In order to demonstrate how new time-sources are added to my
timer-changes patch. Here is my current version of my cyclone-timer
patch for 2.5.41. This uses the infrastructure set up in the
timer-changes_A4 patch set to add the cyclone counter (found on IBM
Summit Based hardware) as a time-source.

        The current code is not enabled as it also depends on James
Cleverdon's 2.5 summit patch, however it illustrates how cleanly new
time-sources can be added.

        This is the final part 3 of 3 of my timer-change patch. Part 3
integrates the moved code (from part 2) into the new infrastructure
(from part 1).

        This is part 2 of 3 of my timer-change patch. Part 2 is just a
bulk move of code out of time.c and into timer_pit.c and timer_tsc.c. No
code is changed, only moved.

        Please note, this code will not compile without the final third
part of this patch collection. This was done for readability alone.

        The i386 time.c code is turning into a mess. We've got multiple
functions that do the same thing, only with different hardware, all
surrounded #ifdefs and even more difficult to follow #ifndefs. George
Anzinger is introducing a new ACPIpm time source, I'm going to attempt
to add the cyclone counter as a time source, and in the future there
will be HPET to deal with. These will not go in cleanly together as
things are now.

        Inspired by suggestions from Alan, this collection of patches
tries to clean up time.c by breaking out the PIT and TSC specific parts
into their own files. Additionally the patch creates an abstract
interface to use these existing time soruces, as well as make it easier
to add future time sources.

        It introduces "struct timer_ops" which gives the time code a
clear interface to use these different time sources. It also allows for
clearer conditional compilation of these various time sources.

        This first patch (part 1 of 3) provides the infrastructure
needed via the timer_ops structure, as well as the select_timer()
function for choosing the best available timer.

Should resolve an ongoing fiasco concerning what we should return to
userspace if they do a readv or writev of zero segments.

SuS is ambiguous, but implies EINVAL.  We're currently returning
EINVAL, but 2.4 returns zero.

I think zero makes more sense, and it is what 2.4 does.

The changed sched_yield() semantics have made ext3's transaction
batching terribly slow.

Apparently a schedule() fixes that, although it probably breaks
transaction batching.

This patch largely fixes my complaints about the new scheduler being
extremely sluggish to interactive applications.  Evidently those
applications were calling fsync() and were spending extremely long
periods in sched_yield().

From Hugh Dickins.

radix_tree_reserve() exists solely for the tmpfs move_to_swap_cache()
and move_from_swap_cache() functions, and yet they don't need it: there
is no problem in the one page being simultaneously listed in two radix
trees (while both locks are held).  Use radix_tree_insert(), and remove
radix_tree_reserve(); also removed a few blank lines.

Fix the raw driver by tricking it into performing O_DIRECT IO against
the bound blockdev.

- rewrite the i_mapping for /dev/raw/raw0 to point at the same thing
  as bdev->bd_inode->i_mapping.  We've performed a bdget() against the
  blockdev, which should pin it for the correct lifetime.

- set the O_DIRECT bit on the caller's file->flags.

The atomicicty fix for move_one_page() was not quite right.

We only do the page_table_present() test if CONFIG_HIGHPTE=y.  Which is
fine, but even with CONFIG_HIGHPTE=n, the pte mapping functions still
do an inc_preempt_count() due to their unconditional kmap_atomic().  So
we get a might_sleep() warning.

The warning is actually bogus, because those pte's are always in
direct-mapped memory.

So hm.  Three fixes suggest themselves:

1: Run the page_table_present() test if CONFIG_HIGHMEM.

   Rejected: penalises non-pte_highmem setups

2: Make kmap_atomic() not do inc_preempt_count() is the page was
   direct mapped.

   Rejected: I don't think we want kmap_atomic side effects to be
   varying according to the page which was passed.

3: Change the pte mapping functions so they don't run kmap_atomic at
   all if CONFIG_HIGHPTE=n

   This is what I did.  And guess what?  For CONFIG_HIGHMEM=y,
   CONFIG_HIGHPTE=n this patch shrinks the kernel by 5 kbytes.  Because
   kmap_atomic is inlined.

   The lesson: we do way too much damn inlining.

I have invented a new software development methodology! You send an
email to Hugh saying "I don't have the foggiest idea why this guy's
kernel is oopsing" and next morning, you get a patch! I shall patent
this.

Since 2.5.3, move_vma() has been passing a freed vma into
move_page_tables().  Fix it to move back to the previous vma in the
list if we're about to delete this one.

Thanks to Morten Helgesen for patient reporting, diagnosis and testing.

This is my latest timer patchset, it makes del_timer_sync() a bit more
robust wrt.  code that re-adds timers from the timer handler.

Other changes in the patch:

 - clean up cascading a bit.

 - do not save flags in __run_timer_list - we enter from an irqs-enabled
   tasklet.

This small patch is extracted from George Anzinger's High-res-timer
patches.

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

Motivation for this modification is that especially on some wireless
network technologies there are delay spikes that trigger RTO even though
no packets are lost. F-RTO sender continues by sending new data after RTO
retransmission in order to avoid unnecessary retransmissions in that case.  
If the sender sees any duplicate acks after the RTO retransmission, it
reverts to traditional slow start retransmissions. If new acks arrive
after forward transmissions, they very likely indicate that the RTO was
indeed spurious and the sender can continue sending new data (because 
only one segment was retransmitted).

into dot.sfbay.redhat.com:/dot/bang/rth/linux/axp-2.5

2 bug fixes for scsi pass through

When talking directly to scsi devices, the driver would
sometimes get two things wrong.  We could set too short
of a timeout. Or, we could confuse the adapter by having
non-zero values in certain fields which we shouldn't have
been using.  This patch corrects these problems.

2 minor bug fixes.

The first section makes sure we limit the size of the
sense_buffer copy to the target buffer's size so that
we don't overflow the sence_buffer.

The other sections remove some pointer arithmatic that
is wrong on 64bit machines do to padding.  Instead, just
call the pci_map functions on the buffer.

This is by far the biggest patch.  It is a rewrite of the
driver's horrid locking.  In addition to the host_lock,
the driver used to have 4 other locks per adapter!
It had a redundant ha_lock and a lock for each of 3
queues.  In a few places it also played with atomic bit
setting. And almost all of it was useless as the
host_lock was already held.

This patch cleans up this locking nightmare.  The driver
now uses the host_lock exclusively.  Only a few places
needed to add calls to lock the host_lock.  Most of
this patch is deletion of useless extra locking.

This patch adds an Adaptec copyright, makes a few
minor comment changes, and removes a few unneeded
macros.

This patch is some simple code consolidation.
A new function ips_abort_init() is created
and consolidates some repeated code that is
used if there is an error during initialization
of the adapter.

This removes several unused header includes and allows
the driver to compile by no longer trying to include
<linux/tqueue.h> . You may have already gotten a patch
to remove tqueue.h from someone else.

This patch also corrects the spelling of my last name
in the MAINTAINERS file.  You'd think I'd be used to
seeing it spelled wrong by now.

into are.twiddle.net:/home/rth/BK/axp-2.5

indent -kr -i8 -bri0 -l255 wd7000.{c,h}

Comment above getpid() is wrong.

This patch fixes it, and expands the comment to explain why on earth
we have getpid() returning ->tgid and not ->pid.

Now that brlocks loop over NR_CPUS, on SMP every br_lock/br_unlock
results in the acquire/release of 32 locks.  This incs/decs the
preempt_count by 32.

Since we only have 7 bits now for actually storing the lock depth, we
cannot nest but 3 locks deep.  I doubt we ever acquire three brlocks
concurrently, but it is still a concern.

Attached patch disables/enables preemption explicitly once and only
once for each lock/unlock.  This is also an optimization as it
removes 31 incs, decs, and conditionals. :)

Problem reported by Andrew Morton.

Not much here, just some tidying/checking. This driver can't alas use NAPI
in 2.5. Note however it has no panics or BUG()s so appears to meet the
carrier grade guidelines ;)

- Clarified authors so I get the mail not Donald
- Added missing MODULE_ bits
- Moved junk into 3c501.h

- new locking
- new_eh
- use ->page/->offset

into penguin.transmeta.com:/home/penguin/torvalds/repositories/kernel/linux

This was accidentally dropped before, but re-added now to completely mimic
behavior of the reboot notifier IDE used to have. 

  
The remove() method is generic for all drives, and set in ide_driver_t::gen_driver.
The call simply forwards the call to ide_driver_t::standby(). 

This obviates the need for IDE reboot notifier. The core iterates over all present
devices in device_shutdown() and unregisters each one. 

  
This adds
      struct device_driver    gen_driver;
  
to ide_driver_t, which is filled in with necessary fields when an ide
driver calls ide_register_driver(). That then registers the driver with
the driver model core. 
  
As a result, this gives us the following output in driverfs:
  
# tree -d /sys/bus/ide/drivers/
/sys/bus/ide/drivers/
|-- ide-cdrom
`-- ide-disk
  
The suspend/resume callbacks in ide-disk.c have been temporarily
disabled until the ide core implements generic methods which forward
the calls to the drive drivers. 

into dot.sfbay.redhat.com:/dot/bang/rth/linux/axp-2.5

into are.twiddle.net:/home/rth/BK/axp-2.5

the netfilter ipt owner module still needs the following to compile