lkml.org 
[lkml]   [2010]   [Oct]   [8]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
From
SubjectRe: [PATCH 2/3] Retry page fault when blocking on disk transfer.
On Tue, Oct 05, 2010 at 03:44:22PM -0700, Michel Lespinasse wrote:
> On Tue, Oct 5, 2010 at 10:38 AM, Rik van Riel <riel@redhat.com> wrote:
> > Looks like it should be relatively easy to do something
> > similar in do_swap_page also.
>
> Good idea. We don't make use of swap too much, which is probably why
> we didn't have that in our kernel, but it seems like a good idea just
> for uniformity. I'll add this in a follow-on patch.

So here's the patch. Sorry for the delay - it did not take long to write,
but I couldn't test it before today.

Please have a look - I'd like to add this to the series I sent earlier.

----------------------------------- 8< ---------------------------------

Retry page fault when blocking on swap in

This change is the cousin of 'Retry page fault when blocking
on disk transfer'. The idea here is to reduce mmap_sem hold times
that are caused by disk transfers when swapping in pages. We drop
mmap_sem while waiting for the page lock, and return the VM_FAULT_RETRY
flag. do_page_fault will then re-acquire mmap_sem and retry the
page fault. It is expected that upon retry the page will now be cached,
and thus the retry will complete with a low mmap_sem hold time.

Signed-off-by: Michel Lespinasse <walken@google.com>

diff --git a/mm/memory.c b/mm/memory.c
index b068c68..0ec70b4 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -2613,6 +2613,21 @@ int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
return 0;
}

+static inline int lock_page_or_retry(struct page *page, struct mm_struct *mm,
+ unsigned int flags)
+{
+ if (trylock_page(page))
+ return 1;
+ if (!(flags & FAULT_FLAG_ALLOW_RETRY)) {
+ __lock_page(page);
+ return 1;
+ }
+
+ up_read(&mm->mmap_sem);
+ wait_on_page_locked(page);
+ return 0;
+}
+
/*
* We enter with non-exclusive mmap_sem (to exclude vma changes,
* but allow concurrent faults), and pte mapped but not yet locked.
@@ -2626,6 +2641,7 @@ static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
struct page *page, *swapcache = NULL;
swp_entry_t entry;
pte_t pte;
+ int locked;
struct mem_cgroup *ptr = NULL;
int exclusive = 0;
int ret = 0;
@@ -2676,8 +2692,12 @@ static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
goto out_release;
}

- lock_page(page);
+ locked = lock_page_or_retry(page, mm, flags);
delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
+ if (!locked) {
+ ret |= VM_FAULT_RETRY;
+ goto out_release;
+ }

/*
* Make sure try_to_free_swap or reuse_swap_page or swapoff did not

--
Michel "Walken" Lespinasse
A program is never fully debugged until the last user dies.


\
 
 \ /
  Last update: 2010-10-08 06:43    [W:0.186 / U:0.364 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site