On Sat, 17 Apr 2010 10:07:26 +0300
Sasha Sirotkin <buildroot@xxxxxxxxxxxxxxx> wrote:
> There is an "early COW" mechanism in __do_fault() which, if the page is
> not present and the fault is FAULT_PAGE_WRITE goes ahead and copies the
> page in order to prevent the next exception.
>
> The question - why the code in __do_fault() does not decrease the shared
> map count of the old page as do_wp_page does ? And while we are at it,
> while this "early COW" code is much more simple than do_wp_page()?
>
> Thanks.
>
IIUC.
Case 1) A task cause a write page fault because pte is not set as PRESENT.
__do_fault() is called. And the kernel found vma is not-SHARED.
Do eary-COW. In this case, the old page was not _mapped_...IOW, the task's
this pte was not accounted into old_page->mapcount.
We just increase new_page->mapcount. Don't touch old_page->mapcount.
Case 2) A task caused a write page fault because pte was not WRITABLE.
do_wp_page() is called because the page was _mapped_.
If the page is shared, decrease old_page->mapcount, increae new_page->mapcount.
Regard,
-Kame
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxxx For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>