Messages in this thread Patch in this message | | | Date | Mon, 24 Dec 2007 11:25:41 -0800 (PST) | From | Christoph Lameter <> | Subject | Re: tlb_finish_mmu() bogisity |
| |
On Sun, 23 Dec 2007, Linus Torvalds wrote:
> > What did you intend here? s/&//, perhaps? > > I think that thing is bogus in other ways. It plays games with > "needs_flush", just because it seems to want the generic code to then call > "check_pgt_cache()", not because it actually wants any flushing to take > place.
It can only free the pages on the quicklist after the flush. So it needs to set need_flush to trigger the flush before check_pgt_cache can trim the quicklist. If need_flush is not set then pages are freed in check_pgt_cache before their TLBs have been flushed.
> Or is there any other reason going on here? That quicklist stuff has been > totally broken, it's done too many totally invalid things to really be > worth even keeping. We should get rid of that unmaintainable hack, and if > it has a real performance upside, we should make it part of the *native* > mmu-gather infrastructure, instead of maintaining it as some separate and > buggy/unmaintainable piece-of-sh*t code.
Yup as we agreed before it would be best to make the TLB pieces part of the native mmu gather infrastructure. But then they vary quite a lot between arches.
Quicklist: Remove bogus &
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Index: linux-2.6/include/asm-generic/tlb.h =================================================================== --- linux-2.6.orig/include/asm-generic/tlb.h 2007-12-24 10:53:52.000000000 -0800 +++ linux-2.6/include/asm-generic/tlb.h 2007-12-24 10:54:01.000000000 -0800 @@ -87,7 +87,7 @@ static inline void tlb_finish_mmu(struct mmu_gather *tlb, unsigned long start, unsigned long end) { #ifdef CONFIG_QUICKLIST - tlb->need_flush += &__get_cpu_var(quicklist)[0].nr_pages != 0; + tlb->need_flush += __get_cpu_var(quicklist)[0].nr_pages != 0; #endif tlb_flush_mmu(tlb, start, end);
| |