Probe and display L3 Cache topology
Add ability to average an added counter
(useful for pre-integrated "counters", such as Watts)
Break the limit of 64 built-in counters.
Assorted bug fixes and minor feature tweaks
Signed-off-by: Len Brown <len.brown@intel.com>
/sys/devices/system/cpu/intel_uncore_frequency/package_X_die_Y/
may be readable by all, but
/sys/devices/system/cpu/intel_uncore_frequency/package_X_die_Y/current_freq_khz
may be readable only by root.
Non-root turbostat users see complaints in this scenario.
Fail probe of the interface if we can't read current_freq_khz.
Reported-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Original-patch-by: Zhang Rui <rui.zhang@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
Together with the RAPL MSRs, there are more MSRs gone on DMR, including
PLR (Perf Limit Reasons), and IRTL (Package cstate Interrupt Response
Time Limit) MSRs. The configurable TDP info should also be retrieved
from TPMI based Intel Speed Select Technology feature.
Remove the access of these MSRs for DMR. Improve the DMR platform
feature table to make it more readable at the same time.
Fixes: 83075bd59d ("tools/power turbostat: Add initial support for DMR")
Signed-off-by: Zhang Rui <rui.zhang@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
External atributes with format "raw" are not printed in summary lines
for nodes/packages (or with option -S). The new format "average"
behaves like "raw" but also adds the summary data
Signed-off-by: Michael Hebenstreit <michael.hebenstreit@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
We have out-grown the ability to use a 64-bit memory location
to inventory every possible built-in counter.
Leverage the the CPU_SET(3) macros to break this barrier.
Also, break the Joules & Watts counters into two,
since we can no longer 'or' them together...
Signed-off-by: Len Brown <len.brown@intel.com>
Kernels configured with CONFIG_MULTIUSER=n have no cap_get_proc().
Check for ENOSYS to recognize this case, and continue on to
attempt to access the requested MSRs (such as temperature).
Signed-off-by: Calvin Owens <calvin@wbinvd.org>
Signed-off-by: Len Brown <len.brown@intel.com>
turbostat.c: In function 'parse_int_file':
turbostat.c:5567:19: error: 'PATH_MAX' undeclared (first use in this function)
5567 | char path[PATH_MAX];
| ^~~~~~~~
turbostat.c: In function 'probe_graphics':
turbostat.c:6787:19: error: 'PATH_MAX' undeclared (first use in this function)
6787 | char path[PATH_MAX];
| ^~~~~~~~
Signed-off-by: Calvin Owens <calvin@wbinvd.org>
Reviewed-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
$ sudo turbostat --quiet --show junk
turbostat: Counter 'junk' can not be added.
Previously, invalid arguments to --show and --hide were silently ignored
Acked-by: Zhang Rui <rui.zhang@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
The new default idle counter groupings broke "--show C1E%" (or any other C-state %)
Also delete a stray debug printf from the same offending commit.
Reported-by: Zhang Rui <rui.zhang@intel.com>
Fixes: ec4acd3166 ("tools/power turbostat: disable "cpuidle" invocation counters, by default")
Signed-off-by: Len Brown <len.brown@intel.com>
Add initial DMR support, which required smarter RAPL probe
Fix AMD MSR RAPL energy reporting
Add RAPL power limit configuration output
Minor fixes
Signed-off-by: Len Brown <len.brown@intel.com>
For the RAPL package energy status counter, Intel and AMD share the same
perf_subsys and perf_name, but with different MSR addresses.
Both rapl_counter_arch_infos[0] and rapl_counter_arch_infos[1] are
introduced to describe this counter for different Vendors.
As a result, the perf counter is probed twice, and causes a failure in
in get_rapl_counters() because expected_read_size and actual_read_size
don't match.
Fix the problem by skipping the already probed counter.
Note, this is not a perfect fix. For example, if different
vendors/platforms use the same MSR value for different purpose, the code
can be fooled when it probes a rapl_counter_arch_infos[] entry that does
not belong to the running Vendor/Platform.
In a long run, better to put rapl_counter_arch_infos[] into the
platform_features so that this becomes Vendor/Platform specific.
Signed-off-by: Zhang Rui <rui.zhang@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
platform_features->rapl_msrs describes the RAPL MSRs supported. While
RAPL Perf counters can be exposed from different kernel backend drivers,
e.g. RAPL MSR I/F driver, or RAPL TPMI I/F driver.
Thus, turbostat should first blindly probe all the available RAPL Perf
counters, and falls back to the RAPL MSR counters if they are listed in
platform_features->rapl_msrs.
With this, platforms that don't have RAPL MSRs can clear the
platform_features->rapl_msrs bits and use RAPL Perf counters only.
Signed-off-by: Zhang Rui <rui.zhang@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
Increase the code readability by moving the no_perf/no_msr flag and the
cai->perf_name/cai->msr sanity checks into the counter probe functions.
No functional change.
Signed-off-by: Zhang Rui <rui.zhang@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
probe_rapl_msr() is reused for probing RAPL MSR counters, cstate MSR
counters and MPERF/APERF/SMI MSR counters, thus its name is misleading.
Similar to add_perf_counter(), introduce add_msr_counter() to probe a
counter via MSR. Introduce wrapper function add_rapl_msr_counter() at
the same time to add extra check for Zero return value for specified
RAPL counters.
No functional change intended.
Signed-off-by: Zhang Rui <rui.zhang@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
As the only caller of add_msr_perf_counter_(), add_msr_perf_counter()
just gives extra debug output on top. There is no need to keep both
functions.
Remove add_msr_perf_counter_() and move all the logic to
add_msr_perf_counter().
No functional change.
Signed-off-by: Zhang Rui <rui.zhang@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
As the only caller of add_cstate_perf_counter_(),
add_cstate_perf_counter() just gives extra debug output on top. There is
no need to keep both functions.
Remove add_cstate_perf_counter_() and move all the logic to
add_cstate_perf_counter().
No functional change.
Signed-off-by: Zhang Rui <rui.zhang@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
As the only caller of add_rapl_perf_counter_(), add_rapl_perf_counter()
just gives extra debug output on top. There is no need to keep both
functions.
Remove add_rapl_perf_counter_() and move all the logic to
add_rapl_perf_counter().
No functional change.
Signed-off-by: Zhang Rui <rui.zhang@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
Quit early for unsupported RAPL counters.
No functional change.
Signed-off-by: Zhang Rui <rui.zhang@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
rapl_joules bit should always be checked even if
platform_features->rapl_msrs is not set or no_msr flag is used.
Signed-off-by: Zhang Rui <rui.zhang@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
commit 05a2f07db8 ("tools/power turbostat: read RAPL counters via
perf") that adds support to read RAPL counters via perf defines the
notion of a RAPL domain_id which is set to physical_core_id on
platforms which support per_core_rapl counters (Eg: AMD processors
Family 17h onwards) and is set to the physical_package_id on all the
other platforms.
However, the physical_core_id is only unique within a package and on
platforms with multiple packages more than one core can have the same
physical_core_id and thus the same domain_id. (For eg, the first cores
of each package have the physical_core_id = 0). This results in all
these cores with the same physical_core_id using the same entry in the
rapl_counter_info_perdomain[]. Since rapl_perf_init() skips the
perf-initialization for cores whose domain_ids have already been
visited, cores that have the same physical_core_id always read the
perf file corresponding to the physical_core_id of the first package
and thus the package-energy is incorrectly reported to be the same
value for different packages.
Note: This issue only arises when RAPL counters are read via perf and
not when they are read via MSRs since in the latter case the MSRs are
read separately on each core.
Fix this issue by associating each CPU with rapl_core_id which is
unique across all the packages in the system.
Fixes: 05a2f07db8 ("tools/power turbostat: read RAPL counters via perf")
Signed-off-by: Gautham R. Shenoy <gautham.shenoy@amd.com>
Signed-off-by: Len Brown <len.brown@intel.com>
Fix typo in the currently unused RAPL_GFX_ALL macro definition.
Signed-off-by: Kaushlendra Kumar <kaushlendra.kumar@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
It uses /dev/msrN device paths on Android instead of /dev/cpu/N/msr,
updates error messages and permission checks to reflect the Android
device path, and wraps platform-specific code with #if defined(ANDROID)
to ensure correct behavior on both Android and non-Android systems.
These changes improve compatibility and usability of turbostat on
Android devices.
Signed-off-by: Kaushlendra Kumar <kaushlendra.kumar@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
Support up to 8192 processors
Add cpuidle governor debug telemetry, disabled by default
Update default output to exclude cpuidle invocation counts
Bug fixes
Signed-off-by: Len Brown <len.brown@intel.com>
Create "pct_idle" counter group, the sofware notion of residency
so it can now be singled out, independent of other counter groups.
Create "cpuidle" group, the cpuidle invocation counts.
Disable "cpuidle", by default.
Create "swidle" = "cpuidle" + "pct_idle".
Undocument "sysfs", the old name for "swidle", but keep it working
for backwards compatibilty.
Create "hwidle", all the HW idle counters
Modify "idle", enabled by default
"idle" = "hwidle" + "pct_idle" (and now excludes "cpuidle")
Signed-off-by: Len Brown <len.brown@intel.com>
Probe cpuidle "sysfs" residency and counts separately,
since soon we will make one disabled on, and the
other disabled off.
Clarify that some BIC (build-in-counters) are actually "groups".
since we're about to re-name some of those groups.
no functional change.
Signed-off-by: Len Brown <len.brown@intel.com>
Do fflush() to discard the buffered data, before each read of the
graphics sysfs knobs.
Fixes: ba99a4fc8c ("tools/power turbostat: Remove unnecessary fflush() call")
Signed-off-by: Zhang Rui <rui.zhang@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
Document that on Intel Granite Rapids Systems,
Uncore domains 0-2 are CPU domains, and
uncore domains 3-4 are IO domains.
Signed-off-by: Len Brown <len.brown@intel.com>
The CoreThr column displays total thermal throttling events
since boot time.
Change it to report events during the measurement interval.
This is more useful for showing a user the current conditions.
Total events since boot time are still available to the user via
/sys/devices/system/cpu/cpu*/thermal_throttle/*
Document CoreThr on turbostat.8
Fixes: eae97e053f ("turbostat: Support thermal throttle count print")
Reported-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
Cc: Chen Yu <yu.c.chen@intel.com>
On systems with >= 1024 cpus (in my case 1152), turbostat fails with the error output:
"turbostat: /sys/fs/cgroup/cpuset.cpus.effective: cpu str malformat 0-1151"
A similar error appears with the use of turbostat --cpu when the inputted cpu
range contains a cpu number >= 1024:
# turbostat -c 1100-1151
"--cpu 1100-1151" malformed
...
Both errors are caused by parse_cpu_str() reaching its limit of CPU_SUBSET_MAXCPUS.
It's a good idea to limit the maximum cpu number being parsed, but 1024 is too low.
For a small increase in compute and allocated memory, increasing CPU_SUBSET_MAXCPUS
brings support for parsing cpu numbers >= 1024.
Increase CPU_SUBSET_MAXCPUS to 8192, a common setting for CONFIG_NR_CPUS on x86_64.
Signed-off-by: Justin Ernst <justin.ernst@hpe.com>
Signed-off-by: Len Brown <len.brown@intel.com>
The idle governor provides the following per-idle state sysfs files:
* above - Indicates overshoots, where a more shallow state should have
been requested (if avaliale and enabled).
* below - Indicates undershoots, where a deeper state should have been
requested (if available and enabled).
These files offer valuable insights into how effectively the Linux kernel
idle governor selects idle states for a given workload. This commit adds
support for these files in turbostat.
Expose the contents of these files with the following naming convention:
* C1: The number of times the C1 state was requested (existing counter).
* C1+: The number of times the idle governor selected C1, but a deeper
idle state should have been selected instead.
* C1-: The number of times the idle governor selected C1, but a shallower
idle state should have been selected instead.
Signed-off-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
Fix the 'find_msrp_by_name()' function which returns incorrect matches for
cases like this:
s1 = "C1-";
find_msrp_by_name(head, s1);
Inside 'find_msrp_by_name()':
...
s2 = "C1"
if !(strcnmp(s1, s2, len(s2)))
// Incorrect match!
return mp;
Full strings should be match istead. Switch to 'strcmp()' to fix the problem.
Signed-off-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
turbostat aborted with below messages on a dual-package system,
turbostat: turbostat.c:3744: rapl_counter_accumulate: Assertion `dst->unit == src->unit' failed.
Aborted
This is because
1. the MSR_DRAM_PERF_STATUS returns Zero for one package, and non-Zero
for another package
2. probe_msr() treats Zero return value as a failure so this feature is
enabled on one package, and disabled for another package.
3. turbostat aborts because the feature is invalid on some package
Unlike the RAPL energy counter registers, MSR_DRAM_PERF_STATUS can
return Zero value, and this should not be treated as a failure.
Fix the problem by allowing Zero return value for RAPL registers other
than the energy counters.
Fixes: 7c6fee25bd ("tools/power turbostat: Check for non-zero value when MSR probing")
Reported-by: Artem Bityutskiy <artem.bityutskiy@intel.com>
Signed-off-by: Zhang Rui <rui.zhang@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
The clustered uncore frequency counters, UMHz*.*
should honor the --show and --hide options.
All non-specified counters should be implicityly hidden.
But when --show was used, UMHz*.* showed up anyway:
$ sudo turbostat -q -S --show Busy%
Busy% UMHz0.0 UMHz1.0 UMHz2.0 UMHz3.0 UMHz4.0
Indeed, there was no string that can be used to explicitly
show or hide clustered uncore counters.
Even through they are dynamically probed and added,
group the clustered UMHz*.* counters with the legacy
built-in-counter "UncMHz" for show/hide.
turbostat --show Busy%
does not show UMHz*.*.
turbostat --show UncMHz
shows either UncMHz or UMHz*.*, if present
turbostat --hide UncMHz
hides either UncMHz or UMHz*.*, if present
Reported-by: Artem Bityutskiy <artem.bityutskiy@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
Tested-by: Artem Bityutskiy <artem.bityutskiy@intel.com>
This CPU was mistakenly given the name INTEL_ATOM_AIRMONT_MID. But it
uses a Silvermont core, not Airmont.
Change #define name to INTEL_ATOM_SILVERMONT_MID2
Reported-by: Christian Ludloff <ludloff@gmail.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com>
Link: https://lore.kernel.org/all/20241007165701.19693-1-tony.luck%40intel.com
Summary of Changes since 2024.11.30:
Fix regression in 2023.11.07 that affinitized forked child
in one-shot mode.
Harden one-shot mode against hotplug online/offline
Enable RAPL SysWatt column by default.
Add initial PTL, CWF platform support.
Harden initial PMT code in response to early use.
Enable first built-in PMT counter: CWF c1e residency
Refuse to run on unsupported platforms without --force,
to encourage updating to a version that supports the system,
and to avoid no-so-useful measurement results.
Signed-off-by: Len Brown <len.brown@intel.com>
Intel Clearwater Forest report PMT telemetry with GUID 0x14421519, which
can be used to obtain module c1e residency counter of type tcore clock.
Add early support for the counter by using heuristic that should work
for the Clearwater Forest platforms.
Signed-off-by: Patryk Wlazlyn <patryk.wlazlyn@linux.intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
when turbostat interval mode can't migrate to a CPU, it complains,
prints no data, re-initializes with the new CPU configuration
and starts a new interval.
But this strategy in the face of a CPU hotplug offline during an interval
doesn't help in one-shot mode. When the missing CPU is discovered
at the end of the interval, the forked program has already returned
and there is nothing left for a new interval to measure.
So instead of aborting get_coutners() and delta_cpu() if a missing CPU
is detected, complain, but carry on and output what statistics are
actually present.
Use the same strategy for delta_cpu when aperf:mperf are observed
to have been reset -- complain, but carry on and print data for
the CPUs that are still present.
Interval mode error handling is unchanged.
One-shot mode can now do this:
$ sudo chcpu -e 1 ; sudo ./turbostat --quiet --show PkgWatt,Busy%,CPU chcpu -d 1
CPU 1 enabled
CPU 1 disabled
get_counters: Could not migrate to CPU 1
./turbostat: Counter reset detected
0.036920 sec
CPU Busy% PkgWatt
- 0.00 10.00
0 99.73 10.00
1 0.00
2 91.53
3 16.83
Suggested-by: Zhang Rui <rui.zhang@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
In "one-shot" mode, turbostat
1. takes a counter snapshot
2. forks and waits for a child
3. takes the end counter snapshot and prints the result.
But turbostat counter snapshots currently use affinity to travel
around the system so that counter reads are "local", and this
affinity must be cleared between #1 and #2 above.
The offending commit removed that reset that allowed the child
to run on cpu_present_set.
Fix that issue, and improve upon the original by using
cpu_possible_set for the child. This allows the child
to also run on CPUs that hotplug online during its runtime.
Reported-by: Zhang Rui <rui.zhang@intel.com>
Fixes: 7bb3fe27ad ("tools/power/turbostat: Obey allowed CPUs during startup")
Signed-off-by: Len Brown <len.brown@intel.com>
Some PMT counters, for example module c1e residency on Intel Clearwater
Forest, are reported using tcore clock type.
Signed-off-by: Patryk Wlazlyn <patryk.wlazlyn@linux.intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
