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
namei.c
/*
* linux/fs/ufs/namei.c
*
* Migration to usage of "page cache" on May 2006 by
* Evgeniy Dushistov <dushistov@mail.ru> based on ext2 code base.
*
* Copyright (C) 1998
* Daniel Pirkl <daniel.pirkl@email.cz>
* Charles University, Faculty of Mathematics and Physics
*
* from
*
* linux/fs/ext2/namei.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)
*
* from
*
* linux/fs/minix/namei.c
*
* Copyright (C) 1991, 1992 Linus Torvalds
*
* Big-endian to little-endian byte-swapping/bitmaps by
* David S. Miller (davem@caip.rutgers.edu), 1995
*/
#include <linux/time.h>
#include <linux/fs.h>
#include "ufs_fs.h"
#include "ufs.h"
#include "util.h"
static inline int ufs_add_nondir(struct dentry *dentry, struct inode *inode)
{
int err = ufs_add_link(dentry, inode);
if (!err) {
unlock_new_inode(inode);
d_instantiate(dentry, inode);
return 0;
}
inode_dec_link_count(inode);
unlock_new_inode(inode);
iput(inode);
return err;
}
static struct dentry *ufs_lookup(struct inode * dir, struct dentry *dentry, unsigned int flags)
{
struct inode * inode = NULL;
ino_t ino;
if (dentry->d_name.len > UFS_MAXNAMLEN)
return ERR_PTR(-ENAMETOOLONG);
ino = ufs_inode_by_name(dir, &dentry->d_name);
if (ino)
inode = ufs_iget(dir->i_sb, ino);
return d_splice_alias(inode, dentry);
}
/*
* By the time this is called, we already have created
* the directory cache entry for the new file, but it
* is so far negative - it has no inode.
*
* If the create succeeds, we fill in the inode information
* with d_instantiate().
*/
static int ufs_create (struct inode * dir, struct dentry * dentry, umode_t mode,
bool excl)
{
struct inode *inode;
inode = ufs_new_inode(dir, mode);
if (IS_ERR(inode))
return PTR_ERR(inode);
inode->i_op = &ufs_file_inode_operations;
inode->i_fop = &ufs_file_operations;
inode->i_mapping->a_ops = &ufs_aops;
mark_inode_dirty(inode);
return ufs_add_nondir(dentry, inode);
}
static int ufs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
{
struct inode *inode;
int err;
if (!old_valid_dev(rdev))
return -EINVAL;
inode = ufs_new_inode(dir, mode);
err = PTR_ERR(inode);
if (!IS_ERR(inode)) {
init_special_inode(inode, mode, rdev);
ufs_set_inode_dev(inode->i_sb, UFS_I(inode), rdev);
mark_inode_dirty(inode);
err = ufs_add_nondir(dentry, inode);
}
return err;
}
static int ufs_symlink (struct inode * dir, struct dentry * dentry,
const char * symname)
{
struct super_block * sb = dir->i_sb;
int err;
unsigned l = strlen(symname)+1;
struct inode * inode;
if (l > sb->s_blocksize)
return -ENAMETOOLONG;
inode = ufs_new_inode(dir, S_IFLNK | S_IRWXUGO);
err = PTR_ERR(inode);
if (IS_ERR(inode))
return err;
if (l > UFS_SB(sb)->s_uspi->s_maxsymlinklen) {
/* slow symlink */
inode->i_op = &page_symlink_inode_operations;
inode_nohighmem(inode);
inode->i_mapping->a_ops = &ufs_aops;
err = page_symlink(inode, symname, l);
if (err)
goto out_fail;
} else {
/* fast symlink */
inode->i_op = &simple_symlink_inode_operations;
inode->i_link = (char *)UFS_I(inode)->i_u1.i_symlink;
memcpy(inode->i_link, symname, l);
inode->i_size = l-1;
}
mark_inode_dirty(inode);
return ufs_add_nondir(dentry, inode);
out_fail:
inode_dec_link_count(inode);
unlock_new_inode(inode);
iput(inode);
return err;
}
static int ufs_link (struct dentry * old_dentry, struct inode * dir,
struct dentry *dentry)
{
struct inode *inode = d_inode(old_dentry);
int error;
inode->i_ctime = current_time(inode);
inode_inc_link_count(inode);
ihold(inode);
error = ufs_add_link(dentry, inode);
if (error) {
inode_dec_link_count(inode);
iput(inode);
} else
d_instantiate(dentry, inode);
return error;
}
static int ufs_mkdir(struct inode * dir, struct dentry * dentry, umode_t mode)
{
struct inode * inode;
int err;
inode_inc_link_count(dir);
inode = ufs_new_inode(dir, S_IFDIR|mode);
err = PTR_ERR(inode);
if (IS_ERR(inode))
goto out_dir;
inode->i_op = &ufs_dir_inode_operations;
inode->i_fop = &ufs_dir_operations;
inode->i_mapping->a_ops = &ufs_aops;
inode_inc_link_count(inode);
err = ufs_make_empty(inode, dir);
if (err)
goto out_fail;
err = ufs_add_link(dentry, inode);
if (err)
goto out_fail;
unlock_new_inode(inode);
d_instantiate(dentry, inode);
return 0;
out_fail:
inode_dec_link_count(inode);
inode_dec_link_count(inode);
unlock_new_inode(inode);
iput (inode);
out_dir:
inode_dec_link_count(dir);
return err;
}
static int ufs_unlink(struct inode *dir, struct dentry *dentry)
{
struct inode * inode = d_inode(dentry);
struct ufs_dir_entry *de;
struct page *page;
int err = -ENOENT;
de = ufs_find_entry(dir, &dentry->d_name, &page);
if (!de)
goto out;
err = ufs_delete_entry(dir, de, page);
if (err)
goto out;
inode->i_ctime = dir->i_ctime;
inode_dec_link_count(inode);
err = 0;
out:
return err;
}
static int ufs_rmdir (struct inode * dir, struct dentry *dentry)
{
struct inode * inode = d_inode(dentry);
int err= -ENOTEMPTY;
if (ufs_empty_dir (inode)) {
err = ufs_unlink(dir, dentry);
if (!err) {
inode->i_size = 0;
inode_dec_link_count(inode);
inode_dec_link_count(dir);
}
}
return err;
}
static int ufs_rename(struct inode *old_dir, struct dentry *old_dentry,
struct inode *new_dir, struct dentry *new_dentry,
unsigned int flags)
{
struct inode *old_inode = d_inode(old_dentry);
struct inode *new_inode = d_inode(new_dentry);
struct page *dir_page = NULL;
struct ufs_dir_entry * dir_de = NULL;
struct page *old_page;
struct ufs_dir_entry *old_de;
int err = -ENOENT;
if (flags & ~RENAME_NOREPLACE)
return -EINVAL;
old_de = ufs_find_entry(old_dir, &old_dentry->d_name, &old_page);
if (!old_de)
goto out;
if (S_ISDIR(old_inode->i_mode)) {
err = -EIO;
dir_de = ufs_dotdot(old_inode, &dir_page);
if (!dir_de)
goto out_old;
}
if (new_inode) {
struct page *new_page;
struct ufs_dir_entry *new_de;
err = -ENOTEMPTY;
if (dir_de && !ufs_empty_dir(new_inode))
goto out_dir;
err = -ENOENT;
new_de = ufs_find_entry(new_dir, &new_dentry->d_name, &new_page);
if (!new_de)
goto out_dir;
ufs_set_link(new_dir, new_de, new_page, old_inode, 1);
new_inode->i_ctime = current_time(new_inode);
if (dir_de)
drop_nlink(new_inode);
inode_dec_link_count(new_inode);
} else {
err = ufs_add_link(new_dentry, old_inode);
if (err)
goto out_dir;
if (dir_de)
inode_inc_link_count(new_dir);
}
/*
* Like most other Unix systems, set the ctime for inodes on a
* rename.
*/
old_inode->i_ctime = current_time(old_inode);
ufs_delete_entry(old_dir, old_de, old_page);
mark_inode_dirty(old_inode);
if (dir_de) {
if (old_dir != new_dir)
ufs_set_link(old_inode, dir_de, dir_page, new_dir, 0);
else {
kunmap(dir_page);
put_page(dir_page);
}
inode_dec_link_count(old_dir);
}
return 0;
out_dir:
if (dir_de) {
kunmap(dir_page);
put_page(dir_page);
}
out_old:
kunmap(old_page);
put_page(old_page);
out:
return err;
}
const struct inode_operations ufs_dir_inode_operations = {
.create = ufs_create,
.lookup = ufs_lookup,
.link = ufs_link,
.unlink = ufs_unlink,
.symlink = ufs_symlink,
.mkdir = ufs_mkdir,
.rmdir = ufs_rmdir,
.mknod = ufs_mknod,
.rename = ufs_rename,
};
![swh spinner](/static/img/swh-spinner.gif)
Computing file changes ...