openwrt/staging/blogic.git
18 years agoMerge master.kernel.org:/home/rmk/linux-2.6-mmc
Linus Torvalds [Fri, 17 Feb 2006 16:12:08 +0000 (08:12 -0800)]
Merge master.kernel.org:/home/rmk/linux-2.6-mmc

18 years ago[PATCH] x86_64: Always pass full number of nodes to NUMA hash computation
Andi Kleen [Thu, 16 Feb 2006 22:42:16 +0000 (23:42 +0100)]
[PATCH] x86_64: Always pass full number of nodes to NUMA hash computation

Previously the numa hash code would be confused by holes in the node space
and stop early. This is the first part of the fix for the non boot issue
with empty nodes on Opterons.

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] x86_64: Relax SRAT covers all memory check a bit
Andi Kleen [Thu, 16 Feb 2006 22:42:13 +0000 (23:42 +0100)]
[PATCH] x86_64: Relax SRAT covers all memory check a bit

Code was refusing good SRATs because about 12K got lost somewhere.
Allow less than 1MB of difference before rejecting it.

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] x86_64: Resolve the RIP of an early exception using kallsyms
Andi Kleen [Thu, 16 Feb 2006 22:42:10 +0000 (23:42 +0100)]
[PATCH] x86_64: Resolve the RIP of an early exception using kallsyms

But do it after everything else to risk less from recursive
crashes.

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] x86_64: Disable tsc when apicpmtimer is active
Andi Kleen [Thu, 16 Feb 2006 22:42:07 +0000 (23:42 +0100)]
[PATCH] x86_64: Disable tsc when apicpmtimer is active

Otherwise it has no effect anyways.

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] x86_64: Don't enable ATI apicmaintimer workaround when the machine has C2...
Andi Kleen [Thu, 16 Feb 2006 22:42:04 +0000 (23:42 +0100)]
[PATCH] x86_64: Don't enable ATI apicmaintimer workaround when the machine has C2 or C3

Many laptops have problems with ticking the local APIC timer in C2/C3.
The code added earlier to use it by default on ATI didn't really work
for them. Don't enable it when the system supports C2/C3.

This doesn't fix the problem fully, but at least it's not worse than before.

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] x86_64: Don't call do_exit with interrupts disabled after IRET exception
Andi Kleen [Thu, 16 Feb 2006 22:42:01 +0000 (23:42 +0100)]
[PATCH] x86_64: Don't call do_exit with interrupts disabled after IRET exception

This caused a sigreturn with bad argument on a preemptible kernel
to complain with

Debug: sleeping function called from invalid context at /home/lsrc/quilt/linux/include/linux/rwsem.h:43
in_atomic():0, irqs_disabled():1

Call Trace: {__might_sleep+190} {profile_task_exit+21}
       {__do_exit+34} {do_wait+0}

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] x86_64: Add boot option to disable randomized mappings and cleanup
Andi Kleen [Thu, 16 Feb 2006 22:41:58 +0000 (23:41 +0100)]
[PATCH] x86_64: Add boot option to disable randomized mappings and cleanup

AMD SimNow!'s JIT doesn't like them at all in the guest. For distribution
installation it's easiest if it's a boot time option.

Also I moved the variable to a more appropiate place and make
it independent from sysctl

And marked __read_mostly which it is.

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] x86_64: make touch_nmi_watchdog() not touch impossible cpus' private data
Jan Beulich [Thu, 16 Feb 2006 22:41:55 +0000 (23:41 +0100)]
[PATCH] x86_64: make touch_nmi_watchdog() not touch impossible cpus' private data

Along with that, also suppress the memory touching altogether when the
watchdog is not running, to eliminate needless crosstalk. Plus ad a call
to it to make things consistent (one could also consider removing the call
in enable_timer_nmi_watchdog()).

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] x86_64: Update defconfig
Andi Kleen [Thu, 16 Feb 2006 22:41:52 +0000 (23:41 +0100)]
[PATCH] x86_64: Update defconfig

... and enable 1394 by default.

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years agoMerge branch 'upstream-linus' of git://oss.oracle.com/home/sourcebo/git/ocfs2
Linus Torvalds [Thu, 16 Feb 2006 22:04:08 +0000 (14:04 -0800)]
Merge branch 'upstream-linus' of git://oss.oracle.com/home/sourcebo/git/ocfs2

18 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/jejb/scsi-rc-fixes-2.6
Linus Torvalds [Thu, 16 Feb 2006 20:47:44 +0000 (12:47 -0800)]
Merge /linux/kernel/git/jejb/scsi-rc-fixes-2.6

18 years ago[PATCH] ocfs2: detach from heartbeat events before freeing mle
Kurt Hackel [Tue, 14 Feb 2006 19:45:21 +0000 (11:45 -0800)]
[PATCH] ocfs2: detach from heartbeat events before freeing mle

Signed-off-by: Kurt Hackel <Kurt.Hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
18 years ago[PATCH] ocfs2: only checkpoint journal when asked to
Mark Fasheh [Thu, 9 Feb 2006 21:23:39 +0000 (13:23 -0800)]
[PATCH] ocfs2: only checkpoint journal when asked to

Disable automatic checkpointing of the journal - this is a relic from older
ocfs2 days. Worth quite a bit of performance on longer running single node
tests.

Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
18 years ago[PATCH] ocfs2: manually grant remote recovery lock
Kurt Hackel [Thu, 19 Jan 2006 01:07:47 +0000 (17:07 -0800)]
[PATCH] ocfs2: manually grant remote recovery lock

* fix a hang in recovery that occurred in dlmlock_remote.  the $RECOVERY
  lock was never moved to the granted queue even after getting DLM_NORMAL
  back from the master node.

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
18 years ago[PATCH] ocfs2: add dlm_wait_for_node_death
Kurt Hackel [Thu, 19 Jan 2006 01:05:38 +0000 (17:05 -0800)]
[PATCH] ocfs2: add dlm_wait_for_node_death

* add dlm_wait_for_node_death function to be used after receiving a network
  error.  this will wait for the given timeout to allow the heartbeat
  callbacks to update the domain map.  without this, some paths may spin
  and consume enough cpu that the heartbeat gets starved and never updates.

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
18 years ago[PATCH] ocfs2: fix release of ast never reserved
Kurt Hackel [Thu, 19 Jan 2006 01:02:56 +0000 (17:02 -0800)]
[PATCH] ocfs2: fix release of ast never reserved

* fix a bug in dlm_convert_lock_handler where dlm_lockres_release_ast was
  being called even if no ast was ever reserved

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
18 years ago[PATCH] ocfs2: recheck recovery state after getting lock
Kurt Hackel [Thu, 19 Jan 2006 01:01:25 +0000 (17:01 -0800)]
[PATCH] ocfs2: recheck recovery state after getting lock

* after successfully taking the $RECOVERY lock in EX mode, recheck to make
  sure that recovery has not already begun or completed on another node

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
18 years ago[MMC] mmci: allow small data transfers
Russell King [Thu, 16 Feb 2006 16:48:31 +0000 (16:48 +0000)]
[MMC] mmci: allow small data transfers

If a data transfer is small (less than a FIFO size) we would
hang waiting for the data to be read due to the PIO interrupt
not occuring.  We allowed for this in our PIO interrupt handler,
but not when setting up a data transfer.

Apply the "fix" when setting up a data transfer as well.

Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
18 years ago[PATCH] Fix over-zealous tag clearing in radix_tree_delete
NeilBrown [Thu, 16 Feb 2006 03:43:01 +0000 (14:43 +1100)]
[PATCH] Fix over-zealous tag clearing in radix_tree_delete

If a tag is set for a node being deleted from a radix_tree, then that
tag gets cleared from the parent of the node, even if it is set for some
siblings of the node begin deleted.

This patch changes the logic to include a test for any_tag_set similar
to the logic a little futher down.  Care is taken to ensure that
'nr_cleared_tags' remains equals to the number of entries in the 'tags'
array which are set to '0' (which means that this tag is not set in the
tree below pathp->node, and should be cleared at pathp->node and
possibly above.

[ Nick says: "Linus FYI, I was able to modify the radix tree test
  harness to catch the bug and can no longer trigger it after the fix.
  Resulting code passes all other harness tests as well of course." ]

Signed-off-by: Neil Brown <neilb@suse.de>
Acked-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/dtor/input
Linus Torvalds [Thu, 16 Feb 2006 04:00:48 +0000 (20:00 -0800)]
Merge /pub/scm/linux/kernel/git/dtor/input

18 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
Linus Torvalds [Thu, 16 Feb 2006 03:56:33 +0000 (19:56 -0800)]
Merge /pub/scm/linux/kernel/git/davem/net-2.6

18 years ago[PATCH] mmconfig: add kernel parameter documentation
Bjorn Helgaas [Wed, 15 Feb 2006 23:17:43 +0000 (15:17 -0800)]
[PATCH] mmconfig: add kernel parameter documentation

Mention the "pci=nommconf" option in kernel-parameters.txt.

Signed-off-by: Bjorn Helgaas <bjorn.helgaas@hp.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] swsusp: nuke noisy message
Andrew Morton [Wed, 15 Feb 2006 23:17:42 +0000 (15:17 -0800)]
[PATCH] swsusp: nuke noisy message

I get about 88 squillion of these when suspending an old ad450nx server.

Cc: Pavel Roskin <proski@gnu.org>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] x86_64: early initialization of cpu_to_node
Daniel Yeisley [Wed, 15 Feb 2006 23:17:41 +0000 (15:17 -0800)]
[PATCH] x86_64: early initialization of cpu_to_node

The early initialization of cpu_to_node code as it is now only updates the
cpu_to_node array, and does not update cpu_pda()->nodemember.  This will
cause numa_node_id() to return 0 on systems where CPU 0 is not on Node 0.
This leads to a kernel panic in slab.c.

I've tested the patch below on a 16 processor x86_64 ES7000-600 server, and
no longer see the panic I saw with the original 2.6.16-rc3.

Signed-off-by: Dan Yeisley <dan.yeisley@unisys.com>
Acked-by: Andi Kleen <ak@muc.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] hrtimer: fix multiple macro argument expansion
Roman Zippel [Wed, 15 Feb 2006 23:17:40 +0000 (15:17 -0800)]
[PATCH] hrtimer: fix multiple macro argument expansion

For two macros the arguments were expanded twice, change them to inline
functions to avoid it.

Signed-off-by: Roman Zippel <zippel@linux-m68k.org>
Acked-by: Ingo Molnar <mingo@elte.hu>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] add asm-generic/mman.h
Michael S. Tsirkin [Wed, 15 Feb 2006 23:17:39 +0000 (15:17 -0800)]
[PATCH] add asm-generic/mman.h

Make new MADV_REMOVE, MADV_DONTFORK, MADV_DOFORK consistent across all
arches.  The idea is to make it possible to use them portably even before
distros include them in libc headers.

Move common flags to asm-generic/mman.h

Signed-off-by: Michael S. Tsirkin <mst@mellanox.co.il>
Cc: Roland Dreier <rolandd@cisco.com>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] cpuset: oops in exit on null cpuset fix
Paul Jackson [Wed, 15 Feb 2006 23:17:38 +0000 (15:17 -0800)]
[PATCH] cpuset: oops in exit on null cpuset fix

Fix a latent bug in cpuset_exit() handling.  If a task tried to allocate
memory after calling cpuset_exit(), it oops'd in
cpuset_update_task_memory_state() on a NULL cpuset pointer.

So set the exiting tasks cpuset to the root cpuset instead of to NULL.

A distro kernel hit this with an added kernel package that had just such a
hook (allocating memory) in the exit code path.

Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] ide: touch softlockup detector during pio
Andrew Morton [Wed, 15 Feb 2006 23:17:37 +0000 (15:17 -0800)]
[PATCH] ide: touch softlockup detector during pio

We're getting some softlockup false positives during heavy PIO operations.  So
poke the lockup detector.

Cc: Bartlomiej Zolnierkiewicz <B.Zolnierkiewicz@elka.pw.edu.pl>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] kbuild: revert "fix make -jN with multiple targets with O=..."
Benjamin LaHaise [Wed, 15 Feb 2006 23:17:35 +0000 (15:17 -0800)]
[PATCH] kbuild: revert "fix make -jN with multiple targets with O=..."

Commit 296e0855b0f9a4ec9be17106ac541745a55b2ce1:

    "kbuild: fix make -jN with multiple targets with O=..."

causes a ~95% increase in build time for the kernel.  Before: 4m21s
after: 8m1.403s.  Can we revert this until another approach is found?

Cc: Sam Ravnborg <sam@ravnborg.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] neofb: avoid resetting display config on unblank (v2)
Christian Trefzer [Wed, 15 Feb 2006 23:17:34 +0000 (15:17 -0800)]
[PATCH] neofb: avoid resetting display config on unblank (v2)

There were two mistakes in the register-read-on-(un)blank approach.

- First, without proper register (un)locking the value read back will always
  be zero, and this is what I missed entirely until just now.  Due to this,
  the logic could not be verified at all and I tried some bogus checks which
  are completely stupid.

- Second, the LCD status bit will always be set to zero when the backlight
  has been turned off.  Reading the value back during unblank will disable the
  LCD unconditionally, regardless of the state it is supposed to be in, since
  we set it to zero beforehand.

So this is what we do now:

- create a new variable in struct neofb_par, and use that to determine
  whether to read back registers (initialized to true)

- before actually blanking the screen, read back the register to sense any
  possible change made through Fn key combo

- use proper neoUnlock() / neoLock() to actually read something

- every call to neofb_blank() determines if we read back next time: blanking
  disables readback, unblanking (FB_BLANK_UNBLANK) enables it

This should give us a nice and clean state machine.  Has been thoroughly
tested on a Dell Latitude CPiA / NM220 Chip docked to a C/Dock2 with attached
CRT in all possible combinations of LCD/CRT on/off.  I changed the config via
Fn key, let the console blank, unblanked by keypress - works flawlessly.

Signed-off-by: Christian Trefzer <ctrefzer@gmx.de>
Cc: "Antonino A. Daplas" <adaplas@pol.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[NETFILTER]: nf_conntrack: Fix TCP/UDP HW checksum handling for IPv6 packet
Yasuyuki Kozakai [Wed, 15 Feb 2006 23:25:18 +0000 (15:25 -0800)]
[NETFILTER]: nf_conntrack: Fix TCP/UDP HW checksum handling for IPv6 packet

If skb->ip_summed is CHECKSUM_HW here, skb->csum includes checksum
of actual IPv6 header and extension headers. Then such excess
checksum must be subtruct when nf_conntrack calculates TCP/UDP checksum
with pseudo IPv6 header. Spotted by Ben Skeggs.

Signed-off-by: Yasuyuki Kozakai <yasuyuki.kozakai@toshiba.co.jp>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[NETFILTER]: nf_conntrack: attach conntrack to locally generated ICMPv6 error
Yasuyuki Kozakai [Wed, 15 Feb 2006 23:24:15 +0000 (15:24 -0800)]
[NETFILTER]: nf_conntrack: attach conntrack to locally generated ICMPv6 error

Locally generated ICMPv6 errors should be associated with the conntrack
of the original packet. Since the conntrack entry may not be in the hash
tables (for the first packet), it must be manually attached.

Signed-off-by: Yasuyuki Kozakai <yasuyuki.kozakai@toshiba.co.jp>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[NETFILTER]: nf_conntrack: attach conntrack to TCP RST generated by ip6t_REJECT
Yasuyuki Kozakai [Wed, 15 Feb 2006 23:23:28 +0000 (15:23 -0800)]
[NETFILTER]: nf_conntrack: attach conntrack to TCP RST generated by ip6t_REJECT

TCP RSTs generated by the REJECT target should be associated with the
conntrack of the original TCP packet. Since the conntrack entry is
usually not is the hash tables, it must be manually attached.

Signed-off-by: Yasuyuki Kozakai <yasuyuki.kozakai@toshiba.co.jp>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[NETFILTER]: nf_conntrack: move registration of __nf_ct_attach
Yasuyuki Kozakai [Wed, 15 Feb 2006 23:22:21 +0000 (15:22 -0800)]
[NETFILTER]: nf_conntrack: move registration of __nf_ct_attach

Move registration of __nf_ct_attach to nf_conntrack_core to make it usable
for IPv6 connection tracking as well.

Signed-off-by: Yasuyuki Kozakai <yasuyuki.kozakai@toshiba.co.jp>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[NETFILTER]: x_tables: fix dependencies of conntrack related modules
Yasuyuki Kozakai [Wed, 15 Feb 2006 23:21:31 +0000 (15:21 -0800)]
[NETFILTER]: x_tables: fix dependencies of conntrack related modules

NF_CONNTRACK_MARK is bool and depends on NF_CONNTRACK which is
tristate.  If a variable depends on NF_CONNTRACK_MARK and doesn't take
care about NF_CONNTRACK, it can be y even if NF_CONNTRACK isn't y.
NF_CT_ACCT have same issue, too.

Signed-off-by: Yasuyuki Kozakai <yasuyuki.kozakai@toshiba.co.jp>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[NETFILTER]: Don't invoke okfn in CONFIG_NETFILTER=n variant of nf_hook()
Patrick McHardy [Wed, 15 Feb 2006 23:18:19 +0000 (15:18 -0800)]
[NETFILTER]: Don't invoke okfn in CONFIG_NETFILTER=n variant of nf_hook()

nf_hook() is supposed to call the netfilter hook and return control of the
packet back to the caller in case it may pass, the okfn is only used for
queueing.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[XFRM]: Fix SNAT-related crash in xfrm4_output_finish
Patrick McHardy [Wed, 15 Feb 2006 23:10:22 +0000 (15:10 -0800)]
[XFRM]: Fix SNAT-related crash in xfrm4_output_finish

When a packet matching an IPsec policy is SNATed so it doesn't match any
policy anymore it looses its xfrm bundle, which makes xfrm4_output_finish
crash because of a NULL pointer dereference.

This patch directs these packets to the original output path instead. Since
the packets have already passed the POST_ROUTING hook, but need to start at
the beginning of the original output path which includes another
POST_ROUTING invocation, a flag is added to the IPCB to indicate that the
packet was rerouted and doesn't need to pass the POST_ROUTING hook again.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[PATCH] fix zap_thread's ptrace related problems
Oleg Nesterov [Wed, 15 Feb 2006 19:50:10 +0000 (22:50 +0300)]
[PATCH] fix zap_thread's ptrace related problems

1. The tracee can go from ptrace_stop() to do_signal_stop()
   after __ptrace_unlink(p).

2. It is unsafe to __ptrace_unlink(p) while p->parent may wait
   for tasklist_lock in ptrace_detach().

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Roland McGrath <roland@redhat.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] fix kill_proc_info() vs fork() theoretical race
Oleg Nesterov [Wed, 15 Feb 2006 19:13:26 +0000 (22:13 +0300)]
[PATCH] fix kill_proc_info() vs fork() theoretical race

copy_process:

attach_pid(p, PIDTYPE_PID, p->pid);
attach_pid(p, PIDTYPE_TGID, p->tgid);

What if kill_proc_info(p->pid) happens in between?

copy_process() holds current->sighand.siglock, so we are safe
in CLONE_THREAD case, because current->sighand == p->sighand.

Otherwise, p->sighand is unlocked, the new process is already
visible to the find_task_by_pid(), but have a copy of parent's
'struct pid' in ->pids[PIDTYPE_TGID].

This means that __group_complete_signal() may hang while doing

do ... while (next_thread() != p)

We can solve this problem if we reverse these 2 attach_pid()s:

attach_pid() does wmb()

group_send_sig_info() calls spin_lock(), which
provides a read barrier. // Yes ?

I don't think we can hit this race in practice, but still.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Roland McGrath <roland@redhat.com>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] fix kill_proc_info() vs CLONE_THREAD race
Oleg Nesterov [Wed, 15 Feb 2006 19:13:24 +0000 (22:13 +0300)]
[PATCH] fix kill_proc_info() vs CLONE_THREAD race

There is a window after copy_process() unlocks ->sighand.siglock
and before it adds the new thread to the thread list.

In that window __group_complete_signal(SIGKILL) will not see the
new thread yet, so this thread will start running while the whole
thread group was supposed to exit.

I beleive we have another good reason to place attach_pid(PID/TGID)
under ->sighand.siglock. We can do the same for

release_task()->__unhash_process()

de_thread()->switch_exec_pids()

After that we don't need tasklist_lock to iterate over the thread
list, and we can simplify things, see for example do_sigaction()
or sys_times().

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Roland McGrath <roland@redhat.com>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
Linus Torvalds [Wed, 15 Feb 2006 16:49:23 +0000 (08:49 -0800)]
Merge /pub/scm/linux/kernel/git/davem/net-2.6

18 years ago[BRIDGE]: Fix deadlock in br_stp_disable_bridge
Adrian Drzewiecki [Wed, 15 Feb 2006 09:47:48 +0000 (01:47 -0800)]
[BRIDGE]: Fix deadlock in br_stp_disable_bridge

Looks like somebody forgot to use the _bh spin_lock variant. We ran into a
deadlock where br->hello_timer expired while br_stp_disable_br() walked
br->port_list.

Signed-off-by: Adrian Drzewiecki <z@drze.net>
Signed-off-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[NETFILTER]: Fix xfrm lookup after SNAT
Patrick McHardy [Wed, 15 Feb 2006 09:34:23 +0000 (01:34 -0800)]
[NETFILTER]: Fix xfrm lookup after SNAT

To find out if a packet needs to be handled by IPsec after SNAT, packets
are currently rerouted in POST_ROUTING and a new xfrm lookup is done. This
breaks SNAT of non-unicast packets to non-local addresses because the
packet is routed as incoming packet and no neighbour entry is bound to the
dst_entry. In general, it seems to be a bad idea to replace the dst_entry
after the packet was already sent to the output routine because its state
might not match what's expected.

This patch changes the xfrm lookup in POST_ROUTING to re-use the original
dst_entry without routing the packet again. This means no policy routing
can be used for transport mode transforms (which keep the original route)
when packets are SNATed to match the policy, but it looks like the best
we can do for now.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years agoInput: kill remnants of 98kbd{,-io} and 98spkr
Arthur Othieno [Wed, 15 Feb 2006 05:49:48 +0000 (00:49 -0500)]
Input: kill remnants of 98kbd{,-io} and 98spkr

98kbd{,-io} and 98spkr all went out with PC98 subarch.  Remove stale Makefile
entries that remained.

Signed-off-by: Arthur Othieno <apgo@patchbomb.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Dmitry Torokhov <dtor@mail.ru>
18 years agoInput: ads7846 - assorted updates
David Brownell [Wed, 15 Feb 2006 05:49:35 +0000 (00:49 -0500)]
Input: ads7846 - assorted updates

This updates the ads7846 touchscreen driver:
  - to allow faster clocking (this driver doesn't push sample rates);
  - bugfixes the conversion of spi_transfer to lists;
  - some dma-unsafe command buffers are fixed.

Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Dmitry Torokhov <dtor@mail.ru>
18 years agoInput: ads7846 - convert to to dynamic input_dev allocation
Dmitry Torokhov [Wed, 15 Feb 2006 05:49:22 +0000 (00:49 -0500)]
Input: ads7846 - convert to to dynamic input_dev allocation

Signed-off-by: Dmitry Torokhov <dtor@mail.ru>
18 years agoInput: trackpoint - enable devices connected to external port
Dmitry Torokhov [Wed, 15 Feb 2006 05:49:09 +0000 (00:49 -0500)]
Input: trackpoint - enable devices connected to external port

Signed-off-by: Dmitry Torokhov <dtor@mail.ru>
18 years agoInput: logips2pp - add new signature (99)
Meelis Roos [Wed, 15 Feb 2006 05:48:58 +0000 (00:48 -0500)]
Input: logips2pp - add new signature (99)

Add Logitech mouse type 99 (Premium Optical Wheel Mouse, model M-BT58,
plain 3 buttons + wheel) to cure the following message: logips2pp: Detected
unknown logitech mouse model 99

Signed-off-by: Meelis Roos <mroos@linux.ee>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Dmitry Torokhov <dtor@mail.ru>
18 years agoInput: ixp4xx-beeper - fix compile error
Alessandro Zummo [Wed, 15 Feb 2006 05:48:40 +0000 (00:48 -0500)]
Input: ixp4xx-beeper - fix compile error

Signed-off-by: Alessandro Zummo <a.zummo@towertech.it>
Signed-off-by: Dmitry Torokhov <dtor@mail.ru>
18 years ago[PATCH] CIFS: fix cifs_user_read oops when null SMB response on forcedirectio mount
Steve French [Wed, 15 Feb 2006 04:30:52 +0000 (22:30 -0600)]
[PATCH] CIFS: fix cifs_user_read oops when null SMB response on forcedirectio mount

This patch fixes an oops reported by Adrian Bunk in cifs_user_read when a null
read response is returned on a forcedirectio mount.

Signed-off-by: Dave Kleikamp <shaggy@austin.ibm.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] neofb: avoid resetting display config on unblank
Christian Trefzer [Tue, 14 Feb 2006 21:53:26 +0000 (13:53 -0800)]
[PATCH] neofb: avoid resetting display config on unblank

Fix issues with the NeoMagic framebuffer driver.

It nicely complements my previous fix already in linus' tree.  The only
thing missing now is that the external CRT will not be activated at neofb
init when external-only is selected, either by register read or
module/kernel parameter.

Testing was done on a Dell Latitude CPi-A/NM2200 chip.

Previous behaviour:
- before booting linux, set the preferred display config X via FN+F8

- boot linux, neofb stores the register values in a private
  variable

- change the display config to Y via keystroke

- leave the machine in peace until display is blanked

- touching any key will result in display config X being restored

- booting up, the BIOS will acknowledge config Y, though...

Current behaviour:
At the time of unblanking, config Y is honoured because we now read back
register contents instead of just overwriting them with outdated values.

Signed-off by: Christian Trefzer <ctrefzer@gmx.de>
Cc: "Antonino A. Daplas" <adaplas@pol.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] x86: gitignore some autogenerated files for i386
Thomas Meyer [Tue, 14 Feb 2006 21:53:21 +0000 (13:53 -0800)]
[PATCH] x86: gitignore some autogenerated files for i386

Add some more gitignore files for i386 architecture.  This files are
created during the build process of a i386 kernel.

Signed-off-by: Thomas Meyer <thomas@m3y3r.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] x86: document sysenter path
Albert D. Cahalan [Tue, 14 Feb 2006 21:53:20 +0000 (13:53 -0800)]
[PATCH] x86: document sysenter path

This path isn't obvious.  It looks as if the kernel will be taking three
args from the user stack, but it only takes one from there.

Signed-off-by: Albert Cahalan <acahalan@gmail.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] FRV: Use virtual interrupt disablement
David Howells [Tue, 14 Feb 2006 21:53:20 +0000 (13:53 -0800)]
[PATCH] FRV: Use virtual interrupt disablement

Make the FRV arch use virtual interrupt disablement because accesses to the
processor status register (PSR) are relatively slow and because we will
soon have the need to deal with multiple interrupt controls at the same
time (separate h/w and inter-core interrupts).

The way this is done is to dedicate one of the four integer condition code
registers (ICC2) to maintaining a virtual interrupt disablement state
whilst inside the kernel.  This uses the ICC2.Z flag (Zero) to indicate
whether the interrupts are virtually disabled and the ICC2.C flag (Carry)
to indicate whether the interrupts are physically disabled.

ICC2.Z is set to indicate interrupts are virtually disabled.  ICC2.C is set
to indicate interrupts are physically enabled.  Under normal running
conditions Z==0 and C==1.

Disabling interrupts with local_irq_disable() doesn't then actually
physically disable interrupts - it merely sets ICC2.Z to 1.  Should an
interrupt then happen, the exception prologue will note ICC2.Z is set and
branch out of line using one instruction (an unlikely BEQ).  Here it will
physically disable interrupts and clear ICC2.C.

When it comes time to enable interrupts (local_irq_enable()), this simply
clears the ICC2.Z flag and invokes a trap #2 if both Z and C flags are
clear (the HI integer condition).  This can be done with the TIHI
conditional trap instruction.

The trap then physically reenables interrupts and sets ICC2.C again.  Upon
returning the interrupt will be taken as interrupts will then be enabled.
Note that whilst processing the trap, the whole exceptions system is
disabled, and so an interrupt can't happen till it returns.

If no pending interrupt had happened, ICC2.C would still be set, the HI
condition would not be fulfilled, and no trap will happen.

Saving interrupts (local_irq_save) is simply a matter of pulling the ICC2.Z
flag out of the CCR register, shifting it down and masking it off.  This
gives a result of 0 if interrupts were enabled and 1 if they weren't.

Restoring interrupts (local_irq_restore) is then a matter of taking the
saved value mentioned previously and XOR'ing it against 1.  If it was one,
the result will be zero, and if it was zero the result will be non-zero.
This result is then used to affect the ICC2.Z flag directly (it is a
condition code flag after all).  An XOR instruction does not affect the
Carry flag, and so that bit of state is unchanged.  The two flags can then
be sampled to see if they're both zero using the trap (TIHI) as for the
unconditional reenablement (local_irq_enable).

This patch also:

 (1) Modifies the debugging stub (break.S) to handle single-stepping crossing
     into the trap #2 handler and into virtually disabled interrupts.

 (2) Removes superseded fixup pointers from the second instructions in the trap
     tables (there's no a separate fixup table for this).

 (3) Declares the trap #3 vector for use in .org directives in the trap table.

 (4) Moves irq_enter() and irq_exit() in do_IRQ() to avoid problems with
     virtual interrupt handling, and removes the duplicate code that has now
     been folded into irq_exit() (softirq and preemption handling).

 (5) Tells the compiler in the arch Makefile that ICC2 is now reserved.

 (6) Documents the in-kernel ABI, including the virtual interrupts.

 (7) Renames the old irq management functions to different names.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] FRV: Miscellaneous fixes
David Howells [Tue, 14 Feb 2006 21:53:18 +0000 (13:53 -0800)]
[PATCH] FRV: Miscellaneous fixes

Make various alterations and fixes to the FRV arch:

 (1) Resyncs the FRV system call collection with the i386 arch.

 (2) Discards __iounmap() as it's not used.

 (3) Fixes the use of the SWAP/SWAPI instruction to get the arguments the right
     way around in atomic.h, and also to get the asm constraints correct.

 (4) Moves copy_to/from_user_page() to asm/cacheflush.h to be consistent with
     other archs.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] hrtimer: round up relative start time on low-res arches
Ingo Molnar [Tue, 14 Feb 2006 21:53:15 +0000 (13:53 -0800)]
[PATCH] hrtimer: round up relative start time on low-res arches

CONFIG_TIME_LOW_RES is a temporary way for architectures to signal that
they simply return xtime in do_gettimeoffset().  In this corner-case we
want to round up by resolution when starting a relative timer, to avoid
short timeouts.  This will go away with the GTOD framework.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cc: Roman Zippel <zippel@linux-m68k.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] s390: fix __delay implementation
Heiko Carstens [Tue, 14 Feb 2006 21:53:14 +0000 (13:53 -0800)]
[PATCH] s390: fix __delay implementation

Fix __delay implementation.  Called with an argument "1" or "0" it would
loop nearly forever (since (1/2)-1 = 0xffffffff).

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] fix a typo in the CPU_H8300H dependencies
Adrian Bunk [Tue, 14 Feb 2006 21:53:10 +0000 (13:53 -0800)]
[PATCH] fix a typo in the CPU_H8300H dependencies

Jean-Luc Leger <reiga@dspnet.fr.eu.org> found this obvious typo.

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] sched: revert "filter affine wakeups"
Chen, Kenneth W [Tue, 14 Feb 2006 21:53:10 +0000 (13:53 -0800)]
[PATCH] sched: revert "filter affine wakeups"

Revert commit d7102e95b7b9c00277562c29aad421d2d521c5f6:

    [PATCH] sched: filter affine wakeups

Apparently caused more than 10% performance regression for aim7 benchmark.
The setup in use is 16-cpu HP rx8620, 64Gb of memory and 12 MSA1000s with 144
disks.  Each disk is 72Gb with a single ext3 filesystem (courtesy of HP, who
supplied benchmark results).

The problem is, for aim7, the wake-up pattern is random, but it still needs
load balancing action in the wake-up path to achieve best performance.  With
the above commit, lack of load balancing hurts that workload.

However, for workloads like database transaction processing, the requirement
is exactly opposite.  In the wake up path, best performance is achieved with
absolutely zero load balancing.  We simply wake up the process on the CPU that
it was previously run.  Worst performance is obtained when we do load
balancing at wake up.

There isn't an easy way to auto detect the workload characteristics.  Ingo's
earlier patch that detects idle CPU and decide whether to load balance or not
doesn't perform with aim7 either since all CPUs are busy (it causes even
bigger perf.  regression).

Revert commit d7102e95b7b9c00277562c29aad421d2d521c5f6, which causes more
than 10% performance regression with aim7.

Signed-off-by: Ken Chen <kenneth.w.chen@intel.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] madvise MADV_DONTFORK/MADV_DOFORK
Michael S. Tsirkin [Tue, 14 Feb 2006 21:53:08 +0000 (13:53 -0800)]
[PATCH] madvise MADV_DONTFORK/MADV_DOFORK

Currently, copy-on-write may change the physical address of a page even if the
user requested that the page is pinned in memory (either by mlock or by
get_user_pages).  This happens if the process forks meanwhile, and the parent
writes to that page.  As a result, the page is orphaned: in case of
get_user_pages, the application will never see any data hardware DMA's into
this page after the COW.  In case of mlock'd memory, the parent is not getting
the realtime/security benefits of mlock.

In particular, this affects the Infiniband modules which do DMA from and into
user pages all the time.

This patch adds madvise options to control whether memory range is inherited
across fork.  Useful e.g.  for when hardware is doing DMA from/into these
pages.  Could also be useful to an application wanting to speed up its forks
by cutting large areas out of consideration.

Signed-off-by: Michael S. Tsirkin <mst@mellanox.co.il>
Acked-by: Hugh Dickins <hugh@veritas.com>
Cc: Michael Kerrisk <mtk-manpages@gmx.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] kprobes: Update Documentation/kprobes.txt
Jim Keniston [Tue, 14 Feb 2006 21:53:06 +0000 (13:53 -0800)]
[PATCH] kprobes: Update Documentation/kprobes.txt

Update Documentation/kprobes.txt to reflect Kprobes enhancements and other
recent developments.

Acked-by: Ananth Mavinakayanahalli <mananth@in.ibm.com>
Signed-off-by: Jim Keniston <jkenisto@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Fix NULL pointer dereference in isdn_tty_at_cout
Karsten Keil [Tue, 14 Feb 2006 21:53:06 +0000 (13:53 -0800)]
[PATCH] Fix NULL pointer dereference in isdn_tty_at_cout

The changes in the tty related code introduced wrong parenthesis in a if
condition in the isdn_tty_at_cout function.  This caused access to index -1
in the dev->drv[] array.  This patch change it back to the correct
condition from the previous versions.

Signed-off-by: Karsten Keil <kkeil@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] fix x86 topology export in sysfs for subarchitectures
James Bottomley [Tue, 14 Feb 2006 21:53:05 +0000 (13:53 -0800)]
[PATCH] fix x86 topology export in sysfs for subarchitectures

The correct way to export hyperthreading based functions is to predicate
them on CONFIG_X86_HT.  Without this, the topology exporting patch breaks
the build on all non-PC x86 subarchitectures.

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] NLM: Fix the NLM_GRANTED callback checks
Trond Myklebust [Tue, 14 Feb 2006 21:53:04 +0000 (13:53 -0800)]
[PATCH] NLM: Fix the NLM_GRANTED callback checks

If 2 threads attached to the same process are blocking on different locks on
different files (maybe even on different servers) but have the same lock
arguments (i.e.  same offset+length - actually quite common, since most
processes try to lock the entire file) then the first GRANTED call that wakes
one up will also wake the other.

Currently when the NLM_GRANTED callback comes in, lockd walks the list of
blocked locks in search of a match to the lock that the NLM server has
granted.  Although it checks the lock pid, start and end, it fails to check
the filehandle and the server address.

By checking the filehandle and server IP address, we ensure that this only
happens if the locks truly are referencing the same file.

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] jbd: revert checkpoint list changes
Mark Fasheh [Tue, 14 Feb 2006 21:53:03 +0000 (13:53 -0800)]
[PATCH] jbd: revert checkpoint list changes

This patch reverts commit f93ea411b73594f7d144855fd34278bcf34a9afc:
  [PATCH] jbd: split checkpoint lists

This broke journal_flush() for OCFS2, which is its method of being sure
that metadata is sent to disk for another node.

And two related commits 8d3c7fce2d20ecc3264c8d8c91ae3beacdeaed1b and
43c3e6f5abdf6acac9b90c86bf03f995bf7d3d92 with the subjects:
  [PATCH] jbd: log_do_checkpoint fix
  [PATCH] jbd: remove_transaction fix

These seem to be incremental bugfixes on the original patch and as such are
no longer needed.

Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
Cc: Jan Kara <jack@ucw.cz>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] HPET: handle multiple ACPI EXTENDED_IRQ resources
Bjorn Helgaas [Tue, 14 Feb 2006 21:53:02 +0000 (13:53 -0800)]
[PATCH] HPET: handle multiple ACPI EXTENDED_IRQ resources

When the _CRS for a single HPET contains multiple EXTENDED_IRQ resources,
we overwrote hdp->hd_nirqs every time we found one.

So the driver worked when all the IRQs were described in a single
EXTENDED_IRQ resource, but failed when multiple resources were used.
(Strictly speaking, I think the latter is actually more correct, but both
styles have been used.)

Someday we should remove all the ACPI stuff from hpet.c and use PNP driver
registration instead.  But currently PNP_MAX_IRQ is 2, and HPETs often have
more IRQs.  Hint, hint, Adam :-)

Signed-off-by: Bjorn Helgaas <bjorn.helgaas@hp.com>
Acked-by: Bob Picco <robert.picco@hp.com>
Cc: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Cc: Len Brown <len.brown@intel.com>
Cc: Adam Belay <ambx1@neo.rr.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] tty reference count fix
Paul Fulghum [Tue, 14 Feb 2006 21:53:00 +0000 (13:53 -0800)]
[PATCH] tty reference count fix

Fix hole where tty structure can be released when reference count is non
zero.  Existing code can sleep without tty_sem protection between deciding
to release the tty structure (setting local variables tty_closing and
otty_closing) and setting TTY_CLOSING to prevent further opens.  An open
can occur during this interval causing release_dev() to free the tty
structure while it is still referenced.

This should fix bugzilla.kernel.org [Bug 6041] New: Unable to handle kernel
paging request

In Bug 6041, tty_open() oopes on accessing the tty structure it has
successfully claimed.  Bug was on SMP machine with the same tty being
opened and closed by multiple processes, and DEBUG_PAGEALLOC enabled.

Signed-off-by: Paul Fulghum <paulkf@microgate.com>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: Jesper Juhl <jesper.juhl@gmail.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] compound page: no access_process_vm check
Hugh Dickins [Tue, 14 Feb 2006 21:52:59 +0000 (13:52 -0800)]
[PATCH] compound page: no access_process_vm check

The PageCompound check before access_process_vm's set_page_dirty_lock is no
longer necessary, so remove it.  But leave the PageCompound checks in
bio_set_pages_dirty, dio_bio_complete and nfs_free_user_pages: at least some
of those were introduced as a little optimization on hugetlb pages.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] compound page: default destructor
Hugh Dickins [Tue, 14 Feb 2006 21:52:59 +0000 (13:52 -0800)]
[PATCH] compound page: default destructor

Somehow I imagined that calling a NULL destructor would free a compound page
rather than oopsing.  No, we must supply a default destructor, __free_pages_ok
using the order noted by prep_compound_page.  hugetlb can still replace this
as before with its own free_huge_page pointer.

The case that needs this is not common: rarely does put_compound_page's
put_page_testzero bring the count down to 0.  But if get_user_pages is applied
to some part of a compound page, without immediate release (e.g.  AIO or
Infiniband), then it's possible for its put_page to come after the containing
vma has been unmapped and the driver done its free_pages.

That's just the kind of case compound pages are supposed to be guarding
against (but Nick points out, nor did PageReserved handle this right).

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] compound page: use page[1].lru
Hugh Dickins [Tue, 14 Feb 2006 21:52:58 +0000 (13:52 -0800)]
[PATCH] compound page: use page[1].lru

If a compound page has its own put_page_testzero destructor (the only current
example is free_huge_page), that is noted in page[1].mapping of the compound
page.  But that's rather a poor place to keep it: functions which call
set_page_dirty_lock after get_user_pages (e.g.  Infiniband's
__ib_umem_release) ought to be checking first, otherwise set_page_dirty is
liable to crash on what's not the address of a struct address_space.

And now I'm about to make that worse: it turns out that every compound page
needs a destructor, so we can no longer rely on hugetlb pages going their own
special way, to avoid further problems of page->mapping reuse.  For example,
not many people know that: on 50% of i386 -Os builds, the first tail page of a
compound page purports to be PageAnon (when its destructor has an odd
address), which surprises page_add_file_rmap.

Keep the compound page destructor in page[1].lru.next instead.  And to free up
the common pairing of mapping and index, also move compound page order from
index to lru.prev.  Slab reuses page->lru too: but if we ever need slab to use
compound pages, it can easily stack its use above this.

(akpm: decoded version of the above: the tail pages of a compound page now
have ->mapping==NULL, so there's no need for the set_page_dirty[_lock]()
caller to check that they're not compund pages before doing the dirty).

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] pktcdvd: Reduce stack usage
Peter Osterlund [Tue, 14 Feb 2006 21:52:56 +0000 (13:52 -0800)]
[PATCH] pktcdvd: Reduce stack usage

Reduce stack usage in the pkt_start_write() function.  Even though it's not
currently a real problem, the pages and offsets arrays can be eliminated,
which saves approximately 1000 bytes of stack space.

Signed-off-by: Peter Osterlund <petero2@telia.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] pktcdvd: Don't unlock the door if the disc is in use
Peter Osterlund [Tue, 14 Feb 2006 21:52:56 +0000 (13:52 -0800)]
[PATCH] pktcdvd: Don't unlock the door if the disc is in use

Unlocking the door when the disc is in use is obviously not good, because then
it's possible to eject the disc at the wrong time and cause severe disc data
corruption.

Signed-off-by: Peter Osterlund <petero2@telia.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] pktcdvd: Allow non-writable media to be mounted
Peter Osterlund [Tue, 14 Feb 2006 21:52:55 +0000 (13:52 -0800)]
[PATCH] pktcdvd: Allow non-writable media to be mounted

If opening for write fails, the open method should return -EROFS.  This makes
"mount" try again with a read-only mount, instead of just giving up.

Signed-off-by: Peter Osterlund <petero2@telia.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] pktcdvd: Don't spam the kernel log when nothing is wrong
Peter Osterlund [Tue, 14 Feb 2006 21:52:54 +0000 (13:52 -0800)]
[PATCH] pktcdvd: Don't spam the kernel log when nothing is wrong

Change some messages that don't indicate an error so that they are only
printed when debugging is enabled.

Signed-off-by: Peter Osterlund <petero2@telia.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland...
Linus Torvalds [Tue, 14 Feb 2006 21:49:37 +0000 (13:49 -0800)]
Merge branch 'for-linus' of git://git./linux/kernel/git/roland/infiniband

18 years ago[MIPS] Fix typo in _sys32_rt_sigreturn and _sysn32_rt_sigreturn.
Atsushi Nemoto [Tue, 14 Feb 2006 13:40:45 +0000 (22:40 +0900)]
[MIPS] Fix typo in _sys32_rt_sigreturn and _sysn32_rt_sigreturn.

Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Fix CPU type bitmasks for MIPS III, IV and V.
Maciej W. Rozycki [Mon, 13 Feb 2006 09:15:49 +0000 (09:15 +0000)]
[MIPS] Fix CPU type bitmasks for MIPS III, IV and V.

Signed-off-by: Maciej W. Rozycki <macro@linux-mips.org>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] RM9000: Fix buggy I-cache workaround.
Thomas Koeller [Fri, 10 Feb 2006 16:36:27 +0000 (17:36 +0100)]
[MIPS] RM9000: Fix buggy I-cache workaround.

Signed-off-by: Thomas Koeller <thomas.koeller@baslerweb.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] MT: Propagate config7 into VPE.
Ralf Baechle [Fri, 10 Feb 2006 16:03:47 +0000 (16:03 +0000)]
[MIPS] MT: Propagate config7 into VPE.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] MT: Fix c0 back-to-back hazard.
Ralf Baechle [Fri, 10 Feb 2006 14:25:16 +0000 (14:25 +0000)]
[MIPS] MT: Fix c0 back-to-back hazard.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Get rid of kludgery needed to keep stdargs of old compilers working.
Ralf Baechle [Fri, 10 Feb 2006 14:13:08 +0000 (14:13 +0000)]
[MIPS] Get rid of kludgery needed to keep stdargs of old compilers working.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] More uaccess.h fixes with gcc >= 4.0.1.
Ralf Baechle [Fri, 10 Feb 2006 01:31:24 +0000 (01:31 +0000)]
[MIPS] More uaccess.h fixes with gcc >= 4.0.1.

From Richard Sandiford <richard@codesourcery.com>:

This patch caused a miscompilation of the restore_gp_regs() block
in restore_sigcontext().  This was in a 32-bit kernel compiled with
GCC CVS head.

restore_gp_regs() copies 64-bit user fields into 32-bit variables,
and in this combination, the new __get_user_asm_ll32() clobbers too
many registers.  It says:

/*
 * Get a long long 64 using 32 bit registers.
 */
{ \
__asm__ __volatile__( \
"1: lw %1, (%3) \n" \
"2: lw %D1, 4(%3) \n" \
" move %0, $0 \n" \
"3: .section .fixup,\"ax\" \n" \
"4: li %0, %4 \n" \
" move %1, $0 \n" \
" move %D1, $0 \n" \
" j 3b \n" \
" .previous \n" \
" .section __ex_table,\"a\" \n" \
" " __UA_ADDR " 1b, 4b \n" \
" " __UA_ADDR " 2b, 4b \n" \
" .previous \n" \
: "=r" (__gu_err), "=&r" (val) \
: "0" (0), "r" (addr), "i" (-EFAULT)); \
}

and this requires val (%1) to be a 64-bit value.  In the case I saw,
gcc was using $3 for the 32-bit val, and wasn't expecting $4 to be
clobbered.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Add protected_blast_icache_range, blast_icache_range, etc.
Atsushi Nemoto [Thu, 9 Feb 2006 15:39:06 +0000 (00:39 +0900)]
[MIPS] Add protected_blast_icache_range, blast_icache_range, etc.

Add blast_xxx_range(), protected_blast_xxx_range() etc. for common
use.  They are built by __BUILD_BLAST_CACHE_RANGE().
Use protected_cache_op() macro for various protected_ routines.
Output code should be logically same.

Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Rewrite get_wchan and its helper functions using kallsyms_lookup.
Atsushi Nemoto [Tue, 7 Feb 2006 16:48:03 +0000 (01:48 +0900)]
[MIPS] Rewrite get_wchan and its helper functions using kallsyms_lookup.

Implement get_wchan() and frame_info_init() using kallsyms_lookup().
This fixes problem with static sched/lock functions and mfinfo[]
maintenance issue.  If CONFIG_KALLSYMS was disabled, get_wchan() just
returns thread_saved_pc() value.

Also unwind stackframe based on "addiu sp,-imm" analysis instead of
frame pointer.  This fixes problem with functions compiled without
-fomit-frame-pointer.

Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Remove commented out function prom_build_cpu_map.
Ralf Baechle [Thu, 9 Feb 2006 12:26:35 +0000 (12:26 +0000)]
[MIPS] Remove commented out function prom_build_cpu_map.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Fold non-__mips64 case into CONFIG_32BIT case.
Ralf Baechle [Thu, 9 Feb 2006 12:13:28 +0000 (12:13 +0000)]
[MIPS] Fold non-__mips64 case into CONFIG_32BIT case.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Update docs to reflect the latest status of the Alchemy IDE driver.
Ralf Baechle [Tue, 14 Feb 2006 19:13:23 +0000 (19:13 +0000)]
[MIPS] Update docs to reflect the latest status of the Alchemy IDE driver.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] RM200: Give RM200 it's own timex.h.
Ralf Baechle [Mon, 28 Nov 2005 13:10:54 +0000 (13:10 +0000)]
[MIPS] RM200: Give RM200 it's own timex.h.

So we can get rid of config.h and the #ifdef crapola in the generic
timex.h.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[PATCH] Fix locking error in esp
Alan Cox [Tue, 14 Feb 2006 17:50:54 +0000 (17:50 +0000)]
[PATCH] Fix locking error in esp

Noted by Al Viro.

Also remove unused tmp_buf

Signed-off-by: Alan Cox <alan@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[SCSI] fix wrong context bugs in SCSI
James Bottomley [Tue, 14 Feb 2006 16:48:46 +0000 (10:48 -0600)]
[SCSI] fix wrong context bugs in SCSI

There's a bug in releasing scsi_device where the release function
actually frees the block queue.  However, the block queue release
calls flush_work(), which requires process context (the scsi_device
structure may release from irq context).  Update the release function
to invoke via the execute_in_process_context() API.

Also clean up the scsi_target structure releasing via this API.

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
18 years ago[PATCH] add scsi_execute_in_process_context() API
James Bottomley [Tue, 14 Feb 2006 16:42:07 +0000 (10:42 -0600)]
[PATCH] add scsi_execute_in_process_context() API

We have several points in the SCSI stack (primarily for our device
functions) where we need to guarantee process context, but (given the
place where the last reference was released) we cannot guarantee this.

This API gets around the issue by executing the function directly if
the caller has process context, but scheduling a workqueue to execute
in process context if the caller doesn't have it.  Unfortunately, it
requires memory allocation in interrupt context, but it's better than
what we have previously.  The true solution will require a bit of
re-engineering, so isn't appropriate for 2.6.16.

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
18 years ago[PATCH] x86: fix oprofile kernel callgraph regression
Gerald Britton [Tue, 14 Feb 2006 15:19:04 +0000 (10:19 -0500)]
[PATCH] x86: fix oprofile kernel callgraph regression

Fix x86 oprofile regression introduced by:
  commit c34d1b4d165c67b966bca4aba026443d7ff161eb
  [PATCH] mm: kill check_user_page_readable

That commit reorganized tests for the userspace stack walking moving all
those tests into dump_backtrace(), however, dump_backtrace() was used for
both userspace and kernel stalk walking.  The result is typically no
recorded callgraph information for kernel samples.

Revive the original function as dump_kernel_backtrace() and rename the
other to dump_user_backtrace() to avoid future confusion.

Signed-off-by: Gerald Britton <gbritton@alum.mit.edu>
Apology-from: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/gregkh/i2c-2.6
Linus Torvalds [Tue, 14 Feb 2006 16:22:57 +0000 (08:22 -0800)]
Merge /pub/scm/linux/kernel/git/gregkh/i2c-2.6

18 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/gregkh/usb-2.6
Linus Torvalds [Tue, 14 Feb 2006 16:21:57 +0000 (08:21 -0800)]
Merge /pub/scm/linux/kernel/git/gregkh/usb-2.6

18 years ago[PATCH] i2c: Drop outdated probe/remove code in i2c-isa
Jean Delvare [Wed, 8 Feb 2006 19:38:29 +0000 (20:38 +0100)]
[PATCH] i2c: Drop outdated probe/remove code in i2c-isa

Probe and remove methods are now defined at bus level. No more need to
redefine them at driver level in i2c-isa.

This lets us get rid of these annoying messages:
Driver 'it87-isa' needs updating - please use bus_type methods

Thanks to Nicolas Mailhot for reporting the problem and testing the fix.

Signed-off-by: Jean Delvare <khali@linux-fr.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
18 years ago[PATCH] it87: Fix oops on removal
Jean Delvare [Tue, 7 Feb 2006 16:53:32 +0000 (17:53 +0100)]
[PATCH] it87: Fix oops on removal

Fix an oops on it87 module removal when no supported hardware was
found.

Signed-off-by: Jean Delvare <khali@linux-fr.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
18 years ago[PATCH] w83627hf: Document the reset module parameter
Jean Delvare [Sun, 5 Feb 2006 22:17:34 +0000 (23:17 +0100)]
[PATCH] w83627hf: Document the reset module parameter

Document the reset module parameter which was recently added to the
w83627hf driver.

Signed-off-by: Jean Delvare <khali@linux-fr.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
18 years ago[PATCH] w83781d: Use real-time status registers
Jean Delvare [Sun, 5 Feb 2006 22:13:48 +0000 (23:13 +0100)]
[PATCH] w83781d: Use real-time status registers

Use the real-time status registers of the Winbond W83782D, W83783S and
W83627HF chips, instead of the interrupt status registers. Interrupts
cannot be trusted at least for voltage inputs, as they are two-times
triggers (as opposed to comparator mode, which we want.) The w83627hf
driver was fixed in a similar way some times ago.

Signed-off-by: Jean Delvare <khali@linux-fr.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
18 years ago[PATCH] vt8231: Fix sysfs temperature interface
Jean Delvare [Sun, 5 Feb 2006 22:11:16 +0000 (23:11 +0100)]
[PATCH] vt8231: Fix sysfs temperature interface

The VT8231 low temperature limits are actually hysteresis temperatures
to the high limits.

Signed-off-by: Jean Delvare <khali@linux-fr.org>
Cc: Roger Lucas <roger@planbit.co.uk>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>