[lkml]   [2011]   [Jul]   [8]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [PATCH 1/1] mmap: Don't count shmem pages as free in __vm_enough_memory
On Sun, 3 Jul 2011, Dmitry Fink wrote:

> shmem pages can't be reclaimed and if they are swapped out
> that doesn't affect the overall available memory in the system,
> so don't count them along with the rest of the file backed pages.
> Signed-off-by: Dmitry Fink <>

That's a good point. You can add
Acked-by: Hugh Dickins <>
if you'll let me burble on for a while...

1. Your block comment style differs from kernel coding style, and
the nearby comments do follow kernel coding style: please adjust.

2. If you're wondering why this was missed for so long, it's simply
that we didn't have that separate NR_SHMEM count until 2.6.32.

3. There's a danger that this change will fail some large mappings
that were allowed before; but I guess we run that risk every
time we release a larger kernel than before, so let's grant you
the patch... but it might have to be reverted if anyone complains.

4. i915 GEM uses shmem pages that _can_ (sometimes) be freed under
memory pressure; but then, on the other side, some of the pages
counted as "free" may actually be locked or pinned in some way.

5. The OVERCOMMIT_GUESS estimation is at best something of a joke
(a thousand concurrent tasks would each be allowed to make their
own separate maximal mappings), and any change appears to dignify
it somewhat; but you are right, so let's do it.

6. I was worried about additional overhead, and puzzled where the
actual free pages get counted: ah, lower down, with a comment that
nr_free_pages() is very expensive on large systems... and what does
nr_free_pages() do? It does global_page_state(NR_FREE_PAGES): so
does that imply that your additional global_page_state(NR_SHMEM)
will be expensive>? No, the comment, and the peculiar placing of
the nr_free_pages() call, date from when it was a loop over all
zones (hence all nodes) in the system. Now, tell me to go away
and make a separate patch of my own if you wish, fair enough;
but I suggest you tidy that up too (and clearer if it explicitly
says global_page_state(NR_FREE_PAGES) rather than nr_free_pages()):

free = global_page_state(NR_FREE_PAGES);
free += global_page_state(NR_FILE_PAGES);

7. There's an almost identical copy of this code in mm/nommu.c:
please update that one too to keep them in synch. I suppose it
would be better to keep one copy of it somewhere else, but by
now I've probably exhausted your patience, plus I've a nasty
feeling that if I suggest somewhere, I'll be tricking you
into a build error with this or that config. Another time...

You can see why I don't like reviewing more than one-line changes,
can't you :-?


> ---
> mm/mmap.c | 7 +++++++
> 1 files changed, 7 insertions(+), 0 deletions(-)
> diff --git a/mm/mmap.c b/mm/mmap.c
> index b88624f..3a34dc2 100644
> --- a/mm/mmap.c
> +++ b/mm/mmap.c
> @@ -119,6 +119,13 @@ int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
> unsigned long n;
> free = global_page_state(NR_FILE_PAGES);
> +
> + /* shmem pages shouldn't be counted as free in this
> + * case, they can't be purged, only swapped out, and
> + * that won't affect the overall amount of available
> + * memory in the system. */
> + free -= global_page_state(NR_SHMEM);
> +
> free += nr_swap_pages;
> /*
> --

 \ /
  Last update: 2011-07-09 00:25    [W:0.067 / U:1.176 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site