| Message ID | 20171214154347.25392-1-marta.lofstedt@intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
https://patchwork.freedesktop.org/series/35365/ Judging from PW results, THRESHOLD_TOTAL 95 is still not enough for KBL-shards. That is interesting. /Marta > -----Original Message----- > From: Lofstedt, Marta > Sent: Thursday, December 14, 2017 5:44 PM > To: intel-gfx@lists.freedesktop.org > Cc: Latvala, Petri <petri.latvala@intel.com>; Taylor, Clinton A > <clinton.a.taylor@intel.com>; Lofstedt, Marta <marta.lofstedt@intel.com> > Subject: [PATCH i-g-t] test/kms_sysfs_edid_timing : Increase > THRESHOLD_PER_CONNECTOR and THRESHOLD_TOTAL > > The discussion on how to fix the issues has been stagnant for a long time, see > the Bugzilla below and: > https://patchwork.freedesktop.org/patch/170429/ > > After only changing display cable the result of my machine went from always > failing to hit a WARN. This is also illustrated on CI-shards, where the shard-apl > flip-flop between pass and WARN and the shard-kbl always fail. This made > me realize that the THRESHOLD_PER_CONNECTOR also needed to be > updated, while previous discussion only touched the THRESHOLD_TOTAL. As > discussed in provided links the motivation for this test is to discover big > anomalies in edid timings, but with the current tight thresholds we are hitting > the issues too often. > > Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=100047 > Signed-off-by: Marta Lofstedt <marta.lofstedt@intel.com> > --- > tests/kms_sysfs_edid_timing.c | 4 ++-- > 1 file changed, 2 insertions(+), 2 deletions(-) > > diff --git a/tests/kms_sysfs_edid_timing.c b/tests/kms_sysfs_edid_timing.c > index 12013881..7dbd3fe6 100644 > --- a/tests/kms_sysfs_edid_timing.c > +++ b/tests/kms_sysfs_edid_timing.c > @@ -26,8 +26,8 @@ > #include <fcntl.h> > #include <sys/stat.h> > > -#define THRESHOLD_PER_CONNECTOR 10 > -#define THRESHOLD_TOTAL 50 > +#define THRESHOLD_PER_CONNECTOR 15 > +#define THRESHOLD_TOTAL 95 > #define CHECK_TIMES 15 > > IGT_TEST_DESCRIPTION("This check the time we take to read the content of > all " > -- > 2.11.0
It is possoble to get very mysterious results from kms_sysfs_edid timing
Since there is no try-bot for IGT I am using IGT PW to get more data.
Marta Lofstedt (2):
test/kms_sysfs_edid_timing : Increase THRESHOLD_PER_CONNECTOR and
THRESHOLD_TOTAL
HACK to get kms_sysfs_edid_timing run on more machines
tests/intel-ci/fast-feedback.testlist | 2 +-
tests/kms_sysfs_edid_timing.c | 14 +++++++-------
2 files changed, 8 insertions(+), 8 deletions(-)
diff --git a/tests/kms_sysfs_edid_timing.c b/tests/kms_sysfs_edid_timing.c index 12013881..7dbd3fe6 100644 --- a/tests/kms_sysfs_edid_timing.c +++ b/tests/kms_sysfs_edid_timing.c @@ -26,8 +26,8 @@ #include <fcntl.h> #include <sys/stat.h> -#define THRESHOLD_PER_CONNECTOR 10 -#define THRESHOLD_TOTAL 50 +#define THRESHOLD_PER_CONNECTOR 15 +#define THRESHOLD_TOTAL 95 #define CHECK_TIMES 15 IGT_TEST_DESCRIPTION("This check the time we take to read the content of all "
The discussion on how to fix the issues has been stagnant for a long time, see the Bugzilla below and: https://patchwork.freedesktop.org/patch/170429/ After only changing display cable the result of my machine went from always failing to hit a WARN. This is also illustrated on CI-shards, where the shard-apl flip-flop between pass and WARN and the shard-kbl always fail. This made me realize that the THRESHOLD_PER_CONNECTOR also needed to be updated, while previous discussion only touched the THRESHOLD_TOTAL. As discussed in provided links the motivation for this test is to discover big anomalies in edid timings, but with the current tight thresholds we are hitting the issues too often. Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=100047 Signed-off-by: Marta Lofstedt <marta.lofstedt@intel.com> --- tests/kms_sysfs_edid_timing.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)