ksm: unmerge is an origin of OOMs
Hugh Dickins [Tue, 22 Sep 2009 00:02:27 +0000 (17:02 -0700)]
Just as the swapoff system call allocates many pages of RAM to various
processes, perhaps triggering OOM, so "echo 2 >/sys/kernel/mm/ksm/run"
(unmerge) is liable to allocate many pages of RAM to various processes,
perhaps triggering OOM; and each is normally run from a modest admin
process (swapoff or shell), easily repeated until it succeeds.

So treat unmerge_and_remove_all_rmap_items() in the same way that we treat
try_to_unuse(): generalize PF_SWAPOFF to PF_OOM_ORIGIN, and bracket both
with that, to ask the OOM killer to kill them first, to prevent them from
spawning more and more OOM kills.

Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Acked-by: Izik Eidus <ieidus@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

include/linux/sched.h
mm/ksm.c
mm/oom_kill.c
mm/swapfile.c

index 8f3e63c..899d730 100644 (file)
@@ -1720,7 +1720,7 @@ extern cputime_t task_gtime(struct task_struct *p);
 #define PF_FROZEN      0x00010000      /* frozen for system suspend */
 #define PF_FSTRANS     0x00020000      /* inside a filesystem transaction */
 #define PF_KSWAPD      0x00040000      /* I am kswapd */
-#define PF_SWAPOFF     0x00080000      /* I am in swapoff */
+#define PF_OOM_ORIGIN  0x00080000      /* Allocating much memory to others */
 #define PF_LESS_THROTTLE 0x00100000    /* Throttle me less: I clean memory */
 #define PF_KTHREAD     0x00200000      /* I am a kernel thread */
 #define PF_RANDOMIZE   0x00400000      /* randomize virtual address space */
index e11e7a5..37cc373 100644 (file)
--- a/mm/ksm.c
+++ b/mm/ksm.c
@@ -1557,7 +1557,9 @@ static ssize_t run_store(struct kobject *kobj, struct kobj_attribute *attr,
        if (ksm_run != flags) {
                ksm_run = flags;
                if (flags & KSM_RUN_UNMERGE) {
+                       current->flags |= PF_OOM_ORIGIN;
                        err = unmerge_and_remove_all_rmap_items();
+                       current->flags &= ~PF_OOM_ORIGIN;
                        if (err) {
                                ksm_run = KSM_RUN_STOP;
                                count = err;
index a7b2460..da4c342 100644 (file)
@@ -79,7 +79,7 @@ unsigned long badness(struct task_struct *p, unsigned long uptime)
        /*
         * swapoff can easily use up all memory, so kill those first.
         */
-       if (p->flags & PF_SWAPOFF)
+       if (p->flags & PF_OOM_ORIGIN)
                return ULONG_MAX;
 
        /*
index 74f1102..f1bf19d 100644 (file)
@@ -1575,9 +1575,9 @@ SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
        p->flags &= ~SWP_WRITEOK;
        spin_unlock(&swap_lock);
 
-       current->flags |= PF_SWAPOFF;
+       current->flags |= PF_OOM_ORIGIN;
        err = try_to_unuse(type);
-       current->flags &= ~PF_SWAPOFF;
+       current->flags &= ~PF_OOM_ORIGIN;
 
        if (err) {
                /* re-insert swap space back into swap_list */