Re: [PATCH] mm: PageBuddy and mapcount underflows robustness

[Linus Torvalds <torvalds@linux-foundation.org>]

On Mon, Mar 14, 2011 at 9:59 AM, Andrea Arcangeli <aarcange@redhat.com> wrote:
>
> +#define PAGE_BUDDY_MAPCOUNT_VALUE (-1024*1024)

I realize that this is a nitpick, but from a code generation
standpoint, large random constants like these are just nasty.

I would suggest aiming for constants that are easy to generate and/or
fit better in the code stream. In many encoding schemes (eg x86), -128
is much easier to generate, since it fits in a signed byte and allows
small instructions etc. And in most RISC encodings, 8- or 16-bit
constants can be encoded much more easily than something like your
current one, and bigger ones often end up resulting in a load from
memory or at least several immediate-building instructions.

> -       __ClearPageBuddy(page);
> +       if (PageBuddy(page)) /* __ClearPageBuddy VM_BUG_ON(!PageBuddy(page)) */
> +               __ClearPageBuddy(page);

Also, this is just disgusting. It adds no safety here to have that
VM_BUG_ON(), so it's just unnecessary code generation to do this.
Also, we don't even WANT to do that stupid "__ClearPageBuddy()" in the
first place! What those two code-sites actually want are just a simple

    reset_page_mapcount(page);

which does the right thing in _general_, and not just for the buddy
case - we want to reset the mapcount for other reasons than just
pagebuddy (ie the underflow/overflow case).

And it avoids the VM_BUG_ON() too, making the crazy conditionals be not needed.

No?

                        Linus

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>