| Message ID | 20230920215624.3482244-1-vinay.belgaumkar@intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | drm/i915/gem: Allow users to disable waitboost | expand |
On 20/09/2023 22:56, Vinay Belgaumkar wrote: > Provide a bit to disable waitboost while waiting on a gem object. > Waitboost results in increased power consumption by requesting RP0 > while waiting for the request to complete. Add a bit in the gem_wait() > IOCTL where this can be disabled. > > This is related to the libva API change here - > Link: https://github.com/XinfengZhang/libva/commit/3d90d18c67609a73121bb71b20ee4776b54b61a7 This link does not appear to lead to userspace code using this uapi? > > Cc: Rodrigo Vivi <rodrigo.vivi@intel.com> > Signed-off-by: Vinay Belgaumkar <vinay.belgaumkar@intel.com> > --- > drivers/gpu/drm/i915/gem/i915_gem_wait.c | 9 ++++++--- > drivers/gpu/drm/i915/i915_request.c | 3 ++- > drivers/gpu/drm/i915/i915_request.h | 1 + > include/uapi/drm/i915_drm.h | 1 + > 4 files changed, 10 insertions(+), 4 deletions(-) > > diff --git a/drivers/gpu/drm/i915/gem/i915_gem_wait.c b/drivers/gpu/drm/i915/gem/i915_gem_wait.c > index d4b918fb11ce..955885ec859d 100644 > --- a/drivers/gpu/drm/i915/gem/i915_gem_wait.c > +++ b/drivers/gpu/drm/i915/gem/i915_gem_wait.c > @@ -72,7 +72,8 @@ i915_gem_object_wait_reservation(struct dma_resv *resv, > struct dma_fence *fence; > long ret = timeout ?: 1; > > - i915_gem_object_boost(resv, flags); > + if (!(flags & I915_WAITBOOST_DISABLE)) > + i915_gem_object_boost(resv, flags); > > dma_resv_iter_begin(&cursor, resv, > dma_resv_usage_rw(flags & I915_WAIT_ALL)); > @@ -236,7 +237,7 @@ i915_gem_wait_ioctl(struct drm_device *dev, void *data, struct drm_file *file) > ktime_t start; > long ret; > > - if (args->flags != 0) > + if (args->flags != 0 || args->flags != I915_GEM_WAITBOOST_DISABLE) > return -EINVAL; > > obj = i915_gem_object_lookup(file, args->bo_handle); > @@ -248,7 +249,9 @@ i915_gem_wait_ioctl(struct drm_device *dev, void *data, struct drm_file *file) > ret = i915_gem_object_wait(obj, > I915_WAIT_INTERRUPTIBLE | > I915_WAIT_PRIORITY | > - I915_WAIT_ALL, > + I915_WAIT_ALL | > + (args->flags & I915_GEM_WAITBOOST_DISABLE ? > + I915_WAITBOOST_DISABLE : 0), > to_wait_timeout(args->timeout_ns)); > > if (args->timeout_ns > 0) { > diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c > index f59081066a19..2957409b4b2a 100644 > --- a/drivers/gpu/drm/i915/i915_request.c > +++ b/drivers/gpu/drm/i915/i915_request.c > @@ -2044,7 +2044,8 @@ long i915_request_wait_timeout(struct i915_request *rq, > * but at a cost of spending more power processing the workload > * (bad for battery). > */ > - if (flags & I915_WAIT_PRIORITY && !i915_request_started(rq)) > + if (!(flags & I915_WAITBOOST_DISABLE) && (flags & I915_WAIT_PRIORITY) && > + !i915_request_started(rq)) > intel_rps_boost(rq); > > wait.tsk = current; > diff --git a/drivers/gpu/drm/i915/i915_request.h b/drivers/gpu/drm/i915/i915_request.h > index 0ac55b2e4223..3cc00e8254dc 100644 > --- a/drivers/gpu/drm/i915/i915_request.h > +++ b/drivers/gpu/drm/i915/i915_request.h > @@ -445,6 +445,7 @@ long i915_request_wait(struct i915_request *rq, > #define I915_WAIT_INTERRUPTIBLE BIT(0) > #define I915_WAIT_PRIORITY BIT(1) /* small priority bump for the request */ > #define I915_WAIT_ALL BIT(2) /* used by i915_gem_object_wait() */ > +#define I915_WAITBOOST_DISABLE BIT(3) /* used by i915_gem_object_wait() */ > > void i915_request_show(struct drm_printer *m, > const struct i915_request *rq, > diff --git a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h > index 7000e5910a1d..4adee70e39cf 100644 > --- a/include/uapi/drm/i915_drm.h > +++ b/include/uapi/drm/i915_drm.h > @@ -1928,6 +1928,7 @@ struct drm_i915_gem_wait { > /** Handle of BO we shall wait on */ > __u32 bo_handle; > __u32 flags; > +#define I915_GEM_WAITBOOST_DISABLE (1u<<0) Probably would be good to avoid mentioning waitboost in the uapi since so far it wasn't an explicit feature/contract. Something like I915_GEM_WAIT_BACKGROUND_PRIORITY? Low priority? I also wonder if there could be a possible angle to help Rob (+cc) upstream the syncobj/fence deadline code if our media driver might make use of that somehow. Like if either we could wire up the deadline into GEM_WAIT (in a backward compatible manner), or if media could use sync fd wait instead. Assuming they have an out fence already, which may not be true. Regards, Tvrtko > /** Number of nanoseconds to wait, Returns time remaining. */ > __s64 timeout_ns; > };
Hello,
kernel test robot noticed a -3.2% regression of phoronix-test-suite.paraview.WaveletContour.1024x768.mipolys___sec on:
commit: 54fef7ea35dadd66193b98805b0bc42ef2b279db ("[PATCH] drm/i915/gem: Allow users to disable waitboost")
url: https://github.com/intel-lab-lkp/linux/commits/Vinay-Belgaumkar/drm-i915-gem-Allow-users-to-disable-waitboost/20230921-060357
base: git://anongit.freedesktop.org/drm-intel for-linux-next
patch link: https://lore.kernel.org/all/20230920215624.3482244-1-vinay.belgaumkar@intel.com/
patch subject: [PATCH] drm/i915/gem: Allow users to disable waitboost
testcase: phoronix-test-suite
test machine: 12 threads 1 sockets Intel(R) Core(TM) i7-8700 CPU @ 3.20GHz (Coffee Lake) with 32G memory
parameters:
need_x: true
test: paraview-1.0.2
option_a: Wavelet Contour
option_b: 1024 x 768
cpufreq_governor: performance
In addition to that, the commit also has significant impact on the following tests:
+------------------+----------------------------------------------------------------------------------------------------------------+
| testcase: change | phoronix-test-suite: phoronix-test-suite.x11perf.PutImageXY500x500Square.operations___second 12.8% improvement |
| test machine | 12 threads 1 sockets Intel(R) Core(TM) i7-8700 CPU @ 3.20GHz (Coffee Lake) with 32G memory |
| test parameters | cpufreq_governor=performance |
| | need_x=true |
| | option_a=PutImage XY 500x500 Square |
| | test=x11perf-1.1.1 |
+------------------+----------------------------------------------------------------------------------------------------------------+
| testcase: change | phoronix-test-suite: phoronix-test-suite.openarena.2560x1440.milliseconds -12.2% regression |
| test machine | 12 threads 1 sockets Intel(R) Core(TM) i7-8700 CPU @ 3.20GHz (Coffee Lake) with 32G memory |
| test parameters | cpufreq_governor=performance |
| | need_x=true |
| | option_a=2560 x 1440 |
| | test=openarena-1.5.5 |
+------------------+----------------------------------------------------------------------------------------------------------------+
If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <oliver.sang@intel.com>
| Closes: https://lore.kernel.org/oe-lkp/202309261055.b74df987-oliver.sang@intel.com
Details are as below:
-------------------------------------------------------------------------------------------------->
The kernel config and materials to reproduce are available at:
https://download.01.org/0day-ci/archive/20230926/202309261055.b74df987-oliver.sang@intel.com
=========================================================================================
compiler/cpufreq_governor/kconfig/need_x/option_a/option_b/rootfs/tbox_group/test/testcase:
gcc-12/performance/x86_64-rhel-8.3/true/Wavelet Contour/1024 x 768/debian-x86_64-phoronix/lkp-cfl-d2/paraview-1.0.2/phoronix-test-suite
commit:
16a9359401 ("drm/i915: Implement transcoder LRR for TGL+")
54fef7ea35 ("drm/i915/gem: Allow users to disable waitboost")
16a9359401edcbc0 54fef7ea35dadd66193b98805b0
---------------- ---------------------------
%stddev %change %stddev
\ | \
0.05 ± 4% +0.0 0.06 ± 2% mpstat.cpu.all.soft%
66.17 ± 60% +145.6% 162.50 ± 33% turbostat.C10
28.61 -3.3% 27.68 phoronix-test-suite.paraview.WaveletContour.1024x768.frames___sec
298.15 -3.2% 288.49 phoronix-test-suite.paraview.WaveletContour.1024x768.mipolys___sec
535005 +8.6% 580810 phoronix-test-suite.time.minor_page_faults
6278 +8.3% 6797 phoronix-test-suite.time.voluntary_context_switches
801166 +5.6% 845675 proc-vmstat.numa_hit
799382 +5.1% 840353 proc-vmstat.numa_local
59648 +2.6% 61211 proc-vmstat.pgactivate
1539307 +2.7% 1580759 proc-vmstat.pgalloc_normal
734297 +6.8% 783862 proc-vmstat.pgfault
1343231 +3.1% 1385353 proc-vmstat.pgfree
39042 +3.7% 40480 ± 2% proc-vmstat.pgreuse
1.106e+08 +2.1% 1.129e+08 perf-stat.i.cache-references
4.255e+09 +1.9% 4.336e+09 perf-stat.i.cpu-cycles
147872 +2.4% 151392 perf-stat.i.dTLB-store-misses
230242 +2.2% 235419 perf-stat.i.iTLB-load-misses
569455 +2.4% 583234 perf-stat.i.iTLB-loads
0.35 +1.9% 0.36 perf-stat.i.metric.GHz
12547 +8.6% 13625 perf-stat.i.minor-faults
609443 ± 2% +4.0% 633739 ± 2% perf-stat.i.node-loads
1701083 ± 2% +5.0% 1786794 ± 2% perf-stat.i.node-stores
12566 +8.6% 13644 perf-stat.i.page-faults
1.085e+08 +2.1% 1.108e+08 perf-stat.ps.cache-references
4.179e+09 +1.9% 4.259e+09 perf-stat.ps.cpu-cycles
225995 +2.3% 231096 perf-stat.ps.iTLB-load-misses
558694 +2.4% 572272 perf-stat.ps.iTLB-loads
12320 +8.6% 13380 perf-stat.ps.minor-faults
598059 ± 2% +4.0% 621901 ± 2% perf-stat.ps.node-loads
1670891 ± 2% +5.0% 1754927 ± 2% perf-stat.ps.node-stores
12339 +8.6% 13399 perf-stat.ps.page-faults
14.86 ± 42% -11.7 3.12 ±152% perf-profile.calltrace.cycles-pp.cpu_startup_entry.start_secondary.secondary_startup_64_no_verify
14.86 ± 42% -11.7 3.12 ±152% perf-profile.calltrace.cycles-pp.do_idle.cpu_startup_entry.start_secondary.secondary_startup_64_no_verify
14.86 ± 42% -11.7 3.12 ±152% perf-profile.calltrace.cycles-pp.start_secondary.secondary_startup_64_no_verify
14.86 ± 42% -11.0 3.82 ±161% perf-profile.calltrace.cycles-pp.secondary_startup_64_no_verify
11.53 ± 67% -10.1 1.39 ±223% perf-profile.calltrace.cycles-pp.cpuidle_enter.cpuidle_idle_call.do_idle.cpu_startup_entry.start_secondary
11.53 ± 67% -9.4 2.08 ±223% perf-profile.calltrace.cycles-pp.cpuidle_enter_state.cpuidle_enter.cpuidle_idle_call.do_idle.cpu_startup_entry
11.53 ± 67% -9.1 2.43 ±143% perf-profile.calltrace.cycles-pp.cpuidle_idle_call.do_idle.cpu_startup_entry.start_secondary.secondary_startup_64_no_verify
6.02 ± 95% -5.3 0.70 ±223% perf-profile.calltrace.cycles-pp.intel_idle.cpuidle_enter_state.cpuidle_enter.cpuidle_idle_call.do_idle
5.51 ± 56% -4.8 0.70 ±223% perf-profile.calltrace.cycles-pp.intel_idle_ibrs.cpuidle_enter_state.cpuidle_enter.cpuidle_idle_call.do_idle
7.02 ±128% -2.6 4.44 ±147% perf-profile.calltrace.cycles-pp.do_group_exit.get_signal.arch_do_signal_or_restart.exit_to_user_mode_loop.exit_to_user_mode_prepare
7.02 ±128% -2.6 4.44 ±147% perf-profile.calltrace.cycles-pp.do_exit.do_group_exit.get_signal.arch_do_signal_or_restart.exit_to_user_mode_loop
7.02 ±128% -2.6 4.44 ±147% perf-profile.calltrace.cycles-pp.exit_to_user_mode_prepare.syscall_exit_to_user_mode.do_syscall_64.entry_SYSCALL_64_after_hwframe
7.02 ±128% -2.6 4.44 ±147% perf-profile.calltrace.cycles-pp.exit_to_user_mode_loop.exit_to_user_mode_prepare.syscall_exit_to_user_mode.do_syscall_64.entry_SYSCALL_64_after_hwframe
7.02 ±128% -2.6 4.44 ±147% perf-profile.calltrace.cycles-pp.arch_do_signal_or_restart.exit_to_user_mode_loop.exit_to_user_mode_prepare.syscall_exit_to_user_mode.do_syscall_64
7.02 ±128% -2.6 4.44 ±147% perf-profile.calltrace.cycles-pp.get_signal.arch_do_signal_or_restart.exit_to_user_mode_loop.exit_to_user_mode_prepare.syscall_exit_to_user_mode
14.86 ± 42% -11.7 3.12 ±152% perf-profile.children.cycles-pp.start_secondary
14.86 ± 42% -11.0 3.82 ±161% perf-profile.children.cycles-pp.secondary_startup_64_no_verify
14.86 ± 42% -11.0 3.82 ±161% perf-profile.children.cycles-pp.cpu_startup_entry
14.86 ± 42% -11.0 3.82 ±161% perf-profile.children.cycles-pp.do_idle
11.53 ± 67% -9.4 2.08 ±223% perf-profile.children.cycles-pp.cpuidle_enter
11.53 ± 67% -9.4 2.08 ±223% perf-profile.children.cycles-pp.cpuidle_enter_state
11.53 ± 67% -8.4 3.12 ±152% perf-profile.children.cycles-pp.cpuidle_idle_call
6.02 ± 95% -5.3 0.70 ±223% perf-profile.children.cycles-pp.intel_idle
5.51 ± 56% -4.8 0.70 ±223% perf-profile.children.cycles-pp.intel_idle_ibrs
7.02 ±128% -2.6 4.44 ±147% perf-profile.children.cycles-pp.exit_to_user_mode_prepare
7.02 ±128% -2.6 4.44 ±147% perf-profile.children.cycles-pp.exit_to_user_mode_loop
7.02 ±128% -2.6 4.44 ±147% perf-profile.children.cycles-pp.arch_do_signal_or_restart
7.02 ±128% -2.6 4.44 ±147% perf-profile.children.cycles-pp.get_signal
6.02 ± 95% -5.3 0.70 ±223% perf-profile.self.cycles-pp.intel_idle
5.51 ± 56% -4.8 0.70 ±223% perf-profile.self.cycles-pp.intel_idle_ibrs
***************************************************************************************************
lkp-cfl-d2: 12 threads 1 sockets Intel(R) Core(TM) i7-8700 CPU @ 3.20GHz (Coffee Lake) with 32G memory
=========================================================================================
compiler/cpufreq_governor/kconfig/need_x/option_a/rootfs/tbox_group/test/testcase:
gcc-12/performance/x86_64-rhel-8.3/true/PutImage XY 500x500 Square/debian-x86_64-phoronix/lkp-cfl-d2/x11perf-1.1.1/phoronix-test-suite
commit:
16a9359401 ("drm/i915: Implement transcoder LRR for TGL+")
54fef7ea35 ("drm/i915/gem: Allow users to disable waitboost")
16a9359401edcbc0 54fef7ea35dadd66193b98805b0
---------------- ---------------------------
%stddev %change %stddev
\ | \
11950 ± 3% +93.9% 23173 ± 3% meminfo.Unevictable
1.02 ± 3% -0.6 0.43 ± 3% mpstat.cpu.all.iowait%
21273 -5.3% 20141 vmstat.system.in
4887 ± 32% -67.7% 1579 ± 32% phoronix-test-suite.time.involuntary_context_switches
147212 +26.1% 185677 phoronix-test-suite.time.minor_page_faults
76.83 +10.6% 85.00 phoronix-test-suite.time.percent_of_cpu_this_job_got
96.48 ± 4% +15.0% 110.93 ± 2% phoronix-test-suite.time.user_time
106.83 +12.8% 120.50 phoronix-test-suite.x11perf.PutImageXY500x500Square.operations___second
3.73 ± 3% -0.6 3.12 ± 2% turbostat.C1E%
58316 ± 5% -20.6% 46326 ± 3% turbostat.C3
0.86 ± 2% -0.1 0.74 ± 3% turbostat.C3%
1.24 ± 8% -18.9% 1.01 ± 10% turbostat.CPU%c3
38.67 ± 2% +5.0 43.70 ± 2% turbostat.CPUGFX%
23.44 +4.0% 24.38 turbostat.CorWatt
2.30 ± 4% -71.6% 0.65 ± 2% turbostat.GFXWatt
26.19 -2.7% 25.49 turbostat.PkgWatt
1.41 +2.4% 1.44 turbostat.RAMWatt
61.83 ± 23% -20.2 41.62 ± 31% perf-profile.calltrace.cycles-pp.intel_idle_ibrs.cpuidle_enter_state.cpuidle_enter.cpuidle_idle_call.do_idle
61.85 ± 23% -20.2 41.64 ± 31% perf-profile.children.cycles-pp.intel_idle_ibrs
1.24 ± 59% -1.0 0.23 ± 27% perf-profile.children.cycles-pp.worker_thread
0.84 ± 60% -0.7 0.18 ± 48% perf-profile.children.cycles-pp.asm_common_interrupt
0.83 ± 61% -0.7 0.18 ± 48% perf-profile.children.cycles-pp.common_interrupt
0.35 ± 74% -0.3 0.06 ± 85% perf-profile.children.cycles-pp.__common_interrupt
0.35 ± 75% -0.3 0.06 ± 85% perf-profile.children.cycles-pp.handle_edge_irq
0.30 ± 78% -0.3 0.05 ± 81% perf-profile.children.cycles-pp.handle_irq_event
0.37 ± 18% -0.2 0.18 ± 46% perf-profile.children.cycles-pp.newidle_balance
0.12 ± 32% -0.1 0.04 ± 79% perf-profile.children.cycles-pp.exit_mm
0.12 ± 30% -0.1 0.05 ± 73% perf-profile.children.cycles-pp.native_apic_msr_eoi
0.03 ±144% +0.1 0.10 ± 38% perf-profile.children.cycles-pp.dma_resv_iter_first_unlocked
0.04 ±154% +0.1 0.19 ± 40% perf-profile.children.cycles-pp.i915_gem_busy_ioctl
61.82 ± 23% -20.2 41.63 ± 31% perf-profile.self.cycles-pp.intel_idle_ibrs
0.12 ± 26% -0.1 0.05 ± 74% perf-profile.self.cycles-pp.native_apic_msr_eoi
31861 +5.9% 33748 proc-vmstat.nr_active_file
29007 +7.6% 31200 proc-vmstat.nr_mapped
183598 +1.6% 186520 proc-vmstat.nr_shmem
2987 ± 3% +93.9% 5793 ± 3% proc-vmstat.nr_unevictable
31861 +5.9% 33748 proc-vmstat.nr_zone_active_file
2987 ± 3% +93.9% 5793 ± 3% proc-vmstat.nr_zone_unevictable
721353 ± 2% +13.3% 816958 proc-vmstat.numa_hit
721406 ± 2% +13.2% 816970 proc-vmstat.numa_local
21301 ± 4% +13.1% 24099 proc-vmstat.pgactivate
2459642 ± 3% +15.5% 2840123 ± 2% proc-vmstat.pgalloc_normal
530507 ± 3% +8.0% 573060 proc-vmstat.pgfault
2336660 ± 3% +16.3% 2717873 ± 2% proc-vmstat.pgfree
9973 ± 2% +143.6% 24292 ± 3% proc-vmstat.unevictable_pgs_culled
9875 +146.0% 24292 ± 3% proc-vmstat.unevictable_pgs_rescued
9876 +146.1% 24304 ± 3% proc-vmstat.unevictable_pgs_scanned
1.413e+09 +12.4% 1.588e+09 perf-stat.i.branch-instructions
2.09 -0.1 1.99 ± 2% perf-stat.i.branch-miss-rate%
23305261 +1.8% 23715959 perf-stat.i.branch-misses
4.57 ± 4% +1.5 6.06 ± 2% perf-stat.i.cache-miss-rate%
4971151 ± 5% +71.3% 8515308 ± 2% perf-stat.i.cache-misses
1.771e+08 +6.4% 1.884e+08 perf-stat.i.cache-references
0.91 -3.5% 0.88 ± 2% perf-stat.i.cpi
4.936e+09 +7.8% 5.319e+09 perf-stat.i.cpu-cycles
39.95 ± 3% -15.1% 33.90 ± 2% perf-stat.i.cpu-migrations
1671 ± 4% -34.5% 1095 ± 6% perf-stat.i.cycles-between-cache-misses
2.815e+09 +11.8% 3.147e+09 perf-stat.i.dTLB-loads
0.02 ± 2% -0.0 0.02 ± 4% perf-stat.i.dTLB-store-miss-rate%
6.478e+08 +10.4% 7.153e+08 perf-stat.i.dTLB-stores
40.15 ± 2% -2.9 37.29 perf-stat.i.iTLB-load-miss-rate%
1.063e+10 +12.2% 1.192e+10 perf-stat.i.instructions
40912 ± 4% +25.5% 51361 ± 2% perf-stat.i.instructions-per-iTLB-miss
2.18 +2.7% 2.23 perf-stat.i.ipc
0.41 +7.8% 0.44 perf-stat.i.metric.GHz
421.03 +11.6% 469.88 perf-stat.i.metric.M/sec
3358 +5.5% 3542 perf-stat.i.minor-faults
0.00 ± 11% -0.0 0.00 ± 6% perf-stat.i.node-load-miss-rate%
6.40 ± 8% +17.6% 7.52 ± 5% perf-stat.i.node-load-misses
216221 ± 6% +61.9% 350134 perf-stat.i.node-loads
0.00 ± 16% +0.1 0.12 ±220% perf-stat.i.node-store-miss-rate%
6.72 ± 6% +6.9e+12% 4.615e+11 ±223% perf-stat.i.node-store-misses
694301 ± 9% -14.3% 594881 ± 3% perf-stat.i.node-stores
3361 +5.5% 3545 perf-stat.i.page-faults
0.47 ± 5% +52.6% 0.71 ± 2% perf-stat.overall.MPKI
1.65 -0.2 1.49 perf-stat.overall.branch-miss-rate%
2.81 ± 6% +1.7 4.52 ± 2% perf-stat.overall.cache-miss-rate%
0.46 -4.0% 0.45 perf-stat.overall.cpi
994.89 ± 5% -37.2% 624.75 ± 2% perf-stat.overall.cycles-between-cache-misses
0.03 -0.0 0.02 perf-stat.overall.dTLB-load-miss-rate%
0.01 -0.0 0.01 perf-stat.overall.dTLB-store-miss-rate%
30.99 ± 3% -2.3 28.68 ± 4% perf-stat.overall.iTLB-load-miss-rate%
25510 ± 11% +20.2% 30672 ± 5% perf-stat.overall.instructions-per-iTLB-miss
2.15 +4.1% 2.24 perf-stat.overall.ipc
0.00 ± 8% -0.0 0.00 ± 5% perf-stat.overall.node-load-miss-rate%
0.00 ± 8% +16.7 16.67 ±223% perf-stat.overall.node-store-miss-rate%
1.403e+09 +12.5% 1.578e+09 perf-stat.ps.branch-instructions
23153579 +1.7% 23558574 perf-stat.ps.branch-misses
4938806 ± 5% +71.3% 8457880 ± 2% perf-stat.ps.cache-misses
1.758e+08 +6.4% 1.871e+08 perf-stat.ps.cache-references
4.901e+09 +7.8% 5.282e+09 perf-stat.ps.cpu-cycles
39.66 ± 3% -15.1% 33.67 ± 2% perf-stat.ps.cpu-migrations
2.795e+09 +11.8% 3.125e+09 perf-stat.ps.dTLB-loads
6.432e+08 +10.4% 7.103e+08 perf-stat.ps.dTLB-stores
1.055e+10 +12.2% 1.184e+10 perf-stat.ps.instructions
3336 +5.5% 3518 perf-stat.ps.minor-faults
6.35 ± 8% +17.6% 7.47 ± 5% perf-stat.ps.node-load-misses
214756 ± 6% +61.9% 347761 perf-stat.ps.node-loads
6.67 ± 6% +6.9e+12% 4.577e+11 ±223% perf-stat.ps.node-store-misses
689455 ± 9% -14.3% 590802 ± 3% perf-stat.ps.node-stores
3338 +5.5% 3521 perf-stat.ps.page-faults
1.467e+12 ± 4% +15.7% 1.698e+12 ± 2% perf-stat.total.instructions
***************************************************************************************************
lkp-cfl-d2: 12 threads 1 sockets Intel(R) Core(TM) i7-8700 CPU @ 3.20GHz (Coffee Lake) with 32G memory
=========================================================================================
compiler/cpufreq_governor/kconfig/need_x/option_a/rootfs/tbox_group/test/testcase:
gcc-12/performance/x86_64-rhel-8.3/true/2560 x 1440/debian-x86_64-phoronix/lkp-cfl-d2/openarena-1.5.5/phoronix-test-suite
commit:
16a9359401 ("drm/i915: Implement transcoder LRR for TGL+")
54fef7ea35 ("drm/i915/gem: Allow users to disable waitboost")
16a9359401edcbc0 54fef7ea35dadd66193b98805b0
---------------- ---------------------------
%stddev %change %stddev
\ | \
1459 ± 27% +34.9% 1967 ± 13% sched_debug.cpu.curr->pid.max
1.63 -0.5 1.14 ± 2% mpstat.cpu.all.iowait%
1.17 -0.1 1.06 mpstat.cpu.all.irq%
189.00 ± 11% -53.6% 87.67 ± 17% perf-c2c.DRAM.local
20.67 ± 28% -87.9% 2.50 ± 72% perf-c2c.HITM.local
6016 +5.7% 6358 vmstat.io.bi
30168 -5.0% 28666 vmstat.system.cs
27678 -2.7% 26933 vmstat.system.in
337.58 +13.2% 382.30 phoronix-test-suite.openarena.2560x1440.frames_per_second
2.95 -12.2% 2.59 phoronix-test-suite.openarena.2560x1440.milliseconds
191369 +23.8% 236938 phoronix-test-suite.time.minor_page_faults
51.00 +6.5% 54.33 phoronix-test-suite.time.percent_of_cpu_this_job_got
56192 -63.1% 20758 phoronix-test-suite.time.voluntary_context_switches
90640 -64.0% 32673 ± 7% turbostat.C1
0.68 ± 2% -0.3 0.38 ± 5% turbostat.C1%
46385 -12.3% 40688 ± 2% turbostat.C3
1.72 -0.2 1.51 ± 2% turbostat.C3%
3.58 ± 6% -17.6% 2.95 ± 11% turbostat.CPU%c3
5.71 -11.1% 5.08 turbostat.GFXWatt
29983 ± 4% -63.0% 11105 ± 24% turbostat.POLL
0.06 ± 8% -0.0 0.01 ± 35% turbostat.POLL%
1.31 ± 15% +70.2% 2.23 ± 7% turbostat.Pkg%pc2
29.56 -1.4% 29.14 turbostat.PkgWatt
2123 -5.7% 2001 proc-vmstat.nr_active_anon
37869 +2.8% 38912 proc-vmstat.nr_active_file
22828 -5.8% 21501 proc-vmstat.nr_unevictable
2123 -5.7% 2001 proc-vmstat.nr_zone_active_anon
37869 +2.8% 38912 proc-vmstat.nr_zone_active_file
22828 -5.8% 21501 proc-vmstat.nr_zone_unevictable
624578 +4.2% 650937 proc-vmstat.numa_hit
624578 +4.2% 650937 proc-vmstat.numa_local
771624 +3.5% 798755 proc-vmstat.pgalloc_normal
429393 +7.2% 460476 proc-vmstat.pgfault
620059 +4.1% 645242 proc-vmstat.pgfree
8.52 ±126% -7.9 0.62 ±223% perf-profile.calltrace.cycles-pp.wp_page_copy.__handle_mm_fault.handle_mm_fault.do_user_addr_fault.exc_page_fault
6.09 ±114% -5.4 0.67 ±223% perf-profile.calltrace.cycles-pp.intel_idle.cpuidle_enter_state.cpuidle_enter.cpuidle_idle_call.do_idle
3.76 ±110% -3.8 0.00 perf-profile.calltrace.cycles-pp.error_entry
8.52 ±126% -7.9 0.62 ±223% perf-profile.children.cycles-pp.wp_page_copy
6.09 ±114% -5.4 0.67 ±223% perf-profile.children.cycles-pp.intel_idle
5.79 ±100% -4.6 1.24 ±223% perf-profile.children.cycles-pp.do_filp_open
5.79 ±100% -4.6 1.24 ±223% perf-profile.children.cycles-pp.path_openat
3.76 ±110% -3.8 0.00 perf-profile.children.cycles-pp.error_entry
3.76 ±110% -2.6 1.11 ±223% perf-profile.children.cycles-pp.asm_sysvec_apic_timer_interrupt
3.76 ±110% -2.6 1.11 ±223% perf-profile.children.cycles-pp.sysvec_apic_timer_interrupt
3.76 ±110% -2.6 1.11 ±223% perf-profile.children.cycles-pp.__sysvec_apic_timer_interrupt
3.76 ±110% -2.6 1.11 ±223% perf-profile.children.cycles-pp.hrtimer_interrupt
6.09 ±114% -5.4 0.67 ±223% perf-profile.self.cycles-pp.intel_idle
3.76 ±110% -3.8 0.00 perf-profile.self.cycles-pp.error_entry
4.97 ±109% -2.2 2.78 ±223% perf-profile.self.cycles-pp.zap_pte_range
8.07e+08 +2.7% 8.29e+08 perf-stat.i.branch-instructions
31612011 +4.7% 33095236 perf-stat.i.cache-misses
31337 -4.0% 30085 perf-stat.i.context-switches
1.57 +7.7% 1.69 ± 2% perf-stat.i.cpi
4.798e+09 +3.2% 4.95e+09 perf-stat.i.cpu-cycles
0.14 ± 3% +0.0 0.16 ± 3% perf-stat.i.dTLB-load-miss-rate%
1.131e+09 +4.1% 1.177e+09 perf-stat.i.dTLB-loads
0.03 ± 2% +0.0 0.03 ± 2% perf-stat.i.dTLB-store-miss-rate%
134590 +6.1% 142835 perf-stat.i.dTLB-store-misses
5.907e+08 +4.5% 6.175e+08 perf-stat.i.dTLB-stores
3858714 +6.5% 4109072 perf-stat.i.iTLB-loads
4.646e+09 +3.6% 4.812e+09 perf-stat.i.instructions
0.40 +3.2% 0.41 perf-stat.i.metric.GHz
824.17 +6.4% 876.81 perf-stat.i.metric.K/sec
224.04 +3.7% 232.30 perf-stat.i.metric.M/sec
5577 +13.3% 6319 perf-stat.i.minor-faults
1661010 +8.8% 1806501 perf-stat.i.node-loads
4199874 +5.4% 4426929 perf-stat.i.node-stores
5582 +13.3% 6325 perf-stat.i.page-faults
151.84 -1.5% 149.61 perf-stat.overall.cycles-between-cache-misses
7.953e+08 +2.6% 8.16e+08 perf-stat.ps.branch-instructions
31131128 +4.6% 32563072 perf-stat.ps.cache-misses
30863 -4.1% 29604 perf-stat.ps.context-switches
4.727e+09 +3.1% 4.871e+09 perf-stat.ps.cpu-cycles
1.114e+09 +4.0% 1.158e+09 perf-stat.ps.dTLB-loads
132539 +6.0% 140536 perf-stat.ps.dTLB-store-misses
5.818e+08 +4.4% 6.076e+08 perf-stat.ps.dTLB-stores
3798756 +6.4% 4042216 perf-stat.ps.iTLB-loads
4.577e+09 +3.5% 4.736e+09 perf-stat.ps.instructions
5496 +13.2% 6220 perf-stat.ps.minor-faults
1635550 +8.7% 1777301 perf-stat.ps.node-loads
4135301 +5.3% 4355124 perf-stat.ps.node-stores
5501 +13.2% 6226 perf-stat.ps.page-faults
3.045e+11 -2.8% 2.961e+11 perf-stat.total.instructions
Disclaimer:
Results have been estimated based on internal Intel analysis and are provided
for informational purposes only. Any difference in system hardware or software
design or configuration may affect actual performance.
On 9/21/2023 3:41 AM, Tvrtko Ursulin wrote: > > On 20/09/2023 22:56, Vinay Belgaumkar wrote: >> Provide a bit to disable waitboost while waiting on a gem object. >> Waitboost results in increased power consumption by requesting RP0 >> while waiting for the request to complete. Add a bit in the gem_wait() >> IOCTL where this can be disabled. >> >> This is related to the libva API change here - >> Link: >> https://github.com/XinfengZhang/libva/commit/3d90d18c67609a73121bb71b20ee4776b54b61a7 > > This link does not appear to lead to userspace code using this uapi? We have asked Carl (cc'd) to post a patch for the same. > >> >> Cc: Rodrigo Vivi <rodrigo.vivi@intel.com> >> Signed-off-by: Vinay Belgaumkar <vinay.belgaumkar@intel.com> >> --- >> drivers/gpu/drm/i915/gem/i915_gem_wait.c | 9 ++++++--- >> drivers/gpu/drm/i915/i915_request.c | 3 ++- >> drivers/gpu/drm/i915/i915_request.h | 1 + >> include/uapi/drm/i915_drm.h | 1 + >> 4 files changed, 10 insertions(+), 4 deletions(-) >> >> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_wait.c >> b/drivers/gpu/drm/i915/gem/i915_gem_wait.c >> index d4b918fb11ce..955885ec859d 100644 >> --- a/drivers/gpu/drm/i915/gem/i915_gem_wait.c >> +++ b/drivers/gpu/drm/i915/gem/i915_gem_wait.c >> @@ -72,7 +72,8 @@ i915_gem_object_wait_reservation(struct dma_resv >> *resv, >> struct dma_fence *fence; >> long ret = timeout ?: 1; >> - i915_gem_object_boost(resv, flags); >> + if (!(flags & I915_WAITBOOST_DISABLE)) >> + i915_gem_object_boost(resv, flags); >> dma_resv_iter_begin(&cursor, resv, >> dma_resv_usage_rw(flags & I915_WAIT_ALL)); >> @@ -236,7 +237,7 @@ i915_gem_wait_ioctl(struct drm_device *dev, void >> *data, struct drm_file *file) >> ktime_t start; >> long ret; >> - if (args->flags != 0) >> + if (args->flags != 0 || args->flags != I915_GEM_WAITBOOST_DISABLE) >> return -EINVAL; >> obj = i915_gem_object_lookup(file, args->bo_handle); >> @@ -248,7 +249,9 @@ i915_gem_wait_ioctl(struct drm_device *dev, void >> *data, struct drm_file *file) >> ret = i915_gem_object_wait(obj, >> I915_WAIT_INTERRUPTIBLE | >> I915_WAIT_PRIORITY | >> - I915_WAIT_ALL, >> + I915_WAIT_ALL | >> + (args->flags & I915_GEM_WAITBOOST_DISABLE ? >> + I915_WAITBOOST_DISABLE : 0), >> to_wait_timeout(args->timeout_ns)); >> if (args->timeout_ns > 0) { >> diff --git a/drivers/gpu/drm/i915/i915_request.c >> b/drivers/gpu/drm/i915/i915_request.c >> index f59081066a19..2957409b4b2a 100644 >> --- a/drivers/gpu/drm/i915/i915_request.c >> +++ b/drivers/gpu/drm/i915/i915_request.c >> @@ -2044,7 +2044,8 @@ long i915_request_wait_timeout(struct >> i915_request *rq, >> * but at a cost of spending more power processing the workload >> * (bad for battery). >> */ >> - if (flags & I915_WAIT_PRIORITY && !i915_request_started(rq)) >> + if (!(flags & I915_WAITBOOST_DISABLE) && (flags & >> I915_WAIT_PRIORITY) && >> + !i915_request_started(rq)) >> intel_rps_boost(rq); >> wait.tsk = current; >> diff --git a/drivers/gpu/drm/i915/i915_request.h >> b/drivers/gpu/drm/i915/i915_request.h >> index 0ac55b2e4223..3cc00e8254dc 100644 >> --- a/drivers/gpu/drm/i915/i915_request.h >> +++ b/drivers/gpu/drm/i915/i915_request.h >> @@ -445,6 +445,7 @@ long i915_request_wait(struct i915_request *rq, >> #define I915_WAIT_INTERRUPTIBLE BIT(0) >> #define I915_WAIT_PRIORITY BIT(1) /* small priority bump for the >> request */ >> #define I915_WAIT_ALL BIT(2) /* used by >> i915_gem_object_wait() */ >> +#define I915_WAITBOOST_DISABLE BIT(3) /* used by >> i915_gem_object_wait() */ >> void i915_request_show(struct drm_printer *m, >> const struct i915_request *rq, >> diff --git a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h >> index 7000e5910a1d..4adee70e39cf 100644 >> --- a/include/uapi/drm/i915_drm.h >> +++ b/include/uapi/drm/i915_drm.h >> @@ -1928,6 +1928,7 @@ struct drm_i915_gem_wait { >> /** Handle of BO we shall wait on */ >> __u32 bo_handle; >> __u32 flags; >> +#define I915_GEM_WAITBOOST_DISABLE (1u<<0) > > Probably would be good to avoid mentioning waitboost in the uapi since > so far it wasn't an explicit feature/contract. Something like > I915_GEM_WAIT_BACKGROUND_PRIORITY? Low priority? sure. > > I also wonder if there could be a possible angle to help Rob (+cc) > upstream the syncobj/fence deadline code if our media driver might > make use of that somehow. > > Like if either we could wire up the deadline into GEM_WAIT (in a > backward compatible manner), or if media could use sync fd wait > instead. Assuming they have an out fence already, which may not be true. Makes sense. We could add a SET_DEADLINE flag or something similar and pass in the deadline when appropriate. Thanks, Vinay. > > Regards, > > Tvrtko > >> /** Number of nanoseconds to wait, Returns time remaining. */ >> __s64 timeout_ns; >> };
On 27/09/2023 20:34, Belgaumkar, Vinay wrote: > > On 9/21/2023 3:41 AM, Tvrtko Ursulin wrote: >> >> On 20/09/2023 22:56, Vinay Belgaumkar wrote: >>> Provide a bit to disable waitboost while waiting on a gem object. >>> Waitboost results in increased power consumption by requesting RP0 >>> while waiting for the request to complete. Add a bit in the gem_wait() >>> IOCTL where this can be disabled. >>> >>> This is related to the libva API change here - >>> Link: >>> https://github.com/XinfengZhang/libva/commit/3d90d18c67609a73121bb71b20ee4776b54b61a7 >> >> This link does not appear to lead to userspace code using this uapi? > We have asked Carl (cc'd) to post a patch for the same. Ack. >>> Cc: Rodrigo Vivi <rodrigo.vivi@intel.com> >>> Signed-off-by: Vinay Belgaumkar <vinay.belgaumkar@intel.com> >>> --- >>> drivers/gpu/drm/i915/gem/i915_gem_wait.c | 9 ++++++--- >>> drivers/gpu/drm/i915/i915_request.c | 3 ++- >>> drivers/gpu/drm/i915/i915_request.h | 1 + >>> include/uapi/drm/i915_drm.h | 1 + >>> 4 files changed, 10 insertions(+), 4 deletions(-) >>> >>> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_wait.c >>> b/drivers/gpu/drm/i915/gem/i915_gem_wait.c >>> index d4b918fb11ce..955885ec859d 100644 >>> --- a/drivers/gpu/drm/i915/gem/i915_gem_wait.c >>> +++ b/drivers/gpu/drm/i915/gem/i915_gem_wait.c >>> @@ -72,7 +72,8 @@ i915_gem_object_wait_reservation(struct dma_resv >>> *resv, >>> struct dma_fence *fence; >>> long ret = timeout ?: 1; >>> - i915_gem_object_boost(resv, flags); >>> + if (!(flags & I915_WAITBOOST_DISABLE)) >>> + i915_gem_object_boost(resv, flags); >>> dma_resv_iter_begin(&cursor, resv, >>> dma_resv_usage_rw(flags & I915_WAIT_ALL)); >>> @@ -236,7 +237,7 @@ i915_gem_wait_ioctl(struct drm_device *dev, void >>> *data, struct drm_file *file) >>> ktime_t start; >>> long ret; >>> - if (args->flags != 0) >>> + if (args->flags != 0 || args->flags != I915_GEM_WAITBOOST_DISABLE) >>> return -EINVAL; >>> obj = i915_gem_object_lookup(file, args->bo_handle); >>> @@ -248,7 +249,9 @@ i915_gem_wait_ioctl(struct drm_device *dev, void >>> *data, struct drm_file *file) >>> ret = i915_gem_object_wait(obj, >>> I915_WAIT_INTERRUPTIBLE | >>> I915_WAIT_PRIORITY | >>> - I915_WAIT_ALL, >>> + I915_WAIT_ALL | >>> + (args->flags & I915_GEM_WAITBOOST_DISABLE ? >>> + I915_WAITBOOST_DISABLE : 0), >>> to_wait_timeout(args->timeout_ns)); >>> if (args->timeout_ns > 0) { >>> diff --git a/drivers/gpu/drm/i915/i915_request.c >>> b/drivers/gpu/drm/i915/i915_request.c >>> index f59081066a19..2957409b4b2a 100644 >>> --- a/drivers/gpu/drm/i915/i915_request.c >>> +++ b/drivers/gpu/drm/i915/i915_request.c >>> @@ -2044,7 +2044,8 @@ long i915_request_wait_timeout(struct >>> i915_request *rq, >>> * but at a cost of spending more power processing the workload >>> * (bad for battery). >>> */ >>> - if (flags & I915_WAIT_PRIORITY && !i915_request_started(rq)) >>> + if (!(flags & I915_WAITBOOST_DISABLE) && (flags & >>> I915_WAIT_PRIORITY) && >>> + !i915_request_started(rq)) >>> intel_rps_boost(rq); >>> wait.tsk = current; >>> diff --git a/drivers/gpu/drm/i915/i915_request.h >>> b/drivers/gpu/drm/i915/i915_request.h >>> index 0ac55b2e4223..3cc00e8254dc 100644 >>> --- a/drivers/gpu/drm/i915/i915_request.h >>> +++ b/drivers/gpu/drm/i915/i915_request.h >>> @@ -445,6 +445,7 @@ long i915_request_wait(struct i915_request *rq, >>> #define I915_WAIT_INTERRUPTIBLE BIT(0) >>> #define I915_WAIT_PRIORITY BIT(1) /* small priority bump for the >>> request */ >>> #define I915_WAIT_ALL BIT(2) /* used by >>> i915_gem_object_wait() */ >>> +#define I915_WAITBOOST_DISABLE BIT(3) /* used by >>> i915_gem_object_wait() */ >>> void i915_request_show(struct drm_printer *m, >>> const struct i915_request *rq, >>> diff --git a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h >>> index 7000e5910a1d..4adee70e39cf 100644 >>> --- a/include/uapi/drm/i915_drm.h >>> +++ b/include/uapi/drm/i915_drm.h >>> @@ -1928,6 +1928,7 @@ struct drm_i915_gem_wait { >>> /** Handle of BO we shall wait on */ >>> __u32 bo_handle; >>> __u32 flags; >>> +#define I915_GEM_WAITBOOST_DISABLE (1u<<0) >> >> Probably would be good to avoid mentioning waitboost in the uapi since >> so far it wasn't an explicit feature/contract. Something like >> I915_GEM_WAIT_BACKGROUND_PRIORITY? Low priority? > sure. >> >> I also wonder if there could be a possible angle to help Rob (+cc) >> upstream the syncobj/fence deadline code if our media driver might >> make use of that somehow. >> >> Like if either we could wire up the deadline into GEM_WAIT (in a >> backward compatible manner), or if media could use sync fd wait >> instead. Assuming they have an out fence already, which may not be true. > > Makes sense. We could add a SET_DEADLINE flag or something similar and > pass in the deadline when appropriate. Rob - do you have time and motivation to think about this angle at all currently? If not I guess we just proceed with the new flag for our GEM_WAIT. Regards, Tvrtko > > Thanks, > > Vinay. > >> >> Regards, >> >> Tvrtko >> >>> /** Number of nanoseconds to wait, Returns time remaining. */ >>> __s64 timeout_ns; >>> };
On Thu, Sep 28, 2023 at 01:48:34PM +0100, Tvrtko Ursulin wrote: > > On 27/09/2023 20:34, Belgaumkar, Vinay wrote: > > > > On 9/21/2023 3:41 AM, Tvrtko Ursulin wrote: > > > > > > On 20/09/2023 22:56, Vinay Belgaumkar wrote: > > > > Provide a bit to disable waitboost while waiting on a gem object. > > > > Waitboost results in increased power consumption by requesting RP0 > > > > while waiting for the request to complete. Add a bit in the gem_wait() > > > > IOCTL where this can be disabled. > > > > > > > > This is related to the libva API change here - > > > > Link: https://github.com/XinfengZhang/libva/commit/3d90d18c67609a73121bb71b20ee4776b54b61a7 > > > > > > This link does not appear to lead to userspace code using this uapi? > > We have asked Carl (cc'd) to post a patch for the same. > > Ack. I'm glad to see that we will have the end-to-end flow of the high-level API. > > > > > Cc: Rodrigo Vivi <rodrigo.vivi@intel.com> > > > > Signed-off-by: Vinay Belgaumkar <vinay.belgaumkar@intel.com> > > > > --- > > > > drivers/gpu/drm/i915/gem/i915_gem_wait.c | 9 ++++++--- > > > > drivers/gpu/drm/i915/i915_request.c | 3 ++- > > > > drivers/gpu/drm/i915/i915_request.h | 1 + > > > > include/uapi/drm/i915_drm.h | 1 + > > > > 4 files changed, 10 insertions(+), 4 deletions(-) > > > > > > > > diff --git a/drivers/gpu/drm/i915/gem/i915_gem_wait.c > > > > b/drivers/gpu/drm/i915/gem/i915_gem_wait.c > > > > index d4b918fb11ce..955885ec859d 100644 > > > > --- a/drivers/gpu/drm/i915/gem/i915_gem_wait.c > > > > +++ b/drivers/gpu/drm/i915/gem/i915_gem_wait.c > > > > @@ -72,7 +72,8 @@ i915_gem_object_wait_reservation(struct > > > > dma_resv *resv, > > > > struct dma_fence *fence; > > > > long ret = timeout ?: 1; > > > > - i915_gem_object_boost(resv, flags); > > > > + if (!(flags & I915_WAITBOOST_DISABLE)) > > > > + i915_gem_object_boost(resv, flags); > > > > dma_resv_iter_begin(&cursor, resv, > > > > dma_resv_usage_rw(flags & I915_WAIT_ALL)); > > > > @@ -236,7 +237,7 @@ i915_gem_wait_ioctl(struct drm_device *dev, > > > > void *data, struct drm_file *file) > > > > ktime_t start; > > > > long ret; > > > > - if (args->flags != 0) > > > > + if (args->flags != 0 || args->flags != I915_GEM_WAITBOOST_DISABLE) > > > > return -EINVAL; > > > > obj = i915_gem_object_lookup(file, args->bo_handle); > > > > @@ -248,7 +249,9 @@ i915_gem_wait_ioctl(struct drm_device *dev, > > > > void *data, struct drm_file *file) > > > > ret = i915_gem_object_wait(obj, > > > > I915_WAIT_INTERRUPTIBLE | > > > > I915_WAIT_PRIORITY | > > > > - I915_WAIT_ALL, > > > > + I915_WAIT_ALL | > > > > + (args->flags & I915_GEM_WAITBOOST_DISABLE ? > > > > + I915_WAITBOOST_DISABLE : 0), > > > > to_wait_timeout(args->timeout_ns)); > > > > if (args->timeout_ns > 0) { > > > > diff --git a/drivers/gpu/drm/i915/i915_request.c > > > > b/drivers/gpu/drm/i915/i915_request.c > > > > index f59081066a19..2957409b4b2a 100644 > > > > --- a/drivers/gpu/drm/i915/i915_request.c > > > > +++ b/drivers/gpu/drm/i915/i915_request.c > > > > @@ -2044,7 +2044,8 @@ long i915_request_wait_timeout(struct > > > > i915_request *rq, > > > > * but at a cost of spending more power processing the workload > > > > * (bad for battery). > > > > */ > > > > - if (flags & I915_WAIT_PRIORITY && !i915_request_started(rq)) > > > > + if (!(flags & I915_WAITBOOST_DISABLE) && (flags & > > > > I915_WAIT_PRIORITY) && > > > > + !i915_request_started(rq)) > > > > intel_rps_boost(rq); > > > > wait.tsk = current; > > > > diff --git a/drivers/gpu/drm/i915/i915_request.h > > > > b/drivers/gpu/drm/i915/i915_request.h > > > > index 0ac55b2e4223..3cc00e8254dc 100644 > > > > --- a/drivers/gpu/drm/i915/i915_request.h > > > > +++ b/drivers/gpu/drm/i915/i915_request.h > > > > @@ -445,6 +445,7 @@ long i915_request_wait(struct i915_request *rq, > > > > #define I915_WAIT_INTERRUPTIBLE BIT(0) > > > > #define I915_WAIT_PRIORITY BIT(1) /* small priority bump > > > > for the request */ > > > > #define I915_WAIT_ALL BIT(2) /* used by > > > > i915_gem_object_wait() */ > > > > +#define I915_WAITBOOST_DISABLE BIT(3) /* used by maybe name it I915_WAIT_NO_BOOST to align a bit better with the existent ones? > > > > i915_gem_object_wait() */ > > > > void i915_request_show(struct drm_printer *m, > > > > const struct i915_request *rq, > > > > diff --git a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h > > > > index 7000e5910a1d..4adee70e39cf 100644 > > > > --- a/include/uapi/drm/i915_drm.h > > > > +++ b/include/uapi/drm/i915_drm.h > > > > @@ -1928,6 +1928,7 @@ struct drm_i915_gem_wait { > > > > /** Handle of BO we shall wait on */ > > > > __u32 bo_handle; > > > > __u32 flags; > > > > +#define I915_GEM_WAITBOOST_DISABLE (1u<<0) > > > > > > Probably would be good to avoid mentioning waitboost in the uapi > > > since so far it wasn't an explicit feature/contract. Something like > > > I915_GEM_WAIT_BACKGROUND_PRIORITY? Low priority? > > sure. > > > > > > I also wonder if there could be a possible angle to help Rob (+cc) > > > upstream the syncobj/fence deadline code if our media driver might > > > make use of that somehow. > > > > > > Like if either we could wire up the deadline into GEM_WAIT (in a > > > backward compatible manner), or if media could use sync fd wait > > > instead. Assuming they have an out fence already, which may not be > > > true. > > > > Makes sense. We could add a SET_DEADLINE flag or something similar and > > pass in the deadline when appropriate. > > Rob - do you have time and motivation to think about this angle at all > currently? If not I guess we just proceed with the new flag for our > GEM_WAIT. Well, this could be the first user for that uapi that Rob was proposing indeed. The downside is probably because we should implement the deadline in i915 and consider all the deadline as 0 (urgent) and boost, unless in this case where before the gem_wait the UMD would use the set_deadline to something higher (max?). Well, if we have a clarity on how to proceed with the deadline we should probably go there. But for simplicity I would be in favor of this proposed gem_wait flag as is, because this already solves many real important cases. > > Regards, > > Tvrtko > > > > > Thanks, > > > > Vinay. > > > > > > > > Regards, > > > > > > Tvrtko > > > > > > > /** Number of nanoseconds to wait, Returns time remaining. */ > > > > __s64 timeout_ns; > > > > };
On 13/10/2023 21:51, Rodrigo Vivi wrote: > On Thu, Sep 28, 2023 at 01:48:34PM +0100, Tvrtko Ursulin wrote: >> >> On 27/09/2023 20:34, Belgaumkar, Vinay wrote: >>> >>> On 9/21/2023 3:41 AM, Tvrtko Ursulin wrote: >>>> >>>> On 20/09/2023 22:56, Vinay Belgaumkar wrote: >>>>> Provide a bit to disable waitboost while waiting on a gem object. >>>>> Waitboost results in increased power consumption by requesting RP0 >>>>> while waiting for the request to complete. Add a bit in the gem_wait() >>>>> IOCTL where this can be disabled. >>>>> >>>>> This is related to the libva API change here - >>>>> Link: https://github.com/XinfengZhang/libva/commit/3d90d18c67609a73121bb71b20ee4776b54b61a7 >>>> >>>> This link does not appear to lead to userspace code using this uapi? >>> We have asked Carl (cc'd) to post a patch for the same. >> >> Ack. > > I'm glad to see that we will have the end-to-end flow of the high-level API. > >> >>>>> Cc: Rodrigo Vivi <rodrigo.vivi@intel.com> >>>>> Signed-off-by: Vinay Belgaumkar <vinay.belgaumkar@intel.com> >>>>> --- >>>>> drivers/gpu/drm/i915/gem/i915_gem_wait.c | 9 ++++++--- >>>>> drivers/gpu/drm/i915/i915_request.c | 3 ++- >>>>> drivers/gpu/drm/i915/i915_request.h | 1 + >>>>> include/uapi/drm/i915_drm.h | 1 + >>>>> 4 files changed, 10 insertions(+), 4 deletions(-) >>>>> >>>>> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_wait.c >>>>> b/drivers/gpu/drm/i915/gem/i915_gem_wait.c >>>>> index d4b918fb11ce..955885ec859d 100644 >>>>> --- a/drivers/gpu/drm/i915/gem/i915_gem_wait.c >>>>> +++ b/drivers/gpu/drm/i915/gem/i915_gem_wait.c >>>>> @@ -72,7 +72,8 @@ i915_gem_object_wait_reservation(struct >>>>> dma_resv *resv, >>>>> struct dma_fence *fence; >>>>> long ret = timeout ?: 1; >>>>> - i915_gem_object_boost(resv, flags); >>>>> + if (!(flags & I915_WAITBOOST_DISABLE)) >>>>> + i915_gem_object_boost(resv, flags); >>>>> dma_resv_iter_begin(&cursor, resv, >>>>> dma_resv_usage_rw(flags & I915_WAIT_ALL)); >>>>> @@ -236,7 +237,7 @@ i915_gem_wait_ioctl(struct drm_device *dev, >>>>> void *data, struct drm_file *file) >>>>> ktime_t start; >>>>> long ret; >>>>> - if (args->flags != 0) >>>>> + if (args->flags != 0 || args->flags != I915_GEM_WAITBOOST_DISABLE) >>>>> return -EINVAL; >>>>> obj = i915_gem_object_lookup(file, args->bo_handle); >>>>> @@ -248,7 +249,9 @@ i915_gem_wait_ioctl(struct drm_device *dev, >>>>> void *data, struct drm_file *file) >>>>> ret = i915_gem_object_wait(obj, >>>>> I915_WAIT_INTERRUPTIBLE | >>>>> I915_WAIT_PRIORITY | >>>>> - I915_WAIT_ALL, >>>>> + I915_WAIT_ALL | >>>>> + (args->flags & I915_GEM_WAITBOOST_DISABLE ? >>>>> + I915_WAITBOOST_DISABLE : 0), >>>>> to_wait_timeout(args->timeout_ns)); >>>>> if (args->timeout_ns > 0) { >>>>> diff --git a/drivers/gpu/drm/i915/i915_request.c >>>>> b/drivers/gpu/drm/i915/i915_request.c >>>>> index f59081066a19..2957409b4b2a 100644 >>>>> --- a/drivers/gpu/drm/i915/i915_request.c >>>>> +++ b/drivers/gpu/drm/i915/i915_request.c >>>>> @@ -2044,7 +2044,8 @@ long i915_request_wait_timeout(struct >>>>> i915_request *rq, >>>>> * but at a cost of spending more power processing the workload >>>>> * (bad for battery). >>>>> */ >>>>> - if (flags & I915_WAIT_PRIORITY && !i915_request_started(rq)) >>>>> + if (!(flags & I915_WAITBOOST_DISABLE) && (flags & >>>>> I915_WAIT_PRIORITY) && >>>>> + !i915_request_started(rq)) >>>>> intel_rps_boost(rq); >>>>> wait.tsk = current; >>>>> diff --git a/drivers/gpu/drm/i915/i915_request.h >>>>> b/drivers/gpu/drm/i915/i915_request.h >>>>> index 0ac55b2e4223..3cc00e8254dc 100644 >>>>> --- a/drivers/gpu/drm/i915/i915_request.h >>>>> +++ b/drivers/gpu/drm/i915/i915_request.h >>>>> @@ -445,6 +445,7 @@ long i915_request_wait(struct i915_request *rq, >>>>> #define I915_WAIT_INTERRUPTIBLE BIT(0) >>>>> #define I915_WAIT_PRIORITY BIT(1) /* small priority bump >>>>> for the request */ >>>>> #define I915_WAIT_ALL BIT(2) /* used by >>>>> i915_gem_object_wait() */ >>>>> +#define I915_WAITBOOST_DISABLE BIT(3) /* used by > > maybe name it I915_WAIT_NO_BOOST to align a bit better with the existent ones? I thought it would be better to not mention wait boost in the uapi, but leave it as implementation detail. My suggestion was along the lines of I915_GEM_WAIT_BACKGROUND/IDLE. In essence saying allowing userspace to say this is not an important wait. Yes, it implies that other waits are (more) important, but I think this is still better than starting to mention wait boost in the uapi. Since that would kind of cement it exists, while we always just viewed it as an "go faster" driver internal heuristics and could freely decide not to employ it even for default waits. Historically even we had a period when waits were getting elevated scheduling priority. We removed it, would have to dig through git and email history to remember exactly why, but probably along the lines that it is not always justified. Same as waitboost is not always justified and can be harmful. So I think a generic name for the uapi leaves more freedom for the driver. Might be a wrong name that I am suggesting and should be something else, not sure. [Comes back later.] eec39e441c29 ("drm/i915: Remove wait priority boosting") So it seems we only removed it because corner cases with the current scheduler were hard. Unfortunately improved deadline based scheduler never got in despite being ready so we can not restore this now. >>>>> i915_gem_object_wait() */ >>>>> void i915_request_show(struct drm_printer *m, >>>>> const struct i915_request *rq, >>>>> diff --git a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h >>>>> index 7000e5910a1d..4adee70e39cf 100644 >>>>> --- a/include/uapi/drm/i915_drm.h >>>>> +++ b/include/uapi/drm/i915_drm.h >>>>> @@ -1928,6 +1928,7 @@ struct drm_i915_gem_wait { >>>>> /** Handle of BO we shall wait on */ >>>>> __u32 bo_handle; >>>>> __u32 flags; >>>>> +#define I915_GEM_WAITBOOST_DISABLE (1u<<0) >>>> >>>> Probably would be good to avoid mentioning waitboost in the uapi >>>> since so far it wasn't an explicit feature/contract. Something like >>>> I915_GEM_WAIT_BACKGROUND_PRIORITY? Low priority? >>> sure. >>>> >>>> I also wonder if there could be a possible angle to help Rob (+cc) >>>> upstream the syncobj/fence deadline code if our media driver might >>>> make use of that somehow. >>>> >>>> Like if either we could wire up the deadline into GEM_WAIT (in a >>>> backward compatible manner), or if media could use sync fd wait >>>> instead. Assuming they have an out fence already, which may not be >>>> true. >>> >>> Makes sense. We could add a SET_DEADLINE flag or something similar and >>> pass in the deadline when appropriate. >> >> Rob - do you have time and motivation to think about this angle at all >> currently? If not I guess we just proceed with the new flag for our >> GEM_WAIT. > > Well, this could be the first user for that uapi that Rob was proposing > indeed. > > The downside is probably because we should implement the deadline in i915 > and consider all the deadline as 0 (urgent) and boost, unless in this > case where before the gem_wait the UMD would use the set_deadline to > something higher (max?). > > Well, if we have a clarity on how to proceed with the deadline we should > probably go there. But for simplicity I would be in favor of this proposed > gem_wait flag as is, because this already solves many real important cases. Yes I don't think we had consensus on the semantics of when no deadline is set, so it does sound better to proceed with i915 specific solution for now. The two wouldn't be incompatible if deadlines were later added. Regards, Tvrtko > >> >> Regards, >> >> Tvrtko >> >>> >>> Thanks, >>> >>> Vinay. >>> >>>> >>>> Regards, >>>> >>>> Tvrtko >>>> >>>>> /** Number of nanoseconds to wait, Returns time remaining. */ >>>>> __s64 timeout_ns; >>>>> };
On Mon, Oct 16, 2023 at 09:02:38AM +0100, Tvrtko Ursulin wrote: > > On 13/10/2023 21:51, Rodrigo Vivi wrote: > > On Thu, Sep 28, 2023 at 01:48:34PM +0100, Tvrtko Ursulin wrote: > > > > > > On 27/09/2023 20:34, Belgaumkar, Vinay wrote: > > > > > > > > On 9/21/2023 3:41 AM, Tvrtko Ursulin wrote: > > > > > > > > > > On 20/09/2023 22:56, Vinay Belgaumkar wrote: > > > > > > Provide a bit to disable waitboost while waiting on a gem object. > > > > > > Waitboost results in increased power consumption by requesting RP0 > > > > > > while waiting for the request to complete. Add a bit in the gem_wait() > > > > > > IOCTL where this can be disabled. > > > > > > > > > > > > This is related to the libva API change here - > > > > > > Link: https://github.com/XinfengZhang/libva/commit/3d90d18c67609a73121bb71b20ee4776b54b61a7 > > > > > > > > > > This link does not appear to lead to userspace code using this uapi? > > > > We have asked Carl (cc'd) to post a patch for the same. > > > > > > Ack. > > > > I'm glad to see that we will have the end-to-end flow of the high-level API. > > > > > > > > > > > Cc: Rodrigo Vivi <rodrigo.vivi@intel.com> > > > > > > Signed-off-by: Vinay Belgaumkar <vinay.belgaumkar@intel.com> > > > > > > --- > > > > > > drivers/gpu/drm/i915/gem/i915_gem_wait.c | 9 ++++++--- > > > > > > drivers/gpu/drm/i915/i915_request.c | 3 ++- > > > > > > drivers/gpu/drm/i915/i915_request.h | 1 + > > > > > > include/uapi/drm/i915_drm.h | 1 + > > > > > > 4 files changed, 10 insertions(+), 4 deletions(-) > > > > > > > > > > > > diff --git a/drivers/gpu/drm/i915/gem/i915_gem_wait.c > > > > > > b/drivers/gpu/drm/i915/gem/i915_gem_wait.c > > > > > > index d4b918fb11ce..955885ec859d 100644 > > > > > > --- a/drivers/gpu/drm/i915/gem/i915_gem_wait.c > > > > > > +++ b/drivers/gpu/drm/i915/gem/i915_gem_wait.c > > > > > > @@ -72,7 +72,8 @@ i915_gem_object_wait_reservation(struct > > > > > > dma_resv *resv, > > > > > > struct dma_fence *fence; > > > > > > long ret = timeout ?: 1; > > > > > > - i915_gem_object_boost(resv, flags); > > > > > > + if (!(flags & I915_WAITBOOST_DISABLE)) > > > > > > + i915_gem_object_boost(resv, flags); > > > > > > dma_resv_iter_begin(&cursor, resv, > > > > > > dma_resv_usage_rw(flags & I915_WAIT_ALL)); > > > > > > @@ -236,7 +237,7 @@ i915_gem_wait_ioctl(struct drm_device *dev, > > > > > > void *data, struct drm_file *file) > > > > > > ktime_t start; > > > > > > long ret; > > > > > > - if (args->flags != 0) > > > > > > + if (args->flags != 0 || args->flags != I915_GEM_WAITBOOST_DISABLE) > > > > > > return -EINVAL; > > > > > > obj = i915_gem_object_lookup(file, args->bo_handle); > > > > > > @@ -248,7 +249,9 @@ i915_gem_wait_ioctl(struct drm_device *dev, > > > > > > void *data, struct drm_file *file) > > > > > > ret = i915_gem_object_wait(obj, > > > > > > I915_WAIT_INTERRUPTIBLE | > > > > > > I915_WAIT_PRIORITY | > > > > > > - I915_WAIT_ALL, > > > > > > + I915_WAIT_ALL | > > > > > > + (args->flags & I915_GEM_WAITBOOST_DISABLE ? > > > > > > + I915_WAITBOOST_DISABLE : 0), > > > > > > to_wait_timeout(args->timeout_ns)); > > > > > > if (args->timeout_ns > 0) { > > > > > > diff --git a/drivers/gpu/drm/i915/i915_request.c > > > > > > b/drivers/gpu/drm/i915/i915_request.c > > > > > > index f59081066a19..2957409b4b2a 100644 > > > > > > --- a/drivers/gpu/drm/i915/i915_request.c > > > > > > +++ b/drivers/gpu/drm/i915/i915_request.c > > > > > > @@ -2044,7 +2044,8 @@ long i915_request_wait_timeout(struct > > > > > > i915_request *rq, > > > > > > * but at a cost of spending more power processing the workload > > > > > > * (bad for battery). > > > > > > */ > > > > > > - if (flags & I915_WAIT_PRIORITY && !i915_request_started(rq)) > > > > > > + if (!(flags & I915_WAITBOOST_DISABLE) && (flags & > > > > > > I915_WAIT_PRIORITY) && > > > > > > + !i915_request_started(rq)) > > > > > > intel_rps_boost(rq); > > > > > > wait.tsk = current; > > > > > > diff --git a/drivers/gpu/drm/i915/i915_request.h > > > > > > b/drivers/gpu/drm/i915/i915_request.h > > > > > > index 0ac55b2e4223..3cc00e8254dc 100644 > > > > > > --- a/drivers/gpu/drm/i915/i915_request.h > > > > > > +++ b/drivers/gpu/drm/i915/i915_request.h > > > > > > @@ -445,6 +445,7 @@ long i915_request_wait(struct i915_request *rq, > > > > > > #define I915_WAIT_INTERRUPTIBLE BIT(0) > > > > > > #define I915_WAIT_PRIORITY BIT(1) /* small priority bump > > > > > > for the request */ > > > > > > #define I915_WAIT_ALL BIT(2) /* used by > > > > > > i915_gem_object_wait() */ > > > > > > +#define I915_WAITBOOST_DISABLE BIT(3) /* used by > > > > maybe name it I915_WAIT_NO_BOOST to align a bit better with the existent ones? > > I thought it would be better to not mention wait boost in the uapi, but > leave it as implementation detail. > > My suggestion was along the lines of I915_GEM_WAIT_BACKGROUND/IDLE. good idea! > > In essence saying allowing userspace to say this is not an important wait. > Yes, it implies that other waits are (more) important, but I think this is > still better than starting to mention wait boost in the uapi. Since that > would kind of cement it exists, while we always just viewed it as an "go > faster" driver internal heuristics and could freely decide not to employ it > even for default waits. > > Historically even we had a period when waits were getting elevated > scheduling priority. We removed it, would have to dig through git and email > history to remember exactly why, but probably along the lines that it is not > always justified. Same as waitboost is not always justified and can be > harmful. > > So I think a generic name for the uapi leaves more freedom for the driver. > Might be a wrong name that I am suggesting and should be something else, not > sure. > > [Comes back later.] > > eec39e441c29 ("drm/i915: Remove wait priority boosting") > > So it seems we only removed it because corner cases with the current > scheduler were hard. Unfortunately improved deadline based scheduler never > got in despite being ready so we can not restore this now. > > > > > > > i915_gem_object_wait() */ > > > > > > void i915_request_show(struct drm_printer *m, > > > > > > const struct i915_request *rq, > > > > > > diff --git a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h > > > > > > index 7000e5910a1d..4adee70e39cf 100644 > > > > > > --- a/include/uapi/drm/i915_drm.h > > > > > > +++ b/include/uapi/drm/i915_drm.h > > > > > > @@ -1928,6 +1928,7 @@ struct drm_i915_gem_wait { > > > > > > /** Handle of BO we shall wait on */ > > > > > > __u32 bo_handle; > > > > > > __u32 flags; > > > > > > +#define I915_GEM_WAITBOOST_DISABLE (1u<<0) > > > > > > > > > > Probably would be good to avoid mentioning waitboost in the uapi > > > > > since so far it wasn't an explicit feature/contract. Something like > > > > > I915_GEM_WAIT_BACKGROUND_PRIORITY? Low priority? > > > > sure. > > > > > > > > > > I also wonder if there could be a possible angle to help Rob (+cc) > > > > > upstream the syncobj/fence deadline code if our media driver might > > > > > make use of that somehow. > > > > > > > > > > Like if either we could wire up the deadline into GEM_WAIT (in a > > > > > backward compatible manner), or if media could use sync fd wait > > > > > instead. Assuming they have an out fence already, which may not be > > > > > true. > > > > > > > > Makes sense. We could add a SET_DEADLINE flag or something similar and > > > > pass in the deadline when appropriate. > > > > > > Rob - do you have time and motivation to think about this angle at all > > > currently? If not I guess we just proceed with the new flag for our > > > GEM_WAIT. > > > > Well, this could be the first user for that uapi that Rob was proposing > > indeed. > > > > The downside is probably because we should implement the deadline in i915 > > and consider all the deadline as 0 (urgent) and boost, unless in this > > case where before the gem_wait the UMD would use the set_deadline to > > something higher (max?). > > > > Well, if we have a clarity on how to proceed with the deadline we should > > probably go there. But for simplicity I would be in favor of this proposed > > gem_wait flag as is, because this already solves many real important cases. > > Yes I don't think we had consensus on the semantics of when no deadline is > set, so it does sound better to proceed with i915 specific solution for now. > The two wouldn't be incompatible if deadlines were later added. > > Regards, > > Tvrtko > > > > > > > > > Regards, > > > > > > Tvrtko > > > > > > > > > > > Thanks, > > > > > > > > Vinay. > > > > > > > > > > > > > > Regards, > > > > > > > > > > Tvrtko > > > > > > > > > > > /** Number of nanoseconds to wait, Returns time remaining. */ > > > > > > __s64 timeout_ns; > > > > > > };
Hello,
kernel test robot noticed "assertion_failure" on:
commit: 54fef7ea35dadd66193b98805b0bc42ef2b279db ("[PATCH] drm/i915/gem: Allow users to disable waitboost")
url: https://github.com/intel-lab-lkp/linux/commits/Vinay-Belgaumkar/drm-i915-gem-Allow-users-to-disable-waitboost/20230921-060357
base: git://anongit.freedesktop.org/drm-intel for-linux-next
patch link: https://lore.kernel.org/all/20230920215624.3482244-1-vinay.belgaumkar@intel.com/
patch subject: [PATCH] drm/i915/gem: Allow users to disable waitboost
in testcase: igt
version: igt-x86_64-0f075441-1_20230520
with following parameters:
group: group-23
compiler: gcc-12
test machine: 20 threads 1 sockets (Commet Lake) with 16G memory
(please refer to attached dmesg/kmsg for entire log/backtrace)
If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <oliver.sang@intel.com>
| Closes: https://lore.kernel.org/oe-lkp/202310271326.abb748d0-oliver.sang@intel.com
user :notice: [ 42.847071] Starting subtest: engines-cleanup
user :notice: [ 42.854776] Starting dynamic subtest: rcs0
user :notice: [ 42.863938] (gem_ctx_persistence:833) CRITICAL: Test assertion failure function test_nonpersistent_cleanup, file ../tests/i915/gem_ctx_persistence.c:283:
user :notice: [ 42.882029] (gem_ctx_persistence:833) CRITICAL: Failed assertion: gem_wait(i915, spin->handle, &timeout) == 0
user :notice: [ 42.895541] (gem_ctx_persistence:833) CRITICAL: error: -22 != 0
The kernel config and materials to reproduce are available at:
https://download.01.org/0day-ci/archive/20231027/202310271326.abb748d0-oliver.sang@intel.com
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_wait.c b/drivers/gpu/drm/i915/gem/i915_gem_wait.c index d4b918fb11ce..955885ec859d 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_wait.c +++ b/drivers/gpu/drm/i915/gem/i915_gem_wait.c @@ -72,7 +72,8 @@ i915_gem_object_wait_reservation(struct dma_resv *resv, struct dma_fence *fence; long ret = timeout ?: 1; - i915_gem_object_boost(resv, flags); + if (!(flags & I915_WAITBOOST_DISABLE)) + i915_gem_object_boost(resv, flags); dma_resv_iter_begin(&cursor, resv, dma_resv_usage_rw(flags & I915_WAIT_ALL)); @@ -236,7 +237,7 @@ i915_gem_wait_ioctl(struct drm_device *dev, void *data, struct drm_file *file) ktime_t start; long ret; - if (args->flags != 0) + if (args->flags != 0 || args->flags != I915_GEM_WAITBOOST_DISABLE) return -EINVAL; obj = i915_gem_object_lookup(file, args->bo_handle); @@ -248,7 +249,9 @@ i915_gem_wait_ioctl(struct drm_device *dev, void *data, struct drm_file *file) ret = i915_gem_object_wait(obj, I915_WAIT_INTERRUPTIBLE | I915_WAIT_PRIORITY | - I915_WAIT_ALL, + I915_WAIT_ALL | + (args->flags & I915_GEM_WAITBOOST_DISABLE ? + I915_WAITBOOST_DISABLE : 0), to_wait_timeout(args->timeout_ns)); if (args->timeout_ns > 0) { diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c index f59081066a19..2957409b4b2a 100644 --- a/drivers/gpu/drm/i915/i915_request.c +++ b/drivers/gpu/drm/i915/i915_request.c @@ -2044,7 +2044,8 @@ long i915_request_wait_timeout(struct i915_request *rq, * but at a cost of spending more power processing the workload * (bad for battery). */ - if (flags & I915_WAIT_PRIORITY && !i915_request_started(rq)) + if (!(flags & I915_WAITBOOST_DISABLE) && (flags & I915_WAIT_PRIORITY) && + !i915_request_started(rq)) intel_rps_boost(rq); wait.tsk = current; diff --git a/drivers/gpu/drm/i915/i915_request.h b/drivers/gpu/drm/i915/i915_request.h index 0ac55b2e4223..3cc00e8254dc 100644 --- a/drivers/gpu/drm/i915/i915_request.h +++ b/drivers/gpu/drm/i915/i915_request.h @@ -445,6 +445,7 @@ long i915_request_wait(struct i915_request *rq, #define I915_WAIT_INTERRUPTIBLE BIT(0) #define I915_WAIT_PRIORITY BIT(1) /* small priority bump for the request */ #define I915_WAIT_ALL BIT(2) /* used by i915_gem_object_wait() */ +#define I915_WAITBOOST_DISABLE BIT(3) /* used by i915_gem_object_wait() */ void i915_request_show(struct drm_printer *m, const struct i915_request *rq, diff --git a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h index 7000e5910a1d..4adee70e39cf 100644 --- a/include/uapi/drm/i915_drm.h +++ b/include/uapi/drm/i915_drm.h @@ -1928,6 +1928,7 @@ struct drm_i915_gem_wait { /** Handle of BO we shall wait on */ __u32 bo_handle; __u32 flags; +#define I915_GEM_WAITBOOST_DISABLE (1u<<0) /** Number of nanoseconds to wait, Returns time remaining. */ __s64 timeout_ns; };
Provide a bit to disable waitboost while waiting on a gem object. Waitboost results in increased power consumption by requesting RP0 while waiting for the request to complete. Add a bit in the gem_wait() IOCTL where this can be disabled. This is related to the libva API change here - Link: https://github.com/XinfengZhang/libva/commit/3d90d18c67609a73121bb71b20ee4776b54b61a7 Cc: Rodrigo Vivi <rodrigo.vivi@intel.com> Signed-off-by: Vinay Belgaumkar <vinay.belgaumkar@intel.com> --- drivers/gpu/drm/i915/gem/i915_gem_wait.c | 9 ++++++--- drivers/gpu/drm/i915/i915_request.c | 3 ++- drivers/gpu/drm/i915/i915_request.h | 1 + include/uapi/drm/i915_drm.h | 1 + 4 files changed, 10 insertions(+), 4 deletions(-)