kms_pipe_crc_basic: Only test one pipe in the basic set
PATCH 1/2 kms_pipe_crc_basic: Only test one pipe in the basic set
From: Tvrtko Ursulin <[email protected]>
I would argue it is enough to test one pipe in the BAT set.
--- tests/kms_pipe_crc_basic.c | 23 +++++++++++++---------- 1 file changed, 13 insertions(+), 10 deletions(-)
tests/kms_pipe_crc_basic.c +13 -10
@@ -180,6 +180,9 @@ static void test_read_crc(data_t *data, int pipe, unsigned flags)data_t data = {0 , }; +#define test_prefix (i) ((i) == 0 ? "basic-" : "") +#define pipe_name (i) ((i) + 'A') + igt_main {igt_skip_on_simulation ();
@@ -196,39 +199,39 @@ igt_mainigt_display_init (&data.display, data.drm_fd); } - igt_subtest("bad-pipe") + igt_subtest("basic-bad-pipe")test_bad_command (&data,"pipe D none" ); - igt_subtest("bad-source") + igt_subtest("basic-bad-source")test_bad_command (&data,"pipe A foo" ); - igt_subtest("bad-nb-words-1") + igt_subtest("basic-bad-nb-words-1")test_bad_command (&data,"pipe foo" ); - igt_subtest("bad-nb-words-3") + igt_subtest("basic-bad-nb-words-3")test_bad_command (&data,"pipe A none option" );for (int i =0 ; i <3 ; i++) { - igt_subtest_f("read-crc-pipe-%c", 'A'+i) + igt_subtest_f("%sread-crc-pipe-%c", test_prefix(i), pipe_name(i))test_read_crc (&data, i,0 ); - igt_subtest_f("read-crc-pipe-%c-frame-sequence", 'A'+i) + igt_subtest_f("%sread-crc-pipe-%c-frame-sequence", test_prefix(i), pipe_name(i))test_read_crc (&data, i,TEST_SEQUENCE ); - igt_subtest_f("nonblocking-crc-pipe-%c", 'A'+i) + igt_subtest_f("%snonblocking-crc-pipe-%c", test_prefix(i), pipe_name(i))test_read_crc (&data, i,TEST_NONBLOCK ); - igt_subtest_f("nonblocking-crc-pipe-%c-frame-sequence", 'A'+i) + igt_subtest_f("%snonblocking-crc-pipe-%c-frame-sequence", test_prefix(i), pipe_name(i))test_read_crc (&data, i,TEST_SEQUENCE |TEST_NONBLOCK ); - igt_subtest_f("suspend-read-crc-pipe-%c", 'A'+i) { + igt_subtest_f("%ssuspend-read-crc-pipe-%c", test_prefix(i), pipe_name(i)) {igt_skip_on (i >= data.display.n_pipes);igt_system_suspend_autoresume ();test_read_crc (&data, i,0 ); } - igt_subtest_f("hang-read-crc-pipe-%c", 'A'+i) { + igt_subtest_f("%shang-read-crc-pipe-%c", test_prefix(i), pipe_name(i)) {igt_hang_ring_t hang =igt_hang_ring (data.drm_fd,I915_EXEC_RENDER );test_read_crc (&data, i,0 );
Signature · 6 lines
1.9.1 _______________________________________________ Intel-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/intel-gfx
PATCH 2/2 kms_pipe_crc_basic: Rename test to kms_pipe_crc
From: Tvrtko Ursulin <[email protected]>
Previous patch has marked individual subtests as basic so now rename the test binary name accordingly.
--- tests/Makefile.sources | 2 +- tests/kms_pipe_crc.c | 246 +++++++++++++++++++++++++++++++++++++++++++++ tests/kms_pipe_crc_basic.c | 246 --------------------------------------------- 3 files changed, 247 insertions(+), 247 deletions(-) create mode 100644 tests/kms_pipe_crc.c delete mode 100644 tests/kms_pipe_crc_basic.c
tests/Makefile.sources +1 -1
@@ -92,7 +92,7 @@ TESTS_progs_M = \ kms_mmio_vs_cs_flip \ kms_pipe_b_c_ivb \ kms_pipe_color \ - kms_pipe_crc_basic \ + kms_pipe_crc \ kms_plane \ kms_psr_sink_crc \ kms_render \
tests/kms_pipe_crc.c new file +246 -0
246 lines not on this page · raw message
tests/kms_pipe_crc_basic.c deleted +0 -246
246 lines not on this page · raw message
Signature · 1 line
1.9.1
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Quoted 36 lines · Tvrtko Ursulin
From: Tvrtko Ursulin <[email protected]>
I would argue it is enough to test one pipe in the BAT set.
Signed-off-by Tvrtko Ursulin <[email protected]>--- tests/kms_pipe_crc_basic.c | 23 +++++++++++++---------- 1 file changed, 13 insertions(+), 10 deletions(-)tests/kms_pipe_crc_basic.c +5
-3@@ -180,6 +180,9 @@ static void test_read_crc(data_t *data, int pipe, unsigned flags)data_t data = {0 , }; +#define test_prefix (i) ((i) == 0 ? "basic-" : "") +#define pipe_name (i) ((i) + 'A') + igt_main {igt_skip_on_simulation ();@@ -196,39 +199,39 @@ igt_mainigt_display_init (&data.display, data.drm_fd); } - igt_subtest("bad-pipe") + igt_subtest("basic-bad-pipe")test_bad_command (&data,"pipe D none" ); - igt_subtest("bad-source") + igt_subtest("basic-bad-source")test_bad_command (&data,"pipe A foo" ); -igt_subtest ("bad-nb-words-1" )
tests/kms_pipe_crc_basic.c igt_main · line 215
Quoted 10 lines
tests/kms_pipe_crc_basic.c +3
-2+ igt_subtest("basic-bad-nb-words-1")test_bad_command (&data,"pipe foo" ); - igt_subtest("bad-nb-words-3") + igt_subtest("basic-bad-nb-words-3")test_bad_command (&data,"pipe A none option" );for (int i =0 ; i <3 ; i++) { -igt_subtest_f ("read-crc-pipe-%c" ,'A' +i) +igt_subtest_f ("%sread-crc-pipe-%c" ,test_prefix (i),pipe_name (i))
So the CRC is the backchannel through which we measure the output on the screen matches expectations. Do any of the following belong in the basic test set? Having demonstrated that the CRC is functional, all the rest are components of other tests - and if a bug if found in any of the other basic tests, one can run the entire kms_pipe_crc to sanity check the test suite itself.
As far as I can see this is a test that tests the BAT and not the proposed changes.
-Chris
Signature · 5 lines
Chris Wilson, Intel Open Source Technology Centre _______________________________________________ Intel-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Quoted 44 lines · Chris Wilson
Quoted 43 lines · Tvrtko Ursulin
From: Tvrtko Ursulin <[email protected]>
I would argue it is enough to test one pipe in the BAT set.
Signed-off-by Tvrtko Ursulin <[email protected]>--- tests/kms_pipe_crc_basic.c | 23 +++++++++++++---------- 1 file changed, 13 insertions(+), 10 deletions(-)tests/kms_pipe_crc_basic.c +7
-4@@ -180,6 +180,9 @@ static void test_read_crc(data_t *data, int pipe, unsigned flags)data_t data = {0 , }; +#define test_prefix (i) ((i) == 0 ? "basic-" : "") +#define pipe_name (i) ((i) + 'A') + igt_main {igt_skip_on_simulation ();@@ -196,39 +199,39 @@ igt_mainigt_display_init (&data.display, data.drm_fd); } - igt_subtest("bad-pipe") + igt_subtest("basic-bad-pipe")test_bad_command (&data,"pipe D none" ); - igt_subtest("bad-source") + igt_subtest("basic-bad-source")test_bad_command (&data,"pipe A foo" ); - igt_subtest("bad-nb-words-1") + igt_subtest("basic-bad-nb-words-1")test_bad_command (&data,"pipe foo" ); - igt_subtest("bad-nb-words-3") + igt_subtest("basic-bad-nb-words-3")test_bad_command (&data,"pipe A none option" );
Quoted 10 lines
tests/kms_pipe_crc_basic.c igt_main · line 215
Quoted 3 lines
tests/kms_pipe_crc_basic.c +1
-1for (int i =0 ; i <3 ; i++) { -igt_subtest_f ("read-crc-pipe-%c" ,'A' +i) +igt_subtest_f ("%sread-crc-pipe-%c" ,test_prefix (i),pipe_name (i))So the CRC is the backchannel through which we measure the output on the screen matches expectations. Do any of the following belong in the basic test set? Having demonstrated that the CRC is functional, all the rest are components of other tests - and if a bug if found in any of the other basic tests, one can run the entire
kms_pipe_crcto sanity check the test suite itself.
The reason I wanted to include kms_pipe_crc_basic was not to validate the kernel driver (it's a side-effect), but to make sure the CRC support is really stable. Otherwise we can't rely on a full BAT run at all.
And yes every single subtest (including testing on all the pipes) included in there is for a bug in the CRC code that actually happened.
I know that we're not there yet with CI, and we don't yet run all the nice kms_*crc tests we have. But imo this is crucial prep work for that, and we really shouldn't reduce test coverage in this area. Same reason we have 1 edp sink crc testcase, essentially it's to make sure that we have a reasonable basis to run the full test set.
Imo BAT should guarantee two things:
- no insta-fireworks
- all the validation stuff actually works and running more tests will give you valid results
From the above reasons I'm against this patch.
-Daniel
Signature · 7 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch _______________________________________________ Intel-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Quoted 56 lines · Daniel Vetter
Quoted 44 lines · Chris Wilson
Quoted 43 lines · Tvrtko Ursulin
From: Tvrtko Ursulin <[email protected]>
I would argue it is enough to test one pipe in the BAT set.
Signed-off-by Tvrtko Ursulin <[email protected]>--- tests/kms_pipe_crc_basic.c | 23 +++++++++++++---------- 1 file changed, 13 insertions(+), 10 deletions(-)tests/kms_pipe_crc_basic.c +7
-4@@ -180,6 +180,9 @@ static void test_read_crc(data_t *data, int pipe, unsigned flags)data_t data = {0 , }; +#define test_prefix (i) ((i) == 0 ? "basic-" : "") +#define pipe_name (i) ((i) + 'A') + igt_main {igt_skip_on_simulation ();@@ -196,39 +199,39 @@ igt_mainigt_display_init (&data.display, data.drm_fd); } - igt_subtest("bad-pipe") + igt_subtest("basic-bad-pipe")test_bad_command (&data,"pipe D none" ); - igt_subtest("bad-source") + igt_subtest("basic-bad-source")test_bad_command (&data,"pipe A foo" ); - igt_subtest("bad-nb-words-1") + igt_subtest("basic-bad-nb-words-1")test_bad_command (&data,"pipe foo" ); - igt_subtest("bad-nb-words-3") + igt_subtest("basic-bad-nb-words-3")test_bad_command (&data,"pipe A none option" );Quoted 10 lines
tests/kms_pipe_crc_basic.c igt_main · line 215
Quoted 3 lines
tests/kms_pipe_crc_basic.c +1
-1for (int i =0 ; i <3 ; i++) { -igt_subtest_f ("read-crc-pipe-%c" ,'A' +i) +igt_subtest_f ("%sread-crc-pipe-%c" ,test_prefix (i),pipe_name (i))So the CRC is the backchannel through which we measure the output on the screen matches expectations. Do any of the following belong in the basic test set? Having demonstrated that the CRC is functional, all the rest are components of other tests - and if a bug if found in any of the other basic tests, one can run the entire
kms_pipe_crcto sanity check the test suite itself.
Quoted 3 lines
The reason I wanted to include
kms_pipe_crc_basicwas not to validate the kernel driver (it's a side-effect), but to make sure the CRC support is really stable. Otherwise we can't rely on a full BAT run at all.
Okay, so these patches are more about defining on what full BAT is, not trying to break full BAT.
Quoted 9 lines
And yes every single subtest (including testing on all the pipes) included in there is for a bug in the CRC code that actually happened.
I know that we're not there yet with CI, and we don't yet run all the nice kms_*crc tests we have. But imo this is crucial prep work for that, and we really shouldn't reduce test coverage in this area. Same reason we have 1 edp sink crc testcase, essentially it's to make sure that we have a reasonable basis to run the full test set.
Quoted 6 lines
Imo BAT should guarantee two things:
- no insta-fireworks
- all the validation stuff actually works and running more tests will give you valid resultsFrom the above reasons I'm against this patch.
My general idea was that the per-patch BAT (todays basic set) could get away with running on one pipe, and then the overnight one (which we do not have yet) would run the more extensive set.
So even if BAT runs everything on pipe A, it is still self-validating. It is not that some tests would run on pipe A and some on all pipes, that at least was the idea.
Motivation simply was that the current BAT run did 80 backlight on-off cycles on my BDW RVP and just waiting for eDP to turn on 80 times takes ~50 seconds in total. The test runtime was around eight minutes here.
When I turned off legacy fbdev that went to ~4 minutes. And then with these two (and other two) patches I shaved off 1-2 minute more.
Considering the time spent vs the typical system usage, which is fewer display on-offs vs. much much more GEM operations, the current state seemed like a sub-optimal balance to me. Especially in the context of not allowing more GEM basic tests since the whole suite takes so long.
Also, for the kms_flip patches I posted I should probably have added cloned test cases for the ones that I modified. So that we have for example basic-plain-flip (one pipe) and plain-flip (all pipes as today).
Regards,
Tvrtko
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Em Seg, 2016-04-25 às 10:45 +0100, Tvrtko Ursulin escreveu:
Quoted 115 lines · Daniel Vetter
Quoted 66 lines · Chris Wilson
Quoted 49 lines · Tvrtko Ursulin
Quoted 47 lines
From: Tvrtko Ursulin <[email protected]>
I would argue it is enough to test one pipe in the BAT set.
Signed-off-by Tvrtko Ursulin <[email protected]>---
tests/kms_pipe_crc_basic.c| 23 +++++++++++++----------1 file changed, 13 insertions(+), 10 deletions(-)tests/kms_pipe_crc_basic.c +3
-0@@ -180,6 +180,9 @@ static void test_read_crc(data_t *data, int pipe, unsigned flags) data_t data = {0, }; +#define test_prefix (i) ((i) == 0 ? "basic-" : "") +#define pipe_name (i) ((i) + 'A') +
igt_main
{
igt_skip_on_simulation();tests/kms_pipe_crc_basic.c +4
-4@@ -196,39 +199,39 @@ igt_main igt_display_init(&data.display, data.drm_fd); } - igt_subtest("bad-pipe") + igt_subtest("basic-bad-pipe") test_bad_command(&data, "pipe D none"); - igt_subtest("bad-source") + igt_subtest("basic-bad-source") test_bad_command(&data, "pipe A foo"); - igt_subtest("bad-nb-words-1") + igt_subtest("basic-bad-nb-words-1") test_bad_command(&data, "pipe foo"); -igt_subtest ("bad-nb-words-3" ) +igt_subtest ("basic-bad-nb-words-3" )
test_bad_command(&data, "pipe A none option");Quoted 15 lines
Quoted 4 lines
for (int i = 0; i < 3; i++) {
tests/kms_pipe_crc_basic.c +1
-1-igt_subtest_f ("read-crc-pipe-%c" ,'A' +i) + igt_subtest_f("%sread-crc-pipe-%c",
test_prefix(i),pipe_name(i))So the CRC is the backchannel through which we measure the output on the
screen matches expectations. Do any of the following belong in the basic
test set? Having demonstrated that the CRC is functional, all the rest
are components of other tests - and if a bug if found in any of the
other basic tests, one can run the entirekms_pipe_crcto sanity check
the test suite itself.Quoted 5 lines
The reason I wanted to include
kms_pipe_crc_basicwas not to validate the
kernel driver (it's a side-effect), but to make sure the CRC support is
really stable. Otherwise we can't rely on a full BAT run at all.Okay, so these patches are more about defining on what full BAT is, not
trying to break full BAT.Quoted 15 lines
And yes every single subtest (including testing on all the pipes) included
in there is for a bug in the CRC code that actually happened.I know that we're not there yet with CI, and we don't yet run all the nice
kms_*crc tests we have. But imo this is crucial prep work for that, and
we really shouldn't reduce test coverage in this area. Same reason we have
1 edp sink crc testcase, essentially it's to make sure that we have a
reasonable basis to run the full test set.Quoted 7 lines
Imo BAT should guarantee two things:
- no insta-fireworks
- all the validation stuff actually works and running more tests will give
you valid resultsFrom the above reasons I'm against this patch.
My general idea was that the per-patch BAT (todays basic set) could get
away with running on one pipe, and then the overnight one (which we do
not have yet) would run the more extensive set.So even if BAT runs everything on pipe A, it is still self- validating.
It is not that some tests would run on pipe A and some on all pipes, that at least was the idea.Motivation simply was that the current BAT run did 80 backlight on- off
cycles on my BDW RVP and just waiting for eDP to turn on 80 times takes
~50 seconds in total. The test runtime was around eight minutes here.
Quoted 3 lines
When I turned off legacy fbdev that went to ~4 minutes. And then with
these two (and other two) patches I shaved off 1-2 minute more.
One of the things about our driver is that it tries to be very very conservative regarding the eDP timings, leading to waits longer than necessary for pretty much everyone. We tried to optimize this in the past, but due to the fact that we have a single code base to support every product released since gen 2, pretty much any sort of optimization attempt tends to break someone's machine because maybe they have a bad BIOS or a bad VBT or a bad panel or something else. Now that this problem is impacting our ability to run tests, maybe it is time to start thinking about these optimizations again?
If you want to investigate this, you could start by looking at the patches stored here:
https://cgit.freedesktop.org/~pzanoni/linux/?h=edp-reverts
The first patch adds an option that you could try to play it, and the other patches contain references to my previous optimization attempts that ended up being reverted. You'll have to search the mailing list for the actual regression reports. Feel free to adopt the patches and change them.
Perhaps we could also think about some local changes that would only optimize QA's systems, like a module option used by them? The problem is that this would mean QA wouldn't be testing exactly what's upstream. Still, it could be worth it, depending on how much time it saves.
Quoted 14 lines
Considering the time spent vs the typical system usage, which is fewer
display on-offs vs. much much more GEM operations, the current state seemed like a sub-optimal balance to me. Especially in the context of
not allowing more GEM basic tests since the whole suite takes so long.Also, for the
kms_flippatches I posted I should probably have added cloned test cases for the ones that I modified. So that we have for example basic-plain-flip (one pipe) and plain-flip (all pipes as today).
Quoted 8 lines
Regards,
Tvrtko
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Quoted 21 lines · Paulo R
One of the things about our driver is that it tries to be very very conservative regarding the eDP timings, leading to waits longer than necessary for pretty much everyone. We tried to optimize this in the past, but due to the fact that we have a single code base to support every product released since gen 2, pretty much any sort of optimization attempt tends to break someone's machine because maybe they have a bad BIOS or a bad VBT or a bad panel or something else. Now that this problem is impacting our ability to run tests, maybe it is time to start thinking about these optimizations again?
If you want to investigate this, you could start by looking at the patches stored here:
https://cgit.freedesktop.org/~pzanoni/linux/?h=edp-reverts
The first patch adds an option that you could try to play it, and the other patches contain references to my previous optimization attempts that ended up being reverted. You'll have to search the mailing list for the actual regression reports. Feel free to adopt the patches and change them.
Quoted 4 lines
Perhaps we could also think about some local changes that would only optimize QA's systems, like a module option used by them? The problem is that this would mean QA wouldn't be testing exactly what's upstream. Still, it could be worth it, depending on how much time it saves.
Broken record, but I don't like code disabled by default. What we could try is to get this merged & enabled, but only for recent-ish vbt version. That way the support nightmare should be manageable, and hopefully we can figure out for recent firmware how it's supposed to work.
CI could then still use the option to enable fast timings everywhere (or at least on those machines where it works).
Thanks, Daniel
Signature · 7 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch _______________________________________________ Intel-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Quoted 3 lines · Tvrtko Ursulin
From: Tvrtko Ursulin <[email protected]>
I would argue it is enough to test one pipe in the BAT set.
FWIW BYT/CHV DSI may be fixed to port C which is fixed to pipe B, skipping the test for pipe A.
BR,
Jani.
Quoted 70 lines
Signed-off-by Tvrtko Ursulin <[email protected]>--- tests/kms_pipe_crc_basic.c | 23 +++++++++++++---------- 1 file changed, 13 insertions(+), 10 deletions(-)tests/kms_pipe_crc_basic.c +13
-10@@ -180,6 +180,9 @@ static void test_read_crc(data_t *data, int pipe, unsigned flags)data_t data = {0 , }; +#define test_prefix (i) ((i) == 0 ? "basic-" : "") +#define pipe_name (i) ((i) + 'A') + igt_main {igt_skip_on_simulation ();@@ -196,39 +199,39 @@ igt_mainigt_display_init (&data.display, data.drm_fd); } - igt_subtest("bad-pipe") + igt_subtest("basic-bad-pipe")test_bad_command (&data,"pipe D none" ); - igt_subtest("bad-source") + igt_subtest("basic-bad-source")test_bad_command (&data,"pipe A foo" ); - igt_subtest("bad-nb-words-1") + igt_subtest("basic-bad-nb-words-1")test_bad_command (&data,"pipe foo" ); - igt_subtest("bad-nb-words-3") + igt_subtest("basic-bad-nb-words-3")test_bad_command (&data,"pipe A none option" );for (int i =0 ; i <3 ; i++) { - igt_subtest_f("read-crc-pipe-%c", 'A'+i) + igt_subtest_f("%sread-crc-pipe-%c", test_prefix(i), pipe_name(i))test_read_crc (&data, i,0 ); - igt_subtest_f("read-crc-pipe-%c-frame-sequence", 'A'+i) + igt_subtest_f("%sread-crc-pipe-%c-frame-sequence", test_prefix(i), pipe_name(i))test_read_crc (&data, i,TEST_SEQUENCE ); - igt_subtest_f("nonblocking-crc-pipe-%c", 'A'+i) + igt_subtest_f("%snonblocking-crc-pipe-%c", test_prefix(i), pipe_name(i))test_read_crc (&data, i,TEST_NONBLOCK ); - igt_subtest_f("nonblocking-crc-pipe-%c-frame-sequence", 'A'+i) + igt_subtest_f("%snonblocking-crc-pipe-%c-frame-sequence", test_prefix(i), pipe_name(i))test_read_crc (&data, i,TEST_SEQUENCE |TEST_NONBLOCK ); - igt_subtest_f("suspend-read-crc-pipe-%c", 'A'+i) { + igt_subtest_f("%ssuspend-read-crc-pipe-%c", test_prefix(i), pipe_name(i)) {igt_skip_on (i >= data.display.n_pipes);igt_system_suspend_autoresume ();test_read_crc (&data, i,0 ); } - igt_subtest_f("hang-read-crc-pipe-%c", 'A'+i) { + igt_subtest_f("%shang-read-crc-pipe-%c", test_prefix(i), pipe_name(i)) {igt_hang_ring_t hang =igt_hang_ring (data.drm_fd,I915_EXEC_RENDER );test_read_crc (&data, i,0 );
Signature · 5 lines
Jani Nikula, Intel Open Source Technology Center _______________________________________________ Intel-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/intel-gfx