cpu_to_le32(...) assigned to 16bit fields.

Signed-off-by: Tom Rini <trini@kernel.crashing.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Just test the end case inside the loop, rather than trying to
be clever and getting it wrong.

The previous single-step patch ("make single-step into signal delivery
stop in handler") took things a little too far. 

It left TF set in the sigcontext on the stack, so a PTRACE_CONT after
stopping at the handler entry will step instead of resume.  This
additional patch fixes it. 
Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

timer_create leaks task_structs.  I probably introduced this bug when I did
the cleanup making posix-timers properly per-process.  This patch fixes it.

There is also a fixup for a random indentation snafu at the end.
Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

    Helps to keep the 'sendmail' doctor away.
Signed-off-by: Trond Myklebust <trond.myklebust@fys.uio.no>
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 flint.arm.linux.org.uk:/usr/src/bk/linux-2.6-rmk

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

Patch from Ian Campbell

Please ignore previous patch: set_irq_type takes the IRQ number and
not the GPIO number so using any variant on IRQ_TO_GPIO* is incorrect
anyway.
Signed-of-by: Ian Campbell <icampbell@arcom.com>

Patch from Nicolas Pitre

- Don't chain assignment to volatile registers.
  Doing otherwise generates extra needless code to reload the value
  for the next assignment.

- Don't touch PXA27x registers on a PXA25x build.
  (spotted by Ian Campbell)

Signed-off-by: Nicolas Pitre 

 CONFIG_ARCH_IOP3XX

Patch from Nicolas Pitre

Signed-off-by: Nicolas Pitre 

Basic serial ports, acpi and intel sound.

Can you guess what my laptop has in it?

There's tons of mis-use of PCI memory-mapped IO that is used
as if it was regular memory. That fails disastrously on a number
of architectures, and it doesn't help that it just happens to
work on regular x86 boxes.

This makes makes us do much stricter type-checking. Some of it
visible to the regular compiler, but the bulk of it is for sparse.

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: Hideaki YOSHIFUJI <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

We run into a potential problem if we are doing
TSO and we reduce the congestion window.  We might
create a case where the TSO packet is what we need
to resend but the congestion window is not large
enough to allow it through.

The fix is very simple, since tcp_retransmit_skb()
is going to chop the size of the packet down to size
of the normal non-TSO MSS, we can pretend at the top
level that each SKB is composed of only one real MSS
worth of data.
Signed-off-by: David S. Miller <davem@davemloft.net>

Calls to tcp_fragment() change the tso_factor of
an SKB, so we need to deal with that.
Signed-off-by: David S. Miller <davem@davemloft.net>

Do not reference the tso_factor of SKB buff until
it is initialized.
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Hideaki YOSHIFUJI <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

If we don't have IPv6 default routes, we assume all ipv6 destinations 
are on-link as specified in RFC2461. It, however, is considered harmful now;
it is problematic with IPv6-capable nodes that do not have off-link
IPv6 connectivity (eg no default routers) and such nodes will take
a few seconds until they fall back to use IPv4.

See <draft-ietf-v6ops-onlinkassumption-02.txt> for details.

From: KUNITAKE Koichi <kunitake@anchor.jp>
Signed-off-by: KUNITAKE Koichi <kunitake@anchor.jp>
Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Here is a really ugly patch to get IPCOMP to use per-cpu buffers.  But
I'm afraid it really is necessary.  At 300K per SA IPCOMP isn't very
affordable at all.

With per-cpu buffers this goes down to 300K per CPU.

I've also turned the kmalloc'ed scratch space into a vmalloc'ed one
since people may be loading the ipcomp module after the system has
been running for a while.  On an i386 machine with 64M of RAM or less
this can often cause a 64K kmalloc to fail.

The crypto deflate buffer space are vmalloc'ed already as well.

Part of the ugliness comes from the lazy allocation.  However we need
the lazy initialisation since new IPCOMP algorithms may be introduced
in future.  That means we can't allocate space for every single IPCOMP
algorithm at module-load time.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

When larval states are generated along with ACQUIRE messages, we should
use the sequence to find the corresponding larval state when creating
states with ADD_SA or ALLOC_SPI.

If we don't do that, then it may take down an unrelated larval state
with the same parameters (think different TCP sessions).  This not only
leaves behind a larval state that shouldn't be there, it may also cause
another ACQUIRE message to be sent unnecessarily.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

include/net/pkt_act.h: In function 'tcf_hash_search':
include/net/pkt_act.h:218: warning: return makes integer from pointer without a cast

gcc-2.95 is stupid.
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

gem_tx() is always invoked with IRQs disabled.
Signed-off-by: David S. Miller <davem@davemloft.net>

into plexity.net:/home/dsaxena/src/linux-2.6-for-rmk