Created attachment 101142 [details] dmesg ==System Environment== -------------------------- 3.15-kernel-2014Q2-release (fail) commit:1860e379875dfe7271c649058aeddffe5afd9d0d ==kernel== -------------------------- -nightly: b438e8793c6d18b0998d75cdf8115a5898ed21d1 (works) drm-intel-nightly: 2014y-06m-13d-22h-09m-12s integration manifest -queued: 868d665b43473e230d560d5186535270a3d57a19 (works) drm/i915: Fix memory leak in intel_dsi_init() error path -fixes: 223a6f2b975ab35d93270ea1d4fb6e0ac6b27fe6 (works) drm/i915/bdw: remove erroneous chv specific workarounds from bdw code ==Bug detailed description== ----------------------------- GFX device consume same Watt no matter when it's idle or active. [root@x-bdw05 tests]# /GFX/Test/P304 frames in 5.0 seconds = 60.730 FPS erformance/performance_pnp/301 frames in 5.0 seconds = 60.001 FPS turbostat 301 frames in 5.0 seconds = 60.001 FPS Core CPU Avg_MHz %Busy Bzy_MHz TSC_MHz SMI CPU%c1 CPU%c3 CPU%c6 CPU%c7 CoreTmp PkgTmp Pkg%pc2 Pkg%pc3 Pkg%pc6 Pkg%pc7 Pkg%pc8 Pkg%pc9 Pk%pc10 PkgWatt CorWatt GFXWatt - - 7 0.43 1631 1796 0 99.57 0.00 0.00 0.00 35 36 0.00 0.00 0.00 0.00 0.00 0.00 0.00 2.23 0.06 0.06 0 0 12 0.77 1559 1796 0 99.23 0.00 0.00 0.00 35 36 0.00 0.00 0.00 0.00 0.00 0.00 0.00 2.23 0.06 0.06 0 2 0 0.01 1500 1796 0 99.99 1 1 16 0.92 1694 1796 0 99.08 0.00 0.00 0.00 34 1 3 0 0.01 1378 1796 0 99.99 ==Reproduce steps== ---------------------------- xinit & glxgears & ./turbostat
glxgears is not exactly a busy workload. Your table is impossible to read due to line wrapping, please attach.
Created attachment 101146 [details] turbost result
(In reply to comment #1) > glxgears is not exactly a busy workload. Even though glxgears is not a busy workload,but GFX device should consume more power rather than the same power when it's Idle >Your table is impossible to read > due to line wrapping, please attach. I change workload ,and attach its result.
It's measuring 60mW which would be close to idle non-rc6 power. (Estimating from earlier generations). That is either a measurement error or you simply aren't loading the system enough to produce a significant load.
(In reply to comment #4) > It's measuring 60mW which would be close to idle non-rc6 power. (Estimating > from earlier generations). That is either a measurement error or you simply > aren't loading the system enough to produce a significant load. For measurement solution: The GFXWatt got from turbostat tool is a kind of RAPL(Running Average Power Limit) power measurement solution, which should be reliable test method. For workload: My expected result is we should get different power consumption when I load some workload comparing with idle scenario. I encountered problem is GFXwatt no change whatever idle or other gfx workload. Pls correct and educate me if my understanding is wrong, thanks!
Your hypothesis is valid, I worry that either the measurement itself is faulty (I haven't checked if the RAPL registers for the GPU have been moved for bdw) or if simply that you haven't generated enough GPU load for it to consume significantly more power - i.e. it we are not hitting rc6 and if the test stays at low frequency throughout, it is plausible for you to not measure any difference. So repeat your test with a busy workload and see if that then detects a difference in GPU power consumption (to validate your testing methodology).
(In reply to comment #6) > Your hypothesis is valid, I worry that either the measurement itself is > faulty (I haven't checked if the RAPL registers for the GPU have been moved > for bdw) or if simply that you haven't generated enough GPU load for it to > consume significantly more power Do you have a recommended workload which would generate enough power meter? >- i.e. it we are not hitting rc6 and if the > test stays at low frequency throughout, it is plausible for you to not > measure any difference. > > So repeat your test with a busy workload and see if that then detects a > difference in GPU power consumption (to validate your testing methodology). In fact, I have tested three different game demos(3DMMES2_taiji, cs, OglDrvCtx), and get the same consumption as idle.
The game workloads would definitely cause more power, as would any of the 2D tests. If you are still seeing no change in power consumption, I would guess that the rapl monitor needs to be adjusted for bdw.
When we file this bug, we verified below kernels, they are working well, so we'd like to assume RAPL solution is workable on BDW, should be 2014Q2 release kernel(commit:1860e379875dfe7271c649058aeddffe5afd9d0d) problem, thanks. -------------------------- -nightly: b438e8793c6d18b0998d75cdf8115a5898ed21d1 (works) drm-intel-nightly: 2014y-06m-13d-22h-09m-12s integration manifest -queued: 868d665b43473e230d560d5186535270a3d57a19 (works) drm/i915: Fix memory leak in intel_dsi_init() error path -fixes: 223a6f2b975ab35d93270ea1d4fb6e0ac6b27fe6 (works) drm/i915/bdw: remove erroneous chv specific workarounds from bdw code
Wendy, I'm a bit confused by your statement. Can we close the bug then?
(In reply to comment #10) > Wendy, I'm a bit confused by your statement. Can we close the bug then? We found this bug only on release kernel
Reporter, what is the status of this bug on current kernels?
(In reply to comment #12) > Reporter, what is the status of this bug on current kernels? This issue doesn't exist on latest three branches kernels.
Mystery bug comes and goes.
Closing old verified.
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.