2005-04-16 18:20:36 -04:00
|
|
|
/*
|
|
|
|
* linux/drivers/cpufreq/cpufreq.c
|
|
|
|
*
|
|
|
|
* Copyright (C) 2001 Russell King
|
|
|
|
* (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
|
|
|
|
*
|
2005-10-30 17:59:54 -05:00
|
|
|
* Oct 2005 - Ashok Raj <ashok.raj@intel.com>
|
2006-02-28 00:43:23 -05:00
|
|
|
* Added handling for CPU hotplug
|
2006-03-05 03:37:23 -05:00
|
|
|
* Feb 2006 - Jacob Shin <jacob.shin@amd.com>
|
|
|
|
* Fix handling for CPU hotplug -- affected CPUs
|
2005-10-30 17:59:54 -05:00
|
|
|
*
|
2005-04-16 18:20:36 -04:00
|
|
|
* 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.
|
|
|
|
*
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <linux/kernel.h>
|
|
|
|
#include <linux/module.h>
|
|
|
|
#include <linux/init.h>
|
|
|
|
#include <linux/notifier.h>
|
|
|
|
#include <linux/cpufreq.h>
|
|
|
|
#include <linux/delay.h>
|
|
|
|
#include <linux/interrupt.h>
|
|
|
|
#include <linux/spinlock.h>
|
|
|
|
#include <linux/device.h>
|
|
|
|
#include <linux/slab.h>
|
|
|
|
#include <linux/cpu.h>
|
|
|
|
#include <linux/completion.h>
|
2006-01-13 18:54:22 -05:00
|
|
|
#include <linux/mutex.h>
|
2005-04-16 18:20:36 -04:00
|
|
|
|
|
|
|
#define dprintk(msg...) cpufreq_debug_printk(CPUFREQ_DEBUG_CORE, "cpufreq-core", msg)
|
|
|
|
|
|
|
|
/**
|
2006-08-11 17:59:28 -04:00
|
|
|
* The "cpufreq driver" - the arch- or hardware-dependent low
|
2005-04-16 18:20:36 -04:00
|
|
|
* level driver of CPUFreq support, and its spinlock. This lock
|
|
|
|
* also protects the cpufreq_cpu_data array.
|
|
|
|
*/
|
2006-02-02 17:03:42 -05:00
|
|
|
static struct cpufreq_driver *cpufreq_driver;
|
|
|
|
static struct cpufreq_policy *cpufreq_cpu_data[NR_CPUS];
|
2005-04-16 18:20:36 -04:00
|
|
|
static DEFINE_SPINLOCK(cpufreq_driver_lock);
|
|
|
|
|
|
|
|
/* internal prototypes */
|
|
|
|
static int __cpufreq_governor(struct cpufreq_policy *policy, unsigned int event);
|
|
|
|
static void handle_update(void *data);
|
|
|
|
|
|
|
|
/**
|
2006-02-28 00:43:23 -05:00
|
|
|
* Two notifier lists: the "policy" list is involved in the
|
|
|
|
* validation process for a new CPU frequency policy; the
|
2005-04-16 18:20:36 -04:00
|
|
|
* "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 04:16:30 -05:00
|
|
|
static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
|
2006-10-04 05:17:06 -04:00
|
|
|
static struct srcu_notifier_head cpufreq_transition_notifier_list;
|
2005-04-16 18:20:36 -04:00
|
|
|
|
2006-10-04 05:17:06 -04:00
|
|
|
static int __init init_cpufreq_transition_notifier_list(void)
|
|
|
|
{
|
|
|
|
srcu_init_notifier_head(&cpufreq_transition_notifier_list);
|
|
|
|
return 0;
|
|
|
|
}
|
2006-11-20 14:47:18 -05:00
|
|
|
pure_initcall(init_cpufreq_transition_notifier_list);
|
2005-04-16 18:20:36 -04:00
|
|
|
|
|
|
|
static LIST_HEAD(cpufreq_governor_list);
|
2006-02-02 17:03:42 -05:00
|
|
|
static DEFINE_MUTEX (cpufreq_governor_mutex);
|
2005-04-16 18:20:36 -04:00
|
|
|
|
2006-02-02 17:03:42 -05:00
|
|
|
struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
|
2005-04-16 18:20:36 -04:00
|
|
|
{
|
|
|
|
struct cpufreq_policy *data;
|
|
|
|
unsigned long flags;
|
|
|
|
|
|
|
|
if (cpu >= NR_CPUS)
|
|
|
|
goto err_out;
|
|
|
|
|
|
|
|
/* get the cpufreq driver */
|
|
|
|
spin_lock_irqsave(&cpufreq_driver_lock, flags);
|
|
|
|
|
|
|
|
if (!cpufreq_driver)
|
|
|
|
goto err_out_unlock;
|
|
|
|
|
|
|
|
if (!try_module_get(cpufreq_driver->owner))
|
|
|
|
goto err_out_unlock;
|
|
|
|
|
|
|
|
|
|
|
|
/* get the CPU */
|
|
|
|
data = cpufreq_cpu_data[cpu];
|
|
|
|
|
|
|
|
if (!data)
|
|
|
|
goto err_out_put_module;
|
|
|
|
|
|
|
|
if (!kobject_get(&data->kobj))
|
|
|
|
goto err_out_put_module;
|
|
|
|
|
|
|
|
spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
|
|
|
return data;
|
|
|
|
|
2006-02-02 17:03:42 -05:00
|
|
|
err_out_put_module:
|
2005-04-16 18:20:36 -04:00
|
|
|
module_put(cpufreq_driver->owner);
|
2006-02-02 17:03:42 -05:00
|
|
|
err_out_unlock:
|
2005-04-16 18:20:36 -04:00
|
|
|
spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
2006-02-02 17:03:42 -05:00
|
|
|
err_out:
|
2005-04-16 18:20:36 -04:00
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
|
|
|
|
|
2006-02-02 17:03:42 -05:00
|
|
|
|
2005-04-16 18:20:36 -04:00
|
|
|
void cpufreq_cpu_put(struct cpufreq_policy *data)
|
|
|
|
{
|
|
|
|
kobject_put(&data->kobj);
|
|
|
|
module_put(cpufreq_driver->owner);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
|
|
|
|
|
|
|
|
|
|
|
|
/*********************************************************************
|
|
|
|
* UNIFIED DEBUG HELPERS *
|
|
|
|
*********************************************************************/
|
|
|
|
#ifdef CONFIG_CPU_FREQ_DEBUG
|
|
|
|
|
|
|
|
/* what part(s) of the CPUfreq subsystem are debugged? */
|
|
|
|
static unsigned int debug;
|
|
|
|
|
|
|
|
/* is the debug output ratelimit'ed using printk_ratelimit? User can
|
|
|
|
* set or modify this value.
|
|
|
|
*/
|
|
|
|
static unsigned int debug_ratelimit = 1;
|
|
|
|
|
|
|
|
/* is the printk_ratelimit'ing enabled? It's enabled after a successful
|
|
|
|
* loading of a cpufreq driver, temporarily disabled when a new policy
|
|
|
|
* is set, and disabled upon cpufreq driver removal
|
|
|
|
*/
|
|
|
|
static unsigned int disable_ratelimit = 1;
|
|
|
|
static DEFINE_SPINLOCK(disable_ratelimit_lock);
|
|
|
|
|
2006-01-14 16:20:43 -05:00
|
|
|
static void cpufreq_debug_enable_ratelimit(void)
|
2005-04-16 18:20:36 -04:00
|
|
|
{
|
|
|
|
unsigned long flags;
|
|
|
|
|
|
|
|
spin_lock_irqsave(&disable_ratelimit_lock, flags);
|
|
|
|
if (disable_ratelimit)
|
|
|
|
disable_ratelimit--;
|
|
|
|
spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
|
|
|
|
}
|
|
|
|
|
2006-01-14 16:20:43 -05:00
|
|
|
static void cpufreq_debug_disable_ratelimit(void)
|
2005-04-16 18:20:36 -04:00
|
|
|
{
|
|
|
|
unsigned long flags;
|
|
|
|
|
|
|
|
spin_lock_irqsave(&disable_ratelimit_lock, flags);
|
|
|
|
disable_ratelimit++;
|
|
|
|
spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
|
|
|
|
}
|
|
|
|
|
|
|
|
void cpufreq_debug_printk(unsigned int type, const char *prefix, const char *fmt, ...)
|
|
|
|
{
|
|
|
|
char s[256];
|
|
|
|
va_list args;
|
|
|
|
unsigned int len;
|
|
|
|
unsigned long flags;
|
2006-02-28 00:43:23 -05:00
|
|
|
|
2005-04-16 18:20:36 -04:00
|
|
|
WARN_ON(!prefix);
|
|
|
|
if (type & debug) {
|
|
|
|
spin_lock_irqsave(&disable_ratelimit_lock, flags);
|
|
|
|
if (!disable_ratelimit && debug_ratelimit && !printk_ratelimit()) {
|
|
|
|
spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
|
|
|
|
|
|
|
|
len = snprintf(s, 256, KERN_DEBUG "%s: ", prefix);
|
|
|
|
|
|
|
|
va_start(args, fmt);
|
|
|
|
len += vsnprintf(&s[len], (256 - len), fmt, args);
|
|
|
|
va_end(args);
|
|
|
|
|
|
|
|
printk(s);
|
|
|
|
|
|
|
|
WARN_ON(len < 5);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(cpufreq_debug_printk);
|
|
|
|
|
|
|
|
|
|
|
|
module_param(debug, uint, 0644);
|
|
|
|
MODULE_PARM_DESC(debug, "CPUfreq debugging: add 1 to debug core, 2 to debug drivers, and 4 to debug governors.");
|
|
|
|
|
|
|
|
module_param(debug_ratelimit, uint, 0644);
|
|
|
|
MODULE_PARM_DESC(debug_ratelimit, "CPUfreq debugging: set to 0 to disable ratelimiting.");
|
|
|
|
|
|
|
|
#else /* !CONFIG_CPU_FREQ_DEBUG */
|
|
|
|
|
|
|
|
static inline void cpufreq_debug_enable_ratelimit(void) { return; }
|
|
|
|
static inline void cpufreq_debug_disable_ratelimit(void) { return; }
|
|
|
|
|
|
|
|
#endif /* CONFIG_CPU_FREQ_DEBUG */
|
|
|
|
|
|
|
|
|
|
|
|
/*********************************************************************
|
|
|
|
* 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 00:43:23 -05:00
|
|
|
* systems as each CPU might be scaled differently. So, use the arch
|
2005-04-16 18:20:36 -04:00
|
|
|
* per-CPU loops_per_jiffy value wherever possible.
|
|
|
|
*/
|
|
|
|
#ifndef CONFIG_SMP
|
|
|
|
static unsigned long l_p_j_ref;
|
|
|
|
static unsigned int l_p_j_ref_freq;
|
|
|
|
|
2006-01-14 16:20:43 -05:00
|
|
|
static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
|
2005-04-16 18:20:36 -04:00
|
|
|
{
|
|
|
|
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;
|
|
|
|
dprintk("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n", l_p_j_ref, l_p_j_ref_freq);
|
|
|
|
}
|
|
|
|
if ((val == CPUFREQ_PRECHANGE && ci->old < ci->new) ||
|
|
|
|
(val == CPUFREQ_POSTCHANGE && ci->old > ci->new) ||
|
2005-04-29 10:40:12 -04:00
|
|
|
(val == CPUFREQ_RESUMECHANGE || val == CPUFREQ_SUSPENDCHANGE)) {
|
2005-04-16 18:20:36 -04:00
|
|
|
loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq, ci->new);
|
|
|
|
dprintk("scaling loops_per_jiffy to %lu for frequency %u kHz\n", loops_per_jiffy, ci->new);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
#else
|
|
|
|
static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci) { return; }
|
|
|
|
#endif
|
|
|
|
|
|
|
|
|
|
|
|
/**
|
2006-01-31 18:53:55 -05:00
|
|
|
* cpufreq_notify_transition - call notifier chain and adjust_jiffies
|
|
|
|
* on frequency transition.
|
2005-04-16 18:20:36 -04:00
|
|
|
*
|
2006-01-31 18:53:55 -05:00
|
|
|
* This function calls the transition notifiers and the "adjust_jiffies"
|
|
|
|
* function. It is called twice on all CPU frequency changes that have
|
2006-02-28 00:43:23 -05:00
|
|
|
* external effects.
|
2005-04-16 18:20:36 -04:00
|
|
|
*/
|
|
|
|
void cpufreq_notify_transition(struct cpufreq_freqs *freqs, unsigned int state)
|
|
|
|
{
|
2006-01-31 18:53:55 -05:00
|
|
|
struct cpufreq_policy *policy;
|
|
|
|
|
2005-04-16 18:20:36 -04:00
|
|
|
BUG_ON(irqs_disabled());
|
|
|
|
|
|
|
|
freqs->flags = cpufreq_driver->flags;
|
2006-01-31 18:53:55 -05:00
|
|
|
dprintk("notification %u of frequency transition to %u kHz\n",
|
|
|
|
state, freqs->new);
|
2005-04-16 18:20:36 -04:00
|
|
|
|
2006-01-31 18:53:55 -05:00
|
|
|
policy = cpufreq_cpu_data[freqs->cpu];
|
2005-04-16 18:20:36 -04:00
|
|
|
switch (state) {
|
2006-01-31 18:53:55 -05:00
|
|
|
|
2005-04-16 18:20:36 -04:00
|
|
|
case CPUFREQ_PRECHANGE:
|
2006-02-28 00:43:23 -05:00
|
|
|
/* detect if the driver reported a value as "old frequency"
|
2006-01-31 18:53:55 -05:00
|
|
|
* which is not equal to what the cpufreq core thinks is
|
|
|
|
* "old frequency".
|
2005-04-16 18:20:36 -04:00
|
|
|
*/
|
|
|
|
if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
|
2006-01-31 18:53:55 -05:00
|
|
|
if ((policy) && (policy->cpu == freqs->cpu) &&
|
|
|
|
(policy->cur) && (policy->cur != freqs->old)) {
|
2006-04-28 07:47:13 -04:00
|
|
|
dprintk("Warning: CPU frequency is"
|
2006-01-31 18:53:55 -05:00
|
|
|
" %u, cpufreq assumed %u kHz.\n",
|
|
|
|
freqs->old, policy->cur);
|
|
|
|
freqs->old = policy->cur;
|
2005-04-16 18:20:36 -04:00
|
|
|
}
|
|
|
|
}
|
2006-10-04 05:17:06 -04:00
|
|
|
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 04:16:30 -05:00
|
|
|
CPUFREQ_PRECHANGE, freqs);
|
2005-04-16 18:20:36 -04:00
|
|
|
adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
|
|
|
|
break;
|
2006-01-31 18:53:55 -05:00
|
|
|
|
2005-04-16 18:20:36 -04:00
|
|
|
case CPUFREQ_POSTCHANGE:
|
|
|
|
adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
|
2006-10-04 05:17:06 -04:00
|
|
|
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 04:16:30 -05:00
|
|
|
CPUFREQ_POSTCHANGE, freqs);
|
2006-01-31 18:53:55 -05:00
|
|
|
if (likely(policy) && likely(policy->cpu == freqs->cpu))
|
|
|
|
policy->cur = freqs->new;
|
2005-04-16 18:20:36 -04:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_notify_transition);
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
/*********************************************************************
|
|
|
|
* SYSFS INTERFACE *
|
|
|
|
*********************************************************************/
|
|
|
|
|
2006-07-06 15:30:26 -04:00
|
|
|
static struct cpufreq_governor *__find_governor(const char *str_governor)
|
|
|
|
{
|
|
|
|
struct cpufreq_governor *t;
|
|
|
|
|
|
|
|
list_for_each_entry(t, &cpufreq_governor_list, governor_list)
|
|
|
|
if (!strnicmp(str_governor,t->name,CPUFREQ_NAME_LEN))
|
|
|
|
return t;
|
|
|
|
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
2005-04-16 18:20:36 -04:00
|
|
|
/**
|
|
|
|
* cpufreq_parse_governor - parse a governor string
|
|
|
|
*/
|
|
|
|
static int cpufreq_parse_governor (char *str_governor, unsigned int *policy,
|
|
|
|
struct cpufreq_governor **governor)
|
|
|
|
{
|
2006-07-06 15:30:26 -04:00
|
|
|
int err = -EINVAL;
|
|
|
|
|
2005-04-16 18:20:36 -04:00
|
|
|
if (!cpufreq_driver)
|
2006-07-06 15:30:26 -04:00
|
|
|
goto out;
|
|
|
|
|
2005-04-16 18:20:36 -04:00
|
|
|
if (cpufreq_driver->setpolicy) {
|
|
|
|
if (!strnicmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
|
|
|
|
*policy = CPUFREQ_POLICY_PERFORMANCE;
|
2006-07-06 15:30:26 -04:00
|
|
|
err = 0;
|
2005-04-16 18:20:36 -04:00
|
|
|
} else if (!strnicmp(str_governor, "powersave", CPUFREQ_NAME_LEN)) {
|
|
|
|
*policy = CPUFREQ_POLICY_POWERSAVE;
|
2006-07-06 15:30:26 -04:00
|
|
|
err = 0;
|
2005-04-16 18:20:36 -04:00
|
|
|
}
|
2006-07-06 15:30:26 -04:00
|
|
|
} else if (cpufreq_driver->target) {
|
2005-04-16 18:20:36 -04:00
|
|
|
struct cpufreq_governor *t;
|
2006-07-06 15:30:26 -04:00
|
|
|
|
2006-01-13 18:54:22 -05:00
|
|
|
mutex_lock(&cpufreq_governor_mutex);
|
2006-07-06 15:30:26 -04:00
|
|
|
|
|
|
|
t = __find_governor(str_governor);
|
|
|
|
|
2006-07-06 15:32:01 -04:00
|
|
|
if (t == NULL) {
|
|
|
|
char *name = kasprintf(GFP_KERNEL, "cpufreq_%s", str_governor);
|
|
|
|
|
|
|
|
if (name) {
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
mutex_unlock(&cpufreq_governor_mutex);
|
|
|
|
ret = request_module(name);
|
|
|
|
mutex_lock(&cpufreq_governor_mutex);
|
|
|
|
|
|
|
|
if (ret == 0)
|
|
|
|
t = __find_governor(str_governor);
|
|
|
|
}
|
|
|
|
|
|
|
|
kfree(name);
|
|
|
|
}
|
|
|
|
|
2006-07-06 15:30:26 -04:00
|
|
|
if (t != NULL) {
|
|
|
|
*governor = t;
|
|
|
|
err = 0;
|
2005-04-16 18:20:36 -04:00
|
|
|
}
|
2006-07-06 15:30:26 -04:00
|
|
|
|
2006-01-13 18:54:22 -05:00
|
|
|
mutex_unlock(&cpufreq_governor_mutex);
|
2005-04-16 18:20:36 -04:00
|
|
|
}
|
2006-07-06 15:30:26 -04:00
|
|
|
out:
|
|
|
|
return err;
|
2005-04-16 18:20:36 -04:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/* drivers/base/cpu.c */
|
|
|
|
extern struct sysdev_class cpu_sysdev_class;
|
|
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
* cpufreq_per_cpu_attr_read() / show_##file_name() - print out cpufreq information
|
|
|
|
*
|
|
|
|
* Write out information from cpufreq_driver->policy[cpu]; object must be
|
|
|
|
* "unsigned int".
|
|
|
|
*/
|
|
|
|
|
2006-02-28 00:43:23 -05:00
|
|
|
#define show_one(file_name, object) \
|
|
|
|
static ssize_t show_##file_name \
|
|
|
|
(struct cpufreq_policy * policy, char *buf) \
|
|
|
|
{ \
|
|
|
|
return sprintf (buf, "%u\n", policy->object); \
|
2005-04-16 18:20:36 -04:00
|
|
|
}
|
|
|
|
|
|
|
|
show_one(cpuinfo_min_freq, cpuinfo.min_freq);
|
|
|
|
show_one(cpuinfo_max_freq, cpuinfo.max_freq);
|
|
|
|
show_one(scaling_min_freq, min);
|
|
|
|
show_one(scaling_max_freq, max);
|
|
|
|
show_one(scaling_cur_freq, cur);
|
|
|
|
|
2006-04-13 09:14:04 -04:00
|
|
|
static int __cpufreq_set_policy(struct cpufreq_policy *data, struct cpufreq_policy *policy);
|
|
|
|
|
2005-04-16 18:20:36 -04:00
|
|
|
/**
|
|
|
|
* cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
|
|
|
|
*/
|
|
|
|
#define store_one(file_name, object) \
|
|
|
|
static ssize_t store_##file_name \
|
|
|
|
(struct cpufreq_policy * policy, const char *buf, size_t count) \
|
|
|
|
{ \
|
|
|
|
unsigned int ret = -EINVAL; \
|
|
|
|
struct cpufreq_policy new_policy; \
|
|
|
|
\
|
|
|
|
ret = cpufreq_get_policy(&new_policy, policy->cpu); \
|
|
|
|
if (ret) \
|
|
|
|
return -EINVAL; \
|
|
|
|
\
|
|
|
|
ret = sscanf (buf, "%u", &new_policy.object); \
|
|
|
|
if (ret != 1) \
|
|
|
|
return -EINVAL; \
|
|
|
|
\
|
2006-07-26 09:40:07 -04:00
|
|
|
lock_cpu_hotplug(); \
|
2006-04-13 09:14:04 -04:00
|
|
|
mutex_lock(&policy->lock); \
|
|
|
|
ret = __cpufreq_set_policy(policy, &new_policy); \
|
|
|
|
policy->user_policy.object = policy->object; \
|
|
|
|
mutex_unlock(&policy->lock); \
|
2006-07-26 09:40:07 -04:00
|
|
|
unlock_cpu_hotplug(); \
|
2005-04-16 18:20:36 -04:00
|
|
|
\
|
|
|
|
return ret ? ret : count; \
|
|
|
|
}
|
|
|
|
|
|
|
|
store_one(scaling_min_freq,min);
|
|
|
|
store_one(scaling_max_freq,max);
|
|
|
|
|
|
|
|
/**
|
|
|
|
* show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
|
|
|
|
*/
|
|
|
|
static ssize_t show_cpuinfo_cur_freq (struct cpufreq_policy * policy, char *buf)
|
|
|
|
{
|
|
|
|
unsigned int cur_freq = cpufreq_get(policy->cpu);
|
|
|
|
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
|
|
|
|
*/
|
|
|
|
static ssize_t show_scaling_governor (struct cpufreq_policy * policy, char *buf)
|
|
|
|
{
|
|
|
|
if(policy->policy == CPUFREQ_POLICY_POWERSAVE)
|
|
|
|
return sprintf(buf, "powersave\n");
|
|
|
|
else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
|
|
|
|
return sprintf(buf, "performance\n");
|
|
|
|
else if (policy->governor)
|
|
|
|
return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", policy->governor->name);
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
* store_scaling_governor - store policy for the specified CPU
|
|
|
|
*/
|
2006-02-28 00:43:23 -05:00
|
|
|
static ssize_t store_scaling_governor (struct cpufreq_policy * policy,
|
|
|
|
const char *buf, size_t count)
|
2005-04-16 18:20:36 -04:00
|
|
|
{
|
|
|
|
unsigned int ret = -EINVAL;
|
|
|
|
char str_governor[16];
|
|
|
|
struct cpufreq_policy new_policy;
|
|
|
|
|
|
|
|
ret = cpufreq_get_policy(&new_policy, policy->cpu);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
ret = sscanf (buf, "%15s", str_governor);
|
|
|
|
if (ret != 1)
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
if (cpufreq_parse_governor(str_governor, &new_policy.policy, &new_policy.governor))
|
|
|
|
return -EINVAL;
|
|
|
|
|
2006-07-07 12:31:27 -04:00
|
|
|
lock_cpu_hotplug();
|
|
|
|
|
2006-04-13 09:14:04 -04:00
|
|
|
/* Do not use cpufreq_set_policy here or the user_policy.max
|
|
|
|
will be wrongly overridden */
|
|
|
|
mutex_lock(&policy->lock);
|
|
|
|
ret = __cpufreq_set_policy(policy, &new_policy);
|
|
|
|
|
|
|
|
policy->user_policy.policy = policy->policy;
|
|
|
|
policy->user_policy.governor = policy->governor;
|
|
|
|
mutex_unlock(&policy->lock);
|
|
|
|
|
2006-07-07 12:31:27 -04:00
|
|
|
unlock_cpu_hotplug();
|
|
|
|
|
2005-04-16 18:20:36 -04:00
|
|
|
return ret ? ret : count;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* show_scaling_driver - show the cpufreq driver currently loaded
|
|
|
|
*/
|
|
|
|
static ssize_t show_scaling_driver (struct cpufreq_policy * policy, char *buf)
|
|
|
|
{
|
|
|
|
return scnprintf(buf, CPUFREQ_NAME_LEN, "%s\n", cpufreq_driver->name);
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* show_scaling_available_governors - show the available CPUfreq governors
|
|
|
|
*/
|
|
|
|
static ssize_t show_scaling_available_governors (struct cpufreq_policy * policy,
|
|
|
|
char *buf)
|
|
|
|
{
|
|
|
|
ssize_t i = 0;
|
|
|
|
struct cpufreq_governor *t;
|
|
|
|
|
|
|
|
if (!cpufreq_driver->target) {
|
|
|
|
i += sprintf(buf, "performance powersave");
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
list_for_each_entry(t, &cpufreq_governor_list, governor_list) {
|
|
|
|
if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char)) - (CPUFREQ_NAME_LEN + 2)))
|
|
|
|
goto out;
|
|
|
|
i += scnprintf(&buf[i], CPUFREQ_NAME_LEN, "%s ", t->name);
|
|
|
|
}
|
2006-02-02 17:03:42 -05:00
|
|
|
out:
|
2005-04-16 18:20:36 -04:00
|
|
|
i += sprintf(&buf[i], "\n");
|
|
|
|
return i;
|
|
|
|
}
|
|
|
|
/**
|
|
|
|
* show_affected_cpus - show the CPUs affected by each transition
|
|
|
|
*/
|
|
|
|
static ssize_t show_affected_cpus (struct cpufreq_policy * policy, char *buf)
|
|
|
|
{
|
|
|
|
ssize_t i = 0;
|
|
|
|
unsigned int cpu;
|
|
|
|
|
|
|
|
for_each_cpu_mask(cpu, policy->cpus) {
|
|
|
|
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))
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
i += sprintf(&buf[i], "\n");
|
|
|
|
return i;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
#define define_one_ro(_name) \
|
|
|
|
static struct freq_attr _name = \
|
|
|
|
__ATTR(_name, 0444, show_##_name, NULL)
|
|
|
|
|
|
|
|
#define define_one_ro0400(_name) \
|
|
|
|
static struct freq_attr _name = \
|
|
|
|
__ATTR(_name, 0400, show_##_name, NULL)
|
|
|
|
|
|
|
|
#define define_one_rw(_name) \
|
|
|
|
static struct freq_attr _name = \
|
|
|
|
__ATTR(_name, 0644, show_##_name, store_##_name)
|
|
|
|
|
|
|
|
define_one_ro0400(cpuinfo_cur_freq);
|
|
|
|
define_one_ro(cpuinfo_min_freq);
|
|
|
|
define_one_ro(cpuinfo_max_freq);
|
|
|
|
define_one_ro(scaling_available_governors);
|
|
|
|
define_one_ro(scaling_driver);
|
|
|
|
define_one_ro(scaling_cur_freq);
|
|
|
|
define_one_ro(affected_cpus);
|
|
|
|
define_one_rw(scaling_min_freq);
|
|
|
|
define_one_rw(scaling_max_freq);
|
|
|
|
define_one_rw(scaling_governor);
|
|
|
|
|
|
|
|
static struct attribute * default_attrs[] = {
|
|
|
|
&cpuinfo_min_freq.attr,
|
|
|
|
&cpuinfo_max_freq.attr,
|
|
|
|
&scaling_min_freq.attr,
|
|
|
|
&scaling_max_freq.attr,
|
|
|
|
&affected_cpus.attr,
|
|
|
|
&scaling_governor.attr,
|
|
|
|
&scaling_driver.attr,
|
|
|
|
&scaling_available_governors.attr,
|
|
|
|
NULL
|
|
|
|
};
|
|
|
|
|
|
|
|
#define to_policy(k) container_of(k,struct cpufreq_policy,kobj)
|
|
|
|
#define to_attr(a) container_of(a,struct freq_attr,attr)
|
|
|
|
|
|
|
|
static ssize_t show(struct kobject * kobj, struct attribute * attr ,char * buf)
|
|
|
|
{
|
|
|
|
struct cpufreq_policy * policy = to_policy(kobj);
|
|
|
|
struct freq_attr * fattr = to_attr(attr);
|
|
|
|
ssize_t ret;
|
|
|
|
policy = cpufreq_cpu_get(policy->cpu);
|
|
|
|
if (!policy)
|
|
|
|
return -EINVAL;
|
2005-04-29 02:27:34 -04:00
|
|
|
ret = fattr->show ? fattr->show(policy,buf) : -EIO;
|
2005-04-16 18:20:36 -04:00
|
|
|
cpufreq_cpu_put(policy);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2006-02-28 00:43:23 -05:00
|
|
|
static ssize_t store(struct kobject * kobj, struct attribute * attr,
|
2005-04-16 18:20:36 -04:00
|
|
|
const char * buf, size_t count)
|
|
|
|
{
|
|
|
|
struct cpufreq_policy * policy = to_policy(kobj);
|
|
|
|
struct freq_attr * fattr = to_attr(attr);
|
|
|
|
ssize_t ret;
|
|
|
|
policy = cpufreq_cpu_get(policy->cpu);
|
|
|
|
if (!policy)
|
|
|
|
return -EINVAL;
|
2005-04-29 02:27:34 -04:00
|
|
|
ret = fattr->store ? fattr->store(policy,buf,count) : -EIO;
|
2005-04-16 18:20:36 -04:00
|
|
|
cpufreq_cpu_put(policy);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void cpufreq_sysfs_release(struct kobject * kobj)
|
|
|
|
{
|
|
|
|
struct cpufreq_policy * policy = to_policy(kobj);
|
|
|
|
dprintk("last reference is dropped\n");
|
|
|
|
complete(&policy->kobj_unregister);
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct sysfs_ops sysfs_ops = {
|
|
|
|
.show = show,
|
|
|
|
.store = store,
|
|
|
|
};
|
|
|
|
|
|
|
|
static struct kobj_type ktype_cpufreq = {
|
|
|
|
.sysfs_ops = &sysfs_ops,
|
|
|
|
.default_attrs = default_attrs,
|
|
|
|
.release = cpufreq_sysfs_release,
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
* cpufreq_add_dev - add a CPU device
|
|
|
|
*
|
2006-02-28 00:43:23 -05:00
|
|
|
* Adds the cpufreq interface for a CPU device.
|
2005-04-16 18:20:36 -04:00
|
|
|
*/
|
|
|
|
static int cpufreq_add_dev (struct sys_device * sys_dev)
|
|
|
|
{
|
|
|
|
unsigned int cpu = sys_dev->id;
|
|
|
|
int ret = 0;
|
|
|
|
struct cpufreq_policy new_policy;
|
|
|
|
struct cpufreq_policy *policy;
|
|
|
|
struct freq_attr **drv_attr;
|
2006-03-05 03:37:23 -05:00
|
|
|
struct sys_device *cpu_sys_dev;
|
2005-04-16 18:20:36 -04:00
|
|
|
unsigned long flags;
|
|
|
|
unsigned int j;
|
2006-03-05 03:37:23 -05:00
|
|
|
#ifdef CONFIG_SMP
|
|
|
|
struct cpufreq_policy *managed_policy;
|
|
|
|
#endif
|
2005-04-16 18:20:36 -04:00
|
|
|
|
2005-10-30 17:59:54 -05:00
|
|
|
if (cpu_is_offline(cpu))
|
|
|
|
return 0;
|
|
|
|
|
2005-04-16 18:20:36 -04:00
|
|
|
cpufreq_debug_disable_ratelimit();
|
|
|
|
dprintk("adding CPU %u\n", cpu);
|
|
|
|
|
|
|
|
#ifdef CONFIG_SMP
|
|
|
|
/* check whether a different CPU already registered this
|
|
|
|
* CPU because it is in the same boat. */
|
|
|
|
policy = cpufreq_cpu_get(cpu);
|
|
|
|
if (unlikely(policy)) {
|
2006-03-05 03:37:23 -05:00
|
|
|
cpufreq_cpu_put(policy);
|
2005-04-16 18:20:36 -04:00
|
|
|
cpufreq_debug_enable_ratelimit();
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
|
|
|
if (!try_module_get(cpufreq_driver->owner)) {
|
|
|
|
ret = -EINVAL;
|
|
|
|
goto module_out;
|
|
|
|
}
|
|
|
|
|
2005-10-20 18:17:43 -04:00
|
|
|
policy = kzalloc(sizeof(struct cpufreq_policy), GFP_KERNEL);
|
2005-04-16 18:20:36 -04:00
|
|
|
if (!policy) {
|
|
|
|
ret = -ENOMEM;
|
|
|
|
goto nomem_out;
|
|
|
|
}
|
|
|
|
|
|
|
|
policy->cpu = cpu;
|
|
|
|
policy->cpus = cpumask_of_cpu(cpu);
|
|
|
|
|
2006-01-14 10:01:49 -05:00
|
|
|
mutex_init(&policy->lock);
|
|
|
|
mutex_lock(&policy->lock);
|
2005-04-16 18:20:36 -04:00
|
|
|
init_completion(&policy->kobj_unregister);
|
|
|
|
INIT_WORK(&policy->update, handle_update, (void *)(long)cpu);
|
|
|
|
|
|
|
|
/* call driver. From then on the cpufreq must be able
|
|
|
|
* to accept all calls to ->verify and ->setpolicy for this CPU
|
|
|
|
*/
|
|
|
|
ret = cpufreq_driver->init(policy);
|
|
|
|
if (ret) {
|
|
|
|
dprintk("initialization failed\n");
|
2006-01-18 16:40:54 -05:00
|
|
|
mutex_unlock(&policy->lock);
|
2005-04-16 18:20:36 -04:00
|
|
|
goto err_out;
|
|
|
|
}
|
|
|
|
|
2006-03-05 03:37:23 -05:00
|
|
|
#ifdef CONFIG_SMP
|
|
|
|
for_each_cpu_mask(j, policy->cpus) {
|
|
|
|
if (cpu == j)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
/* check for existing affected CPUs. They may not be aware
|
|
|
|
* of it due to CPU Hotplug.
|
|
|
|
*/
|
|
|
|
managed_policy = cpufreq_cpu_get(j);
|
|
|
|
if (unlikely(managed_policy)) {
|
|
|
|
spin_lock_irqsave(&cpufreq_driver_lock, flags);
|
|
|
|
managed_policy->cpus = policy->cpus;
|
|
|
|
cpufreq_cpu_data[cpu] = managed_policy;
|
|
|
|
spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
|
|
|
|
|
|
|
dprintk("CPU already managed, adding link\n");
|
|
|
|
sysfs_create_link(&sys_dev->kobj,
|
|
|
|
&managed_policy->kobj, "cpufreq");
|
|
|
|
|
|
|
|
cpufreq_debug_enable_ratelimit();
|
|
|
|
mutex_unlock(&policy->lock);
|
|
|
|
ret = 0;
|
|
|
|
goto err_out_driver_exit; /* call driver->exit() */
|
|
|
|
}
|
|
|
|
}
|
|
|
|
#endif
|
2005-04-16 18:20:36 -04:00
|
|
|
memcpy(&new_policy, policy, sizeof(struct cpufreq_policy));
|
|
|
|
|
|
|
|
/* prepare interface data */
|
|
|
|
policy->kobj.parent = &sys_dev->kobj;
|
|
|
|
policy->kobj.ktype = &ktype_cpufreq;
|
|
|
|
strlcpy(policy->kobj.name, "cpufreq", KOBJ_NAME_LEN);
|
|
|
|
|
|
|
|
ret = kobject_register(&policy->kobj);
|
2006-01-18 16:40:54 -05:00
|
|
|
if (ret) {
|
|
|
|
mutex_unlock(&policy->lock);
|
2005-08-25 16:14:06 -04:00
|
|
|
goto err_out_driver_exit;
|
2006-01-18 16:40:54 -05:00
|
|
|
}
|
2005-04-16 18:20:36 -04:00
|
|
|
/* set up files for this cpu device */
|
|
|
|
drv_attr = cpufreq_driver->attr;
|
|
|
|
while ((drv_attr) && (*drv_attr)) {
|
|
|
|
sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
|
|
|
|
drv_attr++;
|
|
|
|
}
|
|
|
|
if (cpufreq_driver->get)
|
|
|
|
sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
|
|
|
|
if (cpufreq_driver->target)
|
|
|
|
sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
|
|
|
|
|
|
|
|
spin_lock_irqsave(&cpufreq_driver_lock, flags);
|
|
|
|
for_each_cpu_mask(j, policy->cpus)
|
|
|
|
cpufreq_cpu_data[j] = policy;
|
|
|
|
spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
2006-03-05 03:37:23 -05:00
|
|
|
|
|
|
|
/* symlink affected CPUs */
|
|
|
|
for_each_cpu_mask(j, policy->cpus) {
|
|
|
|
if (j == cpu)
|
|
|
|
continue;
|
|
|
|
if (!cpu_online(j))
|
|
|
|
continue;
|
|
|
|
|
2006-03-29 01:40:04 -05:00
|
|
|
dprintk("CPU %u already managed, adding link\n", j);
|
2006-03-05 03:37:23 -05:00
|
|
|
cpufreq_cpu_get(cpu);
|
|
|
|
cpu_sys_dev = get_cpu_sysdev(j);
|
|
|
|
sysfs_create_link(&cpu_sys_dev->kobj, &policy->kobj,
|
|
|
|
"cpufreq");
|
|
|
|
}
|
|
|
|
|
2005-04-16 18:20:36 -04:00
|
|
|
policy->governor = NULL; /* to assure that the starting sequence is
|
|
|
|
* run in cpufreq_set_policy */
|
2006-01-14 10:01:49 -05:00
|
|
|
mutex_unlock(&policy->lock);
|
2006-03-29 01:48:37 -05:00
|
|
|
|
2005-04-16 18:20:36 -04:00
|
|
|
/* set default policy */
|
|
|
|
ret = cpufreq_set_policy(&new_policy);
|
|
|
|
if (ret) {
|
|
|
|
dprintk("setting policy failed\n");
|
|
|
|
goto err_out_unregister;
|
|
|
|
}
|
|
|
|
|
|
|
|
module_put(cpufreq_driver->owner);
|
|
|
|
dprintk("initialization complete\n");
|
|
|
|
cpufreq_debug_enable_ratelimit();
|
2006-03-29 01:48:37 -05:00
|
|
|
|
2005-04-16 18:20:36 -04:00
|
|
|
return 0;
|
|
|
|
|
|
|
|
|
|
|
|
err_out_unregister:
|
|
|
|
spin_lock_irqsave(&cpufreq_driver_lock, flags);
|
|
|
|
for_each_cpu_mask(j, policy->cpus)
|
|
|
|
cpufreq_cpu_data[j] = NULL;
|
|
|
|
spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
|
|
|
|
|
|
|
kobject_unregister(&policy->kobj);
|
|
|
|
wait_for_completion(&policy->kobj_unregister);
|
|
|
|
|
2005-08-25 16:14:06 -04:00
|
|
|
err_out_driver_exit:
|
|
|
|
if (cpufreq_driver->exit)
|
|
|
|
cpufreq_driver->exit(policy);
|
|
|
|
|
2005-04-16 18:20:36 -04:00
|
|
|
err_out:
|
|
|
|
kfree(policy);
|
|
|
|
|
|
|
|
nomem_out:
|
|
|
|
module_put(cpufreq_driver->owner);
|
2005-10-30 17:59:54 -05:00
|
|
|
module_out:
|
2005-04-16 18:20:36 -04:00
|
|
|
cpufreq_debug_enable_ratelimit();
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
* cpufreq_remove_dev - remove a CPU device
|
|
|
|
*
|
|
|
|
* Removes the cpufreq interface for a CPU device.
|
|
|
|
*/
|
|
|
|
static int cpufreq_remove_dev (struct sys_device * sys_dev)
|
|
|
|
{
|
|
|
|
unsigned int cpu = sys_dev->id;
|
|
|
|
unsigned long flags;
|
|
|
|
struct cpufreq_policy *data;
|
|
|
|
#ifdef CONFIG_SMP
|
2005-11-22 00:32:28 -05:00
|
|
|
struct sys_device *cpu_sys_dev;
|
2005-04-16 18:20:36 -04:00
|
|
|
unsigned int j;
|
|
|
|
#endif
|
|
|
|
|
|
|
|
cpufreq_debug_disable_ratelimit();
|
|
|
|
dprintk("unregistering CPU %u\n", cpu);
|
|
|
|
|
|
|
|
spin_lock_irqsave(&cpufreq_driver_lock, flags);
|
|
|
|
data = cpufreq_cpu_data[cpu];
|
|
|
|
|
|
|
|
if (!data) {
|
|
|
|
spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
|
|
|
cpufreq_debug_enable_ratelimit();
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
cpufreq_cpu_data[cpu] = NULL;
|
|
|
|
|
|
|
|
|
|
|
|
#ifdef CONFIG_SMP
|
|
|
|
/* if this isn't the CPU which is the parent of the kobj, we
|
2006-02-28 00:43:23 -05:00
|
|
|
* only need to unlink, put and exit
|
2005-04-16 18:20:36 -04:00
|
|
|
*/
|
|
|
|
if (unlikely(cpu != data->cpu)) {
|
|
|
|
dprintk("removing link\n");
|
2006-03-05 03:37:23 -05:00
|
|
|
cpu_clear(cpu, data->cpus);
|
2005-04-16 18:20:36 -04:00
|
|
|
spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
|
|
|
sysfs_remove_link(&sys_dev->kobj, "cpufreq");
|
|
|
|
cpufreq_cpu_put(data);
|
|
|
|
cpufreq_debug_enable_ratelimit();
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
|
|
|
|
|
|
|
if (!kobject_get(&data->kobj)) {
|
|
|
|
spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
|
|
|
cpufreq_debug_enable_ratelimit();
|
2006-02-28 00:43:23 -05:00
|
|
|
return -EFAULT;
|
2005-04-16 18:20:36 -04:00
|
|
|
}
|
|
|
|
|
|
|
|
#ifdef CONFIG_SMP
|
|
|
|
/* if we have other CPUs still registered, we need to unlink them,
|
|
|
|
* or else wait_for_completion below will lock up. Clean the
|
|
|
|
* cpufreq_cpu_data[] while holding the lock, and remove the sysfs
|
|
|
|
* links afterwards.
|
|
|
|
*/
|
|
|
|
if (unlikely(cpus_weight(data->cpus) > 1)) {
|
|
|
|
for_each_cpu_mask(j, data->cpus) {
|
|
|
|
if (j == cpu)
|
|
|
|
continue;
|
|
|
|
cpufreq_cpu_data[j] = NULL;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
|
|
|
|
|
|
|
if (unlikely(cpus_weight(data->cpus) > 1)) {
|
|
|
|
for_each_cpu_mask(j, data->cpus) {
|
|
|
|
if (j == cpu)
|
|
|
|
continue;
|
|
|
|
dprintk("removing link for cpu %u\n", j);
|
2005-10-30 17:59:52 -05:00
|
|
|
cpu_sys_dev = get_cpu_sysdev(j);
|
|
|
|
sysfs_remove_link(&cpu_sys_dev->kobj, "cpufreq");
|
2005-04-16 18:20:36 -04:00
|
|
|
cpufreq_cpu_put(data);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
#else
|
|
|
|
spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
|
|
|
#endif
|
|
|
|
|
2006-01-14 10:01:49 -05:00
|
|
|
mutex_lock(&data->lock);
|
2005-04-16 18:20:36 -04:00
|
|
|
if (cpufreq_driver->target)
|
|
|
|
__cpufreq_governor(data, CPUFREQ_GOV_STOP);
|
2006-01-14 10:01:49 -05:00
|
|
|
mutex_unlock(&data->lock);
|
2005-04-16 18:20:36 -04:00
|
|
|
|
|
|
|
kobject_unregister(&data->kobj);
|
|
|
|
|
|
|
|
kobject_put(&data->kobj);
|
|
|
|
|
|
|
|
/* we need to make sure that the underlying kobj is actually
|
2006-02-28 00:43:23 -05:00
|
|
|
* not referenced anymore by anybody before we proceed with
|
2005-04-16 18:20:36 -04:00
|
|
|
* unloading.
|
|
|
|
*/
|
|
|
|
dprintk("waiting for dropping of refcount\n");
|
|
|
|
wait_for_completion(&data->kobj_unregister);
|
|
|
|
dprintk("wait complete\n");
|
|
|
|
|
|
|
|
if (cpufreq_driver->exit)
|
|
|
|
cpufreq_driver->exit(data);
|
|
|
|
|
|
|
|
kfree(data);
|
|
|
|
|
|
|
|
cpufreq_debug_enable_ratelimit();
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void handle_update(void *data)
|
|
|
|
{
|
|
|
|
unsigned int cpu = (unsigned int)(long)data;
|
|
|
|
dprintk("handle_update for cpu %u called\n", cpu);
|
|
|
|
cpufreq_update_policy(cpu);
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* cpufreq_out_of_sync - If actual and saved CPU frequency differs, we're in deep trouble.
|
|
|
|
* @cpu: cpu number
|
|
|
|
* @old_freq: CPU frequency the kernel thinks the CPU runs at
|
|
|
|
* @new_freq: CPU frequency the CPU actually runs at
|
|
|
|
*
|
|
|
|
* We adjust to current frequency first, and need to clean up later. So either call
|
|
|
|
* to cpufreq_update_policy() or schedule handle_update()).
|
|
|
|
*/
|
|
|
|
static void cpufreq_out_of_sync(unsigned int cpu, unsigned int old_freq, unsigned int new_freq)
|
|
|
|
{
|
|
|
|
struct cpufreq_freqs freqs;
|
|
|
|
|
2006-04-28 07:47:13 -04:00
|
|
|
dprintk("Warning: CPU frequency out of sync: cpufreq and timing "
|
2005-04-16 18:20:36 -04:00
|
|
|
"core thinks of %u, is %u kHz.\n", old_freq, new_freq);
|
|
|
|
|
|
|
|
freqs.cpu = cpu;
|
|
|
|
freqs.old = old_freq;
|
|
|
|
freqs.new = new_freq;
|
|
|
|
cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
|
|
|
|
cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2006-02-28 00:43:23 -05:00
|
|
|
/**
|
2005-12-02 13:43:20 -05:00
|
|
|
* cpufreq_quick_get - get the CPU frequency (in kHz) frpm policy->cur
|
|
|
|
* @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)
|
|
|
|
{
|
|
|
|
struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
|
|
|
|
unsigned int ret = 0;
|
|
|
|
|
|
|
|
if (policy) {
|
2006-01-14 10:01:49 -05:00
|
|
|
mutex_lock(&policy->lock);
|
2005-12-02 13:43:20 -05:00
|
|
|
ret = policy->cur;
|
2006-01-14 10:01:49 -05:00
|
|
|
mutex_unlock(&policy->lock);
|
2005-12-02 13:43:20 -05:00
|
|
|
cpufreq_cpu_put(policy);
|
|
|
|
}
|
|
|
|
|
|
|
|
return (ret);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(cpufreq_quick_get);
|
|
|
|
|
|
|
|
|
2006-02-28 00:43:23 -05:00
|
|
|
/**
|
2005-04-16 18:20:36 -04:00
|
|
|
* 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)
|
|
|
|
{
|
|
|
|
struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
|
|
|
|
unsigned int ret = 0;
|
|
|
|
|
|
|
|
if (!policy)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
if (!cpufreq_driver->get)
|
|
|
|
goto out;
|
|
|
|
|
2006-01-14 10:01:49 -05:00
|
|
|
mutex_lock(&policy->lock);
|
2005-04-16 18:20:36 -04:00
|
|
|
|
|
|
|
ret = cpufreq_driver->get(cpu);
|
|
|
|
|
2006-02-02 17:03:42 -05:00
|
|
|
if (ret && policy->cur && !(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
|
2005-04-16 18:20:36 -04:00
|
|
|
/* verify no discrepancy between actual and saved value exists */
|
|
|
|
if (unlikely(ret != policy->cur)) {
|
|
|
|
cpufreq_out_of_sync(cpu, policy->cur, ret);
|
|
|
|
schedule_work(&policy->update);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2006-01-14 10:01:49 -05:00
|
|
|
mutex_unlock(&policy->lock);
|
2005-04-16 18:20:36 -04:00
|
|
|
|
2006-02-02 17:03:42 -05:00
|
|
|
out:
|
2005-04-16 18:20:36 -04:00
|
|
|
cpufreq_cpu_put(policy);
|
|
|
|
|
|
|
|
return (ret);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(cpufreq_get);
|
|
|
|
|
|
|
|
|
2005-04-29 10:40:12 -04:00
|
|
|
/**
|
|
|
|
* cpufreq_suspend - let the low level driver prepare for suspend
|
|
|
|
*/
|
|
|
|
|
2005-07-07 20:56:42 -04:00
|
|
|
static int cpufreq_suspend(struct sys_device * sysdev, pm_message_t pmsg)
|
2005-04-29 10:40:12 -04:00
|
|
|
{
|
|
|
|
int cpu = sysdev->id;
|
|
|
|
unsigned int ret = 0;
|
|
|
|
unsigned int cur_freq = 0;
|
|
|
|
struct cpufreq_policy *cpu_policy;
|
|
|
|
|
2006-09-26 23:02:34 -04:00
|
|
|
dprintk("suspending cpu %u\n", cpu);
|
2005-04-29 10:40:12 -04:00
|
|
|
|
|
|
|
if (!cpu_online(cpu))
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
/* we may be lax here as interrupts are off. Nonetheless
|
|
|
|
* we need to grab the correct cpu policy, as to check
|
|
|
|
* whether we really run on this CPU.
|
|
|
|
*/
|
|
|
|
|
|
|
|
cpu_policy = cpufreq_cpu_get(cpu);
|
|
|
|
if (!cpu_policy)
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
/* only handle each CPU group once */
|
|
|
|
if (unlikely(cpu_policy->cpu != cpu)) {
|
|
|
|
cpufreq_cpu_put(cpu_policy);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (cpufreq_driver->suspend) {
|
2005-07-07 20:56:42 -04:00
|
|
|
ret = cpufreq_driver->suspend(cpu_policy, pmsg);
|
2005-04-29 10:40:12 -04:00
|
|
|
if (ret) {
|
|
|
|
printk(KERN_ERR "cpufreq: suspend failed in ->suspend "
|
|
|
|
"step on CPU %u\n", cpu_policy->cpu);
|
|
|
|
cpufreq_cpu_put(cpu_policy);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
if (cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
if (cpufreq_driver->get)
|
|
|
|
cur_freq = cpufreq_driver->get(cpu_policy->cpu);
|
|
|
|
|
|
|
|
if (!cur_freq || !cpu_policy->cur) {
|
|
|
|
printk(KERN_ERR "cpufreq: suspend failed to assert current "
|
|
|
|
"frequency is what timing core thinks it is.\n");
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (unlikely(cur_freq != cpu_policy->cur)) {
|
|
|
|
struct cpufreq_freqs freqs;
|
|
|
|
|
|
|
|
if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
|
2006-04-28 07:47:13 -04:00
|
|
|
dprintk("Warning: CPU frequency is %u, "
|
2005-04-29 10:40:12 -04:00
|
|
|
"cpufreq assumed %u kHz.\n",
|
|
|
|
cur_freq, cpu_policy->cur);
|
|
|
|
|
|
|
|
freqs.cpu = cpu;
|
|
|
|
freqs.old = cpu_policy->cur;
|
|
|
|
freqs.new = cur_freq;
|
|
|
|
|
2006-10-04 05:17:06 -04:00
|
|
|
srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
|
2005-04-29 10:40:12 -04:00
|
|
|
CPUFREQ_SUSPENDCHANGE, &freqs);
|
|
|
|
adjust_jiffies(CPUFREQ_SUSPENDCHANGE, &freqs);
|
|
|
|
|
|
|
|
cpu_policy->cur = cur_freq;
|
|
|
|
}
|
|
|
|
|
2006-02-02 17:03:42 -05:00
|
|
|
out:
|
2005-04-29 10:40:12 -04:00
|
|
|
cpufreq_cpu_put(cpu_policy);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2005-04-16 18:20:36 -04:00
|
|
|
/**
|
|
|
|
* cpufreq_resume - restore proper CPU frequency handling after resume
|
|
|
|
*
|
|
|
|
* 1.) resume CPUfreq hardware support (cpufreq_driver->resume())
|
|
|
|
* 2.) if ->target and !CPUFREQ_CONST_LOOPS: verify we're in sync
|
2005-04-29 10:40:12 -04:00
|
|
|
* 3.) schedule call cpufreq_update_policy() ASAP as interrupts are
|
|
|
|
* restored.
|
2005-04-16 18:20:36 -04:00
|
|
|
*/
|
|
|
|
static int cpufreq_resume(struct sys_device * sysdev)
|
|
|
|
{
|
|
|
|
int cpu = sysdev->id;
|
|
|
|
unsigned int ret = 0;
|
|
|
|
struct cpufreq_policy *cpu_policy;
|
|
|
|
|
|
|
|
dprintk("resuming cpu %u\n", cpu);
|
|
|
|
|
|
|
|
if (!cpu_online(cpu))
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
/* we may be lax here as interrupts are off. Nonetheless
|
|
|
|
* we need to grab the correct cpu policy, as to check
|
|
|
|
* whether we really run on this CPU.
|
|
|
|
*/
|
|
|
|
|
|
|
|
cpu_policy = cpufreq_cpu_get(cpu);
|
|
|
|
if (!cpu_policy)
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
/* only handle each CPU group once */
|
|
|
|
if (unlikely(cpu_policy->cpu != cpu)) {
|
|
|
|
cpufreq_cpu_put(cpu_policy);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (cpufreq_driver->resume) {
|
|
|
|
ret = cpufreq_driver->resume(cpu_policy);
|
|
|
|
if (ret) {
|
|
|
|
printk(KERN_ERR "cpufreq: resume failed in ->resume "
|
|
|
|
"step on CPU %u\n", cpu_policy->cpu);
|
|
|
|
cpufreq_cpu_put(cpu_policy);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
|
|
|
|
unsigned int cur_freq = 0;
|
|
|
|
|
|
|
|
if (cpufreq_driver->get)
|
|
|
|
cur_freq = cpufreq_driver->get(cpu_policy->cpu);
|
|
|
|
|
|
|
|
if (!cur_freq || !cpu_policy->cur) {
|
2005-04-29 10:40:12 -04:00
|
|
|
printk(KERN_ERR "cpufreq: resume failed to assert "
|
|
|
|
"current frequency is what timing core "
|
|
|
|
"thinks it is.\n");
|
2005-04-16 18:20:36 -04:00
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (unlikely(cur_freq != cpu_policy->cur)) {
|
|
|
|
struct cpufreq_freqs freqs;
|
|
|
|
|
2005-05-02 02:25:10 -04:00
|
|
|
if (!(cpufreq_driver->flags & CPUFREQ_PM_NO_WARN))
|
2006-04-28 07:47:13 -04:00
|
|
|
dprintk("Warning: CPU frequency"
|
2005-05-02 02:25:10 -04:00
|
|
|
"is %u, cpufreq assumed %u kHz.\n",
|
|
|
|
cur_freq, cpu_policy->cur);
|
2005-04-16 18:20:36 -04:00
|
|
|
|
|
|
|
freqs.cpu = cpu;
|
|
|
|
freqs.old = cpu_policy->cur;
|
|
|
|
freqs.new = cur_freq;
|
|
|
|
|
2006-10-04 05:17:06 -04:00
|
|
|
srcu_notifier_call_chain(
|
[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 04:16:30 -05:00
|
|
|
&cpufreq_transition_notifier_list,
|
2005-04-29 10:40:12 -04:00
|
|
|
CPUFREQ_RESUMECHANGE, &freqs);
|
2005-04-16 18:20:36 -04:00
|
|
|
adjust_jiffies(CPUFREQ_RESUMECHANGE, &freqs);
|
|
|
|
|
|
|
|
cpu_policy->cur = cur_freq;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
out:
|
|
|
|
schedule_work(&cpu_policy->update);
|
|
|
|
cpufreq_cpu_put(cpu_policy);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct sysdev_driver cpufreq_sysdev_driver = {
|
|
|
|
.add = cpufreq_add_dev,
|
|
|
|
.remove = cpufreq_remove_dev,
|
2005-04-29 10:40:12 -04:00
|
|
|
.suspend = cpufreq_suspend,
|
2005-04-16 18:20:36 -04:00
|
|
|
.resume = cpufreq_resume,
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
|
|
/*********************************************************************
|
|
|
|
* 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 00:43:23 -05:00
|
|
|
* Add a driver to one of two lists: either a list of drivers that
|
2005-04-16 18:20:36 -04:00
|
|
|
* 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 04:16:30 -05:00
|
|
|
* blocking_notifier_chain_register.
|
2005-04-16 18:20:36 -04:00
|
|
|
*/
|
|
|
|
int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
|
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
switch (list) {
|
|
|
|
case CPUFREQ_TRANSITION_NOTIFIER:
|
2006-10-04 05:17:06 -04:00
|
|
|
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 04:16:30 -05:00
|
|
|
&cpufreq_transition_notifier_list, nb);
|
2005-04-16 18:20:36 -04:00
|
|
|
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 04:16:30 -05:00
|
|
|
ret = blocking_notifier_chain_register(
|
|
|
|
&cpufreq_policy_notifier_list, nb);
|
2005-04-16 18:20:36 -04:00
|
|
|
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
|
|
|
|
* @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER
|
|
|
|
*
|
|
|
|
* 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 04:16:30 -05:00
|
|
|
* blocking_notifier_chain_unregister.
|
2005-04-16 18:20:36 -04:00
|
|
|
*/
|
|
|
|
int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
|
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
switch (list) {
|
|
|
|
case CPUFREQ_TRANSITION_NOTIFIER:
|
2006-10-04 05:17:06 -04:00
|
|
|
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 04:16:30 -05:00
|
|
|
&cpufreq_transition_notifier_list, nb);
|
2005-04-16 18:20:36 -04:00
|
|
|
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 04:16:30 -05:00
|
|
|
ret = blocking_notifier_chain_unregister(
|
|
|
|
&cpufreq_policy_notifier_list, nb);
|
2005-04-16 18:20:36 -04:00
|
|
|
break;
|
|
|
|
default:
|
|
|
|
ret = -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(cpufreq_unregister_notifier);
|
|
|
|
|
|
|
|
|
|
|
|
/*********************************************************************
|
|
|
|
* GOVERNORS *
|
|
|
|
*********************************************************************/
|
|
|
|
|
|
|
|
|
2006-07-26 09:40:07 -04:00
|
|
|
/* Must be called with lock_cpu_hotplug held */
|
2005-04-16 18:20:36 -04:00
|
|
|
int __cpufreq_driver_target(struct cpufreq_policy *policy,
|
|
|
|
unsigned int target_freq,
|
|
|
|
unsigned int relation)
|
|
|
|
{
|
|
|
|
int retval = -EINVAL;
|
2005-10-30 17:59:54 -05:00
|
|
|
|
2005-04-16 18:20:36 -04:00
|
|
|
dprintk("target for CPU %u: %u kHz, relation %u\n", policy->cpu,
|
|
|
|
target_freq, relation);
|
|
|
|
if (cpu_online(policy->cpu) && cpufreq_driver->target)
|
|
|
|
retval = cpufreq_driver->target(policy, target_freq, relation);
|
2005-11-09 00:34:24 -05:00
|
|
|
|
2005-04-16 18:20:36 -04:00
|
|
|
return retval;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
|
|
|
|
|
|
|
|
int cpufreq_driver_target(struct cpufreq_policy *policy,
|
|
|
|
unsigned int target_freq,
|
|
|
|
unsigned int relation)
|
|
|
|
{
|
2005-07-28 12:43:56 -04:00
|
|
|
int ret;
|
2005-04-16 18:20:36 -04:00
|
|
|
|
|
|
|
policy = cpufreq_cpu_get(policy->cpu);
|
|
|
|
if (!policy)
|
|
|
|
return -EINVAL;
|
|
|
|
|
2006-07-26 09:40:07 -04:00
|
|
|
lock_cpu_hotplug();
|
2006-01-14 10:01:49 -05:00
|
|
|
mutex_lock(&policy->lock);
|
2005-04-16 18:20:36 -04:00
|
|
|
|
|
|
|
ret = __cpufreq_driver_target(policy, target_freq, relation);
|
|
|
|
|
2006-01-14 10:01:49 -05:00
|
|
|
mutex_unlock(&policy->lock);
|
2006-07-26 09:40:07 -04:00
|
|
|
unlock_cpu_hotplug();
|
2005-04-16 18:20:36 -04:00
|
|
|
|
|
|
|
cpufreq_cpu_put(policy);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_driver_target);
|
|
|
|
|
2006-07-26 09:40:07 -04:00
|
|
|
/*
|
|
|
|
* Locking: Must be called with the lock_cpu_hotplug() lock held
|
|
|
|
* when "event" is CPUFREQ_GOV_LIMITS
|
|
|
|
*/
|
2005-04-16 18:20:36 -04:00
|
|
|
|
|
|
|
static int __cpufreq_governor(struct cpufreq_policy *policy, unsigned int event)
|
|
|
|
{
|
2005-07-28 12:43:56 -04:00
|
|
|
int ret;
|
2005-04-16 18:20:36 -04:00
|
|
|
|
|
|
|
if (!try_module_get(policy->governor->owner))
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
dprintk("__cpufreq_governor for CPU %u, event %u\n", policy->cpu, event);
|
|
|
|
ret = policy->governor->governor(policy, event);
|
|
|
|
|
|
|
|
/* we keep one module reference alive for each CPU governed by this CPU */
|
|
|
|
if ((event != CPUFREQ_GOV_START) || ret)
|
|
|
|
module_put(policy->governor->owner);
|
|
|
|
if ((event == CPUFREQ_GOV_STOP) && !ret)
|
|
|
|
module_put(policy->governor->owner);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
int cpufreq_register_governor(struct cpufreq_governor *governor)
|
|
|
|
{
|
2006-07-06 15:30:26 -04:00
|
|
|
int err;
|
2005-04-16 18:20:36 -04:00
|
|
|
|
|
|
|
if (!governor)
|
|
|
|
return -EINVAL;
|
|
|
|
|
2006-01-13 18:54:22 -05:00
|
|
|
mutex_lock(&cpufreq_governor_mutex);
|
2006-02-28 00:43:23 -05:00
|
|
|
|
2006-07-06 15:30:26 -04:00
|
|
|
err = -EBUSY;
|
|
|
|
if (__find_governor(governor->name) == NULL) {
|
|
|
|
err = 0;
|
|
|
|
list_add(&governor->governor_list, &cpufreq_governor_list);
|
2005-04-16 18:20:36 -04:00
|
|
|
}
|
|
|
|
|
2006-02-28 00:43:23 -05:00
|
|
|
mutex_unlock(&cpufreq_governor_mutex);
|
2006-07-06 15:30:26 -04:00
|
|
|
return err;
|
2005-04-16 18:20:36 -04:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_register_governor);
|
|
|
|
|
|
|
|
|
|
|
|
void cpufreq_unregister_governor(struct cpufreq_governor *governor)
|
|
|
|
{
|
|
|
|
if (!governor)
|
|
|
|
return;
|
|
|
|
|
2006-01-13 18:54:22 -05:00
|
|
|
mutex_lock(&cpufreq_governor_mutex);
|
2005-04-16 18:20:36 -04:00
|
|
|
list_del(&governor->governor_list);
|
2006-01-13 18:54:22 -05:00
|
|
|
mutex_unlock(&cpufreq_governor_mutex);
|
2005-04-16 18:20:36 -04:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
/*********************************************************************
|
|
|
|
* POLICY INTERFACE *
|
|
|
|
*********************************************************************/
|
|
|
|
|
|
|
|
/**
|
|
|
|
* cpufreq_get_policy - get the current cpufreq_policy
|
|
|
|
* @policy: struct cpufreq_policy into which the current cpufreq_policy is written
|
|
|
|
*
|
|
|
|
* 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;
|
|
|
|
|
2006-01-14 10:01:49 -05:00
|
|
|
mutex_lock(&cpu_policy->lock);
|
2005-04-16 18:20:36 -04:00
|
|
|
memcpy(policy, cpu_policy, sizeof(struct cpufreq_policy));
|
2006-01-14 10:01:49 -05:00
|
|
|
mutex_unlock(&cpu_policy->lock);
|
2005-04-16 18:20:36 -04:00
|
|
|
|
|
|
|
cpufreq_cpu_put(cpu_policy);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(cpufreq_get_policy);
|
|
|
|
|
|
|
|
|
2006-07-26 09:40:07 -04:00
|
|
|
/*
|
|
|
|
* Locking: Must be called with the lock_cpu_hotplug() lock held
|
|
|
|
*/
|
2005-04-16 18:20:36 -04:00
|
|
|
static int __cpufreq_set_policy(struct cpufreq_policy *data, struct cpufreq_policy *policy)
|
|
|
|
{
|
|
|
|
int ret = 0;
|
|
|
|
|
|
|
|
cpufreq_debug_disable_ratelimit();
|
|
|
|
dprintk("setting new policy for CPU %u: %u - %u kHz\n", policy->cpu,
|
|
|
|
policy->min, policy->max);
|
|
|
|
|
2006-02-02 17:03:42 -05:00
|
|
|
memcpy(&policy->cpuinfo, &data->cpuinfo, sizeof(struct cpufreq_cpuinfo));
|
2005-04-16 18:20:36 -04:00
|
|
|
|
2006-07-05 17:12:20 -04:00
|
|
|
if (policy->min > data->min && policy->min > policy->max) {
|
|
|
|
ret = -EINVAL;
|
|
|
|
goto error_out;
|
|
|
|
}
|
|
|
|
|
2005-04-16 18:20:36 -04:00
|
|
|
/* verify the cpu speed can be set within this limit */
|
|
|
|
ret = cpufreq_driver->verify(policy);
|
|
|
|
if (ret)
|
|
|
|
goto error_out;
|
|
|
|
|
|
|
|
/* 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 04:16:30 -05:00
|
|
|
blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
|
|
|
|
CPUFREQ_ADJUST, policy);
|
2005-04-16 18:20:36 -04:00
|
|
|
|
|
|
|
/* 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 04:16:30 -05:00
|
|
|
blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
|
|
|
|
CPUFREQ_INCOMPATIBLE, policy);
|
2005-04-16 18:20:36 -04:00
|
|
|
|
|
|
|
/* verify the cpu speed can be set within this limit,
|
|
|
|
which might be different to the first one */
|
|
|
|
ret = cpufreq_driver->verify(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 04:16:30 -05:00
|
|
|
if (ret)
|
2005-04-16 18:20:36 -04:00
|
|
|
goto error_out;
|
|
|
|
|
|
|
|
/* 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 04:16:30 -05:00
|
|
|
blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
|
|
|
|
CPUFREQ_NOTIFY, policy);
|
2005-04-16 18:20:36 -04:00
|
|
|
|
2006-02-02 17:03:42 -05:00
|
|
|
data->min = policy->min;
|
|
|
|
data->max = policy->max;
|
2005-04-16 18:20:36 -04:00
|
|
|
|
|
|
|
dprintk("new min and max freqs are %u - %u kHz\n", data->min, data->max);
|
|
|
|
|
|
|
|
if (cpufreq_driver->setpolicy) {
|
|
|
|
data->policy = policy->policy;
|
|
|
|
dprintk("setting range\n");
|
|
|
|
ret = cpufreq_driver->setpolicy(policy);
|
|
|
|
} else {
|
|
|
|
if (policy->governor != data->governor) {
|
|
|
|
/* save old, working values */
|
|
|
|
struct cpufreq_governor *old_gov = data->governor;
|
|
|
|
|
|
|
|
dprintk("governor switch\n");
|
|
|
|
|
|
|
|
/* end old governor */
|
|
|
|
if (data->governor)
|
|
|
|
__cpufreq_governor(data, CPUFREQ_GOV_STOP);
|
|
|
|
|
|
|
|
/* start new governor */
|
|
|
|
data->governor = policy->governor;
|
|
|
|
if (__cpufreq_governor(data, CPUFREQ_GOV_START)) {
|
|
|
|
/* new governor failed, so re-start old one */
|
|
|
|
dprintk("starting governor %s failed\n", data->governor->name);
|
|
|
|
if (old_gov) {
|
|
|
|
data->governor = old_gov;
|
|
|
|
__cpufreq_governor(data, CPUFREQ_GOV_START);
|
|
|
|
}
|
|
|
|
ret = -EINVAL;
|
|
|
|
goto error_out;
|
|
|
|
}
|
|
|
|
/* might be a policy change, too, so fall through */
|
|
|
|
}
|
|
|
|
dprintk("governor: change or update limits\n");
|
|
|
|
__cpufreq_governor(data, CPUFREQ_GOV_LIMITS);
|
|
|
|
}
|
|
|
|
|
2006-02-02 17:03:42 -05:00
|
|
|
error_out:
|
2005-04-16 18:20:36 -04:00
|
|
|
cpufreq_debug_enable_ratelimit();
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* cpufreq_set_policy - set a new CPUFreq policy
|
|
|
|
* @policy: policy to be set.
|
|
|
|
*
|
|
|
|
* Sets a new CPU frequency and voltage scaling policy.
|
|
|
|
*/
|
|
|
|
int cpufreq_set_policy(struct cpufreq_policy *policy)
|
|
|
|
{
|
|
|
|
int ret = 0;
|
|
|
|
struct cpufreq_policy *data;
|
|
|
|
|
|
|
|
if (!policy)
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
data = cpufreq_cpu_get(policy->cpu);
|
|
|
|
if (!data)
|
|
|
|
return -EINVAL;
|
|
|
|
|
2006-07-26 09:40:07 -04:00
|
|
|
lock_cpu_hotplug();
|
|
|
|
|
2005-04-16 18:20:36 -04:00
|
|
|
/* lock this CPU */
|
2006-01-14 10:01:49 -05:00
|
|
|
mutex_lock(&data->lock);
|
2005-04-16 18:20:36 -04:00
|
|
|
|
|
|
|
ret = __cpufreq_set_policy(data, policy);
|
|
|
|
data->user_policy.min = data->min;
|
|
|
|
data->user_policy.max = data->max;
|
|
|
|
data->user_policy.policy = data->policy;
|
|
|
|
data->user_policy.governor = data->governor;
|
|
|
|
|
2006-01-14 10:01:49 -05:00
|
|
|
mutex_unlock(&data->lock);
|
2006-07-26 09:40:07 -04:00
|
|
|
|
|
|
|
unlock_cpu_hotplug();
|
2005-04-16 18:20:36 -04:00
|
|
|
cpufreq_cpu_put(data);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(cpufreq_set_policy);
|
|
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
* cpufreq_update_policy - re-evaluate an existing cpufreq policy
|
|
|
|
* @cpu: CPU which shall be re-evaluated
|
|
|
|
*
|
|
|
|
* Usefull for policy notifiers which have different necessities
|
|
|
|
* at different times.
|
|
|
|
*/
|
|
|
|
int cpufreq_update_policy(unsigned int cpu)
|
|
|
|
{
|
|
|
|
struct cpufreq_policy *data = cpufreq_cpu_get(cpu);
|
|
|
|
struct cpufreq_policy policy;
|
|
|
|
int ret = 0;
|
|
|
|
|
|
|
|
if (!data)
|
|
|
|
return -ENODEV;
|
|
|
|
|
2006-07-26 09:40:07 -04:00
|
|
|
lock_cpu_hotplug();
|
2006-01-14 10:01:49 -05:00
|
|
|
mutex_lock(&data->lock);
|
2005-04-16 18:20:36 -04:00
|
|
|
|
|
|
|
dprintk("updating policy for CPU %u\n", cpu);
|
2006-02-02 17:03:42 -05:00
|
|
|
memcpy(&policy, data, sizeof(struct cpufreq_policy));
|
2005-04-16 18:20:36 -04:00
|
|
|
policy.min = data->user_policy.min;
|
|
|
|
policy.max = data->user_policy.max;
|
|
|
|
policy.policy = data->user_policy.policy;
|
|
|
|
policy.governor = data->user_policy.governor;
|
|
|
|
|
2006-01-26 12:46:33 -05:00
|
|
|
/* BIOS might change freq behind our back
|
|
|
|
-> ask driver for current freq and notify governors about a change */
|
|
|
|
if (cpufreq_driver->get) {
|
|
|
|
policy.cur = cpufreq_driver->get(cpu);
|
2006-02-01 05:36:04 -05:00
|
|
|
if (!data->cur) {
|
|
|
|
dprintk("Driver did not initialize current freq");
|
|
|
|
data->cur = policy.cur;
|
|
|
|
} else {
|
|
|
|
if (data->cur != policy.cur)
|
|
|
|
cpufreq_out_of_sync(cpu, data->cur, policy.cur);
|
|
|
|
}
|
2006-01-26 12:46:33 -05:00
|
|
|
}
|
|
|
|
|
2005-04-16 18:20:36 -04:00
|
|
|
ret = __cpufreq_set_policy(data, &policy);
|
|
|
|
|
2006-01-14 10:01:49 -05:00
|
|
|
mutex_unlock(&data->lock);
|
2006-07-26 09:40:07 -04:00
|
|
|
unlock_cpu_hotplug();
|
2005-04-16 18:20:36 -04:00
|
|
|
cpufreq_cpu_put(data);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(cpufreq_update_policy);
|
|
|
|
|
2006-06-27 05:54:08 -04:00
|
|
|
#ifdef CONFIG_HOTPLUG_CPU
|
|
|
|
static int cpufreq_cpu_callback(struct notifier_block *nfb,
|
2005-10-30 17:59:54 -05:00
|
|
|
unsigned long action, void *hcpu)
|
|
|
|
{
|
|
|
|
unsigned int cpu = (unsigned long)hcpu;
|
|
|
|
struct cpufreq_policy *policy;
|
|
|
|
struct sys_device *sys_dev;
|
|
|
|
|
|
|
|
sys_dev = get_cpu_sysdev(cpu);
|
|
|
|
|
|
|
|
if (sys_dev) {
|
|
|
|
switch (action) {
|
|
|
|
case CPU_ONLINE:
|
|
|
|
cpufreq_add_dev(sys_dev);
|
|
|
|
break;
|
|
|
|
case CPU_DOWN_PREPARE:
|
|
|
|
/*
|
|
|
|
* We attempt to put this cpu in lowest frequency
|
|
|
|
* possible before going down. This will permit
|
|
|
|
* hardware-managed P-State to switch other related
|
|
|
|
* threads to min or higher speeds if possible.
|
|
|
|
*/
|
|
|
|
policy = cpufreq_cpu_data[cpu];
|
|
|
|
if (policy) {
|
|
|
|
cpufreq_driver_target(policy, policy->min,
|
|
|
|
CPUFREQ_RELATION_H);
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
case CPU_DEAD:
|
|
|
|
cpufreq_remove_dev(sys_dev);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return NOTIFY_OK;
|
|
|
|
}
|
|
|
|
|
2006-06-27 05:54:09 -04:00
|
|
|
static struct notifier_block __cpuinitdata cpufreq_cpu_notifier =
|
2005-10-30 17:59:54 -05:00
|
|
|
{
|
|
|
|
.notifier_call = cpufreq_cpu_callback,
|
|
|
|
};
|
2006-06-27 05:54:08 -04:00
|
|
|
#endif /* CONFIG_HOTPLUG_CPU */
|
2005-04-16 18:20:36 -04:00
|
|
|
|
|
|
|
/*********************************************************************
|
|
|
|
* 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.
|
|
|
|
*
|
2006-02-28 00:43:23 -05:00
|
|
|
* Registers a CPU Frequency driver to this core code. This code
|
2005-04-16 18:20:36 -04:00
|
|
|
* returns zero on success, -EBUSY when another driver got here first
|
2006-02-28 00:43:23 -05:00
|
|
|
* (and isn't unregistered in the meantime).
|
2005-04-16 18:20:36 -04:00
|
|
|
*
|
|
|
|
*/
|
|
|
|
int cpufreq_register_driver(struct cpufreq_driver *driver_data)
|
|
|
|
{
|
|
|
|
unsigned long flags;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
if (!driver_data || !driver_data->verify || !driver_data->init ||
|
|
|
|
((!driver_data->setpolicy) && (!driver_data->target)))
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
dprintk("trying to register driver %s\n", driver_data->name);
|
|
|
|
|
|
|
|
if (driver_data->setpolicy)
|
|
|
|
driver_data->flags |= CPUFREQ_CONST_LOOPS;
|
|
|
|
|
|
|
|
spin_lock_irqsave(&cpufreq_driver_lock, flags);
|
|
|
|
if (cpufreq_driver) {
|
|
|
|
spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
|
|
|
return -EBUSY;
|
|
|
|
}
|
|
|
|
cpufreq_driver = driver_data;
|
|
|
|
spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
|
|
|
|
|
|
|
ret = sysdev_driver_register(&cpu_sysdev_class,&cpufreq_sysdev_driver);
|
|
|
|
|
|
|
|
if ((!ret) && !(cpufreq_driver->flags & CPUFREQ_STICKY)) {
|
|
|
|
int i;
|
|
|
|
ret = -ENODEV;
|
|
|
|
|
|
|
|
/* check for at least one working CPU */
|
|
|
|
for (i=0; i<NR_CPUS; i++)
|
|
|
|
if (cpufreq_cpu_data[i])
|
|
|
|
ret = 0;
|
|
|
|
|
|
|
|
/* if all ->init() calls failed, unregister */
|
|
|
|
if (ret) {
|
|
|
|
dprintk("no CPU initialized for driver %s\n", driver_data->name);
|
|
|
|
sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
|
|
|
|
|
|
|
|
spin_lock_irqsave(&cpufreq_driver_lock, flags);
|
|
|
|
cpufreq_driver = NULL;
|
|
|
|
spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!ret) {
|
2006-06-27 05:54:08 -04:00
|
|
|
register_hotcpu_notifier(&cpufreq_cpu_notifier);
|
2005-04-16 18:20:36 -04:00
|
|
|
dprintk("driver %s up and running\n", driver_data->name);
|
|
|
|
cpufreq_debug_enable_ratelimit();
|
|
|
|
}
|
|
|
|
|
|
|
|
return (ret);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_register_driver);
|
|
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
* cpufreq_unregister_driver - unregister the current CPUFreq driver
|
|
|
|
*
|
2006-02-28 00:43:23 -05:00
|
|
|
* Unregister the current CPUFreq driver. Only call this if you have
|
2005-04-16 18:20:36 -04:00
|
|
|
* 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.
|
|
|
|
*/
|
|
|
|
int cpufreq_unregister_driver(struct cpufreq_driver *driver)
|
|
|
|
{
|
|
|
|
unsigned long flags;
|
|
|
|
|
|
|
|
cpufreq_debug_disable_ratelimit();
|
|
|
|
|
|
|
|
if (!cpufreq_driver || (driver != cpufreq_driver)) {
|
|
|
|
cpufreq_debug_enable_ratelimit();
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
dprintk("unregistering driver %s\n", driver->name);
|
|
|
|
|
|
|
|
sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
|
2006-06-27 05:54:08 -04:00
|
|
|
unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
|
2005-04-16 18:20:36 -04:00
|
|
|
|
|
|
|
spin_lock_irqsave(&cpufreq_driver_lock, flags);
|
|
|
|
cpufreq_driver = NULL;
|
|
|
|
spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
|