| 1 | |
| 2 |
|
| 3 | ==================================================
|
| 4 | Collaborative Processor Performance Control (CPPC)
|
| 5 | ==================================================
|
| 6 |
|
| 7 | .. _cppc_sysfs:
|
| 8 |
|
| 9 | CPPC
|
| 10 | ====
|
| 11 |
|
| 12 | CPPC defined in the ACPI spec describes a mechanism for the OS to manage the
|
| 13 | performance of a logical processor on a contiguous and abstract performance
|
| 14 | scale. CPPC exposes a set of registers to describe abstract performance scale,
|
| 15 | to request performance levels and to measure per-cpu delivered performance.
|
| 16 |
|
| 17 | For more details on CPPC please refer to the ACPI specification at:
|
| 18 |
|
| 19 | http://uefi.org/specifications
|
| 20 |
|
| 21 | Some of the CPPC registers are exposed via sysfs under::
|
| 22 |
|
| 23 | /sys/devices/system/cpu/cpuX/acpi_cppc/
|
| 24 |
|
| 25 | for each cpu X::
|
| 26 |
|
| 27 | $ ls -lR /sys/devices/system/cpu/cpu0/acpi_cppc/
|
| 28 | /sys/devices/system/cpu/cpu0/acpi_cppc/:
|
| 29 | total 0
|
| 30 | -r--r--r-- 1 root root 65536 Mar 5 19:38 feedback_ctrs
|
| 31 | -r--r--r-- 1 root root 65536 Mar 5 19:38 highest_perf
|
| 32 | -r--r--r-- 1 root root 65536 Mar 5 19:38 lowest_freq
|
| 33 | -r--r--r-- 1 root root 65536 Mar 5 19:38 lowest_nonlinear_perf
|
| 34 | -r--r--r-- 1 root root 65536 Mar 5 19:38 lowest_perf
|
| 35 | -r--r--r-- 1 root root 65536 Mar 5 19:38 nominal_freq
|
| 36 | -r--r--r-- 1 root root 65536 Mar 5 19:38 nominal_perf
|
| 37 | -r--r--r-- 1 root root 65536 Mar 5 19:38 reference_perf
|
| 38 | -r--r--r-- 1 root root 65536 Mar 5 19:38 wraparound_time
|
| 39 |
|
| 40 | * highest_perf : Highest performance of this processor (abstract scale).
|
| 41 | * nominal_perf : Highest sustained performance of this processor
|
| 42 | (abstract scale).
|
| 43 | * lowest_nonlinear_perf : Lowest performance of this processor with nonlinear
|
| 44 | power savings (abstract scale).
|
| 45 | * lowest_perf : Lowest performance of this processor (abstract scale).
|
| 46 |
|
| 47 | * lowest_freq : CPU frequency corresponding to lowest_perf (in MHz).
|
| 48 | * nominal_freq : CPU frequency corresponding to nominal_perf (in MHz).
|
| 49 | The above frequencies should only be used to report processor performance in
|
| 50 | frequency instead of abstract scale. These values should not be used for any
|
| 51 | functional decisions.
|
| 52 |
|
| 53 | * feedback_ctrs : Includes both Reference and delivered performance counter.
|
| 54 | Reference counter ticks up proportional to processor's reference performance.
|
| 55 | Delivered counter ticks up proportional to processor's delivered performance.
|
| 56 | * wraparound_time: Minimum time for the feedback counters to wraparound
|
| 57 | (seconds).
|
| 58 | * reference_perf : Performance level at which reference performance counter
|
| 59 | accumulates (abstract scale).
|
| 60 |
|
| 61 |
|
| 62 | Computing Average Delivered Performance
|
| 63 | =======================================
|
| 64 |
|
| 65 | Below describes the steps to compute the average performance delivered by
|
| 66 | taking two different snapshots of feedback counters at time T1 and T2.
|
| 67 |
|
| 68 | T1: Read feedback_ctrs as fbc_t1
|
| 69 | Wait or run some workload
|
| 70 |
|
| 71 | T2: Read feedback_ctrs as fbc_t2
|
| 72 |
|
| 73 | ::
|
| 74 |
|
| 75 | delivered_counter_delta = fbc_t2[del] - fbc_t1[del]
|
| 76 | reference_counter_delta = fbc_t2[ref] - fbc_t1[ref]
|
| 77 |
|
| 78 | delivered_perf = (reference_perf x delivered_counter_delta) / reference_counter_delta
|