Rework the handling of work items by dbs_update_util_handler() and
dbs_work_handler() so the former (which is executed in scheduler
paths) only uses atomic operations when absolutely necessary. That
is, when the policy is shared and dbs_update_util_handler() has
already decided that this is the time to queue up a work item.
In particular, this avoids the atomic ops entirely on platforms where
policy objects are never shared.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
The ondemand governor already updates sample_delay_ns immediately on
updates to the sampling rate, but conservative doesn't do that.
It was left out earlier as the code was really too complex to get
that done easily. Things are sorted out very well now, however, and
the conservative governor can be modified to follow ondemand in that
respect.
Moreover, since the code needed to implement that in the
conservative governor would be identical to the corresponding
ondemand governor's code, make that code common and change both
governors to use it.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Tested-by: Juri Lelli <juri.lelli@arm.com>
Tested-by: Shilpasri G Bhat <shilpa.bhat@linux.vnet.ibm.com>
[ rjw: Changelog ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
We used to drop policy->rwsem just before calling __cpufreq_governor()
in some cases earlier and so it was possible that __cpufreq_governor()
ran concurrently via separate threads for the same policy.
In order to guarantee valid state transitions for governors,
'governor_enabled' was required to be protected using some locking
and cpufreq_governor_lock was added for that.
But now __cpufreq_governor() is always called under policy->rwsem,
and 'governor_enabled' is protected against races even without
cpufreq_governor_lock.
Get rid of the extra lock now.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Tested-by: Juri Lelli <juri.lelli@arm.com>
Tested-by: Shilpasri G Bhat <shilpa.bhat@linux.vnet.ibm.com>
[ rjw : Changelog ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The dbs_data_mutex lock is currently used in two places. First,
cpufreq_governor_dbs() uses it to guarantee mutual exclusion between
invocations of governor operations from the core. Second, it is used by
ondemand governor's update_sampling_rate() to ensure the stability of
data structures walked by it.
The second usage is quite problematic, because update_sampling_rate() is
called from a governor sysfs attribute's ->store callback and that leads
to a deadlock scenario involving cpufreq_governor_exit() which runs
under dbs_data_mutex. Thus it is better to rework the code so
update_sampling_rate() doesn't need to acquire dbs_data_mutex.
To that end, rework update_sampling_rate() to walk a list of policy_dbs
objects supported by the dbs_data one it has been called for (instead of
walking cpu_dbs_info object for all CPUs). The list manipulation is
protected with dbs_data->mutex which also is held around the execution
of update_sampling_rate(), it is not necessary to hold dbs_data_mutex in
that function any more.
Reported-by: Juri Lelli <juri.lelli@arm.com>
Reported-by: Shilpasri G Bhat <shilpa.bhat@linux.vnet.ibm.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
[ rjw: Subject & changelog ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The previous commit introduced a new set of macros for creating sysfs
attributes that represent governor tunables and the old macros used for
this purpose are not needed any more, so drop them.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Tested-by: Juri Lelli <juri.lelli@arm.com>
Tested-by: Shilpasri G Bhat <shilpa.bhat@linux.vnet.ibm.com>
[ rjw: Subject & changelog ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The ondemand and conservative governors use the global-attr or freq-attr
structures to represent sysfs attributes corresponding to their tunables
(which of them is actually used depends on whether or not different
policy objects can use the same governor with different tunables at the
same time and, consequently, on where those attributes are located in
sysfs).
Unfortunately, in the freq-attr case, the standard cpufreq show/store
sysfs attribute callbacks are applied to the governor tunable attributes
and they always acquire the policy->rwsem lock before carrying out the
operation. That may lead to an ABBA deadlock if governor tunable
attributes are removed under policy->rwsem while one of them is being
accessed concurrently (if sysfs attributes removal wins the race, it
will wait for the access to complete with policy->rwsem held while the
attribute callback will block on policy->rwsem indefinitely).
We attempted to address this issue by dropping policy->rwsem around
governor tunable attributes removal (that is, around invocations of the
->governor callback with the event arg equal to CPUFREQ_GOV_POLICY_EXIT)
in cpufreq_set_policy(), but that opened up race conditions that had not
been possible with policy->rwsem held all the time. Therefore
policy->rwsem cannot be dropped in cpufreq_set_policy() at any point,
but the deadlock situation described above must be avoided too.
To that end, use the observation that in principle governor tunables may
be represented by the same data type regardless of whether the governor
is system-wide or per-policy and introduce a new structure, struct
governor_attr, for representing them and new corresponding macros for
creating show/store sysfs callbacks for them. Also make their parent
kobject use a new kobject type whose default show/store callbacks are
not related to the standard core cpufreq ones in any way (and they don't
acquire policy->rwsem in particular).
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Tested-by: Juri Lelli <juri.lelli@arm.com>
Tested-by: Shilpasri G Bhat <shilpa.bhat@linux.vnet.ibm.com>
[ rjw: Subject & changelog + rebase ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
There are a few common tunables shared between the ondemand and
conservative governors. Move them to struct dbs_data to simplify
code.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Tested-by: Juri Lelli <juri.lelli@arm.com>
Tested-by: Shilpasri G Bhat <shilpa.bhat@linux.vnet.ibm.com>
[ rjw: Changelog ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Some tunables are present in governor-specific structures, whereas one
(min_sampling_rate) is located directly in struct dbs_data.
There is a special macro for creating its sysfs attribute and the
show/store callbacks, but since more tunables are going to be moved
to struct dbs_data, a new generic macro for such cases will be useful,
so add it and use it for min_sampling_rate.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Tested-by: Juri Lelli <juri.lelli@arm.com>
Tested-by: Shilpasri G Bhat <shilpa.bhat@linux.vnet.ibm.com>
[ rjw: Subject & changelog ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The skip_work field in struct policy_dbs_info technically is a
counter, so give it a new name to reflect that.
No functional changes.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
The struct policy_dbs_info objects representing per-policy governor
data are not accessible directly from the corresponding policy
objects. To access them, one has to get a pointer to the
struct cpu_dbs_info of policy->cpu and use the policy_dbs field of
that which isn't really straightforward.
To address that rearrange the governor data structures so the
governor_data pointer in struct cpufreq_policy will point to
struct policy_dbs_info (instead of struct dbs_data) and that will
contain a pointer to struct dbs_data.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Since policy->cpu is always passed as the second argument to
dbs_check_cpu(), it is not really necessary to pass it, because
the function can obtain that value via its first argument just fine.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
The struct cpu_common_dbs_info structure represents the per-policy
part of the governor data (for the ondemand and conservative
governors), but its name doesn't reflect its purpose.
Rename it to struct policy_dbs_info and rename variables related to
it accordingly.
No functional changes.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Since it is possible to obtain a pointer to struct dbs_governor
from a pointer to the struct governor embedded in it with the help
of container_of(), the additional gov pointer in struct dbs_data
isn't really necessary.
Drop that pointer and make the code using it reach the dbs_governor
object via policy->governor.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Since it is possible to obtain a pointer to struct dbs_governor
from a pointer to the struct governor embedded in it via
container_of(), the second argument of cpufreq_governor_init()
is not necessary. Accordingly, cpufreq_governor_dbs() doesn't
need its second argument either and the ->governor callbacks
for both the ondemand and conservative governors may be set
to cpufreq_governor_dbs() directly. Make that happen.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Saravana Kannan <skannan@codeaurora.org>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
The ondemand and conservative governors are represented by
struct common_dbs_data whose name doesn't reflect the purpose it
is used for, so rename it to struct dbs_governor and rename
variables of that type accordingly.
No functional changes.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
For the ondemand and conservative governors (generally, governors
that use the common code in cpufreq_governor.c), there are two static
data structures representing the governor, the struct governor
structure (the interface to the cpufreq core) and the struct
common_dbs_data one (the interface to the cpufreq_governor.c code).
There's no fundamental reason why those two structures have to be
separate. Moreover, if the struct governor one is included into
struct common_dbs_data, it will be possible to reach the latter from
the policy via its policy->governor pointer, so it won't be necessary
to pass a separate pointer to it around. For this reason, embed
struct governor in struct common_dbs_data.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Saravana Kannan <skannan@codeaurora.org>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Every governor relying on the common code in cpufreq_governor.c
has to provide its own mutex in struct common_dbs_data. However,
there actually is no need to have a separate mutex per governor
for this purpose, they may be using the same global mutex just
fine. Accordingly, introduce a single common mutex for that and
drop the mutex field from struct common_dbs_data.
That at least will ensure that the mutex is always present and
initialized regardless of what the particular governors do.
Another benefit is that the common code does not need a pointer to
a governor-related structure to get to the mutex which sometimes
helps.
Finally, it makes the code generally easier to follow.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Saravana Kannan <skannan@codeaurora.org>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Instead of using a per-CPU deferrable timer for queuing up governor
work items, register a utilization update callback that will be
invoked from the scheduler on utilization changes.
The sampling rate is still the same as what was used for the
deferrable timers and the added irq_work overhead should be offset by
the eliminated timers overhead, so in theory the functional impact of
this patch should not be significant.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Tested-by: Gautham R. Shenoy <ego@linux.vnet.ibm.com>
It is possible to get rid of the timer_lock spinlock used by the
governor timer function for synchronization, but a couple of races
need to be avoided.
The first race is between multiple dbs_timer_handler() instances
that may be running in parallel with each other on different
CPUs. Namely, one of them has to queue up the work item, but it
cannot be queued up more than once. To achieve that,
atomic_inc_return() can be used on the skip_work field of
struct cpu_common_dbs_info.
The second race is between an already running dbs_timer_handler()
and gov_cancel_work(). In that case the dbs_timer_handler() might
not notice the skip_work incrementation in gov_cancel_work() and
it might queue up its work item after gov_cancel_work() had
returned (and that work item would corrupt skip_work going
forward). To prevent that from happening, gov_cancel_work()
can be made wait for the timer function to complete (on all CPUs)
right after skip_work has been incremented.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
cpufreq governors evaluate load at sampling rate and based on that they
update frequency for a group of CPUs belonging to the same cpufreq
policy.
This is required to be done in a single thread for all policy->cpus, but
because we don't want to wakeup idle CPUs to do just that, we use
deferrable work for this. If we would have used a single delayed
deferrable work for the entire policy, there were chances that the CPU
required to run the handler can be in idle and we might end up not
changing the frequency for the entire group with load variations.
And so we were forced to keep per-cpu works, and only the one that
expires first need to do the real work and others are rescheduled for
next sampling time.
We have been using the more complex solution until now, where we used a
delayed deferrable work for this, which is a combination of a timer and
a work.
This could be made lightweight by keeping per-cpu deferred timers with a
single work item, which is scheduled by the first timer that expires.
This patch does just that and here are important changes:
- The timer handler will run in irq context and so we need to use a
spin_lock instead of the timer_mutex. And so a separate timer_lock is
created. This also makes the use of the mutex and lock quite clear, as
we know what exactly they are protecting.
- A new field 'skip_work' is added to track when the timer handlers can
queue a work. More comments present in code.
Suggested-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Ashwin Chaugule <ashwin.chaugule@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Pass 'policy' as argument to ->gov_dbs_timer() instead of cdbs and
dbs_data.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Conservative governor has its own 'enable' field to check if
conservative governor is used for a CPU or not
This can be checked by policy->governor with 'cpufreq_gov_conservative'
and so this field can be dropped.
Because its not guaranteed that dbs_info->cdbs.shared will a valid
pointer for all CPUs (will be NULL for CPUs that don't use
ondemand/conservative governors), we can't use it anymore. Lets get
policy with cpufreq_cpu_get_raw() instead.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Some part of cs_dbs_timer() and od_dbs_timer() is exactly same and is
unnecessarily duplicated.
Create the real work-handler in cpufreq_governor.c and put the common
code in this routine (dbs_timer()).
Shouldn't make any functional change.
Reviewed-and-tested-by: Preeti U Murthy <preeti@linux.vnet.ibm.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Some information is common to all CPUs belonging to a policy, but are
kept on per-cpu basis. Lets keep that in another structure common to all
policy->cpus. That will make updates/reads to that less complex and less
error prone.
The memory for cpu_common_dbs_info is allocated/freed at INIT/EXIT, so
that it we don't reallocate it for STOP/START sequence. It will be also
be used (in next patch) while the governor is stopped and so must not be
freed that early.
Reviewed-and-tested-by: Preeti U Murthy <preeti@linux.vnet.ibm.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Just call it 'policy', cur_policy is unnecessarily long and doesn't
have any special meaning.
Reviewed-by: Preeti U Murthy <preeti@linux.vnet.ibm.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Its not common info to all CPUs, but a structure representing common
type of cpu info to both governor types. Lets drop 'common_' from its
name.
Reviewed-by: Preeti U Murthy <preeti@linux.vnet.ibm.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Its not used at all, drop it.
Reviewed-by: Preeti U Murthy <preeti@linux.vnet.ibm.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Delayed work was named as 'work' and to access work within it we do
work.work. Not much readable. Rename delayed_work as 'dwork'.
Reviewed-by: Preeti U Murthy <preeti@linux.vnet.ibm.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
There are several races reported in cpufreq core around governors (only
ondemand and conservative) by different people.
There are at least two race scenarios present in governor code:
(a) Concurrent access/updates of governor internal structures.
It is possible that fields such as 'dbs_data->usage_count', etc. are
accessed simultaneously for different policies using same governor
structure (i.e. CPUFREQ_HAVE_GOVERNOR_PER_POLICY flag unset). And
because of this we can dereference bad pointers.
For example consider a system with two CPUs with separate 'struct
cpufreq_policy' instances. CPU0 governor: ondemand and CPU1: powersave.
CPU0 switching to powersave and CPU1 to ondemand:
CPU0 CPU1
store* store*
cpufreq_governor_exit() cpufreq_governor_init()
dbs_data = cdata->gdbs_data;
if (!--dbs_data->usage_count)
kfree(dbs_data);
dbs_data->usage_count++;
*Bad pointer dereference*
There are other races possible between EXIT and START/STOP/LIMIT as
well. Its really complicated.
(b) Switching governor state in bad sequence:
For example trying to switch a governor to START state, when the
governor is in EXIT state. There are some checks present in
__cpufreq_governor() but they aren't sufficient as they compare events
against 'policy->governor_enabled', where as we need to take governor's
state into account, which can be used by multiple policies.
These two issues need to be solved separately and the responsibility
should be properly divided between cpufreq and governor core.
The first problem is more about the governor core, as it needs to
protect its structures properly. And the second problem should be fixed
in cpufreq core instead of governor, as its all about sequence of
events.
This patch is trying to solve only the first problem.
There are two types of data we need to protect,
- 'struct common_dbs_data': No matter what, there is going to be a
single copy of this per governor.
- 'struct dbs_data': With CPUFREQ_HAVE_GOVERNOR_PER_POLICY flag set, we
will have per-policy copy of this data, otherwise a single copy.
Because of such complexities, the mutex present in 'struct dbs_data' is
insufficient to solve our problem. For example we need to protect
fetching of 'dbs_data' from different structures at the beginning of
cpufreq_governor_dbs(), to make sure it isn't currently being updated.
This can be fixed if we can guarantee serialization of event parsing
code for an individual governor. This is best solved with a mutex per
governor, and the placeholder for that is 'struct common_dbs_data'.
And so this patch moves the mutex from 'struct dbs_data' to 'struct
common_dbs_data' and takes it at the beginning and drops it at the end
of cpufreq_governor_dbs().
Tested with and without following configuration options:
CONFIG_LOCKDEP_SUPPORT=y
CONFIG_DEBUG_RT_MUTEXES=y
CONFIG_DEBUG_PI_LIST=y
CONFIG_DEBUG_SPINLOCK=y
CONFIG_DEBUG_MUTEXES=y
CONFIG_DEBUG_LOCK_ALLOC=y
CONFIG_PROVE_LOCKING=y
CONFIG_LOCKDEP=y
CONFIG_DEBUG_ATOMIC_SLEEP=y
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Preeti U Murthy <preeti@linux.vnet.ibm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Notifiers are required only for conservative governor and the common
governor code is unnecessarily polluted with that. Handle that from
cs_init/exit() instead of cpufreq_governor_dbs().
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Preeti U Murthy <preeti@linux.vnet.ibm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
'copy_prev_load' was recently added by commit: 18b46ab (cpufreq: governor: Be
friendly towards latency-sensitive bursty workloads).
It actually is a bit redundant as we also have 'prev_load' which can store any
integer value and can be used instead of 'copy_prev_load' by setting it zero.
True load can also turn out to be zero during long idle intervals (and hence the
actual value of 'prev_load' and the overloaded value can clash). However this is
not a problem because, if the true load was really zero in the previous
interval, it makes sense to evaluate the load afresh for the current interval
rather than copying the previous load.
So, drop 'copy_prev_load' and use 'prev_load' instead.
Update comments as well to make it more clear.
There is another change here which was probably missed by Srivatsa during the
last version of updates he made. The unlikely in the 'if' statement was covering
only half of the condition and the whole line should actually come under it.
Also checkpatch is made more silent as it was reporting this (--strict option):
CHECK: Alignment should match open parenthesis
+ if (unlikely(wall_time > (2 * sampling_rate) &&
+ j_cdbs->prev_load)) {
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Acked-by: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Cpufreq governors like the ondemand governor calculate the load on the CPU
periodically by employing deferrable timers. A deferrable timer won't fire
if the CPU is completely idle (and there are no other timers to be run), in
order to avoid unnecessary wakeups and thus save CPU power.
However, the load calculation logic is agnostic to all this, and this can
lead to the problem described below.
Time (ms) CPU 1
100 Task-A running
110 Governor's timer fires, finds load as 100% in the last
10ms interval and increases the CPU frequency.
110.5 Task-A running
120 Governor's timer fires, finds load as 100% in the last
10ms interval and increases the CPU frequency.
125 Task-A went to sleep. With nothing else to do, CPU 1
went completely idle.
200 Task-A woke up and started running again.
200.5 Governor's deferred timer (which was originally programmed
to fire at time 130) fires now. It calculates load for the
time period 120 to 200.5, and finds the load is almost zero.
Hence it decreases the CPU frequency to the minimum.
210 Governor's timer fires, finds load as 100% in the last
10ms interval and increases the CPU frequency.
So, after the workload woke up and started running, the frequency was suddenly
dropped to absolute minimum, and after that, there was an unnecessary delay of
10ms (sampling period) to increase the CPU frequency back to a reasonable value.
And this pattern repeats for every wake-up-from-cpu-idle for that workload.
This can be quite undesirable for latency- or response-time sensitive bursty
workloads. So we need to fix the governor's logic to detect such wake-up-from-
cpu-idle scenarios and start the workload at a reasonably high CPU frequency.
One extreme solution would be to fake a load of 100% in such scenarios. But
that might lead to undesirable side-effects such as frequency spikes (which
might also need voltage changes) especially if the previous frequency happened
to be very low.
We just want to avoid the stupidity of dropping down the frequency to a minimum
and then enduring a needless (and long) delay before ramping it up back again.
So, let us simply carry forward the previous load - that is, let us just pretend
that the 'load' for the current time-window is the same as the load for the
previous window. That way, the frequency and voltage will continue to be set
to whatever values they were set at previously. This means that bursty workloads
will get a chance to influence the CPU frequency at which they wake up from
cpu-idle, based on their past execution history. Thus, they might be able to
avoid suffering from slow wakeups and long response-times.
However, we should take care not to over-do this. For example, such a "copy
previous load" logic will benefit cases like this: (where # represents busy
and . represents idle)
##########.........#########.........###########...........##########........
but it will be detrimental in cases like the one shown below, because it will
retain the high frequency (copied from the previous interval) even in a mostly
idle system:
##########.........#.................#.....................#...............
(i.e., the workload finished and the remaining tasks are such that their busy
periods are smaller than the sampling interval, which causes the timer to
always get deferred. So, this will make the copy-previous-load logic copy
the initial high load to subsequent idle periods over and over again, thus
keeping the frequency high unnecessarily).
So, we modify this copy-previous-load logic such that it is used only once
upon every wakeup-from-idle. Thus if we have 2 consecutive idle periods, the
previous load won't get blindly copied over; cpufreq will freshly evaluate the
load in the second idle interval, thus ensuring that the system comes back to
its normal state.
[ The right way to solve this whole problem is to teach the CPU frequency
governors to also track load on a per-task basis, not just a per-CPU basis,
and then use both the data sources intelligently to set the appropriate
frequency on the CPUs. But that involves redesigning the cpufreq subsystem,
so this patch should make the situation bearable until then. ]
Experimental results:
+-------------------+
I ran a modified version of ebizzy (called 'sleeping-ebizzy') that sleeps in
between its execution such that its total utilization can be a user-defined
value, say 10% or 20% (higher the utilization specified, lesser the amount of
sleeps injected). This ebizzy was run with a single-thread, tied to CPU 8.
Behavior observed with tracing (sample taken from 40% utilization runs):
------------------------------------------------------------------------
Without patch:
~~~~~~~~~~~~~~
kworker/8:2-12137 416.335742: cpu_frequency: state=2061000 cpu_id=8
kworker/8:2-12137 416.335744: sched_switch: prev_comm=kworker/8:2 ==> next_comm=ebizzy
<...>-40753 416.345741: sched_switch: prev_comm=ebizzy ==> next_comm=kworker/8:2
kworker/8:2-12137 416.345744: cpu_frequency: state=4123000 cpu_id=8
kworker/8:2-12137 416.345746: sched_switch: prev_comm=kworker/8:2 ==> next_comm=ebizzy
<...>-40753 416.355738: sched_switch: prev_comm=ebizzy ==> next_comm=kworker/8:2
<snip> --------------------------------------------------------------------- <snip>
<...>-40753 416.402202: sched_switch: prev_comm=ebizzy ==> next_comm=swapper/8
<idle>-0 416.502130: sched_switch: prev_comm=swapper/8 ==> next_comm=ebizzy
<...>-40753 416.505738: sched_switch: prev_comm=ebizzy ==> next_comm=kworker/8:2
kworker/8:2-12137 416.505739: cpu_frequency: state=2061000 cpu_id=8
kworker/8:2-12137 416.505741: sched_switch: prev_comm=kworker/8:2 ==> next_comm=ebizzy
<...>-40753 416.515739: sched_switch: prev_comm=ebizzy ==> next_comm=kworker/8:2
kworker/8:2-12137 416.515742: cpu_frequency: state=4123000 cpu_id=8
kworker/8:2-12137 416.515744: sched_switch: prev_comm=kworker/8:2 ==> next_comm=ebizzy
Observation: Ebizzy went idle at 416.402202, and started running again at
416.502130. But cpufreq noticed the long idle period, and dropped the frequency
at 416.505739, only to increase it back again at 416.515742, realizing that the
workload is in-fact CPU bound. Thus ebizzy needlessly ran at the lowest frequency
for almost 13 milliseconds (almost 1 full sample period), and this pattern
repeats on every sleep-wakeup. This could hurt latency-sensitive workloads quite
a lot.
With patch:
~~~~~~~~~~~
kworker/8:2-29802 464.832535: cpu_frequency: state=2061000 cpu_id=8
<snip> --------------------------------------------------------------------- <snip>
kworker/8:2-29802 464.962538: sched_switch: prev_comm=kworker/8:2 ==> next_comm=ebizzy
<...>-40738 464.972533: sched_switch: prev_comm=ebizzy ==> next_comm=kworker/8:2
kworker/8:2-29802 464.972536: cpu_frequency: state=4123000 cpu_id=8
kworker/8:2-29802 464.972538: sched_switch: prev_comm=kworker/8:2 ==> next_comm=ebizzy
<...>-40738 464.982531: sched_switch: prev_comm=ebizzy ==> next_comm=kworker/8:2
<snip> --------------------------------------------------------------------- <snip>
kworker/8:2-29802 465.022533: sched_switch: prev_comm=kworker/8:2 ==> next_comm=ebizzy
<...>-40738 465.032531: sched_switch: prev_comm=ebizzy ==> next_comm=kworker/8:2
kworker/8:2-29802 465.032532: sched_switch: prev_comm=kworker/8:2 ==> next_comm=ebizzy
<...>-40738 465.035797: sched_switch: prev_comm=ebizzy ==> next_comm=swapper/8
<idle>-0 465.240178: sched_switch: prev_comm=swapper/8 ==> next_comm=ebizzy
<...>-40738 465.242533: sched_switch: prev_comm=ebizzy ==> next_comm=kworker/8:2
kworker/8:2-29802 465.242535: sched_switch: prev_comm=kworker/8:2 ==> next_comm=ebizzy
<...>-40738 465.252531: sched_switch: prev_comm=ebizzy ==> next_comm=kworker/8:2
Observation: Ebizzy went idle at 465.035797, and started running again at
465.240178. Since ebizzy was the only real workload running on this CPU,
cpufreq retained the frequency at 4.1Ghz throughout the run of ebizzy, no
matter how many times ebizzy slept and woke-up in-between. Thus, ebizzy
got the 10ms worth of 4.1 Ghz benefit during every sleep-wakeup (as compared
to the run without the patch) and this boost gave a modest improvement in total
throughput, as shown below.
Sleeping-ebizzy records-per-second:
-----------------------------------
Utilization Without patch With patch Difference (Absolute and % values)
10% 274767 277046 + 2279 (+0.829%)
20% 543429 553484 + 10055 (+1.850%)
40% 1090744 1107959 + 17215 (+1.578%)
60% 1634908 1662018 + 27110 (+1.658%)
A rudimentary and somewhat approximately latency-sensitive workload such as
sleeping-ebizzy itself showed a consistent, noticeable performance improvement
with this patch. Hence, workloads that are truly latency-sensitive will benefit
quite a bit from this change. Moreover, this is an overall win-win since this
patch does not hurt power-savings at all (because, this patch does not reduce
the idle time or idle residency; and the high frequency of the CPU when it goes
to cpu-idle does not affect/hurt the power-savings of deep idle states).
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Reviewed-by: Gautham R. Shenoy <ego@linux.vnet.ibm.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
When a CPU is hot removed we'll cancel all the delayed work items via
gov_cancel_work(). Sometimes the delayed work function determines that
it should adjust the delay for all other CPUs that the policy is
managing. If this scenario occurs, the canceling CPU will cancel its own
work but queue up the other CPUs works to run.
Commit 3617f2 (cpufreq: Fix timer/workqueue corruption due to double
queueing) has tried to fix this, but reading governor_enabled is not
protected by cpufreq_governor_lock. Even though od_dbs_timer() checks
governor_enabled before gov_queue_work(), this scenario may occur. For
example:
CPU0 CPU1
---- ----
cpu_down()
... <work runs>
__cpufreq_remove_dev() od_dbs_timer()
__cpufreq_governor() policy->governor_enabled
policy->governor_enabled = false;
cpufreq_governor_dbs()
case CPUFREQ_GOV_STOP:
gov_cancel_work(dbs_data, policy);
cpu0 work is canceled
timer is canceled
cpu1 work is canceled
<waits for cpu1>
gov_queue_work(*, *, true);
cpu0 work queued
cpu1 work queued
cpu2 work queued
...
cpu1 work is canceled
cpu2 work is canceled
...
At the end of the GOV_STOP case cpu0 still has a work queued to
run although the code is expecting all of the works to be
canceled. __cpufreq_remove_dev() will then proceed to
re-initialize all the other CPUs works except for the CPU that is
going down. The CPUFREQ_GOV_START case in cpufreq_governor_dbs()
will trample over the queued work and debugobjects will spit out
a warning:
WARNING: at lib/debugobjects.c:260 debug_print_object+0x94/0xbc()
ODEBUG: init active (active state 0) object type: timer_list hint: delayed_work_timer_fn+0x0/0x14
Modules linked in:
CPU: 1 PID: 1205 Comm: sh Tainted: G W 3.10.0 #200
[<c01144f0>] (unwind_backtrace+0x0/0xf8) from [<c0111d98>] (show_stack+0x10/0x14)
[<c0111d98>] (show_stack+0x10/0x14) from [<c01272cc>] (warn_slowpath_common+0x4c/0x68)
[<c01272cc>] (warn_slowpath_common+0x4c/0x68) from [<c012737c>] (warn_slowpath_fmt+0x30/0x40)
[<c012737c>] (warn_slowpath_fmt+0x30/0x40) from [<c034c640>] (debug_print_object+0x94/0xbc)
[<c034c640>] (debug_print_object+0x94/0xbc) from [<c034c7f8>] (__debug_object_init+0xc8/0x3c0)
[<c034c7f8>] (__debug_object_init+0xc8/0x3c0) from [<c01360e0>] (init_timer_key+0x20/0x104)
[<c01360e0>] (init_timer_key+0x20/0x104) from [<c04872ac>] (cpufreq_governor_dbs+0x1dc/0x68c)
[<c04872ac>] (cpufreq_governor_dbs+0x1dc/0x68c) from [<c04833a8>] (__cpufreq_governor+0x80/0x1b0)
[<c04833a8>] (__cpufreq_governor+0x80/0x1b0) from [<c0483704>] (__cpufreq_remove_dev.isra.12+0x22c/0x380)
[<c0483704>] (__cpufreq_remove_dev.isra.12+0x22c/0x380) from [<c0692f38>] (cpufreq_cpu_callback+0x48/0x5c)
[<c0692f38>] (cpufreq_cpu_callback+0x48/0x5c) from [<c014fb40>] (notifier_call_chain+0x44/0x84)
[<c014fb40>] (notifier_call_chain+0x44/0x84) from [<c012ae44>] (__cpu_notify+0x2c/0x48)
[<c012ae44>] (__cpu_notify+0x2c/0x48) from [<c068dd40>] (_cpu_down+0x80/0x258)
[<c068dd40>] (_cpu_down+0x80/0x258) from [<c068df40>] (cpu_down+0x28/0x3c)
[<c068df40>] (cpu_down+0x28/0x3c) from [<c068e4c0>] (store_online+0x30/0x74)
[<c068e4c0>] (store_online+0x30/0x74) from [<c03a7308>] (dev_attr_store+0x18/0x24)
[<c03a7308>] (dev_attr_store+0x18/0x24) from [<c0256fe0>] (sysfs_write_file+0x100/0x180)
[<c0256fe0>] (sysfs_write_file+0x100/0x180) from [<c01fec9c>] (vfs_write+0xbc/0x184)
[<c01fec9c>] (vfs_write+0xbc/0x184) from [<c01ff034>] (SyS_write+0x40/0x68)
[<c01ff034>] (SyS_write+0x40/0x68) from [<c010e200>] (ret_fast_syscall+0x0/0x48)
In gov_queue_work(), lock cpufreq_governor_lock before gov_queue_work,
and unlock it after __gov_queue_work(). In this way, governor_enabled
is guaranteed not changed in gov_queue_work().
Signed-off-by: Jane Li <jiel@marvell.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Use cpufreq_driver->flags to mark CPUFREQ_HAVE_GOVERNOR_PER_POLICY instead
of a separate field within cpufreq_driver. This will save some bytes of
memory.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
- 'Governer' should be 'Governor'.
- 'S' is used for Siemens (electrical conductance) in SI units,
so use small 's' for seconds.
Signed-off-by: Stratos Karafotis <stratosk@semaphore.gr>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
They are called policy, cur_policy, new_policy, data, etc. Just call
them policy wherever possible.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
This patch addresses the following issues in the header files in the
cpufreq core:
- Include headers in ascending order, so that we don't add same
many times by mistake.
- <asm/> must be included after <linux/>, so that they override
whatever they need to.
- Remove unnecessary includes.
- Don't include files already included by cpufreq.h or
cpufreq_governor.h.
[rjw: Changelog]
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
This sysfs file was called ignore_nice_load earlier and commit
4d5dcc4 (cpufreq: governor: Implement per policy instances of
governors) changed its name to ignore_nice by mistake.
Lets get it renamed back to its original name.
Reported-by: Martin von Gagern <Martin.vGagern@gmx.net>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Cc: 3.10+ <stable@vger.kernel.org> # 3.10+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The ondemand governor calculates load in terms of frequency and
increases it only if load_freq is greater than up_threshold
multiplied by the current or average frequency. This appears to
produce oscillations of frequency between min and max because,
for example, a relatively small load can easily saturate minimum
frequency and lead the CPU to the max. Then, it will decrease
back to the min due to small load_freq.
Change the calculation method of load and target frequency on the
basis of the following two observations:
- Load computation should not depend on the current or average
measured frequency. For example, absolute load of 80% at 100MHz
is not necessarily equivalent to 8% at 1000MHz in the next
sampling interval.
- It should be possible to increase the target frequency to any
value present in the frequency table proportional to the absolute
load, rather than to the max only, so that:
Target frequency = C * load
where we take C = policy->cpuinfo.max_freq / 100.
Tested on Intel i7-3770 CPU @ 3.40GHz and on Quad core 1500MHz Krait.
Phoronix benchmark of Linux Kernel Compilation 3.1 test shows an
increase ~1.5% in performance. cpufreq_stats (time_in_state) shows
that middle frequencies are used more, with this patch. Highest
and lowest frequencies were used less by ~9%.
[rjw: We have run multiple other tests on kernels with this
change applied and in the vast majority of cases it turns out
that the resulting performance improvement also leads to reduced
consumption of energy. The change is additionally justified by
the overall simplification of the code in question.]
Signed-off-by: Stratos Karafotis <stratosk@semaphore.gr>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
There were a few noticeable formatting issues in core cpufreq code.
This cleans them up to make code look better. The changes include:
- Whitespace cleanup.
- Rearrangements of code.
- Multiline comments fixes.
- Formatting changes to fit 80 columns.
Copyright information in cpufreq.c is also updated to include my name
for 2013.
[rjw: Changelog]
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Governors other than ondemand and conservative can also use
get_cpu_idle_time() and they aren't required to compile
cpufreq_governor.c. So, move these independent routines to
cpufreq.c instead.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
There are two types of INIT/EXIT activities that we need to do for
governors:
- Done only once per governor (doesn't depend how many instances of
the governor there are). eg: cpufreq_register_notifier() for
conservative governor.
- Done per governor instance, eg: sysfs_{create|remove}_group().
There were some corner cases where current code isn't able to handle
them separately and so failing for some test cases.
We use two separate variables now for keeping track of above two
requirements.
- governor->initialized for first one
- dbs_data->usage_count for per governor instance
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
* pm-cpufreq: (57 commits)
cpufreq: MAINTAINERS: Add co-maintainer
cpufreq: pxa2xx: initialize variables
ARM: S5pv210: compiling issue, ARM_S5PV210_CPUFREQ needs CONFIG_CPU_FREQ_TABLE=y
cpufreq: cpu0: Put cpu parent node after using it
cpufreq: ARM big LITTLE: Adapt to latest cpufreq updates
cpufreq: ARM big LITTLE: put DT nodes after using them
cpufreq: Don't call __cpufreq_governor() for drivers without target()
cpufreq: exynos5440: Protect OPP search calls with RCU lock
cpufreq: dbx500: Round to closest available freq
cpufreq: Call __cpufreq_governor() with correct policy->cpus mask
cpufreq / intel_pstate: Optimize intel_pstate_set_policy
cpufreq: OMAP: instantiate omap-cpufreq as a platform_driver
arm: exynos: Enable OPP library support for exynos5440
cpufreq: exynos: Remove error return even if no soc is found
cpufreq: exynos: Add cpufreq driver for exynos5440
cpufreq: AMD "frequency sensitivity feedback" powersave bias for ondemand governor
cpufreq: ondemand: allow custom powersave_bias_target handler to be registered
cpufreq: convert cpufreq_driver to using RCU
cpufreq: powerpc/platforms/cell: move cpufreq driver to drivers/cpufreq
cpufreq: sparc: move cpufreq driver to drivers/cpufreq
...
Conflicts:
MAINTAINERS (with commit a8e39c3 from pm-cpuidle)
drivers/cpufreq/cpufreq_governor.h (with commit beb0ff3)
This allows for another [arch specific] driver to hook into existing
powersave bias function of the ondemand governor. i.e. This allows AMD
specific powersave bias function (in a separate AMD specific driver)
to aid ondemand governor's frequency transition decisions.
Signed-off-by: Jacob Shin <jacob.shin@amd.com>
Acked-by: Thomas Renninger <trenn@suse.de>
Acked-by: Borislav Petkov <bp@suse.de>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
It should be "governor".
Signed-off-by: Borislav Petkov <bp@suse.de>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Currently we always calculate the CPU iowait time and add it to idle time.
If we are in ondemand and we use io_is_busy, we re-calculate iowait time
and we subtract it from idle time.
With this patch iowait time is calculated only when necessary avoiding
the double call to get_cpu_iowait_time_us. We use a parameter in
function get_cpu_idle_time to distinguish when the iowait time will be
added to idle time or not, without the need of keeping the prev_io_wait.
Signed-off-by: Stratos Karafotis <stratosk@semaphore.gr>
Acked-by: Viresh Kumar <viresh.kumar@linaro.,org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Following patch has introduced per cpu timers or works for ondemand and
conservative governors.
commit 2abfa876f1
Author: Rickard Andersson <rickard.andersson@stericsson.com>
Date: Thu Dec 27 14:55:38 2012 +0000
cpufreq: handle SW coordinated CPUs
This causes additional unnecessary interrupts on all cpus when the load is
recently evaluated by any other cpu. i.e. When load is recently evaluated by cpu
x, we don't really need any other cpu to evaluate this load again for the next
sampling_rate time.
Some sort of code is present to avoid that but we are still getting timer
interrupts for all cpus. A good way of avoiding this would be to modify delays
for all cpus (policy->cpus) whenever any cpu has evaluated load.
This patch does this change and some related code cleanup.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Currently MIN_LATENCY_MULTIPLIER is set defined as 100 and so on a system with
transition latency of 1 ms, the minimum sampling time comes to be around 100 ms.
That is quite big if you want to get better performance for your system.
Redefine MIN_LATENCY_MULTIPLIER to 20 so that we can support 20ms sampling rate
for such platforms.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Currently, there can't be multiple instances of single governor_type.
If we have a multi-package system, where we have multiple instances
of struct policy (per package), we can't have multiple instances of
same governor. i.e. We can't have multiple instances of ondemand
governor for multiple packages.
Governors directory in sysfs is created at /sys/devices/system/cpu/cpufreq/
governor-name/. Which again reflects that there can be only one
instance of a governor_type in the system.
This is a bottleneck for multicluster system, where we want different
packages to use same governor type, but with different tunables.
This patch uses the infrastructure provided by earlier patch and
implements init/exit routines for ondemand and conservative
governors.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Rafael J. Wysocki