[PATCH] Fix keyctl usage of strnlen_user()

In the small window between strnlen_user() and copy_from_user() userspace
could alter the terminating `\0' character.

Signed-off-by: Davi Arnaut <[email protected]>
Cc: David Howells <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] edac_mc: Remove include of version.h

By including version.h edac_mc was rebuilding on every incremental build.
Which defeats the point of incremental builds.

Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Alan Cox <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] fat: Fix truncate() write ordering

The truncate() should write the file size before writing the new EOF entry.
This patch fixes it.

This bug was pointed out by Machida Hiroyuki.

Signed-off-by: OGAWA Hirofumi <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] Trivial optimization of ll_rw_block()

The ll_rw_block() needs to get ref-count only if it submits a buffer().  This
patch avoids the needless get/put of ref-count.

Signed-off-by: OGAWA Hirofumi <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] fat: Replace an own implementation with ll_rw_block(SWRITE,)

This patch replaces an own implementation with LL_RW_BLOCK(SWRITE,) which was
newly added.

Signed-off-by: OGAWA Hirofumi <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] uninline __sigqueue_free()

Five callsites.  I dunno how all this crap got back in there :(

Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] new tty buffering locking fix

Change locking in the new tty buffering facility from using tty->read_lock,
which is currently ignored by drivers and thus ineffective.  New locking
uses a new tty buffering specific lock enforced centrally in the tty
buffering code.

Two drivers (esp and cyclades) are updated to use the tty buffering
functions instead of accessing tty buffering internals directly.  This is
required for the new locking to work.

Minor checks for NULL buffers added to
tty_prepare_flip_string/tty_prepare_flip_string_flags

Signed-off-by: Paul Fulghum <[email protected]>
Cc: Alan Cox <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] ipmi: mem_{in,out}[bwl] => intf_mem_{in,out}[bwl]

On mips:

drivers/char/ipmi/ipmi_si_intf.c:1274: error: conflicting types for 'mem_inb'
include/asm/io.h:436: error: previous definition of 'mem_inb' was here

Don't look at line 436 unless you really know what you're doing.

Move those static functions out of more or less generic namespace.

Signed-off-by: Alexey "## should be banned" Dobriyan <[email protected]>
Acked-by: Corey Minyard <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] Compilation of kexec/kdump broken

The compilation of kexec/kdump seems to be broken for x86_64.  Remove the
dependency of kexec on CONFIG_IA32_EMULATION.

Signed-off-by: Fernando Vazquez <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] Fix comment to synchronize_sched()

Fix to broken comment to synchronize_rcu() noted by Keith Owens.  Also add
sentence noting that synchronize_sched() and synchronize_rcu() are not
necessarily identical.

Signed-off-by: Paul E. McKenney <[email protected]>
Cc: Keith Owens <[email protected]>
Cc: Stephen Hemminger <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] quota: fix error code for ext2_new_inode()

The quota check in ext2_new_inode() returns ENOSPC where it should return
EDQUOT instead.

Signed-off-by: Herbert Pötzl <[email protected]>
Acked-by: Jan Kara <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] Export cpu topology in sysfs

The patch implements cpu topology exportation by sysfs.

Items (attributes) are similar to /proc/cpuinfo.

1) /sys/devices/system/cpu/cpuX/topology/physical_package_id:
represent the physical package id of  cpu X;
2) /sys/devices/system/cpu/cpuX/topology/core_id:
represent the cpu core id to cpu X;
3) /sys/devices/system/cpu/cpuX/topology/thread_siblings:
represent the thread siblings to cpu X in the same core;
4) /sys/devices/system/cpu/cpuX/topology/core_siblings:
represent the thread siblings to cpu X in the same physical package;

To implement it in an architecture-neutral way, a new source file,
driver/base/topology.c, is to export the 5 attributes.

If one architecture wants to support this feature, it just needs to
implement 4 defines, typically in file include/asm-XXX/topology.h.
The 4 defines are:
#define topology_physical_package_id(cpu)
#define topology_core_id(cpu)
#define topology_thread_siblings(cpu)
#define topology_core_siblings(cpu)

The type of **_id is int.
The type of siblings is cpumask_t.

To be consistent on all architectures, the 4 attributes should have
deafult values if their values are unavailable. Below is the rule.

1) physical_package_id: If cpu has no physical package id, -1 is the
default value.

2) core_id: If cpu doesn't support multi-core, its core id is 0.

3) thread_siblings: Just include itself, if the cpu doesn't support
HT/multi-thread.

4) core_siblings: Just include itself, if the cpu doesn't support
multi-core and HT/Multi-thread.

So be careful when declaring the 4 defines in include/asm-XXX/topology.h.

If an attribute isn't defined on an architecture, it won't be exported.

Thank Nathan, Greg, Andi, Paul and Venki.

The patch provides defines for i386/x86_64/ia64.

Signed-off-by: Zhang, Yanmin <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Nick Piggin <[email protected]>
Cc: Greg KH <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] ixj: fix writing silence check

j->write_buffer_rp is a pointer.

Signed-off-by: Alexey Dobriyan <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] lib: Fix bug in int_sqrt() for 64 bit longs

The implementation of int_sqrt() assumes that longs have 32 bits.  On
systems that have 64 bit longs this will result in gross errors when the
argument to the function is greater than 2^32 - 1 on such systems.  I doubt
whether any such use is currently made of int_sqrt() but the attached patch
fixes the problem anyway.

Signed-off-by: Peter Williams <[email protected]>
Cc: Dave Jones <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] quota: remove unused sync_dquots_dev()

The already removed sync_dquots_dev(dev,type) is still defined in the
no-quota case.

Signed-off-by: Herbert Pötzl <[email protected]>
Acked-by: Jan Kara <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] drivers/serial/jsm/: cleanups

- jsm_driver.c: remove the now unused jsm_rawreadok module_param
- jsm_tty.c: remove a now unused variable

Is there any problem with removing the now useless jsm_rawreadok
module_param?

Signed-off-by: Adrian Bunk <[email protected]>
Cc: V. Ananda Krishnan <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] jsm: update for tty buffering revamp

Signed-off-by: V. Ananda Krishnan <[email protected]>
Cc: Alan Cox <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] fcntl F_SETFL and read-only IS_APPEND files

There is code in setfl() which attempts to preserve the O_APPEND flag on
IS_APPEND files...  however IS_APPEND files could also be opened O_RDONLY
and in that case setfl() should not require O_APPEND...

coreutils 5.93 tail -f attempts to set O_NONBLOCK even on regular files...
unfortunately if you try this on an append-only log file the result is
this:

fcntl64(3, F_GETFL)                     = 0x8000 (flags O_RDONLY|O_LARGEFILE)
fcntl64(3, F_SETFL, O_RDONLY|O_NONBLOCK|O_LARGEFILE) = -1 EPERM (Operation not permitted)

I offer up the patch below as one way of fixing the problem...  i've tested
it fixes the problem with tail -f but haven't really tested beyond that.

(I also reported the coreutils bug upstream... it shouldn't fail imho...
<https://savannah.gnu.org/bugs/index.php?func=detailitem&item_id=15473>)

Signed-off-by: dean gaudet <[email protected]>
Cc: Al Viro <[email protected]>
Acked-by: Christoph Hellwig <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] I2O: fix and workaround for Motorola/Freescale controller

- This controller violates the I2O spec for the I/O registers.  The patch
  contains a workaround which moves the registers to the proper location.
  (originally author: Matthew Starzewski)

- If a message frame is beyond the mapped address range a error is
  returned.

Signed-off-by: Markus Lidel <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] I2O: don't disable PCI device if it is enabled before probing

If PCI device is enabled before probing, it will not be disabled at exit.

Signed-off-by: Markus Lidel <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] fix O_DIRECT read of last block in a sparse file

Currently, if you open a file O_DIRECT, truncate it to a size that is not a
multiple of the disk block size, and then try to read the last block in the
file, the read will return 0.  The problem is in do_direct_IO, here:

        /* Handle holes */
        if (!buffer_mapped(map_bh)) {
                char *kaddr;

...

                if (dio->block_in_file >=
                        i_size_read(dio->inode)>>blkbits) {
                        /* We hit eof */
                        page_cache_release(page);
                        goto out;
                }

We shift off any remaining bytes in the final block of the I/O, resulting
in a 0-sized read.  I've attached a patch that fixes this.  I'm not happy
about how ugly the math is getting, so suggestions are more than welcome.

I've tested this with a simple program that performs the steps outlined for
reproducing the problem above.  Without the patch, we get a 0-sized result
from read.  With the patch, we get the correct return value from the short
read.

Signed-off-by: Jeff Moyer <[email protected]>
Cc: Badari Pulavarty <[email protected]>
Cc: Suparna Bhattacharya <[email protected]>
Cc: Mingming Cao <[email protected]>
Cc: Joel Becker <[email protected]>
Cc: "Chen, Kenneth W" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] ext2: print xip mount option in ext2_show_options

In case we have CONFIG_FS_XIP, ext2_show_options shows "xip" if
EXT2_MOUNT_XIP mount flag is set.

Signed-off-by: Carsten Otte <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] disable per cpu intr in /proc/stat

Don't compute and display the per-irq sums on ia64 either, too much
overhead for mostly useless figures.

Cc: Olaf Hering <[email protected]>
Acked-by: "Luck, Tony" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] edac: use C99 initializers (sparse warnings)

drivers/edac/e752x_edac.c:1042:7: warning: obsolete struct initializer, use C99 syntax

Signed-off-by: Randy Dunlap <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] cpuset: fix sparse warning

kernel/cpuset.c:644:38: warning: non-ANSI function declaration of function 'cpuset_update_task_memory_state'

Signed-off-by: Randy Dunlap <[email protected]>
Acked-by: Paul Jackson <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] parport: remove dead address in MAINTAINERS

Remove dead address for David Campbell in MAINTAINERS.

Signed-off-by: Arnaud Giersch <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] parport: fix documentation

Fix documentation to actually match the code.

Signed-off-by: Arnaud Giersch <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] Normalize timespec for negative values in ns_to_timespec

- In case of a negative nsec value the result of the division must be
  normalized.

- Remove inline from an exported function.

Signed-off-by: George Anzinger <[email protected]>
Signed-off-by: Thomas Gleixner <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] v9fs: fix corner cases when flushing request

When v9fs_mux_rpc sends a 9P message, it may be put in the queue of unsent
request.  If the user process receives a signal, v9fs_mux_rpc sets the
request error to ERREQFLUSH and assigns NULL to request's send message.  If
the message was still in the unsent queue, v9fs_write_work would produce an
oops while processing it.

The patch makes sure that requests that are being flushed are moved to the
pending requests queue safely.

If a request is being flushed, don't remove it from the list of pending
requests even if it receives a reply before the flush is acknoledged.  The
request will be removed during from the Rflush handler (v9fs_mux_flush_cb).

Signed-off-by: Latchesar Ionkov <[email protected]>
Cc: Eric Van Hensbergen <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] v9fs: v9fs_put_str fix

v9fs_put_str used to store pointer to the source string, instead of the
cbuf copy.  This patch corrects it.

Signed-off-by: Latchesar Ionkov <[email protected]>
Cc: Eric Van Hensbergen <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] v9fs: symlink support fixes

Two symlink fixes, v9fs_readlink didn't copy the last character of the
symlink name, v9fs_vfs_follow_link incorrectly called strlen of newly
allocated buffer instead of PATH_MAX.

Signed-off-by: Latchesar Ionkov <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] parport: add parallel port support for SGI O2

Add support for the built-in parallel port on SGI O2 (a.k.a.  IP32).
Define a new configuration option: PARPORT_IP32.  The module is named
parport_ip32.

Hardware support for SPP, EPP and ECP modes along with DMA support when
available are currently implemented.

Signed-off-by: Arnaud Giersch <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] fs/jffs/intrep.c: 255 is unsigned char

Signed-off-by: Felix Oxley <[email protected]>
Signed-off-by: Alexey Dobriyan <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] drivers/isdn/sc/ioctl.c: copy_from_user() size fix

A few lines above the patch we have:

char *srec;
srec = kmalloc(SCIOC_SRECSIZE, GFP_KERNEL);

sizeof pointer is probably not meant here.

Signed-off-by: Domen Puncer <[email protected]>
Signed-off-by: Alexey Dobriyan <[email protected]>
Cc: Karsten Keil <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] umem: check pci_set_dma_mask return value correctly

Signed-off-by: Tobias Klauser <[email protected]>
Signed-off-by: Alexey Dobriyan <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] Fix two ext[23] uninitialized warnings

There is a code path that passed size to ext2_xattr_set
(ext3_xattr_set_handle) before initializing it.  The callees don't use the
value in that case, but gcc cannot tell.  Always initialize size to get rid
of the warnings.

Signed-off-by: Andreas Gruenbacher <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] ufs: fix char vs. __s8 clash in ufs

Fix this warning:

fs/ufs/super.c: In function â\80\98ufs_fill_superâ\80\99:
fs/ufs/super.c:858: warning: case label value exceeds maximum value for type

which happens because __s8 != char.  These macros are used for struct
ufs_super_block.fs_clean which is declared as __s8.

Signed-off-by: Andreas Schwab <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] EDAC config cleanup

The AMD76x chipsets aren't used in 64-bit, so don't offer the driver to the
user.

Signed-off-by: Dave Jones <[email protected]>
Acked-by: Alan Cox <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] OProfile: fixed x86_64 incorrect kernel call graphs

Fix the problem in kernel 2.6.15.1 (and early versions) that OProfile on
x86_64 does not correctly collect the stack traces for kernel functions.

The original code in valid_kernel_stack() in arch/i386/oprofile/backtrace.c
assumes that the frame pointer (headaddr) should be greater than stack
(i.e., regs).

This assumption is wrong for x86_64 because NMIs in x86_64 use a seperate
stack different from the kernel stack.  Therefore, the variable stack now
points to some location on the NMI stack, which turns out to be at a higher
address than the frame pointer (headaddr) on the kernel stack.  The correct
comparison here should be between headaddr and regs->rsp for x86_64.

Signed-off-by: Tong Li <[email protected]>
Cc: John Levon <[email protected]>
Cc: Philippe Elie <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] Fix some ucLinux breakage from the tty updates

Breakage reported by Adrian Bunk

Untested (no hardware)

Signed-off-by: Alan Cox <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] rio cleanups

INKERNEL is always defined
HOST is never defined
therefore RTA is also never defined

Strip the relevant garbage out of the headers on this basis.

Signed-off-by: Alan Cox <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] make "struct d_cookie" depend on CONFIG_PROFILING

Shrinks "struct dentry" from 128 bytes to 124 on x86, allowing 31 objects
per slab instead of 30.

Cc: John Levon <[email protected]>
Cc: Philippe Elie <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] ufs: fix hang during `rm'

This fixes the code like this:

bh = sb_find_get_block (sb, tmp + j);
if ((bh && DATA_BUFFER_USED(bh)) || tmp != fs32_to_cpu(sb, *p)) {
retry = 1;
brelse (bh);
goto next1;
}
bforget (bh);

sb_find_get_block() ordinarily returns a buffer_head with b_count>=2, and
this code assume that in case if "b_count>1" buffer is used, so this caused
infinite loop.

(akpm: that is-the-buffer-busy code is incomprehensible.  Good riddance.  Use
of block_truncate_page() seems sane).

Signed-off-by: Evgeniy Dushistov <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] ufs: fix oops with `ufs1' type

"rm" command, on file system with "ufs1" type cause system hang up.  This
is, in fact, not so bad as it seems to be, because of after that in "kernel
control path" there are 3-4 places which may cause "oops".

So the first patch fix oopses, and the second patch fix "kernel hang up".

"oops" appears because of reading of group's summary info partly wrong, and
access to not first group's summary info cause "oops".

Signed-off-by: Evgeniy Dushistov <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] sx.c printk warning fixes

drivers/char/sx.c: In function `sx_set_real_termios':
drivers/char/sx.c:934: warning: long unsigned int format, different type arg (arg 2)
drivers/char/sx.c:961: warning: long unsigned int format, different type arg (arg 2)
drivers/char/sx.c:976: warning: long unsigned int format, different type arg (arg 2)

Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] quota_v2: printk warning fixes

fs/quota_v2.c: In function `v2_check_quota_file':
fs/quota_v2.c:39: warning: int format, different type arg (arg 2)
fs/quota_v2.c:39: warning: int format, different type arg (arg 3)

Cc: Jan Kara <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] parport_serial: printk warning fix

drivers/parport/parport_serial.c: In function `parport_register':
drivers/parport/parport_serial.c:334: warning: unsigned int format, different type arg (arg 3)

Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] someone broke reiserfs V3 mount options, this fixes it

Signed-off-by: Hans Reiser <[email protected]>
Signed-off-by: Vitaly Fertman <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] don't allow users to set CONFIG_BROKEN=y

Do not allow people to create configurations with CONFIG_BROKEN=y.

The sole reason for CONFIG_BROKEN=y would be if you are working on fixing a
broken driver, but in this case editing the Kconfig file is trivial.

Never ever should a user enable CONFIG_BROKEN.

Signed-off-by: Adrian Bunk <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] parport: fix printk format warning

Fix printk format warning:
drivers/parport/probe.c:205: warning: format '%d' expects type 'int', but argument 3 has type 'size_t'

Signed-off-by: Randy Dunlap <[email protected]>
Signed-off-by: Arnaud Giersch <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] SBC EPX does not check/claim I/O ports it uses (2nd Edition)

Signed-off-by: Alan Cox <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] sx.c warning fixes

drivers/char/sx.c: In function `sx_set_real_termios':
drivers/char/sx.c:934: warning: int format, long unsigned int arg (arg 2)
drivers/char/sx.c:961: warning: unsigned int format, tcflag_t arg (arg 2)
drivers/char/sx.c:976: warning: unsigned int format, tcflag_t arg (arg 2)

Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] 3c59x: collision statistic fix

Count the total number of packets with collisions during transmission in
vp->stats.collisions.

Signed-off-by: Steffen Klassert <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] list.h: don't evaluate macro args multiple times

I noticed that list.h init functions were evaluating macro arguments
multiple times and thought it might be nice to protect the unsuspecting
caller.  Converting the macros to inline functions seems to reduce code
size, too.  A i386 defconfig build with gcc 3.3.3 from fc4:

   text    data     bss     dec     hex filename
3573148  565664  188828 4327640  4208d8 vmlinux.before
3572177  565664  188828 4326669  42050d vmlinux

add/remove: 0/0 grow/shrink: 11/144 up/down: 88/-1016 (-928)

There was no difference in checkstack output.

Signed-off-by: Zach Brown <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] include/asm-*/bitops.h: fix more "~0UL >> size" typos

"[PATCH] m68knommu: fix find_next_zero_bit in bitops.h" fixed a typo in
m68knommu implementation of find_next_zero_bit().

grep(1) shows that cris, frv, h8300, v850 are also affected.

Signed-off-by: Alexey Dobriyan <[email protected]>
Cc: Mikael Starvik <[email protected]>
Cc: David Howells <[email protected]>
Cc: Yoshinori Sato <[email protected]>
Cc: Miles Bader <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] Tell kallsyms_lookup_name() to ignore type U entries

When one module exports a function symbol and another module uses that
symbol then kallsyms shows the symbol twice.  Once from the consumer with a
type of 'U' and once from the provider with a type of 't' or 'T'.  On most
architectures, both entries have the same address so it does not matter
which one is returned by kallsyms_lookup_name().  But on architectures with
function descriptors, the 'U' entry points to the descriptor, not to the
code body, which is not what we want.

IA64 # grep -w qla2x00_remove_one /proc/kallsyms
a000000208c25ef8 U qla2x00_remove_one   [qla2300]   <= descriptor
a000000208bf44c0 t qla2x00_remove_one   [qla2xxx]   <= function body

Tell kallsyms_lookup_name() to ignore type U entries in modules.

Signed-off-by: Keith Owens <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] s390: fix to_channelpath macro

Fix broken to_channelpath macro (fortunately worked in all current cases...).

Signed-off-by: Cornelia Huck <[email protected]>
Signed-off-by: Heiko Carstens <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] s390: fix compat syscall wrapper

Patch 9ad11ab48b1ad618bf47076e9e579f267f5306c2 changes the type of the first
argument of some compat syscalls from int to unsigned int.  Add these changes
to the s390 compat wrapper as well.

Signed-off-by: Heiko Carstens <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] s390: compile fix: missing defines in asm-s390/io.h

Compile fix: add missing __raw_read* and __raw_write* defines to
include/asm-s390/io.h.
These are mandatory since patch c27a0d75b33c030965cc97d3d7f571107a673fb4
was merged.

Signed-off-by: Heiko Carstens <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] s390: timer interface visibility

Avoid visibility of kernel internal interface to user space.

Signed-off-by: Jan Glauber <[email protected]>
Signed-off-by: Heiko Carstens <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] s390: dasd extended error reporting module

The DASD extended error reporting is a facility that allows to get detailed
information about certain problems in the DASD I/O.  This information can be
used to implement fail-over applications that can recover these problems.

Signed-off-by: Stefan Weinhuber <[email protected]>
Signed-off-by: Martin Schwidefsky <[email protected]>
Signed-off-by: Heiko Carstens <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] ide: restore support for AEC6280M cards in aec62xx.c

This patch adds missing initialization sequence, necessary to get the
"Macintosh" version of AEC6280 cards to work in Linux.  Without this patch,
the driver hangs for several minutes trying to initialize the card and the
kernel is left in an unstable state.  This patch has been tested fine on
ppc and i386.

Signed-off-by: Thibaut VARENE <[email protected]>
Acked-by: Bartlomiej Zolnierkiewicz <[email protected]>
Cc: Alan Cox <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] Fix compilation errors in maps/dc21285.c

  CC      drivers/mtd/maps/dc21285.o
drivers/mtd/maps/dc21285.c: In function `dc21285_copy_to_32':
drivers/mtd/maps/dc21285.c:113: error: invalid lvalue in increment
drivers/mtd/maps/dc21285.c: In function `dc21285_copy_to_16':
drivers/mtd/maps/dc21285.c:124: error: invalid lvalue in increment
drivers/mtd/maps/dc21285.c: In function `dc21285_copy_to_8':
drivers/mtd/maps/dc21285.c:134: error: invalid lvalue in increment
make[3]: *** [drivers/mtd/maps/dc21285.o] Error 1

Signed-off-by: Martin Michlmayr <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] fix generic_fls64()

Noticed by Rune Torgersen.

Fix generic_fls64().  tcp_cubic is using fls64().

Signed-off-by: Akinobu Mita <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] Documentation: Updated PCI Error Recovery

This patch is a cleanup/restructuring/clarification of the PCI error
handling doc.  It should look rather professional at this point.

Signed-off-by: Linas Vepstas <[email protected]>
Cc: Greg KH <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] Fix build failure in recent pm_prepare_* changes.

kernel/power/power.h:49: error: static declaration of 'pm_prepare_console' follows non-static declaration
include/linux/suspend.h:46: error: previous declaration of 'pm_prepare_console' was here
kernel/power/power.h:50: error: static declaration of 'pm_restore_console' follows non-static declaration
include/linux/suspend.h:47: error: previous declaration of 'pm_restore_console' was here

Signed-off-by: Dave Jones <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] Kprobes: Fix deadlock in function-return probes

When two function-return probes are inserted on kfree()[1] and the second
on say, sys_link()[2], and later [2] is unregistered, we have a deadlock as
kfree is called with the kretprobe_lock held and the function-return probe
on kfree will also try to grab the same lock.

However, we can move the kfree() during unregistration to outside the
spinlock as we are sure that no instances from the free list will be used
after synchronized_sched() returns during the unregistration process.
Thanks to Masami Hiramatsu for spotting this.

Signed-off-by: Ananth N Mavinakayanahalli <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] kernel/kprobes.c: fix a warning #ifndef ARCH_SUPPORTS_KRETPROBES

kernel/kprobes.c:353: warning: 'pre_handler_kretprobe' defined but not used

Signed-off-by: Adrian Bunk <[email protected]>
Acked-by: Ananth N Mavinakayanahalli <[email protected]>
Acked-by: "Keshavamurthy, Anil S" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] md: Make sure rdev->size gets set for version-1 superblocks

Sometimes it doesn't so make the code more like the version-0 code which
works.

Signed-off-by: Neil Brown <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] md: Assorted little md fixes

- version-1 superblock
  + The default_bitmap_offset is in sectors, not bytes.
  + the 'size' field in the superblock is in sectors, not KB
- raid0_run should return a negative number on error, not '1'
- raid10_read_balance should not return a valid 'disk' number if
     ->rdev turned out to be NULL
- kmem_cache_destroy doesn't like being passed a NULL.

Signed-off-by: Neil Brown <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] md: Handle overflow of mdu_array_info_t->size better

mdu_array_info_t->size is 'int', which isn't big enough for the size (in KB of
each component in) some arrays.

So rather than a random overflow, set size to -1 when it cannot be set
correctly.

To update aspect on an array, userspace will sometimes:
  get_array_info
  change one field
  set_array_info

in this case, we don't want the '-1' in 'size' to change to size, or look like
a size change at all.  So test for that in update_array_info.

Signed-off-by: Neil Brown <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] Clean up Documentation/driver-model/overview.txt

Edits to the driver-model documentation for grammar, clarity and content.

These docs haven't been updated in years, and some of the technical content
and discussion has become stale; this patch updates these.  In addition,
some of the language is awkward.  Fix this.

(I'm trying to cleanup the other files in this directory also,
patches for these will come a bit later).

Signed-off-by: Linas Vepstas <[email protected]>
Acked-by: Patrick Mochel <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] Fix RocketPort driver

Call "ld->receive_buf" using the start of the character and flag buffers,
rather than the ends.

Signed-off-by: Michal Ostrowski <[email protected]>
Acked-by: Alan Cox <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

Linux v2.6.16-rc2

Merge branch 'master' of git://git./linux/kernel/git/torvalds/linux-2.6

[PATCH] device-mapper log bitset: fix big endian find_next_zero_bit

This is a fix to the device-mapper-log-bitset-fix-endian patch that
switched to ext2_* versions of the set and clear bit functions.  The
find_next_zero_bit function also has to be the ext2 one.  Otherwise the
mirror target tries to recover non-existent regions beyond the end of
device.

Signed-off-by: Stefan Bader <[email protected]>
Signed-off-by: Alasdair G Kergon <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] md: Add sysfs access to raid6 stripe cache size

.. just as we already have for raid5.

Signed-off-by: Neil Brown <[email protected]>
Cc: Greg KH <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] md: Don't remove bitmap from md array when switching to read-only

While a read-only array doesn't not really need a bitmap, we should
not remove the bitmap when switching an array to read-only because
 a/ There is no code to re-add the bitmap which switching to read-write,
 b/ There is insufficient locking - the bitmap could be accessed while it is
    being removed.

Cc: Reuben Farrelly <[email protected]>
Signed-off-by: Neil Brown <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] md: Make sure array geometry changes persist with version-1 superblocks

super_1_sync only updates fields in the superblock that might have changed.

'raid_disks' and 'size' could have changed, but this information doesn't get
updated....  until this patch.

Signed-off-by: Neil Brown <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] md: Fix device-size updates in md

As 'array_size' is a 'sector_t', it may overflow inappropriately when shifted
10 bits.  So We should cast it to a loff_t first.

There are two places with this problem, but the second (in update_raid_disks)
isn't needed so just remove it:
  The only personality that handles ->reshape currently is raid1,
  and it doesn't change the size of the array.
  When added for raid5/6, reshape again won't change the size of the array,
  at least not straight away.
  This code might be need for reshaping 'linear' but linear->shape,
  if implemented, should probably do the i_size_write itself.

Signed-off-by: Neil Brown <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

Merge branch 'drm-linus' of git://git./linux/kernel/git/airlied/drm-2.6

[PATCH] Fix sgiioc4 DMA timeout problem with 64KiB s/g elements.

Problem caused by the fact that the code used to only pick the low 16
bits of the bytecount.  That may be how some controllers act on it (byte
count of 0 means 0x10000), but not for this particular hardware.

Signed-off-by: Jeremy Higdon <[email protected]>
Acked-by: Jes Sorensen <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

sem2mutex: drivers/char/drm/

From: Arjan van de Ven <[email protected]>

Semaphore to mutex conversion.

The conversion was generated via scripts, and the result was validated
automatically via a script as well.

Signed-off-by: Arjan van de Ven <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Dave Airlie <[email protected]>

drm: drivers/char/drm/: make some functions static

From: Adrian Bunk <[email protected]>

This patch makes some needlessly global functions static.

Signed-off-by: Adrian Bunk <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Dave Airlie <[email protected]>

Merge git://git./linux/kernel/git/perex/alsa

Merge branch 'release' of git://git./linux/kernel/git/lenb/linux-acpi-2.6

[PATCH] compat: fix compat_sys_openat and friends

Most of the 64 bit architectures will zero extend the first argument to
compat_sys_{openat,newfstatat,futimesat} which will fail if the 32 bit
syscall was passed AT_FDCWD (which is a small negative number).  Declare
the first argument to be an unsigned int which will force the correct
sign extension when the internal functions are called in each case.

Also, do some small white space cleanups in fs/compat.c.

Signed-off-by: Stephen Rothwell <[email protected]>
Acked-by: David S. Miller <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

Merge /pub/scm/linux/kernel/git/sfrench/cifs-2.6

[PATCH] PowerPC/PCI Hotplug build break

Build break: Building PCI hotplug on PowerPC results in
a build break, due to failure to export symbols.

Reported today by Dave Jones <[email protected]>:
drivers/pci/hotplug/rpaphp.ko needs unknown symbol pcibios_add_pci_devices

This patch fixes same problem in drivers/pci tree
Previous patch fixes the break in the arch/powerpc tree.

Signed-off-by: Linas Vepstas <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] PowerPC/PCI Hotplug build break

Build break: Building PCI hotplug on PowerPC results in
a build break, due to failure to export symbols.

Reported today by Dave Jones <[email protected]>:
drivers/pci/hotplug/rpaphp.ko needs unknown symbol pcibios_add_pci_devices

This patch fixes the break in the arch/powerpc tree.
Next patch fixes same problem in drivers/pci tree

Signed-off-by: Linas Vepstas <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

Merge ... /linux/kernel/git/torvalds/linux-2.6.git

Signed-off-by: Steve French <[email protected]>

SUNRPC: Remove obsolete rpcauth #defines

 RPCAUTH_CRED_LOCKED, and RPC_AUTH_PROC_CREDS are unused. Kill them.

Signed-off-by: Trond Myklebust <[email protected]>

NFSv3: fix sync_retry in direct i/o NFS

 Only do a sync_retry if the memcmp failed.

Signed-off-by: Dirk Mueller <[email protected]>
Signed-off-by: Trond Myklebust <[email protected]>

SUNRPC: Move upcall out of auth->au_ops->crcreate()

 This fixes a bug whereby if two processes try to look up the same auth_gss
 credential, they may end up creating two creds, and triggering two upcalls
 because the upcall is performed before the credential is added to the
 credcache.

Signed-off-by: Trond Myklebust <[email protected]>

SUNRPC: Remove the deprecated function lookup_hash() from rpc_pipefs code

Signed-off-by: Trond Myklebust <[email protected]>

SUNRPC: rpc_timeout_upcall_queue should not sleep

 The function rpc_timeout_upcall_queue runs from a workqueue, and hence
 sleeping is not recommended. Convert the protection of the upcall queue
 from being mutex-based to being spinlock-based.

Signed-off-by: Trond Myklebust <[email protected]>

SUNRPC: Fix a lock recursion in the auth_gss downcall

 When we look up a new cred in the auth_gss downcall so that we can stuff
 the credcache, we do not want that lookup to queue up an upcall in order
 to initialise it. To do an upcall here not only redundant, but since we
 are already holding the inode->i_mutex, it will trigger a lock recursion.

 This patch allows rpcauth cache searches to indicate that they can cope
 with uninitialised credentials.

Signed-off-by: Trond Myklebust <[email protected]>

NLM: Ensure that nlmclnt_cancel_callback() doesn't loop forever

 If the server returns NLM_LCK_DENIED_NOLOCKS, we currently retry the
 entire NLM_CANCEL request. This may end up looping forever unless the
 server changes its mind (why would it do that, though?).

 Ensure that we limit the number of retries (to 3).

 See bug# 5957 in bugzilla.kernel.org.

Signed-off-by: Trond Myklebust <[email protected]>

NLM: Fix arguments to NLM_CANCEL call

 The OpenGroup docs state that the arguments "block", "exclusive" and
 "alock" must exactly match the arguments for the lock call that we are
 trying to cancel.
 Currently, "block" is always set to false, which is wrong.

 See bug# 5956 on bugzilla.kernel.org.

Signed-off-by: Trond Myklebust <[email protected]>

[PATCH] kernel-doc: clean up the script (whitespace)

Remove lots of trailing whitespace.  Nothing else.

Signed-off-by: Randy Dunlap <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>