Commit a729617c authored by Len Brown's avatar Len Brown

tools/power turbostat: relax dependency on APERF_MSR

While turbostat is significantly less useful on systems
with no APERF_MSR, it seems more friendly
to run on such systems and report what we can,
rather than refusing to run.

Update man page to reflect recent changes.
Signed-off-by: default avatarLen Brown <len.brown@intel.com>
parent d7899447
......@@ -12,16 +12,16 @@ turbostat \- Report processor frequency and idle statistics
.RB [ "\-i interval_sec" ]
.SH DESCRIPTION
\fBturbostat \fP reports processor topology, frequency,
idle power-state statistics, temperature and power on modern X86 processors.
Either \fBcommand\fP is forked and statistics are printed
upon its completion, or statistics are printed periodically.
\fBturbostat \fP
must be run on root, and
minimally requires that the processor
supports an "invariant" TSC, plus the APERF and MPERF MSRs.
Additional information is reported depending on hardware counter support.
idle power-state statistics, temperature and power on X86 processors.
There are two ways to invoke turbostat.
The first method is to supply a
\fBcommand\fP, which is forked and statistics are printed
upon its completion.
The second method is to omit the command,
and turbodstat will print statistics every 5 seconds.
The 5-second interval can changed using the -i option.
Some information is not availalbe on older processors.
.SS Options
The \fB-p\fP option limits output to the 1st thread in 1st core of each package.
.PP
......@@ -130,12 +130,13 @@ cpu3: MSR_IA32_THERM_STATUS: 0x884e0000 (27 C +/- 1)
...
.fi
The \fBmax efficiency\fP frequency, a.k.a. Low Frequency Mode, is the frequency
available at the minimum package voltage. The \fBTSC frequency\fP is the nominal
maximum frequency of the processor if turbo-mode were not available. This frequency
available at the minimum package voltage. The \fBTSC frequency\fP is the base
frequency of the processor -- this should match the brand string
in /proc/cpuinfo. This base frequency
should be sustainable on all CPUs indefinitely, given nominal power and cooling.
The remaining rows show what maximum turbo frequency is possible
depending on the number of idle cores. Note that this information is
not available on all processors.
depending on the number of idle cores. Note that not all information is
available on all processors.
.SH FORK EXAMPLE
If turbostat is invoked with a command, it will fork that command
and output the statistics gathered when the command exits.
......@@ -176,6 +177,11 @@ not including any non-busy idle time.
.B "turbostat "
must be run as root.
Alternatively, non-root users can be enabled to run turbostat this way:
# setcap cap_sys_rawio=ep ./turbostat
# chmod +r /dev/cpu/*/msr
.B "turbostat "
reads hardware counters, but doesn't write them.
......@@ -184,15 +190,33 @@ multiple invocations of itself.
\fBturbostat \fP
may work poorly on Linux-2.6.20 through 2.6.29,
as \fBacpi-cpufreq \fPperiodically cleared the APERF and MPERF
as \fBacpi-cpufreq \fPperiodically cleared the APERF and MPERF MSRs
in those kernels.
If the TSC column does not make sense, then
the other numbers will also make no sense.
Turbostat is lightweight, and its data collection is not atomic.
These issues are usually caused by an extremely short measurement
interval (much less than 1 second), or system activity that prevents
turbostat from being able to run on all CPUS to quickly collect data.
AVG_MHz = APERF_delta/measurement_interval. This is the actual
number of elapsed cycles divided by the entire sample interval --
including idle time. Note that this calculation is resiliant
to systems lacking a non-stop TSC.
TSC_MHz = TSC_delta/measurement_interval.
On a system with an invariant TSC, this value will be constant
and will closely match the base frequency value shown
in the brand string in /proc/cpuinfo. On a system where
the TSC stops in idle, TSC_MHz will drop
below the processor's base frequency.
%Busy = MPERF_delta/TSC_delta
Bzy_MHz = TSC_delta/APERF_delta/MPERF_delta/measurement_interval
Note that these calculations depend on TSC_delta, so they
are not reliable during intervals when TSC_MHz is not running at the base frequency.
Turbostat data collection is not atomic.
Extremely short measurement intervals (much less than 1 second),
or system activity that prevents turbostat from being able
to run on all CPUS to quickly collect data, will result in
inconsistent results.
The APERF, MPERF MSRs are defined to count non-halted cycles.
Although it is not guaranteed by the architecture, turbostat assumes
......
......@@ -673,6 +673,7 @@ delta_thread(struct thread_data *new, struct thread_data *old,
old->c1 = new->c1 - old->c1;
if (has_aperf) {
if ((new->aperf > old->aperf) && (new->mperf > old->mperf)) {
old->aperf = new->aperf - old->aperf;
old->mperf = new->mperf - old->mperf;
......@@ -692,6 +693,7 @@ delta_thread(struct thread_data *new, struct thread_data *old,
skip_c0 = 1;
skip_c1 = 1;
}
}
if (use_c1_residency_msr) {
......@@ -2244,14 +2246,11 @@ void check_cpuid()
has_epb = ecx & (1 << 3);
if (verbose)
fprintf(stderr, "CPUID(6): %s%s%s%s\n",
has_aperf ? "APERF" : "No APERF!",
do_dts ? ", DTS" : "",
do_ptm ? ", PTM": "",
has_epb ? ", EPB": "");
if (!has_aperf)
errx(-1, "No APERF");
fprintf(stderr, "CPUID(6): %sAPERF, %sDTS, %sPTM, %sEPB\n",
has_aperf ? "" : "No ",
do_dts ? "" : "No ",
do_ptm ? "" : "No ",
has_epb ? "" : "No ");
do_nhm_platform_info = do_nhm_cstates = do_smi = has_nhm_msrs(family, model);
do_snb_cstates = has_snb_msrs(family, model);
......@@ -2632,7 +2631,7 @@ int main(int argc, char **argv)
cmdline(argc, argv);
if (verbose)
fprintf(stderr, "turbostat v3.8 14-Aug 2014"
fprintf(stderr, "turbostat v3.9 23-Jan, 2015"
" - Len Brown <lenb@kernel.org>\n");
turbostat_init();
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment