cpusets: rework guarantee_online_cpus() to fix deadlock with cpu_down()
PATCH 2/3 cpusets: rework guarantee_online_cpus() to fix deadlock with cpu_down()
Suppose that task T bound to CPU takes callback_mutex. If cpu_down(CPU) happens before T drops callback_mutex we have a deadlock.
stop_machine() preempts T, then migration_call(CPU_DEAD) tries to take cpuset_lock() and hangs forever because CPU is already dead and thus T can't be scheduled.
To simplify the testing, I added schedule_timeout_interruptible(3*HZ) to cpuset_sprintf_cpulist() under mutex_lock(callback_mutex). Then,
mkdir -p /dev/cpuset
mount -t cgroup -ocpuset cpuset /dev/cpuset mkdir -p /dev/cpuset/XXX
taskset -p 2 $$
cat /dev/cpuset/XXX/cpuset.cpus
and, on another console,
echo 0 >> /sys/devices/system/cpu/cpu1/online
Before this patch cpuset/hotplug subsytems deadlock, after this patch everething is OK.
Since the previous patch we can access cs->cpus_allowed safely either under the global callback_mutex or cs->cpumask_lock.
This patch:
- kills cpuset_lock() and cpuset_cpus_allowed_locked()
- converts move_task_off_dead_cpu() to use cpuset_cpus_allowed()
Then we change guarantee_online_cpus(),
- take cs->cpumask_lock instead of callback_mutex
- do NOT scan cs->parent cpusets. If there are no online CPUs in cs->cpus_allowed, we use cpu_online_mask. This is only possible when we are called by cpu_down() hooks, in that case cpuset_track_online_cpus(CPU_DEAD) will fix things later.
In particular, this all means sched_setaffinity() takes the per cpuset spinlock instead of global callback_mutex, and hotplug can forget about complications added by cpusets.
---
include/linux/cpuset.h | 13 ----------- kernel/sched.c | 4 --- kernel/cpuset.c | 56 ++++++++++--------------------------------------- 3 files changed, 13 insertions(+), 60 deletions(-)
include/linux/cpuset.h renamed from include/linux/cpuset.h~2_KILL_CPUSET_LOCK +0 -13
@@ -21,8 +21,6 @@ extern int number_of_cpusets; /* How manextern int cpuset_init (void );extern void cpuset_init_smp (void );extern void cpuset_cpus_allowed (struct task_struct *p,struct cpumask *mask); -extern void cpuset_cpus_allowed_locked (struct task_struct *p, -struct cpumask *mask);extern nodemask_t cpuset_mems_allowed (struct task_struct *p);#define cpuset_current_mems_allowed (current->mems_allowed)void cpuset_init_current_mems_allowed (void );
@@ -69,9 +67,6 @@ struct seq_file;extern void cpuset_task_status_allowed (struct seq_file *m,struct task_struct *task); -extern void cpuset_lock (void ); -extern void cpuset_unlock (void ); -extern int cpuset_mem_spread_node (void );static inline int cpuset_do_page_mem_spread (void )
@@ -105,11 +100,6 @@ static inline void cpuset_cpus_allowed(s {cpumask_copy (mask, cpu_possible_mask); } -static inline void cpuset_cpus_allowed_locked (struct task_struct *p, -struct cpumask *mask) -{ -cpumask_copy (mask, cpu_possible_mask); -}static inline nodemask_t cpuset_mems_allowed (struct task_struct *p) {
@@ -157,9 +147,6 @@ static inline void cpuset_task_status_al { } -static inline void cpuset_lock (void ) {} -static inline void cpuset_unlock (void ) {} -static inline int cpuset_mem_spread_node (void ) {return 0 ;
kernel/sched.c renamed from kernel/sched.c~2_KILL_CPUSET_LOCK +1 -3
@@ -7136,7 +7136,7 @@ again:/* No more Mr. Nice Guy. */ if (dest_cpu >= nr_cpu_ids) { -cpuset_cpus_allowed_locked (p, &p->cpus_allowed); +cpuset_cpus_allowed (p, &p->cpus_allowed); dest_cpu =cpumask_any_and (cpu_online_mask, &p->cpus_allowed);/*
@@ -7550,7 +7550,6 @@ migration_call(struct notifier_block *nfcase CPU_DEAD :case CPU_DEAD_FROZEN : -cpuset_lock ();/* around calls to cpuset_cpus_allowed_lock() */ migrate_live_tasks (cpu); rq =cpu_rq (cpu);kthread_stop (rq->migration_thread);
@@ -7565,7 +7564,6 @@ migration_call(struct notifier_block *nf rq->idle->sched_class = &idle_sched_class;migrate_dead_tasks (cpu);spin_unlock_irq (&rq->lock); -cpuset_unlock ();migrate_nr_uninterruptible (rq);BUG_ON (rq->nr_running !=0 );calc_global_load_remove (rq);
kernel/cpuset.c renamed from kernel/cpuset.c~2_KILL_CPUSET_LOCK +12 -44
81 lines not on this page · raw message