linux.git/tools/power/cpupower/utils, branch v6.5

cpupower: Make TSC read per CPU for Mperf monitor

2023-05-08T16:46:49+00:00

System-wide TSC read could cause a drift in C0 percentage calculation.
Because if first TSC is read and then one by one mperf is read for all
cpus, this introduces drift between mperf reading of later CPUs and TSC
reading.  To lower this drift read TSC per CPU and also just after mperf
read.  This technique improves C0 percentage calculation in Mperf monitor.

Before fix: (System 100% busy)

              | Mperf              || RAPL        || Idle_Stats
 PKG|CORE| CPU| C0   | Cx   | Freq  || pack | core  || POLL | C1   | C2
   0|   0|   0| 87.15| 12.85|  2695||168659003|3970468||  0.00|  0.00| 0.00
   0|   0| 256| 84.62| 15.38|  2695||168659003|3970468||  0.00|  0.00| 0.00
   0|   1|   1| 87.15| 12.85|  2695||168659003|3970468||  0.00|  0.00| 0.00
   0|   1| 257| 84.08| 15.92|  2695||168659003|3970468||  0.00|  0.00| 0.00
   0|   2|   2| 86.61| 13.39|  2695||168659003|3970468||  0.00|  0.00| 0.00
   0|   2| 258| 83.26| 16.74|  2695||168659003|3970468||  0.00|  0.00| 0.00
   0|   3|   3| 86.61| 13.39|  2695||168659003|3970468||  0.00|  0.00| 0.00
   0|   3| 259| 83.60| 16.40|  2695||168659003|3970468||  0.00|  0.00| 0.00
   0|   4|   4| 86.33| 13.67|  2695||168659003|3970468||  0.00|  0.00| 0.00
   0|   4| 260| 83.33| 16.67|  2695||168659003|3970468||  0.00|  0.00| 0.00
   0|   5|   5| 86.06| 13.94|  2695||168659003|3970468||  0.00|  0.00| 0.00
   0|   5| 261| 83.05| 16.95|  2695||168659003|3970468||  0.00|  0.00| 0.00
   0|   6|   6| 85.51| 14.49|  2695||168659003|3970468||  0.00|  0.00| 0.00

After fix: (System 100% busy)

             | Mperf              || RAPL        || Idle_Stats
 PKG|CORE| CPU| C0   | Cx   | Freq  || pack | core  || POLL | C1   | C2
   0|   0|   0| 98.03|  1.97|  2415||163295480|3811189||  0.00|  0.00| 0.00
   0|   0| 256| 98.50|  1.50|  2394||163295480|3811189||  0.00|  0.00| 0.00
   0|   1|   1| 99.99|  0.01|  2401||163295480|3811189||  0.00|  0.00| 0.00
   0|   1| 257| 99.99|  0.01|  2375||163295480|3811189||  0.00|  0.00| 0.00
   0|   2|   2| 99.99|  0.01|  2401||163295480|3811189||  0.00|  0.00| 0.00
   0|   2| 258|100.00|  0.00|  2401||163295480|3811189||  0.00|  0.00| 0.00
   0|   3|   3|100.00|  0.00|  2401||163295480|3811189||  0.00|  0.00| 0.00
   0|   3| 259| 99.99|  0.01|  2435||163295480|3811189||  0.00|  0.00| 0.00
   0|   4|   4|100.00|  0.00|  2401||163295480|3811189||  0.00|  0.00| 0.00
   0|   4| 260|100.00|  0.00|  2435||163295480|3811189||  0.00|  0.00| 0.00
   0|   5|   5| 99.99|  0.01|  2401||163295480|3811189||  0.00|  0.00| 0.00
   0|   5| 261|100.00|  0.00|  2435||163295480|3811189||  0.00|  0.00| 0.00
   0|   6|   6|100.00|  0.00|  2401||163295480|3811189||  0.00|  0.00| 0.00
   0|   6| 262|100.00|  0.00|  2435||163295480|3811189||  0.00|  0.00| 0.00

Cc: Thomas Renninger 
Cc: Shuah Khan 
Cc: Dominik Brodowski 

Fixes: 7fe2f6399a84 ("cpupowerutils - cpufrequtils extended with quite some features")
Signed-off-by: Wyes Karny 
Signed-off-by: Shuah Khan

cpupower: rapl monitor - shows the used power consumption in uj for each rapl domain

2022-11-30T23:48:34+00:00

This CPU power monitor shows the power consumption
as exposed by the powercap subsystem, cmp with:
Documentation/power/powercap/powercap.rst

cpupower monitor -m RAPL
    | RAPL
 CPU| pack | core | unco
   0|6853926|967832|442381
   8|6853926|967832|442381
   1|6853926|967832|442381
   9|6853926|967832|442381

Unfortunately RAPL domains cannot be directly mapped to the corresponding
CPU socket/package, core it belongs to.
Not sure this is possible at all with the current data exposed from the
kernel.

Still it can be worthful information for developers trying to optimize
power consumption of workloads or their system in general.

Signed-off-by: Thomas Renninger 
CC: Zhang Rui 
CC: Shuah Khan 
Signed-off-by: Shuah Khan

cpupower: Introduce powercap intel-rapl library and powercap-info command

2022-11-30T23:48:28+00:00

Read out powercap zone information via:
cpupower powercap-info
and show the zone hierarchy to the user:

./cpupower powercap-info
Driver: intel-rapl
Powercap domain hierarchy:

Zone: package-0 (enabled)
Power consumption can be monitored in micro Watts

        Zone: core (disabled)
        Power consumption can be monitored in micro Watts

        Zone: uncore (disabled)
        Power consumption can be monitored in micro Watts

        Zone: dram (disabled)
        Power consumption can be monitored in micro Watts

There is a dummy -a option for powercap-info which can/should be used to
show more detailed info later. Like that other args can be added easily
later as well.

A enable/disable option via powercap-set subcommand is also an enhancement
for later.

Also not all RAPL domains are shown. The func walking through RAPL
subdomains is restricted and hardcoded to: "intel-rapl/intel-rapl:0"
On my system above powercap domains map to:
intel-rapl/intel-rapl:0
-> pack (age-0)
intel-rapl/intel-rapl:0/intel-rapl:0:0
-> core
intel-rapl/intel-rapl:0/intel-rapl:0:1
-> uncore

Missing ones on my system are:
intel-rapl-mmio/intel-rapl-mmio:0
-> pack (age-0)

intel-rapl/intel-rapl:1
-> psys

This could get enhanced in:
struct powercap_zone *powercap_init_zones()
and adopted to walk through all intel-rapl zones, but
also to other powercap drivers like dtpm
(Dynamic Thermal Power Management framework),
cmp with: drivers/powercap/dtpm_*

Signed-off-by: Thomas Renninger 
CC: Shuah Khan 
Signed-off-by: Shuah Khan

tools/cpupower: Choose base_cpu to display default cpupower details

2022-11-25T17:50:04+00:00

The default output of cpupower info utils shows unexpected output
when CPU 0 is disabled.

Considering a case where CPU 0 is disabled, output of cpupower idle-info:

Before change:
cpupower idle-info
CPUidle driver: pseries_idle
CPUidle governor: menu
analyzing CPU 0:
 *is offline

After change:
./cpupower idle-info
CPUidle driver: pseries_idle
CPUidle governor: menu
analyzing CPU 50:

Number of idle states: 2
Available idle states: snooze CEDE
snooze:
Flags/Description: snooze
Latency: 0
Usage: 101748
Duration: 2724058
CEDE:
Flags/Description: CEDE
Latency: 12
Usage: 270004
Duration: 283019526849

If -c option is not passed, CPU 0 was chosen as the default chosen CPU to
display details. However when CPU 0 is offline, it results in showing
unexpected output. This commit chooses the base_cpu
instead of CPU 0, hence keeping the output more relevant in all cases.
The base_cpu is the number of CPU on which the calling thread is
currently executing.

Signed-off-by: Saket Kumar Bhaskar 
Signed-off-by: Shuah Khan

cpupower: Add "perf" option to print AMD P-State information

2022-02-24T00:46:21+00:00

Add "-c --perf" option in cpupower-frequency-info to get the performance
and frequency values for AMD P-State.

Commit message amended:
Shuah Khan 

Reviewed-by: Shuah Khan 
Signed-off-by: Huang Rui 
Signed-off-by: Shuah Khan

cpupower: Add function to print AMD P-State performance capabilities

2022-02-23T01:37:12+00:00

AMD P-State kernel module is using the fine grain frequency instead of
acpi hardware pstate. So add a function to print performance and
frequency values.

Reviewed-by: Shuah Khan 
Signed-off-by: Huang Rui 
Signed-off-by: Shuah Khan

cpupower: Move print_speed function into misc helper

2022-02-23T01:37:07+00:00

The print_speed can be as a common function, and expose it into misc
helper header. Then it can be used on other helper files as well.

Reviewed-by: Shuah Khan 
Signed-off-by: Huang Rui 
Signed-off-by: Shuah Khan

cpupower: Enable boost state support for AMD P-State module

2022-02-23T01:37:01+00:00

The legacy ACPI hardware P-States function has 3 P-States on ACPI table,
the CPU frequency only can be switched between the 3 P-States. While the
processor supports the boost state, it will have another boost state
that the frequency can be higher than P0 state, and the state can be
decoded by the function of decode_pstates() and read by
amd_pci_get_num_boost_states().

However, the new AMD P-State function is different than legacy ACPI
hardware P-State on AMD processors. That has a finer grain frequency
range between the highest and lowest frequency. And boost frequency is
actually the frequency which is mapped on highest performance ratio. The
similar previous P0 frequency is mapped on nominal performance ratio.
If the highest performance on the processor is higher than nominal
performance, then we think the current processor supports the boost
state. And it uses amd_pstate_boost_init() to initialize boost for AMD
P-State function.

Reviewed-by: Shuah Khan 
Signed-off-by: Huang Rui 
Signed-off-by: Shuah Khan

cpupower: Add AMD P-State sysfs definition and access helper

2022-02-23T01:36:55+00:00

Introduce the marco definitions and access helper function for
AMD P-State sysfs interfaces such as each performance goals and frequency
levels in amd helper file. They will be used to read the sysfs attribute
from AMD P-State cpufreq driver for cpupower utilities.

Reviewed-by: Shuah Khan 
Signed-off-by: Huang Rui 
Signed-off-by: Shuah Khan

cpupower: Initial AMD P-State capability

2022-02-23T01:36:39+00:00

If kernel starts the AMD P-State module, the cpupower will initial the
capability flag as CPUPOWER_CAP_AMD_PSTATE. And once AMD P-State
capability is set, it won't need to set legacy ACPI relative
capabilities anymore.

Reviewed-by: Shuah Khan 
Signed-off-by: Huang Rui 
Signed-off-by: Shuah Khan