From 2550950c6c0faba9185504e84e2712df5b653b9c Mon Sep 17 00:00:00 2001 Message-Id: <2550950c6c0faba9185504e84e2712df5b653b9c.1601675151.git.zanussi@kernel.org> In-Reply-To: <5b5a156f9808b1acf1205606e03da117214549ea.1601675151.git.zanussi@kernel.org> References: <5b5a156f9808b1acf1205606e03da117214549ea.1601675151.git.zanussi@kernel.org> From: Luiz Capitulino Date: Fri, 27 May 2016 15:03:28 +0200 Subject: [PATCH 078/333] mm: perform lru_add_drain_all() remotely Origin: https://www.kernel.org/pub/linux/kernel/projects/rt/4.19/older/patches-4.19.148-rt64.tar.xz lru_add_drain_all() works by scheduling lru_add_drain_cpu() to run on all CPUs that have non-empty LRU pagevecs and then waiting for the scheduled work to complete. However, workqueue threads may never have the chance to run on a CPU that's running a SCHED_FIFO task. This causes lru_add_drain_all() to block forever. This commit solves this problem by changing lru_add_drain_all() to drain the LRU pagevecs of remote CPUs. This is done by grabbing swapvec_lock and calling lru_add_drain_cpu(). PS: This is based on an idea and initial implementation by Rik van Riel. Signed-off-by: Rik van Riel Signed-off-by: Luiz Capitulino Signed-off-by: Sebastian Andrzej Siewior --- mm/swap.c | 36 ++++++++++++++++++++++++++++++------ 1 file changed, 30 insertions(+), 6 deletions(-) diff --git a/mm/swap.c b/mm/swap.c index 92f994b962f0..3885645a45ce 100644 --- a/mm/swap.c +++ b/mm/swap.c @@ -585,9 +585,15 @@ void lru_add_drain_cpu(int cpu) unsigned long flags; /* No harm done if a racing interrupt already did this */ +#ifdef CONFIG_PREEMPT_RT_BASE + local_lock_irqsave_on(rotate_lock, flags, cpu); + pagevec_move_tail(pvec); + local_unlock_irqrestore_on(rotate_lock, flags, cpu); +#else local_lock_irqsave(rotate_lock, flags); pagevec_move_tail(pvec); local_unlock_irqrestore(rotate_lock, flags); +#endif } pvec = &per_cpu(lru_deactivate_file_pvecs, cpu); @@ -657,6 +663,16 @@ void lru_add_drain(void) #ifdef CONFIG_SMP +#ifdef CONFIG_PREEMPT_RT_BASE +static inline void remote_lru_add_drain(int cpu, struct cpumask *has_work) +{ + local_lock_on(swapvec_lock, cpu); + lru_add_drain_cpu(cpu); + local_unlock_on(swapvec_lock, cpu); +} + +#else + static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work); static void lru_add_drain_per_cpu(struct work_struct *dummy) @@ -664,6 +680,16 @@ static void lru_add_drain_per_cpu(struct work_struct *dummy) lru_add_drain(); } +static inline void remote_lru_add_drain(int cpu, struct cpumask *has_work) +{ + struct work_struct *work = &per_cpu(lru_add_drain_work, cpu); + + INIT_WORK(work, lru_add_drain_per_cpu); + queue_work_on(cpu, mm_percpu_wq, work); + cpumask_set_cpu(cpu, has_work); +} +#endif + /* * Doesn't need any cpu hotplug locking because we do rely on per-cpu * kworkers being shut down before our page_alloc_cpu_dead callback is @@ -688,21 +714,19 @@ void lru_add_drain_all(void) cpumask_clear(&has_work); for_each_online_cpu(cpu) { - struct work_struct *work = &per_cpu(lru_add_drain_work, cpu); if (pagevec_count(&per_cpu(lru_add_pvec, cpu)) || pagevec_count(&per_cpu(lru_rotate_pvecs, cpu)) || pagevec_count(&per_cpu(lru_deactivate_file_pvecs, cpu)) || pagevec_count(&per_cpu(lru_lazyfree_pvecs, cpu)) || - need_activate_page_drain(cpu)) { - INIT_WORK(work, lru_add_drain_per_cpu); - queue_work_on(cpu, mm_percpu_wq, work); - cpumask_set_cpu(cpu, &has_work); - } + need_activate_page_drain(cpu)) + remote_lru_add_drain(cpu, &has_work); } +#ifndef CONFIG_PREEMPT_RT_BASE for_each_cpu(cpu, &has_work) flush_work(&per_cpu(lru_add_drain_work, cpu)); +#endif mutex_unlock(&lock); } -- 2.17.1