On Tue, May 17, 2011 at 11:24 AM, Hugh Dickins <hughd@xxxxxxxxxx> wrote:
Acked-by: Ying Han<yinghan@xxxxxxxxxx>mem_cgroup_count_vm_event() should update the PGMAJFAULT count for the
target mm, not for current mm (but of course they're usually the same).
We don't know the target mm in shmem_getpage(), so do it at the outer
level in shmem_fault(); and it's easier to follow if we move the
count_vm_event(PGMAJFAULT) there too.
Hah, it was using __count_vm_event() before, sneaking that update into
the unpreemptible section under info->lock: well, it comes to the same
on x86 at least, and I still think it's best to keep these together.
Signed-off-by: Hugh Dickins <hughd@xxxxxxxxxx>
---
mm/shmem.c | 13 ++++++-------
1 file changed, 6 insertions(+), 7 deletions(-)
--- mmotm/mm/shmem.c 2011-05-13 14:57:45.367884578 -0700
+++ linux/mm/shmem.c 2011-05-17 10:27:19.901934756 -0700
@@ -1293,14 +1293,10 @@ repeat:
swappage = lookup_swap_cache(swap);
if (!swappage) {
shmem_swp_unmap(entry);
+ spin_unlock(&info->lock);
/* here we actually do the io */
- if (type && !(*type & VM_FAULT_MAJOR)) {
- __count_vm_event(PGMAJFAULT);
- mem_cgroup_count_vm_event(current->mm,
- PGMAJFAULT);
+ if (type)
*type |= VM_FAULT_MAJOR;
- }
- spin_unlock(&info->lock);
swappage = shmem_swapin(swap, gfp, info, idx);
if (!swappage) {
spin_lock(&info->lock);
@@ -1539,7 +1535,10 @@ static int shmem_fault(struct vm_area_st
error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
if (error)
return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
-
+ if (ret & VM_FAULT_MAJOR) {
+ count_vm_event(PGMAJFAULT);
+ mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
+ }
return ret | VM_FAULT_LOCKED;
}
Thank you Hugh for the fix.
--Ying