Messages in this thread |  | | | Date | Mon, 25 Feb 2008 22:53:41 -0800 (PST) | | From | David Rientjes <> | | Subject | Re: [patch 3/6] mempolicy: add MPOL_F_STATIC_NODES flag |
On Mon, 25 Feb 2008, Paul Jackson wrote:
> $ grep mpol_store_user_nodemask mm/mempolicy.c
> static inline int mpol_store_user_nodemask(const struct mempolicy *pol)
> if (mpol_store_user_nodemask(policy))
> if (!mpol_store_user_nodemask(a))
> if (!mpol_store_user_nodemask(pol) &&
>
> So I see no need to waste the instructions needed (in the three copies
> of this code, since it's static inline) to convert a non-zero value to
> exactly the value 1.
>
Done, thanks.
> Hmmm ... speaking of static inline ... I can knock 600 bytes (that's
> IA64 bytes, so equivalent to about 300 x86 bytes) off the kernel text
> size by not inlining the mm/mempolicy.c routines check_pgd_range() and
> interleave_nid(). I wonder if that would be worth doing. Perhaps
> those two routines are in sufficiently tight corners that the duplicate
> copies of them is needed.
>
It seems like a worthwhile change to me even though gcc will pass the
actuals to check_pgd_range() on the stack. The callers to check_range()
shouldn't be in any fast paths: migrate_to_node() can take a long time
depending on the length of the page list and do_mbind() sleeps on
mmap_sem.
text data bss dec hex filename
11695 24 24 11743 2ddf mm/mempolicy.o.before
11215 24 24 11263 2bff mm/mempolicy.o.after
|  |