diff for duplicates of <20161004131417.GC1862@esperanza> diff --git a/a/1.txt b/N1/1.txt index 3149d27..0c324eb 100644 --- a/a/1.txt +++ b/N1/1.txt @@ -42,3 +42,78 @@ On Mon, Oct 03, 2016 at 03:19:31PM +0200, Michal Hocko wrote: Yeah, you're right - I thought max_active implies ordering, but it doesn't. Then we can use an ordered workqueue. Here's the updated patch: + +>From 10f5f126800912c6a4b78a8b615138c1322694ad Mon Sep 17 00:00:00 2001 +From: Vladimir Davydov <vdavydov.dev@gmail.com> +Date: Sat, 1 Oct 2016 16:39:09 +0300 +Subject: [PATCH] mm: memcontrol: use special workqueue for creating per-memcg + caches + +Creating a lot of cgroups at the same time might stall all worker +threads with kmem cache creation works, because kmem cache creation is +done with the slab_mutex held. The problem was amplified by commits +801faf0db894 ("mm/slab: lockless decision to grow cache") in case of +SLAB and 81ae6d03952c ("mm/slub.c: replace kick_all_cpus_sync() with +synchronize_sched() in kmem_cache_shrink()") in case of SLUB, which +increased the maximal time the slab_mutex can be held. + +To prevent that from happening, let's use a special ordered single +threaded workqueue for kmem cache creation. This shouldn't introduce any +functional changes regarding how kmem caches are created, as the work +function holds the global slab_mutex during its whole runtime anyway, +making it impossible to run more than one work at a time. By using a +single threaded workqueue, we just avoid creating a thread per each +work. Ordering is required to avoid a situation when a cgroup's work is +put off indefinitely because there are other cgroups to serve, in other +words to guarantee fairness. + +Link: https://bugzilla.kernel.org/show_bug.cgi?id=172981 +Signed-off-by: Vladimir Davydov <vdavydov.dev@gmail.com> +Reported-by: Doug Smythies <dsmythies@telus.net> +Cc: Christoph Lameter <cl@linux.com> +Cc: David Rientjes <rientjes@google.com> +Cc: Johannes Weiner <hannes@cmpxchg.org> +Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> +Cc: Michal Hocko <mhocko@kernel.org> +Cc: Pekka Enberg <penberg@kernel.org> + +diff --git a/mm/memcontrol.c b/mm/memcontrol.c +index 4be518d4e68a..8d753d87ca37 100644 +--- a/mm/memcontrol.c ++++ b/mm/memcontrol.c +@@ -2175,6 +2175,8 @@ struct memcg_kmem_cache_create_work { + struct work_struct work; + }; + ++static struct workqueue_struct *memcg_kmem_cache_create_wq; ++ + static void memcg_kmem_cache_create_func(struct work_struct *w) + { + struct memcg_kmem_cache_create_work *cw = +@@ -2206,7 +2208,7 @@ static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, + cw->cachep = cachep; + INIT_WORK(&cw->work, memcg_kmem_cache_create_func); + +- schedule_work(&cw->work); ++ queue_work(memcg_kmem_cache_create_wq, &cw->work); + } + + static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, +@@ -5794,6 +5796,17 @@ static int __init mem_cgroup_init(void) + { + int cpu, node; + ++#ifndef CONFIG_SLOB ++ /* ++ * Kmem cache creation is mostly done with the slab_mutex held, ++ * so use a special workqueue to avoid stalling all worker ++ * threads in case lots of cgroups are created simultaneously. ++ */ ++ memcg_kmem_cache_create_wq = ++ alloc_ordered_workqueue("memcg_kmem_cache_create", 0); ++ BUG_ON(!memcg_kmem_cache_create_wq); ++#endif ++ + hotcpu_notifier(memcg_cpu_hotplug_callback, 0); + + for_each_possible_cpu(cpu) diff --git a/a/content_digest b/N1/content_digest index 0eceefb..6993df7 100644 --- a/a/content_digest +++ b/N1/content_digest @@ -54,6 +54,81 @@ "\n" "Yeah, you're right - I thought max_active implies ordering, but it\n" "doesn't. Then we can use an ordered workqueue. Here's the updated\n" - patch: + "patch:\n" + "\n" + ">From 10f5f126800912c6a4b78a8b615138c1322694ad Mon Sep 17 00:00:00 2001\n" + "From: Vladimir Davydov <vdavydov.dev@gmail.com>\n" + "Date: Sat, 1 Oct 2016 16:39:09 +0300\n" + "Subject: [PATCH] mm: memcontrol: use special workqueue for creating per-memcg\n" + " caches\n" + "\n" + "Creating a lot of cgroups at the same time might stall all worker\n" + "threads with kmem cache creation works, because kmem cache creation is\n" + "done with the slab_mutex held. The problem was amplified by commits\n" + "801faf0db894 (\"mm/slab: lockless decision to grow cache\") in case of\n" + "SLAB and 81ae6d03952c (\"mm/slub.c: replace kick_all_cpus_sync() with\n" + "synchronize_sched() in kmem_cache_shrink()\") in case of SLUB, which\n" + "increased the maximal time the slab_mutex can be held.\n" + "\n" + "To prevent that from happening, let's use a special ordered single\n" + "threaded workqueue for kmem cache creation. This shouldn't introduce any\n" + "functional changes regarding how kmem caches are created, as the work\n" + "function holds the global slab_mutex during its whole runtime anyway,\n" + "making it impossible to run more than one work at a time. By using a\n" + "single threaded workqueue, we just avoid creating a thread per each\n" + "work. Ordering is required to avoid a situation when a cgroup's work is\n" + "put off indefinitely because there are other cgroups to serve, in other\n" + "words to guarantee fairness.\n" + "\n" + "Link: https://bugzilla.kernel.org/show_bug.cgi?id=172981\n" + "Signed-off-by: Vladimir Davydov <vdavydov.dev@gmail.com>\n" + "Reported-by: Doug Smythies <dsmythies@telus.net>\n" + "Cc: Christoph Lameter <cl@linux.com>\n" + "Cc: David Rientjes <rientjes@google.com>\n" + "Cc: Johannes Weiner <hannes@cmpxchg.org>\n" + "Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>\n" + "Cc: Michal Hocko <mhocko@kernel.org>\n" + "Cc: Pekka Enberg <penberg@kernel.org>\n" + "\n" + "diff --git a/mm/memcontrol.c b/mm/memcontrol.c\n" + "index 4be518d4e68a..8d753d87ca37 100644\n" + "--- a/mm/memcontrol.c\n" + "+++ b/mm/memcontrol.c\n" + "@@ -2175,6 +2175,8 @@ struct memcg_kmem_cache_create_work {\n" + " \tstruct work_struct work;\n" + " };\n" + " \n" + "+static struct workqueue_struct *memcg_kmem_cache_create_wq;\n" + "+\n" + " static void memcg_kmem_cache_create_func(struct work_struct *w)\n" + " {\n" + " \tstruct memcg_kmem_cache_create_work *cw =\n" + "@@ -2206,7 +2208,7 @@ static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,\n" + " \tcw->cachep = cachep;\n" + " \tINIT_WORK(&cw->work, memcg_kmem_cache_create_func);\n" + " \n" + "-\tschedule_work(&cw->work);\n" + "+\tqueue_work(memcg_kmem_cache_create_wq, &cw->work);\n" + " }\n" + " \n" + " static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,\n" + "@@ -5794,6 +5796,17 @@ static int __init mem_cgroup_init(void)\n" + " {\n" + " \tint cpu, node;\n" + " \n" + "+#ifndef CONFIG_SLOB\n" + "+\t/*\n" + "+\t * Kmem cache creation is mostly done with the slab_mutex held,\n" + "+\t * so use a special workqueue to avoid stalling all worker\n" + "+\t * threads in case lots of cgroups are created simultaneously.\n" + "+\t */\n" + "+\tmemcg_kmem_cache_create_wq =\n" + "+\t\talloc_ordered_workqueue(\"memcg_kmem_cache_create\", 0);\n" + "+\tBUG_ON(!memcg_kmem_cache_create_wq);\n" + "+#endif\n" + "+\n" + " \thotcpu_notifier(memcg_cpu_hotplug_callback, 0);\n" + " \n" + " \tfor_each_possible_cpu(cpu)" -8c035af9453efd66a15e88e25e6b36b9e2dcda6a90bfa2a1ef88519cec1e8d4b +5cfefbeda560426abe9813bdcb27abbd09e76f8db089446b7a71063e29101b46
This is an external index of several public inboxes, see mirroring instructions on how to clone and mirror all data and code used by this external index.