Bug 16513
| Summary: | CPUs doesn't stay in deepest C-state with intel_idle | ||
|---|---|---|---|
| Product: | Power Management | Reporter: | Daniel Eklöf (daniel) |
| Component: | intel_idle | Assignee: | power-management_intel_idle (power-management_intel_idle) |
| Status: | CLOSED UNREPRODUCIBLE | ||
| Severity: | normal | CC: | akpm |
| Priority: | P1 | ||
| Hardware: | All | ||
| OS: | Linux | ||
| Kernel Version: | 2.6.35 | Subsystem: | |
| Regression: | No | Bisected commit-id: | |
| Attachments: |
Kernel log
Powertop output w/ intel-idle Powertop output w/ acpi-idle |
||
Created attachment 27341 [details]
Powertop output w/ intel-idle
Created attachment 27342 [details]
Powertop output w/ acpi-idle
Recategorised to PM, assigned to Len. can you reproduce this issue with 2.6.36? No, I can't. Tested with 2.6.36-rc8 and I get 99.9% residency in C4, and a wakeup count below 10. I.e. everything is as it should be. Looks like we can close this one! |
Created attachment 27340 [details] Kernel log When using the new intel-idle driver, the CPUs never reach 100% residency in the deepest C-state. Instead, it's spread across C1-C4 (C4 being the deepest C-state). Powertop also shows a large number of wakeups: Cn Avg residency C0 (cpu running) ( 0.0%) polling 0.0ms ( 0.0%) C1 mwait 61.8ms (24.5%) C2 mwait 5.3ms (34.8%) C4 mwait 0.8ms (41.1%) P-states (frequencies) 1.67 Ghz 2.6% 1333 Mhz 0.1% 1000 Mhz 97.4% Wakeups-from-idle per second : 595.0 interval: 15.0s no ACPI power usage estimate available Top causes for wakeups: 42.6% ( 12.2) [extra timer interrupt] 14.0% ( 4.0) [kernel core] usb_hcd_poll_rh_status (rh_timer_func) 9.8% ( 2.8) [i915] <interrupt> When using the ACPI driver, things are also somewhat interresting; directly after boot, there are no C-states available. After a couple of minutes (~3 minutes), the C-states become available (only up to C2 when running on AC, C4 when running on battery). At this point one can also see a large drop in power usage. So, to summarize: acpi-idle: works, but it takes ~3 minutes for the C-states to become available after boot intel-idle: C-states are available right away, but something seems to be interferring, and we are never able to stay in the deepest C-state (regardless of how long time I wait). intel-idle output: Aug 4 12:16:07 lappy kernel: intel_idle: MWAIT substates: 0x20220 Aug 4 12:16:07 lappy kernel: intel_idle: v0.4 model 0x1C Aug 4 12:16:07 lappy kernel: intel_idle: lapic_timer_reliable_states 0x6 Aug 4 12:16:07 lappy kernel: ACPI: acpi_idle yielding to intel_idle Hardware: Samsung N150 (Atom N450), BIOS v.08JI