mm, oom: remove oom_lock from oom_reaper
authorMichal Hocko <mhocko@suse.com>
Wed, 22 Aug 2018 04:52:37 +0000 (21:52 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Wed, 22 Aug 2018 17:52:44 +0000 (10:52 -0700)
oom_reaper used to rely on the oom_lock since e2fe14564d33 ("oom_reaper:
close race with exiting task").  We do not really need the lock anymore
though.  212925802454 ("mm: oom: let oom_reap_task and exit_mmap run
concurrently") has removed serialization with the exit path based on the
mm reference count and so we do not really rely on the oom_lock anymore.

Tetsuo was arguing that at least MMF_OOM_SKIP should be set under the lock
to prevent from races when the page allocator didn't manage to get the
freed (reaped) memory in __alloc_pages_may_oom but it sees the flag later
on and move on to another victim.  Although this is possible in principle
let's wait for it to actually happen in real life before we make the
locking more complex again.

Therefore remove the oom_lock for oom_reaper paths (both exit_mmap and
oom_reap_task_mm).  The reaper serializes with exit_mmap by mmap_sem +
MMF_OOM_SKIP flag.  There is no synchronization with out_of_memory path
now.

[mhocko@kernel.org: oom_reap_task_mm should return false when __oom_reap_task_mm did]
Link: http://lkml.kernel.org/r/20180724141747.GP28386@dhcp22.suse.cz
Link: http://lkml.kernel.org/r/20180719075922.13784-1-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Suggested-by: David Rientjes <rientjes@google.com>
Acked-by: David Rientjes <rientjes@google.com>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/mmap.c
mm/oom_kill.c

index bb2a7e097c7d32d4b86372727005d296335ba72c..5f2b2b184c604b203f38142d8e638db87487f5f4 100644 (file)
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -3063,9 +3063,7 @@ void exit_mmap(struct mm_struct *mm)
                 * which clears VM_LOCKED, otherwise the oom reaper cannot
                 * reliably test it.
                 */
-               mutex_lock(&oom_lock);
                (void)__oom_reap_task_mm(mm);
-               mutex_unlock(&oom_lock);
 
                set_bit(MMF_OOM_SKIP, &mm->flags);
                down_write(&mm->mmap_sem);
index be31a1e0fe78ff23319441d71ad92d3744972f78..66a86dd049a0670e31e6c7a508c245115da100cc 100644 (file)
@@ -535,28 +535,9 @@ bool __oom_reap_task_mm(struct mm_struct *mm)
 
 static bool oom_reap_task_mm(struct task_struct *tsk, struct mm_struct *mm)
 {
-       bool ret = true;
-
-       /*
-        * We have to make sure to not race with the victim exit path
-        * and cause premature new oom victim selection:
-        * oom_reap_task_mm             exit_mm
-        *   mmget_not_zero
-        *                                mmput
-        *                                  atomic_dec_and_test
-        *                                exit_oom_victim
-        *                              [...]
-        *                              out_of_memory
-        *                                select_bad_process
-        *                                  # no TIF_MEMDIE task selects new victim
-        *  unmap_page_range # frees some memory
-        */
-       mutex_lock(&oom_lock);
-
        if (!down_read_trylock(&mm->mmap_sem)) {
-               ret = false;
                trace_skip_task_reaping(tsk->pid);
-               goto unlock_oom;
+               return false;
        }
 
        /*
@@ -568,7 +549,7 @@ static bool oom_reap_task_mm(struct task_struct *tsk, struct mm_struct *mm)
        if (test_bit(MMF_OOM_SKIP, &mm->flags)) {
                up_read(&mm->mmap_sem);
                trace_skip_task_reaping(tsk->pid);
-               goto unlock_oom;
+               return true;
        }
 
        trace_start_task_reaping(tsk->pid);
@@ -576,8 +557,7 @@ static bool oom_reap_task_mm(struct task_struct *tsk, struct mm_struct *mm)
        /* failed to reap part of the address space. Try again later */
        if (!__oom_reap_task_mm(mm)) {
                up_read(&mm->mmap_sem);
-               ret = false;
-               goto unlock_oom;
+               return false;
        }
 
        pr_info("oom_reaper: reaped process %d (%s), now anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n",
@@ -588,9 +568,7 @@ static bool oom_reap_task_mm(struct task_struct *tsk, struct mm_struct *mm)
        up_read(&mm->mmap_sem);
 
        trace_finish_task_reaping(tsk->pid);
-unlock_oom:
-       mutex_unlock(&oom_lock);
-       return ret;
+       return true;
 }
 
 #define MAX_OOM_REAP_RETRIES 10