edd: switch to pci_get based API

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

Update zilog timeout

Update zilog timeout, thanks Peter Fuerst.

Signed-off-by: Thiemo Seufer <[email protected]>
Cc: Martin Michlmayr <[email protected]>
Cc: Peter Fuerst <[email protected]>
Cc: Russell King <[email protected]>
Cc: Alan Cox <[email protected]>
Cc: Ralf Baechle <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

fix compiler warnings in acorn.c

warning: 'adfs_partition' defined but not used
warning: 'riscix_partition' defined but not used
warning: 'linux_partition' defined but not used

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

drivers/block/z2ram: Remove TRUE/FALSE defines

Remove defines of TRUE and FALSE
  * not used in the file
  * the file is not included somewhere else

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

Remove unnecessary includes of spinlock.h under include/linux

Remove the obviously unnecessary includes of <linux/spinlock.h> under the
include/linux/ directory, and fix the couple errors that are introduced as
a result of that.

Signed-off-by: Robert P. J. Day <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

fat: gcc 4.3 warning fix

This patch fixes the following warnings.

fs/fat/dir.c: In function 'fat_parse_long':
include/linux/msdos_fs.h:294: warning: array subscript is above array bounds
include/linux/msdos_fs.h:295: warning: array subscript is above array bounds
include/linux/msdos_fs.h:295: warning: array subscript is above array bounds

The ->name is defined as "name[8], ext[3]", but fat_checksum() uses
those as name[11]. There is no actual problem, but it's not a good manner.

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

Make NFS client use seq_list_xxx helpers

This includes /proc/fs/nfsfs/servers and /proc/fs/nfsfs/volumes entries.

Both need to show the header and use the list_head.

Signed-off-by: Pavel Emelianov <[email protected]>
Acked-by: Trond Myklebust <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

Make /proc/self/mounts(tats) use seq_list_xxx helpers

One more simple and stupid switching to the new API.

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

Make /proc/tty/drivers use seq_list_xxx helpers

Simple and stupid like some previous ones.  Just use new API.

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

Make /proc/modules use seq_list_xxx helpers

Here there is not need even in .show callback altering.  The original code
passes list_head in *v.

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

Make /proc/misc use seq_list_xxx helpers

Simple and stupid - just use the helpers.

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

Make crypto API use seq_list_xxx helpers

Simple and stupid - just use the same code from another place in the kernel.

Signed-off-by: Pavel Emelianov <[email protected]>
Acked-by: Herbert Xu <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

Make AFS use seq_list_xxx helpers

These proc files show some header before dumping the list, so the
seq_list_start_head() is used.

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

percpu_counters: use for_each_online_cpu()

Now that we have implemented hotunplug-time counter spilling,
percpu_counter_sum() only needs to look at online CPUs.

Cc: Gautham R Shenoy <[email protected]>
Cc: Oleg Nesterov <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

percpu_counters(): use cpu notifiers

per-cpu counters presently must iterate over all possible CPUs in the
exhaustive percpu_counter_sum().

But it can be much better to only iterate over the presently-online CPUs.  To
do this, we must arrange for an offlined CPU's count to be spilled into the
counter's central count.

We can do this for all percpu_counters in the machine by linking them into a
single global list and walking that list at CPU_DEAD time.

(I hope.  Might have race windows in which the percpu_counter_sum() count is
inaccurate?)

Cc: Gautham R Shenoy <[email protected]>
Cc: Oleg Nesterov <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

vxfs warning fixes

gcc-4.3:

fs/freevxfs/vxfs_lookup.c: In function 'vxfs_find_entry':
fs/freevxfs/vxfs_lookup.c:139: warning: cast from pointer to integer of different size
fs/freevxfs/vxfs_lookup.c: In function 'vxfs_readdir':
fs/freevxfs/vxfs_lookup.c:294: warning: cast from pointer to integer of different size

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

fuse warning fix

gcc-4.3:

fs/fuse/dir.c: In function 'parse_dirfile':
fs/fuse/dir.c:833: warning: cast from pointer to integer of different size
fs/fuse/dir.c:835: warning: cast from pointer to integer of different size

[[email protected]: use offsetof]
Acked-by: Miklos Szeredi <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

Use mutexes instead of semaphores in I2O driver

The I2O driver uses two semaphores as mutexes.  Use the mutex API instead of
the (binary) semaphores.

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

cpu hotplug: fix ksoftirqd termination on cpu hotplug with naughty realtime process

Fix ksoftirqd termination on cpu hotplug with naughty real time process.

Assuming the following case:

 - Try to hot remove CPU2 from CPU1.
 - There is a real time process on CPU2, and that process doesn't sleep at all.
 - That rt process and ksoftirqd/2 is migrated to the CPU0

Then ksoftirqd/2 can't stop becasue that rt process runs everlastingly on
CPU0, and CPU1 waiting the ksoftirqd/2's termination hangs up.  To fix this
problem, set the priority of ksoftirqd/2 to max one before kthread_stop().

[[email protected]: fix warning]
Signed-off-by: Satoru Takeuchi <[email protected]>
Cc: Rusty Russell <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Oleg Nesterov <[email protected]>
Cc: Ashok Raj <[email protected]>
Cc: Gautham R Shenoy <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

Fix stop_machine_run problem with naughty real time process

stop_machine_run() does its work on "kstopmachine" thread having max
priority.  However that thread get such priority after woken up.
Therefore, in the following case ...

  - "kstopmachine" try to run on CPU1

  - There is a real time process which doesn't relinquish CPU time
    voluntary on CPU1

...  "kstopmachine" can't start to run and the CPU on which
    stop_machine_run() is runing hangs up.  To fix this problem, call
    sched_setscheduler() before waking up that thread.

Signed-off-by: Satoru Takeuchi <[email protected]>
Cc: Rusty Russell <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Oleg Nesterov <[email protected]>
Cc: Ashok Raj <[email protected]>
Cc: Gautham R Shenoy <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

UDF: check for allocated memory for inode data

This patch adds checking for granted memory while filling up inode data to
prevent possible NULL pointer usage.  If there is not enough memory to fill
inode data we just mark it as "bad".  Also some whitespace cleanup.

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

Prevent an O_NDELAY writer from blocking when a tty write is blocked by the tty atomic writer mutex

Without this a tty write could block if a previous blocking tty write was
in progress on the same tty and blocked by a line discipline or hardware
event.  Originally found and reported by Dave Johnson.

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

UDF: check for allocated memory for data of new inodes

Add checking for granted memory for inode data at the moment of its
creation.

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

Use boot based time for uptime in /proc

Commit 411187fb05cd11676b0979d9fbf3291db69dbce2 caused uptime not to increase
during suspend.  This may cause confusion so I restore the old behaviour by
using the boot based time instead of monotonic for uptime.

Signed-off-by: Tomas Janousek <[email protected]>
Acked-by: John Stultz <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

Use boot based time for process start time and boot time in /proc

Commit 411187fb05cd11676b0979d9fbf3291db69dbce2 caused boot time to move and
process start times to become invalid after suspend.  Using boot based time
for those restores the old behaviour and fixes the issue.

[[email protected]: little cleanup]
Signed-off-by: Tomas Janousek <[email protected]>
Cc: Tomas Smetana <[email protected]>
Acked-by: John Stultz <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Ingo Molnar <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

Introduce boot based time

The commits

  411187fb05cd11676b0979d9fbf3291db69dbce2 (GTOD: persistent clock support)
  c1d370e167d66b10bca3b602d3740405469383de (i386: use GTOD persistent clock
    support)

changed the monotonic time so that it no longer jumps after resume, but it's
not possible to use it for boot time and process start time calculations then.
 Also, the uptime no longer increases during suspend.

I add a variable to track the wall_to_monotonic changes, a function to get the
real boot time and a function to get the boot based time from the monotonic
one.

[[email protected]: remove exports, add comment]
Signed-off-by: Tomas Janousek <[email protected]>
Cc: Tomas Smetana <[email protected]>
Cc: John Stultz <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Ingo Molnar <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

use no_pci_devices() in pci/search.c

We have an API function for this now.

Cc: Zhang Yanmin <[email protected]>
Cc: Greg KH <[email protected]>
Cc: Bartlomiej Zolnierkiewicz <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

fix jvc cdrom drive lockup

Before calling init_hwif_default, ide_unregister gets lock ide_lock and
disables irq.  init_hwif_default calls ide_default_io_base which calls
pci_get_device and later pci_get_subsys tries to apply for semaphore
pci_bus_sem and goes to sleep.

Mostly, pci_get_device should be called when irq is turned on.

ide_default_io_base just needs find if list pci_devices is empty.

Signed-off-by: Zhang Yanmin <[email protected]>
Cc: Greg KH <[email protected]>
Cc: Bartlomiej Zolnierkiewicz <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

Use menuconfig objects: W1

Use menuconfigs instead of menus, so the whole menu can be disabled at once
instead of going through all options.

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

Use menuconfig objects: PNP

Use menuconfigs instead of menus, so the whole menu can be disabled at once
instead of going through all options.

Signed-off-by: Jan Engelhardt <[email protected]>
Cc: Adam Belay <[email protected]>
Cc: Bjorn Helgaas <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

Use menuconfig objects: parport

Use menuconfigs instead of menus, so the whole menu can be disabled at once
instead of going through all options.

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

Use menuconfig objects: I2O

Use menuconfigs instead of menus, so the whole menu can be disabled at once
instead of going through all options.

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

Use menuconfig objects: crypto hw

Use menuconfigs instead of menus, so the whole menu can be disabled at once
instead of going through all options.

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

Use menuconfig objects: connector

Use menuconfigs instead of menus, so the whole menu can be disabled at once
instead of going through all options.

Signed-off-by: Jan Engelhardt <[email protected]>
Acked-by: Evgeniy Polyakov <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

Use menuconfig objects II - TPM

Change Kconfig objects from "menu, config" into "menuconfig" so
that the user can disable the whole feature without having to
enter the menu first.

Signed-off-by: Jan Engelhardt <[email protected]>
Cc: Kylene Hall <[email protected]>
Cc: Marcel Selhorst <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

Use menuconfig objects II - Telephony

Change Kconfig objects from "menu, config" into "menuconfig" so
that the user can disable the whole feature without having to
enter the menu first.

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

Use menuconfig objects II - oprofile

Make a "menuconfig" out of the Kconfig objects "menu, ..., endmenu",
so that the user can disable all the options in that menu at once
instead of having to disable each option separately.

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

Use menuconfig objects II - module menu

Change menuconfig objects from "menu, config" into "menuconfig" so that the
user can disable the whole feature without entering its menu first.

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

Use menuconfig objects II - misc strange dev

Make a "menuconfig" out of the Kconfig objects "menu, ..., endmenu",
so that the user can disable all the options in that menu at once
instead of having to disable each option separately.

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

Use menuconfig objects II - IPMI

Change Kconfig objects from "menu, config" into "menuconfig" so
that the user can disable the whole feature without having to
enter the menu first.

Signed-off-by: Jan Engelhardt <[email protected]>
Acked-by: Corey Minyard <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

Use menuconfig objects II - EDAC

Change Kconfig objects from "menu, config" into "menuconfig" so
that the user can disable the whole feature without having to
enter the menu first.

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

Use menuconfig objects II - auxdisplay

Make a "menuconfig" out of the Kconfig objects "menu, ..., endmenu",
so that the user can disable all the options in that menu at once
instead of having to disable each option separately.

Signed-off-by: Jan Engelhardt <[email protected]>
Acked-by: Miguel Ojeda Sandonis <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

Use write_trylock_irqsave in ptrace_attach

This patch makes ptrace_attach use write_trylock_irqsave().

[[email protected]: remove unneeded initialisation]
Signed-off-by: Sripathi Kodi <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Oleg Nesterov <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

introduce write_trylock_irqsave()

Introduce a write_trylock_irqsave() implementation.  Similar in style to
the implementation of spin_trylock_irqsave() in mainline.

Signed-off-by: Satyam Sharma <[email protected]>
Cc: Sripathi Kodi <[email protected]>
Cc: Ingo Molnar <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

more scheduled OSS driver removal

This patch contains the scheduled removal of OSS drivers that:
- have ALSA drivers for the same hardware without known regressions and
- whose Kconfig options have been removed in 2.6.20.

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

Fix rmmod/read/write races in /proc entries

Fix following races:
===========================================
1. Write via ->write_proc sleeps in copy_from_user(). Module disappears
   meanwhile. Or, more generically, system call done on /proc file, method
   supplied by module is called, module dissapeares meanwhile.

   pde = create_proc_entry()
   if (!pde)
return -ENOMEM;
   pde->write_proc = ...
open
write
copy_from_user
   pde = create_proc_entry();
   if (!pde) {
remove_proc_entry();
return -ENOMEM;
/* module unloaded */
   }
*boom*
==========================================
2. bogo-revoke aka proc_kill_inodes()

  remove_proc_entry vfs_read
  proc_kill_inodes [check ->f_op validness]
[check ->f_op->read validness]
[verify_area, security permissions checks]
->f_op = NULL;
if (file->f_op->read)
/* ->f_op dereference, boom */

NOTE, NOTE, NOTE: file_operations are proxied for regular files only. Let's
see how this scheme behaves, then extend if needed for directories.
Directories creators in /proc only set ->owner for them, so proxying for
directories may be unneeded.

NOTE, NOTE, NOTE: methods being proxied are ->llseek, ->read, ->write,
->poll, ->unlocked_ioctl, ->ioctl, ->compat_ioctl, ->open, ->release.
If your in-tree module uses something else, yell on me. Full audit pending.

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

v850: enable arbitary speed tty ioctls

Adding the defines/macros activates the existing code in the tty layer
and allows this platform to use the arbitary speed ioctl setting layer

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

uml: remove dead file

I forgot this file was here.  It hasn't been used since UML has been in
mainline.

Thanks to Jesper for finding something that needed doing to it, thus reminding
me of its existence.

Signed-off-by: Jeff Dike <[email protected]>
Cc: Paolo 'Blaisorblade' Giarrusso <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

uml: limit request size on COWed devices

COWed devices can't handle more than 32 (64 on x86_64) sectors in one request
due to the size of the bitmap being carried around in the io_thread_req.

Enforce that by telling the block layer not to put too many sectors in
requests to COWed devices.

Signed-off-by: Jeff Dike <[email protected]>
Cc: Paolo 'Blaisorblade' Giarrusso <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

uml: export hostfs symbols

Add some exports for hostfs that are required after Alberto Bertogli's fixes
for accessing unlinked host files.

Also did some style cleanups while I was here.

Signed-off-by: Jeff Dike <[email protected]>
Cc: Paolo 'Blaisorblade' Giarrusso <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

uml: Eliminate kernel allocator wrappers

UML had two wrapper procedures for kmalloc, um_kmalloc and um_kmalloc_atomic
because the flag constants weren't available in userspace code.
kern_constants.h had made kernel constants available for a long time, so there
is no need for these wrappers any more.  Rather, userspace code calls kmalloc
directly with the userspace versions of the gfp flags.

kmalloc isn't a real procedure, so I had to essentially copy the inline
wrapper around __kmalloc.

vmalloc also had its own wrapper for no good reason.  This is now gone.

Signed-off-by: Jeff Dike <[email protected]>
Cc: Paolo 'Blaisorblade' Giarrusso <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

uml: simplify helper stack handling

run_helper and run_helper_thread had arguments which were the same in all
callers.  run_helper's stack_out was always NULL and run_helper_thread's
stack_order was always 0.  These are now gone, and the constants folded
into the code.

Also fixed leaks of the helper stack in the AIO and SIGIO code.

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

uml: SIGIO support cleanup

Cleanup of the SIGWINCH support.

Some code and comment reformatting.

The stack used for SIGWINCH threads was leaked.  This is now fixed by storing
it with the pid and other information, and freeing it when the thread is
killed.

If something goes wrong with a WIGWINCH thread, and this is discovered in the
interrupt handler, the winch record would leak.  It is now freed, except that
the IRQ isn't freed.  This is hard to do from interrupt context.  This has the
side-effect that the IRQ system maintains a reference to the freed structure,
but that shouldn't cause a problem since the descriptor is disabled.

register_winch_irq is now much better about cleaning up after an
initialization failure.

Signed-off-by: Jeff Dike <[email protected]>
Cc: Paolo 'Blaisorblade' Giarrusso <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

uml: handle errors on opening host side of consoles

If the host side of a console can't be opened, this will now produce visible
error messages.

enable_chan now returns a status and this is passed up to con_open and
ssl_open, which will complain if anything went wrong.

The default host device for the serial line driver is now a pts device rather
than a pty device since lots of hosts have LEGACY_PTYS disabled.  This had
always been failing on such hosts, but silently.

Signed-off-by: Jeff Dike <[email protected]>
Cc: Paolo 'Blaisorblade' Giarrusso <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

uml: pty channel tidying

Cleanup, mostly style violations.

Tidied the includes.

getmaster returns a real errno, which pty_open returns if there's a
problem.

The printks now have severity.

Changed os_* calls to call libc directly.

Signed-off-by: Jeff Dike <[email protected]>
Cc: Paolo 'Blaisorblade' Giarrusso <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

uml: xterm driver tidying

Major tidying of the xterm console driver:
got rid of the tt-mode gdb support
tidied up the includes
fixed lots of style violations
replaced os_* calls with glibc calls in xterm.c
all printk calls now have a severity indicator
the error paths of xterm_open are closer to being right

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

uml: DEBUG_SHIRQ fixes

DEBUG_SHIRQ generates spurious interrupts, triggering handlers such as
mconsole_interrupt() or line_interrupt().  They expect data to be available to
be read from their sockets/pipes, but in the case of spurious interrupts, the
host didn't actually send anything, so UML hangs in read() and friends.
Setting those fd's as O_NONBLOCK makes DEBUG_SHIRQ-enabled UML kernels boot
and run correctly.

Signed-off-by: Eduard-Gabriel Munteanu <[email protected]>
Signed-off-by: Jeff Dike <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

Add generic exit-time stack-depth checking to CONFIG_DEBUG_STACK_USAGE

Add generic exit-time stack-depth checking to CONFIG_DEBUG_STACK_USAGE.

This also adds UML support.

Tested on UML and i386.

[[email protected]: cleanups, speedups, tweaks]
Signed-off-by: Jeff Dike <[email protected]>
Cc: Oleg Nesterov <[email protected]>
Cc: Ingo Molnar <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

uml: use get_free_pages to allocate kernel stacks

For some reason, I was using kmalloc instead of get_free_pages for kernel
stacks.

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

uml: fix request->sector update

It is theoretically possible for a request to finish and be freed between
writing it to the I/O thread and updating the sector count.  In this case, the
update will dereference a freed pointer.

To avoid this, I delay the update until processing the next sg segment, when
the request pointer is known to be good.

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

CRIS: replace old-style member inits with designated inits

Replace the old-style structure member initializers with designated
initializers.

Signed-off-by: Robert P. J. Day <[email protected]>
Cc: Mikael Starvik <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

etrax: enable arbitary speed setting on tty ports

Add the needed constants and bits. The actual code is already in the tty
layer and turned on by the definitions

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

m32r: A MAINTAINERS entry for the M32R architecture

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

m32r: enable arbitary speed tty rate setting

Add the defines and constants needed for the M32R platform to support the
arbitary speed tty ioctls.

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

arm26: remove broken and unused macro

Signed-off-by: Mariusz Kozlowski <[email protected]>
Cc: Ian Molton <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

ARM26: enable arbitary speed tty ioctls and split input/output speed

Add the ioctls and values needed for this to the ARM26/ARM32 ports.  The
actual code has been in the base kernel for a while and automatically turns
on when a port sets the required defines.

Signed-off-by: Alan Cox <[email protected]>
Cc: Russell King <[email protected]>
Cc: Ian Molton <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

fix alpha ISA support

isa_bus_to_virt() is still needed in a few places (lance.c, at least).  When
we switch the kernel to using -Werror-implicit-function-declaration, the lack
of isa_bus_to_virt() breaks alpha allmodconfig builds.

Add isa_bus_to_virt() and deprecate the ezisting ISA APIs, though it might be
better to define these functions as BUG(), since virt_to_bus/bus_to_virt just
do wrong things on a number of machines.

[[email protected]: build fix]
Signed-off-by: Ivan Kokshaysky <[email protected]>
Cc: Richard Henderson <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

alpha: fix trivial section mismatch warnings

Fix the following section mismatch warnings:
WARNING: arch/alpha/kernel/built-in.o(.text+0x7c78): Section mismatch: reference to .init.text:init_rtc_irq (between 'common_init_rtc' and 'timer_interrupt')
WARNING: arch/alpha/kernel/built-in.o(.text+0x7c7c): Section mismatch: reference to .init.text:init_rtc_irq (between 'common_init_rtc' and 'timer_interrupt')
WARNING: arch/alpha/kernel/built-in.o(.data+0x2c30): Section mismatch: reference to .init.text:srm_console_setup (between 'srmcons' and 'tsunami_pci_ops')

In all three cases functions marked __init was called outside __init context.
So the fix was to just drop the __init attribute.

Signed-off-by: Sam Ravnborg <[email protected]>
Cc: Meelis Roos <[email protected]>
Cc: Ivan Kokshaysky <[email protected]>
Cc: Richard Henderson <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

h8300 entry.S update

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

h8300: remove unused file

arch/h8300/kernel/ints.c is unused.

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

h8300 zImage support update

- Add missing files
- Add Makefile target
- Change image base
- Style fix

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

h8300: enable arbitary speed tty port setup

Add the needed constants and defines to activate the new tty code on this
platform

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

m68knommu: remove old cache management cruft from mm code

Remove cache management cruft.  This code is dead, all the cache manangement
functions for the ColdFire exist in the header file
include/asm-m68knommu/cacheflush.h.

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

m68knommu: remove cruft from setup code

Clean out cruft.

. remove include files not needed
. remove not used CAT_ROMARRAY code
. remove generic machine pointers not used
. remove unused functions
. fix email address in copyrights

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

m68knommu: use TRHEAD_SIZE instead of hard constant

Use THREAD_SIZE instead of a hard constant.

Signed-off-by: Philippe De Muyter <[email protected]>
Signed-off-by: Greg Ungerer <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

nommu: stub expand_stack() for nommu case

Be consistent with VM mmap, implement expand_stack().  We can't actually do
anything other than return an error in the no MMU case though.

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

FRV: Remove some dead code

Remove some dead chunks of code that are bounded by preprocessor conditionals
controlled by apparently no-longer available config options.

These are:

CONFIG_BLK_DEV_BLKMEM
CONFIG_CHR_DEV_FLASH
CONFIG_BLK_DEV_FLASH
CONFIG_CONSOLE

[Found by Robert P. J. Day <[email protected]>]

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

FRV: Be (self-)consistent and use CONFIG_GDB_CONSOLE everywhere

Be (self-)consistent and use CONFIG_GDB_CONSOLE everywhere rather than using
CONFIG_GDBSTUB_CONSOLE in some places and not others.  This is also then
consistent with other archs.

Also remove the gdbstub console device() op which doesn't seem to be necessary
now (especially as it doesn't compile).

[Found by Robert P. J. Day <[email protected]>]

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

FRV: Connect up new syscalls

Connect up new system calls.

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

split mmap

This is a straightforward split of do_mmap_pgoff() into two functions:

 - do_mmap_pgoff() checks the parameters, and calculates the vma
   flags.  Then it calls

 - mmap_region(), which does the actual mapping

Signed-off-by: Miklos Szeredi <[email protected]>
Acked-by: Peter Zijlstra <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

NeilBrown <[email protected]>

The do_loop_readv_writev implementation of readv breaks out of the loop as
soon as a single read request didn't fill it's buffer:

if (nr != len)
break;

The generic_file_aio_read version doesn't.  So if it hits EOF before the end
of the list of buffers, it will try again on the next buffer.  If the file was
extended in the mean time, this will produce a bad result.

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

do not limit locked memory when RLIMIT_MEMLOCK is RLIM_INFINITY

Fix a bug in mm/mlock.c on 32-bit architectures that prevents a user from
locking more than 4GB of shared memory, or allocating more than 4GB of
shared memory in hugepages, when rlim[RLIMIT_MEMLOCK] is set to
RLIM_INFINITY.

Signed-off-by: Herbert van den Bergh <[email protected]>
Acked-by: Chris Mason <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

slob: sparsemem support

Currently slob is disabled if we're using sparsemem, due to an earlier
patch from Goto-san.  Slob and static sparsemem work without any trouble as
it is, and the only hiccup is a missing slab_is_available() in the case of
sparsemem extreme.  With this, we're rid of the last set of restrictions
for slob usage.

Signed-off-by: Paul Mundt <[email protected]>
Acked-by: Pekka Enberg <[email protected]>
Acked-by: Matt Mackall <[email protected]>
Cc: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

mspec_mmap: don't set VM_LOCKED

mspec_mmap was setting VM_LOCKED (without adjusting locked_vm): don't do
that, it serves no purpose in 2.6, other than to mess up the locked_vm
accounting - mspec's pages won't get reclaimed anyway.  Thanks to Dmitry
Monakhov for raising the issue.

Signed-off-by: Hugh Dickins <[email protected]>
Acked-by: Jes Sorensen <[email protected]>
Cc: Dmitry Monakhov <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

mm/page_alloc.c: lower printk severity

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

slob: initial NUMA support

This adds preliminary NUMA support to SLOB, primarily aimed at systems with
small nodes (tested all the way down to a 128kB SRAM block), whether
asymmetric or otherwise.

We follow the same conventions as SLAB/SLUB, preferring current node
placement for new pages, or with explicit placement, if a node has been
specified.  Presently on UP NUMA this has the side-effect of preferring
node#0 allocations (since numa_node_id() == 0, though this could be
reworked if we could hand off a pfn to determine node placement), so
single-CPU NUMA systems will want to place smaller nodes further out in
terms of node id.  Once a page has been bound to a node (via explicit node
id typing), we only do block allocations from partial free pages that have
a matching node id in the page flags.

The current implementation does have some scalability problems, in that all
partial free pages are tracked in the global freelist (with contention due
to the single spinlock).  However, these are things that are being reworked
for SMP scalability first, while things like per-node freelists can easily
be built on top of this sort of functionality once it's been added.

More background can be found in:

http://marc.info/?l=linux-mm&m=118117916022379&w=2
http://marc.info/?l=linux-mm&m=118170446306199&w=2
http://marc.info/?l=linux-mm&m=118187859420048&w=2

and subsequent threads.

Acked-by: Christoph Lameter <[email protected]>
Acked-by: Matt Mackall <[email protected]>
Signed-off-by: Paul Mundt <[email protected]>
Acked-by: Nick Piggin <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

speed up madvise_need_mmap_write() usage

In the new madvise_need_mmap_write() call we can avoid an extra case
statement and function call as follows.

Signed-off-by: Jason Baron <[email protected]>
Cc: Nishanth Aravamudan <[email protected]>
Cc: Christoph Hellwig <[email protected]>
Cc: Nick Piggin <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

mm/slab.c: start_cpu_timer() should be __cpuinit

start_cpu_timer() should be __cpuinit (which also matches what it's
callers are).

__devinit didn't cause problems, it simply wasted a few bytes of memory
for the common CONFIG_HOTPLUG_CPU=n case.

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

mm: more __meminit annotations

Currently zone_spanned_pages_in_node() and zone_absent_pages_in_node() are
non-static for ARCH_POPULATES_NODE_MAP and static otherwise.  However, only
the non-static versions are __meminit annotated, despite only being called
from __meminit functions in either case.

zone_init_free_lists() is currently non-static and not __meminit annotated
either, despite only being called once in the entire tree by
init_currently_empty_zone(), which too is __meminit.  So make it static and
properly annotated.

Signed-off-by: Paul Mundt <[email protected]>
Cc: Yasunori Goto <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

kill vmalloc_earlyreserve

This symbol got orphaned quite a while ago.

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

page table handling cleanup

Kill pte_rdprotect(), pte_exprotect(), pte_mkread(), pte_mkexec(), pte_read(),
pte_exec(), and pte_user() except where arch-specific code is making use of
them.

Signed-off-by: Jan Beulich <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: Christoph Hellwig <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

mm: fix improper .init-type section references

.. which modpost started warning about.

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

numa: mempolicy: trivial debug fixes.

Enabling debugging fails to build due to the nodemask variable in
do_mbind() having changed names, and then oopses on boot due to the
assumption that the nodemask can be dereferenced -- which doesn't work out
so well when the policy is changed to MPOL_DEFAULT with a NULL nodemask by
numa_default_policy().

This fixes it up, and switches from PDprintk() to pr_debug() while
we're at it.

Signed-off-by: Paul Mundt <[email protected]>
Cc: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

oom: stop allocating user memory if TIF_MEMDIE is set

get_user_pages() can try to allocate a nearly unlimited amount of memory on
behalf of a user process, even if that process has been OOM killed.  The
OOM kill occurs upon return to user space via a SIGKILL, but
get_user_pages() will try allocate all its memory before returning.  Change
get_user_pages() to check for TIF_MEMDIE, and if set then return
immediately.

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

numa: mempolicy: dynamic interleave map for system init

This converts the default system init memory policy to use a dynamically
created node map instead of defaulting to all online nodes.  Nodes of a
certain size (>= 16MB) are judged to be suitable for interleave, and are added
to the map.  If all nodes are smaller in size, the largest one is
automatically selected.

Without this, tiny nodes find themselves out of memory before we even make it
to userspace.  Systems with large nodes will notice no change.

Only the system init policy is effected by this change, the regular
MPOL_DEFAULT policy is still switched to later on in the boot process as
normal.

Signed-off-by: Paul Mundt <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: Christoph Lameter <[email protected]>
Cc: Hugh Dickins <[email protected]>
Cc: Lee Schermerhorn <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

SLUB: support slub_debug on by default

Add a new configuration variable

CONFIG_SLUB_DEBUG_ON

If set then the kernel will be booted by default with slab debugging
switched on. Similar to CONFIG_SLAB_DEBUG. By default slab debugging
is available but must be enabled by specifying "slub_debug" as a
kernel parameter.

Also add support to switch off slab debugging for a kernel that was
built with CONFIG_SLUB_DEBUG_ON. This works by specifying

slub_debug=-

as a kernel parameter.

Dave Jones wanted this feature.
http://marc.info/?l=linux-kernel&m=118072189913045&w=2

[[email protected]: clean up switch statement]
Signed-off-by: Christoph Lameter <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

invalidate_mapping_pages(): add cond_resched

invalidate_mapping_pages() can sometimes take a long time (millions of pages
to free).  Long enough for the softlockup detector to trigger.

We used to have a cond_resched() in there but I took it out because the
drop_caches code calls invalidate_mapping_pages() under inode_lock.

The patch adds a nasty flag and puts the cond_resched() back.

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

mm: debug check for the fault vs invalidate race

Add a bugcheck for Andrea's pagefault vs invalidate race.  This is triggerable
for both linear and nonlinear pages with a userspace test harness (using
direct IO and truncate, respectively).

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

hugetlb: fix race in alloc_fresh_huge_page()

That static `nid' index needs locking.  Without it we can end up calling
alloc_pages_node() with an illegal node ID and the kernel crashes.

Acked-by: gurudas pai <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

vmscan: fix comments related to shrink_list()

Fix the shrink_list name on some files under mm/ directory.

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