Re: [Intel-gfx] [PATCH 0/4] Enable GuC submission by default on DG1
On Fri, Aug 06, 2021 at 01:34:33PM +0200, Thomas Hellström (Intel) wrote: > Hi, > > On 8/3/21 7:26 PM, Matthew Brost wrote: > > On Tue, Aug 03, 2021 at 02:15:13PM +0200, Daniel Vetter wrote: > > > On Tue, Aug 3, 2021 at 6:53 AM Matthew Brost > > > wrote: > > > > Minimum set of patches to enable GuC submission on DG1 and enable it by > > > > default. > > > > > > > > A little difficult to test as IGTs do not work with DG1 due to a bunch > > > > of uAPI features being disabled (e.g. relocations, caching memory > > > > options, etc...). > > > Matt Auld has an igt series which fixes a lot of this stuff, would be > > > good to do at least a Test-With run with that. > > > > It looks like Maarten now merged Matt's series to IGT. Great. > > There is a series on IGT trybot with pending work to have some igt tests > support relocations, > > https://patchwork.freedesktop.org/series/92043/ > Will take a look but at the moment we are blocked because SLPC won't init on the DG1 in CI but it works just fine on all other parts I've tried in RIL. We suspect the DG1 in CI is an early stepping and we may be missing workarounds. Have a possible fix, just need to try it out. We also might just want to replace the DG1 part in CI with a newer stepping so we don't have to upstream WAs for non-shipping parts. > One of the tests that have WIP fixes is gem_exec_whisper, and that > particular test has historically shown occasional hangs with GuC submission > on DG1 so it would be very desirable if we could make that test in > particular work (I haven't verified that that's the case) reliably. > I just ran gem_exec_whisper on DG1, with GuC submission, and kernel hacked to allow relocs. It passed for me. Only 1 run though so it is possible there are still intermittent issues. We really need to get CI up and running ASAP on all platforms where GuC submission is POR. Perhaps on all gen11+ platforms as well because GuC submission is supported and the more coverage we can get, the better. > Also the following series: > > https://patchwork.freedesktop.org/series/93455/ I'll check this series out too. Matt > > tries a bit harder to get some more tests running, squashing the above > series on top of latest IGT. > > Thanks, > /Thomas > >
Re: [Intel-gfx] [PATCH 0/4] Enable GuC submission by default on DG1
On 8/6/21 1:34 PM, Thomas Hellström (Intel) wrote: Hi, On 8/3/21 7:26 PM, Matthew Brost wrote: On Tue, Aug 03, 2021 at 02:15:13PM +0200, Daniel Vetter wrote: On Tue, Aug 3, 2021 at 6:53 AM Matthew Brost wrote: Minimum set of patches to enable GuC submission on DG1 and enable it by default. A little difficult to test as IGTs do not work with DG1 due to a bunch of uAPI features being disabled (e.g. relocations, caching memory options, etc...). Matt Auld has an igt series which fixes a lot of this stuff, would be good to do at least a Test-With run with that. It looks like Maarten now merged Matt's series to IGT. There is a series on IGT trybot with pending work to have some igt tests support relocations, https://patchwork.freedesktop.org/series/92043/ One of the tests that have WIP fixes is gem_exec_whisper, and that particular test has historically shown occasional hangs with GuC submission on DG1 so it would be very desirable if we could make that test in particular work (I haven't verified that that's the case) reliably. Also the following series: https://patchwork.freedesktop.org/series/93455/ tries a bit harder to get some more tests running, squashing the above series on top of latest IGT. Thanks, /Thomas Just verified that gem-exec-whisper is running now on DG1 on latest igt and the above series. Haven't tried with GuC submission, though. /Thomas
Re: [Intel-gfx] [PATCH 0/4] Enable GuC submission by default on DG1
On 8/6/21 1:34 PM, Thomas Hellström (Intel) wrote: Hi, On 8/3/21 7:26 PM, Matthew Brost wrote: On Tue, Aug 03, 2021 at 02:15:13PM +0200, Daniel Vetter wrote: On Tue, Aug 3, 2021 at 6:53 AM Matthew Brost wrote: Minimum set of patches to enable GuC submission on DG1 and enable it by default. A little difficult to test as IGTs do not work with DG1 due to a bunch of uAPI features being disabled (e.g. relocations, caching memory options, etc...). Matt Auld has an igt series which fixes a lot of this stuff, would be good to do at least a Test-With run with that. It looks like Maarten now merged Matt's series to IGT. There is a series on IGT trybot with pending work to have some igt tests support relocations, https://patchwork.freedesktop.org/series/92043/ One of the tests that have WIP fixes is gem_exec_whisper, and that particular test has historically shown occasional hangs with GuC submission on DG1 so it would be very desirable if we could make that test in particular work (I haven't verified that that's the case) reliably. Also the following series: https://patchwork.freedesktop.org/series/93455/ tries a bit harder to get some more tests running, squashing the above series on top of latest IGT. Thanks, /Thomas And also while we're working on getting igt adapted to uapi changes and to get more LMEM coverage in there, an IMO relevant test case to run manually is "piglit quick"on top of DG1-enabled OpenGL checking for regressions and significant changes in execution time. /Thomas
Re: [Intel-gfx] [PATCH 0/4] Enable GuC submission by default on DG1
Hi, On 8/3/21 7:26 PM, Matthew Brost wrote: On Tue, Aug 03, 2021 at 02:15:13PM +0200, Daniel Vetter wrote: On Tue, Aug 3, 2021 at 6:53 AM Matthew Brost wrote: Minimum set of patches to enable GuC submission on DG1 and enable it by default. A little difficult to test as IGTs do not work with DG1 due to a bunch of uAPI features being disabled (e.g. relocations, caching memory options, etc...). Matt Auld has an igt series which fixes a lot of this stuff, would be good to do at least a Test-With run with that. It looks like Maarten now merged Matt's series to IGT. There is a series on IGT trybot with pending work to have some igt tests support relocations, https://patchwork.freedesktop.org/series/92043/ One of the tests that have WIP fixes is gem_exec_whisper, and that particular test has historically shown occasional hangs with GuC submission on DG1 so it would be very desirable if we could make that test in particular work (I haven't verified that that's the case) reliably. Also the following series: https://patchwork.freedesktop.org/series/93455/ tries a bit harder to get some more tests running, squashing the above series on top of latest IGT. Thanks, /Thomas
Re: [PATCH 0/4] Enable GuC submission by default on DG1
On Tue, Aug 03, 2021 at 02:15:13PM +0200, Daniel Vetter wrote: > On Tue, Aug 3, 2021 at 6:53 AM Matthew Brost wrote: > > > > Minimum set of patches to enable GuC submission on DG1 and enable it by > > default. > > > > A little difficult to test as IGTs do not work with DG1 due to a bunch > > of uAPI features being disabled (e.g. relocations, caching memory > > options, etc...). > > Matt Auld has an igt series which fixes a lot of this stuff, would be > good to do at least a Test-With run with that. > If I'm understanding his series correct it only fixes the mmap issues not relocs so I don't think it will help all that much. I probably could send a trybot series letting relocs work on DG1 + his series and that might work? I guess I'll find out. > Also I'm assuming that for ADL-P we'll get this equivalent patch set GuC submission is enabled by default on ADL-P. We don't have ADL-P in pre-merge CI currently though :(, just offline runs of about 1800 tests. Still a awaiting results for that though. For what it is worth I've run all 1800 locally, via a script, and gotten solid results back with my IGT updates. There are some failures but I believe I understand all of them as either test issues or known differences in behavior between execlists and GuC submission with are opens (e.g. persistence, semaphores, SUBMIT fence, etc...). Matt > soon, and there we should be able to get real results? > -Daniel > > > > > Tested with the loading the driver and 'live' selftests. Submissions > > seem to work. > > > > Signed-off-by: Matthew Brost > > > > Daniele Ceraolo Spurio (1): > > drm/i915/guc: put all guc objects in lmem when available > > > > Matthew Brost (2): > > drm/i915/guc: Add DG1 GuC / HuC firmware defs > > drm/i915/guc: Enable GuC submission by default on DG1 > > > > Venkata Sandeep Dhanalakota (1): > > drm/i915: Do not define vma on stack > > > > drivers/gpu/drm/i915/gem/i915_gem_lmem.c | 26 +++ > > drivers/gpu/drm/i915/gem/i915_gem_lmem.h | 4 + > > drivers/gpu/drm/i915/gt/uc/intel_guc.c| 9 ++- > > drivers/gpu/drm/i915/gt/uc/intel_guc_fw.c | 11 ++- > > drivers/gpu/drm/i915/gt/uc/intel_huc.c| 14 +++- > > drivers/gpu/drm/i915/gt/uc/intel_uc.c | 2 +- > > drivers/gpu/drm/i915/gt/uc/intel_uc_fw.c | 90 --- > > drivers/gpu/drm/i915/gt/uc/intel_uc_fw.h | 2 + > > 8 files changed, 138 insertions(+), 20 deletions(-) > > > > -- > > 2.28.0 > > > > > -- > Daniel Vetter > Software Engineer, Intel Corporation > http://blog.ffwll.ch
Re: [PATCH 0/4] Enable GuC submission by default on DG1
On Tue, Aug 3, 2021 at 6:53 AM Matthew Brost wrote: > > Minimum set of patches to enable GuC submission on DG1 and enable it by > default. > > A little difficult to test as IGTs do not work with DG1 due to a bunch > of uAPI features being disabled (e.g. relocations, caching memory > options, etc...). Matt Auld has an igt series which fixes a lot of this stuff, would be good to do at least a Test-With run with that. Also I'm assuming that for ADL-P we'll get this equivalent patch set soon, and there we should be able to get real results? -Daniel > > Tested with the loading the driver and 'live' selftests. Submissions > seem to work. > > Signed-off-by: Matthew Brost > > Daniele Ceraolo Spurio (1): > drm/i915/guc: put all guc objects in lmem when available > > Matthew Brost (2): > drm/i915/guc: Add DG1 GuC / HuC firmware defs > drm/i915/guc: Enable GuC submission by default on DG1 > > Venkata Sandeep Dhanalakota (1): > drm/i915: Do not define vma on stack > > drivers/gpu/drm/i915/gem/i915_gem_lmem.c | 26 +++ > drivers/gpu/drm/i915/gem/i915_gem_lmem.h | 4 + > drivers/gpu/drm/i915/gt/uc/intel_guc.c| 9 ++- > drivers/gpu/drm/i915/gt/uc/intel_guc_fw.c | 11 ++- > drivers/gpu/drm/i915/gt/uc/intel_huc.c| 14 +++- > drivers/gpu/drm/i915/gt/uc/intel_uc.c | 2 +- > drivers/gpu/drm/i915/gt/uc/intel_uc_fw.c | 90 --- > drivers/gpu/drm/i915/gt/uc/intel_uc_fw.h | 2 + > 8 files changed, 138 insertions(+), 20 deletions(-) > > -- > 2.28.0 > -- Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch
[PATCH 0/4] Enable GuC submission by default on DG1
Minimum set of patches to enable GuC submission on DG1 and enable it by default. A little difficult to test as IGTs do not work with DG1 due to a bunch of uAPI features being disabled (e.g. relocations, caching memory options, etc...). Tested with the loading the driver and 'live' selftests. Submissions seem to work. Signed-off-by: Matthew Brost Daniele Ceraolo Spurio (1): drm/i915/guc: put all guc objects in lmem when available Matthew Brost (2): drm/i915/guc: Add DG1 GuC / HuC firmware defs drm/i915/guc: Enable GuC submission by default on DG1 Venkata Sandeep Dhanalakota (1): drm/i915: Do not define vma on stack drivers/gpu/drm/i915/gem/i915_gem_lmem.c | 26 +++ drivers/gpu/drm/i915/gem/i915_gem_lmem.h | 4 + drivers/gpu/drm/i915/gt/uc/intel_guc.c| 9 ++- drivers/gpu/drm/i915/gt/uc/intel_guc_fw.c | 11 ++- drivers/gpu/drm/i915/gt/uc/intel_huc.c| 14 +++- drivers/gpu/drm/i915/gt/uc/intel_uc.c | 2 +- drivers/gpu/drm/i915/gt/uc/intel_uc_fw.c | 90 --- drivers/gpu/drm/i915/gt/uc/intel_uc_fw.h | 2 + 8 files changed, 138 insertions(+), 20 deletions(-) -- 2.28.0