[Intel-gfx] [RFC 0/8] Per client engine busyness
Tvrtko Ursulin
tvrtko.ursulin at linux.intel.com
Fri Feb 7 16:13:25 UTC 2020
From: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
Another re-spin of the per-client engine busyness series. Highlights from this
version:
* Refactor of the i915_drm_client concept based on feedback from Chris.
* Tracking contexts belonging to a client had a flaw where reaped contexts
would stop contributing to the busyness making the counters go backwards.
I was simply broken. In this version I had to track per-client stats at the
same call-sites where per-context is tracked. It's a little bit more runtime
cost but not too bad.
* GuC support is out - needs more time than I have at the moment to properly
wire it up with the latest changes. Plus there is a monotonicity issue with
either the value stored in pphwsp by the GPU or a bug in my patch which also
needs to be debugged.
Internally we track time spent on engines for each struct intel_context. This
can serve as a building block for several features from the want list:
smarter scheduler decisions, getrusage(2)-like per-GEM-context functionality
wanted by some customers, cgroups controller, dynamic SSEU tuning,...
Externally, in sysfs, we expose time spent on GPU per client and per engine
class.
Sysfs interface enables us to implement a "top-like" tool for GPU tasks. Or with
a "screenshot":
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
intel-gpu-top - 906/ 955 MHz; 0% RC6; 5.30 Watts; 933 irqs/s
IMC reads: 4414 MiB/s
IMC writes: 3805 MiB/s
ENGINE BUSY MI_SEMA MI_WAIT
Render/3D/0 93.46% |████████████████████████████████▋ | 0% 0%
Blitter/0 0.00% | | 0% 0%
Video/0 0.00% | | 0% 0%
VideoEnhance/0 0.00% | | 0% 0%
PID NAME Render/3D Blitter Video VideoEnhance
2733 neverball |██████▌ || || || |
2047 Xorg |███▊ || || || |
2737 glxgears |█▍ || || || |
2128 xfwm4 | || || || |
2047 Xorg | || || || |
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Implementation wise we add a a bunch of files in sysfs like:
# cd /sys/class/drm/card0/clients/
# tree
.
├── 7
│ ├── busy
│ │ ├── 0
│ │ ├── 1
│ │ ├── 2
│ │ └── 3
│ ├── name
│ └── pid
├── 8
│ ├── busy
│ │ ├── 0
│ │ ├── 1
│ │ ├── 2
│ │ └── 3
│ ├── name
│ └── pid
└── 9
├── busy
│ ├── 0
│ ├── 1
│ ├── 2
│ └── 3
├── name
└── pid
Files in 'busy' directories are numbered using the engine class ABI values and
they contain accumulated nanoseconds each client spent on engines of a
respective class.
It is stil a RFC since it misses dedicated test cases to ensure things really
work as advertised.
Tvrtko Ursulin (6):
drm/i915: Expose list of clients in sysfs
drm/i915: Update client name on context create
drm/i915: Make GEM contexts track DRM clients
drm/i915: Track per-context engine busyness
drm/i915: Track per drm client engine class busyness
drm/i915: Expose per-engine client busyness
drivers/gpu/drm/i915/Makefile | 3 +-
drivers/gpu/drm/i915/gem/i915_gem_context.c | 31 +-
.../gpu/drm/i915/gem/i915_gem_context_types.h | 13 +-
drivers/gpu/drm/i915/gt/intel_context.c | 20 +
drivers/gpu/drm/i915/gt/intel_context.h | 35 ++
drivers/gpu/drm/i915/gt/intel_context_types.h | 9 +
drivers/gpu/drm/i915/gt/intel_engine_cs.c | 16 +-
drivers/gpu/drm/i915/gt/intel_lrc.c | 65 +++-
drivers/gpu/drm/i915/i915_debugfs.c | 31 +-
drivers/gpu/drm/i915/i915_drm_client.c | 350 ++++++++++++++++++
drivers/gpu/drm/i915/i915_drm_client.h | 88 +++++
drivers/gpu/drm/i915/i915_drv.h | 5 +
drivers/gpu/drm/i915/i915_gem.c | 35 +-
drivers/gpu/drm/i915/i915_gpu_error.c | 21 +-
drivers/gpu/drm/i915/i915_sysfs.c | 8 +
15 files changed, 673 insertions(+), 57 deletions(-)
create mode 100644 drivers/gpu/drm/i915/i915_drm_client.c
create mode 100644 drivers/gpu/drm/i915/i915_drm_client.h
--
2.20.1
More information about the Intel-gfx
mailing list