safe/jmp/linux-2.6
18 years agodrm: fix quiescent locking
Dave Airlie [Thu, 24 Nov 2005 10:41:14 +0000 (21:41 +1100)]
drm: fix quiescent locking

A fix for a locking bug which is triggered when a client tries to lock with
flag DMA_QUIESCENT (typically the X server), but gets interrupted by a signal.
The locking IOCTL should then return an error, but if DMA_QUIESCENT succeeds
it returns 0, and the client falsely thinks it has the lock. In addition
The client waits for DMA_QUISCENT and possibly DMA_READY without having the lock.

From: Thomas Hellstrom
Signed-off-by: Dave Airlie <airlied@linux.ie>
18 years agoMerge branch 'drm-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/airlied...
Linus Torvalds [Thu, 24 Nov 2005 07:10:47 +0000 (23:10 -0800)]
Merge branch 'drm-linus' of git://git./linux/kernel/git/airlied/drm-2.6

18 years ago[PATCH] usb serial: remove redundant include
Nicolas Kaiser [Wed, 23 Nov 2005 23:45:58 +0000 (15:45 -0800)]
[PATCH] usb serial: remove redundant include

remove redundant include

Signed-off-by: Nicolas Kaiser <nikai@nikai.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] USB: kernel-doc for linux/usb.h
Randy Dunlap [Wed, 23 Nov 2005 23:45:53 +0000 (15:45 -0800)]
[PATCH] USB: kernel-doc for linux/usb.h

Fix kernel-doc warning in linux/usb.h.

Signed-off-by: Randy Dunlap <rdunlap@xenotime.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] USB: fix USB key generates ioctl_internal_command errors issue
David Härdeman [Wed, 23 Nov 2005 23:45:49 +0000 (15:45 -0800)]
[PATCH] USB: fix USB key generates ioctl_internal_command errors issue

On Wed, Nov 16, 2005 at 06:34:24PM -0800, Pete Zaitcev wrote:
>On Wed, 16 Nov 2005 23:52:32 +0100, David Härdeman <david@2gen.com> wrote:
>> usb-storage: waiting for device to settle before scanning
>>   Vendor: I0MEGA    Model: UMni1GB*IOM2K4    Rev: 1.01
>>   Type:   Direct-Access                      ANSI SCSI revision: 02
>> SCSI device sda: 2048000 512-byte hdwr sectors (1049 MB)
>> sda: Write Protect is off
>> sda: Mode Sense: 00 00 00 00
>> sda: assuming drive cache: write through
>> ioctl_internal_command: <8 0 0 0> return code = 8000002
>>    : Current: sense key=0x0
>>     ASC=0x0 ASCQ=0x0
>> SCSI device sda: 2048000 512-byte hdwr sectors (1049 MB)
>
>I think it's harmless. I saw things like that, and initially I plugged
>them with workarounds like this:

Thanks for the pointer, and yes, it is harmless, but it floods the
console with the messages which hides other (potentially important)
messages...following your example I've made a patch which fixes the
problem.

Signed-off-by: David Härdeman <david@2gen.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] USB: ohci, move ppc asic tweaks nearer pci
David Brownell [Wed, 23 Nov 2005 23:45:42 +0000 (15:45 -0800)]
[PATCH] USB: ohci, move ppc asic tweaks nearer pci

This should fix a suspend/resume issues that appear with OHCI on some
PPC hardware.  The PCI layer should doesn't have the hooks needed for
such ASIC-specific hooks (in this case, software clock gating), so
this moves the code to do that into hcd-pci.c ... where it can be
done after the relevant PCI PM state transition (to/from D3).

Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] USB: EHCI updates split init/reinit logic for resume
David Brownell [Wed, 23 Nov 2005 23:45:37 +0000 (15:45 -0800)]
[PATCH] USB: EHCI updates split init/reinit logic for resume

Moving the PCI-specific parts of the EHCI driver into their own file
created a few issues ... notably on resume paths which (like swsusp)
require re-initializing the controller.  This patch:

 - Splits the EHCI startup code into run-once HCD setup code and
   separate "init the hardware" reinit code.  (That reinit code is
   a superset of the "early usb handoff" code.)

 - Then it makes the PCI init code run both, and the resume code only
   run the reinit code.

 - It also removes needless pci wrappers around EHCI start/stop methods.

 - Removes a byteswap issue that would be seen on big-endian hardware.

The HCD glue still doesn't actually provide a good way to do all this
run-one init stuff in one place though.

Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] USB: EHCI updates mostly whitespace cleanups
David Brownell [Wed, 23 Nov 2005 23:45:32 +0000 (15:45 -0800)]
[PATCH] USB: EHCI updates mostly whitespace cleanups

This cleans up the recent updates to EHCI PCI support:

  - Gets rid of checks for "is this a PCI device", they're no
    longer needed since this is now all PCI-only code.

  - Reduce log spamming:  MWI is only interesting in the atypical
    case that it can actually be used.

  - Whitespace cleanup, as appropriate for a new file with no
    other pending patches.

So other than that minor logging change, no functional updates.

Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] USB: EHCI updates
David Brownell [Wed, 23 Nov 2005 23:45:28 +0000 (15:45 -0800)]
[PATCH] USB: EHCI updates

This fixes some bugs in EHCI suspend/resume that joined us over the past
few releases (as usbcore, PCI, pmcore, and other components evolved):

  - Removes suspend and resume recursion from the EHCI driver, getting
    rid of the USB_SUSPEND special casing.

  - Updates the wakeup mechanism to work again; there's a newish usbcore
    call it needs to use.

  - Provide simpler tests for "do we need to restart from scratch", to
    address another case where PCI Vaux was lost.  (In this case it was
    restoring a swsusp snapshot, but there could be others.)

Un-exports a symbol that was temporarily exported.

A notable change from previous version is that this doesn't move
the spinlock init, so there's still a resume/reinit path bug.

Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] USB: ftdi_sio: new IDs for KOBIL devices
Ian Abbott [Wed, 23 Nov 2005 23:45:23 +0000 (15:45 -0800)]
[PATCH] USB: ftdi_sio: new IDs for KOBIL devices

This patch adds two new devices to the ftdi_sio driver's device ID
table.  The device IDs were supplied by Stefan Nies of KOBIL Systems for
two of their devices using the FTDI chip.

Signed-off-by: Ian Abbott <abbotti@mev.co.uk>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] USB: SN9C10x driver - bad page state fix
Damian Wrobel [Wed, 23 Nov 2005 23:45:17 +0000 (15:45 -0800)]
[PATCH] USB: SN9C10x driver - bad page state fix

This patch solves the following problem I've already discovered on the
latest 2.6.15-rc1-git1 kernel:

Nov 13 07:37:28 wrobel kernel: Bad page state at free_hot_cold_page (in process 'motion', page c164e020)
Nov 13 07:37:28 wrobel kernel: flags:0x40000400 mapping:00000000 mapcount:0 count:0
Nov 13 07:37:28 wrobel kernel: Backtrace:
Nov 13 07:37:28 wrobel kernel:  [<c0146d86>] bad_page+0x85/0xbe
Nov 13 07:37:28 wrobel kernel:  [<c0147629>] free_hot_cold_page+0x54/0x129
Nov 13 07:37:28 wrobel kernel:  [<c01598c6>] __vunmap+0xa9/0xfe
Nov 13 07:37:28 wrobel kernel:  [<c0154114>] vmalloc_to_page+0x34/0x55
Nov 13 07:37:28 wrobel kernel:  [<c0159942>] vfree+0x27/0x35
Nov 13 07:37:28 wrobel kernel:  [<f8a20292>]  sn9c102_release_buffers+0x30/0x3f [sn9c102]
Nov 13 07:37:28 wrobel kernel:  [<f8a231c2>] sn9c102_release+0x37/0xeb [sn9c102]
Nov 13 07:37:28 wrobel kernel:  [<c0163e74>] __fput+0xa9/0x1aa
Nov 13 07:37:28 wrobel kernel:  [<c01624f7>] filp_close+0x49/0x6d
Nov 13 07:37:30 wrobel kernel:  [<c016258f>] sys_close+0x74/0x95
Nov 13 07:37:30 wrobel kernel:  [<c0102ef9>] syscall_call+0x7/0xb
Nov 13 07:37:31 wrobel kernel: Trying to fix it up, but a reboot is needed

Signed-off-by: Damian Wrobel <dwrobel@ertel.com.pl>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] PCI: direct.c: DBG
Daniel Marjamäki [Wed, 23 Nov 2005 23:45:09 +0000 (15:45 -0800)]
[PATCH] PCI: direct.c: DBG

The DBG() call where updated with the appropriate KERN_* symbol.

Signed-off-by: Daniel Marjamäki <daniel.marjamaki@comhem.se>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] PCI: kernel-doc fix for pci-acpi.c
Randy Dunlap [Wed, 23 Nov 2005 23:45:04 +0000 (15:45 -0800)]
[PATCH] PCI: kernel-doc fix for pci-acpi.c

Fix kernel-doc warning in pci/pci-acpi.c.

Signed-off-by: Randy Dunlap <rdunlap@xenotime.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] PCI: remove bogus resource collision error
Rajesh Shah [Wed, 23 Nov 2005 23:44:59 +0000 (15:44 -0800)]
[PATCH] PCI: remove bogus resource collision error

When attempting to hotadd a PCI card with a bridge on it, I saw
the kernel reporting resource collision errors even when there were
really no collisions. The problem is that the code doesn't skip
over "invalid" resources with their resource type flag not set.
Others have reported similar problems at boot time and for
non-bridge PCI card hotplug too, where the code flags a
resource collision for disabled ROMs. This patch fixes both
problems.

Signed-off-by: Rajesh Shah <rajesh.shah@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] PCI Express Hotplug: clear sticky power-fault bit
Rajesh Shah [Wed, 23 Nov 2005 23:44:54 +0000 (15:44 -0800)]
[PATCH] PCI Express Hotplug: clear sticky power-fault bit

Per the PCI Express spec, the power-fault-detected bit in the
slot status register can be set anytime hardware detects a power
fault, regardless of whether the slot has a device populated in
it or not. This bit is sticky and must be explicitly cleared.
This patch is needed to allow hot-add after such a power fault
has been detected.

Signed-off-by: Rajesh Shah <rajesh.shah@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] PCI: trivial printk updates in common.c
Daniel Marjamäkia [Wed, 23 Nov 2005 23:44:49 +0000 (15:44 -0800)]
[PATCH] PCI: trivial printk updates in common.c

Modified common.c so it's using the appropriate KERN_* in printk() calls.

Signed-off-by: Daniel Marjamäkia <daniel.marjamaki@comhem.se>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] hwmon: hdaps missing an axis
Eugeniy Meshcheryakov [Wed, 23 Nov 2005 23:44:35 +0000 (15:44 -0800)]
[PATCH] hwmon: hdaps missing an axis

Trivial patch to report both hdaps axises to the joystick device, not
just the X axis.

Signed-off-by: Robert Love <rml@novell.com>
Signed-off-by: Jean Delvare <khali@linux-fr.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] hwmon: Fix missing it87 fan div init
Jean Delvare [Wed, 23 Nov 2005 23:44:31 +0000 (15:44 -0800)]
[PATCH] hwmon: Fix missing it87 fan div init

Fix a bug where setting the low fan speed limits will not work if no
data was ever read through the sysfs interface and the fan clock
dividers have not been explicitely set yet either. The reason is that
data->fan_div[nr] may currently be used before it is initialized from
the chip register values. The fix is to explicitely initialize
data->fan_div[nr] before using it.

Bug reported, and fix tested, by Nicolas Mailhot.

Signed-off-by: Jean Delvare <khali@linux-fr.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] hwmon: Fix lm78 VID conversion
Jean Delvare [Wed, 23 Nov 2005 23:44:26 +0000 (15:44 -0800)]
[PATCH] hwmon: Fix lm78 VID conversion

Fix the lm78 VID reading, which I accidentally broke while making
this driver use the common vid_from_reg function rather than
reimplementing its own in 2.6.14-rc1.

I'm not proud of it, trust me.

Signed-off-by: Jean Delvare <khali@linux-fr.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] hwmon: Fix missing boundary check when setting W83627THF in0 limits
Yuan Mu [Wed, 23 Nov 2005 23:44:21 +0000 (15:44 -0800)]
[PATCH] hwmon: Fix missing boundary check when setting W83627THF in0 limits

Add SENSORS_LIMIT in store VCore limit functions. This fixes a potential
u8 overflow on out-of-range user input.

Signed-off-by: Jean Delvare <khali@linux-fr.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Add more SCM trees to MAINTAINERS
Josh Boyer [Wed, 23 Nov 2005 23:44:15 +0000 (15:44 -0800)]
[PATCH] Add more SCM trees to MAINTAINERS

Greg requested a patch to update MAINTAINERS with more SCM entries.
The patch below is what I've found so far.

Signed-off-by: Josh Boyer <jwboyer@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Clarify T: field in MAINTAINERS
Jody McIntyre [Wed, 23 Nov 2005 23:44:03 +0000 (15:44 -0800)]
[PATCH] Clarify T: field in MAINTAINERS

Pavel Machek points out that for git repos, what we include is not
actually a URL.  It is undesirable to use a URL since git repos can be
accessed in many different ways.

Signed-off-by: Jody McIntyre <scjody@steamballoon.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] MTD git tree location added to MAINTAINERS
Josh Boyer [Wed, 23 Nov 2005 23:43:57 +0000 (15:43 -0800)]
[PATCH] MTD git tree location added to MAINTAINERS

Here's the MTD one.  More later as I find them.

Signed-off-by: Josh Boyer <jwboyer@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Small fixes to driver core
Alan Stern [Wed, 23 Nov 2005 23:43:50 +0000 (15:43 -0800)]
[PATCH] Small fixes to driver core

This patch (as603) makes a few small fixes to the driver core:

Change spin_lock_irq for a klist lock to spin_lock;

Fix reference count leaks;

Minor spelling and formatting changes.

Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Acked-by Patrick Mochel <mochel@digitalimplant.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[BRIDGE]: recompute features when adding a new device
Olaf Rempel [Thu, 24 Nov 2005 03:04:08 +0000 (19:04 -0800)]
[BRIDGE]: recompute features when adding a new device

We must recompute bridge features everytime the list of underlying
devices changes, or we might end up with features that are not
supported by all devices (eg. NETIF_F_TSO)
This patch adds the missing recompute when adding a device to the bridge.

Signed-off-by: Olaf Rempel <razzor@kopf-tisch.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[NETFILTER]: ip_conntrack_netlink.c needs linux/interrupt.h
Benoit Boissinot [Thu, 24 Nov 2005 03:03:46 +0000 (19:03 -0800)]
[NETFILTER]: ip_conntrack_netlink.c needs linux/interrupt.h

net/ipv4/netfilter/ip_conntrack_netlink.c: In function 'ctnetlink_dump_table':
net/ipv4/netfilter/ip_conntrack_netlink.c:409: warning: implicit declaration of function 'local_bh_disable'
net/ipv4/netfilter/ip_conntrack_netlink.c:427: warning: implicit declaration of function 'local_bh_enable'

Signed-off-by: Benoit Boissinot <benoit.boissinot@ens-lyon.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[PATCH] 32bit integer overflow in invalidate_inode_pages2()
Oleg Drokin [Wed, 23 Nov 2005 21:37:47 +0000 (13:37 -0800)]
[PATCH] 32bit integer overflow in invalidate_inode_pages2()

Fix a 32 bit integer overflow in invalidate_inode_pages2_range.

Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] jffs2 debug gcc-2.9x fix
Andrew Morton [Wed, 23 Nov 2005 21:37:47 +0000 (13:37 -0800)]
[PATCH] jffs2 debug gcc-2.9x fix

Work around gcc-2.95.x macro expansion bug.

Cc: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] powerpc: fix for hugepage areas straddling 4GB boundary
David Gibson [Wed, 23 Nov 2005 21:37:45 +0000 (13:37 -0800)]
[PATCH] powerpc: fix for hugepage areas straddling 4GB boundary

Commit 7d24f0b8a53261709938ffabe3e00f88f6498df9 fixed bugs in the ppc64 SLB
miss handler with respect to hugepage handling, and in the process tweaked
the semantics of the hugepage address masks in mm_context_t.

Unfortunately, it left out a couple of necessary changes to go with that
change.  First, the in_hugepage_area() macro was not updated to match,
second prepare_hugepage_range() was not updated to correctly handle
hugepages regions which straddled the 4GB point.

The latter appears only to cause process-hangs when attempting to map such
a region, but the former can cause oopses if a get_user_pages() is
triggered at the wrong point.  This patch addresses both bugs.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Fix crash in unregister_console()
Benjamin Herrenschmidt [Wed, 23 Nov 2005 21:37:44 +0000 (13:37 -0800)]
[PATCH] Fix crash in unregister_console()

If unregister_console() is inadvertently called while no consoles are
registered, it will crash trying to dereference NULL pointer.  It is
necessary to fix that because register_console() provides no indication
that it actually registered the console passed in.  In fact, it may well
decide not to register it based on various things...

(akpm: It'd be better to make register_console() return something and fix the
callers.  All 106 of them...)

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] fix do_wait() vs exec() race
Oleg Nesterov [Wed, 23 Nov 2005 21:37:43 +0000 (13:37 -0800)]
[PATCH] fix do_wait() vs exec() race

When non-leader thread does exec, de_thread adds old leader to the init's
->children list in EXIT_ZOMBIE state and drops tasklist_lock.

This means that release_task(leader) in de_thread() is racy vs do_wait()
from init task.

I think de_thread() should set old leader's state to EXIT_DEAD instead.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: george anzinger <george@mvista.com>
Cc: Roland Dreier <rolandd@cisco.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] kprobes: Fix return probes on sys_execve
Jim Keniston [Wed, 23 Nov 2005 21:37:42 +0000 (13:37 -0800)]
[PATCH] kprobes: Fix return probes on sys_execve

Fix a bug in kprobes that can cause an Oops or even a crash when a return
probe is installed on one of the following functions: sys_execve,
do_execve, load_*_binary, flush_old_exec, or flush_thread.  The fix is to
remove the call to kprobe_flush_task() in flush_thread().  This fix has
been tested on all architectures for which the return-probes feature has
been implemented (i386, x86_64, ppc64, ia64).  Please apply.

BACKGROUND

Up to now, we have called kprobe_flush_task() under two situations: when a
task exits, and when it execs.  Flushing kretprobe_instances on exit is
correct because (a) do_exit() doesn't return, and (b) one or more
return-probed functions may be active when a task calls do_exit().  Neither
is the case for sys_execve() and its callees.

Initially, the mistaken call to kprobe_flush_task() on exec was harmless
because we put the "real" return address of each active probed function
back in the stack, just to be safe, when we recycled its
kretprobe_instance.  When support for ppc64 and ia64 was added, this safety
measure couldn't be employed, and was eventually dropped even for i386 and
x86_64.  sys_execve() and its callees were informally blacklisted for
return probes until this fix was developed.

Acked-by: Prasanna S Panchamukhi <prasanna@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] mm: fill arch atomic64 gaps
Hugh Dickins [Wed, 23 Nov 2005 21:37:40 +0000 (13:37 -0800)]
[PATCH] mm: fill arch atomic64 gaps

alpha, sparc64, x86_64 are each missing some primitives from their atomic64
support: fill in the gaps I've noticed by extrapolating asm, follow the
groupings in each file.  But powerpc and parisc still lack atomic64.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Andi Kleen <ak@muc.de>
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] mm: powerpc init_mm without ptlock
Hugh Dickins [Wed, 23 Nov 2005 21:37:39 +0000 (13:37 -0800)]
[PATCH] mm: powerpc init_mm without ptlock

Restore an earlier mod which went missing in the powerpc reshuffle: the 4xx
mmu_mapin_ram does not need to take init_mm.page_table_lock.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] mm: powerpc ptlock comments
Hugh Dickins [Wed, 23 Nov 2005 21:37:39 +0000 (13:37 -0800)]
[PATCH] mm: powerpc ptlock comments

Update comments (only) on page_table_lock and mmap_sem in arch/powerpc.
Removed the comment on page_table_lock from hash_huge_page: since it's no
longer taking page_table_lock itself, it's irrelevant whether others are; but
how it is safe (even against huge file truncation?) I can't say.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] mm: unbloat get_futex_key
Hugh Dickins [Wed, 23 Nov 2005 21:37:38 +0000 (13:37 -0800)]
[PATCH] mm: unbloat get_futex_key

The follow_page changes in get_futex_key have left it with two almost
identical blocks, when handling the rare case of a futex in a nonlinear vma.
get_user_pages will itself do that follow_page, and its additional
find_extend_vma is hardly any overhead since the vma is already cached.  Let's
just delete the follow_page block and let get_user_pages do it.

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] mm: update split ptlock Kconfig
Hugh Dickins [Wed, 23 Nov 2005 21:37:37 +0000 (13:37 -0800)]
[PATCH] mm: update split ptlock Kconfig

Closer attention to the arithmetic shows that neither ppc64 nor sparc really
uses one page for multiple page tables: how on earth could they, while
pte_alloc_one returns just a struct page pointer, with no offset?

Well, arm26 manages it by returning a pte_t pointer cast to a struct page
pointer, harumph, then compensating in its pmd_populate.  But arm26 is never
SMP, so it's not a problem for split ptlock either.

And the PA-RISC situation has been recently improved: CONFIG_PA20 works
without the 16-byte alignment which inflated its spinlock_t.  But the current
union of spinlock_t with private does make the 7xxx struct page significantly
larger, even without debug, so disable its split ptlock.

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] revert floppy-fix-read-only-handling
Andrew Morton [Wed, 23 Nov 2005 21:37:36 +0000 (13:37 -0800)]
[PATCH] revert floppy-fix-read-only-handling

This fix causes problems on the very first floppy access - we haven't yet
talked to the FDC so we don't know which state the write-protect tab is in.

Revert for now.

Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Check the irq number is within bounds
Matthew Wilcox [Thu, 3 Nov 2005 14:51:18 +0000 (07:51 -0700)]
[PATCH] Check the irq number is within bounds

Most of the functions already check. Do the ones that didn't.

Signed-off-by: Matthew Wilcox <matthew@wil.cx>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years agoRevert "[NET]: Shut up warnings in net/core/flow.c"
Linus Torvalds [Wed, 23 Nov 2005 16:44:05 +0000 (08:44 -0800)]
Revert "[NET]: Shut up warnings in net/core/flow.c"

This reverts commit af2b4079ab154bd12e8c12b02db5f31b31babe63

Changing the #define to an inline function breaks on non-SMP builds,
since wuite a few places in the kernel do not implement the ipi handler
when compiling for UP.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years agodrm: move is_pci to the end of the structure
Dave Airlie [Wed, 23 Nov 2005 11:12:59 +0000 (22:12 +1100)]
drm: move is_pci to the end of the structure

We memset the structure across opens except for the flags. The correct
fix is more intrusive but this should fix a problem with bad iounmaps
seen on AGP radeons acting like PCI ones.

Signed-off-by: Dave Airlie <airlied@linux.ie>
18 years agodrm: add __GFP_COMP to the drm_alloc_pages
Dave Airlie [Wed, 23 Nov 2005 11:09:13 +0000 (22:09 +1100)]
drm: add __GFP_COMP to the drm_alloc_pages

The DRM only uses drm_alloc_pages for non-SG PCI cards using DRM.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Dave Airlie <airlied@linux.ie>
18 years agoI think that if a PCI bus is a root bus, attached to a host bridge not a
Dave Airlie [Wed, 23 Nov 2005 10:45:43 +0000 (21:45 +1100)]
I think that if a PCI bus is a root bus, attached to a host bridge not a
PCI->PCI bridge, then bus->self is allowed to be NULL. Certainly that's
the case on my Pegasos, and it makes the MGA DRM driver oops...

Signed-off-by: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Dave Airlie <airlied@linux.ie>
18 years agocompat-ioctl.c: fix compile with no CONFIG_JBD
Linus Torvalds [Wed, 23 Nov 2005 05:58:37 +0000 (21:58 -0800)]
compat-ioctl.c: fix compile with no CONFIG_JBD

The ext3 compat-ioctl translation wants to translate data structures
that <linux/jbd.h> only declared when CONFIG_JBD was enabled.

So make <linux/jbd.h> play nicely even when we don't actually end up
using it.

Acked-by: Andrew Morton <akpm@osdl.org>
Acked-by: Jeffrey Hundstad <jeffrey.hundstad@mnsu.edu>
Acked-by: Zan Lynx <zlynx@acm.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years agoFix up GFP_ZONEMASK for GFP_DMA32 usage
Linus Torvalds [Wed, 23 Nov 2005 03:39:30 +0000 (19:39 -0800)]
Fix up GFP_ZONEMASK for GFP_DMA32 usage

There was some confusion about the different zone usage, this should fix
up the resulting mess in the GFP zonemask handling.

The different zone usage is still confusing (it's very easy to mix up
the individual zone numbers with the GFP zone _list_ numbers), so we
might want to clean up some of this in the future, but in the meantime
this should fix the actual problems.

Acked-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6
Linus Torvalds [Wed, 23 Nov 2005 00:51:33 +0000 (16:51 -0800)]
Merge /pub/scm/linux/kernel/git/davem/sparc-2.6

18 years ago[SPARC]: drivers/sbus/char/aurora.c: "extern inline" -> "static inline"
Adrian Bunk [Tue, 22 Nov 2005 23:30:29 +0000 (15:30 -0800)]
[SPARC]: drivers/sbus/char/aurora.c:  "extern inline" -> "static inline"

"extern inline" doesn't make much sense.

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[NET]: Fix ifenslave to not fail on lack of IP information
Neil Horman [Tue, 22 Nov 2005 22:56:32 +0000 (14:56 -0800)]
[NET]: Fix ifenslave to not fail on lack of IP information

Patch to ifenslave so that under older ABI versions, a failure to propogate ip
information from master to slave does not result in a filure to enslave the
slave device.

Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[NETFILTER] ctnetlink: Fix refcount leak ip_conntrack/nat_proto
Pablo Neira Ayuso [Tue, 22 Nov 2005 22:54:34 +0000 (14:54 -0800)]
[NETFILTER] ctnetlink: Fix refcount leak ip_conntrack/nat_proto

Remove proto == NULL checking since ip_conntrack_[nat_]proto_find_get
always returns a valid pointer.

Fix missing ip_conntrack_proto_put in some paths.

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Harald Welte <laforge@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[IPV4]: Fix secondary IP addresses after promotion
Jamal Hadi Salim [Tue, 22 Nov 2005 22:47:37 +0000 (14:47 -0800)]
[IPV4]: Fix secondary IP addresses after promotion

This patch fixes the problem with promoting aliases when:
a) a single primary and > 1 secondary addresses
b) multiple primary addresses each with at least one secondary address

Based on earlier efforts from Brian Pomerantz <bapper@piratehaven.org>,
Patrick McHardy <kaber@trash.net> and Thomas Graf <tgraf@suug.ch>

Signed-off-by: Jamal Hadi Salim <hadi@cyberus.ca>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[NETLINK]: Use tgid instead of pid for nlmsg_pid
Herbert Xu [Tue, 22 Nov 2005 22:41:50 +0000 (14:41 -0800)]
[NETLINK]: Use tgid instead of pid for nlmsg_pid

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[NET]: Shut up warnings in net/core/flow.c
Russell King [Tue, 22 Nov 2005 22:38:04 +0000 (14:38 -0800)]
[NET]: Shut up warnings in net/core/flow.c

Not really a network problem, more a !SMP issue.

net/core/flow.c:295: warning: statement with no effect

flow.c:295:        smp_call_function(flow_cache_flush_per_cpu, &info, 1, 0);

Fix this by converting the macro to an inline function, which
also increases the typechecking for !SMP builds.

Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
18 years ago[PATCH] prefer pkg-config for the QT check
Roman Zippel [Tue, 22 Nov 2005 05:32:38 +0000 (21:32 -0800)]
[PATCH] prefer pkg-config for the QT check

This makes pkg-config now the prefered way to configure QT and properly
fixes the recent Fedora breakage and leaves the old QT detection as
fallback mechanism.

Signed-off-by: Roman Zippel <zippel@linux-m68k.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] device-mapper raid1: drop mark_region spinlock fix
Jonathan E Brassow [Tue, 22 Nov 2005 05:32:37 +0000 (21:32 -0800)]
[PATCH] device-mapper raid1: drop mark_region spinlock fix

The spinlock region_lock is held while calling mark_region which can sleep.
Drop the spinlock before calling that function.

A region's state and inclusion in the clean list are altered by rh_inc and
rh_dec.  The state variable is set to RH_CLEAN in rh_dec, but only if
'pending' is zero.  It is set to RH_DIRTY in rh_inc, but not if it is already
so.  The changes to 'pending', the state, and the region's inclusion in the
clean list need to be atomicly.

Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] device-mapper snapshot: bio_list fix
jblunck@suse.de [Tue, 22 Nov 2005 05:32:36 +0000 (21:32 -0800)]
[PATCH] device-mapper snapshot: bio_list fix

bio_list_merge() should do nothing if the second list is empty - not oops.

Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] device-mapper dm-mpath: endio spinlock fix
Stefan Bader [Tue, 22 Nov 2005 05:32:35 +0000 (21:32 -0800)]
[PATCH] device-mapper dm-mpath: endio spinlock fix

do_end_io() can be called without interrupts blocked.

Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] device-mapper: mirror log bitset fix
Alasdair G Kergon [Tue, 22 Nov 2005 05:32:34 +0000 (21:32 -0800)]
[PATCH] device-mapper: mirror log bitset fix

The linux bitset operators (test_bit, set_bit etc) work on arrays of "unsigned
long".  dm-log uses such bitsets but treats them as arrays of uint32_t, only
allocating and zeroing a multiple of 4 bytes (as 'clean_bits' is a uint32_t).

The patch below fixes this problem.

The problem is specific to 64-bit big endian machines such as s390x or ppc-64
and can prevent pvmove terminating.

In the simplest case, if "region_count" were (say) 30, then
bitset_size (below) would be 4 and bitset_uint32_count would be 1.
Thus the memory for this butset, after allocation and zeroing would
be
   0 0 0 0 X X X X
On a bigendian 64bit machine, bit 0 for this bitset is in the 8th
byte! (and every bit that dm-log would use would be in the X area).

   0 0 0 0 X X X X
                 ^
                 here

which hasn't been cleared properly.

As the dm-raid1 code only syncs and counts regions which have a 0 in the
'sync_bits' bitset, and only finishes when it has counted high enough, a large
number of 1's among those 'X's will cause the sync to not complete.

It is worth noting that the code uses the same bitsets for in-memory and
on-disk logs.  As these bitsets are host-endian and host-sized, this means
that they cannot safely be moved between computers with

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] device-mapper: list_versions fix
Alasdair G Kergon [Tue, 22 Nov 2005 05:32:33 +0000 (21:32 -0800)]
[PATCH] device-mapper: list_versions fix

In some circumstances the LIST_VERSIONS output is truncated because the size
calculation forgets about a 'uint32_t' in each structure - but the inclusion
of the whole of ALIGN_MASK frequently compensates for the omission.

This is a quick workaround to use an upper bound.  (The code ought to be fixed
to supply the actual size.)

Running 'dmsetup targets' may demonstrate the problem: when I run it, the last
line comes out as 'erro' instead of 'error'.  Consequently, 'lvcreate --type
error' doesn't work.

Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] device-mapper dm-ioctl: missing put in table load error case
Kiyoshi Ueda [Tue, 22 Nov 2005 05:32:32 +0000 (21:32 -0800)]
[PATCH] device-mapper dm-ioctl: missing put in table load error case

An error path in table_load() forgets to release a table that won't now be
referenced.

Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] kernel Doc/ URL corrections
Randy Dunlap [Tue, 22 Nov 2005 05:32:31 +0000 (21:32 -0800)]
[PATCH] kernel Doc/ URL corrections

Correct lots of URLs in Documentation/ Also a few minor whitespace cleanups
and typo/spello fixes.  Sadly there are still a lot of bad URLs remaining.

Signed-off-by: Randy Dunlap <rdunlap@xenotime.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] dell_rbu driver depends on x86[64]
Dave Jones [Tue, 22 Nov 2005 05:32:30 +0000 (21:32 -0800)]
[PATCH] dell_rbu driver depends on x86[64]

This driver only appears on IA32 & EM64T boxes.

Signed-off-by: Dave Jones <davej@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Fix a bug in scsi_get_command
Matthew Dobson [Tue, 22 Nov 2005 05:32:29 +0000 (21:32 -0800)]
[PATCH] Fix a bug in scsi_get_command

scsi_get_command() attempts to write into a structure that may not have
been successfully allocated.  Move this write inside the if statement that
ensures we won't panic the kernel with a NULL pointer dereference.

Signed-off-by: Matthew Dobson <colpatch@us.ibm.com>
Cc: 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] cpufreq: silence cpufreq for UP
Grant Coady [Tue, 22 Nov 2005 05:32:28 +0000 (21:32 -0800)]
[PATCH] cpufreq: silence cpufreq for UP

drivers/cpufreq/cpufreq.c: In function `cpufreq_remove_dev':
drivers/cpufreq/cpufreq.c:696: warning: unused variable `cpu_sys_dev'

Signed-off-by: Grant Coady <gcoady@gmail.com>
Cc: Dave Jones <davej@codemonkey.org.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] hugetlb: fix race in set_max_huge_pages for multiple updaters of nr_huge_pages
Eric Paris [Tue, 22 Nov 2005 05:32:28 +0000 (21:32 -0800)]
[PATCH] hugetlb: fix race in set_max_huge_pages for multiple updaters of nr_huge_pages

If there are multiple updaters to /proc/sys/vm/nr_hugepages simultaneously
it is possible for the nr_huge_pages variable to become incorrect.  There
is no locking in the set_max_huge_pages function around
alloc_fresh_huge_page which is able to update nr_huge_pages.  Two callers
to alloc_fresh_huge_page could race against each other as could a call to
alloc_fresh_huge_page and a call to update_and_free_page.  This patch just
expands the area covered by the hugetlb_lock to cover the call into
alloc_fresh_huge_page.  I'm not sure how we could say that a sysctl section
is performance critical where more specific locking would be needed.

My reproducer was to run a couple copies of the following script
simultaneously

while [ true ]; do
echo 1000 > /proc/sys/vm/nr_hugepages
echo 500 > /proc/sys/vm/nr_hugepages
echo 750 > /proc/sys/vm/nr_hugepages
echo 100 > /proc/sys/vm/nr_hugepages
echo 0 > /proc/sys/vm/nr_hugepages
done

and then watch /proc/meminfo and eventually you will see things like

HugePages_Total:     100
HugePages_Free:      109

After applying the patch all seemed well.

Signed-off-by: Eric Paris <eparis@redhat.com>
Acked-by: William Irwin <wli@holomorphy.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] vgacon: Fix usage of stale height value on vc initialization
Antonino A. Daplas [Tue, 22 Nov 2005 05:32:26 +0000 (21:32 -0800)]
[PATCH] vgacon: Fix usage of stale height value on vc initialization

Reported by: Wayne E. Harlan

"[1.] One line summary of the problem:
When the kernel option "vga=1" is used, additional tty's (alt+control+Fx
with x=2,3,4,5, etc) do not provide the full 50 lines of output.  The first
one does have 50 lines, however.

[2.] Full description of the problem/report:
These addtitional tty's show only 39 lines plus the top pixel of the 40-th
line.  The remaining lines are black and not shown.  Kernel version
2.6.13.4 does not show this problem."

This bug is caused by using a stale font height value on vgacon_init.

Booting with vga=1 gives an 80x50 screen with an 8x8 font.  Somewhere
during the initialization, the font was changed to 8x9 and the first
vc was correctly resized to 80x44.  However, the rest of the vc's were
not allocated yet, and when they were subsequently initialized, they
still used a font height of 8 (instead of 9) causing the mentioned bug.

Fix by saving the new font height to vga_video_font_height.

Signed-off-by: Antonino Daplas <adaplas@pol.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] fbcon: Console Rotation - Fix wrong shift calculation
Antonino A. Daplas [Tue, 22 Nov 2005 05:32:25 +0000 (21:32 -0800)]
[PATCH] fbcon: Console Rotation - Fix wrong shift calculation

The shift value (amount to shift the bitmap so first pixel starts at
origin(0,0)) is incorrect.  This causes corrupted characters or a kernel crash
if fontwidth is not divisible by 8 at 270 degrees, or fontheight not divisible
by 8 at 180 degrees.

Report and part of the fix contributed by Knut Petersen.

Signed-off-by: Antonino Daplas <adaplas@pol.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Fix hugetlbfs_statfs() reporting of block limits
David Gibson [Tue, 22 Nov 2005 05:32:24 +0000 (21:32 -0800)]
[PATCH] Fix hugetlbfs_statfs() reporting of block limits

Currently, if a hugetlbfs is mounted without limits (the default), statfs()
will return -1 for max/free/used blocks.  This does not appear to be in
line with normal convention: simple_statfs() and shmem_statfs() both return
0 in similar cases.  Worse, it confuses the translation logic in
put_compat_statfs(), causing it to return -EOVERFLOW on such a mount.

This patch alters hugetlbfs_statfs() to return 0 for max/free/used blocks
on a mount without limits.  Note that we need the test in the patch below,
rather than just using 0 in the sbinfo structure, because the -1 marked in
the free blocks field is used internally to tell the

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Fix error handling with put_compat_statfs()
David Gibson [Tue, 22 Nov 2005 05:32:23 +0000 (21:32 -0800)]
[PATCH] Fix error handling with put_compat_statfs()

In fs/compat.c, whenever put_compat_statfs() returns an error, the
containing syscall returns -EFAULT.  This is presumably by analogy with the
non-compat case, where any non-zero code from copy_to_user() should be
translated into an EFAULT.  However, put_compat_statfs() is also return
-EOVERFLOW.  The same applies for put_compat_statfs64().

This bug can be observed with a statfs() on a hugetlbfs directory.
hugetlbfs, when mounted without limits reports available, free and total
blocks as -1 (itself a bug, another patch coming).  statfs() will
mysteriously return EFAULT although it's parameters are perfectly valid
addresses.

This patch causes the compat versions of statfs() and statfs64() to
correctly propogate the return values from put_compat_statfs() and
put_compat_statfs64().

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] unpaged: fix sound Bad page states
Hugh Dickins [Tue, 22 Nov 2005 05:32:22 +0000 (21:32 -0800)]
[PATCH] unpaged: fix sound Bad page states

Earlier I unifdefed PageCompound, so that snd_pcm_mmap_control_nopage and
others can give out a 0-order component of a higher-order page, which won't
be mistakenly freed when zap_pte_range unmaps it.  But many Bad page states
reported a PG_reserved was freed after all: I had missed that we need to
say __GFP_COMP to get compound page behaviour.

Some of these higher-order pages are allocated by snd_malloc_pages, some by
snd_malloc_dev_pages; or if SBUS, by sbus_alloc_consistent - but that has
no gfp arg, so add __GFP_COMP into its sparc32/64 implementations.

I'm still rather puzzled that DRM seems not to need a similar change.

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] unpaged: copy_page_range vma
Hugh Dickins [Tue, 22 Nov 2005 05:32:20 +0000 (21:32 -0800)]
[PATCH] unpaged: copy_page_range vma

For copy_one_pte's print_bad_pte to show the task correctly (instead of
"???"), dup_mmap must pass down parent vma rather than child vma.

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] unpaged: PG_reserved bad_page
Hugh Dickins [Tue, 22 Nov 2005 05:32:20 +0000 (21:32 -0800)]
[PATCH] unpaged: PG_reserved bad_page

It used to be the case that PG_reserved pages were silently never freed, but
in 2.6.15-rc1 they may be freed with a "Bad page state" message.  We should
work through such cases as they appear, fixing the code; but for now it's
safer to issue the message without freeing the page, leaving PG_reserved set.

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] unpaged: ZERO_PAGE in VM_UNPAGED
Hugh Dickins [Tue, 22 Nov 2005 05:32:19 +0000 (21:32 -0800)]
[PATCH] unpaged: ZERO_PAGE in VM_UNPAGED

It's strange enough to be looking out for anonymous pages in VM_UNPAGED areas,
let's not insert the ZERO_PAGE there - though whether it would matter will
depend on what we decide about ZERO_PAGE refcounting.

But whereas do_anonymous_page may (exceptionally) be called on a VM_UNPAGED
area, do_no_page should never be: just BUG_ON.

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] unpaged: anon in VM_UNPAGED
Hugh Dickins [Tue, 22 Nov 2005 05:32:18 +0000 (21:32 -0800)]
[PATCH] unpaged: anon in VM_UNPAGED

copy_one_pte needs to copy the anonymous COWed pages in a VM_UNPAGED area,
zap_pte_range needs to free them, do_wp_page needs to COW them: just like
ordinary pages, not like the unpaged.

But recognizing them is a little subtle: because PageReserved is no longer a
condition for remap_pfn_range, we can now mmap all of /dev/mem (whether the
distro permits, and whether it's advisable on this or that architecture, is
another matter).  So if we can see a PageAnon, it may not be ours to mess with
(or may be ours from elsewhere in the address space).  I suspect there's an
entertaining insoluble self-referential problem here, but the page_is_anon
function does a good practical job, and MAP_PRIVATE PROT_WRITE VM_UNPAGED will
always be an odd choice.

In updating the comment on page_address_in_vma, noticed a potential NULL
dereference, in a path we don't actually take, but fixed it.

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] unpaged: COW on VM_UNPAGED
Hugh Dickins [Tue, 22 Nov 2005 05:32:17 +0000 (21:32 -0800)]
[PATCH] unpaged: COW on VM_UNPAGED

Remove the BUG_ON(vma->vm_flags & VM_UNPAGED) from do_wp_page, and let it do
Copy-On-Write without touching the VM_UNPAGED's page counts - but this is
incomplete, because the anonymous page it inserts will itself need to be
handled, here and in other functions - next patch.

We still don't copy the page if the pfn is invalid, because the
copy_user_highpage interface does not allow it.  But that's not been a problem
in the past: can be added in later if the need arises.

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] unpaged: VM_NONLINEAR VM_RESERVED
Hugh Dickins [Tue, 22 Nov 2005 05:32:16 +0000 (21:32 -0800)]
[PATCH] unpaged: VM_NONLINEAR VM_RESERVED

There's one peculiar use of VM_RESERVED which the previous patch left behind:
because VM_NONLINEAR's try_to_unmap_cluster uses vm_private_data as a swapout
cursor, but should never meet VM_RESERVED vmas, it was a way of extending
VM_NONLINEAR to VM_RESERVED vmas using vm_private_data for some other purpose.
 But that's an empty set - they don't have the populate function required.  So
just throw away those VM_RESERVED tests.

But one more interesting in rmap.c has to go too: try_to_unmap_one will want
to swap out an anonymous page from VM_RESERVED or VM_UNPAGED area.

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] unpaged: VM_UNPAGED
Hugh Dickins [Tue, 22 Nov 2005 05:32:15 +0000 (21:32 -0800)]
[PATCH] unpaged: VM_UNPAGED

Although we tend to associate VM_RESERVED with remap_pfn_range, quite a few
drivers set VM_RESERVED on areas which are then populated by nopage.  The
PageReserved removal in 2.6.15-rc1 changed VM_RESERVED not to free pages in
zap_pte_range, without changing those drivers not to set it: so their pages
just leak away.

Let's not change miscellaneous drivers now: introduce VM_UNPAGED at the core,
to flag the special areas where the ptes may have no struct page, or if they
have then it's not to be touched.  Replace most instances of VM_RESERVED in
core mm by VM_UNPAGED.  Force it on in remap_pfn_range, and the sparc and
sparc64 io_remap_pfn_range.

Revert addition of VM_RESERVED to powerpc vdso, it's not needed there.  Is it
needed anywhere?  It still governs the mm->reserved_vm statistic, and special
vmas not to be merged, and areas not to be core dumped; but could probably be
eliminated later (the drivers are probably specifying it because in 2.4 it
kept swapout off the vma, but in 2.6 we work from the LRU, which these pages
don't get on).

Use the VM_SHM slot for VM_UNPAGED, and define VM_SHM to 0: it serves no
purpose whatsoever, and should be removed from drivers when we clean up.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Acked-by: William Irwin <wli@holomorphy.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] unpaged: unifdefed PageCompound
Hugh Dickins [Tue, 22 Nov 2005 05:32:14 +0000 (21:32 -0800)]
[PATCH] unpaged: unifdefed PageCompound

It looks like snd_xxx is not the only nopage to be using PageReserved as a way
of holding a high-order page together: which no longer works, but is masked by
our failure to free from VM_RESERVED areas.  We cannot fix that bug without
first substituting another way to hold the high-order page together, while
farming out the 0-order pages from within it.

That's just what PageCompound is designed for, but it's been kept under
CONFIG_HUGETLB_PAGE.  Remove the #ifdefs: which saves some space (out- of-line
put_page), doesn't slow down what most needs to be fast (already using
hugetlb), and unifies the way we handle high-order 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] unpaged: sound nopage get_page
Hugh Dickins [Tue, 22 Nov 2005 05:32:13 +0000 (21:32 -0800)]
[PATCH] unpaged: sound nopage get_page

Something noticed when studying use of VM_RESERVED in different drivers:
snd_usX2Y_hwdep_pcm_vm_nopage omitted to get_page: fixed.

And how did this work before?  Aargh!  That nopage is returning a page from
within a buffer allocated by snd_malloc_pages, which allocates a high-order
page, then does SetPageReserved on each 0-order page within.

That would have worked in 2.6.14, because when the area was unmapped,
PageReserved inhibited put_page.  2.6.15-rc1 removed that inhibition (while
leaving ineffective PageReserveds around for now), but it hasn't caused
trouble because..  we've not been freeing from VM_RESERVED at all.

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] unpaged: private write VM_RESERVED
Hugh Dickins [Tue, 22 Nov 2005 05:32:12 +0000 (21:32 -0800)]
[PATCH] unpaged: private write VM_RESERVED

The PageReserved removal in 2.6.15-rc1 issued a "deprecated" message when you
tried to mmap or mprotect MAP_PRIVATE PROT_WRITE a VM_RESERVED, and failed
with -EACCES: because do_wp_page lacks the refinement to COW pages in those
areas, nor do we expect to find anonymous pages in them; and it seemed just
bloat to add code for handling such a peculiar case.  But immediately it
caused vbetool and ddcprobe (using lrmi) to fail.

So revert the "deprecated" messages, letting mmap and mprotect succeed.  But
leave do_wp_page's BUG_ON(vma->vm_flags & VM_RESERVED) in place until we've
added the code to do it right: so this particular patch is only good if the
app doesn't really need to write to that private area.

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] unpaged: get_user_pages VM_RESERVED
Hugh Dickins [Tue, 22 Nov 2005 05:32:11 +0000 (21:32 -0800)]
[PATCH] unpaged: get_user_pages VM_RESERVED

The PageReserved removal in 2.6.15-rc1 prohibited get_user_pages on the areas
flagged VM_RESERVED in place of PageReserved.  That is correct in theory - we
ought not to interfere with struct pages in such a reserved area; but in
practice it broke BTTV for one.

So revert to prohibiting only on VM_IO: if someone gets into trouble with
get_user_pages on VM_RESERVED, it'll just be a "don't do that".

You can argue that videobuf_mmap_mapper shouldn't set VM_RESERVED in the first
place, but now's not the time for breaking drivers without notice.

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] uml: eliminate use of libc PAGE_SIZE
Jeff Dike [Tue, 22 Nov 2005 05:32:10 +0000 (21:32 -0800)]
[PATCH] uml: eliminate use of libc PAGE_SIZE

On some systems, libc PAGE_SIZE calls getpagesize, which can't happen from a
stub.  So, I use UM_KERN_PAGE_SIZE, which is less variable in its definition,
instead.

Signed-off-by: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] uml: properly invoke x86_64 system calls
Jeff Dike [Tue, 22 Nov 2005 05:32:09 +0000 (21:32 -0800)]
[PATCH] uml: properly invoke x86_64 system calls

This patch makes stub_segv use the stub_syscall macros.  This was needed
anyway, but the bug that prompted this was the discovery that gcc was storing
stuff in RCX, which is trashed across a system call.  This is exactly the sort
of problem that the new macros fix.

There is a stub_syscall0 for getpid.  stub_segv was changed to be a libc file,
and that caused some include changes.

Signed-off-by: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] uml: eliminate anonymous union and clean up symlink lossage
Jeff Dike [Tue, 22 Nov 2005 05:32:08 +0000 (21:32 -0800)]
[PATCH] uml: eliminate anonymous union and clean up symlink lossage

This gives a name to the anonymous union introduced in skas-hold-own-ldt,
allowing to build on a wider range of gccs.

It also removes ldt.h, which somehow became real, and replaces it with a
symlink, and creates ldt-x86_64.h as a copy of ldt-i386.h for now.

Signed-off-by: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] uml: eliminate use of local in clone stub
Jeff Dike [Tue, 22 Nov 2005 05:32:04 +0000 (21:32 -0800)]
[PATCH] uml: eliminate use of local in clone stub

We have a bug in the i386 stub_syscall6 which pushes ebp before the system
call and pops it afterwards.  Because we use syscall6 to remap the stack, the
old contents of the stack (and the former value of ebp) are no longer
available.  Some versions of gcc make from a real local, accessed through ebp,
despite my efforts to make it obvious that references to from are really
constants.  This patch attempts to make it even more obvious by eliminating
from and using a macro to access the stub's data explicitly with constants.

My original thinking on this was to replace syscall6 with a remap_stack
interface which saved ebp someplace and restored it afterwards.  The problem
is that there are no registers to put it in, except for esp.  That could work,
since we can store a constant in esp after the mmap because we just replaced
the stack.  However, this approach seems a tad cleaner.

Signed-off-by: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] net: Fix compiler-error on dgrs.c when !CONFIG_PCI
Richard Knutsson [Tue, 22 Nov 2005 05:32:03 +0000 (21:32 -0800)]
[PATCH] net: Fix compiler-error on dgrs.c when !CONFIG_PCI

drivers/net/dgrs.c: In function `dgrs_init_module':
drivers/net/dgrs.c:1598: `dgrs_pci_driver' undeclared (first use in this function)

Signed-off-by: Richard Knutsson <ricknu-0@student.ltu.se>
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/scjody...
Linus Torvalds [Tue, 22 Nov 2005 03:04:44 +0000 (19:04 -0800)]
Merge branch 'for-linus' of git://git./linux/kernel/git/scjody/ieee1394

18 years agoMerge master.kernel.org:/home/rmk/linux-2.6-serial
Linus Torvalds [Tue, 22 Nov 2005 03:00:43 +0000 (19:00 -0800)]
Merge master.kernel.org:/home/rmk/linux-2.6-serial

18 years agoMerge master.kernel.org:/home/rmk/linux-2.6-arm
Linus Torvalds [Tue, 22 Nov 2005 03:00:05 +0000 (19:00 -0800)]
Merge master.kernel.org:/home/rmk/linux-2.6-arm

18 years agoMerge branch 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6
Linus Torvalds [Tue, 22 Nov 2005 01:44:56 +0000 (17:44 -0800)]
Merge branch 'release' of git://git./linux/kernel/git/aegl/linux-2.6

18 years ago[IA64-SGI] bte_copy nasid_index fix
Russ Anderson [Fri, 11 Nov 2005 22:52:02 +0000 (16:52 -0600)]
[IA64-SGI] bte_copy nasid_index fix

The nasid_index was not being incremented if the
pointer was null, causing an infinite loop.

Signed-off-by: Russ Anderson (rja@sgi.com)
Signed-off-by: Tony Luck <tony.luck@intel.com>
18 years ago[IA64-SGI] support for older versions of PROM
Jack Steiner [Fri, 18 Nov 2005 22:11:27 +0000 (16:11 -0600)]
[IA64-SGI] support for older versions of PROM

Add support for old versions of the SN PROMs. Eventually this
support will be deleted but it is useful right now to continue
supporting older PROMs.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
18 years ago[IA64] fix bug in sn/ia64 for sparse CPU numbering
hawkes@sgi.com [Fri, 18 Nov 2005 19:30:34 +0000 (11:30 -0800)]
[IA64] fix bug in sn/ia64 for sparse CPU numbering

The kernel's use of the for_each_*cpu(i) macros has allowed for sparse CPU
numbering.  When I hacked the kernel to test sparse cpu_present_map[] and
cpu_possible_map[] cpumasks, I discovered one remaining spot, in
sn_hwperf_ioctl() during sn initialization, that needs to be fixed.

Signed-off-by: John Hawkes <hawkes@sgi.com>
Signed-off-by: Dean Roe <roe@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
18 years ago[IA64] Prevent sn2 ptc code from executing on all ia64 subarches
Prarit Bhargava [Mon, 21 Nov 2005 18:00:38 +0000 (13:00 -0500)]
[IA64] Prevent sn2 ptc code from executing on all ia64 subarches

Patch to prevent sn2_ptc_init code from attempting to load on non-sn2 systems
when sn2_smp.c is built-in to generic kernel.

Signed-off-by: Prarit Bhargava <prarit@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
18 years ago[PATCH] as-iosched: remove state assertion in as_add_request()
Jens Axboe [Mon, 21 Nov 2005 18:49:41 +0000 (19:49 +0100)]
[PATCH] as-iosched: remove state assertion in as_add_request()

Kill the arq->state poison statement in as_add_request(), it can trigger
for perfectly valid code that just reuses a request after io completion
instead of freeing it and allocating a new one. We probably should
introduce a blk_init_request() to start from scratch, but for now just
kill it as we will be removing the as specific poisoning soon.

Signed-off-by: Jens Axboe <axboe@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/davej/agpgart
Linus Torvalds [Mon, 21 Nov 2005 18:56:30 +0000 (10:56 -0800)]
Merge /pub/scm/linux/kernel/git/davej/agpgart

18 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
Linus Torvalds [Mon, 21 Nov 2005 18:46:34 +0000 (10:46 -0800)]
Merge /pub/scm/linux/kernel/git/davem/net-2.6

18 years ago[SERIAL] imx: Fix missed platform_driver_unregister
Russell King [Mon, 21 Nov 2005 17:05:21 +0000 (17:05 +0000)]
[SERIAL] imx: Fix missed platform_driver_unregister

Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
18 years ago[ARM] Shut up gcc warning in clps7500 core.c
Russell King [Mon, 21 Nov 2005 17:03:15 +0000 (17:03 +0000)]
[ARM] Shut up gcc warning in clps7500 core.c

core.c:263: warning: initialization makes integer from pointer without a cast

Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
18 years ago[ARM] Shut up gcc warning in assabet.c
Russell King [Mon, 21 Nov 2005 17:01:13 +0000 (17:01 +0000)]
[ARM] Shut up gcc warning in assabet.c

assabet.c:291: warning: 'scr' may be used uninitialized in this function

Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
18 years ago[ARM] ebsa110: __arch_ioremap should be 3 args
Russell King [Mon, 21 Nov 2005 15:26:52 +0000 (15:26 +0000)]
[ARM] ebsa110: __arch_ioremap should be 3 args

Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>