On Fri, Dec 10, 2004 at 11:04:22AM +1100, Nathan Scott wrote:
> The bug# for the write deadlock I mentioned earlier -- 925836.

I finally found some time to look at it, and this fix is almost trivial.

XFS doesn't need the i_alloc_sem at all, so we should avoid taking it in
direct-io.c completely.  As a side-effect it makes the code a little bit
simpler even.
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

M68k: Remove nowhere referenced files
Signed-off-by: Domen Puncer <domen@coderock.org>
Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

MMC_WBSD depends on ISA (needs isa_virt_to_bus())
Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

M68k: Update defconfigs for 2.6.10
Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

We, Linux/M32R project members, decided to change the kernel API/ABI.  This
modification is not small, but if we don't change it now, perhaps we have
no chance to change them hereafter.

* Why change the m32r kernel API/ABI?
- The m32r port has many old-style syscall interfaces,
  because we made m32r port refering to the other traditional archs.
  Some old syscalls are no longer used or can be safely removed
  by upgrading the GNU C library.
- To make the m32r kernel more secure, it is preferable to prevent
  stack region from being executed. (e.g. stack overflow)

* API/ABI changes
- include/asm-m32r/unistd.h: Upgrade to the new kernel API.
- arch/m32r/entry.S: Minimum update to the new ABI.
- Don't use UID16 syscalls.
- To make stack noexecutable:
  1) Don't use trampoline for signal handlers for kernel space (cf. sparc64):
    sys_signal: remove.
    sys_sigaction, sys_rt_sigaction: use glibc's restorer.
  2) Don't generate trampoline code by GCC in userspace:
    Support non-executable stack by the m32r gcc.
    --> done (for gcc-3.4.3/gcc-4.0)

* New userland
- This modification does *not* keep backward compatibility.
  So we have been prepared new userland, based on the new API/ABI.
  Already, more than 200 new Debian deb binary packages are available
  on the Linux/M32R site:
  http://debian.linux-m32r.org/dists/04_ordovician/ (for this new ABI)
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
Signed-off-by: Hirokazu Takata <takata@linux-m32r.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

During resume, my previous patch switches over to the saved swsusp image
without suspending all devices first.  This patch fixes that oversight, so
that the state of the hardware upon resume more closely matches the state
it had at suspend time.

While my previous patch alone seemed to work fine in my testing, it is
not fully correct without this as well.
Signed-off-by: Barry K. Nathan <barryn@pobox.com>
Acked-by: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Since at least kernel 2.6.9, if not earlier, swsusp fails to properly
suspend and resume all devices.

The most notable effect is that resuming fails to properly reconfigure
interrupt routers.  In 2.6.9 this was obscured by other kernel code, but in
2.6.10 this often causes post-resume APIC errors and near-total failure of
some PCI devices (e.g.  network, sound and USB controllers).

Even in cases where interrupt routing is unaffected, this bug causes other
problems.  For instance, on one of my systems I have to run "ifdown
eth0;ifup eth0" after resume in order to have functional networking, if I
do not apply this patch.

By itself, this patch is not theoretically complete; my next patch fixes
that.  However, this patch is the critical one for fixing swsusp's behavior
in the real world.
Signed-off-by: Barry K. Nathan <barryn@pobox.com>
Acked-by: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

This introduces pm_message_t.  For now, it is only good for type-safety and
sparse checking, but plan is to turn pm_message_t into structure soon.
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

The old code cuts off the higher bits in CR4 (such as X86_CR4_OSFXSR,
X86_CR4_OSXMMEXCPT).

Signed-off-by: leendert@watson.ibm.com
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Comment on why iounmap is not appropriate in the early-boot legacy
ISA region dmi scan.
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

This patch makes some cleanups to the #defines for various fields in the
MMCR0 performance monitor control register.  Specifically, the names of a
couple of bits are changed so that: a) they are a bit less cumbersomely
long and b) they match the names used in the hardware documentation.
Signed-off-by: David Gibson <dwg@au1.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

In the recent IOMMU cleanup, the new LPAR code assumes that all PHBs must
have a dma window assigned to it.  On some machines we don't have a window
assinged unless there's an adapter in the slot.

In other words, a PHB without a ibm,dma-window property is not a bug and
must be tolerated.  This patch fixes that, and also removes a redundant
check for the dma-window being defined.
Signed-off-by: Olof Johansson <olof@austin.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

The vm_enough_memory functionality was replicated in three separate places,
and not always kept in sync.  It also used capable() for authorization checks.
This caused any process which ends up checking for this permission to have
PF_SUPERPRIV set (inappropriately), and caused poor dependencies between
stacked modules, since each LSM was generically asked to moderate
capable(CAP_SYS_ADMIN) without knowing why.
Signed-off-by: Serge Hallyn <serue@us.ibm.com>
Signed-off-by: Chris Wright <chrisw@osdl.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

The following patch splits bprm_apply_creds into two functions,
bprm_apply_creds and bprm_post_apply_creds.  The latter is called after the
task_lock has been dropped.  Without this patch, SELinux must drop the
task_lock and re-acquire it during apply_creds, making the 'unsafe' flag
meaningless to any later security modules.  Please apply.
Signed-off-by: Serge Hallyn <serue@us.ibm.com>
Signed-off-by: Stephen Smalley <sds@epoch.ncsc.mil>
Signed-off-by: Chris Wright <chrisw@osdl.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Recent changes mean nodemask.h is required in places it previously wasn't.
Signed-off-by: Chris Wedgwood <cw@f00f.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Even though these 2.4. interfaces are already gone in Dave Jones' cpufreq
bitkeeper tree, here's a patch which properly announces it in
Documentation/feature-removal-schedule.txt:

Add meaningful content concerning the removal of deprecated interfaces to
the cpufreq core.
Signed-off-by: Dominik Brodowski <linux@brodo.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

This fixes the entry in the MAINTAINERS file. I guess Marco agrees with
such change as he's not doing a maintaince for some time.

The patch below makes some needlessly global code static.

The most interesting part is that dquot_cachep can become static, since 
it isn't used outside of dquot.c .
Signed-off-by: Adrian Bunk <bunk@stusta.de>
Acked-by: Jan Kara <jack@suse.cz>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

When a thread stops for ptrace exit tracing, it cannot be resumed by
SIGKILL.  Once PF_EXITING is set, SIGKILL will not cause a wakeup from stop
(see wants_signal in kernel/signal.c).  This patch moves the ptrace stop
for exit tracing before the setting of PF_EXITING.
Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Upon reevaluation we think it is indeed safe to permit the race between
a ptrace call and the traced thread waking up, as long as it will never
get back to user mode.  This patch makes SIGKILL wake up threads in
TASK_TRACED.  That alone resolves most of the deadlock issues that
became possible with the introduction of TASK_TRACED, getting us back to
the killing behavior of 2.6.8 and before. 

This patch also further cleans up ptrace detaching, so that threads are
left in TASK_STOPPED only if a job control stop is actually in effect,
and otherwise resume.  This removes the past nuisances requiring a
SIGCONT to resume a thread even when it had a pending SIGKILL. 
Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

- allow CAP_IPC_LOCK to override mlock rlimit during stack expansion as
  in all other cases
Signed-off-by: Chris Wright <chrisw@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Signed-off-by: Chris Wright <chrisw@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

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

Patch from Richard Purdie

The Sharp SL-C7xx series of PDAs (Corgi, Shepherd, Husky) don't set
the machine type upon startup. Fixing this isn't possible as the
devices have been produced and the bootloaders are in ROM. 

This header code detects the differences between them and sets the
machine type accordingly. By doing this one kernel can support the
series of PXA25x machines.

The code also detects Poodle (Sharp SL-5600) which is also a PXA25x
device allowing a common kernel to support that device as well. (The
base machine code is under development for this device).

Signed-off-by: Richard Purdie
Signed-off-by: Russell King

Patch from Richard Purdie

Add base machine support for the Sharp SL-C7xx series of PDAs 
(Corgi - SL-C700, Shepherd - SL-C750 and Husky - SL-C760). See 
linux-arm-kernel for a support commitment for this code.

Depends on 2370/1 (SCOOP Driver).

Signed-off-by: Richard Purdie
Signed-off-by: Russell King

Patch from Richard Purdie

Add common driver code for the SCOOP I/O interface found on various 
Sharp PDAs (Collie, Poodle and Corgi models). It will be used by 
both SA1100 and PXA machines. 

[A patch to update Collie to use this driver will follow.]

Signed-off-by: Richard Purdie
Signed-off-by: Russell King

Patch from Ben Dooks

Remove code for decoding LCD Frame/FIFO IRQs
from the include/asm-arm/arch-s3c2410/entry-macro.S
file.

These are not currently implemented, and should be
decoded by a chained IRQ handler from the irq
code.

Signed-off-by: Ben Dooks
Signed-off-by: Russell King

Patch from Ben Dooks

If we read the INTPEND register and get the wrong
result (details below) then it looks like a possible
loop to go back and re-read it after calculating
the possible correct value for it.

It seems that the INTPEND register, which should
show the unique number of the interrupt that we
need to handle. However, it seems that this register
can show the wrong interrupt under certain conditions.

Shannon's research shows that it is very possible that
the occurence of an external interrupt can end up
merging with the base interrupt number, causing
the wrong result in the register.

This patch also fixes the end position of the #ifdef
block and ensures as many registers as possible are
stacked before calling printk() for debug.

Original patch for 2.6.9 by Shannon Holland re-written
for 2.6.10 by Ben Dooks.

Signed-off-by: Shannon Holland                             
Signed-off-by: Ben Dooks                                 
Signed-off-by: Russell King

Patch from Ben Dooks

This is an old definition, and is no longer used anywhere.
Add a comment to that effect in the header file
include/asm-arm/arch-s3c2410/timex.h and remove all other
references.

Thanks to Klaus Fetscher for pointing out that it isn't
used anymore

Signed-off-by: Ben Dooks
Signed-off-by: Russell King

Signed-off-by: Russell King <rmk@arm.linux.org.uk>

The __ptrace_unlink code that checks for TASK_TRACED fixed the problem of a
thread being left in TASK_TRACED when no longer being ptraced. 

However, an oversight in the original fix made it fail to handle the
case where the child is ptraced by its real parent.

Fixed thus.
Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

As Alan says:

  "The '1.8' version is for the Moxa mxser driver.  Moxa haven't
   released any 2.6 driver for the ancient 'smartio' hardware"

so the smartio driver stil remains valid.

Cset exclude: james4765@gmail.com[torvalds]|ChangeSet|20050108220924|25457

David handed over the keys a few months ago.  Time to make sure everyone
knows to send stuff to me.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

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

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

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

into kroah.com:/home/greg/linux/BK/i2c-2.6

into kroah.com:/home/greg/linux/BK/usb-2.6

The grows-up and grows-down cases had all the same issues, but
differered in the details. Additionlly, historical evolution of
the tests had caused the result to be pretty unreadable with some
rather long and complex conditionals.

Fix it all up in a more readable helper function.

This also adds the missing RLIMIT_MEMLOCK test.