[PATCH] tmpfs: scheduling-while-atomic fix

Nick has tracked scheduling-while-atomic errors to shmem's fragile kmap
avoidance: the root error appears to lie deeper, but rework that fragility.
Plus I've been indicted for war crimes at the end of shmem_swp_entry: my
apologia scorned, so now hide the evidence.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

diff --git a/mm/shmem.c b/mm/shmem.c
index e59de1a98bdaa2dbf43f2bb23dfda1ed6c416996..6581c5d6e4f239a96f2cfaeb5a8ef112d679ae8d 100644 (file)
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -104,23 +104,25 @@ static inline void shmem_dir_unmap(struct page **dir)
 }
 
 static swp_entry_t *shmem_swp_map(struct page *page)
+{
+       return (swp_entry_t *)kmap_atomic(page, KM_USER1);
+}
+
+static inline void shmem_swp_balance_unmap(void)
 {
        /*
-        * We have to avoid the unconditional inc_preempt_count()
-        * in kmap_atomic(), since shmem_swp_unmap() will also be
-        * applied to the low memory addresses within i_direct[].
-        * PageHighMem and high_memory tests are good for all arches
-        * and configs: highmem_start_page and FIXADDR_START are not.
+        * When passing a pointer to an i_direct entry, to code which
+        * also handles indirect entries and so will shmem_swp_unmap,
+        * we must arrange for the preempt count to remain in balance.
+        * What kmap_atomic of a lowmem page does depends on config
+        * and architecture, so pretend to kmap_atomic some lowmem page.
         */
-       return PageHighMem(page)?
-               (swp_entry_t *)kmap_atomic(page, KM_USER1):
-               (swp_entry_t *)page_address(page);
+       (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
 }
 
 static inline void shmem_swp_unmap(swp_entry_t *entry)
 {
-       if (entry >= (swp_entry_t *)high_memory)
-               kunmap_atomic(entry, KM_USER1);
+       kunmap_atomic(entry, KM_USER1);
 }
 
 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
@@ -263,8 +265,10 @@ static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long
        struct page **dir;
        struct page *subdir;
 
-       if (index < SHMEM_NR_DIRECT)
+       if (index < SHMEM_NR_DIRECT) {
+               shmem_swp_balance_unmap();
                return info->i_direct+index;
+       }
        if (!info->i_indirect) {
                if (page) {
                        info->i_indirect = *page;
@@ -306,17 +310,7 @@ static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long
                *page = NULL;
        }
        shmem_dir_unmap(dir);
-
-       /*
-        * With apologies... caller shmem_swp_alloc passes non-NULL
-        * page (though perhaps NULL *page); and now we know that this
-        * indirect page has been allocated, we can shortcut the final
-        * kmap if we know it contains no swap entries, as is commonly
-        * the case: return pointer to a 0 which doesn't need kmapping.
-        */
-       return (page && !subdir->nr_swapped)?
-               (swp_entry_t *)&subdir->nr_swapped:
-               shmem_swp_map(subdir) + offset;
+       return shmem_swp_map(subdir) + offset;
 }
 
 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
@@ -650,8 +644,10 @@ static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, s
        if (size > SHMEM_NR_DIRECT)
                size = SHMEM_NR_DIRECT;
        offset = shmem_find_swp(entry, ptr, ptr+size);
-       if (offset >= 0)
+       if (offset >= 0) {
+               shmem_swp_balance_unmap();
                goto found;
+       }
        if (!info->i_indirect)
                goto lost2;
        /* we might be racing with shmem_truncate */