1.. SPDX-License-Identifier: GPL-2.0
2
3==================================================
4Collaborative Processor Performance Control (CPPC)
5==================================================
6
7CPPC
8====
9
10CPPC defined in the ACPI spec describes a mechanism for the OS to manage the
11performance of a logical processor on a contiguous and abstract performance
12scale. CPPC exposes a set of registers to describe abstract performance scale,
13to request performance levels and to measure per-cpu delivered performance.
14
15For more details on CPPC please refer to the ACPI specification at:
16
17http://uefi.org/specifications
18
19Some of the CPPC registers are exposed via sysfs under::
20
21  /sys/devices/system/cpu/cpuX/acpi_cppc/
22
23for each cpu X::
24
25  $ ls -lR  /sys/devices/system/cpu/cpu0/acpi_cppc/
26  /sys/devices/system/cpu/cpu0/acpi_cppc/:
27  total 0
28  -r--r--r-- 1 root root 65536 Mar  5 19:38 feedback_ctrs
29  -r--r--r-- 1 root root 65536 Mar  5 19:38 highest_perf
30  -r--r--r-- 1 root root 65536 Mar  5 19:38 lowest_freq
31  -r--r--r-- 1 root root 65536 Mar  5 19:38 lowest_nonlinear_perf
32  -r--r--r-- 1 root root 65536 Mar  5 19:38 lowest_perf
33  -r--r--r-- 1 root root 65536 Mar  5 19:38 nominal_freq
34  -r--r--r-- 1 root root 65536 Mar  5 19:38 nominal_perf
35  -r--r--r-- 1 root root 65536 Mar  5 19:38 reference_perf
36  -r--r--r-- 1 root root 65536 Mar  5 19:38 wraparound_time
37
38* highest_perf : Highest performance of this processor (abstract scale).
39* nominal_perf : Highest sustained performance of this processor
40  (abstract scale).
41* lowest_nonlinear_perf : Lowest performance of this processor with nonlinear
42  power savings (abstract scale).
43* lowest_perf : Lowest performance of this processor (abstract scale).
44
45* lowest_freq : CPU frequency corresponding to lowest_perf (in MHz).
46* nominal_freq : CPU frequency corresponding to nominal_perf (in MHz).
47  The above frequencies should only be used to report processor performance in
48  frequency instead of abstract scale. These values should not be used for any
49  functional decisions.
50
51* feedback_ctrs : Includes both Reference and delivered performance counter.
52  Reference counter ticks up proportional to processor's reference performance.
53  Delivered counter ticks up proportional to processor's delivered performance.
54* wraparound_time: Minimum time for the feedback counters to wraparound
55  (seconds).
56* reference_perf : Performance level at which reference performance counter
57  accumulates (abstract scale).
58
59
60Computing Average Delivered Performance
61=======================================
62
63Below describes the steps to compute the average performance delivered by
64taking two different snapshots of feedback counters at time T1 and T2.
65
66  T1: Read feedback_ctrs as fbc_t1
67      Wait or run some workload
68
69  T2: Read feedback_ctrs as fbc_t2
70
71::
72
73  delivered_counter_delta = fbc_t2[del] - fbc_t1[del]
74  reference_counter_delta = fbc_t2[ref] - fbc_t1[ref]
75
76  delivered_perf = (refernce_perf x delivered_counter_delta) / reference_counter_delta
77