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

Many binutils versions over the last year appear to silently build
assembler files with undefined constants, and able to successfully
create executables from such files.  The assembler appears to add
undefined symbols to the symbol table without any corresponding
relocation information.  Obviously this is bad news since the
resulting executable may not be what the programmer intended.
Work around the problem by forcing the linker to fail if there are
any undefined symbols in the final object(s).

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

If we want to trap NULL vma's, we'd better be sure
that we don't dereference it first..

Lots of occurences of the same bug..

From: Meelis Roos <mroos@linux.ee>

This is self-explanatory - former SunDisk renamed itself to SanDisk and
now there are flash disks with both names.

 - we could pass MS_ACTIVE in mount flags and it would be passed into
   ->get_sb(), leading to interesting failure modes.  This flag is only
   for internal use (it's set once fill_super is complete and reset
   before the inode eviction on umount); made sure that we never get
   tricked into having it set it too early.

 - a bunch of filesystems force MS_NODIRATIME on mount but forgot to do
   the same on remount.  Fixed.

 - a bunch of r/o filesystems did force MS_RDONLY on mount but forgot to
   do the same on remount.  Fixed.

 - jff2->remount_fs() was buggy - it played with sb->s_flags instead of
   doing modifications to *flags (->s_flags will be overwritten using
   *flags right after the call of ->remount_fs()).  Moreover, it tried
   to do the wrong thing - it should just enforce noatime and be done
   with that.  Fixed, ACKed by maintainer.

 - we should force noatime both on mount and remount.  Fixed.

 - same problem as with sysv - mount-time checks for fs being good for
   writing are absent on remount.  Check added.

 - several variants of sysv fs are supported only r/o.  Driver does
   force r/o on mount, but doesn't do anything on remount.  As the
   result, one can remount them r/w and results are Not Pretty(tm).
   Missing checks added, code cleaned up. 

 - we had double-brelse() in v7fs - if sanity checks on root inode will
   succeed, but allocation of root dentry fails, we brelse() the same
   buffer_head twice.  Fixed.

into nuts.davemloft.net:/disk1/BK/tg3-2.6

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

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

The low level kernel interface to RTAS (the firmware runtime services)
was plagued with races that could cause from bogus results of RTAS
operations to total machine crashes in some circumstances. This patch
fix the ones I could identify, hoping I didn't miss any. I also added
a WARN_ON (well, it's asm equivalent) to enter_rtas to make sure we
never _ever_ try to call that with interrupts enabled.

This fix comes from ppc32. Always initialise dn->type and dn->name so
that code doesnt have to check for NULL everywhere. There is at least
one bug report where we oopsed because of this.

We have been debugging some strange fails where we branch to 0 in real
mode. At the moment this results in the cpu running through the
initialisation code and failing somewhere well into it.

The following patch uses the featuring nop'ing code to remove the branch
at real address 0 so it falls through to a trap instruction and gets
caught early.

This adda nmi_watchdog=2 support to the Pentium M processor.  The P-M is
a P6 chip, but it shares some chipset logic with the Pentium 4, so it
requires this workaround to function.

Without this patch, NMI gets stuck after 1 count.  With it, the NMI
fires and breaks me out of UHCI-related hard lockups. 

This patch is basically a modified version of the same patch for
oprofile. See the threaded discussion here:

	http://lkml.org/lkml/2004/2/12/181

into debian.org:/usr/src/kernel/ieee1394-2.6

Jeff Garzik notes that the previous cleanup highlights a bug:
>
> 	static const struct card_info {
> 	    zorro_id id;
> 	    const char *name;
> 	    unsigned int offset;
> 	} cards[] __initdata = {
>
> and the lone user is __devinit:
>
> static int __devinit zorro8390_init_one(struct zorro_dev *z,
>                                        const struct zorro_device_id *ent)

Here's the fix..

The author apparently didn't understand that only the mach-default
include directory is included by fallback for header files only.  You
can't stick a .c file in mach-default and expect all subarchs to be able
to use it.

The correct fix is to put std_resources.c in the kernel directory and
give it its own Kconfig symbol for conditional compile so that subarchs
may choose to include it or not.

The last copy_from_user patch to the vicam driver broke compilation with
VICAM_DEBUG on.

There is also another copy_from_user missing in case VIDIOCSPICT. 

This fixes both issues.

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

There's a bug in the x86 code in that it sets the boot CPU to zero.

This isn't correct since some subarch's use physically indexed CPUs. 
However, subarchs have either set the boot cpu before irq_INIT() (or
just inherited the default zero from INIT_THREAD_INFO()), so it's safe
to believe current_thread_info()->cpu about the boot cpu.

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

This adds IOMMU support & IOMU virtual merging to the default g5 config.

This will not impair performances of machines that don't need the iommu
(the kernel will only enable it if you have more than 2Gb of RAM, though
you can explicitely enable it using a command line argument).

The current code has a subtle race where 2 hash PTEs can be inserted
for the same virtual address for a short period of time. There should
not be a stale one as the "old" one ultimately gets flushed, but the
architecture specifies that having two hash PTE is illegal and can
result in undefined behaviour.

This patch fixes it by never clearing the _PAGE_HASHPTE bit when
doing test_and_clear_{young,dirty}. That means that subsequent faults
on those pages will have a bit more overhead to "discover" that the
hash entry was indeed evicted.

It also adds a small optisation to avoid doing the atomic operation
and the hash flush in test_and_clear_dirty when the page isn't dirty
or when setting write protect while it's already set.

From: Martin Schwidefsky <schwidefsky@de.ibm.com>

The new message queue interface needs the following patch to get it working
on s390 (31-bit, 64-bit and 31-bit compat).