deliverable/linux.git
19 years ago[NET]: Allow controlling NAPI device weight with sysfs
Stephen Hemminger [Wed, 8 Jun 2005 21:55:42 +0000 (14:55 -0700)] 
[NET]: Allow controlling NAPI device weight with sysfs

Simple interface to allow changing network device scheduling weight
with sysfs. Please consider this for 2.6.12, since risk/impact is small.

Signed-off-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 years ago[IPV6]: Update parm.link in ip6ip6_tnl_change()
Gabor Fekete [Wed, 8 Jun 2005 21:54:38 +0000 (14:54 -0700)] 
[IPV6]: Update parm.link in ip6ip6_tnl_change()

Signed-off-by: Gabor Fekete <gfekete@cc.jyu.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 years agoMerge master.kernel.org:/home/rmk/linux-2.6-arm
Linus Torvalds [Wed, 8 Jun 2005 18:43:17 +0000 (11:43 -0700)] 
Merge master.kernel.org:/home/rmk/linux-2.6-arm

19 years agoMerge master.kernel.org:/home/rmk/linux-2.6-serial
Linus Torvalds [Wed, 8 Jun 2005 18:42:44 +0000 (11:42 -0700)] 
Merge master.kernel.org:/home/rmk/linux-2.6-serial

19 years ago[PATCH] Serial: remove unused variable in sa1100 driver
Russell King [Wed, 8 Jun 2005 18:26:47 +0000 (19:26 +0100)] 
[PATCH] Serial: remove unused variable in sa1100 driver

Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
19 years ago[PATCH] ARM: 2664/2: add support for atomic ops on pre-ARMv6 SMP systems
Nicolas Pitre [Wed, 8 Jun 2005 18:00:47 +0000 (19:00 +0100)] 
[PATCH] ARM: 2664/2: add support for atomic ops on pre-ARMv6 SMP systems

Patch from Nicolas Pitre

Not that there might be many of them on the planet, but at least RMK
apparently has one.

Signed-off-by: Nicolas Pitre
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
19 years ago[PATCH] ARM: 2705/1: fix writesw for misaligned source pointer
Nicolas Pitre [Wed, 8 Jun 2005 18:00:16 +0000 (19:00 +0100)] 
[PATCH] ARM: 2705/1: fix writesw for misaligned source pointer

Patch from Nicolas Pitre

Signed-off-by: Nicolas Pitre
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
19 years ago[PATCH] ARM: 2703/1: pxa-regs.h: complete I2S GPIO alternate functions for PXA27x
Giorgio Padrin [Wed, 8 Jun 2005 18:00:15 +0000 (19:00 +0100)] 
[PATCH] ARM: 2703/1: pxa-regs.h: complete I2S GPIO alternate functions for PXA27x

Patch from Giorgio Padrin

The patch completes I2S GPIO alternate functions for PXA27x, adding I2S_SYSCLK.
File: pxa-regs.h .

Signed-off-by: Giorgio Padrin
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
19 years ago[PATCH] ppc64 kprobes: don't eat dabr/iabr exceptions
Ananth N Mavinakayanahalli [Wed, 8 Jun 2005 05:33:43 +0000 (15:33 +1000)] 
[PATCH] ppc64 kprobes: don't eat dabr/iabr exceptions

Kprobes was eating the hardware instruction and data address
breakpoint exceptions.  This patch fixes it; kprobes doesn't use those
exceptions at all and should ignore them.

Signed-off-by: Ananth N Mavinakayanahalli <amavin@redhat.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] ppc64: print negative numbers correctly in boot wrapper
Olaf Hering [Wed, 8 Jun 2005 05:12:00 +0000 (15:12 +1000)] 
[PATCH] ppc64: print negative numbers correctly in boot wrapper

if num has a value of -1, accessing the digits[] array will fail and the
format string will be printed in funny way, or not at all. This happens if
one prints negative numbers.
Just change the code to match lib/vsprintf.c
asm/div64.h cant be used because u64 maps to u32 for this build.

Signed-off-by: Olaf Hering <olh@suse.de>
Signed-off-by: Paul Mackerras <paulus@samba.org>
19 years agoAutomatic merge of master.kernel.org:/home/rmk/linux-2.6-arm
Linus Torvalds [Wed, 8 Jun 2005 14:57:17 +0000 (07:57 -0700)] 
Merge ... kernel.org:/home/rmk/linux-2.6-arm

19 years ago[PATCH] ARM: Fix Xscale copy_page implementation
Russell King [Wed, 8 Jun 2005 14:28:24 +0000 (15:28 +0100)] 
[PATCH] ARM: Fix Xscale copy_page implementation

The ARM copypage changes in 2.6.12-rc4-git1 removed the preempt locking
from the copypage functions which broke the XScale implementation.
This patch fixes the locking on XScale and removes the now unneeded
minicache code.

Signed-off-by: Russell King <rmk@arm.linux.org.uk>
Checked-by: Richard Purdie
19 years ago[PATCH] NFS: Fix lookup intent handling
Trond Myklebust [Tue, 7 Jun 2005 22:37:01 +0000 (18:37 -0400)] 
[PATCH] NFS: Fix lookup intent handling

We should never apply a lookup intent to anything other than the last
path component in an open(), create() or access() call.

Introduce the helper nfs_lookup_check_intent() which always returns
zero if LOOKUP_CONTINUE or LOOKUP_PARENT are set, and returns the
intent flags if we're on the last component of the lookup.
By doing so, we fix a bug in open(O_EXCL), where we may end up
optimizing away a real lookup of the parent directory.

Problem noticed by Linda Dunaphant <linda.dunaphant@ccur.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years agoMerge of master.kernel.org:/pub/scm/linux/kernel/git/davej/agpgart
Linus Torvalds [Tue, 7 Jun 2005 20:41:30 +0000 (13:41 -0700)] 
Merge ... /pub/scm/linux/kernel/git/davej/agpgart

19 years ago[PATCH] PCI: do VIA IRQ fixup always, not just in PIC mode
Bjorn Helgaas [Tue, 7 Jun 2005 20:22:18 +0000 (13:22 -0700)] 
[PATCH] PCI: do VIA IRQ fixup always, not just in PIC mode

At least some VIA chipsets require the fixup even in IO-APIC mode.

This was found and debugged with the patient assistance of Stian
Jordet <liste@jordet.nu> on an Asus CUV266-DLS motherboard.

Signed-off-by: Bjorn Helgaas <bjorn.helgaas@hp.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] input: disable scroll feature on AT keyboards
Vojtech Pavlik [Tue, 7 Jun 2005 20:22:14 +0000 (13:22 -0700)] 
[PATCH] input: disable scroll feature on AT keyboards

This patch disables the scroll feature on AT keyboards by default, because
it causes the numbers of mouse devices to shift, breaking user setups.

Signed-off-by: Vojtech Pavlik <vojtech@suse.cz>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] ppc32: add 405EP cpu_spec entry
Eugene Surovegin [Tue, 7 Jun 2005 20:22:09 +0000 (13:22 -0700)] 
[PATCH] ppc32: add 405EP cpu_spec entry

Add a definition for PPC 405EP which was lost somehow during 2.4 -> 2.6
transition.

Recent change to arch/ppc/kernel/misc.S ("Fix incorrect CPU_FTR fixup usage
for unified caches") triggered this bug and 405EP boards don't boot
anymore.

Signed-off-by: Eugene Surovegin <ebs@ebshome.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] send_IPI_mask_sequence() warning fix
Matthew Dobson [Tue, 7 Jun 2005 20:22:05 +0000 (13:22 -0700)] 
[PATCH] send_IPI_mask_sequence() warning fix

In file included from arch/i386/kernel/smp.c:235:
include/asm-i386/mach-numaq/mach_ipi.h:4: warning: `send_IPI_mask_sequence'
declared inline after its definition

Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] Replace check_bridge_mode() with (bridge->mode & AGSTAT_MODE_3_0).
David Mosberger [Mon, 4 Apr 2005 20:29:43 +0000 (13:29 -0700)] 
[PATCH] Replace check_bridge_mode() with (bridge->mode & AGSTAT_MODE_3_0).

[AGPGART] Replace check_bridge_mode() with (bridge->mode & AGSTAT_MODE_3_0).

As mentioned earlier, the current check_bridge_mode() code assumes
that AGP bridges are PCI devices.  This isn't always true.  Definitely
not for HP zx1 chipset and the same seems to be the case for SGI's AGP
bridge.

The patch below fixes the problem by picking up the AGP_MODE_3_0 bit
from bridge->mode.  I feel like I may be missing something, since I
can't see any reason why check_bridge_mode() wasn't doing that in the
first place.  According to the AGP 3.0 specs, the AGP_MODE_3_0 bit is
determined during the hardware reset and cannot be changed, so it
seems to me it should be safe to pick it up from bridge->mode.

With the patch applied, I can definitely use AGP acceleration both
with AGP 2.0 and AGP 3.0 (one with an Nvidia card, the other with an
ATI FireGL card).

Unless someone spots a problem, please apply this patch so 3d
acceleration can work on zx1 boxes again.

This makes AGP work again on machines with an AGP bridge that isn't a
PCI device.

Signed-off-by: David Mosberger-Tang <davidm@hpl.hp.com>
Signed-off-by: Dave Jones <davej@redhat.com>
19 years ago[PATCH] AGP fix for Xen VMM
Keir Fraser [Wed, 30 Mar 2005 21:17:04 +0000 (13:17 -0800)] 
[PATCH] AGP fix for Xen VMM

When Linux is running on the Xen virtual machine monitor, physical
addresses are virtualised and cannot be directly referenced by the AGP
GART.  This patch fixes the GART driver for Xen by adding a layer of
abstraction between physical addresses and 'GART addresses'.

Architecture-specific functions are also defined for allocating and freeing
the GATT.  Xen requires this to ensure that table really is contiguous from
the point of view of the GART.

These extra interface functions are defined as 'no-ops' for all existing
architectures that use the GART driver.

Signed-off-by: Keir Fraser <keir@xensource.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Dave Jones <davej@redhat.com>
19 years ago[PATCH] sgi-agp: fixes a problem with accessing GART memory in sgi_tioca_insert_memor...
Michael Werner [Mon, 28 Mar 2005 06:08:42 +0000 (22:08 -0800)] 
[PATCH] sgi-agp: fixes a problem with accessing GART memory in sgi_tioca_insert_memory and sgi_tioca_remove_memory

This patch fixes a problem with accessing GART memory in
sgi_tioca_insert_memory and sgi_tioca_remove_memory.

 sgi-agp.c |   12 +++++++++---
 1 files changed, 9 insertions(+), 3 deletions(-)

Signed-off-by: Mike Werner <werner@sgi.com>
Signed-off-by: Dave Jones <davej@redhat.com>
19 years ago[PATCH] i945G patch for agpgart
Alan Hourihane [Tue, 31 May 2005 18:50:49 +0000 (19:50 +0100)] 
[PATCH] i945G patch for agpgart

Attached is a small patch for i945G support against 2.6.11.11.

From: Alan Hourihane <alanh@fairlite.demon.co.uk>
Signed-off-by: Dave Jones <davej@redhat.com>
19 years ago[PATCH] Include <linux/config.h> before testing CONFIG_ACPI
David Mosberger [Mon, 6 Jun 2005 22:50:09 +0000 (15:50 -0700)] 
[PATCH] Include <linux/config.h> before testing CONFIG_ACPI

I'm not sure why this issue is suddenly showing, but without this
patchlet, the zx1 config won't compile anymore (e.g., to see the
compilation-error, look for "***" in [1]).

[1] http://www.gelato.unsw.edu.au/kerncomp/results//2005-06-06-17-00/zx1_defconfig-log.html

Signed-off-by: David Mosberger-Tang <davidm@hpl.hp.com>
Cc: "Brown, Len" <len.brown@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] ppc32: add <linux/compiler.h> to <asm/sigcontext.h>
Tom Rini [Mon, 6 Jun 2005 22:50:08 +0000 (15:50 -0700)] 
[PATCH] ppc32: add <linux/compiler.h> to <asm/sigcontext.h>

On ppc32, <asm/sigcontext.h> uses __user, but doesn't directly include
<linux/compiler.h>.  This adds that in.  Without this, glibc will not
compile.

Signed-off-by: Tom Rini <trini@kernel.crashing.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years agoMerge of rsync://rsync.kernel.org/pub/scm/linux/kernel/git/davem/tg3-2.6
Linus Torvalds [Mon, 6 Jun 2005 23:59:55 +0000 (16:59 -0700)] 
Merge ... /pub/scm/linux/kernel/git/davem/tg3-2.6

19 years agoMerge of rsync://rsync.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
Linus Torvalds [Mon, 6 Jun 2005 23:58:53 +0000 (16:58 -0700)] 
Merge ... /pub/scm/linux/kernel/git/davem/net-2.6

19 years ago[TG3]: Update driver version and release date.
David S. Miller [Mon, 6 Jun 2005 22:22:56 +0000 (15:22 -0700)] 
[TG3]: Update driver version and release date.

Signed-off-by: David S. Miller <davem@davemloft.net>
19 years ago[TG3] Fix link failure in 5701
Michael Chan [Mon, 6 Jun 2005 22:16:20 +0000 (15:16 -0700)] 
[TG3] Fix link failure in 5701

On some 5701 devices with older bootcode, the LED configuration bits in
SRAM may be invalid with value zero. The fix is to check for invalid
bits (0) and default to PHY 1 mode. Incorrect LED mode will lead to
error in programming the PHY.

Thanks to Grant Grundler for debugging the problem.

>From Grant:
| In May, 2004,  tg3 v3.4 changed how MAC_LED_CTRL (0x40c) was getting
| programmed and how to determine what to program into LED_CTRL. The new
| code trusted NIC_SRAM_DATA_CFG (0x00000b58) to indicate what to write
| to LED_CTRL and MII EXT_CTRL registers. On "IOX Core Lan", SRAM was
| saying MODE_MAC (0x0) and that doesn't work.

Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 years ago[TG3]: Add TSO firmware license
Michael Chan [Mon, 6 Jun 2005 22:15:17 +0000 (15:15 -0700)] 
[TG3]: Add TSO firmware license

Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 years ago[TG3]: Update pci.ids for BCM5752
John W. Linville [Mon, 6 Jun 2005 22:14:35 +0000 (15:14 -0700)] 
[TG3]: Update pci.ids for BCM5752

Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
19 years ago[ETHTOOL]: Check correct pointer in ethtool_set_coalesce().
David S. Miller [Mon, 6 Jun 2005 22:07:19 +0000 (15:07 -0700)] 
[ETHTOOL]: Check correct pointer in ethtool_set_coalesce().

It was checking the "GET" function pointer instead of
the "SET" one.  Looks like a cut&paste error :-)

Signed-off-by: David S. Miller <davem@davemloft.net>
19 years ago[PATCH] binfmt_flat mmap flag fix
Yoshinori Sato [Mon, 6 Jun 2005 21:46:32 +0000 (14:46 -0700)] 
[PATCH] binfmt_flat mmap flag fix

Make sure that binfmt_flat passes the correct flags into do_mmap().  nommu's
validate_mmap_request() will simple return -EINVAL if we try and pass it a
flags value of zero.

Signed-off-by: Yoshinori Sato <ysato@users.sourceforge.jp>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] namei fixes (19/19)
Al Viro [Mon, 6 Jun 2005 20:36:14 +0000 (13:36 -0700)] 
[PATCH] namei fixes (19/19)

__do_follow_link() passes potentially worng vfsmount to touch_atime().  It
matters only in (currently impossible) case of symlink mounted on something,
but it's trivial to fix and that actually makes more sense.

Signed-off-by: Al Viro <viro@parcelfarce.linux.theplanet.co.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] namei fixes (18/19)
Al Viro [Mon, 6 Jun 2005 20:36:13 +0000 (13:36 -0700)] 
[PATCH] namei fixes (18/19)

Cosmetical cleanups - __follow_mount() calls in __link_path_walk() absorbed
into do_lookup().

Obviously equivalent transformation.

Signed-off-by: Al Viro <viro@parcelfarce.linux.theplanet.co.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] namei fixes (17/19)
Al Viro [Mon, 6 Jun 2005 20:36:13 +0000 (13:36 -0700)] 
[PATCH] namei fixes (17/19)

follow_mount() made void, reordered dput()/mntput() in it.

follow_dotdot() switched from struct vfmount ** + struct dentry ** to
struct nameidata *; callers updated.

Equivalent transformation + fix for too-early-mntput() race.

Signed-off-by: Al Viro <viro@parcelfarce.linux.theplanet.co.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] namei fixes (16/19)
Al Viro [Mon, 6 Jun 2005 20:36:12 +0000 (13:36 -0700)] 
[PATCH] namei fixes (16/19)

Conditional mntput() moved into __do_follow_link().  There it collapses with
unconditional mntget() on the same sucker, closing another too-early-mntput()
race.

Signed-off-by: Al Viro <viro@parcelfarce.linux.theplanet.co.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] namei fixes (15/19)
Al Viro [Mon, 6 Jun 2005 20:36:11 +0000 (13:36 -0700)] 
[PATCH] namei fixes (15/19)

Getting rid of sloppy logics:

a) in do_follow_link() we have the wrong vfsmount dropped if our symlink
had been mounted on something.  Currently it worls only because we never
get such situation (modulo filesystem playing dirty tricks on us).  And
it obfuscates already convoluted logics...

b) same goes for open_namei().

c) in __link_path_walk() we have another "it should never happen" sloppiness -
out_dput: there does double-free on underlying vfsmount and leaks the covering
one if we hit it just after crossing a mountpoint.  Again, wrong vfsmount
getting dropped.

d) another too-early-mntput() race - in do_follow_mount() we need to postpone
conditional mntput(path->mnt) until after dput(path->dentry).  Again, this one
happens only in it-currently-never-happens-unless-some-fs-plays-dirty
scenario...

Signed-off-by: Al Viro <viro@parcelfarce.linux.theplanet.co.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] namei fixes (14/19)
Al Viro [Mon, 6 Jun 2005 20:36:10 +0000 (13:36 -0700)] 
[PATCH] namei fixes (14/19)

shifted conditional mntput() into do_follow_link() - all callers were doing
the same thing.

Obviously equivalent transformation.

Signed-off-by: Al Viro <viro@parcelfarce.linux.theplanet.co.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] namei fixes (13/19)
Al Viro [Mon, 6 Jun 2005 20:36:08 +0000 (13:36 -0700)] 
[PATCH] namei fixes (13/19)

In open_namei() exit_dput: we have mntput() done in the wrong order -
if nd->mnt != path.mnt we end up doing
mntput(nd->mnt);
nd->mnt = path.mnt;
dput(nd->dentry);
mntput(nd->mnt);
which drops nd->dentry too late.  Fixed by having path.mnt go first.
That allows to switch O_NOFOLLOW under if (__follow_mount(...)) back
to exit_dput, while we are at it.

Fix for early-mntput() race + equivalent transformation.

Signed-off-by: Al Viro <viro@parcelfarce.linux.theplanet.co.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] namei fixes (12/19)
Al Viro [Mon, 6 Jun 2005 20:36:08 +0000 (13:36 -0700)] 
[PATCH] namei fixes (12/19)

In open_namei() we take mntput(nd->mnt);nd->mnt=path.mnt; out of the if
(__follow_mount(...)), making it conditional on nd->mnt != path.mnt instead.

Then we shift the result downstream.

Equivalent transformations.

Signed-off-by: Al Viro <viro@parcelfarce.linux.theplanet.co.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] namei fixes (11/19)
Al Viro [Mon, 6 Jun 2005 20:36:07 +0000 (13:36 -0700)] 
[PATCH] namei fixes (11/19)

shifted conditional mntput() calls in __link_path_walk() downstream.

Obviously equivalent transformation.

Signed-off-by: Al Viro <viro@parcelfarce.linux.theplanet.co.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] namei fixes (10/19)
Al Viro [Mon, 6 Jun 2005 20:36:06 +0000 (13:36 -0700)] 
[PATCH] namei fixes (10/19)

In open_namei(), __follow_down() loop turned into __follow_mount().
Instead of
if we are on a mountpoint dentry
if O_NOFOLLOW checks fail
drop path.dentry
drop nd
return
do equivalent of follow_mount(&path.mnt, &path.dentry)
nd->mnt = path.mnt
we do
if __follow_mount(path) had, indeed, traversed mountpoint
/* now both nd->mnt and path.mnt are pinned down */
if O_NOFOLLOW checks fail
drop path.dentry
drop path.mnt
drop nd
return
mntput(nd->mnt)
nd->mnt = path.mnt

Now __follow_down() can be folded into follow_down() - no other callers left.
We need to reorder dput()/mntput() there - same problem as in follow_mount().

Equivalent transformation + fix for a bug in O_NOFOLLOW handling - we used to
get -ELOOP if we had the same fs mounted on /foo and /bar, had something bound
on /bar/baz and tried to open /foo/baz with O_NOFOLLOW.  And fix of
too-early-mntput() race in follow_down()

Signed-off-by: Al Viro <viro@parcelfarce.linux.theplanet.co.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] namei fixes (9/19)
Al Viro [Mon, 6 Jun 2005 20:36:05 +0000 (13:36 -0700)] 
[PATCH] namei fixes (9/19)

New helper: __follow_mount(struct path *path).  Same as follow_mount(), except
that we do *not* do mntput() after the first lookup_mnt().

IOW, original path->mnt stays pinned down.  We also take care to do dput()
before mntput() in the loop body (follow_mount() also needs that reordering,
but that will be done later in the series).

The following are equivalent, assuming that path.mnt == x:
(1)
follow_mount(&path.mnt, &path.dentry)
(2)
__follow_mount(&path);
if (path->mnt != x)
mntput(x);
(3)
if (__follow_mount(&path))
mntput(x);

Callers of follow_mount() in __link_path_walk() converted to (2).

Equivalent transformation + fix for too-late-mntput() race in __follow_mount()
loop.

Signed-off-by: Al Viro <viro@parcelfarce.linux.theplanet.co.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] namei fixes (8/19)
Al Viro [Mon, 6 Jun 2005 20:36:04 +0000 (13:36 -0700)] 
[PATCH] namei fixes (8/19)

In open_namei() we never use path.mnt or path.dentry after exit: or ok:.
Assignment of path.dentry in case of LAST_BIND is dead code and only
obfuscates already convoluted function; assignment of path.mnt after
__do_follow_link() can be moved down to the place where we set path.dentry.

Obviously equivalent transformations, just to clean the air a bit in that
region.

Signed-off-by: Al Viro <viro@parcelfarce.linux.theplanet.co.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] namei fixes (7/19)
Al Viro [Mon, 6 Jun 2005 20:36:03 +0000 (13:36 -0700)] 
[PATCH] namei fixes (7/19)

The first argument of __do_follow_link() switched to struct path *
(__do_follow_link(path->dentry, ...) -> __do_follow_link(path, ...)).

All callers have the same calls of mntget() right before and dput()/mntput()
right after __do_follow_link(); these calls have been moved inside.

Obviously equivalent transformations.

Signed-off-by: Al Viro <viro@parcelfarce.linux.theplanet.co.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] namei fixes (6/19)
Al Viro [Mon, 6 Jun 2005 20:36:02 +0000 (13:36 -0700)] 
[PATCH] namei fixes (6/19)

mntget(path->mnt) in do_follow_link() moved down to right before the
__do_follow_link() call and rigth after loop: resp.

dput()+mntput() on non-ELOOP branch moved up to right after __do_follow_link()
call.

resulting
loop:
mntget(path->mnt);
path_release(nd);
dput(path->mnt);
mntput(path->mnt);
replaced with equivalent
dput(path->mnt);
path_release(nd);

Equivalent transformations - the reason why we have that mntget() is that
__do_follow_link() can drop a reference to nd->mnt and that's what holds
path->mnt.  So that call can happen at any point prior to __do_follow_link()
touching nd->mnt.  The rest is obvious.

NOTE: current tree relies on symlinks *never* being mounted on anything.  It's
not hard to get rid of that assumption (actually, that will come for free
later in the series).  For now we are just not making the situation worse than
it is.

Signed-off-by: Al Viro <viro@parcelfarce.linux.theplanet.co.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] namei fixes (5/19)
Al Viro [Mon, 6 Jun 2005 20:36:01 +0000 (13:36 -0700)] 
[PATCH] namei fixes (5/19)

fix for too early mntput() in open_namei() - we pin path.mnt down for the
duration of __do_follow_link().  Otherwise we could get the fs where our
symlink lived unmounted while we were in __do_follow_link().  That would end
up with dentry of symlink staying pinned down through the fs shutdown.

Signed-off-by: Al Viro <viro@parcelfarce.linux.theplanet.co.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] namei fixes (4/19)
Al Viro [Mon, 6 Jun 2005 20:36:01 +0000 (13:36 -0700)] 
[PATCH] namei fixes (4/19)

path.mnt in open_namei() set to mirror nd->mnt.

nd->mnt is set in 3 places in that function - path_lookup() in the beginning,
__follow_down() loop after do_last: and __do_follow_link() call after
do_link:.

We set path.mnt to nd->mnt after path_lookup() and __do_follow_link().  In
__follow_down() loop we use &path.mnt instead of &nd->mnt and set nd->mnt to
path.mnt immediately after that loop.

Obviously equivalent transformation.

Signed-off-by: Al Viro <viro@parcelfarce.linux.theplanet.co.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] namei fixes (3/19)
Al Viro [Mon, 6 Jun 2005 20:36:00 +0000 (13:36 -0700)] 
[PATCH] namei fixes (3/19)

Replaced struct dentry *dentry in namei with struct path path.  All uses of
dentry replaced with path.dentry there.

Obviously equivalent transformation.

Signed-off-by: Al Viro <viro@parcelfarce.linux.theplanet.co.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] namei fixes (2/19)
Al Viro [Mon, 6 Jun 2005 20:35:59 +0000 (13:35 -0700)] 
[PATCH] namei fixes (2/19)

All callers of do_follow_link() do mntget() right before it and
dput()+mntput() right after.  These calls are moved inside do_follow_link()
now.

Obviously equivalent transformation.

Signed-off-by: Al Viro <viro@parcelfarce.linux.theplanet.co.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] namei fixes
Al Viro [Mon, 6 Jun 2005 20:35:58 +0000 (13:35 -0700)] 
[PATCH] namei fixes

OK, here comes a patch series that hopefully should close all
too-early-mntput() races in fs/namei.c.  Entire area is convoluted as hell, so
I'm splitting that series into _very_ small chunks.

Patches alread in the tree close only (very wide) races in following symlinks
(see "busy inodes after umount" thread some time ago).  Unfortunately, quite a
few narrower races of the same nature were not closed.  Hopefully this should
take care of all of them.

Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] ppc32: Fix incorrect CPU_FTR fixup usage for unified caches
Kumar Gala [Mon, 6 Jun 2005 20:35:57 +0000 (13:35 -0700)] 
[PATCH] ppc32: Fix incorrect CPU_FTR fixup usage for unified caches

Runtime feature support for unified caches was testing a userland feature
flag (PPC_FEATURE_UNIFIED_CACHE) instead of a cpu feature flag
(CPU_FTR_SPLIT_ID_CACHE).  Luckily the current defined bit mask for cpu
features and userland features do not overlap so this only causes an issue
on machines with a unified cache, which is extremely rare on PPC today.

Signed-off-by: Kumar Gala <kumar.gala@freescale.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] h8300 build error fix
Yoshinori Sato [Mon, 6 Jun 2005 20:35:56 +0000 (13:35 -0700)] 
[PATCH] h8300 build error fix

h8300 was missing a few definitions.

Signed-off-by: Yoshinori Sato <ysato@users.sourceforge.jp>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] moxa: do not ignore input
Denis Vlasenko [Mon, 6 Jun 2005 20:35:55 +0000 (13:35 -0700)] 
[PATCH] moxa: do not ignore input

Stop using tty internal structure in mxser_receive_chars(), use
tty_insert_flip_char(tty, ch flag); instead.

Without this change driver ignores any rx'ed chars.

Run tested.

Cc: Alan Cox <alan@redhat.com>
Cc: Russell King <rmk@arm.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] broken fault_in_pages_readable call in generic_file_buffered_write()
Martin Schwidefsky [Mon, 6 Jun 2005 20:35:54 +0000 (13:35 -0700)] 
[PATCH] broken fault_in_pages_readable call in generic_file_buffered_write()

fault_in_pages_readable() is being passed an incorrect `end' address, which
can result in writes accidentally faulting in pages which will not be affected
by the write() call.

Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years agoLinux 2.6.12-rc6
Linus Torvalds [Mon, 6 Jun 2005 15:22:29 +0000 (08:22 -0700)] 
Linux 2.6.12-rc6

Getting ready for the real release..

19 years ago[PATCH] serial: update NEC VR4100 series serial support
Yoichi Yuasa [Sat, 4 Jun 2005 22:43:34 +0000 (15:43 -0700)] 
[PATCH] serial: update NEC VR4100 series serial support

- Changed the return value of unknown type to NULL.

- Deleted the NULL check of dev_id in siu_interrupt().

- Deleted the NULL check of port->membase in siu_shutdown().

- Added the NULL check of port->membase to siu_startup().

- Removed early_uart_ops. Now using vr41xx_siu standerd one.

- Changed KSEG1ADDR() in siu_console_setup() to ioremap().

- When uart_add_one_port() failed, changed to set NULL to port->dev.

Signed-off-by: Yoichi Yuasa <yuasa@hh.iij4u.or.jp>
Cc: Russell King <rmk@arm.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] s390: deadlock in appldata
Gerald Schaefer [Sat, 4 Jun 2005 22:43:33 +0000 (15:43 -0700)] 
[PATCH] s390: deadlock in appldata

The system might hang when using appldata_mem with high I/O traffic and a
large number of devices.  The spinlocks bdev_lock and swaplock are acquired
via calls to si_meminfo() and si_swapinfo() from a tasklet, i.e.  interrupt
context, which can lead to a deadlock.  Replace tasklet with work queue.

Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] s390: in_interrupt vs. in_atomic
Martin Schwidefsky [Sat, 4 Jun 2005 22:43:32 +0000 (15:43 -0700)] 
[PATCH] s390: in_interrupt vs. in_atomic

The condition for no context in do_exception checks for hard and soft
interrupts by using in_interrupt() but not for preemption.  This is bad for
the users of __copy_from/to_user_inatomic because the fault handler might call
schedule although the preemption count is != 0.  Use in_atomic() instead
in_interrupt().

Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] s390: uml ptrace fixes
Bodo Stroesser [Sat, 4 Jun 2005 22:43:32 +0000 (15:43 -0700)] 
[PATCH] s390: uml ptrace fixes

To make UML build and run on s390, I needed to do these two little
changes:

1) UML includes some of the subarch's (s390) headers. I had to
   change one of them with the following one-liner, to make this
   compile. AFAICS, this change doesn't break compilation of s390
   itself.

2) UML needs to intercept syscalls via ptrace to invalidate the syscall,
   read syscall's parameters and write the result with the result of
   UML's syscall processing. Also, UML needs to make sure, that the host
   does no syscall restart processing. On i386 for example, this can be
   done by writing -1 to orig_eax on the 2nd syscall interception
   (orig_eax is the syscall number, which after the interception is used
   as a "interrupt was a syscall" flag only.
   Unfortunately, s390 holds syscall number and syscall result in gpr2 and
   its "interrupt was a syscall" flag (trap) is unreachable via ptrace.
   So I changed the host to set trap to -1, if the syscall number is changed
   to an invalid value on the first syscall interception.

Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] s390: ptrace peek and poke
Martin Schwidefsky [Sat, 4 Jun 2005 22:43:30 +0000 (15:43 -0700)] 
[PATCH] s390: ptrace peek and poke

The special cases of peek and poke on acrs[15] and the fpc register are not
handled correctly.  A poke on acrs[15] will clobber the 4 bytes after the
access registers in the thread_info structure.  That happens to be the kernel
stack pointer.  A poke on the fpc with an invalid value is not caught by the
validity check.  On the next context switch the broken fpc value will cause a
program check in the kernel.  Improving the checks in peek and poke fixes
this.

Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] mpage_end_io_write() I/O error handling fix
Qu Fuping [Sat, 4 Jun 2005 22:43:29 +0000 (15:43 -0700)] 
[PATCH] mpage_end_io_write() I/O error handling fix

When fsync() runs wait_on_page_writeback_range() it only inspects pages which
are actually under I/O (PAGECACHE_TAG_WRITEBACK).  If a page completed I/O
prior to wait_on_page_writeback_range() looking at it, it is supposed to have
recorded its I/O error state in the address_space.

But mpage_mpage_end_io_write() forgot to set the address_space error flag in
this case.

Signed-off-by: Qu Fuping <fs@ercist.iscas.ac.cn>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years agoAutomatic merge of 'misc-fixes' branch from
Linus Torvalds [Sat, 4 Jun 2005 15:18:39 +0000 (08:18 -0700)] 
Automatic merge of 'misc-fixes' branch from

rsync://rsync.kernel.org/pub/scm/linux/kernel/git/jgarzik/netdev-2.6

19 years agoAutomatic merge of /spare/repo/netdev-2.6 branch r8169-fix