The ACC (automatic C-state conversion) feature was available on Sky Lake and
Cascade Lake Xeons (SKX and CLX), but it is not available on Ice Lake and
Sapphire Rapids Xeons (ICX and SPR). Therefore, stop decoding it for ICX and
SPR.
Signed-off-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
Sapphire Rapids Xeon (SPR) supports 2 flavors of PC6 - PC6N (non-retention) and
PC6R (retention). Before this patch we used ICX package C-state limits, which
was wrong, because ICX has only one PC6 flavor. With this patch, we use SKX PC6
limits for SPR, because they are the same.
Signed-off-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
The 'automatic_cstate_conversion_probe()' function has a too long 'if'
statement, convert it to a 'switch' statement in order to improve code
readability a bit.
Signed-off-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
Before this patch, SPR platform was considered identical to ICX platform. This
patch separates SPR support from ICX.
This patch is a preparation for adding SPR-specific package C-state limits
support.
Signed-off-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Reviewed-by: Chen Yu <yu.c.chen@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
Intel Performance Hybrid processors have a 2nd MSR
describing the turbo limits enforced on the Ecores.
Note, TRL and Secondary-TRL are usually R/O information,
but on overclock-capable parts, they can be written.
Signed-off-by: Len Brown <len.brown@intel.com>
When CONFIG_INTEL_UNCORE_FREQ_CONTROL is effective,
(Linux 5.9 and later), print the current (and default)
min and max uncore frequency limits.
When that driver provides the current uncore frequency
(Linux 5.18 and later), print a UncMHz column
reflecting the current uncore frequency.
Note that UncMHz is an instantaneous sample, not an average.
eg.
$ sudo ./turbostat -S --show frequency
...
Uncore Frequency pkg0 die0: 800 - 3900 MHz (800 - 3900 MHz)
...
Avg_MHz Busy% Bzy_MHz TSC_MHz UncMHz
28 0.70 4049 3095 3900
Signed-off-by: Len Brown <len.brown@intel.com>
Currently if a fscanf fails then an early return leaks an open
file pointer. Fix this by fclosing the file before the return.
Detected using static analysis with cppcheck:
tools/power/x86/turbostat/turbostat.c:2039:3: error: Resource leak: fp [resourceLeak]
Fixes: eae97e053f ("tools/power turbostat: Support thermal throttle count print")
Signed-off-by: Colin Ian King <colin.i.king@gmail.com>
Acked-by: Chen Yu <yu.c.chen@intel.com>
Reviewed-by: Tom Rix <trix@redhat.com>
Signed-off-by: Len Brown <len.brown@intel.com>
Using strncmp for a single character comparison is overly complicated,
just use a simpler single character comparison instead. Also stops
static analyzers (such as cppcheck) from complaining about strncmp on
non-null terminated strings.
Signed-off-by: Colin Ian King <colin.i.king@gmail.com>
Signed-off-by: Len Brown <len.brown@intel.com>
It would be handy to have cmdline in turbostat output. For example,
according to the turbostat output, there are no C-states requested.
In this case the user is very curious if something like
intel_idle.max_cstate=0 was used, or may be idle=none too. It is
also curious whether things like intel_pstate=nohwp were used.
Print the boot command line accordingly:
turbostat version 21.05.04 - Len Brown <lenb@kernel.org>
Kernel command line: BOOT_IMAGE=/boot/vmlinuz-5.16.0+ root=UUID=
b42359ed-1e05-42eb-8757-6bf2a1c19070 ro quiet splash vt.handoff=7
Suggested-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Signed-off-by: Chen Yu <yu.c.chen@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
Pull turbostat changes for 5.19 from Len Brown:
"Chen Yu (1):
tools/power turbostat: Support thermal throttle count print
Dan Merillat (1):
tools/power turbostat: fix dump for AMD cpus
Len Brown (5):
tools/power turbostat: tweak --show and --hide capability
tools/power turbostat: fix ICX DRAM power numbers
tools/power turbostat: be more useful as non-root
tools/power turbostat: No build warnings with -Wextra
tools/power turbostat: version 2022.04.16
Sumeet Pawnikar (2):
tools/power turbostat: Add Power Limit4 support
tools/power turbostat: print power values upto three decimal
Zephaniah E. Loss-Cutler-Hull (2):
tools/power turbostat: Allow -e for all names.
tools/power turbostat: Allow printing header every N iterations"
* 'turbostat' of git://git.kernel.org/pub/scm/linux/kernel/git/lenb/linux:
tools/power turbostat: version 2022.04.16
tools/power turbostat: No build warnings with -Wextra
tools/power turbostat: be more useful as non-root
tools/power turbostat: fix ICX DRAM power numbers
tools/power turbostat: Support thermal throttle count print
tools/power turbostat: Allow printing header every N iterations
tools/power turbostat: Allow -e for all names.
tools/power turbostat: print power values upto three decimal
tools/power turbostat: Add Power Limit4 support
tools/power turbostat: fix dump for AMD cpus
tools/power turbostat: tweak --show and --hide capability
Don't exit if used this way:
sudo setcap cap_sys_nice,cap_sys_rawio=+ep ./turbostat
sudo chmod +r /dev/cpu/*/msr
./turbostat
note: cap_sys_admin is now also needed for the perf IPC counter:
sudo setcap cap_sys_admin,cap_sys_nice,cap_sys_rawio=+ep ./turbostat
Reported-by: Artem S. Tashkinov <aros@gmx.com>
Reported-by: Toby Broom <tbroom@outlook.com>
Signed-off-by: Len Brown <len.brown@intel.com>
ICX (and its duplicates) require special hard-coded DRAM RAPL units,
rather than using the generic RAPL energy units.
Reported-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
The turbostat data is collected by end user for power evaluationit. However
it looks like we are missing enough thermal context there. Already a couple of
time we found that power management developer asking something like this:
grep -r . /sys/devices/system/cpu/cpu*/thermal_throttle/*
Print the per core thermal throttle count so as to get suffificent thermal
context.
turbostat -i 5 -s Core,CPU,CoreThr
Core CPU CoreThr
- - 104
0 0 61
0 4
1 1 0
1 5
2 2 104
2 6
3 3 7
3 7
Suggested-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Signed-off-by: Chen Yu <yu.c.chen@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
This gives the ability to reprint the header every N iterations, so you
can ensure that a scrolling display always has the header visible
somewhere on the screen.
Signed-off-by: Zephaniah E. Loss-Cutler-Hull <zephaniah@gmail.com>
Signed-off-by: Len Brown <len.brown@intel.com>
Currently, there are a number of variables which are displayed by
default, enabled with -e all, and listed by --list, but which you can
not give to --enable/-e.
So you can enable CPU0c1 (in the bic array), but you can't enable C1 or
C1% (not in the bic array, but exists in sysfs).
This runs counter to both the documentation and user expectations, and
it's just not very user friendly.
As such, the mechanism used by --hide has been duplicated, and is now
also used by --enable, so we can handle unknown names gracefully.
Note: One impact of this is that truly unknown fields given to --enable
will no longer generate errors, they will be silently ignored, as --hide
does.
Signed-off-by: Zephaniah E. Loss-Cutler-Hull <zephaniah@gmail.com>
Signed-off-by: Len Brown <len.brown@intel.com>
Print power values upto three decimal places in watts.
Suggested-by: Tony Luck <tony.luck@intel.com>
Signed-off-by: Sumeet Pawnikar <sumeet.r.pawnikar@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
turbostat --Dump exits early with status 243 (-13)
get_counters() calls get_msr_sum() on zen CPUS
for MSR_PKG_ENERGY_STAT, but per_cpu_msr_sum
has not been initialized.
Signed-off-by: Dan Merillat <git@dan.eginity.com>
Signed-off-by: Len Brown <len.brown@intel.com>
'MSR_PKG_CST_CONFIG_CONTROL' encodes the deepest allowed package C-state limit,
and turbostat decodes it.
Before this patch: turbostat does not recognize value "3" on Ice Lake Xeon
(ICX) and Sapphire Rapids Xeon (SPR), treats it as "unknown", and does not
display any package C-states in the results table.
After this patch: turbostat recognizes value 3 on ICX and SPR, treats it as
"PC6", and correctly displays package C-states in the results table.
Signed-off-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
As idle, in particular, can have many columns on some machines...
Make it easy to ignore them all at once.
Signed-off-by: Len Brown <len.brown@intel.com>
There are two TCC activation temeprature.
One is the default TCC activation temperature, also known as TJ_MAX.
Another one is the effective TCC activation temperature, which is the
subtraction of default TCC activation temperature and TCC offset.
The name of variable tcc_activation_temp might be misleading here.
Thus rename tcc_activation_temp to tj_max, and use tcc_default and
tcc_offset to calculate the effective TCC activation temperature.
No functional change in this patch.
Signed-off-by: Zhang Rui <rui.zhang@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
The length of TCC Offset bits varies on different platforms.
Decode TCC Offset bits only for the platforms that we have verified.
For the others, only show default TCC activation temperature.
Signed-off-by: Zhang Rui <rui.zhang@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
CPU model may get changed in intel_model_duplicates() for code reuse.
But there are still some cases we need the original CPU model to handle
minor differences between generations.
Thus save the original CPU model.
Signed-off-by: Zhang Rui <rui.zhang@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
For Atom CPUs that have core cstate deeper than C6,
MSR_CORE_C6_RESIDENCY actually returns the residency for both CC6 and
deeper Core cstates.
Thus, the real Core C6 residency should be the subtraction of
MSR_CORE_C6_RESIDENCY return value and MSR_CORE_C6_RESIDENCY return value.
Signed-off-by: Zhang Rui <rui.zhang@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
C-state pre-wake setting[1] is an optimization for some Intel CPUs to
be woken up from deep C-states in order to reduce latency. According to
the spec, the BIT30 is the C-state Pre-wake Disable. Expose this setting
accordingly.
Sample output from turbostat:
...
cpu51: MSR_IA32_POWER_CTL: 0x1a00a40059 (C1E auto-promotion: DISabled)
C-state Pre-wake: ENabled
cpu51: MSR_TURBO_RATIO_LIMIT: 0x2021212121212224
...
[1] https://intel.github.io/wult/#c-state-pre-wake
Signed-off-by: Chen Yu <yu.c.chen@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
It was found that on Elkhart Lake the TSC frequency is driven by
a separate crystal-clock domain, which is different from the
BCLK domain which includes mperf. This has result in small different
speed thus inconsistence between TSC and the mperf, which caused the
Busy% to be higher than 100%. On this platform it seems that the mperf
runs faster than tsc when the CPU is 100% utilized:
delta tsc(18815473183) < delta mperf(18958403680) for 10 seconds.
To align TSC with mperf, leverage the tsc_tweak mechanism introduced for
cores newer than Skylake, so that TSC and mperf would be calculated in
the same domain.
Reported-by: Zhang Rui <rui.zhang@intel.com>
Signed-off-by: Chen Yu <yu.c.chen@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
Do not mark a comment as kernel-doc notation when it is not
meant to be in kernel-doc notation.
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: Len Brown <len.brown@intel.com>
Currently the turbostat treats ICX the same way as SKX and shares the
code among them. But one difference is that ICX does not support Package
C6 Retention, unlike SKX and CLX.
So this patch:
1. Splitting SKX and ICX in turbostat.
2. Removing Package C6 Rentention for ICX.
And after this split, it would be easier to cutomize Ice Lake Server
in turbostat in the future.
Suggested-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Signed-off-by: Chen Yu <yu.c.chen@intel.com>
Reviewed-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Tested-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
The idx_to_offset() function returns type int (32-bit signed), but
MSR_PKG_ENERGY_STAT is u32 and would be interpreted as a negative number.
The end result is that it hits the if (offset < 0) check in update_msr_sum()
which prevents the timer callback from updating the stat in the background when
long durations are used. The similar issue exists in offset_to_idx() and
update_msr_sum(). Fix this issue by converting the 'int' to 'off_t' accordingly.
Fixes: 9972d5d84d ("tools/power turbostat: Enable accumulate RAPL display")
Signed-off-by: Calvin Walton <calvin.walton@kepstin.ca>
Signed-off-by: Len Brown <len.brown@intel.com>
It was reported that on Zen+ system turbostat started exiting,
which was tracked down to the MSR_PKG_ENERGY_STAT read failing because
offset_to_idx wasn't returning a non-negative index.
This patch combined the modification from Bingsong Si and
Bas Nieuwenhuizen and addd the MSR to the index system as alternative for
MSR_PKG_ENERGY_STATUS.
Fixes: 9972d5d84d ("tools/power turbostat: Enable accumulate RAPL display")
Reported-by: youling257 <youling257@gmail.com>
Tested-by: youling257 <youling257@gmail.com>
Tested-by: Kurt Garloff <kurt@garloff.de>
Tested-by: Bingsong Si <owen.si@ucloud.cn>
Tested-by: Artem S. Tashkinov <aros@gmx.com>
Co-developed-by: Bingsong Si <owen.si@ucloud.cn>
Co-developed-by: Terry Bowman <terry.bowman@amd.com>
Signed-off-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Chen Yu <yu.c.chen@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
This reverts commit 6ff7cb371c.
Apparently the TCC offset should not be used to adjust what temperature
we show the user after all.
(on most systems, TCC offset is 0, FWIW)
Fixes: 6ff7cb371c
Signed-off-by: Len Brown <len.brown@intel.com>
Ice Lake D is low-end server version of Ice Lake X, reuse
the code accordingly.
Tested-by: Wendy Wang <wendy.wang@intel.com>
Signed-off-by: Chen Yu <yu.c.chen@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
Share the code between Alder Lake Mobile and Alder Lake Desktop.
Signed-off-by: Chen Yu <yu.c.chen@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
Use linux-perf to access the hardware instructions-retired counter.
This is necessary because the counter is not enabled by default,
and also the counter is prone to roll-over -- both of which
perf manages.
It is not necessary to use perf for the cycle counter,
because turbostat already needs to collect delta-aperf
to calcuate frequency.
Signed-off-by: Len Brown <len.brown@intel.com>
