[i915] i915/vcs1-sema exceeds i915/vcs1-busy value
Submitted by andrey.orlov@intel.com
Assigned to Intel GFX Bugs mailing list
Link to original bug (#111788)
Description
Values I got from perf tool (very often we see i915/vcs1-sema > i915/vcs1-busy):
{code}
./perf stat -I 100 -a -e i915/vcs1-busy/,i915/vcs1-sema/,i915/vcs1-wait/ ./sample_multi_transcode -par 0002_FHD23i_1.par
libva info: VA-API version 1.6.0
libva info: va_getDriverName() returns 0
libva info: User requested driver 'iHD'
libva info: Trying to open /opt/intel/mediasdk/lib64/iHD_drv_video.so
libva info: Found init function __vaDriverInit_1_6
libva info: va_openDriver() returns 0
# time counts unit events
0.100160810 34,209,689 ns i915/vcs1-busy/
0.100160810 15,046,281 ns i915/vcs1-sema/
0.100160810 0 ns i915/vcs1-wait/
0.200365878 81,586,469 ns i915/vcs1-busy/
0.200365878 75,075,011 ns i915/vcs1-sema/
0.200365878 0 ns i915/vcs1-wait/
0.300568998 85,209,534 ns i915/vcs1-busy/
0.300568998 44,998,752 ns i915/vcs1-sema/
0.300568998 0 ns i915/vcs1-wait/
0.400733002 85,725,483 ns i915/vcs1-busy/
0.400733002 90,037,698 ns i915/vcs1-sema/
0.400733002 0 ns i915/vcs1-wait/
0.500939741 84,982,430 ns i915/vcs1-busy/
0.500939741 80,097,984 ns i915/vcs1-sema/
0.500939741 0 ns i915/vcs1-wait/
0.601132522 85,518,925 ns i915/vcs1-busy/
0.601132522 90,071,194 ns i915/vcs1-sema/
0.601132522 0 ns i915/vcs1-wait/
0.701333967 83,735,465 ns i915/vcs1-busy/
0.701333967 85,106,697 ns i915/vcs1-sema/
0.701333967 0 ns i915/vcs1-wait/
0.801501148 84,875,175 ns i915/vcs1-busy/
0.801501148 45,109,484 ns i915/vcs1-sema/
0.801501148 0 ns i915/vcs1-wait/
0.901658275 85,498,038 ns i915/vcs1-busy/
0.901658275 30,055,031 ns i915/vcs1-sema/
0.901658275 0 ns i915/vcs1-wait/
1.001859013 86,232,348 ns i915/vcs1-busy/
1.001859013 89,996,108 ns i915/vcs1-sema/
1.001859013 0 ns i915/vcs1-wait/
1.102076724 85,695,796 ns i915/vcs1-busy/
1.102076724 94,994,151 ns i915/vcs1-sema/
1.102076724 0 ns i915/vcs1-wait/
1.202270626 85,122,835 ns i915/vcs1-busy/
1.202270626 69,968,796 ns i915/vcs1-sema/
1.202270626 0 ns i915/vcs1-wait/
1.302443248 85,881,238 ns i915/vcs1-busy/
1.302443248 95,000,911 ns i915/vcs1-sema/
1.302443248 0 ns i915/vcs1-wait/
1.402619770 86,196,687 ns i915/vcs1-busy/
1.402619770 80,003,731 ns i915/vcs1-sema/
1.402619770 0 ns i915/vcs1-wait/
1.502802220 84,797,945 ns i915/vcs1-busy/
1.502802220 100,000,569 ns i915/vcs1-sema/
1.502802220 0 ns i915/vcs1-wait/
1.603005224 84,732,615 ns i915/vcs1-busy/
1.603005224 100,000,407 ns i915/vcs1-sema/
1.603005224 0 ns i915/vcs1-wait/
1.703178374 84,980,237 ns i915/vcs1-busy/
1.703178374 85,010,242 ns i915/vcs1-sema/
1.703178374 0 ns i915/vcs1-wait/
1.803356979 84,558,765 ns i915/vcs1-busy/
1.803356979 35,035,493 ns i915/vcs1-sema/
1.803356979 0 ns i915/vcs1-wait/
1.903561910 85,307,891 ns i915/vcs1-busy/
1.903561910 80,113,763 ns i915/vcs1-sema/
1.903561910 0 ns i915/vcs1-wait/
2.003732306 85,147,056 ns i915/vcs1-busy/
2.003732306 40,049,492 ns i915/vcs1-sema/
2.003732306 0 ns i915/vcs1-wait/
2.103939868 85,385,397 ns i915/vcs1-busy/
2.103939868 5,002,618 ns i915/vcs1-sema/
2.103939868 0 ns i915/vcs1-wait/
2.204145664 86,464,111 ns i915/vcs1-busy/
2.204145664 34,987,714 ns i915/vcs1-sema/
2.204145664 0 ns i915/vcs1-wait/
2.304341045 85,311,024 ns i915/vcs1-busy/
2.304341045 94,976,653 ns i915/vcs1-sema/
2.304341045 0 ns i915/vcs1-wait/
2.404520380 85,852,106 ns i915/vcs1-busy/
2.404520380 34,996,772 ns i915/vcs1-sema/
2.404520380 0 ns i915/vcs1-wait/
2.504729578 85,753,160 ns i915/vcs1-busy/
2.504729578 44,995,847 ns i915/vcs1-sema/
2.504729578 0 ns i915/vcs1-wait/
# time counts unit events
2.604946693 83,258,753 ns i915/vcs1-busy/
2.604946693 74,990,707 ns i915/vcs1-sema/
2.604946693 0 ns i915/vcs1-wait/
2.705148192 85,853,133 ns i915/vcs1-busy/
2.705148192 85,146,362 ns i915/vcs1-sema/
2.705148192 0 ns i915/vcs1-wait/
2.805344742 85,636,079 ns i915/vcs1-busy/
2.805344742 25,033,601 ns i915/vcs1-sema/
2.805344742 0 ns i915/vcs1-wait/
2.905514029 83,719,125 ns i915/vcs1-busy/
2.905514029 65,043,622 ns i915/vcs1-sema/
2.905514029 0 ns i915/vcs1-wait/
3.005722647 85,310,040 ns i915/vcs1-busy/
3.005722647 30,037,930 ns i915/vcs1-sema/
3.005722647 0 ns i915/vcs1-wait/
3.105932982 85,260,192 ns i915/vcs1-busy/
3.105932982 30,031,303 ns i915/vcs1-sema/
3.105932982 0 ns i915/vcs1-wait/
3.206102810 85,288,760 ns i915/vcs1-busy/
3.206102810 50,078,485 ns i915/vcs1-sema/
3.206102810 0 ns i915/vcs1-wait/
3.306279745 85,587,333 ns i915/vcs1-busy/
3.306279745 75,027,713 ns i915/vcs1-sema/
3.306279745 0 ns i915/vcs1-wait/
3.406493454 85,765,888 ns i915/vcs1-busy/
3.406493454 95,011,091 ns i915/vcs1-sema/
3.406493454 0 ns i915/vcs1-wait/
3.506691238 85,091,538 ns i915/vcs1-busy/
3.506691238 85,025,489 ns i915/vcs1-sema/
3.506691238 0 ns i915/vcs1-wait/
3.606898304 85,016,700 ns i915/vcs1-busy/
3.606898304 94,998,820 ns i915/vcs1-sema/
3.606898304 0 ns i915/vcs1-wait/
3.707108156 85,224,696 ns i915/vcs1-busy/
3.707108156 55,041,010 ns i915/vcs1-sema/
3.707108156 0 ns i915/vcs1-wait/
3.807313350 84,935,598 ns i915/vcs1-busy/
3.807313350 99,971,882 ns i915/vcs1-sema/
3.807313350 0 ns i915/vcs1-wait/
3.907520124 86,416,994 ns i915/vcs1-busy/
3.907520124 60,218,944 ns i915/vcs1-sema/
3.907520124 0 ns i915/vcs1-wait/
4.007730814 85,646,119 ns i915/vcs1-busy/
4.007730814 50,073,911 ns i915/vcs1-sema/
4.007730814 0 ns i915/vcs1-wait/
4.107904563 85,931,492 ns i915/vcs1-busy/
4.107904563 80,000,899 ns i915/vcs1-sema/
4.107904563 0 ns i915/vcs1-wait/
4.208073659 85,295,903 ns i915/vcs1-busy/
4.208073659 84,978,887 ns i915/vcs1-sema/
4.208073659 0 ns i915/vcs1-wait/
4.308284750 85,051,978 ns i915/vcs1-busy/
4.308284750 99,963,217 ns i915/vcs1-sema/
4.308284750 0 ns i915/vcs1-wait/
4.408481753 86,220,293 ns i915/vcs1-busy/
4.408481753 100,000,425 ns i915/vcs1-sema/
4.408481753 0 ns i915/vcs1-wait/
4.508689997 85,022,693 ns i915/vcs1-busy/
4.508689997 65,139,017 ns i915/vcs1-sema/
4.508689997 0 ns i915/vcs1-wait/
4.608895935 84,892,627 ns i915/vcs1-busy/
4.608895935 50,069,240 ns i915/vcs1-sema/
4.608895935 0 ns i915/vcs1-wait/
4.709103951 85,038,229 ns i915/vcs1-busy/
4.709103951 80,109,594 ns i915/vcs1-sema/
4.709103951 0 ns i915/vcs1-wait/
4.809279557 85,325,514 ns i915/vcs1-busy/
4.809279557 85,050,025 ns i915/vcs1-sema/
4.809279557 0 ns i915/vcs1-wait/
4.909499959 85,595,340 ns i915/vcs1-busy/
4.909499959 29,999,102 ns i915/vcs1-sema/
4.909499959 0 ns i915/vcs1-wait/
5.009715840 86,633,730 ns i915/vcs1-busy/
5.009715840 94,996,584 ns i915/vcs1-sema/
5.009715840 0 ns i915/vcs1-wait/
# time counts unit events
5.109927801 85,274,181 ns i915/vcs1-busy/
5.109927801 94,968,127 ns i915/vcs1-sema/
5.109927801 0 ns i915/vcs1-wait/
5.210138711 84,449,606 ns i915/vcs1-busy/
5.210138711 40,170,670 ns i915/vcs1-sema/
5.210138711 0 ns i915/vcs1-wait/
5.310351659 85,629,044 ns i915/vcs1-busy/
5.310351659 45,012,072 ns i915/vcs1-sema/
5.310351659 0 ns i915/vcs1-wait/
5.410568218 86,679,014 ns i915/vcs1-busy/
5.410568218 99,994,658 ns i915/vcs1-sema/
5.410568218 0 ns i915/vcs1-wait/
5.510780105 85,682,007 ns i915/vcs1-busy/
5.510780105 0 ns i915/vcs1-sema/
5.510780105 0 ns i915/vcs1-wait/
5.610987172 84,631,229 ns i915/vcs1-busy/
5.610987172 50,027,357 ns i915/vcs1-sema/
5.610987172 0 ns i915/vcs1-wait/
5.711192665 85,008,656 ns i915/vcs1-busy/
5.711192665 80,007,406 ns i915/vcs1-sema/
5.711192665 0 ns i915/vcs1-wait/
5.811410934 84,773,228 ns i915/vcs1-busy/
5.811410934 35,052,545 ns i915/vcs1-sema/
5.811410934 0 ns i915/vcs1-wait/
5.911616718 84,900,639 ns i915/vcs1-busy/
5.911616718 30,006,841 ns i915/vcs1-sema/
5.911616718 0 ns i915/vcs1-wait/
6.011827360 86,234,742 ns i915/vcs1-busy/
6.011827360 89,983,850 ns i915/vcs1-sema/
6.011827360 0 ns i915/vcs1-wait/
6.081667753 38,056,952 ns i915/vcs1-busy/
6.081667753 40,003,467 ns i915/vcs1-sema/
6.081667753 0 ns i915/vcs1-wait/
{code}
We have performed one another experiment with GPU frequency (only gt_max_freq_mhz, gt_boost_freq_mhz): default (950), 900, 800, 700, 600, 500, 400.
Example:
echo 800 > /sys/class/drm/card0/gt_max_freq_mhz
echo 800 > /sys/class/drm/card0/gt_boost_freq_mhz
Other settings - system defaults.
And only one case, where I see big amount of bad samples - 700 MHz.
As already was discussed in our internal bug, the fix for more-precise-sampling helps:
{code}
From 0a1252963a958aefb09d0113afd0caf3359a4462 Mon Sep 17 00:00:00 2001
From: Tvrtko Ursulin
Date: Wed, 11 Sep 2019 17:52:31 +0100
Subject: [PATCH] more precise sampling
Signed-off-by: Tvrtko Ursulin
---
drivers/gpu/drm/i915/i915_pmu.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/i915/i915_pmu.c b/drivers/gpu/drm/i915/i915_pmu.c
index 623ad32303a1..9f0bb3b85401 100644
--- a/drivers/gpu/drm/i915/i915_pmu.c
+++ b/drivers/gpu/drm/i915/i915_pmu.c
@@ -17,7 +17,7 @@
#include "intel_pm.h"
/* Frequency for the sampling timer for events which need it. */
-#define FREQUENCY 200
+#define FREQUENCY 1000
#define PERIOD max_t(u64, 10000, NSEC_PER_SEC / FREQUENCY)
#define ENGINE_SAMPLE_MASK \
--
2.20.1
{code}