[+CC Michal Hocko] On 07/10/2017 12:49 AM, Joel Fernandes wrote: > Certain allocation paths such as the ftrace ring buffer allocator > want to try hard to allocate but not trigger OOM killer and de-stabilize > the system. Currently the ring buffer uses __GFP_NO_RETRY to prevent > the OOM killer from triggering situation however this has an issue. > Its possible the system is in a state where: > a) retrying can make the allocation succeed. > b) there's plenty of memory available in the page cache to satisfy > the request and just retrying is needed. Even though direct reclaim > makes progress, it still couldn't find free page from the free list. > > This patch adds a new GFP flag (__GFP_DONTOOM) to handle the situation > where we want the retry behavior but still want to bail out before going > to OOM killer if retries couldn't satisfy the allocation.
Michal recently turned __GFP_REPEAT into __GFP_RETRY_MAYFAIL [1][2] which I think does exactly what you want. Try hard as long as reclaim/compaction makes progress, but fail the allocation instead of triggering OOM killer. Can you check it out? It's in mmotm/linux-next. [1] http://www.ozlabs.org/~akpm/mmotm/broken-out/mm-tree-wide-replace-__gfp_repeat-by-__gfp_retry_mayfail-with-more-useful-semantic.patch [2] http://lkml.kernel.org/r/[email protected] > Cc: Alexander Duyck <[email protected]> > Cc: Mel Gorman <[email protected]> > Cc: Hao Lee <[email protected]> > Cc: Vladimir Davydov <[email protected]> > Cc: Johannes Weiner <[email protected]> > Cc: Joonsoo Kim <[email protected]> > Cc: Steven Rostedt <[email protected]> > Cc: [email protected] > Cc: [email protected] > Signed-off-by: Joel Fernandes <[email protected]> > --- > include/linux/gfp.h | 6 +++++- > mm/page_alloc.c | 7 +++++++ > 2 files changed, 12 insertions(+), 1 deletion(-) > > diff --git a/include/linux/gfp.h b/include/linux/gfp.h > index 4c6656f1fee7..beaabd110008 100644 > --- a/include/linux/gfp.h > +++ b/include/linux/gfp.h > @@ -40,6 +40,7 @@ struct vm_area_struct; > #define ___GFP_DIRECT_RECLAIM 0x400000u > #define ___GFP_WRITE 0x800000u > #define ___GFP_KSWAPD_RECLAIM 0x1000000u > +#define ___GFP_DONTOOM 0x2000000u > #ifdef CONFIG_LOCKDEP > #define ___GFP_NOLOCKDEP 0x2000000u > #else > @@ -149,6 +150,8 @@ struct vm_area_struct; > * return NULL when direct reclaim and memory compaction have failed to > allow > * the allocation to succeed. The OOM killer is not called with the > current > * implementation. > + * > + * __GFP_DONTOOM: The VM implementation must not OOM if retries have > exhausted. > */ > #define __GFP_IO ((__force gfp_t)___GFP_IO) > #define __GFP_FS ((__force gfp_t)___GFP_FS) > @@ -158,6 +161,7 @@ struct vm_area_struct; > #define __GFP_REPEAT ((__force gfp_t)___GFP_REPEAT) > #define __GFP_NOFAIL ((__force gfp_t)___GFP_NOFAIL) > #define __GFP_NORETRY ((__force gfp_t)___GFP_NORETRY) > +#define __GFP_DONTOOM ((__force gfp_t)___GFP_DONTOOM) > > /* > * Action modifiers > @@ -188,7 +192,7 @@ struct vm_area_struct; > #define __GFP_NOLOCKDEP ((__force gfp_t)___GFP_NOLOCKDEP) > > /* Room for N __GFP_FOO bits */ > -#define __GFP_BITS_SHIFT (25 + IS_ENABLED(CONFIG_LOCKDEP)) > +#define __GFP_BITS_SHIFT (26 + IS_ENABLED(CONFIG_LOCKDEP)) > #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1)) > > /* > diff --git a/mm/page_alloc.c b/mm/page_alloc.c > index bd65b60939b6..970a5c380bb6 100644 > --- a/mm/page_alloc.c > +++ b/mm/page_alloc.c > @@ -3908,6 +3908,13 @@ __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int > order, > if (check_retry_cpuset(cpuset_mems_cookie, ac)) > goto retry_cpuset; > > + /* > + * Its possible that retries failed but we still don't want OOM > + * killer to trigger and can just try again later. > + */ > + if (gfp_mask & __GFP_DONTOOM) > + goto nopage; > + > /* Reclaim has failed us, start killing things */ > page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress); > if (page) >
