OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>

This speeds up directory reads for large FAT partitions, if the buffercache
has to be filled from the drive. Following values were taken from:

        $ time find path_to_freshly_mounted_fat > /dev/null

on an otherwise idle system.

FAT with 16KB Clusters on IDE attached drive:   Factor  2
FAT with 32KB Clusters on USB2 attached drive:  Factor 10 (!)
Its less than 1/10 slower, if the buffercache is uptodate.

The patch introduces the new function fat_dir_readahead().

fat_dir_readahead() calls sb_breadahead() to readahead a whole cluster,
if the requested sector is the first one in a cluster.
It is usefull to do this, because on FAT directories occupy whole
clusters, with the exception of FAT12/FAT16 root dirs.

Readahead is only done, if the cluster's first sector is not uptodate
to avoid overhead, when the buffer cache is already uptodate.
Note that under memory pressure, the maximal byte count wasted
(read: has to be red from disk twice) is 1 cluster's size.  Thats 64KB.

fat_dir_readahead() is called from fat__get_entry().

There is also an unrelated cleanup at one spot:

        if (bh)
                brelse(bh);

is replaced with:

        brelse(bh);

brelse() can handle NULL pointer arguments by itself.
Signed-off-by: NKarsten Wiese <annabellesgarden@yahoo.de>
Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Change the sequence of operations performed during module loading to flush
the instruction cache before module parameters are processed.  If a module
has parameters of an unusual type that cannot be handled using the standard
accessor functions param_set_xxx and param_get_xxx, it has to to provide a
set of accessor functions for this type.  This requires module code to be
executed during parameter processing, which is of course only possible
after the icache has been flushed.
Signed-off-by: NThomas Koeller <thomas@koeller.dyndns.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Change the /proc/sysvipc/shm|sem|msg files to use the generic seq_file
implementation for struct ipc_ids.
Signed-off-by: NMike Waychison <mikew@google.com>
Cc: Manfred Spraul <manfred@colorfullife.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

The following two patches convert /proc/sysvipc/* to use seq_file.

This gives us the following:

 - Self-consistent IPC records in proc.
 - O(n) reading of the files themselves.

This patch:

Add a generic method for ipc types to be displayed using seq_file.  This
patch abstracts out seq_file iterating over struct ipc_ids into ipc/util.c
Signed-off-by: NMike Waychison <mikew@google.com>
Cc: Manfred Spraul <manfred@colorfullife.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

When registering an RPC cache, cache_register() always sets the owner as the
sunrpc module.  However, there are RPC caches owned by other modules.  With
the incorrect owner setting, the real owning module can be removed potentially
with an open reference to the cache from userspace.

For example, if one were to stop the nfs server and unmount the nfsd
filesystem, the nfsd module could be removed eventhough rpc.idmapd had
references to the idtoname and nametoid caches (i.e.
/proc/net/rpc/nfs4.<cachename>/channel is still open).  This resulted in a
system panic on one of our machines when attempting to restart the nfs
services after reloading the nfsd module.

The following patch adds a 'struct module *owner' field in struct
cache_detail.  The owner is further assigned to the struct proc_dir_entry
in cache_register() so that the module cannot be unloaded while user-space
daemons have an open reference on the associated file under /proc.
Signed-off-by: NBruce Allan <bwa@us.ibm.com>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Use the DMA_32BIT_MASK constant from dma-mapping.h when calling
pci_set_dma_mask() or pci_set_consistent_dma_mask() This patch includes
dma-mapping.h explicitly because it caused errors on some architectures
otherwise.  See http://marc.theaimsgroup.com/?t=108001993000001&r=1&w=2 for
details
Signed-off-by: NTobias Klauser <tklauser@nuerscht.ch>
Signed-off-by: NStelian Pop <stelian@popies.net>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

We don't reset the cache hit count until after readahead does a successful
readahead.  This seems to leave a corner case open where we miss in cache,
but don't restart the readhead right away.
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Seems pointless to require .c files to test CONFIG_PNP_DEBUG and
conditionally define DEBUG before including <linux/pnp.h>.  Just test
CONFIG_PNP_DEBUG directly in pnp.h.
Signed-off-by: NBjorn Helgaas <bjorn.helgaas@hp.com>
Cc: Adam Belay <ambx1@neo.rr.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

      Christoph Lameter <clameter@engr.sgi.com>

When using a time interpolator that is susceptible to jitter there's
potentially contention over a cmpxchg used to prevent time from going
backwards.  This is unnecessary when the caller holds the xtime write
seqlock as all readers will be blocked from returning until the write is
complete.  We can therefore allow writers to insert a new value and exit
rather than fight with CPUs who only hold a reader lock.
Signed-off-by: NAlex Williamson <alex.williamson@hp.com>
Signed-off-by: NChristoph Lameter <clameter@sgi.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Remove old obsolete event.
Signed-off-by: NStelian Pop <stelian@popies.net>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Newer Sony VAIO models (VGN-S480, VGN-S460, VGN-S3XP etc) use a new method to
initialize the SPIC device.  The new way to initialize (and disable) the
device comes directly from the AML code in the _CRS, _SRS and _DIS methods
from the DSDT table.  This patch adds support for the new models.
Signed-off-by: NErik Waling <erikw@acc.umu.se>
Signed-off-by: NStelian Pop <stelian@popies.net>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

These no longer have any users.
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

this is the last serial driver not using initcalls.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Cc: <jeff@uclinux.org>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

this gets rid of the last two explicit initializations in misc.c
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Acked-by: NGeert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

I passed init=/mylinuxrc to the kernel on the command line.  The kernel
silently dropped down to exec /sbin/init.  It turned out that /mylinuxrc
had improper permissions.  Without any warning message from the kernel that
something was wrong it took awhile to find the issue.  The patch below adds
a warning.
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

If /etc/mtab is a regular file all of the mount options (of a file system)
are written to /etc/mtab by the mount command.  The quota tools look there
for the quota strings for their operation.  If, however, /etc/mtab is a
symlink to /proc/mounts (a "good thing" in some environments) the tools
don't write anything - they assume the kernel will take care of things.

While the quota options are sent down to the kernel via the mount system
call and the file system codes handle them properly unfortunately there is
no code to echo the quota strings into /proc/mounts and the quota tools
fail in the symlink case.

The attached patchs modify the EXT[2|3] and JFS codes to add the necessary
hooks.  The show_options function of each file system in these patches
currently deal with only those things that seemed related to quotas;
especially in the EXT3 case more can be done (later?).

Jan Kara also noted the difficulty in moving these changes above the FS
codes responding similarly to myself to Andrew's comment about possible
VFS migration. Issue summary:

 - FS codes have to process the entire string of options anyway.

 - Only FS codes that use quotas must have a show_options function (for
   quotas to work properly) however quotas are only used in a small number
   of FS.

 - Since most of the quota using FS support other options these FS codes
   should have the a show_options function to show those options - and the
   quota echoing becomes virtually negligible.

Based on feedback I have modified my patches from the original:

   JFS a missing patch has been restored to the posting
   EXT[2|3] and JFS always use the show_options function
       - Each FS has at least one FS specific option displayed
       - QUOTA output is under a CONFIG_QUOTA ifdef
       - a follow-on patch will add a multitude of options for each FS
   EXT[2|3] and JFS "quota" is treated as "usrquota"
   EXT3 journalled data check for journalled quota removed
   EXT[2|3] mount when quota specified but not compiled in

 - no changes from my original patch.  I tested the patch and the codes
   warn but

 - still mount.  With all due respection I believe the comments
   otherwise were a

 - misread of the patch.  Please reread/test and comment.  XFS patch
   removed - the XFS team already made the necessary changes EXT3 mixing
   old and new quotas are handled differently (not purely exclusive)

 - if old and new quotas for the same type are used together the old
   type is silently depricated for compatability (e.g.  usrquota and
   usrjquota)

 - mixing of old and new quotas is an error (e.g.  usrjquota and
   grpquota)
Signed-off-by: NMark Bellon <mbellon@mvista.com>
Acked-by: NDave Kleikamp <shaggy@austin.ibm.com>
Cc: Jan Kara <jack@ucw.cz>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Ensure that sound/arm/Makefile is sanely organised so that additions to it
don't break all other patches out there.  This means I only have to adjust
the line numbers in my patch queue rather than having to re-generate by
hand those which touch this file.
Signed-off-by: NRussell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

ISAPNP, PNPBIOS, and PNPACPI all had their own kmalloc wrappers that
reimplemented kcalloc().  Remove the wrappers and just use kcalloc()
directly.

Note that this also removes the PNPBIOS error message when the kmalloc
fails.
Signed-off-by: NBjorn Helgaas <bjorn.helgaas@hp.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

ToPIC95 brides (and maybe some other too) require to use the ExCA registers
to power up the socket if a 16bit card is pluged.  allow socket drivers to
set a flag so that yenta does just that.  also clean up yenta_get_status()
a bit to use the new yenta_get_power() function.

Side note: ToPIC97 bridges (at least in Rev.5 i have) don't require this.

Ryan Underwood <nemesis-lists@icequake.net> said:

 According to the mail that David Hinds received from a Toshiba engineer,
 ToPIC95 and 97 do require this, and ToPIC100 does not.  Maybe you have a
 later revision.

 For all chips, 16-bit cards can be enabled through ExCA.  So doesn't it
 make sense just to make this the default behavior for all Toshiba chips,
 to avoid corner cases showing up later?

Daniel responded:

 I disagree with ryan to change anything for topic97 bridges.  they work.
 and I couldn't find (read google) any report of a topic97 breaking on
 applying power with the CB registers.

 I'm having several toshba notebooks at work (and home) with topic95,97,100
 bridges.  Only the ones with a topic95 didn't work.
Signed-off-by: NDaniel Ritz <daniel.ritz@gmx.ch>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

The size of auxiliary vector is fixed at 42 in linux/sched.h.  But it isn't
very obvious when looking at linux/elf.h.  This patch adds AT_VECTOR_SIZE
so that we can change it if necessary when a new vector is added.

Because of include file ordering problems, doing this necessitated the
extraction of the AT_* symbols into a standalone header file.
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Simple patch to radix_tree_tag_get() to return different values for non
present node and tag unset.

The function is not used by any in-kernel callers (yet), but this
information is definitely useful.
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

- There is frequent use of indirections in the radix code. This patch
  removes those indirections, makes the code more readable and allows
  the compilers to generate better code.

- Removing indirections allows the removal of several casts.

- Removing indirections allows the reduction of the radix_tree_path
  size from 3 to 2 words.

- Use pathp-> consistently.

- Remove unnecessary tmp variable in radix_tree_insert

- Separate the upper layer processing from the lowest layer in __lookup()
  in order to make it easier to understand what is going on and allow
  compilers to generate better code for the loop.
Signed-off-by: NChristoph Lameter <clameter@sgi.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: James Bottomley <James.Bottomley@steeleye.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

This adds support for powering Zaurus's video up and down.  PDA without
screen is kind of useless, so it is quite important...  I'll have to figure
out how to really control the frontlight, because LCD without that is quite
hard to read.
Signed-off-by: NPavel Machek <pavel@suse.cz>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Signed-off-by: NAlan Cox <alan@redhat.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

All users have been converted.
Signed-off-by: NAdrian Bunk <bunk@stusta.de>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Jens:

->bi_set is totally unnecessary bloat of struct bio.  Just define a proper
destructor for the bio and it already knows what bio_set it belongs too.

Peter:

Fixed the bugs.
Signed-off-by: NJens Axboe <axboe@suse.de>
Signed-off-by: NPeter Osterlund <petero2@telia.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Local symbols generated by gcc start with a `$'; no point in including them
in the kernel.
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

This patch contains the following cleanups:
- make needlessly global functions static
- journal.c: remove the unused global function __journal_internal_check
             and move the check to journal_init
- remove the following write-only global variable:
  - journal.c: current_journal
- remove the following unneeded EXPORT_SYMBOL:
  - journal.c: journal_recover
Signed-off-by: NAdrian Bunk <bunk@stusta.de>
Acked-by: NAndreas Dilger <adilger@clusterfs.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

When I first wrote the compat layer patches, I was somewhat cavalier about
the definition of compat_uid_t and compat_gid_t (or maybe I just
misunderstood :-)).  This patch makes the compat types much more consistent
with the types we are being compatible with and hopefully will fix a few
bugs along the way.

	compat type		type in compat arch
	__compat_[ug]id_t	__kernel_[ug]id_t
	__compat_[ug]id32_t	__kernel_[ug]id32_t
	compat_[ug]id_t		[ug]id_t

The difference is that compat_uid_t is always 32 bits (for the archs we
care about) but __compat_uid_t may be 16 bits on some.
Signed-off-by: NStephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

In sound/isa/Kconfig, select ISAPNP and depend on ISAPNP are intermixed,
resulting in funny behaviour.  (Soundcarts get selectable if other
soundcards are selected).

This patch changes the "depend on ISAPNP"s to select.
Signed-Off-By: NBodo Eggert <7eggert@gmx.de>
Signed-off-by: NAdrian Bunk <bunk@stusta.de>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Bypass an inotify-related fastpath spinlock and several function calls on
systems which have no inotify watches registered.
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Uses of RCU for dynamically changeable NMI handlers need to use the new
rcu_dereference() and rcu_assign_pointer() facilities.  This change makes
it clear that these uses are safe from a memory-barrier viewpoint, but the
main purpose is to document exactly what operations are being protected by
RCU.  This has been tested on x86 and x86-64, which are the only
architectures affected by this change.

Signed-off-by: <paulmck@us.ibm.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

The attached patch prevents oopses interleaving with characters from
other printks on other CPUs by only breaking the lock if the oops is
happening on the machine holding the lock.

It might be better if the oops generator got the lock and then called an
inner vprintk routine that assumed the caller holds the lock, thus
making oops reports "atomic".
Signed-Off-By: NDavid Howells <dhowells@redhat.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Move some more frequently read variables that showed up during some of our
performance tests as sometimes ending up in hot cachelines to the
read_mostly section.

Fix: Move the __read_mostly from before hpet_usec_quotient to follow the
variable like the other uses of __read_mostly.
Signed-off-by: NAlok N Kataria <alokk@calsoftinc.com>
Signed-off-by: NChristoph Lameter <christoph@scalex86.org>
Signed-off-by: NShai Fultheim <shai@scalex86.org>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

This patch changes the way the compression algorithm works.  The base
algorithm is similiar to the previous but we force the compressed token
size to 2.

Having a fixed size compressed token allows for a lot of optimizations, and
that in turn allows this code to run over *all* the symbols faster than it
did before over just a subset.

Having it work over all the symbols will make it behave better when symbols
change positions between passes, and the "inconsistent kallsyms" messages
should become less frequent.

In my tests the compression ratio was degraded by about 0.5%, but the
results will depend greatly on the number of symbols to compress.
Signed-off-by: NPaulo Marques <pmarques@grupopie.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Here's the latest version of relayfs, against linux-2.6.11-mm2.  I'm hoping
you'll consider putting this version back into your tree - the previous
rounds of comment seem to have shaken out all the API issues and the number
of comments on the code itself have also steadily dwindled.

This patch is essentially the same as the relayfs redux part 5 patch, with
some minor changes based on reviewer comments.  Thanks again to Pekka
Enberg for those.  The patch size without documentation is now a little
smaller at just over 40k.  Here's a detailed list of the changes:

- removed the attribute_flags in relay open and changed it to a
  boolean specifying either overwrite or no-overwrite mode, and removed
  everything referencing the attribute flags.
- added a check for NULL names in relayfs_create_entry()
- got rid of the unnecessary multiple labels in relay_create_buf()
- some minor simplification of relay_alloc_buf() which got rid of a
  couple params
- updated the Documentation

In addition, this version (through code contained in the relay-apps tarball
linked to below, not as part of the relayfs patch) tries to make it as easy
as possible to create the cooperating kernel/user pieces of a typical and
common type of logging application, one where kernel logging is kicked off
when a user space data collection app starts and stops when the collection
app exits, with the data being automatically logged to disk in between.  To
create this type of application, you basically just include a header file
(relay-app.h, included in the relay-apps tarball) in your kernel module,
define a couple of callbacks and call an initialization function, and on
the user side call a single function that sets up and continuously monitors
the buffers, and writes data to files as it becomes available.  Channels
are created when the collection app is started and destroyed when it exits,
not when the kernel module is inserted, so different channel buffer sizes
can be specified for each separate run via command-line options.  See the
README in the relay-apps tarball for details.

Also included in the relay-apps tarball are a couple examples
demonstrating how you can use this to create quick and dirty kernel
logging/debugging applications.  They are:

- tprintk, short for 'tee printk', which temporarily puts a kprobe on
  printk() and writes a duplicate stream of printk output to a relayfs
  channel.  This could be used anywhere there's printk() debugging code
  in the kernel which you'd like to exercise, but would rather not have
  your system logs cluttered with debugging junk.  You'd probably want
  to kill klogd while you do this, otherwise there wouldn't be much
  point (since putting a kprobe on printk() doesn't change the output
  of printk()).  I've used this method to temporarily divert the packet
  logging output of the iptables LOG target from the system logs to
  relayfs files instead, for instance.

- klog, which just provides a printk-like formatted logging function
  on top of relayfs.  Again, you can use this to keep stuff out of your
  system logs if used in place of printk.

The example applications can be found here:

http://prdownloads.sourceforge.net/dprobes/relay-apps.tar.gz?download

From: Christoph Hellwig <hch@lst.de>

  avoid lookup_hash usage in relayfs
Signed-off-by: NTom Zanussi <zanussi@us.ibm.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

This patch adds a new kernel debug feature: CONFIG_DETECT_SOFTLOCKUP.

When enabled then per-CPU watchdog threads are started, which try to run
once per second.  If they get delayed for more than 10 seconds then a
callback from the timer interrupt detects this condition and prints out a
warning message and a stack dump (once per lockup incident).  The feature
is otherwise non-intrusive, it doesnt try to unlock the box in any way, it
only gets the debug info out, automatically, and on all CPUs affected by
the lockup.
Signed-off-by: NIngo Molnar <mingo@elte.hu>
Signed-off-by: NNishanth Aravamudan <nacc@us.ibm.com>
Signed-Off-By: NMatthias Urlichs <smurf@smurf.noris.de>
Signed-off-by: NRichard Purdie <rpurdie@rpsys.net>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

ATM pthread_cond_signal is unnecessarily slow, because it wakes one waiter
(which at least on UP usually means an immediate context switch to one of
the waiter threads).  This waiter wakes up and after a few instructions it
attempts to acquire the cv internal lock, but that lock is still held by
the thread calling pthread_cond_signal.  So it goes to sleep and eventually
the signalling thread is scheduled in, unlocks the internal lock and wakes
the waiter again.

Now, before 2003-09-21 NPTL was using FUTEX_REQUEUE in pthread_cond_signal
to avoid this performance issue, but it was removed when locks were
redesigned to the 3 state scheme (unlocked, locked uncontended, locked
contended).

Following scenario shows why simply using FUTEX_REQUEUE in
pthread_cond_signal together with using lll_mutex_unlock_force in place of
lll_mutex_unlock is not enough and probably why it has been disabled at
that time:

The number is value in cv->__data.__lock.
        thr1            thr2            thr3
0       pthread_cond_wait
1       lll_mutex_lock (cv->__data.__lock)
0       lll_mutex_unlock (cv->__data.__lock)
0       lll_futex_wait (&cv->__data.__futex, futexval)
0                       pthread_cond_signal
1                       lll_mutex_lock (cv->__data.__lock)
1                                       pthread_cond_signal
2                                       lll_mutex_lock (cv->__data.__lock)
2                                         lll_futex_wait (&cv->__data.__lock, 2)
2                       lll_futex_requeue (&cv->__data.__futex, 0, 1, &cv->__data.__lock)
                          # FUTEX_REQUEUE, not FUTEX_CMP_REQUEUE
2                       lll_mutex_unlock_force (cv->__data.__lock)
0                         cv->__data.__lock = 0
0                         lll_futex_wake (&cv->__data.__lock, 1)
1       lll_mutex_lock (cv->__data.__lock)
0       lll_mutex_unlock (cv->__data.__lock)
          # Here, lll_mutex_unlock doesn't know there are threads waiting
          # on the internal cv's lock

Now, I believe it is possible to use FUTEX_REQUEUE in pthread_cond_signal,
but it will cost us not one, but 2 extra syscalls and, what's worse, one of
these extra syscalls will be done for every single waiting loop in
pthread_cond_*wait.

We would need to use lll_mutex_unlock_force in pthread_cond_signal after
requeue and lll_mutex_cond_lock in pthread_cond_*wait after lll_futex_wait.

Another alternative is to do the unlocking pthread_cond_signal needs to do
(the lock can't be unlocked before lll_futex_wake, as that is racy) in the
kernel.

I have implemented both variants, futex-requeue-glibc.patch is the first
one and futex-wake_op{,-glibc}.patch is the unlocking inside of the kernel.
 The kernel interface allows userland to specify how exactly an unlocking
operation should look like (some atomic arithmetic operation with optional
constant argument and comparison of the previous futex value with another
constant).

It has been implemented just for ppc*, x86_64 and i?86, for other
architectures I'm including just a stub header which can be used as a
starting point by maintainers to write support for their arches and ATM
will just return -ENOSYS for FUTEX_WAKE_OP.  The requeue patch has been
(lightly) tested just on x86_64, the wake_op patch on ppc64 kernel running
32-bit and 64-bit NPTL and x86_64 kernel running 32-bit and 64-bit NPTL.

With the following benchmark on UP x86-64 I get:

for i in nptl-orig nptl-requeue nptl-wake_op; do echo time elf/ld.so --library-path .:$i /tmp/bench; \
for j in 1 2; do echo ( time elf/ld.so --library-path .:$i /tmp/bench ) 2>&1; done; done
time elf/ld.so --library-path .:nptl-orig /tmp/bench
real 0m0.655s user 0m0.253s sys 0m0.403s
real 0m0.657s user 0m0.269s sys 0m0.388s
time elf/ld.so --library-path .:nptl-requeue /tmp/bench
real 0m0.496s user 0m0.225s sys 0m0.271s
real 0m0.531s user 0m0.242s sys 0m0.288s
time elf/ld.so --library-path .:nptl-wake_op /tmp/bench
real 0m0.380s user 0m0.176s sys 0m0.204s
real 0m0.382s user 0m0.175s sys 0m0.207s

The benchmark is at:
http://sourceware.org/ml/libc-alpha/2005-03/txt00001.txt
Older futex-requeue-glibc.patch version is at:
http://sourceware.org/ml/libc-alpha/2005-03/txt00002.txt
Older futex-wake_op-glibc.patch version is at:
http://sourceware.org/ml/libc-alpha/2005-03/txt00003.txt
Will post a new version (just x86-64 fixes so that the patch
applies against pthread_cond_signal.S) to libc-hacker ml soon.

Attached is the kernel FUTEX_WAKE_OP patch as well as a simple-minded
testcase that will not test the atomicity of the operation, but at least
check if the threads that should have been woken up are woken up and
whether the arithmetic operation in the kernel gave the expected results.
Acked-by: NIngo Molnar <mingo@redhat.com>
Cc: Ulrich Drepper <drepper@redhat.com>
Cc: Jamie Lokier <jamie@shareable.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: NYoichi Yuasa <yuasa@hh.iij4u.or.jp>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>