linux-2.6
16 years agoalpha: fix compile failures with gcc-4.3 (bug #10438)
Ivan Kokshaysky [Fri, 20 Jun 2008 23:28:31 +0000 (03:28 +0400)] 
alpha: fix compile failures with gcc-4.3 (bug #10438)

Vast majority of these build failures are gcc-4.3 warnings
about static functions and objects being referenced from
non-static (read: "extern inline") functions, in conjunction
with our -Werror.

We cannot just convert "extern inline" to "static inline",
as people keep suggesting all the time, because "extern inline"
logic is crucial for generic kernel build.
So
- just make sure that all callees of critical "extern inline"
  functions are also "extern inline";
- use "static inline", wherever it's possible.

traps.c: work around gcc-4.3 being too smart about array
bounds-checking.

TODO: add "gnu_inline" attribute to all our "extern inline"
functions to ensure desired behaviour with future compilers.

Signed-off-by: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoalpha: link failure fix
Ivan Kokshaysky [Fri, 20 Jun 2008 23:26:21 +0000 (03:26 +0400)] 
alpha: link failure fix

With built-in scsi disk driver, the final link fails with a following
error:
`.exit.text' referenced in section `.rodata' of drivers/built-in.o:
defined in discarded section `.exit.text' of drivers/built-in.o

This happens with -Os (CONFIG_CC_OPTIMIZE_FOR_SIZE=y) with all gcc-4
versions, and also with -O2 and gcc-4.3.

The problem is in sd.c:sd_major() being inlined into __exit function
exit_sd(), and the compiler generating a jump table in .rodata section
for the 'switch' statement in sd_major(). So we have references to
discarded section.

Fixed with a big hammer in the form of -fno-jump-tables.

Note that jump tables vs. discarded sections is a generic problem,
other architectures are just lucky not to suffer from it. But with
a slightly more complex switch/case statement it can be reproduced
on x86 as well. So maybe at some point we should consider
-fno-jump-tables as a generic compile option...

Signed-off-by: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoalpha: fix module load failures on smp (bug #10926)
Ivan Kokshaysky [Fri, 20 Jun 2008 23:25:39 +0000 (03:25 +0400)] 
alpha: fix module load failures on smp (bug #10926)

To calculate addresses of locally defined variables, GCC uses 32-bit
displacement from the GP. Which doesn't work for per cpu variables in
modules, as an offset to the kernel per cpu area is way above 4G.

The workaround is to force allocation of a GOT entry for per cpu variable
using ldq instruction with a 'literal' relocation.
I had to use custom asm/percpu.h, as a required argument magic doesn't
work with asm-generic/percpu.h macros.

Signed-off-by: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoLinux 2.6.26-rc7 v2.6.26-rc7
Linus Torvalds [Fri, 20 Jun 2008 23:19:44 +0000 (16:19 -0700)] 
Linux 2.6.26-rc7

16 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/bart/ide-2.6
Linus Torvalds [Fri, 20 Jun 2008 19:46:47 +0000 (12:46 -0700)] 
Merge git://git./linux/kernel/git/bart/ide-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/bart/ide-2.6:
  BAST: Remove old IDE driver
  pcmcia ide kingston compactflash's have a new manufacturer id
  pcmcia: add another pata/ide ID
  pcmcia: add an pata/ide ID
  ide: increase timeout in wait_drive_not_busy()
  palm_bk3710: fix resource management

16 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ieee1394...
Linus Torvalds [Fri, 20 Jun 2008 19:41:10 +0000 (12:41 -0700)] 
Merge branch 'for-linus' of git://git./linux/kernel/git/ieee1394/linux1394-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ieee1394/linux1394-2.6:
  ieee1394: Kconfig menu touch-up
  firewire: Kconfig menu touch-up
  firewire: deadline for PHY config transmission
  firewire: fw-ohci: unify printk prefixes
  firewire: fill_bus_reset_event needs lock protection
  firewire: fw-ohci: write selfIDBufferPtr before LinkControl.rcvSelfID
  firewire: fw-ohci: disable PHY packet reception into AR context
  firewire: fw-ohci: use of uninitialized data in AR handler
  firewire: don't panic on invalid AR request buffer

16 years agoMerge branch 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/lenb/linux...
Linus Torvalds [Fri, 20 Jun 2008 19:39:12 +0000 (12:39 -0700)] 
Merge branch 'release' of git://git./linux/kernel/git/lenb/linux-acpi-2.6

* 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/lenb/linux-acpi-2.6:
  ACPI: no AC status notification
  ACPI Exception (video-1721): UNKNOWN_STATUS_CODE, Cant attach device

16 years agoMerge branch 'drm-patches' of git://git.kernel.org/pub/scm/linux/kernel/git/airlied...
Linus Torvalds [Fri, 20 Jun 2008 19:38:18 +0000 (12:38 -0700)] 
Merge branch 'drm-patches' of git://git./linux/kernel/git/airlied/drm-2.6

* 'drm-patches' of git://git.kernel.org/pub/scm/linux/kernel/git/airlied/drm-2.6: (21 commits)
  drm: only trust core drm ioctls - driver ioctls are a mess.
  drm/i915: add support for Intel series 4 chipsets.
  drm/radeon: add hier-z registers for r300 and r500 chipsets
  drm/radeon: use DSTCACHE_CTLSTAT rather than RB2D_DSTCACHE_CTLSTAT
  drm/radeon: switch IGP gart to use radeon_write_agp_base()
  drm/radeon: Restore sw interrupt on resume
  drm/r500: add support for AGP based cards.
  drm/radeon: fix texture uploads with large 3d textures (bug 13980)
  drm/radeon: add initial r500 support.
  drm/radeon: init pipe setup in kernel code.
  drm/radeon: fixup radeon_do_engine_reset
  drm/radeon: fix pixcache and purge/cache flushing registers
  drm/radeon: write AGP_BASE_2 on chips that support it.
  drm/radeon: merge IGP chip setup and fixup RS400 vs RS480 support
  drm/radeon: IGP clean up register and magic numbers.
  drm/rs690: set base 2 to 0.
  drm/rs690: set all of gart base address.
  radeon: add production microcode from AMD
  drm: pcigart use proper pci map interfaces.
  drm: the sg alloc ioctl should write back the handle to userspace
  ...

16 years agoMerge branch 'agp-patches' of git://git.kernel.org/pub/scm/linux/kernel/git/airlied...
Linus Torvalds [Fri, 20 Jun 2008 19:37:55 +0000 (12:37 -0700)] 
Merge branch 'agp-patches' of git://git./linux/kernel/git/airlied/agp-2.6

* 'agp-patches' of git://git.kernel.org/pub/scm/linux/kernel/git/airlied/agp-2.6:
  [agp]: fixup chipset flush for new Intel G4x.
  agp: brown paper bag patch - put back the two lines it took out.

16 years agoMerge branch 'core-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Fri, 20 Jun 2008 19:37:13 +0000 (12:37 -0700)] 
Merge branch 'core-fixes-for-linus' of git://git./linux/kernel/git/tip/linux-2.6-tip

* 'core-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
  softlockup: fix NMI hangs due to lock race - 2.6.26-rc regression
  rcupreempt: remove export of rcu_batches_completed_bh
  cpuset: limit the input of cpuset.sched_relax_domain_level

16 years agoMerge branch 'sched-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Fri, 20 Jun 2008 19:36:55 +0000 (12:36 -0700)] 
Merge branch 'sched-fixes-for-linus' of git://git./linux/kernel/git/tip/linux-2.6-tip

* 'sched-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
  sched, delay accounting: fix incorrect delay time when constantly waiting on runqueue
  sched: CPU hotplug events must not destroy scheduler domains created by the cpusets
  sched: rt-group: fix RR buglet
  sched: rt-group: heirarchy aware throttle
  sched: rt-group: fix hierarchy
  sched: NULL pointer dereference while setting sched_rt_period_us
  sched: fix defined-but-unused warning

16 years agoMerge branch 'x86-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Fri, 20 Jun 2008 19:36:38 +0000 (12:36 -0700)] 
Merge branch 'x86-fixes-for-linus' of git://git./linux/kernel/git/tip/linux-2.6-tip

* 'x86-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
  x86, geode: add a VSA2 ID for General Software
  x86: use BOOTMEM_EXCLUSIVE on 32-bit
  x86, 32-bit: fix boot failure on TSC-less processors
  x86: fix NULL pointer deref in __switch_to
  x86: set PAE PHYSICAL_MASK_SHIFT to 44 bits.

16 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/cooloney...
Linus Torvalds [Fri, 20 Jun 2008 19:34:43 +0000 (12:34 -0700)] 
Merge branch 'for-linus' of git://git./linux/kernel/git/cooloney/blackfin-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/cooloney/blackfin-2.6:
  Blackfin Serial Driver: Use timer to poll CTS PIN instead of workqueue.
  Blackfin arch: fix typo error in bf548 serial header file

16 years agoMerge branch 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jgarzi...
Linus Torvalds [Fri, 20 Jun 2008 19:31:03 +0000 (12:31 -0700)] 
Merge branch 'upstream-linus' of git://git./linux/kernel/git/jgarzik/libata-dev

* 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jgarzik/libata-dev:
  ahci: sis can't do PMP
  ata_piix: add TECRA M4 to broken suspend list
  LIBATA: Add HAVE_PATA_PLATFORM to select PATA_PLATFORM driver
  sata_mv: warn on PIO with multiple DRQs
  sata_mv: enable async_notify for 60x1 Rev.C0 and higher
  libata: don't check whether to use DMA or not for no data commands
  ahci: jmb361 has only one port

16 years ago[watchdog] hpwdt: fix use of inline assembly
Linus Torvalds [Fri, 20 Jun 2008 19:19:28 +0000 (12:19 -0700)] 
[watchdog] hpwdt: fix use of inline assembly

The inline assembly in drivers/watchdog/hpwdt.c was incredibly broken,
and included all the function prologue and epilogue stuff, even though
it was itself then inside a C function where the compiler would add its
own prologue and epilogue on top of it all.

This then just _happened_ to work if you had exactly the right compiler
version and exactly the right compiler flags, so that gcc just happened
to not create any prologue at all (the gcc-generated epilogue wouldn't
matter, since it would never be reached).

But the more proper way to fix it is to simply not do this.  Move the
inline asm to the top level, with no surrounding function at all (the
better alternative would be to remove the prologue and make it actually
use proper description of the arguments to the inline asm, but that's a
bigger change than the one I'm willing to make right now).

Tested-by: S.Çağlar Onur <caglar@pardus.org.tr>
Acked-by: Thomas Mingarelli <Thomas.Mingarelli@hp.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoBAST: Remove old IDE driver
Ben Dooks [Fri, 20 Jun 2008 18:53:35 +0000 (20:53 +0200)] 
BAST: Remove old IDE driver

Remove the old BAST IDE driver, as we are now using the platform-pata
support.

Signed-off-by: Ben Dooks <ben-linux@fluff.org>
Cc: Jeff Garzik <jgarzik@pobox.com>
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
16 years agopcmcia ide kingston compactflash's have a new manufacturer id
Christophe Niclaes [Fri, 20 Jun 2008 18:53:34 +0000 (20:53 +0200)] 
pcmcia ide kingston compactflash's have a new manufacturer id

Up to now, Kingston compactflash cards (ab)used the Toshiba Manufacturer's ID,
In their new CF cards, they use a new one.  Let's the ide subsystem
recognize CF cards with the new id.

Signed-off-by: Christophe Niclaes <cniclaes@develtech.com>
Acked-by: Philippe De Muyter <phdm@macqel.be>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: Dominik Brodowski <linux@dominikbrodowski.net>
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
16 years agopcmcia: add another pata/ide ID
Kristoffer Ericson [Fri, 20 Jun 2008 18:53:34 +0000 (20:53 +0200)] 
pcmcia: add another pata/ide ID

Addition of Transcend 1GB 45x id so that it is properly detected.

[bart: fix typo in ide-cs's ID spotted by Alan Cox]

Signed-off-by: William Peters <w1ll14@gmail.com>
Signed-off-by: Kristoffer Ericson <Kristoffer_e1@hotmail.com>
CC: Alan Cox <alan@lxorguk.ukuu.org.uk>
CC: linux-ide@vger.kernel.org
Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net>
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
16 years agopcmcia: add an pata/ide ID
Matt Reimer [Fri, 20 Jun 2008 18:53:34 +0000 (20:53 +0200)] 
pcmcia: add an pata/ide ID

Add an id for:

product info: "M-Systems", "CF300", ""
manfid: 0x000a, 0x0000
function: 4 (fixed disk)

Signed-off-by: Matt Reimer <mreimer@vpop.net>
CC: Alan Cox <alan@lxorguk.ukuu.org.uk>
CC: linux-ide@vger.kernel.org
Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net>
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
16 years agoide: increase timeout in wait_drive_not_busy()
Bartlomiej Zolnierkiewicz [Fri, 20 Jun 2008 18:53:33 +0000 (20:53 +0200)] 
ide: increase timeout in wait_drive_not_busy()

Some ATAPI devices take longer than the current max timeout value to
become ready (i.e. TEAC DV-W28ECW takes 6 ms) so increase the timeout
value to 10 ms.

This fixes kernel.org bugzilla bug #10887:
http://bugzilla.kernel.org/show_bug.cgi?id=10887

Reported-by: Masanari Iida <standby24x7@gmail.com>
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
16 years agopalm_bk3710: fix resource management
Sergei Shtylyov [Fri, 20 Jun 2008 18:53:32 +0000 (20:53 +0200)] 
palm_bk3710: fix resource management

The driver expected a *virtual* address in the IDE platform device's memory
resource and didn't request the memory region for the register block. Fix this
taking into account the fact that DaVinci SoC devices are fixed-mapped to the
virtual memory early and we can get their virtual addresses using IO_ADDRESS()
macro, not having to call ioremap()...

While at it, also do some cosmetic changes...

Signed-off-by: Sergei Shtylyov <sshtylyov@ru.mvista.com>
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
16 years agoReinstate ZERO_PAGE optimization in 'get_user_pages()' and fix XIP
Linus Torvalds [Fri, 20 Jun 2008 18:18:25 +0000 (11:18 -0700)] 
Reinstate ZERO_PAGE optimization in 'get_user_pages()' and fix XIP

KAMEZAWA Hiroyuki and Oleg Nesterov point out that since the commit
557ed1fa2620dc119adb86b34c614e152a629a80 ("remove ZERO_PAGE") removed
the ZERO_PAGE from the VM mappings, any users of get_user_pages() will
generally now populate the VM with real empty pages needlessly.

We used to get the ZERO_PAGE when we did the "handle_mm_fault()", but
since fault handling no longer uses ZERO_PAGE for new anonymous pages,
we now need to handle that special case in follow_page() instead.

In particular, the removal of ZERO_PAGE effectively removed the core
file writing optimization where we would skip writing pages that had not
been populated at all, and increased memory pressure a lot by allocating
all those useless newly zeroed pages.

This reinstates the optimization by making the unmapped PTE case the
same as for a non-existent page table, which already did this correctly.

While at it, this also fixes the XIP case for follow_page(), where the
caller could not differentiate between the case of a page that simply
could not be used (because it had no "struct page" associated with it)
and a page that just wasn't mapped.

We do that by simply returning an error pointer for pages that could not
be turned into a "struct page *".  The error is arbitrarily picked to be
EFAULT, since that was what get_user_pages() already used for the
equivalent IO-mapped page case.

[ Also removed an impossible test for pte_offset_map_lock() failing:
  that's not how that function works ]

Acked-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: Nick Piggin <npiggin@suse.de>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoMerge branch 'bugzilla-9761' into release
Len Brown [Fri, 20 Jun 2008 06:47:16 +0000 (02:47 -0400)] 
Merge branch 'bugzilla-9761' into release

16 years agoMerge branch 'bugzilla-10695' into release
Len Brown [Fri, 20 Jun 2008 06:45:05 +0000 (02:45 -0400)] 
Merge branch 'bugzilla-10695' into release

16 years agodrm: only trust core drm ioctls - driver ioctls are a mess.
Dave Airlie [Fri, 20 Jun 2008 05:42:38 +0000 (15:42 +1000)] 
drm: only trust core drm ioctls - driver ioctls are a mess.

So driver ioctls need a full auditing before we can make this change.

Signed-off-by: Dave Airlie <airlied@redhat.com>
16 years agodrm/i915: add support for Intel series 4 chipsets.
Zhenyu Wang [Fri, 20 Jun 2008 02:12:56 +0000 (12:12 +1000)] 
drm/i915: add support for Intel series 4 chipsets.

Signed-off-by: Zhenyu Wang <zhenyu.z.wang@intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
16 years ago[agp]: fixup chipset flush for new Intel G4x.
Zhenyu Wang [Fri, 20 Jun 2008 01:48:06 +0000 (11:48 +1000)] 
[agp]: fixup chipset flush for new Intel G4x.

Signed-off-by: Zhenyu Wang <zhenyu.z.wang@intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
16 years agox86, geode: add a VSA2 ID for General Software
Jordan Crouse [Wed, 18 Jun 2008 17:34:38 +0000 (11:34 -0600)] 
x86, geode: add a VSA2 ID for General Software

General Software writes their own VSA2 module for their version
of the Geode BIOS, which returns a different ID then the standard
VSA2.  This was causing the framebuffer driver to break for most
GSW boards.

Signed-off-by: Jordan Crouse <jordan.crouse@amd.com>
Cc: tglx@linutronix.de
Cc: linux-geode@lists.infradead.org
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agosched, delay accounting: fix incorrect delay time when constantly waiting on runqueue
Bharath Ravi [Mon, 16 Jun 2008 09:41:01 +0000 (15:11 +0530)] 
sched, delay accounting: fix incorrect delay time when constantly waiting on runqueue

This patch corrects the incorrect value of per process run-queue wait
time reported by delay statistics. The anomaly was due to the following
reason. When a process leaves the CPU and immediately starts waiting for
CPU on the runqueue (which means it remains in the TASK_RUNNABLE state),
the time of re-entry into the run-queue is never recorded. Due to this,
the waiting time on the runqueue from this point of re-entry upto the
next time it hits the CPU is not accounted for. This is solved by
recording the time of re-entry of a process leaving the CPU in the
sched_info_depart() function IF the process will go back to waiting on
the run-queue. This IF condition is verified by checking whether the
process is still in the TASK_RUNNABLE state.

The patch was tested on 2.6.26-rc6 using two simple CPU hog programs.
The values noted prior to the fix did not account for the time spent on
the runqueue waiting. After the fix, the correct values were reported
back to user space.

Signed-off-by: Bharath Ravi <bharathravi1@gmail.com>
Signed-off-by: Madhava K R <madhavakr@gmail.com>
Cc: dhaval@linux.vnet.ibm.com
Cc: vatsa@in.ibm.com
Cc: balbir@in.ibm.com
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoBlackfin Serial Driver: Use timer to poll CTS PIN instead of workqueue.
Sonic Zhang [Thu, 19 Jun 2008 09:46:39 +0000 (17:46 +0800)] 
Blackfin Serial Driver: Use timer to poll CTS PIN instead of workqueue.

This allows other threads to run when the serial driver polls the CTS
PIN in a loop.

Signed-off-by: Sonic Zhang <sonic.zhang@analog.com>
Signed-off-by: Bryan Wu <cooloney@kernel.org>
16 years agoBlackfin arch: fix typo error in bf548 serial header file
Sonic Zhang [Thu, 19 Jun 2008 09:07:15 +0000 (17:07 +0800)] 
Blackfin arch: fix typo error in bf548 serial header file

Signed-off-by: Sonic Zhang <sonic.zhang@analog.com>
Signed-off-by: Bryan Wu <cooloney@kernel.org>
16 years agox86: use BOOTMEM_EXCLUSIVE on 32-bit
Bernhard Walle [Sun, 8 Jun 2008 14:16:07 +0000 (16:16 +0200)] 
x86: use BOOTMEM_EXCLUSIVE on 32-bit

This patch uses the BOOTMEM_EXCLUSIVE for crashkernel reservation also for
i386 and prints a error message on failure.

The patch is still for 2.6.26 since it is only bug fixing. The unification
of reserve_crashkernel() between i386 and x86_64 should be done for 2.6.27.

Signed-off-by: Bernhard Walle <bwalle@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cc: <stable@kernel.org>
16 years agox86, 32-bit: fix boot failure on TSC-less processors
Mikael Pettersson [Sun, 15 Jun 2008 00:19:56 +0000 (02:19 +0200)] 
x86, 32-bit: fix boot failure on TSC-less processors

Booting 2.6.26-rc6 on my 486 DX/4 fails with a "BUG: Int 6"
(invalid opcode) and a kernel halt immediately after the
kernel has been uncompressed. The BUG shows EIP pointing
to an rdtsc instruction in native_read_tsc(), invoked from
native_sched_clock().

(This error occurs so early that not even the serial console
can capture it.)

A bisection showed that this bug first occurs in 2.6.26-rc3-git7,
via commit 9ccc906c97e34fd91dc6aaf5b69b52d824386910:

>x86: distangle user disabled TSC from unstable
>
>tsc_enabled is set to 0 from the command line switch "notsc" and from
>the mark_tsc_unstable code. Seperate those functionalities and replace
>tsc_enable with tsc_disable. This makes also the native_sched_clock()
>decision when to use TSC understandable.
>
>Preparatory patch to solve the sched_clock() issue on 32 bit.
>
>Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

The core reason for this bug is that native_sched_clock() gets
called before tsc_init().

Before the commit above, tsc_32.c used a "tsc_enabled" variable
which defaulted to 0 == disabled, and which only got enabled late
in tsc_init(). Thus early calls to native_sched_clock() would skip
the TSC and use jiffies instead.

After the commit above, tsc_32.c uses a "tsc_disabled" variable
which defaults to 0, meaning that the TSC is Ok to use. Early calls
to native_sched_clock() now erroneously try to use the TSC on
!cpu_has_tsc processors, leading to invalid opcode exceptions.

My proposed fix is to initialise tsc_disabled to a "soft disabled"
state distinct from the hard disabled state set up by the "notsc"
kernel option. This fixes the native_sched_clock() problem. It also
allows tsc_init() to be simplified: instead of setting tsc_disabled = 1
on every error return, we just set tsc_disabled = 0 once when all
checks have succeeded.

I've verified that this lets my 486 boot again. I've also verified
that a Core2 machine still uses the TSC as clocksource after the patch.

Signed-off-by: Mikael Pettersson <mikpe@it.uu.se>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agox86: fix NULL pointer deref in __switch_to
Suresh Siddha [Fri, 13 Jun 2008 22:47:12 +0000 (15:47 -0700)] 
x86: fix NULL pointer deref in __switch_to

Patrick McHardy reported a crash:

> > I get this oops once a day, its apparently triggered by something
> > run by cron, but the process is a different one each time.
> >
> > Kernel is -git from yesterday shortly before the -rc6 release
> > (last commit is the usb-2.6 merge, the x86 patches are missing),
> > .config is attached.
> >
> > I'll retry with current -git, but the patches that have gone in
> > since I last updated don't look related.
> >
> > [62060.043009] BUG: unable to handle kernel NULL pointer dereference at
> > 000001ff
> > [62060.043009] IP: [<c0102a9b>] __switch_to+0x2f/0x118
> > [62060.043009] *pde = 00000000
> > [62060.043009] Oops: 0002 [#1] PREEMPT

Vegard Nossum analyzed it:

> This decodes to
>
>    0:   0f ae 00                fxsave (%eax)
>
> so it's related to the floating-point context. This is the exact
> location of the crash:
>
> $ addr2line -e arch/x86/kernel/process_32.o -i ab0
> include/asm/i387.h:232
> include/asm/i387.h:262
> arch/x86/kernel/process_32.c:595
>
> ...so it looks like prev_task->thread.xstate->fxsave has become NULL.
> Or maybe it never had any other value.

Somehow (as described below) TS_USEDFPU is set but the fpu is not
allocated or freed.

Another possible FPU pre-emption issue with the sleazy FPU optimization
which was benign before but not so anymore, with the dynamic FPU allocation
patch.

New task is getting exec'd and it is prempted at the below point.

flush_thread() {
...
/*
* Forget coprocessor state..
*/
clear_fpu(tsk);
<----- Preemption point
clear_used_math();
...
}

Now when it context switches in again, as the used_math() is still set
and fpu_counter can be > 5, we will do a math_state_restore() which sets
the task's TS_USEDFPU. After it continues from the above preemption point
it does clear_used_math() and much later free_thread_xstate().

Now, at the next context switch, it is quite possible that xstate is
null, used_math() is not set and TS_USEDFPU is still set. This will
trigger unlazy_fpu() causing kernel oops.

Fix this  by clearing tsk's fpu_counter before clearing task's fpu.

Reported-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agox86: set PAE PHYSICAL_MASK_SHIFT to 44 bits.
Jeremy Fitzhardinge [Fri, 6 Jun 2008 09:21:39 +0000 (10:21 +0100)] 
x86: set PAE PHYSICAL_MASK_SHIFT to 44 bits.

When a 64-bit x86 processor runs in 32-bit PAE mode, a pte can
potentially have the same number of physical address bits as the
64-bit host ("Enhanced Legacy PAE Paging").  This means, in theory,
we could have up to 52 bits of physical address in a pte.

The 32-bit kernel uses a 32-bit unsigned long to represent a pfn.
This means that it can only represent physical addresses up to 32+12=44
bits wide.  Rather than widening pfns everywhere, just set 2^44 as the
Linux x86_32-PAE architectural limit for physical address size.

This is a bugfix for two cases:
1. running a 32-bit PAE kernel on a machine with
  more than 64GB RAM.
2. running a 32-bit PAE Xen guest on a host machine with
  more than 64GB RAM

In both cases, a pte could need to have more than 36 bits of physical,
and masking it to 36-bits will cause fairly severe havoc.

Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Cc: Jan Beulich <jbeulich@novell.com>
Cc: <stable@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agosoftlockup: fix NMI hangs due to lock race - 2.6.26-rc regression
Jason Wessel [Tue, 27 May 2008 17:23:29 +0000 (12:23 -0500)] 
softlockup: fix NMI hangs due to lock race - 2.6.26-rc regression

The touch_nmi_watchdog() routine on x86 ultimately calls
touch_softlockup_watchdog().  The problem is that to touch the
softlockup watchdog, the cpu_clock code has to be called which could
involve multiple cpu locks and can lead to a hard hang if one of the
locks is held by a processor that is not going to return anytime soon
(such as could be the case with kgdb or perhaps even with some other
kind of exception).

This patch causes the public version of the
touch_softlockup_watchdog() to defer the cpu clock access to a later
point.

The test case for this problem is to use the following kernel config
options:

CONFIG_KGDB_TESTS=y
CONFIG_KGDB_TESTS_ON_BOOT=y
CONFIG_KGDB_TESTS_BOOT_STRING="V1F100I100000"

It should be noted that kgdb test suite and these options were not
available until 2.6.26-rc2, so it was necessary to patch the kgdb
test suite during the bisection.

I would consider this patch a regression fix because the problem first
appeared in commit 27ec4407790d075c325e1f4da0a19c56953cce23 when some
logic was added to try to periodically sync the clocks.  It was
possible to work around this particular problem by simply not
performing the sync anytime the system was in a critical context.
This was ok until commit 3e51f33fcc7f55e6df25d15b55ed10c8b4da84cd,
which added config option CONFIG_HAVE_UNSTABLE_SCHED_CLOCK and some
multi-cpu locks to sync the clocks.  It became clear that accessing
this code from an nmi was the source of the lockups.  Avoiding the
access to the low level clock code from an code inside the NMI
processing also fixed the problem with the 27ec44... commit.

Signed-off-by: Jason Wessel <jason.wessel@windriver.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agorcupreempt: remove export of rcu_batches_completed_bh
Steven Rostedt [Thu, 22 May 2008 18:18:17 +0000 (14:18 -0400)] 
rcupreempt: remove export of rcu_batches_completed_bh

In rcupreempt, rcu_batches_completed_bh is defined as a static inline in
the header file. This does not need to be exported, and not only that,
this breaks my PPC build.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Cc: paulus@samba.org
Cc: linuxppc-dev@ozlabs.org
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agocpuset: limit the input of cpuset.sched_relax_domain_level
Li Zefan [Tue, 13 May 2008 02:27:17 +0000 (10:27 +0800)] 
cpuset: limit the input of cpuset.sched_relax_domain_level

We allow the inputs to be [-1 ... SD_LV_MAX), and return -EINVAL
for inputs outside this range.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Paul Menage <menage@google.com>
Acked-by: Paul Jackson <pj@sgi.com>
Acked-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agoMerge branch 'linus' into sched/urgent
Ingo Molnar [Thu, 19 Jun 2008 07:37:31 +0000 (09:37 +0200)] 
Merge branch 'linus' into sched/urgent

16 years agosched: CPU hotplug events must not destroy scheduler domains created by the cpusets
Max Krasnyansky [Thu, 29 May 2008 18:17:01 +0000 (11:17 -0700)] 
sched: CPU hotplug events must not destroy scheduler domains created by the cpusets

First issue is not related to the cpusets. We're simply leaking doms_cur.
It's allocated in arch_init_sched_domains() which is called for every
hotplug event. So we just keep reallocation doms_cur without freeing it.
I introduced free_sched_domains() function that cleans things up.

Second issue is that sched domains created by the cpusets are
completely destroyed by the CPU hotplug events. For all CPU hotplug
events scheduler attaches all CPUs to the NULL domain and then puts
them all into the single domain thereby destroying domains created
by the cpusets (partition_sched_domains).
The solution is simple, when cpusets are enabled scheduler should not
create default domain and instead let cpusets do that. Which is
exactly what the patch does.

Signed-off-by: Max Krasnyansky <maxk@qualcomm.com>
Cc: pj@sgi.com
Cc: menage@google.com
Cc: rostedt@goodmis.org
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
16 years agosched: rt-group: fix RR buglet
Peter Zijlstra [Thu, 19 Jun 2008 07:06:59 +0000 (09:06 +0200)] 
sched: rt-group: fix RR buglet

In tick_task_rt() we first call update_curr_rt() which can dequeue a runqueue
due to it running out of runtime, and then we try to requeue it, of it also
having exhausted its RR quota. Obviously requeueing something that is no longer
on the runqueue will not have the expected result.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Tested-by: Daniel K. <dk@uw.no>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agosched: rt-group: heirarchy aware throttle
Peter Zijlstra [Thu, 19 Jun 2008 07:06:57 +0000 (09:06 +0200)] 
sched: rt-group: heirarchy aware throttle

The bandwidth throttle code dequeues a group when it runs out of quota, and
re-queues it once the period rolls over and the quota gets refreshed.

Sadly it failed to take the hierarchy into consideration. Share more of the
enqueue/dequeue code with regular task opterations.

Also, some operations like sched_setscheduler() can dequeue/enqueue tasks that
are in throttled runqueues, we should not inadvertly re-enqueue empty runqueues
so check for that.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Tested-by: Daniel K. <dk@uw.no>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agosched: rt-group: fix hierarchy
Peter Zijlstra [Thu, 19 Jun 2008 07:06:56 +0000 (09:06 +0200)] 
sched: rt-group: fix hierarchy

Don't re-set the entity's runqueue to the wrong rq after we've set it
to the right one.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Tested-by: Daniel K. <dk@uw.no>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agosched: NULL pointer dereference while setting sched_rt_period_us
Dario Faggioli [Wed, 18 Jun 2008 07:18:38 +0000 (09:18 +0200)] 
sched: NULL pointer dereference while setting sched_rt_period_us

When CONFIG_RT_GROUP_SCHED and CONFIG_CGROUP_SCHED are enabled, with:

 echo 10000 > /proc/sys/kernel/sched_rt_period_us

We get this:

 BUG: unable to handle kernel NULL pointer dereference at 0000008c
 [  947.682233] IP: [<c0216b72>] __rt_schedulable+0x12/0x160
 [  947.683123] *pde = 00000000=20
 [  947.683782] Oops: 0000 [#1]
 [  947.684307] Modules linked in:
 [  947.684308]
 [  947.684308] Pid: 2359, comm: bash Not tainted (2.6.26-rc6 #8)
 [  947.684308] EIP: 0060:[<c0216b72>] EFLAGS: 00000246 CPU: 0
 [  947.684308] EIP is at __rt_schedulable+0x12/0x160
 [  947.684308] EAX: 00000000 EBX: 00000000 ECX: 00000000 EDX: 00000001
 [  947.684308] ESI: c0521db4 EDI: 00000001 EBP: c6cc9f00 ESP: c6cc9ed0
 [  947.684308]  DS: 007b ES: 007b FS: 0000 GS: 0033 SS: 0068
 [  947.684308] Process bash (pid: 2359, tiÆcc8000 taskÇa54f00=20 task.tiÆcc8000)
 [  947.684308] Stack: c0222790 00000000 080f8c08 c0521db4 c6cc9f00 00000001 00000000 00000000
 [  947.684308]        c6cc9f9c 00000000 c0521db4 00000001 c6cc9f28 c0216d40 00000000 00000000
 [  947.684308]        c6cc9f9c 000f4240 000e7ef0 ffffffff c0521db4 c79dfb60 c6cc9f58 c02af2cc
 [  947.684308] Call Trace:
 [  947.684308]  [<c0222790>] ? do_proc_dointvec_conv+0x0/0x50
 [  947.684308]  [<c0216d40>] ? sched_rt_handler+0x80/0x110
 [  947.684308]  [<c02af2cc>] ? proc_sys_call_handler+0x9c/0xb0
 [  947.684308]  [<c02af2fa>] ? proc_sys_write+0x1a/0x20
 [  947.684308]  [<c0273c36>] ? vfs_write+0x96/0x160
 [  947.684308]  [<c02af2e0>] ? proc_sys_write+0x0/0x20
 [  947.684308]  [<c027423d>] ? sys_write+0x3d/0x70
 [  947.684308]  [<c0202ef5>] ? sysenter_past_esp+0x6a/0x91
 [  947.684308]  =======================
 [  947.684308] Code: 24 04 e8 62 b1 0e 00 89 c7 89 f8 8b 5d f4 8b 75
 f8 8b 7d fc 89 ec 5d c3 90 55 89 e5 57 56 53 83 ec 24 89 45 ec 89 55 e4
 89 4d e8 <8b> b8 8c 00 00 00 85 ff 0f 84 c9 00 00 00 8b 57 24 39 55 e8
 8b
 [  947.684308] EIP: [<c0216b72>] __rt_schedulable+0x12/0x160 SS:ESP  0068:c6cc9ed0

We think the following patch solves the issue.

Signed-off-by: Dario Faggioli <raistlin@linux.it>
Signed-off-by: Michael Trimarchi <trimarchimichael@yahoo.it>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoagp: brown paper bag patch - put back two lines that got lost
Dave Airlie [Thu, 19 Jun 2008 04:57:31 +0000 (14:57 +1000)] 
agp: brown paper bag patch - put back two lines that got lost

Commit 62c96b9d0917894c164aa3e474a3ff3bca1554ae ("agp/intel: cleanup
some serious whitespace badness") didn't just fix whitespace.  It also
lost two lines.

Noticed by Linus. No more whitespace diffs for me.

Signed-off-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoagp: brown paper bag patch - put back the two lines it took out.
Dave Airlie [Thu, 19 Jun 2008 04:57:31 +0000 (14:57 +1000)] 
agp: brown paper bag patch - put back the two lines it took out.

no more whitespace diffs for me.

Signed-off-by: Dave Airlie <airlied@redhat.com>
16 years agoMerge branch 'agp-patches' of git://git.kernel.org/pub/scm/linux/kernel/git/airlied...
Linus Torvalds [Thu, 19 Jun 2008 04:52:35 +0000 (21:52 -0700)] 
Merge branch 'agp-patches' of git://git./linux/kernel/git/airlied/agp-2.6

* 'agp-patches' of git://git.kernel.org/pub/scm/linux/kernel/git/airlied/agp-2.6:
  agp/intel: cleanup some serious whitespace badness
  [AGP] intel_agp: Add support for Intel 4 series chipsets
  [AGP] intel_agp: extra stolen mem size available for IGD_GM chipset
  agp: more boolean conversions.
  drivers/char/agp - use bool
  agp: two-stage page destruction issue
  agp/via: fixup pci ids

16 years agoagp/intel: cleanup some serious whitespace badness
Dave Airlie [Thu, 19 Jun 2008 04:27:53 +0000 (14:27 +1000)] 
agp/intel: cleanup some serious whitespace badness

Signed-off-by: Dave Airlie <airlied@redhat.com>
16 years ago[AGP] intel_agp: Add support for Intel 4 series chipsets
Zhenyu Wang [Thu, 19 Jun 2008 04:17:58 +0000 (14:17 +1000)] 
[AGP] intel_agp: Add support for Intel 4 series chipsets

Signed-off-by: Zhenyu Wang <zhenyu.z.wang@intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
16 years ago[AGP] intel_agp: extra stolen mem size available for IGD_GM chipset
Zhenyu Wang [Thu, 19 Jun 2008 04:00:37 +0000 (14:00 +1000)] 
[AGP] intel_agp: extra stolen mem size available for IGD_GM chipset

This adds missing stolen memory size detect for IGD_GM, be sure to
detect right size as current X intel driver (2.3.2) which has already
worked out.

Signed-off-by: Zhenyu Wang <zhenyu.z.wang@intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
16 years agodrm/radeon: add hier-z registers for r300 and r500 chipsets
Dave Airlie [Thu, 19 Jun 2008 03:01:58 +0000 (13:01 +1000)] 
drm/radeon: add hier-z registers for r300 and r500 chipsets

16 years agodrm/radeon: use DSTCACHE_CTLSTAT rather than RB2D_DSTCACHE_CTLSTAT
Alex Deucher [Thu, 19 Jun 2008 02:39:23 +0000 (12:39 +1000)] 
drm/radeon: use DSTCACHE_CTLSTAT rather than RB2D_DSTCACHE_CTLSTAT

According to the hw guys, you should use DSTCACHE_CTLSTAT to flush
the 2D dst cache rather than RB2D_DSTCACHE_CTLSTAT.

Signed-off-by: Dave Airlie <airlied@redhat.com>
16 years agodrm/radeon: switch IGP gart to use radeon_write_agp_base()
Alex Deucher [Thu, 19 Jun 2008 02:38:29 +0000 (12:38 +1000)] 
drm/radeon: switch IGP gart to use radeon_write_agp_base()

Signed-off-by: Dave Airlie <airlied@redhat.com>
16 years agodrm/radeon: Restore sw interrupt on resume
Dennis Kasprzyk [Thu, 19 Jun 2008 02:36:55 +0000 (12:36 +1000)] 
drm/radeon: Restore sw interrupt on resume

Fixes performance drop after suspend/resume on some systems.

Signed-off-by: Dave Airlie <airlied@redhat.com>
16 years agodrm/r500: add support for AGP based cards.
Dave Airlie [Thu, 19 Jun 2008 01:40:44 +0000 (11:40 +1000)] 
drm/r500: add support for AGP based cards.

AGP registers weren't programmed properly for r500 cards.

Signed-off-by: Dave Airlie <airlied@redhat.com>
16 years agodrm/radeon: fix texture uploads with large 3d textures (bug 13980)
Roland Scheidegger [Thu, 19 Jun 2008 01:36:04 +0000 (11:36 +1000)] 
drm/radeon: fix texture uploads with large 3d textures (bug 13980)

Texture uploads could hit the blitter coordinate limit, adjust the texture
offset when uploading the pieces. Make sure to check the end address of the
upload too.

Signed-off-by: Dave Airlie <airlied@redhat.com>
16 years agodrm/radeon: add initial r500 support.
Dave Airlie [Wed, 28 May 2008 03:52:28 +0000 (13:52 +1000)] 
drm/radeon: add initial r500 support.

This contains all the command buffer processing for the r500 cards.
It doesn't yet contain vblank support.

Signed-off-by: Dave Airlie <airlied@redhat.com>
16 years agodrm/radeon: init pipe setup in kernel code.
Alex Deucher [Wed, 28 May 2008 01:57:40 +0000 (11:57 +1000)] 
drm/radeon: init pipe setup in kernel code.

This inits the card pipes in the kernel and lets userspace getparam
the correct setup.

Signed-off-by: Dave Airlie <airlied@redhat.com>
16 years agodrm/radeon: fixup radeon_do_engine_reset
Alex Deucher [Wed, 28 May 2008 01:54:06 +0000 (11:54 +1000)] 
drm/radeon: fixup radeon_do_engine_reset

Cleanup do engine reset for different chip families.

Signed-off-by: Dave Airlie <airlied@redhat.com>
16 years agodrm/radeon: fix pixcache and purge/cache flushing registers
Alex Deucher [Wed, 28 May 2008 01:51:12 +0000 (11:51 +1000)] 
drm/radeon: fix pixcache and purge/cache flushing registers

Signed-off-by: Dave Airlie <airlied@redhat.com>
16 years agodrm/radeon: write AGP_BASE_2 on chips that support it.
Alex Deucher [Wed, 28 May 2008 01:46:36 +0000 (11:46 +1000)] 
drm/radeon: write AGP_BASE_2 on chips that support it.

Signed-off-by: Dave Airlie <airlied@redhat.com>
16 years agodrm/radeon: merge IGP chip setup and fixup RS400 vs RS480 support
Alex Deucher [Wed, 28 May 2008 03:28:59 +0000 (13:28 +1000)] 
drm/radeon: merge IGP chip setup and fixup RS400 vs RS480 support

We only support RS480 (AMD based IGP) at the moment not
RS400 (Intel based IGP) ones.

Signed-off-by: Dave Airlie <airlied@redhat.com>
16 years agodrm/radeon: IGP clean up register and magic numbers.
Alex Deucher [Wed, 28 May 2008 02:54:16 +0000 (12:54 +1000)] 
drm/radeon: IGP clean up register and magic numbers.

Signed-off-by: Dave Airlie <airlied@redhat.com>
16 years agodrm/rs690: set base 2 to 0.
Dave Airlie [Wed, 28 May 2008 01:28:27 +0000 (11:28 +1000)] 
drm/rs690: set base 2 to 0.

Signed-off-by: Dave Airlie <airlied@redhat.com>
16 years agodrm/rs690: set all of gart base address.
Dave Airlie [Wed, 28 May 2008 01:27:01 +0000 (11:27 +1000)] 
drm/rs690: set all of gart base address.

Docs state bits 4-11 maps to bits 32-39 of the 40-bit range

Signed-off-by: Dave Airlie <airlied@redhat.com>
16 years agoradeon: add production microcode from AMD
Alex Deucher [Wed, 28 May 2008 01:21:25 +0000 (11:21 +1000)] 
radeon: add production microcode from AMD

This adds production microcode for r100->r500 from AMD.

Signed-off-by: Dave Airlie <airlied@redhat.com>
16 years agodrm: pcigart use proper pci map interfaces.
Dave Airlie [Thu, 19 Jun 2008 01:27:23 +0000 (11:27 +1000)] 
drm: pcigart use proper pci map interfaces.

Switch to using more correct pci dma mapping interfaces.

Signed-off-by: Dave Airlie <airlied@redhat.com>
16 years agoahci: sis can't do PMP
Tejun Heo [Thu, 19 Jun 2008 00:56:58 +0000 (20:56 -0400)] 
ahci: sis can't do PMP

From: Piter PUNK <piterpunk@slackware.com>

SiS AHCIs say they can do PMP but can't and fail detection if SRST w/
pmp==15 is used.  Turn off PMP support.

tj: added patch description, adapted patch to #upstream-fixes and
    renamed board_ahci_sis to board_ahci_nopmp.

Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
16 years agoagp: more boolean conversions.
Dave Airlie [Thu, 19 Jun 2008 00:42:17 +0000 (10:42 +1000)] 
agp: more boolean conversions.

Signed-off-by: Dave Airlie <airlied@redhat.com>
16 years agoata_piix: add TECRA M4 to broken suspend list
Tejun Heo [Fri, 13 Jun 2008 09:05:02 +0000 (18:05 +0900)] 
ata_piix: add TECRA M4 to broken suspend list

TOSHIBA also used "TECRA M4" in additon to "Tecra M4", add it.

Signed-off-by: Tejun Heo <htejun@gmail.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
16 years agoLIBATA: Add HAVE_PATA_PLATFORM to select PATA_PLATFORM driver
Ben Dooks [Mon, 16 Jun 2008 11:16:26 +0000 (12:16 +0100)] 
LIBATA: Add HAVE_PATA_PLATFORM to select PATA_PLATFORM driver

Add HAVE_PATA_PLATFORM to select the pata platform driver
to ensure that we do not end up with a long 'depends on' list
when other users of this driver turn up.

Signed-off-by: Ben Dooks <ben-linux@fluff.org>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
16 years agosata_mv: warn on PIO with multiple DRQs
Mark Lord [Wed, 18 Jun 2008 16:13:02 +0000 (12:13 -0400)] 
sata_mv: warn on PIO with multiple DRQs

Chip errata sometimes prevents reliable use of PIO commands which involve
more than a single DRQ (data request).  In normal operation, libata should
not generate such PIO commands (uses DMA instead), but they could be sent
in via SG_IO from userspace.

A full workaround might be to break up such commands into sequences
of single DRQ ones, but that's just way too complex for something
that doesn't normally happen in real life.

So, allow the attempt (it often works, despite the errata),
but log the event for reference when somebody screams.

Signed-off-by: Mark Lord <mlord@pobox.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
16 years agosata_mv: enable async_notify for 60x1 Rev.C0 and higher
Mark Lord [Wed, 18 Jun 2008 16:11:16 +0000 (12:11 -0400)] 
sata_mv: enable async_notify for 60x1 Rev.C0 and higher

The early chipsets cannot safely handle Async Notification (AN),
but 6041/6081 chip revision "C0" (and newer) can handle it.

So allow AN for "C0" and higher.

This enables use of hotplug on PMP ports for the 6041/6081 PCI Rev.9 chips.

Signed-off-by: Mark Lord <mlord@pobox.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
16 years agolibata: don't check whether to use DMA or not for no data commands
Tejun Heo [Tue, 17 Jun 2008 03:36:26 +0000 (12:36 +0900)] 
libata: don't check whether to use DMA or not for no data commands

There's no reason to check whether to use DMA or not for no data
commands.  Don't do it.  While at it, make local variable using_pio in
atapi_xlat() set iff ATAPI_PROT_PIO is going to be used and rename
ata_check_atapi_dma() to atapi_check_dma() for consistency.

Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
16 years agoahci: jmb361 has only one port
Tejun Heo [Tue, 17 Jun 2008 03:46:30 +0000 (12:46 +0900)] 
ahci: jmb361 has only one port

JMB361 has only one port but reports it has two causing longish probe
failure on the second one.  Quirk it.

Reported by Gajo Petrovic in bz 10911.

Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Gajo Petrovic <gajo01@gmail.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
16 years agodrivers/char/agp - use bool
Joe Perches [Wed, 26 Mar 2008 21:10:02 +0000 (14:10 -0700)] 
drivers/char/agp - use bool

Use boolean in AGP instead of having own TRUE/FALSE

--
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
16 years agoagp: two-stage page destruction issue
Jan Beulich [Wed, 18 Jun 2008 08:28:00 +0000 (09:28 +0100)] 
agp: two-stage page destruction issue

besides it apparently being useful only in 2.6.24 (the changes in 2.6.25
really mean that it could be converted back to a single-stage mechanism),
I'm seeing an issue in Xen Dom0 kernels, which is caused by the calling
of gart_to_virt() in the second stage invocations of the destroy function.
I think that besides this being a real issue with Xen (where
unmap_page_from_agp() is not just a page table attribute change), this
also is invalid from a theoretical perspective: One should not assume that
gart_to_virt() is still valid after unmapping a page. So minimally (keeping
the 2-stage mechanism) a patch like the one below would be needed.

Jan

Signed-off-by: Dave Airlie <airlied@redhat.com>
16 years agoagp/via: fixup pci ids
Greg KH [Wed, 18 Jun 2008 23:52:26 +0000 (09:52 +1000)] 
agp/via: fixup pci ids

add a new PCI ID and remove an old dodgy one, include the explaination
in the commented code so nobody readds later.

(davej also sent the pci id addition).

Signed-off-by: Dave Airlie <airlied@redhat.com>
16 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland...
Linus Torvalds [Wed, 18 Jun 2008 23:08:59 +0000 (16:08 -0700)] 
Merge branch 'for-linus' of git://git./linux/kernel/git/roland/infiniband

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband:
  IB/uverbs: Fix check of is_closed flag check in ib_uverbs_async_handler()
  RDMA/nes: Fix off-by-one in nes_reg_user_mr() error path

16 years agoIB/uverbs: Fix check of is_closed flag check in ib_uverbs_async_handler()
Jack Morgenstein [Wed, 18 Jun 2008 22:36:38 +0000 (15:36 -0700)] 
IB/uverbs: Fix check of is_closed flag check in ib_uverbs_async_handler()

Commit 1ae5c187 ("IB/uverbs: Don't store struct file * for event
files") changed the way that closed files are handled in the uverbs
code.  However, after the conversion, is_closed flag is checked
incorrectly in ib_uverbs_async_handler().  As a result, no async
events are ever passed to applications.

Found by: Ronni Zimmerman <ronniz@mellanox.co.il>

Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoieee1394: Kconfig menu touch-up
Stefan Richter [Sun, 15 Jun 2008 23:39:28 +0000 (01:39 +0200)] 
ieee1394: Kconfig menu touch-up

Rename and reorder some prompts and modify some help texts.
The result:

  -------------------- IEEE 1394 (FireWire) support --------------------
  *** Enable only one of the two stacks, unless you know what you are doing ***
  New FireWire stack, EXPERIMENTAL
    OHCI-1394 controllers
    Storage devices (SBP-2 protocol)
  Stable FireWire stack
    OHCI-1394 controllers
    PCILynx controller
    Storage devices (SBP-2 protocol)
      Enable replacement for physical DMA in SBP2
    IP over 1394
    raw1394 userspace interface
    video1394 userspace interface
    dv1394 userspace interface (deprecated)
    Excessive debugging output

The old prompts for reference:

  -------------------- IEEE 1394 (FireWire) support --------------------
  IEEE 1394 (FireWire) support - alternative stack, EXPERIMENTAL
    Support for OHCI FireWire host controllers
    Support for storage devices (SBP-2 protocol driver)
  IEEE 1394 (FireWire) support
    *** Subsystem Options ***
    Excessive debugging output
    *** Controllers ***
    Texas Instruments PCILynx support
    OHCI-1394 support
    *** Protocols ***
    OHCI-1394 Video support
    SBP-2 support (Harddisks etc.)
      Enable replacement for physical DMA in SBP2
    IP over 1394
    OHCI-DV I/O support (deprecated)
    Raw IEEE1394 I/O support

Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
16 years agofirewire: Kconfig menu touch-up
Stefan Richter [Sat, 14 Jun 2008 12:24:53 +0000 (14:24 +0200)] 
firewire: Kconfig menu touch-up

Emphasize the recommendation to build only one stack.
Trim the prompts to better fit into short attention spans.

Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
16 years agofirewire: deadline for PHY config transmission
Stefan Richter [Wed, 18 Jun 2008 16:20:45 +0000 (18:20 +0200)] 
firewire: deadline for PHY config transmission

If the low-level driver failed to initialize a card properly without
noticing it, fw-core was blocked indefinitely when trying to send a
PHY config packet.  This hung up the events kernel thread, e.g. locked
up keyboard input.
https://bugzilla.redhat.com/show_bug.cgi?id=444694
https://bugzilla.redhat.com/show_bug.cgi?id=446763

This problem was introduced between 2.6.25 and 2.6.26-rc1 by commit
2a0a2590498be7b92e3e76409c9b8ee722e23c8f "firewire: wait until PHY
configuration packet was transmitted (fix bus reset loop)".

The solution is to wait with timeout.  I tested it with 7 different
working controllers and 1 non-working controller.  On the working ones,
the packet callback complete()s usually --- but not always --- before a
timeout of 10ms.  Hence I chose a safer timeout of 100ms.

On the few tests with the non-working controller ALi M5271, PHY config
packet transmission always timed out so far.  (Fw-ohci needs to be fixed
for this controller independently of this deadline fix.  Often the core
doesn't even attempt to send a phy config because not even self ID
reception works.)

Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
16 years agofirewire: fw-ohci: unify printk prefixes
Stefan Richter [Sat, 14 Jun 2008 12:23:43 +0000 (14:23 +0200)] 
firewire: fw-ohci: unify printk prefixes

The messages which can be enabled by fw-ohci's debug module parameter
are changed from KERN_DEBUG to KERN_NOTICE level and uniformly prefixed
with "firewire_ohci: ".  This further simplifies communication with
users when we ask them to capture debug messages.

Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
16 years agofirewire: fill_bus_reset_event needs lock protection
Stefan Richter [Fri, 6 Jun 2008 20:11:30 +0000 (22:11 +0200)] 
firewire: fill_bus_reset_event needs lock protection

Callers of fill_bus_reset_event() have to take card->lock.  Otherwise
access to node data may oops if node removal is in progress.

A lockless alternative would be

- event->local_node_id = card->local_node->node_id;
+ tmp = fw_node_get(card->local_node);
+ event->local_node_id = tmp->node_id;
+ fw_node_put(tmp);

and ditto with the other node pointers which fill_bus_reset_event()
accesses.  But I went the locked route because one of the two callers
already holds the lock.  As a bonus, we don't need the memory barrier
anymore because device->generation and device->node_id are written in
a card->lock protected section.

Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
Signed-off-by: Kristian Høgsberg <krh@redhat.com>
16 years agofirewire: fw-ohci: write selfIDBufferPtr before LinkControl.rcvSelfID
Stefan Richter [Thu, 5 Jun 2008 18:50:53 +0000 (20:50 +0200)] 
firewire: fw-ohci: write selfIDBufferPtr before LinkControl.rcvSelfID

OHCI 1.1 clause 5.10 requires that selfIDBufferPtr is valid when a 1 is
written into LinkControl.rcvSelfID.

This driver bug has so far not been known to cause harm because most
chips obviously accept a later selfIDBufferPtr write, at least before
HCControl.linkEnable is written.

Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
Signed-off-by: Jarod Wilson <jwilson@redhat.com>
Signed-off-by: Kristian Høgsberg <krh@redhat.com>
16 years agofirewire: fw-ohci: disable PHY packet reception into AR context
Stefan Richter [Thu, 5 Jun 2008 18:49:38 +0000 (20:49 +0200)] 
firewire: fw-ohci: disable PHY packet reception into AR context

We want the rcvPhyPkt bit in LinkControl off before we start using the
chip.  However, the spec says that the reset value of it is undefined.
Hence switch it explicitly off.

https://bugzilla.redhat.com/show_bug.cgi?id=244576#c48 shows that for
example the nForce2 integrated FireWire controller seems to have it on
by default.

Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
Signed-off-by: Jarod Wilson <jwilson@redhat.com>
16 years agofirewire: fw-ohci: use of uninitialized data in AR handler
Stefan Richter [Sat, 31 May 2008 17:36:06 +0000 (19:36 +0200)] 
firewire: fw-ohci: use of uninitialized data in AR handler

header_length and payload_length are filled with random data if an
unknown tcode was read from the AR buffer (i.e. if the AR buffer
contained invalid data).

We still need a better strategy to recover from this, but at least
handle_ar_packet now doesn't return out of bound buffer addresses
anymore.

Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
16 years agofirewire: don't panic on invalid AR request buffer
Stefan Richter [Sat, 31 May 2008 17:01:26 +0000 (19:01 +0200)] 
firewire: don't panic on invalid AR request buffer

BUG() at this place is wrong.  (Unless if the low level driver would
already do higher-level input validation of incoming request headers.)

Invalid incoming requests or bugs in the controller which corrupt the
AR-req buffer needlessly crashed the box because this is run in tasklet
context.

Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
16 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/wim/linux-2.6-watchdog
Linus Torvalds [Wed, 18 Jun 2008 19:55:00 +0000 (12:55 -0700)] 
Merge git://git./linux/kernel/git/wim/linux-2.6-watchdog

* git://git.kernel.org/pub/scm/linux/kernel/git/wim/linux-2.6-watchdog:
  Revert "[WATCHDOG] hpwdt: Fix NMI handling."
  [WATCHDOG] hpwdt: Add CFLAGS to get driver working
  Revert "[WATCHDOG] make watchdog/hpwdt.c:asminline_call() static"

16 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi-rc-fixes-2.6
Linus Torvalds [Wed, 18 Jun 2008 18:55:43 +0000 (11:55 -0700)] 
Merge git://git./linux/kernel/git/jejb/scsi-rc-fixes-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi-rc-fixes-2.6:
  [SCSI] dpt_i2o: Add PROC_IA64 define
  [SCSI] scsi_host regression: fix scsi host leak
  [SCSI] sr: fix corrupt CD data after media change and delay

16 years agoMerge branch 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc
Linus Torvalds [Wed, 18 Jun 2008 18:55:19 +0000 (11:55 -0700)] 
Merge branch 'merge' of git://git./linux/kernel/git/paulus/powerpc

* 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc:
  [POWERPC] Clear sub-page HPTE present bits when demoting page size
  [POWERPC] 4xx: Clear new TLB cache attribute bits in Data Storage vector

16 years agoMerge branch 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jack/linux...
Linus Torvalds [Wed, 18 Jun 2008 18:55:03 +0000 (11:55 -0700)] 
Merge branch 'for_linus' of git://git./linux/kernel/git/jack/linux-udf-2.6

* 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jack/linux-udf-2.6:
  udf: restore UDFFS_DEBUG to being undefined by default

16 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
Linus Torvalds [Wed, 18 Jun 2008 18:48:40 +0000 (11:48 -0700)] 
Merge git://git./linux/kernel/git/davem/net-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6: (43 commits)
  netlink: genl: fix circular locking
  Revert "mac80211: Use skb_header_cloned() on TX path."
  af_unix: fix 'poll for write'/ connected DGRAM sockets
  tun: Proper handling of IPv6 header in tun driver when TUN_NO_PI is set
  atl1: relax eeprom mac address error check
  net/enc28j60: low power mode
  net/enc28j60: section fix
  sky2: 88E8040T pci device id
  netxen: download firmware in pci probe
  netxen: cleanup debug messages
  netxen: remove global physical_port array
  netxen: fix portnum for hp mezz cards
  ibm_newemac: select CRC32 in Kconfig
  xfrm: fix fragmentation for ipv4 xfrm tunnel
  netfilter: nf_conntrack_h323: fix module unload crash
  netfilter: nf_conntrack_h323: fix memory leak in module initialization error path
  netfilter: nf_nat: fix RCU races
  atm: [he] send idle cells instead of unassigned when in SDH mode
  atm: [he] limit queries to the device's register space
  atm: [br2864] fix routed vcmux support
  ...

16 years agoRevert "[WATCHDOG] hpwdt: Fix NMI handling."
Wim Van Sebroeck [Wed, 18 Jun 2008 16:22:48 +0000 (16:22 +0000)] 
Revert "[WATCHDOG] hpwdt: Fix NMI handling."

The old setup works better.

Signed-off-by: Thomas Mingarelli <Thomas.Mingarelli@hp.com>
Signed-off-by: Wim Van Sebroeck <wim@iguana.be>
16 years ago[POWERPC] Clear sub-page HPTE present bits when demoting page size
Paul Mackerras [Wed, 18 Jun 2008 06:40:35 +0000 (16:40 +1000)] 
[POWERPC] Clear sub-page HPTE present bits when demoting page size

When we demote a slice from 64k to 4k, and we are about to insert an
HPTE for a 4k subpage and we notice that there is an existing 64k
HPTE, we first invalidate that HPTE before inserting the new 4k
subpage HPTE.  Since the bits that encode which hash bucket the old
HPTE was in overlap with the bits that encode which of the 16 subpages
have HPTEs, we need to clear out the subpage HPTE-present bits before
starting to insert HPTEs for the 4k subpages.  If we don't do that, we
can erroneously think that a subpage already has an HPTE when it
doesn't.

That in itself wouldn't be such a problem except that when we go to
update the HPTE that we think is present on machines with a
hypervisor, the hypervisor can tell us that the HPTE we think is there
is actually there even though it isn't, which can lead to a process
getting stuck in a loop, continually faulting.  The reason for the
confusion is that the AVPN (abbreviated virtual page number) we are
looking for in the HPTE for a 4k subpage can actually match the AVPN
in a stale HPTE for another 64k page.  For example, the HPTE for
the 4k subpage at 0x84000f000 will be in the same hash bucket and have
the same AVPN as the HPTE for the 64k page at 0x8400f0000.

This fixes the code to clear out the subpage HPTE-present bits.

Signed-off-by: Paul Mackerras <paulus@samba.org>
16 years ago[POWERPC] 4xx: Clear new TLB cache attribute bits in Data Storage vector
Josh Boyer [Tue, 17 Jun 2008 22:34:39 +0000 (08:34 +1000)] 
[POWERPC] 4xx: Clear new TLB cache attribute bits in Data Storage vector

A recent commit added support for the new 440x6 and 464 cores that have the
added WL1, IL1I, IL1D, IL2I, and ILD2 bits for the caching attributes in the
TLBs.  The new bits were cleared in the finish_tlb_load function, however a
similar bit of code was missed in the DataStorage interrupt vector.

Signed-off-by: Josh Boyer <jwboyer@linux.vnet.ibm.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>
16 years agonetlink: genl: fix circular locking
Patrick McHardy [Wed, 18 Jun 2008 09:07:07 +0000 (02:07 -0700)] 
netlink: genl: fix circular locking

genetlink has a circular locking dependency when dumping the registered
families:

- dump start:
genl_rcv()            : take genl_mutex
genl_rcv_msg()        : call netlink_dump_start() while holding genl_mutex
netlink_dump_start(),
netlink_dump()        : take nlk->cb_mutex
ctrl_dumpfamily()     : try to detect this case and not take genl_mutex a
                        second time

- dump continuance:
netlink_rcv()         : call netlink_dump
netlink_dump          : take nlk->cb_mutex
ctrl_dumpfamily()     : take genl_mutex

Register genl_lock as callback mutex with netlink to fix this. This slightly
widens an already existing module unload race, the genl ops used during the
dump might go away when the module is unloaded. Thomas Graf is working on a
seperate fix for this.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years agoRevert "mac80211: Use skb_header_cloned() on TX path."
David S. Miller [Wed, 18 Jun 2008 08:19:51 +0000 (01:19 -0700)] 
Revert "mac80211: Use skb_header_cloned() on TX path."

This reverts commit 608961a5eca8d3c6bd07172febc27b5559408c5d.

The problem is that the mac80211 stack not only needs to be able to
muck with the link-level headers, it also might need to mangle all of
the packet data if doing sw wireless encryption.

This fixes kernel bugzilla #10903.  Thanks to Didier Raboud (for the
bugzilla report), Andrew Prince (for bisecting), Johannes Berg (for
bringing this bisection analysis to my attention), and Ilpo (for
trying to analyze this purely from the TCP side).

In 2.6.27 we can take another stab at this, by using something like
skb_cow_data() when the TX path of mac80211 ends up with a non-NULL
tx->key.  The ESP protocol code in the IPSEC stack can be used as a
model for implementation.

Signed-off-by: David S. Miller <davem@davemloft.net>
16 years agoaf_unix: fix 'poll for write'/ connected DGRAM sockets
Rainer Weikusat [Wed, 18 Jun 2008 05:28:05 +0000 (22:28 -0700)] 
af_unix: fix 'poll for write'/ connected DGRAM sockets

The unix_dgram_sendmsg routine implements a (somewhat crude)
form of receiver-imposed flow control by comparing the length of the
receive queue of the 'peer socket' with the max_ack_backlog value
stored in the corresponding sock structure, either blocking
the thread which caused the send-routine to be called or returning
EAGAIN. This routine is used by both SOCK_DGRAM and SOCK_SEQPACKET
sockets. The poll-implementation for these socket types is
datagram_poll from core/datagram.c. A socket is deemed to be writeable
by this routine when the memory presently consumed by datagrams
owned by it is less than the configured socket send buffer size. This
is always wrong for connected PF_UNIX non-stream sockets when the
abovementioned receive queue is currently considered to be full.
'poll' will then return, indicating that the socket is writeable, but
a subsequent write result in EAGAIN, effectively causing an
(usual) application to 'poll for writeability by repeated send request
with O_NONBLOCK set' until it has consumed its time quantum.

The change below uses a suitably modified variant of the datagram_poll
routines for both type of PF_UNIX sockets, which tests if the
recv-queue of the peer a socket is connected to is presently
considered to be 'full' as part of the 'is this socket
writeable'-checking code. The socket being polled is additionally
put onto the peer_wait wait queue associated with its peer, because the
unix_dgram_sendmsg routine does a wake up on this queue after a
datagram was received and the 'other wakeup call' is done implicitly
as part of skb destruction, meaning, a process blocked in poll
because of a full peer receive queue could otherwise sleep forever
if no datagram owned by its socket was already sitting on this queue.
Among this change is a small (inline) helper routine named
'unix_recvq_full', which consolidates the actual testing code (in three
different places) into a single location.

Signed-off-by: Rainer Weikusat <rweikusat@mssgmbh.com>
Signed-off-by: David S. Miller <davem@davemloft.net>