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

The following patch fixes the dentry refcounting, during
sysfs_remove_group() and also adds the missing dput() for the "extra" ref
taken during sysfs_create() for the sub-directory dentry corresponding to
attribute group.

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

I have re-done the patch fixing the race between sysfs_remove_dir() and
dcache_readdir().  If you recall, sysfs_remove_dir(kobj) manipulates the
->d_subdirs list for the dentry corresponding to the sysfs directory being
removed.  It can end up deleting the cursor dentry which is added to the
->d_subdirs list during a concurrent dcache_dir_open() ==> dcache_readdir()
for the same directory.  And as a result dcache_readdir() can loop for ever
holding dcache_lock.

The earlier patch which was included in -mm1 created problems which
resulted in list_del() BUG hits in prune_dcache().  The reason I think is
that in the main loop in sysfs_remove_dir(), dcache_lock is dropped and
re-acquired, and this could result in inconsistent ->d_subdirs list and
prune_dcache() may try to delete an already deleted dentry.  I have
corrected this in the new patch as below.

I could do sysfs_remove_dir() more neatly on sysfs backing store patch set
as there I don't use the ->d_subdirs list.  Instead the list of children
sysfs_dirent works out well.  But untill sysfs backing store patch is
picked up the existing code suffer from this race.  This can be easily
tested by running following two loops on a SMP box

# while true; do insmod drivers/net/dummy.ko; rmmod dummy; done
# while true; do find /sys/class/net > /dev/null; done


o This patch fixes sysfs_remove_dir race with dcache_readdir.  There is
  no need for sysfs_remove_dir to modify the d_subdirs list for the
  directory being deleted as it is taken care in the final dput.  Modifying
  this list results in inconsistent d_subdirs list and causes infinite loop
  in concurrently occurring dcache_readdir.

o The main loop is restarted every time, dcache_lock is re-acquired in
  order to maintain consistency.

From: Anton Blanchard <anton@samba.org>

Introduce dma_error() and pci_dma_error() which are used to detect failures
in pci_map_single.

From: Anton Blanchard <anton@samba.org>

- Fix for machines with 3GB IO holes (eg nighthawk).
- Increase the maximum number of PHBs and warn if we exceed this (we used
  to walk off the end of the array)
- Only allocate an 8MB TCE table on POWER4

From: Paul Mackerras <paulus@samba.org>

Recently we found a particularly nasty bug in the segment handling in the
ppc64 kernel.  It would only happen rarely under heavy load, but when it
did the machine would lock up with the whole of memory filled with
exception stack frames.

The primary cause was that we were losing the translation for the kernel
stack from the SLB, but we still had it in the ERAT for a while longer.
Now, there is a critical region in various exception exit paths where we
have loaded the SRR0 and SRR1 registers from GPRs and we are loading those
GPRs and the stack pointer from the exception frame on the kernel stack.
If we lose the ERAT entry for the kernel stack in that region, we take an
SLB miss on the next access to the kernel stack.  Taking the exception
overwrites the values we have put into SRR0 and SRR1, which means we lose
state.  In fact we ended up repeating that last section of the exception
exit path, but using the user stack pointer this time.  That caused another
exception (or if it didn't, we loaded a new value from the user stack and
then went around and tried to use that).  And it spiralled downwards from
there.

The patch below fixes the primary problem by making sure that we really
never cast out the SLB entry for the kernel stack.  It also improves
debuggability in case anything like this happens again by:

- In our exception exit paths, we now check whether the RI bit in the
  SRR1 value is 0.  We already set the RI bit to 0 before starting the
  critical region, but we never checked it.  Now, if we do ever get an
  exception in one of the critical regions, we will detect it before
  returning to the critical region, and instead we will print a nasty
  message and oops.

- In the exception entry code, we now check that the kernel stack pointer
  value we're about to use isn't a userspace address.  If it is, we print a
  nasty message and oops.

This has been tested on G5 and pSeries (both with and without hypervisor)
and compile-tested on iSeries.

From: Anton Blanchard <anton@samba.org>

Add numa=off command line option to disable NUMA support at runtime.
Useful if there are issues with our parsing of the NUMA toplogy or for
testing NUMA gains.

From: Anton Blanchard <anton@samba.org>

Remove the old __IO_DEBUG stuff and add some nice comments courtesy of x86.

From: Stephen Rothwell <sfr@canb.auug.org.au>

This patch adds the driver for the PPC64 iSeries virtual tape.

Preparation for per-mountpoint noatime, nodiratime and later -
per-mountpoint r/o.  Depends on file_accessed() patch, should go after
it.

New helper - touch_atime(mnt, dentry).  It's a wrapper for
update_atime() and that's where all future per-mountpoint checks will
go.

New inlined helper - file_accessed(file) (wrapper for update_atime())

Make sure that we don't end up with symlink mounted over something

(mount --bind is safe since we use LOOKUP_FOLLOW in pathname resolution
there).

into nuts.davemloft.net:/disk1/BK/sparc-2.6

This fixes a minor problem with fcntl.

get_close_on_exec() uses FD_ISSET() to determine the fd state, but this
is not guaranteed to be either 0 of FD_CLOEXEC.  Make that explicit.

Also, the argument of set_close_on_exec() is being AND'ed with the
literal constant 1.  Make it use an explicit FD_CLOEXEC test.

(The broken macro only triggers for non-gcc compiles, but
still..)

Original patch from Hannes Reinecke <hare@suse.de>.

This is required to have modular IDE drivers announce themselves
properly in modules.pcimap.

Implement migrate_all_tasks() which moves tasks off cpu while machine
is stopped.

The registration and unregistration of CPU notifiers should be done
under the cpucontrol sem.  They should also be exported.

include files moved to fs/hpfs/, gratitious #include removed, stuff that
doesn't have to be global made static, misindented chunk of
hpfs_readdir() put in place, etc.

	Fixed the locking scheme.  The need of extra locking was caused by
the fact that hpfs_write_inode() must update directory entry; since HPFS
directories are implemented as b-trees, we must provide protection both
against rename() (to make sure that we update the entry in right directory)
and against rebalancing of the parent.

	Old scheme had both deadlocks and races - to start with, we had no
protection against rename()/unlink()/rmdir(), since (a) locking parent
was done without any warranties that it will remain our parent and (b)
check that we still have a directory entry (== have positive nlink) was
done before we tried to lock the parent.  Moreover, iget serialization
killed two steps ago gave immediate deadlocks if iget() of parent had
triggered another hpfs_write_inode().

	New scheme introduces another per-inode semaphore (hpfs-only,
obviously) protecting the reference to parent.  It's taken on
rename/rmdir/unlink victims and inode being moved by rename.  Old semaphores
are taken only on parent(s) and only after we grab one(s) of the new kind.
hpfs_write_inode() gets the new semaphore on our inode, checks nlink and
if it's non-zero grabs parent and takes the old semaphore on it.

	Order among the semaphores of the same kind is arbitrary - the only
function that might take more than one of the same kind is hpfs_rename()
and it's serialized by VFS.

	We might get away with only one semaphore, but then the ordering
issues would bite us big way - we would have to make sure that child is
always locked before parent (hpfs_write_inode() leaves no other choice)
and while that's easy to do for almost all operations, rename() is a bitch -
as always.  And per-superblock rwsem giving rename() vs. write_inode()
exclusion on hpfs would make the entire thing too baroque for my taste.
	->readdir() takes no locks at all (protection against directory
modifications is provided by VFS exclusion), ditto for ->lookup().
	->llseek() on directories switched to use of (VFS) ->i_sem, so
it's safe from directory modifications and ->readdir() is safe from it -
no hpfs locks are needed here.

We used to have GFP_KERNEL kmalloc() done by the code that held hpfs
lock on directory.  That could trigger a call of hpfs_write_inode() and
deadlock; fixed by switch to GFP_NOFS.  Same for hpfs inodes themselves
- hpfs_write_inode() calls iget() and that could trigger both the
deadlocks (avoidable with very baroque locking scheme) and stack
overflows (unavoidable unless we kill potential recursion here).

Killed the nightmares in hpfs iget handling.  Since in some (fairly
frequent) cases hpfs_read_inode() could avoid any IO (basically, lookup
hitting a native HPFS regular file can get all data from directory
entry) hpfs had a flag passed to that sucker.  Said flag had been
protected by a semaphore lookalike made out of spit and duct-tape and
callers of iget looked like

	hpfs_lock_iget(sb, flag);
	result = iget(sb, ino);
	hpfs_unlock_iget(sb);

Since now we are calling hpfs_read_inode() directly (note that calling
it without hpfs_lock_iget() would simply break) we can forget all that
crap and get rid of the flag - caller knows what it wants to call.

BTW, that had killed one of the last sleep_on() users in fs/*/*.

	Preparation to hpfs iget locking cleanup - remaining iget() callers
replaced with explicit iget_locked() + call hpfs_read_inode()/unlock_new_inode()
if inode is new.

	1) common initialization for all paths in hpfs_read_inode() taken into
a separate helper (hpfs_init_inode())
	2) hpfs mkdir(),create(),mknod() and symlink() do not bother with
iget() anymore - they call new_inode(), do initializations and insert new
inode into icache.  Handling of OOM failures cleaned up - if we can't
allocate in-core inode, bail instead of corrupting the filesystem.
Allocating in-core inode early also avoids one of the deadlocks here
(hpfs_write_inode() from memory pressure by kmem_cache_alloc() could
deadlock on attempt to lock our directory).
	3) hpfs_write_inode() marks the inode dirty again in case if it
fails to iget() its parent directory.  Again, OOM could trigger fs corruption
here.

	hpfs_{lock,unlock}_{2,3}inodes() killed; all places that take more than
one lock have ->i_sem held by VFS on all inodes involved and all hpfs per-inode
locks are of the same type.  IOW, we can replace these guys with multiple
hpfs_lock_inode() - order doesn't matter here.

Failure exits in hpfs/namei.c merged and cleaned up.

From: Armin Schindler <armin@melware.de>

Fixed NULL pointer reference in recv_handler()

From: Armin Schindler <armin@melware.de>

Use the notifier workqueue in a cleaner way.

From: Armin Schindler <armin@melware.de>

Show debug messages if debug is enabled only.

From: "Jose R. Santos" <jrsantos@austin.ibm.com>

After discussing it with Neil, he fell that the original justification for
taking the kernel_lock on find_exported_dentry() is not longer valid and
should be safe to remove.

This patch fixes an issue while running SpecSFS where under memory
pressure, shrinking dcache cause find_exported_dentry() to allocate
disconnected dentries that later needed to be properly connected.  The
connecting part of the code was done with BKL taken which cause a sharp
drop in performance during iterations and profiles showing 75% time spent
on find_exported_dentry().  After applying the patch, time spent on the
function is reduce to <1%.

I have tested this on an 8-way machine with 56 filesystems for several days
now with no problems using ext2, ext3, xfs and jfs.

Don't panic if the journal superblock is wrecked: just fail the mount.

From: Anton Blanchard <anton@samba.org>

From: Robert Love <rml@ximian.com>

arch/ppc64/Kconfig's entry for CONFIG_PREEMPT is missing the description
after the "bool" statement, so the entry does not show up.

Also, the help description mentions a restriction that is not [any longer]
true.

From: Anton Blanchard <anton@samba.org>

Sometimes we just want to pass the error up to the kernel and let it oops.
X it is.

From: Anton Blanchard <anton@samba.org>

- remove now unused kernel syscalls.
- wrap recently added defines in #ifdef __KERNEL__, fixes glibc
  compile issue
- some of our extra syscalls used asmlinkage, some did not. Make them
  consistent

From: Tom Rini <trini@kernel.crashing.org>

The following patch comes from Paul Mackerras.  Earlier on in 2.6,
arch/ppc/boot/utils/mkprep.c was changed slightly so that it would build
and work on Solaris.  Doing this required changing from filling out
pointers to an area to filling out a local copy of the struct.  However, a
memcpy was left out, and the info is only needed on some machines to boot.
The following adds in the missing memcpy and allows for IBM PRePs to boot
from a raw floppy again.

From: Olaf Hering <olh@suse.de>

Current Linus tree adds an extra space and dot to the mkprep options.
`make all' with an smp config doesnt work.  This patch fixes it.

into ppc970.osdl.org:/home/torvalds/v2.6/linux

Please apply this fix to backout an erroneous change in loopback.c
The statistics structure is allocated separately from the
loopback_dev structure, and the current code overwrites something
other than the statistics.  In my case the scsi_cmd_pool structure.