max[2 ];igt_spin_t *spin;@@ -1572,10 +1575,11 @@ test_frequency(int gem_fd)* Set GPU to min frequency and read PMU counters. */ igt_require (igt_sysfs_set_u32 (sysfs,"gt_min_freq_mhz" , min_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) == min_freq);igt_require (igt_sysfs_set_u32 (sysfs,"gt_max_freq_mhz" , min_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == min_freq);igt_require (igt_sysfs_set_u32 (sysfs,"gt_boost_freq_mhz" , min_freq)); +wait_freq_set (); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) == min_freq); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == min_freq);igt_require (igt_sysfs_get_u32 (sysfs,"gt_boost_freq_mhz" ) == min_freq);gem_quiescent_gpu (gem_fd);/* Idle to be sure the change takes effect */
tests/i915/perf_pmu.c test_frequency(int gem_fd) · line 1593
Quoted 3 lines
tests/i915/perf_pmu.c +1
-0@@ -1587,6 +1591,7 @@ test_frequency(int gem_fd) min[0 ] =1e9 *(val[0 ] - start[0 ]) / slept; min[1 ] =1e9 *(val[1 ] - start[1 ]) / slept; + min_req =igt_sysfs_get_u32 (sysfs,"gt_cur_freq_mhz" );This really leaves a bad taste for me still, sorry. First of all, it may make it more palatable if readback was immediately after
wait_freq_set(), or as part of it. Otherwise it seems like even more confused test and sysfs ABI.
Quoted 5 lines
Did we close on my question of whether we can make readback of
gt_max_freq_mhzactually represent the real max? Correct me please if I got confused - with min freq checking here we write max=300 and get 350? If so can we makegt_max_freq_mhzreturn 350 after the write of 300? Or return an error?
Did you read Vinay's email on v1 here:
https://patchwork.freedesktop.org/patch/510304/?series=110574&rev=1#comment_921571
As Vinay says, SLPC is requesting an entirely different freq (the effecient freq rather than the set 'min == max freq') and no amount of waiting will result in the requested freq being the same as the set 'min == max freq'. Also the efficient freq cannot be controlled from user space.
(So in the example above min and max are set to 300 MHz and SLPC is requesting efficient freq which is 350 or 400 MHz. So min and max values are correct at 300 MHz and requested/actual freq will show 350/400 MHz).
Let me also explain what happens in GuC FW when we change min/max freq:
* i915 receives min/max freq from sysfs and sends a message to GuC to change min/max freq.
* The message itself might be queued behind other messages in the host to GuC queue. i915 is still blocked in the sysfs handler.
* Now when GuC receives the message interrupt, in its top-half handler it validates the freq parameters and sends a message back to host saying the freq change has been accepted. But min/max freq is actually changed later in the bottom-half handler in GuC FW. So the top half only validates the freq value.
* When i915 receives the message from GuC top half, the sysfs handler unblocks and starts reflecting the new min/max freq value.
So the delay for the new min/max freq values to take effect is between returning from the sysfs handler and when the min/max freq are really set in GuC bottom half (an indeterminate possibly large amount of time if GuC is busy).
So as you see this introduces an indeterminacy due to to GuC interface not being synchronous. If this needs to be fixed, it needs to be fixed in GuC FW rather than waiting etc. in i915. GuC needs to implement another notification to host /after/ the min/max freq change has been applied in its bottom half. Doing this is not a priority for the GuC team at this point.
However, this bug is not due to these delays but due to GuC requesting the efficient freq instead of the set 'min == max freq' as mentioned above.
If all these delays are confusing we can revert to v1 of the patch:
https://patchwork.freedesktop.org/patch/510304/?series=110574&rev=1
v1 is simple and does not introduce these delays but just reads the requested freq from sysfs after the 500 ms PMU measurement but while the spinner is still running (which we do in v3 too). Without load SLPC will only request RPn so to compare with PMU measurement we must read the requested freq from sysfs with the spinner load. v1 is sufficient to fix the bug but I thought to be completely safe it is better to add wait_freq_set() after changing min/max freq (and hence v3).
I hope this clarifies things somewhat. If not please let me know.
Thanks.
--
Ashutosh
Quoted 60 lines
Quoted 17 lines
tests/i915/perf_pmu.c +3
-3igt_spin_free (gem_fd, spin);gem_quiescent_gpu (gem_fd);/* Don't leak busy bo into the next phase */ @@ -1597,11 +1602,11 @@ test_frequency(int gem_fd)* Set GPU to max frequency and read PMU counters. */ igt_require (igt_sysfs_set_u32 (sysfs,"gt_max_freq_mhz" , max_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == max_freq);igt_require (igt_sysfs_set_u32 (sysfs,"gt_boost_freq_mhz" , boost_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_boost_freq_mhz" ) == boost_freq); -igt_require (igt_sysfs_set_u32 (sysfs,"gt_min_freq_mhz" , max_freq)); +wait_freq_set (); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == max_freq); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_boost_freq_mhz" ) == boost_freq);igt_require (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) == max_freq);gem_quiescent_gpu (gem_fd);tests/i915/perf_pmu.c test_frequency(int gem_fd) · line 1620
Quoted 3 lines
tests/i915/perf_pmu.c +1
-0@@ -1613,6 +1618,7 @@ test_frequency(int gem_fd) max[0 ] =1e9 *(val[0 ] - start[0 ]) / slept; max[1 ] =1e9 *(val[1 ] - start[1 ]) / slept; + max_req =igt_sysfs_get_u32 (sysfs,"gt_cur_freq_mhz" );For the max freq use case we had 15% downward tolerance to account for thermally challenged.
a) Is that not enough with SLPC? Like there is a new "failure" mode?
b) Is the extra 10% needed with the sysfs readback added? Aka do we need both?
Quoted 11 lines
tests/i915/perf_pmu.c +1
-0igt_spin_free (gem_fd, spin);gem_quiescent_gpu (gem_fd);@@ -1621,6 +1627,7 @@ test_frequency(int gem_fd)* Restore min/max. */ igt_sysfs_set_u32 (sysfs,"gt_min_freq_mhz" , min_freq); +wait_freq_set ();if (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) != min_freq)igt_warn ("Unable to restore min frequency to saved value [%u MHz], now %u MHz\n" , min_freq,igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ));tests/i915/perf_pmu.c test_frequency(int gem_fd) · line 1648
Quoted 10 lines
tests/i915/perf_pmu.c +2
-1@@ -1633,12 +1640,12 @@ test_frequency(int gem_fd)igt_info ("Max frequency: requested %.1f, actual %.1f\n" , max[0 ], max[1 ]); -assert_within_epsilon (min[0 ], min_freq, tolerance); +assert_within_epsilon (min[0 ], min_req, tolerance);/* * On thermally throttled devices we cannot be sure maximum frequency * can be reached so use larger tolerance downards. */ -__assert_within_epsilon (max[0 ], max_freq, tolerance,0.15f ); +__assert_within_epsilon (max[0 ], max_req, tolerance,0.15f );This 0.15f is what I was talking about above.
Regards,
Tvrtko
Quoted 2 lines
tests/i915/perf_pmu.c +0
-0} static void
Quoted 87 lines · Ashutosh
Quoted 1 line · Tvrtko Ursulin
Hi Tvrtko,
I am only offering an overall clarification below first, not answering all of your points for now.
Quoted 73 lines · Ashutosh Dixit
Quoted 26 lines
With SLPC, even when we set the same min and max freq's, the requested and actual freq's can differ from the min/max freq set. For example "efficient freq" (when in effect) can override set min/max freq. In general FW is the final arbiter in determining freq and can override set values.
Therefore in igt@perf_pmu@frequency subtest, compare the requested freq reported by PMU not against the set freq's but against the requested freq reported in sysfs. Also add a delay after setting freq's to account for messaging delays in setting freq's in GuC.
v2: Introduce a 100 ms delay after setting freq
v3: Update commit message, code identical to v2Cc Vinay Belgaumkar <[email protected]>Cc Tvrtko Ursulin <[email protected]>Signed-off-by Ashutosh Dixit <[email protected]>--- tests/i915/perf_pmu.c | 23 +++++++++++++++-------- 1 file changed, 15 insertions(+), 8 deletions(-)tests/i915/perf_pmu.c +0
-0tests/i915/perf_pmu.c test_interrupts_sync(int gem_fd) · line 1546
Quoted 4 lines
tests/i915/perf_pmu.c +1
-0@@ -1543,10 +1543,13 @@ test_interrupts_sync(int gem_fd)igt_assert_lte (target, busy); } +/* Wait for GuC SLPC freq changes to take effect */ +#define wait_freq_set () usleep(100000)A future task of maybe adding a
drop_cachesflag to wait for this?Alternatively, if we have a sysfs or debugfs file which reflects the value once GuC has processed it, keep reading it until it is updated? Even if to something other than it was before the write, to handle the inability to set the exact requested frequency sometimes?
Quoted 25 lines
tests/i915/perf_pmu.c +5
-3+static void test_frequency (int gem_fd) { -uint32_t min_freq, max_freq, boost_freq; +uint32_t min_freq, max_freq, boost_freq, min_req, max_req;uint64_t val[2 ], start[2 ], slept;double min[2 ], max[2 ];igt_spin_t *spin;@@ -1572,10 +1575,11 @@ test_frequency(int gem_fd)* Set GPU to min frequency and read PMU counters. */ igt_require (igt_sysfs_set_u32 (sysfs,"gt_min_freq_mhz" , min_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) == min_freq);igt_require (igt_sysfs_set_u32 (sysfs,"gt_max_freq_mhz" , min_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == min_freq);igt_require (igt_sysfs_set_u32 (sysfs,"gt_boost_freq_mhz" , min_freq)); +wait_freq_set (); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) == min_freq); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == min_freq);igt_require (igt_sysfs_get_u32 (sysfs,"gt_boost_freq_mhz" ) == min_freq);gem_quiescent_gpu (gem_fd);/* Idle to be sure the change takes effect */
tests/i915/perf_pmu.c test_frequency(int gem_fd) · line 1593
Quoted 3 lines
tests/i915/perf_pmu.c +1
-0@@ -1587,6 +1591,7 @@ test_frequency(int gem_fd) min[0 ] =1e9 *(val[0 ] - start[0 ]) / slept; min[1 ] =1e9 *(val[1 ] - start[1 ]) / slept; + min_req =igt_sysfs_get_u32 (sysfs,"gt_cur_freq_mhz" );This really leaves a bad taste for me still, sorry. First of all, it may make it more palatable if readback was immediately after
wait_freq_set(), or as part of it. Otherwise it seems like even more confused test and sysfs ABI.Quoted 5 lines
Did we close on my question of whether we can make readback of
gt_max_freq_mhzactually represent the real max? Correct me please if I got confused - with min freq checking here we write max=300 and get 350? If so can we makegt_max_freq_mhzreturn 350 after the write of 300? Or return an error?
Quoted 8 lines
Did you read Vinay's email on v1 here:
https://patchwork.freedesktop.org/patch/510304/?series=110574&rev=1#comment_921571
As Vinay says, SLPC is requesting an entirely different freq (the effecient freq rather than the set 'min == max freq') and no amount of waiting will result in the requested freq being the same as the set 'min == max freq'. Also the efficient freq cannot be controlled from user space.
Why is that not a break in our ABI? What is the point in allowing users to write into this field if the desired value can be silently ignored?
Quoted 17 lines
(So in the example above min and max are set to 300 MHz and SLPC is requesting efficient freq which is 350 or 400 MHz. So min and max values are correct at 300 MHz and requested/actual freq will show 350/400 MHz).
Let me also explain what happens in GuC FW when we change min/max freq:
* i915 receives min/max freq from sysfs and sends a message to GuC to change min/max freq.
* The message itself might be queued behind other messages in the host to GuC queue. i915 is still blocked in the sysfs handler.
* Now when GuC receives the message interrupt, in its top-half handler it validates the freq parameters and sends a message back to host saying the freq change has been accepted. But min/max freq is actually changed later in the bottom-half handler in GuC FW. So the top half only validates the freq value.
* When i915 receives the message from GuC top half, the sysfs handler unblocks and starts reflecting the new min/max freq value.
Quoted 4 lines
So the delay for the new min/max freq values to take effect is between returning from the sysfs handler and when the min/max freq are really set in GuC bottom half (an indeterminate possibly large amount of time if GuC is busy).
Do we have a drop_caches flag which would force waiting of idle GuC, which would cover the bottom half as well? Could we or should we add it if not? Would beat the usleep approach..
Quoted 6 lines
So as you see this introduces an indeterminacy due to to GuC interface not being synchronous. If this needs to be fixed, it needs to be fixed in GuC FW rather than waiting etc. in i915. GuC needs to implement another notification to host /after/ the min/max freq change has been applied in its bottom half. Doing this is not a priority for the GuC team at this point.
Is there a method for querying the real min from GuC, which would maybe work after waiting for GuC to idle?
Quoted 16 lines
However, this bug is not due to these delays but due to GuC requesting the efficient freq instead of the set 'min == max freq' as mentioned above.
If all these delays are confusing we can revert to v1 of the patch:
https://patchwork.freedesktop.org/patch/510304/?series=110574&rev=1
v1 is simple and does not introduce these delays but just reads the requested freq from sysfs after the 500 ms PMU measurement but while the spinner is still running (which we do in v3 too). Without load SLPC will only request RPn so to compare with PMU measurement we must read the requested freq from sysfs with the spinner load. v1 is sufficient to fix the bug but I thought to be completely safe it is better to add
wait_freq_set()after changing min/max freq (and hence v3).
Quoted 1 line
I hope this clarifies things somewhat. If not please let me know.
It clarifies a large chunk thank you.
Another question which you missed is why is it needed for the max mode? There we allow a 15% error (more than 5% default). Is that not enough with SLPC?
Regards,
Tvrtko
Quoted 66 lines
Thanks.
--
AshutoshQuoted 60 lines
Quoted 17 lines
tests/i915/perf_pmu.c +3
-3igt_spin_free (gem_fd, spin);gem_quiescent_gpu (gem_fd);/* Don't leak busy bo into the next phase */ @@ -1597,11 +1602,11 @@ test_frequency(int gem_fd)* Set GPU to max frequency and read PMU counters. */ igt_require (igt_sysfs_set_u32 (sysfs,"gt_max_freq_mhz" , max_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == max_freq);igt_require (igt_sysfs_set_u32 (sysfs,"gt_boost_freq_mhz" , boost_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_boost_freq_mhz" ) == boost_freq); -igt_require (igt_sysfs_set_u32 (sysfs,"gt_min_freq_mhz" , max_freq)); +wait_freq_set (); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == max_freq); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_boost_freq_mhz" ) == boost_freq);igt_require (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) == max_freq);gem_quiescent_gpu (gem_fd);tests/i915/perf_pmu.c test_frequency(int gem_fd) · line 1620
Quoted 3 lines
tests/i915/perf_pmu.c +1
-0@@ -1613,6 +1618,7 @@ test_frequency(int gem_fd) max[0 ] =1e9 *(val[0 ] - start[0 ]) / slept; max[1 ] =1e9 *(val[1 ] - start[1 ]) / slept; + max_req =igt_sysfs_get_u32 (sysfs,"gt_cur_freq_mhz" );For the max freq use case we had 15% downward tolerance to account for thermally challenged.
a) Is that not enough with SLPC? Like there is a new "failure" mode?
b) Is the extra 10% needed with the sysfs readback added? Aka do we need both?
Quoted 11 lines
tests/i915/perf_pmu.c +1
-0igt_spin_free (gem_fd, spin);gem_quiescent_gpu (gem_fd);@@ -1621,6 +1627,7 @@ test_frequency(int gem_fd)* Restore min/max. */ igt_sysfs_set_u32 (sysfs,"gt_min_freq_mhz" , min_freq); +wait_freq_set ();if (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) != min_freq)igt_warn ("Unable to restore min frequency to saved value [%u MHz], now %u MHz\n" , min_freq,igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ));tests/i915/perf_pmu.c test_frequency(int gem_fd) · line 1648
Quoted 10 lines
tests/i915/perf_pmu.c +2
-1@@ -1633,12 +1640,12 @@ test_frequency(int gem_fd)igt_info ("Max frequency: requested %.1f, actual %.1f\n" , max[0 ], max[1 ]); -assert_within_epsilon (min[0 ], min_freq, tolerance); +assert_within_epsilon (min[0 ], min_req, tolerance);/* * On thermally throttled devices we cannot be sure maximum frequency * can be reached so use larger tolerance downards. */ -__assert_within_epsilon (max[0 ], max_freq, tolerance,0.15f ); +__assert_within_epsilon (max[0 ], max_req, tolerance,0.15f );This 0.15f is what I was talking about above.
Regards,
Tvrtko
Quoted 2 lines
tests/i915/perf_pmu.c +0
-0} static void
Quoted 5 lines · Tvrtko Ursulin
Quoted 3 lines · Ashutosh
Quoted 1 line · Tvrtko Ursulin
Hi Tvrtko,
Sorry for the delay in replying on this. We are discussing this but still no conclusion. I have replied what I can below, will update again after I know more. Wanted to send a reply before disappearing for the holidays. Please see below.
Quoted 88 lines
Quoted 88 lines
Hi Tvrtko,
I am only offering an overall clarification below first, not answering all of your points for now.
Quoted 73 lines · Ashutosh Dixit
Quoted 26 lines
With SLPC, even when we set the same min and max freq's, the requested and actual freq's can differ from the min/max freq set. For example "efficient freq" (when in effect) can override set min/max freq. In general FW is the final arbiter in determining freq and can override set values.
Therefore in igt@perf_pmu@frequency subtest, compare the requested freq reported by PMU not against the set freq's but against the requested freq reported in sysfs. Also add a delay after setting freq's to account for messaging delays in setting freq's in GuC.
v2: Introduce a 100 ms delay after setting freq
v3: Update commit message, code identical to v2Cc Vinay Belgaumkar <[email protected]>Cc Tvrtko Ursulin <[email protected]>Signed-off-by Ashutosh Dixit <[email protected]>--- tests/i915/perf_pmu.c | 23 +++++++++++++++-------- 1 file changed, 15 insertions(+), 8 deletions(-)tests/i915/perf_pmu.c +0
-0tests/i915/perf_pmu.c test_interrupts_sync(int gem_fd) · line 1546
Quoted 4 lines
tests/i915/perf_pmu.c +1
-0@@ -1543,10 +1543,13 @@ test_interrupts_sync(int gem_fd)igt_assert_lte (target, busy); } +/* Wait for GuC SLPC freq changes to take effect */ +#define wait_freq_set () usleep(100000)A future task of maybe adding a
drop_cachesflag to wait for this?Alternatively, if we have a sysfs or debugfs file which reflects the value once GuC has processed it, keep reading it until it is updated? Even if to something other than it was before the write, to handle the inability to set the exact requested frequency sometimes?
Quoted 25 lines
tests/i915/perf_pmu.c +5
-3+static void test_frequency (int gem_fd) { -uint32_t min_freq, max_freq, boost_freq; +uint32_t min_freq, max_freq, boost_freq, min_req, max_req;uint64_t val[2 ], start[2 ], slept;double min[2 ], max[2 ];igt_spin_t *spin;@@ -1572,10 +1575,11 @@ test_frequency(int gem_fd)* Set GPU to min frequency and read PMU counters. */ igt_require (igt_sysfs_set_u32 (sysfs,"gt_min_freq_mhz" , min_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) == min_freq);igt_require (igt_sysfs_set_u32 (sysfs,"gt_max_freq_mhz" , min_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == min_freq);igt_require (igt_sysfs_set_u32 (sysfs,"gt_boost_freq_mhz" , min_freq)); +wait_freq_set (); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) == min_freq); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == min_freq);igt_require (igt_sysfs_get_u32 (sysfs,"gt_boost_freq_mhz" ) == min_freq);gem_quiescent_gpu (gem_fd);/* Idle to be sure the change takes effect */
tests/i915/perf_pmu.c test_frequency(int gem_fd) · line 1593
Quoted 3 lines
tests/i915/perf_pmu.c +1
-0@@ -1587,6 +1591,7 @@ test_frequency(int gem_fd) min[0 ] =1e9 *(val[0 ] - start[0 ]) / slept; min[1 ] =1e9 *(val[1 ] - start[1 ]) / slept; + min_req =igt_sysfs_get_u32 (sysfs,"gt_cur_freq_mhz" );This really leaves a bad taste for me still, sorry. First of all, it may make it more palatable if readback was immediately after
wait_freq_set(), or as part of it. Otherwise it seems like even more confused test and sysfs ABI.Quoted 5 lines
Did we close on my question of whether we can make readback of
gt_max_freq_mhzactually represent the real max? Correct me please if I got confused - with min freq checking here we write max=300 and get 350? If so can we makegt_max_freq_mhzreturn 350 after the write of 300? Or return an error?Did you read Vinay's email on v1 here:
https://patchwork.freedesktop.org/patch/510304/?series=110574&rev=1#comment_921571
Quoted 7 lines
Quoted 4 lines
As Vinay says, SLPC is requesting an entirely different freq (the effecient freq rather than the set 'min == max freq') and no amount of waiting will result in the requested freq being the same as the set 'min == max freq'. Also the efficient freq cannot be controlled from user space.
Why is that not a break in our ABI? What is the point in allowing users to write into this field if the desired value can be silently ignored?
Yes, this is in discussion. FW is doing this and we need to see if FW can be made to obey the i915 ABI. IMHO what matters is performance per watt and if the ABI has to be tossed so be it. I will keep this list posted about what happens here. Merging any patches for these issues (GL #6806 and #6786) is blocked till the ABI breakage issue is resolved one way or another.
Quoted 18 lines
Quoted 18 lines
(So in the example above min and max are set to 300 MHz and SLPC is requesting efficient freq which is 350 or 400 MHz. So min and max values are correct at 300 MHz and requested/actual freq will show 350/400 MHz).
Let me also explain what happens in GuC FW when we change min/max freq:
* i915 receives min/max freq from sysfs and sends a message to GuC to change min/max freq.
* The message itself might be queued behind other messages in the host to GuC queue. i915 is still blocked in the sysfs handler.
* Now when GuC receives the message interrupt, in its top-half handler it validates the freq parameters and sends a message back to host saying the freq change has been accepted. But min/max freq is actually changed later in the bottom-half handler in GuC FW. So the top half only validates the freq value.
* When i915 receives the message from GuC top half, the sysfs handler unblocks and starts reflecting the new min/max freq value.
Quoted 8 lines
Quoted 4 lines
So the delay for the new min/max freq values to take effect is between returning from the sysfs handler and when the min/max freq are really set in GuC bottom half (an indeterminate possibly large amount of time if GuC is busy).
Do we have a
drop_cachesflag which would force waiting of idle GuC, which would cover the bottom half as well? Could we or should we add it if not? Would beat the usleep approach..
Agree, waiting for GuC to idle would do it, but not sure "how" to do it. As I said there is no notification after the bottom half is done applying the freq settings. We may need to schedule something after the freq setting host to guc command and wait for that to be done. A little bit out of my depth on this, let me know if you have any ideas here. I'll ask the GuC guys too. Currently we don't have a drop_caches flag for this.
Quoted 10 lines
Quoted 7 lines
So as you see this introduces an indeterminacy due to to GuC interface not being synchronous. If this needs to be fixed, it needs to be fixed in GuC FW rather than waiting etc. in i915. GuC needs to implement another notification to host /after/ the min/max freq change has been applied in its bottom half. Doing this is not a priority for the GuC team at this point.
Is there a method for querying the real min from GuC, which would maybe work after waiting for GuC to idle?
Will find out about this too. Afaik, at present querying the real min from GuC is not possible. As soon as the top half has completed the freq settings are updated so querying will return the new settings even though they might not have been applied by the bottom half yet.
Quoted 17 lines
Quoted 17 lines
However, this bug is not due to these delays but due to GuC requesting the efficient freq instead of the set 'min == max freq' as mentioned above.
If all these delays are confusing we can revert to v1 of the patch:
https://patchwork.freedesktop.org/patch/510304/?series=110574&rev=1
v1 is simple and does not introduce these delays but just reads the requested freq from sysfs after the 500 ms PMU measurement but while the spinner is still running (which we do in v3 too). Without load SLPC will only request RPn so to compare with PMU measurement we must read the requested freq from sysfs with the spinner load. v1 is sufficient to fix the bug but I thought to be completely safe it is better to add
wait_freq_set()after changing min/max freq (and hence v3).
Quoted 7 lines
Quoted 1 line
I hope this clarifies things somewhat. If not please let me know.
It clarifies a large chunk thank you.
Another question which you missed is why is it needed for the max mode? There we allow a 15% error (more than 5% default). Is that not enough with SLPC?
Yes I think you are right, should not be needed for the max mode. The real issue is that efficient freq can exceed the max freq set but I think that should not be an issue if the max is RP0.
Thanks.
--
Ashutosh
Quoted 61 lines
Quoted 61 lines
Quoted 61 lines
Quoted 18 lines
tests/i915/perf_pmu.c +3
-3igt_spin_free (gem_fd, spin);gem_quiescent_gpu (gem_fd);/* Don't leak busy bo into the next phase */ @@ -1597,11 +1602,11 @@ test_frequency(int gem_fd)* Set GPU to max frequency and read PMU counters. */ igt_require (igt_sysfs_set_u32 (sysfs,"gt_max_freq_mhz" , max_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == max_freq);igt_require (igt_sysfs_set_u32 (sysfs,"gt_boost_freq_mhz" , boost_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_boost_freq_mhz" ) == boost_freq); -igt_require (igt_sysfs_set_u32 (sysfs,"gt_min_freq_mhz" , max_freq)); +wait_freq_set (); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == max_freq); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_boost_freq_mhz" ) == boost_freq);igt_require (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) == max_freq);gem_quiescent_gpu (gem_fd);tests/i915/perf_pmu.c test_frequency(int gem_fd) · line 1620
Quoted 3 lines
tests/i915/perf_pmu.c +1
-0@@ -1613,6 +1618,7 @@ test_frequency(int gem_fd) max[0 ] =1e9 *(val[0 ] - start[0 ]) / slept; max[1 ] =1e9 *(val[1 ] - start[1 ]) / slept; + max_req =igt_sysfs_get_u32 (sysfs,"gt_cur_freq_mhz" );For the max freq use case we had 15% downward tolerance to account for thermally challenged.
a) Is that not enough with SLPC? Like there is a new "failure" mode?
b) Is the extra 10% needed with the sysfs readback added? Aka do we need both?
Quoted 11 lines
tests/i915/perf_pmu.c +1
-0igt_spin_free (gem_fd, spin);gem_quiescent_gpu (gem_fd);@@ -1621,6 +1627,7 @@ test_frequency(int gem_fd)* Restore min/max. */ igt_sysfs_set_u32 (sysfs,"gt_min_freq_mhz" , min_freq); +wait_freq_set ();if (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) != min_freq)igt_warn ("Unable to restore min frequency to saved value [%u MHz], now %u MHz\n" , min_freq,igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ));tests/i915/perf_pmu.c test_frequency(int gem_fd) · line 1648
Quoted 10 lines
tests/i915/perf_pmu.c +2
-1@@ -1633,12 +1640,12 @@ test_frequency(int gem_fd)igt_info ("Max frequency: requested %.1f, actual %.1f\n" , max[0 ], max[1 ]); -assert_within_epsilon (min[0 ], min_freq, tolerance); +assert_within_epsilon (min[0 ], min_req, tolerance);/* * On thermally throttled devices we cannot be sure maximum frequency * can be reached so use larger tolerance downards. */ -__assert_within_epsilon (max[0 ], max_freq, tolerance,0.15f ); +__assert_within_epsilon (max[0 ], max_req, tolerance,0.15f );This 0.15f is what I was talking about above.
Regards,
Tvrtko
Quoted 2 lines
tests/i915/perf_pmu.c +0
-0} static void
Quoted 110 lines · Ashutosh
Quoted 5 lines · Tvrtko Ursulin
Quoted 3 lines · Ashutosh
Quoted 1 line · Tvrtko Ursulin
Hi Tvrtko,
Sorry for the delay in replying on this. We are discussing this but still no conclusion. I have replied what I can below, will update again after I know more. Wanted to send a reply before disappearing for the holidays. Please see below.
Quoted 88 lines
Quoted 88 lines
Hi Tvrtko,
I am only offering an overall clarification below first, not answering all of your points for now.
Quoted 73 lines · Ashutosh Dixit
Quoted 26 lines
With SLPC, even when we set the same min and max freq's, the requested and actual freq's can differ from the min/max freq set. For example "efficient freq" (when in effect) can override set min/max freq. In general FW is the final arbiter in determining freq and can override set values.
Therefore in igt@perf_pmu@frequency subtest, compare the requested freq reported by PMU not against the set freq's but against the requested freq reported in sysfs. Also add a delay after setting freq's to account for messaging delays in setting freq's in GuC.
v2: Introduce a 100 ms delay after setting freq
v3: Update commit message, code identical to v2Cc Vinay Belgaumkar <[email protected]>Cc Tvrtko Ursulin <[email protected]>Signed-off-by Ashutosh Dixit <[email protected]>--- tests/i915/perf_pmu.c | 23 +++++++++++++++-------- 1 file changed, 15 insertions(+), 8 deletions(-)tests/i915/perf_pmu.c +0
-0tests/i915/perf_pmu.c test_interrupts_sync(int gem_fd) · line 1546
Quoted 4 lines
tests/i915/perf_pmu.c +1
-0@@ -1543,10 +1543,13 @@ test_interrupts_sync(int gem_fd)igt_assert_lte (target, busy); } +/* Wait for GuC SLPC freq changes to take effect */ +#define wait_freq_set () usleep(100000)A future task of maybe adding a
drop_cachesflag to wait for this?Alternatively, if we have a sysfs or debugfs file which reflects the value once GuC has processed it, keep reading it until it is updated? Even if to something other than it was before the write, to handle the inability to set the exact requested frequency sometimes?
Quoted 25 lines
tests/i915/perf_pmu.c +5
-3+static void test_frequency (int gem_fd) { -uint32_t min_freq, max_freq, boost_freq; +uint32_t min_freq, max_freq, boost_freq, min_req, max_req;uint64_t val[2 ], start[2 ], slept;double min[2 ], max[2 ];igt_spin_t *spin;@@ -1572,10 +1575,11 @@ test_frequency(int gem_fd)* Set GPU to min frequency and read PMU counters. */ igt_require (igt_sysfs_set_u32 (sysfs,"gt_min_freq_mhz" , min_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) == min_freq);igt_require (igt_sysfs_set_u32 (sysfs,"gt_max_freq_mhz" , min_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == min_freq);igt_require (igt_sysfs_set_u32 (sysfs,"gt_boost_freq_mhz" , min_freq)); +wait_freq_set (); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) == min_freq); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == min_freq);igt_require (igt_sysfs_get_u32 (sysfs,"gt_boost_freq_mhz" ) == min_freq);gem_quiescent_gpu (gem_fd);/* Idle to be sure the change takes effect */
tests/i915/perf_pmu.c test_frequency(int gem_fd) · line 1593
Quoted 3 lines
tests/i915/perf_pmu.c +1
-0@@ -1587,6 +1591,7 @@ test_frequency(int gem_fd) min[0 ] =1e9 *(val[0 ] - start[0 ]) / slept; min[1 ] =1e9 *(val[1 ] - start[1 ]) / slept; + min_req =igt_sysfs_get_u32 (sysfs,"gt_cur_freq_mhz" );This really leaves a bad taste for me still, sorry. First of all, it may make it more palatable if readback was immediately after
wait_freq_set(), or as part of it. Otherwise it seems like even more confused test and sysfs ABI.Quoted 5 lines
Did we close on my question of whether we can make readback of
gt_max_freq_mhzactually represent the real max? Correct me please if I got confused - with min freq checking here we write max=300 and get 350? If so can we makegt_max_freq_mhzreturn 350 after the write of 300? Or return an error?Did you read Vinay's email on v1 here:
https://patchwork.freedesktop.org/patch/510304/?series=110574&rev=1#comment_921571
Quoted 7 lines
Quoted 4 lines
As Vinay says, SLPC is requesting an entirely different freq (the effecient freq rather than the set 'min == max freq') and no amount of waiting will result in the requested freq being the same as the set 'min == max freq'. Also the efficient freq cannot be controlled from user space.
Why is that not a break in our ABI? What is the point in allowing users to write into this field if the desired value can be silently ignored?
Quoted 6 lines
Yes, this is in discussion. FW is doing this and we need to see if FW can be made to obey the i915 ABI. IMHO what matters is performance per watt and if the ABI has to be tossed so be it. I will keep this list posted about what happens here. Merging any patches for these issues (GL #6806 and #6786) is blocked till the ABI breakage issue is resolved one way or another.
Two thoughts here, one has been mentioned before.
New one is that performance per watt perhaps does not matter (is not a primary concern at least) for use cases where "users" (for some definition) tweak the min/max in sysfs. Presumably in those cases user wants control, for some reason, so as long as it is safe and within the limits why not give it to them.
Same argument from a different angle - what is the use case for allowing min/max control which does not really do what it says on the tin?
Old thought is this - if SLPC refuses to be controlled using our existing controls then we could just say -ENODEV on attempts to modify them. Whether for any input value, or some, depending on what SLPC can support (or wants to support).
Important to note I am primarily focusing on the min freq here, which AFAIR we concluded can behave that user writes in 300, but SLPC decides it will only go as low as 350. Because that is different from the existing, should be well known, situation where we can write a max which cannot be achieved due thermal throttling.
Perhaps we need a table, or a summary of some kind, which explains the possible behaviours for both min and max settings.
Quoted 26 lines
Quoted 17 lines
Quoted 17 lines
(So in the example above min and max are set to 300 MHz and SLPC is requesting efficient freq which is 350 or 400 MHz. So min and max values are correct at 300 MHz and requested/actual freq will show 350/400 MHz).
Let me also explain what happens in GuC FW when we change min/max freq:
* i915 receives min/max freq from sysfs and sends a message to GuC to change min/max freq.
* The message itself might be queued behind other messages in the host to GuC queue. i915 is still blocked in the sysfs handler.
* Now when GuC receives the message interrupt, in its top-half handler it validates the freq parameters and sends a message back to host saying the freq change has been accepted. But min/max freq is actually changed later in the bottom-half handler in GuC FW. So the top half only validates the freq value.
* When i915 receives the message from GuC top half, the sysfs handler unblocks and starts reflecting the new min/max freq value.Quoted 8 lines
Quoted 4 lines
So the delay for the new min/max freq values to take effect is between returning from the sysfs handler and when the min/max freq are really set in GuC bottom half (an indeterminate possibly large amount of time if GuC is busy).
Do we have a
drop_cachesflag which would force waiting of idle GuC, which would cover the bottom half as well? Could we or should we add it if not? Would beat the usleep approach..
Quoted 6 lines
Agree, waiting for GuC to idle would do it, but not sure "how" to do it. As I said there is no notification after the bottom half is done applying the freq settings. We may need to schedule something after the freq setting host to guc command and wait for that to be done. A little bit out of my depth on this, let me know if you have any ideas here. I'll ask the GuC guys too. Currently we don't have a
drop_cachesflag for this.
Okay, I was wondering if DROP_IDLE would do it, but perhaps this bottom half can run independently from those flows so it was just a thought, not a claim.
Quoted 7 lines
Quoted 7 lines
Quoted 6 lines
So as you see this introduces an indeterminacy due to to GuC interface not being synchronous. If this needs to be fixed, it needs to be fixed in GuC FW rather than waiting etc. in i915. GuC needs to implement another notification to host /after/ the min/max freq change has been applied in its bottom half. Doing this is not a priority for the GuC team at this point.
Quoted 7 lines
Quoted 2 lines
Is there a method for querying the real min from GuC, which would maybe work after waiting for GuC to idle?
Will find out about this too. Afaik, at present querying the real min from GuC is not possible. As soon as the top half has completed the freq settings are updated so querying will return the new settings even though they might not have been applied by the bottom half yet.
And it will return the written value, or the real one? Going back to the 300 written in, 350 actual min freq example. Presumably 300 and 350 SLPC picks is completely invisible to i915 via the FW interface?
Quoted 19 lines
Quoted 19 lines
Quoted 15 lines
However, this bug is not due to these delays but due to GuC requesting the efficient freq instead of the set 'min == max freq' as mentioned above.
If all these delays are confusing we can revert to v1 of the patch:
https://patchwork.freedesktop.org/patch/510304/?series=110574&rev=1
v1 is simple and does not introduce these delays but just reads the requested freq from sysfs after the 500 ms PMU measurement but while the spinner is still running (which we do in v3 too). Without load SLPC will only request RPn so to compare with PMU measurement we must read the requested freq from sysfs with the spinner load. v1 is sufficient to fix the bug but I thought to be completely safe it is better to add
wait_freq_set()after changing min/max freq (and hence v3).Quoted 1 line
I hope this clarifies things somewhat. If not please let me know.
It clarifies a large chunk thank you.
Quoted 7 lines
Quoted 3 lines
Another question which you missed is why is it needed for the max mode? There we allow a 15% error (more than 5% default). Is that not enough with SLPC?
Yes I think you are right, should not be needed for the max mode. The real issue is that efficient freq can exceed the max freq set but I think that should not be an issue if the max is RP0.
Ah okay, so SLPC not respecting max same as with min. I think the question of what is the use case for min/max controls in the SLPC world is the one to answer then.
Regards,
Tvrtko
Quoted 127 lines · Tvrtko Ursulin
Quoted 110 lines · Ashutosh
Quoted 5 lines · Tvrtko Ursulin
Quoted 3 lines · Ashutosh
Quoted 1 line · Tvrtko Ursulin
Hi Tvrtko,
Sorry for the delay in replying on this. We are discussing this but still no conclusion. I have replied what I can below, will update again after I know more. Wanted to send a reply before disappearing for the holidays. Please see below.
Quoted 88 lines
Quoted 88 lines
Hi Tvrtko,
I am only offering an overall clarification below first, not answering all of your points for now.
Quoted 73 lines · Ashutosh Dixit
Quoted 26 lines
With SLPC, even when we set the same min and max freq's, the requested and actual freq's can differ from the min/max freq set. For example "efficient freq" (when in effect) can override set min/max freq. In general FW is the final arbiter in determining freq and can override set values.
Therefore in igt@perf_pmu@frequency subtest, compare the requested freq reported by PMU not against the set freq's but against the requested freq reported in sysfs. Also add a delay after setting freq's to account for messaging delays in setting freq's in GuC.
v2: Introduce a 100 ms delay after setting freq
v3: Update commit message, code identical to v2Cc Vinay Belgaumkar <[email protected]>Cc Tvrtko Ursulin <[email protected]>Signed-off-by Ashutosh Dixit <[email protected]>--- tests/i915/perf_pmu.c | 23 +++++++++++++++-------- 1 file changed, 15 insertions(+), 8 deletions(-)tests/i915/perf_pmu.c +0
-0tests/i915/perf_pmu.c test_interrupts_sync(int gem_fd) · line 1546
Quoted 4 lines
tests/i915/perf_pmu.c +1
-0@@ -1543,10 +1543,13 @@ test_interrupts_sync(int gem_fd)igt_assert_lte (target, busy); } +/* Wait for GuC SLPC freq changes to take effect */ +#define wait_freq_set () usleep(100000)A future task of maybe adding a
drop_cachesflag to wait for this?Alternatively, if we have a sysfs or debugfs file which reflects the value once GuC has processed it, keep reading it until it is updated? Even if to something other than it was before the write, to handle the inability to set the exact requested frequency sometimes?
Quoted 25 lines
tests/i915/perf_pmu.c +5
-3+static void test_frequency (int gem_fd) { -uint32_t min_freq, max_freq, boost_freq; +uint32_t min_freq, max_freq, boost_freq, min_req, max_req;uint64_t val[2 ], start[2 ], slept;double min[2 ], max[2 ];igt_spin_t *spin;@@ -1572,10 +1575,11 @@ test_frequency(int gem_fd)* Set GPU to min frequency and read PMU counters. */ igt_require (igt_sysfs_set_u32 (sysfs,"gt_min_freq_mhz" , min_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) == min_freq);igt_require (igt_sysfs_set_u32 (sysfs,"gt_max_freq_mhz" , min_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == min_freq);igt_require (igt_sysfs_set_u32 (sysfs,"gt_boost_freq_mhz" , min_freq)); +wait_freq_set (); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) == min_freq); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == min_freq);igt_require (igt_sysfs_get_u32 (sysfs,"gt_boost_freq_mhz" ) == min_freq);gem_quiescent_gpu (gem_fd);/* Idle to be sure the change takes effect */
tests/i915/perf_pmu.c test_frequency(int gem_fd) · line 1593
Quoted 3 lines
tests/i915/perf_pmu.c +1
-0@@ -1587,6 +1591,7 @@ test_frequency(int gem_fd) min[0 ] =1e9 *(val[0 ] - start[0 ]) / slept; min[1 ] =1e9 *(val[1 ] - start[1 ]) / slept; + min_req =igt_sysfs_get_u32 (sysfs,"gt_cur_freq_mhz" );This really leaves a bad taste for me still, sorry. First of all, it may make it more palatable if readback was immediately after
wait_freq_set(), or as part of it. Otherwise it seems like even more confused test and sysfs ABI.Quoted 5 lines
Did we close on my question of whether we can make readback of
gt_max_freq_mhzactually represent the real max? Correct me please if I got confused - with min freq checking here we write max=300 and get 350? If so can we makegt_max_freq_mhzreturn 350 after the write of 300? Or return an error?Did you read Vinay's email on v1 here:
https://patchwork.freedesktop.org/patch/510304/?series=110574&rev=1#comment_921571
Quoted 7 lines
Quoted 4 lines
As Vinay says, SLPC is requesting an entirely different freq (the effecient freq rather than the set 'min == max freq') and no amount of waiting will result in the requested freq being the same as the set 'min == max freq'. Also the efficient freq cannot be controlled from user space.
Why is that not a break in our ABI? What is the point in allowing users to write into this field if the desired value can be silently ignored?
Quoted 6 lines
Yes, this is in discussion. FW is doing this and we need to see if FW can be made to obey the i915 ABI. IMHO what matters is performance per watt and if the ABI has to be tossed so be it. I will keep this list posted about what happens here. Merging any patches for these issues (GL #6806 and #6786) is blocked till the ABI breakage issue is resolved one way or another.
Two thoughts here, one has been mentioned before.
New one is that performance per watt perhaps does not matter (is not a primary concern at least) for use cases where "users" (for some definition) tweak the min/max in sysfs. Presumably in those cases user wants control, for some reason, so as long as it is safe and within the limits why not give it to them.
Quoted 7 lines
Same argument from a different angle - what is the use case for allowing min/max control which does not really do what it says on the tin?
Old thought is this - if SLPC refuses to be controlled using our existing controls then we could just say
-ENODEVon attempts to modify them. Whether for any input value, or some, depending on what SLPC can support (or wants to support).
The problem is not just SLPC. The freq is ultimately decided by PCODE and it
has sudo and all the rights to disrespect what drivers (including SLPC as a a driver from PCODE perspective) are requesting.
Quoted 4 lines
Important to note I am primarily focusing on the min freq here, which AFAIR we concluded can behave that user writes in 300, but SLPC decides it will only go as low as 350.
The other caveat here with the efficient freq is that it is dynamic and PCODE decides it with runtime conditions. It may start with 300, then move to 350, then back to 300. i915 is only stashing the initial value.
We could potentially update our rpe everytime any freq control is touched.
Just to highlight how bad this is, I've seen cases where rpe in runtime goes below the minimal fused freq (our rpn).
Quoted 3 lines
Because that is different from the existing, should
be well known, situation where we can write a max which cannot be achieved due thermal throttling.
yeap, usually when things got disrespected by PCODE there will be some thermal throttling information. But I have seen cases where this indicators simply do not exist. What complicate the api expectation to a very bad level.
Quoted 3 lines
Perhaps we need a table, or a summary of some kind, which explains the possible behaviours for both min and max settings.
This is probably a good idea. At least document everything and what and when things can be disrespected, etc...
(and update our rpe info dynamically)
Quoted 91 lines
Quoted 27 lines
Quoted 18 lines
Quoted 18 lines
(So in the example above min and max are set to 300 MHz and SLPC is requesting efficient freq which is 350 or 400 MHz. So min and max values are correct at 300 MHz and requested/actual freq will show 350/400 MHz).
Let me also explain what happens in GuC FW when we change min/max freq:
* i915 receives min/max freq from sysfs and sends a message to GuC to change min/max freq.
* The message itself might be queued behind other messages in the host to GuC queue. i915 is still blocked in the sysfs handler.
* Now when GuC receives the message interrupt, in its top-half handler it validates the freq parameters and sends a message back to host saying the freq change has been accepted. But min/max freq is actually changed later in the bottom-half handler in GuC FW. So the top half only validates the freq value.
* When i915 receives the message from GuC top half, the sysfs handler unblocks and starts reflecting the new min/max freq value.Quoted 8 lines
Quoted 4 lines
So the delay for the new min/max freq values to take effect is between returning from the sysfs handler and when the min/max freq are really set in GuC bottom half (an indeterminate possibly large amount of time if GuC is busy).
Do we have a
drop_cachesflag which would force waiting of idle GuC, which would cover the bottom half as well? Could we or should we add it if not? Would beat the usleep approach..Quoted 6 lines
Agree, waiting for GuC to idle would do it, but not sure "how" to do it. As I said there is no notification after the bottom half is done applying the freq settings. We may need to schedule something after the freq setting host to guc command and wait for that to be done. A little bit out of my depth on this, let me know if you have any ideas here. I'll ask the GuC guys too. Currently we don't have a
drop_cachesflag for this.Okay, I was wondering if
DROP_IDLEwould do it, but perhaps this bottom half can run independently from those flows so it was just a thought, not a claim.Quoted 7 lines
Quoted 7 lines
Quoted 6 lines
So as you see this introduces an indeterminacy due to to GuC interface not being synchronous. If this needs to be fixed, it needs to be fixed in GuC FW rather than waiting etc. in i915. GuC needs to implement another notification to host /after/ the min/max freq change has been applied in its bottom half. Doing this is not a priority for the GuC team at this point.
Quoted 7 lines
Quoted 2 lines
Is there a method for querying the real min from GuC, which would maybe work after waiting for GuC to idle?
Will find out about this too. Afaik, at present querying the real min from GuC is not possible. As soon as the top half has completed the freq settings are updated so querying will return the new settings even though they might not have been applied by the bottom half yet.
And it will return the written value, or the real one? Going back to the 300 written in, 350 actual min freq example. Presumably 300 and 350 SLPC picks is completely invisible to i915 via the FW interface?
Quoted 19 lines
Quoted 19 lines
Quoted 15 lines
However, this bug is not due to these delays but due to GuC requesting the efficient freq instead of the set 'min == max freq' as mentioned above.
If all these delays are confusing we can revert to v1 of the patch:
https://patchwork.freedesktop.org/patch/510304/?series=110574&rev=1
v1 is simple and does not introduce these delays but just reads the requested freq from sysfs after the 500 ms PMU measurement but while the spinner is still running (which we do in v3 too). Without load SLPC will only request RPn so to compare with PMU measurement we must read the requested freq from sysfs with the spinner load. v1 is sufficient to fix the bug but I thought to be completely safe it is better to add
wait_freq_set()after changing min/max freq (and hence v3).Quoted 1 line
I hope this clarifies things somewhat. If not please let me know.
It clarifies a large chunk thank you.
Quoted 7 lines
Quoted 3 lines
Another question which you missed is why is it needed for the max mode? There we allow a 15% error (more than 5% default). Is that not enough with SLPC?
Yes I think you are right, should not be needed for the max mode. The real issue is that efficient freq can exceed the max freq set but I think that should not be an issue if the max is RP0.
Ah okay, so SLPC not respecting max same as with min. I think the question of what is the use case for min/max controls in the SLPC world is the one to answer then.
Regards,
Tvrtko
Quoted 145 lines · Rodrigo Vivi
Quoted 127 lines · Tvrtko Ursulin
Quoted 110 lines · Ashutosh
Quoted 5 lines · Tvrtko Ursulin
Quoted 3 lines · Ashutosh
Quoted 1 line · Tvrtko Ursulin
Hi Tvrtko,
Sorry for the delay in replying on this. We are discussing this but still no conclusion. I have replied what I can below, will update again after I know more. Wanted to send a reply before disappearing for the holidays. Please see below.
Quoted 88 lines
Quoted 88 lines
Hi Tvrtko,
I am only offering an overall clarification below first, not answering all of your points for now.
Quoted 73 lines · Ashutosh Dixit
Quoted 26 lines
With SLPC, even when we set the same min and max freq's, the requested and actual freq's can differ from the min/max freq set. For example "efficient freq" (when in effect) can override set min/max freq. In general FW is the final arbiter in determining freq and can override set values.
Therefore in igt@perf_pmu@frequency subtest, compare the requested freq reported by PMU not against the set freq's but against the requested freq reported in sysfs. Also add a delay after setting freq's to account for messaging delays in setting freq's in GuC.
v2: Introduce a 100 ms delay after setting freq
v3: Update commit message, code identical to v2Cc Vinay Belgaumkar <[email protected]>Cc Tvrtko Ursulin <[email protected]>Signed-off-by Ashutosh Dixit <[email protected]>--- tests/i915/perf_pmu.c | 23 +++++++++++++++-------- 1 file changed, 15 insertions(+), 8 deletions(-)tests/i915/perf_pmu.c +0
-0tests/i915/perf_pmu.c test_interrupts_sync(int gem_fd) · line 1546
Quoted 4 lines
tests/i915/perf_pmu.c +1
-0@@ -1543,10 +1543,13 @@ test_interrupts_sync(int gem_fd)igt_assert_lte (target, busy); } +/* Wait for GuC SLPC freq changes to take effect */ +#define wait_freq_set () usleep(100000)A future task of maybe adding a
drop_cachesflag to wait for this?Alternatively, if we have a sysfs or debugfs file which reflects the value once GuC has processed it, keep reading it until it is updated? Even if to something other than it was before the write, to handle the inability to set the exact requested frequency sometimes?
Quoted 25 lines
tests/i915/perf_pmu.c +5
-3+static void test_frequency (int gem_fd) { -uint32_t min_freq, max_freq, boost_freq; +uint32_t min_freq, max_freq, boost_freq, min_req, max_req;uint64_t val[2 ], start[2 ], slept;double min[2 ], max[2 ];igt_spin_t *spin;@@ -1572,10 +1575,11 @@ test_frequency(int gem_fd)* Set GPU to min frequency and read PMU counters. */ igt_require (igt_sysfs_set_u32 (sysfs,"gt_min_freq_mhz" , min_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) == min_freq);igt_require (igt_sysfs_set_u32 (sysfs,"gt_max_freq_mhz" , min_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == min_freq);igt_require (igt_sysfs_set_u32 (sysfs,"gt_boost_freq_mhz" , min_freq)); +wait_freq_set (); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) == min_freq); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == min_freq);igt_require (igt_sysfs_get_u32 (sysfs,"gt_boost_freq_mhz" ) == min_freq);gem_quiescent_gpu (gem_fd);/* Idle to be sure the change takes effect */
tests/i915/perf_pmu.c test_frequency(int gem_fd) · line 1593
Quoted 3 lines
tests/i915/perf_pmu.c +1
-0@@ -1587,6 +1591,7 @@ test_frequency(int gem_fd) min[0 ] =1e9 *(val[0 ] - start[0 ]) / slept; min[1 ] =1e9 *(val[1 ] - start[1 ]) / slept; + min_req =igt_sysfs_get_u32 (sysfs,"gt_cur_freq_mhz" );This really leaves a bad taste for me still, sorry. First of all, it may make it more palatable if readback was immediately after
wait_freq_set(), or as part of it. Otherwise it seems like even more confused test and sysfs ABI.Quoted 5 lines
Did we close on my question of whether we can make readback of
gt_max_freq_mhzactually represent the real max? Correct me please if I got confused - with min freq checking here we write max=300 and get 350? If so can we makegt_max_freq_mhzreturn 350 after the write of 300? Or return an error?Did you read Vinay's email on v1 here:
https://patchwork.freedesktop.org/patch/510304/?series=110574&rev=1#comment_921571
Quoted 7 lines
Quoted 4 lines
As Vinay says, SLPC is requesting an entirely different freq (the effecient freq rather than the set 'min == max freq') and no amount of waiting will result in the requested freq being the same as the set 'min == max freq'. Also the efficient freq cannot be controlled from user space.
Why is that not a break in our ABI? What is the point in allowing users to write into this field if the desired value can be silently ignored?
Quoted 6 lines
Yes, this is in discussion. FW is doing this and we need to see if FW can be made to obey the i915 ABI. IMHO what matters is performance per watt and if the ABI has to be tossed so be it. I will keep this list posted about what happens here. Merging any patches for these issues (GL #6806 and #6786) is blocked till the ABI breakage issue is resolved one way or another.
Two thoughts here, one has been mentioned before.
New one is that performance per watt perhaps does not matter (is not a primary concern at least) for use cases where "users" (for some definition) tweak the min/max in sysfs. Presumably in those cases user wants control, for some reason, so as long as it is safe and within the limits why not give it to them.
Quoted 7 lines
Same argument from a different angle - what is the use case for allowing min/max control which does not really do what it says on the tin?
Old thought is this - if SLPC refuses to be controlled using our existing controls then we could just say
-ENODEVon attempts to modify them. Whether for any input value, or some, depending on what SLPC can support (or wants to support).The problem is not just SLPC. The freq is ultimately decided by PCODE and it
has sudo and all the rights to disrespect what drivers (including SLPC as a a driver from PCODE perspective) are requesting.Quoted 4 lines
Important to note I am primarily focusing on the min freq here, which AFAIR we concluded can behave that user writes in 300, but SLPC decides it will only go as low as 350.
Quoted 6 lines
The other caveat here with the efficient freq is that it is dynamic and PCODE decides it with runtime conditions. It may start with 300, then move to 350, then back to 300. i915 is only stashing the initial value.
We could potentially update our rpe everytime any freq control is touched.
We thought of doing this but two points:
1. i915 now reads the same register as GuC for rpe but I never see the value in this register change.
2. Even if we could know the real rpe, the freq's might be set before a workload start running and rpe might change after the workload starts running. So knowing rpe before the workload might not be very useful.
I am of the opposite opinion that anyway the previous kernel ABI is probably now irrecoverably bust, so we should give FW free reign and stop disabling efficient freq the way we do now when min_freq < rpe_freq.
Cheers.
--
Ashutosh
Quoted 111 lines
Just to highlight how bad this is, I've seen cases where rpe in runtime goes below the minimal fused freq (our rpn).
Quoted 3 lines
Because that is different from the existing, should
be well known, situation where we can write a max which cannot be achieved due thermal throttling.yeap, usually when things got disrespected by PCODE there will be some thermal throttling information. But I have seen cases where this indicators simply do not exist. What complicate the api expectation to a very bad level.
Quoted 3 lines
Perhaps we need a table, or a summary of some kind, which explains the possible behaviours for both min and max settings.
This is probably a good idea. At least document everything and what and when things can be disrespected, etc...
(and update our rpe info dynamically)
Quoted 91 lines
Quoted 27 lines
Quoted 18 lines
Quoted 18 lines
(So in the example above min and max are set to 300 MHz and SLPC is requesting efficient freq which is 350 or 400 MHz. So min and max values are correct at 300 MHz and requested/actual freq will show 350/400 MHz).
Let me also explain what happens in GuC FW when we change min/max freq:
* i915 receives min/max freq from sysfs and sends a message to GuC to change min/max freq.
* The message itself might be queued behind other messages in the host to GuC queue. i915 is still blocked in the sysfs handler.
* Now when GuC receives the message interrupt, in its top-half handler it validates the freq parameters and sends a message back to host saying the freq change has been accepted. But min/max freq is actually changed later in the bottom-half handler in GuC FW. So the top half only validates the freq value.
* When i915 receives the message from GuC top half, the sysfs handler unblocks and starts reflecting the new min/max freq value.Quoted 8 lines
Quoted 4 lines
So the delay for the new min/max freq values to take effect is between returning from the sysfs handler and when the min/max freq are really set in GuC bottom half (an indeterminate possibly large amount of time if GuC is busy).
Do we have a
drop_cachesflag which would force waiting of idle GuC, which would cover the bottom half as well? Could we or should we add it if not? Would beat the usleep approach..Quoted 6 lines
Agree, waiting for GuC to idle would do it, but not sure "how" to do it. As I said there is no notification after the bottom half is done applying the freq settings. We may need to schedule something after the freq setting host to guc command and wait for that to be done. A little bit out of my depth on this, let me know if you have any ideas here. I'll ask the GuC guys too. Currently we don't have a
drop_cachesflag for this.Okay, I was wondering if
DROP_IDLEwould do it, but perhaps this bottom half can run independently from those flows so it was just a thought, not a claim.Quoted 7 lines
Quoted 7 lines
Quoted 6 lines
So as you see this introduces an indeterminacy due to to GuC interface not being synchronous. If this needs to be fixed, it needs to be fixed in GuC FW rather than waiting etc. in i915. GuC needs to implement another notification to host /after/ the min/max freq change has been applied in its bottom half. Doing this is not a priority for the GuC team at this point.
Quoted 7 lines
Quoted 2 lines
Is there a method for querying the real min from GuC, which would maybe work after waiting for GuC to idle?
Will find out about this too. Afaik, at present querying the real min from GuC is not possible. As soon as the top half has completed the freq settings are updated so querying will return the new settings even though they might not have been applied by the bottom half yet.
And it will return the written value, or the real one? Going back to the 300 written in, 350 actual min freq example. Presumably 300 and 350 SLPC picks is completely invisible to i915 via the FW interface?
Quoted 19 lines
Quoted 19 lines
Quoted 15 lines
However, this bug is not due to these delays but due to GuC requesting the efficient freq instead of the set 'min == max freq' as mentioned above.
If all these delays are confusing we can revert to v1 of the patch:
https://patchwork.freedesktop.org/patch/510304/?series=110574&rev=1
v1 is simple and does not introduce these delays but just reads the requested freq from sysfs after the 500 ms PMU measurement but while the spinner is still running (which we do in v3 too). Without load SLPC will only request RPn so to compare with PMU measurement we must read the requested freq from sysfs with the spinner load. v1 is sufficient to fix the bug but I thought to be completely safe it is better to add
wait_freq_set()after changing min/max freq (and hence v3).Quoted 1 line
I hope this clarifies things somewhat. If not please let me know.
It clarifies a large chunk thank you.
Quoted 7 lines
Quoted 3 lines
Another question which you missed is why is it needed for the max mode? There we allow a 15% error (more than 5% default). Is that not enough with SLPC?
Yes I think you are right, should not be needed for the max mode. The real issue is that efficient freq can exceed the max freq set but I think that should not be an issue if the max is RP0.
Ah okay, so SLPC not respecting max same as with min. I think the question of what is the use case for min/max controls in the SLPC world is the one to answer then.
Regards,
Tvrtko
Quoted 161 lines · Ashutosh
Quoted 145 lines · Rodrigo Vivi
Quoted 127 lines · Tvrtko Ursulin
Quoted 110 lines · Ashutosh
Quoted 5 lines · Tvrtko Ursulin
Quoted 3 lines · Ashutosh
Quoted 1 line · Tvrtko Ursulin
Hi Tvrtko,
Sorry for the delay in replying on this. We are discussing this but still no conclusion. I have replied what I can below, will update again after I know more. Wanted to send a reply before disappearing for the holidays. Please see below.
Quoted 88 lines
Quoted 88 lines
Hi Tvrtko,
I am only offering an overall clarification below first, not answering all of your points for now.
Quoted 73 lines · Ashutosh Dixit
Quoted 26 lines
With SLPC, even when we set the same min and max freq's, the requested and actual freq's can differ from the min/max freq set. For example "efficient freq" (when in effect) can override set min/max freq. In general FW is the final arbiter in determining freq and can override set values.
Therefore in igt@perf_pmu@frequency subtest, compare the requested freq reported by PMU not against the set freq's but against the requested freq reported in sysfs. Also add a delay after setting freq's to account for messaging delays in setting freq's in GuC.
v2: Introduce a 100 ms delay after setting freq
v3: Update commit message, code identical to v2Cc Vinay Belgaumkar <[email protected]>Cc Tvrtko Ursulin <[email protected]>Signed-off-by Ashutosh Dixit <[email protected]>--- tests/i915/perf_pmu.c | 23 +++++++++++++++-------- 1 file changed, 15 insertions(+), 8 deletions(-)tests/i915/perf_pmu.c +0
-0tests/i915/perf_pmu.c test_interrupts_sync(int gem_fd) · line 1546
Quoted 4 lines
tests/i915/perf_pmu.c +1
-0@@ -1543,10 +1543,13 @@ test_interrupts_sync(int gem_fd)igt_assert_lte (target, busy); } +/* Wait for GuC SLPC freq changes to take effect */ +#define wait_freq_set () usleep(100000)A future task of maybe adding a
drop_cachesflag to wait for this?Alternatively, if we have a sysfs or debugfs file which reflects the value once GuC has processed it, keep reading it until it is updated? Even if to something other than it was before the write, to handle the inability to set the exact requested frequency sometimes?
Quoted 25 lines
tests/i915/perf_pmu.c +5
-3+static void test_frequency (int gem_fd) { -uint32_t min_freq, max_freq, boost_freq; +uint32_t min_freq, max_freq, boost_freq, min_req, max_req;uint64_t val[2 ], start[2 ], slept;double min[2 ], max[2 ];igt_spin_t *spin;@@ -1572,10 +1575,11 @@ test_frequency(int gem_fd)* Set GPU to min frequency and read PMU counters. */ igt_require (igt_sysfs_set_u32 (sysfs,"gt_min_freq_mhz" , min_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) == min_freq);igt_require (igt_sysfs_set_u32 (sysfs,"gt_max_freq_mhz" , min_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == min_freq);igt_require (igt_sysfs_set_u32 (sysfs,"gt_boost_freq_mhz" , min_freq)); +wait_freq_set (); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) == min_freq); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == min_freq);igt_require (igt_sysfs_get_u32 (sysfs,"gt_boost_freq_mhz" ) == min_freq);gem_quiescent_gpu (gem_fd);/* Idle to be sure the change takes effect */
tests/i915/perf_pmu.c test_frequency(int gem_fd) · line 1593
Quoted 3 lines
tests/i915/perf_pmu.c +1
-0@@ -1587,6 +1591,7 @@ test_frequency(int gem_fd) min[0 ] =1e9 *(val[0 ] - start[0 ]) / slept; min[1 ] =1e9 *(val[1 ] - start[1 ]) / slept; + min_req =igt_sysfs_get_u32 (sysfs,"gt_cur_freq_mhz" );This really leaves a bad taste for me still, sorry. First of all, it may make it more palatable if readback was immediately after
wait_freq_set(), or as part of it. Otherwise it seems like even more confused test and sysfs ABI.Quoted 5 lines
Did we close on my question of whether we can make readback of
gt_max_freq_mhzactually represent the real max? Correct me please if I got confused - with min freq checking here we write max=300 and get 350? If so can we makegt_max_freq_mhzreturn 350 after the write of 300? Or return an error?Did you read Vinay's email on v1 here:
https://patchwork.freedesktop.org/patch/510304/?series=110574&rev=1#comment_921571
Quoted 7 lines
Quoted 4 lines
As Vinay says, SLPC is requesting an entirely different freq (the effecient freq rather than the set 'min == max freq') and no amount of waiting will result in the requested freq being the same as the set 'min == max freq'. Also the efficient freq cannot be controlled from user space.
Why is that not a break in our ABI? What is the point in allowing users to write into this field if the desired value can be silently ignored?
Quoted 6 lines
Yes, this is in discussion. FW is doing this and we need to see if FW can be made to obey the i915 ABI. IMHO what matters is performance per watt and if the ABI has to be tossed so be it. I will keep this list posted about what happens here. Merging any patches for these issues (GL #6806 and #6786) is blocked till the ABI breakage issue is resolved one way or another.
Two thoughts here, one has been mentioned before.
New one is that performance per watt perhaps does not matter (is not a primary concern at least) for use cases where "users" (for some definition) tweak the min/max in sysfs. Presumably in those cases user wants control, for some reason, so as long as it is safe and within the limits why not give it to them.
Quoted 7 lines
Same argument from a different angle - what is the use case for allowing min/max control which does not really do what it says on the tin?
Old thought is this - if SLPC refuses to be controlled using our existing controls then we could just say
-ENODEVon attempts to modify them. Whether for any input value, or some, depending on what SLPC can support (or wants to support).The problem is not just SLPC. The freq is ultimately decided by PCODE and it
has sudo and all the rights to disrespect what drivers (including SLPC as a a driver from PCODE perspective) are requesting.Quoted 4 lines
Important to note I am primarily focusing on the min freq here, which AFAIR we concluded can behave that user writes in 300, but SLPC decides it will only go as low as 350.
Quoted 6 lines
The other caveat here with the efficient freq is that it is dynamic and PCODE decides it with runtime conditions. It may start with 300, then move to 350, then back to 300. i915 is only stashing the initial value.
We could potentially update our rpe everytime any freq control is touched.
We thought of doing this but two points:
1. i915 now reads the same register as GuC for rpe but I never see the value in this register change.
2. Even if we could know the real rpe, the freq's might be set before a workload start running and rpe might change after the workload starts running. So knowing rpe before the workload might not be very useful.
Quoted 3 lines
I am of the opposite opinion that anyway the previous kernel ABI is probably now irrecoverably bust, so we should give FW free reign and stop disabling efficient freq the way we do now when
min_freq<rpe_freq.
We have to answer the question of why we need to keep exposing the sysfs controls if they don't/can't/won't do what they say on the tin. Could we make a break with DG2 for instance? Some use cases might suffer but it's probably better than having them think it does what they think it does, or file bugs when it does not.
Regards,
Tvrtko
Quoted 116 lines
Cheers.
--
AshutoshQuoted 111 lines
Just to highlight how bad this is, I've seen cases where rpe in runtime goes below the minimal fused freq (our rpn).
Quoted 3 lines
Because that is different from the existing, should
be well known, situation where we can write a max which cannot be achieved due thermal throttling.yeap, usually when things got disrespected by PCODE there will be some thermal throttling information. But I have seen cases where this indicators simply do not exist. What complicate the api expectation to a very bad level.
Quoted 3 lines
Perhaps we need a table, or a summary of some kind, which explains the possible behaviours for both min and max settings.
This is probably a good idea. At least document everything and what and when things can be disrespected, etc...
(and update our rpe info dynamically)
Quoted 91 lines
Quoted 27 lines
Quoted 18 lines
Quoted 18 lines
(So in the example above min and max are set to 300 MHz and SLPC is requesting efficient freq which is 350 or 400 MHz. So min and max values are correct at 300 MHz and requested/actual freq will show 350/400 MHz).
Let me also explain what happens in GuC FW when we change min/max freq:
* i915 receives min/max freq from sysfs and sends a message to GuC to change min/max freq.
* The message itself might be queued behind other messages in the host to GuC queue. i915 is still blocked in the sysfs handler.
* Now when GuC receives the message interrupt, in its top-half handler it validates the freq parameters and sends a message back to host saying the freq change has been accepted. But min/max freq is actually changed later in the bottom-half handler in GuC FW. So the top half only validates the freq value.
* When i915 receives the message from GuC top half, the sysfs handler unblocks and starts reflecting the new min/max freq value.Quoted 8 lines
Quoted 4 lines
So the delay for the new min/max freq values to take effect is between returning from the sysfs handler and when the min/max freq are really set in GuC bottom half (an indeterminate possibly large amount of time if GuC is busy).
Do we have a
drop_cachesflag which would force waiting of idle GuC, which would cover the bottom half as well? Could we or should we add it if not? Would beat the usleep approach..Quoted 6 lines
Agree, waiting for GuC to idle would do it, but not sure "how" to do it. As I said there is no notification after the bottom half is done applying the freq settings. We may need to schedule something after the freq setting host to guc command and wait for that to be done. A little bit out of my depth on this, let me know if you have any ideas here. I'll ask the GuC guys too. Currently we don't have a
drop_cachesflag for this.Okay, I was wondering if
DROP_IDLEwould do it, but perhaps this bottom half can run independently from those flows so it was just a thought, not a claim.Quoted 7 lines
Quoted 7 lines
Quoted 6 lines
So as you see this introduces an indeterminacy due to to GuC interface not being synchronous. If this needs to be fixed, it needs to be fixed in GuC FW rather than waiting etc. in i915. GuC needs to implement another notification to host /after/ the min/max freq change has been applied in its bottom half. Doing this is not a priority for the GuC team at this point.
Quoted 7 lines
Quoted 2 lines
Is there a method for querying the real min from GuC, which would maybe work after waiting for GuC to idle?
Will find out about this too. Afaik, at present querying the real min from GuC is not possible. As soon as the top half has completed the freq settings are updated so querying will return the new settings even though they might not have been applied by the bottom half yet.
And it will return the written value, or the real one? Going back to the 300 written in, 350 actual min freq example. Presumably 300 and 350 SLPC picks is completely invisible to i915 via the FW interface?
Quoted 19 lines
Quoted 19 lines
Quoted 15 lines
However, this bug is not due to these delays but due to GuC requesting the efficient freq instead of the set 'min == max freq' as mentioned above.
If all these delays are confusing we can revert to v1 of the patch:
https://patchwork.freedesktop.org/patch/510304/?series=110574&rev=1
v1 is simple and does not introduce these delays but just reads the requested freq from sysfs after the 500 ms PMU measurement but while the spinner is still running (which we do in v3 too). Without load SLPC will only request RPn so to compare with PMU measurement we must read the requested freq from sysfs with the spinner load. v1 is sufficient to fix the bug but I thought to be completely safe it is better to add
wait_freq_set()after changing min/max freq (and hence v3).Quoted 1 line
I hope this clarifies things somewhat. If not please let me know.
It clarifies a large chunk thank you.
Quoted 7 lines
Quoted 3 lines
Another question which you missed is why is it needed for the max mode? There we allow a 15% error (more than 5% default). Is that not enough with SLPC?
Yes I think you are right, should not be needed for the max mode. The real issue is that efficient freq can exceed the max freq set but I think that should not be an issue if the max is RP0.
Ah okay, so SLPC not respecting max same as with min. I think the question of what is the use case for min/max controls in the SLPC world is the one to answer then.
Regards,
Tvrtko
Quoted 167 lines · Tvrtko Ursulin
Quoted 161 lines · Ashutosh
Quoted 145 lines · Rodrigo Vivi
Quoted 127 lines · Tvrtko Ursulin
Quoted 110 lines · Ashutosh
Quoted 5 lines · Tvrtko Ursulin
Quoted 3 lines · Ashutosh
Quoted 1 line · Tvrtko Ursulin
Hi Tvrtko,
Sorry for the delay in replying on this. We are discussing this but still no conclusion. I have replied what I can below, will update again after I know more. Wanted to send a reply before disappearing for the holidays. Please see below.
Quoted 88 lines
Quoted 88 lines
Hi Tvrtko,
I am only offering an overall clarification below first, not answering all of your points for now.
Quoted 73 lines · Ashutosh Dixit
Quoted 26 lines
With SLPC, even when we set the same min and max freq's, the requested and actual freq's can differ from the min/max freq set. For example "efficient freq" (when in effect) can override set min/max freq. In general FW is the final arbiter in determining freq and can override set values.
Therefore in igt@perf_pmu@frequency subtest, compare the requested freq reported by PMU not against the set freq's but against the requested freq reported in sysfs. Also add a delay after setting freq's to account for messaging delays in setting freq's in GuC.
v2: Introduce a 100 ms delay after setting freq
v3: Update commit message, code identical to v2Cc Vinay Belgaumkar <[email protected]>Cc Tvrtko Ursulin <[email protected]>Signed-off-by Ashutosh Dixit <[email protected]>--- tests/i915/perf_pmu.c | 23 +++++++++++++++-------- 1 file changed, 15 insertions(+), 8 deletions(-)tests/i915/perf_pmu.c +0
-0tests/i915/perf_pmu.c test_interrupts_sync(int gem_fd) · line 1546
Quoted 4 lines
tests/i915/perf_pmu.c +1
-0@@ -1543,10 +1543,13 @@ test_interrupts_sync(int gem_fd)igt_assert_lte (target, busy); } +/* Wait for GuC SLPC freq changes to take effect */ +#define wait_freq_set () usleep(100000)A future task of maybe adding a
drop_cachesflag to wait for this?Alternatively, if we have a sysfs or debugfs file which reflects the value once GuC has processed it, keep reading it until it is updated? Even if to something other than it was before the write, to handle the inability to set the exact requested frequency sometimes?
Quoted 25 lines
tests/i915/perf_pmu.c +5
-3+static void test_frequency (int gem_fd) { -uint32_t min_freq, max_freq, boost_freq; +uint32_t min_freq, max_freq, boost_freq, min_req, max_req;uint64_t val[2 ], start[2 ], slept;double min[2 ], max[2 ];igt_spin_t *spin;@@ -1572,10 +1575,11 @@ test_frequency(int gem_fd)* Set GPU to min frequency and read PMU counters. */ igt_require (igt_sysfs_set_u32 (sysfs,"gt_min_freq_mhz" , min_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) == min_freq);igt_require (igt_sysfs_set_u32 (sysfs,"gt_max_freq_mhz" , min_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == min_freq);igt_require (igt_sysfs_set_u32 (sysfs,"gt_boost_freq_mhz" , min_freq)); +wait_freq_set (); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) == min_freq); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == min_freq);igt_require (igt_sysfs_get_u32 (sysfs,"gt_boost_freq_mhz" ) == min_freq);gem_quiescent_gpu (gem_fd);/* Idle to be sure the change takes effect */
tests/i915/perf_pmu.c test_frequency(int gem_fd) · line 1593
Quoted 3 lines
tests/i915/perf_pmu.c +1
-0@@ -1587,6 +1591,7 @@ test_frequency(int gem_fd) min[0 ] =1e9 *(val[0 ] - start[0 ]) / slept; min[1 ] =1e9 *(val[1 ] - start[1 ]) / slept; + min_req =igt_sysfs_get_u32 (sysfs,"gt_cur_freq_mhz" );This really leaves a bad taste for me still, sorry. First of all, it may make it more palatable if readback was immediately after
wait_freq_set(), or as part of it. Otherwise it seems like even more confused test and sysfs ABI.Quoted 5 lines
Did we close on my question of whether we can make readback of
gt_max_freq_mhzactually represent the real max? Correct me please if I got confused - with min freq checking here we write max=300 and get 350? If so can we makegt_max_freq_mhzreturn 350 after the write of 300? Or return an error?Did you read Vinay's email on v1 here:
https://patchwork.freedesktop.org/patch/510304/?series=110574&rev=1#comment_921571
Quoted 7 lines
Quoted 4 lines
As Vinay says, SLPC is requesting an entirely different freq (the effecient freq rather than the set 'min == max freq') and no amount of waiting will result in the requested freq being the same as the set 'min == max freq'. Also the efficient freq cannot be controlled from user space.
Why is that not a break in our ABI? What is the point in allowing users to write into this field if the desired value can be silently ignored?
Quoted 6 lines
Yes, this is in discussion. FW is doing this and we need to see if FW can be made to obey the i915 ABI. IMHO what matters is performance per watt and if the ABI has to be tossed so be it. I will keep this list posted about what happens here. Merging any patches for these issues (GL #6806 and #6786) is blocked till the ABI breakage issue is resolved one way or another.
Two thoughts here, one has been mentioned before.
New one is that performance per watt perhaps does not matter (is not a primary concern at least) for use cases where "users" (for some definition) tweak the min/max in sysfs. Presumably in those cases user wants control, for some reason, so as long as it is safe and within the limits why not give it to them.
Quoted 7 lines
Same argument from a different angle - what is the use case for allowing min/max control which does not really do what it says on the tin?
Old thought is this - if SLPC refuses to be controlled using our existing controls then we could just say
-ENODEVon attempts to modify them. Whether for any input value, or some, depending on what SLPC can support (or wants to support).The problem is not just SLPC. The freq is ultimately decided by PCODE and it
has sudo and all the rights to disrespect what drivers (including SLPC as a a driver from PCODE perspective) are requesting.Quoted 4 lines
Important to note I am primarily focusing on the min freq here, which AFAIR we concluded can behave that user writes in 300, but SLPC decides it will only go as low as 350.
Quoted 6 lines
The other caveat here with the efficient freq is that it is dynamic and PCODE decides it with runtime conditions. It may start with 300, then move to 350, then back to 300. i915 is only stashing the initial value.
We could potentially update our rpe everytime any freq control is touched.
We thought of doing this but two points:
1. i915 now reads the same register as GuC for rpe but I never see the value in this register change.
2. Even if we could know the real rpe, the freq's might be set before a workload start running and rpe might change after the workload starts running. So knowing rpe before the workload might not be very useful.Quoted 3 lines
I am of the opposite opinion that anyway the previous kernel ABI is probably now irrecoverably bust, so we should give FW free reign and stop disabling efficient freq the way we do now when
min_freq<rpe_freq.
Quoted 5 lines
We have to answer the question of why we need to keep exposing the sysfs controls if they don't/can't/won't do what they say on the tin. Could we make a break with DG2 for instance? Some use cases might suffer but it's probably better than having them think it does what they think it does, or file bugs when it does not.
That's a good question. But if we have no control at all people will still file requests for something our hardware and low level firmwares cannot promise.
There are many folks who run important/critical experiments using these interfaces at the same time they are monitoring throttle conditions and all.
Quoted 121 lines
Regards,
Tvrtko
Quoted 116 lines
Cheers.
--
AshutoshQuoted 111 lines
Just to highlight how bad this is, I've seen cases where rpe in runtime goes below the minimal fused freq (our rpn).
Quoted 3 lines
Because that is different from the existing, should
be well known, situation where we can write a max which cannot be achieved due thermal throttling.yeap, usually when things got disrespected by PCODE there will be some thermal throttling information. But I have seen cases where this indicators simply do not exist. What complicate the api expectation to a very bad level.
Quoted 3 lines
Perhaps we need a table, or a summary of some kind, which explains the possible behaviours for both min and max settings.
This is probably a good idea. At least document everything and what and when things can be disrespected, etc...
(and update our rpe info dynamically)
Quoted 91 lines
Quoted 27 lines
Quoted 18 lines
Quoted 18 lines
(So in the example above min and max are set to 300 MHz and SLPC is requesting efficient freq which is 350 or 400 MHz. So min and max values are correct at 300 MHz and requested/actual freq will show 350/400 MHz).
Let me also explain what happens in GuC FW when we change min/max freq:
* i915 receives min/max freq from sysfs and sends a message to GuC to change min/max freq.
* The message itself might be queued behind other messages in the host to GuC queue. i915 is still blocked in the sysfs handler.
* Now when GuC receives the message interrupt, in its top-half handler it validates the freq parameters and sends a message back to host saying the freq change has been accepted. But min/max freq is actually changed later in the bottom-half handler in GuC FW. So the top half only validates the freq value.
* When i915 receives the message from GuC top half, the sysfs handler unblocks and starts reflecting the new min/max freq value.Quoted 8 lines
Quoted 4 lines
So the delay for the new min/max freq values to take effect is between returning from the sysfs handler and when the min/max freq are really set in GuC bottom half (an indeterminate possibly large amount of time if GuC is busy).
Do we have a
drop_cachesflag which would force waiting of idle GuC, which would cover the bottom half as well? Could we or should we add it if not? Would beat the usleep approach..Quoted 6 lines
Agree, waiting for GuC to idle would do it, but not sure "how" to do it. As I said there is no notification after the bottom half is done applying the freq settings. We may need to schedule something after the freq setting host to guc command and wait for that to be done. A little bit out of my depth on this, let me know if you have any ideas here. I'll ask the GuC guys too. Currently we don't have a
drop_cachesflag for this.Okay, I was wondering if
DROP_IDLEwould do it, but perhaps this bottom half can run independently from those flows so it was just a thought, not a claim.Quoted 7 lines
Quoted 7 lines
Quoted 6 lines
So as you see this introduces an indeterminacy due to to GuC interface not being synchronous. If this needs to be fixed, it needs to be fixed in GuC FW rather than waiting etc. in i915. GuC needs to implement another notification to host /after/ the min/max freq change has been applied in its bottom half. Doing this is not a priority for the GuC team at this point.
Quoted 7 lines
Quoted 2 lines
Is there a method for querying the real min from GuC, which would maybe work after waiting for GuC to idle?
Will find out about this too. Afaik, at present querying the real min from GuC is not possible. As soon as the top half has completed the freq settings are updated so querying will return the new settings even though they might not have been applied by the bottom half yet.
And it will return the written value, or the real one? Going back to the 300 written in, 350 actual min freq example. Presumably 300 and 350 SLPC picks is completely invisible to i915 via the FW interface?
Quoted 19 lines
Quoted 19 lines
Quoted 15 lines
However, this bug is not due to these delays but due to GuC requesting the efficient freq instead of the set 'min == max freq' as mentioned above.
If all these delays are confusing we can revert to v1 of the patch:
https://patchwork.freedesktop.org/patch/510304/?series=110574&rev=1
v1 is simple and does not introduce these delays but just reads the requested freq from sysfs after the 500 ms PMU measurement but while the spinner is still running (which we do in v3 too). Without load SLPC will only request RPn so to compare with PMU measurement we must read the requested freq from sysfs with the spinner load. v1 is sufficient to fix the bug but I thought to be completely safe it is better to add
wait_freq_set()after changing min/max freq (and hence v3).Quoted 1 line
I hope this clarifies things somewhat. If not please let me know.
It clarifies a large chunk thank you.
Quoted 7 lines
Quoted 3 lines
Another question which you missed is why is it needed for the max mode? There we allow a 15% error (more than 5% default). Is that not enough with SLPC?
Yes I think you are right, should not be needed for the max mode. The real issue is that efficient freq can exceed the max freq set but I think that should not be an issue if the max is RP0.
Ah okay, so SLPC not respecting max same as with min. I think the question of what is the use case for min/max controls in the SLPC world is the one to answer then.
Regards,
Tvrtko
Quoted 174 lines · Rodrigo Vivi
Quoted 167 lines · Tvrtko Ursulin
Quoted 161 lines · Ashutosh
Quoted 145 lines · Rodrigo Vivi
Quoted 127 lines · Tvrtko Ursulin
Quoted 110 lines · Ashutosh
Quoted 5 lines · Tvrtko Ursulin
Quoted 3 lines · Ashutosh
Quoted 1 line · Tvrtko Ursulin
Hi Tvrtko,
Sorry for the delay in replying on this. We are discussing this but still no conclusion. I have replied what I can below, will update again after I know more. Wanted to send a reply before disappearing for the holidays. Please see below.
Quoted 88 lines
Quoted 88 lines
Hi Tvrtko,
I am only offering an overall clarification below first, not answering all of your points for now.
Quoted 73 lines · Ashutosh Dixit
Quoted 26 lines
With SLPC, even when we set the same min and max freq's, the requested and actual freq's can differ from the min/max freq set. For example "efficient freq" (when in effect) can override set min/max freq. In general FW is the final arbiter in determining freq and can override set values.
Therefore in igt@perf_pmu@frequency subtest, compare the requested freq reported by PMU not against the set freq's but against the requested freq reported in sysfs. Also add a delay after setting freq's to account for messaging delays in setting freq's in GuC.
v2: Introduce a 100 ms delay after setting freq
v3: Update commit message, code identical to v2Cc Vinay Belgaumkar <[email protected]>Cc Tvrtko Ursulin <[email protected]>Signed-off-by Ashutosh Dixit <[email protected]>--- tests/i915/perf_pmu.c | 23 +++++++++++++++-------- 1 file changed, 15 insertions(+), 8 deletions(-)tests/i915/perf_pmu.c +0
-0tests/i915/perf_pmu.c test_interrupts_sync(int gem_fd) · line 1546
Quoted 4 lines
tests/i915/perf_pmu.c +1
-0@@ -1543,10 +1543,13 @@ test_interrupts_sync(int gem_fd)igt_assert_lte (target, busy); } +/* Wait for GuC SLPC freq changes to take effect */ +#define wait_freq_set () usleep(100000)A future task of maybe adding a
drop_cachesflag to wait for this?Alternatively, if we have a sysfs or debugfs file which reflects the value once GuC has processed it, keep reading it until it is updated? Even if to something other than it was before the write, to handle the inability to set the exact requested frequency sometimes?
Quoted 25 lines
tests/i915/perf_pmu.c +5
-3+static void test_frequency (int gem_fd) { -uint32_t min_freq, max_freq, boost_freq; +uint32_t min_freq, max_freq, boost_freq, min_req, max_req;uint64_t val[2 ], start[2 ], slept;double min[2 ], max[2 ];igt_spin_t *spin;@@ -1572,10 +1575,11 @@ test_frequency(int gem_fd)* Set GPU to min frequency and read PMU counters. */ igt_require (igt_sysfs_set_u32 (sysfs,"gt_min_freq_mhz" , min_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) == min_freq);igt_require (igt_sysfs_set_u32 (sysfs,"gt_max_freq_mhz" , min_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == min_freq);igt_require (igt_sysfs_set_u32 (sysfs,"gt_boost_freq_mhz" , min_freq)); +wait_freq_set (); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) == min_freq); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == min_freq);igt_require (igt_sysfs_get_u32 (sysfs,"gt_boost_freq_mhz" ) == min_freq);gem_quiescent_gpu (gem_fd);/* Idle to be sure the change takes effect */
tests/i915/perf_pmu.c test_frequency(int gem_fd) · line 1593
Quoted 3 lines
tests/i915/perf_pmu.c +1
-0@@ -1587,6 +1591,7 @@ test_frequency(int gem_fd) min[0 ] =1e9 *(val[0 ] - start[0 ]) / slept; min[1 ] =1e9 *(val[1 ] - start[1 ]) / slept; + min_req =igt_sysfs_get_u32 (sysfs,"gt_cur_freq_mhz" );This really leaves a bad taste for me still, sorry. First of all, it may make it more palatable if readback was immediately after
wait_freq_set(), or as part of it. Otherwise it seems like even more confused test and sysfs ABI.Quoted 5 lines
Did we close on my question of whether we can make readback of
gt_max_freq_mhzactually represent the real max? Correct me please if I got confused - with min freq checking here we write max=300 and get 350? If so can we makegt_max_freq_mhzreturn 350 after the write of 300? Or return an error?Did you read Vinay's email on v1 here:
https://patchwork.freedesktop.org/patch/510304/?series=110574&rev=1#comment_921571
Quoted 7 lines
Quoted 4 lines
As Vinay says, SLPC is requesting an entirely different freq (the effecient freq rather than the set 'min == max freq') and no amount of waiting will result in the requested freq being the same as the set 'min == max freq'. Also the efficient freq cannot be controlled from user space.
Why is that not a break in our ABI? What is the point in allowing users to write into this field if the desired value can be silently ignored?
Quoted 6 lines
Yes, this is in discussion. FW is doing this and we need to see if FW can be made to obey the i915 ABI. IMHO what matters is performance per watt and if the ABI has to be tossed so be it. I will keep this list posted about what happens here. Merging any patches for these issues (GL #6806 and #6786) is blocked till the ABI breakage issue is resolved one way or another.
Two thoughts here, one has been mentioned before.
New one is that performance per watt perhaps does not matter (is not a primary concern at least) for use cases where "users" (for some definition) tweak the min/max in sysfs. Presumably in those cases user wants control, for some reason, so as long as it is safe and within the limits why not give it to them.
Quoted 7 lines
Same argument from a different angle - what is the use case for allowing min/max control which does not really do what it says on the tin?
Old thought is this - if SLPC refuses to be controlled using our existing controls then we could just say
-ENODEVon attempts to modify them. Whether for any input value, or some, depending on what SLPC can support (or wants to support).The problem is not just SLPC. The freq is ultimately decided by PCODE and it
has sudo and all the rights to disrespect what drivers (including SLPC as a a driver from PCODE perspective) are requesting.Quoted 4 lines
Important to note I am primarily focusing on the min freq here, which AFAIR we concluded can behave that user writes in 300, but SLPC decides it will only go as low as 350.
Quoted 6 lines
The other caveat here with the efficient freq is that it is dynamic and PCODE decides it with runtime conditions. It may start with 300, then move to 350, then back to 300. i915 is only stashing the initial value.
We could potentially update our rpe everytime any freq control is touched.
We thought of doing this but two points:
1. i915 now reads the same register as GuC for rpe but I never see the value in this register change.
2. Even if we could know the real rpe, the freq's might be set before a workload start running and rpe might change after the workload starts running. So knowing rpe before the workload might not be very useful.Quoted 3 lines
I am of the opposite opinion that anyway the previous kernel ABI is probably now irrecoverably bust, so we should give FW free reign and stop disabling efficient freq the way we do now when
min_freq<rpe_freq.Quoted 5 lines
We have to answer the question of why we need to keep exposing the sysfs controls if they don't/can't/won't do what they say on the tin. Could we make a break with DG2 for instance? Some use cases might suffer but it's probably better than having them think it does what they think it does, or file bugs when it does not.
Quoted 5 lines
That's a good question. But if we have no control at all people will still file requests for something our hardware and low level firmwares cannot promise.
There are many folks who run important/critical experiments using these interfaces at the same time they are monitoring throttle conditions and all.
Aren't the two paragraphs a bit in contradiction? Important and critical experiments using controls which are not doing what they think they are.
I don't claim I know for what all use cases do they get used. I *think* there was one in the media transcoding world one where people wanted truly fixed frequency so they can compare like-for-like between something.
So bah.. but okay for the perf_pmu changes, how about instead of this:
rpN = read RPn from sysfs set_min_freq(rpN)
sleepreal_min = read actual from sysfs actual = sample pmu for t assert real_min is actual
We do this:
rpN = read RPn from sysfs set_min_freq(rpN)
sleepfw_min = read actual from sysfs set_min_freq(fw_min)
sleep
actual = sample pmu for t assert fw_min is actual
That's a tiny difference which both "documents" and verifies that even if we cannot set the min to RPn, firmware will somewhat consistently (stable) respect the minimum it prefers.
Is that somewhat acceptable?
Regards,
Tvrtko
Quoted 176 lines · Tvrtko Ursulin
Quoted 174 lines · Rodrigo Vivi
Quoted 172 lines · Tvrtko Ursulin
Quoted 161 lines · Ashutosh
Quoted 145 lines · Rodrigo Vivi
Quoted 127 lines · Tvrtko Ursulin
Quoted 110 lines · Ashutosh
Quoted 5 lines · Tvrtko Ursulin
Quoted 3 lines · Ashutosh
Quoted 1 line · Tvrtko Ursulin
Hi Tvrtko,
Sorry for the delay in replying on this. We are discussing this but still no conclusion. I have replied what I can below, will update again after I know more. Wanted to send a reply before disappearing for the holidays. Please see below.
Quoted 88 lines
Quoted 88 lines
Hi Tvrtko,
I am only offering an overall clarification below first, not answering all of your points for now.
Quoted 73 lines · Ashutosh Dixit
Quoted 26 lines
With SLPC, even when we set the same min and max freq's, the requested and actual freq's can differ from the min/max freq set. For example "efficient freq" (when in effect) can override set min/max freq. In general FW is the final arbiter in determining freq and can override set values.
Therefore in igt@perf_pmu@frequency subtest, compare the requested freq reported by PMU not against the set freq's but against the requested freq reported in sysfs. Also add a delay after setting freq's to account for messaging delays in setting freq's in GuC.
v2: Introduce a 100 ms delay after setting freq
v3: Update commit message, code identical to v2Cc Vinay Belgaumkar <[email protected]>Cc Tvrtko Ursulin <[email protected]>Signed-off-by Ashutosh Dixit <[email protected]>--- tests/i915/perf_pmu.c | 23 +++++++++++++++-------- 1 file changed, 15 insertions(+), 8 deletions(-)tests/i915/perf_pmu.c +0
-0tests/i915/perf_pmu.c test_interrupts_sync(int gem_fd) · line 1546
Quoted 4 lines
tests/i915/perf_pmu.c +1
-0@@ -1543,10 +1543,13 @@ test_interrupts_sync(int gem_fd)igt_assert_lte (target, busy); } +/* Wait for GuC SLPC freq changes to take effect */ +#define wait_freq_set () usleep(100000)A future task of maybe adding a
drop_cachesflag to wait for this?Alternatively, if we have a sysfs or debugfs file which reflects the value once GuC has processed it, keep reading it until it is updated? Even if to something other than it was before the write, to handle the inability to set the exact requested frequency sometimes?
Quoted 25 lines
tests/i915/perf_pmu.c +5
-3+static void test_frequency (int gem_fd) { -uint32_t min_freq, max_freq, boost_freq; +uint32_t min_freq, max_freq, boost_freq, min_req, max_req;uint64_t val[2 ], start[2 ], slept;double min[2 ], max[2 ];igt_spin_t *spin;@@ -1572,10 +1575,11 @@ test_frequency(int gem_fd)* Set GPU to min frequency and read PMU counters. */ igt_require (igt_sysfs_set_u32 (sysfs,"gt_min_freq_mhz" , min_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) == min_freq);igt_require (igt_sysfs_set_u32 (sysfs,"gt_max_freq_mhz" , min_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == min_freq);igt_require (igt_sysfs_set_u32 (sysfs,"gt_boost_freq_mhz" , min_freq)); +wait_freq_set (); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) == min_freq); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == min_freq);igt_require (igt_sysfs_get_u32 (sysfs,"gt_boost_freq_mhz" ) == min_freq);gem_quiescent_gpu (gem_fd);/* Idle to be sure the change takes effect */
tests/i915/perf_pmu.c test_frequency(int gem_fd) · line 1593
Quoted 3 lines
tests/i915/perf_pmu.c +1
-0@@ -1587,6 +1591,7 @@ test_frequency(int gem_fd) min[0 ] =1e9 *(val[0 ] - start[0 ]) / slept; min[1 ] =1e9 *(val[1 ] - start[1 ]) / slept; + min_req =igt_sysfs_get_u32 (sysfs,"gt_cur_freq_mhz" );This really leaves a bad taste for me still, sorry. First of all, it may make it more palatable if readback was immediately after
wait_freq_set(), or as part of it. Otherwise it seems like even more confused test and sysfs ABI.Quoted 5 lines
Did we close on my question of whether we can make readback of
gt_max_freq_mhzactually represent the real max? Correct me please if I got confused - with min freq checking here we write max=300 and get 350? If so can we makegt_max_freq_mhzreturn 350 after the write of 300? Or return an error?Did you read Vinay's email on v1 here:
https://patchwork.freedesktop.org/patch/510304/?series=110574&rev=1#comment_921571
Quoted 7 lines
Quoted 4 lines
As Vinay says, SLPC is requesting an entirely different freq (the effecient freq rather than the set 'min == max freq') and no amount of waiting will result in the requested freq being the same as the set 'min == max freq'. Also the efficient freq cannot be controlled from user space.
Why is that not a break in our ABI? What is the point in allowing users to write into this field if the desired value can be silently ignored?
Quoted 6 lines
Yes, this is in discussion. FW is doing this and we need to see if FW can be made to obey the i915 ABI. IMHO what matters is performance per watt and if the ABI has to be tossed so be it. I will keep this list posted about what happens here. Merging any patches for these issues (GL #6806 and #6786) is blocked till the ABI breakage issue is resolved one way or another.
Two thoughts here, one has been mentioned before.
New one is that performance per watt perhaps does not matter (is not a primary concern at least) for use cases where "users" (for some definition) tweak the min/max in sysfs. Presumably in those cases user wants control, for some reason, so as long as it is safe and within the limits why not give it to them.
Quoted 7 lines
Same argument from a different angle - what is the use case for allowing min/max control which does not really do what it says on the tin?
Old thought is this - if SLPC refuses to be controlled using our existing controls then we could just say
-ENODEVon attempts to modify them. Whether for any input value, or some, depending on what SLPC can support (or wants to support).The problem is not just SLPC. The freq is ultimately decided by PCODE and it
has sudo and all the rights to disrespect what drivers (including SLPC as a a driver from PCODE perspective) are requesting.Quoted 4 lines
Important to note I am primarily focusing on the min freq here, which AFAIR we concluded can behave that user writes in 300, but SLPC decides it will only go as low as 350.
Quoted 6 lines
The other caveat here with the efficient freq is that it is dynamic and PCODE decides it with runtime conditions. It may start with 300, then move to 350, then back to 300. i915 is only stashing the initial value.
We could potentially update our rpe everytime any freq control is touched.
We thought of doing this but two points:
1. i915 now reads the same register as GuC for rpe but I never see the value in this register change.
2. Even if we could know the real rpe, the freq's might be set before a workload start running and rpe might change after the workload starts running. So knowing rpe before the workload might not be very useful.Quoted 3 lines
I am of the opposite opinion that anyway the previous kernel ABI is probably now irrecoverably bust, so we should give FW free reign and stop disabling efficient freq the way we do now when
min_freq<rpe_freq.We have to answer the question of why we need to keep exposing the sysfs controls if they don't/can't/won't do what they say on the tin. Could we make a break with DG2 for instance? Some use cases might suffer but it's probably better than having them think it does what they think it does, or file bugs when it does not.
Quoted 8 lines
Quoted 5 lines
That's a good question. But if we have no control at all people will still file requests for something our hardware and low level firmwares cannot promise.
There are many folks who run important/critical experiments using these interfaces at the same time they are monitoring throttle conditions and all.
Aren't the two paragraphs a bit in contradiction? Important and critical experiments using controls which are not doing what they think they are.
Well, not really.
Most of the times it will work. We just don't have a warranty.
We have some indications when this were disrespected like the throttle_reasons.
Those limited times doesn't exclude the importance of the usage.
Quoted 25 lines
I don't claim I know for what all use cases do they get used. I *think* there was one in the media transcoding world one where people wanted truly fixed frequency so they can compare like-for-like between something.
So bah.. but okay for the
perf_pmuchanges, how about instead of this:rpN = read RPn from sysfs
set_min_freq(rpN)
sleepreal_min= read actual from sysfs
actual = sample pmu for t
assertreal_minis actualWe do this:
rpN = read RPn from sysfs
set_min_freq(rpN)
sleepfw_min= read actual from sysfsset_min_freq(fw_min)
sleep
actual = sample pmu for t
assertfw_minis actual
Quoted 5 lines
That's a tiny difference which both "documents" and verifies that even if we cannot set the min to RPn, firmware will somewhat consistently (stable) respect the minimum it prefers.
Is that somewhat acceptable?
yeap, this sounds a better approach for a sane test.
Quoted 4 lines
Regards,
Tvrtko
Quoted 184 lines · Rodrigo Vivi
Quoted 182 lines · Tvrtko Ursulin
Quoted 179 lines · Rodrigo Vivi
Quoted 167 lines · Tvrtko Ursulin
Quoted 161 lines · Ashutosh
Quoted 145 lines · Rodrigo Vivi
Quoted 127 lines · Tvrtko Ursulin
Quoted 110 lines · Ashutosh
Quoted 5 lines · Tvrtko Ursulin
Quoted 3 lines · Ashutosh
Quoted 1 line · Tvrtko Ursulin
Hi Tvrtko,
Sorry for the delay in replying on this. We are discussing this but still no conclusion. I have replied what I can below, will update again after I know more. Wanted to send a reply before disappearing for the holidays. Please see below.
Quoted 88 lines
Quoted 88 lines
Hi Tvrtko,
I am only offering an overall clarification below first, not answering all of your points for now.
Quoted 73 lines · Ashutosh Dixit
Quoted 26 lines
With SLPC, even when we set the same min and max freq's, the requested and actual freq's can differ from the min/max freq set. For example "efficient freq" (when in effect) can override set min/max freq. In general FW is the final arbiter in determining freq and can override set values.
Therefore in igt@perf_pmu@frequency subtest, compare the requested freq reported by PMU not against the set freq's but against the requested freq reported in sysfs. Also add a delay after setting freq's to account for messaging delays in setting freq's in GuC.
v2: Introduce a 100 ms delay after setting freq
v3: Update commit message, code identical to v2Cc Vinay Belgaumkar <[email protected]>Cc Tvrtko Ursulin <[email protected]>Signed-off-by Ashutosh Dixit <[email protected]>--- tests/i915/perf_pmu.c | 23 +++++++++++++++-------- 1 file changed, 15 insertions(+), 8 deletions(-)tests/i915/perf_pmu.c +0
-0tests/i915/perf_pmu.c test_interrupts_sync(int gem_fd) · line 1546
Quoted 4 lines
tests/i915/perf_pmu.c +1
-0@@ -1543,10 +1543,13 @@ test_interrupts_sync(int gem_fd)igt_assert_lte (target, busy); } +/* Wait for GuC SLPC freq changes to take effect */ +#define wait_freq_set () usleep(100000)A future task of maybe adding a
drop_cachesflag to wait for this?Alternatively, if we have a sysfs or debugfs file which reflects the value once GuC has processed it, keep reading it until it is updated? Even if to something other than it was before the write, to handle the inability to set the exact requested frequency sometimes?
Quoted 25 lines
tests/i915/perf_pmu.c +5
-3+static void test_frequency (int gem_fd) { -uint32_t min_freq, max_freq, boost_freq; +uint32_t min_freq, max_freq, boost_freq, min_req, max_req;uint64_t val[2 ], start[2 ], slept;double min[2 ], max[2 ];igt_spin_t *spin;@@ -1572,10 +1575,11 @@ test_frequency(int gem_fd)* Set GPU to min frequency and read PMU counters. */ igt_require (igt_sysfs_set_u32 (sysfs,"gt_min_freq_mhz" , min_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) == min_freq);igt_require (igt_sysfs_set_u32 (sysfs,"gt_max_freq_mhz" , min_freq)); -igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == min_freq);igt_require (igt_sysfs_set_u32 (sysfs,"gt_boost_freq_mhz" , min_freq)); +wait_freq_set (); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_min_freq_mhz" ) == min_freq); +igt_require (igt_sysfs_get_u32 (sysfs,"gt_max_freq_mhz" ) == min_freq);igt_require (igt_sysfs_get_u32 (sysfs,"gt_boost_freq_mhz" ) == min_freq);gem_quiescent_gpu (gem_fd);/* Idle to be sure the change takes effect */
tests/i915/perf_pmu.c test_frequency(int gem_fd) · line 1593
Quoted 3 lines
tests/i915/perf_pmu.c +1
-0@@ -1587,6 +1591,7 @@ test_frequency(int gem_fd) min[0 ] =1e9 *(val[0 ] - start[0 ]) / slept; min[1 ] =1e9 *(val[1 ] - start[1 ]) / slept; + min_req =igt_sysfs_get_u32 (sysfs,"gt_cur_freq_mhz" );This really leaves a bad taste for me still, sorry. First of all, it may make it more palatable if readback was immediately after
wait_freq_set(), or as part of it. Otherwise it seems like even more confused test and sysfs ABI.Quoted 5 lines
Did we close on my question of whether we can make readback of
gt_max_freq_mhzactually represent the real max? Correct me please if I got confused - with min freq checking here we write max=300 and get 350? If so can we makegt_max_freq_mhzreturn 350 after the write of 300? Or return an error?Did you read Vinay's email on v1 here:
https://patchwork.freedesktop.org/patch/510304/?series=110574&rev=1#comment_921571
Quoted 7 lines
Quoted 4 lines
As Vinay says, SLPC is requesting an entirely different freq (the effecient freq rather than the set 'min == max freq') and no amount of waiting will result in the requested freq being the same as the set 'min == max freq'. Also the efficient freq cannot be controlled from user space.
Why is that not a break in our ABI? What is the point in allowing users to write into this field if the desired value can be silently ignored?
Quoted 6 lines
Yes, this is in discussion. FW is doing this and we need to see if FW can be made to obey the i915 ABI. IMHO what matters is performance per watt and if the ABI has to be tossed so be it. I will keep this list posted about what happens here. Merging any patches for these issues (GL #6806 and #6786) is blocked till the ABI breakage issue is resolved one way or another.
Two thoughts here, one has been mentioned before.
New one is that performance per watt perhaps does not matter (is not a primary concern at least) for use cases where "users" (for some definition) tweak the min/max in sysfs. Presumably in those cases user wants control, for some reason, so as long as it is safe and within the limits why not give it to them.
Quoted 7 lines
Same argument from a different angle - what is the use case for allowing min/max control which does not really do what it says on the tin?
Old thought is this - if SLPC refuses to be controlled using our existing controls then we could just say
-ENODEVon attempts to modify them. Whether for any input value, or some, depending on what SLPC can support (or wants to support).The problem is not just SLPC. The freq is ultimately decided by PCODE and it
has sudo and all the rights to disrespect what drivers (including SLPC as a a driver from PCODE perspective) are requesting.Quoted 4 lines
Important to note I am primarily focusing on the min freq here, which AFAIR we concluded can behave that user writes in 300, but SLPC decides it will only go as low as 350.
Quoted 6 lines
The other caveat here with the efficient freq is that it is dynamic and PCODE decides it with runtime conditions. It may start with 300, then move to 350, then back to 300. i915 is only stashing the initial value.
We could potentially update our rpe everytime any freq control is touched.
We thought of doing this but two points:
1. i915 now reads the same register as GuC for rpe but I never see the value in this register change.
2. Even if we could know the real rpe, the freq's might be set before a workload start running and rpe might change after the workload starts running. So knowing rpe before the workload might not be very useful.Quoted 3 lines
I am of the opposite opinion that anyway the previous kernel ABI is probably now irrecoverably bust, so we should give FW free reign and stop disabling efficient freq the way we do now when
min_freq<rpe_freq.Quoted 5 lines
We have to answer the question of why we need to keep exposing the sysfs controls if they don't/can't/won't do what they say on the tin. Could we make a break with DG2 for instance? Some use cases might suffer but it's probably better than having them think it does what they think it does, or file bugs when it does not.
That's a good question. But if we have no control at all people will still file requests for something our hardware and low level firmwares cannot promise.
There are many folks who run important/critical experiments using these interfaces at the same time they are monitoring throttle conditions and all.
Quoted 6 lines
Quoted 2 lines
Aren't the two paragraphs a bit in contradiction? Important and critical experiments using controls which are not doing what they think they are.
Well, not really.
Most of the times it will work. We just don't have a warranty.
I think it would help for the user to know when it will work and when it will not.
Quoted 2 lines
We have some indications when this were disrespected like the
throttle_reasons.
There is no indication related to rpe_freq in throttle_reasons, isn't it?
Quoted 30 lines
Those limited times doesn't exclude the importance of the usage.
Quoted 28 lines
I don't claim I know for what all use cases do they get used. I *think* there was one in the media transcoding world one where people wanted truly fixed frequency so they can compare like-for-like between something.
So bah.. but okay for the
perf_pmuchanges, how about instead of this:rpN = read RPn from sysfs
set_min_freq(rpN)
sleepreal_min= read actual from sysfs
actual = sample pmu for t
assertreal_minis actualWe do this:
rpN = read RPn from sysfs
set_min_freq(rpN)
sleepfw_min= read actual from sysfsset_min_freq(fw_min)
sleep
actual = sample pmu for t
assertfw_minis actualThat's a tiny difference which both "documents" and verifies that even if we cannot set the min to RPn, firmware will somewhat consistently (stable) respect the minimum it prefers.
Note it's not the min but the max freq which is not honored.
Quoted 4 lines
Quoted 2 lines
Is that somewhat acceptable?
yeap, this sounds a better approach for a sane test.
We talked with the GuC guys and learnt a few things:
* PCODE continuously calculates and updates the current RPe (efficient freq) value
* GuC samples the RPe values from PCODE every 1 ms and uses these to request operating freq. So the requested freq is always set by GuC but can be overridden say in cases of thermal throttling etc.
* GuC ignores the max freq and directly uses RPe freq when requesting an operating freq. This is beacuse RPe is supposedly more efficient that a lower max freq.
* GuC does have a separate control to ignore RPe entirely.
* Regarding the delays I added in the patch, some time back GuC FW was changed to move setting min/max freq's to the top half. As a result the delays are no longer required.
The most recent version the patches in light of above points are here:
https://patchwork.freedesktop.org/series/111282/#rev3
(The patches are similar to Tvrtko's first (rather than the second) pseudo-code above).
These patches improve things but even with these patches unfortunately there are still sporadic failures. For example:
Original failure:
* https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8303/bat-dg2-11/igt@[email protected]
* RPn 300 MHz, PMU measures 397 MHz
New failure:
* https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8304/bat-dg2-11/igt@[email protected]
* PMU measures 371 MHz, requested freq measured after PMU measurement 400 MHz
So these sporadic failures seems to be due to dynamically changing RPe.
It may be possible to get rid of the failures by either increasing the tolerance or sampling the requested freq sysfs more often (rather than just once after the PMU measurement) though maybe it's too much.
So the best way to fix such sporadic failures seems to be to expose a 'rpe_enable' sysfs to turn off RPe for such tests/usages where strict userspace control of freq limits is required.
@Rodrigo Vivi: so we need to see if we can make this uapi change and then go from there I think.
Thanks.
--
Ashutosh
Quoted 188 lines · Ashutosh
Quoted 187 lines · Rodrigo Vivi
Quoted 184 lines · Tvrtko Ursulin
Quoted 174 lines · Rodrigo Vivi
Quoted 167 lines · Tvrtko Ursulin
Quoted 161 lines · Ashutosh
Quoted 145 lines · Rodrigo Vivi
Quoted 127 lines · Tvrtko Ursulin
Quoted 110 lines · Ashutosh
Quoted 5 lines · Tvrtko Ursulin
Quoted 3 lines · Ashutosh
Quoted 1 line · Tvrtko Ursulin
Hi Tvrtko,
Sorry for the delay in replying on this. We are discussing this but still no conclusion. I have replied what I can below, will update again after I know more. Wanted to send a reply before disappearing for the holidays. Please see below.
Quoted 88 lines
Quoted 88 lines
Hi Tvrtko,
I am only offering an overall clarification below first, not answering all of your points for now.
Quoted 73 lines · Ashutosh Dixit
Quoted 30 lines
With SLPC, even when we set the same min and max freq's, the requested and
actual freq's can differ from the min/max freq set. For example "efficient
freq" (when in effect) can override set min/max freq. In general FW is the
final arbiter in determining freq and can override set values.Therefore in igt@perf_pmu@frequency subtest, compare the requested freq
reported by PMU not against the set freq's but against the requested freq
reported in sysfs. Also add a delay after setting freq's to account for
messaging delays in setting freq's in GuC.v2: Introduce a 100 ms delay after setting freq
v3: Update commit message, code identical to v2Bug: https://gitlab.freedesktop.org/drm/intel/-/issues/6806
Cc: Vinay Belgaumkar <[email protected]>
Cc: Tvrtko Ursulin <[email protected]>
Signed-off-by: Ashutosh Dixit <[email protected]>
---
tests/i915/perf_pmu.c | 23 +++++++++++++++--------
1 file changed, 15 insertions(+), 8 deletions(-)diff --git a/tests/i915/perf_pmu.c b/tests/i915/perf_pmu.c
index f363db2ba13..02f6ae989b1 100644
--- a/tests/i915/perf_pmu.c
+++ b/tests/i915/perf_pmu.c
@@ -1543,10 +1543,13 @@ test_interrupts_sync(int gem_fd)
igt_assert_lte(target, busy);
}
+/* Wait for GuC SLPC freq changes to take effect */
+#define wait_freq_set() usleep(100000)A future task of maybe adding a
drop_cachesflag to wait for this?Alternatively, if we have a sysfs or debugfs file which reflects the value once GuC has processed it, keep reading it until it is updated? Even if to something other than it was before the write, to handle the inability to set the exact requested frequency sometimes?
Quoted 28 lines
+
static void
test_frequency(int gem_fd)
{
- uint32_t min_freq, max_freq, boost_freq;
+ uint32_t min_freq, max_freq, boost_freq, min_req, max_req;
uint64_t val[2], start[2], slept;
double min[2], max[2];
igt_spin_t *spin;
@@ -1572,10 +1575,11 @@ test_frequency(int gem_fd)
* Set GPU to min frequency and read PMU counters.
*/
igt_require(igt_sysfs_set_u32(sysfs, "gt_min_freq_mhz", min_freq));
- igt_require(igt_sysfs_get_u32(sysfs, "gt_min_freq_mhz") == min_freq);
igt_require(igt_sysfs_set_u32(sysfs, "gt_max_freq_mhz", min_freq));
- igt_require(igt_sysfs_get_u32(sysfs, "gt_max_freq_mhz") == min_freq);
igt_require(igt_sysfs_set_u32(sysfs, "gt_boost_freq_mhz", min_freq));
+ wait_freq_set();
+ igt_require(igt_sysfs_get_u32(sysfs, "gt_min_freq_mhz") == min_freq);
+ igt_require(igt_sysfs_get_u32(sysfs, "gt_max_freq_mhz") == min_freq);
igt_require(igt_sysfs_get_u32(sysfs, "gt_boost_freq_mhz") == min_freq);
gem_quiescent_gpu(gem_fd); /* Idle to be sure the change takes
effect */
@@ -1587,6 +1591,7 @@ test_frequency(int gem_fd)
min[0] = 1e9*(val[0] - start[0]) / slept;
min[1] = 1e9*(val[1] - start[1]) / slept;
+ min_req = igt_sysfs_get_u32(sysfs, "gt_cur_freq_mhz");This really leaves a bad taste for me still, sorry. First of all, it may make it more palatable if readback was immediately after
wait_freq_set(), or as part of it. Otherwise it seems like even more confused test and sysfs ABI.Quoted 5 lines
Did we close on my question of whether we can make readback of
gt_max_freq_mhzactually represent the real max? Correct me please if I got confused - with min freq checking here we write max=300 and get 350? If so can we makegt_max_freq_mhzreturn 350 after the write of 300? Or return an error?Did you read Vinay's email on v1 here:
https://patchwork.freedesktop.org/patch/510304/?series=110574&rev=1#comment_921571
Quoted 7 lines
Quoted 4 lines
As Vinay says, SLPC is requesting an entirely different freq (the effecient freq rather than the set 'min == max freq') and no amount of waiting will result in the requested freq being the same as the set 'min == max freq'. Also the efficient freq cannot be controlled from user space.
Why is that not a break in our ABI? What is the point in allowing users to write into this field if the desired value can be silently ignored?
Quoted 6 lines
Yes, this is in discussion. FW is doing this and we need to see if FW can be made to obey the i915 ABI. IMHO what matters is performance per watt and if the ABI has to be tossed so be it. I will keep this list posted about what happens here. Merging any patches for these issues (GL #6806 and #6786) is blocked till the ABI breakage issue is resolved one way or another.
Two thoughts here, one has been mentioned before.
New one is that performance per watt perhaps does not matter (is not a primary concern at least) for use cases where "users" (for some definition) tweak the min/max in sysfs. Presumably in those cases user wants control, for some reason, so as long as it is safe and within the limits why not give it to them.
Quoted 7 lines
Same argument from a different angle - what is the use case for allowing min/max control which does not really do what it says on the tin?
Old thought is this - if SLPC refuses to be controlled using our existing controls then we could just say
-ENODEVon attempts to modify them. Whether for any input value, or some, depending on what SLPC can support (or wants to support).The problem is not just SLPC. The freq is ultimately decided by PCODE and it
has sudo and all the rights to disrespect what drivers (including SLPC as a a driver from PCODE perspective) are requesting.Quoted 4 lines
Important to note I am primarily focusing on the min freq here, which AFAIR we concluded can behave that user writes in 300, but SLPC decides it will only go as low as 350.
Quoted 6 lines
The other caveat here with the efficient freq is that it is dynamic and PCODE decides it with runtime conditions. It may start with 300, then move to 350, then back to 300. i915 is only stashing the initial value.
We could potentially update our rpe everytime any freq control is touched.
We thought of doing this but two points:
1. i915 now reads the same register as GuC for rpe but I never see the value in this register change.
2. Even if we could know the real rpe, the freq's might be set before a workload start running and rpe might change after the workload starts running. So knowing rpe before the workload might not be very useful.Quoted 3 lines
I am of the opposite opinion that anyway the previous kernel ABI is probably now irrecoverably bust, so we should give FW free reign and stop disabling efficient freq the way we do now when
min_freq<rpe_freq.Quoted 5 lines
We have to answer the question of why we need to keep exposing the sysfs controls if they don't/can't/won't do what they say on the tin. Could we make a break with DG2 for instance? Some use cases might suffer but it's probably better than having them think it does what they think it does, or file bugs when it does not.
Quoted 5 lines
That's a good question. But if we have no control at all people will still file requests for something our hardware and low level firmwares cannot promise.
There are many folks who run important/critical experiments using these interfaces at the same time they are monitoring throttle conditions and all.
Aren't the two paragraphs a bit in contradiction? Important and critical experiments using controls which are not doing what they think they are.
Quoted 6 lines
Quoted 3 lines
Well, not really.
Most of the times it will work. We just don't have a warranty.
I think it would help for the user to know when it will work and when it will not.
Yeap. But mostly this lays on improving our throttle reasons infra.
I'm also wondering now if upon freq reads if we also couldn't internally check the throttle reason and put a debug message?!
Quoted 5 lines
Quoted 3 lines
We have some indications when this were disrespected like the
throttle_reasons.There is no indication related to
rpe_freqinthrottle_reasons, isn't it?
no.
Quoted 68 lines
Quoted 31 lines
Those limited times doesn't exclude the importance of the usage.
Quoted 28 lines
I don't claim I know for what all use cases do they get used. I *think* there was one in the media transcoding world one where people wanted truly fixed frequency so they can compare like-for-like between something.
So bah.. but okay for the
perf_pmuchanges, how about instead of this:rpN = read RPn from sysfs
set_min_freq(rpN)
sleepreal_min= read actual from sysfs
actual = sample pmu for t
assertreal_minis actualWe do this:
rpN = read RPn from sysfs
set_min_freq(rpN)
sleepfw_min= read actual from sysfsset_min_freq(fw_min)
sleep
actual = sample pmu for t
assertfw_minis actualThat's a tiny difference which both "documents" and verifies that even if we cannot set the min to RPn, firmware will somewhat consistently (stable) respect the minimum it prefers.
Note it's not the min but the max freq which is not honored.
Quoted 4 lines
Quoted 2 lines
Is that somewhat acceptable?
yeap, this sounds a better approach for a sane test.
We talked with the GuC guys and learnt a few things:
* PCODE continuously calculates and updates the current RPe (efficient freq) value
* GuC samples the RPe values from PCODE every 1 ms and uses these to request operating freq. So the requested freq is always set by GuC but can be overridden say in cases of thermal throttling etc.
* GuC ignores the max freq and directly uses RPe freq when requesting an operating freq. This is beacuse RPe is supposedly more efficient that a lower max freq.
* GuC does have a separate control to ignore RPe entirely.
* Regarding the delays I added in the patch, some time back GuC FW was changed to move setting min/max freq's to the top half. As a result the delays are no longer required.The most recent version the patches in light of above points are here:
https://patchwork.freedesktop.org/series/111282/#rev3
(The patches are similar to Tvrtko's first (rather than the second) pseudo-code above).
These patches improve things but even with these patches unfortunately there are still sporadic failures. For example:
Original failure:
* https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8303/bat-dg2-11/igt@[email protected]
* RPn 300 MHz, PMU measures 397 MHz
Quoted 5 lines
New failure:
* https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8304/bat-dg2-11/igt@[email protected]
* PMU measures 371 MHz, requested freq measured after PMU measurement 400 MHzSo these sporadic failures seems to be due to dynamically changing RPe.
Yes, they are.
Quoted 4 lines
It may be possible to get rid of the failures by either increasing the tolerance or sampling the requested freq sysfs more often (rather than just once after the PMU measurement) though maybe it's too much.
On Xe driver I workarounded it by 'constantly' updating our cached rpe. By constantly I mean whenever anyone is touching (reading or writing) any frequency.
However I'm aware that this can be racy. Specially for the PMU cases or with some out of band checks.
Quoted 7 lines
So the best way to fix such sporadic failures seems to be to expose a '
rpe_enable' sysfs to turn off RPe for such tests/usages where strict userspace control of freq limits is required.@Rodrigo Vivi: so we need to see if we can make this uapi change and then go from there I think.
I believe this is the most sane option right now. We need to document this well and have test cases for both.
Probably this plus the constant update of the cached RPe.
Quoted 4 lines
Thanks.
--
Ashutosh
Quoted 259 lines · Rodrigo Vivi
Quoted 162 lines · Ashutosh
Quoted 161 lines · Rodrigo Vivi
Quoted 160 lines · Tvrtko Ursulin
Quoted 159 lines · Rodrigo Vivi
Quoted 145 lines · Tvrtko Ursulin
Quoted 144 lines · Ashutosh
Quoted 129 lines · Rodrigo Vivi
Quoted 114 lines · Tvrtko Ursulin
Quoted 97 lines · Ashutosh
Quoted 2 lines · Tvrtko Ursulin
Quoted 1 line · Ashutosh
On Mon, 21 Nov 2022 01:09:55 -0800, Tvrtko Ursulin wrote:
Hi Tvrtko,
Sorry for the delay in replying on this. We are discussing this but still no conclusion. I have replied what I can below, will update again after I know more. Wanted to send a reply before disappearing for the holidays. Please see below.
Quoted 87 lines
Quoted 87 lines
Hi Tvrtko,
I am only offering an overall clarification below first, not answering all of your points for now.
Quoted 74 lines · Ashutosh Dixit
With SLPC, even when we set the same min and max freq's, the requested and
actual freq's can differ from the min/max freq set. For example "efficient
freq" (when in effect) can override set min/max freq. In general FW is the
final arbiter in determining freq and can override set values.Therefore in igt@perf_pmu@frequency subtest, compare the requested freq
reported by PMU not against the set freq's but against the requested freq
reported in sysfs. Also add a delay after setting freq's to account for
messaging delays in setting freq's in GuC.v2: Introduce a 100 ms delay after setting freq
v3: Update commit message, code identical to v2Bug: https://gitlab.freedesktop.org/drm/intel/-/issues/6806
Cc: Vinay Belgaumkar <[email protected]>
Cc: Tvrtko Ursulin <[email protected]>
Signed-off-by: Ashutosh Dixit <[email protected]>
---
tests/i915/perf_pmu.c | 23 +++++++++++++++--------
1 file changed, 15 insertions(+), 8 deletions(-)diff --git a/tests/i915/perf_pmu.c b/tests/i915/perf_pmu.c
index f363db2ba13..02f6ae989b1 100644
--- a/tests/i915/perf_pmu.c
+++ b/tests/i915/perf_pmu.c
@@ -1543,10 +1543,13 @@ test_interrupts_sync(int gem_fd)
igt_assert_lte(target, busy);
}
+/* Wait for GuC SLPC freq changes to take effect */
+#define wait_freq_set() usleep(100000)
A future task of maybe adding a drop_caches flag to wait for this?Alternatively, if we have a sysfs or debugfs file which reflects the value
once GuC has processed it, keep reading it until it is updated? Even if to
something other than it was before the write, to handle the inability to
set the exact requested frequency sometimes?
+
static void
test_frequency(int gem_fd)
{
- uint32_t min_freq, max_freq, boost_freq;
+ uint32_t min_freq, max_freq, boost_freq, min_req, max_req;
uint64_t val[2], start[2], slept;
double min[2], max[2];
igt_spin_t *spin;
@@ -1572,10 +1575,11 @@ test_frequency(int gem_fd)
* Set GPU to min frequency and read PMU counters.
*/
igt_require(igt_sysfs_set_u32(sysfs, "gt_min_freq_mhz", min_freq));
- igt_require(igt_sysfs_get_u32(sysfs, "gt_min_freq_mhz") == min_freq);
igt_require(igt_sysfs_set_u32(sysfs, "gt_max_freq_mhz", min_freq));
- igt_require(igt_sysfs_get_u32(sysfs, "gt_max_freq_mhz") == min_freq);
igt_require(igt_sysfs_set_u32(sysfs, "gt_boost_freq_mhz", min_freq));
+ wait_freq_set();
+ igt_require(igt_sysfs_get_u32(sysfs, "gt_min_freq_mhz") == min_freq);
+ igt_require(igt_sysfs_get_u32(sysfs, "gt_max_freq_mhz") == min_freq);
igt_require(igt_sysfs_get_u32(sysfs, "gt_boost_freq_mhz") == min_freq);
gem_quiescent_gpu(gem_fd); /* Idle to be sure the change takes
effect */
@@ -1587,6 +1591,7 @@ test_frequency(int gem_fd)
min[0] = 1e9*(val[0] - start[0]) / slept;
min[1] = 1e9*(val[1] - start[1]) / slept;
+ min_req = igt_sysfs_get_u32(sysfs, "gt_cur_freq_mhz");
This really leaves a bad taste for me still, sorry. First of all, it may
make it more palatable if readback was immediately after wait_freq_set(),
or as part of it. Otherwise it seems like even more confused test and sysfs
ABI.Did we close on my question of whether we can make readback of
gt_max_freq_mhz actually represent the real max? Correct me please if I got
confused - with min freq checking here we write max=300 and get 350? If so
can we make gt_max_freq_mhz return 350 after the write of 300? Or return an
error?Did you read Vinay's email on v1 here:
https://patchwork.freedesktop.org/patch/510304/?series=110574&rev=1#comment_921571
As Vinay says, SLPC is requesting an entirely different freq (the effecient freq rather than the set 'min == max freq') and no amount of waiting will result in the requested freq being the same as the set 'min == max freq'. Also the efficient freq cannot be controlled from user space.
Quoted 8 lines
Quoted 2 lines
Why is that not a break in our ABI? What is the point in allowing users to write into this field if the desired value can be silently ignored?
Yes, this is in discussion. FW is doing this and we need to see if FW can be made to obey the i915 ABI. IMHO what matters is performance per watt and if the ABI has to be tossed so be it. I will keep this list posted about what happens here. Merging any patches for these issues (GL #6806 and #6786) is blocked till the ABI breakage issue is resolved one way or another.
Two thoughts here, one has been mentioned before.
New one is that performance per watt perhaps does not matter (is not a primary concern at least) for use cases where "users" (for some definition) tweak the min/max in sysfs. Presumably in those cases user wants control, for some reason, so as long as it is safe and within the limits why not give it to them.
Quoted 7 lines
Same argument from a different angle - what is the use case for allowing min/max control which does not really do what it says on the tin?
Old thought is this - if SLPC refuses to be controlled using our existing controls then we could just say
-ENODEVon attempts to modify them. Whether for any input value, or some, depending on what SLPC can support (or wants to support).The problem is not just SLPC. The freq is ultimately decided by PCODE and it
has sudo and all the rights to disrespect what drivers (including SLPC as a a driver from PCODE perspective) are requesting.Quoted 3 lines
Important to note I am primarily focusing on the min freq here, which AFAIR we concluded can behave that user writes in 300, but SLPC decides it will only go as low as 350.
Quoted 6 lines
The other caveat here with the efficient freq is that it is dynamic and PCODE decides it with runtime conditions. It may start with 300, then move to 350, then back to 300. i915 is only stashing the initial value.
We could potentially update our rpe everytime any freq control is touched.
We thought of doing this but two points:
1. i915 now reads the same register as GuC for rpe but I never see the value in this register change.
2. Even if we could know the real rpe, the freq's might be set before a workload start running and rpe might change after the workload starts running. So knowing rpe before the workload might not be very useful.Quoted 8 lines
Quoted 3 lines
I am of the opposite opinion that anyway the previous kernel ABI is probably now irrecoverably bust, so we should give FW free reign and stop disabling efficient freq the way we do now when
min_freq<rpe_freq.We have to answer the question of why we need to keep exposing the sysfs controls if they don't/can't/won't do what they say on the tin. Could we make a break with DG2 for instance? Some use cases might suffer but it's probably better than having them think it does what they think it does, or file bugs when it does not.
That's a good question. But if we have no control at all people will still file requests for something our hardware and low level firmwares cannot promise.
There are many folks who run important/critical experiments using these interfaces at the same time they are monitoring throttle conditions and all.
Quoted 7 lines
Quoted 5 lines
Quoted 2 lines
Aren't the two paragraphs a bit in contradiction? Important and critical experiments using controls which are not doing what they think they are.
Well, not really.
Most of the times it will work. We just don't have a warranty.
I think it would help for the user to know when it will work and when it will not.
Yeap. But mostly this lays on improving our throttle reasons infra.
I'm also wondering now if upon freq reads if we also couldn't internally check the throttle reason and put a debug message?!
Quoted 3 lines
Quoted 2 lines
We have some indications when this were disrespected like the
throttle_reasons.There is no indication related to
rpe_freqinthrottle_reasons, isn't it?no.
Quoted 62 lines
Quoted 29 lines
Those limited times doesn't exclude the importance of the usage.
Quoted 27 lines
I don't claim I know for what all use cases do they get used. I *think* there was one in the media transcoding world one where people wanted truly fixed frequency so they can compare like-for-like between something.
So bah.. but okay for the
perf_pmuchanges, how about instead of this:rpN = read RPn from sysfs
set_min_freq(rpN)
sleepreal_min= read actual from sysfs
actual = sample pmu for t
assertreal_minis actualWe do this:
rpN = read RPn from sysfs
set_min_freq(rpN)
sleepfw_min= read actual from sysfsset_min_freq(fw_min)
sleep
actual = sample pmu for t
assertfw_minis actualThat's a tiny difference which both "documents" and verifies that even if we cannot set the min to RPn, firmware will somewhat consistently (stable) respect the minimum it prefers.
Note it's not the min but the max freq which is not honored.
Quoted 2 lines
Quoted 1 line
Is that somewhat acceptable?
yeap, this sounds a better approach for a sane test.
We talked with the GuC guys and learnt a few things:
* PCODE continuously calculates and updates the current RPe (efficient freq) value
* GuC samples the RPe values from PCODE every 1 ms and uses these to request operating freq. So the requested freq is always set by GuC but can be overridden say in cases of thermal throttling etc.
* GuC ignores the max freq and directly uses RPe freq when requesting an operating freq. This is beacuse RPe is supposedly more efficient that a lower max freq.
* GuC does have a separate control to ignore RPe entirely.
* Regarding the delays I added in the patch, some time back GuC FW was changed to move setting min/max freq's to the top half. As a result the delays are no longer required.The most recent version the patches in light of above points are here:
https://patchwork.freedesktop.org/series/111282/#rev3
(The patches are similar to Tvrtko's first (rather than the second) pseudo-code above).
These patches improve things but even with these patches unfortunately there are still sporadic failures. For example:
Original failure:
* https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8303/bat-dg2-11/igt@[email protected]
* RPn 300 MHz, PMU measures 397 MHzQuoted 5 lines
New failure:
* https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8304/bat-dg2-11/igt@[email protected]
* PMU measures 371 MHz, requested freq measured after PMU measurement 400 MHzSo these sporadic failures seems to be due to dynamically changing RPe.
Yes, they are.
Quoted 3 lines
It may be possible to get rid of the failures by either increasing the tolerance or sampling the requested freq sysfs more often (rather than just once after the PMU measurement) though maybe it's too much.
On Xe driver I workarounded it by 'constantly' updating our cached rpe. By constantly I mean whenever anyone is touching (reading or writing) any frequency.
However I'm aware that this can be racy. Specially for the PMU cases or with some out of band checks.
Quoted 8 lines
Quoted 6 lines
So the best way to fix such sporadic failures seems to be to expose a '
rpe_enable' sysfs to turn off RPe for such tests/usages where strict userspace control of freq limits is required.@Rodrigo Vivi: so we need to see if we can make this uapi change and then go from there I think.
I believe this is the most sane option right now. We need to document this well and have test cases for both.
Either that, or we turn off efficient frequency when user updates max frequency as well (similar to what we do for min).
Quoted 2 lines
Probably this plus the constant update of the cached RPe.
Caching might be easier if GuC does the same. It can update the task state data with the new value for everyone else to read.
Thanks,
Vinay.
Quoted 4 lines
Quoted 4 lines
Thanks.
--
Ashutosh
Quoted 253 lines · Vinay
Quoted 252 lines · Rodrigo Vivi
Quoted 162 lines · Ashutosh
Quoted 161 lines · Rodrigo Vivi
Quoted 160 lines · Tvrtko Ursulin
Quoted 159 lines · Rodrigo Vivi
Quoted 145 lines · Tvrtko Ursulin
Quoted 144 lines · Ashutosh
Quoted 129 lines · Rodrigo Vivi
Quoted 114 lines · Tvrtko Ursulin
Quoted 97 lines · Ashutosh
Quoted 2 lines · Tvrtko Ursulin
Quoted 1 line · Ashutosh
On Mon, 21 Nov 2022 01:09:55 -0800, Tvrtko Ursulin wrote:
Hi Tvrtko,
Sorry for the delay in replying on this. We are discussing this but still no conclusion. I have replied what I can below, will update again after I know more. Wanted to send a reply before disappearing for the holidays. Please see below.
Quoted 87 lines
Quoted 87 lines
Hi Tvrtko,
I am only offering an overall clarification below first, not answering all
of your points for now.On 19/11/2022 02:00, Ashutosh Dixit wrote:
With SLPC, even when we set the same min and max freq's, the requested and
actual freq's can differ from the min/max freq set. For example "efficient
freq" (when in effect) can override set min/max freq. In general FW is the
final arbiter in determining freq and can override set values.Therefore in igt@perf_pmu@frequency subtest, compare the requested freq
reported by PMU not against the set freq's but against the requested freq
reported in sysfs. Also add a delay after setting freq's to account for
messaging delays in setting freq's in GuC.v2: Introduce a 100 ms delay after setting freq
v3: Update commit message, code identical to v2Bug: https://gitlab.freedesktop.org/drm/intel/-/issues/6806
Cc: Vinay Belgaumkar <[email protected]>
Cc: Tvrtko Ursulin <[email protected]>
Signed-off-by: Ashutosh Dixit <[email protected]>
---
tests/i915/perf_pmu.c | 23 +++++++++++++++--------
1 file changed, 15 insertions(+), 8 deletions(-)diff --git a/tests/i915/perf_pmu.c b/tests/i915/perf_pmu.c
index f363db2ba13..02f6ae989b1 100644
--- a/tests/i915/perf_pmu.c
+++ b/tests/i915/perf_pmu.c
@@ -1543,10 +1543,13 @@ test_interrupts_sync(int gem_fd)
igt_assert_lte(target, busy);
}
+/* Wait for GuC SLPC freq changes to take effect */
+#define wait_freq_set() usleep(100000)
A future task of maybe adding a drop_caches flag to wait for this?Alternatively, if we have a sysfs or debugfs file which reflects the value
once GuC has processed it, keep reading it until it is updated? Even if to
something other than it was before the write, to handle the inability to
set the exact requested frequency sometimes?
+
static void
test_frequency(int gem_fd)
{
- uint32_t min_freq, max_freq, boost_freq;
+ uint32_t min_freq, max_freq, boost_freq, min_req, max_req;
uint64_t val[2], start[2], slept;
double min[2], max[2];
igt_spin_t *spin;
@@ -1572,10 +1575,11 @@ test_frequency(int gem_fd)
* Set GPU to min frequency and read PMU counters.
*/
igt_require(igt_sysfs_set_u32(sysfs, "gt_min_freq_mhz", min_freq));
- igt_require(igt_sysfs_get_u32(sysfs, "gt_min_freq_mhz") == min_freq);
igt_require(igt_sysfs_set_u32(sysfs, "gt_max_freq_mhz", min_freq));
- igt_require(igt_sysfs_get_u32(sysfs, "gt_max_freq_mhz") == min_freq);
igt_require(igt_sysfs_set_u32(sysfs, "gt_boost_freq_mhz", min_freq));
+ wait_freq_set();
+ igt_require(igt_sysfs_get_u32(sysfs, "gt_min_freq_mhz") == min_freq);
+ igt_require(igt_sysfs_get_u32(sysfs, "gt_max_freq_mhz") == min_freq);
igt_require(igt_sysfs_get_u32(sysfs, "gt_boost_freq_mhz") == min_freq);
gem_quiescent_gpu(gem_fd); /* Idle to be sure the change takes
effect */
@@ -1587,6 +1591,7 @@ test_frequency(int gem_fd)
min[0] = 1e9*(val[0] - start[0]) / slept;
min[1] = 1e9*(val[1] - start[1]) / slept;
+ min_req = igt_sysfs_get_u32(sysfs, "gt_cur_freq_mhz");
This really leaves a bad taste for me still, sorry. First of all, it may
make it more palatable if readback was immediately after wait_freq_set(),
or as part of it. Otherwise it seems like even more confused test and sysfs
ABI.Did we close on my question of whether we can make readback of
gt_max_freq_mhz actually represent the real max? Correct me please if I got
confused - with min freq checking here we write max=300 and get 350? If so
can we make gt_max_freq_mhz return 350 after the write of 300? Or return an
error?
Did you read Vinay's email on v1 here:https://patchwork.freedesktop.org/patch/510304/?series=110574&rev=1#comment_921571
As Vinay says, SLPC is requesting an entirely different freq (the effecient
freq rather than the set 'min == max freq') and no amount of waiting will
result in the requested freq being the same as the set 'min == max
freq'. Also the efficient freq cannot be controlled from user space.Quoted 8 lines
Quoted 2 lines
Why is that not a break in our ABI? What is the point in allowing users to write into this field if the desired value can be silently ignored?
Yes, this is in discussion. FW is doing this and we need to see if FW can be made to obey the i915 ABI. IMHO what matters is performance per watt and if the ABI has to be tossed so be it. I will keep this list posted about what happens here. Merging any patches for these issues (GL #6806 and #6786) is blocked till the ABI breakage issue is resolved one way or another.
Two thoughts here, one has been mentioned before.
New one is that performance per watt perhaps does not matter (is not a primary concern at least) for use cases where "users" (for some definition) tweak the min/max in sysfs. Presumably in those cases user wants control, for some reason, so as long as it is safe and within the limits why not give it to them.
Quoted 7 lines
Same argument from a different angle - what is the use case for allowing min/max control which does not really do what it says on the tin?
Old thought is this - if SLPC refuses to be controlled using our existing controls then we could just say
-ENODEVon attempts to modify them. Whether for any input value, or some, depending on what SLPC can support (or wants to support).The problem is not just SLPC. The freq is ultimately decided by PCODE and it
has sudo and all the rights to disrespect what drivers (including SLPC as a a driver from PCODE perspective) are requesting.Quoted 3 lines
Important to note I am primarily focusing on the min freq here, which AFAIR we concluded can behave that user writes in 300, but SLPC decides it will only go as low as 350.
Quoted 6 lines
The other caveat here with the efficient freq is that it is dynamic and PCODE decides it with runtime conditions. It may start with 300, then move to 350, then back to 300. i915 is only stashing the initial value.
We could potentially update our rpe everytime any freq control is touched.
We thought of doing this but two points:
1. i915 now reads the same register as GuC for rpe but I never see the value in this register change.
2. Even if we could know the real rpe, the freq's might be set before a workload start running and rpe might change after the workload starts running. So knowing rpe before the workload might not be very useful.Quoted 8 lines
Quoted 3 lines
I am of the opposite opinion that anyway the previous kernel ABI is probably now irrecoverably bust, so we should give FW free reign and stop disabling efficient freq the way we do now when
min_freq<rpe_freq.We have to answer the question of why we need to keep exposing the sysfs controls if they don't/can't/won't do what they say on the tin. Could we make a break with DG2 for instance? Some use cases might suffer but it's probably better than having them think it does what they think it does, or file bugs when it does not.
That's a good question. But if we have no control at all people will still file requests for something our hardware and low level firmwares cannot promise.
There are many folks who run important/critical experiments using these interfaces at the same time they are monitoring throttle conditions and all.
Quoted 7 lines
Quoted 5 lines
Quoted 2 lines
Aren't the two paragraphs a bit in contradiction? Important and critical experiments using controls which are not doing what they think they are.
Well, not really.
Most of the times it will work. We just don't have a warranty.
I think it would help for the user to know when it will work and when it will not.
Yeap. But mostly this lays on improving our throttle reasons infra.
I'm also wondering now if upon freq reads if we also couldn't internally check the throttle reason and put a debug message?!
Quoted 3 lines
Quoted 2 lines
We have some indications when this were disrespected like the
throttle_reasons.There is no indication related to
rpe_freqinthrottle_reasons, isn't it?no.
Quoted 62 lines
Quoted 29 lines
Those limited times doesn't exclude the importance of the usage.
Quoted 27 lines
I don't claim I know for what all use cases do they get used. I *think* there was one in the media transcoding world one where people wanted truly fixed frequency so they can compare like-for-like between something.
So bah.. but okay for the
perf_pmuchanges, how about instead of this:rpN = read RPn from sysfs
set_min_freq(rpN)
sleepreal_min= read actual from sysfs
actual = sample pmu for t
assertreal_minis actualWe do this:
rpN = read RPn from sysfs
set_min_freq(rpN)
sleepfw_min= read actual from sysfsset_min_freq(fw_min)
sleep
actual = sample pmu for t
assertfw_minis actualThat's a tiny difference which both "documents" and verifies that even if we cannot set the min to RPn, firmware will somewhat consistently (stable) respect the minimum it prefers.
Note it's not the min but the max freq which is not honored.
Quoted 2 lines
Quoted 1 line
Is that somewhat acceptable?
yeap, this sounds a better approach for a sane test.
We talked with the GuC guys and learnt a few things:
* PCODE continuously calculates and updates the current RPe (efficient freq) value
* GuC samples the RPe values from PCODE every 1 ms and uses these to request operating freq. So the requested freq is always set by GuC but can be overridden say in cases of thermal throttling etc.
* GuC ignores the max freq and directly uses RPe freq when requesting an operating freq. This is beacuse RPe is supposedly more efficient that a lower max freq.
* GuC does have a separate control to ignore RPe entirely.
* Regarding the delays I added in the patch, some time back GuC FW was changed to move setting min/max freq's to the top half. As a result the delays are no longer required.The most recent version the patches in light of above points are here:
https://patchwork.freedesktop.org/series/111282/#rev3
(The patches are similar to Tvrtko's first (rather than the second) pseudo-code above).
These patches improve things but even with these patches unfortunately there are still sporadic failures. For example:
Original failure:
* https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8303/bat-dg2-11/igt@[email protected]
* RPn 300 MHz, PMU measures 397 MHzQuoted 5 lines
New failure:
* https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8304/bat-dg2-11/igt@[email protected]
* PMU measures 371 MHz, requested freq measured after PMU measurement 400 MHzSo these sporadic failures seems to be due to dynamically changing RPe.
Yes, they are.
Quoted 3 lines
It may be possible to get rid of the failures by either increasing the tolerance or sampling the requested freq sysfs more often (rather than just once after the PMU measurement) though maybe it's too much.
Quoted 6 lines
Quoted 6 lines
On Xe driver I workarounded it by 'constantly' updating our cached rpe. By constantly I mean whenever anyone is touching (reading or writing) any frequency.
However I'm aware that this can be racy. Specially for the PMU cases or with some out of band checks.
Consider 2 cases:
1. We start a spinner load and then update min/max freq
2. We update min/max freq first and then start spinner load
Doing the above, we likely pick up the correct (RPe) freq being used for case 1 but not for case 2. And even for case 1 RPe might be dynamically changing.
Quoted 10 lines
Quoted 8 lines
Quoted 6 lines
So the best way to fix such sporadic failures seems to be to expose a '
rpe_enable' sysfs to turn off RPe for such tests/usages where strict userspace control of freq limits is required.@Rodrigo Vivi: so we need to see if we can make this uapi change and then go from there I think.
I believe this is the most sane option right now. We need to document this well and have test cases for both.
Either that, or we turn off efficient frequency when user updates max frequency as well (similar to what we do for min).
I think I disagree here, turning off RPe will affect PnP. I think we should have explicit disable of RPe when that is needed and should actually remove RPe disable when min is updated. Basically IMO the idea should be to provide the best PnP and then tailor the ABI for the best PnP.
Quoted 5 lines
Quoted 2 lines
Probably this plus the constant update of the cached RPe.
Caching might be easier if GuC does the same. It can update the task state data with the new value for everyone else to read.
i915 will either need to read the register or read the task state from GuC. You mean reading the task state is more efficient than reading the register?
Thanks.
--
Ashutosh
Quoted 261 lines · Ashutosh
Quoted 253 lines · Vinay
Quoted 252 lines · Rodrigo Vivi
Quoted 162 lines · Ashutosh
Quoted 161 lines · Rodrigo Vivi
Quoted 160 lines · Tvrtko Ursulin
Quoted 159 lines · Rodrigo Vivi
Quoted 145 lines · Tvrtko Ursulin
Quoted 144 lines · Ashutosh
Quoted 129 lines · Rodrigo Vivi
Quoted 114 lines · Tvrtko Ursulin
Quoted 97 lines · Ashutosh
Quoted 2 lines · Tvrtko Ursulin
On 23/11/2022 06:03, Dixit, Ashutosh wrote:
On Mon, 21 Nov 2022 01:09:55 -0800, Tvrtko Ursulin wrote:Hi Tvrtko,
Sorry for the delay in replying on this. We are discussing this but still no conclusion. I have replied what I can below, will update again after I know more. Wanted to send a reply before disappearing for the holidays. Please see below.
Quoted 87 lines
Hi Tvrtko,
I am only offering an overall clarification below first, not answering all
of your points for now.On 19/11/2022 02:00, Ashutosh Dixit wrote:
With SLPC, even when we set the same min and max freq's, the requested and
actual freq's can differ from the min/max freq set. For example "efficient
freq" (when in effect) can override set min/max freq. In general FW is the
final arbiter in determining freq and can override set values.Therefore in igt@perf_pmu@frequency subtest, compare the requested freq
reported by PMU not against the set freq's but against the requested freq
reported in sysfs. Also add a delay after setting freq's to account for
messaging delays in setting freq's in GuC.v2: Introduce a 100 ms delay after setting freq
v3: Update commit message, code identical to v2Bug: https://gitlab.freedesktop.org/drm/intel/-/issues/6806
Cc: Vinay Belgaumkar <[email protected]>
Cc: Tvrtko Ursulin <[email protected]>
Signed-off-by: Ashutosh Dixit <[email protected]>
---
tests/i915/perf_pmu.c | 23 +++++++++++++++--------
1 file changed, 15 insertions(+), 8 deletions(-)diff --git a/tests/i915/perf_pmu.c b/tests/i915/perf_pmu.c
index f363db2ba13..02f6ae989b1 100644
--- a/tests/i915/perf_pmu.c
+++ b/tests/i915/perf_pmu.c
@@ -1543,10 +1543,13 @@ test_interrupts_sync(int gem_fd)
igt_assert_lte(target, busy);
}
+/* Wait for GuC SLPC freq changes to take effect */
+#define wait_freq_set() usleep(100000)
A future task of maybe adding a drop_caches flag to wait for this?Alternatively, if we have a sysfs or debugfs file which reflects the value
once GuC has processed it, keep reading it until it is updated? Even if to
something other than it was before the write, to handle the inability to
set the exact requested frequency sometimes?
+
static void
test_frequency(int gem_fd)
{
- uint32_t min_freq, max_freq, boost_freq;
+ uint32_t min_freq, max_freq, boost_freq, min_req, max_req;
uint64_t val[2], start[2], slept;
double min[2], max[2];
igt_spin_t *spin;
@@ -1572,10 +1575,11 @@ test_frequency(int gem_fd)
* Set GPU to min frequency and read PMU counters.
*/
igt_require(igt_sysfs_set_u32(sysfs, "gt_min_freq_mhz", min_freq));
- igt_require(igt_sysfs_get_u32(sysfs, "gt_min_freq_mhz") == min_freq);
igt_require(igt_sysfs_set_u32(sysfs, "gt_max_freq_mhz", min_freq));
- igt_require(igt_sysfs_get_u32(sysfs, "gt_max_freq_mhz") == min_freq);
igt_require(igt_sysfs_set_u32(sysfs, "gt_boost_freq_mhz", min_freq));
+ wait_freq_set();
+ igt_require(igt_sysfs_get_u32(sysfs, "gt_min_freq_mhz") == min_freq);
+ igt_require(igt_sysfs_get_u32(sysfs, "gt_max_freq_mhz") == min_freq);
igt_require(igt_sysfs_get_u32(sysfs, "gt_boost_freq_mhz") == min_freq);
gem_quiescent_gpu(gem_fd); /* Idle to be sure the change takes
effect */
@@ -1587,6 +1591,7 @@ test_frequency(int gem_fd)
min[0] = 1e9*(val[0] - start[0]) / slept;
min[1] = 1e9*(val[1] - start[1]) / slept;
+ min_req = igt_sysfs_get_u32(sysfs, "gt_cur_freq_mhz");
This really leaves a bad taste for me still, sorry. First of all, it may
make it more palatable if readback was immediately after wait_freq_set(),
or as part of it. Otherwise it seems like even more confused test and sysfs
ABI.Did we close on my question of whether we can make readback of
gt_max_freq_mhz actually represent the real max? Correct me please if I got
confused - with min freq checking here we write max=300 and get 350? If so
can we make gt_max_freq_mhz return 350 after the write of 300? Or return an
error?
Did you read Vinay's email on v1 here:https://patchwork.freedesktop.org/patch/510304/?series=110574&rev=1#comment_921571
As Vinay says, SLPC is requesting an entirely different freq (the effecient
freq rather than the set 'min == max freq') and no amount of waiting will
result in the requested freq being the same as the set 'min == max
freq'. Also the efficient freq cannot be controlled from user space.Quoted 8 lines
Quoted 2 lines
Why is that not a break in our ABI? What is the point in allowing users to
write into this field if the desired value can be silently ignored?Yes, this is in discussion. FW is doing this and we need to see if FW can be made to obey the i915 ABI. IMHO what matters is performance per watt and if the ABI has to be tossed so be it. I will keep this list posted about what happens here. Merging any patches for these issues (GL #6806 and #6786) is blocked till the ABI breakage issue is resolved one way or another.
Two thoughts here, one has been mentioned before.
New one is that performance per watt perhaps does not matter (is not a primary concern at least) for use cases where "users" (for some definition) tweak the min/max in sysfs. Presumably in those cases user wants control, for some reason, so as long as it is safe and within the limits why not give it to them.
Quoted 7 lines
Same argument from a different angle - what is the use case for allowing min/max control which does not really do what it says on the tin?
Old thought is this - if SLPC refuses to be controlled using our existing controls then we could just say
-ENODEVon attempts to modify them. Whether for any input value, or some, depending on what SLPC can support (or wants to support).The problem is not just SLPC. The freq is ultimately decided by PCODE and it
has sudo and all the rights to disrespect what drivers (including SLPC as a a driver from PCODE perspective) are requesting.Quoted 3 lines
Important to note I am primarily focusing on the min freq here, which AFAIR we concluded can behave that user writes in 300, but SLPC decides it will only go as low as 350.
Quoted 6 lines
The other caveat here with the efficient freq is that it is dynamic and PCODE decides it with runtime conditions. It may start with 300, then move to 350, then back to 300. i915 is only stashing the initial value.
We could potentially update our rpe everytime any freq control is touched.
We thought of doing this but two points:
1. i915 now reads the same register as GuC for rpe but I never see the value in this register change.
2. Even if we could know the real rpe, the freq's might be set before a workload start running and rpe might change after the workload starts running. So knowing rpe before the workload might not be very useful.Quoted 8 lines
Quoted 3 lines
I am of the opposite opinion that anyway the previous kernel ABI is probably now irrecoverably bust, so we should give FW free reign and stop disabling efficient freq the way we do now when
min_freq<rpe_freq.We have to answer the question of why we need to keep exposing the sysfs controls if they don't/can't/won't do what they say on the tin. Could we make a break with DG2 for instance? Some use cases might suffer but it's probably better than having them think it does what they think it does, or file bugs when it does not.
That's a good question. But if we have no control at all people will still file requests for something our hardware and low level firmwares cannot promise.
There are many folks who run important/critical experiments using these interfaces at the same time they are monitoring throttle conditions and all.
Quoted 7 lines
Quoted 5 lines
Quoted 2 lines
Aren't the two paragraphs a bit in contradiction? Important and critical experiments using controls which are not doing what they think they are.
Well, not really.
Most of the times it will work. We just don't have a warranty.
I think it would help for the user to know when it will work and when it will not.
Yeap. But mostly this lays on improving our throttle reasons infra.
I'm also wondering now if upon freq reads if we also couldn't internally check the throttle reason and put a debug message?!
Quoted 3 lines
Quoted 2 lines
We have some indications when this were disrespected like the
throttle_reasons.There is no indication related to
rpe_freqinthrottle_reasons, isn't it?no.
Quoted 62 lines
Quoted 29 lines
Those limited times doesn't exclude the importance of the usage.
Quoted 27 lines
I don't claim I know for what all use cases do they get used. I *think* there was one in the media transcoding world one where people wanted truly fixed frequency so they can compare like-for-like between something.
So bah.. but okay for the
perf_pmuchanges, how about instead of this:rpN = read RPn from sysfs
set_min_freq(rpN)
sleepreal_min= read actual from sysfs
actual = sample pmu for t
assertreal_minis actualWe do this:
rpN = read RPn from sysfs
set_min_freq(rpN)
sleepfw_min= read actual from sysfsset_min_freq(fw_min)
sleep
actual = sample pmu for t
assertfw_minis actualThat's a tiny difference which both "documents" and verifies that even if we cannot set the min to RPn, firmware will somewhat consistently (stable) respect the minimum it prefers.
Note it's not the min but the max freq which is not honored.
Quoted 2 lines
Quoted 1 line
Is that somewhat acceptable?
yeap, this sounds a better approach for a sane test.
We talked with the GuC guys and learnt a few things:
* PCODE continuously calculates and updates the current RPe (efficient freq) value
* GuC samples the RPe values from PCODE every 1 ms and uses these to request operating freq. So the requested freq is always set by GuC but can be overridden say in cases of thermal throttling etc.
* GuC ignores the max freq and directly uses RPe freq when requesting an operating freq. This is beacuse RPe is supposedly more efficient that a lower max freq.
* GuC does have a separate control to ignore RPe entirely.
* Regarding the delays I added in the patch, some time back GuC FW was changed to move setting min/max freq's to the top half. As a result the delays are no longer required.The most recent version the patches in light of above points are here:
https://patchwork.freedesktop.org/series/111282/#rev3
(The patches are similar to Tvrtko's first (rather than the second) pseudo-code above).
These patches improve things but even with these patches unfortunately there are still sporadic failures. For example:
Original failure:
* https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8303/bat-dg2-11/igt@[email protected]
* RPn 300 MHz, PMU measures 397 MHzQuoted 5 lines
New failure:
* https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8304/bat-dg2-11/igt@[email protected]
* PMU measures 371 MHz, requested freq measured after PMU measurement 400 MHzSo these sporadic failures seems to be due to dynamically changing RPe.
Yes, they are.
Quoted 3 lines
It may be possible to get rid of the failures by either increasing the tolerance or sampling the requested freq sysfs more often (rather than just once after the PMU measurement) though maybe it's too much.
Quoted 6 lines
Quoted 6 lines
On Xe driver I workarounded it by 'constantly' updating our cached rpe. By constantly I mean whenever anyone is touching (reading or writing) any frequency.
However I'm aware that this can be racy. Specially for the PMU cases or with some out of band checks.
Quoted 8 lines
Consider 2 cases:
1. We start a spinner load and then update min/max freq
2. We update min/max freq first and then start spinner loadDoing the above, we likely pick up the correct (RPe) freq being used for case 1 but not for case 2. And even for case 1 RPe might be dynamically changing.
Indeed. It is racy no matter how you look at it. 1 component deciding and 2 other reading and using to take decisions without any lock.
Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
Quoted 7 lines
So the best way to fix such sporadic failures seems to be to expose a '
rpe_enable' sysfs to turn off RPe for such tests/usages where strict userspace control of freq limits is required.@Rodrigo Vivi: so we need to see if we can make this uapi change and then go from there I think.
I believe this is the most sane option right now. We need to document this well and have test cases for both.
Quoted 7 lines
Quoted 2 lines
Either that, or we turn off efficient frequency when user updates max frequency as well (similar to what we do for min).
I think I disagree here, turning off RPe will affect PnP. I think we should have explicit disable of RPe when that is needed and should actually remove RPe disable when min is updated. Basically IMO the idea should be to provide the best PnP and then tailor the ABI for the best PnP.
We don't have any api problem when rpe is enabled but min <= rpe < max. So Vinay is wondering if we can get this scenario to enable the rpe in guc we don't need to create another api. Then on any read or write we would check this condition and decide on the efficient enabled or not.
But as I told, it is racy anyway.
So, in the end if we don't disable rpe we don't have how to guarantee the freq stable freq api. But if we disable we have the pnp impact. If we try any magic it will be racy.
I'm tending to agree with Ashutosh and that adding the rpe_enabled is the best option. But document that very well. And yes, update the cached actual rpe anytime that we try to read that.
Quoted 10 lines
Quoted 6 lines
Quoted 3 lines
Probably this plus the constant update of the cached RPe.
Caching might be easier if GuC does the same. It can update the task state data with the new value for everyone else to read.
i915 will either need to read the register or read the task state from GuC. You mean reading the task state is more efficient than reading the register?
oh, indeed... we can remove the cache and read from pcode directly anytime we need that. Safer indeed.
Quoted 4 lines
Thanks.
--
Ashutosh
Quoted 190 lines · Rodrigo Vivi
Quoted 188 lines · Ashutosh
Quoted 187 lines · Vinay
Quoted 186 lines · Rodrigo Vivi
Quoted 162 lines · Ashutosh
Quoted 161 lines · Rodrigo Vivi
Quoted 160 lines · Tvrtko Ursulin
Quoted 159 lines · Rodrigo Vivi
Quoted 145 lines · Tvrtko Ursulin
Quoted 144 lines · Ashutosh
Quoted 129 lines · Rodrigo Vivi
Quoted 114 lines · Tvrtko Ursulin
Quoted 105 lines · Ashutosh
On Thu, 24 Nov 2022 04:42:08 -0800, Tvrtko Ursulin wrote:
On 23/11/2022 06:03, Dixit, Ashutosh wrote:
On Mon, 21 Nov 2022 01:09:55 -0800, Tvrtko Ursulin wrote:
Hi Tvrtko,Sorry for the delay in replying on this. We are discussing this but still
no conclusion. I have replied what I can below, will update again after I
know more. Wanted to send a reply before disappearing for the
holidays. Please see below.Hi Tvrtko,
I am only offering an overall clarification below first, not answering all
of your points for now.On 19/11/2022 02:00, Ashutosh Dixit wrote:
With SLPC, even when we set the same min and max freq's, the requested and
actual freq's can differ from the min/max freq set. For example "efficient
freq" (when in effect) can override set min/max freq. In general FW is the
final arbiter in determining freq and can override set values.Therefore in igt@perf_pmu@frequency subtest, compare the requested freq
reported by PMU not against the set freq's but against the requested freq
reported in sysfs. Also add a delay after setting freq's to account for
messaging delays in setting freq's in GuC.v2: Introduce a 100 ms delay after setting freq
v3: Update commit message, code identical to v2Bug: https://gitlab.freedesktop.org/drm/intel/-/issues/6806
Cc: Vinay Belgaumkar <[email protected]>
Cc: Tvrtko Ursulin <[email protected]>
Signed-off-by: Ashutosh Dixit <[email protected]>
---
tests/i915/perf_pmu.c | 23 +++++++++++++++--------
1 file changed, 15 insertions(+), 8 deletions(-)diff --git a/tests/i915/perf_pmu.c b/tests/i915/perf_pmu.c
index f363db2ba13..02f6ae989b1 100644
--- a/tests/i915/perf_pmu.c
+++ b/tests/i915/perf_pmu.c
@@ -1543,10 +1543,13 @@ test_interrupts_sync(int gem_fd)
igt_assert_lte(target, busy);
}
+/* Wait for GuC SLPC freq changes to take effect */
+#define wait_freq_set() usleep(100000)
A future task of maybe adding a drop_caches flag to wait for this?Alternatively, if we have a sysfs or debugfs file which reflects the value
once GuC has processed it, keep reading it until it is updated? Even if to
something other than it was before the write, to handle the inability to
set the exact requested frequency sometimes?
+
static void
test_frequency(int gem_fd)
{
- uint32_t min_freq, max_freq, boost_freq;
+ uint32_t min_freq, max_freq, boost_freq, min_req, max_req;
uint64_t val[2], start[2], slept;
double min[2], max[2];
igt_spin_t *spin;
@@ -1572,10 +1575,11 @@ test_frequency(int gem_fd)
* Set GPU to min frequency and read PMU counters.
*/
igt_require(igt_sysfs_set_u32(sysfs, "gt_min_freq_mhz", min_freq));
- igt_require(igt_sysfs_get_u32(sysfs, "gt_min_freq_mhz") == min_freq);
igt_require(igt_sysfs_set_u32(sysfs, "gt_max_freq_mhz", min_freq));
- igt_require(igt_sysfs_get_u32(sysfs, "gt_max_freq_mhz") == min_freq);
igt_require(igt_sysfs_set_u32(sysfs, "gt_boost_freq_mhz", min_freq));
+ wait_freq_set();
+ igt_require(igt_sysfs_get_u32(sysfs, "gt_min_freq_mhz") == min_freq);
+ igt_require(igt_sysfs_get_u32(sysfs, "gt_max_freq_mhz") == min_freq);
igt_require(igt_sysfs_get_u32(sysfs, "gt_boost_freq_mhz") == min_freq);
gem_quiescent_gpu(gem_fd); /* Idle to be sure the change takes
effect */
@@ -1587,6 +1591,7 @@ test_frequency(int gem_fd)
min[0] = 1e9*(val[0] - start[0]) / slept;
min[1] = 1e9*(val[1] - start[1]) / slept;
+ min_req = igt_sysfs_get_u32(sysfs, "gt_cur_freq_mhz");
This really leaves a bad taste for me still, sorry. First of all, it may
make it more palatable if readback was immediately after wait_freq_set(),
or as part of it. Otherwise it seems like even more confused test and sysfs
ABI.Did we close on my question of whether we can make readback of
gt_max_freq_mhz actually represent the real max? Correct me please if I got
confused - with min freq checking here we write max=300 and get 350? If so
can we make gt_max_freq_mhz return 350 after the write of 300? Or return an
error?
Did you read Vinay's email on v1 here:https://patchwork.freedesktop.org/patch/510304/?series=110574&rev=1#comment_921571
As Vinay says, SLPC is requesting an entirely different freq (the effecient
freq rather than the set 'min == max freq') and no amount of waiting will
result in the requested freq being the same as the set 'min == max
freq'. Also the efficient freq cannot be controlled from user space.
Why is that not a break in our ABI? What is the point in allowing users to
write into this field if the desired value can be silently ignored?
Yes, this is in discussion. FW is doing this and we need to see if FW can
be made to obey the i915 ABI. IMHO what matters is performance per watt and
if the ABI has to be tossed so be it. I will keep this list posted about
what happens here. Merging any patches for these issues (GL #6806 and
#6786) is blocked till the ABI breakage issue is resolved one way or
another.Two thoughts here, one has been mentioned before.
New one is that performance per watt perhaps does not matter (is not a primary concern at least) for use cases where "users" (for some definition) tweak the min/max in sysfs. Presumably in those cases user wants control, for some reason, so as long as it is safe and within the limits why not give it to them.
Quoted 7 lines
Same argument from a different angle - what is the use case for allowing min/max control which does not really do what it says on the tin?
Old thought is this - if SLPC refuses to be controlled using our existing controls then we could just say
-ENODEVon attempts to modify them. Whether for any input value, or some, depending on what SLPC can support (or wants to support).The problem is not just SLPC. The freq is ultimately decided by PCODE and it
has sudo and all the rights to disrespect what drivers (including SLPC as a a driver from PCODE perspective) are requesting.Quoted 3 lines
Important to note I am primarily focusing on the min freq here, which AFAIR we concluded can behave that user writes in 300, but SLPC decides it will only go as low as 350.
Quoted 6 lines
The other caveat here with the efficient freq is that it is dynamic and PCODE decides it with runtime conditions. It may start with 300, then move to 350, then back to 300. i915 is only stashing the initial value.
We could potentially update our rpe everytime any freq control is touched.
We thought of doing this but two points:
1. i915 now reads the same register as GuC for rpe but I never see the value in this register change.
2. Even if we could know the real rpe, the freq's might be set before a workload start running and rpe might change after the workload starts running. So knowing rpe before the workload might not be very useful.Quoted 8 lines
Quoted 3 lines
I am of the opposite opinion that anyway the previous kernel ABI is probably now irrecoverably bust, so we should give FW free reign and stop disabling efficient freq the way we do now when
min_freq<rpe_freq.We have to answer the question of why we need to keep exposing the sysfs controls if they don't/can't/won't do what they say on the tin. Could we make a break with DG2 for instance? Some use cases might suffer but it's probably better than having them think it does what they think it does, or file bugs when it does not.
That's a good question. But if we have no control at all people will still file requests for something our hardware and low level firmwares cannot promise.
There are many folks who run important/critical experiments using these interfaces at the same time they are monitoring throttle conditions and all.
Quoted 7 lines
Quoted 5 lines
Quoted 2 lines
Aren't the two paragraphs a bit in contradiction? Important and critical experiments using controls which are not doing what they think they are.
Well, not really.
Most of the times it will work. We just don't have a warranty.
I think it would help for the user to know when it will work and when it will not.
Yeap. But mostly this lays on improving our throttle reasons infra.
I'm also wondering now if upon freq reads if we also couldn't internally check the throttle reason and put a debug message?!
Quoted 3 lines
Quoted 2 lines
We have some indications when this were disrespected like the
throttle_reasons.There is no indication related to
rpe_freqinthrottle_reasons, isn't it?no.
Quoted 6 lines
Quoted 6 lines
Those limited times doesn't exclude the importance of the usage.
Quoted 4 lines
I don't claim I know for what all use cases do they get used. I *think* there was one in the media transcoding world one where people wanted truly fixed frequency so they can compare like-for-like between something.
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
So bah.. but okay for the
perf_pmuchanges, how about instead of this:rpN = read RPn from sysfs
set_min_freq(rpN)
sleepreal_min= read actual from sysfs
actual = sample pmu for t
assertreal_minis actual
I have resubmitted the previous IGT patches based on this approach (with RPe enabled) after incorporating all the previous findings (mentioned previously in this thread) and copied all of you on this mail thread:
https://patchwork.freedesktop.org/series/111282/#rev8
Quoted 12 lines
Quoted 12 lines
Quoted 12 lines
Quoted 12 lines
Quoted 12 lines
Quoted 12 lines
Quoted 12 lines
We do this:
rpN = read RPn from sysfs
set_min_freq(rpN)
sleepfw_min= read actual from sysfsset_min_freq(fw_min)
sleep
actual = sample pmu for t
assertfw_minis actual
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 3 lines
Quoted 3 lines
That's a tiny difference which both "documents" and verifies that even if we cannot set the min to RPn, firmware will somewhat consistently (stable) respect the minimum it prefers.
Note it's not the min but the max freq which is not honored.
I decided not to do this because as mentioned previously we already know that FW does not respect max_freq. But yes by way of documenting, I have explained reasons for the changes in comments in the code (and the commit message history).
Quoted 54 lines
Quoted 54 lines
Quoted 42 lines
Quoted 42 lines
Quoted 32 lines
Quoted 3 lines
Quoted 2 lines
Is that somewhat acceptable?
yeap, this sounds a better approach for a sane test.
We talked with the GuC guys and learnt a few things:
* PCODE continuously calculates and updates the current RPe (efficient freq) value
* GuC samples the RPe values from PCODE every 1 ms and uses these to request operating freq. So the requested freq is always set by GuC but can be overridden say in cases of thermal throttling etc.
* GuC ignores the max freq and directly uses RPe freq when requesting an operating freq. This is beacuse RPe is supposedly more efficient that a lower max freq.
* GuC does have a separate control to ignore RPe entirely.
* Regarding the delays I added in the patch, some time back GuC FW was changed to move setting min/max freq's to the top half. As a result the delays are no longer required.The most recent version the patches in light of above points are here:
https://patchwork.freedesktop.org/series/111282/#rev3
(The patches are similar to Tvrtko's first (rather than the second) pseudo-code above).
These patches improve things but even with these patches unfortunately there are still sporadic failures. For example:
Original failure:
* https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8303/bat-dg2-11/igt@[email protected]
* RPn 300 MHz, PMU measures 397 MHzQuoted 5 lines
New failure:
* https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_8304/bat-dg2-11/igt@[email protected]
* PMU measures 371 MHz, requested freq measured after PMU measurement 400 MHzSo these sporadic failures seems to be due to dynamically changing RPe.
Yes, they are.
Quoted 3 lines
It may be possible to get rid of the failures by either increasing the tolerance or sampling the requested freq sysfs more often (rather than just once after the PMU measurement) though maybe it's too much.
Quoted 6 lines
Quoted 6 lines
On Xe driver I workarounded it by 'constantly' updating our cached rpe. By constantly I mean whenever anyone is touching (reading or writing) any frequency.
However I'm aware that this can be racy. Specially for the PMU cases or with some out of band checks.
Consider 2 cases:
1. We start a spinner load and then update min/max freq
2. We update min/max freq first and then start spinner load
Quoted 16 lines
Quoted 3 lines
Doing the above, we likely pick up the correct (RPe) freq being used for case 1 but not for case 2. And even for case 1 RPe might be dynamically changing.
Indeed. It is racy no matter how you look at it. 1 component deciding and 2 other reading and using to take decisions without any lock.
Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
Quoted 7 lines
So the best way to fix such sporadic failures seems to be to expose a '
rpe_enable' sysfs to turn off RPe for such tests/usages where strict userspace control of freq limits is required.@Rodrigo Vivi: so we need to see if we can make this uapi change and then go from there I think.
I believe this is the most sane option right now. We need to document this well and have test cases for both.
So as mentioned above the current patches are for the case with RPe enabled. Let's try to implement the rpe_enable sysfs and then have separate patches for the RPe disabled case as a next step?
Thanks.
--
Ashutosh
Quoted 41 lines
Quoted 7 lines
Quoted 2 lines
Either that, or we turn off efficient frequency when user updates max frequency as well (similar to what we do for min).
I think I disagree here, turning off RPe will affect PnP. I think we should have explicit disable of RPe when that is needed and should actually remove RPe disable when min is updated. Basically IMO the idea should be to provide the best PnP and then tailor the ABI for the best PnP.
We don't have any api problem when rpe is enabled but min <= rpe < max. So Vinay is wondering if we can get this scenario to enable the rpe in guc we don't need to create another api. Then on any read or write we would check this condition and decide on the efficient enabled or not.
But as I told, it is racy anyway.
So, in the end if we don't disable rpe we don't have how to guarantee the freq stable freq api. But if we disable we have the pnp impact. If we try any magic it will be racy.
I'm tending to agree with Ashutosh and that adding the
rpe_enabledis the best option. But document that very well. And yes, update the cached actual rpe anytime that we try to read that.Quoted 10 lines
Quoted 6 lines
Quoted 3 lines
Probably this plus the constant update of the cached RPe.
Caching might be easier if GuC does the same. It can update the task state data with the new value for everyone else to read.
i915 will either need to read the register or read the task state from GuC. You mean reading the task state is more efficient than reading the register?
oh, indeed... we can remove the cache and read from pcode directly anytime we need that. Safer indeed.
Quoted 4 lines
Thanks.
--
Ashutosh