[PATCH] mm: split page table lock
authorHugh Dickins <hugh@veritas.com>
Sun, 30 Oct 2005 01:16:40 +0000 (18:16 -0700)
committerLinus Torvalds <torvalds@g5.osdl.org>
Sun, 30 Oct 2005 04:40:42 +0000 (21:40 -0700)
commit4c21e2f2441dc5fbb957b030333f5a3f2d02dea7
tree1f76d33bb1d76221c6424bc5fed080a4f91349a6
parentb38c6845b695141259019e2b7c0fe6c32a6e720d
[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>
23 files changed:
arch/arm/mm/mm-armv.c
arch/frv/mm/pgalloc.c
arch/i386/mm/pgtable.c
arch/um/kernel/skas/mmu.c
fs/afs/file.c
fs/buffer.c
fs/jfs/jfs_metapage.c
fs/xfs/linux-2.6/xfs_buf.c
include/linux/buffer_head.h
include/linux/mm.h
kernel/kexec.c
mm/Kconfig
mm/filemap.c
mm/memory.c
mm/mremap.c
mm/page_alloc.c
mm/page_io.c
mm/rmap.c
mm/shmem.c
mm/swap.c
mm/swap_state.c
mm/swapfile.c
mm/vmscan.c