Re. Revised [PATCH 3/3] mm/readahead: Remove the check for ra->ra_pages

[Raghavendra D Prabhu <rprabhu@wnohang.net>]

* On Sun, Jul 10, 2011 at 08:59:06AM -0700, Wu Fengguang <fengguang.wu@intel.com> wrote:
>On Sun, Jul 10, 2011 at 08:59:09PM +0800, Raghavendra D Prabhu wrote:
>> * On Sat, Jul 09, 2011 at 01:53:08PM -0700, Wu Fengguang <fengguang.wu@intel.com> wrote:
>> >On Sun, Jul 10, 2011 at 03:41:20AM +0800, Raghavendra D Prabhu wrote:
>> >>page_cache_sync_readahead checks for ra->ra_pages again, so moving the check after VM_SequentialReadHint.

>> >NAK. This patch adds nothing but overheads.

>> >>--- a/mm/filemap.c
>> >>+++ b/mm/filemap.c
>> >>@@ -1566,8 +1566,6 @@ static void do_sync_mmap_readahead(struct vm_area_struct *vma,
>> >> 	/* If we don't want any read-ahead, don't bother */
>> >> 	if (VM_RandomReadHint(vma))
>> >> 		return;
>> >>-	if (!ra->ra_pages)
>> >>-		return;

>> >> 	if (VM_SequentialReadHint(vma)) {
>> >> 		page_cache_sync_readahead(mapping, ra, file, offset,
>> >>@@ -1575,6 +1573,9 @@ static void do_sync_mmap_readahead(struct vm_area_struct *vma,
>> >> 		return;
>> >> 	}

>> >>+	if (!ra->ra_pages)
>> >>+		return;
>> >>+

>> >page_cache_sync_readahead() has the same

>> >	if (!ra->ra_pages)
>> >		return;
>> 1. Yes, I saw that and that is why I moved it after the condition, so that duplicate checks are
>> not needed -- ie., if VM_SequentialReadHint is true, then
>> (!ra->ra_pages) is checked twice otherwise.
>
>Ok, I see.
>
>> 2. Also, another thought, is the check needed at its original place (if
>> not it can be removed), reasons being -- filesystems like tmpfs which
>> have ra_pages set to 0 don't use filemap_fault in their VMA ops and also
>
>Good point. tmpfs is using shmem_fault().. Can you remove the test?
I have removed that test. Patch attached.
>
>> do_sync_mmap_readahead is called in a major page fault context.
>
>Right. This is irrelevant however, because if pa_pages==0, the
>page faults will normally be major ones.
>
>Thanks,
>Fengguang
>
>> >So the patch adds the call into page_cache_sync_readahead() just to return..

>> >Thanks,
>> >Fengguang

>> --------------------------
>> Raghavendra Prabhu
>> GPG Id : 0xD72BE977
>> Fingerprint: B93F EBCB 8E05 7039 CD3C A4B8 A616 DCA1 D72B E977
>> www: wnohang.net
>
>
======================================================================
  The check for ra->ra_pages is not required since fs like tmpfs which have
  ra_pages set to 0 don't use filemap_fault as part of their VMA ops (it uses
  shmem_fault). Also, page_cache_sync_readahead does its own check for ra_pages.

  Signed-off-by: Raghavendra D Prabhu <rprabhu@wnohang.net>
  ---
   mm/filemap.c |    2 --
   1 files changed, 0 insertions(+), 2 deletions(-)

  diff --git a/mm/filemap.c b/mm/filemap.c
  index 074c23d..0bcd276 100644
  --- a/mm/filemap.c
  +++ b/mm/filemap.c
  @@ -1566,8 +1566,6 @@ static void do_sync_mmap_readahead(struct vm_area_struct *vma,
          /* If we don't want any read-ahead, don't bother */
          if (VM_RandomReadHint(vma))
                  return;
  -       if (!ra->ra_pages)
  -               return;

          if (VM_SequentialReadHint(vma)) {
                  page_cache_sync_readahead(mapping, ra, file, offset,
  --
  1.7.6


--------------------------
Raghavendra Prabhu
GPG Id : 0xD72BE977
Fingerprint: B93F EBCB 8E05 7039 CD3C A4B8 A616 DCA1 D72B E977
www: wnohang.net