openwrt/staging/blogic.git
16 years agomodule: don't ignore vermagic string if module doesn't have modversions
Rusty Russell [Fri, 9 May 2008 06:25:28 +0000 (16:25 +1000)]
module: don't ignore vermagic string if module doesn't have modversions

Linus found a logic bug: we ignore the version number in a module's
vermagic string if we have CONFIG_MODVERSIONS set, but modversions
also lets through a module with no __versions section for modprobe
--force (with tainting, but still).

We should only ignore the start of the vermagic string if the module
actually *has* crcs to check.  Rather than (say) having an
entertaining hissy fit and creating a config option to work around the
buggy code.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agomodule: be more picky about allowing missing module versions
Rusty Russell [Fri, 9 May 2008 06:24:21 +0000 (16:24 +1000)]
module: be more picky about allowing missing module versions

We allow missing __versions sections, because modprobe --force strips
it.  It makes less sense to allow sections where there's no version
for a specific symbol the module uses, so disallow that.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agomodule: put modversions in vermagic
Rusty Russell [Fri, 9 May 2008 06:23:17 +0000 (16:23 +1000)]
module: put modversions in vermagic

Don't allow a module built without versions altogether to be inserted
into a kernel which expects modversions.

modprobe --force will strip vermagic as well as modversions, so it
won't be effected, but this will make sure that a
non-CONFIG_MODVERSIONS module won't be accidentally inserted into a
CONFIG_MODVERSIONS kernel.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
Linus Torvalds [Fri, 9 May 2008 02:03:26 +0000 (19:03 -0700)]
Merge git://git./linux/kernel/git/davem/net-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6: (32 commits)
  net: Added ASSERT_RTNL() to dev_open() and dev_close().
  can: Fix can_send() handling on dev_queue_xmit() failures
  netns: Fix arbitrary net_device-s corruptions on net_ns stop.
  netfilter: Kconfig: default DCCP/SCTP conntrack support to the protocol config values
  netfilter: nf_conntrack_sip: restrict RTP expect flushing on error to last request
  macvlan: Fix memleak on device removal/crash on module removal
  net/ipv4: correct RFC 1122 section reference in comment
  tcp FRTO: SACK variant is errorneously used with NewReno
  e1000e: don't return half-read eeprom on error
  ucc_geth: Don't use RX clock as TX clock.
  cxgb3: Use CAP_SYS_RAWIO for firmware
  pcnet32: delete non NAPI code from driver.
  fs_enet: Fix a memory leak in fs_enet_mdio_probe
  [netdrvr] eexpress: IPv6 fails - multicast problems
  3c59x: use netstats in net_device structure
  3c980-TX needs EXTRA_PREAMBLE
  fix warning in drivers/net/appletalk/cops.c
  e1000e: Add support for BM PHYs on ICH9
  uli526x: fix endianness issues in the setup frame
  uli526x: initialize the hardware prior to requesting interrupts
  ...

16 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc-2.6
Linus Torvalds [Fri, 9 May 2008 02:03:19 +0000 (19:03 -0700)]
Merge git://git./linux/kernel/git/davem/sparc-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc-2.6:
  sparc: Fix SA_ONSTACK signal handling.

16 years agoRevert "PCI: remove default PCI expansion ROM memory allocation"
Linus Torvalds [Fri, 9 May 2008 01:41:48 +0000 (18:41 -0700)]
Revert "PCI: remove default PCI expansion ROM memory allocation"

This reverts commit 9f8daccaa05c14e5643bdd4faf5aed9cc8e6f11e, which was
reported to break X startup (xf86-video-ati-6.8.0). See

http://bugs.freedesktop.org/show_bug.cgi?id=15523

for details.

Reported-by: Laurence Withers <l@lwithers.me.uk>
Cc: Gary Hade <garyhade@us.ibm.com>
Cc: Greg KH <greg@kroah.com>
Cc: Jan Beulich <jbeulich@novell.com>
Cc: "Jun'ichi Nomura" <j-nomura@ce.jp.nec.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mingo/linux...
Linus Torvalds [Thu, 8 May 2008 18:31:07 +0000 (11:31 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/mingo/linux-2.6-sched-fixes

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mingo/linux-2.6-sched-fixes:
  sched: fix weight calculations
  semaphore: fix

16 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound-2.6
Linus Torvalds [Thu, 8 May 2008 17:58:45 +0000 (10:58 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/tiwai/sound-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound-2.6:
  [ALSA] soc at91 minor bug fixes
  [ALSA] soc - at91-pcm - Fix line wrapping
  pcspkr: fix dependancies

16 years agoRemove duplicated include in net/sunrpc/svc.c
Huang Weiyi [Thu, 8 May 2008 14:48:31 +0000 (22:48 +0800)]
Remove duplicated include in net/sunrpc/svc.c

<linux/sched.h> we included twice.

Signed-off-by: Huang Weiyi <weiyi.huang@gmail.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agofs/proc/task_mmu.c: remove duplicated include files
Huang Weiyi [Thu, 8 May 2008 14:36:27 +0000 (22:36 +0800)]
fs/proc/task_mmu.c: remove duplicated include files

Removed duplicated include files <linux/ptrace.h> and <linux/seq_file.h> in
fs/proc/task_mmu.c.

Signed-off-by: Huang Weiyi <weiyi.huang@gmail.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoFix drivers/media build for modular builds
Ingo Molnar [Wed, 30 Apr 2008 07:48:07 +0000 (09:48 +0200)]
Fix drivers/media build for modular builds

Fix allmodconfig build bug introduced in latest -git by commit
7c91f0624a9 ("V4L/DVB(7767): Move tuners to common/tuners"):

  LD      kernel/built-in.o
  LD      drivers/built-in.o
  ld: drivers/media/built-in.o: No such file: No such file or directory

which happens if all media drivers are modular:

  http://redhat.com/~mingo/misc/config-Wed_Apr_30_09_24_48_CEST_2008.bad

In that case there's no obj-y rule connecting all the built-in.o files and
the link tree breaks.

The fix is to add a guaranteed obj-y rule for the core vmlinux to build.
(which results in an empty object file if all media drivers are modular)

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Acked-by: Sam Ravnborg <sam@ravnborg.org>
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland...
Linus Torvalds [Thu, 8 May 2008 17:50:34 +0000 (10:50 -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/ehca: Wait for async events to finish before destroying QP
  IB/ipath: Fix SDMA error recovery in absence of link status change
  IB/ipath: Need to always request and handle PIO avail interrupts
  IB/ipath: Fix count of packets received by kernel
  IB/ipath: Return the correct opcode for RDMA WRITE with immediate
  IB/ipath: Fix bug that can leave sends disabled after freeze recovery
  IB/ipath: Only increment SSN if WQE is put on send queue
  IB/ipath: Only warn about prototype chip during init
  RDMA/cxgb3: Fix severe limit on userspace memory registration size
  RDMA/cxgb3: Don't add PBL memory to gen_pool in chunks

16 years agoMN10300: Make cpu_relax() invoke barrier()
David Howells [Wed, 7 May 2008 14:31:54 +0000 (15:31 +0100)]
MN10300: Make cpu_relax() invoke barrier()

Make cpu_relax() invoke barrier() to be the same as other arches.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoMerge branch 'for-linus' of git://git.kernel.dk/linux-2.6-block
Linus Torvalds [Thu, 8 May 2008 17:48:36 +0000 (10:48 -0700)]
Merge branch 'for-linus' of git://git.kernel.dk/linux-2.6-block

* 'for-linus' of git://git.kernel.dk/linux-2.6-block:
  Revert "relay: fix splice problem"
  docbook: fix bio missing parameter
  block: use unitialized_var() in bio_alloc_bioset()
  block: avoid duplicate calls to get_part() in disk stat code
  cfq-iosched: make io priorities inherit CPU scheduling class as well as nice
  block: optimize generic_unplug_device()
  block: get rid of likely/unlikely predictions in merge logic
  vfs: splice remove_suid() cleanup
  cfq-iosched: fix RCU race in the cfq io_context destructor handling
  block: adjust tagging function queue bit locking
  block: sysfs store function needs to grab queue_lock and use queue_flag_*()

16 years agoMerge branch 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jack/linux...
Linus Torvalds [Thu, 8 May 2008 17:48:03 +0000 (10:48 -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: Fix memory corruption when fs mounted with noadinicb option
  udf: Make udf exportable
  udf: fs/udf/partition.c:udf_get_pblock() mustn't be inline

16 years agoMerge branch 'for-linus' of git://git390.osdl.marist.edu/pub/scm/linux-2.6
Linus Torvalds [Thu, 8 May 2008 17:47:39 +0000 (10:47 -0700)]
Merge branch 'for-linus' of git://git390.osdl.marist.edu/linux-2.6

* 'for-linus' of git://git390.osdl.marist.edu/pub/scm/linux-2.6:
  [S390] guest page hinting light
  [S390] tty3270: fix put_char fail/success conversion.
  [S390] compat ptrace cleanup
  [S390] s390mach compile warning
  [S390] cio: Fix parsing mechanism for blacklisted devices.
  [S390] cio: Remove cio_msg kernel parameter.
  [S390] s390-kvm: leave sie context on work. Removes preemption requirement
  [S390] s390: Optimize user and work TIF check

16 years agodrivers/scsi/dpt_i2o.c: fix build on alpha
Andrew Morton [Wed, 7 May 2008 03:42:42 +0000 (20:42 -0700)]
drivers/scsi/dpt_i2o.c: fix build on alpha

alpha:

drivers/scsi/dpt_i2o.c:1997: error: implicit declaration of function 'adpt_alpha_info'
drivers/scsi/dpt_i2o.c: At top level:
drivers/scsi/dpt_i2o.c:2032: warning: conflicting types for 'adpt_alpha_info'
drivers/scsi/dpt_i2o.c:2032: error: static declaration of 'adpt_alpha_info' follows non-static declaration
drivers/scsi/dpt_i2o.c:1997: error: previous implicit declaration of 'adpt_alpha_info' was here

Due to a copy-n-paste error in drivers/scsi/dpti.h.

Fix that up and remove some of the many daft static-declarations-in-a-header
which this driver enjoys.

Cc: Miquel van Smoorenburg <miquels@cistron.nl>
Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoFix cpuset sched_relax_domain_level control file
Paul Menage [Wed, 7 May 2008 03:42:41 +0000 (20:42 -0700)]
Fix cpuset sched_relax_domain_level control file

Due to a merge conflict, the sched_relax_domain_level control file was marked
as being handled by cpuset_read/write_u64, but the code to handle it was
actually in cpuset_common_file_read/write.

Since the value being written/read is in fact a signed integer, it should be
treated as such; this patch adds cpuset_read/write_s64 functions, and uses
them to handle the sched_relax_domain_level file.

With this patch, the sched_relax_domain_level can be read and written, and the
correct contents seen/updated.

Signed-off-by: Paul Menage <menage@google.com>
Cc: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Cc: Paul Jackson <pj@sgi.com>
Cc: Ingo Molnar <mingo@elte.hu>
Reviewed-by: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoslub: fix atomic usage in any_slab_objects()
Benjamin Herrenschmidt [Wed, 7 May 2008 03:42:39 +0000 (20:42 -0700)]
slub: fix atomic usage in any_slab_objects()

any_slab_objects() does an atomic_read on an atomic_long_t, this
fixes it to use atomic_long_read instead.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agosys_pipe(): fix file descriptor leaks
Ulrich Drepper [Wed, 7 May 2008 03:42:38 +0000 (20:42 -0700)]
sys_pipe(): fix file descriptor leaks

Remember to close the files if copy_to_user() failed.

Spotted by dm.n9107@gmail.com.

Signed-off-by: Ulrich Drepper <drepper@redhat.com>
Cc: DM <dm.n9107@gmail.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agovt: fix canonical input in UTF-8 mode
Samuel Thibault [Wed, 7 May 2008 03:42:37 +0000 (20:42 -0700)]
vt: fix canonical input in UTF-8 mode

For e.g.  proper TTY canonical support, IUTF8 termios flag has to be set as
appropriate.  Linux used to not care about setting that flag for VT TTYs.

This patch fixes that by activating it according to the current mode of the
VT, and sets the default value according to the vt.default_utf8 parameter.

Signed-off-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Cc: Willy Tarreau <w@1wt.eu>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agousb/asix: add Buffalo LUA-U2-GT 10/100/1000
Mattia Dongili [Wed, 7 May 2008 03:42:35 +0000 (20:42 -0700)]
usb/asix: add Buffalo LUA-U2-GT 10/100/1000

The USB net adapter Buffalo LUA-U2-GT (0411:006e) carries a AX88178 chip.
Tested on the above HW.

Signed-off-by: Mattia Dongili <malattia@linux.it>
Acked-off-by: David Hollis <dhollis@davehollis.com>
Cc: Greg KH <greg@kroah.com>
Acked-by: Jeff Garzik <jeff@garzik.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agosx.c: fix printk warnings on sparc32
Andrew Morton [Wed, 7 May 2008 03:42:35 +0000 (20:42 -0700)]
sx.c: fix printk warnings on sparc32

drivers/char/sx.c: In function 'sx_set_real_termios':
drivers/char/sx.c:973: warning: format '%u' expects type 'unsigned int', but argument 2 has type 'long unsigned int'
drivers/char/sx.c:999: warning: format '%x' expects type 'unsigned int', but argument 2 has type 'tcflag_t'
drivers/char/sx.c:1012: warning: format '%x' expects type 'unsigned int', but argument 2 has type 'tcflag_t'

sparc32 seems to use weird types for its tty things.

[ Fine by me but this is ancient debug and most of the debug in sx just
  wants deleting eventually.  - Alan ]

Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Alan Cox <alan@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agouml: fix inconsistence due to tty_operation change
WANG Cong [Wed, 7 May 2008 03:42:33 +0000 (20:42 -0700)]
uml: fix inconsistence due to tty_operation change

'put_char' of 'struct tty_operations' has changed from 'void' into 'int'.
This can also shut up compiler warnings.

Cc: Jeff Dike <jdike@addtoit.com>
Signed-off-by: WANG Cong <wangcong@zeuux.org>
Acked-by: Alan Cox <alan@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agomisc: fix integer as NULL pointer warnings
Harvey Harrison [Wed, 7 May 2008 03:42:32 +0000 (20:42 -0700)]
misc: fix integer as NULL pointer warnings

drivers/md/raid10.c:889:17: warning: Using plain integer as NULL pointer
drivers/media/video/cx18/cx18-driver.c:616:12: warning: Using plain integer as NULL pointer
sound/oss/kahlua.c:70:12: warning: Using plain integer as NULL pointer

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Mauro Carvalho Chehab <mchehab@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agofix irq flags for iuu_phoenix.c
Steven Rostedt [Wed, 7 May 2008 03:42:31 +0000 (20:42 -0700)]
fix irq flags for iuu_phoenix.c

The file drivers/usb/serial/iuu_phoenix.c uses "int" for flags.  This can
cause hard to find bugs on some architectures.  This patch converts the flags
to use "long" instead.

This bug was discovered by doing an allyesconfig make on the -rt kernel where
checks are done to ensure all flags are of size sizeof(long).

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agofix irq flags in rtc-ds1511
Steven Rostedt [Wed, 7 May 2008 03:42:30 +0000 (20:42 -0700)]
fix irq flags in rtc-ds1511

The file in drivers/rtc/rtc-ds1551.c uses "int" for flags.  This can cause
hard to find bugs on some architectures.  This patch converts the flags to use
"long" instead.

This bug was discovered by doing an allyesconfig make on the -rt kernel where
checks are done to ensure all flags are of size sizeof(long).

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Cc: Alessandro Zummo <a.zummo@towertech.it>
Cc: David Brownell <david-b@pacbell.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agofix irq flags in saa7134
Steven Rostedt [Wed, 7 May 2008 03:42:29 +0000 (20:42 -0700)]
fix irq flags in saa7134

Some files in the drivers/media/video/saa7134 directory uses "int" for flags.
This can cause hard to find bugs on some architectures.  This patch converts
the flags to use "long" instead.

This bug was discovered by doing an allyesconfig make on the -rt kernel where
checks are done to ensure all flags are of size sizeof(long).

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Acked-by: Mauro Carvalho Chehab <mchehab@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agofix irq flags in mac80211 code
Steven Rostedt [Wed, 7 May 2008 03:42:28 +0000 (20:42 -0700)]
fix irq flags in mac80211 code

A file in the net/mac80211 directory uses "int" for flags.  This can cause
hard to find bugs on some architectures.  This patch converts the flags to use
"long" instead.

This bug was discovered by doing an allyesconfig make on the -rt kernel where
checks are done to ensure all flags are of size sizeof(long).

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Cc: "John W. Linville" <linville@tuxdriver.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoserial: access after NULL check in uart_flush_buffer()
Tetsuo Handa [Wed, 7 May 2008 03:42:27 +0000 (20:42 -0700)]
serial: access after NULL check in uart_flush_buffer()

I noticed that

  static void uart_flush_buffer(struct tty_struct *tty)
  {
   struct uart_state *state = tty->driver_data;
   struct uart_port *port = state->port;
   unsigned long flags;

   /*
    * This means you called this function _after_ the port was
    * closed.  No cookie for you.
    */
   if (!state || !state->info) {
   WARN_ON(1);
   return;
   }

is too late for checking state != NULL.

Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoKconfig: improved help for CONFIG_ACCESSIBILITY
Samuel Thibault [Wed, 7 May 2008 03:42:26 +0000 (20:42 -0700)]
Kconfig: improved help for CONFIG_ACCESSIBILITY

Add a small explanation of what accessibility is.

Signed-off-by: Samuel Thibault <samuel.thibault@ens-lyon.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agosched: fix weight calculations
Mike Galbraith [Thu, 8 May 2008 15:00:42 +0000 (17:00 +0200)]
sched: fix weight calculations

The conversion between virtual and real time is as follows:

  dvt = rw/w * dt <=> dt = w/rw * dvt

Since we want the fair sleeper granularity to be in real time, we actually
need to do:

  dvt = - rw/w * l

This bug could be related to the regression reported by Yanmin Zhang:

| Comparing with kernel 2.6.25, sysbench+mysql(oltp, readonly) has lots
| of regressions with 2.6.26-rc1:
|
| 1) 8-core stoakley: 28%;
| 2) 16-core tigerton: 20%;
| 3) Itanium Montvale: 50%.

Reported-by: "Zhang, Yanmin" <yanmin_zhang@linux.intel.com>
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agosemaphore: fix
Ingo Molnar [Thu, 8 May 2008 09:53:48 +0000 (11:53 +0200)]
semaphore: fix

Yanmin Zhang reported:

| Comparing with kernel 2.6.25, ï»¿AIM7 (use tmpfs) has ï»¿more th
| regression under 2.6.26-rc1 on my 8-core stoakley, 16-core tigerton,
| and Itanium Montecito. Bisect located the patch below:
|
64ac24e738823161693bf791f87adc802cf529ff is first bad commit
| commit 64ac24e738823161693bf791f87adc802cf529ff
| Author: Matthew Wilcox <matthew@wil.cx>
| Date:   Fri Mar 7 21:55:58 2008 -0500
|
|     Generic semaphore implementation
|
| After I manually reverted the patch against 2.6.26-rc1 while fixing
| lots of conflicts/errors, aim7 regression became less than 2%.

i reproduced the AIM7 workload and can confirm Yanmin's findings that
-.26-rc1 regresses over .25 - by over 67% here.

Looking at the workload i found and fixed what i believe to be the real
bug causing the AIM7 regression: it was inefficient wakeup / scheduling
/ locking behavior of the new generic semaphore code, causing suboptimal
performance.

The problem comes from the following code. The new semaphore code does
this on down():

        spin_lock_irqsave(&sem->lock, flags);
        if (likely(sem->count > 0))
                sem->count--;
        else
                __down(sem);
        spin_unlock_irqrestore(&sem->lock, flags);

and this on up():

        spin_lock_irqsave(&sem->lock, flags);
        if (likely(list_empty(&sem->wait_list)))
                sem->count++;
        else
                __up(sem);
        spin_unlock_irqrestore(&sem->lock, flags);

where __up() does:

        list_del(&waiter->list);
        waiter->up = 1;
        wake_up_process(waiter->task);

and where __down() does this in essence:

        list_add_tail(&waiter.list, &sem->wait_list);
        waiter.task = task;
        waiter.up = 0;
        for (;;) {
                [...]
                spin_unlock_irq(&sem->lock);
                timeout = schedule_timeout(timeout);
                spin_lock_irq(&sem->lock);
                if (waiter.up)
                        return 0;
        }

the fastpath looks good and obvious, but note the following property of
the contended path: if there's a task on the ->wait_list, the up() of
the current owner will "pass over" ownership to that waiting task, in a
wake-one manner, via the waiter->up flag and by removing the waiter from
the wait list.

That is all and fine in principle, but as implemented in
kernel/semaphore.c it also creates a nasty, hidden source of contention!

The contention comes from the following property of the new semaphore
code: the new owner owns the semaphore exclusively, even if it is not
running yet.

So if the old owner, even if just a few instructions later, does a
down() [lock_kernel()] again, it will be blocked and will have to wait
on the new owner to eventually be scheduled (possibly on another CPU)!
Or if another task gets to lock_kernel() sooner than the "new owner"
scheduled, it will be blocked unnecessarily and for a very long time
when there are 2000 tasks running.

I.e. the implementation of the new semaphores code does wake-one and
lock ownership in a very restrictive way - it does not allow
opportunistic re-locking of the lock at all and keeps the scheduler from
picking task order intelligently.

This kind of scheduling, with 2000 AIM7 processes running, creates awful
cross-scheduling between those 2000 tasks, causes reduced parallelism, a
throttled runqueue length and a lot of idle time. With increasing number
of CPUs it causes an exponentially worse behavior in AIM7, as the chance
for a newly woken new-owner task to actually run anytime soon is less
and less likely.

Note that it takes just a tiny bit of contention for the 'new-semaphore
catastrophy' to happen: the wakeup latencies get added to whatever small
contention there is, and quickly snowball out of control!

I believe Yanmin's findings and numbers support this analysis too.

The best fix for this problem is to use the same scheduling logic that
the kernel/mutex.c code uses: keep the wake-one behavior (that is OK and
wanted because we do not want to over-schedule), but also allow
opportunistic locking of the lock even if a wakee is already "in
flight".

The patch below implements this new logic. With this patch applied the
AIM7 regression is largely fixed on my quad testbox:

  # v2.6.25 vanilla:
  ..................
  Tasks   Jobs/Min        JTI     Real    CPU     Jobs/sec/task
  2000    56096.4         91      207.5   789.7   0.4675
  2000    55894.4         94      208.2   792.7   0.4658

  # v2.6.26-rc1-166-gc0a1811 vanilla:
  ...................................
  Tasks   Jobs/Min        JTI     Real    CPU     Jobs/sec/task
  2000    33230.6         83      350.3   784.5   0.2769
  2000    31778.1         86      366.3   783.6   0.2648

  # v2.6.26-rc1-166-gc0a1811 + semaphore-speedup:
  ...............................................
  Tasks   Jobs/Min        JTI     Real    CPU     Jobs/sec/task
  2000    55707.1         92      209.0   795.6   0.4642
  2000    55704.4         96      209.0   796.0   0.4642

i.e. a 67% speedup. We are now back to within 1% of the v2.6.25
performance levels and have zero idle time during the test, as expected.

Btw., interactivity also improved dramatically with the fix - for
example console-switching became almost instantaneous during this
workload (which after all is running 2000 tasks at once!), without the
patch it was stuck for a minute at times.

There's another nice side-effect of this speedup patch, the new generic
semaphore code got even smaller:

   text    data     bss     dec     hex filename
   1241       0       0    1241     4d9 semaphore.o.before
   1207       0       0    1207     4b7 semaphore.o.after

(because the waiter.up complication got removed.)

Longer-term we should look into using the mutex code for the generic
semaphore code as well - but i's not easy due to legacies and it's
outside of the scope of v2.6.26 and outside the scope of this patch as
well.

Bisected-by: "Zhang, Yanmin" <yanmin_zhang@linux.intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoRevert "relay: fix splice problem"
Jens Axboe [Thu, 8 May 2008 12:06:19 +0000 (14:06 +0200)]
Revert "relay: fix splice problem"

This reverts commit c3270e577c18b3d0e984c3371493205a4807db9d.

16 years ago[ALSA] soc at91 minor bug fixes
Patrik Sevallius [Thu, 8 May 2008 12:04:08 +0000 (14:04 +0200)]
[ALSA] soc at91 minor bug fixes

Found these two bugs while browsing through the code.  The first one is
a cut-n-paste bug, instead of disabling the clock when request_irq()
fails, it enabled it once more.  The second one fixes a debug printout,
AT91_SSC_IER is write only, AT91_SSC_IMR is readable (the printed string
actually says imr).

Frank Mandarino was busy so he asked me to send these to this list.

/Patrik

Signed-off-by: Patrik Sevallius <patrik.sevallius@enea.com>
Acked-by: Frank Mandarino <fmandarino@endrelia.com>
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
16 years ago[ALSA] soc - at91-pcm - Fix line wrapping
Mark Brown [Thu, 8 May 2008 12:03:30 +0000 (14:03 +0200)]
[ALSA] soc - at91-pcm - Fix line wrapping

There's more checkpatch stuff to fix in the driver, this just fixes the
minimum required for the following patch to be clean.

Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
16 years agonet: Added ASSERT_RTNL() to dev_open() and dev_close().
Ben Hutchings [Thu, 8 May 2008 09:53:17 +0000 (02:53 -0700)]
net: Added ASSERT_RTNL() to dev_open() and dev_close().

dev_open() and dev_close() must be called holding the RTNL, since they
call device functions and netdevice notifiers that are promised the RTNL.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years agocan: Fix can_send() handling on dev_queue_xmit() failures
Oliver Hartkopp [Thu, 8 May 2008 09:49:55 +0000 (02:49 -0700)]
can: Fix can_send() handling on dev_queue_xmit() failures

The tx packet counting and the local loopback of CAN frames should
only happen in the case that the CAN frame has been enqueued to the
netdevice tx queue successfully.

Thanks to Andre Naujoks <nautsch@gmail.com> for reporting this issue.

Signed-off-by: Oliver Hartkopp <oliver@hartkopp.net>
Signed-off-by: Urs Thuermann <urs@isnogud.escape.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years agoMerge branch 'upstream-davem' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik...
David S. Miller [Thu, 8 May 2008 09:35:54 +0000 (02:35 -0700)]
Merge branch 'upstream-davem' of /linux/kernel/git/jgarzik/netdev-2.6

16 years agonetns: Fix arbitrary net_device-s corruptions on net_ns stop.
Pavel Emelyanov [Thu, 8 May 2008 08:24:25 +0000 (01:24 -0700)]
netns: Fix arbitrary net_device-s corruptions on net_ns stop.

When a net namespace is destroyed, some devices (those, not killed
on ns stop explicitly) are moved back to init_net.

The problem, is that this net_ns change has one point of failure -
the __dev_alloc_name() may be called if a name collision occurs (and
this is easy to trigger). This allocator performs a likely-to-fail
GFP_ATOMIC allocation to find a suitable number. Other possible
conditions that may cause error (for device being ns local or not
registered) are always false in this case.

So, when this call fails, the device is unregistered. But this is
*not* the right thing to do, since after this the device may be
released (and kfree-ed) improperly. E. g. bridges require more
actions (sysfs update, timer disarming, etc.), some other devices
want to remove their private areas from lists, etc.

I. e. arbitrary use-after-free cases may occur.

The proposed fix is the following: since the only reason for the
dev_change_net_namespace to fail is the name generation, we may
give it a unique fall-back name w/o %d-s in it - the dev<ifindex>
one, since ifindexes are still unique.

So make this change, raise the failure-case printk loglevel to
EMERG and replace the unregister_netdevice call with BUG().

[ Use snprintf() -DaveM ]

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years agonetfilter: Kconfig: default DCCP/SCTP conntrack support to the protocol config values
Patrick McHardy [Thu, 8 May 2008 08:16:04 +0000 (01:16 -0700)]
netfilter: Kconfig: default DCCP/SCTP conntrack support to the protocol config values

When conntrack and DCCP/SCTP protocols are enabled, chances are good
that people also want DCCP/SCTP conntrack and NAT support.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years agonetfilter: nf_conntrack_sip: restrict RTP expect flushing on error to last request
Patrick McHardy [Thu, 8 May 2008 08:15:21 +0000 (01:15 -0700)]
netfilter: nf_conntrack_sip: restrict RTP expect flushing on error to last request

Some Inovaphone PBXs exhibit very stange behaviour: when dialing for
example "123", the device sends INVITE requests for "1", "12" and
"123" back to back.  The first requests will elicit error responses
from the receiver, causing the SIP helper to flush the RTP
expectations even though we might still see a positive response.

Note the sequence number of the last INVITE request that contained a
media description and only flush the expectations when receiving a
negative response for that sequence number.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years agomacvlan: Fix memleak on device removal/crash on module removal
Patrick McHardy [Thu, 8 May 2008 08:13:31 +0000 (01:13 -0700)]
macvlan: Fix memleak on device removal/crash on module removal

As noticed by Ben Greear, macvlan crashes the kernel when unloading the
module. The reason is that it tries to clean up the macvlan_port pointer
on the macvlan device itself instead of the underlying device. A non-NULL
pointer is taken as indication that the macvlan_handle_frame_hook is
valid, when receiving the next packet on the underlying device it tries
to call the NULL hook and crashes.

Clean up the macvlan_port on the correct device to fix this.

Signed-off-by; Patrick McHardy <kaber@trash.net>
Tested-by: Ben Greear <greearb@candelatech.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years agonet/ipv4: correct RFC 1122 section reference in comment
J.H.M. Dassen (Ray) [Thu, 8 May 2008 08:11:04 +0000 (01:11 -0700)]
net/ipv4: correct RFC 1122 section reference in comment

RFC 1122 does not have a section 3.1.2.2. The requirement to silently
discard datagrams with a bad checksum is in section 3.2.1.2 instead.

Addresses http://bugzilla.kernel.org/show_bug.cgi?id=10611

Signed-off-by: J.H.M. Dassen (Ray) <jdassen@debian.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years agotcp FRTO: SACK variant is errorneously used with NewReno
Ilpo Järvinen [Thu, 8 May 2008 08:09:11 +0000 (01:09 -0700)]
tcp FRTO: SACK variant is errorneously used with NewReno

Note: there's actually another bug in FRTO's SACK variant, which
is the causing failure in NewReno case because of the error
that's fixed here. I'll fix the SACK case separately (it's
a separate bug really, though related, but in order to fix that
I need to audit tp->snd_nxt usage a bit).

There were two places where SACK variant of FRTO is getting
incorrectly used even if SACK wasn't negotiated by the TCP flow.
This leads to incorrect setting of frto_highmark with NewReno
if a previous recovery was interrupted by another RTO.

An eventual fallback to conventional recovery then incorrectly
considers one or couple of segments as forward transmissions
though they weren't, which then are not LOST marked during
fallback making them "non-retransmittable" until the next RTO.
In a bad case, those segments are really lost and are the only
one left in the window. Thus TCP needs another RTO to continue.
The next FRTO, however, could again repeat the same events
making the progress of the TCP flow extremely slow.

In order for these events to occur at all, FRTO must occur
again in FRTOs step 3 while the key segments must be lost as
well, which is not too likely in practice. It seems to most
frequently with some small devices such as network printers
that *seem* to accept TCP segments only in-order. In cases
were key segments weren't lost, things get automatically
resolved because those wrongly marked segments don't need to be
retransmitted in order to continue.

I found a reproducer after digging up relevant reports (few
reports in total, none at netdev or lkml I know of), some
cases seemed to indicate middlebox issues which seems now
to be a false assumption some people had made. Bugzilla
#10063 _might_ be related. Damon L. Chesser <damon@damtek.com>
had a reproducable case and was kind enough to tcpdump it
for me. With the tcpdump log it was quite trivial to figure
out.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years agosparc: Fix SA_ONSTACK signal handling.
David S. Miller [Thu, 8 May 2008 01:54:05 +0000 (18:54 -0700)]
sparc: Fix SA_ONSTACK signal handling.

We need to be more liberal about the alignment of the buffer given to
us by sigaltstack().  The user should not need to be mindful of all of
the alignment constraints we have for the stack frame.

This mirrors how we handle this situation in clone() as well.

Also, we align the stack even in non-SA_ONSTACK cases so that signals
due to bad stack alignment can be delivered properly.  This makes such
errors easier to debug and recover from.

Finally, add the sanity check x86 has to make sure we won't overflow
the signal stack.

This fixes glibc testcases nptl/tst-cancel20.c and
nptl/tst-cancelx20.c

Signed-off-by: David S. Miller <davem@davemloft.net>
16 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc-2.6
Linus Torvalds [Thu, 8 May 2008 00:04:49 +0000 (17:04 -0700)]
Merge git://git./linux/kernel/git/davem/sparc-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc-2.6:
  sparc: Fix fork/clone/vfork system call restart.
  sparc: Fix mmap VA span checking.

16 years agosparc: Fix fork/clone/vfork system call restart.
David S. Miller [Wed, 7 May 2008 23:21:28 +0000 (16:21 -0700)]
sparc: Fix fork/clone/vfork system call restart.

We clobber %i1 as well as %i0 for these system calls,
because they give two return values.

Therefore, on error, we have to restore %i1 properly
or else the restart explodes since it uses the wrong
arguments.

This fixes glibc's nptl/tst-eintr1.c testcase.

Signed-off-by: David S. Miller <davem@davemloft.net>
16 years ago[MAINTAINERS] New maintainer for Intel ethernet adapters
Auke Kok [Wed, 7 May 2008 20:42:33 +0000 (13:42 -0700)]
[MAINTAINERS] New maintainer for Intel ethernet adapters

I'm handing over maintainership to Jeff Kirsher and moving on
to other Linux/Open Source work within Intel. Good luck to Jeff ;)

Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoIB/ehca: Wait for async events to finish before destroying QP
Stefan Roscher [Wed, 7 May 2008 18:35:06 +0000 (11:35 -0700)]
IB/ehca: Wait for async events to finish before destroying QP

This is necessary because, in a multicore environment, a race between
uverbs async handler and destroy QP could occur.

Signed-off-by: Stefan Roscher <stefan.roscher at de.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ipath: Fix SDMA error recovery in absence of link status change
John Gregor [Wed, 7 May 2008 18:01:10 +0000 (11:01 -0700)]
IB/ipath: Fix SDMA error recovery in absence of link status change

What's fixed:

    in ipath_cancel_sends()

        We need to unconditionally set ABORTING.  So, swap the tests
        so the set_bit() isn't shadowed by the &&.

        If we've disarmed the piobufs, then we need to unconditionally
        set DISARMED.  So, move it out from the overly protective if
        at the bottom.

    in sdma_abort_task()

        Abort_task was written knowing that the SDMA engine would always
        be reset (and restarted) on error.  A recent change broke that
        fundamental assumption by taking the restart portion and making
        it conditional on a link status change.  But, SDMA can go boom
        without a link status change in some conditions.

Signed-off-by: John Gregor <john.gregor@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ipath: Need to always request and handle PIO avail interrupts
Dave Olson [Wed, 7 May 2008 18:00:15 +0000 (11:00 -0700)]
IB/ipath: Need to always request and handle PIO avail interrupts

Now that we always use PIO for vl15 on 7220, we could get stuck forever
if we happened to run out of PIO buffers from the verbs code, because
the setup code wouldn't run; the interrupt was also ignored if SDMA was
supported.  We also have to reduce the pio update threshold if we have
fewer kernel buffers than the existing threshold.

Clean up the initialization a bit to get ordering safer and more
sensible, and use the existing ipath_chg_kernavail call to do init,
rather than doing it separately.

Drop unnecessary clearing of pio buffer on pio parity error.

Drop incorrect updating of pioavailshadow when exitting freeze mode
(software state may not match chip state if buffer has been allocated
and not yet written).

If we couldn't get a kernel buffer for a while, make sure we are
in sync with hardware, mainly to handle the exitting freeze case.

Signed-off-by: Dave Olson <dave.olson@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ipath: Fix count of packets received by kernel
Michael Albaugh [Wed, 7 May 2008 17:59:23 +0000 (10:59 -0700)]
IB/ipath: Fix count of packets received by kernel

The loop in ipath_kreceive() that processes packets increments the
loop-index 'i' once too often, because the exit condition does not
depend on it, and is checked after the increment. By adding a check for
!last to the iterator in the for loop, we correct that in a way that is
not so likely to be re-broken by changes in the loop body.

Signed-off-by: Michael Albaugh <micheal.albaugh@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ipath: Return the correct opcode for RDMA WRITE with immediate
Ralph Campbell [Wed, 7 May 2008 17:58:50 +0000 (10:58 -0700)]
IB/ipath: Return the correct opcode for RDMA WRITE with immediate

This patch fixes a bug in the RC responder which generates a completion
entry with the wrong opcode when an RDMA WRITE with immediate is received.

Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ipath: Fix bug that can leave sends disabled after freeze recovery
Dave Olson [Wed, 7 May 2008 17:57:48 +0000 (10:57 -0700)]
IB/ipath: Fix bug that can leave sends disabled after freeze recovery

The semantics of cancel_sends changed, but the code using it was missed.
Don't leave sends and pioavail updates disabled, and add a comment as to
why the force update is needed.

Signed-off-by: Dave Olson <dave.olson@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ipath: Only increment SSN if WQE is put on send queue
Ralph Campbell [Wed, 7 May 2008 17:57:14 +0000 (10:57 -0700)]
IB/ipath: Only increment SSN if WQE is put on send queue

If a send work request has immediate errors and is not put on the
send queue, we shouldn't update any of the QP state.

The increment of the SSN wasn't obeying this.

Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/ipath: Only warn about prototype chip during init
Michael Albaugh [Wed, 7 May 2008 17:56:47 +0000 (10:56 -0700)]
IB/ipath: Only warn about prototype chip during init

We warn about prototype chips, but the function that checks for
support is also called as a result of a get_portinfo request, which
can clutter the logs.

Restrict warning to only appear during initialization.

Signed-off-by: Michael Albaugh <michael.albaugh@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agodocbook: fix bio missing parameter
Randy Dunlap [Wed, 30 Apr 2008 07:08:54 +0000 (09:08 +0200)]
docbook: fix bio missing parameter

Fix fs/bio.c kernel-doc parameter warning:
Warning(linux-2.6.25-git14//fs/bio.c:972): No description found for parameter 'reading'

Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: use unitialized_var() in bio_alloc_bioset()
Jens Axboe [Wed, 7 May 2008 11:26:27 +0000 (13:26 +0200)]
block: use unitialized_var() in bio_alloc_bioset()

Better than setting idx to some random value and it silences the
same bogus gcc warning.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agopcspkr: fix dependancies
Stas Sergeev [Wed, 7 May 2008 10:39:56 +0000 (12:39 +0200)]
pcspkr: fix dependancies

fix pcspkr dependancies: make the pcspkr platform
drivers to depend on a platform device, and
not the other way around.

Signed-off-by: Stas Sergeev <stsp@aknet.ru>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Dmitry Torokhov <dtor@mail.ru>
CC: Vojtech Pavlik <vojtech@suse.cz>
CC: Michael Opdenacker <michael-lists@free-electrons.com>
[fixed for 2.6.26-rc1 by tiwai]
Signed-off-by: Takashi Iwai <tiwai@suse.de>
16 years agosparc: Fix mmap VA span checking.
David S. Miller [Wed, 7 May 2008 09:24:28 +0000 (02:24 -0700)]
sparc: Fix mmap VA span checking.

We should not conditionalize VA range checks on MAP_FIXED.

Signed-off-by: David S. Miller <davem@davemloft.net>
16 years agoblock: avoid duplicate calls to get_part() in disk stat code
Jens Axboe [Wed, 7 May 2008 08:15:46 +0000 (10:15 +0200)]
block: avoid duplicate calls to get_part() in disk stat code

get_part() is fairly expensive, as it O(N) loops over partitions
to find the right one. In lots of normal IO paths we end up looking
up the partition twice, to make matters even worse. Change the
stat add code to accept a passed in partition instead.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agocfq-iosched: make io priorities inherit CPU scheduling class as well as nice
Jens Axboe [Wed, 7 May 2008 07:51:23 +0000 (09:51 +0200)]
cfq-iosched: make io priorities inherit CPU scheduling class as well as nice

We currently set all processes to the best-effort scheduling class,
regardless of what CPU scheduling class they belong to. Improve that
so that we correctly track idle and rt scheduling classes as well.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoudf: Fix memory corruption when fs mounted with noadinicb option
Jan Kara [Tue, 6 May 2008 16:26:17 +0000 (18:26 +0200)]
udf: Fix memory corruption when fs mounted with noadinicb option

When UDF filesystem is mounted with noadinicb mount option, it
happens that we extend an empty directory with a block. A code in
udf_add_entry() didn't count with this possibility and used
uninitialized data leading to memory and filesystem corruption.
Add a check whether file already has some extents before operating
on them.

Signed-off-by: Jan Kara <jack@suse.cz>
16 years agoudf: Make udf exportable
Rasmus Rohde [Wed, 30 Apr 2008 15:22:06 +0000 (17:22 +0200)]
udf: Make udf exportable

Cc: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Rasmus Rohde <rohde@duff.dk>
Signed-off-by: Jan Kara <jack@suse.cz>
16 years agoblock: optimize generic_unplug_device()
Jens Axboe [Wed, 7 May 2008 07:48:17 +0000 (09:48 +0200)]
block: optimize generic_unplug_device()

Original patch from Mikulas Patocka <mpatocka@redhat.com>

Mike Anderson was doing an OLTP benchmark on a computer with 48 physical
disks mapped to one logical device via device mapper.

He found that there was a slowdown on request_queue->lock in function
generic_unplug_device. The slowdown is caused by the fact that when some
code calls unplug on the device mapper, device mapper calls unplug on all
physical disks. These unplug calls take the lock, find that the queue is
already unplugged, release the lock and exit.

With the below patch, performance of the benchmark was increased by 18%
(the whole OLTP application, not just block layer microbenchmarks).

So I'm submitting this patch for upstream. I think the patch is correct,
because when more threads call simultaneously plug and unplug, it is
unspecified, if the queue is or isn't plugged (so the patch can't make
this worse). And the caller that plugged the queue should unplug it
anyway. (if it doesn't, there's 3ms timeout).

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: get rid of likely/unlikely predictions in merge logic
Jens Axboe [Wed, 7 May 2008 07:33:55 +0000 (09:33 +0200)]
block: get rid of likely/unlikely predictions in merge logic

They tend to depend a lot on the workload, so not a clear-cut
likely or unlikely fit.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agovfs: splice remove_suid() cleanup
Miklos Szeredi [Wed, 7 May 2008 07:22:39 +0000 (09:22 +0200)]
vfs: splice remove_suid() cleanup

generic_file_splice_write() duplicates remove_suid() just because it
doesn't hold i_mutex.  But it grabs i_mutex inside splice_from_pipe()
anyway, so this is rather pointless.

Move locking to generic_file_splice_write() and call remove_suid() and
__splice_from_pipe() instead.

Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agocfq-iosched: fix RCU race in the cfq io_context destructor handling
Jens Axboe [Wed, 7 May 2008 07:17:12 +0000 (09:17 +0200)]
cfq-iosched: fix RCU race in the cfq io_context destructor handling

put_io_context() drops the RCU read lock before calling into cfq_dtor(),
however we need to hold off freeing there before grabbing and
dereferencing the first object on the list.

So extend the rcu_read_lock() scope to cover the calling of cfq_dtor(),
and optimize cfq_free_io_context() to use a new variant for
call_for_each_cic() that assumes the RCU read lock is already held.

Hit in the wild by Alexey Dobriyan <adobriyan@gmail.com>

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: adjust tagging function queue bit locking
Jens Axboe [Wed, 7 May 2008 07:27:43 +0000 (09:27 +0200)]
block: adjust tagging function queue bit locking

For most initialization purposes, calling blk_queue_init_tags() without
the queue lock held is OK. Only if called for resizing an existing map
must the lock be held. Ditto for tag cleanup, the maps are reference
counted.

So switch the general queue flag setting to the unlocked variant, but
retain the locked variant for resizing.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years ago[S390] guest page hinting light
Martin Schwidefsky [Wed, 7 May 2008 07:22:59 +0000 (09:22 +0200)]
[S390] guest page hinting light

Use the existing arch_alloc_page/arch_free_page callbacks to do
the guest page state transitions between stable and unused.

Acked-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
16 years ago[S390] tty3270: fix put_char fail/success conversion.
Heiko Carstens [Wed, 7 May 2008 07:22:58 +0000 (09:22 +0200)]
[S390] tty3270: fix put_char fail/success conversion.

The wrong function got coverted ;)

  CC      drivers/s390/char/tty3270.o
drivers/s390/char/tty3270.c:1747:
 warning: initialization from incompatible pointer type

Acked-by: Alan Cox <alan@redhat.com>
Cc: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
16 years ago[S390] compat ptrace cleanup
Roland McGrath [Wed, 7 May 2008 07:22:57 +0000 (09:22 +0200)]
[S390] compat ptrace cleanup

This removes redundant arch code for generic ptrace requests
already handled by ptrace_request and compat_ptrace_request.
It simplifies things to just have the standard entry points,
and use the generic compat_sys_ptrace.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
16 years ago[S390] s390mach compile warning
Martin Schwidefsky [Wed, 7 May 2008 07:22:56 +0000 (09:22 +0200)]
[S390] s390mach compile warning

Fix the following compile warning:

drivers/s390/s390mach.c: In function 's390_collect_crw_info':
drivers/s390/s390mach.c:77: warning: ignoring return value of 'down_interruptibl

Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
16 years ago[S390] cio: Fix parsing mechanism for blacklisted devices.
Michael Ernst [Wed, 7 May 2008 07:22:55 +0000 (09:22 +0200)]
[S390] cio: Fix parsing mechanism for blacklisted devices.

New format cssid.ssid.devno is now parsed correctly.

Signed-off-by: Michael Ernst <mernst@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
16 years ago[S390] cio: Remove cio_msg kernel parameter.
Michael Ernst [Wed, 7 May 2008 07:22:54 +0000 (09:22 +0200)]
[S390] cio: Remove cio_msg kernel parameter.

The only sporadically used CIO_DEBUG messages are replaced by ordinary
CIO_MSG_EVENT messages. The CIO_MSG_EVENT messages debug levels are
consolidated.

Signed-off-by: Michael Ernst <mernst@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
16 years ago[S390] s390-kvm: leave sie context on work. Removes preemption requirement
Christian Borntraeger [Wed, 7 May 2008 07:22:53 +0000 (09:22 +0200)]
[S390] s390-kvm: leave sie context on work. Removes preemption requirement

From: Martin Schwidefsky <schwidefsky@de.ibm.com>

This patch fixes a bug with cpu bound guest on kvm-s390. Sometimes it
was impossible to deliver a signal to a spinning guest. We used
preemption as a circumvention. The preemption notifiers called
vcpu_load, which checked for pending signals and triggered a host
intercept. But even with preemption, a sigkill was not delivered
immediately.

This patch changes the low level host interrupt handler to check for the
SIE  instruction, if TIF_WORK is set. In that case we change the
instruction pointer of the return PSW to rerun the vcpu_run loop. The kvm
code sees an intercept reason 0 if that happens. This patch adds accounting
for these types of intercept as well.

The advantages:
- works with and without preemption
- signals are delivered immediately
- much better host latencies without preemption

Acked-by: Carsten Otte <cotte@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
16 years ago[S390] s390: Optimize user and work TIF check
Martin Schwidefsky [Wed, 7 May 2008 07:22:52 +0000 (09:22 +0200)]
[S390] s390: Optimize user and work TIF check

On return from syscall or interrupt, we have to check if we return to
userspace (likely) and if there is work todo (less likely) to decide
if we handle the work. We can optimize this check: we first check for
the less likely work case and then check for userspace.

This patch is also a preparation for an additional patch, that fixes a bug
in KVM dealing with cpu bound guests.

Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
16 years agoblock: sysfs store function needs to grab queue_lock and use queue_flag_*()
Jens Axboe [Wed, 7 May 2008 07:09:39 +0000 (09:09 +0200)]
block: sysfs store function needs to grab queue_lock and use queue_flag_*()

Concurrency isn't a big deal here since we have requests in flight
at this point, but do the locked variant to set a better example.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc-2.6
Linus Torvalds [Wed, 7 May 2008 01:18:43 +0000 (18:18 -0700)]
Merge git://git./linux/kernel/git/davem/sparc-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc-2.6:
  sparc64: Fix initrd regression.
  usb: Sparc build fix, make USB_ISP1760_OF depend on PPC_OF
  sparc64: remove online_page()
  sparc64: use compat_sys_utimes instead of home-grown local copy.
  sbus: Fix bpp driver build.
  sparc video: make blank use proper constant
  Revert "[SPARC64]: Wrap SMP IPIs with irq_enter()/irq_exit()."
  sparc: tcx.c remove unnecessary function

16 years agoRevert "uml: fix gcc problem"
Linus Torvalds [Wed, 7 May 2008 00:09:27 +0000 (17:09 -0700)]
Revert "uml: fix gcc problem"

This reverts commit 22eecde2f9034764a3fd095eecfa3adfb8ec9a98.  Uli
reports that it breaks UML on x86-64 with the Fedora 8 gcc (gcc 4.1.2),
causing a crash on startup. See

http://marc.info/?l=linux-kernel&m=121011722806093&w=2

for a trace.

Reported-by: Ulrich Drepper <drepper@redhat.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoRDMA/cxgb3: Fix severe limit on userspace memory registration size
Roland Dreier [Tue, 6 May 2008 22:56:22 +0000 (15:56 -0700)]
RDMA/cxgb3: Fix severe limit on userspace memory registration size

Currently, iw_cxgb3 is severely limited on the amount of userspace
memory that can be registered in in a single memory region, which
causes big problems for applications that expect to be able to
register 100s of MB.

The problem is that the driver uses a single kmalloc()ed buffer to
hold the physical buffer list (PBL) for the entire memory region
during registration, which means that 8 bytes of contiguous memory are
required for each page of memory being registered.  For example, a 64
MB registration will require 128 KB of contiguous memory with 4 KB
pages, and it unlikely that such an allocation will succeed on a busy
system.

This is purely a driver problem: the temporary page list buffer is not
needed by the hardware, so we can fix this by writing the PBL to the
hardware in page-sized chunks rather than all at once.  We do this by
splitting the memory registration operation up into several steps:

 - Allocate PBL space in adapter memory for the full registration
 - Copy PBL to adapter memory in chunks
 - Allocate STag and enable memory region

This also allows several other cleanups to the __cxio_tpt_op()
interface and related parts of the driver.

This change leaves the reregister memory region and memory window
operations broken, but they already didn't work due to other
longstanding bugs, so fixing them will be left to a later patch.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agosparc64: Fix initrd regression.
David S. Miller [Tue, 6 May 2008 22:19:54 +0000 (15:19 -0700)]
sparc64: Fix initrd regression.

We die because we forget to convert initrd_start and
initrd_end to virtual addresses.

Reported by Mikael Pettersson

Signed-off-by: David S. Miller <davem@davemloft.net>
16 years agousb: Sparc build fix, make USB_ISP1760_OF depend on PPC_OF
David S. Miller [Tue, 6 May 2008 22:15:12 +0000 (15:15 -0700)]
usb: Sparc build fix, make USB_ISP1760_OF depend on PPC_OF

Sparc doesn't have some of the OF interfaces this driver
wants to use.

Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years agoRDMA/cxgb3: Don't add PBL memory to gen_pool in chunks
Roland Dreier [Tue, 6 May 2008 22:03:38 +0000 (15:03 -0700)]
RDMA/cxgb3: Don't add PBL memory to gen_pool in chunks

Current iw_cxgb3 code adds PBL memory to the driver's gen_pool in 2 MB
chunks.  This limits the largest single allocation that can be done to
the same size, which means that with 4 KB pages, each of which takes 8
bytes of PBL memory, the largest memory region that can be allocated
is 1 GB (256K PBL entries * 4 KB/entry).

Remove this limit by adding all the PBL memory in a single gen_pool
chunk, if possible.  Add code that falls back to smaller chunks if
gen_pool_add() fails, which can happen if there is not sufficient
contiguous lowmem for the internal gen_pool bitmap.

Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoFix bogus warning in sysdev_driver_register()
OGAWA Hirofumi [Tue, 6 May 2008 19:02:53 +0000 (04:02 +0900)]
Fix bogus warning in sysdev_driver_register()

        if ((drv->entry.next != drv->entry.prev) ||
            (drv->entry.next != NULL)) {

warns list_empty(&drv->entry).

Signed-off-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Cc: Greg KH <gregkh@suse.de>
Cc: Len Brown <lenb@kernel.org>
[ Version 2 totally redone based on suggestions from Linus & Greg ]
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoVFS: fix unused variable warning
Linus Torvalds [Tue, 6 May 2008 20:13:37 +0000 (13:13 -0700)]
VFS: fix unused variable warning

Commit 33dcdac2df54e66c447ae03f58c95c7251aa5649 ("kill ->put_inode")
removed the final use of i_op->put_inode, but left the now totally
unused "op" variable in iput().

Get rid of it.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agox86: fix PAE pmd_bad bootup warning
Hugh Dickins [Tue, 6 May 2008 19:49:23 +0000 (20:49 +0100)]
x86: fix PAE pmd_bad bootup warning

Fix warning from pmd_bad() at bootup on a HIGHMEM64G HIGHPTE x86_32.

That came from 9fc34113f6880b215cbea4e7017fc818700384c2 x86: debug pmd_bad();
but we understand now that the typecasting was wrong for PAE in the previous
version: pagetable pages above 4GB looked bad and stopped Arjan from booting.

And revert that cded932b75ab0a5f9181ee3da34a0a488d1a14fd x86: fix pmd_bad
and pud_bad to support huge pages.  It was the wrong way round: we shouldn't
weaken every pmd_bad and pud_bad check to let huge pages slip through - in
part they check that we _don't_ have a huge page where it's not expected.

Put the x86 pmd_bad() and pud_bad() definitions back to what they have long
been: they can be improved (x86_32 should use PTE_MASK, to stop PAE thinking
junk in the upper word is good; and x86_64 should follow x86_32's stricter
comparison, to stop thinking any subset of required bits is good); but that
should be a later patch.

Fix Hans' good observation that follow_page() will never find pmd_huge()
because that would have already failed the pmd_bad test: test pmd_huge in
between the pmd_none and pmd_bad tests.  Tighten x86's pmd_huge() check?
No, once it's a hugepage entry, it can get quite far from a good pmd: for
example, PROT_NONE leaves it with only ACCESSED of the KERN_PGTABLE bits.

However... though follow_page() contains this and another test for huge
pages, so it's nice to keep it working on them, where does it actually get
called on a huge page?  get_user_pages() checks is_vm_hugetlb_page(vma) to
to call alternative hugetlb processing, as does unmap_vmas() and others.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Earlier-version-tested-by: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Jeff Chua <jeff.chua.linux@gmail.com>
Cc: Hans Rosenfeld <hans.rosenfeld@amd.com>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs-2.6
Linus Torvalds [Tue, 6 May 2008 18:39:57 +0000 (11:39 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/viro/vfs-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs-2.6:
  [PATCH] fix SMP ordering hole in fcntl_setlk()
  [PATCH] kill ->put_inode
  [PATCH] fix reservation discarding in affs

16 years ago[PATCH] fix SMP ordering hole in fcntl_setlk()
Al Viro [Tue, 6 May 2008 17:58:34 +0000 (13:58 -0400)]
[PATCH] fix SMP ordering hole in fcntl_setlk()

fcntl_setlk()/close() race prevention has a subtle hole - we need to
make sure that if we *do* have an fcntl/close race on SMP box, the
access to descriptor table and inode->i_flock won't get reordered.

As it is, we get STORE inode->i_flock, LOAD descriptor table entry vs.
STORE descriptor table entry, LOAD inode->i_flock with not a single
lock in common on both sides.  We do have BKL around the first STORE,
but check in locks_remove_posix() is outside of BKL and for a good
reason - we don't want BKL on common path of close(2).

Solution is to hold ->file_lock around fcheck() in there; that orders
us wrt removal from descriptor table that preceded locks_remove_posix()
on close path and we either come first (in which case eviction will be
handled by the close side) or we'll see the effect of close and do
eviction ourselves.  Note that even though it's read-only access,
we do need ->file_lock here - rcu_read_lock() won't be enough to
order the things.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
16 years ago[PATCH] kill ->put_inode
Christoph Hellwig [Tue, 29 Apr 2008 15:46:26 +0000 (17:46 +0200)]
[PATCH] kill ->put_inode

And with that last patch to affs killing the last put_inode instance we
can finally, after many years of transition kill this racy and awkward
interface.

(It's kinda funny that even the description in
Documentation/filesystems/vfs.txt was entirely wrong..)

Also remove a very misleading comment above the defintion of
struct super_operations.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
16 years ago[PATCH] fix reservation discarding in affs
Roman Zippel [Tue, 29 Apr 2008 15:02:20 +0000 (17:02 +0200)]
[PATCH] fix reservation discarding in affs

- remove affs_put_inode, so preallocations aren't discared unnecessarily
  often.
- remove affs_drop_inode, it's called with a spinlock held, so it can't
  use a mutex.
- make i_opencnt atomic
- avoid direct b_count manipulations
- a few allocation failure fixes, so that these are more gracefully
  handled now.
Signed-off-by: Roman Zippel <zippel@linux-m68k.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
16 years agoe1000e: don't return half-read eeprom on error
Kok, Auke [Tue, 29 Apr 2008 18:18:55 +0000 (11:18 -0700)]
e1000e: don't return half-read eeprom on error

On a read error, e1000e might have returned uninitialized block of
eeprom data back to userspace. The convention is that 0xff is "empty",
so mark the entire eeprom as empty in case of an error.

Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
16 years agoucc_geth: Don't use RX clock as TX clock.
Joakim Tjernlund [Tue, 29 Apr 2008 11:03:57 +0000 (13:03 +0200)]
ucc_geth: Don't use RX clock as TX clock.

Commit 9fb1e350e16164d56990dde036ae9c0a2fd3f634,
ucc_geth: use rx-clock-name and tx-clock-name device tree properties
Introduced a typo that made the driver use the RX clock
as TX clock, causing massive TX errors.

Signed-off-by: Joakim Tjernlund <Joakim.Tjernlund@transmode.se>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
16 years agocxgb3: Use CAP_SYS_RAWIO for firmware
Alan Cox [Tue, 29 Apr 2008 13:29:30 +0000 (14:29 +0100)]
cxgb3: Use CAP_SYS_RAWIO for firmware

Otherwise theoretically at least

CAP_NET_ADMIN
Reload new firmware
Wait..
Firmware patches kernel

So it should be CAY_SYS_RAWIO - not that I suspect this is in fact a
credible attack vector!

Signed-off-by: Alan Cox <alan@redhat.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
16 years agopcnet32: delete non NAPI code from driver.
Don Fry [Tue, 29 Apr 2008 20:49:58 +0000 (13:49 -0700)]
pcnet32: delete non NAPI code from driver.

Delete the non-napi code from the driver and Kconfig.
Tested x86_64.  Apply at next open opportunity.

Signed-off-by: Don Fry <pcnet32@verizon.net>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
16 years agofs_enet: Fix a memory leak in fs_enet_mdio_probe
Scott Wood [Fri, 2 May 2008 18:42:41 +0000 (13:42 -0500)]
fs_enet: Fix a memory leak in fs_enet_mdio_probe

There are more memory leaks in the !PPC_CPM_NEW_BINDING case, but that code
will disappear soon along with arch/ppc.

Reported by Daniel Marjamki <danielm77@spray.se> at
http://bugzilla.kernel.org/show_bug.cgi?id=10591

Signed-off-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
16 years ago[netdrvr] eexpress: IPv6 fails - multicast problems
Bruce Robson [Fri, 2 May 2008 20:40:53 +0000 (13:40 -0700)]
[netdrvr] eexpress: IPv6 fails - multicast problems

Taken from http://bugzilla.kernel.org/show_bug.cgi?id=10577

I was unable to access a computer containing an Intel EtherExpress 16 network
card using IPv6.

I traced this to failure of neighbour discovery.  When I used an "ip -6 neigh
add" command, on the computer attempting access, to insert a binding between
the IPv6 address of the computer with the Intel EtherExpress 16 network card
and the card's ethernet address, I was able to access that computer using
IPv6.

Neighbour discovery requires working multicast.  The driver sources file
eexpress.c contains an approximately 30 line function eexp_setup_filter used
when loading multicast addresses.

I found 3 problems in this function

1) It wrote the number of multicast addresses to the card instead of the
    number of bytes in the multicast addresses.

2) When loading multiple multicast addresses it loaded the first one
    provided multiple times instead of loading each one once.

3) The setting of pointer 'data' from 'dmi->dmi_addr' occured before the
    test for the error situation of 'dmi' being NULL.

Correcting these problems allows the computer with the Intel EtherExpress 16
network card to found by IPv6 neighbour discovery.

p.s. There is some information on the Intel EtherExpress 16 at
http://www.intel.com/support/etherexpress/vintage/sb/cs-013500.htm
Datasheet for the Intel 82586 ethernet controller used by the card
http://www.datasheetcatalog.com/datasheets_pdf/8/2/5/8/82586.shtml

Signed-off-by: Bruce Robson <bns_robson@hotmail.com>
Cc: Jeff Garzik <jeff@garzik.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
16 years ago3c59x: use netstats in net_device structure
Paulius Zaleckas [Mon, 5 May 2008 12:01:29 +0000 (14:01 +0200)]
3c59x: use netstats in net_device structure

Use net_device_stats from net_device structure instead of local.

Signed-off-by: Paulius Zaleckas <paulius.zaleckas@teltonika.lt>
Acked-by: Steffen Klassert <klassert@mathematik.tu-chemnitz.de>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
16 years ago3c980-TX needs EXTRA_PREAMBLE
Gunnar Larisch [Mon, 5 May 2008 12:01:28 +0000 (14:01 +0200)]
3c980-TX needs EXTRA_PREAMBLE

The ethernet card 3c980-TX needs a mdio_sync() to initialize the ethernet
properly. This is forced by adding an EXTRA_PREAMBLE to its drv_flags.

Without this, the driver did not reconnect after a link loss.

Signed-off-by: Gunnar Larisch <Gunnar.Larisch@gmx.de>
Acked-by: Steffen Klassert <klassert@mathematik.tu-chemnitz.de>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>