Change all iterators of devices to:
- use list_for_each
- check return of get_device_locked
- don't break until we hold the lock if we get an error

When a device's reference count hits 0, remove it from all lists, including
bus and driver lists.

Between the iterator algorithm and the guaranteed removal from the lists,
there should never be a device in a list with a reference count of 0. 
So, whenever we're iterating over the lists, we'll always have a valid 
device. We don't decrement the refcount until the next iteration of the
loop, so we're also guaranteed to get the correct next item in the list.

When adding to the global device list, we were adding devices just after their parent, while we wanted to
add them just before. That way when we iterate over the list on suspend and shutdown, we'll hit all
children before the parents.

Make sure dev->driver_list and dev->bus_list are initialized when the device is registered

Remove device from global and parent's list if registration failed.

This updates the device model locking to use device_lock when accessing all
lists (the global list, the bus' lists and the drivers' lists). Before the latter
two would use their own rwlocks. 

This also updates get_device() to return a pointer to the struct device if it 
can successfully increment the reference count. 

Between these two changes, this should prevent anything gaining an invalid 
reference to a device that is in the process of being removed:

If a device is being removed, it's reference count is 0, but it hasn't 
necessarily hasn't been removed from its bus's list. If the bus list iterator
attempts to access the device, it will take the lock, but will continue on to 
the next device because the refcount is 0 (and drop the lock).

Well, theoretically; the bus iterators still need to be changed, but that's 
coming next..

into osdl.org:/home/mochel/src/kernel/devel/linux-2.5-driverfs

Well the optimum solution there would be to create and use
`inc_preempt_count_non_preempt()'.  I don't see any
way of embedding this in kmap_atomic() or copy_to_user_atomic()
without loss of flexibility or incurring a double-inc somewhere.

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

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

Fix a race between set_page_dirty() and truncate.

The page could have been removed from the mapping while this CPU is
spinning on the lock.  __free_pages_ok() will go BUG.

This has not been observed in practice - most callers of
set_page_dirty() hold the page lock which gives exclusion from
truncate.  But zap_pte_range() does not.

A fix for this has been sent to Marcelo also.

Some direct IO fixes from Badari Pulavarty.

- off-by-one in the bounds checking in blkdev_get_blocks().

- When adding more blocks into a bio_vec, account for the current
  offset into that bio_vec.

- Fix a total ballsup in the code which calculates the total number
  of pages which are about to be put under IO.

Forward port of get_user_pages() change from 2.4.

- If the vma is marked as VM_IO area then fail the map.

  This prevents kernel deadlocks which occur when applications which
  have frame buffers mapped try to dump core.  Also prevents a kernel
  oops when a debugger is attached to a process which has an IO mmap.

- Check that the mapped page is inside mem_map[] (pfn_valid).

- inline follow_page() and remove the preempt_disable()s.  It has
  only a single callsite and is called under spinloclk.

The patch from Stephen Tweedie allows users to modify the journal
commit interval for the ext3 filesystem.

The commit interval is normally five seconds.  For portable computers
with spun-down drives it is advantageous to be able to increase the
commit interval.

There may also be advantages in decreasing the commit interval for
specialised applications such as heavily-loaded NFS servers which are
using synchronous exports.

The laptop users will also need to increase the pdflush periodic
writeback interval (/proc/sys/vm/dirty_writeback_centisecs), because
the `kupdate' activity also forces a commit.

To specify the commit interval, use

	mount -o commit=30 /dev/hda1 /mnt/whatever

or
	mount -o remount,commit=30 /dev/hda1

The commit interval is specified in units of seconds.

This is the first of three patches which reduce the amount of
kmap/kunmap traffic on highmem machines.

The workload which was tested was RAM-only dbench.  This is dominated
by copy_*_user() costs.

The three patches speed up my 4xPIII by 3%

The three patches speed up a 16P NUMA-Q by 100 to 150%

The first two patches (copy_strings and pagecache reads) speed up an
8-way by 15%.  I expect that all three patches will speed up the 8-way
by 40%.

Some of the benefit is from reduced pressure on kmap_lock.  Most of it
is from reducing the number of global TLB invalidations.


This patch fixes up copy_strings().  copy_strings does a huge amount of
kmapping.  Martin Bligh has noted that across a kernel compile this
function is the second or third largest user of kmaps in the kernel.

The fix is pretty simple: just hang onto the previous kmap as we we go
around the loop.  It reduces the number of kmappings from copy_strings
by a factor of 30.

Patch from Zwane which fixes a transceiver problem on his 3c905B.

Apparently the 905B's MII status register is saying that it doesn't
need preamble, but the datasheet says that it does.  So add a 905B
override for that in the device table.

This could break other 3c905B's.  I don't know.  There's only one way
to find out.

This patch has been verified to apply cleanly to 2.5.30

This patch fixes a build warning in smp.h.  register_cpu_notifier uses
struct notifier_block in its argument list.  Unfortunately, there are
places where smp.h is included before the definition of this structure.

pci driver's resume must not be called during RESUME_POWER_ON because
interrupts are still off and i8259A is not initialized [OHCI kills
machine in such case, cardbus probably too.  PCI drivers just assume
initialized interrupts.]

Second hunk fixes device_resume calls to be okay according to
documentation.

The prototype name is wrong, not the name called by the actual
implementation.  Noticed by David Miller.

Forgot to export it to modules :/

Small, but tricky: fix for check_disk_change() deadlocks.
What we do is
	a) opening block device shifted from check_partition() to
	   grok_partitions(); check_partitions() takes opened
	   struct block_device.
	b) all callers of check_disk_change() fall in two groups -
	   ones that are called only from some ->open() and ones
	   that are _never_ called from ->open().  There is no
	   middle ground.  We split the thing in two functions -
	   check_disk_change() for the first class and full_check_....
	   for the second.  The former (ones inside ->open()) doesn't
	   touch partition tables but marks the bdev as "had been
	   invalidated".  In the end of do_open() we check if
	   bdev is marked and call wipe_partitions()/check_partition()
	   if it is - at that point bdev is fully set up and ready.
	c) ->bd_part_sem kludge is gone - we use ->bd_sem instead.
	   That is, do_open() on a partition grabs ->bd_sem on entire
	   disk and picks partition data while under it; do_open() on
	   entire disk rereads partition if needed before dropping
	   ->bd_sem (right before dropping it); BLKRRPART does
	   trylock on ->bd_sem and then checks ->bd_part_count -
	   same logics as before, except that we use ->bd_sem instead
	   of ->bd_part_sem.

That kills recursive open(), gives us the same exclusion rules as
we had and makes sure that actual IO (including rereading partition
tables) is done only when we are ready to do it.

It actually sounds a lot nastier than it is.  do_open() is a one sick
puppy right now, but we have everything in one place and _out_ of drivers
(and 20-odd equally sick puppies are gone from them, along with about
the same number of races).

Now we are almost ready to clean it up for good - all that remains to
do before that is to get the rest of drivers (cciss, DAC960, i2o and
a couple of ancients - xd and acsi) using per-disk gendisks.  Then
most of that crap will disappear.

BTW, the only generic ioctl remaining in the drivers is HDIO_GETGEO -
a lot of foo_ioctl() starts with if (cmd != HDIO_GETGEO) return -EINVAL; ;-)

ps2esdi.c switched to per-disk gendisks

cpqarray.c switched to per-disk gendisks

Big One.  Flushing/rereading partition tables is taken from
->revalidate() for partitioned devices; now it's done in the
caller (check_disk_change()).  BLKRRPART handling also moved
out of drivers - they are still allowed to override it (DAC960
and i2o are the only remaining ones), but common case is handled
in fs/block_dev.c.

Note: we are still only shifting stuff - bd_sem deadlocks in
check_disk_change() are still there.  However, now we have all
relevant code outside of drivers and that will allow to fix the
thing (see next patches).

->attach() for ide subdrivers explicitly calls register_disk()
instead of ata_revalidate() now; revalidate_drives() is gone -
it's not needed anymore (we _know_ that we'll read partition
table as soon as driver claims the drive; no need to mess with
bogus rereading).

->major_name for per-disk gendisks set to full name - i.e.
IDE gendisks have "hda", "hdb", etc. instead of "hd".
As the result, we kill a lot of crap in check.c::disk_name().
In particular, now we can afford ->minor_shift set to 0
for ide-cd (disk_name() was the only obstacle)

check_disk_change() converted to passing struct block_device.

Old variant is still needed for a couple of places; wrapper
is provided (__check_disk_change(kdev)).  do_open() logics
with setting ->bd_op sanitized - now we do that before calling
->open().

fix for embarrassing braino in /proc/partitions - size in kilobytes
is _half_ the size in secotrs, not twice that size...

Present both in 2.4 and 2.5 ;-/

- inline grab_cache_page() in pagemap.h, it's just a simple wrapper
  around find_or_create_page()
- rename (__)remove_inode_page to (__)remove_from_page_cache and
  move them from mm.h and swap.h to pagemap.h because they reverse
  add_to_page_cache and that's where they belong.

- __down_[read,write]_trylock, __downgrade_write implemented;
- __builtin_expect replaced with unlikely().

Set of small fixes:
- pcibios_init() must be int;
- fls() - ctlz on ev67, generic on others. This was required for
  something several kernel releases back, now it seems to be unused.
  Anyway, it shouldn't hurt, so included here.
- missing #includes, missing #if RTC_IRQ in drivers/char/rtc.c;
- define USER_HZ;
From Jeff Wiedemeier:
- rename alpha-specific config section 'General setup' to 'System setup'
  to avoid confusion with generic 'General setup';
- fix the 'bootpfile' build.

- osf_getrusage() updated for new utime/stime fields of the task_struct;
- compatibility wrappers for OSF/1 v4 readv/writev syscalls:
  forward port from 2.4.19.

Generic per-cpu areas; wrappers for SMP boot process.

This one is large mostly because of massive code deletion.
- cli, sti an so on go away;
- irq_smp.c goes to /dev/null; the only leftover (synchronize_irq)
  moved to irq.c;
- hardirq count field in the preemption counter extended to 12 bits -
  one more than required for wildfire.

From Jay Estabrook:
CIA rev 1 can't use DAC and windows 1,2 for SG.

Historically, assembly routines included libc header <alpha/regdef.h>
for OSF/1 register names. With the new kernel build system
it doesn't work anymore. Make our own copy in <include/asm>.

Hardware cpu_id to logical cpu mapping is gone.
Converted to cpu_online() etc.

- send_ipi_message() fix from Jeff Wiedemeier:
  The 2.5.30 IPI algorithm (with the to_whom == set test) incorrectly sends
  IPI messages to CPU 0 in a SMP system running with one processor. In this
  case to_whom is often 0 (cpu_present_mask & ~1UL << smp_processor_id()) which
  ends up triggering the to_whom == set case.
- migration IPI removed;

This starts a large set of alpha patches accumulated since 2.5.18 or
even earlier. All of this was reasonably well tested.
Thanks to Jeff Wiedemeier for SMP testing and fixes.

- sync up with (2.5.18?) pte/pfn/page/tlb etc. macros;
- asm-generic/tlb.h: loading unsigned long constant to unsigned int
  tlb->nr causes compiler warnings on 64 bit platforms.