[MIPS] Ignore unresolved weak symbols in modules.

Signed-off-by: Atsushi Nemoto <[email protected]>
Signed-off-by: Ralf Baechle <[email protected]>

[MIPS] Fix SMP now that fixup_cpu_present_map is gone.

Signed-off-by: Ralf Baechle <[email protected]>

[MIPS] Remove duplicate declaration of cpu_online_map.

Signed-off-by: Ralf Baechle <[email protected]>

Merge /linux/kernel/git/brodo/pcmcia-fixes-2.6

* master.kernel.org:/pub/scm/linux/kernel/git/brodo/pcmcia-fixes-2.6:
  [PATCH] pcmcia: fix zeroing of cm4000_cs.c data
  [PATCH] pcmcia: missing pcmcia_get_socket() result check

[PATCH] uml: add -ffreestanding to CFLAGS

From: Jeff Dike <[email protected]>

This fixes the undefined reference to strcpy seen when building modules on
i386.  Tracked down by Al Viro.

Signed-off-by: Jeff Dike <[email protected]>
Cc: Paolo 'Blaisorblade' Giarrusso <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] uml: more __user annotations

From: Al Viro <[email protected]>

uml __user annotations

Signed-off-by: Al Viro <[email protected]>
Signed-off-by: Jeff Dike <[email protected]>
Cc: Paolo 'Blaisorblade' Giarrusso <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] uml: __user annotation in arch_prctl

From: Al Viro <[email protected]>

fix uml/amd64 prctl()

put_user() there should go to (long __user *)addr, not &addr

Signed-off-by: Al Viro <[email protected]>
Signed-off-by: Jeff Dike <[email protected]>
Cc: Paolo 'Blaisorblade' Giarrusso <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] uml: fix a typo in do_uml_initcalls

From: Jeff Dike <[email protected]>

We had a spurious semicolon somehow.

Signed-off-by: Jeff Dike <[email protected]>
Cc: Paolo 'Blaisorblade' Giarrusso <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] uml: fix wall_to_monotonic initialization

From: Jeff Dike <[email protected]>

Initialize wall_to_monotonic correctly.  This fixes a problem where sleeps
lasted about one secone less than they should.  This also called for a bit of
code restructuring, following a patch which Blaisorblade had been keeping.

Signed-off-by: Jeff Dike <[email protected]>
Cc: Paolo 'Blaisorblade' Giarrusso <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] uml: add asm/irqflags.h

From: Jeff Dike <[email protected]>

Add an empty asm/irqflags.h, which seems to satisfy the lock validator enough
that UML builds.

Signed-off-by: Jeff Dike <[email protected]>
Cc: Paolo 'Blaisorblade' Giarrusso <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] m48t86: ia64 build fix

From: Andrew Morton <[email protected]>

drivers/rtc/rtc-m48t86.c: In function `m48t86_rtc_read_time':
drivers/rtc/rtc-m48t86.c:51: error: structure has no member named `ia64_mv'
drivers/rtc/rtc-m48t86.c:55: error: structure has no member named `ia64_mv'
drivers/rtc/rtc-m48t86.c:56: error: structure has no member named `ia64_mv'
drivers/rtc/rtc-m48t86.c:57: error: structure has no member named `ia64_mv'
drivers/rtc/rtc-m48t86.c:58: error: structure has no member named `ia64_mv'
drivers/rtc/rtc-m48t86.c:60: error: structure has no member named `ia64_mv'

readb() and writeb() are macros on ia64.

Cc: Alessandro Zummo <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] sata_sil24: SII3124 sata driver endian problem

From: "Rune Torgersen" <[email protected]>

Fix an endian issue in the sil24 driver.

Signed-off-by: Rune Torgersen <[email protected]>
Acked-by: Jeff Garzik <[email protected]>
Cc: Tejun Heo <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] sbp2: fix check of return value of hpsb_allocate_and_register_addrspace()

From: Stefan Richter <[email protected]>

I added a failure check in patch "sbp2: variable status FIFO address (fix
login timeout)" --- alas for a wrong error value.  This is a bug since
Linux 2.6.16.  Leads to NULL pointer dereference if the call failed, and
bogus failure handling if call succeeded.

Signed-off-by: Stefan Richter <[email protected]>
Cc: <[email protected]>
Cc: Ben Collins <[email protected]>
Cc: Jody McIntyre <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] MAINTAINERS: Add entries for BNX2 and TG3

From: "Michael Chan" <[email protected]>

Add maintainer entries for Broadcom BNX2 and TG3 drivers.

Signed-off-by: Michael Chan <[email protected]>
Cc: "David S. Miller" <[email protected]>
Acked-by: Jeff Garzik <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] Implement get / set tso for forcedeth driver

From: Zachary Amsden <[email protected]>

Signed-off-by: Zachary Amsden <[email protected]>
Cc: Ayaz Abdulla <[email protected]>
Cc: Manfred Spraul <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] pmf_register_irq_client() gives sleep with locks held warning

From: Benjamin Herrenschmidt <[email protected]>

This fixes request_irq() potentially called from atomic context.

Signed-off-by: Benjamin Herrenschmidt <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] fs/namei.c: Call to file_permission() under a spinlock in do_lookup_path()

From: Trond Myklebust <[email protected]>

We're presently running lock_kernel() under fs_lock via nfs's ->permission
handler.  That's a ranking bug and sometimes a sleep-in-spinlock bug.  This
problem was introduced in the openat() patchset.

We should not need to hold the current->fs->lock for a codepath that doesn't
use current->fs.

[[email protected]: fix error path]
Signed-off-by: Trond Myklebust <[email protected]>
Cc: Al Viro <[email protected]>
Signed-off-by: Sergey Vlasov <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] alpha: SMP IRQ routing fix

From: Ivan Kokshaysky <[email protected]>

After removal of fixup_cpu_present_map() function Alpha ended up with an empty
cpu_present_map, so secondary CPUs on SMP systems are not being started.

Worse, on some platforms we route interrupts to secondary CPUs using
cpu_possible_map which is still populated properly.  As a result, these
interrupts go nowhere so the machines like DP264 aren't able to boot even with
a primary CPU.

Fixed basically by s/cpu_present_mask/cpu_present_map/.

Thanks to Ernst Herzberg for reporting the bug and testing the fix.

Cc: Ernst Herzberg <[email protected]>
Cc: Richard Henderson <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] selinux: fix sb_lock/sb_security_lock nesting

From: Stephen Smalley <[email protected]>

Fix unsafe nesting of sb_lock inside sb_security_lock in
selinux_complete_init.  Detected by the kernel locking validator.

Signed-off-by: Stephen Smalley <[email protected]>
Acked-by: James Morris <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] Sparsemem build fix

From: Ralf Baechle <[email protected]>

<linux/mmzone.h> uses PAGE_SIZE, PAGE_SHIFT from <asm/page.h> without
including that header itself.  For some sparsemem configurations this may
result in build errors like:

  CC      init/initramfs.o
In file included from include/linux/gfp.h:4,
                 from include/linux/slab.h:15,
                 from include/linux/percpu.h:4,
                 from include/linux/rcupdate.h:41,
                 from include/linux/dcache.h:10,
                 from include/linux/fs.h:226,
                 from init/initramfs.c:2:
include/linux/mmzone.h:498:22: warning: "PAGE_SHIFT" is not defined
In file included from include/linux/gfp.h:4,
                 from include/linux/slab.h:15,
                 from include/linux/percpu.h:4,
                 from include/linux/rcupdate.h:41,
                 from include/linux/dcache.h:10,
                 from include/linux/fs.h:226,
                 from init/initramfs.c:2:
include/linux/mmzone.h:526: error: `PAGE_SIZE' undeclared here (not in a function)
include/linux/mmzone.h: In function `__pfn_to_section':
include/linux/mmzone.h:573: error: `PAGE_SHIFT' undeclared (first use in this function)
include/linux/mmzone.h:573: error: (Each undeclared identifier is reported only once
include/linux/mmzone.h:573: error: for each function it appears in.)
include/linux/mmzone.h: In function `pfn_valid':
include/linux/mmzone.h:578: error: `PAGE_SHIFT' undeclared (first use in this function)
make[1]: *** [init/initramfs.o] Error 1
make: *** [init] Error 2

Signed-off-by: Ralf Baechle <[email protected]>
Seems-reasonable-to: Andy Whitcroft <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] s390: cio non-unique path group ids

From: Peter Oberparleiter <[email protected]>

The path grouping can fail due to non-unique pathgroup-IDs.  The source for
the CPU-ID part of the ID was incorrectly specified on 64 bit systems.
Additionally, the length of the ID was too large due to incorrect data packing
declaration.  Fix CPU-ID lowcore address and add missing packing declaration.

Signed-off-by: Peter Oberparleiter <[email protected]>
Signed-off-by: Martin Schwidefsky <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] s390: irb memcpy argument swap

From: Cornelia Huck <[email protected]>

Swapped memcpy arguments in ccw_device_irq() when doing basic sense after
unsolicited interrupt.

Signed-off-by: Cornelia Huck <[email protected]>
Signed-off-by: Martin Schwidefsky <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] nmclan_cs: dereferencing skb after netif_rx()

From: Florin Malita <[email protected]>

The skb may be gone after netif_rx(), we can't use 'skb->len' to update the
stats.  'pkt_len' should work instead.

Coverity CID: 911.

Signed-off-by: Florin Malita <[email protected]>
Cc: Dominik Brodowski <[email protected]>
Acked-by: Jeff Garzik <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

Merge /pub/scm/linux/kernel/git/davem/sparc-2.6

* master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6:
  [SPARC64]: Fix missing fold at end of checksums.

IPoIB: Fix AH leak at interface down

When ipoib_stop() is called it first calls netif_stop_queue() to stop
the kernel from passing more packets to the network driver. However,
the completion handler may call netif_wake_queue() re-enabling packet
transfer.

This might result in leaks (we see AH leaks which we think can be
attributed to this bug) as new packets get posted while the interface
is going down.

Signed-off-by: Eli Cohen <[email protected]>
Signed-off-by: Michael Tsirkin <[email protected]>
Signed-off-by: Roland Dreier <[email protected]>

[SPARC64]: Fix missing fold at end of checksums.

Both csum_partial() and the csum_partial_copy*() family of routines
forget to do a final fold on the computed checksum value on sparc64.
So do the standard Sparc "add + set condition codes, add carry"
sequence, then make sure the high 32-bits of the return value are
clear.

Based upon some excellent detective work and debugging done by
Richard Braun and Samuel Thibault.

Signed-off-by: David S. Miller <[email protected]>

Merge /linux/kernel/git/jejb/scsi-rc-fixes-2.6

* master.kernel.org:/pub/scm/linux/kernel/git/jejb/scsi-rc-fixes-2.6:
  [SCSI] scsi_lib.c: properly count the number of pages in scsi_req_map_sg()
  [SCSI] scsi_transport_sas: make write attrs writeable
  [SCSI] scsi_transport_sas; fix user_scan
  [SCSI] ppa: fix for machines with highmem
  [SCSI] mptspi: reset handler shouldn't be called for other bus protocols
  [SCSI] Blacklist entry for HP dat changer

[TCP] tcp_highspeed: Fix problem observed by Xiaoliang (David) Wei

When snd_cwnd is smaller than 38 and the connection is in
congestion avoidance phase (snd_cwnd > snd_ssthresh), the snd_cwnd
seems to stop growing.

The additive increase was confused because C array's are 0 based.

Signed-off-by: Stephen Hemminger <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

Merge master.kernel.org:/home/rmk/linux-2.6-serial

* master.kernel.org:/home/rmk/linux-2.6-serial:
  [SERIAL] Update parity handling documentation

Merge master.kernel.org:/home/rmk/linux-2.6-arm

* master.kernel.org:/home/rmk/linux-2.6-arm:
  [ARM] 3540/1: ixp23xx: deal with gap in interrupt bitmasks
  [ARM] 3539/1: ixp23xx: fix __arch_ixp23xx_is_coherent() for A1 stepping

Merge /pub/scm/linux/kernel/git/davem/sparc-2.6

* master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6:
  [SPARC64]: Fix D-cache corruption in mremap
  [SPARC64]: Make smp_processor_id() functional before start_kernel()

[ARM] 3540/1: ixp23xx: deal with gap in interrupt bitmasks

Patch from Lennert Buytenhek

On the ixp23xx, the microengine thread interrupt sources are numbered
56..119, but their mask/status bits are located in bit positions 64..127
in the various registers in the interrupt controller (bit positions
56..63 are unused.)

We don't deal with this, so currently, when asked to enable IRQ 64, we
will enable IRQ 56 instead.

The only interrupts >= 64 are the thread interrupt sources, and there
are no in-tree users of those yet, so this is fortunately not a big
problem, but this needs fixing anyway.

Signed-off-by: Lennert Buytenhek <[email protected]>
Signed-off-by: Russell King <[email protected]>

[ARM] 3539/1: ixp23xx: fix __arch_ixp23xx_is_coherent() for A1 stepping

Patch from Lennert Buytenhek

The current __ixp23xx_arch_is_coherent() check assumes that the
lower byte of IXP23XX_PRODUCT_ID is identical to the lower byte of
processor_id, but this is not the case, and because of this we were
incorrectly enabling coherency on A1 stepping CPUs.

Stepping A1 of the ixp2350, which has a PRODUCT_ID of 0x401, has '02'
in the lower byte of processor_id, while A2, with a PRODUCT_ID of
0x402, has '04' in the lower byte of processor_id.

So, to check for >= A2, we really need to check the lower byte of
processor_id against >= 4.

Signed-off-by: Lennert Buytenhek <[email protected]>
Signed-off-by: Russell King <[email protected]>

[PATCH] slab.c: fix offslab_limit bug

mm/slab.c's offlab_limit logic is totally broken.

Firstly, "offslab_limit" is a global variable while it should either be
calculated in situ or should be passed in as a parameter.

Secondly, the more serious problem with it is that the condition for
calculating it:

               if (!(OFF_SLAB(sizes->cs_cachep))) {
                       offslab_limit = sizes->cs_size - sizeof(struct slab);
                       offslab_limit /= sizeof(kmem_bufctl_t);

is in total disconnect with the condition that makes use of it:

               /* More than offslab_limit objects will cause problems */
               if ((flags & CFLGS_OFF_SLAB) && num > offslab_limit)
                       break;

but due to offslab_limit being a global variable this breakage was
hidden.

Up until lockdep came along and perturbed the slab sizes sufficiently so
that the first off-slab cache would still see a (non-calculated) zero
value for offslab_limit and would panic with:

  kmem_cache_create: couldn't create cache size-512.

  Call Trace:
   [<ffffffff8020a5b9>] show_trace+0x96/0x1c8
   [<ffffffff8020a8f0>] dump_stack+0x13/0x15
   [<ffffffff8022994f>] panic+0x39/0x21a
   [<ffffffff80270814>] kmem_cache_create+0x5a0/0x5d0
   [<ffffffff80aced62>] kmem_cache_init+0x193/0x379
   [<ffffffff80abf779>] start_kernel+0x17f/0x218
   [<ffffffff80abf263>] _sinittext+0x263/0x26a

  Kernel panic - not syncing: kmem_cache_create(): failed to create slab `size-512'

Paolo Ornati's config on x86_64 managed to trigger it.

The fix is to move the calculation to the place that makes use of it.
This also makes slab.o 54 bytes smaller.

Btw., the check itself is quite silly. Its intention is to test whether
the number of objects per slab would be higher than the number of slab
control pointers possible. In theory it could be triggered: if someone
tried to allocate 4-byte objects cache and explicitly requested with
CFLGS_OFF_SLAB. So i kept the check.

Out of historic interest i checked how old this bug was and it's
ancient, 10 years old! It is the oldest hidden and then truly triggering
bugs i ever saw being fixed in the kernel!

Signed-off-by: Ingo Molnar <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[SERIAL] Update parity handling documentation

Update documentation to match reality. INPCK controls whether input
parity checking is enabled.

Signed-off-by: Peter Korsgaard <[email protected]>
Signed-off-by: Russell King <[email protected]>

[SPARC64]: Fix D-cache corruption in mremap

If we move a mapping from one virtual address to another,
and this changes the virtual color of the mapping to those
pages, we can see corrupt data due to D-cache aliasing.

Check for and deal with this by overriding the move_pte()
macro.  Set things up so that other platforms can cleanly
override the move_pte() macro too.

Signed-off-by: David S. Miller <[email protected]>

[PATCH] pcmcia: fix zeroing of cm4000_cs.c data

Fix the incorrect calculation of how much to zero out in struct cm4000_dev
on device initialization.

Signed-off-by: Dominik Brodowski <[email protected]>

[PATCH] pcmcia: missing pcmcia_get_socket() result check

The result of pcmcia_get_socket() may be NULL but ds_event() uses it
without checking.

Coverity CID: 436.

Signed-off-by: Florin Malita <[email protected]>
Signed-off-by: Dominik Brodowski <[email protected]>

[PATCH] cfq-iosched: busy_rr fairness fix

Now that we select busy_rr for possible service, insert entries at the
back of that list instead of at the front.

Signed-off-by: Jens Axboe <[email protected]>

[SCSI] scsi_lib.c: properly count the number of pages in scsi_req_map_sg()

The calculation of nr_pages in scsi_req_map_sg() doesn't account for
the fact that the first page could have an offset that pushes the end
of the buffer onto a new page.

Signed-off-by: Bryan Holty <[email protected]>
Signed-off-by: James Bottomley <[email protected]>

[PATCH] cfq-iosched: fix bug in timer handling for the idle class

There's a small window from when the timer is entered and we grab
the queue lock, where cfq_set_active_queue() could be rearming the
timer for us. Seen in the wild on a 12-way ppc box. Fix this by
just using mod_timer(), which will do the right thing for us.

Signed-off-by: Jens Axboe <[email protected]>

[PATCH] cfq-iosched: Detect hardware queueing

If the hardware is doing real queueing, decide that it's worthless to
idle the hardware. It does reasonable simultaneous io in that case
anyways, and the idling hurts some work loads.

Signed-off-by: Jens Axboe <[email protected]>

[PATCH] cfq-iosched: Detect idle process issuing async request

If we are anticipating a sync request from this process and we are
waiting for that and see an async request come in, expire that slice
and move on.

Signed-off-by: Jens Axboe <[email protected]>

[PATCH] cfq-iosched: check busy queues before deciding we are idle

For just one busy queue (like async write out), we often overlooked
that we could queue more io and decided we were idle instead. This causes
us quite a bit of performance loss.

Signed-off-by: Jens Axboe <[email protected]>

Merge branch 'upstream' of git://ftp.linux-mips.org/upstream-linus

* 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus:
  [MIPS] Treat R14000 like R10000.
  [MIPS] Remove EXPERIMENTAL from PAGE_SIZE_16KB
  [MIPS] Update/Fix instruction definitions
  [MIPS] DSP and MDMX share the same config flag bit.
  [MIPS] Fix deadlock on MP with cache aliases.
  [MIPS] Use generic STABS_DEBUG macro.
  [MIPS] Create consistency in "system type" selection.
  [MIPS] Use generic DWARF_DEBUG
  [MIPS] Fix kgdb exception handler from user mode.
  [MIPS] Update struct sigcontext member names
  [MIPS] Update/fix futex assembly
  [MIPS] Remove support for sysmips(2) SETNAME and MIPS_RDNVRAM operations.
  [MIPS] Fix detection and handling of the 74K processor.
  [MIPS] Add missing 34K processor IDs
  [MIPS] Fix marking buddy of pte global for MIPS32 w/36-bit physical address
  [MIPS] AU1xxx mips_timer_interrupt() fixes
  [MIPS] Fix typo

[MIPS] Treat R14000 like R10000.

Signed-off-by: Joshua Kinard <[email protected]>
Signed-off-by: Ralf Baechle <[email protected]>

[MIPS] Remove EXPERIMENTAL from PAGE_SIZE_16KB

This is known to be working fine for a while.  While at it also update
and fix the help texts.

Signed-off-by: Ralf Baechle <[email protected]>

[MIPS] Update/Fix instruction definitions

A small bugfix for up to now unused instruction definitions, and a
somewhat larger update to cover MIPS32R2 instructions.

Signed-off-by: Thiemo Seufer <[email protected]>
Signed-off-by: Ralf Baechle <[email protected]>

[MIPS] DSP and MDMX share the same config flag bit.

Clarify comment.

Signed-off-by: Thiemo Seufer <[email protected]>
Signed-off-by: Ralf Baechle <[email protected]>

[MIPS] Fix deadlock on MP with cache aliases.

A proper fix would involve introducing the notion of shared caches but
at this stage of 2.6.17 that's going to be too intrusive and not needed
for current hardware; aside I think some discussion will be needed.

So for now on the affected SMP configurations which happen to suffer from
cache aliases we make use of the fact that a single cache will be shared
by all processors.  This solves the deadlock issue and will improve
performance by getting rid of the smp_call_function overhead.

Signed-off-by: Ralf Baechle <[email protected]>

[MIPS] Use generic STABS_DEBUG macro.

Signed-off-by: Atsushi Nemoto <[email protected]>
Signed-off-by: Ralf Baechle <[email protected]>

[MIPS] Create consistency in "system type" selection.

The "system type" Kconfig options on MIPS are not consistent.  For
some platforms, only the name is listed while other entries are
prepended with "Support for".  Remove this as it doesn't make sense
when describing the "system type".

Signed-off-by: Martin Michlmayr <[email protected]>
Signed-off-by: Ralf Baechle <[email protected]>

[MIPS] Use generic DWARF_DEBUG

When debugging a kernel compiled by gcc 4.1 with gdb 6.4, gdb could
not show filename, linenumber, etc.  It seems fixed if I used generic
DWARF_DEBUG macro.  Although gcc 3.x seems work without this change,
it would be better to use the generic macro unless there were
something MIPS specific.

Signed-off-by: Atsushi Nemoto <[email protected]>
Signed-off-by: Ralf Baechle <[email protected]>

[MIPS] Fix kgdb exception handler from user mode.

Fix a calculation of saved vector address in trap_low.

(damage done by lmo f4c72cc737561aab0d9c7f877abbc0a853f1c465)

Signed-off-by: Ralf Baechle <[email protected]>

[MIPS] Update struct sigcontext member names

Rename the 64-bit sc_hi and sc_lo arrays to use the same names
as the 32-bit struct sigcontext (sc_mdhi, sc_hi1, et cetera).

Signed-off-by: Daniel Jacobowitz <[email protected]>
Signed-off-by: Ralf Baechle <[email protected]>

[MIPS] Update/fix futex assembly

 o Implement futex_atomic_op_inuser() operation
 o Don't use the R10000-ll/sc bug workaround version for every processor.
   branch likely is deprecated and some historic ll/sc processors don't
   implement it.  In any case it's slow.

Signed-off-by: Ralf Baechle <[email protected]>

[MIPS] Remove support for sysmips(2) SETNAME and MIPS_RDNVRAM operations.

SETNAME only had a minor defect but probably never had a user and
MIPS_RDNVRAM was unimplemented anyway.

Signed-off-by: Ralf Baechle <[email protected]>

[MIPS] Fix detection and handling of the 74K processor.

Nothing exciting; Linux just didn't know it yet so this is most adding
a value to a case statement.

Signed-off-by: Chris Dearman <[email protected]>
Signed-off-by: Ralf Baechle <[email protected]>

[MIPS] Add missing 34K processor IDs

The 34K is very much like a 24K on steroids.

Signed-off-by: Ralf Baechle <[email protected]>

[MIPS] Fix marking buddy of pte global for MIPS32 w/36-bit physical address

In case of CONFIG_64BIT_PHYS_ADDR, set_pte() and pte_clear() functions
only set _PAGE_GLOBAL bit in the pte_low field of the buddy PTEs,
forgetting to propagate ito to pte_high. Thus, the both pages might not
really be made global for the CPU (since it AND's the G-bit of the
odd / even PTEs together to decide whether they're global or not). Thus,
if only a single page is allocated via vmalloc() or ioremap(), it's not
really global for CPU (and it must be, since this is kernel mapping),
and thus its ASID is compared against the current process' one -- so,
we'll get into trouble sooner or later...  Also, pte_none() will fail
on global pages because _PAGE_GLOBAL bit is set in both pte_low and
pte_high, and pte_val() will return u64 value consisting of those fields
concateneted.

Signed-off-by: Sergei Shtylyov <[email protected]>
Signed-off-by: Ralf Baechle <[email protected]>

[MIPS] AU1xxx mips_timer_interrupt() fixes

common/au1000/irq.c was missing a mips_timer_interrupt() prototype,
whereas in common/au1000/time.c the actual mips_timer_interrupt()
implementation was missing an irq_exit() invocation, causing a
preempt_count() leak.

Signed-off-by: Herbert Valerio Riedel <[email protected]>
Signed-off-by: Ralf Baechle <[email protected]>

[MIPS] Fix typo

Found by Chris Dearman ([email protected]).

Signed-off-by: Ralf Baechle <[email protected]>

[PATCH] ARM: Fix XScale PMD setting

The ARM Architecture Reference Manual lists bit 4 of the PMD as "implementation
defined" and it must be set to zero on Intel XScale CPUs or the cache does
not behave properly. Found by Mike Rapoport while debugging a flash issue
on the PXA255:

http://marc.10east.com/?l=linux-arm-kernel&m=114845287600782&w=1

Signed-off-by: Deepak Saxena <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] net/compat.h build fix

From: Andrew Morton <[email protected]>

Move the forward decl outside the ifdef, since we use it in both legs.

Should fix the spacr64 build error reported in
http://bugzilla.kernel.org/show_bug.cgi?id=6625

Acked-by: "David S. Miller" <[email protected]>
Cc: Cedric Pellerin <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] md: Fix badness in sysfs_notify caused by md_new_event

From: NeilBrown <[email protected]>

If an error is reported by a drive in a RAID array (which is done via
bi_end_io - in interrupt context), we call md_error and md_new_event which
calls sysfs_notify.  However sysfs_notify grabs a mutex and so cannot be
called in interrupt context.

This patch just creates a variant of md_new_event which avoids the sysfs
call, and uses that.  A better fix for later is to arrange for the event to
be called from user-context.

Note: avoiding the sysfs call isn't a problem as an error will not, by
itself, modify the sync_action attribute.  (We do still need to
wake_up(&md_event_waiters) as an error by itself will modify /proc/mdstat).

Signed-off-by: Neil Brown <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] sgiioc4: use mmio ops instead of port io

From: Jeremy Higdon <[email protected]>

This patch fixes a bug in sgiioc4 where it was using the default IDE port
I/O operations instead of MMIO.

The IDE part of the IOC4 chip uses MMIO to map the chip registers.
Unfortunately, the sgiioc4 driver uses the default port IO operations,
which happens to have worked for the past few years.  That's about to
change, however, thus this change from inX/outX to readX/writeX.

Signed-off-by: Jeremy Higdon <[email protected]>
Cc: Bartlomiej Zolnierkiewicz <[email protected]>
Cc: Alan Cox <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] maxinefb: Fix compilation error

From: Martin Michlmayr <[email protected]>

Fix the following compilation error:

CC      drivers/video/maxinefb.o
drivers/video/maxinefb.c:58: warning: initializer-string for array of chars is too long
drivers/video/maxinefb.c:58: warning: (near initialization for \u2018maxinefb_fix.id\u2019)
drivers/video/maxinefb.c:110: error: unknown field \u2018fb_get_fix\u2019 specified in initializer
drivers/video/maxinefb.c:110: error: \u2018gen_get_fix\u2019 undeclared here (not in a function)
drivers/video/maxinefb.c:111: error: unknown field \u2018fb_get_var\u2019 specified in initializer
drivers/video/maxinefb.c:111: error: \u2018gen_get_var\u2019 undeclared here (not in a function)
make[2]: *** [drivers/video/maxinefb.o] Error 1

Signed-off-by: Martin Michlmayr <[email protected]>
Signed-off-by: Antonino Daplas <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] au1100fb: Fix compilation

From: Rodolfo Giometti <[email protected]>

Fix the following warning on compilation:

drivers/video/au1100fb.c: In function `au1100fb_fb_setcolreg':
drivers/video/au1100fb.c:219: warning: ISO C90 forbids mixed declarations and code
drivers/video/au1100fb.c: In function `au1100fb_fb_pan_display':
drivers/video/au1100fb.c:321: warning: ISO C90 forbids mixed declarations and code
drivers/video/au1100fb.c: In function `au1100fb_fb_mmap':
drivers/video/au1100fb.c:387: warning: ISO C90 forbids mixed declarations and code
drivers/video/au1100fb.c: In function `au1100fb_drv_probe':
drivers/video/au1100fb.c:471: warning: unsigned int format, long unsigned int arg (arg 2)
drivers/video/au1100fb.c: At top level:
drivers/video/au1100fb.c:617: warning: initialization from incompatible pointer type
drivers/video/au1100fb.c:618: warning: initialization from incompatible pointer type

Signed-off-by: Rodolfo Giometti <[email protected]>
Signed-off-by: Antonino Daplas <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] powerpc: Fix boot on eMac

From: Benjamin Herrenschmidt <[email protected]>

Prevent calling of some platform functions on the clock chips of the eMac
as it seems to cause it to lockup at boot.  For now, add a quirk to prevent
that from happening.  Later, I might find out what's wrong and fix it but
that doesn't seem to be important as the machine appear to work fine
without running those.  It's possible that Darwin doesn't run them.

Signed-off-by: Benjamin Herrenschmidt <[email protected]>
Cc: Nathan Pilatzke <[email protected]>
Cc: Paul Mackerras <[email protected]>
Cc: Jean Delvare <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] hrtimer: export symbols

From: Stephen Hemminger <[email protected]>

I want to use the hrtimer's in the netem (Network Emulator) qdisc.  But the
necessary symbols aren't exported for module use.

Also needed by SystemTap.

Signed-off-by: Stephen Hemminger <[email protected]>
Acked-by: Ingo Molnar <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: "Stone, Joshua I" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] revert "swsusp add check for suspension of X controlled devices"

From: Andrew Morton <[email protected]>

Revert commit ff4da2e262d2509fe1bacff70dd00934be569c66.

It broke APM suspend, probably because APM doesn't switch back to a VT
when suspending.

Tracked down by Matt Mackall <[email protected]>

Rafael sayeth:
  "It only fixed the theoretical issue that a quick-handed user could
   switch to X after processes have been frozen and before the devices
   are suspended.

   With the current userland suspend tools it shouldn't be necessary."

Cc: Pavel Machek <[email protected]>
Cc: "Rafael J. Wysocki" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] IPMI: reserve I/O ports separately

From: Corey Minyard <[email protected]>

This patch is pretty important to get in for IPMI, new systems have been
changing the way ACPI and IPMI interact, and this works around the problems
for now.  This is a temporary fix until we get proper ACPI handling in
IPMI.

Fixed releasing already-allocated regions when a later request fails, and
forward-ported it to HEAD.

Some BIOSes reserve disjoint I/O regions in their ACPI tables for the IPMI
controller.  This causes problems when trying to register the entire I/O
region.  Therefore we must register each I/O port separately.

Signed-off-by: Jordan Hargrave <[email protected]>
Signed-off-by: Matt Domsch <[email protected]>
Signed-off-by: Corey Minyard <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] tpm: more bios log parsing fixes

From: Seiji Munetoh <[email protected]>

Change the binary output format to actual ACPI TCPA log structure since the
current format does not contain all event-data information that need to
verify the PCRs in TPM.  tpm_binary_bios_measurements_show() uses
get_event_name() to convert the binary event-data to ascii format, and puts
them as binary.  However, to verify the PCRs, the event-data must be a
actual binary event-data used by SHA1 calc.  in BIOS.

So, I think actual ACPI TCPA log is good for this binary output format.
That way, any userland tools easily parse this data with reference to TCG
PC specification.

Signed-off-by: Seiji Munetoh <[email protected]>
Acked-by: Kylene Hall <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] tpm: bios log parsing fixes

From: Seiji Munetoh <[email protected]>

Fix "tcpa_pc_event" misalignment between enum, strings and TCG PC spec and
output of the event which contains a hash data.

Signed-off-by: Seiji Munetoh <[email protected]>
Acked-by: Kylene Hall <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] spanned_pages is not updated at a case of memory hot-add

From: Yasunori Goto <[email protected]>

If hot-added memory's address is smaller than old area, spanned_pages will
not be updated.  It must be fixed.

example) Old zone_start_pfn = 0x60000, and spanned_pages = 0x10000
         Added new memory's start_pfn = 0x50000, and end_pfn = 0x60000

  new spanned_pages will be still 0x10000 by old code.
  (It should be updated to 0x20000.) Because old_zone_end_pfn will be
  0x70000, and end_pfn smaller than it. So, spanned_pages will not be
  updated.

In current code, spanned_pages is updated only when end_pfn is updated.
But, it should be updated by subtraction between bigger end_pfn and new
zone_start_pfn.

Signed-off-by: Yasunori Goto <[email protected]>
Signed-off-by: Dave Hansen <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] fbcon: fix scrollback with logo issue immediately after boot

From: David Hollister <[email protected]>

After the system boots with the logo, if the first action is a scrollback, the
screen may become garbled.  This patch ensures that the softback_curr value is
updated along with softback_in following the scrollback.

Signed-off-by: David Hollister <[email protected]>
Signed-off-by: Jordan Crouse <[email protected]>
Cc: "Antonino A. Daplas" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] ext3 resize: fix double unlock_super()

From: Andrew Morton <[email protected]>

Spotted by Jan Capek <[email protected]>

Cc: "Stephen C. Tweedie" <[email protected]>
Cc: Andreas Dilger <[email protected]>
Cc: Jan Capek <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[SPARC64]: Make smp_processor_id() functional before start_kernel()

Uses of smp_processor_id() get pushed earlier and earlier in
the start_kernel() sequence.  So just get it working before
we call start_kernel() to avoid all possible problems.

Signed-off-by: David S. Miller <[email protected]>

[PATCH] ARM: explicitly disable BTB on ixp2350

We don't enable the BTB on the ixp2350 as that can cause weird
crashes (erratum #42.)  However, some bootloaders enable the BTB,
which means that we have to disable the BTB explicitly.

Found thanks to Tom Rini.

Signed-off-by: Lennert Buytenhek <[email protected]>
Signed-off-by: Deepak Saxena <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

Revert "[PATCH] i386/x86_64: Force pci=noacpi on HP XW9300"

This reverts commit 5491d0f3e206beb95eeb506510d62a1dab462df1.

As per Andi:

  "After some discussion with people who have the affected system it
   seems best to revert for 2.6.17.  It broke a common BIOS workaround
   and PCI-X still doesn't work.  Alternative is for people to change
   the BIOS which seems to be better right now."

Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] x86_64: Don't do syscall exit tracing twice

int_ret_from_syscall already does syscall exit tracing, so
no need to do it again in the caller.

This caused problems for UML and some other special programs doing
syscall interception.

Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] x86_64: Fix off by one in bad_addr checking in find_e820_area

From: Robert Hentosh <[email protected]>

Actually, we just stumbled on a different bug found in find_e820_area() in
e820.c.  The following code does not handle the edge condition correctly:

   while (bad_addr(&addr, size) && addr+size < ei->addr + ei->size)
       ;
   last = addr + size;
   if ( last > ei->addr + ei->size )
       continue;

The second statement in the while loop needs to be a <= b so that it is the
logical negavite of the if (a > b) outside it. It needs to read:

   while (bad_addr(&addr, size) && addr+size <= ei->addr + ei->size)
       ;

In the case that failed bad_addr was returning an address that is exactly size
bellow the end of the e820 range.

AK: Again together with the earlier avoid edma fix this fixes
boot on a Dell PE6850/16GB

Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] x86_64: Handle empty node zero

From: Daniel Yeisley <[email protected]>

It is possible to boot a Unisys ES7000 with CPUs from multiple cells, and not
also include the memory from those cells.  This can create a scenario where
node 0 has cpus, but no associated memory.  The system will boot fine in a
configuration where node 0 has memory, but nodes 2 and 3 do not.

[AK: I rechecked the code and generic code seems to indeed handle that already.
Dan's original patch had a change for mm/slab.c that seems to be already in now.]

Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] x86_64: fix last_tsc calculation of PM timer

From: "Jan Beulich" <[email protected]>

The PM timer code updates vxtime.last_tsc, but this update was done
incorrectly in two ways:
- offset_delay being in microseconds requires multiplying with cpu_mhz
  rather than cpu_khz
- the multiplication of offset_delay and cpu_khz (both being 32-bit
  values) on most current CPUs would overflow (observed value of the
  delay was approximately 4000us, yielding an overflow for frequencies
  starting a little above 1GHz)

Signed-off-by: Jan Beulich <[email protected]>
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] i386: apic= command line option should always be

From: "Jan Beulich" <[email protected]>

When using apic= on the kernel command line, this had no effect for machines
matched by either the ACPI MADT or the MPS OEM table scan. However, when such
option is specified, it should also take effect for this set of systems.

Signed-off-by: Jan Beulich <[email protected]>
Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] x86_64: Fix no IOMMU warning in PCI-GART driver

Complaining about the IOMMU not compiled in doesn't make sense
here because it is clearly compiled in.

Signed-off-by: Andi Kleen <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] x86_64: Fix stack/mmap randomization for compat tasks

ia32_setup_arg_pages would ignore the passed in random stack top
and use its own static value.

Now it uses the 8bit of randomness native i386 would use too.

This indirectly fixes mmap randomization for 32bit processes too,
which depends on the stack randomization.

Should also give slightly better virtual cache colouring and
possibly better performance with HyperThreading.

Signed-off-by: Andi Kleen <[email protected]>
Acked-by: Ingo Molnar <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

[PATCH] cfq-iosched: fixup locking and ->queue_list list management

- Drop cic from the list when seen as dead.
- Fixup the locking, just use a simple spinlock.

Signed-off-by: Jens Axboe <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

Merge /pub/scm/linux/kernel/git/davej/agpgart

* master.kernel.org:/pub/scm/linux/kernel/git/davej/agpgart:
  [AGPGART] VIA PT880 Ultra support.
  [AGPGART] Fix Nforce3 suspend on amd64.
  [AGPGART] Enable SIS AGP driver on x86-64 for EM64T systems

Merge git://git./linux/kernel/git/sfrench/cifs-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/sfrench/cifs-2.6:
  [[CIFS] Pass truncate open flag through on file open in case setattr fails
  [CIFS] Fix typos in previous fix
  [CIFS] endian fix for new POSIX byte range lock support
  [CIFS] fix memory leak in cifs session info struct on reconnect
  [CIFS] ACPI suspend oops
  [CIFS] Do not limit the length of share names (was 100 for whole UNC name)
  [CIFS] Fix new POSIX Locking for setting lock_type correctly on unlock

Merge /pub/scm/linux/kernel/git/davem/net-2.6

* master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6:
  [NETFILTER]: PPTP helper: fix sstate/cstate typo
  [NETFILTER]: mark H.323 helper experimental
  [NETFILTER]: Fix small information leak in SO_ORIGINAL_DST (CVE-2006-1343)

[[CIFS] Pass truncate open flag through on file open in case setattr fails

on set size to zero.

Signed-off-by: Sebastian Voitzsch <sebastoam/[email protected]>
Signed-off-by: Steve French <[email protected]>

[CIFS] Fix typos in previous fix

Signed-off-by: Steve French <[email protected]>

[CIFS] endian fix for new POSIX byte range lock support

Signed-off-by: Alexey Dobriyan <[email protected]>
Signed-off-by: Steve French <[email protected]>

[CIFS] fix memory leak in cifs session info struct on reconnect

Signed-off-by: Steve French <[email protected]>

[CIFS] ACPI suspend oops

Wasn't able to reproduce a hard hang, but was able to get an oops if
suspended the machine during a copy to the cifs mount.  This led to some
things hanging, including a "sync".  Also got I/O errors when trying to
access the mount afterwards (even when didn't see the oops), and had
to unmount and remount in order to access the filesystem.

This patch fixed the oops.

Signed-off-by: Dave Kleikamp <[email protected]>
Signed-off-by: Steve French <[email protected]>

[CIFS] Do not limit the length of share names (was 100 for whole UNC name)
during mount. Especially important for some non-Western languages.

Signed-off-by: Steve French <[email protected]>

[CIFS] Fix new POSIX Locking for setting lock_type correctly on unlock

Signed-off-by: Steve French <[email protected]>

Input: psmouse - DMI updates for lifebook protocol

Added different lifebook-versions and the CF-18 to the corresponding
dmi-table.

Signed-off-by: Kenan Esau <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>

Input: change from numbered to named switches

Remove the numbered SW_* entries from the input system and assign names
to the existing users.

Signed-off-by: Richard Purdie <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>