CPU hotplug and suspend
Subsystems (20)
Hi,
Currently, we use the CPU hotplug to disable nonboot CPUs in the suspend code paths, but with the recent change of code ordering (ie. nonboot CPUs are disabled after freezing tasks _and_ devices) it has become quite troublesome. The reason of this is that there are some CPU hotplug notifiers registered and called on each run of cpu_up()/cpu_down() that assume the system to be fully functional, which is not the case during the suspend. Moreover, at least some of them do things that are not really necessary for disabling or enabling the nonboot CPUs.
For example, it doesn't seem to be necessary to stop worker threads bound to the nonboot CPUs when they are disabled, because these CPUs most likely reappear during the resume. This particular problem has caused us to make all workqueus nonfreezable, although at least some of them should be freezable, as far as the suspend is concerned.
The advantage of using the CPU hotplug (in its current form) for suspending is that if some CPUs don't reappear during the resume, we are safe. Still, I think it would be more appropriate, and simpler in the long run, to notify the interested subsystems _only_ if one (or more) CPUs are not functional after the resume. In fact, with the current code ordering the subsystems don't even need to know that we have disabled and enabled the nonboot CPUs, unless something goes wrong.
For this reason, I'd like to change the suspend code to use a simplified CPU management, sharing some low-level code with the current CPU hotplug, that won't call all of the CPU hotplug notifiers at all, but will be able to call some other special notifiers in case one (or more) of the nonboot CPUs cannot be enabled. Of course, that would require the subsystems to register separate CPU notifiers for the resume, but I think they may share some code with the current CPU hotplug notifiers.
It seems to me that we should separate the special case of suspend from the "run-time" CPU hotplug, or things will get more and more complicated over time. Still, that would be quite radical redesign, so I'm not sure if it's generally acceptable.
Please advise.
Greetings,
Rafael
Quoted 33 lines · Rafael J. Wysocki
Hi,
Currently, we use the CPU hotplug to disable nonboot CPUs in the suspend code paths, but with the recent change of code ordering (ie. nonboot CPUs are disabled after freezing tasks _and_ devices) it has become quite troublesome. The reason of this is that there are some CPU hotplug notifiers registered and called on each run of cpu_up()/cpu_down() that assume the system to be fully functional, which is not the case during the suspend. Moreover, at least some of them do things that are not really necessary for disabling or enabling the nonboot CPUs.
For example, it doesn't seem to be necessary to stop worker threads bound to the nonboot CPUs when they are disabled, because these CPUs most likely reappear during the resume. This particular problem has caused us to make all workqueus nonfreezable, although at least some of them should be freezable, as far as the suspend is concerned.
The advantage of using the CPU hotplug (in its current form) for suspending is that if some CPUs don't reappear during the resume, we are safe. Still, I think it would be more appropriate, and simpler in the long run, to notify the interested subsystems _only_ if one (or more) CPUs are not functional after the resume. In fact, with the current code ordering the subsystems don't even need to know that we have disabled and enabled the nonboot CPUs, unless something goes wrong.
For this reason, I'd like to change the suspend code to use a simplified CPU management, sharing some low-level code with the current CPU hotplug, that won't call all of the CPU hotplug notifiers at all, but will be able to call some other special notifiers in case one (or more) of the nonboot CPUs cannot be enabled. Of course, that would require the subsystems to register separate CPU notifiers for the resume, but I think they may share some code with the current CPU hotplug notifiers.
Quoted 4 lines
It seems to me that we should separate the special case of suspend from the "run-time" CPU hotplug, or things will get more and more complicated over time. Still, that would be quite radical redesign, so I'm not sure if it's generally acceptable.
My two cents.
cpu hotplug up/down semantics with respect to irqs do not appear to be implementable in a race free way on x86, with the current generation of hardware.
The suspend/resume semantics for disabling irqs seem perfect reasonable. (We tell the drivers to shut off the irqs before we even consider migrating or turning them off).
We already have suspend/resume callbacks for everything else, why not some of the core subsystems.
Because of the global disable nature of suspend/resume I suspect it is actually easier to implement a suspend/resume callback.
If we are only talking core subsystems here there is much less of problem for duplicating functions then if this was at the driver level because core subsystems everyone uses, so there are more eyse on the code.
cpu hotplug is still CONFIG_EXPERIMENTAL suspend/resume is not.
Eric
Hi!
Quoted 8 lines
Currently, we use the CPU hotplug to disable nonboot CPUs in the suspend code paths, but with the recent change of code ordering (ie. nonboot CPUs are disabled after freezing tasks _and_ devices) it has become quite troublesome. The reason of this is that there are some CPU hotplug notifiers registered and called on each run of cpu_up()/cpu_down() that assume the system to be fully functional, which is not the case during the suspend. Moreover, at least some of them do things that are not really necessary for disabling or enabling the nonboot CPUs.
Right.
Quoted 5 lines
The advantage of using the CPU hotplug (in its current form) for suspending is that if some CPUs don't reappear during the resume, we are safe. Still, I think it would be more appropriate, and simpler in the long run, to notify the interested subsystems _only_ if one (or more) CPUs are not functional after the resume.
I'm afraid that adding 'cpu not there so simulate unplug' path will make it complex, and prone to failure, as _noone_ is going to test it.
Pavel
Signature · 2 lines
(english) http://www.livejournal.com/~pavelmachek (cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
Hi,
Quoted 11 lines · Pavel Machek
Quoted 8 lines
Currently, we use the CPU hotplug to disable nonboot CPUs in the suspend code paths, but with the recent change of code ordering (ie. nonboot CPUs are disabled after freezing tasks _and_ devices) it has become quite troublesome. The reason of this is that there are some CPU hotplug notifiers registered and called on each run of cpu_up()/cpu_down() that assume the system to be fully functional, which is not the case during the suspend. Moreover, at least some of them do things that are not really necessary for disabling or enabling the nonboot CPUs.
Right.
Quoted 8 lines
Quoted 5 lines
The advantage of using the CPU hotplug (in its current form) for suspending is that if some CPUs don't reappear during the resume, we are safe. Still, I think it would be more appropriate, and simpler in the long run, to notify the interested subsystems _only_ if one (or more) CPUs are not functional after the resume.
I'm afraid that adding 'cpu not there so simulate unplug' path will make it complex, and prone to failure, as _noone_ is going to test it.
Does it mean you think we should stick with the current approach and sort out all issues as they show up, or should we go for not using the CPU hotplug for suspending without implementing the 'cpu not there so simulate unplug' path at all (eg. we can fail the resume instead)?
Rafael
Hi!
Quoted 17 lines
Quoted 17 lines
Quoted 8 lines
Currently, we use the CPU hotplug to disable nonboot CPUs in the suspend code paths, but with the recent change of code ordering (ie. nonboot CPUs are disabled after freezing tasks _and_ devices) it has become quite troublesome. The reason of this is that there are some CPU hotplug notifiers registered and called on each run of cpu_up()/cpu_down() that assume the system to be fully functional, which is not the case during the suspend. Moreover, at least some of them do things that are not really necessary for disabling or enabling the nonboot CPUs.
Right.
Quoted 5 lines
The advantage of using the CPU hotplug (in its current form) for suspending is that if some CPUs don't reappear during the resume, we are safe. Still, I think it would be more appropriate, and simpler in the long run, to notify the interested subsystems _only_ if one (or more) CPUs are not functional after the resume.
Quoted 7 lines
Quoted 2 lines
I'm afraid that adding 'cpu not there so simulate unplug' path will make it complex, and prone to failure, as _noone_ is going to test it.
Does it mean you think we should stick with the current approach and sort out all issues as they show up, or should we go for not using the CPU hotplug for suspending without implementing the 'cpu not there so simulate unplug' path at all (eg. we can fail the resume instead)?
I'd fix the current approach, but...
Pavel
Signature · 2 lines
(english) http://www.livejournal.com/~pavelmachek (cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
RFC PATCH 1/1 CPU hotplug with frozen tasks
In reply to Rafael J. WysockiHi,
As I said before, we have a problem with using the CPU hotplug for suspending because of the notifiers that are called from within cpu_up()/cpu_down() and (sometimes) assume that the system is fully functional.
One obvious solution of this problem would be to make the notifiers behave differently if tasks are frozen, but for this purpose we'd need to tell them that this is the case. In principle, we could do it in many different ways (eg. by using a global variable, with the help of suspend notifiers etc.), but IMO one of the cleanest methods woud be to use some special values for the notifications occuring while tasks are frozen (eg. CPU_DEAD_FROZEN instead of CPU_DEAD etc.). In that case the notifiers could react in some special ways to the "FROZEN" notfifications and that would allow us to simplify some code paths (eg. in the microcode driver).
The appended patch introduces such "FROZEN" notfifications, modifies the CPU hotplug core to use them and updates all of the users of CPU hotplug notifiers to recognize them. For now, they are treated in the same way as the corresponding "normal" notifications, but I'm going to modify the microcode driver to really use them and I believe that some other subsystems can benefit from using them as well.
The patch is totally experimental and untested, although it's been successfully compiled on x86_64 and it's main purpose is to show what exactly I mean. :-)
Comments welcome.
Greetings,
Rafael
--- Documentation/cpu-hotplug.txt | 8 ++++++-- arch/i386/kernel/cpu/intel_cacheinfo.c | 2 ++ arch/i386/kernel/cpu/mcheck/therm_throt.c | 2 ++ arch/i386/kernel/cpuid.c | 2 ++ arch/i386/kernel/microcode.c | 3 +++ arch/i386/kernel/msr.c | 2 ++ arch/ia64/kernel/palinfo.c | 2 ++ arch/ia64/kernel/salinfo.c | 2 ++ arch/ia64/kernel/topology.c | 2 ++ arch/powerpc/kernel/sysfs.c | 2 ++ arch/powerpc/mm/numa.c | 3 +++ arch/s390/appldata/appldata_base.c | 2 ++ arch/x86_64/kernel/mce.c | 2 ++ arch/x86_64/kernel/mce_amd.c | 2 ++ arch/x86_64/kernel/vsyscall.c | 2 +- block/ll_rw_blk.c | 2 +- drivers/base/topology.c | 3 +++ drivers/cpufreq/cpufreq.c | 3 +++ drivers/cpufreq/cpufreq_stats.c | 2 ++ drivers/infiniband/hw/ehca/ehca_irq.c | 6 ++++++ drivers/kvm/kvm_main.c | 3 +++ fs/buffer.c | 2 +- fs/xfs/xfs_mount.c | 3 +++ include/linux/notifier.h | 9 +++++++++ kernel/cpu.c | 26 ++++++++++++++------------ kernel/hrtimer.c | 2 ++ kernel/profile.c | 4 ++++ kernel/rcupdate.c | 2 ++ kernel/relay.c | 2 ++ kernel/sched.c | 10 ++++++++++ kernel/softirq.c | 4 ++++ kernel/softlockup.c | 4 ++++ kernel/timer.c | 2 ++ kernel/workqueue.c | 6 ++++++ lib/radix-tree.c | 2 +- mm/page_alloc.c | 5 ++++- mm/slab.c | 6 ++++++ mm/swap.c | 2 +- mm/vmscan.c | 2 +- mm/vmstat.c | 3 +++ net/core/dev.c | 2 +- net/core/flow.c | 2 +- net/iucv/iucv.c | 6 ++++++ 43 files changed, 140 insertions(+), 23 deletions(-)
include/linux/notifier.h +9 -0
@@ -186,6 +186,15 @@ extern int srcu_notifier_call_chain(stru#define CPU_DOWN_PREPARE 0x0005/* CPU (unsigned)v going down */ #define CPU_DOWN_FAILED 0x0006/* CPU (unsigned)v NOT going down */ #define CPU_DEAD 0x0007/* CPU (unsigned)v dead */ +/* The following values are used for CPU hotplug events occuring while tasks are +* frozen (eg. during a suspend) +*/ +#define CPU_ONLINE_FROZEN 0x000A/* CPU (unsigned)v is up */ +#define CPU_UP_PREPARE_FROZEN 0x000B/* CPU (unsigned)v coming up */ +#define CPU_UP_CANCELED_FROZEN 0x000C/* CPU (unsigned)v NOT coming up */ +#define CPU_DOWN_PREPARE_FROZEN 0x000D/* CPU (unsigned)v going down */ +#define CPU_DOWN_FAILED_FROZEN 0x000E/* CPU (unsigned)v NOT going down */ +#define CPU_DEAD_FROZEN 0x000F/* CPU (unsigned)v dead */ #endif /* __KERNEL__ */ #endif /* _LINUX_NOTIFIER_H */
kernel/cpu.c +14 -12
@@ -120,11 +120,12 @@ static int take_cpu_down(void *unused) }/* Requires cpu_add_remove_lock to be held */ -static int _cpu_down (unsigned int cpu) +static int _cpu_down (unsigned int cpu,int tasks_frozen) {int err;struct task_struct *p;cpumask_t old_allowed, tmp; +unsigned long mod = tasks_frozen ?0x0008 :0 ;if (num_online_cpus () ==1 )return -EBUSY ;
@@ -132,7 +133,7 @@ static int _cpu_down(unsigned int cpu)if (!cpu_online (cpu))return -EINVAL ; - err =raw_notifier_call_chain (&cpu_chain,CPU_DOWN_PREPARE , + err =raw_notifier_call_chain (&cpu_chain,CPU_DOWN_PREPARE | mod, (void *)(long )cpu);if (err ==NOTIFY_BAD ) { printk("%s: attempt to take down CPU %u failed\n",
@@ -152,7 +153,7 @@ static int _cpu_down(unsigned int cpu)if (IS_ERR (p) ||cpu_online (cpu)) {/* CPU didn't die: tell everyone. Can't complain. */ -if (raw_notifier_call_chain (&cpu_chain,CPU_DOWN_FAILED , +if (raw_notifier_call_chain (&cpu_chain,CPU_DOWN_FAILED | mod, (void *)(long )cpu) ==NOTIFY_BAD )BUG ();
@@ -175,7 +176,7 @@ static int _cpu_down(unsigned int cpu)put_cpu ();/* CPU is completely dead: tell everyone. Too late to complain. */ -if (raw_notifier_call_chain (&cpu_chain,CPU_DEAD , +if (raw_notifier_call_chain (&cpu_chain,CPU_DEAD | mod, (void *)(long )cpu) ==NOTIFY_BAD )BUG ();
@@ -196,7 +197,7 @@ int cpu_down(unsigned int cpu)if (cpu_hotplug_disabled) err = -EBUSY ;else - err =_cpu_down (cpu); + err =_cpu_down (cpu,0 );mutex_unlock (&cpu_add_remove_lock);return err;
@@ -204,15 +205,16 @@ int cpu_down(unsigned int cpu)#endif /*CONFIG_HOTPLUG_CPU*/ /* Requires cpu_add_remove_lock to be held */ -static int __cpuinit _cpu_up (unsigned int cpu) +static int __cpuinit _cpu_up (unsigned int cpu,int tasks_frozen) {int ret;void *hcpu = (void *)(long )cpu; +unsigned long mod = tasks_frozen ?0x0008 :0 ;if (cpu_online (cpu) || !cpu_present (cpu))return -EINVAL ; - ret =raw_notifier_call_chain (&cpu_chain,CPU_UP_PREPARE , hcpu); + ret =raw_notifier_call_chain (&cpu_chain,CPU_UP_PREPARE | mod, hcpu);if (ret ==NOTIFY_BAD ) {printk ("%s: attempt to bring up CPU %u failed\n" , __FUNCTION__, cpu);
@@ -229,12 +231,12 @@ static int __cpuinit _cpu_up(unsigned inBUG_ON (!cpu_online (cpu));/* Now call notifier in preparation. */ -raw_notifier_call_chain (&cpu_chain,CPU_ONLINE , hcpu); +raw_notifier_call_chain (&cpu_chain,CPU_ONLINE | mod, hcpu); out_notify:if (ret !=0 )raw_notifier_call_chain (&cpu_chain, -CPU_UP_CANCELED , hcpu); +CPU_UP_CANCELED | mod, hcpu);return ret; }
@@ -247,7 +249,7 @@ int __cpuinit cpu_up(unsigned int cpu)if (cpu_hotplug_disabled) err = -EBUSY ;else - err =_cpu_up (cpu); + err =_cpu_up (cpu,0 );mutex_unlock (&cpu_add_remove_lock);return err;
@@ -277,7 +279,7 @@ int disable_nonboot_cpus(void)for_each_online_cpu (cpu) {if (cpu == first_cpu)continue ; - error =_cpu_down (cpu); + error =_cpu_down (cpu,1 );if (!error) {cpu_set (cpu, frozen_cpus);printk ("CPU%d is down\n" , cpu);
@@ -312,7 +314,7 @@ void enable_nonboot_cpus(void) suspend_cpu_hotplug =1 ;printk ("Enabling non-boot CPUs ...\n" );for_each_cpu_mask (cpu, frozen_cpus) { - error =_cpu_up (cpu); + error =_cpu_up (cpu,1 );if (!error) {printk ("CPU%d is up\n" , cpu);continue ;
arch/i386/kernel/cpu/intel_cacheinfo.c +2 -0
@@ -733,9 +733,11 @@ static int __cpuinit cacheinfo_cpu_callb sys_dev =get_cpu_sysdev (cpu);switch (action) {case CPU_ONLINE : +case CPU_ONLINE_FROZEN :cache_add_dev (sys_dev);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :cache_remove_dev (sys_dev);break ; }
arch/i386/kernel/cpu/mcheck/therm_throt.c +2 -0
@@ -137,10 +137,12 @@ static __cpuinit int thermal_throttle_cpmutex_lock (&therm_cpu_lock);switch (action) {case CPU_ONLINE : +case CPU_ONLINE_FROZEN : err =thermal_throttle_add_dev (sys_dev);WARN_ON (err);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :thermal_throttle_remove_dev (sys_dev);break ; }
arch/i386/kernel/cpuid.c +2 -0
@@ -169,9 +169,11 @@ static int cpuid_class_cpu_callback(struswitch (action) {case CPU_ONLINE : +case CPU_ONLINE_FROZEN :cpuid_device_create (cpu);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :device_destroy (cpuid_class,MKDEV (CPUID_MAJOR , cpu));break ; }
arch/i386/kernel/microcode.c +3 -0
@@ -775,10 +775,13 @@ mc_cpu_callback(struct notifier_block *n sys_dev =get_cpu_sysdev (cpu);switch (action) {case CPU_ONLINE : +case CPU_ONLINE_FROZEN :case CPU_DOWN_FAILED : +case CPU_DOWN_FAILED_FROZEN :mc_sysdev_add (sys_dev);break ;case CPU_DOWN_PREPARE : +case CPU_DOWN_PREPARE_FROZEN :mc_sysdev_remove (sys_dev);break ; }
arch/i386/kernel/msr.c +2 -0
@@ -251,9 +251,11 @@ static int msr_class_cpu_callback(structswitch (action) {case CPU_ONLINE : +case CPU_ONLINE_FROZEN :msr_device_create (cpu);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :device_destroy (msr_class,MKDEV (MSR_MAJOR , cpu));break ; }
arch/ia64/kernel/palinfo.c +2 -0
@@ -975,9 +975,11 @@ static int palinfo_cpu_callback(struct nswitch (action) {case CPU_ONLINE : +case CPU_ONLINE_FROZEN :create_palinfo_proc_entries (hotcpu);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :remove_palinfo_proc_entries (hotcpu);break ; }
arch/ia64/kernel/salinfo.c +2 -0
@@ -583,6 +583,7 @@ salinfo_cpu_callback(struct notifier_blostruct salinfo_data *data;switch (action) {case CPU_ONLINE : +case CPU_ONLINE_FROZEN :spin_lock_irqsave (&data_saved_lock, flags); for (i = 0, data = salinfo_data; i < ARRAY_SIZE(salinfo_data);
@@ -593,6 +594,7 @@ salinfo_cpu_callback(struct notifier_blospin_unlock_irqrestore (&data_saved_lock, flags);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :spin_lock_irqsave (&data_saved_lock, flags); for (i = 0, data = salinfo_data; i < ARRAY_SIZE(salinfo_data);
arch/ia64/kernel/topology.c +2 -0
@@ -412,9 +412,11 @@ static int __cpuinit cache_cpu_callback( sys_dev =get_cpu_sysdev (cpu);switch (action) {case CPU_ONLINE : +case CPU_ONLINE_FROZEN :cache_add_dev (sys_dev);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :cache_remove_dev (sys_dev);break ; }
arch/powerpc/kernel/sysfs.c +2 -0
@@ -341,10 +341,12 @@ static int __cpuinit sysfs_cpu_notify(stswitch (action) {case CPU_ONLINE : +case CPU_ONLINE_FROZEN :register_cpu_online (cpu);break ;#ifdef CONFIG_HOTPLUG_CPU case CPU_DEAD : +case CPU_DEAD_FROZEN :unregister_cpu_online (cpu);break ;#endif
arch/powerpc/mm/numa.c +3 -0
@@ -252,12 +252,15 @@ static int __cpuinit cpu_numa_callback(sswitch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN :numa_setup_cpu (lcpu); ret =NOTIFY_OK ;break ;#ifdef CONFIG_HOTPLUG_CPU case CPU_DEAD : +case CPU_DEAD_FROZEN :case CPU_UP_CANCELED : +case CPU_UP_CANCELED_FROZEN :unmap_cpu_from_node (lcpu);break ; ret =NOTIFY_OK ;
arch/s390/appldata/appldata_base.c +2 -0
@@ -567,9 +567,11 @@ appldata_cpu_notify(struct notifier_bloc {switch (action) {case CPU_ONLINE : +case CPU_ONLINE_FROZEN :appldata_online_cpu ((long ) hcpu);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :appldata_offline_cpu ((long ) hcpu);break ;default :
arch/x86_64/kernel/mce.c +2 -0
@@ -704,9 +704,11 @@ mce_cpu_callback(struct notifier_block *switch (action) {case CPU_ONLINE : +case CPU_ONLINE_FROZEN :mce_create_device (cpu);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :mce_remove_device (cpu);break ; }
arch/x86_64/kernel/mce_amd.c +2 -0
@@ -654,9 +654,11 @@ static int threshold_cpu_callback(structswitch (action) {case CPU_ONLINE : +case CPU_ONLINE_FROZEN :threshold_create_device (cpu);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :threshold_remove_device (cpu);break ;default :
arch/x86_64/kernel/vsyscall.c +1 -1
@@ -301,7 +301,7 @@ static int __cpuinit cpu_vsyscall_notifier(struct notifier_block *n, unsigned long action, void *arg) { long cpu = (long)arg; -if (action ==CPU_ONLINE ) +if (action ==CPU_ONLINE || action ==CPU_ONLINE_FROZEN )smp_call_function_single (cpu, cpu_vsyscall_init,NULL ,0 ,1 );return NOTIFY_DONE ; }
block/ll_rw_blk.c +1 -1
@@ -3505,7 +3505,7 @@ static int blk_cpu_notify(struct notifie* If a CPU goes away, splice its entries to the current CPU * and trigger a run of the softirq */ -if (action ==CPU_DEAD ) { +if (action ==CPU_DEAD || action ==CPU_DEAD_FROZEN ) {int cpu = (unsigned long ) hcpu;local_irq_disable ();
drivers/base/topology.c +3 -0
@@ -126,10 +126,13 @@ static int __cpuinit topology_cpu_callbaswitch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN : rc =topology_add_dev (cpu);break ;case CPU_UP_CANCELED : +case CPU_UP_CANCELED_FROZEN :case CPU_DEAD : +case CPU_DEAD_FROZEN :topology_remove_dev (cpu);break ; }
drivers/cpufreq/cpufreq.c +3 -0
@@ -1716,9 +1716,11 @@ static int cpufreq_cpu_callback(struct nif (sys_dev) {switch (action) {case CPU_ONLINE : +case CPU_ONLINE_FROZEN :cpufreq_add_dev (sys_dev);break ;case CPU_DOWN_PREPARE : +case CPU_DOWN_PREPARE_FROZEN :if (unlikely (lock_policy_rwsem_write (cpu)))BUG ();
@@ -1730,6 +1732,7 @@ static int cpufreq_cpu_callback(struct n__cpufreq_remove_dev (sys_dev);break ;case CPU_DOWN_FAILED : +case CPU_DOWN_FAILED_FROZEN :cpufreq_add_dev (sys_dev);break ; }
drivers/cpufreq/cpufreq_stats.c +2 -0
@@ -313,9 +313,11 @@ static int cpufreq_stat_cpu_callback(strswitch (action) {case CPU_ONLINE : +case CPU_ONLINE_FROZEN :cpufreq_update_policy (cpu);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :cpufreq_stats_free_table (cpu);break ; }
drivers/infiniband/hw/ehca/ehca_irq.c +6 -0
@@ -745,6 +745,7 @@ static int comp_pool_callback(struct notswitch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN :ehca_gen_dbg ("CPU: %x (CPU_PREPARE)" , cpu);if (!create_comp_task (pool, cpu)) {ehca_gen_err ("Can't create comp_task for cpu: %x" , cpu);
@@ -752,24 +753,29 @@ static int comp_pool_callback(struct not }break ;case CPU_UP_CANCELED : +case CPU_UP_CANCELED_FROZEN :ehca_gen_dbg ("CPU: %x (CPU_CANCELED)" , cpu); cct =per_cpu_ptr (pool->cpu_comp_tasks, cpu);kthread_bind (cct->task,any_online_cpu (cpu_online_map));destroy_comp_task (pool, cpu);break ;case CPU_ONLINE : +case CPU_ONLINE_FROZEN :ehca_gen_dbg ("CPU: %x (CPU_ONLINE)" , cpu); cct =per_cpu_ptr (pool->cpu_comp_tasks, cpu);kthread_bind (cct->task, cpu);wake_up_process (cct->task);break ;case CPU_DOWN_PREPARE : +case CPU_DOWN_PREPARE_FROZEN :ehca_gen_dbg ("CPU: %x (CPU_DOWN_PREPARE)" , cpu);break ;case CPU_DOWN_FAILED : +case CPU_DOWN_FAILED_FROZEN :ehca_gen_dbg ("CPU: %x (CPU_DOWN_FAILED)" , cpu);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :ehca_gen_dbg ("CPU: %x (CPU_DEAD)" , cpu);destroy_comp_task (pool, cpu);take_over_work (pool, cpu);
drivers/kvm/kvm_main.c +3 -0
@@ -2363,7 +2363,9 @@ static int kvm_cpu_hotplug(struct notifiswitch (val) {case CPU_DOWN_PREPARE : +case CPU_DOWN_PREPARE_FROZEN :case CPU_UP_CANCELED : +case CPU_UP_CANCELED_FROZEN :printk (KERN_INFO "kvm: disabling virtualization on CPU%d\n" , cpu);decache_vcpus_on_cpu (cpu);
@@ -2371,6 +2373,7 @@ static int kvm_cpu_hotplug(struct notifi NULL, 0, 1);break ;case CPU_ONLINE : +case CPU_ONLINE_FROZEN :printk (KERN_INFO "kvm: enabling virtualization on CPU%d\n" , cpu); smp_call_function_single(cpu, kvm_arch_ops->hardware_enable,
fs/buffer.c +1 -1
@@ -2994,7 +2994,7 @@ static void buffer_exit_cpu(int cpu)static int buffer_cpu_notify (struct notifier_block *self,unsigned long action,void *hcpu) { -if (action ==CPU_DEAD ) +if (action ==CPU_DEAD || action ==CPU_DEAD_FROZEN )buffer_exit_cpu ((unsigned long )hcpu);return NOTIFY_OK ; }
fs/xfs/xfs_mount.c +3 -0
@@ -1734,11 +1734,13 @@ xfs_icsb_cpu_notify(per_cpu_ptr (mp->m_sb_cnts, (unsigned long )hcpu);switch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN :/* Easy Case - initialize the area and locks, and * then rebalance when online does everything else for us. */ memset (cntp,0 ,sizeof (xfs_icsb_cnts_t));break ;case CPU_ONLINE : +case CPU_ONLINE_FROZEN :xfs_icsb_lock (mp);xfs_icsb_balance_counter (mp,XFS_SBS_ICOUNT ,0 ,0 );xfs_icsb_balance_counter (mp,XFS_SBS_IFREE ,0 ,0 );
@@ -1746,6 +1748,7 @@ xfs_icsb_cpu_notify(xfs_icsb_unlock (mp);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :/* Disable all the counters, then fold the dead cpu's * count into the total on the global superblock and * re-enable the counters. */
kernel/hrtimer.c +2 -0
@@ -1395,11 +1395,13 @@ static int __cpuinit hrtimer_cpu_notify(switch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN :init_hrtimers_cpu (cpu);break ;#ifdef CONFIG_HOTPLUG_CPU case CPU_DEAD : +case CPU_DEAD_FROZEN :clockevents_notify (CLOCK_EVT_NOTIFY_CPU_DEAD , &cpu);migrate_hrtimers (cpu);break ;
kernel/profile.c +4 -0
@@ -340,6 +340,7 @@ static int __devinit profile_cpu_callbacswitch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN : node =cpu_to_node (cpu);per_cpu (cpu_profile_flip, cpu) =0 ;if (!per_cpu (cpu_profile_hits, cpu)[1 ]) {
@@ -365,10 +366,13 @@ static int __devinit profile_cpu_callbac__free_page (page);return NOTIFY_BAD ;case CPU_ONLINE : +case CPU_ONLINE_FROZEN :cpu_set (cpu, prof_cpu_mask);break ;case CPU_UP_CANCELED : +case CPU_UP_CANCELED_FROZEN :case CPU_DEAD : +case CPU_DEAD_FROZEN :cpu_clear (cpu, prof_cpu_mask);if (per_cpu (cpu_profile_hits, cpu)[0 ]) { page =virt_to_page (per_cpu (cpu_profile_hits, cpu)[0 ]);
kernel/rcupdate.c +2 -0
@@ -558,9 +558,11 @@ static int __cpuinit rcu_cpu_notify(strulong cpu = (long )hcpu;switch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN :rcu_online_cpu (cpu);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :rcu_offline_cpu (cpu);break ;default :
kernel/relay.c +2 -0
@@ -490,6 +490,7 @@ static int __cpuinit relay_hotcpu_callbaswitch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN :mutex_lock (&relay_channels_mutex);list_for_each_entry (chan, &relay_channels, list) {if (chan->buf[hotcpu])
@@ -506,6 +507,7 @@ static int __cpuinit relay_hotcpu_callbamutex_unlock (&relay_channels_mutex);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :/* No need to flush the cpu : will be flushed upon * final relay_flush() call. */ break ;
kernel/sched.c +10 -0
@@ -5192,6 +5192,7 @@ migration_call(struct notifier_block *nfswitch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN : p =kthread_create (migration_thread, hcpu,"migration/%d" ,cpu);if (IS_ERR (p))return NOTIFY_BAD ;
@@ -5205,12 +5206,14 @@ migration_call(struct notifier_block *nfbreak ;case CPU_ONLINE : +case CPU_ONLINE_FROZEN :/* Strictly unneccessary, as first user will wake it. */ wake_up_process (cpu_rq (cpu)->migration_thread);break ;#ifdef CONFIG_HOTPLUG_CPU case CPU_UP_CANCELED : +case CPU_UP_CANCELED_FROZEN :if (!cpu_rq (cpu)->migration_thread)break ;/* Unbind it from offline cpu so it can run. Fall thru. */
@@ -5221,6 +5224,7 @@ migration_call(struct notifier_block *nfbreak ;case CPU_DEAD : +case CPU_DEAD_FROZEN :migrate_live_tasks (cpu); rq =cpu_rq (cpu);kthread_stop (rq->migration_thread);
@@ -6702,14 +6706,20 @@ static int update_sched_domains(struct n {switch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN :case CPU_DOWN_PREPARE : +case CPU_DOWN_PREPARE_FROZEN :detach_destroy_domains (&cpu_online_map);return NOTIFY_OK ;case CPU_UP_CANCELED : +case CPU_UP_CANCELED_FROZEN :case CPU_DOWN_FAILED : +case CPU_DOWN_FAILED_FROZEN :case CPU_ONLINE : +case CPU_ONLINE_FROZEN :case CPU_DEAD : +case CPU_DEAD_FROZEN :/* * Fall through and re-initialise the domains. */
kernel/softirq.c +4 -0
@@ -593,6 +593,7 @@ static int __cpuinit cpu_callback(structswitch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN : p =kthread_create (ksoftirqd, hcpu,"ksoftirqd/%d" , hotcpu);if (IS_ERR (p)) {printk ("ksoftirqd for %i failed\n" , hotcpu);
@@ -602,16 +603,19 @@ static int __cpuinit cpu_callback(structper_cpu (ksoftirqd, hotcpu) = p;break ;case CPU_ONLINE : +case CPU_ONLINE_FROZEN :wake_up_process (per_cpu (ksoftirqd, hotcpu));break ;#ifdef CONFIG_HOTPLUG_CPU case CPU_UP_CANCELED : +case CPU_UP_CANCELED_FROZEN :if (!per_cpu (ksoftirqd, hotcpu))break ;/* Unbind so it can run. Fall thru. */ kthread_bind (per_cpu (ksoftirqd, hotcpu),any_online_cpu (cpu_online_map));case CPU_DEAD : +case CPU_DEAD_FROZEN : p =per_cpu (ksoftirqd, hotcpu);per_cpu (ksoftirqd, hotcpu) =NULL ;kthread_stop (p);
kernel/softlockup.c +4 -0
@@ -112,6 +112,7 @@ cpu_callback(struct notifier_block *nfb,switch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN :BUG_ON (per_cpu (watchdog_task, hotcpu)); p =kthread_create (watchdog, hcpu,"watchdog/%d" , hotcpu);if (IS_ERR (p)) {
@@ -123,16 +124,19 @@ cpu_callback(struct notifier_block *nfb,kthread_bind (p, hotcpu);break ;case CPU_ONLINE : +case CPU_ONLINE_FROZEN :wake_up_process (per_cpu (watchdog_task, hotcpu));break ;#ifdef CONFIG_HOTPLUG_CPU case CPU_UP_CANCELED : +case CPU_UP_CANCELED_FROZEN :if (!per_cpu (watchdog_task, hotcpu))break ;/* Unbind so it can run. Fall thru. */ kthread_bind (per_cpu (watchdog_task, hotcpu),any_online_cpu (cpu_online_map));case CPU_DEAD : +case CPU_DEAD_FROZEN : p =per_cpu (watchdog_task, hotcpu);per_cpu (watchdog_task, hotcpu) =NULL ;kthread_stop (p);
kernel/timer.c +2 -0
@@ -1699,11 +1699,13 @@ static int __cpuinit timer_cpu_notify(stlong cpu = (long )hcpu;switch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN :if (init_timers_cpu (cpu) <0 )return NOTIFY_BAD ;break ;#ifdef CONFIG_HOTPLUG_CPU case CPU_DEAD : +case CPU_DEAD_FROZEN :migrate_timers (cpu);break ;#endif
kernel/workqueue.c +6 -0
@@ -757,6 +757,7 @@ static int __devinit workqueue_cpu_callbswitch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN :mutex_lock (&workqueue_mutex);/* Create a new workqueue thread for it. */ list_for_each_entry (wq, &workqueues, list) {
@@ -768,6 +769,7 @@ static int __devinit workqueue_cpu_callbbreak ;case CPU_ONLINE : +case CPU_ONLINE_FROZEN :/* Kick off worker threads. */ list_for_each_entry (wq, &workqueues, list) {struct cpu_workqueue_struct *cwq;
@@ -780,6 +782,7 @@ static int __devinit workqueue_cpu_callbbreak ;case CPU_UP_CANCELED : +case CPU_UP_CANCELED_FROZEN :list_for_each_entry (wq, &workqueues, list) {if (!per_cpu_ptr (wq->cpu_wq, hotcpu)->thread)continue ;
@@ -792,14 +795,17 @@ static int __devinit workqueue_cpu_callbbreak ;case CPU_DOWN_PREPARE : +case CPU_DOWN_PREPARE_FROZEN :mutex_lock (&workqueue_mutex);break ;case CPU_DOWN_FAILED : +case CPU_DOWN_FAILED_FROZEN :mutex_unlock (&workqueue_mutex);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :list_for_each_entry (wq, &workqueues, list)cleanup_workqueue_thread (wq, hotcpu);list_for_each_entry (wq, &workqueues, list)
lib/radix-tree.c +1 -1
@@ -1004,7 +1004,7 @@ static int radix_tree_callback(struct nostruct radix_tree_preload *rtp;/* Free per-cpu pool of perloaded nodes */ -if (action ==CPU_DEAD ) { +if (action ==CPU_DEAD || action ==CPU_DEAD_FROZEN ) { rtp = &per_cpu (radix_tree_preloads, cpu);while (rtp->nr) { kmem_cache_free(radix_tree_node_cachep,
mm/page_alloc.c +4 -1
@@ -2143,11 +2143,14 @@ static int __cpuinit pageset_cpuup_callbswitch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN :if (process_zones (cpu)) ret =NOTIFY_BAD ;break ;case CPU_UP_CANCELED : +case CPU_UP_CANCELED_FROZEN :case CPU_DEAD : +case CPU_DEAD_FROZEN :free_zone_pagesets (cpu);break ;default :
@@ -3007,7 +3010,7 @@ static int page_alloc_cpu_notify(struct {int cpu = (unsigned long )hcpu; -if (action ==CPU_DEAD ) { +if (action ==CPU_DEAD || action ==CPU_DEAD_FROZEN ) {local_irq_disable ();__drain_pages (cpu);vm_events_fold_cpu (cpu);
mm/swap.c +1 -1
@@ -488,7 +488,7 @@ static int cpu_swap_callback(struct notilong *committed; committed = &per_cpu (committed_space, (long )hcpu); -if (action ==CPU_DEAD ) { +if (action ==CPU_DEAD || action ==CPU_DEAD_FROZEN ) {atomic_add (*committed, &vm_committed_space); *committed =0 ;__lru_add_drain ((long )hcpu);
mm/vmscan.c +1 -1
@@ -1527,7 +1527,7 @@ static int __devinit cpu_callback(structpg_data_t *pgdat;cpumask_t mask; -if (action ==CPU_ONLINE ) { +if (action ==CPU_ONLINE || action ==CPU_ONLINE_FROZEN ) {for_each_online_pgdat (pgdat) { mask =node_to_cpumask (pgdat->node_id);if (any_online_cpu (mask) !=NR_CPUS )
mm/vmstat.c +3 -0
@@ -650,8 +650,11 @@ static int __cpuinit vmstat_cpuup_callba {switch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN :case CPU_UP_CANCELED : +case CPU_UP_CANCELED_FROZEN :case CPU_DEAD : +case CPU_DEAD_FROZEN :refresh_zone_stat_thresholds ();break ;default :
net/core/dev.c +1 -1
@@ -3344,7 +3344,7 @@ static int dev_cpu_callback(struct notifunsigned int cpu, oldcpu = (unsigned long )ocpu;struct softnet_data *sd, *oldsd; -if (action !=CPU_DEAD ) +if (action !=CPU_DEAD && action !=CPU_DEAD_FROZEN )return NOTIFY_OK ;local_irq_disable ();
net/core/flow.c +1 -1
@@ -338,7 +338,7 @@ static int flow_cache_cpu(struct notifie unsigned long action, void *hcpu) { -if (action ==CPU_DEAD ) +if (action ==CPU_DEAD || action ==CPU_DEAD_FROZEN )__flow_cache_shrink ((unsigned long )hcpu,0 );return NOTIFY_OK ; }
net/iucv/iucv.c +6 -0
@@ -528,6 +528,7 @@ static int __cpuinit iucv_cpu_notify(strswitch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN :if (!percpu_populate (iucv_irq_data,sizeof (struct iucv_irq_data ),GFP_KERNEL |GFP_DMA , cpu))
@@ -539,15 +540,20 @@ static int __cpuinit iucv_cpu_notify(str }break ;case CPU_UP_CANCELED : +case CPU_UP_CANCELED_FROZEN :case CPU_DEAD : +case CPU_DEAD_FROZEN :percpu_depopulate (iucv_param, cpu);percpu_depopulate (iucv_irq_data, cpu);break ;case CPU_ONLINE : +case CPU_ONLINE_FROZEN :case CPU_DOWN_FAILED : +case CPU_DOWN_FAILED_FROZEN :smp_call_function_on (iucv_declare_cpu,NULL ,0 ,1 , cpu);break ;case CPU_DOWN_PREPARE : +case CPU_DOWN_PREPARE_FROZEN : cpumask = iucv_buffer_cpumask;cpu_clear (cpu, cpumask);if (cpus_empty (cpumask))
Documentation/cpu-hotplug.txt +6 -2
@@ -217,14 +217,16 @@ Q: What happens when a CPU is being logi A: The following happen, listed in no particular order :-) - A notification is sent to in-kernel registered modules by sending an event - CPU_DOWN_PREPARE + CPU_DOWN_PREPARE or CPU_DOWN_PREPARE_FROZEN, depending on whether or not the + CPU is being offlined while tasks are frozen (eg. during a suspend) - All process is migrated away from this outgoing CPU to a new CPU - All interrupts targeted to this CPU is migrated to a new CPU - timers/bottom half/task lets are also migrated to a new CPU - Once all services are migrated, kernel calls an arch specific routine __cpu_disable() to perform arch specific cleanup. - Once this is successful, an event for successful cleanup is sent by an event - CPU_DEAD. + CPU_DEAD (or CPU_DEAD_FROZEN if tasks are frozen while the CPU is being + offlined). "It is expected that each service cleans up when the CPU_DOWN_PREPARE notifier is called, when CPU_DEAD is called its expected there is nothing
@@ -242,9 +244,11 @@ A: This is what you would need in your k switch (action) { case CPU_ONLINE: + case CPU_ONLINE_FROZEN: foobar_online_action(cpu); break; case CPU_DEAD: + case CPU_DEAD_FROZEN: foobar_dead_action(cpu); break; }
mm/slab.c +6 -0
@@ -1180,6 +1180,7 @@ static int __cpuinit cpuup_callback(struswitch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN :mutex_lock (&cache_chain_mutex);/* * We need to do this right in the beginning since
@@ -1266,17 +1267,21 @@ static int __cpuinit cpuup_callback(stru }break ;case CPU_ONLINE : +case CPU_ONLINE_FROZEN :mutex_unlock (&cache_chain_mutex);start_cpu_timer (cpu);break ;#ifdef CONFIG_HOTPLUG_CPU case CPU_DOWN_PREPARE : +case CPU_DOWN_PREPARE_FROZEN :mutex_lock (&cache_chain_mutex);break ;case CPU_DOWN_FAILED : +case CPU_DOWN_FAILED_FROZEN :mutex_unlock (&cache_chain_mutex);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :/* * Even if all the cpus of a node are down, we don't free the * kmem_list3 of any cache. This to avoid a race between
@@ -1288,6 +1293,7 @@ static int __cpuinit cpuup_callback(stru/* fall thru */ #endif case CPU_UP_CANCELED : +case CPU_UP_CANCELED_FROZEN :list_for_each_entry (cachep, &cache_chain, next) {struct array_cache *nc;struct array_cache *shared;
Re: [RFC][PATCH][EXPERIMENTAL] CPU hotplug with frozen tasks
Hi!
Quoted 21 lines
As I said before, we have a problem with using the CPU hotplug for suspending because of the notifiers that are called from within cpu_up()/cpu_down() and (sometimes) assume that the system is fully functional.
One obvious solution of this problem would be to make the notifiers behave differently if tasks are frozen, but for this purpose we'd need to tell them that this is the case. In principle, we could do it in many different ways (eg. by using a global variable, with the help of suspend notifiers etc.), but IMO one of the cleanest methods woud be to use some special values for the notifications occuring while tasks are frozen (eg.
CPU_DEAD_FROZENinstead ofCPU_DEADetc.). In that case the notifiers could react in some special ways to the "FROZEN" notfifications and that would allow us to simplify some code paths (eg. in the microcode driver).The appended patch introduces such "FROZEN" notfifications, modifies the CPU hotplug core to use them and updates all of the users of CPU hotplug notifiers to recognize them. For now, they are treated in the same way as the corresponding "normal" notifications, but I'm going to modify the microcode driver to really use them and I believe that some other subsystems can benefit from using them as well.
Quoted 3 lines
The patch is totally experimental and untested, although it's been successfully compiled on
x86_64and it's main purpose is to show what exactly I mean. :-)
Looks sane to me.
Quoted 6 lines
Index:
linux-2.6.21-rc6/kernel/cpu.c
===================================================================kernel/cpu.c +0
-0@@ -120,11 +120,12 @@ static int take_cpu_down(void *unused) }
kernel/cpu.c static int take_cpu_down(void *unused) · line 129
Quoted 10 lines
kernel/cpu.c +2
-1/* Requires cpu_add_remove_lock to be held */ -static int _cpu_down (unsigned int cpu) +static int _cpu_down (unsigned int cpu,int tasks_frozen) {int err;struct task_struct *p;cpumask_t old_allowed, tmp; +unsigned long mod = tasks_frozen ?0x0008 :0 ;
Can we get constant instead of 0x0008 here?
Pavel
Signature · 2 lines
(english) http://www.livejournal.com/~pavelmachek (cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
Re: [RFC][PATCH][EXPERIMENTAL] CPU hotplug with frozen tasks
Quoted 36 lines · Pavel Machek
Hi!
Quoted 21 lines
As I said before, we have a problem with using the CPU hotplug for suspending because of the notifiers that are called from within cpu_up()/cpu_down() and (sometimes) assume that the system is fully functional.
One obvious solution of this problem would be to make the notifiers behave differently if tasks are frozen, but for this purpose we'd need to tell them that this is the case. In principle, we could do it in many different ways (eg. by using a global variable, with the help of suspend notifiers etc.), but IMO one of the cleanest methods woud be to use some special values for the notifications occuring while tasks are frozen (eg.
CPU_DEAD_FROZENinstead ofCPU_DEADetc.). In that case the notifiers could react in some special ways to the "FROZEN" notfifications and that would allow us to simplify some code paths (eg. in the microcode driver).The appended patch introduces such "FROZEN" notfifications, modifies the CPU hotplug core to use them and updates all of the users of CPU hotplug notifiers to recognize them. For now, they are treated in the same way as the corresponding "normal" notifications, but I'm going to modify the microcode driver to really use them and I believe that some other subsystems can benefit from using them as well.
Quoted 3 lines
The patch is totally experimental and untested, although it's been successfully compiled on
x86_64and it's main purpose is to show what exactly I mean. :-)Looks sane to me.
Quoted 7 lines
Index:
linux-2.6.21-rc6/kernel/cpu.c
===================================================================kernel/cpu.c +0
-0@@ -120,11 +120,12 @@ static int take_cpu_down(void *unused) }
Quoted 11 lines
kernel/cpu.c static int take_cpu_down(void *unused) · line 129
Quoted 9 lines
kernel/cpu.c +2
-1/* Requires cpu_add_remove_lock to be held */ -static int _cpu_down (unsigned int cpu) +static int _cpu_down (unsigned int cpu,int tasks_frozen) {int err;struct task_struct *p;cpumask_t old_allowed, tmp; +unsigned long mod = tasks_frozen ?0x0008 :0 ;Can we get constant instead of
0x0008here?
Sure. Updated patch is in the reply to Gautham.
Greetings,
Rafael
Re: [RFC][PATCH][EXPERIMENTAL] CPU hotplug with frozen tasks
In reply to Rafael J. WysockiHi Rafael,
Quoted 6 lines · Rafael J. Wysocki
Hi,
As I said before, we have a problem with using the CPU hotplug for suspending because of the notifiers that are called from within cpu_up()/cpu_down() and (sometimes) assume that the system is fully functional.
Right. In order to use freezer for CPU hotplug, we need to perform that audit anyway.
Quoted 10 lines
One obvious solution of this problem would be to make the notifiers behave differently if tasks are frozen, but for this purpose we'd need to tell them that this is the case. In principle, we could do it in many different ways (eg. by using a global variable, with the help of suspend notifiers etc.), but IMO one of the cleanest methods woud be to use some special values for the notifications occuring while tasks are frozen (eg.
CPU_DEAD_FROZENinstead ofCPU_DEADetc.). In that case the notifiers could react in some special ways to the "FROZEN" notfifications and that would allow us to simplify some code paths (eg. in the microcode driver).
Agreed.
Quoted 7 lines
The appended patch introduces such "FROZEN" notfifications, modifies the CPU hotplug core to use them and updates all of the users of CPU hotplug notifiers to recognize them. For now, they are treated in the same way as the corresponding "normal" notifications, but I'm going to modify the microcode driver to really use them and I believe that some other subsystems can benefit from using them as well.
Ok. A minor doubt.
When you say FROZEN, do you mean frozen due to suspend ? If yes, then it makes sense. Otherwise once cpu-hotplug starts using the freezer (hopefully it will someday soon
:-)) won't this patch become redundant ? [Except of course fixing a few glitches due to the assumption that the system is fully functional, when it's actually frozen.]
I am of the opinion that we should have notifications which help the cpu-hotplug aware
subsystems differentiate between a normal cpu-hotplug and a cpu-hotplug initiated by
suspend. Thereby they can handle it accordingly and not destroy any percpu resources
and reuse them instead during resume.
Am I missing something?
Quoted 5 lines
The patch is totally experimental and untested, although it's been successfully compiled on
x86_64and it's main purpose is to show what exactly I mean. :-)Comments welcome.
Other than that, I am ok with the patch.
Quoted 3 lines
Greetings,
Rafael
Thanks and Regards
gautham.
Re: [RFC][PATCH][EXPERIMENTAL] CPU hotplug with frozen tasks
Quoted 34 lines · Gautham Shenoy
Hi Rafael,
Quoted 6 lines · Rafael J. Wysocki
Hi,
As I said before, we have a problem with using the CPU hotplug for suspending because of the notifiers that are called from within cpu_up()/cpu_down() and (sometimes) assume that the system is fully functional.
Right. In order to use freezer for CPU hotplug, we need to perform that audit anyway.
Quoted 10 lines
One obvious solution of this problem would be to make the notifiers behave differently if tasks are frozen, but for this purpose we'd need to tell them that this is the case. In principle, we could do it in many different ways (eg. by using a global variable, with the help of suspend notifiers etc.), but IMO one of the cleanest methods woud be to use some special values for the notifications occuring while tasks are frozen (eg.
CPU_DEAD_FROZENinstead ofCPU_DEADetc.). In that case the notifiers could react in some special ways to the "FROZEN" notfifications and that would allow us to simplify some code paths (eg. in the microcode driver).Agreed.
Quoted 7 lines
The appended patch introduces such "FROZEN" notfifications, modifies the CPU hotplug core to use them and updates all of the users of CPU hotplug notifiers to recognize them. For now, they are treated in the same way as the corresponding "normal" notifications, but I'm going to modify the microcode driver to really use them and I believe that some other subsystems can benefit from using them as well.
Quoted 3 lines
Ok. A minor doubt.
When you say FROZEN, do you mean frozen due to suspend ?
Yes.
I have modified the patch to stress that in the documentation (and in the comment in notifier.h).
Quoted 2 lines
If yes, then it makes sense. Otherwise once cpu-hotplug starts using the freezer (hopefully it will someday soon :-)) won't this patch become redundant ?
Well, I'm not entirely sure.
Quoted 11 lines
[Except of course fixing a few glitches
due to the assumption that the system is fully functional, when it's actually frozen.]I am of the opinion that we should have notifications which help the cpu-hotplug aware
subsystems differentiate between a normal cpu-hotplug and a cpu-hotplug initiated by
suspend. Thereby they can handle it accordingly and not destroy any percpu resources
and reuse them instead during resume.
[Unless, of course, the CPU in question refuses to go online during the resume.]
I agree.
Initially I thought the patch might cover some scenarios other than just the suspend, but then I changed my mind. For this reason I'm not sure if the "_FROZEN" parts of the new constants names are exactly right, but I have no better ideas anyway.
Quoted 1 line
Am I missing something?
No, I don't think so. :-)
Appended is the updated version of the patch (in addition to the changes mentioned above I've eliminated the magic constant 0x0008 from cpu.c by changing the new definitions in notifier.h).
Greetings,
Rafael
--- Documentation/cpu-hotplug.txt | 9 +++++++-- arch/i386/kernel/cpu/intel_cacheinfo.c | 2 ++ arch/i386/kernel/cpu/mcheck/therm_throt.c | 2 ++ arch/i386/kernel/cpuid.c | 2 ++ arch/i386/kernel/microcode.c | 3 +++ arch/i386/kernel/msr.c | 2 ++ arch/ia64/kernel/palinfo.c | 2 ++ arch/ia64/kernel/salinfo.c | 2 ++ arch/ia64/kernel/topology.c | 2 ++ arch/powerpc/kernel/sysfs.c | 2 ++ arch/powerpc/mm/numa.c | 3 +++ arch/s390/appldata/appldata_base.c | 2 ++ arch/x86_64/kernel/mce.c | 2 ++ arch/x86_64/kernel/mce_amd.c | 2 ++ arch/x86_64/kernel/vsyscall.c | 2 +- block/ll_rw_blk.c | 2 +- drivers/base/topology.c | 3 +++ drivers/cpufreq/cpufreq.c | 3 +++ drivers/cpufreq/cpufreq_stats.c | 2 ++ drivers/infiniband/hw/ehca/ehca_irq.c | 6 ++++++ drivers/kvm/kvm_main.c | 3 +++ fs/buffer.c | 2 +- fs/xfs/xfs_mount.c | 3 +++ include/linux/notifier.h | 12 ++++++++++++ kernel/cpu.c | 26 ++++++++++++++------------ kernel/hrtimer.c | 2 ++ kernel/profile.c | 4 ++++ kernel/rcupdate.c | 2 ++ kernel/relay.c | 2 ++ kernel/sched.c | 10 ++++++++++ kernel/softirq.c | 4 ++++ kernel/softlockup.c | 4 ++++ kernel/timer.c | 2 ++ kernel/workqueue.c | 6 ++++++ lib/radix-tree.c | 2 +- mm/page_alloc.c | 5 ++++- mm/slab.c | 6 ++++++ mm/swap.c | 2 +- mm/vmscan.c | 2 +- mm/vmstat.c | 3 +++ net/core/dev.c | 2 +- net/core/flow.c | 2 +- net/iucv/iucv.c | 6 ++++++ 43 files changed, 144 insertions(+), 23 deletions(-)
include/linux/notifier.h +12 -0
@@ -187,5 +187,17 @@ extern int srcu_notifier_call_chain(stru#define CPU_DOWN_FAILED 0x0006/* CPU (unsigned)v NOT going down */ #define CPU_DEAD 0x0007/* CPU (unsigned)v dead */ +/* Used for CPU hotplug events occuring while tasks are frozen due to a suspend +* operation in progress +*/ +#define CPU_TASKS_FROZEN 0x0008 + +#define CPU_ONLINE_FROZEN (CPU_ONLINE | CPU_TASKS_FROZEN) +#define CPU_UP_PREPARE_FROZEN (CPU_UP_PREPARE | CPU_TASKS_FROZEN) +#define CPU_UP_CANCELED_FROZEN (CPU_UP_CANCELED | CPU_TASKS_FROZEN) +#define CPU_DOWN_PREPARE_FROZEN (CPU_DOWN_PREPARE | CPU_TASKS_FROZEN) +#define CPU_DOWN_FAILED_FROZEN (CPU_DOWN_FAILED | CPU_TASKS_FROZEN) +#define CPU_DEAD_FROZEN (CPU_DEAD | CPU_TASKS_FROZEN) +#endif /* __KERNEL__ */ #endif /* _LINUX_NOTIFIER_H */
kernel/cpu.c +14 -12
@@ -120,11 +120,12 @@ static int take_cpu_down(void *unused) }/* Requires cpu_add_remove_lock to be held */ -static int _cpu_down (unsigned int cpu) +static int _cpu_down (unsigned int cpu,int tasks_frozen) {int err;struct task_struct *p;cpumask_t old_allowed, tmp; +unsigned long mod = tasks_frozen ?CPU_TASKS_FROZEN :0 ;if (num_online_cpus () ==1 )return -EBUSY ;
@@ -132,7 +133,7 @@ static int _cpu_down(unsigned int cpu)if (!cpu_online (cpu))return -EINVAL ; - err =raw_notifier_call_chain (&cpu_chain,CPU_DOWN_PREPARE , + err =raw_notifier_call_chain (&cpu_chain,CPU_DOWN_PREPARE | mod, (void *)(long )cpu);if (err ==NOTIFY_BAD ) { printk("%s: attempt to take down CPU %u failed\n",
@@ -152,7 +153,7 @@ static int _cpu_down(unsigned int cpu)if (IS_ERR (p) ||cpu_online (cpu)) {/* CPU didn't die: tell everyone. Can't complain. */ -if (raw_notifier_call_chain (&cpu_chain,CPU_DOWN_FAILED , +if (raw_notifier_call_chain (&cpu_chain,CPU_DOWN_FAILED | mod, (void *)(long )cpu) ==NOTIFY_BAD )BUG ();
@@ -175,7 +176,7 @@ static int _cpu_down(unsigned int cpu)put_cpu ();/* CPU is completely dead: tell everyone. Too late to complain. */ -if (raw_notifier_call_chain (&cpu_chain,CPU_DEAD , +if (raw_notifier_call_chain (&cpu_chain,CPU_DEAD | mod, (void *)(long )cpu) ==NOTIFY_BAD )BUG ();
@@ -196,7 +197,7 @@ int cpu_down(unsigned int cpu)if (cpu_hotplug_disabled) err = -EBUSY ;else - err =_cpu_down (cpu); + err =_cpu_down (cpu,0 );mutex_unlock (&cpu_add_remove_lock);return err;
@@ -204,15 +205,16 @@ int cpu_down(unsigned int cpu)#endif /*CONFIG_HOTPLUG_CPU*/ /* Requires cpu_add_remove_lock to be held */ -static int __cpuinit _cpu_up (unsigned int cpu) +static int __cpuinit _cpu_up (unsigned int cpu,int tasks_frozen) {int ret;void *hcpu = (void *)(long )cpu; +unsigned long mod = tasks_frozen ?CPU_TASKS_FROZEN :0 ;if (cpu_online (cpu) || !cpu_present (cpu))return -EINVAL ; - ret =raw_notifier_call_chain (&cpu_chain,CPU_UP_PREPARE , hcpu); + ret =raw_notifier_call_chain (&cpu_chain,CPU_UP_PREPARE | mod, hcpu);if (ret ==NOTIFY_BAD ) {printk ("%s: attempt to bring up CPU %u failed\n" , __FUNCTION__, cpu);
@@ -229,12 +231,12 @@ static int __cpuinit _cpu_up(unsigned inBUG_ON (!cpu_online (cpu));/* Now call notifier in preparation. */ -raw_notifier_call_chain (&cpu_chain,CPU_ONLINE , hcpu); +raw_notifier_call_chain (&cpu_chain,CPU_ONLINE | mod, hcpu); out_notify:if (ret !=0 )raw_notifier_call_chain (&cpu_chain, -CPU_UP_CANCELED , hcpu); +CPU_UP_CANCELED | mod, hcpu);return ret; }
@@ -247,7 +249,7 @@ int __cpuinit cpu_up(unsigned int cpu)if (cpu_hotplug_disabled) err = -EBUSY ;else - err =_cpu_up (cpu); + err =_cpu_up (cpu,0 );mutex_unlock (&cpu_add_remove_lock);return err;
@@ -277,7 +279,7 @@ int disable_nonboot_cpus(void)for_each_online_cpu (cpu) {if (cpu == first_cpu)continue ; - error =_cpu_down (cpu); + error =_cpu_down (cpu,1 );if (!error) {cpu_set (cpu, frozen_cpus);printk ("CPU%d is down\n" , cpu);
@@ -312,7 +314,7 @@ void enable_nonboot_cpus(void) suspend_cpu_hotplug =1 ;printk ("Enabling non-boot CPUs ...\n" );for_each_cpu_mask (cpu, frozen_cpus) { - error =_cpu_up (cpu); + error =_cpu_up (cpu,1 );if (!error) {printk ("CPU%d is up\n" , cpu);continue ;
arch/i386/kernel/cpu/intel_cacheinfo.c +2 -0
@@ -733,9 +733,11 @@ static int __cpuinit cacheinfo_cpu_callb sys_dev =get_cpu_sysdev (cpu);switch (action) {case CPU_ONLINE : +case CPU_ONLINE_FROZEN :cache_add_dev (sys_dev);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :cache_remove_dev (sys_dev);break ; }
arch/i386/kernel/cpu/mcheck/therm_throt.c +2 -0
@@ -137,10 +137,12 @@ static __cpuinit int thermal_throttle_cpmutex_lock (&therm_cpu_lock);switch (action) {case CPU_ONLINE : +case CPU_ONLINE_FROZEN : err =thermal_throttle_add_dev (sys_dev);WARN_ON (err);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :thermal_throttle_remove_dev (sys_dev);break ; }
arch/i386/kernel/cpuid.c +2 -0
@@ -169,9 +169,11 @@ static int cpuid_class_cpu_callback(struswitch (action) {case CPU_ONLINE : +case CPU_ONLINE_FROZEN :cpuid_device_create (cpu);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :device_destroy (cpuid_class,MKDEV (CPUID_MAJOR , cpu));break ; }
arch/i386/kernel/microcode.c +3 -0
@@ -775,10 +775,13 @@ mc_cpu_callback(struct notifier_block *n sys_dev =get_cpu_sysdev (cpu);switch (action) {case CPU_ONLINE : +case CPU_ONLINE_FROZEN :case CPU_DOWN_FAILED : +case CPU_DOWN_FAILED_FROZEN :mc_sysdev_add (sys_dev);break ;case CPU_DOWN_PREPARE : +case CPU_DOWN_PREPARE_FROZEN :mc_sysdev_remove (sys_dev);break ; }
arch/i386/kernel/msr.c +2 -0
@@ -251,9 +251,11 @@ static int msr_class_cpu_callback(structswitch (action) {case CPU_ONLINE : +case CPU_ONLINE_FROZEN :msr_device_create (cpu);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :device_destroy (msr_class,MKDEV (MSR_MAJOR , cpu));break ; }
arch/ia64/kernel/palinfo.c +2 -0
@@ -975,9 +975,11 @@ static int palinfo_cpu_callback(struct nswitch (action) {case CPU_ONLINE : +case CPU_ONLINE_FROZEN :create_palinfo_proc_entries (hotcpu);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :remove_palinfo_proc_entries (hotcpu);break ; }
arch/ia64/kernel/salinfo.c +2 -0
@@ -583,6 +583,7 @@ salinfo_cpu_callback(struct notifier_blostruct salinfo_data *data;switch (action) {case CPU_ONLINE : +case CPU_ONLINE_FROZEN :spin_lock_irqsave (&data_saved_lock, flags); for (i = 0, data = salinfo_data; i < ARRAY_SIZE(salinfo_data);
@@ -593,6 +594,7 @@ salinfo_cpu_callback(struct notifier_blospin_unlock_irqrestore (&data_saved_lock, flags);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :spin_lock_irqsave (&data_saved_lock, flags); for (i = 0, data = salinfo_data; i < ARRAY_SIZE(salinfo_data);
arch/ia64/kernel/topology.c +2 -0
@@ -412,9 +412,11 @@ static int __cpuinit cache_cpu_callback( sys_dev =get_cpu_sysdev (cpu);switch (action) {case CPU_ONLINE : +case CPU_ONLINE_FROZEN :cache_add_dev (sys_dev);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :cache_remove_dev (sys_dev);break ; }
arch/powerpc/kernel/sysfs.c +2 -0
@@ -341,10 +341,12 @@ static int __cpuinit sysfs_cpu_notify(stswitch (action) {case CPU_ONLINE : +case CPU_ONLINE_FROZEN :register_cpu_online (cpu);break ;#ifdef CONFIG_HOTPLUG_CPU case CPU_DEAD : +case CPU_DEAD_FROZEN :unregister_cpu_online (cpu);break ;#endif
arch/powerpc/mm/numa.c +3 -0
@@ -252,12 +252,15 @@ static int __cpuinit cpu_numa_callback(sswitch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN :numa_setup_cpu (lcpu); ret =NOTIFY_OK ;break ;#ifdef CONFIG_HOTPLUG_CPU case CPU_DEAD : +case CPU_DEAD_FROZEN :case CPU_UP_CANCELED : +case CPU_UP_CANCELED_FROZEN :unmap_cpu_from_node (lcpu);break ; ret =NOTIFY_OK ;
arch/s390/appldata/appldata_base.c +2 -0
@@ -567,9 +567,11 @@ appldata_cpu_notify(struct notifier_bloc {switch (action) {case CPU_ONLINE : +case CPU_ONLINE_FROZEN :appldata_online_cpu ((long ) hcpu);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :appldata_offline_cpu ((long ) hcpu);break ;default :
arch/x86_64/kernel/mce.c +2 -0
@@ -704,9 +704,11 @@ mce_cpu_callback(struct notifier_block *switch (action) {case CPU_ONLINE : +case CPU_ONLINE_FROZEN :mce_create_device (cpu);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :mce_remove_device (cpu);break ; }
arch/x86_64/kernel/mce_amd.c +2 -0
@@ -654,9 +654,11 @@ static int threshold_cpu_callback(structswitch (action) {case CPU_ONLINE : +case CPU_ONLINE_FROZEN :threshold_create_device (cpu);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :threshold_remove_device (cpu);break ;default :
arch/x86_64/kernel/vsyscall.c +1 -1
@@ -301,7 +301,7 @@ static int __cpuinit cpu_vsyscall_notifier(struct notifier_block *n, unsigned long action, void *arg) { long cpu = (long)arg; -if (action ==CPU_ONLINE ) +if (action ==CPU_ONLINE || action ==CPU_ONLINE_FROZEN )smp_call_function_single (cpu, cpu_vsyscall_init,NULL ,0 ,1 );return NOTIFY_DONE ; }
block/ll_rw_blk.c +1 -1
@@ -3505,7 +3505,7 @@ static int blk_cpu_notify(struct notifie* If a CPU goes away, splice its entries to the current CPU * and trigger a run of the softirq */ -if (action ==CPU_DEAD ) { +if (action ==CPU_DEAD || action ==CPU_DEAD_FROZEN ) {int cpu = (unsigned long ) hcpu;local_irq_disable ();
drivers/base/topology.c +3 -0
@@ -126,10 +126,13 @@ static int __cpuinit topology_cpu_callbaswitch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN : rc =topology_add_dev (cpu);break ;case CPU_UP_CANCELED : +case CPU_UP_CANCELED_FROZEN :case CPU_DEAD : +case CPU_DEAD_FROZEN :topology_remove_dev (cpu);break ; }
drivers/cpufreq/cpufreq.c +3 -0
@@ -1716,9 +1716,11 @@ static int cpufreq_cpu_callback(struct nif (sys_dev) {switch (action) {case CPU_ONLINE : +case CPU_ONLINE_FROZEN :cpufreq_add_dev (sys_dev);break ;case CPU_DOWN_PREPARE : +case CPU_DOWN_PREPARE_FROZEN :if (unlikely (lock_policy_rwsem_write (cpu)))BUG ();
@@ -1730,6 +1732,7 @@ static int cpufreq_cpu_callback(struct n__cpufreq_remove_dev (sys_dev);break ;case CPU_DOWN_FAILED : +case CPU_DOWN_FAILED_FROZEN :cpufreq_add_dev (sys_dev);break ; }
drivers/cpufreq/cpufreq_stats.c +2 -0
@@ -313,9 +313,11 @@ static int cpufreq_stat_cpu_callback(strswitch (action) {case CPU_ONLINE : +case CPU_ONLINE_FROZEN :cpufreq_update_policy (cpu);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :cpufreq_stats_free_table (cpu);break ; }
drivers/infiniband/hw/ehca/ehca_irq.c +6 -0
@@ -745,6 +745,7 @@ static int comp_pool_callback(struct notswitch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN :ehca_gen_dbg ("CPU: %x (CPU_PREPARE)" , cpu);if (!create_comp_task (pool, cpu)) {ehca_gen_err ("Can't create comp_task for cpu: %x" , cpu);
@@ -752,24 +753,29 @@ static int comp_pool_callback(struct not }break ;case CPU_UP_CANCELED : +case CPU_UP_CANCELED_FROZEN :ehca_gen_dbg ("CPU: %x (CPU_CANCELED)" , cpu); cct =per_cpu_ptr (pool->cpu_comp_tasks, cpu);kthread_bind (cct->task,any_online_cpu (cpu_online_map));destroy_comp_task (pool, cpu);break ;case CPU_ONLINE : +case CPU_ONLINE_FROZEN :ehca_gen_dbg ("CPU: %x (CPU_ONLINE)" , cpu); cct =per_cpu_ptr (pool->cpu_comp_tasks, cpu);kthread_bind (cct->task, cpu);wake_up_process (cct->task);break ;case CPU_DOWN_PREPARE : +case CPU_DOWN_PREPARE_FROZEN :ehca_gen_dbg ("CPU: %x (CPU_DOWN_PREPARE)" , cpu);break ;case CPU_DOWN_FAILED : +case CPU_DOWN_FAILED_FROZEN :ehca_gen_dbg ("CPU: %x (CPU_DOWN_FAILED)" , cpu);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :ehca_gen_dbg ("CPU: %x (CPU_DEAD)" , cpu);destroy_comp_task (pool, cpu);take_over_work (pool, cpu);
drivers/kvm/kvm_main.c +3 -0
@@ -2363,7 +2363,9 @@ static int kvm_cpu_hotplug(struct notifiswitch (val) {case CPU_DOWN_PREPARE : +case CPU_DOWN_PREPARE_FROZEN :case CPU_UP_CANCELED : +case CPU_UP_CANCELED_FROZEN :printk (KERN_INFO "kvm: disabling virtualization on CPU%d\n" , cpu);decache_vcpus_on_cpu (cpu);
@@ -2371,6 +2373,7 @@ static int kvm_cpu_hotplug(struct notifi NULL, 0, 1);break ;case CPU_ONLINE : +case CPU_ONLINE_FROZEN :printk (KERN_INFO "kvm: enabling virtualization on CPU%d\n" , cpu); smp_call_function_single(cpu, kvm_arch_ops->hardware_enable,
fs/buffer.c +1 -1
@@ -2994,7 +2994,7 @@ static void buffer_exit_cpu(int cpu)static int buffer_cpu_notify (struct notifier_block *self,unsigned long action,void *hcpu) { -if (action ==CPU_DEAD ) +if (action ==CPU_DEAD || action ==CPU_DEAD_FROZEN )buffer_exit_cpu ((unsigned long )hcpu);return NOTIFY_OK ; }
fs/xfs/xfs_mount.c +3 -0
@@ -1734,11 +1734,13 @@ xfs_icsb_cpu_notify(per_cpu_ptr (mp->m_sb_cnts, (unsigned long )hcpu);switch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN :/* Easy Case - initialize the area and locks, and * then rebalance when online does everything else for us. */ memset (cntp,0 ,sizeof (xfs_icsb_cnts_t));break ;case CPU_ONLINE : +case CPU_ONLINE_FROZEN :xfs_icsb_lock (mp);xfs_icsb_balance_counter (mp,XFS_SBS_ICOUNT ,0 ,0 );xfs_icsb_balance_counter (mp,XFS_SBS_IFREE ,0 ,0 );
@@ -1746,6 +1748,7 @@ xfs_icsb_cpu_notify(xfs_icsb_unlock (mp);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :/* Disable all the counters, then fold the dead cpu's * count into the total on the global superblock and * re-enable the counters. */
kernel/hrtimer.c +2 -0
@@ -1407,11 +1407,13 @@ static int __cpuinit hrtimer_cpu_notify(switch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN :init_hrtimers_cpu (cpu);break ;#ifdef CONFIG_HOTPLUG_CPU case CPU_DEAD : +case CPU_DEAD_FROZEN :clockevents_notify (CLOCK_EVT_NOTIFY_CPU_DEAD , &cpu);migrate_hrtimers (cpu);break ;
kernel/profile.c +4 -0
@@ -340,6 +340,7 @@ static int __devinit profile_cpu_callbacswitch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN : node =cpu_to_node (cpu);per_cpu (cpu_profile_flip, cpu) =0 ;if (!per_cpu (cpu_profile_hits, cpu)[1 ]) {
@@ -365,10 +366,13 @@ static int __devinit profile_cpu_callbac__free_page (page);return NOTIFY_BAD ;case CPU_ONLINE : +case CPU_ONLINE_FROZEN :cpu_set (cpu, prof_cpu_mask);break ;case CPU_UP_CANCELED : +case CPU_UP_CANCELED_FROZEN :case CPU_DEAD : +case CPU_DEAD_FROZEN :cpu_clear (cpu, prof_cpu_mask);if (per_cpu (cpu_profile_hits, cpu)[0 ]) { page =virt_to_page (per_cpu (cpu_profile_hits, cpu)[0 ]);
kernel/rcupdate.c +2 -0
@@ -558,9 +558,11 @@ static int __cpuinit rcu_cpu_notify(strulong cpu = (long )hcpu;switch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN :rcu_online_cpu (cpu);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :rcu_offline_cpu (cpu);break ;default :
kernel/relay.c +2 -0
@@ -490,6 +490,7 @@ static int __cpuinit relay_hotcpu_callbaswitch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN :mutex_lock (&relay_channels_mutex);list_for_each_entry (chan, &relay_channels, list) {if (chan->buf[hotcpu])
@@ -506,6 +507,7 @@ static int __cpuinit relay_hotcpu_callbamutex_unlock (&relay_channels_mutex);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :/* No need to flush the cpu : will be flushed upon * final relay_flush() call. */ break ;
kernel/sched.c +10 -0
@@ -5158,6 +5158,7 @@ migration_call(struct notifier_block *nfswitch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN : p =kthread_create (migration_thread, hcpu,"migration/%d" ,cpu);if (IS_ERR (p))return NOTIFY_BAD ;
@@ -5171,12 +5172,14 @@ migration_call(struct notifier_block *nfbreak ;case CPU_ONLINE : +case CPU_ONLINE_FROZEN :/* Strictly unneccessary, as first user will wake it. */ wake_up_process (cpu_rq (cpu)->migration_thread);break ;#ifdef CONFIG_HOTPLUG_CPU case CPU_UP_CANCELED : +case CPU_UP_CANCELED_FROZEN :if (!cpu_rq (cpu)->migration_thread)break ;/* Unbind it from offline cpu so it can run. Fall thru. */
@@ -5187,6 +5190,7 @@ migration_call(struct notifier_block *nfbreak ;case CPU_DEAD : +case CPU_DEAD_FROZEN :migrate_live_tasks (cpu); rq =cpu_rq (cpu);kthread_stop (rq->migration_thread);
@@ -6668,14 +6672,20 @@ static int update_sched_domains(struct n {switch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN :case CPU_DOWN_PREPARE : +case CPU_DOWN_PREPARE_FROZEN :detach_destroy_domains (&cpu_online_map);return NOTIFY_OK ;case CPU_UP_CANCELED : +case CPU_UP_CANCELED_FROZEN :case CPU_DOWN_FAILED : +case CPU_DOWN_FAILED_FROZEN :case CPU_ONLINE : +case CPU_ONLINE_FROZEN :case CPU_DEAD : +case CPU_DEAD_FROZEN :/* * Fall through and re-initialise the domains. */
kernel/softirq.c +4 -0
@@ -593,6 +593,7 @@ static int __cpuinit cpu_callback(structswitch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN : p =kthread_create (ksoftirqd, hcpu,"ksoftirqd/%d" , hotcpu);if (IS_ERR (p)) {printk ("ksoftirqd for %i failed\n" , hotcpu);
@@ -602,16 +603,19 @@ static int __cpuinit cpu_callback(structper_cpu (ksoftirqd, hotcpu) = p;break ;case CPU_ONLINE : +case CPU_ONLINE_FROZEN :wake_up_process (per_cpu (ksoftirqd, hotcpu));break ;#ifdef CONFIG_HOTPLUG_CPU case CPU_UP_CANCELED : +case CPU_UP_CANCELED_FROZEN :if (!per_cpu (ksoftirqd, hotcpu))break ;/* Unbind so it can run. Fall thru. */ kthread_bind (per_cpu (ksoftirqd, hotcpu),any_online_cpu (cpu_online_map));case CPU_DEAD : +case CPU_DEAD_FROZEN : p =per_cpu (ksoftirqd, hotcpu);per_cpu (ksoftirqd, hotcpu) =NULL ;kthread_stop (p);
kernel/softlockup.c +4 -0
@@ -112,6 +112,7 @@ cpu_callback(struct notifier_block *nfb,switch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN :BUG_ON (per_cpu (watchdog_task, hotcpu)); p =kthread_create (watchdog, hcpu,"watchdog/%d" , hotcpu);if (IS_ERR (p)) {
@@ -123,16 +124,19 @@ cpu_callback(struct notifier_block *nfb,kthread_bind (p, hotcpu);break ;case CPU_ONLINE : +case CPU_ONLINE_FROZEN :wake_up_process (per_cpu (watchdog_task, hotcpu));break ;#ifdef CONFIG_HOTPLUG_CPU case CPU_UP_CANCELED : +case CPU_UP_CANCELED_FROZEN :if (!per_cpu (watchdog_task, hotcpu))break ;/* Unbind so it can run. Fall thru. */ kthread_bind (per_cpu (watchdog_task, hotcpu),any_online_cpu (cpu_online_map));case CPU_DEAD : +case CPU_DEAD_FROZEN : p =per_cpu (watchdog_task, hotcpu);per_cpu (watchdog_task, hotcpu) =NULL ;kthread_stop (p);
kernel/timer.c +2 -0
@@ -1699,11 +1699,13 @@ static int __cpuinit timer_cpu_notify(stlong cpu = (long )hcpu;switch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN :if (init_timers_cpu (cpu) <0 )return NOTIFY_BAD ;break ;#ifdef CONFIG_HOTPLUG_CPU case CPU_DEAD : +case CPU_DEAD_FROZEN :migrate_timers (cpu);break ;#endif
kernel/workqueue.c +6 -0
@@ -757,6 +757,7 @@ static int __devinit workqueue_cpu_callbswitch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN :mutex_lock (&workqueue_mutex);/* Create a new workqueue thread for it. */ list_for_each_entry (wq, &workqueues, list) {
@@ -768,6 +769,7 @@ static int __devinit workqueue_cpu_callbbreak ;case CPU_ONLINE : +case CPU_ONLINE_FROZEN :/* Kick off worker threads. */ list_for_each_entry (wq, &workqueues, list) {struct cpu_workqueue_struct *cwq;
@@ -780,6 +782,7 @@ static int __devinit workqueue_cpu_callbbreak ;case CPU_UP_CANCELED : +case CPU_UP_CANCELED_FROZEN :list_for_each_entry (wq, &workqueues, list) {if (!per_cpu_ptr (wq->cpu_wq, hotcpu)->thread)continue ;
@@ -792,14 +795,17 @@ static int __devinit workqueue_cpu_callbbreak ;case CPU_DOWN_PREPARE : +case CPU_DOWN_PREPARE_FROZEN :mutex_lock (&workqueue_mutex);break ;case CPU_DOWN_FAILED : +case CPU_DOWN_FAILED_FROZEN :mutex_unlock (&workqueue_mutex);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :list_for_each_entry (wq, &workqueues, list)cleanup_workqueue_thread (wq, hotcpu);list_for_each_entry (wq, &workqueues, list)
lib/radix-tree.c +1 -1
@@ -1004,7 +1004,7 @@ static int radix_tree_callback(struct nostruct radix_tree_preload *rtp;/* Free per-cpu pool of perloaded nodes */ -if (action ==CPU_DEAD ) { +if (action ==CPU_DEAD || action ==CPU_DEAD_FROZEN ) { rtp = &per_cpu (radix_tree_preloads, cpu);while (rtp->nr) { kmem_cache_free(radix_tree_node_cachep,
mm/page_alloc.c +4 -1
@@ -2143,11 +2143,14 @@ static int __cpuinit pageset_cpuup_callbswitch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN :if (process_zones (cpu)) ret =NOTIFY_BAD ;break ;case CPU_UP_CANCELED : +case CPU_UP_CANCELED_FROZEN :case CPU_DEAD : +case CPU_DEAD_FROZEN :free_zone_pagesets (cpu);break ;default :
@@ -3007,7 +3010,7 @@ static int page_alloc_cpu_notify(struct {int cpu = (unsigned long )hcpu; -if (action ==CPU_DEAD ) { +if (action ==CPU_DEAD || action ==CPU_DEAD_FROZEN ) {local_irq_disable ();__drain_pages (cpu);vm_events_fold_cpu (cpu);
mm/swap.c +1 -1
@@ -488,7 +488,7 @@ static int cpu_swap_callback(struct notilong *committed; committed = &per_cpu (committed_space, (long )hcpu); -if (action ==CPU_DEAD ) { +if (action ==CPU_DEAD || action ==CPU_DEAD_FROZEN ) {atomic_add (*committed, &vm_committed_space); *committed =0 ;__lru_add_drain ((long )hcpu);
mm/vmscan.c +1 -1
@@ -1527,7 +1527,7 @@ static int __devinit cpu_callback(structpg_data_t *pgdat;cpumask_t mask; -if (action ==CPU_ONLINE ) { +if (action ==CPU_ONLINE || action ==CPU_ONLINE_FROZEN ) {for_each_online_pgdat (pgdat) { mask =node_to_cpumask (pgdat->node_id);if (any_online_cpu (mask) !=NR_CPUS )
mm/vmstat.c +3 -0
@@ -650,8 +650,11 @@ static int __cpuinit vmstat_cpuup_callba {switch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN :case CPU_UP_CANCELED : +case CPU_UP_CANCELED_FROZEN :case CPU_DEAD : +case CPU_DEAD_FROZEN :refresh_zone_stat_thresholds ();break ;default :
net/core/dev.c +1 -1
@@ -3344,7 +3344,7 @@ static int dev_cpu_callback(struct notifunsigned int cpu, oldcpu = (unsigned long )ocpu;struct softnet_data *sd, *oldsd; -if (action !=CPU_DEAD ) +if (action !=CPU_DEAD && action !=CPU_DEAD_FROZEN )return NOTIFY_OK ;local_irq_disable ();
net/core/flow.c +1 -1
@@ -338,7 +338,7 @@ static int flow_cache_cpu(struct notifie unsigned long action, void *hcpu) { -if (action ==CPU_DEAD ) +if (action ==CPU_DEAD || action ==CPU_DEAD_FROZEN )__flow_cache_shrink ((unsigned long )hcpu,0 );return NOTIFY_OK ; }
net/iucv/iucv.c +6 -0
@@ -528,6 +528,7 @@ static int __cpuinit iucv_cpu_notify(strswitch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN :if (!percpu_populate (iucv_irq_data,sizeof (struct iucv_irq_data ),GFP_KERNEL |GFP_DMA , cpu))
@@ -539,15 +540,20 @@ static int __cpuinit iucv_cpu_notify(str }break ;case CPU_UP_CANCELED : +case CPU_UP_CANCELED_FROZEN :case CPU_DEAD : +case CPU_DEAD_FROZEN :percpu_depopulate (iucv_param, cpu);percpu_depopulate (iucv_irq_data, cpu);break ;case CPU_ONLINE : +case CPU_ONLINE_FROZEN :case CPU_DOWN_FAILED : +case CPU_DOWN_FAILED_FROZEN :smp_call_function_on (iucv_declare_cpu,NULL ,0 ,1 , cpu);break ;case CPU_DOWN_PREPARE : +case CPU_DOWN_PREPARE_FROZEN : cpumask = iucv_buffer_cpumask;cpu_clear (cpu, cpumask);if (cpus_empty (cpumask))
Documentation/cpu-hotplug.txt +7 -2
@@ -217,14 +217,17 @@ Q: What happens when a CPU is being logi A: The following happen, listed in no particular order :-) - A notification is sent to in-kernel registered modules by sending an event - CPU_DOWN_PREPARE + CPU_DOWN_PREPARE or CPU_DOWN_PREPARE_FROZEN, depending on whether or not the + CPU is being offlined while tasks are frozen due to a suspend operation in + progress - All process is migrated away from this outgoing CPU to a new CPU - All interrupts targeted to this CPU is migrated to a new CPU - timers/bottom half/task lets are also migrated to a new CPU - Once all services are migrated, kernel calls an arch specific routine __cpu_disable() to perform arch specific cleanup. - Once this is successful, an event for successful cleanup is sent by an event - CPU_DEAD. + CPU_DEAD (or CPU_DEAD_FROZEN if tasks are frozen due to a suspend while the + CPU is being offlined). "It is expected that each service cleans up when the CPU_DOWN_PREPARE notifier is called, when CPU_DEAD is called its expected there is nothing
@@ -242,9 +245,11 @@ A: This is what you would need in your k switch (action) { case CPU_ONLINE: + case CPU_ONLINE_FROZEN: foobar_online_action(cpu); break; case CPU_DEAD: + case CPU_DEAD_FROZEN: foobar_dead_action(cpu); break; }
mm/slab.c +6 -0
@@ -1180,6 +1180,7 @@ static int __cpuinit cpuup_callback(struswitch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN :mutex_lock (&cache_chain_mutex);/* * We need to do this right in the beginning since
@@ -1266,17 +1267,21 @@ static int __cpuinit cpuup_callback(stru }break ;case CPU_ONLINE : +case CPU_ONLINE_FROZEN :mutex_unlock (&cache_chain_mutex);start_cpu_timer (cpu);break ;#ifdef CONFIG_HOTPLUG_CPU case CPU_DOWN_PREPARE : +case CPU_DOWN_PREPARE_FROZEN :mutex_lock (&cache_chain_mutex);break ;case CPU_DOWN_FAILED : +case CPU_DOWN_FAILED_FROZEN :mutex_unlock (&cache_chain_mutex);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :/* * Even if all the cpus of a node are down, we don't free the * kmem_list3 of any cache. This to avoid a race between
@@ -1288,6 +1293,7 @@ static int __cpuinit cpuup_callback(stru/* fall thru */ #endif case CPU_UP_CANCELED : +case CPU_UP_CANCELED_FROZEN :list_for_each_entry (cachep, &cache_chain, next) {struct array_cache *nc;struct array_cache *shared;
Re: [RFC][PATCH][EXPERIMENTAL] CPU hotplug with frozen tasks
Hi,
The patch looks good to me.
Quoted 6 lines · Rafael J. Wysocki
--- Documentation/cpu-hotplug.txt | 9 +++++++-- arch/i386/kernel/cpu/intel_cacheinfo.c | 2 ++ arch/i386/kernel/cpu/mcheck/therm_throt.c | 2 ++ arch/i386/kernel/cpuid.c | 2 ++
[snip]
Though I am wondering what might be the usecase for microcode! Guess we'll see that patch soon :-)
Regards
gautham.
Signature · 5 lines
Gautham R Shenoy Linux Technology Center IBM India. "Freedom comes with a price tag of responsibility, which is still a bargain, because Freedom is priceless!"
Re: [RFC][PATCH][EXPERIMENTAL] CPU hotplug with frozen tasks
Quoted 12 lines · Gautham R Shenoy
Hi,
The patch looks good to me.
Quoted 6 lines · Rafael J. Wysocki
--- Documentation/cpu-hotplug.txt | 9 +++++++-- arch/i386/kernel/cpu/intel_cacheinfo.c | 2 ++ arch/i386/kernel/cpu/mcheck/therm_throt.c | 2 ++ arch/i386/kernel/cpuid.c | 2 ++
Quoted 3 lines
[snip]
Though I am wondering what might be the usecase for microcode!
Well, for 2.6.21-rc I had to introduce the variable suspend_cpu_hotplug (in cpu.c) and make the microcode driver use it to distinguish between the 'normal' and suspend-related CPU hotplug. I'd like to get rid of this ugliness ASAP.
Quoted 1 line
Guess we'll see that patch soon :-)
Yes, in a couple of days. I have to make both patches apply to -mm first. :-)
Greetings,
Rafael
Re: [RFC][PATCH][EXPERIMENTAL] CPU hotplug with frozen tasks
In reply to Rafael J. WysockiQuoted 4 lines · Rafael J. Wysocki
Appended is the updated version of the patch (in addition to the changes mentioned above I've eliminated the magic constant
0x0008fromcpu.cby changing the new definitions innotifier.h).
Most sub-systems doesn't care about CPU_TASKS_FROZEN bit. Take for example workqueue.c,
Quoted 34 lines
kernel/workqueue.c +4
-0@@ -757,6 +757,7 @@ static int __devinit workqueue_cpu_callbswitch (action) {case CPU_UP_PREPARE : +case CPU_UP_PREPARE_FROZEN :mutex_lock (&workqueue_mutex);/* Create a new workqueue thread for it. */ list_for_each_entry (wq, &workqueues, list) {@@ -768,6 +769,7 @@ static int __devinit workqueue_cpu_callbbreak ;case CPU_ONLINE : +case CPU_ONLINE_FROZEN :/* Kick off worker threads. */ list_for_each_entry (wq, &workqueues, list) {struct cpu_workqueue_struct *cwq;@@ -780,6 +782,7 @@ static int __devinit workqueue_cpu_callbbreak ;case CPU_UP_CANCELED : +case CPU_UP_CANCELED_FROZEN :list_for_each_entry (wq, &workqueues, list) {if (!per_cpu_ptr (wq->cpu_wq, hotcpu)->thread)continue ;@@ -792,14 +795,17 @@ static int __devinit workqueue_cpu_callbbreak ;case CPU_DOWN_PREPARE : +case CPU_DOWN_PREPARE_FROZEN :mutex_lock (&workqueue_mutex);break ;
kernel/workqueue.c static int __devinit workqueue_cpu_callb · line 811
Quoted 10 lines
kernel/workqueue.c +2
-0case CPU_DOWN_FAILED : +case CPU_DOWN_FAILED_FROZEN :mutex_unlock (&workqueue_mutex);break ;case CPU_DEAD : +case CPU_DEAD_FROZEN :list_for_each_entry (wq, &workqueues, list)cleanup_workqueue_thread (wq, hotcpu);list_for_each_entry (wq, &workqueues, list)
I think it is better to add
action &= ~CPU_TASKS_FROZEN;
at the head of workqueue_cpu_callback() instead. I think this way you can make this patch a lot simpler and smaller.
Oleg.