Messages in this thread |  | | | Date | Thu, 8 May 2008 16:07:14 -0700 (PDT) | | From | Linus Torvalds <> | | Subject | Re: [patch] speed up / fix the new generic semaphore code (fix AIM7 40% regression with 2.6.26-rc1) | |
On Thu, 8 May 2008, Linus Torvalds wrote:
>
> Some of it is that "page_to_pfn(page)", which involves a nasty division
> (divide by sizeof(struct page)). It gets turned into that shift and
> multiply, but it's still quite expensive with big constants etc.
Btw, sparse will complain about those, because the source code *looks*
really cheap.
The normal "page_to_pfn()" looks trivial:
((unsigned long)((page) - mem_map) + ARCH_PFN_OFFSET)
which looks like a trivial subtraction and addition of a constant, but the
subtraction is on C pointers, and basically turns into
((unsigned long)page - (unsigned long)mem_map) / sizeof(struct page)
and because "struct page" is not some nice power-of-two in size, that
division is rather nasty even though it's a constant size.
Linus
|  |