The commit b5810039a5 contains the note
A last caveat: the ZERO_PAGE is now refcounted and managed with rmap
(and thus mapcounted and count towards shared rss). These writes to
the struct page could cause excessive cacheline bouncing on big
systems. There are a number of ways this could be addressed if it is
an issue.
And indeed this cacheline bouncing has shown up on large SGI systems.
There was a situation where an Altix system was essentially livelocked
tearing down ZERO_PAGE pagetables when an HPC app aborted during startup.
This situation can be avoided in userspace, but it does highlight the
potential scalability problem with refcounting ZERO_PAGE, and corner
cases where it can really hurt (we don't want the system to livelock!).
There are several broad ways to fix this problem:
1. add back some special casing to avoid refcounting ZERO_PAGE
2. per-node or per-cpu ZERO_PAGES
3. remove the ZERO_PAGE completely
I will argue for 3. The others should also fix the problem, but they
result in more complex code than does 3, with little or no real benefit
that I can see.
Why? Inserting a ZERO_PAGE for anonymous read faults appears to be a
false optimisation: if an application is performance critical, it would
not be doing many read faults of new memory, or at least it could be
expected to write to that memory soon afterwards. If cache or memory use
is critical, it should not be working with a significant number of
ZERO_PAGEs anyway (a more compact representation of zeroes should be
used).
As a sanity check -- mesuring on my desktop system, there are never many
mappings to the ZERO_PAGE (eg. 2 or 3), thus memory usage here should not
increase much without it.
When running a make -j4 kernel compile on my dual core system, there are
about 1,000 mappings to the ZERO_PAGE created per second, but about 1,000
ZERO_PAGE COW faults per second (less than 1 ZERO_PAGE mapping per second
is torn down without being COWed). So removing ZERO_PAGE will save 1,000
page faults per second when running kbuild, while keeping it only saves
less than 1 page clearing operation per second. 1 page clear is cheaper
than a thousand faults, presumably, so there isn't an obvious loss.
Neither the logical argument nor these basic tests give a guarantee of no
regressions. However, this is a reasonable opportunity to try to remove
the ZERO_PAGE from the pagefault path. If it is found to cause regressions,
we can reintroduce it and just avoid refcounting it.
The /dev/zero ZERO_PAGE usage and TLB tricks also get nuked. I don't see
much use to them except on benchmarks. All other users of ZERO_PAGE are
converted just to use ZERO_PAGE(0) for simplicity. We can look at
replacing them all and maybe ripping out ZERO_PAGE completely when we are
more satisfied with this solution.
Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus "snif" Torvalds <torvalds@linux-foundation.org>
Combine the file_ra_state members
unsigned long prev_index
unsigned int prev_offset
into
loff_t prev_pos
It is more consistent and better supports huge files.
Thanks to Peter for the nice proposal!
[akpm@linux-foundation.org: fix shift overflow]
Cc: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Fengguang Wu <wfg@mail.ustc.edu.cn>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Fix filesystems docbook warnings.
Warning(linux-2.6.23-git8//fs/debugfs/file.c:241): No description found for parameter 'name'
Warning(linux-2.6.23-git8//fs/debugfs/file.c:241): No description found for parameter 'mode'
Warning(linux-2.6.23-git8//fs/debugfs/file.c:241): No description found for parameter 'parent'
Warning(linux-2.6.23-git8//fs/debugfs/file.c:241): No description found for parameter 'value'
Warning(linux-2.6.23-git8//include/linux/jbd.h:404): No description found for parameter 'h_lockdep_map'
Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
* 'locks' of git://linux-nfs.org/~bfields/linux:
nfsd: remove IS_ISMNDLCK macro
Rework /proc/locks via seq_files and seq_list helpers
fs/locks.c: use list_for_each_entry() instead of list_for_each()
NFS: clean up explicit check for mandatory locks
AFS: clean up explicit check for mandatory locks
9PFS: clean up explicit check for mandatory locks
GFS2: clean up explicit check for mandatory locks
Cleanup macros for distinguishing mandatory locks
Documentation: move locks.txt in filesystems/
locks: add warning about mandatory locking races
Documentation: move mandatory locking documentation to filesystems/
locks: Fix potential OOPS in generic_setlease()
Use list_first_entry in locks_wake_up_blocks
locks: fix flock_lock_file() comment
Memory shortage can result in inconsistent flocks state
locks: kill redundant local variable
locks: reverse order of posix_locks_conflict() arguments
* git://git.linux-nfs.org/pub/linux/nfs-2.6: (131 commits)
NFSv4: Fix a typo in nfs_inode_reclaim_delegation
NFS: Add a boot parameter to disable 64 bit inode numbers
NFS: nfs_refresh_inode should clear cache_validity flags on success
NFS: Fix a connectathon regression in NFSv3 and NFSv4
NFS: Use nfs_refresh_inode() in ops that aren't expected to change the inode
SUNRPC: Don't call xprt_release in call refresh
SUNRPC: Don't call xprt_release() if call_allocate fails
SUNRPC: Fix buggy UDP transmission
[23/37] Clean up duplicate includes in
[2.6 patch] net/sunrpc/rpcb_clnt.c: make struct rpcb_program static
SUNRPC: Use correct type in buffer length calculations
SUNRPC: Fix default hostname created in rpc_create()
nfs: add server port to rpc_pipe info file
NFS: Get rid of some obsolete macros
NFS: Simplify filehandle revalidation
NFS: Ensure that nfs_link() returns a hashed dentry
NFS: Be strict about dentry revalidation when doing exclusive create
NFS: Don't zap the readdir caches upon error
NFS: Remove the redundant nfs_reval_fsid()
NFSv3: Always use directory post-op attributes in nfs3_proc_lookup
...
Fix up trivial conflict due to sock_owned_by_user() cleanup manually in
net/sunrpc/xprtsock.c
* 'nfs-server-stable' of git://linux-nfs.org/~bfields/linux:
knfsd: query filesystem for NFSv4 getattr of FATTR4_MAXNAME
knfsd: nfsv4 delegation recall should take reference on client
knfsd: don't shutdown callbacks until nfsv4 client is freed
knfsd: let nfsd manage timing out its own leases
knfsd: Add source address to sunrpc svc errors
knfsd: 64 bit ino support for NFS server
svcgss: move init code into separate function
knfsd: remove code duplication in nfsd4_setclientid()
nfsd warning fix
knfsd: fix callback rpc cred
knfsd: move nfsv4 slab creation/destruction to module init/exit
knfsd: spawn kernel thread to probe callback channel
knfsd: nfs4 name->id mapping not correctly parsing negative downcall
knfsd: demote some printk()s to dprintk()s
knfsd: cleanup of nfsd4 cmp_* functions
knfsd: delete code made redundant by map_new_errors
nfsd: fix horrible indentation in nfsd_setattr
nfsd: remove unused cache_for_each macro
nfsd: tone down inaccurate dprintk
make sync wakeups affine for cache-cold tasks: if a cache-cold task
is woken up by a sync wakeup then use the opportunity to migrate it
straight away. (the two tasks are 'related' because they communicate)
Signed-off-by: Ingo Molnar <mingo@elte.hu>
like for cpustat, introduce the "gtime" (guest time of the task) and
"cgtime" (guest time of the task children) fields for the
tasks. Modify signal_struct and task_struct.
Modify /proc/<pid>/stat to display these new fields.
Signed-off-by: Laurent Vivier <Laurent.Vivier@bull.net>
Acked-by: Avi Kivity <avi@qumranet.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
as recent CPUs introduce a third running state, after "user" and
"system", we need a new field, "guest", in cpustat to store the time
used by the CPU to run virtual CPU. Modify /proc/stat to display this
new field.
Signed-off-by: Laurent Vivier <Laurent.Vivier@bull.net>
Acked-by: Avi Kivity <avi@qumranet.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
rename all 'cnt' fields and variables to the less yucky 'count' name.
yuckage noticed by Andrew Morton.
no change in code, other than the /proc/sched_debug bkl_count string got
a bit larger:
text data bss dec hex filename
38236 3506 24 41766 a326 sched.o.before
38240 3506 24 41770 a32a sched.o.after
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
On Mon, 2007-09-24 at 22:13 -0400, Steven Rostedt wrote:
> The circular lock seems to be this:
>
> #1:
>
> sys_mmap2: down_write(&mm->mmap_sem);
> nfs_revalidate_mapping: mutex_lock(&inode->i_mutex);
>
>
> #0:
>
> vfs_readdir: mutex_lock(&inode->i_mutex);
> - during the readdir (filldir64), we take a user fault (missing page?)
> and call do_page_fault -
> do_page_fault: down_read(&mm->mmap_sem);
>
>
> So it does indeed look like a circular locking. Now the question is, "is
> this a bug?". Looking like the inode of #1 must be a file or something
> else that you can mmap and the inode of #0 seems it must be a directory.
> I would say "no".
>
> Now if you can readdir on a file or mmap a directory, then this could be
> an issue.
>
> Otherwise, I'd love to see someone teach lockdep about this issue! ;-)
Make a distinction between file and dir usage of i_mutex.
The inode should be complete and unused at unlock_new_inode(), re-init
i_mutex depending on its type.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Give each filesystem its own inode lock class. The various filesystems have
different locking order wrt the inode locks; esp. the pseudo filesystems differ
from the rest.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
commit e30408b2a9 ("JFS: fix bio-related
build breakage") removed some "return 0;" statements, rather than
changing them to null returns.
Signed-off-by: Dave Kleikamp <shaggy@linux.vnet.ibm.com>
Cc: Jeff Garzik <jeff@garzik.org>
Cc: Randy Dunlap <randy.dunlap@oracle.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
In three places: summary scan, normal scan, REF_PRISTINE GC.
Just truncate at the NUL, since that was the correct thing to do in the
only case where this (inexplicable) breakage has been seen.
Signed-off-by: David Woodhouse <dwmw2@infradead.org>
In OLPC trac #4184 we found a case where a corrupted node didn't
actually get obsoleted when we tried to garbage-collect it. So we wrote
out many million copies of it, in repeated attempts to obsolete it,
until the flash became full. Don't Do That.
Signed-off-by: David Woodhouse <dwmw2@infradead.org>
Instead of matching resv_blocks_gcmerge, which is only about 3, instead
match resv_blocks_gctrigger, which includes a proportion of the total
device size.
These ought to become tunable from userspace, at some point.
Signed-off-by: David Woodhouse <dwmw2@infradead.org>
* master.kernel.org:/pub/scm/linux/kernel/git/gregkh/driver-2.6: (75 commits)
PM: merge device power-management source files
sysfs: add copyrights
kobject: update the copyrights
kset: add some kerneldoc to help describe what these strange things are
Driver core: rename ktype_edd and ktype_efivar
Driver core: rename ktype_driver
Driver core: rename ktype_device
Driver core: rename ktype_class
driver core: remove subsystem_init()
sysfs: move sysfs file poll implementation to sysfs_open_dirent
sysfs: implement sysfs_open_dirent
sysfs: move sysfs_dirent->s_children into sysfs_dirent->s_dir
sysfs: make sysfs_root a regular directory dirent
sysfs: open code sysfs_attach_dentry()
sysfs: make s_elem an anonymous union
sysfs: make bin attr open get active reference of parent too
sysfs: kill unnecessary NULL pointer check in sysfs_release()
sysfs: kill unnecessary sysfs_get() in open paths
sysfs: reposition sysfs_dirent->s_mode.
sysfs: kill sysfs_update_file()
...
* 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2: (23 commits)
ocfs2: Optionally return filldir errors
ocfs2: Write support for directories with inline data
ocfs2: Read support for directories with inline data
ocfs2: Write support for inline data
ocfs2: Read support for inline data
ocfs2: Structure updates for inline data
ocfs2: Cleanup dirent size check
ocfs2: Rename cleanups
ocfs2: Provide convenience function for ino lookup
ocfs2: Implement ocfs2_empty_dir() as a caller of ocfs2_dir_foreach()
ocfs2: Remove open coded readdir()
ocfs2: Pass raw u64 to filldir
ocfs2: Abstract out core dir listing functionality
ocfs2: Move directory manipulation code into dir.c
ocfs2: Small refactor of truncate zeroing code
ocfs2: move nonsparse hole-filling into ocfs2_write_begin()
ocfs2: Sync ocfs2_fs.h with ocfs2-tools
[PATCH] fs/ocfs2/: removed unneeded initial value and function's return value
ocfs2: Implement show_options()
ocfs2: Clear slot map when umounting a local volume
...
Sysfs file poll implementation is scattered over sysfs and kobject.
Event numbering is done in sysfs_dirent but wait itself is done on
kobject. This not only unecessarily bloats both kobject and
sysfs_dirent but is also buggy - if a sysfs_dirent is removed while
there still are pollers, the associaton betwen the kobject and
sysfs_dirent breaks and kobject may be freed with the pollers still
sleeping on it.
This patch moves whole poll implementation into sysfs_open_dirent.
Each time a sysfs_open_dirent is created, event number restarts from 1
and pollers sleep on sysfs_open_dirent. As event sequence number is
meaningless without any open file and pollers should have open file
and thus sysfs_open_dirent, this ephemeral event counting works and is
a saner implementation.
This patch fixes the dnagling sleepers bug and reduces the sizes of
kobject and sysfs_dirent by one pointer.
Signed-off-by: Tejun Heo <htejun@gmail.com>
Acked-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Implement sysfs_open_dirent which represents an open file (attribute)
sysfs_dirent. A file sysfs_dirent with one or more open files have
one sysfs_dirent and all sysfs_buffers (one for each open instance)
are linked to it.
sysfs_open_dirent doesn't actually do anything yet but will be used to
off-load things which are specific for open file sysfs_dirent from it.
Signed-off-by: Tejun Heo <htejun@gmail.com>
Acked-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Children list head is only meaninful for directory nodes. Move it
into s_dir. This doesn't save any space currently but it will with
further changes.
Signed-off-by: Tejun Heo <htejun@gmail.com>
Acked-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
sysfs_root is different from a regular directory dirent in that it's
of type SYSFS_ROOT and doesn't have a name. These differences aren't
used by anybody and only adds to complexity. Make sysfs_root a
regular directory dirent.
Signed-off-by: Tejun Heo <htejun@gmail.com>
Acked-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
sysfs_attach_dentry() now has only one caller and isn't doing much
other than obfuscating the code. Open code and kill it.
Signed-off-by: Tejun Heo <htejun@gmail.com>
Acked-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Make s_elem an anonymous union. Prefixing with s_elem makes things
needlessly longer without any advantage.
Signed-off-by: Tejun Heo <htejun@gmail.com>
Acked-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
All bin attr operations require active references of itself and its
parent. There's no reason to allow open when its parent has been
deactivated and allowing it is inconsistent with regular sysfs file.
Use sysfs_get_active_two() in bin attribute open function.
Signed-off-by: Tejun Heo <htejun@gmail.com>
Acked-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
In sysfs_release(), sysfs_buffer pointed to by filp->private_data is
guaranteed to exist. Kill the unnecessary NULL check. This also
makes the code more consistent with the counterpart in fs/sysfs/bin.c.
Signed-off-by: Tejun Heo <htejun@gmail.com>
Acked-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
There's no reason to get an extra reference to sysfs_dirent for an
open file. Open file has a reference to the dentry which in turn has
a reference to sysfs_dirent. This is fairly obvious as otherwise open
itself won't be able to access the sysfs_dirent. Kill the extra
sysfs_get() and matching sysfs_put().
Signed-off-by: Tejun Heo <htejun@gmail.com>
Acked-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Move s_mode downward such that it's side-by-side with s_iattr which is
used for the same thing.
Signed-off-by: Tejun Heo <htejun@gmail.com>
Acked-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
sysfs_update_file() depends on inode->i_mtime but sysfs iondes are now
reclaimable making the reported modification time unreliable. There's
only one user (pci hotplug) of this notification mechanism and it
reportedly isn't utilized from userland.
Kill sysfs_update_file().
Signed-off-by: Tejun Heo <htejun@gmail.com>
Acked-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
sysfs is about to go through major overhaul making this a pretty good
opportunity to clean up (out-of-tree changes and pending patches will
need regeneration anyway). Clean up headers.
* Kill space between * and symbolname.
* Move SYSFS_* type constants and flags into fs/sysfs/sysfs.h.
They're internal to sysfs.
* Reformat function prototypes and add argument symbol names.
* Make dummy function definition order match that of function
prototypes.
* Add some comments.
* Reorganize fs/sysfs/sysfs.h according to which file the declared
variable or feature lives in.
This patch does not introduce any behavior change.
Signed-off-by: Tejun Heo <htejun@gmail.com>
Acked-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
sysfs_chmod_file() looked and updated only inode of the target file.
Dentry and inode are reclaimable and the update mode data will go away
when the inode is reclaimed. This patch makes sysfs_chmod_file()
update sd->s_mode too such that the change is permanent.
Signed-off-by: Tejun Heo <htejun@gmail.com>
Acked-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
sysfs_add/remove_one() now link and unlink the target dirent into and
from the children list. Update comments accordingly.
Signed-off-by: Tejun Heo <htejun@gmail.com>
Acked-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
We want to let people know when we create a duplicate sysfs file, as
they need to fix up their code.
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
This patch rewrites sysfs_move_dir to perform it's checks
as much as possible on the underlying sysfs_dirents instead
of the contents of the dcache, making sysfs_move_dir
more like the rest of the sysfs directory modification
code.
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Tejun Heo <htejun@gmail.com>
Cc: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
This patch rewrites sysfs_rename_dir to perform it's checks
as much as possible on the underlying sysfs_dirents instead
of the contents of the dcache. It turns out that this version
is a little simpler, and a little more like the rest of
the sysfs directory modification code.
tj: fixed double locking of sysfs_mutex
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Tejun Heo <htejun@gmail.com>
Cc: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
The only uses of s_dentry left are the code that maintains
s_dentry and trivial users that don't actually need it.
So this patch removes the s_dentry maintenance code and
restructures the trivial uses to use something else.
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Tejun Heo <htejun@gmail.com>
Cc: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Now that we know the sysfs tree structure cannot change under us and
sysfs shadow support is dropped, sysfs_get_dentry() can be simplified
greatly. It can just look up from the root and there's no need to
retry on failure.
Signed-off-by: Tejun Heo <htejun@gmail.com>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Cc: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Looking carefully at the rename code we have a subtle dependency
that the structure of sysfs not change while we are performing
a rename. If the parent directory of the object we are renaming
changes while the rename is being performed nasty things could
happen when we go to release our locks.
So introduce a sysfs_rename_mutex to prevent this highly
unlikely theoretical issue.
In addition hold sysfs_rename_mutex over all calls to
sysfs_get_dentry. Allowing sysfs_get_dentry to be simplified
in the future.
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Tejun Heo <htejun@gmail.com>
Cc: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Currently we find the dentry to drop by looking at sd->s_dentry.
We can just as easily accomplish the same task by looking up the
sysfs inode and finding all of the dentries from there, with the
added bonus that we don't need to play with the sysfs_assoc_lock.
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Tejun Heo <htejun@gmail.com>
Cc: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
