tmpfs: make shmem_unuse more preemptible

shmem_unuse is at present an unbroken search through every swap vector page of
every tmpfs file which might be swapped, all under shmem_swaplist_lock.  This
dates from long ago, when the caller held mmlist_lock over it all too: long
gone, but there's never been much pressure for preemptible swapoff.

Make it a little more preemptible, replacing shmem_swaplist_lock by
shmem_swaplist_mutex, inserting a cond_resched in the main loop, and a
cond_resched_lock (on info->lock) at one convenient point in the
shmem_unuse_inode loop, where it has no outstanding kmap_atomic.

If we're serious about preemptible swapoff, there's much further to go e.g.
I'm stupid to let the kmap_atomics of the decreasingly significant HIGHMEM
case dictate preemptiblility for other configs.  But as in the earlier patch
to make swapoff scan ptes preemptibly, my hidden agenda is really towards
making memcgroups work, hardly about preemptibility at all.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

tmpfs: allocate on read when stacked

tmpfs is expected to limit the memory used (unless mounted with nr_blocks=0 or
size=0).  But if a stacked filesystem such as unionfs gets pages from a sparse
tmpfs file by reading holes, and then writes to them, it can easily exceed any
such limit at present.

So suppress the SGP_READ "don't allocate page" ZERO_PAGE optimization when
reading for the kernel (a KERNEL_DS check, ugh, sorry about that).  Indeed,
pessimistically mark such pages as dirty, so they cannot get reclaimed and
unaccounted by mistake.  The venerable shmem_recalc_inode code (originally to
account for the reclaim of clean pages) suffices to get the accounting right
when swappages are dropped in favour of more uptodate filepages.

This also fixes the NULL shmem_swp_entry BUG or oops in shmem_writepage,
caused by unionfs writing to a very sparse tmpfs file: to minimize memory
allocation in swapout, tmpfs requires the swap vector be allocated upfront,
which wasn't always happening in this stacked case.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

tmpfs: allow filepage alongside swappage

tmpfs has long allowed for a fresh filepage to be created in pagecache, just
before shmem_getpage gets the chance to match it up with the swappage which
already belongs to that offset.  But unionfs_writepage now does a
find_or_create_page, divorced from shmem_getpage, which leaves conflicting
filepage and swappage outstanding indefinitely, when unionfs is over tmpfs.

Therefore shmem_writepage (where a page is swizzled from file to swap) must
now be on the lookout for existing swap, ready to free it in favour of the
more uptodate filepage, instead of BUGging on that clash.  And when the
add_to_page_cache fails in shmem_unuse_inode, it must defer to an uptodate
filepage, otherwise swapoff would hang.  Whereas when add_to_page_cache fails
in shmem_getpage, it should retry in the same way it already does.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

tmpfs: move swap swizzling into shmem

move_to_swap_cache and move_from_swap_cache functions (which swizzle a page
between tmpfs page cache and swap cache, to avoid page copying) are only used
by shmem.c; and our subsequent fix for unionfs needs different treatments in
the two instances of move_from_swap_cache.  Move them from swap_state.c into
their callsites shmem_writepage, shmem_unuse_inode and shmem_getpage, making
add_to_swap_cache externally visible.

shmem.c likes to say set_page_dirty where swap_state.c liked to say
SetPageDirty: respect that diversity, which __set_page_dirty_no_writeback
makes moot (and implies we should lose that "shift page from clean_pages to
dirty_pages list" comment: it's on neither).

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

tmpfs: shuffle add_to_swap_caches

add_to_swap_cache doesn't amount to much: merge it into its sole caller
read_swap_cache_async.  But we'll be needing to call __add_to_swap_cache from
shmem.c, so promote it to the new add_to_swap_cache.  Both were static, so
there's no interface confusion to worry about.

And lose that inappropriate "Anon pages are already on the LRU" comment in the
merging: they're not already on the LRU, as Nick Piggin noticed.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
No-problems-with: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

tmpfs: move swap_state stats update

Both unionfs and memcgroups pose challenges to tmpfs and shmem.  To help fix,
it's best to move the swap swizzling functions from swap_state.c to shmem.c.
As a preliminary to that, move swap stats updating down into
__add_to_swap_cache, which will remain internal to swap_state.c.

Well, actually, just move down the incrementation of add_total: remove
noent_race and exist_race completely, they are relics of my 2.4.11 testing.
Alt-SysRq-m users will be thrilled if 2.6.25 is at last free of "race M+N"s.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

tmpfs: fix mounts when size is less than the page size

When tmpfs is mounted with a size less than one page, the number of blocks
is set to 0 which makes the tmpfs mount unlimited.  This can lead to a
quick and surprising death if someone typos a tmpfs mount command and
writes too much.

tmpfs can still be mounted as unlimited if size or nr_blocks is exactly 0,
as Documentation/filesystems/tmpfs.txt says.

Hugh: do this by rounding size up instead of down in all cases: which
slightly expands other odd-sized tmpfs mounts, but in a consistent way.

Signed-off-by: Michael Marineau <mike@marineau.org>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

shmem: factor out sbi->free_inodes manipulations

The shmem_sb_info structure has a number of free_inodes. This
value is altered in appropriate places under spinlock and with
the sbi->max_inodes != 0 check.

Consolidate these manipulations into two helpers.

This is minus 42 bytes of shmem.o and minus 4 :) lines of code.

[akpm@linux-foundation.org: fix error return values]
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

swapoff: scan ptes preemptibly

Provided that CONFIG_HIGHPTE is not set, unuse_pte_range can reduce latency
in swapoff by scanning the page table preemptibly: so long as unuse_pte is
careful to recheck that entry under pte lock.

(To tell the truth, this patch was not inspired by any cries for lower
latency here: rather, this restructuring permits a future memory controller
patch to allocate with GFP_KERNEL in unuse_pte, where before it could not.
But it would be wrong to tuck this change away inside a memcgroup patch.)

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Tested-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

swapin: fix valid_swaphandles defect

valid_swaphandles is supposed to do a quick pass over the swap map entries
neigbouring the entry which swapin_readahead is targetting, to determine for
it a range worth reading all together.  But since it always starts its search
from the beginning of the swap "cluster", a reject (free entry) there
immediately curtails the readaround, and every swapin_readahead from that
cluster is for just a single page.  Instead scan forwards and backwards around
the target entry.

Use better names for some variables: a swap_info pointer is usually called
"si" not "swapdev".  And at the end, if only the target page should be read,
return count of 0 to disable readaround, to avoid the unnecessarily repeated
call to read_swap_cache_async.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Acked-by: Rik van Riel <riel@surriel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

shmem_file_write is redundant

With the old aops, writing to a tmpfs file had to use its own special method:
the generic method would pass in a fresh page to prepare_write when the right
page was there in swapcache - which was inefficient to handle, even once we'd
concocted the code to handle it.

With the new aops, the generic method uses shmem_write_end, which lets
shmem_getpage find the right page: so now abandon shmem_file_write in favour
of the generic method.  Yes, that does do several things that tmpfs hasn't
really needed (notably balance_dirty_pages_ratelimited, which ramfs also
calls); but more use of common code is preferable.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Acked-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

shmem_getpage return page locked

In the new aops, write_begin is supposed to return the page locked: though
I've seen no ill effects, that's been overlooked in the case of
shmem_write_begin, and should be fixed.  Then shmem_write_end must unlock the
page: do so _after_ updating i_size, as we found to be important in other
filesystems (though since shmem pages don't go the usual writeback route, they
never suffered from that corruption).

For shmem_write_begin to return the page locked, we need shmem_getpage to
return the page locked in SGP_WRITE case as well as SGP_CACHE case: let's
simplify the interface and return it locked even when SGP_READ.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

shmem: SGP_QUICK and SGP_FAULT redundant

Remove SGP_QUICK from the sgp_type enum: it was for shmem_populate and has no
users now.  Remove SGP_FAULT from the enum: SGP_CACHE does just as well (and
shmem_getpage is about to return with page always locked).

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Acked-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

swapin needs gfp_mask for loop on tmpfs

Building in a filesystem on a loop device on a tmpfs file can hang when
swapping, the loop thread caught in that infamous throttle_vm_writeout.

In theory this is a long standing problem, which I've either never seen in
practice, or long ago suppressed the recollection, after discounting my load
and my tmpfs size as unrealistically high.  But now, with the new aops, it has
become easy to hang on one machine.

Loop used to grab_cache_page before the old prepare_write to tmpfs, which
seems to have been enough to free up some memory for any swapin needed; but
the new write_begin lets tmpfs find or allocate the page (much nicer, since
grab_cache_page missed tmpfs pages in swapcache).

When allocating a fresh page, tmpfs respects loop's mapping_gfp_mask, which
has __GFP_IO|__GFP_FS stripped off, and throttle_vm_writeout is designed to
break out when __GFP_IO or GFP_FS is unset; but when tmfps swaps in,
read_swap_cache_async allocates with GFP_HIGHUSER_MOVABLE regardless of the
mapping_gfp_mask - hence the hang.

So, pass gfp_mask down the line from shmem_getpage to shmem_swapin to
swapin_readahead to read_swap_cache_async to add_to_swap_cache.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Acked-by: Rik van Riel <riel@redhat.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

swapin_readahead: move and rearrange args

swapin_readahead has never sat well in mm/memory.c: move it to mm/swap_state.c
beside its kindred read_swap_cache_async.  Why were its args in a different
order?  rearrange them.  And since it was always followed by a
read_swap_cache_async of the target page, fold that in and return struct
page*.  Then CONFIG_SWAP=n no longer needs valid_swaphandles and
read_swap_cache_async stubs.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Acked-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

swapin_readahead: excise NUMA bogosity

For three years swapin_readahead has been cluttered with fanciful CONFIG_NUMA
code, advancing addr, and stepping on to the next vma at the boundary, to line
up the mempolicy for each page allocation.

It _might_ be a good idea to allocate swap more according to vma layout; but
the fact is, that's not how we do it at all, 2.6 even less than 2.4: swap is
allocated as needed for pages as they sink to the bottom of the inactive LRUs.
 Sometimes that may match vma layout, but not so often that it's worth going
to these misleading vma->vm_next lengths: rip all that out.

Originally I intended to retain the incrementation of addr, but correct its
initial value: valid_swaphandles generally supplies an offset below the target
addr (this is readaround rather than readahead), but addr has not been
adjusted accordingly, so in the interleave case it has usually been allocating
the target page from the "wrong" node (though that may not matter very much).

But look at the equivalent shmem_swapin code: either by oversight or by
design, though it has all the apparatus for choosing a new mempolicy per page,
it uses the same idx throughout, choosing the same mempolicy and interleave
node for each page of the cluster.

Which is actually a much better strategy: each node has its own LRUs and its
own kswapd, so if you're betting on any particular relationship between swap
and node, the best bet is that nearby swap entries belong to pages from the
same node - even when the mempolicy of the target page is to interleave.  And
examining a map of nodes corresponding to swap entries on a numa=fake system
bears this out.  (We could later tweak swap allocation to make it even more
likely, but this patch is merely about removing cruft.)

So, neither adjust nor increment addr in swapin_readahead, and then
shmem_swapin can use it too; the pseudo-vma to pass policy need only be set up
once per cluster, and so few fields of pvma are used, let's skip the memset -
from shmem_alloc_page also.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Andi Kleen <ak@suse.de>
Cc: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb: allow sticky directory mount option

Allow sticky directory mount option for hugetlbfs.  This allows admin
to create a shared hugetlbfs mount point for multiple users, while
prevent accidental file deletion that users may step on each other.
It is similiar to default tmpfs mount option, or typical option used
on /tmp.

Signed-off-by: Ken Chen <kenchen@google.com>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Cc: Adam Litke <agl@us.ibm.com>
Cc: David Gibson <hermes@gibson.dropbear.id.au>
Cc: William Lee Irwin III <wli@holomorphy.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

bufferhead: revert constructor removal

The constructor for buffer_head slabs was removed recently.  We need the
constructor back in slab defrag in order to insure that slab objects always
have a definite state even before we allocated them.

I think we mistakenly merged the removal of the constuctor into a cleanup
patch.  You (ie: akpm) had a test that showed that the removal of the
constructor led to a small regression.  The prior state makes things easier
for slab defrag.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

VM: allow get_page_unless_zero on compound pages

Both slab defrag and the large blocksize patches need to ability to take
refcounts on compound pages.  May be useful in other places as well.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

vmalloc: clean up page array indexing

The page array is repeatedly indexed both in vunmap and vmalloc_area_node().
Add a temporary variable to make it easier to read (and easier to patch
later).

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

is_vmalloc_addr(): Check if an address is within the vmalloc boundaries

Checking if an address is a vmalloc address is done in a couple of places.
Define a common version in mm.h and replace the other checks.

Again the include structures suck.  The definition of VMALLOC_START and
VMALLOC_END is not available in vmalloc.h since highmem.c cannot be included
there.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

i386: Resolve dependency of asm-i386/pgtable.h on highmem.h

pgtable.h does not include highmem.h but uses various constants from
highmem.h.  We cannot include highmem.h because highmem.h will in turn include
many other include files that also depend on pgtable.h

So move the definitions from highmem.h into pgtable.h.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

vmalloc: add const to void* parameters

Make vmalloc functions work the same way as kfree() and friends that
take a const void * argument.

[akpm@linux-foundation.org: fix consts, coding-style]
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Move vmalloc_to_page() to mm/vmalloc.

We already have page table manipulation for vmalloc in vmalloc.c. Move the
vmalloc_to_page() function there as well.

Move the definitions for vmalloc related functions in mm.h to a newly created
section.  A better place would be vmalloc.h but mm.h is basic and may depend
on these functions.  An alternative would be to include vmalloc.h in mm.h
(like done for vmstat.h).

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Pagecache zeroing: zero_user_segment, zero_user_segments and zero_user

Simplify page cache zeroing of segments of pages through 3 functions

zero_user_segments(page, start1, end1, start2, end2)

        Zeros two segments of the page. It takes the position where to
        start and end the zeroing which avoids length calculations and
makes code clearer.

zero_user_segment(page, start, end)

        Same for a single segment.

zero_user(page, start, length)

        Length variant for the case where we know the length.

We remove the zero_user_page macro. Issues:

1. Its a macro. Inline functions are preferable.

2. The KM_USER0 macro is only defined for HIGHMEM.

   Having to treat this special case everywhere makes the
   code needlessly complex. The parameter for zeroing is always
   KM_USER0 except in one single case that we open code.

Avoiding KM_USER0 makes a lot of code not having to be dealing
with the special casing for HIGHMEM anymore. Dealing with
kmap is only necessary for HIGHMEM configurations. In those
configurations we use KM_USER0 like we do for a series of other
functions defined in highmem.h.

Since KM_USER0 is depends on HIGHMEM the existing zero_user_page
function could not be a macro. zero_user_* functions introduced
here can be be inline because that constant is not used when these
functions are called.

Also extract the flushing of the caches to be outside of the kmap.

[akpm@linux-foundation.org: fix nfs and ntfs build]
[akpm@linux-foundation.org: fix ntfs build some more]
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Steven French <sfrench@us.ibm.com>
Cc: Michael Halcrow <mhalcrow@us.ibm.com>
Cc: <linux-ext4@vger.kernel.org>
Cc: Steven Whitehouse <swhiteho@redhat.com>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Cc: Anton Altaparmakov <aia21@cantab.net>
Cc: Mark Fasheh <mark.fasheh@oracle.com>
Cc: David Chinner <dgc@sgi.com>
Cc: Michael Halcrow <mhalcrow@us.ibm.com>
Cc: Steven French <sfrench@us.ibm.com>
Cc: Steven Whitehouse <swhiteho@redhat.com>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

gpiolib: avr32 at32ap platform support

Teach AVR32 to use the "GPIO Library" when exposing its GPIOs, so that signals
on external chips (like GPIO expanders) can easily be used.

This mostly reorganizes some existing logic, with two minor changes in
behavior:

 - The PSR registers are used instead of the previous "gpio_mask" values,
   matching AT91 behavior and removing some duplication between that role
   and that of "pinmux_mask".

 - NR_IRQs grew to acommodate a bank of external GPIOs.  Eventually this
   number should probably become a board-specific config option.

There's a debugfs dump of status for the built-in GPIOs, showing which pins
have deglitching, pullups, or open drain drive enabled, as well as the ID
string used when requesting each IRQ.

Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Acked-by: Haavard Skinnemoen <hskinnemoen@atmel.com>
Cc: Jean Delvare <khali@linux-fr.org>
Cc: Eric Miao <eric.miao@marvell.com>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Philipp Zabel <philipp.zabel@gmail.com>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Ben Gardner <bgardner@wabtec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

deprecate obsolete pca9539 driver

Use drivers/gpio/pca9539.c instead.

Signed-off-by: eric miao <eric.miao@marvell.com>
Acked-by: Ben Gardner <bgardner@wabtec.com>
Acked-by: Jean Delvare <khali@linux-fr.org>
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Haavard Skinnemoen <hskinnemoen@atmel.com>
Cc: Philipp Zabel <philipp.zabel@gmail.com>
Cc: Russell King <rmk@arm.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

gpiolib: pca9539 i2c gpio expander support

This adds a new-style I2C driver with basic support for the sixteen bit
PCA9539 GPIO expanders.  These chips have multiple registers, push-pull output
drivers, and (not supported in this patch) pin change interrupts.

Board-specific code must provide "pca9539_platform_data" with each chip's
"i2c_board_info".  That provides the GPIO numbers to be used by that chip, and
callbacks for board-specific setup/teardown logic.

Derived from drivers/i2c/chips/pca9539.c (which has no current known users).
This is faster and simpler; it uses 16-bit register access, and cache the
OUTPUT and DIRECTION registers for fast access

Signed-off-by: eric miao <eric.miao@marvell.com>
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Acked-by: Jean Delvare <khali@linux-fr.org>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Haavard Skinnemoen <hskinnemoen@atmel.com>
Cc: Philipp Zabel <philipp.zabel@gmail.com>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Ben Gardner <bgardner@wabtec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mcp23s08 spi gpio expander support

Basic driver for 8-bit SPI based MCP23S08 GPIO expander, without support for
IRQs or the shared chipselect mechanism.

Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Cc: Jean Delvare <khali@linux-fr.org>
Cc: Eric Miao <eric.miao@marvell.com>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Haavard Skinnemoen <hskinnemoen@atmel.com>
Cc: Philipp Zabel <philipp.zabel@gmail.com>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Ben Gardner <bgardner@wabtec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

gpiolib: pcf857x i2c gpio expander support

This is a new-style I2C driver for most common 8 and 16 bit I2C based
"quasi-bidirectional" GPIO expanders: pcf8574 or pcf8575, and several
compatible models (mostly faster, supporting I2C at up to 1 MHz).

The driver exposes the GPIO signals using the platform-neutral GPIO
programming interface, so they are easily accessed by other kernel code.  The
lack of such a flexible kernel API has been a big factor in the proliferation
of board-specific drivers for these chips...  stuff that rarely makes it
upstream since it's so ugly.  This driver will let such boards use standard
calls.

Since it's a new-style driver, these devices must be configured as part of
board-specific init.  That eliminates the need for error-prone manual
configuration of module parameters, and makes compatibility with legacy
drivers (pcf8574.c, pc8575.c) for these chips easier (there's a clear
either/or disjunction).

Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Acked-by: Jean Delvare <khali@linux-fr.org>
Cc: Eric Miao <eric.miao@marvell.com>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Haavard Skinnemoen <hskinnemoen@atmel.com>
Cc: Philipp Zabel <philipp.zabel@gmail.com>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Ben Gardner <bgardner@wabtec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

gpiolib support for the PXA architecture

This adds gpiolib support for the PXA architecture:
  - move all GPIO API functions from generic.c into gpio.c
  - convert the gpio_get/set_value macros into inline functions

This makes it easier to hook up GPIOs provided by external chips like
ASICs and CPLDs.

Signed-off-by: Philipp Zabel <philipp.zabel@gmail.com>
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Acked-by: Russell King <rmk+kernel@arm.linux.org.uk>
Cc: Jean Delvare <khali@linux-fr.org>
Cc: Eric Miao <eric.miao@marvell.com>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Haavard Skinnemoen <hskinnemoen@atmel.com>
Cc: Ben Gardner <bgardner@wabtec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
[ Minor ARM fixup from David Brownell folded into this ]
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

gpiolib: update Documentation/gpio.txt

Update Documentation/gpio.txt, primarily to include the new "gpiolib"
infrastructure.

Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Cc: Jean Delvare <khali@linux-fr.org>
Cc: Eric Miao <eric.miao@marvell.com>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Haavard Skinnemoen <hskinnemoen@atmel.com>
Cc: Philipp Zabel <philipp.zabel@gmail.com>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Ben Gardner <bgardner@wabtec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

gpiolib: add gpio provider infrastructure

Provide new implementation infrastructure that platforms may choose to use
when implementing the GPIO programming interface.  Platforms can update their
GPIO support to use this.  In many cases the incremental cost to access a
non-inlined GPIO should be less than a dozen instructions, with the memory
cost being about a page (total) of extra data and code.  The upside is:

  * Providing two features which were "want to have (but OK to defer)" when
    GPIO interfaces were first discussed in November 2006:

    - A "struct gpio_chip" to plug in GPIOs that aren't directly supported
by SOC platforms, but come from FPGAs or other multifunction devices
using conventional device registers (like UCB-1x00 or SM501 GPIOs,
and southbridges in PCs with more open specs than usual).

    - Full support for message-based GPIO expanders, where registers are
accessed through sleeping I/O calls.  Previous support for these
"cansleep" calls was just stubs.  (One example: the widely used
pcf8574 I2C chips, with 8 GPIOs each.)

  * Including a non-stub implementation of the gpio_{request,free}() calls,
    making those calls much more useful.  The diagnostic labels are also
    recorded given DEBUG_FS, so /sys/kernel/debug/gpio can show a snapshot
    of all GPIOs known to this infrastructure.

The driver programming interfaces introduced in 2.6.21 do not change at all;
this infrastructure is entirely below those covers.

Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Jean Delvare <khali@linux-fr.org>
Cc: Eric Miao <eric.miao@marvell.com>
Cc: Haavard Skinnemoen <hskinnemoen@atmel.com>
Cc: Philipp Zabel <philipp.zabel@gmail.com>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Ben Gardner <bgardner@wabtec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

gpiolib: add drivers/gpio directory

Add an empty drivers/gpio directory for gpiolib infrastructure and GPIO
expanders.  It will be populated by later patches.

This won't be the only place to hold such gpio_chip code.  Many external chips
add a few GPIOs as secondary functionality (such as MFD drivers) and platform
code frequently needs to closely integrate GPIO and IRQ support.

This is placed *early* in the build/link sequence since it's common for other
drivers to depend on GPIOs to do their work, so they must be initialized early
in the device_initcall() sequence.

Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Acked-by: Jean Delvare <khali@linux-fr.org>
Cc: Eric Miao <eric.miao@marvell.com>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Haavard Skinnemoen <hskinnemoen@atmel.com>
Cc: Philipp Zabel <philipp.zabel@gmail.com>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Ben Gardner <bgardner@wabtec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

iommu sg merging: call dma_set_seg_boundary in __scsi_alloc_queue()

This is a one-line patch to add the following to __scsi_alloc_queue():

dma_set_seg_boundary(dev, shost->dma_boundary);

This is the simplest approach but the result looks odd,
__scsi_alloc_queue() does:

blk_queue_segment_boundary(q, shost->dma_boundary);
dma_set_seg_boundary(dev, shost->dma_boundary);
blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));

I think that it would be better to set up segment boundary in the same
way as we did for the maximum segment size. That is, removing
shost->dma_boundary and LLDs call pci_set_dma_seg_boundary (or its
friends).

Then __scsi_alloc_queue() can set up both limits in the same way:

blk_queue_segment_boundary(q, dma_get_seg_boundary(dev));
blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));

killing dma_boundary in scsi_host_template needs a large patch for
libata (dma_boundary is used by only libata and sym53c8xx). I'll send
a patch to do that if it is acceptable. James and Jeff?

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: James Bottomley <James.Bottomley@steeleye.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Greg KH <greg@kroah.com>
Cc: Jeff Garzik <jeff@garzik.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

iommu sg merging: swiotlb: respect the segment boundary limits

This patch makes swiotlb not allocate a memory area spanning LLD's segment
boundary.

is_span_boundary() judges whether a memory area spans LLD's segment boundary.
If map_single finds such a area, map_single tries to find the next available
memory area.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: James Bottomley <James.Bottomley@steeleye.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Greg KH <greg@kroah.com>
Cc: Jeff Garzik <jeff@garzik.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

iommu sg merging: PCI: add dma segment boundary support

This adds PCI's accessor for segment_boundary_mask in device_dma_parameters.

The default segment_boundary is set to 0xffffffff, same to the block layer's
default value (and the scsi mid layer uses the same value).

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: James Bottomley <James.Bottomley@steeleye.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Greg KH <greg@kroah.com>
Cc: Jeff Garzik <jeff@garzik.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

iommu sg merging: add accessors for segment_boundary_mask in device_dma_parameters()

This adds new accessors for segment_boundary_mask in device_dma_parameters
structure in the same way I did for max_segment_size.  So we can easily change
where to place struct device_dma_parameters in the future.

dma_get_segment boundary returns 0xffffffff if dma_parms in struct device
isn't set up properly.  0xffffffff is the default value used in the block
layer and the scsi mid layer.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: James Bottomley <James.Bottomley@steeleye.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Greg KH <greg@kroah.com>
Cc: Jeff Garzik <jeff@garzik.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

iommu sg: kill __clear_bit_string and find_next_zero_string

This kills unused __clear_bit_string and find_next_zero_string (they
were used by only gart and calgary IOMMUs).

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: Jeff Garzik <jeff@garzik.org>
Cc: James Bottomley <James.Bottomley@steeleye.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Muli Ben-Yehuda <mulix@mulix.org>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

iommu sg: x86: convert gart IOMMU to use the IOMMU helper

This patch converts gart IOMMU to use the IOMMU helper functions. The
IOMMU doesn't allocate a memory area spanning LLD's segment boundary
anymore.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: Jeff Garzik <jeff@garzik.org>
Cc: James Bottomley <James.Bottomley@steeleye.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Muli Ben-Yehuda <mulix@mulix.org>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

iommu sg: x86: convert calgary IOMMU to use the IOMMU helper

This patch converts calgary IOMMU to use the IOMMU helper
functions. The IOMMU doesn't allocate a memory area spanning LLD's
segment boundary anymore.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: Jeff Garzik <jeff@garzik.org>
Cc: James Bottomley <James.Bottomley@steeleye.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Muli Ben-Yehuda <mulix@mulix.org>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

iommu sg: powerpc: remove DMA 4GB boundary protection

Previously, during initialization of the IOMMU tables, the last entry
at each 4GB boundary is marked as used since there are many adapters
which cannot handle DMAing across any 4GB boundary.

The IOMMU doesn't allocate a memory area spanning LLD's segment
boundary anymore. The segment boundary of devices are set to 4GB by
default. So we can remove 4GB boundary protection now.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: Jeff Garzik <jeff@garzik.org>
Cc: James Bottomley <James.Bottomley@steeleye.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

iommu sg: powerpc: convert iommu to use the IOMMU helper

This patch converts PPC's IOMMU to use the IOMMU helper functions.  The IOMMU
doesn't allocate a memory area spanning LLD's segment boundary anymore.

iseries_hv_alloc and iseries_hv_map don't have proper device
struct. 4GB boundary is used for them.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: Jeff Garzik <jeff@garzik.org>
Cc: James Bottomley <James.Bottomley@steeleye.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

iommu sg: add IOMMU helper functions for the free area management

This adds IOMMU helper functions for the free area management.  These
functions take care of LLD's segment boundary limit for IOMMUs.  They would be
useful for IOMMUs that use bitmap for the free area management.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: Jeff Garzik <jeff@garzik.org>
Cc: James Bottomley <James.Bottomley@steeleye.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

iommu sg merging: aacraid: use pci_set_dma_max_seg_size

This sets the segment size limit properly via pci_set_dma_max_seg_size
and remove blk_queue_max_segment_size because scsi-ml calls it.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: Jeff Garzik <jeff@garzik.org>
Cc: James Bottomley <James.Bottomley@steeleye.com>
Acked-by: Jens Axboe <jens.axboe@oracle.com>
Acked-by: "Salyzyn, Mark" <mark_salyzyn@adaptec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

iommu sg merging: sata_inic162x: use pci_set_dma_max_seg_size

This sets the segment size limit properly via pci_set_dma_max_seg_size
and remove blk_queue_max_segment_size because scsi-ml calls it.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: Jeff Garzik <jeff@garzik.org>
Cc: James Bottomley <James.Bottomley@steeleye.com>
Acked-by: Jens Axboe <jens.axboe@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

iommu sg merging: call blk_queue_segment_boundary in __scsi_alloc_queue

request_queue and device struct must have the same value of a segment
size limit. This patch adds blk_queue_segment_boundary in
__scsi_alloc_queue so LLDs don't need to call both
blk_queue_segment_boundary and set_dma_max_seg_size. A LLD can change
the default value (64KB) can call device_dma_parameters accessors like
pci_set_dma_max_seg_size when allocating scsi_host.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Acked-by: Jeff Garzik <jeff@garzik.org>
Cc: James Bottomley <James.Bottomley@steeleye.com>
Acked-by: Jens Axboe <jens.axboe@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

iommu sg merging: parisc: make iommu respect the segment size limits

This patch makes iommu respect segment size limits when merging sg
lists.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: Jeff Garzik <jeff@garzik.org>
Cc: James Bottomley <James.Bottomley@steeleye.com>
Acked-by: Jens Axboe <jens.axboe@oracle.com>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Acked-by: Grant Grundler <grundler@parisc-linux.org>
Cc: Matthew Wilcox <willy@debian.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

iommu sg merging: sparc64: make iommu respect the segment size limits

This patch makes iommu respect segment size limits when merging sg
lists.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: Jeff Garzik <jeff@garzik.org>
Cc: James Bottomley <James.Bottomley@steeleye.com>
Acked-by: Jens Axboe <jens.axboe@oracle.com>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

iommu sg merging: alpha: make pci_iommu respect the segment size limits

This patch makes pci_iommu respect segment size limits when merging sg
lists.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: Jeff Garzik <jeff@garzik.org>
Cc: James Bottomley <James.Bottomley@steeleye.com>
Acked-by: Jens Axboe <jens.axboe@oracle.com>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

iommu sg merging: IA64: make sba_iommu respect the segment size limits

This patch makes sba iommu respect segment size limits when merging sg
lists.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: Jeff Garzik <jeff@garzik.org>
Cc: James Bottomley <James.Bottomley@steeleye.com>
Acked-by: Jens Axboe <jens.axboe@oracle.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

iommu sg merging: ppc: make iommu respect the segment size limits

This patch makes iommu respect segment size limits when merging sg
lists.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: Jeff Garzik <jeff@garzik.org>
Cc: James Bottomley <James.Bottomley@steeleye.com>
Acked-by: Jens Axboe <jens.axboe@oracle.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

iommu sg merging: x86: make pci-gart iommu respect the segment size limits

This patch makes pci-gart iommu respect segment size limits when
merging sg lists.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: Jeff Garzik <jeff@garzik.org>
Cc: James Bottomley <James.Bottomley@steeleye.com>
Acked-by: Jens Axboe <jens.axboe@oracle.com>
Cc: Dave Airlie <airlied@linux.ie>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

iommu sg merging: PCI: add device_dma_parameters support

This adds struct device_dma_parameters in struct pci_dev and properly
sets up a pointer in struct device.

The default max_segment_size is set to 64K, same to the block layer's
default value.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Mostly-acked-by: Jeff Garzik <jeff@garzik.org>
Cc: James Bottomley <James.Bottomley@steeleye.com>
Acked-by: Jens Axboe <jens.axboe@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

iommu sg merging: add device_dma_parameters structure

IOMMUs merges scatter/gather segments without considering a low level
driver's restrictions. The problem is that IOMMUs can't access to the
limitations because they are in request_queue.

This patchset introduces a new structure, device_dma_parameters,
including dma information. A pointer to device_dma_parameters is added
to struct device. The bus specific structures (like pci_dev) includes
device_dma_parameters. Low level drivers can use dma_set_max_seg_size
to tell IOMMUs about the restrictions.

We can move more dma stuff in struct device (like dma_mask) to struct
device_dma_parameters later (needs some cleanups before that).

This includes patches for all the IOMMUs that could merge sg (x86_64,
ppc, IA64, alpha, sparc64, and parisc) though only the ppc patch was
tested. The patches for other IOMMUs are only compile tested.

This patch:

Add a new structure, device_dma_parameters, including dma information.  A
pointer to device_dma_parameters is added to struct device.

- there are only max_segment_size and segment_boundary_mask there but we'll
  move more dma stuff in struct device (like dma_mask) to struct
  device_dma_parameters later.  segment_boundary_mask is not supported yet.

- new accessors for the dma parameters are added.  So we can easily change
  where to place struct device_dma_parameters in the future.

- dma_get_max_seg_size returns 64K if dma_parms in struct device isn't set
  up properly.  64K is the default max_segment_size in the block layer.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Acked-by: Jeff Garzik <jeff@garzik.org>
Cc: James Bottomley <James.Bottomley@steeleye.com>
Acked-by: Jens Axboe <jens.axboe@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

serial: MPSC: set baudrate when BRG divider is set.

The clock to generate the desired baudrate with the MPSC is first divided
by the Baud Rate Generator (BRG) and then by the MPSC itself.  So, when the
BRG divider is changed, the MPSC divider must also be changed to generate
the correct baudrate.  During MPSC initialization, the BRG divider is
changed but the MPSC divider isn't changed until much later.  This results
in some printk's coming out garbled.  To fix that, set the MPSC divider at
the same time that the BRG divider is changed.

Signed-off-by: Mark A. Greer <mgreer@mvista.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

serial: Coding style

Coding style tweaks and printk levels.

Signed-off-by: Alan Cox <alan@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

serial: speed setup failure reporting

Invalid speeds are forced to 9600. Update the code for this to encode new
style baud rates properly.

Signed-off-by: Alan Cox <alan@redhat.com>
Cc: Russell King <rmk@arm.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

serial: avoid stalling suspend if serial port won't drain

Some ports seem to be unable to drain their transmitters on shut down.  Such a
problem can occur if the port is programmed for hardware imposed flow control,
characters are in the FIFO but the CTS signal is inactive.

Normally, this isn't a problem because most places where we wait for the
transmitter to drain have a time-out.  However, there is no timeout in the
suspend path.

Give a port 30ms to drain; this is an arbitary value chosen to avoid long
delays if there are many such ports in the system, while giving a reasonable
chance for a single port to drain.  Should a port not drain within this
timeout, issue a warning.

Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Acked-by: Alan Cox <alan@lxorguk.ukuu.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

serial: avoid waking up closed serial ports on resume

When we boot, serial ports remain in low power mode until they're used either
by userspace or for the kernel console.

However, if you suspend the system, and then resume, all serial ports will be
taken out of low power mode.  This is bad news for embedded devices where this
can mean higher power consumption.

Only bring a serial port out of low power mode if the port is being used as
the kernel console, or is in use by userspace.

Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Acked-by: Alan Cox <alan@lxorguk.ukuu.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

8250.c: support specifying DW APB UARTs in device platform_data

Allow the private_data field to be specified in platform_data for the
standard 8250/16550 UART.  This field is used by DW APB type UARTs and
without this patch it's only possible to set this field when registering
the port by hand.  If private_data is not set then the driver will
potentially oops with a NULL pointer dereference.

Signed-off-by: Will Newton <will.newton@gmail.com>
Acked-by: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: Russell King <rmk@arm.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

serial: add ADDI-DATA GmbH Communication cardsin8250_pci.c and pci_ids.h

Add ADDI-DATA GmbH communication cards to 8250_pci driver.  Supported cards
are:

APCI-7300, APCI-7420, APCI-7500, APCI-7800 APCI-7300-2, APCI-7420-2,
APCI-7500-2 APCI-7300-3, APCI-7420-3, APCI-7500-3, APCI-7800-3

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Krauth J. <krauth.julien@addi-data.com>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Alan Cox <alan@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

drivers/serial/s3c2410.c: remove dead config symbols

Remove dead config symbol.

Signed-off-by: Jiri Olsa <olsajiri@gmail.com>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Ben Dooks <ben@fluff.org>
Cc: Alan Cox <alan@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

serial: keep the DTR setting for serial console.

with reverting "x86, serial: convert legacy COM ports to platform devices",
we will have the serial console before the port is probled again.

uart_add_one_port==>uart_configure_port==>set_mcttrl(port, 0) will clear
the DTR setting by uart_set_options().  then I will lose my output from
serial console again.

So try to keep DTR in uart_configure_port()

Signed-off-by: Yinghai Lu <yinghai.lu@sun.com>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: Andi Kleen <ak@suse.de>
Cc: Bjorn Helgaas <bjorn.helgaas@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

drivers/pcmcia: add missing pci_dev_get

pci_get_slot does a pci_dev_get, so pci_dev_put needs to be called in an
error case.

An extract of the semantic match used to find the problem is as follows:
(http://www.emn.fr/x-info/coccinelle/)

// <smpl>
@@
type find1.T,T1,T2;
identifier find1.E;
statement find1.S;
expression x1,x2,x3;
expression find1.test;
int ret != 0;
@@

  T E;
  ...
(
* E = pci_get_slot(...);
  if (E == NULL) S
|
* if ((E = pci_get_slot(...)) == NULL)
  S
)
  ... when != pci_dev_put(...,(T1)E,...)
      when != if (E != NULL) { ... pci_dev_put(...,(T1)E,...); ...}
      when != x1 = (T1)E
      when != E = x3;
      when any
  if (test) {
    ... when != pci_dev_put(...,(T2)E,...)
        when != if (E != NULL) { ... pci_dev_put(...,(T2)E,...); ...}
        when != x2 = (T2)E
(
*   return;
|
*   return ret;
)
  }
// </smpl>

Signed-off-by: Julia Lawall <julia@diku.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

drivers/pcmcia: Add missing iounmap

of_iomap calls ioremap, and so should be matched with an iounmap.  At the
two error returns, the result of calling of_iomap is only stored in a local
variable, so these error paths need to call iounmap.  Furthermore, this
function ultimately stores the result of of_iomap in an array that is local
to the file.  These values should be iounmapped at some point.  I have
added a corresponding call to iounmap at the end of the function
m8xx_remove.

The problem was found using the following semantic match.
(http://www.emn.fr/x-info/coccinelle/)

// <smpl>
@@
type T,T1,T2;
identifier E;
statement S;
expression x1,x2,x3;
int ret;
@@

  T E;
  ...
* E = of_iomap(...);
  if (E == NULL) S
  ... when != iounmap(...,(T1)E,...)
      when != if (E != NULL) { ... iounmap(...,(T1)E,...); ...}
      when != x1 = (T1)E
      when != E = x3;
      when any
  if (...) {
    ... when != iounmap(...,(T2)E,...)
        when != if (E != NULL) { ... iounmap(...,(T2)E,...); ...}
        when != x2 = (T2)E
(
*   return;
|
*   return ret;
)
  }
// </smpl>

Signed-off-by: Julia Lawall <julia@diku.dk>
Vitaly Bordug <vitb@kernel.crashing.org>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Olof Johansson <olof@lixom.net>
Cc: Dominik Brodowski <linux@dominikbrodowski.net>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Kumar Gala <galak@kernel.crashing.org>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

at91_cf: use generic gpio calls

Update the AT91 CF driver to use the generic GPIO calls instead of the
AT91-specific ones; and request exclusive use of those signals.

Minor tweaks to cleanup code paths: always in reverse order of how the
resources were allocated, with remove() matching the fault paths of
probe().

Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

pcmcia/pcnet_cs: fix 'shadow variable' warning

Fixing:
  CHECK   drivers/net/pcmcia/pcnet_cs.c
drivers/net/pcmcia/pcnet_cs.c:523:15: warning: symbol 'hw_info' shadows an earlier one
drivers/net/pcmcia/pcnet_cs.c:148:18: originally declared here

Signed-off-by: Richard Knutsson <ricknu-0@student.ltu.se>
Acked-by: Alan Cox <alan@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

pcmcia/fmvj18x_cs: fix 'shadow variable' warning

Fixing:
  CHECK   drivers/net/pcmcia/fmvj18x_cs.c
drivers/net/pcmcia/fmvj18x_cs.c:1205:6: warning: symbol 'i' shadows an earlier one
drivers/net/pcmcia/fmvj18x_cs.c:1179:9: originally declared here

Signed-off-by: Richard Knutsson <ricknu-0@student.ltu.se>
Acked-by: Alan Cox <alan@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

pcmcia/axnet_cs: make use of 'max()' instead of handcrafted one

Use 'max(x,y)' instead of 'x < y ? y : x'.

Signed-off-by: Richard Knutsson <ricknu-0@student.ltu.se>
Acked-by: Alan Cox <alan@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

pcmcia/axnet_cs: make functions static

Fixing:
  CHECK   drivers/net/pcmcia/axnet_cs.c
drivers/net/pcmcia/axnet_cs.c:994:5: warning: symbol 'ax_close' was not declared. Should it be static?
drivers/net/pcmcia/axnet_cs.c:1017:6: warning: symbol 'ei_tx_timeout' was not declared. Should it be static?

Signed-off-by: Richard Knutsson <ricknu-0@student.ltu.se>
Acked-by: Alan Cox <alan@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

pcmcia/3c574_cs: fix 'shadow variable' warning

Fixing:
  CHECK   drivers/net/pcmcia/3c574_cs.c
drivers/net/pcmcia/3c574_cs.c:695:7: warning: symbol 'i' shadows an earlier one
drivers/net/pcmcia/3c574_cs.c:636:6: originally declared here

Signed-off-by: Richard Knutson <ricknu-0@student.ltu.se>
Acked-by: Alan Cox <alan@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

pcmcia: include bad CIS filename in error message

- Print the invalid CIS filename in the invalid filename message.
- Use sizeof() instead of hard-coded constant for buffer size.

Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

pcmcia: stop updating dev->power.power_state

This stops the pcmcia core from using dev->power.power_state; that field is
deprecated (overdue for removal) and the only reason to update it was to make
the /sys/devices/.../power/state files (now removed) work better.

Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Cc: Dominik Brodowski <linux@dominikbrodowski.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

pcmcia: replace kio_addr_t with unsigned int everywhere

Remove kio_addr_t, and replace it with unsigned int.  No known architecture
needs more than 32 bits for IO addresses and ports and having a separate type
for it is just messy.

Signed-off-by: Olof Johansson <olof@lixom.net>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Matthew Wilcox <matthew@wil.cx>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: Dominik Brodowski <linux@dominikbrodowski.net>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

pcmcia: convert some internal-only ioaddr_t to unsigned int

Convert the io_req_t members to unsigned int, to allow use on machines with
more than 16 bits worth of IO ports (i.e.  secondary busses on ppc64, etc).

There was only a couple of places in drivers where a change was needed.  I
left printk formats alone (there are lots of %04x-style formats in there),
mostly to not change the format on the platforms that only have 16-bit io
addresses, but also because the padding doesn't really add all that much value
most of the time.

I found only one sprintf of an address, and upsized the string accordingly (I
doubt anyone will have anywhere near INT_MAX as irq value, but at least
there's room for it now).

Signed-off-by: Olof Johansson <olof@lixom.net>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Matthew Wilcox <matthew@wil.cx>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: Dominik Brodowski <linux@dominikbrodowski.net>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

m32r: remove dead config symbols from M32R code

remove dead config symbols from M32R code

Signed-off-by: Jiri Olsa <olsajiri@gmail.com>
Cc: Hirokazu Takata <takata@linux-m32r.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

include/asm-powerpc/nvram.h needs list.h

CC [M]  sound/ppc/awacs.o
In file included from sound/ppc/awacs.c:24:
include/asm/nvram.h:62: error: field 'partition' has incomplete type

Reported-by: Mariusz Kozlowski <m.kozlowski@tuxland.pl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

sdio: fix module device table definition for m68k

FATAL: drivers/bluetooth/btsdio: sizeof(struct sdio_device_id)=12 is not a modulo of the size of section __mod_sdio_device_table=30.
Fix definition of struct sdio_device_id in mod_devicetable.h

m68k has 16bit alignment for unsigned long.

Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Pierre Ossman <drzeus@drzeus.cx>
CC: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

timerfd: un-break CONFIG_TIMERFD

Remove the broken status to CONFIG_TIMERFD.

Signed-off-by: Davide Libenzi <davidel@xmailserver.org>
Cc: Michael Kerrisk <mtk-manpages@gmx.net>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

timerfd: wire the new timerfd API to the x86 family

Wires up the new timerfd API to the x86 family.

Signed-off-by: Davide Libenzi <davidel@xmailserver.org>
Cc: Michael Kerrisk <mtk-manpages@gmx.net>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: <linux-arch@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

timerfd: new timerfd API

This is the new timerfd API as it is implemented by the following patch:

int timerfd_create(int clockid, int flags);
int timerfd_settime(int ufd, int flags,
    const struct itimerspec *utmr,
    struct itimerspec *otmr);
int timerfd_gettime(int ufd, struct itimerspec *otmr);

The timerfd_create() API creates an un-programmed timerfd fd.  The "clockid"
parameter can be either CLOCK_MONOTONIC or CLOCK_REALTIME.

The timerfd_settime() API give new settings by the timerfd fd, by optionally
retrieving the previous expiration time (in case the "otmr" parameter is not
NULL).

The time value specified in "utmr" is absolute, if the TFD_TIMER_ABSTIME bit
is set in the "flags" parameter.  Otherwise it's a relative time.

The timerfd_gettime() API returns the next expiration time of the timer, or
{0, 0} if the timerfd has not been set yet.

Like the previous timerfd API implementation, read(2) and poll(2) are
supported (with the same interface).  Here's a simple test program I used to
exercise the new timerfd APIs:

http://www.xmailserver.org/timerfd-test2.c

[akpm@linux-foundation.org: coding-style cleanups]
[akpm@linux-foundation.org: fix ia64 build]
[akpm@linux-foundation.org: fix m68k build]
[akpm@linux-foundation.org: fix mips build]
[akpm@linux-foundation.org: fix alpha, arm, blackfin, cris, m68k, s390, sparc and sparc64 builds]
[heiko.carstens@de.ibm.com: fix s390]
[akpm@linux-foundation.org: fix powerpc build]
[akpm@linux-foundation.org: fix sparc64 more]
Signed-off-by: Davide Libenzi <davidel@xmailserver.org>
Cc: Michael Kerrisk <mtk-manpages@gmx.net>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Davide Libenzi <davidel@xmailserver.org>
Cc: Michael Kerrisk <mtk-manpages@gmx.net>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Cc: Davide Libenzi <davidel@xmailserver.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

timerfd: introduce a new hrtimer_forward_now() function

I think that advancing the timer against the timer's current "now" can be a
pretty common usage, so, w/out exposing hrtimer's internals, we add a new
hrtimer_forward_now() function.

Signed-off-by: Davide Libenzi <davidel@xmailserver.org>
Cc: Michael Kerrisk <mtk-manpages@gmx.net>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

exec: rework the group exit and fix the race with kill

As Roland pointed out, we have the very old problem with exec.  de_thread()
sets SIGNAL_GROUP_EXIT, kills other threads, changes ->group_leader and then
clears signal->flags.  All signals (even fatal ones) sent in this window
(which is not too small) will be lost.

With this patch exec doesn't abuse SIGNAL_GROUP_EXIT.  signal_group_exit(),
the new helper, should be used to detect exit_group() or exec() in progress.
It can have more users, but this patch does only strictly necessary changes.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Davide Libenzi <davidel@xmailserver.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Robin Holt <holt@sgi.com>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

remove handle_group_stop() in favor of do_signal_stop()

Every time we set SIGNAL_GROUP_EXIT or clear SIGNAL_STOP_DEQUEUED we also
reset ->group_stop_count.

This means that the SIGNAL_GROUP_EXIT check in handle_group_stop() is not
needed, and do_signal_stop() should check SIGNAL_STOP_DEQUEUED only when
->group_stop_count == 0. With these changes handle_group_stop() becomes the
subset of do_signal_stop(), we can kill it and use do_signal_stop() instead.

Also, a preparation for the next patch.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Davide Libenzi <davidel@xmailserver.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Robin Holt <holt@sgi.com>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

__group_complete_signal(): fix coredump with group stop race

When __group_complete_signal() sees sig_kernel_coredump() signal, it starts
the group stop, but sets ->group_exit_task = t in a hope that "t" will
actually dequeue this signal and invoke do_coredump().  However, by the
time "t" enters get_signal_to_deliver() it is possible that the signal was
blocked/ignored or we have another pending !SIG_KERNEL_COREDUMP_MASK signal
which will be dequeued first.  This means the task could be stopped but not
killed.

Remove this code from __group_complete_signal().  Note also this patch
removes the bogus signal_wake_up(t, 1).  This thread can't be
STOPPED/TRACED, note the corresponding check in wants_signal().

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Davide Libenzi <davidel@xmailserver.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Robin Holt <holt@sgi.com>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

clone: prepare to recycle CLONE_STOPPED

Ulrich says that we never used this clone flags and that nothing should be
using it.

As we're down to only a single bit left in clone's flags argument, let's add a
warning to check that no userspace is actually using it.  Hopefully we will
be able to recycle it.

Roland said:

  CLONE_STOPPED was previously used by some NTPL versions when under
  thread_db (i.e.  only when being actively debugged by gdb), but not for a
  long time now, and it never worked reliably when it was used.  Removing it
  seems fine to me.

[akpm@linux-foundation.org: it looks like CLONE_DETACHED is being used]
Cc: Ulrich Drepper <drepper@redhat.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

get_task_comm(): return the result

It was dumb to make get_task_comm() return void.  Change it to return a
pointer to the resulting output for caller convenience.

Cc: Ulrich Drepper <drepper@redhat.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

lockdep: annotate epoll

On Sat, 2008-01-05 at 13:35 -0800, Davide Libenzi wrote:

> I remember I talked with Arjan about this time ago. Basically, since 1)
> you can drop an epoll fd inside another epoll fd 2) callback-based wakeups
> are used, you can see a wake_up() from inside another wake_up(), but they
> will never refer to the same lock instance.
> Think about:
>
>  dfd = socket(...);
>  efd1 = epoll_create();
>  efd2 = epoll_create();
>  epoll_ctl(efd1, EPOLL_CTL_ADD, dfd, ...);
>  epoll_ctl(efd2, EPOLL_CTL_ADD, efd1, ...);
>
> When a packet arrives to the device underneath "dfd", the net code will
> issue a wake_up() on its poll wake list. Epoll (efd1) has installed a
> callback wakeup entry on that queue, and the wake_up() performed by the
> "dfd" net code will end up in ep_poll_callback(). At this point epoll
> (efd1) notices that it may have some event ready, so it needs to wake up
> the waiters on its poll wait list (efd2). So it calls ep_poll_safewake()
> that ends up in another wake_up(), after having checked about the
> recursion constraints. That are, no more than EP_MAX_POLLWAKE_NESTS, to
> avoid stack blasting. Never hit the same queue, to avoid loops like:
>
>  epoll_ctl(efd2, EPOLL_CTL_ADD, efd1, ...);
>  epoll_ctl(efd3, EPOLL_CTL_ADD, efd2, ...);
>  epoll_ctl(efd4, EPOLL_CTL_ADD, efd3, ...);
>  epoll_ctl(efd1, EPOLL_CTL_ADD, efd4, ...);
>
> The code "if (tncur->wq == wq || ..." prevents re-entering the same
> queue/lock.

Since the epoll code is very careful to not nest same instance locks
allow the recursion.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Tested-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
Acked-by: Davide Libenzi <davidel@xmailserver.org>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

drivers/net/wireless/b43/main.c needs io.h

m68k:

drivers/net/wireless/b43/main.c:251: error: implicit declaration of function 'mmiowb'

Cc: "John W. Linville" <linville@tuxdriver.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

sys_remap_file_pages: fix ->vm_file accounting

Fix ->vm_file accounting, mmap_region() may do do_munmap().

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kvm: i386 fix

arch/x86/kvm/x86.c: In function 'emulator_cmpxchg_emulated':
arch/x86/kvm/x86.c:1746: warning: passing argument 2 of 'vcpu->arch.mmu.gva_to_gpa' makes integer from pointer without a cast
arch/x86/kvm/x86.c:1746: warning: 'addr' is used uninitialized in this function

Is true.  Local variable `addr' shadows incoming arg `addr'.  Avi is on
vacation for a while, so...

Cc: Avi Kivity <avi@qumranet.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Merge branch 'for-linus' of /home/rmk/linux-2.6-arm

* 'for-linus' of master.kernel.org:/home/rmk/linux-2.6-arm: (44 commits)
  [ARM] 4822/1: RealView: Change the REALVIEW_MPCORE configuration option
  [ARM] 4821/1: RealView: Remove the platform dependencies from localtimer.c
  [ARM] 4820/1: RealView: Select the timer IRQ at run-time
  [ARM] 4819/1: RealView: Fix entry-macro.S to work with multiple platforms
  [ARM] 4818/1: RealView: Add core-tile detection
  [ARM] 4817/1: RealView: Move the AMBA resource definitions to realview_eb.c
  [ARM] 4816/1: RealView: Move the platform-specific definitions into board-eb.h
  [ARM] 4815/1: RealView: Add clockevents suport for the local timers
  [ARM] 4814/1: RealView: Add broadcasting clockevents support for ARM11MPCore
  [ARM] 4813/1: Add SMP helper functions for clockevents support
  [ARM] 4812/1: RealView: clockevents support for the RealView platforms
  [ARM] 4811/1: RealView: clocksource support for the RealView platforms
  [ARM] 4736/1: Export atags to userspace and allow kexec to use customised atags
  [ARM] 4798/1: pcm027: fix missing header file
  [ARM] 4803/1: pxa: fix building issue of poodle.c caused by patch 4737/1
  [ARM] 4801/1: pxa: fix building issues of missing pxa2xx-regs.h
  [ARM] pxa: introduce sysdev for pxa3xx static memory controller
  [ARM] pxa: add preliminary suspend/resume code for pxa3xx
  [ARM] pxa: introduce sysdev for GPIO register saving/restoring
  [ARM] pxa: introduce sysdev for IRQ register saving/restoring
  ...

Merge branch 'slub-linus' of git://git./linux/kernel/git/christoph/vm

* 'slub-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/christoph/vm:
  Explain kmem_cache_cpu fields
  SLUB: Do not upset lockdep
  SLUB: Fix coding style violations
  Add parameter to add_partial to avoid having two functions
  SLUB: rename defrag to remote_node_defrag_ratio
  Move count_partial before kmem_cache_shrink
  SLUB: Fix sysfs refcounting
  slub: fix shadowed variable sparse warnings

Explain kmem_cache_cpu fields

Add some comments explaining the fields of the kmem_cache_cpu structure.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

SLUB: Do not upset lockdep

inconsistent {softirq-on-W} -> {in-softirq-W} usage.
swapper/0 [HC0[0]:SC1[1]:HE0:SE0] takes:
 (&n->list_lock){-+..}, at: [<ffffffff802935c1>] add_partial+0x31/0xa0
{softirq-on-W} state was registered at:
  [<ffffffff80259fb8>] __lock_acquire+0x3e8/0x1140
  [<ffffffff80259838>] debug_check_no_locks_freed+0x188/0x1a0
  [<ffffffff8025ad65>] lock_acquire+0x55/0x70
  [<ffffffff802935c1>] add_partial+0x31/0xa0
  [<ffffffff805c76de>] _spin_lock+0x1e/0x30
  [<ffffffff802935c1>] add_partial+0x31/0xa0
  [<ffffffff80296f9c>] kmem_cache_open+0x1cc/0x330
  [<ffffffff805c7984>] _spin_unlock_irq+0x24/0x30
  [<ffffffff802974f4>] create_kmalloc_cache+0x64/0xf0
  [<ffffffff80295640>] init_alloc_cpu_cpu+0x70/0x90
  [<ffffffff8080ada5>] kmem_cache_init+0x65/0x1d0
  [<ffffffff807f1b4e>] start_kernel+0x23e/0x350
  [<ffffffff807f112d>] _sinittext+0x12d/0x140
  [<ffffffffffffffff>] 0xffffffffffffffff

This change isn't really necessary for correctness, but it prevents lockdep
from getting upset and then disabling itself.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: Kamalesh Babulal <kamalesh@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Christoph Lameter <clameter@sgi.com>

SLUB: Fix coding style violations

This fixes most of the obvious coding style violations in mm/slub.c as
reported by checkpatch.

Acked-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Christoph Lameter <clameter@sgi.com>

Add parameter to add_partial to avoid having two functions

Add a parameter to add_partial instead of having separate functions.  The
parameter allows a more detailed control of where the slab pages is placed in
the partial queues.

If we put slabs back to the front then they are likely immediately used for
allocations.  If they are put at the end then we can maximize the time that
the partial slabs spent without being subject to allocations.

When deactivating slab we can put the slabs that had remote objects freed (we
can see that because objects were put on the freelist that requires locks) to
them at the end of the list so that the cachelines of remote processors can
cool down.  Slabs that had objects from the local cpu freed to them (objects
exist in the lockless freelist) are put in the front of the list to be reused
ASAP in order to exploit the cache hot state of the local cpu.

Patch seems to slightly improve tbench speed (1-2%).

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Reviewed-by: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

SLUB: rename defrag to remote_node_defrag_ratio

The NUMA defrag works by allocating objects from partial slabs on remote
nodes.  Rename it to

remote_node_defrag_ratio

to be clear about this.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

Move count_partial before kmem_cache_shrink

Move the counting function for objects in partial slabs so that it is placed
before kmem_cache_shrink.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>