From 93ae4f978ca7f26d17df915ac7afc919c1dd0353 Mon Sep 17 00:00:00 2001 From: "Srivatsa S. Bhat" Date: Tue, 11 Mar 2014 02:04:14 +0530 Subject: CPU hotplug: Provide lockless versions of callback registration functions The following method of CPU hotplug callback registration is not safe due to the possibility of an ABBA deadlock involving the cpu_add_remove_lock and the cpu_hotplug.lock. get_online_cpus(); for_each_online_cpu(cpu) init_cpu(cpu); register_cpu_notifier(&foobar_cpu_notifier); put_online_cpus(); The deadlock is shown below: CPU 0 CPU 1 ----- ----- Acquire cpu_hotplug.lock [via get_online_cpus()] CPU online/offline operation takes cpu_add_remove_lock [via cpu_maps_update_begin()] Try to acquire cpu_add_remove_lock [via register_cpu_notifier()] CPU online/offline operation tries to acquire cpu_hotplug.lock [via cpu_hotplug_begin()] *** DEADLOCK! *** The problem here is that callback registration takes the locks in one order whereas the CPU hotplug operations take the same locks in the opposite order. To avoid this issue and to provide a race-free method to register CPU hotplug callbacks (along with initialization of already online CPUs), introduce new variants of the callback registration APIs that simply register the callbacks without holding the cpu_add_remove_lock during the registration. That way, we can avoid the ABBA scenario. However, we will need to hold the cpu_add_remove_lock throughout the entire critical section, to protect updates to the callback/notifier chain. This can be achieved by writing the callback registration code as follows: cpu_maps_update_begin(); [ or cpu_notifier_register_begin(); see below ] for_each_online_cpu(cpu) init_cpu(cpu); /* This doesn't take the cpu_add_remove_lock */ __register_cpu_notifier(&foobar_cpu_notifier); cpu_maps_update_done(); [ or cpu_notifier_register_done(); see below ] Note that we can't use get_online_cpus() here instead of cpu_maps_update_begin() because the cpu_hotplug.lock is dropped during the invocation of CPU_POST_DEAD notifiers, and hence get_online_cpus() cannot provide the necessary synchronization to protect the callback/notifier chains against concurrent reads and writes. On the other hand, since the cpu_add_remove_lock protects the entire hotplug operation (including CPU_POST_DEAD), we can use cpu_maps_update_begin/done() to guarantee proper synchronization. Also, since cpu_maps_update_begin/done() is like a super-set of get/put_online_cpus(), the former naturally protects the critical sections from concurrent hotplug operations. Since the names cpu_maps_update_begin/done() don't make much sense in CPU hotplug callback registration scenarios, we'll introduce new APIs named cpu_notifier_register_begin/done() and map them to cpu_maps_update_begin/done(). In summary, introduce the lockless variants of un/register_cpu_notifier() and also export the cpu_notifier_register_begin/done() APIs for use by modules. This way, we provide a race-free way to register hotplug callbacks as well as perform initialization for the CPUs that are already online. Cc: Thomas Gleixner Cc: Andrew Morton Cc: Peter Zijlstra Cc: Ingo Molnar Acked-by: Oleg Nesterov Acked-by: Toshi Kani Reviewed-by: Gautham R. Shenoy Signed-off-by: Srivatsa S. Bhat Signed-off-by: Rafael J. Wysocki --- include/linux/cpu.h | 47 +++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 47 insertions(+) (limited to 'include') diff --git a/include/linux/cpu.h b/include/linux/cpu.h index 03e235ad1bba..488d6ebcf6a1 100644 --- a/include/linux/cpu.h +++ b/include/linux/cpu.h @@ -122,26 +122,46 @@ enum { { .notifier_call = fn, .priority = pri }; \ register_cpu_notifier(&fn##_nb); \ } + +#define __cpu_notifier(fn, pri) { \ + static struct notifier_block fn##_nb = \ + { .notifier_call = fn, .priority = pri }; \ + __register_cpu_notifier(&fn##_nb); \ +} #else /* #if defined(CONFIG_HOTPLUG_CPU) || !defined(MODULE) */ #define cpu_notifier(fn, pri) do { (void)(fn); } while (0) +#define __cpu_notifier(fn, pri) do { (void)(fn); } while (0) #endif /* #else #if defined(CONFIG_HOTPLUG_CPU) || !defined(MODULE) */ + #ifdef CONFIG_HOTPLUG_CPU extern int register_cpu_notifier(struct notifier_block *nb); +extern int __register_cpu_notifier(struct notifier_block *nb); extern void unregister_cpu_notifier(struct notifier_block *nb); +extern void __unregister_cpu_notifier(struct notifier_block *nb); #else #ifndef MODULE extern int register_cpu_notifier(struct notifier_block *nb); +extern int __register_cpu_notifier(struct notifier_block *nb); #else static inline int register_cpu_notifier(struct notifier_block *nb) { return 0; } + +static inline int __register_cpu_notifier(struct notifier_block *nb) +{ + return 0; +} #endif static inline void unregister_cpu_notifier(struct notifier_block *nb) { } + +static inline void __unregister_cpu_notifier(struct notifier_block *nb) +{ +} #endif int cpu_up(unsigned int cpu); @@ -149,19 +169,32 @@ void notify_cpu_starting(unsigned int cpu); extern void cpu_maps_update_begin(void); extern void cpu_maps_update_done(void); +#define cpu_notifier_register_begin cpu_maps_update_begin +#define cpu_notifier_register_done cpu_maps_update_done + #else /* CONFIG_SMP */ #define cpu_notifier(fn, pri) do { (void)(fn); } while (0) +#define __cpu_notifier(fn, pri) do { (void)(fn); } while (0) static inline int register_cpu_notifier(struct notifier_block *nb) { return 0; } +static inline int __register_cpu_notifier(struct notifier_block *nb) +{ + return 0; +} + static inline void unregister_cpu_notifier(struct notifier_block *nb) { } +static inline void __unregister_cpu_notifier(struct notifier_block *nb) +{ +} + static inline void cpu_maps_update_begin(void) { } @@ -170,6 +203,14 @@ static inline void cpu_maps_update_done(void) { } +static inline void cpu_notifier_register_begin(void) +{ +} + +static inline void cpu_notifier_register_done(void) +{ +} + #endif /* CONFIG_SMP */ extern struct bus_type cpu_subsys; @@ -183,8 +224,11 @@ extern void put_online_cpus(void); extern void cpu_hotplug_disable(void); extern void cpu_hotplug_enable(void); #define hotcpu_notifier(fn, pri) cpu_notifier(fn, pri) +#define __hotcpu_notifier(fn, pri) __cpu_notifier(fn, pri) #define register_hotcpu_notifier(nb) register_cpu_notifier(nb) +#define __register_hotcpu_notifier(nb) __register_cpu_notifier(nb) #define unregister_hotcpu_notifier(nb) unregister_cpu_notifier(nb) +#define __unregister_hotcpu_notifier(nb) __unregister_cpu_notifier(nb) void clear_tasks_mm_cpumask(int cpu); int cpu_down(unsigned int cpu); @@ -197,9 +241,12 @@ static inline void cpu_hotplug_done(void) {} #define cpu_hotplug_disable() do { } while (0) #define cpu_hotplug_enable() do { } while (0) #define hotcpu_notifier(fn, pri) do { (void)(fn); } while (0) +#define __hotcpu_notifier(fn, pri) do { (void)(fn); } while (0) /* These aren't inline functions due to a GCC bug. */ #define register_hotcpu_notifier(nb) ({ (void)(nb); 0; }) +#define __register_hotcpu_notifier(nb) ({ (void)(nb); 0; }) #define unregister_hotcpu_notifier(nb) ({ (void)(nb); }) +#define __unregister_hotcpu_notifier(nb) ({ (void)(nb); }) #endif /* CONFIG_HOTPLUG_CPU */ #ifdef CONFIG_PM_SLEEP_SMP -- cgit v1.2.3 From f0bdb5e0c72b7347c867da539367138ad95c6b24 Mon Sep 17 00:00:00 2001 From: "Srivatsa S. Bhat" Date: Tue, 11 Mar 2014 02:04:39 +0530 Subject: CPU hotplug, perf: Fix CPU hotplug callback registration Subsystems that want to register CPU hotplug callbacks, as well as perform initialization for the CPUs that are already online, often do it as shown below: get_online_cpus(); for_each_online_cpu(cpu) init_cpu(cpu); register_cpu_notifier(&foobar_cpu_notifier); put_online_cpus(); This is wrong, since it is prone to ABBA deadlocks involving the cpu_add_remove_lock and the cpu_hotplug.lock (when running concurrently with CPU hotplug operations). Instead, the correct and race-free way of performing the callback registration is: cpu_notifier_register_begin(); for_each_online_cpu(cpu) init_cpu(cpu); /* Note the use of the double underscored version of the API */ __register_cpu_notifier(&foobar_cpu_notifier); cpu_notifier_register_done(); Fix the perf subsystem's hotplug notifier by using this latter form of callback registration. Also provide a bare-bones version of perf_cpu_notifier() that doesn't invoke the notifiers for the already online CPUs. This would be useful for subsystems that need to perform a different set of initialization for the already online CPUs, or don't need the initialization altogether. Cc: Peter Zijlstra Cc: Paul Mackerras Cc: Ingo Molnar Cc: Arnaldo Carvalho de Melo Signed-off-by: Srivatsa S. Bhat Signed-off-by: Rafael J. Wysocki --- include/linux/perf_event.h | 16 +++++++++++++++- 1 file changed, 15 insertions(+), 1 deletion(-) (limited to 'include') diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index e56b07f5c9b6..3356abcfff18 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -835,6 +835,8 @@ do { \ { .notifier_call = fn, .priority = CPU_PRI_PERF }; \ unsigned long cpu = smp_processor_id(); \ unsigned long flags; \ + \ + cpu_notifier_register_begin(); \ fn(&fn##_nb, (unsigned long)CPU_UP_PREPARE, \ (void *)(unsigned long)cpu); \ local_irq_save(flags); \ @@ -843,9 +845,21 @@ do { \ local_irq_restore(flags); \ fn(&fn##_nb, (unsigned long)CPU_ONLINE, \ (void *)(unsigned long)cpu); \ - register_cpu_notifier(&fn##_nb); \ + __register_cpu_notifier(&fn##_nb); \ + cpu_notifier_register_done(); \ } while (0) +/* + * Bare-bones version of perf_cpu_notifier(), which doesn't invoke the + * callback for already online CPUs. + */ +#define __perf_cpu_notifier(fn) \ +do { \ + static struct notifier_block fn##_nb = \ + { .notifier_call = fn, .priority = CPU_PRI_PERF }; \ + \ + __register_cpu_notifier(&fn##_nb); \ +} while (0) struct perf_pmu_events_attr { struct device_attribute attr; -- cgit v1.2.3