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

Make the actual use of hwif->data_phase.
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@elka.pw.edu.pl>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Use it for taskfile requests (only PIO/DMA for now) for storing
ide_task_t->data_phase of the active command.

Also add some missing task->data_phase assignments.
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@elka.pw.edu.pl>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@elka.pw.edu.pl>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Affected chipsets support LBA48 but not LBA48 DMA.
Just use DMA for area < 137GB and revert to PIO for > 137GB one.
Also disallow transfers > 256 sectors for better performance.
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@elka.pw.edu.pl>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

tmp_hwif should not be kmalloc()ed in ide_unregister().
If the kmalloc fails, there is no possible error recovery.
This patch makes tmp_hwif static, protected by ide_cfg_sem.
Signed-off-by: Jorn Engel <joern@wh.fh-wedel.de>
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@elka.pw.edu.pl>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

From: Alan Cox <alan@redhat.com>

Extracted from 2.6.8.1-ac1 by me.
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@elka.pw.edu.pl>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

From: Alan Cox <alan@redhat.com>

Extracted from 2.6.8.1-ac1, minor corrections by me.
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@elka.pw.edu.pl>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@elka.pw.edu.pl>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Using a label isn't sufficent anymore for determining the location of the
failed lock. Use __builtin_return_address instead.
Signed-off-by: Zwane Mwaikambo <zwane@fsmlabs.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Using a label isn't sufficent anymore for determining the location of the
failed lock. Use __builtin_return_address instead.
Signed-off-by: Zwane Mwaikambo <zwane@fsmlabs.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

The time interpolator scalability enhancement patch broke the time
interpolator integration for hpet on IA64.  Here is the fix.

The recent sungem NAPI change introduced a bug: dev_kfree_skb() is
called within a softirq context, thus triggers all sort of WARN_ON's
later on down the stack.

This patch changes it to dev_kfree_skb_any().

Acked by David Miller.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

This patch fixes boot on machines with a 970FX CPU, for PPC 32
bits kernels, please apply.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

The recent fbdev changes broke something quite fundamental on ppc, which
is the way offb works. It's a "fallback" driver that is to be used when
no other driver picked up the video, or when "forced" via the video=ofonly
command line option.

The recent changes completely broke that (which is a pretty important
behaviour on ppc since we still have some models regulary with video
cards that don't work properly with the kernel drivers, like some nVidias
or some older stuffs).

This patch moves offb to the end of the Makefile, so at least the behaviour
of taking over as a "fallback" is restored (the current kernel will have
offb take control of the framebuffer before any fbdev has a chance to do
it, which breaks everything).

Apparently, the entire support for "video=ofonly" was removed though,
that NEEDS to be restored in some way, though i'm not yet sure what is
the best path to that yet, I have to look more deeply at the new code.

 - create in_lock_functions() to match in_sched_functions(). Export it
   for use in oprofile.
 - use char __lock_text_start[] instead of long __lock_text_start when
   declaring linker symbols. Rusty fixed a number of these a while ago
   based on advice from rth.
 - Move __preempt_*_lock into kernel/spinlock.c and make it inline. This
   means locks are only one deep.
 - Make in_sched_functions() check in_lock_functions()
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Andries made a declaration static, and this clashed with Al who
changed it for byteorder annotation. Fix it all up.

That one was slightly messier than ext2 - several data structures had
been used both in little-endian (on-disk) and host-endian (in-core)
situations.  They had to be split, the rest was trivial annotations. 

Gets UDF sparse-clean with -Wbitwise.  Note that here we just annotate -
bug fixes from the original version of that patch had already been
merged.
Signed-off-by: Al Viro <viro@parcelfarce.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

AFFS partially annotated - some fields of on-disk structures are never
used and while I'm reasonably sure that they ought to be big-endian, I'd
rather leave them alone for now; annotating them won't change the amount
of noise since nothing in the kernel ever accesses them. 
Signed-off-by: Al Viro <viro@parcelfarce.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Signed-off-by: Al Viro <viro@parcelfarce.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Signed-off-by: Al Viro <viro@parcelfarce.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

This adds the types and annotates conversion functions.  I've converted
the ...p() versions to inlines; AFAICS, everything's still happy...
Signed-off-by: Al Viro <viro@parcelfarce.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

In highmem end_io handling, we need to iterate over the completed bio
from 0, not bio->bi_idx. If not we leak N-1 pages for any bio with N
pages where N > 1.
Signed-off-by: Jens Axboe <axboe@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

This also solved a problem noticed by Thomas Graf
in that RTATTR_MAX had become incorrect recently
due to IFLA_MAX increasing and thus becomming the
largest one.
Signed-off-by: David S. Miller <davem@davemloft.net>

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

into kernel.bkbits.net:/home/davem/net-2.6

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

We divide by the wrong MSS.
Signed-off-by: David S. Miller <davem@davemloft.net>

Unused outside of tcp_output.c
Signed-off-by: David S. Miller <davem@davemloft.net>

Looks like a bug in the cleanup patch :)

From: Andrew Morton <akpm@osdl.org>

Shows how many people are testing ext2.  Let's fix up that whitespace
also. 
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

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