Revision b91e1302ad9b80c174a4855533f7e3aa2873355e authored by Linus Torvalds on 27 December 2016, 19:40:38 UTC, committed by Linus Torvalds on 29 December 2016, 19:03:15 UTC
In commit 62906027091f ("mm: add PageWaiters indicating tasks are
waiting for a page bit") Nick Piggin made our page locking no longer
unconditionally touch the hashed page waitqueue, which not only helps
performance in general, but is particularly helpful on NUMA machines
where the hashed wait queues can bounce around a lot.
However, the "clear lock bit atomically and then test the waiters bit"
sequence turns out to be much more expensive than it needs to be,
because you get a nasty stall when trying to access the same word that
just got updated atomically.
On architectures where locking is done with LL/SC, this would be trivial
to fix with a new primitive that clears one bit and tests another
atomically, but that ends up not working on x86, where the only atomic
operations that return the result end up being cmpxchg and xadd. The
atomic bit operations return the old value of the same bit we changed,
not the value of an unrelated bit.
On x86, we could put the lock bit in the high bit of the byte, and use
"xadd" with that bit (where the overflow ends up not touching other
bits), and look at the other bits of the result. However, an even
simpler model is to just use a regular atomic "and" to clear the lock
bit, and then the sign bit in eflags will indicate the resulting state
of the unrelated bit #7.
So by moving the PageWaiters bit up to bit #7, we can atomically clear
the lock bit and test the waiters bit on x86 too. And architectures
with LL/SC (which is all the usual RISC suspects), the particular bit
doesn't matter, so they are fine with this approach too.
This avoids the extra access to the same atomic word, and thus avoids
the costly stall at page unlock time.
The only downside is that the interface ends up being a bit odd and
specialized: clear a bit in a byte, and test the sign bit. Nick doesn't
love the resulting name of the new primitive, but I'd rather make the
name be descriptive and very clear about the limitation imposed by
trying to work across all relevant architectures than make it be some
generic thing that doesn't make the odd semantics explicit.
So this introduces the new architecture primitive
clear_bit_unlock_is_negative_byte();
and adds the trivial implementation for x86. We have a generic
non-optimized fallback (that just does a "clear_bit()"+"test_bit(7)"
combination) which can be overridden by any architecture that can do
better. According to Nick, Power has the same hickup x86 has, for
example, but some other architectures may not even care.
All these optimizations mean that my page locking stress-test (which is
just executing a lot of small short-lived shell scripts: "make test" in
the git source tree) no longer makes our page locking look horribly bad.
Before all these optimizations, just the unlock_page() costs were just
over 3% of all CPU overhead on "make test". After this, it's down to
0.66%, so just a quarter of the cost it used to be.
(The difference on NUMA is bigger, but there this micro-optimization is
likely less noticeable, since the big issue on NUMA was not the accesses
to 'struct page', but the waitqueue accesses that were already removed
by Nick's earlier commit).
Acked-by: Nick Piggin <npiggin@gmail.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Bob Peterson <rpeterso@redhat.com>
Cc: Steven Whitehouse <swhiteho@redhat.com>
Cc: Andrew Lutomirski <luto@kernel.org>
Cc: Andreas Gruenbacher <agruenba@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
1 parent 2d706e7
ialloc.c
/*
* linux/fs/ufs/ialloc.c
*
* Copyright (c) 1998
* Daniel Pirkl <daniel.pirkl@email.cz>
* Charles University, Faculty of Mathematics and Physics
*
* from
*
* linux/fs/ext2/ialloc.c
*
* Copyright (C) 1992, 1993, 1994, 1995
* Remy Card (card@masi.ibp.fr)
* Laboratoire MASI - Institut Blaise Pascal
* Universite Pierre et Marie Curie (Paris VI)
*
* BSD ufs-inspired inode and directory allocation by
* Stephen Tweedie (sct@dcs.ed.ac.uk), 1993
* Big-endian to little-endian byte-swapping/bitmaps by
* David S. Miller (davem@caip.rutgers.edu), 1995
*
* UFS2 write support added by
* Evgeniy Dushistov <dushistov@mail.ru>, 2007
*/
#include <linux/fs.h>
#include <linux/time.h>
#include <linux/stat.h>
#include <linux/string.h>
#include <linux/buffer_head.h>
#include <linux/sched.h>
#include <linux/bitops.h>
#include <asm/byteorder.h>
#include "ufs_fs.h"
#include "ufs.h"
#include "swab.h"
#include "util.h"
/*
* NOTE! When we get the inode, we're the only people
* that have access to it, and as such there are no
* race conditions we have to worry about. The inode
* is not on the hash-lists, and it cannot be reached
* through the filesystem because the directory entry
* has been deleted earlier.
*
* HOWEVER: we must make sure that we get no aliases,
* which means that we have to call "clear_inode()"
* _before_ we mark the inode not in use in the inode
* bitmaps. Otherwise a newly created file might use
* the same inode number (not actually the same pointer
* though), and then we'd have two inodes sharing the
* same inode number and space on the harddisk.
*/
void ufs_free_inode (struct inode * inode)
{
struct super_block * sb;
struct ufs_sb_private_info * uspi;
struct ufs_cg_private_info * ucpi;
struct ufs_cylinder_group * ucg;
int is_directory;
unsigned ino, cg, bit;
UFSD("ENTER, ino %lu\n", inode->i_ino);
sb = inode->i_sb;
uspi = UFS_SB(sb)->s_uspi;
ino = inode->i_ino;
mutex_lock(&UFS_SB(sb)->s_lock);
if (!((ino > 1) && (ino < (uspi->s_ncg * uspi->s_ipg )))) {
ufs_warning(sb, "ufs_free_inode", "reserved inode or nonexistent inode %u\n", ino);
mutex_unlock(&UFS_SB(sb)->s_lock);
return;
}
cg = ufs_inotocg (ino);
bit = ufs_inotocgoff (ino);
ucpi = ufs_load_cylinder (sb, cg);
if (!ucpi) {
mutex_unlock(&UFS_SB(sb)->s_lock);
return;
}
ucg = ubh_get_ucg(UCPI_UBH(ucpi));
if (!ufs_cg_chkmagic(sb, ucg))
ufs_panic (sb, "ufs_free_fragments", "internal error, bad cg magic number");
ucg->cg_time = cpu_to_fs32(sb, get_seconds());
is_directory = S_ISDIR(inode->i_mode);
if (ubh_isclr (UCPI_UBH(ucpi), ucpi->c_iusedoff, bit))
ufs_error(sb, "ufs_free_inode", "bit already cleared for inode %u", ino);
else {
ubh_clrbit (UCPI_UBH(ucpi), ucpi->c_iusedoff, bit);
if (ino < ucpi->c_irotor)
ucpi->c_irotor = ino;
fs32_add(sb, &ucg->cg_cs.cs_nifree, 1);
uspi->cs_total.cs_nifree++;
fs32_add(sb, &UFS_SB(sb)->fs_cs(cg).cs_nifree, 1);
if (is_directory) {
fs32_sub(sb, &ucg->cg_cs.cs_ndir, 1);
uspi->cs_total.cs_ndir--;
fs32_sub(sb, &UFS_SB(sb)->fs_cs(cg).cs_ndir, 1);
}
}
ubh_mark_buffer_dirty (USPI_UBH(uspi));
ubh_mark_buffer_dirty (UCPI_UBH(ucpi));
if (sb->s_flags & MS_SYNCHRONOUS)
ubh_sync_block(UCPI_UBH(ucpi));
ufs_mark_sb_dirty(sb);
mutex_unlock(&UFS_SB(sb)->s_lock);
UFSD("EXIT\n");
}
/*
* Nullify new chunk of inodes,
* BSD people also set ui_gen field of inode
* during nullification, but we not care about
* that because of linux ufs do not support NFS
*/
static void ufs2_init_inodes_chunk(struct super_block *sb,
struct ufs_cg_private_info *ucpi,
struct ufs_cylinder_group *ucg)
{
struct buffer_head *bh;
struct ufs_sb_private_info *uspi = UFS_SB(sb)->s_uspi;
sector_t beg = uspi->s_sbbase +
ufs_inotofsba(ucpi->c_cgx * uspi->s_ipg +
fs32_to_cpu(sb, ucg->cg_u.cg_u2.cg_initediblk));
sector_t end = beg + uspi->s_fpb;
UFSD("ENTER cgno %d\n", ucpi->c_cgx);
for (; beg < end; ++beg) {
bh = sb_getblk(sb, beg);
lock_buffer(bh);
memset(bh->b_data, 0, sb->s_blocksize);
set_buffer_uptodate(bh);
mark_buffer_dirty(bh);
unlock_buffer(bh);
if (sb->s_flags & MS_SYNCHRONOUS)
sync_dirty_buffer(bh);
brelse(bh);
}
fs32_add(sb, &ucg->cg_u.cg_u2.cg_initediblk, uspi->s_inopb);
ubh_mark_buffer_dirty(UCPI_UBH(ucpi));
if (sb->s_flags & MS_SYNCHRONOUS)
ubh_sync_block(UCPI_UBH(ucpi));
UFSD("EXIT\n");
}
/*
* There are two policies for allocating an inode. If the new inode is
* a directory, then a forward search is made for a block group with both
* free space and a low directory-to-inode ratio; if that fails, then of
* the groups with above-average free space, that group with the fewest
* directories already is chosen.
*
* For other inodes, search forward from the parent directory's block
* group to find a free inode.
*/
struct inode *ufs_new_inode(struct inode *dir, umode_t mode)
{
struct super_block * sb;
struct ufs_sb_info * sbi;
struct ufs_sb_private_info * uspi;
struct ufs_cg_private_info * ucpi;
struct ufs_cylinder_group * ucg;
struct inode * inode;
unsigned cg, bit, i, j, start;
struct ufs_inode_info *ufsi;
int err = -ENOSPC;
UFSD("ENTER\n");
/* Cannot create files in a deleted directory */
if (!dir || !dir->i_nlink)
return ERR_PTR(-EPERM);
sb = dir->i_sb;
inode = new_inode(sb);
if (!inode)
return ERR_PTR(-ENOMEM);
ufsi = UFS_I(inode);
sbi = UFS_SB(sb);
uspi = sbi->s_uspi;
mutex_lock(&sbi->s_lock);
/*
* Try to place the inode in its parent directory
*/
i = ufs_inotocg(dir->i_ino);
if (sbi->fs_cs(i).cs_nifree) {
cg = i;
goto cg_found;
}
/*
* Use a quadratic hash to find a group with a free inode
*/
for ( j = 1; j < uspi->s_ncg; j <<= 1 ) {
i += j;
if (i >= uspi->s_ncg)
i -= uspi->s_ncg;
if (sbi->fs_cs(i).cs_nifree) {
cg = i;
goto cg_found;
}
}
/*
* That failed: try linear search for a free inode
*/
i = ufs_inotocg(dir->i_ino) + 1;
for (j = 2; j < uspi->s_ncg; j++) {
i++;
if (i >= uspi->s_ncg)
i = 0;
if (sbi->fs_cs(i).cs_nifree) {
cg = i;
goto cg_found;
}
}
goto failed;
cg_found:
ucpi = ufs_load_cylinder (sb, cg);
if (!ucpi) {
err = -EIO;
goto failed;
}
ucg = ubh_get_ucg(UCPI_UBH(ucpi));
if (!ufs_cg_chkmagic(sb, ucg))
ufs_panic (sb, "ufs_new_inode", "internal error, bad cg magic number");
start = ucpi->c_irotor;
bit = ubh_find_next_zero_bit (UCPI_UBH(ucpi), ucpi->c_iusedoff, uspi->s_ipg, start);
if (!(bit < uspi->s_ipg)) {
bit = ubh_find_first_zero_bit (UCPI_UBH(ucpi), ucpi->c_iusedoff, start);
if (!(bit < start)) {
ufs_error (sb, "ufs_new_inode",
"cylinder group %u corrupted - error in inode bitmap\n", cg);
err = -EIO;
goto failed;
}
}
UFSD("start = %u, bit = %u, ipg = %u\n", start, bit, uspi->s_ipg);
if (ubh_isclr (UCPI_UBH(ucpi), ucpi->c_iusedoff, bit))
ubh_setbit (UCPI_UBH(ucpi), ucpi->c_iusedoff, bit);
else {
ufs_panic (sb, "ufs_new_inode", "internal error");
err = -EIO;
goto failed;
}
if (uspi->fs_magic == UFS2_MAGIC) {
u32 initediblk = fs32_to_cpu(sb, ucg->cg_u.cg_u2.cg_initediblk);
if (bit + uspi->s_inopb > initediblk &&
initediblk < fs32_to_cpu(sb, ucg->cg_u.cg_u2.cg_niblk))
ufs2_init_inodes_chunk(sb, ucpi, ucg);
}
fs32_sub(sb, &ucg->cg_cs.cs_nifree, 1);
uspi->cs_total.cs_nifree--;
fs32_sub(sb, &sbi->fs_cs(cg).cs_nifree, 1);
if (S_ISDIR(mode)) {
fs32_add(sb, &ucg->cg_cs.cs_ndir, 1);
uspi->cs_total.cs_ndir++;
fs32_add(sb, &sbi->fs_cs(cg).cs_ndir, 1);
}
ubh_mark_buffer_dirty (USPI_UBH(uspi));
ubh_mark_buffer_dirty (UCPI_UBH(ucpi));
if (sb->s_flags & MS_SYNCHRONOUS)
ubh_sync_block(UCPI_UBH(ucpi));
ufs_mark_sb_dirty(sb);
inode->i_ino = cg * uspi->s_ipg + bit;
inode_init_owner(inode, dir, mode);
inode->i_blocks = 0;
inode->i_generation = 0;
inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
ufsi->i_flags = UFS_I(dir)->i_flags;
ufsi->i_lastfrag = 0;
ufsi->i_shadow = 0;
ufsi->i_osync = 0;
ufsi->i_oeftflag = 0;
ufsi->i_dir_start_lookup = 0;
memset(&ufsi->i_u1, 0, sizeof(ufsi->i_u1));
if (insert_inode_locked(inode) < 0) {
err = -EIO;
goto failed;
}
mark_inode_dirty(inode);
if (uspi->fs_magic == UFS2_MAGIC) {
struct buffer_head *bh;
struct ufs2_inode *ufs2_inode;
/*
* setup birth date, we do it here because of there is no sense
* to hold it in struct ufs_inode_info, and lose 64 bit
*/
bh = sb_bread(sb, uspi->s_sbbase + ufs_inotofsba(inode->i_ino));
if (!bh) {
ufs_warning(sb, "ufs_read_inode",
"unable to read inode %lu\n",
inode->i_ino);
err = -EIO;
goto fail_remove_inode;
}
lock_buffer(bh);
ufs2_inode = (struct ufs2_inode *)bh->b_data;
ufs2_inode += ufs_inotofsbo(inode->i_ino);
ufs2_inode->ui_birthtime = cpu_to_fs64(sb, CURRENT_TIME.tv_sec);
ufs2_inode->ui_birthnsec = cpu_to_fs32(sb, CURRENT_TIME.tv_nsec);
mark_buffer_dirty(bh);
unlock_buffer(bh);
if (sb->s_flags & MS_SYNCHRONOUS)
sync_dirty_buffer(bh);
brelse(bh);
}
mutex_unlock(&sbi->s_lock);
UFSD("allocating inode %lu\n", inode->i_ino);
UFSD("EXIT\n");
return inode;
fail_remove_inode:
mutex_unlock(&sbi->s_lock);
clear_nlink(inode);
unlock_new_inode(inode);
iput(inode);
UFSD("EXIT (FAILED): err %d\n", err);
return ERR_PTR(err);
failed:
mutex_unlock(&sbi->s_lock);
make_bad_inode(inode);
iput (inode);
UFSD("EXIT (FAILED): err %d\n", err);
return ERR_PTR(err);
}
Computing file changes ...
