There is clearly wrong output when mperf monitor runs in MAX_FREQ_SYSFS mode:
average frequency shows in kHz unit (despite the intended output to be in MHz),
and percentages for C state information are all wrong (including high/negative
values shown).
The problem is that the max_frequency read on initialization isn't used where it
should have been used on mperf_get_count_percent (to estimate the number of
ticks in the given time period), and the value we read from sysfs is in kHz, so
we must divide it to get the MHz value to use in current calculations.
While at it, also I fixed another small issues in the debug output of
max_frequency value in mperf_get_count_freq.
Signed-off-by: Herton R. Krzesinski <herton@redhat.com>
Acked-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
libpci 3.3.0 introduced an additional member in the pci_filter struct
which needs to be initialized to -1 to get the same behavior as before
the API change. The libpci internal helpers got updated accordingly,
but as the cpupower pci helpers initialized the struct themselves the
behavior changed.
Use the libpci helper pci_filter_init() to fix this and guard against
similar breakages in the future.
This fixes probing of the AMD fam12h/14h cpuidle monitor on systems
with libpci >= 3.3.0.
Signed-off-by: Lucas Stach <dev@lynxeye.de>
Acked-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
sysfs_get_idlestate_count() returns an unsigned int. Returning -ENODEV
is not the right thing to do here, and in any case is handled the same
way as if there are no states found.
Signed-off-by: Prarit Bhargava <prarit@redhat.com>
Acked-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Some operations, like frequency-set, need root privileges. However,
the way that this is detected is not correct. The getuid() is called,
while in fact geteuid() should be. This way we can allow
distributions or users to set SETUID flags on the cpupower binary if
they want to and let regular users change the cpu frequency governor.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Acked-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
This reverts commit 16b7c275c0.
My previous commit 16b7c275c0 ("tools: cpupower: fix return checks for
sysfs_get_idlestate_count()") was not correct. After looking
at the changelog for cpupower I noticed that Thomas had changed the return of
sysfs_get_idlestate_count() to an unsigned int to simplify the code. The
problem is really that both he (in his original change) and I (in my new
change) missed the obvious that sysfs_get_idlestate_count()
can't return -ENODEV. It should just return 0 for "no c-states".
Fixes: 16b7c275c0 (tools: cpupower: fix return checks for ...)
Signed-off-by: Prarit Bhargava <prarit@redhat.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Red Hat and Fedora use a bug reporting tool that gathers data about
"broken" systems called sosreport. Among other things, it includes the
output of 'cpupower idle-info'. Executing 'cpupower idle-info' on a
system that has cpuidle disabled via 'cpuidle.off=1' results in a 300
second hang in the cpupower application.
ie)
[root@intel-brickland-05]# cpupower idle-info
Could not determine cpuidle driver
Analyzing CPU 0:
Number of idle states: -19
[hang]
The problem is that the cpupower code only checks for a zero return from
sysfs_get_idlestate_count(). The function can return -ENODEV (-19) as
above. This patch fixes callers to sysfs_get_idlestate_count() to check
the right return values.
Signed-off-by: Prarit Bhargava <prarit@redhat.com>
Signed-off-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Remove double checks, and move the call to print_error to the
first check. Replace break by return, and return 0 on success.
The simplified version of the coccinelle semantic patch that
fixes this issue is as follows:
// <smpl>
@@
expression E; identifier pr; expression list es;
@@
for(...;...;...){
...
- if (E) break;
+ if (E){
+ pr(es);
+ break;
+ }
...
}
- if(E) pr(es);
// </smpl>
Signed-off-by: Peter Senna Tschudin <peter.senna@gmail.com>
Signed-off-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
In commit ae91d60ba8, a bug was fixed that
involved converting !x & y to !(x & y). The code below shows the same
pattern, and thus should perhaps be fixed in the same way.
The Coccinelle semantic patch that makes this change is as follows:
// <smpl>
@@ expression E1,E2; @@
(
!E1 & !E2
|
- !E1 & E2
+ !(E1 & E2)
)
// </smpl>
Signed-off-by: Himangi Saraogi <himangi774@gmail.com>
Signed-off-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Link: https://bugzilla.kernel.org/show_bug.cgi?id=80621
Reported-by: David Binderman <dcb314@hotmail.com>
Signed-off-by: Andrey Utkin <andrey.krieger.utkin@gmail.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
These kernel interfaces got removed by:
commit 8e7fbcbc22
Author: Peter Zijlstra <peterz@infradead.org>
Date: Mon Jan 9 11:28:35 2012 +0100
sched: Remove stale power aware scheduling remnants and dysfunctional knobs
No need to further keep them as userspace configurations.
Signed-off-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The command "cpupower frequency-info" can be used when using cpupower to
monitor and test processor behaviour to determine if the processor is
behaving as expected. This data can be compared to the output of
/proc/cpuinfo or the output of
/sys/devices/system/cpu/cpuX/cpufreq/scaling_available_frequencies
to determine if the cpu is in an expected state.
When doing this I noticed comparison test failures due to the way the
data is displayed in cpupower. For example,
[root@intel-s3e37-02 cpupower]# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies
2262000 2261000 2128000 1995000 1862000 1729000 1596000 1463000 1330000
1197000 1064000
compared to
[root@intel-s3e37-02 cpupower]# cpupower frequency-info
analyzing CPU 0:
driver: acpi-cpufreq
CPUs which run at the same hardware frequency: 0
CPUs which need to have their frequency coordinated by software: 0
maximum transition latency: 10.0 us.
hardware limits: 1.06 GHz - 2.26 GHz
available frequency steps: 2.26 GHz, 2.26 GHz, 2.13 GHz, 2.00 GHz, 1.86 GHz, 1.73 GHz, 1.60 GHz, 1.46 GHz, 1.33 GHz, 1.20 GHz, 1.06 GHz
available cpufreq governors: conservative, userspace, powersave, ondemand, performance
current policy: frequency should be within 1.06 GHz and 2.26 GHz.
The governor "performance" may decide which speed to use
within this range.
current CPU frequency is 2.26 GHz (asserted by call to hardware).
boost state support:
Supported: yes
Active: yes
shows very different values for the available frequency steps. The cpupower
output rounds off values at 2 decimal points and this causes problems with
test scripts. For example, with the data above,
1.064 is 1.06
1.197 is 1.20
1.596 is 1.60
1.995 is 2.00
2.128 is 2.13
and most confusingly,
2.261 is 2.26
2.262 is 2.26
Truncating these values serves no real purpose other than making the output
pretty. Since the default has been to round off these values I am adding
a -n/--no-rounding option to the cpupower utility that will display the
data without rounding off the still significant digits.
After patch,
analyzing CPU 0:
driver: acpi-cpufreq
CPUs which run at the same hardware frequency: 0
CPUs which need to have their frequency coordinated by software: 0
maximum transition latency: 10.000 us.
hardware limits: 1.064000 GHz - 2.262000 GHz
available frequency steps: 2.262000 GHz, 2.261000 GHz, 2.128000 GHz, 1.995000 GHz, 1.862000 GHz, 1.729000 GHz, 1.596000 GHz, 1.463000 GHz, 1.330000 GHz, 1.197000 GHz, 1.064000 GHz
available cpufreq governors: conservative, userspace, powersave, ondemand, performance
current policy: frequency should be within 1.064000 GHz and 2.262000 GHz.
The governor "performance" may decide which speed to use
within this range.
current CPU frequency is 2.262000 GHz (asserted by call to hardware).
boost state support:
Supported: yes
Active: yes
Acked-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Prarit Bhargava <prarit@redhat.com>
[rjw: Subject]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The cpufreq-set tool has a missing length check. This is basically
just correctness but still should get fixed.
One of a set of sscanf problems reported by Jackie Chang
Signed-off-by: Alan Cox <alan@linux.intel.com>
[rjw: Subject]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
If a user calls 'cpupower set --perf-bias 15', the process will end with
a SIGSEGV in libc because cpupower-set passes a NULL optarg to the atoi
call. This is because the getopt_long structure currently has all of
the options as having an optional_argument when they really have a
required argument. We change the structure to use required_argument to
match the short options and it resolves the issue.
This fixes https://bugzilla.redhat.com/show_bug.cgi?id=1000439
Signed-off-by: Josh Boyer <jwboyer@fedoraproject.org>
Cc: Dominik Brodowski <linux@dominikbrodowski.net>
Cc: Thomas Renninger <trenn@suse.de>
Cc: stable@vger.kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
idlestates in sysfs are counted from 0.
This fixes a wrong error message.
Current behavior on a machine with 4 sleep states is:
cpupower idle-set -e 4
Idlestate 4 enabled on CPU 0
-----Wrong---------------------
cpupower idle-set -e 5
Idlestate enabling not supported by kernel
-----Must and now will be -----
cpupower idle-set -e 5
Idlestate 6 not available on CPU 0
-------------------------------
cpupower idle-set -e 6
Idlestate 6 not available on CPU 0
Signed-off-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
This specific processor supports 3 new package sleep states.
Provide a monitor, so that the user can see their usage.
Signed-off-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Add Haswell model numbers to snb_register() as it also supports the
C-states introduced in SandyBridge processors.
[rjw: Changelog]
Signed-off-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Example:
cpupower idle-set -d 3
will disable C-state 3 on all processors (set commands are active on
all CPUs by default), same as:
cpupower -c all idle-set -d 3
Signed-off-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Latest kernel allows to disable C-states via:
/sys/devices/system/cpu/cpuX/cpuidle/stateY/disable
This patch provides lower level sysfs access functions to make use of
this interface. A later patch will implement the higher level stuff.
Signed-off-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Use unsigned int as the data type for some variables related to CPU
idle states which allows the code to be simplified slightly.
[rjw: Changelog]
Signed-off-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
If an MSR based monitor is run in parallel this is not needed. This is the
default case on all/most Intel machines.
But when only sysfs info is read via cpupower monitor -m Idle_Stats (typically
the case for non root users) or when other monitors are PCI based (AMD),
Idle_Stats, read from sysfs can be totally bogus:
cpupower monitor -m Idle_Stats
PKG |CORE|CPU | POLL | C1-N | C3-N | C6-N
0| 0| 0| 0.00| 0.00| 0.24| 99.81
0| 0| 32| 0.00| 0.00| 0.00| 100.7
...
0| 17| 20| 0.00| 0.00| 0.00| 173.1
0| 17| 52| 0.00| 0.00| 0.07| 173.0
0| 18| 68| 0.00| 0.00| 0.00| 0.00
0| 18| 76| 0.00| 0.00| 0.00| 0.00
...
With the -c option all cores are woken up and the kernel
did update cpuidle statistics before reading out sysfs.
This causes some overhead. Therefore avoid if possible, use
if needed:
cpupower monitor -c -m Idle_Stats
PKG |CORE|CPU | POLL | C1-N | C3-N | C6-N
0| 0| 0| 0.00| 0.00| 0.00| 100.2
0| 0| 32| 0.00| 0.00| 0.00| 100.2
...
0| 8| 8| 0.00| 0.00| 0.00| 99.82
0| 8| 40| 0.00| 0.00| 0.00| 99.81
0| 9| 24| 0.00| 0.00| 0.00| 100.3
0| 9| 56| 0.00| 0.00| 0.00| 100.2
0| 16| 4| 0.00| 0.00| 0.00| 99.75
0| 16| 36| 0.00| 0.00| 0.00| 99.38
...
Signed-off-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The pkgs member of cpupower_topology is being used as the number of
cpu packages. As the comment in get_cpu_topology notes, the package ids
are not guaranteed to be contiguous. So, simply setting pkgs to the value
of the highest physical_package_id doesn't actually provide a count of
the number of cpu packages. Instead, calculate pkgs by setting it to
the number of distinct physical_packge_id values which is pretty easy
to do after the core_info structs are sorted. Calculating pkgs this
way also has the nice benefit of getting rid of a sign comparison warning
that GCC 4.6 was reporting.
Signed-off-by: Palmer Cox <p@lmercox.com>
Signed-off-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The cpu_info member of cpupower_topology was being declared as an unnamed
structure. This member was then being malloced using the size of the
parent cpupower_topology * the number of cpus. This works
because cpu_info is smaller than cpupower_topology. However, there is
no guarantee that will always be the case. Making cpu_info its own
top level structure (named cpuid_core_info) allows for mallocing the actual
size of this structure. This also lets us get rid of a redefinition of
the structure in topology.c with slightly different field names.
Signed-off-by: Palmer Cox <p@lmercox.com>
Signed-off-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Fix a variety of issues with sysfs_topology_read_file:
* The return value of sysfs_topology_read_file function was not properly
being checked for failure.
* The function was reading int valued sysfs variables and then returning
their value. So, even if a function was trying to check the return value
of this function, a caller would not be able to tell an failure code apart
from reading a negative value. This also conflicted with the comment on the
function which said that a return value of 0 indicated success.
* The function was parsing int valued sysfs values with strtoul instead of
strtol.
* The function was non-static even though it was only used in the
file it was declared in.
Signed-off-by: Palmer Cox <p@lmercox.com>
Signed-off-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Fix minor warnings reported with GCC 4.6:
* The sysfs_write_file function is unused - remove it.
* The pr_mon_len in the print_header function is unsed - remove it.
Signed-off-by: Palmer Cox <p@lmercox.com>
Signed-off-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
It's been broken forever (i.e. it's not scheduling in a power
aware fashion), as reported by Suresh and others sending
patches, and nobody cares enough to fix it properly ...
so remove it to make space free for something better.
There's various problems with the code as it stands today, first
and foremost the user interface which is bound to topology
levels and has multiple values per level. This results in a
state explosion which the administrator or distro needs to
master and almost nobody does.
Furthermore large configuration state spaces aren't good, it
means the thing doesn't just work right because it's either
under so many impossibe to meet constraints, or even if
there's an achievable state workloads have to be aware of
it precisely and can never meet it for dynamic workloads.
So pushing this kind of decision to user-space was a bad idea
even with a single knob - it's exponentially worse with knobs
on every node of the topology.
There is a proposal to replace the user interface with a single
3 state knob:
sched_balance_policy := { performance, power, auto }
where 'auto' would be the preferred default which looks at things
like Battery/AC mode and possible cpufreq state or whatever the hw
exposes to show us power use expectations - but there's been no
progress on it in the past many months.
Aside from that, the actual implementation of the various knobs
is known to be broken. There have been sporadic attempts at
fixing things but these always stop short of reaching a mergable
state.
Therefore this wholesale removal with the hopes of spurring
people who care to come forward once again and work on a
coherent replacement.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Suresh Siddha <suresh.b.siddha@intel.com>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Cc: Vincent Guittot <vincent.guittot@linaro.org>
Cc: Vaidyanathan Srinivasan <svaidy@linux.vnet.ibm.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Link: http://lkml.kernel.org/r/1326104915.2442.53.camel@twins
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Looks like some not needed debug code slipped in.
Also this code:
tmp = sysfs_get_idlestate_name(cpu, idlestates - 1);
performs a strdup and the mem was not freed again.
-> delete it.
Signed-off-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net>
The number of idle states was wrong.
The POLL idle state (on X86) was missed out:
Number of idle states: 4
Available idle states: C1-NHM C3-NHM C6-NHM
While the POLL is not a real idle state, its
statistics should still be shown. It's now also
explained in a detailed manpage.
This should fix a bug of missing the first idle
state on other archs.
Signed-off-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net>
The name of the monitor is updated at runtime to the name of the
CPU type.
Signed-off-by: Thomas Renninger <trenn@suse.de>
CC: Andreas Herrmann <herrmann.der.user@googlemail.com>
Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net>
AMD's BKDG (Bios and Kernel Developers Guide) talks in the CPU spec of their
CPU families about PCI registers defined by "device" (slot) and func(tion).
Assuming that CPU specific configuration PCI devices are always on domain
and bus zero a pci_slot_func_init() func which gets the slot and func of
the desired PCI device passed looks like the most convenient way.
This also obsoletes the PCI device id maintenance.
Signed-off-by: Thomas Renninger <trenn@suse.de>
CC: Andreas Herrmann <herrmann.der.user@googlemail.com>
Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net>
Instead of printing something non-formatted to stdout, call
man(1) to show the man page for the proper subcommand.
Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net>
Loosely based on a patch for cpufrequtils, submittted by
Sergey Dryabzhinsky <sergey.dryabzhinsky@gmail.com> and
signed-off-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net>
This allows for example:
cpupower -c 2-4,6 monitor -m Mperf
|Mperf
PKG |CORE|CPU | C0 | Cx | Freq
0| 8| 4| 2.42| 97.58| 1353
0| 16| 2| 14.38| 85.62| 1928
0| 24| 6| 1.76| 98.24| 1442
1| 16| 3| 15.53| 84.47| 1650
CPUs always get resorted for package, core then cpu id if it could get read out
(or however you name these topology levels...).
Still this is a nice way to keep the overview if a test binary is bound to
a specific CPU or if one wants to show all CPUs inside a package or similar.
Still missing: Do not measure not available cores to reduce the overhead
and achieve better results.
Signed-off-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net>
Before, checking for offlined CPUs was done dirty and
it was checked whether topology parsing returned -1 values.
But this is a valid case on a Xen (and possibly other) kernels.
Do proper online/offline checking, also take CONFIG_HOTPLUG_CPU
option into account (no /sys/devices/../cpuX/online file).
Signed-off-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net>
Which makes the implementation independent from cpufreq drivers.
Therefore this would also work on a Xen kernel where the hypervisor
is doing frequency switching and idle entering.
Signed-off-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net>
IA32-Intel Devel guide Volume 3A - 14.3.2.1
-------------------------------------------
...
Opportunistic processor performance operation can be disabled by setting bit 38 of
IA32_MISC_ENABLES. This mechanism is intended for BIOS only. If
IA32_MISC_ENABLES[38] is set, CPUID.06H:EAX[1] will return 0.
Better detect things via cpuid, this cleans up the code a bit
and the MSR parts were not working correctly anyway.
Signed-off-by: Thomas Renninger <trenn@suse.de>
CC: lenb@kernel.org
CC: linux@dominikbrodowski.net
CC: cpufreq@vger.kernel.org
Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net>
This adds the last piece missing from turbostat (if called with -v).
It shows on Intel machines supporting Turbo Boost how many cores
have to be active/idle to enter which boost mode (frequency).
Whether the HW really enters these boost modes can be verified via
./cpupower monitor.
Signed-off-by: Thomas Renninger <trenn@suse.de>
CC: lenb@kernel.org
CC: linux@dominikbrodowski.net
CC: cpufreq@vger.kernel.org
Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net>
As cpupowerutils is intended to be included into the kernel sources,
use the kernel versioning instead of a custom version.
The script utils/version-gen.sh is largely based on the script already
found in tools/perf/util/PERF-VERSION-GEN .
Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net>
CPU power consumption vs performance tuning is no longer
limited to CPU frequency switching anymore: deep sleep states,
traditional dynamic frequency scaling and hidden turbo/boost
frequencies are tied close together and depend on each other.
The first two exist on different architectures like PPC, Itanium and
ARM, the latter (so far) only on X86. On X86 the APU (CPU+GPU) will
only run most efficiently if CPU and GPU has proper power management
in place.
Users and Developers want to have *one* tool to get an overview what
their system supports and to monitor and debug CPU power management
in detail. The tool should compile and work on as many architectures
as possible.
Once this tool stabilizes a bit, it is intended to replace the
Intel-specific tools in tools/power/x86
Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net>
