Whenever we are changing frequency of a cpu, we are calling PRECHANGE and
POSTCHANGE notifiers. They must be serialized. i.e. PRECHANGE or POSTCHANGE
shouldn't be called twice contiguously.
This can happen due to bugs in users of __cpufreq_driver_target() or actual
cpufreq drivers who are sending these notifiers.
This patch adds some protection against this. Now, we keep track of the last
transaction and see if something went wrong.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
__cpufreq_notify_transition() is used only in cpufreq.c,
make it static.
Signed-off-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>
Cpufreq governors' stop and start operations should be carried out
in sequence. Otherwise, there will be unexpected behavior, like in
the example below.
Suppose there are 4 CPUs and policy->cpu=CPU0, CPU1/2/3 are linked
to CPU0. The normal sequence is:
1) Current governor is userspace. An application tries to set the
governor to ondemand. It will call __cpufreq_set_policy() in
which it will stop the userspace governor and then start the
ondemand governor.
2) Current governor is userspace. The online of CPU3 runs on CPU0.
It will call cpufreq_add_policy_cpu() in which it will first
stop the userspace governor, and then start it again.
If the sequence of the above two cases interleaves, it becomes:
1) Application stops userspace governor
2) Hotplug stops userspace governor
which is a problem, because the governor shouldn't be stopped twice
in a row. What happens next is:
3) Application starts ondemand governor
4) Hotplug starts a governor
In step 4, the hotplug is supposed to start the userspace governor,
but now the governor has been changed by the application to ondemand,
so the ondemand governor is started once again, which is incorrect.
The solution is to prevent policy governors from being stopped
multiple times in a row. A governor should only be stopped once for
one policy. After it has been stopped, no more governor stop
operations should be executed.
Also add a mutex to serialize governor operations.
[rjw: Changelog. And you owe me a beverage of my choice.]
Signed-off-by: Xiaoguang Chen <chenxg@marvell.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
struct cpufreq_policy is already passed as argument to some routines
like: __cpufreq_driver_getavg() and so we don't really need to do
cpufreq_cpu_get() before and cpufreq_cpu_put() in them to get a
policy structure.
Remove them.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
When we don't have any file in cpu/cpufreq directory we shouldn't
create it. Specially with the introduction of per-policy governor
instance patchset, even governors are moved to
cpu/cpu*/cpufreq/governor-name directory and so this directory is
just not required.
Lets have it only when required.
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>
get_governor_parent_kobj() can be used by any governor, generic
cpufreq governors or platform specific ones and so must be present in
cpufreq.c instead of cpufreq_governor.c.
This patch moves it to cpufreq.c. This also adds
EXPORT_SYMBOL_GPL(get_governor_parent_kobj) so that modules can use
this function too.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
This patch adds: EXPORT_SYMBOL_GPL(have_governor_per_policy), so that
this routine can be used by modules too.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The file permissions of cpufreq per-cpu sysfs files are not preserved
across suspend/resume because we internally go through the CPU
Hotplug path which reinitializes the file permissions on CPU online.
But the user is not supposed to know that we are using CPU hotplug
internally within suspend/resume (IOW, the kernel should not silently
wreck the user-set file permissions across a suspend cycle).
Therefore, we need to preserve the file permissions as they are
across suspend/resume.
The simplest way to achieve that is to just not touch the sysfs files
at all - ie., just ignore the CPU hotplug notifications in the
suspend/resume path (_FROZEN) in the cpufreq hotplug callback.
Reported-by: Robert Jarzmik <robert.jarzmik@intel.com>
Reported-by: Durgadoss R <durgadoss.r@intel.com>
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
We must call __cpufreq_governor(data, CPUFREQ_GOV_POLICY_EXIT) before
calling cpufreq_cpu_put(data), so that policy kobject have valid
fields. Otherwise, removing last online cpu of policy->cpus causes
this crash for ondemand/conservative governor.
[<c00fb076>] (sysfs_find_dirent+0xe/0xa8) from [<c00fb1bd>] (sysfs_get_dirent+0x21/0x58)
[<c00fb1bd>] (sysfs_get_dirent+0x21/0x58) from [<c00fc259>] (sysfs_remove_group+0x85/0xbc)
[<c00fc259>] (sysfs_remove_group+0x85/0xbc) from [<c02faad9>] (cpufreq_governor_dbs+0x369/0x4a0)
[<c02faad9>] (cpufreq_governor_dbs+0x369/0x4a0) from [<c02f66d7>] (__cpufreq_governor+0x2b/0x8c)
[<c02f66d7>] (__cpufreq_governor+0x2b/0x8c) from [<c02f6893>] (__cpufreq_remove_dev.isra.12+0x15b/0x250)
[<c02f6893>] (__cpufreq_remove_dev.isra.12+0x15b/0x250) from [<c03e91c7>] (cpufreq_cpu_callback+0x2f/0x3c)
[<c03e91c7>] (cpufreq_cpu_callback+0x2f/0x3c) from [<c0036fe1>] (notifier_call_chain+0x45/0x54)
[<c0036fe1>] (notifier_call_chain+0x45/0x54) from [<c001e611>] (__cpu_notify+0x1d/0x34)
[<c001e611>] (__cpu_notify+0x1d/0x34) from [<c03e5833>] (_cpu_down+0x63/0x1ac)
[<c03e5833>] (_cpu_down+0x63/0x1ac) from [<c03e5997>] (cpu_down+0x1b/0x30)
[<c03e5997>] (cpu_down+0x1b/0x30) from [<c03e60eb>] (store_online+0x27/0x54)
[<c03e60eb>] (store_online+0x27/0x54) from [<c0295629>] (dev_attr_store+0x11/0x18)
[<c0295629>] (dev_attr_store+0x11/0x18) from [<c00f9edd>] (sysfs_write_file+0xed/0x114)
[<c00f9edd>] (sysfs_write_file+0xed/0x114) from [<c00b42a9>] (vfs_write+0x65/0xd8)
[<c00b42a9>] (vfs_write+0x65/0xd8) from [<c00b4523>] (sys_write+0x2f/0x50)
[<c00b4523>] (sys_write+0x2f/0x50) from [<c000cdc1>] (ret_fast_syscall+0x1/0x52)
Of course this only impacted drivers which have
have_governor_per_policy set to true. i.e. big LITTLE cpufreq driver.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Commit 5800043 (cpufreq: convert cpufreq_driver to using RCU) causes
the following call trace to be spit on boot:
BUG: sleeping function called from invalid context at /scratch/rafael/work/linux-pm/mm/slab.c:3179
in_atomic(): 0, irqs_disabled(): 0, pid: 292, name: systemd-udevd
2 locks held by systemd-udevd/292:
#0: (subsys mutex){+.+.+.}, at: [<ffffffff8146851a>] subsys_interface_register+0x4a/0xe0
#1: (rcu_read_lock){.+.+.+}, at: [<ffffffff81538210>] cpufreq_add_dev_interface+0x60/0x5e0
Pid: 292, comm: systemd-udevd Not tainted 3.9.0-rc8+ #323
Call Trace:
[<ffffffff81072c90>] __might_sleep+0x140/0x1f0
[<ffffffff811581c2>] kmem_cache_alloc+0x42/0x2b0
[<ffffffff811e7179>] sysfs_new_dirent+0x59/0x130
[<ffffffff811e63cb>] sysfs_add_file_mode+0x6b/0x110
[<ffffffff81538210>] ? cpufreq_add_dev_interface+0x60/0x5e0
[<ffffffff810a3254>] ? __lock_is_held+0x54/0x80
[<ffffffff811e647d>] sysfs_add_file+0xd/0x10
[<ffffffff811e6541>] sysfs_create_file+0x21/0x30
[<ffffffff81538280>] cpufreq_add_dev_interface+0xd0/0x5e0
[<ffffffff81538210>] ? cpufreq_add_dev_interface+0x60/0x5e0
[<ffffffffa000337f>] ? acpi_processor_get_platform_limit+0x32/0xbb [processor]
[<ffffffffa022f540>] ? do_drv_write+0x70/0x70 [acpi_cpufreq]
[<ffffffff810a3254>] ? __lock_is_held+0x54/0x80
[<ffffffff8106c97e>] ? up_read+0x1e/0x40
[<ffffffff8106e632>] ? __blocking_notifier_call_chain+0x72/0xc0
[<ffffffff81538dbd>] cpufreq_add_dev+0x62d/0xae0
[<ffffffff815389b8>] ? cpufreq_add_dev+0x228/0xae0
[<ffffffff81468569>] subsys_interface_register+0x99/0xe0
[<ffffffffa014d000>] ? 0xffffffffa014cfff
[<ffffffff81535d5d>] cpufreq_register_driver+0x9d/0x200
[<ffffffffa014d000>] ? 0xffffffffa014cfff
[<ffffffffa014d0e9>] acpi_cpufreq_init+0xe9/0x1000 [acpi_cpufreq]
[<ffffffff810002fa>] do_one_initcall+0x11a/0x170
[<ffffffff810b4b87>] load_module+0x1cf7/0x2920
[<ffffffff81322580>] ? ddebug_proc_open+0xb0/0xb0
[<ffffffff816baee0>] ? retint_restore_args+0xe/0xe
[<ffffffff810b5887>] sys_init_module+0xd7/0x120
[<ffffffff816bb6d2>] system_call_fastpath+0x16/0x1b
which is quite obvious, because that commit put (multiple instances
of) sysfs_create_file() under rcu_read_lock()/rcu_read_unlock(),
although sysfs_create_file() may cause memory to be allocated with
GFP_KERNEL and that may sleep, which is not permitted in RCU read
critical section.
Revert the buggy commit altogether along with some changes on top
of it.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Some cpufreq drivers implement their own governor and so don't need
us to call generic governors interface via __cpufreq_governor(). Few
recent commits haven't obeyed this law well and we saw some
regressions.
This patch is an attempt to fix the above issue.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reported-and-tested-by: Sedat Dilek <sedat.dilek@gmail.com>
Tested-by: Dirk Brandewie <dirk.brandewie@gmail.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
__cpufreq_governor() must be called with a correct policy->cpus mask.
In __cpufreq_remove_dev() we initially clear policy->cpus with
cpumask_clear_cpu() and then call
__cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT). If the governor
is doing some per-cpu stuff in EXIT callback, this can create
uncertain behavior.
Generic governors in drivers/cpufreq/ doesn't do any per-cpu stuff
in EXIT callback and so we don't face any issues currently. But its
better to keep the code clean, so we don't face any issues in future.
Now, we call cpumask_clear_cpu() only when multiple cpus are managed
by policy.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
We eventually would like to remove the rwlock cpufreq_driver_lock or
convert it back to a spinlock and protect the read sections with RCU.
The first step in that direction is to make cpufreq_driver use RCU.
I don't see an easy wasy to protect the cpufreq_cpu_data structure
with RCU, so I am leaving it with the rwlock for now since under
certain configs __cpufreq_cpu_get is a hot spot with 256+ cores.
[rjw: Subject, changelog, white space]
Signed-off-by: Nathan Zimmer <nzimmer@sgi.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
policy->cpus contains all online cpus that have single shared clock line. And
their frequencies are always updated together.
Many SMP system's cpufreq drivers take care of this in individual drivers but
the best place for this code is in cpufreq core.
This patch modifies cpufreq_notify_transition() to notify frequency change for
all cpus in policy->cpus and hence updates all users of this API.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Acked-by: Stephen Warren <swarren@nvidia.com>
Tested-by: Stephen Warren <swarren@nvidia.com>
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>
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 is inclined towards providing this infrastructure. Because
we are required to allocate governor's resources dynamically now, we
must do it at policy creation and end. And so got
CPUFREQ_GOV_POLICY_INIT/EXIT.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
This eliminates the contention I am seeing in __cpufreq_cpu_get.
It also nicely stages the lock to be replaced by the rcu.
Signed-off-by: Nathan Zimmer <nzimmer@sgi.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
* pm-cpufreq: (55 commits)
cpufreq / intel_pstate: Fix 32 bit build
cpufreq: conservative: Fix typos in comments
cpufreq: ondemand: Fix typos in comments
cpufreq: exynos: simplify .init() for setting policy->cpus
cpufreq: kirkwood: Add a cpufreq driver for Marvell Kirkwood SoCs
cpufreq/x86: Add P-state driver for sandy bridge.
cpufreq_stats: do not remove sysfs files if frequency table is not present
cpufreq: Do not track governor name for scaling drivers with internal governors.
cpufreq: Only call cpufreq_out_of_sync() for driver that implement cpufreq_driver.target()
cpufreq: Retrieve current frequency from scaling drivers with internal governors
cpufreq: Fix locking issues
cpufreq: Create a macro for unlock_policy_rwsem{read,write}
cpufreq: Remove unused HOTPLUG_CPU code
cpufreq: governors: Fix WARN_ON() for multi-policy platforms
cpufreq: ondemand: Replace down_differential tuner with adj_up_threshold
cpufreq / stats: Get rid of CPUFREQ_STATDEVICE_ATTR
cpufreq: Don't check cpu_online(policy->cpu)
cpufreq: add imx6q-cpufreq driver
cpufreq: Don't remove sysfs link for policy->cpu
cpufreq: Remove unnecessary use of policy->shared_type
...
Scaling drivers that implement internal governors do not have governor
structures assocaited with them. Only track the name of the governor
associated with the CPU if the driver does not implement
cpufreq_driver.setpolicy()
Signed-off-by: Dirk Brandewie <dirk.j.brandewie@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Scaling drivers that implement cpufreq_driver.setpolicy() have
internal governors that do not signal changes via
cpufreq_notify_transition() so the frequncy in the policy will almost
certainly be different than the current frequncy. Only call
cpufreq_out_of_sync() when the underlying driver implements
cpufreq_driver.target()
Signed-off-by: Dirk Brandewie <dirk.j.brandewie@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Scaling drivers that implement the cpufreq_driver.setpolicy() versus
the cpufreq_driver.target() interface do not set policy->cur.
Normally policy->cur is set during the call to cpufreq_driver.target()
when the frequnecy request is made by the governor.
If the scaling driver implements cpufreq_driver.setpolicy() and
cpufreq_driver.get() interfaces use cpufreq_driver.get() to retrieve
the current frequency.
Signed-off-by: Dirk Brandewie <dirk.j.brandewie@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
cpufreq core uses two locks:
- cpufreq_driver_lock: General lock for driver and cpufreq_cpu_data array.
- cpu_policy_rwsemfix locking: per CPU reader-writer semaphore designed to cure
all cpufreq/hotplug/workqueue/etc related lock issues.
These locks were not used properly and are placed against their principle
(present before their definition) at various places. This patch is an attempt to
fix their use.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
On the lines of macro: lock_policy_rwsem, we can create another macro for
unlock_policy_rwsem. Lets do it.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Because the sibling cpu of any online cpu is identified very early in
cpufreq_add_dev(), below code is never executed. And so can be removed.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
On multi-policy systems there is a single instance of governor for both the
policies (if same governor is chosen for both policies). With the code update
from following patches:
8eeed09 cpufreq: governors: Get rid of dbs_data->enable field
b394058 cpufreq: governors: Reset tunables only for cpufreq_unregister_governor()
We are creating/removing sysfs directory of governor for for every call to
GOV_START and STOP. This would fail for multi-policy system as there is a
per-policy call to START/STOP.
This patch reuses the governor->initialized variable to detect total users of
governor.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
policy->cpu or cpus in policy->cpus can't be offline anymore. And so we don't
need to check if they are online or not.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
"cpufreq" directory in policy->cpu is never created using
sysfs_create_link(), but using kobject_init_and_add(). And so we
shouldn't call sysfs_remove_link() for policy->cpu(). sysfs stuff
for policy->cpu is automatically removed when we call kobject_put()
for dying policy.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Tested-by: Dirk Brandewie <dirk.brandewie@gmail.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Currently, whenever governor->governor() is called for CPUFRREQ_GOV_START event
we reset few tunables of governor. Which isn't correct, as this routine is
called for every cpu hot-[un]plugging event. We should actually be resetting
these only when the governor module is removed and re-installed.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Currently cpufreq_add_dev() firsts allocates policy, calls
driver->init() and then checks if this CPU is already managed or not.
And if it is already managed, its policy is freed.
We can save all this if we somehow know that CPU is managed or not in
advance. policy->related_cpus contains the list of all valid sibling
CPUs of policy->cpu. We can check this to see if the current CPU is
already managed.
From now on, platforms don't really need to set related_cpus from
their init() routines, as the same work is done by core too.
If a platform driver needs to set the related_cpus mask with some
additional CPUs, other than CPUs present in policy->cpus, they are
free to do it, though, as we don't override anything.
[rjw: Changelog]
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Tested-by: Shawn Guo <shawn.guo@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
This reverts commit 956f339 "cpufreq: Don't use cpu removed during
cpufreq_driver_unregister".
With the addition of the following commit, this change/variable is not
required any more:
commit b9ba2725343ae57add3f324dfa5074167f48de96
Author: Viresh Kumar <viresh.kumar@linaro.org>
Date: Mon Jan 14 13:23:03 2013 +0000
cpufreq: Simplify __cpufreq_remove_dev()
[rjw: Subject and changelog]
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Add a helper function to return cpufreq_driver->name.
Signed-off-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
When disable_cpufreq() is called some exported functions are still
being used that do not have a check for cpufreq being disabled.
Add a disabled check into cpufreq_cpu_get() to return NULL if
cpufreq is disabled this covers most of the exported functions. For
the exported functions that do not call cpufreq_cpu_get() add an
explicit check.
Signed-off-by: Dirk Brandewie <dirk.j.brandewie@intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
__cpufreq_remove_dev() is called on multiple occasions: cpufreq_driver
unregister and cpu removals.
Current implementation of this routine is overly complex without much need. If
the cpu to be removed is the policy->cpu, we remove the policy first and add all
other cpus again from policy->cpus and then finally call __cpufreq_remove_dev()
again to remove the cpu to be deleted. Haahhhh..
There exist a simple solution to removal of a cpu:
- Simply use the old policy structure
- update its fields like: policy->cpu, etc.
- notify any users of cpufreq, which depend on changing policy->cpu
Hence this patch, which tries to implement the above theory. It is tested well
by myself on ARM big.LITTLE TC2 SoC, which has 5 cores (2 A15 and 3 A7). Both
A15's share same struct policy and all A7's share same policy structure.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Tested-by: Shawn Guo <shawn.guo@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
This is how the core works:
cpufreq_driver_unregister()
- subsys_interface_unregister()
- for_each_cpu() call cpufreq_remove_dev(), i.e. 0,1,2,3,4 when we
unregister.
cpufreq_remove_dev():
- Remove policy node
- Call cpufreq_add_dev() for next cpu, sharing mask with removed cpu.
i.e. When cpu 0 is removed, we call it for cpu 1. And when called for cpu 2,
we call it for cpu 3.
- cpufreq_add_dev() would call cpufreq_driver->init()
- init would return mask as AND of 2, 3 and 4 for cluster A7.
- cpufreq core would do online_cpu && policy->cpus
Here is the BUG(). Because cpu hasn't died but we have just unregistered
the cpufreq driver, online cpu would still have cpu 2 in it. And so thing
go bad again.
Solution: Keep cpumask of cpus that are registered with cpufreq core and clear
cpus when we get a call from subsys_interface_unregister() via
cpufreq_remove_dev().
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Because cpufreq core and governors worry only about the online cpus, if a cpu is
hot [un]plugged, we must notify governors about it, otherwise be ready to expect
something unexpected.
We already have notifiers in the form of CPUFREQ_GOV_START/CPUFREQ_GOV_STOP, we
just need to call them now.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
cpufreq core doesn't manage offline cpus and if driver->init() has returned
mask including offline cpus, it may result in unwanted behavior by cpufreq core
or governors.
We need to get only online cpus in this mask. There are two places to fix this
mask, cpufreq core and cpufreq driver. It makes sense to do this at common place
and hence is done in core.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The text in Documentation said it would be removed in 2.6.41;
the text in the Kconfig said removal in the 3.1 release. Either
way you look at it, we are well past both, so push it off a cliff.
Note that the POWER_CSTATE and the POWER_PSTATE are part of the
legacy tracing API. Remove all tracepoints which use these flags.
As can be seen from context, most already have a trace entry via
trace_cpu_idle anyways.
Also, the cpufreq/cpufreq.c PSTATE one is actually unpaired, as
compared to the CSTATE ones which all have a clear start/stop.
As part of this, the trace_power_frequency also becomes orphaned,
so it too is deleted.
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Remove an unnecessary initializer for the 'ret' variable in
__cpufreq_set_policy().
[rjw: Modified the subject and changelog slightly.]
Signed-off-by: Jingoo Han <jg1.han@samsung.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
__cpufreq_driver_target() must not pass target frequency beyond the
limits of current policy.
Today most of cpufreq platform drivers are doing this check in their
target routines. Why not move it to __cpufreq_driver_target()?
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Avoid calling cpufreq driver's target() routine if new frequency is same as
policies current frequency.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
cpufreq_disabled() is a local function, so should be marked static.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
There is no need to do cpufreq_get_cpu() and cpufreq_put_cpu() for drivers that
don't support getavg() routine.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
With debug options on, it is difficult to locate cpufreq core's debug prints.
Fix this by prefixing debug prints with KBUILD_MODNAME.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Arrays for governer and driver name are of size CPUFREQ_NAME_LEN or 16.
i.e. 15 bytes for name and 1 for trailing '\0'.
When cpufreq driver print these names (for sysfs), it includes '\n' or ' ' in
the fmt string and still passes length as CPUFREQ_NAME_LEN. If the driver or
governor names are using all 15 fields allocated to them, then the trailing '\n'
or ' ' will never be printed. And so commands like:
root@linaro-developer# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_driver
will print something like:
cpufreq_foodrvroot@linaro-developer#
Fix this by increasing print length by one character.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
show_bios_limit is mistakenly written as show_scaling_driver in a comment
describing purpose of show_bios_limit() routine.
Fix it.
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Running one program that continuously hotplugs and replugs a cpu
concurrently with another program that continuously writes to the
scaling_setspeed node eventually deadlocks with:
=============================================
[ INFO: possible recursive locking detected ]
3.4.0 #37 Tainted: G W
---------------------------------------------
filemonkey/122 is trying to acquire lock:
(s_active#13){++++.+}, at: [<c01a3d28>] sysfs_remove_dir+0x9c/0xb4
but task is already holding lock:
(s_active#13){++++.+}, at: [<c01a22f0>] sysfs_write_file+0xe8/0x140
other info that might help us debug this:
Possible unsafe locking scenario:
CPU0
----
lock(s_active#13);
lock(s_active#13);
*** DEADLOCK ***
May be due to missing lock nesting notation
2 locks held by filemonkey/122:
#0: (&buffer->mutex){+.+.+.}, at: [<c01a2230>] sysfs_write_file+0x28/0x140
#1: (s_active#13){++++.+}, at: [<c01a22f0>] sysfs_write_file+0xe8/0x140
stack backtrace:
[<c0014fcc>] (unwind_backtrace+0x0/0x120) from [<c00ca600>] (validate_chain+0x6f8/0x1054)
[<c00ca600>] (validate_chain+0x6f8/0x1054) from [<c00cb778>] (__lock_acquire+0x81c/0x8d8)
[<c00cb778>] (__lock_acquire+0x81c/0x8d8) from [<c00cb9c0>] (lock_acquire+0x18c/0x1e8)
[<c00cb9c0>] (lock_acquire+0x18c/0x1e8) from [<c01a3ba8>] (sysfs_addrm_finish+0xd0/0x180)
[<c01a3ba8>] (sysfs_addrm_finish+0xd0/0x180) from [<c01a3d28>] (sysfs_remove_dir+0x9c/0xb4)
[<c01a3d28>] (sysfs_remove_dir+0x9c/0xb4) from [<c02d0e5c>] (kobject_del+0x10/0x38)
[<c02d0e5c>] (kobject_del+0x10/0x38) from [<c02d0f74>] (kobject_release+0xf0/0x194)
[<c02d0f74>] (kobject_release+0xf0/0x194) from [<c0565a98>] (cpufreq_cpu_put+0xc/0x24)
[<c0565a98>] (cpufreq_cpu_put+0xc/0x24) from [<c05683f0>] (store+0x6c/0x74)
[<c05683f0>] (store+0x6c/0x74) from [<c01a2314>] (sysfs_write_file+0x10c/0x140)
[<c01a2314>] (sysfs_write_file+0x10c/0x140) from [<c014af44>] (vfs_write+0xb0/0x128)
[<c014af44>] (vfs_write+0xb0/0x128) from [<c014b06c>] (sys_write+0x3c/0x68)
[<c014b06c>] (sys_write+0x3c/0x68) from [<c000e0e0>] (ret_fast_syscall+0x0/0x3c)
This is because store() in cpufreq.c indirectly calls
kobject_get() via cpufreq_cpu_get() and is the last one to call
kobject_put() via cpufreq_cpu_put(). Sysfs code should not call
kobject_get() or kobject_put() directly (see the comment around
sysfs_schedule_callback() for more information).
Fix this deadlock by introducing two new functions:
struct cpufreq_policy *cpufreq_cpu_get_sysfs(unsigned int cpu)
void cpufreq_cpu_put_sysfs(struct cpufreq_policy *data)
which do the same thing as cpufreq_cpu_{get,put}() but don't call
kobject functions.
To easily trigger this deadlock you can insert an msleep() with a
reasonably large value right after the fail label at the bottom
of the store() function in cpufreq.c and then write
scaling_setspeed in one task and offline the cpu in another. The
first task will hang and be detected by the hung task detector.
Signed-off-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
useful for disabling cpufreq altogether. The cpu frequency
scaling drivers and cpu frequency governors will fail to register.
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Dave Jones <davej@redhat.com>
