2005-04-17 06:20:36 +08:00
|
|
|
/*
|
|
|
|
* linux/drivers/cpufreq/cpufreq.c
|
|
|
|
*
|
|
|
|
* Copyright (C) 2001 Russell King
|
|
|
|
* (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
|
2013-06-19 16:49:33 +08:00
|
|
|
* (C) 2013 Viresh Kumar <viresh.kumar@linaro.org>
|
2005-04-17 06:20:36 +08:00
|
|
|
*
|
2005-10-31 06:59:54 +08:00
|
|
|
* Oct 2005 - Ashok Raj <ashok.raj@intel.com>
|
2006-02-28 13:43:23 +08:00
|
|
|
* Added handling for CPU hotplug
|
2006-03-05 16:37:23 +08:00
|
|
|
* Feb 2006 - Jacob Shin <jacob.shin@amd.com>
|
|
|
|
* Fix handling for CPU hotplug -- affected CPUs
|
2005-10-31 06:59:54 +08:00
|
|
|
*
|
2005-04-17 06:20:36 +08:00
|
|
|
* This program is free software; you can redistribute it and/or modify
|
|
|
|
* it under the terms of the GNU General Public License version 2 as
|
|
|
|
* published by the Free Software Foundation.
|
|
|
|
*/
|
|
|
|
|
2012-10-23 07:29:03 +08:00
|
|
|
#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
|
|
|
|
|
2013-08-07 01:23:03 +08:00
|
|
|
#include <linux/cpu.h>
|
2005-04-17 06:20:36 +08:00
|
|
|
#include <linux/cpufreq.h>
|
|
|
|
#include <linux/delay.h>
|
|
|
|
#include <linux/device.h>
|
2013-08-07 01:23:03 +08:00
|
|
|
#include <linux/init.h>
|
|
|
|
#include <linux/kernel_stat.h>
|
|
|
|
#include <linux/module.h>
|
2006-01-14 07:54:22 +08:00
|
|
|
#include <linux/mutex.h>
|
2013-08-07 01:23:03 +08:00
|
|
|
#include <linux/slab.h>
|
2014-03-04 11:00:26 +08:00
|
|
|
#include <linux/suspend.h>
|
2014-12-24 14:09:48 +08:00
|
|
|
#include <linux/syscore_ops.h>
|
2013-08-07 01:23:03 +08:00
|
|
|
#include <linux/tick.h>
|
2010-04-20 19:17:36 +08:00
|
|
|
#include <trace/events/power.h>
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/**
|
2006-08-12 05:59:28 +08:00
|
|
|
* The "cpufreq driver" - the arch- or hardware-dependent low
|
2005-04-17 06:20:36 +08:00
|
|
|
* level driver of CPUFreq support, and its spinlock. This lock
|
|
|
|
* also protects the cpufreq_cpu_data array.
|
|
|
|
*/
|
2013-04-29 06:08:16 +08:00
|
|
|
static struct cpufreq_driver *cpufreq_driver;
|
2008-03-26 06:06:53 +08:00
|
|
|
static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
|
cpufreq: Preserve policy structure across suspend/resume
To perform light-weight cpu-init and teardown in the cpufreq subsystem
during suspend/resume, we need to separate out the 2 main functionalities
of the cpufreq CPU hotplug callbacks, as outlined below:
1. Init/tear-down of core cpufreq and CPU-specific components, which are
critical to the correct functioning of the cpufreq subsystem.
2. Init/tear-down of cpufreq sysfs files during suspend/resume.
The first part requires accurate updates to the policy structure such as
its ->cpus and ->related_cpus masks, whereas the second part requires that
the policy->kobj structure is not released or re-initialized during
suspend/resume.
To handle both these requirements, we need to allow updates to the policy
structure throughout suspend/resume, but prevent the structure from getting
freed up. Also, we must have a mechanism by which the cpu-up callbacks can
restore the policy structure, without allocating things afresh. (That also
helps avoid memory leaks).
To achieve this, we use 2 schemes:
a. Use a fallback per-cpu storage area for preserving the policy structures
during suspend, so that they can be restored during resume appropriately.
b. Use the 'frozen' flag to determine when to free or allocate the policy
structure vs when to restore the policy from the saved fallback storage.
Thus we can successfully preserve the structure across suspend/resume.
Effectively, this helps us complete the separation of the 'light-weight'
and the 'full' init/tear-down sequences in the cpufreq subsystem, so that
this can be made use of in the suspend/resume scenario.
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>
2013-07-30 06:55:10 +08:00
|
|
|
static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data_fallback);
|
2013-06-19 16:49:33 +08:00
|
|
|
static DEFINE_RWLOCK(cpufreq_driver_lock);
|
2014-01-03 17:17:41 +08:00
|
|
|
DEFINE_MUTEX(cpufreq_governor_lock);
|
2013-08-07 01:23:08 +08:00
|
|
|
static LIST_HEAD(cpufreq_policy_list);
|
2013-06-19 16:49:33 +08:00
|
|
|
|
2007-07-10 02:35:28 +08:00
|
|
|
/* This one keeps track of the previously set governor of a removed CPU */
|
2009-10-05 04:38:55 +08:00
|
|
|
static DEFINE_PER_CPU(char[CPUFREQ_NAME_LEN], cpufreq_cpu_governor);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-03-04 11:00:26 +08:00
|
|
|
/* Flag to suspend/resume CPUFreq governors */
|
|
|
|
static bool cpufreq_suspended;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-10-25 22:15:48 +08:00
|
|
|
static inline bool has_target(void)
|
|
|
|
{
|
|
|
|
return cpufreq_driver->target_index || cpufreq_driver->target;
|
|
|
|
}
|
|
|
|
|
2013-08-07 01:23:11 +08:00
|
|
|
/*
|
|
|
|
* rwsem to guarantee that cpufreq driver module doesn't unload during critical
|
|
|
|
* sections
|
|
|
|
*/
|
|
|
|
static DECLARE_RWSEM(cpufreq_rwsem);
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/* internal prototypes */
|
2009-01-18 14:37:11 +08:00
|
|
|
static int __cpufreq_governor(struct cpufreq_policy *policy,
|
|
|
|
unsigned int event);
|
2015-01-02 15:04:29 +08:00
|
|
|
static unsigned int __cpufreq_get(struct cpufreq_policy *policy);
|
2006-11-22 22:55:48 +08:00
|
|
|
static void handle_update(struct work_struct *work);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/**
|
2006-02-28 13:43:23 +08:00
|
|
|
* Two notifier lists: the "policy" list is involved in the
|
|
|
|
* validation process for a new CPU frequency policy; the
|
2005-04-17 06:20:36 +08:00
|
|
|
* "transition" list for kernel code that needs to handle
|
|
|
|
* changes to devices when the CPU clock speed changes.
|
|
|
|
* The mutex locks both lists.
|
|
|
|
*/
|
[PATCH] Notifier chain update: API changes
The kernel's implementation of notifier chains is unsafe. There is no
protection against entries being added to or removed from a chain while the
chain is in use. The issues were discussed in this thread:
http://marc.theaimsgroup.com/?l=linux-kernel&m=113018709002036&w=2
We noticed that notifier chains in the kernel fall into two basic usage
classes:
"Blocking" chains are always called from a process context
and the callout routines are allowed to sleep;
"Atomic" chains can be called from an atomic context and
the callout routines are not allowed to sleep.
We decided to codify this distinction and make it part of the API. Therefore
this set of patches introduces three new, parallel APIs: one for blocking
notifiers, one for atomic notifiers, and one for "raw" notifiers (which is
really just the old API under a new name). New kinds of data structures are
used for the heads of the chains, and new routines are defined for
registration, unregistration, and calling a chain. The three APIs are
explained in include/linux/notifier.h and their implementation is in
kernel/sys.c.
With atomic and blocking chains, the implementation guarantees that the chain
links will not be corrupted and that chain callers will not get messed up by
entries being added or removed. For raw chains the implementation provides no
guarantees at all; users of this API must provide their own protections. (The
idea was that situations may come up where the assumptions of the atomic and
blocking APIs are not appropriate, so it should be possible for users to
handle these things in their own way.)
There are some limitations, which should not be too hard to live with. For
atomic/blocking chains, registration and unregistration must always be done in
a process context since the chain is protected by a mutex/rwsem. Also, a
callout routine for a non-raw chain must not try to register or unregister
entries on its own chain. (This did happen in a couple of places and the code
had to be changed to avoid it.)
Since atomic chains may be called from within an NMI handler, they cannot use
spinlocks for synchronization. Instead we use RCU. The overhead falls almost
entirely in the unregister routine, which is okay since unregistration is much
less frequent that calling a chain.
Here is the list of chains that we adjusted and their classifications. None
of them use the raw API, so for the moment it is only a placeholder.
ATOMIC CHAINS
-------------
arch/i386/kernel/traps.c: i386die_chain
arch/ia64/kernel/traps.c: ia64die_chain
arch/powerpc/kernel/traps.c: powerpc_die_chain
arch/sparc64/kernel/traps.c: sparc64die_chain
arch/x86_64/kernel/traps.c: die_chain
drivers/char/ipmi/ipmi_si_intf.c: xaction_notifier_list
kernel/panic.c: panic_notifier_list
kernel/profile.c: task_free_notifier
net/bluetooth/hci_core.c: hci_notifier
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_chain
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_expect_chain
net/ipv6/addrconf.c: inet6addr_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_expect_chain
net/netlink/af_netlink.c: netlink_chain
BLOCKING CHAINS
---------------
arch/powerpc/platforms/pseries/reconfig.c: pSeries_reconfig_chain
arch/s390/kernel/process.c: idle_chain
arch/x86_64/kernel/process.c idle_notifier
drivers/base/memory.c: memory_chain
drivers/cpufreq/cpufreq.c cpufreq_policy_notifier_list
drivers/cpufreq/cpufreq.c cpufreq_transition_notifier_list
drivers/macintosh/adb.c: adb_client_list
drivers/macintosh/via-pmu.c sleep_notifier_list
drivers/macintosh/via-pmu68k.c sleep_notifier_list
drivers/macintosh/windfarm_core.c wf_client_list
drivers/usb/core/notify.c usb_notifier_list
drivers/video/fbmem.c fb_notifier_list
kernel/cpu.c cpu_chain
kernel/module.c module_notify_list
kernel/profile.c munmap_notifier
kernel/profile.c task_exit_notifier
kernel/sys.c reboot_notifier_list
net/core/dev.c netdev_chain
net/decnet/dn_dev.c: dnaddr_chain
net/ipv4/devinet.c: inetaddr_chain
It's possible that some of these classifications are wrong. If they are,
please let us know or submit a patch to fix them. Note that any chain that
gets called very frequently should be atomic, because the rwsem read-locking
used for blocking chains is very likely to incur cache misses on SMP systems.
(However, if the chain's callout routines may sleep then the chain cannot be
atomic.)
The patch set was written by Alan Stern and Chandra Seetharaman, incorporating
material written by Keith Owens and suggestions from Paul McKenney and Andrew
Morton.
[jes@sgi.com: restructure the notifier chain initialization macros]
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: Jes Sorensen <jes@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-27 17:16:30 +08:00
|
|
|
static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
|
2006-10-04 17:17:06 +08:00
|
|
|
static struct srcu_notifier_head cpufreq_transition_notifier_list;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2008-02-16 18:41:24 +08:00
|
|
|
static bool init_cpufreq_transition_notifier_list_called;
|
2006-10-04 17:17:06 +08:00
|
|
|
static int __init init_cpufreq_transition_notifier_list(void)
|
|
|
|
{
|
|
|
|
srcu_init_notifier_head(&cpufreq_transition_notifier_list);
|
2008-02-16 18:41:24 +08:00
|
|
|
init_cpufreq_transition_notifier_list_called = true;
|
2006-10-04 17:17:06 +08:00
|
|
|
return 0;
|
|
|
|
}
|
2006-11-21 03:47:18 +08:00
|
|
|
pure_initcall(init_cpufreq_transition_notifier_list);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2012-03-14 07:18:39 +08:00
|
|
|
static int off __read_mostly;
|
2012-10-26 06:51:32 +08:00
|
|
|
static int cpufreq_disabled(void)
|
2012-03-14 07:18:39 +08:00
|
|
|
{
|
|
|
|
return off;
|
|
|
|
}
|
|
|
|
void disable_cpufreq(void)
|
|
|
|
{
|
|
|
|
off = 1;
|
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
static LIST_HEAD(cpufreq_governor_list);
|
2009-01-18 14:37:11 +08:00
|
|
|
static DEFINE_MUTEX(cpufreq_governor_mutex);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-03-27 23:58:58 +08:00
|
|
|
bool have_governor_per_policy(void)
|
|
|
|
{
|
2013-10-02 16:43:18 +08:00
|
|
|
return !!(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY);
|
2013-03-27 23:58:58 +08:00
|
|
|
}
|
2013-05-16 13:09:56 +08:00
|
|
|
EXPORT_SYMBOL_GPL(have_governor_per_policy);
|
2013-03-27 23:58:58 +08:00
|
|
|
|
2013-05-16 13:09:57 +08:00
|
|
|
struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
|
|
|
|
{
|
|
|
|
if (have_governor_per_policy())
|
|
|
|
return &policy->kobj;
|
|
|
|
else
|
|
|
|
return cpufreq_global_kobject;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
|
|
|
|
|
2013-05-17 19:26:32 +08:00
|
|
|
static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
|
|
|
|
{
|
|
|
|
u64 idle_time;
|
|
|
|
u64 cur_wall_time;
|
|
|
|
u64 busy_time;
|
|
|
|
|
|
|
|
cur_wall_time = jiffies64_to_cputime64(get_jiffies_64());
|
|
|
|
|
|
|
|
busy_time = kcpustat_cpu(cpu).cpustat[CPUTIME_USER];
|
|
|
|
busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SYSTEM];
|
|
|
|
busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_IRQ];
|
|
|
|
busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SOFTIRQ];
|
|
|
|
busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_STEAL];
|
|
|
|
busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_NICE];
|
|
|
|
|
|
|
|
idle_time = cur_wall_time - busy_time;
|
|
|
|
if (wall)
|
|
|
|
*wall = cputime_to_usecs(cur_wall_time);
|
|
|
|
|
|
|
|
return cputime_to_usecs(idle_time);
|
|
|
|
}
|
|
|
|
|
|
|
|
u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
|
|
|
|
{
|
|
|
|
u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
|
|
|
|
|
|
|
|
if (idle_time == -1ULL)
|
|
|
|
return get_cpu_idle_time_jiffy(cpu, wall);
|
|
|
|
else if (!io_busy)
|
|
|
|
idle_time += get_cpu_iowait_time_us(cpu, wall);
|
|
|
|
|
|
|
|
return idle_time;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(get_cpu_idle_time);
|
|
|
|
|
2013-10-03 22:59:07 +08:00
|
|
|
/*
|
|
|
|
* This is a generic cpufreq init() routine which can be used by cpufreq
|
|
|
|
* drivers of SMP systems. It will do following:
|
|
|
|
* - validate & show freq table passed
|
|
|
|
* - set policies transition latency
|
|
|
|
* - policy->cpus with all possible CPUs
|
|
|
|
*/
|
|
|
|
int cpufreq_generic_init(struct cpufreq_policy *policy,
|
|
|
|
struct cpufreq_frequency_table *table,
|
|
|
|
unsigned int transition_latency)
|
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
ret = cpufreq_table_validate_and_show(policy, table);
|
|
|
|
if (ret) {
|
|
|
|
pr_err("%s: invalid frequency table: %d\n", __func__, ret);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
policy->cpuinfo.transition_latency = transition_latency;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* The driver only supports the SMP configuartion where all processors
|
|
|
|
* share the clock and voltage and clock.
|
|
|
|
*/
|
|
|
|
cpumask_setall(policy->cpus);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_generic_init);
|
|
|
|
|
2014-01-09 23:08:43 +08:00
|
|
|
unsigned int cpufreq_generic_get(unsigned int cpu)
|
|
|
|
{
|
|
|
|
struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
|
|
|
|
|
|
|
|
if (!policy || IS_ERR(policy->clk)) {
|
2014-03-12 01:03:00 +08:00
|
|
|
pr_err("%s: No %s associated to cpu: %d\n",
|
|
|
|
__func__, policy ? "clk" : "policy", cpu);
|
2014-01-09 23:08:43 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
return clk_get_rate(policy->clk) / 1000;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_generic_get);
|
|
|
|
|
2014-03-10 17:23:33 +08:00
|
|
|
/* Only for cpufreq core internal use */
|
|
|
|
struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
|
|
|
|
{
|
|
|
|
return per_cpu(cpufreq_cpu_data, cpu);
|
|
|
|
}
|
|
|
|
|
2013-08-07 01:23:11 +08:00
|
|
|
struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2013-08-07 01:23:11 +08:00
|
|
|
struct cpufreq_policy *policy = NULL;
|
2005-04-17 06:20:36 +08:00
|
|
|
unsigned long flags;
|
|
|
|
|
2013-08-07 01:23:11 +08:00
|
|
|
if (cpufreq_disabled() || (cpu >= nr_cpu_ids))
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
if (!down_read_trylock(&cpufreq_rwsem))
|
|
|
|
return NULL;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/* get the cpufreq driver */
|
2013-04-29 06:08:16 +08:00
|
|
|
read_lock_irqsave(&cpufreq_driver_lock, flags);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-08-07 01:23:11 +08:00
|
|
|
if (cpufreq_driver) {
|
|
|
|
/* get the CPU */
|
|
|
|
policy = per_cpu(cpufreq_cpu_data, cpu);
|
|
|
|
if (policy)
|
|
|
|
kobject_get(&policy->kobj);
|
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-08-07 01:23:11 +08:00
|
|
|
read_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-08-07 01:23:05 +08:00
|
|
|
if (!policy)
|
2013-08-07 01:23:11 +08:00
|
|
|
up_read(&cpufreq_rwsem);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-08-07 01:23:05 +08:00
|
|
|
return policy;
|
2012-07-21 02:14:38 +08:00
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
|
|
|
|
|
2013-08-07 01:23:05 +08:00
|
|
|
void cpufreq_cpu_put(struct cpufreq_policy *policy)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2013-01-18 00:22:21 +08:00
|
|
|
if (cpufreq_disabled())
|
|
|
|
return;
|
|
|
|
|
2013-08-07 01:23:11 +08:00
|
|
|
kobject_put(&policy->kobj);
|
|
|
|
up_read(&cpufreq_rwsem);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
|
|
|
|
|
|
|
|
/*********************************************************************
|
|
|
|
* EXTERNALLY AFFECTING FREQUENCY CHANGES *
|
|
|
|
*********************************************************************/
|
|
|
|
|
|
|
|
/**
|
|
|
|
* adjust_jiffies - adjust the system "loops_per_jiffy"
|
|
|
|
*
|
|
|
|
* This function alters the system "loops_per_jiffy" for the clock
|
|
|
|
* speed change. Note that loops_per_jiffy cannot be updated on SMP
|
2006-02-28 13:43:23 +08:00
|
|
|
* systems as each CPU might be scaled differently. So, use the arch
|
2005-04-17 06:20:36 +08:00
|
|
|
* per-CPU loops_per_jiffy value wherever possible.
|
|
|
|
*/
|
2006-01-15 05:20:43 +08:00
|
|
|
static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2015-01-02 15:04:34 +08:00
|
|
|
#ifndef CONFIG_SMP
|
|
|
|
static unsigned long l_p_j_ref;
|
|
|
|
static unsigned int l_p_j_ref_freq;
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
if (ci->flags & CPUFREQ_CONST_LOOPS)
|
|
|
|
return;
|
|
|
|
|
|
|
|
if (!l_p_j_ref_freq) {
|
|
|
|
l_p_j_ref = loops_per_jiffy;
|
|
|
|
l_p_j_ref_freq = ci->old;
|
2014-03-12 01:03:00 +08:00
|
|
|
pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
|
|
|
|
l_p_j_ref, l_p_j_ref_freq);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2014-03-19 13:54:58 +08:00
|
|
|
if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
|
2006-10-26 18:50:58 +08:00
|
|
|
loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
|
|
|
|
ci->new);
|
2014-03-12 01:03:00 +08:00
|
|
|
pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
|
|
|
|
loops_per_jiffy, ci->new);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
#endif
|
2015-01-02 15:04:34 +08:00
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-06-19 16:49:34 +08:00
|
|
|
static void __cpufreq_notify_transition(struct cpufreq_policy *policy,
|
2013-03-24 14:26:43 +08:00
|
|
|
struct cpufreq_freqs *freqs, unsigned int state)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
BUG_ON(irqs_disabled());
|
|
|
|
|
2013-01-18 00:22:21 +08:00
|
|
|
if (cpufreq_disabled())
|
|
|
|
return;
|
|
|
|
|
2013-04-29 06:08:16 +08:00
|
|
|
freqs->flags = cpufreq_driver->flags;
|
2011-03-27 21:04:46 +08:00
|
|
|
pr_debug("notification %u of frequency transition to %u kHz\n",
|
2014-03-12 01:03:00 +08:00
|
|
|
state, freqs->new);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
switch (state) {
|
2006-02-01 07:53:55 +08:00
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
case CPUFREQ_PRECHANGE:
|
2006-02-28 13:43:23 +08:00
|
|
|
/* detect if the driver reported a value as "old frequency"
|
2006-02-01 07:53:55 +08:00
|
|
|
* which is not equal to what the cpufreq core thinks is
|
|
|
|
* "old frequency".
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
2013-04-29 06:08:16 +08:00
|
|
|
if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
|
2006-02-01 07:53:55 +08:00
|
|
|
if ((policy) && (policy->cpu == freqs->cpu) &&
|
|
|
|
(policy->cur) && (policy->cur != freqs->old)) {
|
2014-03-12 01:03:00 +08:00
|
|
|
pr_debug("Warning: CPU frequency is %u, cpufreq assumed %u kHz\n",
|
|
|
|
freqs->old, policy->cur);
|
2006-02-01 07:53:55 +08:00
|
|
|
freqs->old = policy->cur;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
}
|
2006-10-04 17:17:06 +08:00
|
|
|
srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
|
[PATCH] Notifier chain update: API changes
The kernel's implementation of notifier chains is unsafe. There is no
protection against entries being added to or removed from a chain while the
chain is in use. The issues were discussed in this thread:
http://marc.theaimsgroup.com/?l=linux-kernel&m=113018709002036&w=2
We noticed that notifier chains in the kernel fall into two basic usage
classes:
"Blocking" chains are always called from a process context
and the callout routines are allowed to sleep;
"Atomic" chains can be called from an atomic context and
the callout routines are not allowed to sleep.
We decided to codify this distinction and make it part of the API. Therefore
this set of patches introduces three new, parallel APIs: one for blocking
notifiers, one for atomic notifiers, and one for "raw" notifiers (which is
really just the old API under a new name). New kinds of data structures are
used for the heads of the chains, and new routines are defined for
registration, unregistration, and calling a chain. The three APIs are
explained in include/linux/notifier.h and their implementation is in
kernel/sys.c.
With atomic and blocking chains, the implementation guarantees that the chain
links will not be corrupted and that chain callers will not get messed up by
entries being added or removed. For raw chains the implementation provides no
guarantees at all; users of this API must provide their own protections. (The
idea was that situations may come up where the assumptions of the atomic and
blocking APIs are not appropriate, so it should be possible for users to
handle these things in their own way.)
There are some limitations, which should not be too hard to live with. For
atomic/blocking chains, registration and unregistration must always be done in
a process context since the chain is protected by a mutex/rwsem. Also, a
callout routine for a non-raw chain must not try to register or unregister
entries on its own chain. (This did happen in a couple of places and the code
had to be changed to avoid it.)
Since atomic chains may be called from within an NMI handler, they cannot use
spinlocks for synchronization. Instead we use RCU. The overhead falls almost
entirely in the unregister routine, which is okay since unregistration is much
less frequent that calling a chain.
Here is the list of chains that we adjusted and their classifications. None
of them use the raw API, so for the moment it is only a placeholder.
ATOMIC CHAINS
-------------
arch/i386/kernel/traps.c: i386die_chain
arch/ia64/kernel/traps.c: ia64die_chain
arch/powerpc/kernel/traps.c: powerpc_die_chain
arch/sparc64/kernel/traps.c: sparc64die_chain
arch/x86_64/kernel/traps.c: die_chain
drivers/char/ipmi/ipmi_si_intf.c: xaction_notifier_list
kernel/panic.c: panic_notifier_list
kernel/profile.c: task_free_notifier
net/bluetooth/hci_core.c: hci_notifier
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_chain
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_expect_chain
net/ipv6/addrconf.c: inet6addr_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_expect_chain
net/netlink/af_netlink.c: netlink_chain
BLOCKING CHAINS
---------------
arch/powerpc/platforms/pseries/reconfig.c: pSeries_reconfig_chain
arch/s390/kernel/process.c: idle_chain
arch/x86_64/kernel/process.c idle_notifier
drivers/base/memory.c: memory_chain
drivers/cpufreq/cpufreq.c cpufreq_policy_notifier_list
drivers/cpufreq/cpufreq.c cpufreq_transition_notifier_list
drivers/macintosh/adb.c: adb_client_list
drivers/macintosh/via-pmu.c sleep_notifier_list
drivers/macintosh/via-pmu68k.c sleep_notifier_list
drivers/macintosh/windfarm_core.c wf_client_list
drivers/usb/core/notify.c usb_notifier_list
drivers/video/fbmem.c fb_notifier_list
kernel/cpu.c cpu_chain
kernel/module.c module_notify_list
kernel/profile.c munmap_notifier
kernel/profile.c task_exit_notifier
kernel/sys.c reboot_notifier_list
net/core/dev.c netdev_chain
net/decnet/dn_dev.c: dnaddr_chain
net/ipv4/devinet.c: inetaddr_chain
It's possible that some of these classifications are wrong. If they are,
please let us know or submit a patch to fix them. Note that any chain that
gets called very frequently should be atomic, because the rwsem read-locking
used for blocking chains is very likely to incur cache misses on SMP systems.
(However, if the chain's callout routines may sleep then the chain cannot be
atomic.)
The patch set was written by Alan Stern and Chandra Seetharaman, incorporating
material written by Keith Owens and suggestions from Paul McKenney and Andrew
Morton.
[jes@sgi.com: restructure the notifier chain initialization macros]
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: Jes Sorensen <jes@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-27 17:16:30 +08:00
|
|
|
CPUFREQ_PRECHANGE, freqs);
|
2005-04-17 06:20:36 +08:00
|
|
|
adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
|
|
|
|
break;
|
2006-02-01 07:53:55 +08:00
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
case CPUFREQ_POSTCHANGE:
|
|
|
|
adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
|
2014-03-12 01:03:00 +08:00
|
|
|
pr_debug("FREQ: %lu - CPU: %lu\n",
|
|
|
|
(unsigned long)freqs->new, (unsigned long)freqs->cpu);
|
2011-01-04 00:50:44 +08:00
|
|
|
trace_cpu_frequency(freqs->new, freqs->cpu);
|
2006-10-04 17:17:06 +08:00
|
|
|
srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
|
[PATCH] Notifier chain update: API changes
The kernel's implementation of notifier chains is unsafe. There is no
protection against entries being added to or removed from a chain while the
chain is in use. The issues were discussed in this thread:
http://marc.theaimsgroup.com/?l=linux-kernel&m=113018709002036&w=2
We noticed that notifier chains in the kernel fall into two basic usage
classes:
"Blocking" chains are always called from a process context
and the callout routines are allowed to sleep;
"Atomic" chains can be called from an atomic context and
the callout routines are not allowed to sleep.
We decided to codify this distinction and make it part of the API. Therefore
this set of patches introduces three new, parallel APIs: one for blocking
notifiers, one for atomic notifiers, and one for "raw" notifiers (which is
really just the old API under a new name). New kinds of data structures are
used for the heads of the chains, and new routines are defined for
registration, unregistration, and calling a chain. The three APIs are
explained in include/linux/notifier.h and their implementation is in
kernel/sys.c.
With atomic and blocking chains, the implementation guarantees that the chain
links will not be corrupted and that chain callers will not get messed up by
entries being added or removed. For raw chains the implementation provides no
guarantees at all; users of this API must provide their own protections. (The
idea was that situations may come up where the assumptions of the atomic and
blocking APIs are not appropriate, so it should be possible for users to
handle these things in their own way.)
There are some limitations, which should not be too hard to live with. For
atomic/blocking chains, registration and unregistration must always be done in
a process context since the chain is protected by a mutex/rwsem. Also, a
callout routine for a non-raw chain must not try to register or unregister
entries on its own chain. (This did happen in a couple of places and the code
had to be changed to avoid it.)
Since atomic chains may be called from within an NMI handler, they cannot use
spinlocks for synchronization. Instead we use RCU. The overhead falls almost
entirely in the unregister routine, which is okay since unregistration is much
less frequent that calling a chain.
Here is the list of chains that we adjusted and their classifications. None
of them use the raw API, so for the moment it is only a placeholder.
ATOMIC CHAINS
-------------
arch/i386/kernel/traps.c: i386die_chain
arch/ia64/kernel/traps.c: ia64die_chain
arch/powerpc/kernel/traps.c: powerpc_die_chain
arch/sparc64/kernel/traps.c: sparc64die_chain
arch/x86_64/kernel/traps.c: die_chain
drivers/char/ipmi/ipmi_si_intf.c: xaction_notifier_list
kernel/panic.c: panic_notifier_list
kernel/profile.c: task_free_notifier
net/bluetooth/hci_core.c: hci_notifier
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_chain
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_expect_chain
net/ipv6/addrconf.c: inet6addr_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_expect_chain
net/netlink/af_netlink.c: netlink_chain
BLOCKING CHAINS
---------------
arch/powerpc/platforms/pseries/reconfig.c: pSeries_reconfig_chain
arch/s390/kernel/process.c: idle_chain
arch/x86_64/kernel/process.c idle_notifier
drivers/base/memory.c: memory_chain
drivers/cpufreq/cpufreq.c cpufreq_policy_notifier_list
drivers/cpufreq/cpufreq.c cpufreq_transition_notifier_list
drivers/macintosh/adb.c: adb_client_list
drivers/macintosh/via-pmu.c sleep_notifier_list
drivers/macintosh/via-pmu68k.c sleep_notifier_list
drivers/macintosh/windfarm_core.c wf_client_list
drivers/usb/core/notify.c usb_notifier_list
drivers/video/fbmem.c fb_notifier_list
kernel/cpu.c cpu_chain
kernel/module.c module_notify_list
kernel/profile.c munmap_notifier
kernel/profile.c task_exit_notifier
kernel/sys.c reboot_notifier_list
net/core/dev.c netdev_chain
net/decnet/dn_dev.c: dnaddr_chain
net/ipv4/devinet.c: inetaddr_chain
It's possible that some of these classifications are wrong. If they are,
please let us know or submit a patch to fix them. Note that any chain that
gets called very frequently should be atomic, because the rwsem read-locking
used for blocking chains is very likely to incur cache misses on SMP systems.
(However, if the chain's callout routines may sleep then the chain cannot be
atomic.)
The patch set was written by Alan Stern and Chandra Seetharaman, incorporating
material written by Keith Owens and suggestions from Paul McKenney and Andrew
Morton.
[jes@sgi.com: restructure the notifier chain initialization macros]
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: Jes Sorensen <jes@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-27 17:16:30 +08:00
|
|
|
CPUFREQ_POSTCHANGE, freqs);
|
2006-02-01 07:53:55 +08:00
|
|
|
if (likely(policy) && likely(policy->cpu == freqs->cpu))
|
|
|
|
policy->cur = freqs->new;
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
2013-06-19 16:49:33 +08:00
|
|
|
|
2013-03-24 14:26:43 +08:00
|
|
|
/**
|
|
|
|
* cpufreq_notify_transition - call notifier chain and adjust_jiffies
|
|
|
|
* on frequency transition.
|
|
|
|
*
|
|
|
|
* This function calls the transition notifiers and the "adjust_jiffies"
|
|
|
|
* function. It is called twice on all CPU frequency changes that have
|
|
|
|
* external effects.
|
|
|
|
*/
|
2014-03-24 16:05:46 +08:00
|
|
|
static void cpufreq_notify_transition(struct cpufreq_policy *policy,
|
2013-03-24 14:26:43 +08:00
|
|
|
struct cpufreq_freqs *freqs, unsigned int state)
|
|
|
|
{
|
|
|
|
for_each_cpu(freqs->cpu, policy->cpus)
|
|
|
|
__cpufreq_notify_transition(policy, freqs, state);
|
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-12-02 13:34:12 +08:00
|
|
|
/* Do post notifications when there are chances that transition has failed */
|
2014-03-24 16:05:46 +08:00
|
|
|
static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
|
2013-12-02 13:34:12 +08:00
|
|
|
struct cpufreq_freqs *freqs, int transition_failed)
|
|
|
|
{
|
|
|
|
cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
|
|
|
|
if (!transition_failed)
|
|
|
|
return;
|
|
|
|
|
|
|
|
swap(freqs->old, freqs->new);
|
|
|
|
cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
|
|
|
|
cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
|
|
|
|
}
|
|
|
|
|
cpufreq: Make sure frequency transitions are serialized
Whenever we change the frequency of a CPU, we call the PRECHANGE and POSTCHANGE
notifiers. They must be serialized, i.e. PRECHANGE and POSTCHANGE notifiers
should strictly alternate, thereby preventing two different sets of PRECHANGE or
POSTCHANGE notifiers from interleaving arbitrarily.
The following examples illustrate why this is important:
Scenario 1:
-----------
A thread reading the value of cpuinfo_cur_freq, will call
__cpufreq_cpu_get()->cpufreq_out_of_sync()->cpufreq_notify_transition()
The ondemand governor can decide to change the frequency of the CPU at the same
time and hence it can end up sending the notifications via ->target().
If the notifiers are not serialized, the following sequence can occur:
- PRECHANGE Notification for freq A (from cpuinfo_cur_freq)
- PRECHANGE Notification for freq B (from target())
- Freq changed by target() to B
- POSTCHANGE Notification for freq B
- POSTCHANGE Notification for freq A
We can see from the above that the last POSTCHANGE Notification happens for freq
A but the hardware is set to run at freq B.
Where would we break then?: adjust_jiffies() in cpufreq.c & cpufreq_callback()
in arch/arm/kernel/smp.c (which also adjusts the jiffies). All the
loops_per_jiffy calculations will get messed up.
Scenario 2:
-----------
The governor calls __cpufreq_driver_target() to change the frequency. At the
same time, if we change scaling_{min|max}_freq from sysfs, it will end up
calling the governor's CPUFREQ_GOV_LIMITS notification, which will also call
__cpufreq_driver_target(). And hence we end up issuing concurrent calls to
->target().
Typically, platforms have the following logic in their ->target() routines:
(Eg: cpufreq-cpu0, omap, exynos, etc)
A. If new freq is more than old: Increase voltage
B. Change freq
C. If new freq is less than old: decrease voltage
Now, if the two concurrent calls to ->target() are X and Y, where X is trying to
increase the freq and Y is trying to decrease it, we get the following race
condition:
X.A: voltage gets increased for larger freq
Y.A: nothing happens
Y.B: freq gets decreased
Y.C: voltage gets decreased
X.B: freq gets increased
X.C: nothing happens
Thus we can end up setting a freq which is not supported by the voltage we have
set. That will probably make the clock to the CPU unstable and the system might
not work properly anymore.
This patch introduces a set of synchronization primitives to serialize frequency
transitions, which are to be used as shown below:
cpufreq_freq_transition_begin();
//Perform the frequency change
cpufreq_freq_transition_end();
The _begin() call sends the PRECHANGE notification whereas the _end() call sends
the POSTCHANGE notification. Also, all the necessary synchronization is handled
within these calls. In particular, even drivers which set the ASYNC_NOTIFICATION
flag can also use these APIs for performing frequency transitions (ie., you can
call _begin() from one task, and call the corresponding _end() from a different
task).
The actual synchronization underneath is not that complicated:
The key challenge is to allow drivers to begin the transition from one thread
and end it in a completely different thread (this is to enable drivers that do
asynchronous POSTCHANGE notification from bottom-halves, to also use the same
interface).
To achieve this, a 'transition_ongoing' flag, a 'transition_lock' spinlock and a
wait-queue are added per-policy. The flag and the wait-queue are used in
conjunction to create an "uninterrupted flow" from _begin() to _end(). The
spinlock is used to ensure that only one such "flow" is in flight at any given
time. Put together, this provides us all the necessary synchronization.
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2014-03-24 16:05:44 +08:00
|
|
|
void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
|
|
|
|
struct cpufreq_freqs *freqs)
|
|
|
|
{
|
cpufreq: Catch double invocations of cpufreq_freq_transition_begin/end
Some cpufreq drivers were redundantly invoking the _begin() and _end()
APIs around frequency transitions, and this double invocation (one from
the cpufreq core and the other from the cpufreq driver) used to result
in a self-deadlock, leading to system hangs during boot. (The _begin()
API makes contending callers wait until the previous invocation is
complete. Hence, the cpufreq driver would end up waiting on itself!).
Now all such drivers have been fixed, but debugging this issue was not
very straight-forward (even lockdep didn't catch this). So let us add a
debug infrastructure to the cpufreq core to catch such issues more easily
in the future.
We add a new field called 'transition_task' to the policy structure, to keep
track of the task which is performing the frequency transition. Using this
field, we make note of this task during _begin() and print a warning if we
find a case where the same task is calling _begin() again, before completing
the previous frequency transition using the corresponding _end().
We have left out ASYNC_NOTIFICATION drivers from this debug infrastructure
for 2 reasons:
1. At the moment, we have no way to avoid a particular scenario where this
debug infrastructure can emit false-positive warnings for such drivers.
The scenario is depicted below:
Task A Task B
/* 1st freq transition */
Invoke _begin() {
...
...
}
Change the frequency
/* 2nd freq transition */
Invoke _begin() {
... //waiting for B to
... //finish _end() for
... //the 1st transition
... | Got interrupt for successful
... | change of frequency (1st one).
... |
... | /* 1st freq transition */
... | Invoke _end() {
... | ...
... V }
...
...
}
This scenario is actually deadlock-free because, once Task A changes the
frequency, it is Task B's responsibility to invoke the corresponding
_end() for the 1st frequency transition. Hence it is perfectly legal for
Task A to go ahead and attempt another frequency transition in the meantime.
(Of course it won't be able to proceed until Task B finishes the 1st _end(),
but this doesn't cause a deadlock or a hang).
The debug infrastructure cannot handle this scenario and will treat it as
a deadlock and print a warning. To avoid this, we exclude such drivers
from the purview of this code.
2. Luckily, we don't _need_ this infrastructure for ASYNC_NOTIFICATION drivers
at all! The cpufreq core does not automatically invoke the _begin() and
_end() APIs during frequency transitions in such drivers. Thus, the driver
alone is responsible for invoking _begin()/_end() and hence there shouldn't
be any conflicts which lead to double invocations. So, we can skip these
drivers, since the probability that such drivers will hit this problem is
extremely low, as outlined above.
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>
2014-05-05 15:22:39 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Catch double invocations of _begin() which lead to self-deadlock.
|
|
|
|
* ASYNC_NOTIFICATION drivers are left out because the cpufreq core
|
|
|
|
* doesn't invoke _begin() on their behalf, and hence the chances of
|
|
|
|
* double invocations are very low. Moreover, there are scenarios
|
|
|
|
* where these checks can emit false-positive warnings in these
|
|
|
|
* drivers; so we avoid that by skipping them altogether.
|
|
|
|
*/
|
|
|
|
WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
|
|
|
|
&& current == policy->transition_task);
|
|
|
|
|
cpufreq: Make sure frequency transitions are serialized
Whenever we change the frequency of a CPU, we call the PRECHANGE and POSTCHANGE
notifiers. They must be serialized, i.e. PRECHANGE and POSTCHANGE notifiers
should strictly alternate, thereby preventing two different sets of PRECHANGE or
POSTCHANGE notifiers from interleaving arbitrarily.
The following examples illustrate why this is important:
Scenario 1:
-----------
A thread reading the value of cpuinfo_cur_freq, will call
__cpufreq_cpu_get()->cpufreq_out_of_sync()->cpufreq_notify_transition()
The ondemand governor can decide to change the frequency of the CPU at the same
time and hence it can end up sending the notifications via ->target().
If the notifiers are not serialized, the following sequence can occur:
- PRECHANGE Notification for freq A (from cpuinfo_cur_freq)
- PRECHANGE Notification for freq B (from target())
- Freq changed by target() to B
- POSTCHANGE Notification for freq B
- POSTCHANGE Notification for freq A
We can see from the above that the last POSTCHANGE Notification happens for freq
A but the hardware is set to run at freq B.
Where would we break then?: adjust_jiffies() in cpufreq.c & cpufreq_callback()
in arch/arm/kernel/smp.c (which also adjusts the jiffies). All the
loops_per_jiffy calculations will get messed up.
Scenario 2:
-----------
The governor calls __cpufreq_driver_target() to change the frequency. At the
same time, if we change scaling_{min|max}_freq from sysfs, it will end up
calling the governor's CPUFREQ_GOV_LIMITS notification, which will also call
__cpufreq_driver_target(). And hence we end up issuing concurrent calls to
->target().
Typically, platforms have the following logic in their ->target() routines:
(Eg: cpufreq-cpu0, omap, exynos, etc)
A. If new freq is more than old: Increase voltage
B. Change freq
C. If new freq is less than old: decrease voltage
Now, if the two concurrent calls to ->target() are X and Y, where X is trying to
increase the freq and Y is trying to decrease it, we get the following race
condition:
X.A: voltage gets increased for larger freq
Y.A: nothing happens
Y.B: freq gets decreased
Y.C: voltage gets decreased
X.B: freq gets increased
X.C: nothing happens
Thus we can end up setting a freq which is not supported by the voltage we have
set. That will probably make the clock to the CPU unstable and the system might
not work properly anymore.
This patch introduces a set of synchronization primitives to serialize frequency
transitions, which are to be used as shown below:
cpufreq_freq_transition_begin();
//Perform the frequency change
cpufreq_freq_transition_end();
The _begin() call sends the PRECHANGE notification whereas the _end() call sends
the POSTCHANGE notification. Also, all the necessary synchronization is handled
within these calls. In particular, even drivers which set the ASYNC_NOTIFICATION
flag can also use these APIs for performing frequency transitions (ie., you can
call _begin() from one task, and call the corresponding _end() from a different
task).
The actual synchronization underneath is not that complicated:
The key challenge is to allow drivers to begin the transition from one thread
and end it in a completely different thread (this is to enable drivers that do
asynchronous POSTCHANGE notification from bottom-halves, to also use the same
interface).
To achieve this, a 'transition_ongoing' flag, a 'transition_lock' spinlock and a
wait-queue are added per-policy. The flag and the wait-queue are used in
conjunction to create an "uninterrupted flow" from _begin() to _end(). The
spinlock is used to ensure that only one such "flow" is in flight at any given
time. Put together, this provides us all the necessary synchronization.
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2014-03-24 16:05:44 +08:00
|
|
|
wait:
|
|
|
|
wait_event(policy->transition_wait, !policy->transition_ongoing);
|
|
|
|
|
|
|
|
spin_lock(&policy->transition_lock);
|
|
|
|
|
|
|
|
if (unlikely(policy->transition_ongoing)) {
|
|
|
|
spin_unlock(&policy->transition_lock);
|
|
|
|
goto wait;
|
|
|
|
}
|
|
|
|
|
|
|
|
policy->transition_ongoing = true;
|
cpufreq: Catch double invocations of cpufreq_freq_transition_begin/end
Some cpufreq drivers were redundantly invoking the _begin() and _end()
APIs around frequency transitions, and this double invocation (one from
the cpufreq core and the other from the cpufreq driver) used to result
in a self-deadlock, leading to system hangs during boot. (The _begin()
API makes contending callers wait until the previous invocation is
complete. Hence, the cpufreq driver would end up waiting on itself!).
Now all such drivers have been fixed, but debugging this issue was not
very straight-forward (even lockdep didn't catch this). So let us add a
debug infrastructure to the cpufreq core to catch such issues more easily
in the future.
We add a new field called 'transition_task' to the policy structure, to keep
track of the task which is performing the frequency transition. Using this
field, we make note of this task during _begin() and print a warning if we
find a case where the same task is calling _begin() again, before completing
the previous frequency transition using the corresponding _end().
We have left out ASYNC_NOTIFICATION drivers from this debug infrastructure
for 2 reasons:
1. At the moment, we have no way to avoid a particular scenario where this
debug infrastructure can emit false-positive warnings for such drivers.
The scenario is depicted below:
Task A Task B
/* 1st freq transition */
Invoke _begin() {
...
...
}
Change the frequency
/* 2nd freq transition */
Invoke _begin() {
... //waiting for B to
... //finish _end() for
... //the 1st transition
... | Got interrupt for successful
... | change of frequency (1st one).
... |
... | /* 1st freq transition */
... | Invoke _end() {
... | ...
... V }
...
...
}
This scenario is actually deadlock-free because, once Task A changes the
frequency, it is Task B's responsibility to invoke the corresponding
_end() for the 1st frequency transition. Hence it is perfectly legal for
Task A to go ahead and attempt another frequency transition in the meantime.
(Of course it won't be able to proceed until Task B finishes the 1st _end(),
but this doesn't cause a deadlock or a hang).
The debug infrastructure cannot handle this scenario and will treat it as
a deadlock and print a warning. To avoid this, we exclude such drivers
from the purview of this code.
2. Luckily, we don't _need_ this infrastructure for ASYNC_NOTIFICATION drivers
at all! The cpufreq core does not automatically invoke the _begin() and
_end() APIs during frequency transitions in such drivers. Thus, the driver
alone is responsible for invoking _begin()/_end() and hence there shouldn't
be any conflicts which lead to double invocations. So, we can skip these
drivers, since the probability that such drivers will hit this problem is
extremely low, as outlined above.
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>
2014-05-05 15:22:39 +08:00
|
|
|
policy->transition_task = current;
|
cpufreq: Make sure frequency transitions are serialized
Whenever we change the frequency of a CPU, we call the PRECHANGE and POSTCHANGE
notifiers. They must be serialized, i.e. PRECHANGE and POSTCHANGE notifiers
should strictly alternate, thereby preventing two different sets of PRECHANGE or
POSTCHANGE notifiers from interleaving arbitrarily.
The following examples illustrate why this is important:
Scenario 1:
-----------
A thread reading the value of cpuinfo_cur_freq, will call
__cpufreq_cpu_get()->cpufreq_out_of_sync()->cpufreq_notify_transition()
The ondemand governor can decide to change the frequency of the CPU at the same
time and hence it can end up sending the notifications via ->target().
If the notifiers are not serialized, the following sequence can occur:
- PRECHANGE Notification for freq A (from cpuinfo_cur_freq)
- PRECHANGE Notification for freq B (from target())
- Freq changed by target() to B
- POSTCHANGE Notification for freq B
- POSTCHANGE Notification for freq A
We can see from the above that the last POSTCHANGE Notification happens for freq
A but the hardware is set to run at freq B.
Where would we break then?: adjust_jiffies() in cpufreq.c & cpufreq_callback()
in arch/arm/kernel/smp.c (which also adjusts the jiffies). All the
loops_per_jiffy calculations will get messed up.
Scenario 2:
-----------
The governor calls __cpufreq_driver_target() to change the frequency. At the
same time, if we change scaling_{min|max}_freq from sysfs, it will end up
calling the governor's CPUFREQ_GOV_LIMITS notification, which will also call
__cpufreq_driver_target(). And hence we end up issuing concurrent calls to
->target().
Typically, platforms have the following logic in their ->target() routines:
(Eg: cpufreq-cpu0, omap, exynos, etc)
A. If new freq is more than old: Increase voltage
B. Change freq
C. If new freq is less than old: decrease voltage
Now, if the two concurrent calls to ->target() are X and Y, where X is trying to
increase the freq and Y is trying to decrease it, we get the following race
condition:
X.A: voltage gets increased for larger freq
Y.A: nothing happens
Y.B: freq gets decreased
Y.C: voltage gets decreased
X.B: freq gets increased
X.C: nothing happens
Thus we can end up setting a freq which is not supported by the voltage we have
set. That will probably make the clock to the CPU unstable and the system might
not work properly anymore.
This patch introduces a set of synchronization primitives to serialize frequency
transitions, which are to be used as shown below:
cpufreq_freq_transition_begin();
//Perform the frequency change
cpufreq_freq_transition_end();
The _begin() call sends the PRECHANGE notification whereas the _end() call sends
the POSTCHANGE notification. Also, all the necessary synchronization is handled
within these calls. In particular, even drivers which set the ASYNC_NOTIFICATION
flag can also use these APIs for performing frequency transitions (ie., you can
call _begin() from one task, and call the corresponding _end() from a different
task).
The actual synchronization underneath is not that complicated:
The key challenge is to allow drivers to begin the transition from one thread
and end it in a completely different thread (this is to enable drivers that do
asynchronous POSTCHANGE notification from bottom-halves, to also use the same
interface).
To achieve this, a 'transition_ongoing' flag, a 'transition_lock' spinlock and a
wait-queue are added per-policy. The flag and the wait-queue are used in
conjunction to create an "uninterrupted flow" from _begin() to _end(). The
spinlock is used to ensure that only one such "flow" is in flight at any given
time. Put together, this provides us all the necessary synchronization.
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2014-03-24 16:05:44 +08:00
|
|
|
|
|
|
|
spin_unlock(&policy->transition_lock);
|
|
|
|
|
|
|
|
cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
|
|
|
|
|
|
|
|
void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
|
|
|
|
struct cpufreq_freqs *freqs, int transition_failed)
|
|
|
|
{
|
|
|
|
if (unlikely(WARN_ON(!policy->transition_ongoing)))
|
|
|
|
return;
|
|
|
|
|
|
|
|
cpufreq_notify_post_transition(policy, freqs, transition_failed);
|
|
|
|
|
|
|
|
policy->transition_ongoing = false;
|
cpufreq: Catch double invocations of cpufreq_freq_transition_begin/end
Some cpufreq drivers were redundantly invoking the _begin() and _end()
APIs around frequency transitions, and this double invocation (one from
the cpufreq core and the other from the cpufreq driver) used to result
in a self-deadlock, leading to system hangs during boot. (The _begin()
API makes contending callers wait until the previous invocation is
complete. Hence, the cpufreq driver would end up waiting on itself!).
Now all such drivers have been fixed, but debugging this issue was not
very straight-forward (even lockdep didn't catch this). So let us add a
debug infrastructure to the cpufreq core to catch such issues more easily
in the future.
We add a new field called 'transition_task' to the policy structure, to keep
track of the task which is performing the frequency transition. Using this
field, we make note of this task during _begin() and print a warning if we
find a case where the same task is calling _begin() again, before completing
the previous frequency transition using the corresponding _end().
We have left out ASYNC_NOTIFICATION drivers from this debug infrastructure
for 2 reasons:
1. At the moment, we have no way to avoid a particular scenario where this
debug infrastructure can emit false-positive warnings for such drivers.
The scenario is depicted below:
Task A Task B
/* 1st freq transition */
Invoke _begin() {
...
...
}
Change the frequency
/* 2nd freq transition */
Invoke _begin() {
... //waiting for B to
... //finish _end() for
... //the 1st transition
... | Got interrupt for successful
... | change of frequency (1st one).
... |
... | /* 1st freq transition */
... | Invoke _end() {
... | ...
... V }
...
...
}
This scenario is actually deadlock-free because, once Task A changes the
frequency, it is Task B's responsibility to invoke the corresponding
_end() for the 1st frequency transition. Hence it is perfectly legal for
Task A to go ahead and attempt another frequency transition in the meantime.
(Of course it won't be able to proceed until Task B finishes the 1st _end(),
but this doesn't cause a deadlock or a hang).
The debug infrastructure cannot handle this scenario and will treat it as
a deadlock and print a warning. To avoid this, we exclude such drivers
from the purview of this code.
2. Luckily, we don't _need_ this infrastructure for ASYNC_NOTIFICATION drivers
at all! The cpufreq core does not automatically invoke the _begin() and
_end() APIs during frequency transitions in such drivers. Thus, the driver
alone is responsible for invoking _begin()/_end() and hence there shouldn't
be any conflicts which lead to double invocations. So, we can skip these
drivers, since the probability that such drivers will hit this problem is
extremely low, as outlined above.
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>
2014-05-05 15:22:39 +08:00
|
|
|
policy->transition_task = NULL;
|
cpufreq: Make sure frequency transitions are serialized
Whenever we change the frequency of a CPU, we call the PRECHANGE and POSTCHANGE
notifiers. They must be serialized, i.e. PRECHANGE and POSTCHANGE notifiers
should strictly alternate, thereby preventing two different sets of PRECHANGE or
POSTCHANGE notifiers from interleaving arbitrarily.
The following examples illustrate why this is important:
Scenario 1:
-----------
A thread reading the value of cpuinfo_cur_freq, will call
__cpufreq_cpu_get()->cpufreq_out_of_sync()->cpufreq_notify_transition()
The ondemand governor can decide to change the frequency of the CPU at the same
time and hence it can end up sending the notifications via ->target().
If the notifiers are not serialized, the following sequence can occur:
- PRECHANGE Notification for freq A (from cpuinfo_cur_freq)
- PRECHANGE Notification for freq B (from target())
- Freq changed by target() to B
- POSTCHANGE Notification for freq B
- POSTCHANGE Notification for freq A
We can see from the above that the last POSTCHANGE Notification happens for freq
A but the hardware is set to run at freq B.
Where would we break then?: adjust_jiffies() in cpufreq.c & cpufreq_callback()
in arch/arm/kernel/smp.c (which also adjusts the jiffies). All the
loops_per_jiffy calculations will get messed up.
Scenario 2:
-----------
The governor calls __cpufreq_driver_target() to change the frequency. At the
same time, if we change scaling_{min|max}_freq from sysfs, it will end up
calling the governor's CPUFREQ_GOV_LIMITS notification, which will also call
__cpufreq_driver_target(). And hence we end up issuing concurrent calls to
->target().
Typically, platforms have the following logic in their ->target() routines:
(Eg: cpufreq-cpu0, omap, exynos, etc)
A. If new freq is more than old: Increase voltage
B. Change freq
C. If new freq is less than old: decrease voltage
Now, if the two concurrent calls to ->target() are X and Y, where X is trying to
increase the freq and Y is trying to decrease it, we get the following race
condition:
X.A: voltage gets increased for larger freq
Y.A: nothing happens
Y.B: freq gets decreased
Y.C: voltage gets decreased
X.B: freq gets increased
X.C: nothing happens
Thus we can end up setting a freq which is not supported by the voltage we have
set. That will probably make the clock to the CPU unstable and the system might
not work properly anymore.
This patch introduces a set of synchronization primitives to serialize frequency
transitions, which are to be used as shown below:
cpufreq_freq_transition_begin();
//Perform the frequency change
cpufreq_freq_transition_end();
The _begin() call sends the PRECHANGE notification whereas the _end() call sends
the POSTCHANGE notification. Also, all the necessary synchronization is handled
within these calls. In particular, even drivers which set the ASYNC_NOTIFICATION
flag can also use these APIs for performing frequency transitions (ie., you can
call _begin() from one task, and call the corresponding _end() from a different
task).
The actual synchronization underneath is not that complicated:
The key challenge is to allow drivers to begin the transition from one thread
and end it in a completely different thread (this is to enable drivers that do
asynchronous POSTCHANGE notification from bottom-halves, to also use the same
interface).
To achieve this, a 'transition_ongoing' flag, a 'transition_lock' spinlock and a
wait-queue are added per-policy. The flag and the wait-queue are used in
conjunction to create an "uninterrupted flow" from _begin() to _end(). The
spinlock is used to ensure that only one such "flow" is in flight at any given
time. Put together, this provides us all the necessary synchronization.
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2014-03-24 16:05:44 +08:00
|
|
|
|
|
|
|
wake_up(&policy->transition_wait);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/*********************************************************************
|
|
|
|
* SYSFS INTERFACE *
|
|
|
|
*********************************************************************/
|
2014-02-27 00:42:42 +08:00
|
|
|
static ssize_t show_boost(struct kobject *kobj,
|
2013-12-20 22:24:49 +08:00
|
|
|
struct attribute *attr, char *buf)
|
|
|
|
{
|
|
|
|
return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
|
|
|
|
}
|
|
|
|
|
|
|
|
static ssize_t store_boost(struct kobject *kobj, struct attribute *attr,
|
|
|
|
const char *buf, size_t count)
|
|
|
|
{
|
|
|
|
int ret, enable;
|
|
|
|
|
|
|
|
ret = sscanf(buf, "%d", &enable);
|
|
|
|
if (ret != 1 || enable < 0 || enable > 1)
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
if (cpufreq_boost_trigger_state(enable)) {
|
2014-03-12 01:03:00 +08:00
|
|
|
pr_err("%s: Cannot %s BOOST!\n",
|
|
|
|
__func__, enable ? "enable" : "disable");
|
2013-12-20 22:24:49 +08:00
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
2014-03-12 01:03:00 +08:00
|
|
|
pr_debug("%s: cpufreq BOOST %s\n",
|
|
|
|
__func__, enable ? "enabled" : "disabled");
|
2013-12-20 22:24:49 +08:00
|
|
|
|
|
|
|
return count;
|
|
|
|
}
|
|
|
|
define_one_global_rw(boost);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2015-01-02 15:04:26 +08:00
|
|
|
static struct cpufreq_governor *find_governor(const char *str_governor)
|
2006-07-07 03:30:26 +08:00
|
|
|
{
|
|
|
|
struct cpufreq_governor *t;
|
|
|
|
|
|
|
|
list_for_each_entry(t, &cpufreq_governor_list, governor_list)
|
2014-09-29 21:50:11 +08:00
|
|
|
if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
|
2006-07-07 03:30:26 +08:00
|
|
|
return t;
|
|
|
|
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/**
|
|
|
|
* cpufreq_parse_governor - parse a governor string
|
|
|
|
*/
|
2008-03-06 03:28:32 +08:00
|
|
|
static int cpufreq_parse_governor(char *str_governor, unsigned int *policy,
|
2005-04-17 06:20:36 +08:00
|
|
|
struct cpufreq_governor **governor)
|
|
|
|
{
|
2006-07-07 03:30:26 +08:00
|
|
|
int err = -EINVAL;
|
2013-04-29 06:08:16 +08:00
|
|
|
|
|
|
|
if (!cpufreq_driver)
|
2006-07-07 03:30:26 +08:00
|
|
|
goto out;
|
|
|
|
|
2013-04-29 06:08:16 +08:00
|
|
|
if (cpufreq_driver->setpolicy) {
|
2014-09-29 21:50:11 +08:00
|
|
|
if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
|
2005-04-17 06:20:36 +08:00
|
|
|
*policy = CPUFREQ_POLICY_PERFORMANCE;
|
2006-07-07 03:30:26 +08:00
|
|
|
err = 0;
|
2014-09-29 21:50:11 +08:00
|
|
|
} else if (!strncasecmp(str_governor, "powersave",
|
2006-10-26 18:50:58 +08:00
|
|
|
CPUFREQ_NAME_LEN)) {
|
2005-04-17 06:20:36 +08:00
|
|
|
*policy = CPUFREQ_POLICY_POWERSAVE;
|
2006-07-07 03:30:26 +08:00
|
|
|
err = 0;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2015-01-02 15:04:27 +08:00
|
|
|
} else {
|
2005-04-17 06:20:36 +08:00
|
|
|
struct cpufreq_governor *t;
|
2006-07-07 03:30:26 +08:00
|
|
|
|
2006-01-14 07:54:22 +08:00
|
|
|
mutex_lock(&cpufreq_governor_mutex);
|
2006-07-07 03:30:26 +08:00
|
|
|
|
2015-01-02 15:04:26 +08:00
|
|
|
t = find_governor(str_governor);
|
2006-07-07 03:30:26 +08:00
|
|
|
|
2006-07-07 03:32:01 +08:00
|
|
|
if (t == NULL) {
|
2011-05-04 23:38:56 +08:00
|
|
|
int ret;
|
2006-07-07 03:32:01 +08:00
|
|
|
|
2011-05-04 23:38:56 +08:00
|
|
|
mutex_unlock(&cpufreq_governor_mutex);
|
|
|
|
ret = request_module("cpufreq_%s", str_governor);
|
|
|
|
mutex_lock(&cpufreq_governor_mutex);
|
2006-07-07 03:32:01 +08:00
|
|
|
|
2011-05-04 23:38:56 +08:00
|
|
|
if (ret == 0)
|
2015-01-02 15:04:26 +08:00
|
|
|
t = find_governor(str_governor);
|
2006-07-07 03:32:01 +08:00
|
|
|
}
|
|
|
|
|
2006-07-07 03:30:26 +08:00
|
|
|
if (t != NULL) {
|
|
|
|
*governor = t;
|
|
|
|
err = 0;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2006-07-07 03:30:26 +08:00
|
|
|
|
2006-01-14 07:54:22 +08:00
|
|
|
mutex_unlock(&cpufreq_governor_mutex);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2009-01-18 14:37:11 +08:00
|
|
|
out:
|
2006-07-07 03:30:26 +08:00
|
|
|
return err;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
2006-10-26 18:50:58 +08:00
|
|
|
* cpufreq_per_cpu_attr_read() / show_##file_name() -
|
|
|
|
* print out cpufreq information
|
2005-04-17 06:20:36 +08:00
|
|
|
*
|
|
|
|
* Write out information from cpufreq_driver->policy[cpu]; object must be
|
|
|
|
* "unsigned int".
|
|
|
|
*/
|
|
|
|
|
2006-02-28 13:43:23 +08:00
|
|
|
#define show_one(file_name, object) \
|
|
|
|
static ssize_t show_##file_name \
|
2008-03-06 03:28:32 +08:00
|
|
|
(struct cpufreq_policy *policy, char *buf) \
|
2006-02-28 13:43:23 +08:00
|
|
|
{ \
|
2009-01-18 14:37:11 +08:00
|
|
|
return sprintf(buf, "%u\n", policy->object); \
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
show_one(cpuinfo_min_freq, cpuinfo.min_freq);
|
|
|
|
show_one(cpuinfo_max_freq, cpuinfo.max_freq);
|
2009-02-04 08:17:41 +08:00
|
|
|
show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
|
2005-04-17 06:20:36 +08:00
|
|
|
show_one(scaling_min_freq, min);
|
|
|
|
show_one(scaling_max_freq, max);
|
2014-10-13 23:37:40 +08:00
|
|
|
|
2015-01-02 15:04:24 +08:00
|
|
|
static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
|
2014-10-13 23:37:40 +08:00
|
|
|
{
|
|
|
|
ssize_t ret;
|
|
|
|
|
|
|
|
if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
|
|
|
|
ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
|
|
|
|
else
|
|
|
|
ret = sprintf(buf, "%u\n", policy->cur);
|
|
|
|
return ret;
|
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-10-02 16:43:16 +08:00
|
|
|
static int cpufreq_set_policy(struct cpufreq_policy *policy,
|
2013-08-07 01:23:05 +08:00
|
|
|
struct cpufreq_policy *new_policy);
|
2006-04-13 21:14:04 +08:00
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/**
|
|
|
|
* cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
|
|
|
|
*/
|
|
|
|
#define store_one(file_name, object) \
|
|
|
|
static ssize_t store_##file_name \
|
2008-03-06 03:28:32 +08:00
|
|
|
(struct cpufreq_policy *policy, const char *buf, size_t count) \
|
2005-04-17 06:20:36 +08:00
|
|
|
{ \
|
2014-11-10 14:14:50 +08:00
|
|
|
int ret, temp; \
|
2005-04-17 06:20:36 +08:00
|
|
|
struct cpufreq_policy new_policy; \
|
|
|
|
\
|
|
|
|
ret = cpufreq_get_policy(&new_policy, policy->cpu); \
|
|
|
|
if (ret) \
|
|
|
|
return -EINVAL; \
|
|
|
|
\
|
2009-01-18 14:37:11 +08:00
|
|
|
ret = sscanf(buf, "%u", &new_policy.object); \
|
2005-04-17 06:20:36 +08:00
|
|
|
if (ret != 1) \
|
|
|
|
return -EINVAL; \
|
|
|
|
\
|
2014-11-10 14:14:50 +08:00
|
|
|
temp = new_policy.object; \
|
2013-10-02 16:43:16 +08:00
|
|
|
ret = cpufreq_set_policy(policy, &new_policy); \
|
2014-11-10 14:14:50 +08:00
|
|
|
if (!ret) \
|
|
|
|
policy->user_policy.object = temp; \
|
2005-04-17 06:20:36 +08:00
|
|
|
\
|
|
|
|
return ret ? ret : count; \
|
|
|
|
}
|
|
|
|
|
2009-01-18 14:37:11 +08:00
|
|
|
store_one(scaling_min_freq, min);
|
|
|
|
store_one(scaling_max_freq, max);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/**
|
|
|
|
* show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
|
|
|
|
*/
|
2008-03-06 03:28:32 +08:00
|
|
|
static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
|
|
|
|
char *buf)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2015-01-02 15:04:29 +08:00
|
|
|
unsigned int cur_freq = __cpufreq_get(policy);
|
2005-04-17 06:20:36 +08:00
|
|
|
if (!cur_freq)
|
|
|
|
return sprintf(buf, "<unknown>");
|
|
|
|
return sprintf(buf, "%u\n", cur_freq);
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* show_scaling_governor - show the current policy for the specified CPU
|
|
|
|
*/
|
2008-03-06 03:28:32 +08:00
|
|
|
static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2009-01-18 14:37:11 +08:00
|
|
|
if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
|
2005-04-17 06:20:36 +08:00
|
|
|
return sprintf(buf, "powersave\n");
|
|
|
|
else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
|
|
|
|
return sprintf(buf, "performance\n");
|
|
|
|
else if (policy->governor)
|
2012-10-23 07:23:43 +08:00
|
|
|
return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
|
2009-01-18 14:37:11 +08:00
|
|
|
policy->governor->name);
|
2005-04-17 06:20:36 +08:00
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* store_scaling_governor - store policy for the specified CPU
|
|
|
|
*/
|
2008-03-06 03:28:32 +08:00
|
|
|
static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
|
|
|
|
const char *buf, size_t count)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2013-09-07 03:54:06 +08:00
|
|
|
int ret;
|
2005-04-17 06:20:36 +08:00
|
|
|
char str_governor[16];
|
|
|
|
struct cpufreq_policy new_policy;
|
|
|
|
|
|
|
|
ret = cpufreq_get_policy(&new_policy, policy->cpu);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
2009-01-18 14:37:11 +08:00
|
|
|
ret = sscanf(buf, "%15s", str_governor);
|
2005-04-17 06:20:36 +08:00
|
|
|
if (ret != 1)
|
|
|
|
return -EINVAL;
|
|
|
|
|
2006-10-26 18:50:58 +08:00
|
|
|
if (cpufreq_parse_governor(str_governor, &new_policy.policy,
|
|
|
|
&new_policy.governor))
|
2005-04-17 06:20:36 +08:00
|
|
|
return -EINVAL;
|
|
|
|
|
2013-10-02 16:43:16 +08:00
|
|
|
ret = cpufreq_set_policy(policy, &new_policy);
|
2006-04-13 21:14:04 +08:00
|
|
|
|
|
|
|
policy->user_policy.policy = policy->policy;
|
|
|
|
policy->user_policy.governor = policy->governor;
|
|
|
|
|
2006-10-26 18:50:58 +08:00
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
else
|
|
|
|
return count;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* show_scaling_driver - show the cpufreq driver currently loaded
|
|
|
|
*/
|
2008-03-06 03:28:32 +08:00
|
|
|
static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2013-04-29 06:08:16 +08:00
|
|
|
return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* show_scaling_available_governors - show the available CPUfreq governors
|
|
|
|
*/
|
2008-03-06 03:28:32 +08:00
|
|
|
static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
|
|
|
|
char *buf)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
ssize_t i = 0;
|
|
|
|
struct cpufreq_governor *t;
|
|
|
|
|
2013-10-25 22:15:48 +08:00
|
|
|
if (!has_target()) {
|
2005-04-17 06:20:36 +08:00
|
|
|
i += sprintf(buf, "performance powersave");
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
|
2009-01-18 14:37:11 +08:00
|
|
|
if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
|
|
|
|
- (CPUFREQ_NAME_LEN + 2)))
|
2005-04-17 06:20:36 +08:00
|
|
|
goto out;
|
2012-10-23 07:23:43 +08:00
|
|
|
i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2006-02-03 06:03:42 +08:00
|
|
|
out:
|
2005-04-17 06:20:36 +08:00
|
|
|
i += sprintf(&buf[i], "\n");
|
|
|
|
return i;
|
|
|
|
}
|
2008-04-19 04:31:12 +08:00
|
|
|
|
2013-06-27 15:08:54 +08:00
|
|
|
ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
ssize_t i = 0;
|
|
|
|
unsigned int cpu;
|
|
|
|
|
2009-01-04 21:18:06 +08:00
|
|
|
for_each_cpu(cpu, mask) {
|
2005-04-17 06:20:36 +08:00
|
|
|
if (i)
|
|
|
|
i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), " ");
|
|
|
|
i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u", cpu);
|
|
|
|
if (i >= (PAGE_SIZE - 5))
|
2009-01-18 14:37:11 +08:00
|
|
|
break;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
i += sprintf(&buf[i], "\n");
|
|
|
|
return i;
|
|
|
|
}
|
2013-06-27 15:08:54 +08:00
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2008-04-19 04:31:12 +08:00
|
|
|
/**
|
|
|
|
* show_related_cpus - show the CPUs affected by each transition even if
|
|
|
|
* hw coordination is in use
|
|
|
|
*/
|
|
|
|
static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
|
|
|
|
{
|
2013-06-27 15:08:54 +08:00
|
|
|
return cpufreq_show_cpus(policy->related_cpus, buf);
|
2008-04-19 04:31:12 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* show_affected_cpus - show the CPUs affected by each transition
|
|
|
|
*/
|
|
|
|
static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
|
|
|
|
{
|
2013-06-27 15:08:54 +08:00
|
|
|
return cpufreq_show_cpus(policy->cpus, buf);
|
2008-04-19 04:31:12 +08:00
|
|
|
}
|
|
|
|
|
2007-10-27 01:18:21 +08:00
|
|
|
static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
|
2008-03-06 03:28:32 +08:00
|
|
|
const char *buf, size_t count)
|
2007-10-27 01:18:21 +08:00
|
|
|
{
|
|
|
|
unsigned int freq = 0;
|
|
|
|
unsigned int ret;
|
|
|
|
|
2008-06-06 13:46:33 +08:00
|
|
|
if (!policy->governor || !policy->governor->store_setspeed)
|
2007-10-27 01:18:21 +08:00
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
ret = sscanf(buf, "%u", &freq);
|
|
|
|
if (ret != 1)
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
policy->governor->store_setspeed(policy, freq);
|
|
|
|
|
|
|
|
return count;
|
|
|
|
}
|
|
|
|
|
|
|
|
static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
|
|
|
|
{
|
2008-06-06 13:46:33 +08:00
|
|
|
if (!policy->governor || !policy->governor->show_setspeed)
|
2007-10-27 01:18:21 +08:00
|
|
|
return sprintf(buf, "<unsupported>\n");
|
|
|
|
|
|
|
|
return policy->governor->show_setspeed(policy, buf);
|
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2009-11-19 19:31:01 +08:00
|
|
|
/**
|
2012-10-23 07:23:33 +08:00
|
|
|
* show_bios_limit - show the current cpufreq HW/BIOS limitation
|
2009-11-19 19:31:01 +08:00
|
|
|
*/
|
|
|
|
static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
|
|
|
|
{
|
|
|
|
unsigned int limit;
|
|
|
|
int ret;
|
2013-04-29 06:08:16 +08:00
|
|
|
if (cpufreq_driver->bios_limit) {
|
|
|
|
ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
|
2009-11-19 19:31:01 +08:00
|
|
|
if (!ret)
|
|
|
|
return sprintf(buf, "%u\n", limit);
|
|
|
|
}
|
|
|
|
return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
|
|
|
|
}
|
|
|
|
|
2010-04-01 03:56:46 +08:00
|
|
|
cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
|
|
|
|
cpufreq_freq_attr_ro(cpuinfo_min_freq);
|
|
|
|
cpufreq_freq_attr_ro(cpuinfo_max_freq);
|
|
|
|
cpufreq_freq_attr_ro(cpuinfo_transition_latency);
|
|
|
|
cpufreq_freq_attr_ro(scaling_available_governors);
|
|
|
|
cpufreq_freq_attr_ro(scaling_driver);
|
|
|
|
cpufreq_freq_attr_ro(scaling_cur_freq);
|
|
|
|
cpufreq_freq_attr_ro(bios_limit);
|
|
|
|
cpufreq_freq_attr_ro(related_cpus);
|
|
|
|
cpufreq_freq_attr_ro(affected_cpus);
|
|
|
|
cpufreq_freq_attr_rw(scaling_min_freq);
|
|
|
|
cpufreq_freq_attr_rw(scaling_max_freq);
|
|
|
|
cpufreq_freq_attr_rw(scaling_governor);
|
|
|
|
cpufreq_freq_attr_rw(scaling_setspeed);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2008-03-06 03:28:32 +08:00
|
|
|
static struct attribute *default_attrs[] = {
|
2005-04-17 06:20:36 +08:00
|
|
|
&cpuinfo_min_freq.attr,
|
|
|
|
&cpuinfo_max_freq.attr,
|
2009-02-04 08:17:41 +08:00
|
|
|
&cpuinfo_transition_latency.attr,
|
2005-04-17 06:20:36 +08:00
|
|
|
&scaling_min_freq.attr,
|
|
|
|
&scaling_max_freq.attr,
|
|
|
|
&affected_cpus.attr,
|
2008-04-19 04:31:12 +08:00
|
|
|
&related_cpus.attr,
|
2005-04-17 06:20:36 +08:00
|
|
|
&scaling_governor.attr,
|
|
|
|
&scaling_driver.attr,
|
|
|
|
&scaling_available_governors.attr,
|
2007-10-27 01:18:21 +08:00
|
|
|
&scaling_setspeed.attr,
|
2005-04-17 06:20:36 +08:00
|
|
|
NULL
|
|
|
|
};
|
|
|
|
|
2009-01-18 14:37:11 +08:00
|
|
|
#define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
|
|
|
|
#define to_attr(a) container_of(a, struct freq_attr, attr)
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2009-01-18 14:37:11 +08:00
|
|
|
static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2008-03-06 03:28:32 +08:00
|
|
|
struct cpufreq_policy *policy = to_policy(kobj);
|
|
|
|
struct freq_attr *fattr = to_attr(attr);
|
2013-10-02 16:43:09 +08:00
|
|
|
ssize_t ret;
|
2013-08-07 01:23:11 +08:00
|
|
|
|
|
|
|
if (!down_read_trylock(&cpufreq_rwsem))
|
2013-10-02 16:43:09 +08:00
|
|
|
return -EINVAL;
|
2007-02-06 08:12:44 +08:00
|
|
|
|
2013-10-18 21:40:15 +08:00
|
|
|
down_read(&policy->rwsem);
|
2007-02-06 08:12:44 +08:00
|
|
|
|
2006-10-26 18:50:58 +08:00
|
|
|
if (fattr->show)
|
|
|
|
ret = fattr->show(policy, buf);
|
|
|
|
else
|
|
|
|
ret = -EIO;
|
|
|
|
|
2013-10-18 21:40:15 +08:00
|
|
|
up_read(&policy->rwsem);
|
2013-08-07 01:23:11 +08:00
|
|
|
up_read(&cpufreq_rwsem);
|
2013-10-02 16:43:09 +08:00
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2008-03-06 03:28:32 +08:00
|
|
|
static ssize_t store(struct kobject *kobj, struct attribute *attr,
|
|
|
|
const char *buf, size_t count)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2008-03-06 03:28:32 +08:00
|
|
|
struct cpufreq_policy *policy = to_policy(kobj);
|
|
|
|
struct freq_attr *fattr = to_attr(attr);
|
2008-03-06 03:22:25 +08:00
|
|
|
ssize_t ret = -EINVAL;
|
2013-08-07 01:23:11 +08:00
|
|
|
|
2013-09-07 03:53:43 +08:00
|
|
|
get_online_cpus();
|
|
|
|
|
|
|
|
if (!cpu_online(policy->cpu))
|
|
|
|
goto unlock;
|
|
|
|
|
2013-08-07 01:23:11 +08:00
|
|
|
if (!down_read_trylock(&cpufreq_rwsem))
|
2013-09-07 03:53:43 +08:00
|
|
|
goto unlock;
|
2007-02-06 08:12:44 +08:00
|
|
|
|
2013-10-18 21:40:15 +08:00
|
|
|
down_write(&policy->rwsem);
|
2007-02-06 08:12:44 +08:00
|
|
|
|
2006-10-26 18:50:58 +08:00
|
|
|
if (fattr->store)
|
|
|
|
ret = fattr->store(policy, buf, count);
|
|
|
|
else
|
|
|
|
ret = -EIO;
|
|
|
|
|
2013-10-18 21:40:15 +08:00
|
|
|
up_write(&policy->rwsem);
|
2013-08-07 01:23:11 +08:00
|
|
|
|
|
|
|
up_read(&cpufreq_rwsem);
|
2013-09-07 03:53:43 +08:00
|
|
|
unlock:
|
|
|
|
put_online_cpus();
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2008-03-06 03:28:32 +08:00
|
|
|
static void cpufreq_sysfs_release(struct kobject *kobj)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2008-03-06 03:28:32 +08:00
|
|
|
struct cpufreq_policy *policy = to_policy(kobj);
|
2011-03-27 21:04:46 +08:00
|
|
|
pr_debug("last reference is dropped\n");
|
2005-04-17 06:20:36 +08:00
|
|
|
complete(&policy->kobj_unregister);
|
|
|
|
}
|
|
|
|
|
2010-01-19 09:58:23 +08:00
|
|
|
static const struct sysfs_ops sysfs_ops = {
|
2005-04-17 06:20:36 +08:00
|
|
|
.show = show,
|
|
|
|
.store = store,
|
|
|
|
};
|
|
|
|
|
|
|
|
static struct kobj_type ktype_cpufreq = {
|
|
|
|
.sysfs_ops = &sysfs_ops,
|
|
|
|
.default_attrs = default_attrs,
|
|
|
|
.release = cpufreq_sysfs_release,
|
|
|
|
};
|
|
|
|
|
2013-05-17 18:39:09 +08:00
|
|
|
struct kobject *cpufreq_global_kobject;
|
|
|
|
EXPORT_SYMBOL(cpufreq_global_kobject);
|
|
|
|
|
|
|
|
static int cpufreq_global_kobject_usage;
|
|
|
|
|
|
|
|
int cpufreq_get_global_kobject(void)
|
|
|
|
{
|
|
|
|
if (!cpufreq_global_kobject_usage++)
|
|
|
|
return kobject_add(cpufreq_global_kobject,
|
|
|
|
&cpu_subsys.dev_root->kobj, "%s", "cpufreq");
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(cpufreq_get_global_kobject);
|
|
|
|
|
|
|
|
void cpufreq_put_global_kobject(void)
|
|
|
|
{
|
|
|
|
if (!--cpufreq_global_kobject_usage)
|
|
|
|
kobject_del(cpufreq_global_kobject);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(cpufreq_put_global_kobject);
|
|
|
|
|
|
|
|
int cpufreq_sysfs_create_file(const struct attribute *attr)
|
|
|
|
{
|
|
|
|
int ret = cpufreq_get_global_kobject();
|
|
|
|
|
|
|
|
if (!ret) {
|
|
|
|
ret = sysfs_create_file(cpufreq_global_kobject, attr);
|
|
|
|
if (ret)
|
|
|
|
cpufreq_put_global_kobject();
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(cpufreq_sysfs_create_file);
|
|
|
|
|
|
|
|
void cpufreq_sysfs_remove_file(const struct attribute *attr)
|
|
|
|
{
|
|
|
|
sysfs_remove_file(cpufreq_global_kobject, attr);
|
|
|
|
cpufreq_put_global_kobject();
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(cpufreq_sysfs_remove_file);
|
|
|
|
|
2009-07-09 05:35:39 +08:00
|
|
|
/* symlink affected CPUs */
|
2013-07-31 20:35:14 +08:00
|
|
|
static int cpufreq_add_dev_symlink(struct cpufreq_policy *policy)
|
2009-07-09 05:35:39 +08:00
|
|
|
{
|
|
|
|
unsigned int j;
|
|
|
|
int ret = 0;
|
|
|
|
|
|
|
|
for_each_cpu(j, policy->cpus) {
|
2011-12-22 06:29:42 +08:00
|
|
|
struct device *cpu_dev;
|
2009-07-09 05:35:39 +08:00
|
|
|
|
2013-07-31 20:35:14 +08:00
|
|
|
if (j == policy->cpu)
|
2009-07-09 05:35:39 +08:00
|
|
|
continue;
|
|
|
|
|
2013-07-31 20:31:33 +08:00
|
|
|
pr_debug("Adding link for CPU: %u\n", j);
|
2011-12-22 06:29:42 +08:00
|
|
|
cpu_dev = get_cpu_device(j);
|
|
|
|
ret = sysfs_create_link(&cpu_dev->kobj, &policy->kobj,
|
2009-07-09 05:35:39 +08:00
|
|
|
"cpufreq");
|
cpufreq: Do not hold driver module references for additional policy CPUs
The cpufreq core is a little inconsistent in the way it uses the
driver module refcount.
Namely, if __cpufreq_add_dev() is called for a CPU that doesn't
share the policy object with any other CPUs, the driver module
refcount it grabs to start with will be dropped by it before
returning and will be equal to whatever it had been before that
function was invoked.
However, if the given CPU does share the policy object with other
CPUs, either cpufreq_add_policy_cpu() is called to link the new CPU
to the existing policy, or cpufreq_add_dev_symlink() is used to link
the other CPUs sharing the policy with it to the just created policy
object. In that case, because both cpufreq_add_policy_cpu() and
cpufreq_add_dev_symlink() call cpufreq_cpu_get() for the given
policy (the latter possibly many times) without the balancing
cpufreq_cpu_put() (unless there is an error), the driver module
refcount will be left by __cpufreq_add_dev() with a nonzero value
(different from the initial one).
To remove that inconsistency make cpufreq_add_policy_cpu() execute
cpufreq_cpu_put() for the given policy before returning, which
decrements the driver module refcount so that it will be equal to its
initial value after __cpufreq_add_dev() returns. Also remove the
cpufreq_cpu_get() call from cpufreq_add_dev_symlink(), since both the
policy refcount and the driver module refcount are nonzero when it is
called and they don't need to be bumped up by it.
Accordingly, drop the cpufreq_cpu_put() from __cpufreq_remove_dev(),
since it is only necessary to balance the cpufreq_cpu_get() called
by cpufreq_add_policy_cpu() or cpufreq_add_dev_symlink().
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reviewed-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2013-08-04 07:19:34 +08:00
|
|
|
if (ret)
|
|
|
|
break;
|
2009-07-09 05:35:39 +08:00
|
|
|
}
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2013-07-31 20:35:14 +08:00
|
|
|
static int cpufreq_add_dev_interface(struct cpufreq_policy *policy,
|
2011-12-22 06:29:42 +08:00
|
|
|
struct device *dev)
|
2009-07-09 06:05:42 +08:00
|
|
|
{
|
|
|
|
struct freq_attr **drv_attr;
|
|
|
|
int ret = 0;
|
|
|
|
|
|
|
|
/* set up files for this cpu device */
|
2013-04-29 06:08:16 +08:00
|
|
|
drv_attr = cpufreq_driver->attr;
|
2015-01-02 15:04:23 +08:00
|
|
|
while (drv_attr && *drv_attr) {
|
2009-07-09 06:05:42 +08:00
|
|
|
ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
|
|
|
|
if (ret)
|
2014-11-24 17:08:03 +08:00
|
|
|
return ret;
|
2009-07-09 06:05:42 +08:00
|
|
|
drv_attr++;
|
|
|
|
}
|
2013-04-29 06:08:16 +08:00
|
|
|
if (cpufreq_driver->get) {
|
2009-07-09 06:05:42 +08:00
|
|
|
ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
|
|
|
|
if (ret)
|
2014-11-24 17:08:03 +08:00
|
|
|
return ret;
|
2009-07-09 06:05:42 +08:00
|
|
|
}
|
2014-10-13 23:37:40 +08:00
|
|
|
|
|
|
|
ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
|
|
|
|
if (ret)
|
2014-11-24 17:08:03 +08:00
|
|
|
return ret;
|
2014-10-13 23:37:40 +08:00
|
|
|
|
2013-04-29 06:08:16 +08:00
|
|
|
if (cpufreq_driver->bios_limit) {
|
2009-11-19 19:31:01 +08:00
|
|
|
ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
|
|
|
|
if (ret)
|
2014-11-24 17:08:03 +08:00
|
|
|
return ret;
|
2009-11-19 19:31:01 +08:00
|
|
|
}
|
2009-07-09 06:05:42 +08:00
|
|
|
|
2014-11-24 17:08:03 +08:00
|
|
|
return cpufreq_add_dev_symlink(policy);
|
2013-07-30 06:54:23 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static void cpufreq_init_policy(struct cpufreq_policy *policy)
|
|
|
|
{
|
2014-03-04 11:43:59 +08:00
|
|
|
struct cpufreq_governor *gov = NULL;
|
2013-07-30 06:54:23 +08:00
|
|
|
struct cpufreq_policy new_policy;
|
|
|
|
int ret = 0;
|
|
|
|
|
2013-08-07 01:23:06 +08:00
|
|
|
memcpy(&new_policy, policy, sizeof(*policy));
|
2013-12-20 06:50:50 +08:00
|
|
|
|
2014-03-04 11:43:59 +08:00
|
|
|
/* Update governor of new_policy to the governor used before hotplug */
|
2015-01-02 15:04:26 +08:00
|
|
|
gov = find_governor(per_cpu(cpufreq_cpu_governor, policy->cpu));
|
2014-03-04 11:43:59 +08:00
|
|
|
if (gov)
|
|
|
|
pr_debug("Restoring governor %s for cpu %d\n",
|
|
|
|
policy->governor->name, policy->cpu);
|
|
|
|
else
|
|
|
|
gov = CPUFREQ_DEFAULT_GOVERNOR;
|
|
|
|
|
|
|
|
new_policy.governor = gov;
|
|
|
|
|
2013-12-20 06:50:50 +08:00
|
|
|
/* Use the default policy if its valid. */
|
|
|
|
if (cpufreq_driver->setpolicy)
|
2014-03-04 11:43:59 +08:00
|
|
|
cpufreq_parse_governor(gov->name, &new_policy.policy, NULL);
|
2009-07-09 06:48:47 +08:00
|
|
|
|
|
|
|
/* set default policy */
|
2013-10-02 16:43:16 +08:00
|
|
|
ret = cpufreq_set_policy(policy, &new_policy);
|
2009-07-09 06:48:47 +08:00
|
|
|
if (ret) {
|
2011-03-27 21:04:46 +08:00
|
|
|
pr_debug("setting policy failed\n");
|
2013-04-29 06:08:16 +08:00
|
|
|
if (cpufreq_driver->exit)
|
|
|
|
cpufreq_driver->exit(policy);
|
2009-07-09 06:48:47 +08:00
|
|
|
}
|
2009-07-09 06:05:42 +08:00
|
|
|
}
|
|
|
|
|
2013-08-04 07:20:07 +08:00
|
|
|
static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy,
|
2013-12-20 23:56:02 +08:00
|
|
|
unsigned int cpu, struct device *dev)
|
2013-01-29 22:39:08 +08:00
|
|
|
{
|
2013-10-25 22:15:48 +08:00
|
|
|
int ret = 0;
|
2013-01-29 22:39:08 +08:00
|
|
|
unsigned long flags;
|
|
|
|
|
2013-10-25 22:15:48 +08:00
|
|
|
if (has_target()) {
|
2013-08-07 01:23:13 +08:00
|
|
|
ret = __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
|
|
|
|
if (ret) {
|
|
|
|
pr_err("%s: Failed to stop governor\n", __func__);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
}
|
2013-01-29 22:39:08 +08:00
|
|
|
|
2013-10-18 21:40:15 +08:00
|
|
|
down_write(&policy->rwsem);
|
2013-02-07 13:25:00 +08:00
|
|
|
|
2013-02-23 00:24:34 +08:00
|
|
|
write_lock_irqsave(&cpufreq_driver_lock, flags);
|
2013-02-07 13:25:00 +08:00
|
|
|
|
2013-01-29 22:39:08 +08:00
|
|
|
cpumask_set_cpu(cpu, policy->cpus);
|
|
|
|
per_cpu(cpufreq_cpu_data, cpu) = policy;
|
2013-02-23 00:24:34 +08:00
|
|
|
write_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
2013-01-29 22:39:08 +08:00
|
|
|
|
2013-10-18 21:40:15 +08:00
|
|
|
up_write(&policy->rwsem);
|
2013-02-07 13:25:00 +08:00
|
|
|
|
2013-10-25 22:15:48 +08:00
|
|
|
if (has_target()) {
|
2014-03-20 05:29:17 +08:00
|
|
|
ret = __cpufreq_governor(policy, CPUFREQ_GOV_START);
|
|
|
|
if (!ret)
|
|
|
|
ret = __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
|
|
|
|
|
|
|
|
if (ret) {
|
2013-08-07 01:23:13 +08:00
|
|
|
pr_err("%s: Failed to start governor\n", __func__);
|
|
|
|
return ret;
|
|
|
|
}
|
2013-04-22 06:48:03 +08:00
|
|
|
}
|
2013-01-29 22:39:08 +08:00
|
|
|
|
2013-12-20 23:56:02 +08:00
|
|
|
return sysfs_create_link(&dev->kobj, &policy->kobj, "cpufreq");
|
2013-01-29 22:39:08 +08:00
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
cpufreq: Preserve policy structure across suspend/resume
To perform light-weight cpu-init and teardown in the cpufreq subsystem
during suspend/resume, we need to separate out the 2 main functionalities
of the cpufreq CPU hotplug callbacks, as outlined below:
1. Init/tear-down of core cpufreq and CPU-specific components, which are
critical to the correct functioning of the cpufreq subsystem.
2. Init/tear-down of cpufreq sysfs files during suspend/resume.
The first part requires accurate updates to the policy structure such as
its ->cpus and ->related_cpus masks, whereas the second part requires that
the policy->kobj structure is not released or re-initialized during
suspend/resume.
To handle both these requirements, we need to allow updates to the policy
structure throughout suspend/resume, but prevent the structure from getting
freed up. Also, we must have a mechanism by which the cpu-up callbacks can
restore the policy structure, without allocating things afresh. (That also
helps avoid memory leaks).
To achieve this, we use 2 schemes:
a. Use a fallback per-cpu storage area for preserving the policy structures
during suspend, so that they can be restored during resume appropriately.
b. Use the 'frozen' flag to determine when to free or allocate the policy
structure vs when to restore the policy from the saved fallback storage.
Thus we can successfully preserve the structure across suspend/resume.
Effectively, this helps us complete the separation of the 'light-weight'
and the 'full' init/tear-down sequences in the cpufreq subsystem, so that
this can be made use of in the suspend/resume scenario.
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>
2013-07-30 06:55:10 +08:00
|
|
|
static struct cpufreq_policy *cpufreq_policy_restore(unsigned int cpu)
|
|
|
|
{
|
|
|
|
struct cpufreq_policy *policy;
|
|
|
|
unsigned long flags;
|
|
|
|
|
2013-09-11 15:05:05 +08:00
|
|
|
read_lock_irqsave(&cpufreq_driver_lock, flags);
|
cpufreq: Preserve policy structure across suspend/resume
To perform light-weight cpu-init and teardown in the cpufreq subsystem
during suspend/resume, we need to separate out the 2 main functionalities
of the cpufreq CPU hotplug callbacks, as outlined below:
1. Init/tear-down of core cpufreq and CPU-specific components, which are
critical to the correct functioning of the cpufreq subsystem.
2. Init/tear-down of cpufreq sysfs files during suspend/resume.
The first part requires accurate updates to the policy structure such as
its ->cpus and ->related_cpus masks, whereas the second part requires that
the policy->kobj structure is not released or re-initialized during
suspend/resume.
To handle both these requirements, we need to allow updates to the policy
structure throughout suspend/resume, but prevent the structure from getting
freed up. Also, we must have a mechanism by which the cpu-up callbacks can
restore the policy structure, without allocating things afresh. (That also
helps avoid memory leaks).
To achieve this, we use 2 schemes:
a. Use a fallback per-cpu storage area for preserving the policy structures
during suspend, so that they can be restored during resume appropriately.
b. Use the 'frozen' flag to determine when to free or allocate the policy
structure vs when to restore the policy from the saved fallback storage.
Thus we can successfully preserve the structure across suspend/resume.
Effectively, this helps us complete the separation of the 'light-weight'
and the 'full' init/tear-down sequences in the cpufreq subsystem, so that
this can be made use of in the suspend/resume scenario.
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>
2013-07-30 06:55:10 +08:00
|
|
|
|
|
|
|
policy = per_cpu(cpufreq_cpu_data_fallback, cpu);
|
|
|
|
|
2013-09-11 15:05:05 +08:00
|
|
|
read_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
cpufreq: Preserve policy structure across suspend/resume
To perform light-weight cpu-init and teardown in the cpufreq subsystem
during suspend/resume, we need to separate out the 2 main functionalities
of the cpufreq CPU hotplug callbacks, as outlined below:
1. Init/tear-down of core cpufreq and CPU-specific components, which are
critical to the correct functioning of the cpufreq subsystem.
2. Init/tear-down of cpufreq sysfs files during suspend/resume.
The first part requires accurate updates to the policy structure such as
its ->cpus and ->related_cpus masks, whereas the second part requires that
the policy->kobj structure is not released or re-initialized during
suspend/resume.
To handle both these requirements, we need to allow updates to the policy
structure throughout suspend/resume, but prevent the structure from getting
freed up. Also, we must have a mechanism by which the cpu-up callbacks can
restore the policy structure, without allocating things afresh. (That also
helps avoid memory leaks).
To achieve this, we use 2 schemes:
a. Use a fallback per-cpu storage area for preserving the policy structures
during suspend, so that they can be restored during resume appropriately.
b. Use the 'frozen' flag to determine when to free or allocate the policy
structure vs when to restore the policy from the saved fallback storage.
Thus we can successfully preserve the structure across suspend/resume.
Effectively, this helps us complete the separation of the 'light-weight'
and the 'full' init/tear-down sequences in the cpufreq subsystem, so that
this can be made use of in the suspend/resume scenario.
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>
2013-07-30 06:55:10 +08:00
|
|
|
|
2014-11-05 00:05:25 +08:00
|
|
|
if (policy)
|
|
|
|
policy->governor = NULL;
|
2014-03-04 11:43:59 +08:00
|
|
|
|
cpufreq: Preserve policy structure across suspend/resume
To perform light-weight cpu-init and teardown in the cpufreq subsystem
during suspend/resume, we need to separate out the 2 main functionalities
of the cpufreq CPU hotplug callbacks, as outlined below:
1. Init/tear-down of core cpufreq and CPU-specific components, which are
critical to the correct functioning of the cpufreq subsystem.
2. Init/tear-down of cpufreq sysfs files during suspend/resume.
The first part requires accurate updates to the policy structure such as
its ->cpus and ->related_cpus masks, whereas the second part requires that
the policy->kobj structure is not released or re-initialized during
suspend/resume.
To handle both these requirements, we need to allow updates to the policy
structure throughout suspend/resume, but prevent the structure from getting
freed up. Also, we must have a mechanism by which the cpu-up callbacks can
restore the policy structure, without allocating things afresh. (That also
helps avoid memory leaks).
To achieve this, we use 2 schemes:
a. Use a fallback per-cpu storage area for preserving the policy structures
during suspend, so that they can be restored during resume appropriately.
b. Use the 'frozen' flag to determine when to free or allocate the policy
structure vs when to restore the policy from the saved fallback storage.
Thus we can successfully preserve the structure across suspend/resume.
Effectively, this helps us complete the separation of the 'light-weight'
and the 'full' init/tear-down sequences in the cpufreq subsystem, so that
this can be made use of in the suspend/resume scenario.
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>
2013-07-30 06:55:10 +08:00
|
|
|
return policy;
|
|
|
|
}
|
|
|
|
|
2013-07-30 06:54:11 +08:00
|
|
|
static struct cpufreq_policy *cpufreq_policy_alloc(void)
|
|
|
|
{
|
|
|
|
struct cpufreq_policy *policy;
|
|
|
|
|
|
|
|
policy = kzalloc(sizeof(*policy), GFP_KERNEL);
|
|
|
|
if (!policy)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
|
|
|
|
goto err_free_policy;
|
|
|
|
|
|
|
|
if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
|
|
|
|
goto err_free_cpumask;
|
|
|
|
|
2013-08-07 01:23:08 +08:00
|
|
|
INIT_LIST_HEAD(&policy->policy_list);
|
2013-10-18 21:40:15 +08:00
|
|
|
init_rwsem(&policy->rwsem);
|
cpufreq: Make sure frequency transitions are serialized
Whenever we change the frequency of a CPU, we call the PRECHANGE and POSTCHANGE
notifiers. They must be serialized, i.e. PRECHANGE and POSTCHANGE notifiers
should strictly alternate, thereby preventing two different sets of PRECHANGE or
POSTCHANGE notifiers from interleaving arbitrarily.
The following examples illustrate why this is important:
Scenario 1:
-----------
A thread reading the value of cpuinfo_cur_freq, will call
__cpufreq_cpu_get()->cpufreq_out_of_sync()->cpufreq_notify_transition()
The ondemand governor can decide to change the frequency of the CPU at the same
time and hence it can end up sending the notifications via ->target().
If the notifiers are not serialized, the following sequence can occur:
- PRECHANGE Notification for freq A (from cpuinfo_cur_freq)
- PRECHANGE Notification for freq B (from target())
- Freq changed by target() to B
- POSTCHANGE Notification for freq B
- POSTCHANGE Notification for freq A
We can see from the above that the last POSTCHANGE Notification happens for freq
A but the hardware is set to run at freq B.
Where would we break then?: adjust_jiffies() in cpufreq.c & cpufreq_callback()
in arch/arm/kernel/smp.c (which also adjusts the jiffies). All the
loops_per_jiffy calculations will get messed up.
Scenario 2:
-----------
The governor calls __cpufreq_driver_target() to change the frequency. At the
same time, if we change scaling_{min|max}_freq from sysfs, it will end up
calling the governor's CPUFREQ_GOV_LIMITS notification, which will also call
__cpufreq_driver_target(). And hence we end up issuing concurrent calls to
->target().
Typically, platforms have the following logic in their ->target() routines:
(Eg: cpufreq-cpu0, omap, exynos, etc)
A. If new freq is more than old: Increase voltage
B. Change freq
C. If new freq is less than old: decrease voltage
Now, if the two concurrent calls to ->target() are X and Y, where X is trying to
increase the freq and Y is trying to decrease it, we get the following race
condition:
X.A: voltage gets increased for larger freq
Y.A: nothing happens
Y.B: freq gets decreased
Y.C: voltage gets decreased
X.B: freq gets increased
X.C: nothing happens
Thus we can end up setting a freq which is not supported by the voltage we have
set. That will probably make the clock to the CPU unstable and the system might
not work properly anymore.
This patch introduces a set of synchronization primitives to serialize frequency
transitions, which are to be used as shown below:
cpufreq_freq_transition_begin();
//Perform the frequency change
cpufreq_freq_transition_end();
The _begin() call sends the PRECHANGE notification whereas the _end() call sends
the POSTCHANGE notification. Also, all the necessary synchronization is handled
within these calls. In particular, even drivers which set the ASYNC_NOTIFICATION
flag can also use these APIs for performing frequency transitions (ie., you can
call _begin() from one task, and call the corresponding _end() from a different
task).
The actual synchronization underneath is not that complicated:
The key challenge is to allow drivers to begin the transition from one thread
and end it in a completely different thread (this is to enable drivers that do
asynchronous POSTCHANGE notification from bottom-halves, to also use the same
interface).
To achieve this, a 'transition_ongoing' flag, a 'transition_lock' spinlock and a
wait-queue are added per-policy. The flag and the wait-queue are used in
conjunction to create an "uninterrupted flow" from _begin() to _end(). The
spinlock is used to ensure that only one such "flow" is in flight at any given
time. Put together, this provides us all the necessary synchronization.
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2014-03-24 16:05:44 +08:00
|
|
|
spin_lock_init(&policy->transition_lock);
|
|
|
|
init_waitqueue_head(&policy->transition_wait);
|
2013-10-18 21:40:15 +08:00
|
|
|
|
2013-07-30 06:54:11 +08:00
|
|
|
return policy;
|
|
|
|
|
|
|
|
err_free_cpumask:
|
|
|
|
free_cpumask_var(policy->cpus);
|
|
|
|
err_free_policy:
|
|
|
|
kfree(policy);
|
|
|
|
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
2013-12-20 23:56:02 +08:00
|
|
|
static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy)
|
|
|
|
{
|
|
|
|
struct kobject *kobj;
|
|
|
|
struct completion *cmp;
|
|
|
|
|
2014-01-07 09:40:10 +08:00
|
|
|
blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
|
|
|
|
CPUFREQ_REMOVE_POLICY, policy);
|
|
|
|
|
2013-12-20 23:56:02 +08:00
|
|
|
down_read(&policy->rwsem);
|
|
|
|
kobj = &policy->kobj;
|
|
|
|
cmp = &policy->kobj_unregister;
|
|
|
|
up_read(&policy->rwsem);
|
|
|
|
kobject_put(kobj);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We need to make sure that the underlying kobj is
|
|
|
|
* actually not referenced anymore by anybody before we
|
|
|
|
* proceed with unloading.
|
|
|
|
*/
|
|
|
|
pr_debug("waiting for dropping of refcount\n");
|
|
|
|
wait_for_completion(cmp);
|
|
|
|
pr_debug("wait complete\n");
|
|
|
|
}
|
|
|
|
|
2013-07-30 06:54:11 +08:00
|
|
|
static void cpufreq_policy_free(struct cpufreq_policy *policy)
|
|
|
|
{
|
|
|
|
free_cpumask_var(policy->related_cpus);
|
|
|
|
free_cpumask_var(policy->cpus);
|
|
|
|
kfree(policy);
|
|
|
|
}
|
|
|
|
|
2014-07-17 13:18:28 +08:00
|
|
|
static int update_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu,
|
|
|
|
struct device *cpu_dev)
|
cpufreq: Fix crash in cpufreq-stats during suspend/resume
Stephen Warren reported that the cpufreq-stats code hits a NULL pointer
dereference during the second attempt to suspend a system. He also
pin-pointed the problem to commit 5302c3f "cpufreq: Perform light-weight
init/teardown during suspend/resume".
That commit actually ensured that the cpufreq-stats table and the
cpufreq-stats sysfs entries are *not* torn down (ie., not freed) during
suspend/resume, which makes it all the more surprising. However, it turns
out that the root-cause is not that we access an already freed memory, but
that the reference to the allocated memory gets moved around and we lose
track of that during resume, leading to the reported crash in a subsequent
suspend attempt.
In the suspend path, during CPU offline, the value of policy->cpu is updated
by choosing one of the surviving CPUs in that policy, as long as there is
atleast one CPU in that policy. And cpufreq_stats_update_policy_cpu() is
invoked to update the reference to the stats structure by assigning it to
the new CPU. However, in the resume path, during CPU online, we end up
assigning a fresh CPU as the policy->cpu, without letting cpufreq-stats
know about this. Thus the reference to the stats structure remains
(incorrectly) associated with the old CPU. So, in a subsequent suspend attempt,
during CPU offline, we end up accessing an incorrect location to get the
stats structure, which eventually leads to the NULL pointer dereference.
Fix this by letting cpufreq-stats know about the update of the policy->cpu
during CPU online in the resume path. (Also, move the update_policy_cpu()
function higher up in the file, so that __cpufreq_add_dev() can invoke
it).
Reported-and-tested-by: Stephen Warren <swarren@nvidia.com>
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2013-09-12 04:12:59 +08:00
|
|
|
{
|
2014-07-17 13:18:28 +08:00
|
|
|
int ret;
|
|
|
|
|
2013-09-12 19:59:09 +08:00
|
|
|
if (WARN_ON(cpu == policy->cpu))
|
2014-07-17 13:18:28 +08:00
|
|
|
return 0;
|
|
|
|
|
|
|
|
/* Move kobject to the new policy->cpu */
|
|
|
|
ret = kobject_move(&policy->kobj, &cpu_dev->kobj);
|
|
|
|
if (ret) {
|
|
|
|
pr_err("%s: Failed to move kobj: %d\n", __func__, ret);
|
|
|
|
return ret;
|
|
|
|
}
|
cpufreq: Prevent problems in update_policy_cpu() if last_cpu == new_cpu
If update_policy_cpu() is invoked with the existing policy->cpu itself
as the new-cpu parameter, then a lot of things can go terribly wrong.
In its present form, update_policy_cpu() always assumes that the new-cpu
is different from policy->cpu and invokes other functions to perform their
respective updates. And those functions implement the actual update like
this:
per_cpu(..., new_cpu) = per_cpu(..., last_cpu);
per_cpu(..., last_cpu) = NULL;
Thus, when new_cpu == last_cpu, the final NULL assignment makes the per-cpu
references vanish into thin air! (memory leak). From there, it leads to more
problems: cpufreq_stats_create_table() now doesn't find the per-cpu reference
and hence tries to create a new sysfs-group; but sysfs already had created
the group earlier, so it complains that it cannot create a duplicate filename.
In short, the repercussions of a rather innocuous invocation of
update_policy_cpu() can turn out to be pretty nasty.
Ideally update_policy_cpu() should handle this situation (new == last)
gracefully, and not lead to such severe problems. So fix it by adding an
appropriate check.
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Tested-by: Stephen Warren <swarren@nvidia.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2013-09-12 04:13:42 +08:00
|
|
|
|
2013-10-18 21:40:15 +08:00
|
|
|
down_write(&policy->rwsem);
|
2013-09-17 12:52:11 +08:00
|
|
|
|
cpufreq: Fix crash in cpufreq-stats during suspend/resume
Stephen Warren reported that the cpufreq-stats code hits a NULL pointer
dereference during the second attempt to suspend a system. He also
pin-pointed the problem to commit 5302c3f "cpufreq: Perform light-weight
init/teardown during suspend/resume".
That commit actually ensured that the cpufreq-stats table and the
cpufreq-stats sysfs entries are *not* torn down (ie., not freed) during
suspend/resume, which makes it all the more surprising. However, it turns
out that the root-cause is not that we access an already freed memory, but
that the reference to the allocated memory gets moved around and we lose
track of that during resume, leading to the reported crash in a subsequent
suspend attempt.
In the suspend path, during CPU offline, the value of policy->cpu is updated
by choosing one of the surviving CPUs in that policy, as long as there is
atleast one CPU in that policy. And cpufreq_stats_update_policy_cpu() is
invoked to update the reference to the stats structure by assigning it to
the new CPU. However, in the resume path, during CPU online, we end up
assigning a fresh CPU as the policy->cpu, without letting cpufreq-stats
know about this. Thus the reference to the stats structure remains
(incorrectly) associated with the old CPU. So, in a subsequent suspend attempt,
during CPU offline, we end up accessing an incorrect location to get the
stats structure, which eventually leads to the NULL pointer dereference.
Fix this by letting cpufreq-stats know about the update of the policy->cpu
during CPU online in the resume path. (Also, move the update_policy_cpu()
function higher up in the file, so that __cpufreq_add_dev() can invoke
it).
Reported-and-tested-by: Stephen Warren <swarren@nvidia.com>
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2013-09-12 04:12:59 +08:00
|
|
|
policy->last_cpu = policy->cpu;
|
|
|
|
policy->cpu = cpu;
|
|
|
|
|
2013-10-18 21:40:15 +08:00
|
|
|
up_write(&policy->rwsem);
|
2013-09-17 12:52:11 +08:00
|
|
|
|
cpufreq: Fix crash in cpufreq-stats during suspend/resume
Stephen Warren reported that the cpufreq-stats code hits a NULL pointer
dereference during the second attempt to suspend a system. He also
pin-pointed the problem to commit 5302c3f "cpufreq: Perform light-weight
init/teardown during suspend/resume".
That commit actually ensured that the cpufreq-stats table and the
cpufreq-stats sysfs entries are *not* torn down (ie., not freed) during
suspend/resume, which makes it all the more surprising. However, it turns
out that the root-cause is not that we access an already freed memory, but
that the reference to the allocated memory gets moved around and we lose
track of that during resume, leading to the reported crash in a subsequent
suspend attempt.
In the suspend path, during CPU offline, the value of policy->cpu is updated
by choosing one of the surviving CPUs in that policy, as long as there is
atleast one CPU in that policy. And cpufreq_stats_update_policy_cpu() is
invoked to update the reference to the stats structure by assigning it to
the new CPU. However, in the resume path, during CPU online, we end up
assigning a fresh CPU as the policy->cpu, without letting cpufreq-stats
know about this. Thus the reference to the stats structure remains
(incorrectly) associated with the old CPU. So, in a subsequent suspend attempt,
during CPU offline, we end up accessing an incorrect location to get the
stats structure, which eventually leads to the NULL pointer dereference.
Fix this by letting cpufreq-stats know about the update of the policy->cpu
during CPU online in the resume path. (Also, move the update_policy_cpu()
function higher up in the file, so that __cpufreq_add_dev() can invoke
it).
Reported-and-tested-by: Stephen Warren <swarren@nvidia.com>
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2013-09-12 04:12:59 +08:00
|
|
|
blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
|
|
|
|
CPUFREQ_UPDATE_POLICY_CPU, policy);
|
2014-07-17 13:18:28 +08:00
|
|
|
|
|
|
|
return 0;
|
cpufreq: Fix crash in cpufreq-stats during suspend/resume
Stephen Warren reported that the cpufreq-stats code hits a NULL pointer
dereference during the second attempt to suspend a system. He also
pin-pointed the problem to commit 5302c3f "cpufreq: Perform light-weight
init/teardown during suspend/resume".
That commit actually ensured that the cpufreq-stats table and the
cpufreq-stats sysfs entries are *not* torn down (ie., not freed) during
suspend/resume, which makes it all the more surprising. However, it turns
out that the root-cause is not that we access an already freed memory, but
that the reference to the allocated memory gets moved around and we lose
track of that during resume, leading to the reported crash in a subsequent
suspend attempt.
In the suspend path, during CPU offline, the value of policy->cpu is updated
by choosing one of the surviving CPUs in that policy, as long as there is
atleast one CPU in that policy. And cpufreq_stats_update_policy_cpu() is
invoked to update the reference to the stats structure by assigning it to
the new CPU. However, in the resume path, during CPU online, we end up
assigning a fresh CPU as the policy->cpu, without letting cpufreq-stats
know about this. Thus the reference to the stats structure remains
(incorrectly) associated with the old CPU. So, in a subsequent suspend attempt,
during CPU offline, we end up accessing an incorrect location to get the
stats structure, which eventually leads to the NULL pointer dereference.
Fix this by letting cpufreq-stats know about the update of the policy->cpu
during CPU online in the resume path. (Also, move the update_policy_cpu()
function higher up in the file, so that __cpufreq_add_dev() can invoke
it).
Reported-and-tested-by: Stephen Warren <swarren@nvidia.com>
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2013-09-12 04:12:59 +08:00
|
|
|
}
|
|
|
|
|
2014-03-10 17:23:35 +08:00
|
|
|
static int __cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2013-01-29 22:39:08 +08:00
|
|
|
unsigned int j, cpu = dev->id;
|
2013-02-07 13:26:03 +08:00
|
|
|
int ret = -ENOMEM;
|
2015-01-02 15:04:32 +08:00
|
|
|
struct cpufreq_policy *policy;
|
2005-04-17 06:20:36 +08:00
|
|
|
unsigned long flags;
|
2014-03-10 17:23:35 +08:00
|
|
|
bool recover_policy = cpufreq_suspended;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2005-10-31 06:59:54 +08:00
|
|
|
if (cpu_is_offline(cpu))
|
|
|
|
return 0;
|
|
|
|
|
2011-03-27 21:04:46 +08:00
|
|
|
pr_debug("adding CPU %u\n", cpu);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/* check whether a different CPU already registered this
|
|
|
|
* CPU because it is in the same boat. */
|
2015-01-02 15:04:33 +08:00
|
|
|
policy = cpufreq_cpu_get_raw(cpu);
|
|
|
|
if (unlikely(policy))
|
2005-04-17 06:20:36 +08:00
|
|
|
return 0;
|
2013-01-29 22:39:08 +08:00
|
|
|
|
2013-08-07 01:23:11 +08:00
|
|
|
if (!down_read_trylock(&cpufreq_rwsem))
|
|
|
|
return 0;
|
|
|
|
|
2013-01-29 22:39:08 +08:00
|
|
|
/* Check if this cpu was hot-unplugged earlier and has siblings */
|
2013-02-23 00:24:34 +08:00
|
|
|
read_lock_irqsave(&cpufreq_driver_lock, flags);
|
2015-01-02 15:04:32 +08:00
|
|
|
list_for_each_entry(policy, &cpufreq_policy_list, policy_list) {
|
|
|
|
if (cpumask_test_cpu(cpu, policy->related_cpus)) {
|
2013-02-23 00:24:34 +08:00
|
|
|
read_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
2015-01-02 15:04:32 +08:00
|
|
|
ret = cpufreq_add_policy_cpu(policy, cpu, dev);
|
2013-08-07 01:23:11 +08:00
|
|
|
up_read(&cpufreq_rwsem);
|
|
|
|
return ret;
|
2013-02-07 13:25:00 +08:00
|
|
|
}
|
2013-01-29 22:39:08 +08:00
|
|
|
}
|
2013-02-23 00:24:34 +08:00
|
|
|
read_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-12-27 08:07:11 +08:00
|
|
|
/*
|
|
|
|
* Restore the saved policy when doing light-weight init and fall back
|
|
|
|
* to the full init if that fails.
|
|
|
|
*/
|
2014-03-10 17:23:35 +08:00
|
|
|
policy = recover_policy ? cpufreq_policy_restore(cpu) : NULL;
|
2013-12-27 08:07:11 +08:00
|
|
|
if (!policy) {
|
2014-03-10 17:23:35 +08:00
|
|
|
recover_policy = false;
|
cpufreq: Preserve policy structure across suspend/resume
To perform light-weight cpu-init and teardown in the cpufreq subsystem
during suspend/resume, we need to separate out the 2 main functionalities
of the cpufreq CPU hotplug callbacks, as outlined below:
1. Init/tear-down of core cpufreq and CPU-specific components, which are
critical to the correct functioning of the cpufreq subsystem.
2. Init/tear-down of cpufreq sysfs files during suspend/resume.
The first part requires accurate updates to the policy structure such as
its ->cpus and ->related_cpus masks, whereas the second part requires that
the policy->kobj structure is not released or re-initialized during
suspend/resume.
To handle both these requirements, we need to allow updates to the policy
structure throughout suspend/resume, but prevent the structure from getting
freed up. Also, we must have a mechanism by which the cpu-up callbacks can
restore the policy structure, without allocating things afresh. (That also
helps avoid memory leaks).
To achieve this, we use 2 schemes:
a. Use a fallback per-cpu storage area for preserving the policy structures
during suspend, so that they can be restored during resume appropriately.
b. Use the 'frozen' flag to determine when to free or allocate the policy
structure vs when to restore the policy from the saved fallback storage.
Thus we can successfully preserve the structure across suspend/resume.
Effectively, this helps us complete the separation of the 'light-weight'
and the 'full' init/tear-down sequences in the cpufreq subsystem, so that
this can be made use of in the suspend/resume scenario.
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>
2013-07-30 06:55:10 +08:00
|
|
|
policy = cpufreq_policy_alloc();
|
2013-12-27 08:07:11 +08:00
|
|
|
if (!policy)
|
|
|
|
goto nomem_out;
|
|
|
|
}
|
cpufreq: Fix crash in cpufreq-stats during suspend/resume
Stephen Warren reported that the cpufreq-stats code hits a NULL pointer
dereference during the second attempt to suspend a system. He also
pin-pointed the problem to commit 5302c3f "cpufreq: Perform light-weight
init/teardown during suspend/resume".
That commit actually ensured that the cpufreq-stats table and the
cpufreq-stats sysfs entries are *not* torn down (ie., not freed) during
suspend/resume, which makes it all the more surprising. However, it turns
out that the root-cause is not that we access an already freed memory, but
that the reference to the allocated memory gets moved around and we lose
track of that during resume, leading to the reported crash in a subsequent
suspend attempt.
In the suspend path, during CPU offline, the value of policy->cpu is updated
by choosing one of the surviving CPUs in that policy, as long as there is
atleast one CPU in that policy. And cpufreq_stats_update_policy_cpu() is
invoked to update the reference to the stats structure by assigning it to
the new CPU. However, in the resume path, during CPU online, we end up
assigning a fresh CPU as the policy->cpu, without letting cpufreq-stats
know about this. Thus the reference to the stats structure remains
(incorrectly) associated with the old CPU. So, in a subsequent suspend attempt,
during CPU offline, we end up accessing an incorrect location to get the
stats structure, which eventually leads to the NULL pointer dereference.
Fix this by letting cpufreq-stats know about the update of the policy->cpu
during CPU online in the resume path. (Also, move the update_policy_cpu()
function higher up in the file, so that __cpufreq_add_dev() can invoke
it).
Reported-and-tested-by: Stephen Warren <swarren@nvidia.com>
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2013-09-12 04:12:59 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* In the resume path, since we restore a saved policy, the assignment
|
|
|
|
* to policy->cpu is like an update of the existing policy, rather than
|
|
|
|
* the creation of a brand new one. So we need to perform this update
|
|
|
|
* by invoking update_policy_cpu().
|
|
|
|
*/
|
2014-07-17 13:18:28 +08:00
|
|
|
if (recover_policy && cpu != policy->cpu)
|
|
|
|
WARN_ON(update_policy_cpu(policy, cpu, dev));
|
|
|
|
else
|
cpufreq: Fix crash in cpufreq-stats during suspend/resume
Stephen Warren reported that the cpufreq-stats code hits a NULL pointer
dereference during the second attempt to suspend a system. He also
pin-pointed the problem to commit 5302c3f "cpufreq: Perform light-weight
init/teardown during suspend/resume".
That commit actually ensured that the cpufreq-stats table and the
cpufreq-stats sysfs entries are *not* torn down (ie., not freed) during
suspend/resume, which makes it all the more surprising. However, it turns
out that the root-cause is not that we access an already freed memory, but
that the reference to the allocated memory gets moved around and we lose
track of that during resume, leading to the reported crash in a subsequent
suspend attempt.
In the suspend path, during CPU offline, the value of policy->cpu is updated
by choosing one of the surviving CPUs in that policy, as long as there is
atleast one CPU in that policy. And cpufreq_stats_update_policy_cpu() is
invoked to update the reference to the stats structure by assigning it to
the new CPU. However, in the resume path, during CPU online, we end up
assigning a fresh CPU as the policy->cpu, without letting cpufreq-stats
know about this. Thus the reference to the stats structure remains
(incorrectly) associated with the old CPU. So, in a subsequent suspend attempt,
during CPU offline, we end up accessing an incorrect location to get the
stats structure, which eventually leads to the NULL pointer dereference.
Fix this by letting cpufreq-stats know about the update of the policy->cpu
during CPU online in the resume path. (Also, move the update_policy_cpu()
function higher up in the file, so that __cpufreq_add_dev() can invoke
it).
Reported-and-tested-by: Stephen Warren <swarren@nvidia.com>
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2013-09-12 04:12:59 +08:00
|
|
|
policy->cpu = cpu;
|
|
|
|
|
2009-01-04 21:18:06 +08:00
|
|
|
cpumask_copy(policy->cpus, cpumask_of(cpu));
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
init_completion(&policy->kobj_unregister);
|
2006-11-22 22:55:48 +08:00
|
|
|
INIT_WORK(&policy->update, handle_update);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/* call driver. From then on the cpufreq must be able
|
|
|
|
* to accept all calls to ->verify and ->setpolicy for this CPU
|
|
|
|
*/
|
2013-04-29 06:08:16 +08:00
|
|
|
ret = cpufreq_driver->init(policy);
|
2005-04-17 06:20:36 +08:00
|
|
|
if (ret) {
|
2011-03-27 21:04:46 +08:00
|
|
|
pr_debug("initialization failed\n");
|
2013-02-07 13:25:00 +08:00
|
|
|
goto err_set_policy_cpu;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2013-01-12 13:14:38 +08:00
|
|
|
|
2014-11-24 17:08:03 +08:00
|
|
|
down_write(&policy->rwsem);
|
|
|
|
|
2014-03-04 11:44:00 +08:00
|
|
|
/* related cpus should atleast have policy->cpus */
|
|
|
|
cpumask_or(policy->related_cpus, policy->related_cpus, policy->cpus);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* affected cpus must always be the one, which are online. We aren't
|
|
|
|
* managing offline cpus here.
|
|
|
|
*/
|
|
|
|
cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
|
|
|
|
|
2014-03-10 17:23:35 +08:00
|
|
|
if (!recover_policy) {
|
2014-03-04 11:44:00 +08:00
|
|
|
policy->user_policy.min = policy->min;
|
|
|
|
policy->user_policy.max = policy->max;
|
2014-11-24 17:08:03 +08:00
|
|
|
|
|
|
|
/* prepare interface data */
|
|
|
|
ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
|
|
|
|
&dev->kobj, "cpufreq");
|
|
|
|
if (ret) {
|
|
|
|
pr_err("%s: failed to init policy->kobj: %d\n",
|
|
|
|
__func__, ret);
|
|
|
|
goto err_init_policy_kobj;
|
|
|
|
}
|
2014-03-04 11:44:00 +08:00
|
|
|
}
|
|
|
|
|
2014-01-09 23:08:43 +08:00
|
|
|
write_lock_irqsave(&cpufreq_driver_lock, flags);
|
|
|
|
for_each_cpu(j, policy->cpus)
|
|
|
|
per_cpu(cpufreq_cpu_data, j) = policy;
|
|
|
|
write_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
|
|
|
|
2014-03-13 04:49:33 +08:00
|
|
|
if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
|
2013-10-03 22:58:30 +08:00
|
|
|
policy->cur = cpufreq_driver->get(policy->cpu);
|
|
|
|
if (!policy->cur) {
|
|
|
|
pr_err("%s: ->get() failed\n", __func__);
|
|
|
|
goto err_get_freq;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2013-12-03 13:50:46 +08:00
|
|
|
/*
|
|
|
|
* Sometimes boot loaders set CPU frequency to a value outside of
|
|
|
|
* frequency table present with cpufreq core. In such cases CPU might be
|
|
|
|
* unstable if it has to run on that frequency for long duration of time
|
|
|
|
* and so its better to set it to a frequency which is specified in
|
|
|
|
* freq-table. This also makes cpufreq stats inconsistent as
|
|
|
|
* cpufreq-stats would fail to register because current frequency of CPU
|
|
|
|
* isn't found in freq-table.
|
|
|
|
*
|
|
|
|
* Because we don't want this change to effect boot process badly, we go
|
|
|
|
* for the next freq which is >= policy->cur ('cur' must be set by now,
|
|
|
|
* otherwise we will end up setting freq to lowest of the table as 'cur'
|
|
|
|
* is initialized to zero).
|
|
|
|
*
|
|
|
|
* We are passing target-freq as "policy->cur - 1" otherwise
|
|
|
|
* __cpufreq_driver_target() would simply fail, as policy->cur will be
|
|
|
|
* equal to target-freq.
|
|
|
|
*/
|
|
|
|
if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
|
|
|
|
&& has_target()) {
|
|
|
|
/* Are we running at unknown frequency ? */
|
|
|
|
ret = cpufreq_frequency_table_get_index(policy, policy->cur);
|
|
|
|
if (ret == -EINVAL) {
|
|
|
|
/* Warn user and fix it */
|
|
|
|
pr_warn("%s: CPU%d: Running at unlisted freq: %u KHz\n",
|
|
|
|
__func__, policy->cpu, policy->cur);
|
|
|
|
ret = __cpufreq_driver_target(policy, policy->cur - 1,
|
|
|
|
CPUFREQ_RELATION_L);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Reaching here after boot in a few seconds may not
|
|
|
|
* mean that system will remain stable at "unknown"
|
|
|
|
* frequency for longer duration. Hence, a BUG_ON().
|
|
|
|
*/
|
|
|
|
BUG_ON(ret);
|
|
|
|
pr_warn("%s: CPU%d: Unlisted initial frequency changed to: %u KHz\n",
|
|
|
|
__func__, policy->cpu, policy->cur);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2008-07-30 13:32:58 +08:00
|
|
|
blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
|
|
|
|
CPUFREQ_START, policy);
|
|
|
|
|
2014-03-10 17:23:35 +08:00
|
|
|
if (!recover_policy) {
|
2013-07-31 20:35:14 +08:00
|
|
|
ret = cpufreq_add_dev_interface(policy, dev);
|
2013-07-30 06:54:49 +08:00
|
|
|
if (ret)
|
|
|
|
goto err_out_unregister;
|
2014-01-07 09:40:10 +08:00
|
|
|
blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
|
|
|
|
CPUFREQ_CREATE_POLICY, policy);
|
2013-07-30 06:54:49 +08:00
|
|
|
}
|
2006-03-05 16:37:23 +08:00
|
|
|
|
2013-08-20 14:38:23 +08:00
|
|
|
write_lock_irqsave(&cpufreq_driver_lock, flags);
|
|
|
|
list_add(&policy->policy_list, &cpufreq_policy_list);
|
|
|
|
write_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
|
|
|
|
2013-07-30 06:54:23 +08:00
|
|
|
cpufreq_init_policy(policy);
|
|
|
|
|
2014-03-10 17:23:35 +08:00
|
|
|
if (!recover_policy) {
|
2013-12-24 09:41:01 +08:00
|
|
|
policy->user_policy.policy = policy->policy;
|
|
|
|
policy->user_policy.governor = policy->governor;
|
|
|
|
}
|
2014-03-04 11:44:01 +08:00
|
|
|
up_write(&policy->rwsem);
|
2013-12-24 09:41:01 +08:00
|
|
|
|
2007-12-18 03:54:39 +08:00
|
|
|
kobject_uevent(&policy->kobj, KOBJ_ADD);
|
2014-11-27 08:37:51 +08:00
|
|
|
|
2013-08-07 01:23:11 +08:00
|
|
|
up_read(&cpufreq_rwsem);
|
|
|
|
|
2014-11-27 08:37:51 +08:00
|
|
|
/* Callback for handling stuff after policy is ready */
|
|
|
|
if (cpufreq_driver->ready)
|
|
|
|
cpufreq_driver->ready(policy);
|
|
|
|
|
2011-03-27 21:04:46 +08:00
|
|
|
pr_debug("initialization complete\n");
|
2006-03-29 14:48:37 +08:00
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
return 0;
|
|
|
|
|
|
|
|
err_out_unregister:
|
2014-01-09 23:08:43 +08:00
|
|
|
err_get_freq:
|
2013-02-23 00:24:34 +08:00
|
|
|
write_lock_irqsave(&cpufreq_driver_lock, flags);
|
2013-08-20 14:38:25 +08:00
|
|
|
for_each_cpu(j, policy->cpus)
|
2008-03-26 06:06:53 +08:00
|
|
|
per_cpu(cpufreq_cpu_data, j) = NULL;
|
2013-02-23 00:24:34 +08:00
|
|
|
write_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-11-24 17:08:03 +08:00
|
|
|
if (!recover_policy) {
|
|
|
|
kobject_put(&policy->kobj);
|
|
|
|
wait_for_completion(&policy->kobj_unregister);
|
|
|
|
}
|
|
|
|
err_init_policy_kobj:
|
2014-09-10 22:12:08 +08:00
|
|
|
up_write(&policy->rwsem);
|
|
|
|
|
2013-10-03 22:58:30 +08:00
|
|
|
if (cpufreq_driver->exit)
|
|
|
|
cpufreq_driver->exit(policy);
|
2013-02-07 13:25:00 +08:00
|
|
|
err_set_policy_cpu:
|
2014-03-10 17:23:35 +08:00
|
|
|
if (recover_policy) {
|
2013-12-27 08:07:11 +08:00
|
|
|
/* Do not leave stale fallback data behind. */
|
|
|
|
per_cpu(cpufreq_cpu_data_fallback, cpu) = NULL;
|
2013-12-20 23:56:02 +08:00
|
|
|
cpufreq_policy_put_kobj(policy);
|
2013-12-27 08:07:11 +08:00
|
|
|
}
|
2013-07-30 06:54:11 +08:00
|
|
|
cpufreq_policy_free(policy);
|
2013-12-20 23:56:02 +08:00
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
nomem_out:
|
2013-08-07 01:23:11 +08:00
|
|
|
up_read(&cpufreq_rwsem);
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2013-07-30 06:54:49 +08:00
|
|
|
/**
|
|
|
|
* cpufreq_add_dev - add a CPU device
|
|
|
|
*
|
|
|
|
* Adds the cpufreq interface for a CPU device.
|
|
|
|
*
|
|
|
|
* The Oracle says: try running cpufreq registration/unregistration concurrently
|
|
|
|
* with with cpu hotplugging and all hell will break loose. Tried to clean this
|
|
|
|
* mess up, but more thorough testing is needed. - Mathieu
|
|
|
|
*/
|
|
|
|
static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
|
|
|
|
{
|
2014-03-10 17:23:35 +08:00
|
|
|
return __cpufreq_add_dev(dev, sif);
|
2013-07-30 06:54:49 +08:00
|
|
|
}
|
|
|
|
|
2013-09-07 03:53:09 +08:00
|
|
|
static int __cpufreq_remove_dev_prepare(struct device *dev,
|
2014-03-10 17:23:35 +08:00
|
|
|
struct subsys_interface *sif)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2013-07-30 06:54:36 +08:00
|
|
|
unsigned int cpu = dev->id, cpus;
|
2014-07-17 13:18:28 +08:00
|
|
|
int ret;
|
2005-04-17 06:20:36 +08:00
|
|
|
unsigned long flags;
|
2013-08-07 01:23:05 +08:00
|
|
|
struct cpufreq_policy *policy;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-01-14 21:23:03 +08:00
|
|
|
pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-02-23 00:24:34 +08:00
|
|
|
write_lock_irqsave(&cpufreq_driver_lock, flags);
|
2013-02-07 13:25:00 +08:00
|
|
|
|
2013-08-07 01:23:05 +08:00
|
|
|
policy = per_cpu(cpufreq_cpu_data, cpu);
|
2013-02-07 13:25:00 +08:00
|
|
|
|
cpufreq: Preserve policy structure across suspend/resume
To perform light-weight cpu-init and teardown in the cpufreq subsystem
during suspend/resume, we need to separate out the 2 main functionalities
of the cpufreq CPU hotplug callbacks, as outlined below:
1. Init/tear-down of core cpufreq and CPU-specific components, which are
critical to the correct functioning of the cpufreq subsystem.
2. Init/tear-down of cpufreq sysfs files during suspend/resume.
The first part requires accurate updates to the policy structure such as
its ->cpus and ->related_cpus masks, whereas the second part requires that
the policy->kobj structure is not released or re-initialized during
suspend/resume.
To handle both these requirements, we need to allow updates to the policy
structure throughout suspend/resume, but prevent the structure from getting
freed up. Also, we must have a mechanism by which the cpu-up callbacks can
restore the policy structure, without allocating things afresh. (That also
helps avoid memory leaks).
To achieve this, we use 2 schemes:
a. Use a fallback per-cpu storage area for preserving the policy structures
during suspend, so that they can be restored during resume appropriately.
b. Use the 'frozen' flag to determine when to free or allocate the policy
structure vs when to restore the policy from the saved fallback storage.
Thus we can successfully preserve the structure across suspend/resume.
Effectively, this helps us complete the separation of the 'light-weight'
and the 'full' init/tear-down sequences in the cpufreq subsystem, so that
this can be made use of in the suspend/resume scenario.
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>
2013-07-30 06:55:10 +08:00
|
|
|
/* Save the policy somewhere when doing a light-weight tear-down */
|
2014-03-10 17:23:35 +08:00
|
|
|
if (cpufreq_suspended)
|
2013-08-07 01:23:05 +08:00
|
|
|
per_cpu(cpufreq_cpu_data_fallback, cpu) = policy;
|
cpufreq: Preserve policy structure across suspend/resume
To perform light-weight cpu-init and teardown in the cpufreq subsystem
during suspend/resume, we need to separate out the 2 main functionalities
of the cpufreq CPU hotplug callbacks, as outlined below:
1. Init/tear-down of core cpufreq and CPU-specific components, which are
critical to the correct functioning of the cpufreq subsystem.
2. Init/tear-down of cpufreq sysfs files during suspend/resume.
The first part requires accurate updates to the policy structure such as
its ->cpus and ->related_cpus masks, whereas the second part requires that
the policy->kobj structure is not released or re-initialized during
suspend/resume.
To handle both these requirements, we need to allow updates to the policy
structure throughout suspend/resume, but prevent the structure from getting
freed up. Also, we must have a mechanism by which the cpu-up callbacks can
restore the policy structure, without allocating things afresh. (That also
helps avoid memory leaks).
To achieve this, we use 2 schemes:
a. Use a fallback per-cpu storage area for preserving the policy structures
during suspend, so that they can be restored during resume appropriately.
b. Use the 'frozen' flag to determine when to free or allocate the policy
structure vs when to restore the policy from the saved fallback storage.
Thus we can successfully preserve the structure across suspend/resume.
Effectively, this helps us complete the separation of the 'light-weight'
and the 'full' init/tear-down sequences in the cpufreq subsystem, so that
this can be made use of in the suspend/resume scenario.
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>
2013-07-30 06:55:10 +08:00
|
|
|
|
2013-02-23 00:24:34 +08:00
|
|
|
write_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-08-07 01:23:05 +08:00
|
|
|
if (!policy) {
|
2013-01-14 21:23:03 +08:00
|
|
|
pr_debug("%s: No cpu_data found\n", __func__);
|
2005-04-17 06:20:36 +08:00
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
2013-10-25 22:15:48 +08:00
|
|
|
if (has_target()) {
|
2013-08-07 01:23:13 +08:00
|
|
|
ret = __cpufreq_governor(policy, CPUFREQ_GOV_STOP);
|
|
|
|
if (ret) {
|
|
|
|
pr_err("%s: Failed to stop governor\n", __func__);
|
|
|
|
return ret;
|
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-02-07 01:02:11 +08:00
|
|
|
strncpy(per_cpu(cpufreq_cpu_governor, cpu),
|
2013-08-07 01:23:05 +08:00
|
|
|
policy->governor->name, CPUFREQ_NAME_LEN);
|
2015-01-02 15:04:25 +08:00
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-10-18 21:40:15 +08:00
|
|
|
down_read(&policy->rwsem);
|
2013-08-07 01:23:05 +08:00
|
|
|
cpus = cpumask_weight(policy->cpus);
|
2013-10-18 21:40:15 +08:00
|
|
|
up_read(&policy->rwsem);
|
2007-07-10 02:35:28 +08:00
|
|
|
|
cpufreq: Restructure if/else block to avoid unintended behavior
In __cpufreq_remove_dev_prepare(), the code which decides whether to remove
the sysfs link or nominate a new policy cpu, is governed by an if/else block
with a rather complex set of conditionals. Worse, they harbor a subtlety
which leads to certain unintended behavior.
The code looks like this:
if (cpu != policy->cpu && !frozen) {
sysfs_remove_link(&dev->kobj, "cpufreq");
} else if (cpus > 1) {
new_cpu = cpufreq_nominate_new_policy_cpu(...);
...
update_policy_cpu(..., new_cpu);
}
The original intention was:
If the CPU going offline is not policy->cpu, just remove the link.
On the other hand, if the CPU going offline is the policy->cpu itself,
handover the policy->cpu job to some other surviving CPU in that policy.
But because the 'if' condition also includes the 'frozen' check, now there
are *two* possibilities by which we can enter the 'else' block:
1. cpu == policy->cpu (intended)
2. cpu != policy->cpu && frozen (unintended)
Due to the second (unintended) scenario, we end up spuriously nominating
a CPU as the policy->cpu, even when the existing policy->cpu is alive and
well. This can cause problems further down the line, especially when we end
up nominating the same policy->cpu as the new one (ie., old == new),
because it totally confuses update_policy_cpu().
To avoid this mess, restructure the if/else block to only do what was
originally intended, and thus prevent any unwelcome surprises.
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Tested-by: Stephen Warren <swarren@nvidia.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2013-09-12 04:13:25 +08:00
|
|
|
if (cpu != policy->cpu) {
|
2014-02-17 17:22:11 +08:00
|
|
|
sysfs_remove_link(&dev->kobj, "cpufreq");
|
2013-02-06 05:21:14 +08:00
|
|
|
} else if (cpus > 1) {
|
2014-07-17 13:18:28 +08:00
|
|
|
/* Nominate new CPU */
|
|
|
|
int new_cpu = cpumask_any_but(policy->cpus, cpu);
|
|
|
|
struct device *cpu_dev = get_cpu_device(new_cpu);
|
2013-07-30 06:54:49 +08:00
|
|
|
|
2014-07-17 13:18:28 +08:00
|
|
|
sysfs_remove_link(&cpu_dev->kobj, "cpufreq");
|
|
|
|
ret = update_policy_cpu(policy, new_cpu, cpu_dev);
|
|
|
|
if (ret) {
|
|
|
|
if (sysfs_create_link(&cpu_dev->kobj, &policy->kobj,
|
|
|
|
"cpufreq"))
|
|
|
|
pr_err("%s: Failed to restore kobj link to cpu:%d\n",
|
|
|
|
__func__, cpu_dev->id);
|
|
|
|
return ret;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2014-07-17 13:18:28 +08:00
|
|
|
|
|
|
|
if (!cpufreq_suspended)
|
|
|
|
pr_debug("%s: policy Kobject moved to cpu: %d from: %d\n",
|
|
|
|
__func__, new_cpu, cpu);
|
2014-09-29 21:47:12 +08:00
|
|
|
} else if (cpufreq_driver->stop_cpu) {
|
2014-03-19 23:45:53 +08:00
|
|
|
cpufreq_driver->stop_cpu(policy);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2013-09-07 03:53:09 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int __cpufreq_remove_dev_finish(struct device *dev,
|
2014-03-10 17:23:35 +08:00
|
|
|
struct subsys_interface *sif)
|
2013-09-07 03:53:09 +08:00
|
|
|
{
|
|
|
|
unsigned int cpu = dev->id, cpus;
|
|
|
|
int ret;
|
|
|
|
unsigned long flags;
|
|
|
|
struct cpufreq_policy *policy;
|
|
|
|
|
|
|
|
read_lock_irqsave(&cpufreq_driver_lock, flags);
|
|
|
|
policy = per_cpu(cpufreq_cpu_data, cpu);
|
|
|
|
read_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
|
|
|
|
|
|
|
if (!policy) {
|
|
|
|
pr_debug("%s: No cpu_data found\n", __func__);
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
2013-10-18 21:40:15 +08:00
|
|
|
down_write(&policy->rwsem);
|
2013-09-07 03:53:09 +08:00
|
|
|
cpus = cpumask_weight(policy->cpus);
|
2013-09-12 19:36:33 +08:00
|
|
|
|
|
|
|
if (cpus > 1)
|
|
|
|
cpumask_clear_cpu(cpu, policy->cpus);
|
2013-10-18 21:40:15 +08:00
|
|
|
up_write(&policy->rwsem);
|
2013-09-07 03:53:09 +08:00
|
|
|
|
2013-01-14 21:23:03 +08:00
|
|
|
/* If cpu is last user of policy, free policy */
|
|
|
|
if (cpus == 1) {
|
2013-10-25 22:15:48 +08:00
|
|
|
if (has_target()) {
|
2013-08-07 01:23:13 +08:00
|
|
|
ret = __cpufreq_governor(policy,
|
|
|
|
CPUFREQ_GOV_POLICY_EXIT);
|
|
|
|
if (ret) {
|
|
|
|
pr_err("%s: Failed to exit governor\n",
|
2014-03-12 01:03:00 +08:00
|
|
|
__func__);
|
2013-08-07 01:23:13 +08:00
|
|
|
return ret;
|
|
|
|
}
|
2013-08-20 14:38:22 +08:00
|
|
|
}
|
cpufreq: Fix cpufreq driver module refcount balance after suspend/resume
Since cpufreq_cpu_put() called by __cpufreq_remove_dev() drops the
driver module refcount, __cpufreq_remove_dev() causes that refcount
to become negative for the cpufreq driver after a suspend/resume
cycle.
This is not the only bad thing that happens there, however, because
kobject_put() should only be called for the policy kobject at this
point if the CPU is not the last one for that policy.
Namely, if the given CPU is the last one for that policy, the
policy kobject's refcount should be 1 at this point, as set by
cpufreq_add_dev_interface(), and only needs to be dropped once for
the kobject to go away. This actually happens under the cpu == 1
check, so it need not be done before by cpufreq_cpu_put().
On the other hand, if the given CPU is not the last one for that
policy, this means that cpufreq_add_policy_cpu() has been called
at least once for that policy and cpufreq_cpu_get() has been
called for it too. To balance that cpufreq_cpu_get(), we need to
call cpufreq_cpu_put() in that case.
Thus, to fix the described problem and keep the reference
counters balanced in both cases, move the cpufreq_cpu_get() call
in __cpufreq_remove_dev() to the code path executed only for
CPUs that share the policy with other CPUs.
Reported-and-tested-by: Toralf Förster <toralf.foerster@gmx.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reviewed-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Cc: 3.10+ <stable@vger.kernel.org>
2013-07-30 06:32:00 +08:00
|
|
|
|
2014-03-10 17:23:35 +08:00
|
|
|
if (!cpufreq_suspended)
|
2013-12-20 23:56:02 +08:00
|
|
|
cpufreq_policy_put_kobj(policy);
|
2009-07-03 08:08:30 +08:00
|
|
|
|
cpufreq: Preserve policy structure across suspend/resume
To perform light-weight cpu-init and teardown in the cpufreq subsystem
during suspend/resume, we need to separate out the 2 main functionalities
of the cpufreq CPU hotplug callbacks, as outlined below:
1. Init/tear-down of core cpufreq and CPU-specific components, which are
critical to the correct functioning of the cpufreq subsystem.
2. Init/tear-down of cpufreq sysfs files during suspend/resume.
The first part requires accurate updates to the policy structure such as
its ->cpus and ->related_cpus masks, whereas the second part requires that
the policy->kobj structure is not released or re-initialized during
suspend/resume.
To handle both these requirements, we need to allow updates to the policy
structure throughout suspend/resume, but prevent the structure from getting
freed up. Also, we must have a mechanism by which the cpu-up callbacks can
restore the policy structure, without allocating things afresh. (That also
helps avoid memory leaks).
To achieve this, we use 2 schemes:
a. Use a fallback per-cpu storage area for preserving the policy structures
during suspend, so that they can be restored during resume appropriately.
b. Use the 'frozen' flag to determine when to free or allocate the policy
structure vs when to restore the policy from the saved fallback storage.
Thus we can successfully preserve the structure across suspend/resume.
Effectively, this helps us complete the separation of the 'light-weight'
and the 'full' init/tear-down sequences in the cpufreq subsystem, so that
this can be made use of in the suspend/resume scenario.
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>
2013-07-30 06:55:10 +08:00
|
|
|
/*
|
|
|
|
* Perform the ->exit() even during light-weight tear-down,
|
|
|
|
* since this is a core component, and is essential for the
|
|
|
|
* subsequent light-weight ->init() to succeed.
|
2013-01-14 21:23:03 +08:00
|
|
|
*/
|
2013-04-29 06:08:16 +08:00
|
|
|
if (cpufreq_driver->exit)
|
2013-08-07 01:23:05 +08:00
|
|
|
cpufreq_driver->exit(policy);
|
[CPUFREQ] CPU hotplug, re-create sysfs directory and symlinks
When we discover CPUs that are affected by each other's
frequency/voltage transitions, the first CPU gets a sysfs directory
created, and rest of the siblings get symlinks. Currently, when we
hotplug off only the first CPU, all of the symlinks and the sysfs
directory gets removed. Even though rest of the siblings are still
online and functional, they are orphaned, and no longer governed by
cpufreq.
This patch, given the above scenario, creates a sysfs directory for
the first sibling and symlinks for the rest of the siblings.
Please note the recursive call, it was rather too ugly to roll it
out. And the removal of redundant NULL setting (it is already taken
care of near the top of the function).
Signed-off-by: Jacob Shin <jacob.shin@amd.com>
Acked-by: Mark Langsdorf <mark.langsdorf@amd.com>
Reviewed-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Dave Jones <davej@redhat.com>
Cc: stable@kernel.org
2011-04-28 02:32:11 +08:00
|
|
|
|
2013-08-20 14:38:23 +08:00
|
|
|
/* Remove policy from list of active policies */
|
|
|
|
write_lock_irqsave(&cpufreq_driver_lock, flags);
|
|
|
|
list_del(&policy->policy_list);
|
|
|
|
write_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
|
|
|
|
2014-03-10 17:23:35 +08:00
|
|
|
if (!cpufreq_suspended)
|
2013-08-07 01:23:05 +08:00
|
|
|
cpufreq_policy_free(policy);
|
2014-03-20 05:29:17 +08:00
|
|
|
} else if (has_target()) {
|
|
|
|
ret = __cpufreq_governor(policy, CPUFREQ_GOV_START);
|
|
|
|
if (!ret)
|
|
|
|
ret = __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
|
|
|
|
|
|
|
|
if (ret) {
|
|
|
|
pr_err("%s: Failed to start governor\n", __func__);
|
|
|
|
return ret;
|
cpufreq: Fix cpufreq driver module refcount balance after suspend/resume
Since cpufreq_cpu_put() called by __cpufreq_remove_dev() drops the
driver module refcount, __cpufreq_remove_dev() causes that refcount
to become negative for the cpufreq driver after a suspend/resume
cycle.
This is not the only bad thing that happens there, however, because
kobject_put() should only be called for the policy kobject at this
point if the CPU is not the last one for that policy.
Namely, if the given CPU is the last one for that policy, the
policy kobject's refcount should be 1 at this point, as set by
cpufreq_add_dev_interface(), and only needs to be dropped once for
the kobject to go away. This actually happens under the cpu == 1
check, so it need not be done before by cpufreq_cpu_put().
On the other hand, if the given CPU is not the last one for that
policy, this means that cpufreq_add_policy_cpu() has been called
at least once for that policy and cpufreq_cpu_get() has been
called for it too. To balance that cpufreq_cpu_get(), we need to
call cpufreq_cpu_put() in that case.
Thus, to fix the described problem and keep the reference
counters balanced in both cases, move the cpufreq_cpu_get() call
in __cpufreq_remove_dev() to the code path executed only for
CPUs that share the policy with other CPUs.
Reported-and-tested-by: Toralf Förster <toralf.foerster@gmx.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reviewed-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Cc: 3.10+ <stable@vger.kernel.org>
2013-07-30 06:32:00 +08:00
|
|
|
}
|
[CPUFREQ] CPU hotplug, re-create sysfs directory and symlinks
When we discover CPUs that are affected by each other's
frequency/voltage transitions, the first CPU gets a sysfs directory
created, and rest of the siblings get symlinks. Currently, when we
hotplug off only the first CPU, all of the symlinks and the sysfs
directory gets removed. Even though rest of the siblings are still
online and functional, they are orphaned, and no longer governed by
cpufreq.
This patch, given the above scenario, creates a sysfs directory for
the first sibling and symlinks for the rest of the siblings.
Please note the recursive call, it was rather too ugly to roll it
out. And the removal of redundant NULL setting (it is already taken
care of near the top of the function).
Signed-off-by: Jacob Shin <jacob.shin@amd.com>
Acked-by: Mark Langsdorf <mark.langsdorf@amd.com>
Reviewed-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Dave Jones <davej@redhat.com>
Cc: stable@kernel.org
2011-04-28 02:32:11 +08:00
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-08-20 14:38:25 +08:00
|
|
|
per_cpu(cpufreq_cpu_data, cpu) = NULL;
|
2005-04-17 06:20:36 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2013-09-07 03:53:09 +08:00
|
|
|
/**
|
2013-10-02 16:43:14 +08:00
|
|
|
* cpufreq_remove_dev - remove a CPU device
|
2013-09-07 03:53:09 +08:00
|
|
|
*
|
|
|
|
* Removes the cpufreq interface for a CPU device.
|
|
|
|
*/
|
2011-12-22 06:29:42 +08:00
|
|
|
static int cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
|
2007-02-06 08:12:44 +08:00
|
|
|
{
|
2011-12-22 06:29:42 +08:00
|
|
|
unsigned int cpu = dev->id;
|
2013-10-02 16:43:14 +08:00
|
|
|
int ret;
|
2007-03-27 03:03:19 +08:00
|
|
|
|
|
|
|
if (cpu_is_offline(cpu))
|
|
|
|
return 0;
|
|
|
|
|
2014-03-10 17:23:35 +08:00
|
|
|
ret = __cpufreq_remove_dev_prepare(dev, sif);
|
2013-10-02 16:43:14 +08:00
|
|
|
|
|
|
|
if (!ret)
|
2014-03-10 17:23:35 +08:00
|
|
|
ret = __cpufreq_remove_dev_finish(dev, sif);
|
2013-10-02 16:43:14 +08:00
|
|
|
|
|
|
|
return ret;
|
2007-02-06 08:12:44 +08:00
|
|
|
}
|
|
|
|
|
2006-11-22 22:55:48 +08:00
|
|
|
static void handle_update(struct work_struct *work)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2006-11-22 22:55:48 +08:00
|
|
|
struct cpufreq_policy *policy =
|
|
|
|
container_of(work, struct cpufreq_policy, update);
|
|
|
|
unsigned int cpu = policy->cpu;
|
2011-03-27 21:04:46 +08:00
|
|
|
pr_debug("handle_update for cpu %u called\n", cpu);
|
2005-04-17 06:20:36 +08:00
|
|
|
cpufreq_update_policy(cpu);
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
2013-06-19 16:49:33 +08:00
|
|
|
* cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're
|
|
|
|
* in deep trouble.
|
2015-01-02 15:04:28 +08:00
|
|
|
* @policy: policy managing CPUs
|
2005-04-17 06:20:36 +08:00
|
|
|
* @new_freq: CPU frequency the CPU actually runs at
|
|
|
|
*
|
2009-01-18 14:37:11 +08:00
|
|
|
* We adjust to current frequency first, and need to clean up later.
|
|
|
|
* So either call to cpufreq_update_policy() or schedule handle_update()).
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
2015-01-02 15:04:28 +08:00
|
|
|
static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
|
2006-10-26 18:50:58 +08:00
|
|
|
unsigned int new_freq)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
struct cpufreq_freqs freqs;
|
2013-03-24 14:26:43 +08:00
|
|
|
|
2014-03-12 01:03:00 +08:00
|
|
|
pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
|
2015-01-02 15:04:28 +08:00
|
|
|
policy->cur, new_freq);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2015-01-02 15:04:28 +08:00
|
|
|
freqs.old = policy->cur;
|
2005-04-17 06:20:36 +08:00
|
|
|
freqs.new = new_freq;
|
2013-03-24 14:26:43 +08:00
|
|
|
|
2014-03-24 16:05:45 +08:00
|
|
|
cpufreq_freq_transition_begin(policy, &freqs);
|
|
|
|
cpufreq_freq_transition_end(policy, &freqs, 0);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2006-02-28 13:43:23 +08:00
|
|
|
/**
|
2006-12-13 17:19:15 +08:00
|
|
|
* cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
|
2005-12-03 02:43:20 +08:00
|
|
|
* @cpu: CPU number
|
|
|
|
*
|
|
|
|
* This is the last known freq, without actually getting it from the driver.
|
|
|
|
* Return value will be same as what is shown in scaling_cur_freq in sysfs.
|
|
|
|
*/
|
|
|
|
unsigned int cpufreq_quick_get(unsigned int cpu)
|
|
|
|
{
|
2013-02-07 01:02:08 +08:00
|
|
|
struct cpufreq_policy *policy;
|
2006-10-26 18:50:58 +08:00
|
|
|
unsigned int ret_freq = 0;
|
2005-12-03 02:43:20 +08:00
|
|
|
|
2013-04-29 06:08:16 +08:00
|
|
|
if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get)
|
|
|
|
return cpufreq_driver->get(cpu);
|
2013-02-07 01:02:08 +08:00
|
|
|
|
|
|
|
policy = cpufreq_cpu_get(cpu);
|
2005-12-03 02:43:20 +08:00
|
|
|
if (policy) {
|
2006-10-26 18:50:58 +08:00
|
|
|
ret_freq = policy->cur;
|
2005-12-03 02:43:20 +08:00
|
|
|
cpufreq_cpu_put(policy);
|
|
|
|
}
|
|
|
|
|
2008-02-08 05:33:49 +08:00
|
|
|
return ret_freq;
|
2005-12-03 02:43:20 +08:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(cpufreq_quick_get);
|
|
|
|
|
2011-06-29 01:59:12 +08:00
|
|
|
/**
|
|
|
|
* cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
|
|
|
|
* @cpu: CPU number
|
|
|
|
*
|
|
|
|
* Just return the max possible frequency for a given CPU.
|
|
|
|
*/
|
|
|
|
unsigned int cpufreq_quick_get_max(unsigned int cpu)
|
|
|
|
{
|
|
|
|
struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
|
|
|
|
unsigned int ret_freq = 0;
|
|
|
|
|
|
|
|
if (policy) {
|
|
|
|
ret_freq = policy->max;
|
|
|
|
cpufreq_cpu_put(policy);
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret_freq;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(cpufreq_quick_get_max);
|
|
|
|
|
2015-01-02 15:04:29 +08:00
|
|
|
static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2006-10-26 18:50:58 +08:00
|
|
|
unsigned int ret_freq = 0;
|
2013-04-04 22:53:25 +08:00
|
|
|
|
2013-04-29 06:08:16 +08:00
|
|
|
if (!cpufreq_driver->get)
|
2008-02-08 05:33:49 +08:00
|
|
|
return ret_freq;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2015-01-02 15:04:29 +08:00
|
|
|
ret_freq = cpufreq_driver->get(policy->cpu);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2006-10-26 18:50:58 +08:00
|
|
|
if (ret_freq && policy->cur &&
|
2013-04-29 06:08:16 +08:00
|
|
|
!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
|
2006-10-26 18:50:58 +08:00
|
|
|
/* verify no discrepancy between actual and
|
|
|
|
saved value exists */
|
|
|
|
if (unlikely(ret_freq != policy->cur)) {
|
2015-01-02 15:04:28 +08:00
|
|
|
cpufreq_out_of_sync(policy, ret_freq);
|
2005-04-17 06:20:36 +08:00
|
|
|
schedule_work(&policy->update);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2008-02-08 05:33:49 +08:00
|
|
|
return ret_freq;
|
2007-02-06 08:12:44 +08:00
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2007-02-06 08:12:44 +08:00
|
|
|
/**
|
|
|
|
* cpufreq_get - get the current CPU frequency (in kHz)
|
|
|
|
* @cpu: CPU number
|
|
|
|
*
|
|
|
|
* Get the CPU current (static) CPU frequency
|
|
|
|
*/
|
|
|
|
unsigned int cpufreq_get(unsigned int cpu)
|
|
|
|
{
|
2014-03-05 04:42:15 +08:00
|
|
|
struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
|
2007-02-06 08:12:44 +08:00
|
|
|
unsigned int ret_freq = 0;
|
|
|
|
|
2014-03-05 04:42:15 +08:00
|
|
|
if (policy) {
|
|
|
|
down_read(&policy->rwsem);
|
2015-01-02 15:04:29 +08:00
|
|
|
ret_freq = __cpufreq_get(policy);
|
2014-03-05 04:42:15 +08:00
|
|
|
up_read(&policy->rwsem);
|
2007-02-06 08:12:44 +08:00
|
|
|
|
2014-03-05 04:42:15 +08:00
|
|
|
cpufreq_cpu_put(policy);
|
|
|
|
}
|
2013-08-07 01:23:11 +08:00
|
|
|
|
2008-02-08 05:33:49 +08:00
|
|
|
return ret_freq;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(cpufreq_get);
|
|
|
|
|
2011-12-22 06:29:42 +08:00
|
|
|
static struct subsys_interface cpufreq_interface = {
|
|
|
|
.name = "cpufreq",
|
|
|
|
.subsys = &cpu_subsys,
|
|
|
|
.add_dev = cpufreq_add_dev,
|
|
|
|
.remove_dev = cpufreq_remove_dev,
|
cpufreq: Use syscore_ops for boot CPU suspend/resume (v2)
The cpufreq subsystem uses sysdev suspend and resume for
executing cpufreq_suspend() and cpufreq_resume(), respectively,
during system suspend, after interrupts have been switched off on the
boot CPU, and during system resume, while interrupts are still off on
the boot CPU. In both cases the other CPUs are off-line at the
relevant point (either they have been switched off via CPU hotplug
during suspend, or they haven't been switched on yet during resume).
For this reason, although it may seem that cpufreq_suspend() and
cpufreq_resume() are executed for all CPUs in the system, they are
only called for the boot CPU in fact, which is quite confusing.
To remove the confusion and to prepare for elimiating sysdev
suspend and resume operations from the kernel enirely, convernt
cpufreq to using a struct syscore_ops object for the boot CPU
suspend and resume and rename the callbacks so that their names
reflect their purpose. In addition, put some explanatory remarks
into their kerneldoc comments.
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
2011-03-24 05:16:32 +08:00
|
|
|
};
|
|
|
|
|
2014-03-04 11:00:27 +08:00
|
|
|
/*
|
|
|
|
* In case platform wants some specific frequency to be configured
|
|
|
|
* during suspend..
|
|
|
|
*/
|
|
|
|
int cpufreq_generic_suspend(struct cpufreq_policy *policy)
|
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
if (!policy->suspend_freq) {
|
|
|
|
pr_err("%s: suspend_freq can't be zero\n", __func__);
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
pr_debug("%s: Setting suspend-freq: %u\n", __func__,
|
|
|
|
policy->suspend_freq);
|
|
|
|
|
|
|
|
ret = __cpufreq_driver_target(policy, policy->suspend_freq,
|
|
|
|
CPUFREQ_RELATION_H);
|
|
|
|
if (ret)
|
|
|
|
pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
|
|
|
|
__func__, policy->suspend_freq, ret);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(cpufreq_generic_suspend);
|
|
|
|
|
2005-04-29 22:40:12 +08:00
|
|
|
/**
|
2014-03-04 11:00:26 +08:00
|
|
|
* cpufreq_suspend() - Suspend CPUFreq governors
|
cpufreq: Use syscore_ops for boot CPU suspend/resume (v2)
The cpufreq subsystem uses sysdev suspend and resume for
executing cpufreq_suspend() and cpufreq_resume(), respectively,
during system suspend, after interrupts have been switched off on the
boot CPU, and during system resume, while interrupts are still off on
the boot CPU. In both cases the other CPUs are off-line at the
relevant point (either they have been switched off via CPU hotplug
during suspend, or they haven't been switched on yet during resume).
For this reason, although it may seem that cpufreq_suspend() and
cpufreq_resume() are executed for all CPUs in the system, they are
only called for the boot CPU in fact, which is quite confusing.
To remove the confusion and to prepare for elimiating sysdev
suspend and resume operations from the kernel enirely, convernt
cpufreq to using a struct syscore_ops object for the boot CPU
suspend and resume and rename the callbacks so that their names
reflect their purpose. In addition, put some explanatory remarks
into their kerneldoc comments.
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
2011-03-24 05:16:32 +08:00
|
|
|
*
|
2014-03-04 11:00:26 +08:00
|
|
|
* Called during system wide Suspend/Hibernate cycles for suspending governors
|
|
|
|
* as some platforms can't change frequency after this point in suspend cycle.
|
|
|
|
* Because some of the devices (like: i2c, regulators, etc) they use for
|
|
|
|
* changing frequency are suspended quickly after this point.
|
2005-04-29 22:40:12 +08:00
|
|
|
*/
|
2014-03-04 11:00:26 +08:00
|
|
|
void cpufreq_suspend(void)
|
2005-04-29 22:40:12 +08:00
|
|
|
{
|
2013-08-07 01:23:05 +08:00
|
|
|
struct cpufreq_policy *policy;
|
2005-04-29 22:40:12 +08:00
|
|
|
|
2014-03-04 11:00:26 +08:00
|
|
|
if (!cpufreq_driver)
|
|
|
|
return;
|
2005-04-29 22:40:12 +08:00
|
|
|
|
2014-03-04 11:00:26 +08:00
|
|
|
if (!has_target())
|
2014-09-30 12:03:17 +08:00
|
|
|
goto suspend;
|
2005-04-29 22:40:12 +08:00
|
|
|
|
2014-03-04 11:00:26 +08:00
|
|
|
pr_debug("%s: Suspending Governors\n", __func__);
|
|
|
|
|
|
|
|
list_for_each_entry(policy, &cpufreq_policy_list, policy_list) {
|
|
|
|
if (__cpufreq_governor(policy, CPUFREQ_GOV_STOP))
|
|
|
|
pr_err("%s: Failed to stop governor for policy: %p\n",
|
|
|
|
__func__, policy);
|
|
|
|
else if (cpufreq_driver->suspend
|
|
|
|
&& cpufreq_driver->suspend(policy))
|
|
|
|
pr_err("%s: Failed to suspend driver: %p\n", __func__,
|
|
|
|
policy);
|
2005-04-29 22:40:12 +08:00
|
|
|
}
|
2014-09-30 12:03:17 +08:00
|
|
|
|
|
|
|
suspend:
|
|
|
|
cpufreq_suspended = true;
|
2005-04-29 22:40:12 +08:00
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/**
|
2014-03-04 11:00:26 +08:00
|
|
|
* cpufreq_resume() - Resume CPUFreq governors
|
2005-04-17 06:20:36 +08:00
|
|
|
*
|
2014-03-04 11:00:26 +08:00
|
|
|
* Called during system wide Suspend/Hibernate cycle for resuming governors that
|
|
|
|
* are suspended with cpufreq_suspend().
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
2014-03-04 11:00:26 +08:00
|
|
|
void cpufreq_resume(void)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2013-08-07 01:23:05 +08:00
|
|
|
struct cpufreq_policy *policy;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-03-04 11:00:26 +08:00
|
|
|
if (!cpufreq_driver)
|
|
|
|
return;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-09-18 15:03:07 +08:00
|
|
|
cpufreq_suspended = false;
|
|
|
|
|
2014-03-04 11:00:26 +08:00
|
|
|
if (!has_target())
|
cpufreq: Use syscore_ops for boot CPU suspend/resume (v2)
The cpufreq subsystem uses sysdev suspend and resume for
executing cpufreq_suspend() and cpufreq_resume(), respectively,
during system suspend, after interrupts have been switched off on the
boot CPU, and during system resume, while interrupts are still off on
the boot CPU. In both cases the other CPUs are off-line at the
relevant point (either they have been switched off via CPU hotplug
during suspend, or they haven't been switched on yet during resume).
For this reason, although it may seem that cpufreq_suspend() and
cpufreq_resume() are executed for all CPUs in the system, they are
only called for the boot CPU in fact, which is quite confusing.
To remove the confusion and to prepare for elimiating sysdev
suspend and resume operations from the kernel enirely, convernt
cpufreq to using a struct syscore_ops object for the boot CPU
suspend and resume and rename the callbacks so that their names
reflect their purpose. In addition, put some explanatory remarks
into their kerneldoc comments.
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
2011-03-24 05:16:32 +08:00
|
|
|
return;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-03-04 11:00:26 +08:00
|
|
|
pr_debug("%s: Resuming Governors\n", __func__);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-03-04 11:00:26 +08:00
|
|
|
list_for_each_entry(policy, &cpufreq_policy_list, policy_list) {
|
2014-03-24 15:00:29 +08:00
|
|
|
if (cpufreq_driver->resume && cpufreq_driver->resume(policy))
|
|
|
|
pr_err("%s: Failed to resume driver: %p\n", __func__,
|
|
|
|
policy);
|
|
|
|
else if (__cpufreq_governor(policy, CPUFREQ_GOV_START)
|
2014-03-04 11:00:26 +08:00
|
|
|
|| __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS))
|
|
|
|
pr_err("%s: Failed to start governor for policy: %p\n",
|
|
|
|
__func__, policy);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-03-04 11:00:26 +08:00
|
|
|
/*
|
|
|
|
* schedule call cpufreq_update_policy() for boot CPU, i.e. last
|
|
|
|
* policy in list. It will verify that the current freq is in
|
|
|
|
* sync with what we believe it to be.
|
|
|
|
*/
|
|
|
|
if (list_is_last(&policy->policy_list, &cpufreq_policy_list))
|
|
|
|
schedule_work(&policy->update);
|
|
|
|
}
|
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-01-20 18:24:28 +08:00
|
|
|
/**
|
|
|
|
* cpufreq_get_current_driver - return current driver's name
|
|
|
|
*
|
|
|
|
* Return the name string of the currently loaded cpufreq driver
|
|
|
|
* or NULL, if none.
|
|
|
|
*/
|
|
|
|
const char *cpufreq_get_current_driver(void)
|
|
|
|
{
|
2013-04-29 06:08:16 +08:00
|
|
|
if (cpufreq_driver)
|
|
|
|
return cpufreq_driver->name;
|
|
|
|
|
|
|
|
return NULL;
|
2013-01-20 18:24:28 +08:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-10-19 17:30:27 +08:00
|
|
|
/**
|
|
|
|
* cpufreq_get_driver_data - return current driver data
|
|
|
|
*
|
|
|
|
* Return the private data of the currently loaded cpufreq
|
|
|
|
* driver, or NULL if no cpufreq driver is loaded.
|
|
|
|
*/
|
|
|
|
void *cpufreq_get_driver_data(void)
|
|
|
|
{
|
|
|
|
if (cpufreq_driver)
|
|
|
|
return cpufreq_driver->driver_data;
|
|
|
|
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/*********************************************************************
|
|
|
|
* NOTIFIER LISTS INTERFACE *
|
|
|
|
*********************************************************************/
|
|
|
|
|
|
|
|
/**
|
|
|
|
* cpufreq_register_notifier - register a driver with cpufreq
|
|
|
|
* @nb: notifier function to register
|
|
|
|
* @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
|
|
|
|
*
|
2006-02-28 13:43:23 +08:00
|
|
|
* Add a driver to one of two lists: either a list of drivers that
|
2005-04-17 06:20:36 +08:00
|
|
|
* are notified about clock rate changes (once before and once after
|
|
|
|
* the transition), or a list of drivers that are notified about
|
|
|
|
* changes in cpufreq policy.
|
|
|
|
*
|
|
|
|
* This function may sleep, and has the same return conditions as
|
[PATCH] Notifier chain update: API changes
The kernel's implementation of notifier chains is unsafe. There is no
protection against entries being added to or removed from a chain while the
chain is in use. The issues were discussed in this thread:
http://marc.theaimsgroup.com/?l=linux-kernel&m=113018709002036&w=2
We noticed that notifier chains in the kernel fall into two basic usage
classes:
"Blocking" chains are always called from a process context
and the callout routines are allowed to sleep;
"Atomic" chains can be called from an atomic context and
the callout routines are not allowed to sleep.
We decided to codify this distinction and make it part of the API. Therefore
this set of patches introduces three new, parallel APIs: one for blocking
notifiers, one for atomic notifiers, and one for "raw" notifiers (which is
really just the old API under a new name). New kinds of data structures are
used for the heads of the chains, and new routines are defined for
registration, unregistration, and calling a chain. The three APIs are
explained in include/linux/notifier.h and their implementation is in
kernel/sys.c.
With atomic and blocking chains, the implementation guarantees that the chain
links will not be corrupted and that chain callers will not get messed up by
entries being added or removed. For raw chains the implementation provides no
guarantees at all; users of this API must provide their own protections. (The
idea was that situations may come up where the assumptions of the atomic and
blocking APIs are not appropriate, so it should be possible for users to
handle these things in their own way.)
There are some limitations, which should not be too hard to live with. For
atomic/blocking chains, registration and unregistration must always be done in
a process context since the chain is protected by a mutex/rwsem. Also, a
callout routine for a non-raw chain must not try to register or unregister
entries on its own chain. (This did happen in a couple of places and the code
had to be changed to avoid it.)
Since atomic chains may be called from within an NMI handler, they cannot use
spinlocks for synchronization. Instead we use RCU. The overhead falls almost
entirely in the unregister routine, which is okay since unregistration is much
less frequent that calling a chain.
Here is the list of chains that we adjusted and their classifications. None
of them use the raw API, so for the moment it is only a placeholder.
ATOMIC CHAINS
-------------
arch/i386/kernel/traps.c: i386die_chain
arch/ia64/kernel/traps.c: ia64die_chain
arch/powerpc/kernel/traps.c: powerpc_die_chain
arch/sparc64/kernel/traps.c: sparc64die_chain
arch/x86_64/kernel/traps.c: die_chain
drivers/char/ipmi/ipmi_si_intf.c: xaction_notifier_list
kernel/panic.c: panic_notifier_list
kernel/profile.c: task_free_notifier
net/bluetooth/hci_core.c: hci_notifier
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_chain
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_expect_chain
net/ipv6/addrconf.c: inet6addr_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_expect_chain
net/netlink/af_netlink.c: netlink_chain
BLOCKING CHAINS
---------------
arch/powerpc/platforms/pseries/reconfig.c: pSeries_reconfig_chain
arch/s390/kernel/process.c: idle_chain
arch/x86_64/kernel/process.c idle_notifier
drivers/base/memory.c: memory_chain
drivers/cpufreq/cpufreq.c cpufreq_policy_notifier_list
drivers/cpufreq/cpufreq.c cpufreq_transition_notifier_list
drivers/macintosh/adb.c: adb_client_list
drivers/macintosh/via-pmu.c sleep_notifier_list
drivers/macintosh/via-pmu68k.c sleep_notifier_list
drivers/macintosh/windfarm_core.c wf_client_list
drivers/usb/core/notify.c usb_notifier_list
drivers/video/fbmem.c fb_notifier_list
kernel/cpu.c cpu_chain
kernel/module.c module_notify_list
kernel/profile.c munmap_notifier
kernel/profile.c task_exit_notifier
kernel/sys.c reboot_notifier_list
net/core/dev.c netdev_chain
net/decnet/dn_dev.c: dnaddr_chain
net/ipv4/devinet.c: inetaddr_chain
It's possible that some of these classifications are wrong. If they are,
please let us know or submit a patch to fix them. Note that any chain that
gets called very frequently should be atomic, because the rwsem read-locking
used for blocking chains is very likely to incur cache misses on SMP systems.
(However, if the chain's callout routines may sleep then the chain cannot be
atomic.)
The patch set was written by Alan Stern and Chandra Seetharaman, incorporating
material written by Keith Owens and suggestions from Paul McKenney and Andrew
Morton.
[jes@sgi.com: restructure the notifier chain initialization macros]
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: Jes Sorensen <jes@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-27 17:16:30 +08:00
|
|
|
* blocking_notifier_chain_register.
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
|
|
|
int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
|
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
|
2013-01-18 00:22:21 +08:00
|
|
|
if (cpufreq_disabled())
|
|
|
|
return -EINVAL;
|
|
|
|
|
2008-02-16 18:41:24 +08:00
|
|
|
WARN_ON(!init_cpufreq_transition_notifier_list_called);
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
switch (list) {
|
|
|
|
case CPUFREQ_TRANSITION_NOTIFIER:
|
2006-10-04 17:17:06 +08:00
|
|
|
ret = srcu_notifier_chain_register(
|
[PATCH] Notifier chain update: API changes
The kernel's implementation of notifier chains is unsafe. There is no
protection against entries being added to or removed from a chain while the
chain is in use. The issues were discussed in this thread:
http://marc.theaimsgroup.com/?l=linux-kernel&m=113018709002036&w=2
We noticed that notifier chains in the kernel fall into two basic usage
classes:
"Blocking" chains are always called from a process context
and the callout routines are allowed to sleep;
"Atomic" chains can be called from an atomic context and
the callout routines are not allowed to sleep.
We decided to codify this distinction and make it part of the API. Therefore
this set of patches introduces three new, parallel APIs: one for blocking
notifiers, one for atomic notifiers, and one for "raw" notifiers (which is
really just the old API under a new name). New kinds of data structures are
used for the heads of the chains, and new routines are defined for
registration, unregistration, and calling a chain. The three APIs are
explained in include/linux/notifier.h and their implementation is in
kernel/sys.c.
With atomic and blocking chains, the implementation guarantees that the chain
links will not be corrupted and that chain callers will not get messed up by
entries being added or removed. For raw chains the implementation provides no
guarantees at all; users of this API must provide their own protections. (The
idea was that situations may come up where the assumptions of the atomic and
blocking APIs are not appropriate, so it should be possible for users to
handle these things in their own way.)
There are some limitations, which should not be too hard to live with. For
atomic/blocking chains, registration and unregistration must always be done in
a process context since the chain is protected by a mutex/rwsem. Also, a
callout routine for a non-raw chain must not try to register or unregister
entries on its own chain. (This did happen in a couple of places and the code
had to be changed to avoid it.)
Since atomic chains may be called from within an NMI handler, they cannot use
spinlocks for synchronization. Instead we use RCU. The overhead falls almost
entirely in the unregister routine, which is okay since unregistration is much
less frequent that calling a chain.
Here is the list of chains that we adjusted and their classifications. None
of them use the raw API, so for the moment it is only a placeholder.
ATOMIC CHAINS
-------------
arch/i386/kernel/traps.c: i386die_chain
arch/ia64/kernel/traps.c: ia64die_chain
arch/powerpc/kernel/traps.c: powerpc_die_chain
arch/sparc64/kernel/traps.c: sparc64die_chain
arch/x86_64/kernel/traps.c: die_chain
drivers/char/ipmi/ipmi_si_intf.c: xaction_notifier_list
kernel/panic.c: panic_notifier_list
kernel/profile.c: task_free_notifier
net/bluetooth/hci_core.c: hci_notifier
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_chain
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_expect_chain
net/ipv6/addrconf.c: inet6addr_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_expect_chain
net/netlink/af_netlink.c: netlink_chain
BLOCKING CHAINS
---------------
arch/powerpc/platforms/pseries/reconfig.c: pSeries_reconfig_chain
arch/s390/kernel/process.c: idle_chain
arch/x86_64/kernel/process.c idle_notifier
drivers/base/memory.c: memory_chain
drivers/cpufreq/cpufreq.c cpufreq_policy_notifier_list
drivers/cpufreq/cpufreq.c cpufreq_transition_notifier_list
drivers/macintosh/adb.c: adb_client_list
drivers/macintosh/via-pmu.c sleep_notifier_list
drivers/macintosh/via-pmu68k.c sleep_notifier_list
drivers/macintosh/windfarm_core.c wf_client_list
drivers/usb/core/notify.c usb_notifier_list
drivers/video/fbmem.c fb_notifier_list
kernel/cpu.c cpu_chain
kernel/module.c module_notify_list
kernel/profile.c munmap_notifier
kernel/profile.c task_exit_notifier
kernel/sys.c reboot_notifier_list
net/core/dev.c netdev_chain
net/decnet/dn_dev.c: dnaddr_chain
net/ipv4/devinet.c: inetaddr_chain
It's possible that some of these classifications are wrong. If they are,
please let us know or submit a patch to fix them. Note that any chain that
gets called very frequently should be atomic, because the rwsem read-locking
used for blocking chains is very likely to incur cache misses on SMP systems.
(However, if the chain's callout routines may sleep then the chain cannot be
atomic.)
The patch set was written by Alan Stern and Chandra Seetharaman, incorporating
material written by Keith Owens and suggestions from Paul McKenney and Andrew
Morton.
[jes@sgi.com: restructure the notifier chain initialization macros]
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: Jes Sorensen <jes@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-27 17:16:30 +08:00
|
|
|
&cpufreq_transition_notifier_list, nb);
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
case CPUFREQ_POLICY_NOTIFIER:
|
[PATCH] Notifier chain update: API changes
The kernel's implementation of notifier chains is unsafe. There is no
protection against entries being added to or removed from a chain while the
chain is in use. The issues were discussed in this thread:
http://marc.theaimsgroup.com/?l=linux-kernel&m=113018709002036&w=2
We noticed that notifier chains in the kernel fall into two basic usage
classes:
"Blocking" chains are always called from a process context
and the callout routines are allowed to sleep;
"Atomic" chains can be called from an atomic context and
the callout routines are not allowed to sleep.
We decided to codify this distinction and make it part of the API. Therefore
this set of patches introduces three new, parallel APIs: one for blocking
notifiers, one for atomic notifiers, and one for "raw" notifiers (which is
really just the old API under a new name). New kinds of data structures are
used for the heads of the chains, and new routines are defined for
registration, unregistration, and calling a chain. The three APIs are
explained in include/linux/notifier.h and their implementation is in
kernel/sys.c.
With atomic and blocking chains, the implementation guarantees that the chain
links will not be corrupted and that chain callers will not get messed up by
entries being added or removed. For raw chains the implementation provides no
guarantees at all; users of this API must provide their own protections. (The
idea was that situations may come up where the assumptions of the atomic and
blocking APIs are not appropriate, so it should be possible for users to
handle these things in their own way.)
There are some limitations, which should not be too hard to live with. For
atomic/blocking chains, registration and unregistration must always be done in
a process context since the chain is protected by a mutex/rwsem. Also, a
callout routine for a non-raw chain must not try to register or unregister
entries on its own chain. (This did happen in a couple of places and the code
had to be changed to avoid it.)
Since atomic chains may be called from within an NMI handler, they cannot use
spinlocks for synchronization. Instead we use RCU. The overhead falls almost
entirely in the unregister routine, which is okay since unregistration is much
less frequent that calling a chain.
Here is the list of chains that we adjusted and their classifications. None
of them use the raw API, so for the moment it is only a placeholder.
ATOMIC CHAINS
-------------
arch/i386/kernel/traps.c: i386die_chain
arch/ia64/kernel/traps.c: ia64die_chain
arch/powerpc/kernel/traps.c: powerpc_die_chain
arch/sparc64/kernel/traps.c: sparc64die_chain
arch/x86_64/kernel/traps.c: die_chain
drivers/char/ipmi/ipmi_si_intf.c: xaction_notifier_list
kernel/panic.c: panic_notifier_list
kernel/profile.c: task_free_notifier
net/bluetooth/hci_core.c: hci_notifier
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_chain
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_expect_chain
net/ipv6/addrconf.c: inet6addr_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_expect_chain
net/netlink/af_netlink.c: netlink_chain
BLOCKING CHAINS
---------------
arch/powerpc/platforms/pseries/reconfig.c: pSeries_reconfig_chain
arch/s390/kernel/process.c: idle_chain
arch/x86_64/kernel/process.c idle_notifier
drivers/base/memory.c: memory_chain
drivers/cpufreq/cpufreq.c cpufreq_policy_notifier_list
drivers/cpufreq/cpufreq.c cpufreq_transition_notifier_list
drivers/macintosh/adb.c: adb_client_list
drivers/macintosh/via-pmu.c sleep_notifier_list
drivers/macintosh/via-pmu68k.c sleep_notifier_list
drivers/macintosh/windfarm_core.c wf_client_list
drivers/usb/core/notify.c usb_notifier_list
drivers/video/fbmem.c fb_notifier_list
kernel/cpu.c cpu_chain
kernel/module.c module_notify_list
kernel/profile.c munmap_notifier
kernel/profile.c task_exit_notifier
kernel/sys.c reboot_notifier_list
net/core/dev.c netdev_chain
net/decnet/dn_dev.c: dnaddr_chain
net/ipv4/devinet.c: inetaddr_chain
It's possible that some of these classifications are wrong. If they are,
please let us know or submit a patch to fix them. Note that any chain that
gets called very frequently should be atomic, because the rwsem read-locking
used for blocking chains is very likely to incur cache misses on SMP systems.
(However, if the chain's callout routines may sleep then the chain cannot be
atomic.)
The patch set was written by Alan Stern and Chandra Seetharaman, incorporating
material written by Keith Owens and suggestions from Paul McKenney and Andrew
Morton.
[jes@sgi.com: restructure the notifier chain initialization macros]
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: Jes Sorensen <jes@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-27 17:16:30 +08:00
|
|
|
ret = blocking_notifier_chain_register(
|
|
|
|
&cpufreq_policy_notifier_list, nb);
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
default:
|
|
|
|
ret = -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(cpufreq_register_notifier);
|
|
|
|
|
|
|
|
/**
|
|
|
|
* cpufreq_unregister_notifier - unregister a driver with cpufreq
|
|
|
|
* @nb: notifier block to be unregistered
|
2013-06-19 16:49:33 +08:00
|
|
|
* @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
|
2005-04-17 06:20:36 +08:00
|
|
|
*
|
|
|
|
* Remove a driver from the CPU frequency notifier list.
|
|
|
|
*
|
|
|
|
* This function may sleep, and has the same return conditions as
|
[PATCH] Notifier chain update: API changes
The kernel's implementation of notifier chains is unsafe. There is no
protection against entries being added to or removed from a chain while the
chain is in use. The issues were discussed in this thread:
http://marc.theaimsgroup.com/?l=linux-kernel&m=113018709002036&w=2
We noticed that notifier chains in the kernel fall into two basic usage
classes:
"Blocking" chains are always called from a process context
and the callout routines are allowed to sleep;
"Atomic" chains can be called from an atomic context and
the callout routines are not allowed to sleep.
We decided to codify this distinction and make it part of the API. Therefore
this set of patches introduces three new, parallel APIs: one for blocking
notifiers, one for atomic notifiers, and one for "raw" notifiers (which is
really just the old API under a new name). New kinds of data structures are
used for the heads of the chains, and new routines are defined for
registration, unregistration, and calling a chain. The three APIs are
explained in include/linux/notifier.h and their implementation is in
kernel/sys.c.
With atomic and blocking chains, the implementation guarantees that the chain
links will not be corrupted and that chain callers will not get messed up by
entries being added or removed. For raw chains the implementation provides no
guarantees at all; users of this API must provide their own protections. (The
idea was that situations may come up where the assumptions of the atomic and
blocking APIs are not appropriate, so it should be possible for users to
handle these things in their own way.)
There are some limitations, which should not be too hard to live with. For
atomic/blocking chains, registration and unregistration must always be done in
a process context since the chain is protected by a mutex/rwsem. Also, a
callout routine for a non-raw chain must not try to register or unregister
entries on its own chain. (This did happen in a couple of places and the code
had to be changed to avoid it.)
Since atomic chains may be called from within an NMI handler, they cannot use
spinlocks for synchronization. Instead we use RCU. The overhead falls almost
entirely in the unregister routine, which is okay since unregistration is much
less frequent that calling a chain.
Here is the list of chains that we adjusted and their classifications. None
of them use the raw API, so for the moment it is only a placeholder.
ATOMIC CHAINS
-------------
arch/i386/kernel/traps.c: i386die_chain
arch/ia64/kernel/traps.c: ia64die_chain
arch/powerpc/kernel/traps.c: powerpc_die_chain
arch/sparc64/kernel/traps.c: sparc64die_chain
arch/x86_64/kernel/traps.c: die_chain
drivers/char/ipmi/ipmi_si_intf.c: xaction_notifier_list
kernel/panic.c: panic_notifier_list
kernel/profile.c: task_free_notifier
net/bluetooth/hci_core.c: hci_notifier
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_chain
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_expect_chain
net/ipv6/addrconf.c: inet6addr_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_expect_chain
net/netlink/af_netlink.c: netlink_chain
BLOCKING CHAINS
---------------
arch/powerpc/platforms/pseries/reconfig.c: pSeries_reconfig_chain
arch/s390/kernel/process.c: idle_chain
arch/x86_64/kernel/process.c idle_notifier
drivers/base/memory.c: memory_chain
drivers/cpufreq/cpufreq.c cpufreq_policy_notifier_list
drivers/cpufreq/cpufreq.c cpufreq_transition_notifier_list
drivers/macintosh/adb.c: adb_client_list
drivers/macintosh/via-pmu.c sleep_notifier_list
drivers/macintosh/via-pmu68k.c sleep_notifier_list
drivers/macintosh/windfarm_core.c wf_client_list
drivers/usb/core/notify.c usb_notifier_list
drivers/video/fbmem.c fb_notifier_list
kernel/cpu.c cpu_chain
kernel/module.c module_notify_list
kernel/profile.c munmap_notifier
kernel/profile.c task_exit_notifier
kernel/sys.c reboot_notifier_list
net/core/dev.c netdev_chain
net/decnet/dn_dev.c: dnaddr_chain
net/ipv4/devinet.c: inetaddr_chain
It's possible that some of these classifications are wrong. If they are,
please let us know or submit a patch to fix them. Note that any chain that
gets called very frequently should be atomic, because the rwsem read-locking
used for blocking chains is very likely to incur cache misses on SMP systems.
(However, if the chain's callout routines may sleep then the chain cannot be
atomic.)
The patch set was written by Alan Stern and Chandra Seetharaman, incorporating
material written by Keith Owens and suggestions from Paul McKenney and Andrew
Morton.
[jes@sgi.com: restructure the notifier chain initialization macros]
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: Jes Sorensen <jes@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-27 17:16:30 +08:00
|
|
|
* blocking_notifier_chain_unregister.
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
|
|
|
int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
|
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
|
2013-01-18 00:22:21 +08:00
|
|
|
if (cpufreq_disabled())
|
|
|
|
return -EINVAL;
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
switch (list) {
|
|
|
|
case CPUFREQ_TRANSITION_NOTIFIER:
|
2006-10-04 17:17:06 +08:00
|
|
|
ret = srcu_notifier_chain_unregister(
|
[PATCH] Notifier chain update: API changes
The kernel's implementation of notifier chains is unsafe. There is no
protection against entries being added to or removed from a chain while the
chain is in use. The issues were discussed in this thread:
http://marc.theaimsgroup.com/?l=linux-kernel&m=113018709002036&w=2
We noticed that notifier chains in the kernel fall into two basic usage
classes:
"Blocking" chains are always called from a process context
and the callout routines are allowed to sleep;
"Atomic" chains can be called from an atomic context and
the callout routines are not allowed to sleep.
We decided to codify this distinction and make it part of the API. Therefore
this set of patches introduces three new, parallel APIs: one for blocking
notifiers, one for atomic notifiers, and one for "raw" notifiers (which is
really just the old API under a new name). New kinds of data structures are
used for the heads of the chains, and new routines are defined for
registration, unregistration, and calling a chain. The three APIs are
explained in include/linux/notifier.h and their implementation is in
kernel/sys.c.
With atomic and blocking chains, the implementation guarantees that the chain
links will not be corrupted and that chain callers will not get messed up by
entries being added or removed. For raw chains the implementation provides no
guarantees at all; users of this API must provide their own protections. (The
idea was that situations may come up where the assumptions of the atomic and
blocking APIs are not appropriate, so it should be possible for users to
handle these things in their own way.)
There are some limitations, which should not be too hard to live with. For
atomic/blocking chains, registration and unregistration must always be done in
a process context since the chain is protected by a mutex/rwsem. Also, a
callout routine for a non-raw chain must not try to register or unregister
entries on its own chain. (This did happen in a couple of places and the code
had to be changed to avoid it.)
Since atomic chains may be called from within an NMI handler, they cannot use
spinlocks for synchronization. Instead we use RCU. The overhead falls almost
entirely in the unregister routine, which is okay since unregistration is much
less frequent that calling a chain.
Here is the list of chains that we adjusted and their classifications. None
of them use the raw API, so for the moment it is only a placeholder.
ATOMIC CHAINS
-------------
arch/i386/kernel/traps.c: i386die_chain
arch/ia64/kernel/traps.c: ia64die_chain
arch/powerpc/kernel/traps.c: powerpc_die_chain
arch/sparc64/kernel/traps.c: sparc64die_chain
arch/x86_64/kernel/traps.c: die_chain
drivers/char/ipmi/ipmi_si_intf.c: xaction_notifier_list
kernel/panic.c: panic_notifier_list
kernel/profile.c: task_free_notifier
net/bluetooth/hci_core.c: hci_notifier
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_chain
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_expect_chain
net/ipv6/addrconf.c: inet6addr_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_expect_chain
net/netlink/af_netlink.c: netlink_chain
BLOCKING CHAINS
---------------
arch/powerpc/platforms/pseries/reconfig.c: pSeries_reconfig_chain
arch/s390/kernel/process.c: idle_chain
arch/x86_64/kernel/process.c idle_notifier
drivers/base/memory.c: memory_chain
drivers/cpufreq/cpufreq.c cpufreq_policy_notifier_list
drivers/cpufreq/cpufreq.c cpufreq_transition_notifier_list
drivers/macintosh/adb.c: adb_client_list
drivers/macintosh/via-pmu.c sleep_notifier_list
drivers/macintosh/via-pmu68k.c sleep_notifier_list
drivers/macintosh/windfarm_core.c wf_client_list
drivers/usb/core/notify.c usb_notifier_list
drivers/video/fbmem.c fb_notifier_list
kernel/cpu.c cpu_chain
kernel/module.c module_notify_list
kernel/profile.c munmap_notifier
kernel/profile.c task_exit_notifier
kernel/sys.c reboot_notifier_list
net/core/dev.c netdev_chain
net/decnet/dn_dev.c: dnaddr_chain
net/ipv4/devinet.c: inetaddr_chain
It's possible that some of these classifications are wrong. If they are,
please let us know or submit a patch to fix them. Note that any chain that
gets called very frequently should be atomic, because the rwsem read-locking
used for blocking chains is very likely to incur cache misses on SMP systems.
(However, if the chain's callout routines may sleep then the chain cannot be
atomic.)
The patch set was written by Alan Stern and Chandra Seetharaman, incorporating
material written by Keith Owens and suggestions from Paul McKenney and Andrew
Morton.
[jes@sgi.com: restructure the notifier chain initialization macros]
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: Jes Sorensen <jes@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-27 17:16:30 +08:00
|
|
|
&cpufreq_transition_notifier_list, nb);
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
case CPUFREQ_POLICY_NOTIFIER:
|
[PATCH] Notifier chain update: API changes
The kernel's implementation of notifier chains is unsafe. There is no
protection against entries being added to or removed from a chain while the
chain is in use. The issues were discussed in this thread:
http://marc.theaimsgroup.com/?l=linux-kernel&m=113018709002036&w=2
We noticed that notifier chains in the kernel fall into two basic usage
classes:
"Blocking" chains are always called from a process context
and the callout routines are allowed to sleep;
"Atomic" chains can be called from an atomic context and
the callout routines are not allowed to sleep.
We decided to codify this distinction and make it part of the API. Therefore
this set of patches introduces three new, parallel APIs: one for blocking
notifiers, one for atomic notifiers, and one for "raw" notifiers (which is
really just the old API under a new name). New kinds of data structures are
used for the heads of the chains, and new routines are defined for
registration, unregistration, and calling a chain. The three APIs are
explained in include/linux/notifier.h and their implementation is in
kernel/sys.c.
With atomic and blocking chains, the implementation guarantees that the chain
links will not be corrupted and that chain callers will not get messed up by
entries being added or removed. For raw chains the implementation provides no
guarantees at all; users of this API must provide their own protections. (The
idea was that situations may come up where the assumptions of the atomic and
blocking APIs are not appropriate, so it should be possible for users to
handle these things in their own way.)
There are some limitations, which should not be too hard to live with. For
atomic/blocking chains, registration and unregistration must always be done in
a process context since the chain is protected by a mutex/rwsem. Also, a
callout routine for a non-raw chain must not try to register or unregister
entries on its own chain. (This did happen in a couple of places and the code
had to be changed to avoid it.)
Since atomic chains may be called from within an NMI handler, they cannot use
spinlocks for synchronization. Instead we use RCU. The overhead falls almost
entirely in the unregister routine, which is okay since unregistration is much
less frequent that calling a chain.
Here is the list of chains that we adjusted and their classifications. None
of them use the raw API, so for the moment it is only a placeholder.
ATOMIC CHAINS
-------------
arch/i386/kernel/traps.c: i386die_chain
arch/ia64/kernel/traps.c: ia64die_chain
arch/powerpc/kernel/traps.c: powerpc_die_chain
arch/sparc64/kernel/traps.c: sparc64die_chain
arch/x86_64/kernel/traps.c: die_chain
drivers/char/ipmi/ipmi_si_intf.c: xaction_notifier_list
kernel/panic.c: panic_notifier_list
kernel/profile.c: task_free_notifier
net/bluetooth/hci_core.c: hci_notifier
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_chain
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_expect_chain
net/ipv6/addrconf.c: inet6addr_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_expect_chain
net/netlink/af_netlink.c: netlink_chain
BLOCKING CHAINS
---------------
arch/powerpc/platforms/pseries/reconfig.c: pSeries_reconfig_chain
arch/s390/kernel/process.c: idle_chain
arch/x86_64/kernel/process.c idle_notifier
drivers/base/memory.c: memory_chain
drivers/cpufreq/cpufreq.c cpufreq_policy_notifier_list
drivers/cpufreq/cpufreq.c cpufreq_transition_notifier_list
drivers/macintosh/adb.c: adb_client_list
drivers/macintosh/via-pmu.c sleep_notifier_list
drivers/macintosh/via-pmu68k.c sleep_notifier_list
drivers/macintosh/windfarm_core.c wf_client_list
drivers/usb/core/notify.c usb_notifier_list
drivers/video/fbmem.c fb_notifier_list
kernel/cpu.c cpu_chain
kernel/module.c module_notify_list
kernel/profile.c munmap_notifier
kernel/profile.c task_exit_notifier
kernel/sys.c reboot_notifier_list
net/core/dev.c netdev_chain
net/decnet/dn_dev.c: dnaddr_chain
net/ipv4/devinet.c: inetaddr_chain
It's possible that some of these classifications are wrong. If they are,
please let us know or submit a patch to fix them. Note that any chain that
gets called very frequently should be atomic, because the rwsem read-locking
used for blocking chains is very likely to incur cache misses on SMP systems.
(However, if the chain's callout routines may sleep then the chain cannot be
atomic.)
The patch set was written by Alan Stern and Chandra Seetharaman, incorporating
material written by Keith Owens and suggestions from Paul McKenney and Andrew
Morton.
[jes@sgi.com: restructure the notifier chain initialization macros]
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: Jes Sorensen <jes@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-27 17:16:30 +08:00
|
|
|
ret = blocking_notifier_chain_unregister(
|
|
|
|
&cpufreq_policy_notifier_list, nb);
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
default:
|
|
|
|
ret = -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(cpufreq_unregister_notifier);
|
|
|
|
|
|
|
|
|
|
|
|
/*********************************************************************
|
|
|
|
* GOVERNORS *
|
|
|
|
*********************************************************************/
|
|
|
|
|
2014-06-03 01:19:28 +08:00
|
|
|
/* Must set freqs->new to intermediate frequency */
|
|
|
|
static int __target_intermediate(struct cpufreq_policy *policy,
|
|
|
|
struct cpufreq_freqs *freqs, int index)
|
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
freqs->new = cpufreq_driver->get_intermediate(policy, index);
|
|
|
|
|
|
|
|
/* We don't need to switch to intermediate freq */
|
|
|
|
if (!freqs->new)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
|
|
|
|
__func__, policy->cpu, freqs->old, freqs->new);
|
|
|
|
|
|
|
|
cpufreq_freq_transition_begin(policy, freqs);
|
|
|
|
ret = cpufreq_driver->target_intermediate(policy, index);
|
|
|
|
cpufreq_freq_transition_end(policy, freqs, ret);
|
|
|
|
|
|
|
|
if (ret)
|
|
|
|
pr_err("%s: Failed to change to intermediate frequency: %d\n",
|
|
|
|
__func__, ret);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2014-05-21 16:59:29 +08:00
|
|
|
static int __target_index(struct cpufreq_policy *policy,
|
|
|
|
struct cpufreq_frequency_table *freq_table, int index)
|
|
|
|
{
|
2014-06-03 01:19:28 +08:00
|
|
|
struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
|
|
|
|
unsigned int intermediate_freq = 0;
|
2014-05-21 16:59:29 +08:00
|
|
|
int retval = -EINVAL;
|
|
|
|
bool notify;
|
|
|
|
|
|
|
|
notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
|
|
|
|
if (notify) {
|
2014-06-03 01:19:28 +08:00
|
|
|
/* Handle switching to intermediate frequency */
|
|
|
|
if (cpufreq_driver->get_intermediate) {
|
|
|
|
retval = __target_intermediate(policy, &freqs, index);
|
|
|
|
if (retval)
|
|
|
|
return retval;
|
|
|
|
|
|
|
|
intermediate_freq = freqs.new;
|
|
|
|
/* Set old freq to intermediate */
|
|
|
|
if (intermediate_freq)
|
|
|
|
freqs.old = freqs.new;
|
|
|
|
}
|
2014-05-21 16:59:29 +08:00
|
|
|
|
2014-06-03 01:19:28 +08:00
|
|
|
freqs.new = freq_table[index].frequency;
|
2014-05-21 16:59:29 +08:00
|
|
|
pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
|
|
|
|
__func__, policy->cpu, freqs.old, freqs.new);
|
|
|
|
|
|
|
|
cpufreq_freq_transition_begin(policy, &freqs);
|
|
|
|
}
|
|
|
|
|
|
|
|
retval = cpufreq_driver->target_index(policy, index);
|
|
|
|
if (retval)
|
|
|
|
pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
|
|
|
|
retval);
|
|
|
|
|
2014-06-03 01:19:28 +08:00
|
|
|
if (notify) {
|
2014-05-21 16:59:29 +08:00
|
|
|
cpufreq_freq_transition_end(policy, &freqs, retval);
|
|
|
|
|
2014-06-03 01:19:28 +08:00
|
|
|
/*
|
|
|
|
* Failed after setting to intermediate freq? Driver should have
|
|
|
|
* reverted back to initial frequency and so should we. Check
|
|
|
|
* here for intermediate_freq instead of get_intermediate, in
|
|
|
|
* case we have't switched to intermediate freq at all.
|
|
|
|
*/
|
|
|
|
if (unlikely(retval && intermediate_freq)) {
|
|
|
|
freqs.old = intermediate_freq;
|
|
|
|
freqs.new = policy->restore_freq;
|
|
|
|
cpufreq_freq_transition_begin(policy, &freqs);
|
|
|
|
cpufreq_freq_transition_end(policy, &freqs, 0);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2014-05-21 16:59:29 +08:00
|
|
|
return retval;
|
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
int __cpufreq_driver_target(struct cpufreq_policy *policy,
|
|
|
|
unsigned int target_freq,
|
|
|
|
unsigned int relation)
|
|
|
|
{
|
2012-10-31 08:28:21 +08:00
|
|
|
unsigned int old_target_freq = target_freq;
|
2014-05-21 16:59:29 +08:00
|
|
|
int retval = -EINVAL;
|
2005-10-31 06:59:54 +08:00
|
|
|
|
2012-03-14 07:18:39 +08:00
|
|
|
if (cpufreq_disabled())
|
|
|
|
return -ENODEV;
|
|
|
|
|
2012-10-31 08:28:21 +08:00
|
|
|
/* Make sure that target_freq is within supported range */
|
|
|
|
if (target_freq > policy->max)
|
|
|
|
target_freq = policy->max;
|
|
|
|
if (target_freq < policy->min)
|
|
|
|
target_freq = policy->min;
|
|
|
|
|
|
|
|
pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
|
2014-03-12 01:03:00 +08:00
|
|
|
policy->cpu, target_freq, relation, old_target_freq);
|
2012-10-31 08:28:15 +08:00
|
|
|
|
2013-10-25 22:15:48 +08:00
|
|
|
/*
|
|
|
|
* This might look like a redundant call as we are checking it again
|
|
|
|
* after finding index. But it is left intentionally for cases where
|
|
|
|
* exactly same freq is called again and so we can save on few function
|
|
|
|
* calls.
|
|
|
|
*/
|
2012-10-31 08:28:15 +08:00
|
|
|
if (target_freq == policy->cur)
|
|
|
|
return 0;
|
|
|
|
|
2014-06-03 01:19:28 +08:00
|
|
|
/* Save last value to restore later on errors */
|
|
|
|
policy->restore_freq = policy->cur;
|
|
|
|
|
2013-04-29 06:08:16 +08:00
|
|
|
if (cpufreq_driver->target)
|
|
|
|
retval = cpufreq_driver->target(policy, target_freq, relation);
|
2013-10-25 22:15:48 +08:00
|
|
|
else if (cpufreq_driver->target_index) {
|
|
|
|
struct cpufreq_frequency_table *freq_table;
|
|
|
|
int index;
|
2005-11-09 13:34:24 +08:00
|
|
|
|
2013-10-25 22:15:48 +08:00
|
|
|
freq_table = cpufreq_frequency_get_table(policy->cpu);
|
|
|
|
if (unlikely(!freq_table)) {
|
|
|
|
pr_err("%s: Unable to find freq_table\n", __func__);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
retval = cpufreq_frequency_table_target(policy, freq_table,
|
|
|
|
target_freq, relation, &index);
|
|
|
|
if (unlikely(retval)) {
|
|
|
|
pr_err("%s: Unable to find matching freq\n", __func__);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2013-08-14 22:08:24 +08:00
|
|
|
if (freq_table[index].frequency == policy->cur) {
|
2013-10-25 22:15:48 +08:00
|
|
|
retval = 0;
|
2013-08-14 22:08:24 +08:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2014-05-21 16:59:29 +08:00
|
|
|
retval = __target_index(policy, freq_table, index);
|
2013-10-25 22:15:48 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
out:
|
2005-04-17 06:20:36 +08:00
|
|
|
return retval;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
|
|
|
|
|
|
|
|
int cpufreq_driver_target(struct cpufreq_policy *policy,
|
|
|
|
unsigned int target_freq,
|
|
|
|
unsigned int relation)
|
|
|
|
{
|
2008-07-26 04:44:53 +08:00
|
|
|
int ret = -EINVAL;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-10-18 21:40:15 +08:00
|
|
|
down_write(&policy->rwsem);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
ret = __cpufreq_driver_target(policy, target_freq, relation);
|
|
|
|
|
2013-10-18 21:40:15 +08:00
|
|
|
up_write(&policy->rwsem);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_driver_target);
|
|
|
|
|
2006-10-26 18:50:58 +08:00
|
|
|
static int __cpufreq_governor(struct cpufreq_policy *policy,
|
|
|
|
unsigned int event)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2005-07-29 00:43:56 +08:00
|
|
|
int ret;
|
2007-10-03 04:28:13 +08:00
|
|
|
|
|
|
|
/* Only must be defined when default governor is known to have latency
|
|
|
|
restrictions, like e.g. conservative or ondemand.
|
|
|
|
That this is the case is already ensured in Kconfig
|
|
|
|
*/
|
|
|
|
#ifdef CONFIG_CPU_FREQ_GOV_PERFORMANCE
|
|
|
|
struct cpufreq_governor *gov = &cpufreq_gov_performance;
|
|
|
|
#else
|
|
|
|
struct cpufreq_governor *gov = NULL;
|
|
|
|
#endif
|
2007-10-03 04:28:12 +08:00
|
|
|
|
2014-03-04 11:00:26 +08:00
|
|
|
/* Don't start any governor operations if we are entering suspend */
|
|
|
|
if (cpufreq_suspended)
|
|
|
|
return 0;
|
2014-12-18 14:28:19 +08:00
|
|
|
/*
|
|
|
|
* Governor might not be initiated here if ACPI _PPC changed
|
|
|
|
* notification happened, so check it.
|
|
|
|
*/
|
|
|
|
if (!policy->governor)
|
|
|
|
return -EINVAL;
|
2014-03-04 11:00:26 +08:00
|
|
|
|
2007-10-03 04:28:12 +08:00
|
|
|
if (policy->governor->max_transition_latency &&
|
|
|
|
policy->cpuinfo.transition_latency >
|
|
|
|
policy->governor->max_transition_latency) {
|
2007-10-03 04:28:13 +08:00
|
|
|
if (!gov)
|
|
|
|
return -EINVAL;
|
|
|
|
else {
|
2014-03-12 01:03:00 +08:00
|
|
|
pr_warn("%s governor failed, too long transition latency of HW, fallback to %s governor\n",
|
|
|
|
policy->governor->name, gov->name);
|
2007-10-03 04:28:13 +08:00
|
|
|
policy->governor = gov;
|
|
|
|
}
|
2007-10-03 04:28:12 +08:00
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-08-07 01:23:10 +08:00
|
|
|
if (event == CPUFREQ_GOV_POLICY_INIT)
|
|
|
|
if (!try_module_get(policy->governor->owner))
|
|
|
|
return -EINVAL;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2011-03-27 21:04:46 +08:00
|
|
|
pr_debug("__cpufreq_governor for CPU %u, event %u\n",
|
2014-03-12 01:03:00 +08:00
|
|
|
policy->cpu, event);
|
cpufreq: Fix governor start/stop race condition
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>
2013-06-19 15:00:07 +08:00
|
|
|
|
|
|
|
mutex_lock(&cpufreq_governor_lock);
|
2013-09-07 03:53:55 +08:00
|
|
|
if ((policy->governor_enabled && event == CPUFREQ_GOV_START)
|
2013-08-31 20:23:40 +08:00
|
|
|
|| (!policy->governor_enabled
|
|
|
|
&& (event == CPUFREQ_GOV_LIMITS || event == CPUFREQ_GOV_STOP))) {
|
cpufreq: Fix governor start/stop race condition
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>
2013-06-19 15:00:07 +08:00
|
|
|
mutex_unlock(&cpufreq_governor_lock);
|
|
|
|
return -EBUSY;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (event == CPUFREQ_GOV_STOP)
|
|
|
|
policy->governor_enabled = false;
|
|
|
|
else if (event == CPUFREQ_GOV_START)
|
|
|
|
policy->governor_enabled = true;
|
|
|
|
|
|
|
|
mutex_unlock(&cpufreq_governor_lock);
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
ret = policy->governor->governor(policy, event);
|
|
|
|
|
2013-03-27 23:58:58 +08:00
|
|
|
if (!ret) {
|
|
|
|
if (event == CPUFREQ_GOV_POLICY_INIT)
|
|
|
|
policy->governor->initialized++;
|
|
|
|
else if (event == CPUFREQ_GOV_POLICY_EXIT)
|
|
|
|
policy->governor->initialized--;
|
cpufreq: Fix governor start/stop race condition
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>
2013-06-19 15:00:07 +08:00
|
|
|
} else {
|
|
|
|
/* Restore original values */
|
|
|
|
mutex_lock(&cpufreq_governor_lock);
|
|
|
|
if (event == CPUFREQ_GOV_STOP)
|
|
|
|
policy->governor_enabled = true;
|
|
|
|
else if (event == CPUFREQ_GOV_START)
|
|
|
|
policy->governor_enabled = false;
|
|
|
|
mutex_unlock(&cpufreq_governor_lock);
|
2013-03-27 23:58:58 +08:00
|
|
|
}
|
2013-02-01 13:42:58 +08:00
|
|
|
|
2013-08-07 01:23:10 +08:00
|
|
|
if (((event == CPUFREQ_GOV_POLICY_INIT) && ret) ||
|
|
|
|
((event == CPUFREQ_GOV_POLICY_EXIT) && !ret))
|
2005-04-17 06:20:36 +08:00
|
|
|
module_put(policy->governor->owner);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
int cpufreq_register_governor(struct cpufreq_governor *governor)
|
|
|
|
{
|
2006-07-07 03:30:26 +08:00
|
|
|
int err;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
if (!governor)
|
|
|
|
return -EINVAL;
|
|
|
|
|
2012-03-14 07:18:39 +08:00
|
|
|
if (cpufreq_disabled())
|
|
|
|
return -ENODEV;
|
|
|
|
|
2006-01-14 07:54:22 +08:00
|
|
|
mutex_lock(&cpufreq_governor_mutex);
|
2006-02-28 13:43:23 +08:00
|
|
|
|
2013-02-01 13:42:58 +08:00
|
|
|
governor->initialized = 0;
|
2006-07-07 03:30:26 +08:00
|
|
|
err = -EBUSY;
|
2015-01-02 15:04:26 +08:00
|
|
|
if (!find_governor(governor->name)) {
|
2006-07-07 03:30:26 +08:00
|
|
|
err = 0;
|
|
|
|
list_add(&governor->governor_list, &cpufreq_governor_list);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2006-02-28 13:43:23 +08:00
|
|
|
mutex_unlock(&cpufreq_governor_mutex);
|
2006-07-07 03:30:26 +08:00
|
|
|
return err;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_register_governor);
|
|
|
|
|
|
|
|
void cpufreq_unregister_governor(struct cpufreq_governor *governor)
|
|
|
|
{
|
2009-11-12 22:18:46 +08:00
|
|
|
int cpu;
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
if (!governor)
|
|
|
|
return;
|
|
|
|
|
2012-03-14 07:18:39 +08:00
|
|
|
if (cpufreq_disabled())
|
|
|
|
return;
|
|
|
|
|
2009-11-12 22:18:46 +08:00
|
|
|
for_each_present_cpu(cpu) {
|
|
|
|
if (cpu_online(cpu))
|
|
|
|
continue;
|
|
|
|
if (!strcmp(per_cpu(cpufreq_cpu_governor, cpu), governor->name))
|
|
|
|
strcpy(per_cpu(cpufreq_cpu_governor, cpu), "\0");
|
|
|
|
}
|
|
|
|
|
2006-01-14 07:54:22 +08:00
|
|
|
mutex_lock(&cpufreq_governor_mutex);
|
2005-04-17 06:20:36 +08:00
|
|
|
list_del(&governor->governor_list);
|
2006-01-14 07:54:22 +08:00
|
|
|
mutex_unlock(&cpufreq_governor_mutex);
|
2005-04-17 06:20:36 +08:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
|
|
|
|
|
|
|
|
|
|
|
|
/*********************************************************************
|
|
|
|
* POLICY INTERFACE *
|
|
|
|
*********************************************************************/
|
|
|
|
|
|
|
|
/**
|
|
|
|
* cpufreq_get_policy - get the current cpufreq_policy
|
2009-01-18 14:37:11 +08:00
|
|
|
* @policy: struct cpufreq_policy into which the current cpufreq_policy
|
|
|
|
* is written
|
2005-04-17 06:20:36 +08:00
|
|
|
*
|
|
|
|
* Reads the current cpufreq policy.
|
|
|
|
*/
|
|
|
|
int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
|
|
|
|
{
|
|
|
|
struct cpufreq_policy *cpu_policy;
|
|
|
|
if (!policy)
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
cpu_policy = cpufreq_cpu_get(cpu);
|
|
|
|
if (!cpu_policy)
|
|
|
|
return -EINVAL;
|
|
|
|
|
2013-08-07 01:23:06 +08:00
|
|
|
memcpy(policy, cpu_policy, sizeof(*policy));
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
cpufreq_cpu_put(cpu_policy);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(cpufreq_get_policy);
|
|
|
|
|
2006-07-26 21:40:07 +08:00
|
|
|
/*
|
2013-10-02 16:43:16 +08:00
|
|
|
* policy : current policy.
|
|
|
|
* new_policy: policy to be set.
|
2006-07-26 21:40:07 +08:00
|
|
|
*/
|
2013-10-02 16:43:16 +08:00
|
|
|
static int cpufreq_set_policy(struct cpufreq_policy *policy,
|
2013-08-07 01:23:05 +08:00
|
|
|
struct cpufreq_policy *new_policy)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2014-02-18 05:56:35 +08:00
|
|
|
struct cpufreq_governor *old_gov;
|
|
|
|
int ret;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-03-12 01:03:00 +08:00
|
|
|
pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
|
|
|
|
new_policy->cpu, new_policy->min, new_policy->max);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-08-07 01:23:06 +08:00
|
|
|
memcpy(&new_policy->cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-02-18 05:56:35 +08:00
|
|
|
if (new_policy->min > policy->max || new_policy->max < policy->min)
|
|
|
|
return -EINVAL;
|
2006-07-06 05:12:20 +08:00
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/* verify the cpu speed can be set within this limit */
|
2013-08-07 01:23:05 +08:00
|
|
|
ret = cpufreq_driver->verify(new_policy);
|
2005-04-17 06:20:36 +08:00
|
|
|
if (ret)
|
2014-02-18 05:56:35 +08:00
|
|
|
return ret;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/* adjust if necessary - all reasons */
|
[PATCH] Notifier chain update: API changes
The kernel's implementation of notifier chains is unsafe. There is no
protection against entries being added to or removed from a chain while the
chain is in use. The issues were discussed in this thread:
http://marc.theaimsgroup.com/?l=linux-kernel&m=113018709002036&w=2
We noticed that notifier chains in the kernel fall into two basic usage
classes:
"Blocking" chains are always called from a process context
and the callout routines are allowed to sleep;
"Atomic" chains can be called from an atomic context and
the callout routines are not allowed to sleep.
We decided to codify this distinction and make it part of the API. Therefore
this set of patches introduces three new, parallel APIs: one for blocking
notifiers, one for atomic notifiers, and one for "raw" notifiers (which is
really just the old API under a new name). New kinds of data structures are
used for the heads of the chains, and new routines are defined for
registration, unregistration, and calling a chain. The three APIs are
explained in include/linux/notifier.h and their implementation is in
kernel/sys.c.
With atomic and blocking chains, the implementation guarantees that the chain
links will not be corrupted and that chain callers will not get messed up by
entries being added or removed. For raw chains the implementation provides no
guarantees at all; users of this API must provide their own protections. (The
idea was that situations may come up where the assumptions of the atomic and
blocking APIs are not appropriate, so it should be possible for users to
handle these things in their own way.)
There are some limitations, which should not be too hard to live with. For
atomic/blocking chains, registration and unregistration must always be done in
a process context since the chain is protected by a mutex/rwsem. Also, a
callout routine for a non-raw chain must not try to register or unregister
entries on its own chain. (This did happen in a couple of places and the code
had to be changed to avoid it.)
Since atomic chains may be called from within an NMI handler, they cannot use
spinlocks for synchronization. Instead we use RCU. The overhead falls almost
entirely in the unregister routine, which is okay since unregistration is much
less frequent that calling a chain.
Here is the list of chains that we adjusted and their classifications. None
of them use the raw API, so for the moment it is only a placeholder.
ATOMIC CHAINS
-------------
arch/i386/kernel/traps.c: i386die_chain
arch/ia64/kernel/traps.c: ia64die_chain
arch/powerpc/kernel/traps.c: powerpc_die_chain
arch/sparc64/kernel/traps.c: sparc64die_chain
arch/x86_64/kernel/traps.c: die_chain
drivers/char/ipmi/ipmi_si_intf.c: xaction_notifier_list
kernel/panic.c: panic_notifier_list
kernel/profile.c: task_free_notifier
net/bluetooth/hci_core.c: hci_notifier
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_chain
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_expect_chain
net/ipv6/addrconf.c: inet6addr_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_expect_chain
net/netlink/af_netlink.c: netlink_chain
BLOCKING CHAINS
---------------
arch/powerpc/platforms/pseries/reconfig.c: pSeries_reconfig_chain
arch/s390/kernel/process.c: idle_chain
arch/x86_64/kernel/process.c idle_notifier
drivers/base/memory.c: memory_chain
drivers/cpufreq/cpufreq.c cpufreq_policy_notifier_list
drivers/cpufreq/cpufreq.c cpufreq_transition_notifier_list
drivers/macintosh/adb.c: adb_client_list
drivers/macintosh/via-pmu.c sleep_notifier_list
drivers/macintosh/via-pmu68k.c sleep_notifier_list
drivers/macintosh/windfarm_core.c wf_client_list
drivers/usb/core/notify.c usb_notifier_list
drivers/video/fbmem.c fb_notifier_list
kernel/cpu.c cpu_chain
kernel/module.c module_notify_list
kernel/profile.c munmap_notifier
kernel/profile.c task_exit_notifier
kernel/sys.c reboot_notifier_list
net/core/dev.c netdev_chain
net/decnet/dn_dev.c: dnaddr_chain
net/ipv4/devinet.c: inetaddr_chain
It's possible that some of these classifications are wrong. If they are,
please let us know or submit a patch to fix them. Note that any chain that
gets called very frequently should be atomic, because the rwsem read-locking
used for blocking chains is very likely to incur cache misses on SMP systems.
(However, if the chain's callout routines may sleep then the chain cannot be
atomic.)
The patch set was written by Alan Stern and Chandra Seetharaman, incorporating
material written by Keith Owens and suggestions from Paul McKenney and Andrew
Morton.
[jes@sgi.com: restructure the notifier chain initialization macros]
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: Jes Sorensen <jes@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-27 17:16:30 +08:00
|
|
|
blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
|
2013-08-07 01:23:05 +08:00
|
|
|
CPUFREQ_ADJUST, new_policy);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/* adjust if necessary - hardware incompatibility*/
|
[PATCH] Notifier chain update: API changes
The kernel's implementation of notifier chains is unsafe. There is no
protection against entries being added to or removed from a chain while the
chain is in use. The issues were discussed in this thread:
http://marc.theaimsgroup.com/?l=linux-kernel&m=113018709002036&w=2
We noticed that notifier chains in the kernel fall into two basic usage
classes:
"Blocking" chains are always called from a process context
and the callout routines are allowed to sleep;
"Atomic" chains can be called from an atomic context and
the callout routines are not allowed to sleep.
We decided to codify this distinction and make it part of the API. Therefore
this set of patches introduces three new, parallel APIs: one for blocking
notifiers, one for atomic notifiers, and one for "raw" notifiers (which is
really just the old API under a new name). New kinds of data structures are
used for the heads of the chains, and new routines are defined for
registration, unregistration, and calling a chain. The three APIs are
explained in include/linux/notifier.h and their implementation is in
kernel/sys.c.
With atomic and blocking chains, the implementation guarantees that the chain
links will not be corrupted and that chain callers will not get messed up by
entries being added or removed. For raw chains the implementation provides no
guarantees at all; users of this API must provide their own protections. (The
idea was that situations may come up where the assumptions of the atomic and
blocking APIs are not appropriate, so it should be possible for users to
handle these things in their own way.)
There are some limitations, which should not be too hard to live with. For
atomic/blocking chains, registration and unregistration must always be done in
a process context since the chain is protected by a mutex/rwsem. Also, a
callout routine for a non-raw chain must not try to register or unregister
entries on its own chain. (This did happen in a couple of places and the code
had to be changed to avoid it.)
Since atomic chains may be called from within an NMI handler, they cannot use
spinlocks for synchronization. Instead we use RCU. The overhead falls almost
entirely in the unregister routine, which is okay since unregistration is much
less frequent that calling a chain.
Here is the list of chains that we adjusted and their classifications. None
of them use the raw API, so for the moment it is only a placeholder.
ATOMIC CHAINS
-------------
arch/i386/kernel/traps.c: i386die_chain
arch/ia64/kernel/traps.c: ia64die_chain
arch/powerpc/kernel/traps.c: powerpc_die_chain
arch/sparc64/kernel/traps.c: sparc64die_chain
arch/x86_64/kernel/traps.c: die_chain
drivers/char/ipmi/ipmi_si_intf.c: xaction_notifier_list
kernel/panic.c: panic_notifier_list
kernel/profile.c: task_free_notifier
net/bluetooth/hci_core.c: hci_notifier
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_chain
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_expect_chain
net/ipv6/addrconf.c: inet6addr_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_expect_chain
net/netlink/af_netlink.c: netlink_chain
BLOCKING CHAINS
---------------
arch/powerpc/platforms/pseries/reconfig.c: pSeries_reconfig_chain
arch/s390/kernel/process.c: idle_chain
arch/x86_64/kernel/process.c idle_notifier
drivers/base/memory.c: memory_chain
drivers/cpufreq/cpufreq.c cpufreq_policy_notifier_list
drivers/cpufreq/cpufreq.c cpufreq_transition_notifier_list
drivers/macintosh/adb.c: adb_client_list
drivers/macintosh/via-pmu.c sleep_notifier_list
drivers/macintosh/via-pmu68k.c sleep_notifier_list
drivers/macintosh/windfarm_core.c wf_client_list
drivers/usb/core/notify.c usb_notifier_list
drivers/video/fbmem.c fb_notifier_list
kernel/cpu.c cpu_chain
kernel/module.c module_notify_list
kernel/profile.c munmap_notifier
kernel/profile.c task_exit_notifier
kernel/sys.c reboot_notifier_list
net/core/dev.c netdev_chain
net/decnet/dn_dev.c: dnaddr_chain
net/ipv4/devinet.c: inetaddr_chain
It's possible that some of these classifications are wrong. If they are,
please let us know or submit a patch to fix them. Note that any chain that
gets called very frequently should be atomic, because the rwsem read-locking
used for blocking chains is very likely to incur cache misses on SMP systems.
(However, if the chain's callout routines may sleep then the chain cannot be
atomic.)
The patch set was written by Alan Stern and Chandra Seetharaman, incorporating
material written by Keith Owens and suggestions from Paul McKenney and Andrew
Morton.
[jes@sgi.com: restructure the notifier chain initialization macros]
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: Jes Sorensen <jes@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-27 17:16:30 +08:00
|
|
|
blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
|
2013-08-07 01:23:05 +08:00
|
|
|
CPUFREQ_INCOMPATIBLE, new_policy);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-06-19 16:49:33 +08:00
|
|
|
/*
|
|
|
|
* verify the cpu speed can be set within this limit, which might be
|
|
|
|
* different to the first one
|
|
|
|
*/
|
2013-08-07 01:23:05 +08:00
|
|
|
ret = cpufreq_driver->verify(new_policy);
|
[PATCH] Notifier chain update: API changes
The kernel's implementation of notifier chains is unsafe. There is no
protection against entries being added to or removed from a chain while the
chain is in use. The issues were discussed in this thread:
http://marc.theaimsgroup.com/?l=linux-kernel&m=113018709002036&w=2
We noticed that notifier chains in the kernel fall into two basic usage
classes:
"Blocking" chains are always called from a process context
and the callout routines are allowed to sleep;
"Atomic" chains can be called from an atomic context and
the callout routines are not allowed to sleep.
We decided to codify this distinction and make it part of the API. Therefore
this set of patches introduces three new, parallel APIs: one for blocking
notifiers, one for atomic notifiers, and one for "raw" notifiers (which is
really just the old API under a new name). New kinds of data structures are
used for the heads of the chains, and new routines are defined for
registration, unregistration, and calling a chain. The three APIs are
explained in include/linux/notifier.h and their implementation is in
kernel/sys.c.
With atomic and blocking chains, the implementation guarantees that the chain
links will not be corrupted and that chain callers will not get messed up by
entries being added or removed. For raw chains the implementation provides no
guarantees at all; users of this API must provide their own protections. (The
idea was that situations may come up where the assumptions of the atomic and
blocking APIs are not appropriate, so it should be possible for users to
handle these things in their own way.)
There are some limitations, which should not be too hard to live with. For
atomic/blocking chains, registration and unregistration must always be done in
a process context since the chain is protected by a mutex/rwsem. Also, a
callout routine for a non-raw chain must not try to register or unregister
entries on its own chain. (This did happen in a couple of places and the code
had to be changed to avoid it.)
Since atomic chains may be called from within an NMI handler, they cannot use
spinlocks for synchronization. Instead we use RCU. The overhead falls almost
entirely in the unregister routine, which is okay since unregistration is much
less frequent that calling a chain.
Here is the list of chains that we adjusted and their classifications. None
of them use the raw API, so for the moment it is only a placeholder.
ATOMIC CHAINS
-------------
arch/i386/kernel/traps.c: i386die_chain
arch/ia64/kernel/traps.c: ia64die_chain
arch/powerpc/kernel/traps.c: powerpc_die_chain
arch/sparc64/kernel/traps.c: sparc64die_chain
arch/x86_64/kernel/traps.c: die_chain
drivers/char/ipmi/ipmi_si_intf.c: xaction_notifier_list
kernel/panic.c: panic_notifier_list
kernel/profile.c: task_free_notifier
net/bluetooth/hci_core.c: hci_notifier
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_chain
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_expect_chain
net/ipv6/addrconf.c: inet6addr_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_expect_chain
net/netlink/af_netlink.c: netlink_chain
BLOCKING CHAINS
---------------
arch/powerpc/platforms/pseries/reconfig.c: pSeries_reconfig_chain
arch/s390/kernel/process.c: idle_chain
arch/x86_64/kernel/process.c idle_notifier
drivers/base/memory.c: memory_chain
drivers/cpufreq/cpufreq.c cpufreq_policy_notifier_list
drivers/cpufreq/cpufreq.c cpufreq_transition_notifier_list
drivers/macintosh/adb.c: adb_client_list
drivers/macintosh/via-pmu.c sleep_notifier_list
drivers/macintosh/via-pmu68k.c sleep_notifier_list
drivers/macintosh/windfarm_core.c wf_client_list
drivers/usb/core/notify.c usb_notifier_list
drivers/video/fbmem.c fb_notifier_list
kernel/cpu.c cpu_chain
kernel/module.c module_notify_list
kernel/profile.c munmap_notifier
kernel/profile.c task_exit_notifier
kernel/sys.c reboot_notifier_list
net/core/dev.c netdev_chain
net/decnet/dn_dev.c: dnaddr_chain
net/ipv4/devinet.c: inetaddr_chain
It's possible that some of these classifications are wrong. If they are,
please let us know or submit a patch to fix them. Note that any chain that
gets called very frequently should be atomic, because the rwsem read-locking
used for blocking chains is very likely to incur cache misses on SMP systems.
(However, if the chain's callout routines may sleep then the chain cannot be
atomic.)
The patch set was written by Alan Stern and Chandra Seetharaman, incorporating
material written by Keith Owens and suggestions from Paul McKenney and Andrew
Morton.
[jes@sgi.com: restructure the notifier chain initialization macros]
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: Jes Sorensen <jes@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-27 17:16:30 +08:00
|
|
|
if (ret)
|
2014-02-18 05:56:35 +08:00
|
|
|
return ret;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/* notification of the new policy */
|
[PATCH] Notifier chain update: API changes
The kernel's implementation of notifier chains is unsafe. There is no
protection against entries being added to or removed from a chain while the
chain is in use. The issues were discussed in this thread:
http://marc.theaimsgroup.com/?l=linux-kernel&m=113018709002036&w=2
We noticed that notifier chains in the kernel fall into two basic usage
classes:
"Blocking" chains are always called from a process context
and the callout routines are allowed to sleep;
"Atomic" chains can be called from an atomic context and
the callout routines are not allowed to sleep.
We decided to codify this distinction and make it part of the API. Therefore
this set of patches introduces three new, parallel APIs: one for blocking
notifiers, one for atomic notifiers, and one for "raw" notifiers (which is
really just the old API under a new name). New kinds of data structures are
used for the heads of the chains, and new routines are defined for
registration, unregistration, and calling a chain. The three APIs are
explained in include/linux/notifier.h and their implementation is in
kernel/sys.c.
With atomic and blocking chains, the implementation guarantees that the chain
links will not be corrupted and that chain callers will not get messed up by
entries being added or removed. For raw chains the implementation provides no
guarantees at all; users of this API must provide their own protections. (The
idea was that situations may come up where the assumptions of the atomic and
blocking APIs are not appropriate, so it should be possible for users to
handle these things in their own way.)
There are some limitations, which should not be too hard to live with. For
atomic/blocking chains, registration and unregistration must always be done in
a process context since the chain is protected by a mutex/rwsem. Also, a
callout routine for a non-raw chain must not try to register or unregister
entries on its own chain. (This did happen in a couple of places and the code
had to be changed to avoid it.)
Since atomic chains may be called from within an NMI handler, they cannot use
spinlocks for synchronization. Instead we use RCU. The overhead falls almost
entirely in the unregister routine, which is okay since unregistration is much
less frequent that calling a chain.
Here is the list of chains that we adjusted and their classifications. None
of them use the raw API, so for the moment it is only a placeholder.
ATOMIC CHAINS
-------------
arch/i386/kernel/traps.c: i386die_chain
arch/ia64/kernel/traps.c: ia64die_chain
arch/powerpc/kernel/traps.c: powerpc_die_chain
arch/sparc64/kernel/traps.c: sparc64die_chain
arch/x86_64/kernel/traps.c: die_chain
drivers/char/ipmi/ipmi_si_intf.c: xaction_notifier_list
kernel/panic.c: panic_notifier_list
kernel/profile.c: task_free_notifier
net/bluetooth/hci_core.c: hci_notifier
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_chain
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_expect_chain
net/ipv6/addrconf.c: inet6addr_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_expect_chain
net/netlink/af_netlink.c: netlink_chain
BLOCKING CHAINS
---------------
arch/powerpc/platforms/pseries/reconfig.c: pSeries_reconfig_chain
arch/s390/kernel/process.c: idle_chain
arch/x86_64/kernel/process.c idle_notifier
drivers/base/memory.c: memory_chain
drivers/cpufreq/cpufreq.c cpufreq_policy_notifier_list
drivers/cpufreq/cpufreq.c cpufreq_transition_notifier_list
drivers/macintosh/adb.c: adb_client_list
drivers/macintosh/via-pmu.c sleep_notifier_list
drivers/macintosh/via-pmu68k.c sleep_notifier_list
drivers/macintosh/windfarm_core.c wf_client_list
drivers/usb/core/notify.c usb_notifier_list
drivers/video/fbmem.c fb_notifier_list
kernel/cpu.c cpu_chain
kernel/module.c module_notify_list
kernel/profile.c munmap_notifier
kernel/profile.c task_exit_notifier
kernel/sys.c reboot_notifier_list
net/core/dev.c netdev_chain
net/decnet/dn_dev.c: dnaddr_chain
net/ipv4/devinet.c: inetaddr_chain
It's possible that some of these classifications are wrong. If they are,
please let us know or submit a patch to fix them. Note that any chain that
gets called very frequently should be atomic, because the rwsem read-locking
used for blocking chains is very likely to incur cache misses on SMP systems.
(However, if the chain's callout routines may sleep then the chain cannot be
atomic.)
The patch set was written by Alan Stern and Chandra Seetharaman, incorporating
material written by Keith Owens and suggestions from Paul McKenney and Andrew
Morton.
[jes@sgi.com: restructure the notifier chain initialization macros]
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: Jes Sorensen <jes@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-27 17:16:30 +08:00
|
|
|
blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
|
2013-08-07 01:23:05 +08:00
|
|
|
CPUFREQ_NOTIFY, new_policy);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-08-07 01:23:05 +08:00
|
|
|
policy->min = new_policy->min;
|
|
|
|
policy->max = new_policy->max;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2011-03-27 21:04:46 +08:00
|
|
|
pr_debug("new min and max freqs are %u - %u kHz\n",
|
2014-03-12 01:03:00 +08:00
|
|
|
policy->min, policy->max);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-04-29 06:08:16 +08:00
|
|
|
if (cpufreq_driver->setpolicy) {
|
2013-08-07 01:23:05 +08:00
|
|
|
policy->policy = new_policy->policy;
|
2011-03-27 21:04:46 +08:00
|
|
|
pr_debug("setting range\n");
|
2014-02-18 05:56:35 +08:00
|
|
|
return cpufreq_driver->setpolicy(new_policy);
|
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-02-18 05:56:35 +08:00
|
|
|
if (new_policy->governor == policy->governor)
|
|
|
|
goto out;
|
2013-03-27 23:58:57 +08:00
|
|
|
|
2014-02-18 05:56:35 +08:00
|
|
|
pr_debug("governor switch\n");
|
|
|
|
|
|
|
|
/* save old, working values */
|
|
|
|
old_gov = policy->governor;
|
|
|
|
/* end old governor */
|
|
|
|
if (old_gov) {
|
|
|
|
__cpufreq_governor(policy, CPUFREQ_GOV_STOP);
|
|
|
|
up_write(&policy->rwsem);
|
2014-03-20 05:29:17 +08:00
|
|
|
__cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
|
2014-02-18 05:56:35 +08:00
|
|
|
down_write(&policy->rwsem);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2014-02-18 05:56:35 +08:00
|
|
|
/* start new governor */
|
|
|
|
policy->governor = new_policy->governor;
|
|
|
|
if (!__cpufreq_governor(policy, CPUFREQ_GOV_POLICY_INIT)) {
|
|
|
|
if (!__cpufreq_governor(policy, CPUFREQ_GOV_START))
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
up_write(&policy->rwsem);
|
|
|
|
__cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT);
|
|
|
|
down_write(&policy->rwsem);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* new governor failed, so re-start old one */
|
|
|
|
pr_debug("starting governor %s failed\n", policy->governor->name);
|
|
|
|
if (old_gov) {
|
|
|
|
policy->governor = old_gov;
|
|
|
|
__cpufreq_governor(policy, CPUFREQ_GOV_POLICY_INIT);
|
|
|
|
__cpufreq_governor(policy, CPUFREQ_GOV_START);
|
|
|
|
}
|
|
|
|
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
out:
|
|
|
|
pr_debug("governor: change or update limits\n");
|
|
|
|
return __cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* cpufreq_update_policy - re-evaluate an existing cpufreq policy
|
|
|
|
* @cpu: CPU which shall be re-evaluated
|
|
|
|
*
|
2011-03-31 09:57:33 +08:00
|
|
|
* Useful for policy notifiers which have different necessities
|
2005-04-17 06:20:36 +08:00
|
|
|
* at different times.
|
|
|
|
*/
|
|
|
|
int cpufreq_update_policy(unsigned int cpu)
|
|
|
|
{
|
2013-08-07 01:23:05 +08:00
|
|
|
struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
|
|
|
|
struct cpufreq_policy new_policy;
|
2008-07-26 04:44:53 +08:00
|
|
|
int ret;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-06-19 02:27:32 +08:00
|
|
|
if (!policy)
|
|
|
|
return -ENODEV;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-10-18 21:40:15 +08:00
|
|
|
down_write(&policy->rwsem);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2011-03-27 21:04:46 +08:00
|
|
|
pr_debug("updating policy for CPU %u\n", cpu);
|
2013-08-07 01:23:06 +08:00
|
|
|
memcpy(&new_policy, policy, sizeof(*policy));
|
2013-08-07 01:23:05 +08:00
|
|
|
new_policy.min = policy->user_policy.min;
|
|
|
|
new_policy.max = policy->user_policy.max;
|
|
|
|
new_policy.policy = policy->user_policy.policy;
|
|
|
|
new_policy.governor = policy->user_policy.governor;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-06-19 16:49:33 +08:00
|
|
|
/*
|
|
|
|
* BIOS might change freq behind our back
|
|
|
|
* -> ask driver for current freq and notify governors about a change
|
|
|
|
*/
|
2014-03-13 04:49:33 +08:00
|
|
|
if (cpufreq_driver->get && !cpufreq_driver->setpolicy) {
|
2013-08-07 01:23:05 +08:00
|
|
|
new_policy.cur = cpufreq_driver->get(cpu);
|
2014-02-25 16:59:44 +08:00
|
|
|
if (WARN_ON(!new_policy.cur)) {
|
|
|
|
ret = -EIO;
|
2014-06-19 02:27:32 +08:00
|
|
|
goto unlock;
|
2014-02-25 16:59:44 +08:00
|
|
|
}
|
|
|
|
|
2013-08-07 01:23:05 +08:00
|
|
|
if (!policy->cur) {
|
2014-03-12 01:03:00 +08:00
|
|
|
pr_debug("Driver did not initialize current freq\n");
|
2013-08-07 01:23:05 +08:00
|
|
|
policy->cur = new_policy.cur;
|
2006-02-01 18:36:04 +08:00
|
|
|
} else {
|
2013-10-25 22:15:48 +08:00
|
|
|
if (policy->cur != new_policy.cur && has_target())
|
2015-01-02 15:04:28 +08:00
|
|
|
cpufreq_out_of_sync(policy, new_policy.cur);
|
2006-02-01 18:36:04 +08:00
|
|
|
}
|
2006-01-27 01:46:33 +08:00
|
|
|
}
|
|
|
|
|
2013-10-02 16:43:16 +08:00
|
|
|
ret = cpufreq_set_policy(policy, &new_policy);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-06-19 02:27:32 +08:00
|
|
|
unlock:
|
2013-10-18 21:40:15 +08:00
|
|
|
up_write(&policy->rwsem);
|
2007-02-06 08:12:44 +08:00
|
|
|
|
2013-08-07 01:23:05 +08:00
|
|
|
cpufreq_cpu_put(policy);
|
2005-04-17 06:20:36 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(cpufreq_update_policy);
|
|
|
|
|
2013-06-20 01:54:04 +08:00
|
|
|
static int cpufreq_cpu_callback(struct notifier_block *nfb,
|
2005-10-31 06:59:54 +08:00
|
|
|
unsigned long action, void *hcpu)
|
|
|
|
{
|
|
|
|
unsigned int cpu = (unsigned long)hcpu;
|
2011-12-22 06:29:42 +08:00
|
|
|
struct device *dev;
|
2005-10-31 06:59:54 +08:00
|
|
|
|
2011-12-22 06:29:42 +08:00
|
|
|
dev = get_cpu_device(cpu);
|
|
|
|
if (dev) {
|
2013-07-30 06:55:25 +08:00
|
|
|
switch (action & ~CPU_TASKS_FROZEN) {
|
2005-10-31 06:59:54 +08:00
|
|
|
case CPU_ONLINE:
|
2014-03-10 17:23:35 +08:00
|
|
|
__cpufreq_add_dev(dev, NULL);
|
2005-10-31 06:59:54 +08:00
|
|
|
break;
|
2013-07-30 06:55:25 +08:00
|
|
|
|
2005-10-31 06:59:54 +08:00
|
|
|
case CPU_DOWN_PREPARE:
|
2014-03-10 17:23:35 +08:00
|
|
|
__cpufreq_remove_dev_prepare(dev, NULL);
|
cpufreq: Invoke __cpufreq_remove_dev_finish() after releasing cpu_hotplug.lock
__cpufreq_remove_dev_finish() handles the kobject cleanup for a CPU going
offline. But because we destroy the kobject towards the end of the CPU offline
phase, there are certain race windows where a task can try to write to a
cpufreq sysfs file (eg: using store_scaling_max_freq()) while we are taking
that CPU offline, and this can bump up the kobject refcount, which in turn might
hinder the CPU offline task from running to completion. (It can also cause
other more serious problems such as trying to acquire a destroyed timer-mutex
etc., depending on the exact stage of the cleanup at which the task managed to
take a new refcount).
To fix the race window, we will need to synchronize those store_*() call-sites
with CPU hotplug, using get_online_cpus()/put_online_cpus(). However, that
in turn can cause a total deadlock because it can end up waiting for the
CPU offline task to complete, with incremented refcount!
Write to sysfs CPU offline task
-------------- ----------------
kobj_refcnt++
Acquire cpu_hotplug.lock
get_online_cpus();
Wait for kobj_refcnt to drop to zero
**DEADLOCK**
A simple way to avoid this problem is to perform the kobject cleanup in the
CPU offline path, with the cpu_hotplug.lock *released*. That is, we can
perform the wait-for-kobj-refcnt-to-drop as well as the subsequent cleanup
in the CPU_POST_DEAD stage of CPU offline, which is run with cpu_hotplug.lock
released. Doing this helps us avoid deadlocks due to holding kobject refcounts
and waiting on each other on the cpu_hotplug.lock.
(Note: We can't move all of the cpufreq CPU offline steps to the
CPU_POST_DEAD stage, because certain things such as stopping the governors
have to be done before the outgoing CPU is marked offline. So retain those
parts in the CPU_DOWN_PREPARE stage itself).
Reported-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2013-09-07 03:53:27 +08:00
|
|
|
break;
|
|
|
|
|
|
|
|
case CPU_POST_DEAD:
|
2014-03-10 17:23:35 +08:00
|
|
|
__cpufreq_remove_dev_finish(dev, NULL);
|
2005-10-31 06:59:54 +08:00
|
|
|
break;
|
2013-07-30 06:55:25 +08:00
|
|
|
|
2007-02-06 08:12:44 +08:00
|
|
|
case CPU_DOWN_FAILED:
|
2014-03-10 17:23:35 +08:00
|
|
|
__cpufreq_add_dev(dev, NULL);
|
2005-10-31 06:59:54 +08:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return NOTIFY_OK;
|
|
|
|
}
|
|
|
|
|
2010-06-23 11:02:44 +08:00
|
|
|
static struct notifier_block __refdata cpufreq_cpu_notifier = {
|
2013-06-19 16:49:33 +08:00
|
|
|
.notifier_call = cpufreq_cpu_callback,
|
2005-10-31 06:59:54 +08:00
|
|
|
};
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-12-20 22:24:49 +08:00
|
|
|
/*********************************************************************
|
|
|
|
* BOOST *
|
|
|
|
*********************************************************************/
|
|
|
|
static int cpufreq_boost_set_sw(int state)
|
|
|
|
{
|
|
|
|
struct cpufreq_frequency_table *freq_table;
|
|
|
|
struct cpufreq_policy *policy;
|
|
|
|
int ret = -EINVAL;
|
|
|
|
|
|
|
|
list_for_each_entry(policy, &cpufreq_policy_list, policy_list) {
|
|
|
|
freq_table = cpufreq_frequency_get_table(policy->cpu);
|
|
|
|
if (freq_table) {
|
|
|
|
ret = cpufreq_frequency_table_cpuinfo(policy,
|
|
|
|
freq_table);
|
|
|
|
if (ret) {
|
|
|
|
pr_err("%s: Policy frequency update failed\n",
|
|
|
|
__func__);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
policy->user_policy.max = policy->max;
|
|
|
|
__cpufreq_governor(policy, CPUFREQ_GOV_LIMITS);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
int cpufreq_boost_trigger_state(int state)
|
|
|
|
{
|
|
|
|
unsigned long flags;
|
|
|
|
int ret = 0;
|
|
|
|
|
|
|
|
if (cpufreq_driver->boost_enabled == state)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
write_lock_irqsave(&cpufreq_driver_lock, flags);
|
|
|
|
cpufreq_driver->boost_enabled = state;
|
|
|
|
write_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
|
|
|
|
|
|
|
ret = cpufreq_driver->set_boost(state);
|
|
|
|
if (ret) {
|
|
|
|
write_lock_irqsave(&cpufreq_driver_lock, flags);
|
|
|
|
cpufreq_driver->boost_enabled = !state;
|
|
|
|
write_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
|
|
|
|
2014-03-12 01:03:00 +08:00
|
|
|
pr_err("%s: Cannot %s BOOST\n",
|
|
|
|
__func__, state ? "enable" : "disable");
|
2013-12-20 22:24:49 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
int cpufreq_boost_supported(void)
|
|
|
|
{
|
|
|
|
if (likely(cpufreq_driver))
|
|
|
|
return cpufreq_driver->boost_supported;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_boost_supported);
|
|
|
|
|
|
|
|
int cpufreq_boost_enabled(void)
|
|
|
|
{
|
|
|
|
return cpufreq_driver->boost_enabled;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/*********************************************************************
|
|
|
|
* REGISTER / UNREGISTER CPUFREQ DRIVER *
|
|
|
|
*********************************************************************/
|
|
|
|
|
|
|
|
/**
|
|
|
|
* cpufreq_register_driver - register a CPU Frequency driver
|
|
|
|
* @driver_data: A struct cpufreq_driver containing the values#
|
|
|
|
* submitted by the CPU Frequency driver.
|
|
|
|
*
|
2013-06-19 16:49:33 +08:00
|
|
|
* Registers a CPU Frequency driver to this core code. This code
|
2005-04-17 06:20:36 +08:00
|
|
|
* returns zero on success, -EBUSY when another driver got here first
|
2006-02-28 13:43:23 +08:00
|
|
|
* (and isn't unregistered in the meantime).
|
2005-04-17 06:20:36 +08:00
|
|
|
*
|
|
|
|
*/
|
2007-02-27 06:55:48 +08:00
|
|
|
int cpufreq_register_driver(struct cpufreq_driver *driver_data)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
unsigned long flags;
|
|
|
|
int ret;
|
|
|
|
|
2012-03-14 07:18:39 +08:00
|
|
|
if (cpufreq_disabled())
|
|
|
|
return -ENODEV;
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
if (!driver_data || !driver_data->verify || !driver_data->init ||
|
2013-10-25 22:15:48 +08:00
|
|
|
!(driver_data->setpolicy || driver_data->target_index ||
|
2014-03-19 19:48:30 +08:00
|
|
|
driver_data->target) ||
|
|
|
|
(driver_data->setpolicy && (driver_data->target_index ||
|
2014-06-03 01:19:28 +08:00
|
|
|
driver_data->target)) ||
|
|
|
|
(!!driver_data->get_intermediate != !!driver_data->target_intermediate))
|
2005-04-17 06:20:36 +08:00
|
|
|
return -EINVAL;
|
|
|
|
|
2011-03-27 21:04:46 +08:00
|
|
|
pr_debug("trying to register driver %s\n", driver_data->name);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-02-23 00:24:34 +08:00
|
|
|
write_lock_irqsave(&cpufreq_driver_lock, flags);
|
2013-04-29 06:08:16 +08:00
|
|
|
if (cpufreq_driver) {
|
2013-02-23 00:24:34 +08:00
|
|
|
write_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
2013-09-19 12:05:20 +08:00
|
|
|
return -EEXIST;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2013-04-29 06:08:16 +08:00
|
|
|
cpufreq_driver = driver_data;
|
2013-02-23 00:24:34 +08:00
|
|
|
write_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2015-01-02 15:04:30 +08:00
|
|
|
if (driver_data->setpolicy)
|
|
|
|
driver_data->flags |= CPUFREQ_CONST_LOOPS;
|
|
|
|
|
2013-12-20 22:24:49 +08:00
|
|
|
if (cpufreq_boost_supported()) {
|
|
|
|
/*
|
|
|
|
* Check if driver provides function to enable boost -
|
|
|
|
* if not, use cpufreq_boost_set_sw as default
|
|
|
|
*/
|
|
|
|
if (!cpufreq_driver->set_boost)
|
|
|
|
cpufreq_driver->set_boost = cpufreq_boost_set_sw;
|
|
|
|
|
|
|
|
ret = cpufreq_sysfs_create_file(&boost.attr);
|
|
|
|
if (ret) {
|
|
|
|
pr_err("%s: cannot register global BOOST sysfs file\n",
|
2014-03-12 01:03:00 +08:00
|
|
|
__func__);
|
2013-12-20 22:24:49 +08:00
|
|
|
goto err_null_driver;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2011-12-22 06:29:42 +08:00
|
|
|
ret = subsys_interface_register(&cpufreq_interface);
|
2011-03-02 00:41:10 +08:00
|
|
|
if (ret)
|
2013-12-20 22:24:49 +08:00
|
|
|
goto err_boost_unreg;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-04-29 06:08:16 +08:00
|
|
|
if (!(cpufreq_driver->flags & CPUFREQ_STICKY)) {
|
2005-04-17 06:20:36 +08:00
|
|
|
int i;
|
|
|
|
ret = -ENODEV;
|
|
|
|
|
|
|
|
/* check for at least one working CPU */
|
2008-03-26 06:06:53 +08:00
|
|
|
for (i = 0; i < nr_cpu_ids; i++)
|
|
|
|
if (cpu_possible(i) && per_cpu(cpufreq_cpu_data, i)) {
|
2005-04-17 06:20:36 +08:00
|
|
|
ret = 0;
|
2008-03-26 06:06:53 +08:00
|
|
|
break;
|
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/* if all ->init() calls failed, unregister */
|
|
|
|
if (ret) {
|
2011-03-27 21:04:46 +08:00
|
|
|
pr_debug("no CPU initialized for driver %s\n",
|
2014-03-12 01:03:00 +08:00
|
|
|
driver_data->name);
|
2011-12-22 06:29:42 +08:00
|
|
|
goto err_if_unreg;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2011-03-02 00:41:10 +08:00
|
|
|
register_hotcpu_notifier(&cpufreq_cpu_notifier);
|
2011-03-27 21:04:46 +08:00
|
|
|
pr_debug("driver %s up and running\n", driver_data->name);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2011-03-02 00:41:10 +08:00
|
|
|
return 0;
|
2011-12-22 06:29:42 +08:00
|
|
|
err_if_unreg:
|
|
|
|
subsys_interface_unregister(&cpufreq_interface);
|
2013-12-20 22:24:49 +08:00
|
|
|
err_boost_unreg:
|
|
|
|
if (cpufreq_boost_supported())
|
|
|
|
cpufreq_sysfs_remove_file(&boost.attr);
|
2011-03-02 00:41:10 +08:00
|
|
|
err_null_driver:
|
2013-02-23 00:24:34 +08:00
|
|
|
write_lock_irqsave(&cpufreq_driver_lock, flags);
|
2013-04-29 06:08:16 +08:00
|
|
|
cpufreq_driver = NULL;
|
2013-02-23 00:24:34 +08:00
|
|
|
write_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
2008-02-08 05:33:49 +08:00
|
|
|
return ret;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_register_driver);
|
|
|
|
|
|
|
|
/**
|
|
|
|
* cpufreq_unregister_driver - unregister the current CPUFreq driver
|
|
|
|
*
|
2013-06-19 16:49:33 +08:00
|
|
|
* Unregister the current CPUFreq driver. Only call this if you have
|
2005-04-17 06:20:36 +08:00
|
|
|
* the right to do so, i.e. if you have succeeded in initialising before!
|
|
|
|
* Returns zero if successful, and -EINVAL if the cpufreq_driver is
|
|
|
|
* currently not initialised.
|
|
|
|
*/
|
2007-02-27 06:55:48 +08:00
|
|
|
int cpufreq_unregister_driver(struct cpufreq_driver *driver)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
unsigned long flags;
|
|
|
|
|
2013-04-29 06:08:16 +08:00
|
|
|
if (!cpufreq_driver || (driver != cpufreq_driver))
|
2005-04-17 06:20:36 +08:00
|
|
|
return -EINVAL;
|
|
|
|
|
2011-03-27 21:04:46 +08:00
|
|
|
pr_debug("unregistering driver %s\n", driver->name);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2011-12-22 06:29:42 +08:00
|
|
|
subsys_interface_unregister(&cpufreq_interface);
|
2013-12-20 22:24:49 +08:00
|
|
|
if (cpufreq_boost_supported())
|
|
|
|
cpufreq_sysfs_remove_file(&boost.attr);
|
|
|
|
|
2006-06-27 17:54:08 +08:00
|
|
|
unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-08-07 01:23:11 +08:00
|
|
|
down_write(&cpufreq_rwsem);
|
2013-02-23 00:24:34 +08:00
|
|
|
write_lock_irqsave(&cpufreq_driver_lock, flags);
|
2013-08-07 01:23:11 +08:00
|
|
|
|
2013-04-29 06:08:16 +08:00
|
|
|
cpufreq_driver = NULL;
|
2013-08-07 01:23:11 +08:00
|
|
|
|
2013-02-23 00:24:34 +08:00
|
|
|
write_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
2013-08-07 01:23:11 +08:00
|
|
|
up_write(&cpufreq_rwsem);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
|
2007-02-06 08:12:44 +08:00
|
|
|
|
2014-12-24 14:09:48 +08:00
|
|
|
/*
|
|
|
|
* Stop cpufreq at shutdown to make sure it isn't holding any locks
|
|
|
|
* or mutexes when secondary CPUs are halted.
|
|
|
|
*/
|
|
|
|
static struct syscore_ops cpufreq_syscore_ops = {
|
|
|
|
.shutdown = cpufreq_suspend,
|
|
|
|
};
|
|
|
|
|
2007-02-06 08:12:44 +08:00
|
|
|
static int __init cpufreq_core_init(void)
|
|
|
|
{
|
2012-03-14 07:18:39 +08:00
|
|
|
if (cpufreq_disabled())
|
|
|
|
return -ENODEV;
|
|
|
|
|
2013-05-17 18:39:09 +08:00
|
|
|
cpufreq_global_kobject = kobject_create();
|
2009-07-24 21:25:05 +08:00
|
|
|
BUG_ON(!cpufreq_global_kobject);
|
|
|
|
|
2014-12-24 14:09:48 +08:00
|
|
|
register_syscore_ops(&cpufreq_syscore_ops);
|
|
|
|
|
2007-02-06 08:12:44 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
core_initcall(cpufreq_core_init);
|