[PATCH] mm: use __GFP_NOMEMALLOC

Use the new __GFP_NOMEMALLOC to simplify the previous handling of
PF_MEMALLOC.

Signed-off-by: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

diff --git a/drivers/md/dm-crypt.c b/drivers/md/dm-crypt.c
index 77619a5..0dd6c2b 100644 (file)
--- a/drivers/md/dm-crypt.c
+++ b/drivers/md/dm-crypt.c
@@ -331,25 +331,19 @@ crypt_alloc_buffer(struct crypt_config *cc, unsigned int size,
        struct bio *bio;
        unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
        int gfp_mask = GFP_NOIO | __GFP_HIGHMEM;
-       unsigned long flags = current->flags;
        unsigned int i;
 
        /*
-        * Tell VM to act less aggressively and fail earlier.
-        * This is not necessary but increases throughput.
+        * Use __GFP_NOMEMALLOC to tell the VM to act less aggressively and
+        * to fail earlier.  This is not necessary but increases throughput.
         * FIXME: Is this really intelligent?
         */
-       current->flags &= ~PF_MEMALLOC;
-
        if (base_bio)
-               bio = bio_clone(base_bio, GFP_NOIO);
+               bio = bio_clone(base_bio, GFP_NOIO|__GFP_NOMEMALLOC);
        else
-               bio = bio_alloc(GFP_NOIO, nr_iovecs);
-       if (!bio) {
-               if (flags & PF_MEMALLOC)
-                       current->flags |= PF_MEMALLOC;
+               bio = bio_alloc(GFP_NOIO|__GFP_NOMEMALLOC, nr_iovecs);
+       if (!bio)
                return NULL;
-       }
 
        /* if the last bio was not complete, continue where that one ended */
        bio->bi_idx = *bio_vec_idx;
@@ -386,9 +380,6 @@ crypt_alloc_buffer(struct crypt_config *cc, unsigned int size,
                size -= bv->bv_len;
        }
 
-       if (flags & PF_MEMALLOC)
-               current->flags |= PF_MEMALLOC;
-
        if (!bio->bi_size) {
                bio_put(bio);
                return NULL;

diff --git a/mm/swap_state.c b/mm/swap_state.c
index a063a90..4f25177 100644 (file)
--- a/mm/swap_state.c
+++ b/mm/swap_state.c
@@ -143,7 +143,6 @@ void __delete_from_swap_cache(struct page *page)
 int add_to_swap(struct page * page)
 {
        swp_entry_t entry;
-       int pf_flags;
        int err;
 
        if (!PageLocked(page))
@@ -154,29 +153,19 @@ int add_to_swap(struct page * page)
                if (!entry.val)
                        return 0;
 
-               /* Radix-tree node allocations are performing
-                * GFP_ATOMIC allocations under PF_MEMALLOC.  
-                * They can completely exhaust the page allocator.  
-                *
-                * So PF_MEMALLOC is dropped here.  This causes the slab 
-                * allocations to fail earlier, so radix-tree nodes will 
-                * then be allocated from the mempool reserves.
+               /*
+                * Radix-tree node allocations from PF_MEMALLOC contexts could
+                * completely exhaust the page allocator. __GFP_NOMEMALLOC
+                * stops emergency reserves from being allocated.
                 *
-                * We're still using __GFP_HIGH for radix-tree node
-                * allocations, so some of the emergency pools are available,
-                * just not all of them.
+                * TODO: this could cause a theoretical memory reclaim
+                * deadlock in the swap out path.
                 */
-
-               pf_flags = current->flags;
-               current->flags &= ~PF_MEMALLOC;
-
                /*
                 * Add it to the swap cache and mark it dirty
                 */
-               err = __add_to_swap_cache(page, entry, GFP_ATOMIC|__GFP_NOWARN);
-
-               if (pf_flags & PF_MEMALLOC)
-                       current->flags |= PF_MEMALLOC;
+               err = __add_to_swap_cache(page, entry,
+                               GFP_ATOMIC|__GFP_NOMEMALLOC|__GFP_NOWARN);
 
                switch (err) {
                case 0:                         /* Success */