Turbostat can be useful on systems that do not support invariant TSC,
so allow it to run on those systgems.
All arithmetic in turbostat using the TSC value is per-processsor,
so it does not depend on the TSC values being in sync acrosss
processors.
Turbostat uses cdtime() for the measurement interval
rather than using the TSC directly, so that key metric
is also immune from variable TSC.
Turbostat has a TSC sanity check gauge:
TSC_MHz = TSC_delta/interval
If this column is constant and is close to the processor
base frequency, then the TSC is behaving properly.
The other key turbostat columns are calculated this way:
Avg_Mhz = APERF_delta/interval
Bzy_MHz = TSC_delta/APERF_delta/MPERF_delta/interval
c0 = MPERF_delta/TSC_delta
This adaptation of the original commit has not been tested on any
old CPU. This patch also re-introduce the boolean "do_smi" as such
old CPUs don't have SMI.
This new gauge contains TSC / interval_float.
If TSC and interval_float are behaving correctly, this is supposed to be
constant. Otherwise, other measurements can be wrong
Jean Delvare [Thu, 1 May 2014 09:40:19 +0000 (11:40 +0200)]
turbostat: Backport 'Drop temperature checks'
The Intel 64 and IA-32 Architectures Software Developer's Manual says
that TjMax is stored in bits 23:16 of MSR_TEMPERATURE TARGET (0x1a2).
That's 8 bits, not 7, so it must be masked with 0xFF rather than 0x7F.
The manual has no mention of which values should be considered valid,
which kind of implies that they all are. Arbitrarily discarding values
outside a specific range is wrong. The upper range check had to be
fixed recently (commit 144b44b1) and the lower range check is just as
wrong. See bug #75071:
https://bugzilla.kernel.org/show_bug.cgi?id=75071
There are many Xeon processor series with TjMax of 70, 71 or 80
degrees Celsius, way below the arbitrary 85 degrees Celsius limit.
There may be other (past or future) models with even lower limits.
So drop this arbitrary check. The only value that would be clearly
invalid is 0. Everything else should be accepted.
After these changes, turbostat is aligned with what the coretemp
driver does.
Signed-off-by: Jean Delvare <jdelvare@suse.de> Cc: Len Brown <len.brown@intel.com> Acked-by: Guenter Roeck <linux@roeck-us.net> Reviewed-by: Josh Triplett <josh@joshtriplett.org> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
[git@lerya.net: Ported to collectd turbostat plugin] Signed-off-by: Vincent Brillault <git@lerya.net>
get_msr doesn't need to be run on the CPU targetted. However, when run
on another CPU, perfomances will decrease as task have to be scheduled
on another CPU. Migrating to another CPU during the setup (where only a
few get_msr are done) is probably not worth it. However, in the
get_counters function, the large number of get_msr query makes it
potentially important.
This patch also introduce the restoration of the scheduling affinity as
it was before the plugin invocation.
In two out of three "get_msr" calls, there is a CPU migration, which
does not seem to be used by something else (in one case, it was used by
the rdtsc call).
Move the migrate call directly into the open_msr function
Reorganization and rewriting of the cpu-probing code:
- Only support genuine_intel CPUs
- Remove variables that are always expected to be true
- Make supported pkg/core cstates more comprehensive
- Simplify rafl code (remove unused warning feature: intervals too long)
Nicolas Iooss [Mon, 23 Jun 2014 16:08:15 +0000 (18:08 +0200)]
Turbostat: make DO_OR_GOTO_ERR look like a function
Signed-off-by: Nicolas Iooss <nicolas.iooss_git@polytechnique.org>
[git@lerya.net: Ported from other git, changed commit name] Signed-off-by: Vincent Brillault <git@lerya.net>
- Remove horrible STATIC_MUST_CHECK macro
- Define __must_check __attribute__((warn_unused_result))
- Refactor function declaration: type only on first line
Eric Sandeen [Wed, 28 Jan 2015 04:36:05 +0000 (22:36 -0600)]
modbus: rename register types once more
As I was looking over my config files again, ModbusRegisterType
kind of stuck out like a sore thumb.
Modbus does talk about "read holding" and "read input" as "commands";
perhaps renaming it as this patch does is more intuitive, so that all
of the Modbus register configurations start with Register.
I'm not wedded to it, but if you like it & agree, here you go.
Vincent Bernat [Fri, 23 Jan 2015 17:55:04 +0000 (18:55 +0100)]
ipmi: use SIGIO instead of SIGUSR2 to avoid conflict with JVM
Java uses SIGUSR2 to suspend/resume threads. The OpenIPMI plugins also
need a signal to resume its event loop when setting a timer. They can't
both use the same signal. We ask OpenIPMI to use SIGIO instead.
Vincent Bernat [Fri, 23 Jan 2015 17:55:04 +0000 (18:55 +0100)]
ipmi: use SIGIO instead of SIGUSR2 to avoid conflict with JVM
Java uses SIGUSR2 to suspend/resume threads. The OpenIPMI plugins also
need a signal to resume its event loop when setting a timer. They can't
both use the same signal. We ask OpenIPMI to use SIGIO instead.
Vincent Bernat [Fri, 23 Jan 2015 10:57:03 +0000 (11:57 +0100)]
libstatgrab: fix detection of libstatgrab >= 0.90
Ensure linking is done after compilation of the test program, otherwise
symbols cannot be found. `LDFLAGS` and `CFLAGS` are placed before the
test program on gcc command line while `LIBS` is placed after. Even if
in our case, `LDFLAGS` also contain the lib, being put before the
program while the linking phase is done with the same command leads to
`sg_init()` never found which is misinterpreted as requiring an
argument.
Marc Fournier [Sat, 17 Jan 2015 15:10:46 +0000 (16:10 +0100)]
write_http: make callback names context-dependent
This allows multiple destinations to work again (fixes #821), using the
same logic as other write plugins.
The callback name would now be something like:
`write_http/http://example.com/endpoint`
which is not very nice.
The next step would be to change this plugin to use `<Node>` blocks
like many others, and pass the URL as a parameter inside each instance
block. So I see this patch as the minimum required to let 5.3 and 5.4
users use this plugin with multiple destinations.
Marc Fournier [Sat, 17 Jan 2015 15:10:46 +0000 (16:10 +0100)]
write_http: make callback names context-dependent
This allows multiple destinations to work again (fixes #821), using the
same logic as other write plugins.
The callback name would now be something like:
`write_http/http://example.com/endpoint`
which is not very nice.
The next step would be to change this plugin to use `<Node>` blocks
like many others, and pass the URL as a parameter inside each instance
block. So I see this patch as the minimum required to let 5.3 and 5.4
users use this plugin with multiple destinations.
projects / thirdparty / collectd.git / log
