linux/drivers/cpufreq/cpufreq.c

2004 lines
50 KiB
C
Raw Normal View History

/*
* linux/drivers/cpufreq/cpufreq.c
*
* Copyright (C) 2001 Russell King
* (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
*
[PATCH] create and destroy cpufreq sysfs entries based on cpu notifiers cpufreq entries in sysfs should only be populated when CPU is online state. When we either boot with maxcpus=x and then boot the other cpus by echoing to sysfs online file, these entries should be created and destroyed when CPU_DEAD is notified. Same treatement as cache entries under sysfs. We place the processor in the lowest frequency, so hw managed P-State transitions can still work on the other threads to save power. Primary goal was to just make these directories appear/disapper dynamically. There is one in this patch i had to do, which i really dont like myself but probably best if someone handling the cpufreq infrastructure could give this code right treatment if this is not acceptable. I guess its probably good for the first cut. - Converting lock_cpu_hotplug()/unlock_cpu_hotplug() to disable/enable preempt. The locking was smack in the middle of the notification path, when the hotplug is already holding the lock. I tried another solution to avoid this so avoid taking locks if we know we are from notification path. The solution was getting very ugly and i decided this was probably good for this iteration until someone who understands cpufreq could do a better job than me. (akpm: export cpucontrol to GPL modules: drivers/cpufreq/cpufreq_stats.c now does lock_cpu_hotplug()) Signed-off-by: Ashok Raj <ashok.raj@intel.com> Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> Cc: Dave Jones <davej@codemonkey.org.uk> Cc: Zwane Mwaikambo <zwane@holomorphy.com> Cc: Greg KH <greg@kroah.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 22:59:54 +00:00
* Oct 2005 - Ashok Raj <ashok.raj@intel.com>
* Added handling for CPU hotplug
* Feb 2006 - Jacob Shin <jacob.shin@amd.com>
* Fix handling for CPU hotplug -- affected CPUs
[PATCH] create and destroy cpufreq sysfs entries based on cpu notifiers cpufreq entries in sysfs should only be populated when CPU is online state. When we either boot with maxcpus=x and then boot the other cpus by echoing to sysfs online file, these entries should be created and destroyed when CPU_DEAD is notified. Same treatement as cache entries under sysfs. We place the processor in the lowest frequency, so hw managed P-State transitions can still work on the other threads to save power. Primary goal was to just make these directories appear/disapper dynamically. There is one in this patch i had to do, which i really dont like myself but probably best if someone handling the cpufreq infrastructure could give this code right treatment if this is not acceptable. I guess its probably good for the first cut. - Converting lock_cpu_hotplug()/unlock_cpu_hotplug() to disable/enable preempt. The locking was smack in the middle of the notification path, when the hotplug is already holding the lock. I tried another solution to avoid this so avoid taking locks if we know we are from notification path. The solution was getting very ugly and i decided this was probably good for this iteration until someone who understands cpufreq could do a better job than me. (akpm: export cpucontrol to GPL modules: drivers/cpufreq/cpufreq_stats.c now does lock_cpu_hotplug()) Signed-off-by: Ashok Raj <ashok.raj@intel.com> Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> Cc: Dave Jones <davej@codemonkey.org.uk> Cc: Zwane Mwaikambo <zwane@holomorphy.com> Cc: Greg KH <greg@kroah.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 22:59:54 +00:00
*
* 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>
#include <linux/mutex.h>
#define dprintk(msg...) cpufreq_debug_printk(CPUFREQ_DEBUG_CORE, \
"cpufreq-core", msg)
/**
* The "cpufreq driver" - the arch- or hardware-dependent low
* level driver of CPUFreq support, and its spinlock. This lock
* also protects the cpufreq_cpu_data array.
*/
static struct cpufreq_driver *cpufreq_driver;
static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
#ifdef CONFIG_HOTPLUG_CPU
/* This one keeps track of the previously set governor of a removed CPU */
static DEFINE_PER_CPU(char[CPUFREQ_NAME_LEN], cpufreq_cpu_governor);
#endif
static DEFINE_SPINLOCK(cpufreq_driver_lock);
/*
* cpu_policy_rwsem is a per CPU reader-writer semaphore designed to cure
* all cpufreq/hotplug/workqueue/etc related lock issues.
*
* The rules for this semaphore:
* - Any routine that wants to read from the policy structure will
* do a down_read on this semaphore.
* - Any routine that will write to the policy structure and/or may take away
* the policy altogether (eg. CPU hotplug), will hold this lock in write
* mode before doing so.
*
* Additional rules:
* - All holders of the lock should check to make sure that the CPU they
* are concerned with are online after they get the lock.
* - Governor routines that can be called in cpufreq hotplug path should not
* take this sem as top level hotplug notifier handler takes this.
* - Lock should not be held across
* __cpufreq_governor(data, CPUFREQ_GOV_STOP);
*/
static DEFINE_PER_CPU(int, cpufreq_policy_cpu);
static DEFINE_PER_CPU(struct rw_semaphore, cpu_policy_rwsem);
#define lock_policy_rwsem(mode, cpu) \
int lock_policy_rwsem_##mode \
(int cpu) \
{ \
int policy_cpu = per_cpu(cpufreq_policy_cpu, cpu); \
BUG_ON(policy_cpu == -1); \
down_##mode(&per_cpu(cpu_policy_rwsem, policy_cpu)); \
if (unlikely(!cpu_online(cpu))) { \
up_##mode(&per_cpu(cpu_policy_rwsem, policy_cpu)); \
return -1; \
} \
\
return 0; \
}
lock_policy_rwsem(read, cpu);
EXPORT_SYMBOL_GPL(lock_policy_rwsem_read);
lock_policy_rwsem(write, cpu);
EXPORT_SYMBOL_GPL(lock_policy_rwsem_write);
void unlock_policy_rwsem_read(int cpu)
{
int policy_cpu = per_cpu(cpufreq_policy_cpu, cpu);
BUG_ON(policy_cpu == -1);
up_read(&per_cpu(cpu_policy_rwsem, policy_cpu));
}
EXPORT_SYMBOL_GPL(unlock_policy_rwsem_read);
void unlock_policy_rwsem_write(int cpu)
{
int policy_cpu = per_cpu(cpufreq_policy_cpu, cpu);
BUG_ON(policy_cpu == -1);
up_write(&per_cpu(cpu_policy_rwsem, policy_cpu));
}
EXPORT_SYMBOL_GPL(unlock_policy_rwsem_write);
/* internal prototypes */
static int __cpufreq_governor(struct cpufreq_policy *policy,
unsigned int event);
static unsigned int __cpufreq_get(unsigned int cpu);
2006-11-22 14:55:48 +00:00
static void handle_update(struct work_struct *work);
/**
* Two notifier lists: the "policy" list is involved in the
* validation process for a new CPU frequency policy; the
* "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 09:16:30 +00:00
static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
static struct srcu_notifier_head cpufreq_transition_notifier_list;
static bool init_cpufreq_transition_notifier_list_called;
static int __init init_cpufreq_transition_notifier_list(void)
{
srcu_init_notifier_head(&cpufreq_transition_notifier_list);
init_cpufreq_transition_notifier_list_called = true;
return 0;
}
pure_initcall(init_cpufreq_transition_notifier_list);
static LIST_HEAD(cpufreq_governor_list);
static DEFINE_MUTEX(cpufreq_governor_mutex);
struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
{
struct cpufreq_policy *data;
unsigned long flags;
if (cpu >= nr_cpu_ids)
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 = per_cpu(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;
err_out_put_module:
module_put(cpufreq_driver->owner);
err_out_unlock:
spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
err_out:
return NULL;
}
EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
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);
static void cpufreq_debug_enable_ratelimit(void)
{
unsigned long flags;
spin_lock_irqsave(&disable_ratelimit_lock, flags);
if (disable_ratelimit)
disable_ratelimit--;
spin_unlock_irqrestore(&disable_ratelimit_lock, flags);
}
static void cpufreq_debug_disable_ratelimit(void)
{
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;
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
* systems as each CPU might be scaled differently. So, use the arch
* 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;
static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
{
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) ||
(val == CPUFREQ_RESUMECHANGE || val == CPUFREQ_SUSPENDCHANGE)) {
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
/**
* cpufreq_notify_transition - call notifier chain and adjust_jiffies
* on frequency transition.
*
* This function calls the transition notifiers and the "adjust_jiffies"
* function. It is called twice on all CPU frequency changes that have
* external effects.
*/
void cpufreq_notify_transition(struct cpufreq_freqs *freqs, unsigned int state)
{
struct cpufreq_policy *policy;
BUG_ON(irqs_disabled());
freqs->flags = cpufreq_driver->flags;
dprintk("notification %u of frequency transition to %u kHz\n",
state, freqs->new);
policy = per_cpu(cpufreq_cpu_data, freqs->cpu);
switch (state) {
case CPUFREQ_PRECHANGE:
/* detect if the driver reported a value as "old frequency"
* which is not equal to what the cpufreq core thinks is
* "old frequency".
*/
if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
if ((policy) && (policy->cpu == freqs->cpu) &&
(policy->cur) && (policy->cur != freqs->old)) {
dprintk("Warning: CPU frequency is"
" %u, cpufreq assumed %u kHz.\n",
freqs->old, policy->cur);
freqs->old = policy->cur;
}
}
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 09:16:30 +00:00
CPUFREQ_PRECHANGE, freqs);
adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
break;
case CPUFREQ_POSTCHANGE:
adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
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 09:16:30 +00:00
CPUFREQ_POSTCHANGE, freqs);
if (likely(policy) && likely(policy->cpu == freqs->cpu))
policy->cur = freqs->new;
break;
}
}
EXPORT_SYMBOL_GPL(cpufreq_notify_transition);
/*********************************************************************
* SYSFS INTERFACE *
*********************************************************************/
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;
}
/**
* cpufreq_parse_governor - parse a governor string
*/
static int cpufreq_parse_governor(char *str_governor, unsigned int *policy,
struct cpufreq_governor **governor)
{
int err = -EINVAL;
if (!cpufreq_driver)
goto out;
if (cpufreq_driver->setpolicy) {
if (!strnicmp(str_governor, "performance", CPUFREQ_NAME_LEN)) {
*policy = CPUFREQ_POLICY_PERFORMANCE;
err = 0;
} else if (!strnicmp(str_governor, "powersave",
CPUFREQ_NAME_LEN)) {
*policy = CPUFREQ_POLICY_POWERSAVE;
err = 0;
}
} else if (cpufreq_driver->target) {
struct cpufreq_governor *t;
mutex_lock(&cpufreq_governor_mutex);
t = __find_governor(str_governor);
if (t == NULL) {
char *name = kasprintf(GFP_KERNEL, "cpufreq_%s",
str_governor);
if (name) {
int ret;
mutex_unlock(&cpufreq_governor_mutex);
ret = request_module("%s", name);
mutex_lock(&cpufreq_governor_mutex);
if (ret == 0)
t = __find_governor(str_governor);
}
kfree(name);
}
if (t != NULL) {
*governor = t;
err = 0;
}
mutex_unlock(&cpufreq_governor_mutex);
}
out:
return err;
}
/**
* 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".
*/
#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); \
}
show_one(cpuinfo_min_freq, cpuinfo.min_freq);
show_one(cpuinfo_max_freq, cpuinfo.max_freq);
show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
show_one(scaling_min_freq, min);
show_one(scaling_max_freq, max);
show_one(scaling_cur_freq, cur);
static int __cpufreq_set_policy(struct cpufreq_policy *data,
struct cpufreq_policy *policy);
[CPUFREQ] If max_freq got reduced (e.g. by _PPC) a write to sysfs scaling_governor let cpufreq core stuck at low max_freq for ever The previous patch had bugs (locking and refcount). This one could also be related to the latest DELL reports. But they only slip into this if a user prog (e.g. powersave daemon does when AC got (un) plugged due to a scheme change) echos something to /sys/../cpufreq/scaling_governor while the frequencies got limited by BIOS. This one works: Subject: Max freq stucks at low freq if reduced by _PPC and sysfs gov access The problem is reproducable by(if machine is limiting freqs via BIOS): - Unplugging AC -> max freq gets limited - echo ${governor} >/sys/.../cpufreq/scaling_governor (policy->user_data.max gets overridden with policy->max and will never come up again.) This patch exchanged the cpufreq_set_policy call to __cpufreq_set_policy and duplicated it's functionality but did not override user_data.max. The same happens with overridding min/max values. If freqs are limited and you override the min freq value, the max freq global value will also get stuck to the limited freq, even if BIOS allows all freqs again. Last scenario does only happen if BIOS does not reduce the frequency to the lowest value (should never happen, just for correctness...) drivers/cpufreq/cpufreq.c | 17 +++++++++++++++-- 1 files changed, 15 insertions(+), 2 deletions(-) Signed-off-by: Thomas Renninger <trenn@suse.de> Signed-off-by: "Pallipadi, Venkatesh" <venkatesh.pallipadi@intel.com> Signed-off-by: Dave Jones <davej@redhat.com>
2006-04-13 13:14:04 +00:00
/**
* 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; \
\
[CPUFREQ] If max_freq got reduced (e.g. by _PPC) a write to sysfs scaling_governor let cpufreq core stuck at low max_freq for ever The previous patch had bugs (locking and refcount). This one could also be related to the latest DELL reports. But they only slip into this if a user prog (e.g. powersave daemon does when AC got (un) plugged due to a scheme change) echos something to /sys/../cpufreq/scaling_governor while the frequencies got limited by BIOS. This one works: Subject: Max freq stucks at low freq if reduced by _PPC and sysfs gov access The problem is reproducable by(if machine is limiting freqs via BIOS): - Unplugging AC -> max freq gets limited - echo ${governor} >/sys/.../cpufreq/scaling_governor (policy->user_data.max gets overridden with policy->max and will never come up again.) This patch exchanged the cpufreq_set_policy call to __cpufreq_set_policy and duplicated it's functionality but did not override user_data.max. The same happens with overridding min/max values. If freqs are limited and you override the min freq value, the max freq global value will also get stuck to the limited freq, even if BIOS allows all freqs again. Last scenario does only happen if BIOS does not reduce the frequency to the lowest value (should never happen, just for correctness...) drivers/cpufreq/cpufreq.c | 17 +++++++++++++++-- 1 files changed, 15 insertions(+), 2 deletions(-) Signed-off-by: Thomas Renninger <trenn@suse.de> Signed-off-by: "Pallipadi, Venkatesh" <venkatesh.pallipadi@intel.com> Signed-off-by: Dave Jones <davej@redhat.com>
2006-04-13 13:14:04 +00:00
ret = __cpufreq_set_policy(policy, &new_policy); \
policy->user_policy.object = policy->object; \
\
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
*/
static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
const char *buf, size_t count)
{
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;
[CPUFREQ] If max_freq got reduced (e.g. by _PPC) a write to sysfs scaling_governor let cpufreq core stuck at low max_freq for ever The previous patch had bugs (locking and refcount). This one could also be related to the latest DELL reports. But they only slip into this if a user prog (e.g. powersave daemon does when AC got (un) plugged due to a scheme change) echos something to /sys/../cpufreq/scaling_governor while the frequencies got limited by BIOS. This one works: Subject: Max freq stucks at low freq if reduced by _PPC and sysfs gov access The problem is reproducable by(if machine is limiting freqs via BIOS): - Unplugging AC -> max freq gets limited - echo ${governor} >/sys/.../cpufreq/scaling_governor (policy->user_data.max gets overridden with policy->max and will never come up again.) This patch exchanged the cpufreq_set_policy call to __cpufreq_set_policy and duplicated it's functionality but did not override user_data.max. The same happens with overridding min/max values. If freqs are limited and you override the min freq value, the max freq global value will also get stuck to the limited freq, even if BIOS allows all freqs again. Last scenario does only happen if BIOS does not reduce the frequency to the lowest value (should never happen, just for correctness...) drivers/cpufreq/cpufreq.c | 17 +++++++++++++++-- 1 files changed, 15 insertions(+), 2 deletions(-) Signed-off-by: Thomas Renninger <trenn@suse.de> Signed-off-by: "Pallipadi, Venkatesh" <venkatesh.pallipadi@intel.com> Signed-off-by: Dave Jones <davej@redhat.com>
2006-04-13 13:14:04 +00:00
/* Do not use cpufreq_set_policy here or the user_policy.max
will be wrongly overridden */
ret = __cpufreq_set_policy(policy, &new_policy);
policy->user_policy.policy = policy->policy;
policy->user_policy.governor = policy->governor;
if (ret)
return ret;
else
return 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);
}
out:
i += sprintf(&buf[i], "\n");
return i;
}
static ssize_t show_cpus(const struct cpumask *mask, char *buf)
{
ssize_t i = 0;
unsigned int cpu;
for_each_cpu(cpu, mask) {
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;
}
/**
* show_related_cpus - show the CPUs affected by each transition even if
* hw coordination is in use
*/
static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
{
if (cpumask_empty(policy->related_cpus))
return show_cpus(policy->cpus, buf);
return show_cpus(policy->related_cpus, buf);
}
/**
* show_affected_cpus - show the CPUs affected by each transition
*/
static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
{
return show_cpus(policy->cpus, buf);
}
[CPUFREQ] Eliminate cpufreq_userspace scaling_setspeed deadlock Eliminate cpufreq_userspace scaling_setspeed deadlock. Luming Yu recently uncovered yet another cpufreq related deadlock. One thread that continuously switches the governors and the other thread that repeatedly cats the contents of cpufreq directory causes both these threads to go into a deadlock. Detailed examination of the deadlock showed the exact flow before the deadlock as: Thread 1 Thread 2 ________ ________ cats files under /sys/devices/.../cpufreq/ Set governor to userspace Adds a new sysfs entry for scaling_setspeed cats files under /sys/devices/.../cpufreq/ Set governor to performance Holds cpufreq_rw_sem in write mode Sends a STOP notify to userspace governor cat /sys/devices/.../cpufreq/scaling_setspeed Gets a handle on the above sysfs entry with sysfs_get_active Blocks while trying to get cpufreq_rw_sem in read mode Remove a sysfs entry for scaling_setspeed Blocks on sysfs_deactivate while waiting for earlier get_active (on other thread) to drain At this point both threads go into deadlock and any other thread that tries to do anything with sysfs cpufreq will also block. There seems to be no easy way to avoid this deadlock as long as cpufreq_userspace adds/removes the sysfs entry under same kobject as cpufreq. Below patch moves scaling_setspeed to cpufreq.c, keeping it always and calling back the governor on read/write. This is the cleanest fix I could think of, even though adding two callbacks in governor structure just for this seems unnecessary. Note that the change makes scaling_setspeed under /sys/.../cpufreq permanent and returns <unsupported> when governor is not userspace. Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> Signed-off-by: Dave Jones <davej@redhat.com>
2007-10-26 17:18:21 +00:00
static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
const char *buf, size_t count)
[CPUFREQ] Eliminate cpufreq_userspace scaling_setspeed deadlock Eliminate cpufreq_userspace scaling_setspeed deadlock. Luming Yu recently uncovered yet another cpufreq related deadlock. One thread that continuously switches the governors and the other thread that repeatedly cats the contents of cpufreq directory causes both these threads to go into a deadlock. Detailed examination of the deadlock showed the exact flow before the deadlock as: Thread 1 Thread 2 ________ ________ cats files under /sys/devices/.../cpufreq/ Set governor to userspace Adds a new sysfs entry for scaling_setspeed cats files under /sys/devices/.../cpufreq/ Set governor to performance Holds cpufreq_rw_sem in write mode Sends a STOP notify to userspace governor cat /sys/devices/.../cpufreq/scaling_setspeed Gets a handle on the above sysfs entry with sysfs_get_active Blocks while trying to get cpufreq_rw_sem in read mode Remove a sysfs entry for scaling_setspeed Blocks on sysfs_deactivate while waiting for earlier get_active (on other thread) to drain At this point both threads go into deadlock and any other thread that tries to do anything with sysfs cpufreq will also block. There seems to be no easy way to avoid this deadlock as long as cpufreq_userspace adds/removes the sysfs entry under same kobject as cpufreq. Below patch moves scaling_setspeed to cpufreq.c, keeping it always and calling back the governor on read/write. This is the cleanest fix I could think of, even though adding two callbacks in governor structure just for this seems unnecessary. Note that the change makes scaling_setspeed under /sys/.../cpufreq permanent and returns <unsupported> when governor is not userspace. Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> Signed-off-by: Dave Jones <davej@redhat.com>
2007-10-26 17:18:21 +00:00
{
unsigned int freq = 0;
unsigned int ret;
if (!policy->governor || !policy->governor->store_setspeed)
[CPUFREQ] Eliminate cpufreq_userspace scaling_setspeed deadlock Eliminate cpufreq_userspace scaling_setspeed deadlock. Luming Yu recently uncovered yet another cpufreq related deadlock. One thread that continuously switches the governors and the other thread that repeatedly cats the contents of cpufreq directory causes both these threads to go into a deadlock. Detailed examination of the deadlock showed the exact flow before the deadlock as: Thread 1 Thread 2 ________ ________ cats files under /sys/devices/.../cpufreq/ Set governor to userspace Adds a new sysfs entry for scaling_setspeed cats files under /sys/devices/.../cpufreq/ Set governor to performance Holds cpufreq_rw_sem in write mode Sends a STOP notify to userspace governor cat /sys/devices/.../cpufreq/scaling_setspeed Gets a handle on the above sysfs entry with sysfs_get_active Blocks while trying to get cpufreq_rw_sem in read mode Remove a sysfs entry for scaling_setspeed Blocks on sysfs_deactivate while waiting for earlier get_active (on other thread) to drain At this point both threads go into deadlock and any other thread that tries to do anything with sysfs cpufreq will also block. There seems to be no easy way to avoid this deadlock as long as cpufreq_userspace adds/removes the sysfs entry under same kobject as cpufreq. Below patch moves scaling_setspeed to cpufreq.c, keeping it always and calling back the governor on read/write. This is the cleanest fix I could think of, even though adding two callbacks in governor structure just for this seems unnecessary. Note that the change makes scaling_setspeed under /sys/.../cpufreq permanent and returns <unsupported> when governor is not userspace. Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> Signed-off-by: Dave Jones <davej@redhat.com>
2007-10-26 17:18:21 +00:00
return -EINVAL;
ret = sscanf(buf, "%u", &freq);
if (ret != 1)
return -EINVAL;
policy->governor->store_setspeed(policy, freq);
return count;
}
static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
{
if (!policy->governor || !policy->governor->show_setspeed)
[CPUFREQ] Eliminate cpufreq_userspace scaling_setspeed deadlock Eliminate cpufreq_userspace scaling_setspeed deadlock. Luming Yu recently uncovered yet another cpufreq related deadlock. One thread that continuously switches the governors and the other thread that repeatedly cats the contents of cpufreq directory causes both these threads to go into a deadlock. Detailed examination of the deadlock showed the exact flow before the deadlock as: Thread 1 Thread 2 ________ ________ cats files under /sys/devices/.../cpufreq/ Set governor to userspace Adds a new sysfs entry for scaling_setspeed cats files under /sys/devices/.../cpufreq/ Set governor to performance Holds cpufreq_rw_sem in write mode Sends a STOP notify to userspace governor cat /sys/devices/.../cpufreq/scaling_setspeed Gets a handle on the above sysfs entry with sysfs_get_active Blocks while trying to get cpufreq_rw_sem in read mode Remove a sysfs entry for scaling_setspeed Blocks on sysfs_deactivate while waiting for earlier get_active (on other thread) to drain At this point both threads go into deadlock and any other thread that tries to do anything with sysfs cpufreq will also block. There seems to be no easy way to avoid this deadlock as long as cpufreq_userspace adds/removes the sysfs entry under same kobject as cpufreq. Below patch moves scaling_setspeed to cpufreq.c, keeping it always and calling back the governor on read/write. This is the cleanest fix I could think of, even though adding two callbacks in governor structure just for this seems unnecessary. Note that the change makes scaling_setspeed under /sys/.../cpufreq permanent and returns <unsupported> when governor is not userspace. Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> Signed-off-by: Dave Jones <davej@redhat.com>
2007-10-26 17:18:21 +00:00
return sprintf(buf, "<unsupported>\n");
return policy->governor->show_setspeed(policy, buf);
}
[ACPI/CPUFREQ] Introduce bios_limit per cpu cpufreq sysfs interface This interface is mainly intended (and implemented) for ACPI _PPC BIOS frequency limitations, but other cpufreq drivers can also use it for similar use-cases. Why is this needed: Currently it's not obvious why cpufreq got limited. People see cpufreq/scaling_max_freq reduced, but this could have happened by: - any userspace prog writing to scaling_max_freq - thermal limitations - hardware (_PPC in ACPI case) limitiations Therefore export bios_limit (in kHz) to: - Point the user that it's the BIOS (broken or intended) which limits frequency - Export it as a sysfs interface for userspace progs. While this was a rarely used feature on laptops, there will appear more and more server implemenations providing "Green IT" features like allowing the service processor to limit the frequency. People want to know about HW/BIOS frequency limitations. All ACPI P-state driven cpufreq drivers are covered with this patch: - powernow-k8 - powernow-k7 - acpi-cpufreq Tested with a patched DSDT which limits the first two cores (_PPC returns 1) via _PPC, exposed by bios_limit: # echo 2200000 >cpu2/cpufreq/scaling_max_freq # cat cpu*/cpufreq/scaling_max_freq 2600000 2600000 2200000 2200000 # #scaling_max_freq shows general user/thermal/BIOS limitations # cat cpu*/cpufreq/bios_limit 2600000 2600000 2800000 2800000 # #bios_limit only shows the HW/BIOS limitation CC: Pallipadi Venkatesh <venkatesh.pallipadi@intel.com> CC: Len Brown <lenb@kernel.org> CC: davej@codemonkey.org.uk CC: linux@dominikbrodowski.net Signed-off-by: Thomas Renninger <trenn@suse.de> Signed-off-by: Dave Jones <davej@redhat.com>
2009-11-19 11:31:01 +00:00
/**
* show_scaling_driver - show the current cpufreq HW/BIOS limitation
*/
static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
{
unsigned int limit;
int ret;
if (cpufreq_driver->bios_limit) {
ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
if (!ret)
return sprintf(buf, "%u\n", limit);
}
return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
}
cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
cpufreq_freq_attr_ro(cpuinfo_min_freq);
cpufreq_freq_attr_ro(cpuinfo_max_freq);
cpufreq_freq_attr_ro(cpuinfo_transition_latency);
cpufreq_freq_attr_ro(scaling_available_governors);
cpufreq_freq_attr_ro(scaling_driver);
cpufreq_freq_attr_ro(scaling_cur_freq);
cpufreq_freq_attr_ro(bios_limit);
cpufreq_freq_attr_ro(related_cpus);
cpufreq_freq_attr_ro(affected_cpus);
cpufreq_freq_attr_rw(scaling_min_freq);
cpufreq_freq_attr_rw(scaling_max_freq);
cpufreq_freq_attr_rw(scaling_governor);
cpufreq_freq_attr_rw(scaling_setspeed);
static struct attribute *default_attrs[] = {
&cpuinfo_min_freq.attr,
&cpuinfo_max_freq.attr,
&cpuinfo_transition_latency.attr,
&scaling_min_freq.attr,
&scaling_max_freq.attr,
&affected_cpus.attr,
&related_cpus.attr,
&scaling_governor.attr,
&scaling_driver.attr,
&scaling_available_governors.attr,
[CPUFREQ] Eliminate cpufreq_userspace scaling_setspeed deadlock Eliminate cpufreq_userspace scaling_setspeed deadlock. Luming Yu recently uncovered yet another cpufreq related deadlock. One thread that continuously switches the governors and the other thread that repeatedly cats the contents of cpufreq directory causes both these threads to go into a deadlock. Detailed examination of the deadlock showed the exact flow before the deadlock as: Thread 1 Thread 2 ________ ________ cats files under /sys/devices/.../cpufreq/ Set governor to userspace Adds a new sysfs entry for scaling_setspeed cats files under /sys/devices/.../cpufreq/ Set governor to performance Holds cpufreq_rw_sem in write mode Sends a STOP notify to userspace governor cat /sys/devices/.../cpufreq/scaling_setspeed Gets a handle on the above sysfs entry with sysfs_get_active Blocks while trying to get cpufreq_rw_sem in read mode Remove a sysfs entry for scaling_setspeed Blocks on sysfs_deactivate while waiting for earlier get_active (on other thread) to drain At this point both threads go into deadlock and any other thread that tries to do anything with sysfs cpufreq will also block. There seems to be no easy way to avoid this deadlock as long as cpufreq_userspace adds/removes the sysfs entry under same kobject as cpufreq. Below patch moves scaling_setspeed to cpufreq.c, keeping it always and calling back the governor on read/write. This is the cleanest fix I could think of, even though adding two callbacks in governor structure just for this seems unnecessary. Note that the change makes scaling_setspeed under /sys/.../cpufreq permanent and returns <unsupported> when governor is not userspace. Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> Signed-off-by: Dave Jones <davej@redhat.com>
2007-10-26 17:18:21 +00:00
&scaling_setspeed.attr,
NULL
};
struct kobject *cpufreq_global_kobject;
EXPORT_SYMBOL(cpufreq_global_kobject);
#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 = -EINVAL;
policy = cpufreq_cpu_get(policy->cpu);
if (!policy)
goto no_policy;
if (lock_policy_rwsem_read(policy->cpu) < 0)
goto fail;
if (fattr->show)
ret = fattr->show(policy, buf);
else
ret = -EIO;
unlock_policy_rwsem_read(policy->cpu);
fail:
cpufreq_cpu_put(policy);
no_policy:
return ret;
}
static ssize_t store(struct kobject *kobj, struct attribute *attr,
const char *buf, size_t count)
{
struct cpufreq_policy *policy = to_policy(kobj);
struct freq_attr *fattr = to_attr(attr);
ssize_t ret = -EINVAL;
policy = cpufreq_cpu_get(policy->cpu);
if (!policy)
goto no_policy;
if (lock_policy_rwsem_write(policy->cpu) < 0)
goto fail;
if (fattr->store)
ret = fattr->store(policy, buf, count);
else
ret = -EIO;
unlock_policy_rwsem_write(policy->cpu);
fail:
cpufreq_cpu_put(policy);
no_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 const 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] Bail out of cpufreq_add_dev if the link for a managed CPU got created Doing: echo 0 >cpu1/online echo 1 >cpu1/online on a managed CPU will result in: Jul 22 15:15:37 linux kernel: [ 80.013864] WARNING: at fs/sysfs/dir.c:487 sysfs_add_one+0xcf/0xe6() Jul 22 15:15:37 linux kernel: [ 80.013866] Hardware name: To Be Filled By O.E.M. Jul 22 15:15:37 linux kernel: [ 80.013868] sysfs: cannot create duplicate filename '/devices/system/cpu/cpu1/cpufreq' Jul 22 15:15:37 linux kernel: [ 80.013870] Modules linked in: powernow_k8 Jul 22 15:15:37 linux kernel: [ 80.013874] Pid: 5750, comm: bash Not tainted 2.6.31-rc2 #40 Jul 22 15:15:37 linux kernel: [ 80.013876] Call Trace: Jul 22 15:15:37 linux kernel: [ 80.013879] [<ffffffff8112ebda>] ? sysfs_add_one+0xcf/0xe6 Jul 22 15:15:37 linux kernel: [ 80.013884] [<ffffffff81041926>] warn_slowpath_common+0x77/0xa4 Jul 22 15:15:37 linux kernel: [ 80.013888] [<ffffffff810419a0>] warn_slowpath_fmt+0x3c/0x3e Jul 22 15:15:37 linux kernel: [ 80.013891] [<ffffffff8112ebda>] sysfs_add_one+0xcf/0xe6 Jul 22 15:15:37 linux kernel: [ 80.013894] [<ffffffff8112f213>] create_dir+0x58/0x87 Jul 22 15:15:37 linux kernel: [ 80.013898] [<ffffffff8112f27a>] sysfs_create_dir+0x38/0x4f Jul 22 15:15:37 linux kernel: [ 80.013902] [<ffffffff811ffb8a>] kobject_add_internal+0x11f/0x1de Jul 22 15:15:37 linux kernel: [ 80.013905] [<ffffffff811ffd21>] kobject_add_varg+0x41/0x4e Jul 22 15:15:37 linux kernel: [ 80.013908] [<ffffffff811ffd7a>] kobject_init_and_add+0x4c/0x57 Jul 22 15:15:37 linux kernel: [ 80.013913] [<ffffffff810667bc>] ? mark_lock+0x22/0x228 Jul 22 15:15:37 linux kernel: [ 80.013918] [<ffffffff813e8a3b>] cpufreq_add_dev_interface+0x40/0x1e4 ... This bug slipped in by git commit: 150b06f7f223cfd0f808737a5243cceca8ea47fa When splitting up cpufreq_add_dev, the whole cpufreq_add_dev function is not left anymore, only cpufreq_add_dev_policy. This patch should reconstruct the identical functionality again as it was before the split. CC: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> Signed-off-by: Thomas Renninger <trenn@suse.de> Signed-off-by: Dave Jones <davej@redhat.com>
2009-07-24 13:25:03 +00:00
/*
* Returns:
* Negative: Failure
* 0: Success
* Positive: When we have a managed CPU and the sysfs got symlinked
*/
static int cpufreq_add_dev_policy(unsigned int cpu,
struct cpufreq_policy *policy,
struct sys_device *sys_dev)
{
int ret = 0;
#ifdef CONFIG_SMP
unsigned long flags;
unsigned int j;
#ifdef CONFIG_HOTPLUG_CPU
struct cpufreq_governor *gov;
gov = __find_governor(per_cpu(cpufreq_cpu_governor, cpu));
if (gov) {
policy->governor = gov;
dprintk("Restoring governor %s for cpu %d\n",
policy->governor->name, cpu);
}
#endif
for_each_cpu(j, policy->cpus) {
struct cpufreq_policy *managed_policy;
if (cpu == j)
continue;
/* Check for existing affected CPUs.
* They may not be aware of it due to CPU Hotplug.
* cpufreq_cpu_put is called when the device is removed
* in __cpufreq_remove_dev()
*/
managed_policy = cpufreq_cpu_get(j);
if (unlikely(managed_policy)) {
/* Set proper policy_cpu */
unlock_policy_rwsem_write(cpu);
per_cpu(cpufreq_policy_cpu, cpu) = managed_policy->cpu;
if (lock_policy_rwsem_write(cpu) < 0) {
/* Should not go through policy unlock path */
if (cpufreq_driver->exit)
cpufreq_driver->exit(policy);
cpufreq_cpu_put(managed_policy);
return -EBUSY;
}
spin_lock_irqsave(&cpufreq_driver_lock, flags);
cpumask_copy(managed_policy->cpus, policy->cpus);
per_cpu(cpufreq_cpu_data, cpu) = managed_policy;
spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
dprintk("CPU already managed, adding link\n");
ret = sysfs_create_link(&sys_dev->kobj,
&managed_policy->kobj,
"cpufreq");
if (ret)
cpufreq_cpu_put(managed_policy);
/*
* Success. We only needed to be added to the mask.
* Call driver->exit() because only the cpu parent of
* the kobj needed to call init().
*/
if (cpufreq_driver->exit)
cpufreq_driver->exit(policy);
[CPUFREQ] Bail out of cpufreq_add_dev if the link for a managed CPU got created Doing: echo 0 >cpu1/online echo 1 >cpu1/online on a managed CPU will result in: Jul 22 15:15:37 linux kernel: [ 80.013864] WARNING: at fs/sysfs/dir.c:487 sysfs_add_one+0xcf/0xe6() Jul 22 15:15:37 linux kernel: [ 80.013866] Hardware name: To Be Filled By O.E.M. Jul 22 15:15:37 linux kernel: [ 80.013868] sysfs: cannot create duplicate filename '/devices/system/cpu/cpu1/cpufreq' Jul 22 15:15:37 linux kernel: [ 80.013870] Modules linked in: powernow_k8 Jul 22 15:15:37 linux kernel: [ 80.013874] Pid: 5750, comm: bash Not tainted 2.6.31-rc2 #40 Jul 22 15:15:37 linux kernel: [ 80.013876] Call Trace: Jul 22 15:15:37 linux kernel: [ 80.013879] [<ffffffff8112ebda>] ? sysfs_add_one+0xcf/0xe6 Jul 22 15:15:37 linux kernel: [ 80.013884] [<ffffffff81041926>] warn_slowpath_common+0x77/0xa4 Jul 22 15:15:37 linux kernel: [ 80.013888] [<ffffffff810419a0>] warn_slowpath_fmt+0x3c/0x3e Jul 22 15:15:37 linux kernel: [ 80.013891] [<ffffffff8112ebda>] sysfs_add_one+0xcf/0xe6 Jul 22 15:15:37 linux kernel: [ 80.013894] [<ffffffff8112f213>] create_dir+0x58/0x87 Jul 22 15:15:37 linux kernel: [ 80.013898] [<ffffffff8112f27a>] sysfs_create_dir+0x38/0x4f Jul 22 15:15:37 linux kernel: [ 80.013902] [<ffffffff811ffb8a>] kobject_add_internal+0x11f/0x1de Jul 22 15:15:37 linux kernel: [ 80.013905] [<ffffffff811ffd21>] kobject_add_varg+0x41/0x4e Jul 22 15:15:37 linux kernel: [ 80.013908] [<ffffffff811ffd7a>] kobject_init_and_add+0x4c/0x57 Jul 22 15:15:37 linux kernel: [ 80.013913] [<ffffffff810667bc>] ? mark_lock+0x22/0x228 Jul 22 15:15:37 linux kernel: [ 80.013918] [<ffffffff813e8a3b>] cpufreq_add_dev_interface+0x40/0x1e4 ... This bug slipped in by git commit: 150b06f7f223cfd0f808737a5243cceca8ea47fa When splitting up cpufreq_add_dev, the whole cpufreq_add_dev function is not left anymore, only cpufreq_add_dev_policy. This patch should reconstruct the identical functionality again as it was before the split. CC: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> Signed-off-by: Thomas Renninger <trenn@suse.de> Signed-off-by: Dave Jones <davej@redhat.com>
2009-07-24 13:25:03 +00:00
if (!ret)
return 1;
else
return ret;
}
}
#endif
return ret;
}
/* symlink affected CPUs */
static int cpufreq_add_dev_symlink(unsigned int cpu,
struct cpufreq_policy *policy)
{
unsigned int j;
int ret = 0;
for_each_cpu(j, policy->cpus) {
struct cpufreq_policy *managed_policy;
struct sys_device *cpu_sys_dev;
if (j == cpu)
continue;
if (!cpu_online(j))
continue;
dprintk("CPU %u already managed, adding link\n", j);
managed_policy = cpufreq_cpu_get(cpu);
cpu_sys_dev = get_cpu_sysdev(j);
ret = sysfs_create_link(&cpu_sys_dev->kobj, &policy->kobj,
"cpufreq");
if (ret) {
cpufreq_cpu_put(managed_policy);
return ret;
}
}
return ret;
}
static int cpufreq_add_dev_interface(unsigned int cpu,
struct cpufreq_policy *policy,
struct sys_device *sys_dev)
{
struct cpufreq_policy new_policy;
struct freq_attr **drv_attr;
unsigned long flags;
int ret = 0;
unsigned int j;
/* prepare interface data */
ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
&sys_dev->kobj, "cpufreq");
if (ret)
return ret;
/* set up files for this cpu device */
drv_attr = cpufreq_driver->attr;
while ((drv_attr) && (*drv_attr)) {
ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
if (ret)
goto err_out_kobj_put;
drv_attr++;
}
if (cpufreq_driver->get) {
ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
if (ret)
goto err_out_kobj_put;
}
if (cpufreq_driver->target) {
ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
if (ret)
goto err_out_kobj_put;
}
[ACPI/CPUFREQ] Introduce bios_limit per cpu cpufreq sysfs interface This interface is mainly intended (and implemented) for ACPI _PPC BIOS frequency limitations, but other cpufreq drivers can also use it for similar use-cases. Why is this needed: Currently it's not obvious why cpufreq got limited. People see cpufreq/scaling_max_freq reduced, but this could have happened by: - any userspace prog writing to scaling_max_freq - thermal limitations - hardware (_PPC in ACPI case) limitiations Therefore export bios_limit (in kHz) to: - Point the user that it's the BIOS (broken or intended) which limits frequency - Export it as a sysfs interface for userspace progs. While this was a rarely used feature on laptops, there will appear more and more server implemenations providing "Green IT" features like allowing the service processor to limit the frequency. People want to know about HW/BIOS frequency limitations. All ACPI P-state driven cpufreq drivers are covered with this patch: - powernow-k8 - powernow-k7 - acpi-cpufreq Tested with a patched DSDT which limits the first two cores (_PPC returns 1) via _PPC, exposed by bios_limit: # echo 2200000 >cpu2/cpufreq/scaling_max_freq # cat cpu*/cpufreq/scaling_max_freq 2600000 2600000 2200000 2200000 # #scaling_max_freq shows general user/thermal/BIOS limitations # cat cpu*/cpufreq/bios_limit 2600000 2600000 2800000 2800000 # #bios_limit only shows the HW/BIOS limitation CC: Pallipadi Venkatesh <venkatesh.pallipadi@intel.com> CC: Len Brown <lenb@kernel.org> CC: davej@codemonkey.org.uk CC: linux@dominikbrodowski.net Signed-off-by: Thomas Renninger <trenn@suse.de> Signed-off-by: Dave Jones <davej@redhat.com>
2009-11-19 11:31:01 +00:00
if (cpufreq_driver->bios_limit) {
ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
if (ret)
goto err_out_kobj_put;
}
spin_lock_irqsave(&cpufreq_driver_lock, flags);
for_each_cpu(j, policy->cpus) {
if (!cpu_online(j))
continue;
per_cpu(cpufreq_cpu_data, j) = policy;
per_cpu(cpufreq_policy_cpu, j) = policy->cpu;
}
spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
ret = cpufreq_add_dev_symlink(cpu, policy);
if (ret)
goto err_out_kobj_put;
memcpy(&new_policy, policy, sizeof(struct cpufreq_policy));
/* assure that the starting sequence is run in __cpufreq_set_policy */
policy->governor = NULL;
/* set default policy */
ret = __cpufreq_set_policy(policy, &new_policy);
policy->user_policy.policy = policy->policy;
policy->user_policy.governor = policy->governor;
if (ret) {
dprintk("setting policy failed\n");
if (cpufreq_driver->exit)
cpufreq_driver->exit(policy);
}
return ret;
err_out_kobj_put:
kobject_put(&policy->kobj);
wait_for_completion(&policy->kobj_unregister);
return ret;
}
/**
* cpufreq_add_dev - add a CPU device
*
* Adds the cpufreq interface for a CPU device.
*
* The Oracle says: try running cpufreq registration/unregistration concurrently
* with with cpu hotplugging and all hell will break loose. Tried to clean this
* mess up, but more thorough testing is needed. - Mathieu
*/
static int cpufreq_add_dev(struct sys_device *sys_dev)
{
unsigned int cpu = sys_dev->id;
int ret = 0, found = 0;
struct cpufreq_policy *policy;
unsigned long flags;
unsigned int j;
#ifdef CONFIG_HOTPLUG_CPU
int sibling;
#endif
[PATCH] create and destroy cpufreq sysfs entries based on cpu notifiers cpufreq entries in sysfs should only be populated when CPU is online state. When we either boot with maxcpus=x and then boot the other cpus by echoing to sysfs online file, these entries should be created and destroyed when CPU_DEAD is notified. Same treatement as cache entries under sysfs. We place the processor in the lowest frequency, so hw managed P-State transitions can still work on the other threads to save power. Primary goal was to just make these directories appear/disapper dynamically. There is one in this patch i had to do, which i really dont like myself but probably best if someone handling the cpufreq infrastructure could give this code right treatment if this is not acceptable. I guess its probably good for the first cut. - Converting lock_cpu_hotplug()/unlock_cpu_hotplug() to disable/enable preempt. The locking was smack in the middle of the notification path, when the hotplug is already holding the lock. I tried another solution to avoid this so avoid taking locks if we know we are from notification path. The solution was getting very ugly and i decided this was probably good for this iteration until someone who understands cpufreq could do a better job than me. (akpm: export cpucontrol to GPL modules: drivers/cpufreq/cpufreq_stats.c now does lock_cpu_hotplug()) Signed-off-by: Ashok Raj <ashok.raj@intel.com> Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> Cc: Dave Jones <davej@codemonkey.org.uk> Cc: Zwane Mwaikambo <zwane@holomorphy.com> Cc: Greg KH <greg@kroah.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 22:59:54 +00:00
if (cpu_is_offline(cpu))
return 0;
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)) {
cpufreq_cpu_put(policy);
cpufreq_debug_enable_ratelimit();
return 0;
}
#endif
if (!try_module_get(cpufreq_driver->owner)) {
ret = -EINVAL;
goto module_out;
}
ret = -ENOMEM;
policy = kzalloc(sizeof(struct cpufreq_policy), GFP_KERNEL);
if (!policy)
goto nomem_out;
if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
goto err_free_policy;
if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
goto err_free_cpumask;
policy->cpu = cpu;
cpumask_copy(policy->cpus, cpumask_of(cpu));
/* Initially set CPU itself as the policy_cpu */
per_cpu(cpufreq_policy_cpu, cpu) = cpu;
ret = (lock_policy_rwsem_write(cpu) < 0);
WARN_ON(ret);
init_completion(&policy->kobj_unregister);
2006-11-22 14:55:48 +00:00
INIT_WORK(&policy->update, handle_update);
/* Set governor before ->init, so that driver could check it */
#ifdef CONFIG_HOTPLUG_CPU
for_each_online_cpu(sibling) {
struct cpufreq_policy *cp = per_cpu(cpufreq_cpu_data, sibling);
if (cp && cp->governor &&
(cpumask_test_cpu(cpu, cp->related_cpus))) {
policy->governor = cp->governor;
found = 1;
break;
}
}
#endif
if (!found)
policy->governor = CPUFREQ_DEFAULT_GOVERNOR;
/* 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");
goto err_unlock_policy;
}
policy->user_policy.min = policy->min;
policy->user_policy.max = policy->max;
blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
CPUFREQ_START, policy);
ret = cpufreq_add_dev_policy(cpu, policy, sys_dev);
[CPUFREQ] Bail out of cpufreq_add_dev if the link for a managed CPU got created Doing: echo 0 >cpu1/online echo 1 >cpu1/online on a managed CPU will result in: Jul 22 15:15:37 linux kernel: [ 80.013864] WARNING: at fs/sysfs/dir.c:487 sysfs_add_one+0xcf/0xe6() Jul 22 15:15:37 linux kernel: [ 80.013866] Hardware name: To Be Filled By O.E.M. Jul 22 15:15:37 linux kernel: [ 80.013868] sysfs: cannot create duplicate filename '/devices/system/cpu/cpu1/cpufreq' Jul 22 15:15:37 linux kernel: [ 80.013870] Modules linked in: powernow_k8 Jul 22 15:15:37 linux kernel: [ 80.013874] Pid: 5750, comm: bash Not tainted 2.6.31-rc2 #40 Jul 22 15:15:37 linux kernel: [ 80.013876] Call Trace: Jul 22 15:15:37 linux kernel: [ 80.013879] [<ffffffff8112ebda>] ? sysfs_add_one+0xcf/0xe6 Jul 22 15:15:37 linux kernel: [ 80.013884] [<ffffffff81041926>] warn_slowpath_common+0x77/0xa4 Jul 22 15:15:37 linux kernel: [ 80.013888] [<ffffffff810419a0>] warn_slowpath_fmt+0x3c/0x3e Jul 22 15:15:37 linux kernel: [ 80.013891] [<ffffffff8112ebda>] sysfs_add_one+0xcf/0xe6 Jul 22 15:15:37 linux kernel: [ 80.013894] [<ffffffff8112f213>] create_dir+0x58/0x87 Jul 22 15:15:37 linux kernel: [ 80.013898] [<ffffffff8112f27a>] sysfs_create_dir+0x38/0x4f Jul 22 15:15:37 linux kernel: [ 80.013902] [<ffffffff811ffb8a>] kobject_add_internal+0x11f/0x1de Jul 22 15:15:37 linux kernel: [ 80.013905] [<ffffffff811ffd21>] kobject_add_varg+0x41/0x4e Jul 22 15:15:37 linux kernel: [ 80.013908] [<ffffffff811ffd7a>] kobject_init_and_add+0x4c/0x57 Jul 22 15:15:37 linux kernel: [ 80.013913] [<ffffffff810667bc>] ? mark_lock+0x22/0x228 Jul 22 15:15:37 linux kernel: [ 80.013918] [<ffffffff813e8a3b>] cpufreq_add_dev_interface+0x40/0x1e4 ... This bug slipped in by git commit: 150b06f7f223cfd0f808737a5243cceca8ea47fa When splitting up cpufreq_add_dev, the whole cpufreq_add_dev function is not left anymore, only cpufreq_add_dev_policy. This patch should reconstruct the identical functionality again as it was before the split. CC: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> Signed-off-by: Thomas Renninger <trenn@suse.de> Signed-off-by: Dave Jones <davej@redhat.com>
2009-07-24 13:25:03 +00:00
if (ret) {
if (ret > 0)
/* This is a managed cpu, symlink created,
exit with 0 */
ret = 0;
goto err_unlock_policy;
[CPUFREQ] Bail out of cpufreq_add_dev if the link for a managed CPU got created Doing: echo 0 >cpu1/online echo 1 >cpu1/online on a managed CPU will result in: Jul 22 15:15:37 linux kernel: [ 80.013864] WARNING: at fs/sysfs/dir.c:487 sysfs_add_one+0xcf/0xe6() Jul 22 15:15:37 linux kernel: [ 80.013866] Hardware name: To Be Filled By O.E.M. Jul 22 15:15:37 linux kernel: [ 80.013868] sysfs: cannot create duplicate filename '/devices/system/cpu/cpu1/cpufreq' Jul 22 15:15:37 linux kernel: [ 80.013870] Modules linked in: powernow_k8 Jul 22 15:15:37 linux kernel: [ 80.013874] Pid: 5750, comm: bash Not tainted 2.6.31-rc2 #40 Jul 22 15:15:37 linux kernel: [ 80.013876] Call Trace: Jul 22 15:15:37 linux kernel: [ 80.013879] [<ffffffff8112ebda>] ? sysfs_add_one+0xcf/0xe6 Jul 22 15:15:37 linux kernel: [ 80.013884] [<ffffffff81041926>] warn_slowpath_common+0x77/0xa4 Jul 22 15:15:37 linux kernel: [ 80.013888] [<ffffffff810419a0>] warn_slowpath_fmt+0x3c/0x3e Jul 22 15:15:37 linux kernel: [ 80.013891] [<ffffffff8112ebda>] sysfs_add_one+0xcf/0xe6 Jul 22 15:15:37 linux kernel: [ 80.013894] [<ffffffff8112f213>] create_dir+0x58/0x87 Jul 22 15:15:37 linux kernel: [ 80.013898] [<ffffffff8112f27a>] sysfs_create_dir+0x38/0x4f Jul 22 15:15:37 linux kernel: [ 80.013902] [<ffffffff811ffb8a>] kobject_add_internal+0x11f/0x1de Jul 22 15:15:37 linux kernel: [ 80.013905] [<ffffffff811ffd21>] kobject_add_varg+0x41/0x4e Jul 22 15:15:37 linux kernel: [ 80.013908] [<ffffffff811ffd7a>] kobject_init_and_add+0x4c/0x57 Jul 22 15:15:37 linux kernel: [ 80.013913] [<ffffffff810667bc>] ? mark_lock+0x22/0x228 Jul 22 15:15:37 linux kernel: [ 80.013918] [<ffffffff813e8a3b>] cpufreq_add_dev_interface+0x40/0x1e4 ... This bug slipped in by git commit: 150b06f7f223cfd0f808737a5243cceca8ea47fa When splitting up cpufreq_add_dev, the whole cpufreq_add_dev function is not left anymore, only cpufreq_add_dev_policy. This patch should reconstruct the identical functionality again as it was before the split. CC: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> Signed-off-by: Thomas Renninger <trenn@suse.de> Signed-off-by: Dave Jones <davej@redhat.com>
2009-07-24 13:25:03 +00:00
}
ret = cpufreq_add_dev_interface(cpu, policy, sys_dev);
if (ret)
goto err_out_unregister;
[CPUFREQ] fix double unlock of cpu_policy_rwsem in drivers/cpufreq/cpufreq.c In drivers/cpufreq/cpufreq.c the function cpufreq_add_dev() takes the error exit 'err_out_unregister' from different places once with the 'cpu_policy_rwsem' lock held, once with the lock released: | if (ret) | goto err_out_unregister; | } | | policy->governor = NULL; /* to assure that the starting sequence is | * run in cpufreq_set_policy */ | | /* set default policy */ | ret = __cpufreq_set_policy(policy, &new_policy); | policy->user_policy.policy = policy->policy; | policy->user_policy.governor = policy->governor; | | unlock_policy_rwsem_write(cpu); | | if (ret) { | dprintk("setting policy failed\n"); | goto err_out_unregister; | } This leads to the following error message in case of a failing __cpufreq_set_policy() call: ===================================== [ BUG: bad unlock balance detected! ] ------------------------------------- swapper/1 is trying to release lock (&per_cpu(cpu_policy_rwsem, cpu)) at: [<c01b4564>] unlock_policy_rwsem_write+0x30/0x40 but there are no more locks to release! other info that might help us debug this: 1 lock held by swapper/1: #0: (sysdev_drivers_lock){--..}, at: [<c018fd18>] sysdev_driver_register+0x74/0x130 stack backtrace: [<c002f588>] (dump_stack+0x0/0x14) from [<c00692fc>] (print_unlock_inbalance_bug+0xc8/0x104) [<c0069234>] (print_unlock_inbalance_bug+0x0/0x104) from [<c006b7ac>] (lock_release_non_nested+0xc4/0x19c) r6:00000028 r5:c3c1ab80 r4:c01b4564 [<c006b6e8>] (lock_release_non_nested+0x0/0x19c) from [<c006b9e0>] (lock_release+0x15c/0x18c) r8:60000013 r7:00000001 r6:c01b4564 r5:c0541bb4 r4:c3c1ab80 [<c006b884>] (lock_release+0x0/0x18c) from [<c0061ba0>] (up_write+0x24/0x30) r8:c0541b80 r7:00000000 r6:ffffffea r5:c3c34828 r4:c0541b8c [<c0061b7c>] (up_write+0x0/0x30) from [<c01b4564>] (unlock_policy_rwsem_write+0x30/0x40) r4:c3c34884 [<c01b4534>] (unlock_policy_rwsem_write+0x0/0x40) from [<c01b4c40>] (cpufreq_add_dev+0x324/0x398) [<c01b491c>] (cpufreq_add_dev+0x0/0x398) from [<c018fd64>] (sysdev_driver_register+0xc0/0x130) [<c018fca4>] (sysdev_driver_register+0x0/0x130) from [<c01b3574>] (cpufreq_register_driver+0xbc/0x174) Signed-off-by: Lothar Waßmann <LW@KARO-electronics.de> Signed-off-by: Dave Jones <davej@redhat.com>
2008-05-29 15:54:52 +00:00
unlock_policy_rwsem_write(cpu);
kobject_uevent(&policy->kobj, KOBJ_ADD);
module_put(cpufreq_driver->owner);
dprintk("initialization complete\n");
cpufreq_debug_enable_ratelimit();
return 0;
err_out_unregister:
spin_lock_irqsave(&cpufreq_driver_lock, flags);
for_each_cpu(j, policy->cpus)
per_cpu(cpufreq_cpu_data, j) = NULL;
spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
kobject_put(&policy->kobj);
wait_for_completion(&policy->kobj_unregister);
err_unlock_policy:
unlock_policy_rwsem_write(cpu);
err_free_cpumask:
free_cpumask_var(policy->cpus);
err_free_policy:
kfree(policy);
nomem_out:
module_put(cpufreq_driver->owner);
[PATCH] create and destroy cpufreq sysfs entries based on cpu notifiers cpufreq entries in sysfs should only be populated when CPU is online state. When we either boot with maxcpus=x and then boot the other cpus by echoing to sysfs online file, these entries should be created and destroyed when CPU_DEAD is notified. Same treatement as cache entries under sysfs. We place the processor in the lowest frequency, so hw managed P-State transitions can still work on the other threads to save power. Primary goal was to just make these directories appear/disapper dynamically. There is one in this patch i had to do, which i really dont like myself but probably best if someone handling the cpufreq infrastructure could give this code right treatment if this is not acceptable. I guess its probably good for the first cut. - Converting lock_cpu_hotplug()/unlock_cpu_hotplug() to disable/enable preempt. The locking was smack in the middle of the notification path, when the hotplug is already holding the lock. I tried another solution to avoid this so avoid taking locks if we know we are from notification path. The solution was getting very ugly and i decided this was probably good for this iteration until someone who understands cpufreq could do a better job than me. (akpm: export cpucontrol to GPL modules: drivers/cpufreq/cpufreq_stats.c now does lock_cpu_hotplug()) Signed-off-by: Ashok Raj <ashok.raj@intel.com> Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> Cc: Dave Jones <davej@codemonkey.org.uk> Cc: Zwane Mwaikambo <zwane@holomorphy.com> Cc: Greg KH <greg@kroah.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 22:59:54 +00:00
module_out:
cpufreq_debug_enable_ratelimit();
return ret;
}
/**
* __cpufreq_remove_dev - remove a CPU device
*
* Removes the cpufreq interface for a CPU device.
* Caller should already have policy_rwsem in write mode for this CPU.
* This routine frees the rwsem before returning.
*/
static int __cpufreq_remove_dev(struct sys_device *sys_dev)
{
unsigned int cpu = sys_dev->id;
unsigned long flags;
struct cpufreq_policy *data;
struct kobject *kobj;
struct completion *cmp;
#ifdef CONFIG_SMP
struct sys_device *cpu_sys_dev;
unsigned int j;
#endif
cpufreq_debug_disable_ratelimit();
dprintk("unregistering CPU %u\n", cpu);
spin_lock_irqsave(&cpufreq_driver_lock, flags);
data = per_cpu(cpufreq_cpu_data, cpu);
if (!data) {
spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
cpufreq_debug_enable_ratelimit();
unlock_policy_rwsem_write(cpu);
return -EINVAL;
}
per_cpu(cpufreq_cpu_data, cpu) = NULL;
#ifdef CONFIG_SMP
/* if this isn't the CPU which is the parent of the kobj, we
* only need to unlink, put and exit
*/
if (unlikely(cpu != data->cpu)) {
dprintk("removing link\n");
cpumask_clear_cpu(cpu, data->cpus);
spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
kobj = &sys_dev->kobj;
cpufreq_cpu_put(data);
cpufreq_debug_enable_ratelimit();
unlock_policy_rwsem_write(cpu);
sysfs_remove_link(kobj, "cpufreq");
return 0;
}
#endif
#ifdef CONFIG_SMP
#ifdef CONFIG_HOTPLUG_CPU
strncpy(per_cpu(cpufreq_cpu_governor, cpu), data->governor->name,
CPUFREQ_NAME_LEN);
#endif
/* if we have other CPUs still registered, we need to unlink them,
* or else wait_for_completion below will lock up. Clean the
* per_cpu(cpufreq_cpu_data) while holding the lock, and remove
* the sysfs links afterwards.
*/
if (unlikely(cpumask_weight(data->cpus) > 1)) {
for_each_cpu(j, data->cpus) {
if (j == cpu)
continue;
per_cpu(cpufreq_cpu_data, j) = NULL;
}
}
spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
if (unlikely(cpumask_weight(data->cpus) > 1)) {
for_each_cpu(j, data->cpus) {
if (j == cpu)
continue;
dprintk("removing link for cpu %u\n", j);
#ifdef CONFIG_HOTPLUG_CPU
strncpy(per_cpu(cpufreq_cpu_governor, j),
data->governor->name, CPUFREQ_NAME_LEN);
#endif
cpu_sys_dev = get_cpu_sysdev(j);
kobj = &cpu_sys_dev->kobj;
unlock_policy_rwsem_write(cpu);
sysfs_remove_link(kobj, "cpufreq");
lock_policy_rwsem_write(cpu);
cpufreq_cpu_put(data);
}
}
#else
spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
#endif
if (cpufreq_driver->target)
__cpufreq_governor(data, CPUFREQ_GOV_STOP);
kobj = &data->kobj;
cmp = &data->kobj_unregister;
unlock_policy_rwsem_write(cpu);
kobject_put(kobj);
/* we need to make sure that the underlying kobj is actually
* not referenced anymore by anybody before we proceed with
* unloading.
*/
dprintk("waiting for dropping of refcount\n");
wait_for_completion(cmp);
dprintk("wait complete\n");
lock_policy_rwsem_write(cpu);
if (cpufreq_driver->exit)
cpufreq_driver->exit(data);
unlock_policy_rwsem_write(cpu);
free_cpumask_var(data->related_cpus);
free_cpumask_var(data->cpus);
kfree(data);
per_cpu(cpufreq_cpu_data, cpu) = NULL;
cpufreq_debug_enable_ratelimit();
return 0;
}
static int cpufreq_remove_dev(struct sys_device *sys_dev)
{
unsigned int cpu = sys_dev->id;
int retval;
if (cpu_is_offline(cpu))
return 0;
if (unlikely(lock_policy_rwsem_write(cpu)))
BUG();
retval = __cpufreq_remove_dev(sys_dev);
return retval;
}
2006-11-22 14:55:48 +00:00
static void handle_update(struct work_struct *work)
{
2006-11-22 14:55:48 +00:00
struct cpufreq_policy *policy =
container_of(work, struct cpufreq_policy, update);
unsigned int cpu = policy->cpu;
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;
dprintk("Warning: CPU frequency out of sync: cpufreq and timing "
"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);
}
/**
* cpufreq_quick_get - get the CPU frequency (in kHz) from 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_freq = 0;
if (policy) {
ret_freq = policy->cur;
cpufreq_cpu_put(policy);
}
return ret_freq;
}
EXPORT_SYMBOL(cpufreq_quick_get);
static unsigned int __cpufreq_get(unsigned int cpu)
{
struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
unsigned int ret_freq = 0;
if (!cpufreq_driver->get)
return ret_freq;
ret_freq = cpufreq_driver->get(cpu);
if (ret_freq && policy->cur &&
!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) {
/* verify no discrepancy between actual and
saved value exists */
if (unlikely(ret_freq != policy->cur)) {
cpufreq_out_of_sync(cpu, policy->cur, ret_freq);
schedule_work(&policy->update);
}
}
return ret_freq;
}
/**
* 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)
{
unsigned int ret_freq = 0;
struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
if (!policy)
goto out;
if (unlikely(lock_policy_rwsem_read(cpu)))
goto out_policy;
ret_freq = __cpufreq_get(cpu);
unlock_policy_rwsem_read(cpu);
out_policy:
cpufreq_cpu_put(policy);
out:
return ret_freq;
}
EXPORT_SYMBOL(cpufreq_get);
/**
* cpufreq_suspend - let the low level driver prepare for suspend
*/
static int cpufreq_suspend(struct sys_device *sysdev, pm_message_t pmsg)
{
int ret = 0;
int cpu = sysdev->id;
struct cpufreq_policy *cpu_policy;
dprintk("suspending 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))
goto out;
if (cpufreq_driver->suspend) {
ret = cpufreq_driver->suspend(cpu_policy, pmsg);
[CPUFREQ] Re-enable cpufreq suspend and resume code Commit 4bc5d3413503 is broken and causes regressions: (1) cpufreq_driver->resume() and ->suspend() were only called on __powerpc__, but you could set them on all architectures. In fact, ->resume() was defined and used before the PPC-related commit 42d4dc3f4e1e complained about in 4bc5d3413503. (2) Therfore, the resume functions in acpi_cpufreq and speedstep-smi would never be called. (3) This means speedstep-smi would be unusuable after suspend or resume. The _real_ problem was calling cpufreq_driver->get() with interrupts off, but it re-enabling interrupts on some platforms. Why is ->get() necessary? Some systems like to change the CPU frequency behind our back, especially during BIOS-intensive operations like suspend or resume. If such systems also use a CPU frequency-dependant timing loop, delays might be off by large factors. Therefore, we need to ascertain as soon as possible that the CPU frequency is indeed at the speed we think it is. You can do this two ways: either setting it anew, or trying to get it. The latter is what was done, the former also has the same IRQ issue. So, let's try something different: defer the checking to after interrupts are re-enabled, by calling cpufreq_update_policy() (via schedule_work()). Timings may be off until this later stage, so let's watch out for resume regressions caused by the deferred handling of frequency changes behind the kernel's back. Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net> Signed-off-by: Dave Jones <davej@redhat.com>
2009-08-07 20:58:51 +00:00
if (ret)
printk(KERN_ERR "cpufreq: suspend failed in ->suspend "
"step on CPU %u\n", cpu_policy->cpu);
}
out:
cpufreq_cpu_put(cpu_policy);
return ret;
}
/**
* cpufreq_resume - restore proper CPU frequency handling after resume
*
* 1.) resume CPUfreq hardware support (cpufreq_driver->resume())
[CPUFREQ] Re-enable cpufreq suspend and resume code Commit 4bc5d3413503 is broken and causes regressions: (1) cpufreq_driver->resume() and ->suspend() were only called on __powerpc__, but you could set them on all architectures. In fact, ->resume() was defined and used before the PPC-related commit 42d4dc3f4e1e complained about in 4bc5d3413503. (2) Therfore, the resume functions in acpi_cpufreq and speedstep-smi would never be called. (3) This means speedstep-smi would be unusuable after suspend or resume. The _real_ problem was calling cpufreq_driver->get() with interrupts off, but it re-enabling interrupts on some platforms. Why is ->get() necessary? Some systems like to change the CPU frequency behind our back, especially during BIOS-intensive operations like suspend or resume. If such systems also use a CPU frequency-dependant timing loop, delays might be off by large factors. Therefore, we need to ascertain as soon as possible that the CPU frequency is indeed at the speed we think it is. You can do this two ways: either setting it anew, or trying to get it. The latter is what was done, the former also has the same IRQ issue. So, let's try something different: defer the checking to after interrupts are re-enabled, by calling cpufreq_update_policy() (via schedule_work()). Timings may be off until this later stage, so let's watch out for resume regressions caused by the deferred handling of frequency changes behind the kernel's back. Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net> Signed-off-by: Dave Jones <davej@redhat.com>
2009-08-07 20:58:51 +00:00
* 2.) schedule call cpufreq_update_policy() ASAP as interrupts are
* restored. It will verify that the current freq is in sync with
* what we believe it to be. This is a bit later than when it
* should be, but nonethteless it's better than calling
* cpufreq_driver->get() here which might re-enable interrupts...
*/
static int cpufreq_resume(struct sys_device *sysdev)
{
int ret = 0;
int cpu = sysdev->id;
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))
goto fail;
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);
goto fail;
}
}
schedule_work(&cpu_policy->update);
[CPUFREQ] Re-enable cpufreq suspend and resume code Commit 4bc5d3413503 is broken and causes regressions: (1) cpufreq_driver->resume() and ->suspend() were only called on __powerpc__, but you could set them on all architectures. In fact, ->resume() was defined and used before the PPC-related commit 42d4dc3f4e1e complained about in 4bc5d3413503. (2) Therfore, the resume functions in acpi_cpufreq and speedstep-smi would never be called. (3) This means speedstep-smi would be unusuable after suspend or resume. The _real_ problem was calling cpufreq_driver->get() with interrupts off, but it re-enabling interrupts on some platforms. Why is ->get() necessary? Some systems like to change the CPU frequency behind our back, especially during BIOS-intensive operations like suspend or resume. If such systems also use a CPU frequency-dependant timing loop, delays might be off by large factors. Therefore, we need to ascertain as soon as possible that the CPU frequency is indeed at the speed we think it is. You can do this two ways: either setting it anew, or trying to get it. The latter is what was done, the former also has the same IRQ issue. So, let's try something different: defer the checking to after interrupts are re-enabled, by calling cpufreq_update_policy() (via schedule_work()). Timings may be off until this later stage, so let's watch out for resume regressions caused by the deferred handling of frequency changes behind the kernel's back. Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net> Signed-off-by: Dave Jones <davej@redhat.com>
2009-08-07 20:58:51 +00:00
fail:
cpufreq_cpu_put(cpu_policy);
return ret;
}
static struct sysdev_driver cpufreq_sysdev_driver = {
.add = cpufreq_add_dev,
.remove = cpufreq_remove_dev,
.suspend = cpufreq_suspend,
.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
*
* Add a driver to one of two lists: either a list of drivers that
* 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 09:16:30 +00:00
* blocking_notifier_chain_register.
*/
int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
{
int ret;
WARN_ON(!init_cpufreq_transition_notifier_list_called);
switch (list) {
case CPUFREQ_TRANSITION_NOTIFIER:
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 09:16:30 +00:00
&cpufreq_transition_notifier_list, nb);
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 09:16:30 +00:00
ret = blocking_notifier_chain_register(
&cpufreq_policy_notifier_list, nb);
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 09:16:30 +00:00
* blocking_notifier_chain_unregister.
*/
int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
{
int ret;
switch (list) {
case CPUFREQ_TRANSITION_NOTIFIER:
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 09:16:30 +00:00
&cpufreq_transition_notifier_list, nb);
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 09:16:30 +00:00
ret = blocking_notifier_chain_unregister(
&cpufreq_policy_notifier_list, nb);
break;
default:
ret = -EINVAL;
}
return ret;
}
EXPORT_SYMBOL(cpufreq_unregister_notifier);
/*********************************************************************
* GOVERNORS *
*********************************************************************/
int __cpufreq_driver_target(struct cpufreq_policy *policy,
unsigned int target_freq,
unsigned int relation)
{
int retval = -EINVAL;
[PATCH] create and destroy cpufreq sysfs entries based on cpu notifiers cpufreq entries in sysfs should only be populated when CPU is online state. When we either boot with maxcpus=x and then boot the other cpus by echoing to sysfs online file, these entries should be created and destroyed when CPU_DEAD is notified. Same treatement as cache entries under sysfs. We place the processor in the lowest frequency, so hw managed P-State transitions can still work on the other threads to save power. Primary goal was to just make these directories appear/disapper dynamically. There is one in this patch i had to do, which i really dont like myself but probably best if someone handling the cpufreq infrastructure could give this code right treatment if this is not acceptable. I guess its probably good for the first cut. - Converting lock_cpu_hotplug()/unlock_cpu_hotplug() to disable/enable preempt. The locking was smack in the middle of the notification path, when the hotplug is already holding the lock. I tried another solution to avoid this so avoid taking locks if we know we are from notification path. The solution was getting very ugly and i decided this was probably good for this iteration until someone who understands cpufreq could do a better job than me. (akpm: export cpucontrol to GPL modules: drivers/cpufreq/cpufreq_stats.c now does lock_cpu_hotplug()) Signed-off-by: Ashok Raj <ashok.raj@intel.com> Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> Cc: Dave Jones <davej@codemonkey.org.uk> Cc: Zwane Mwaikambo <zwane@holomorphy.com> Cc: Greg KH <greg@kroah.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 22:59:54 +00:00
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);
return retval;
}
EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
int cpufreq_driver_target(struct cpufreq_policy *policy,
unsigned int target_freq,
unsigned int relation)
{
int ret = -EINVAL;
policy = cpufreq_cpu_get(policy->cpu);
if (!policy)
goto no_policy;
if (unlikely(lock_policy_rwsem_write(policy->cpu)))
goto fail;
ret = __cpufreq_driver_target(policy, target_freq, relation);
unlock_policy_rwsem_write(policy->cpu);
fail:
cpufreq_cpu_put(policy);
no_policy:
return ret;
}
EXPORT_SYMBOL_GPL(cpufreq_driver_target);
int __cpufreq_driver_getavg(struct cpufreq_policy *policy, unsigned int cpu)
{
int ret = 0;
policy = cpufreq_cpu_get(policy->cpu);
if (!policy)
return -EINVAL;
if (cpu_online(cpu) && cpufreq_driver->getavg)
ret = cpufreq_driver->getavg(policy, cpu);
cpufreq_cpu_put(policy);
return ret;
}
EXPORT_SYMBOL_GPL(__cpufreq_driver_getavg);
/*
* when "event" is CPUFREQ_GOV_LIMITS
*/
static int __cpufreq_governor(struct cpufreq_policy *policy,
unsigned int event)
{
int ret;
/* Only must be defined when default governor is known to have latency
restrictions, like e.g. conservative or ondemand.
That this is the case is already ensured in Kconfig
*/
#ifdef CONFIG_CPU_FREQ_GOV_PERFORMANCE
struct cpufreq_governor *gov = &cpufreq_gov_performance;
#else
struct cpufreq_governor *gov = NULL;
#endif
if (policy->governor->max_transition_latency &&
policy->cpuinfo.transition_latency >
policy->governor->max_transition_latency) {
if (!gov)
return -EINVAL;
else {
printk(KERN_WARNING "%s governor failed, too long"
" transition latency of HW, fallback"
" to %s governor\n",
policy->governor->name,
gov->name);
policy->governor = gov;
}
}
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)
{
int err;
if (!governor)
return -EINVAL;
mutex_lock(&cpufreq_governor_mutex);
err = -EBUSY;
if (__find_governor(governor->name) == NULL) {
err = 0;
list_add(&governor->governor_list, &cpufreq_governor_list);
}
mutex_unlock(&cpufreq_governor_mutex);
return err;
}
EXPORT_SYMBOL_GPL(cpufreq_register_governor);
void cpufreq_unregister_governor(struct cpufreq_governor *governor)
{
#ifdef CONFIG_HOTPLUG_CPU
int cpu;
#endif
if (!governor)
return;
#ifdef CONFIG_HOTPLUG_CPU
for_each_present_cpu(cpu) {
if (cpu_online(cpu))
continue;
if (!strcmp(per_cpu(cpufreq_cpu_governor, cpu), governor->name))
strcpy(per_cpu(cpufreq_cpu_governor, cpu), "\0");
}
#endif
mutex_lock(&cpufreq_governor_mutex);
list_del(&governor->governor_list);
mutex_unlock(&cpufreq_governor_mutex);
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;
memcpy(policy, cpu_policy, sizeof(struct cpufreq_policy));
cpufreq_cpu_put(cpu_policy);
return 0;
}
EXPORT_SYMBOL(cpufreq_get_policy);
/*
* data : current policy.
* policy : policy to be set.
*/
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);
memcpy(&policy->cpuinfo, &data->cpuinfo,
sizeof(struct cpufreq_cpuinfo));
cpufreq: fix obvious condition statement error The function __cpufreq_set_policy in file drivers/cpufreq/cpufreq.c has a very obvious error: if (policy->min > data->min && policy->min > policy->max) { ret = -EINVAL; goto error_out; } This condtion statement is wrong because it returns -EINVAL only if policy->min is greater than policy->max (in this case, "policy->min > data->min" is true for ever.). In fact, it should return -EINVAL as well if policy->max is less than data->min. The correct condition should be: if (policy->min > data->max || policy->max < data->min) { The following test result testifies the above conclusion: Before applying this patch: [root@yangyi-dev /]# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies 2394000 1596000 [root@yangyi-dev /]# echo 1596000 > /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq [root@yangyi-dev /]# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq 1596000 [root@yangyi-dev /]# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_min_freq 1596000 [root@yangyi-dev /]# echo "2000000" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_min_freq -bash: echo: write error: Invalid argument [root@yangyi-dev /]# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_min_freq 1596000 [root@yangyi-dev /]# echo "0" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq [root@yangyi-dev /]# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq 1596000 [root@yangyi-dev /]# echo "1595000" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq [root@yangyi-dev /]# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq 1596000 [root@yangyi-dev /]# After applying this patch: [root@yangyi-dev /]# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies 2394000 1596000 [root@yangyi-dev /]# echo 1596000 > /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq [root@yangyi-dev /]# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq 1596000 [root@yangyi-dev /]# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_min_freq 1596000 [root@localhost /]# echo "2000000" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_min_freq -bash: echo: write error: Invalid argument [root@localhost /]# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_min_freq 1596000 [root@localhost /]# echo "0" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq -bash: echo: write error: Invalid argument [root@localhost /]# echo "1595000" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq -bash: echo: write error: Invalid argument [root@localhost /]# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq 1596000 [root@localhost /]# echo "1596000" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq [root@localhost /]# echo "2394000" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq [root@localhost /]# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq 2394000 [root@localhost /] Signed-off-by: Yi Yang <yi.y.yang@intel.com> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-01-30 12:33:34 +00:00
if (policy->min > data->max || policy->max < data->min) {
ret = -EINVAL;
goto error_out;
}
/* 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 09:16:30 +00:00
blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
CPUFREQ_ADJUST, policy);
/* 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 09:16:30 +00:00
blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
CPUFREQ_INCOMPATIBLE, policy);
/* 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 09:16:30 +00:00
if (ret)
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 09:16:30 +00:00
blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
CPUFREQ_NOTIFY, policy);
data->min = policy->min;
data->max = policy->max;
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);
}
error_out:
cpufreq_debug_enable_ratelimit();
return ret;
}
/**
* 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;
if (!data) {
ret = -ENODEV;
goto no_policy;
}
if (unlikely(lock_policy_rwsem_write(cpu))) {
ret = -EINVAL;
goto fail;
}
dprintk("updating policy for CPU %u\n", cpu);
memcpy(&policy, data, sizeof(struct cpufreq_policy));
policy.min = data->user_policy.min;
policy.max = data->user_policy.max;
policy.policy = data->user_policy.policy;
policy.governor = data->user_policy.governor;
/* 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);
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);
}
}
ret = __cpufreq_set_policy(data, &policy);
unlock_policy_rwsem_write(cpu);
fail:
cpufreq_cpu_put(data);
no_policy:
return ret;
}
EXPORT_SYMBOL(cpufreq_update_policy);
static int __cpuinit cpufreq_cpu_callback(struct notifier_block *nfb,
[PATCH] create and destroy cpufreq sysfs entries based on cpu notifiers cpufreq entries in sysfs should only be populated when CPU is online state. When we either boot with maxcpus=x and then boot the other cpus by echoing to sysfs online file, these entries should be created and destroyed when CPU_DEAD is notified. Same treatement as cache entries under sysfs. We place the processor in the lowest frequency, so hw managed P-State transitions can still work on the other threads to save power. Primary goal was to just make these directories appear/disapper dynamically. There is one in this patch i had to do, which i really dont like myself but probably best if someone handling the cpufreq infrastructure could give this code right treatment if this is not acceptable. I guess its probably good for the first cut. - Converting lock_cpu_hotplug()/unlock_cpu_hotplug() to disable/enable preempt. The locking was smack in the middle of the notification path, when the hotplug is already holding the lock. I tried another solution to avoid this so avoid taking locks if we know we are from notification path. The solution was getting very ugly and i decided this was probably good for this iteration until someone who understands cpufreq could do a better job than me. (akpm: export cpucontrol to GPL modules: drivers/cpufreq/cpufreq_stats.c now does lock_cpu_hotplug()) Signed-off-by: Ashok Raj <ashok.raj@intel.com> Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> Cc: Dave Jones <davej@codemonkey.org.uk> Cc: Zwane Mwaikambo <zwane@holomorphy.com> Cc: Greg KH <greg@kroah.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 22:59:54 +00:00
unsigned long action, void *hcpu)
{
unsigned int cpu = (unsigned long)hcpu;
struct sys_device *sys_dev;
sys_dev = get_cpu_sysdev(cpu);
if (sys_dev) {
switch (action) {
case CPU_ONLINE:
case CPU_ONLINE_FROZEN:
[PATCH] create and destroy cpufreq sysfs entries based on cpu notifiers cpufreq entries in sysfs should only be populated when CPU is online state. When we either boot with maxcpus=x and then boot the other cpus by echoing to sysfs online file, these entries should be created and destroyed when CPU_DEAD is notified. Same treatement as cache entries under sysfs. We place the processor in the lowest frequency, so hw managed P-State transitions can still work on the other threads to save power. Primary goal was to just make these directories appear/disapper dynamically. There is one in this patch i had to do, which i really dont like myself but probably best if someone handling the cpufreq infrastructure could give this code right treatment if this is not acceptable. I guess its probably good for the first cut. - Converting lock_cpu_hotplug()/unlock_cpu_hotplug() to disable/enable preempt. The locking was smack in the middle of the notification path, when the hotplug is already holding the lock. I tried another solution to avoid this so avoid taking locks if we know we are from notification path. The solution was getting very ugly and i decided this was probably good for this iteration until someone who understands cpufreq could do a better job than me. (akpm: export cpucontrol to GPL modules: drivers/cpufreq/cpufreq_stats.c now does lock_cpu_hotplug()) Signed-off-by: Ashok Raj <ashok.raj@intel.com> Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> Cc: Dave Jones <davej@codemonkey.org.uk> Cc: Zwane Mwaikambo <zwane@holomorphy.com> Cc: Greg KH <greg@kroah.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 22:59:54 +00:00
cpufreq_add_dev(sys_dev);
break;
case CPU_DOWN_PREPARE:
case CPU_DOWN_PREPARE_FROZEN:
if (unlikely(lock_policy_rwsem_write(cpu)))
BUG();
__cpufreq_remove_dev(sys_dev);
[PATCH] create and destroy cpufreq sysfs entries based on cpu notifiers cpufreq entries in sysfs should only be populated when CPU is online state. When we either boot with maxcpus=x and then boot the other cpus by echoing to sysfs online file, these entries should be created and destroyed when CPU_DEAD is notified. Same treatement as cache entries under sysfs. We place the processor in the lowest frequency, so hw managed P-State transitions can still work on the other threads to save power. Primary goal was to just make these directories appear/disapper dynamically. There is one in this patch i had to do, which i really dont like myself but probably best if someone handling the cpufreq infrastructure could give this code right treatment if this is not acceptable. I guess its probably good for the first cut. - Converting lock_cpu_hotplug()/unlock_cpu_hotplug() to disable/enable preempt. The locking was smack in the middle of the notification path, when the hotplug is already holding the lock. I tried another solution to avoid this so avoid taking locks if we know we are from notification path. The solution was getting very ugly and i decided this was probably good for this iteration until someone who understands cpufreq could do a better job than me. (akpm: export cpucontrol to GPL modules: drivers/cpufreq/cpufreq_stats.c now does lock_cpu_hotplug()) Signed-off-by: Ashok Raj <ashok.raj@intel.com> Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> Cc: Dave Jones <davej@codemonkey.org.uk> Cc: Zwane Mwaikambo <zwane@holomorphy.com> Cc: Greg KH <greg@kroah.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 22:59:54 +00:00
break;
case CPU_DOWN_FAILED:
case CPU_DOWN_FAILED_FROZEN:
cpufreq_add_dev(sys_dev);
[PATCH] create and destroy cpufreq sysfs entries based on cpu notifiers cpufreq entries in sysfs should only be populated when CPU is online state. When we either boot with maxcpus=x and then boot the other cpus by echoing to sysfs online file, these entries should be created and destroyed when CPU_DEAD is notified. Same treatement as cache entries under sysfs. We place the processor in the lowest frequency, so hw managed P-State transitions can still work on the other threads to save power. Primary goal was to just make these directories appear/disapper dynamically. There is one in this patch i had to do, which i really dont like myself but probably best if someone handling the cpufreq infrastructure could give this code right treatment if this is not acceptable. I guess its probably good for the first cut. - Converting lock_cpu_hotplug()/unlock_cpu_hotplug() to disable/enable preempt. The locking was smack in the middle of the notification path, when the hotplug is already holding the lock. I tried another solution to avoid this so avoid taking locks if we know we are from notification path. The solution was getting very ugly and i decided this was probably good for this iteration until someone who understands cpufreq could do a better job than me. (akpm: export cpucontrol to GPL modules: drivers/cpufreq/cpufreq_stats.c now does lock_cpu_hotplug()) Signed-off-by: Ashok Raj <ashok.raj@intel.com> Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> Cc: Dave Jones <davej@codemonkey.org.uk> Cc: Zwane Mwaikambo <zwane@holomorphy.com> Cc: Greg KH <greg@kroah.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 22:59:54 +00:00
break;
}
}
return NOTIFY_OK;
}
static struct notifier_block __refdata cpufreq_cpu_notifier =
[PATCH] create and destroy cpufreq sysfs entries based on cpu notifiers cpufreq entries in sysfs should only be populated when CPU is online state. When we either boot with maxcpus=x and then boot the other cpus by echoing to sysfs online file, these entries should be created and destroyed when CPU_DEAD is notified. Same treatement as cache entries under sysfs. We place the processor in the lowest frequency, so hw managed P-State transitions can still work on the other threads to save power. Primary goal was to just make these directories appear/disapper dynamically. There is one in this patch i had to do, which i really dont like myself but probably best if someone handling the cpufreq infrastructure could give this code right treatment if this is not acceptable. I guess its probably good for the first cut. - Converting lock_cpu_hotplug()/unlock_cpu_hotplug() to disable/enable preempt. The locking was smack in the middle of the notification path, when the hotplug is already holding the lock. I tried another solution to avoid this so avoid taking locks if we know we are from notification path. The solution was getting very ugly and i decided this was probably good for this iteration until someone who understands cpufreq could do a better job than me. (akpm: export cpucontrol to GPL modules: drivers/cpufreq/cpufreq_stats.c now does lock_cpu_hotplug()) Signed-off-by: Ashok Raj <ashok.raj@intel.com> Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> Cc: Dave Jones <davej@codemonkey.org.uk> Cc: Zwane Mwaikambo <zwane@holomorphy.com> Cc: Greg KH <greg@kroah.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-10-30 22:59:54 +00:00
{
.notifier_call = cpufreq_cpu_callback,
};
/*********************************************************************
* 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.
*
* Registers a CPU Frequency driver to this core code. This code
* returns zero on success, -EBUSY when another driver got here first
* (and isn't unregistered in the meantime).
*
*/
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_cpu_ids; i++)
if (cpu_possible(i) && per_cpu(cpufreq_cpu_data, i)) {
ret = 0;
break;
}
/* 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) {
register_hotcpu_notifier(&cpufreq_cpu_notifier);
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
*
* Unregister the current CPUFreq driver. Only call this if you have
* 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);
unregister_hotcpu_notifier(&cpufreq_cpu_notifier);
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);
static int __init cpufreq_core_init(void)
{
int cpu;
for_each_possible_cpu(cpu) {
per_cpu(cpufreq_policy_cpu, cpu) = -1;
init_rwsem(&per_cpu(cpu_policy_rwsem, cpu));
}
cpufreq_global_kobject = kobject_create_and_add("cpufreq",
&cpu_sysdev_class.kset.kobj);
BUG_ON(!cpufreq_global_kobject);
return 0;
}
core_initcall(cpufreq_core_init);