Created attachment 106351 [details] dmesg On my Dell XPS 15 (9530 model, late 2013) that uses vendor_id : GenuineIntel cpu family : 6 model : 60 model name : Intel(R) Core(TM) i7-4702HQ CPU @ 2.20GHz stepping : 3 microcode : 0x1a cpu MHz : 2899.960 cache size : 6144 KB there is a problem that after resume from ram the power usage rises a lot, about 7-10W (from 26-30W to 36-40W). It happens with minimal userspace, so I don't blame userspace. powertop estimates that 17W is used by "CPU core" (I assume intel gpu hides under this, too). "perf top" doesn't show any obvious suspects. When the problem happens registers look like this: --- power-ok 2014-09-15 22:30:20.615917605 +0200 +++ power-bad 2014-09-15 22:31:43.383356688 +0200 @@ -130,12 +130,12 @@ TRANS_VSYNC_A: 0x00000000 (1 start, 1 end) TRANS_VSYNCSHIFT_A: 0x00000000 TRANSACONF: 0x00000000 (disable, inactive, progressive) - FDI_RXA_MISC: 0x0a200090 (FDI Delay 144) + FDI_RXA_MISC: 0x00200080 (FDI Delay 128) FDI_RXA_TUSIZE1: 0x7e000000 FDI_RXA_IIR: 0x00000000 FDI_RXA_IMR: 0x00000fff BLC_PWM_CPU_CTL2: 0xe0000000 (enable 1, pipe EDP, blinking 0, granularity 128) - BLC_PWM_CPU_CTL: 0x149906e9 (cycle 1769, freq 5273) + BLC_PWM_CPU_CTL: 0x00000cc7 (cycle 3271, freq 0) BLC_PWM2_CPU_CTL2: 0x60000000 (enable 0, pipe EDP, blinking 0, granularity 128) BLC_PWM2_CPU_CTL: 0x00000000 (cycle 0, freq 0) BLC_MISC_CTL: 0x00000000 (PWM1-PCH PWM2-CPU) @@ -149,7 +149,7 @@ PCH_PP_DIVISOR: 0x00186906 PIXCLK_GATE: 0x00000000 SDEISR: 0x00800000 (port d:1, port c:0, port b:0, crt:0) - RC6_RESIDENCY_TIME: 0x771f52d8 + RC6_RESIDENCY_TIME: 0x02ee80e1 FENCE START 0: 0x02cc0001 FENCE END 0: 0x042b8063 FENCE START 1: 0x00000000 For me that happens under 3.17.0-rc5 and 3.14.18. Going to test other versions soon. Note that there was similar problem for SNB but not sure if these are related: Bug 54089 - [SNB regression] Power consumption goes postal after resume
(In reply to comment #0) > It happens with minimal userspace, so I don't blame userspace. powertop > estimates that 17W is used by "CPU core" (I assume intel gpu hides under > this, too). "perf top" doesn't show any obvious suspects. The GPU itself is measured separately, but that only really measures the EU power, not the display or the influence of the GPU on the L3 cache ring. (I would expect the ring power to be tallied under the CPU, so it is entirely possible the GPU is causing it to run at maximum frequency and we end up blaming the CPU.) Best hope is that this is an easily identifiable kernel regression. *fingers crossed*
Looks like GPU is innocent. This machine has 8 cores (4 normal and 4 HT). I was using maxcpus=4 (due to Dell XPS 15 bugs like whining coil). My current theory is that after resume from ram these 4 HT cores are eating power. Why? Because if not using maxcpus option then everything fine and there is no issue at all. It could be called kernel bug to not shut up these additional cores after resume but well...
... and workaround that gets power usage to normal when using maxcpus=X for cpu in $(grep -l 0 /sys/devices/system/cpu/cpu*/online); do echo 1 > $cpu sleep 1 echo 0 > $cpu done
Arkadiusz, thanks for following up with this. I'd like to ask you to do one more thing, though: please file a bug on https://bugzilla.kernel.org/ or the linux-kernel mailing list. Thanks.
https://bugzilla.kernel.org/show_bug.cgi?id=84741
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.