Here is the cleanup patch I promised back in February.  Sorry it took a
while.

The effects should be purely cosmetic in 2.5.66.  However, the new
interface for the proper way to send thread-specific of process-global
signals from inside the kernel is needed for correct implementation of
some fixes to timer stuff that Ulrich told me about.

This cleans up some obsolete comments and macros in kernel/signal.c,
restores send_sig_info to its original behavior, and adds a global entry
point send_group_sig_info.  I checked all the uses of send_sig and
send_sig_info and changed a few to send_group_sig_info.

I think it would be cleanest if the whole mess of *_sig* entry points were
reduced to two or three, but I did the change that minimized the number of
callers I had to fix up.

There should be no discernible difference, since the 2.5.66 send_sig_info
function did group semantics for those signals by number already.  The only
exception to that is pdeath_signal, which I guess can be any signal number
but I deemed ought to be process-wide.

I did not change any of the calls using SIGKILL, though that does have
process-wide semantics.  There is no need to change it since SIGKILL always
kills the whole group, though the code path for send_sig(SIGKILL,...) calls
in multithreaded processes will be different now.

OK, so I screwed up - didn't notice the late_initcall() that was
introduced, which was obviously bogus. This one should build OK for the
module case. I've tested insmod/rmmod alongside a mounted sysfs.

I think the built-in case is OK: oprofile/ is after kernel/ in the link
order. I tested that too.

ACPI is performing a spin_lock() on a `void *'.  That's OK when spin_lock is
implemented via an inline function.  But when it is implemented via macros
it causes compile-time breakage.

So cast it to the right type. It really should be fixed not to use opaque
handles, though.

This fixes some performance problems.  Some vendors implement firmware
updates over IPMI, and this speeds up that process quite a bit.

 * Improve the "send - wait for response - send -wait for response -
   etc" performance when using high-res timers.  Before, an ~10ms delay
   would be added to each message, because it didn't restart the timer
   if nothing was happing when a new message was started.

 * Add some checking for leaked messages.

We really need a memory barrier in smp_call_function(), so that the
other cpu's get the updated value when they get IPI'd immediately
afterwards.

This seems to be true on some old (and arguably broken) hardware where
the IPI generation and reception doesn't synchronize enough.

into home.transmeta.com:/home/torvalds/v2.5/linux

into nuts.ninka.net:/home/davem/src/BK/net-2.5

into nuts.ninka.net:/home/davem/src/BK/net-2.5

Stops <6> appearing in middles of line and improves output.

into nuts.ninka.net:/home/davem/src/BK/sparc-2.5

1) Use PTREGS_OFF consistently
2) Define it to allocate STACKFRAME_SZ instead of REGWIN_SZ
3) Kill off REGWIN_SZ, replace with sizeof(struct reg_window).

into home.transmeta.com:/home/torvalds/v2.5/linux

into home.transmeta.com:/home/torvalds/v2.5/linux

From: Hua Zhong <hzhong@cisco.com>

The current ext3 totally ignores I/O errors that happened during a
journal_force_commit time, causing user space to falsely believe it has
succeeded, which actually did not.

This patch  checks IO error during  journal_commit_transaction. and aborts
the journal when there is I/O error.

Originally I thought about reporting the error without doing aborting the
journal, but it probably needs a new flag. Aborting the journal seems to be
the easy way to  signal "hey sth is wrong..".

ext3 allocates and frees at least one handle structure for each system call.
kmalloc and kfree are apparent in the profiles.

Adding a slab cache for these objects takes the overhead for a write() from
1.63 microseconds down to 1.56.

For an appending write, ext3_commit_write() will call the expensive
ext3_mark_inode_dirty() twice.  Once in generic_commit_write()'s extension of
i_size and once in ext3_commit_write() itself where i_disksize is updated.

But by updating i_disksize _before_ calling generic_commit_write() these can
be piggybacked.

The patch takes the overhead of a write() from 1.96 microseconds down to
1.63.

ext3_mark_inode_dirty() (and several other callers) use the
ext3_reserve_inode_write() and ext3_mark_ioc_dirty() pair for journalling an
inode's backing block.

Because ext3_reserve_inode_write() gets journalling access to the block there
is no need for ext3_mark_iloc_dirty() to do it as well.

This change reduces the overhead of a write() from 2.7 microseconds to 1.95
on a 2.7G P4.

From: Stephen Tweedie <sct@redhat.com>

The buffer_uptodate flag gets cleared on IO failure, and this can panic jbd
when it tries to write such a buffer.  Relax the panic to be just a warning.

From: Stephen Tweedie <sct@redhat.com>

Add a new form of assert failure in ext3 which allows us to flag events which
are *usually* bugs, but which can be legally triggered in the presence of IO
failures.  Don't panic the kernel on such errors unless we've defined
#JBD_PARANOID_IOFAIL, which will normally be set only for testing purposes.

The big SMP machines are seeing quite some contention in dnotify_parent()
(via vfs_write).  This function is hammering the global dparent_lock.

However we don't actually need a global dparent_lock for pinning down
dentry->d_parent.  We can use dentry->d_lock for this.  That is already being
held across d_move.

This patch speeds up SDET on the 16-way by 5% and wipes dnotify_parent() off
the profiles.

It also uninlines dnofity_parent().

It also uses spin_lock(), which is faster than read_lock().

I'm not sure that we need to take both the source and target dentry's d_lock
in d_move.

The patch also does lots of s/__inline__/inline/ in dcache.h

From: Maneesh Soni <maneesh@in.ibm.com>

Here is a patch to use seqlock for real_lookup race with d_lookup as suggested
by Linus. The race condition can result in duplicate dentry when d_lookup
fails due concurrent d_move in some unrelated directory.

Apart from real_lookup, lookup_hash()->cached_lookup() can also fail due
to same reason. So, for that I am doing the d_lookup again.

Now we have __d_lookup (called from do_lookup() during pathwalk) and
d_lookup which uses seqlock to protect againt rename race.

dcachebench numbers (lower is better) don't have much difference on a 4-way
PIII xeon SMP box.

base-2565
Average usec/iteration  19059.4
Standard Deviation      503.07

base-2565 + seq_lock
Average usec/iteration  18843.2
Standard Deviation      450.57

From: Neil Brown and myself.

Don't do dput() inside read_lock().  It can sleep.

From: Andries.Brouwer@cwi.nl

Make sure structs used by umsdos ioctls do not change size when the size of
dev_t is changed.

From: Andre Landwehr <andre.landwehr@gmx.net>

with / on an IDE harddisk the disks partitions do not appear in
devfs, only the disc device. This is due to rescan_partitions
being called twice and deleting but not re-creating the entries
during the second call.

hch has acked this.

From: Samuel Thibault <Samuel.Thibault@ens-lyon.fr>

There is no way for a braille device driven by brltty (userland root-owned
daemon) to scrollback the virtual console, the only way is to use the pc
keyboard.  A very simple TIOCLINUX ioctl meets this need (tested).

Also add a command for bringing the last console to the top, as keyboard.c's
lastcons() does when pressing alt - down arrow.

This changes the fadvise(FADV_DONTNEED) operation to start async writeout of
any dirty pages in the file.

The thinking is that if the application doesn't want to use those pages in
the future, we may as well get IO underway against them so they can be freed
up on the next call to fadvise().

The POSIX spec does not go into any detail as to whether this is the right or
wrong behaviour.

This provides a nice way for applications whihc are writing streaming data
(the main users of fadvise) to keep the amount of dirty pagecache under
control without having to resort to system-wide VM tuning.

It also provides an "async fsync()".  If the application passes in a length
of zero, fadvise will start async writeout of the pages, but will not
invalidate any of the file's pagecache.

From: "J.A. Magallon" <jamagallon@able.es>

Adds support for a couple of 3c980 variants which are in pci.ids, but not in
the driver.