From 17ef899c75aa15aeb7818b3b1fec7a84518d10e5 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Mon, 6 Jun 2011 12:20:33 +0200 Subject: [PATCH] sched: Move mmdrop to RCU on RT Takes sleeping locks and calls into the memory allocator, so nothing we want to do in task switch and oder atomic contexts. Signed-off-by: Thomas Gleixner diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h index ca3e517980a0..986c97b1a5bf 100644 --- a/include/linux/mm_types.h +++ b/include/linux/mm_types.h @@ -11,6 +11,7 @@ #include #include #include +#include #include #include #include @@ -507,6 +508,9 @@ struct mm_struct { bool tlb_flush_pending; #endif struct uprobes_state uprobes_state; +#ifdef CONFIG_PREEMPT_RT_BASE + struct rcu_head delayed_drop; +#endif #ifdef CONFIG_X86_INTEL_MPX /* address of the bounds directory */ void __user *bd_addr; diff --git a/include/linux/sched.h b/include/linux/sched.h index 8f50f34904f4..89319443d322 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -2757,6 +2757,7 @@ extern struct mm_struct * mm_alloc(void); /* mmdrop drops the mm and the page tables */ extern void __mmdrop(struct mm_struct *); + static inline void mmdrop(struct mm_struct *mm) { if (unlikely(atomic_dec_and_test(&mm->mm_count))) @@ -2768,6 +2769,17 @@ static inline bool mmget_not_zero(struct mm_struct *mm) return atomic_inc_not_zero(&mm->mm_users); } +#ifdef CONFIG_PREEMPT_RT_BASE +extern void __mmdrop_delayed(struct rcu_head *rhp); +static inline void mmdrop_delayed(struct mm_struct *mm) +{ + if (atomic_dec_and_test(&mm->mm_count)) + call_rcu(&mm->delayed_drop, __mmdrop_delayed); +} +#else +# define mmdrop_delayed(mm) mmdrop(mm) +#endif + /* mmput gets rid of the mappings and all user-space */ extern void mmput(struct mm_struct *); #ifdef CONFIG_MMU diff --git a/kernel/fork.c b/kernel/fork.c index bec1b14130e7..113771be0045 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -701,6 +701,19 @@ struct mm_struct *mm_alloc(void) return mm_init(mm, current); } +#ifdef CONFIG_PREEMPT_RT_BASE +/* + * RCU callback for delayed mm drop. Not strictly rcu, but we don't + * want another facility to make this work. + */ +void __mmdrop_delayed(struct rcu_head *rhp) +{ + struct mm_struct *mm = container_of(rhp, struct mm_struct, delayed_drop); + + __mmdrop(mm); +} +#endif + /* * Called when the last reference to the mm * is dropped: either by a lazy thread or by diff --git a/kernel/sched/core.c b/kernel/sched/core.c index 37b89c4a46db..56f08d6edcd0 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -2731,8 +2731,12 @@ static struct rq *finish_task_switch(struct task_struct *prev) finish_arch_post_lock_switch(); fire_sched_in_preempt_notifiers(current); + /* + * We use mmdrop_delayed() here so we don't have to do the + * full __mmdrop() when we are the last user. + */ if (mm) - mmdrop(mm); + mmdrop_delayed(mm); if (unlikely(prev_state == TASK_DEAD)) { if (prev->sched_class->task_dead) prev->sched_class->task_dead(prev); @@ -5443,6 +5447,8 @@ void sched_setnuma(struct task_struct *p, int nid) #endif /* CONFIG_NUMA_BALANCING */ #ifdef CONFIG_HOTPLUG_CPU +static DEFINE_PER_CPU(struct mm_struct *, idle_last_mm); + /* * Ensures that the idle task is using init_mm right before its cpu goes * offline. @@ -5457,7 +5463,11 @@ void idle_task_exit(void) switch_mm_irqs_off(mm, &init_mm, current); finish_arch_post_lock_switch(); } - mmdrop(mm); + /* + * Defer the cleanup to an alive cpu. On RT we can neither + * call mmdrop() nor mmdrop_delayed() from here. + */ + per_cpu(idle_last_mm, smp_processor_id()) = mm; } /* @@ -7328,6 +7338,10 @@ int sched_cpu_dying(unsigned int cpu) BUG_ON(rq->nr_running != 1); raw_spin_unlock_irqrestore(&rq->lock, flags); calc_load_migrate(rq); + if (per_cpu(idle_last_mm, cpu)) { + mmdrop(per_cpu(idle_last_mm, cpu)); + per_cpu(idle_last_mm, cpu) = NULL; + } update_max_interval(); nohz_balance_exit_idle(cpu); hrtick_clear(rq); -- 2.5.0