| Message ID | 20210803201349.31031-1-umesh.nerlige.ramappa@intel.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | Enable triggered perf query for Xe_HP | expand |
+ Joonas On Tue, Aug 03, 2021 at 01:13:41PM -0700, Umesh Nerlige Ramappa wrote: >This is a revival of the patch series to support triggered perf query reports >from here - https://patchwork.freedesktop.org/series/83831/ > >The patches were not pushed earlier because corresponding UMD changes were >missing. This revival addresses UMD changes in GPUvis for this series. GPUvis >uses the perf library in igt-gpu-tools. Changes to the library are here - >https://patchwork.freedesktop.org/series/93355/ > >GPUvis changes will be posted as a PR once the above library and kernel changes >are pushed. > >Summary of the feature: > >Current platforms provide MI_REPORT_PERF_COUNT to query a snapshot of perf >counters from a batch. This mechanism does not have consistent support on all >engines for newer platforms. To support perf query, all new platforms use a >mechanism to trigger OA report snapshots into the OA buffer by writing to a HW >register from a batch. To lookup this report in the OA buffer quickly, the OA >buffer is mmapped into the user space. > >This series implements the new query mechanism. > >Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com> > >Chris Wilson (3): > drm/i915/gt: Refactor _wa_add to reuse wa_index and wa_list_grow > drm/i915/gt: Check for conflicting RING_NONPRIV > drm/i915/gt: Enable dynamic adjustment of RING_NONPRIV > >Piotr Maciejewski (1): > drm/i915/perf: Ensure observation logic is not clock gated > >Umesh Nerlige Ramappa (4): > drm/i915/gt: Lock intel_engine_apply_whitelist with uncore->lock > drm/i915/perf: Whitelist OA report trigger registers > drm/i915/perf: Whitelist OA counter and buffer registers > drm/i915/perf: Map OA buffer to user space for gen12 performance query > > drivers/gpu/drm/i915/gem/i915_gem_mman.c | 2 +- > drivers/gpu/drm/i915/gem/i915_gem_mman.h | 2 + > drivers/gpu/drm/i915/gt/intel_workarounds.c | 269 +++++++++++++----- > drivers/gpu/drm/i915/gt/intel_workarounds.h | 7 + > .../gpu/drm/i915/gt/selftest_workarounds.c | 237 +++++++++++++++ > drivers/gpu/drm/i915/i915_perf.c | 228 ++++++++++++++- > drivers/gpu/drm/i915/i915_perf_types.h | 8 + > drivers/gpu/drm/i915/i915_reg.h | 30 +- > include/uapi/drm/i915_drm.h | 33 +++ > 9 files changed, 728 insertions(+), 88 deletions(-) > >-- >2.20.1 >
On Tue, Aug 03, 2021 at 01:18:38PM -0700, Umesh Nerlige Ramappa wrote: >+ Joonas > >On Tue, Aug 03, 2021 at 01:13:41PM -0700, Umesh Nerlige Ramappa wrote: >>This is a revival of the patch series to support triggered perf query reports >>from here - https://patchwork.freedesktop.org/series/83831/ >> >>The patches were not pushed earlier because corresponding UMD changes were >>missing. This revival addresses UMD changes in GPUvis for this series. GPUvis >>uses the perf library in igt-gpu-tools. Changes to the library are here - >>https://patchwork.freedesktop.org/series/93355/ >> >>GPUvis changes will be posted as a PR once the above library and kernel changes >>are pushed. GPUvis changes: https://github.com/unerlige/gpuvis/commit/1c19c134a64564f7b8d7ca3b46449324040a40be >> >>Summary of the feature: >> >>Current platforms provide MI_REPORT_PERF_COUNT to query a snapshot of perf >>counters from a batch. This mechanism does not have consistent support on all >>engines for newer platforms. To support perf query, all new platforms use a >>mechanism to trigger OA report snapshots into the OA buffer by writing to a HW >>register from a batch. To lookup this report in the OA buffer quickly, the OA >>buffer is mmapped into the user space. >> >>This series implements the new query mechanism. >> >>Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com> >> >>Chris Wilson (3): >> drm/i915/gt: Refactor _wa_add to reuse wa_index and wa_list_grow >> drm/i915/gt: Check for conflicting RING_NONPRIV >> drm/i915/gt: Enable dynamic adjustment of RING_NONPRIV >> >>Piotr Maciejewski (1): >> drm/i915/perf: Ensure observation logic is not clock gated >> >>Umesh Nerlige Ramappa (4): >> drm/i915/gt: Lock intel_engine_apply_whitelist with uncore->lock >> drm/i915/perf: Whitelist OA report trigger registers >> drm/i915/perf: Whitelist OA counter and buffer registers >> drm/i915/perf: Map OA buffer to user space for gen12 performance query >> >>drivers/gpu/drm/i915/gem/i915_gem_mman.c | 2 +- >>drivers/gpu/drm/i915/gem/i915_gem_mman.h | 2 + >>drivers/gpu/drm/i915/gt/intel_workarounds.c | 269 +++++++++++++----- >>drivers/gpu/drm/i915/gt/intel_workarounds.h | 7 + >>.../gpu/drm/i915/gt/selftest_workarounds.c | 237 +++++++++++++++ >>drivers/gpu/drm/i915/i915_perf.c | 228 ++++++++++++++- >>drivers/gpu/drm/i915/i915_perf_types.h | 8 + >>drivers/gpu/drm/i915/i915_reg.h | 30 +- >>include/uapi/drm/i915_drm.h | 33 +++ >>9 files changed, 728 insertions(+), 88 deletions(-) >> >>-- >>2.20.1 >>
This is a revival of the patch series to support triggered perf query reports from here - https://patchwork.freedesktop.org/series/83831/ The patches were not pushed earlier because corresponding UMD changes were missing. This revival addresses UMD changes in GPUvis for this series. GPUvis uses the perf library in igt-gpu-tools. Changes to the library are here - https://patchwork.freedesktop.org/series/93355/ GPUvis changes will be posted as a PR once the above library and kernel changes are pushed. Summary of the feature: Current platforms provide MI_REPORT_PERF_COUNT to query a snapshot of perf counters from a batch. This mechanism does not have consistent support on all engines for newer platforms. To support perf query, all new platforms use a mechanism to trigger OA report snapshots into the OA buffer by writing to a HW register from a batch. To lookup this report in the OA buffer quickly, the OA buffer is mmapped into the user space. This series implements the new query mechanism. Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com> Chris Wilson (3): drm/i915/gt: Refactor _wa_add to reuse wa_index and wa_list_grow drm/i915/gt: Check for conflicting RING_NONPRIV drm/i915/gt: Enable dynamic adjustment of RING_NONPRIV Piotr Maciejewski (1): drm/i915/perf: Ensure observation logic is not clock gated Umesh Nerlige Ramappa (4): drm/i915/gt: Lock intel_engine_apply_whitelist with uncore->lock drm/i915/perf: Whitelist OA report trigger registers drm/i915/perf: Whitelist OA counter and buffer registers drm/i915/perf: Map OA buffer to user space for gen12 performance query drivers/gpu/drm/i915/gem/i915_gem_mman.c | 2 +- drivers/gpu/drm/i915/gem/i915_gem_mman.h | 2 + drivers/gpu/drm/i915/gt/intel_workarounds.c | 269 +++++++++++++----- drivers/gpu/drm/i915/gt/intel_workarounds.h | 7 + .../gpu/drm/i915/gt/selftest_workarounds.c | 237 +++++++++++++++ drivers/gpu/drm/i915/i915_perf.c | 228 ++++++++++++++- drivers/gpu/drm/i915/i915_perf_types.h | 8 + drivers/gpu/drm/i915/i915_reg.h | 30 +- include/uapi/drm/i915_drm.h | 33 +++ 9 files changed, 728 insertions(+), 88 deletions(-)