Bug 11170
| Summary: | 2.6.24 w/ nohz=off runs cooler than 2.6.25 tickless | ||
|---|---|---|---|
| Product: | ACPI | Reporter: | James Ettle (james) |
| Component: | Power-Processor | Assignee: | cpufreq (cpufreq) |
| Status: | CLOSED INSUFFICIENT_DATA | ||
| Severity: | normal | CC: | acpi-bugzilla, bunk, rui.zhang, trenn |
| Priority: | P1 | ||
| Hardware: | All | ||
| OS: | Linux | ||
| Kernel Version: | 2.6.25.12-62.fc8 | Subsystem: | |
| Regression: | Yes | Bisected commit-id: | |
| Attachments: |
acpidump output for this machine
dmesg output with kernel 2.6.24.7-92.fc8 cpu0cst from kernel-2.6.25.14-67.fc8 cpu0ist from kernel-2.6.24.7-92.fc8 cpu0ist from kernel-2.6.25.14-67.fc8 cpu1cst from kernel-2.6.25.14-67.fc8 cpu1ist from kernel-2.6.25.14-67.fc8 2.6.24 output, AC power 2.6.26 output, AC power 2.6.24 output, battery power 2.6.26 output, battery power Comment #18 requested data on 2.6.27.4-26.fc9.x86_64 w/ ondemand governor Comment #18 requested data on 2.6.27.4-26.fc9.x86_64 w/ powersave governor |
||
Please provide acpidump and dmesg. This should be assigned to the ACPI and not Power Management component. I cannot change it. Created attachment 17030 [details]
acpidump output for this machine
Created attachment 17031 [details]
dmesg output with kernel 2.6.24.7-92.fc8
Will you please attach the following outputs?
acpidump --addr 0x7f6d4e23 --length 0x2A8 -o cpu0ist
acpidump --addr 0x7f6d50cb --length 0xb1 -o cpu1ist
acpidump --addr 0x7f6d48f4 --length 0x4aa -o cpu0cst
acpidump --addr 0x7f6d4d9e --length 0x85 -o cpu1cst
Thanks.
I ran the above commands under both kernels 2.6.24.7-92.fc8 and 2.6.25.14-67.fc8. cpu0ist was different between kernel versions, so I have posted both here. The other files were the same and I've posted the ones obtained under 2.6.25.14. Created attachment 17080 [details]
cpu0cst from kernel-2.6.25.14-67.fc8
Created attachment 17081 [details]
cpu0ist from kernel-2.6.24.7-92.fc8
Created attachment 17082 [details]
cpu0ist from kernel-2.6.25.14-67.fc8
Created attachment 17083 [details]
cpu1cst from kernel-2.6.25.14-67.fc8
Created attachment 17084 [details]
cpu1ist from kernel-2.6.25.14-67.fc8
Some extra info, don't know if it's related. Another thing I've just noticed with 2.6.25.14-67.fc8 is that it no longer limits the CPU clock to 1.2 GHz on battery power or when the "smart power" button is pressed. please attach the result of "grep . /sys/devices/system/cpu/cpu0/cpufreq/*" on both 2.6.25 and 2.6.24 kernel. I've attached the requested data from kernel-2.6.24.7-92.fc8 and kernel-2.6.26.3-14.fc8, both on battery and AC power. Created attachment 17743 [details]
2.6.24 output, AC power
Created attachment 17744 [details]
2.6.26 output, AC power
Created attachment 17745 [details]
2.6.24 output, battery power
Created attachment 17746 [details]
2.6.26 output, battery power
comment #14 and #15 shows that cpu is running in 800000 in 2.6.26 and 21010000 in 2.6.24. First, please use the "powersave" governor in both 2.6.24 and 2.6.26. if the problem still exists, please run "grep . /sys/devices/system/cpu/cpu0/cpuidle/*/*" and attach the result. no response from bug reporter for over two months. James, please re-open it if you can provide the info listed in comment #18 Hi --- sorry for neglecting you. I've just started a new job and this has kept me rather busy. On this bug: it *appears* to be cured in (at least) 2.6.26.7 on x86-64; I'm now using 2.6.27.4 which at idle cools down to the point where the fan turns off. I'll attached the above-requested anyway in case it is of some use. Created attachment 18726 [details] Comment #18 requested data on 2.6.27.4-26.fc9.x86_64 w/ ondemand governor Created attachment 18727 [details] Comment #18 requested data on 2.6.27.4-26.fc9.x86_64 w/ powersave governor Unfortunately, since upgrading to Fedora 9 I no longer have access to the 2.6.24 kernel... I'll close it a fixed for the time being. For further reference: this appears to be fixed for now. Thanks! |
Latest working kernel version: 2.6.24.7-92.fc8 Earliest failing kernel version: 2.6.25 Distribution: Fedora 8 Hardware Environment: Intel Core 2 Duo T8100 on Clevo M720R notebook Software Environment: i686 Problem Description: Kernel 2.6.24.7 runs cooler on this machine, even when booted "nohz=off", than Kernel 2.6.25.* in tickless mode. The difference, in both coretemp and ACPI thermal zone, is typically 4-6 C. This is all the more odd since 2.6.25 seems to better support ACPI features like the lid and sleep button, which were detected under 2.6.24 but didn't generate events, and the "smart power" button which limits the clock to 1.2 GHz. [I'm booting "nohz=off" due to another problem whereby the machine seems to "pause" (*not* freeze) when there are few interrupts happening.] Under 2.6.24.7, powertop reports 100% "C0 (cpu running)" residency, just altering the clock rate, and around 100-200 wups. With 2.6.25.12, it shows two C0 states, "C0 (cpu running)" and just "C0"; the residency figures don't make much sense, and I suspect powertop is reporting garbage in both cases (sometimes things like 8 wups). Furthermore, I don't really understand what I see in /proc/acpi/processor/*/power --- with 2.6.24.7, only C1 shows any usage, and the number keeps resetting after a few seconds. With 2.6.25.12, C1--C3 show some usage, again with the numbers resetting. Another thing (I don't know if it's related) is that under 2.6.25.12, I get the following in dmesg: ACPI Exception (thermal-0457): AE_ERROR, ACPI thermal trip point state changed I also see a lot of ACPI: Invalid passive threshold in dmesg for 2.6.24.7 if I cold-boot 2.6.25.12 first. Furthermore, cold-booting 2.6.25.12 and then re-booting into 2.6.24.7 makes the lid work under the latter.