aboutsummaryrefslogtreecommitdiffstats
path: root/arch/i386
diff options
context:
space:
mode:
authorHugh Dickins <hugh@veritas.com>2005-10-29 21:16:40 -0400
committerLinus Torvalds <torvalds@g5.osdl.org>2005-10-30 00:40:42 -0400
commit4c21e2f2441dc5fbb957b030333f5a3f2d02dea7 (patch)
tree1f76d33bb1d76221c6424bc5fed080a4f91349a6 /arch/i386
parentb38c6845b695141259019e2b7c0fe6c32a6e720d (diff)
[PATCH] mm: split page table lock
Christoph Lameter demonstrated very poor scalability on the SGI 512-way, with a many-threaded application which concurrently initializes different parts of a large anonymous area. This patch corrects that, by using a separate spinlock per page table page, to guard the page table entries in that page, instead of using the mm's single page_table_lock. (But even then, page_table_lock is still used to guard page table allocation, and anon_vma allocation.) In this implementation, the spinlock is tucked inside the struct page of the page table page: with a BUILD_BUG_ON in case it overflows - which it would in the case of 32-bit PA-RISC with spinlock debugging enabled. Splitting the lock is not quite for free: another cacheline access. Ideally, I suppose we would use split ptlock only for multi-threaded processes on multi-cpu machines; but deciding that dynamically would have its own costs. So for now enable it by config, at some number of cpus - since the Kconfig language doesn't support inequalities, let preprocessor compare that with NR_CPUS. But I don't think it's worth being user-configurable: for good testing of both split and unsplit configs, split now at 4 cpus, and perhaps change that to 8 later. There is a benefit even for singly threaded processes: kswapd can be attacking one part of the mm while another part is busy faulting. Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Diffstat (limited to 'arch/i386')
-rw-r--r--arch/i386/mm/pgtable.c8
1 files changed, 4 insertions, 4 deletions
diff --git a/arch/i386/mm/pgtable.c b/arch/i386/mm/pgtable.c
index dcdce2c6c532..39c099f15b5f 100644
--- a/arch/i386/mm/pgtable.c
+++ b/arch/i386/mm/pgtable.c
@@ -188,19 +188,19 @@ static inline void pgd_list_add(pgd_t *pgd)
188 struct page *page = virt_to_page(pgd); 188 struct page *page = virt_to_page(pgd);
189 page->index = (unsigned long)pgd_list; 189 page->index = (unsigned long)pgd_list;
190 if (pgd_list) 190 if (pgd_list)
191 pgd_list->private = (unsigned long)&page->index; 191 set_page_private(pgd_list, (unsigned long)&page->index);
192 pgd_list = page; 192 pgd_list = page;
193 page->private = (unsigned long)&pgd_list; 193 set_page_private(page, (unsigned long)&pgd_list);
194} 194}
195 195
196static inline void pgd_list_del(pgd_t *pgd) 196static inline void pgd_list_del(pgd_t *pgd)
197{ 197{
198 struct page *next, **pprev, *page = virt_to_page(pgd); 198 struct page *next, **pprev, *page = virt_to_page(pgd);
199 next = (struct page *)page->index; 199 next = (struct page *)page->index;
200 pprev = (struct page **)page->private; 200 pprev = (struct page **)page_private(page);
201 *pprev = next; 201 *pprev = next;
202 if (next) 202 if (next)
203 next->private = (unsigned long)pprev; 203 set_page_private(next, (unsigned long)pprev);
204} 204}
205 205
206void pgd_ctor(void *pgd, kmem_cache_t *cache, unsigned long unused) 206void pgd_ctor(void *pgd, kmem_cache_t *cache, unsigned long unused)