kernel/kernel-generic.git
16 years ago[MIPS] Replace 40c7869b693b18412491fdcff64682215b739f9e kludge
Atsushi Nemoto [Mon, 7 Jan 2008 15:41:13 +0000 (00:41 +0900)]
[MIPS] Replace 40c7869b693b18412491fdcff64682215b739f9e kludge

Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
16 years ago[MIPS] Lasat: Fix built in separate object directory.
WANG Cong [Wed, 2 Jan 2008 06:21:36 +0000 (14:21 +0800)]
[MIPS] Lasat: Fix built in separate object directory.

Signed-off-by: WANG Cong <xiyou.wangcong@gmail.com>
[Ralf: The LDSCRIPT script needed fixing, too]

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
16 years ago[MIPS] Malta: Fix software reset on big endian
Dmitri Vorobiev [Tue, 8 Jan 2008 03:44:00 +0000 (06:44 +0300)]
[MIPS] Malta: Fix software reset on big endian

I noticed that the commit f197465384bf7ef1af184c2ed1a4e268911a91e3
(MIPS Tech: Get rid of volatile in core code) broke the software
reset functionality for MIPS Malta boards in big-endian mode.

According to the MIPS Malta board user's manual, writing the magic
32-bit GORESET value into the SOFTRES register initiates board soft
reset. My experimentation has shown that the endianness of the GORESET
integer should thereby be the same as the endianness, which has been
set for the CPU itself. The writew() function used to write the magic
value in the code introduced by the commit mentioned above, however,
swaps bytes for big-endian kernels and transfers 16 bits instead of 32.

The patch below replaces the writew() function by the __raw_writel()
routine, which leaves the byte order intact and transfers the whole
MIPS machine word. Trivial code cleanup (replacing spaces by a tab
and cutting oversized lines to make checkpatch.pl happy) is also
included.

The patch was tested using a Malta evaluation board running in both
BE and LE modes. For both modes, software reset was fully functional
after the change.

P.S. I suspect that the same commit broke the "standby" functionality
for MIPS Atlas boards. However, I did not touch the Atlas code as I
don't have such board at my disposal and also because the linux-mips.org
Web site claims that Atlas support is scheduled for removal.

Signed-off-by: Dmitri Vorobiev <dmitri.vorobiev@gmail.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
16 years ago[MIPS] pnx8xxx: move to clocksource
Vitaly Wool [Thu, 10 Jan 2008 14:10:05 +0000 (17:10 +0300)]
[MIPS] pnx8xxx: move to clocksource

This patch converts PNX8XXX system timer to clocksource restoring PNX8550
support back to live.

Signed-off-by: Vitaly Wool <vitalywool@gmail.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
16 years ago[MIPS] Wrong CONFIG option prevents setup of DMA zone.
Thomas Bogendoerfer [Sun, 30 Dec 2007 11:45:40 +0000 (12:45 +0100)]
[MIPS] Wrong CONFIG option prevents setup of DMA zone.

Signed-off-by: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
16 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6
Linus Torvalds [Fri, 11 Jan 2008 00:09:34 +0000 (16:09 -0800)]
Merge git://git./linux/kernel/git/herbert/crypto-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6:
  [CRYPTO] padlock: Fix alignment fault in aes_crypt_copy

16 years agoMerge branch 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jgarzi...
Linus Torvalds [Thu, 10 Jan 2008 22:53:37 +0000 (14:53 -0800)]
Merge branch 'upstream-linus' of git://git./linux/kernel/git/jgarzik/libata-dev

* 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jgarzik/libata-dev:
  libata and starting/stopping ATAPI floppy devices
  sata_sil24: prevent hba lockup when pass-through ATA commands are used
  Update kernel parameter document for libata DMA mode setting knobs.
  libata: don't normalize UNKNOWN to NONE after reset
  libata-pmp: propagate timeout to host link
  libata-pmp: 4726 hates SRST
  pata_ixp4xx_cf: fix compilation introduced by ata_port_desc() conversion
  pata_pdc202xx_old: Further fixups
  libata-sff: PCI IRQ handling fix
  sata_qstor: use hardreset instead of softreset

16 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/bart/ide-2.6
Linus Torvalds [Thu, 10 Jan 2008 22:31:16 +0000 (14:31 -0800)]
Merge git://git./linux/kernel/git/bart/ide-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/bart/ide-2.6:
  trm290: do hook dma_host_{on,off} methods (take 2)
  ide: fix cable detection for SATA bridges
  ide: workaround suspend bug for ACPI IDE

16 years agolibata and starting/stopping ATAPI floppy devices
Ondrej Zary [Thu, 10 Jan 2008 22:13:19 +0000 (17:13 -0500)]
libata and starting/stopping ATAPI floppy devices

Prevent libata from starting/stopping non-ATA devices (like ATAPI floppy
drives) as they don't seem to like it:

sd 1:0:1:0: [sdb] Starting disk
ata2.01: configured for PIO2
sd 1:0:1:0: [sdb] Result: hostbyte=0x00 driverbyte=0x08
sd 1:0:1:0: [sdb] Sense Key : 0x2 [current]
sd 1:0:1:0: [sdb] ASC=0x3a ASCQ=0x0

Signed-off-by: Ondrej Zary <linux@rainbow-software.org>
Signed-off-by: Tejun Heo <htejun@gmail.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
16 years agotrm290: do hook dma_host_{on,off} methods (take 2)
Sergei Shtylyov [Thu, 10 Jan 2008 22:03:42 +0000 (23:03 +0100)]
trm290: do hook dma_host_{on,off} methods (take 2)

Using default methods caused the chip's DMA PRD count registers, inadvertently
starting DMA!  While fixing it, also do:

- get rid of the 'ide_' prefixes in several functions for which the prefix in
  the method's name has been 'ide_' ectomized already;

- align the code hooking the IDE DMA methods in init_hwif_trm290()...

Signed-off-by: Sergei Shtylyov <sshtylyov@ru.mvista.com>
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
16 years agoide: fix cable detection for SATA bridges
George Kibardin [Thu, 10 Jan 2008 22:03:42 +0000 (23:03 +0100)]
ide: fix cable detection for SATA bridges

Signed-off-by: George Kibardin <george-kibardin@yandex.ru>
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
16 years agoide: workaround suspend bug for ACPI IDE
Shaohua Li [Thu, 10 Jan 2008 22:03:42 +0000 (23:03 +0100)]
ide: workaround suspend bug for ACPI IDE

http://bugzilla.kernel.org/show_bug.cgi?id=9673

ACPI _PS3 cause S4 breaks in the second attempt. The system has a _PS3
method for IDE, which will call into SMM mode. Currently we haven't clue
why just the second attempt fails, as it's totally in BIOS code, so
blacklist the system so far for 2.6.24.

A possible suspect is ACPI NVS isn't save/restore, we will revisit the
bug after linux does ACPI NVS save/restore.

Bart:
- fix scripts/checkpatch.pl complaints
- const-ify ide_acpi_dmi_table[]

Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Reported-by: Mikko Vinni <mmvinni@yahoo.com>
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
16 years agosata_sil24: prevent hba lockup when pass-through ATA commands are used
Gwendal Grignou [Thu, 10 Jan 2008 06:47:56 +0000 (15:47 +0900)]
sata_sil24: prevent hba lockup when pass-through ATA commands are used

Fix commands timeout with Sil3124/3132 based HBA when pass-through ATA
commands [where ATA_QCFLAG_RESULT_TF is set] are used while other
commands are active on other devices connected to the same port with a
Port Multiplier.  Due to a hardware bug, these commands must be sent
alone, like ATAPI commands.

Signed-off-by: Gwendal Grignou <gwendal@google.com>
Acked-by: Tejun Heo <htejun@gmail.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
16 years agoUpdate kernel parameter document for libata DMA mode setting knobs.
FD Cami [Sun, 6 Jan 2008 18:08:56 +0000 (19:08 +0100)]
Update kernel parameter document for libata DMA mode setting knobs.

Signed-off-by: Jeff Garzik <jeff@garzik.org>
16 years agolibata: don't normalize UNKNOWN to NONE after reset
Tejun Heo [Tue, 8 Jan 2008 11:26:12 +0000 (20:26 +0900)]
libata: don't normalize UNKNOWN to NONE after reset

After non-classifying reset, ehc->classes[] could contain
ATA_DEV_UNKNOWN which used to be normalized to ATA_DEV_NONE for
consistency.  However, this causes unfortunate side effect for drivers
which have non-classifying hardresets (e.g. sata_nv) by making
hardreset report ATA_DEV_NONE for non-classifying resets and thus
makes EH believe that the port is unoccupied and recovery can be
skipped.  The end result is that after a device is swapped with
another one, the new device isn't attached after the old one is
detached.

This patch makes ata_eh_reset() not normalize UNKNOWN to NONE after
non-classifying resets.  This fixes the above problem.  As UNKNOWN and
NONE are handled differently by only EH hotplug logic, this doesn't
cause other behavior changes.

Signed-off-by: Tejun Heo <htejun@gmail.com>
Cc: Robert Hancock <hancockr@shaw.ca>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
16 years agolibata-pmp: propagate timeout to host link
Tejun Heo [Thu, 10 Jan 2008 04:41:23 +0000 (13:41 +0900)]
libata-pmp: propagate timeout to host link

Timeout on downstream command may indicate transmission problem on
host link.  Propagate timeouts to host link.

Signed-off-by: Tejun Heo <htejun@gmail.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
16 years agolibata-pmp: 4726 hates SRST
Tejun Heo [Thu, 10 Jan 2008 04:38:46 +0000 (13:38 +0900)]
libata-pmp: 4726 hates SRST

4726 hates SRST even on non-config ports.  Don't use it.

Signed-off-by: Tejun Heo <htejun@gmail.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
16 years agopata_ixp4xx_cf: fix compilation introduced by ata_port_desc() conversion
Rod Whitby [Sun, 6 Jan 2008 01:05:28 +0000 (10:05 +0900)]
pata_ixp4xx_cf: fix compilation introduced by ata_port_desc() conversion

Fixes a compilation error caused by ata_port_desc() conversion
(cbcdd87593a1d85c5c4b259945a3a09eee12814d).

Signed-off-by: Rod Whitby <rod@whitby.id.au>
Signed-off-by: Tejun Heo <htejun@gmail.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
16 years agopata_pdc202xx_old: Further fixups
Alan Cox [Fri, 4 Jan 2008 00:08:49 +0000 (00:08 +0000)]
pata_pdc202xx_old: Further fixups

Turns out distros always enabled burst mode and it is pretty essential so
do the same. Also sort out the post DMA mode restore properly.

My 20263 card now seems happy but needs some four drive tests done yet
(when I've persuaded the kernel not to hang in the edd boot code if I
plug them in ..)

Signed-off-by: Alan Cox <alan@redhat.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
16 years agolibata-sff: PCI IRQ handling fix
Alan Cox [Thu, 3 Jan 2008 17:22:28 +0000 (17:22 +0000)]
libata-sff: PCI IRQ handling fix

It is legitimate (although annoying and silly) for a PCI IDE controller
not to be assigned an interrupt and to be polled. The libata-sff code
should therefore not try and request IRQ 0 in this case.

Signed-off-by: Alan Cox <alan@redhat.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
16 years agosata_qstor: use hardreset instead of softreset
Tejun Heo [Thu, 3 Jan 2008 15:22:19 +0000 (00:22 +0900)]
sata_qstor: use hardreset instead of softreset

During conversion to new EH, sata_qstor was accidentaly changed to use
softreset, which is buggy on this chip, instead of hardreset.  This
patch updates sata_qstor such that it uses hardreset again.

This fixes bugzilla bug 9631.

Signed-off-by: Tejun Heo <htejun@gmail.com>
Cc: Mark Lord <mlord@pobox.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
16 years agoNFSv4: Give the lock stateid its own sequence queue
Trond Myklebust [Thu, 10 Jan 2008 21:07:54 +0000 (16:07 -0500)]
NFSv4: Give the lock stateid its own sequence queue

Sharing the open sequence queue causes a deadlock when we try to take
both a lock sequence id and and open sequence id.

This fixes the regression reported by Dimitri Puzin and Jeff Garzik: See

http://bugzilla.kernel.org/show_bug.cgi?id=9712

for details.

Reported-and-tested-by: Dimitri Puzin <bugs@psycast.de>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Tested-by: Jeff Garzik <jgarzik@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years ago[CRYPTO] padlock: Fix alignment fault in aes_crypt_copy
Herbert Xu [Thu, 10 Jan 2008 21:09:35 +0000 (08:09 +1100)]
[CRYPTO] padlock: Fix alignment fault in aes_crypt_copy

The previous patch fixed spurious read faults from occuring by copying
the data if we happen to have a single block at the end of a page.  It
appears that gcc cannot guarantee 16-byte alignment in the kernel with
__attribute__.  The following report from Torben Viets shows a buffer
that's only 8-byte aligned:

> eneral protection fault: 0000 [#1]
> Modules linked in: xt_TCPMSS xt_tcpmss iptable_mangle ipt_MASQUERADE
> xt_tcpudp xt_mark xt_state iptable_nat nf_nat nf_conntrack_ipv4
> iptable_filter ip_tables x_tables pppoe pppox af_packet ppp_generic slhc
> aes_i586
> CPU:    0
> EIP:    0060:[<c035b828>]    Not tainted VLI
> EFLAGS: 00010292   (2.6.23.12 #7)
> EIP is at aes_crypt_copy+0x28/0x40
> eax: f7639ff0   ebx: f6c24050   ecx: 00000001   edx: f6c24030
> esi: f7e89dc8   edi: f7639ff0   ebp: 00010000   esp: f7e89dc8

Since the hardware must have 16-byte alignment, the following patch fixes
this by open coding the alignment adjustment.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
16 years ago[SCSI] qla1280: fix 32 bit segment code
FUJITA Tomonori [Tue, 8 Jan 2008 14:07:01 +0000 (23:07 +0900)]
[SCSI] qla1280: fix 32 bit segment code

There's an error remaining in the 32 bit descriptor code after the
conversion to dma accessors:  req_cnt is left uninitialised.

qla1280_32bit_start_scsi gives the following warnings:

drivers/scsi/qla1280.c: In function 'qla1280_32bit_start_scsi':
drivers/scsi/qla1280.c:3044: warning: unused variable 'dma_handle'
drivers/scsi/qla1280.c: In function 'qla1280_queuecommand':
drivers/scsi/qla1280.c:3060: warning: 'req_cnt' is used uninitialized in this function
drivers/scsi/qla1280.c:3042: note: 'req_cnt' was declared here

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
16 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
Linus Torvalds [Wed, 9 Jan 2008 16:05:12 +0000 (08:05 -0800)]
Merge git://git./linux/kernel/git/davem/net-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6: (36 commits)
  [ATM]: Check IP header validity in mpc_send_packet
  [IPV6]: IPV6_MULTICAST_IF setting is ignored on link-local connect()
  [CONNECTOR]: Don't touch queue dev after decrement of ref count.
  [SOCK]: Adds a rcu_dereference() in sk_filter
  [XFRM]: xfrm_algo_clone() allocates too much memory
  [FORCEDETH]: Fix reversing the MAC address on suspend.
  [NET]: mcs7830 passes msecs instead of jiffies to usb_control_msg
  [LRO] Fix lro_mgr->features checks
  [NET]: Clone the sk_buff 'iif' field in __skb_clone()
  [IPV4] ROUTE: ip_rt_dump() is unecessary slow
  [NET]: kaweth was forgotten in msec switchover of usb_start_wait_urb
  [NET] Intel ethernet drivers: update MAINTAINERS
  [NET]: Make ->poll() breakout consistent in Intel ethernet drivers.
  [NET]: Stop polling when napi_disable() is pending.
  [NET]: Fix drivers to handle napi_disable() disabling interrupts.
  [NETXEN]: Fix ->poll() done logic.
  mac80211: return an error when SIWRATE doesn't match any rate
  ssb: Fix probing of PCI cores if PCI and PCIE core is available
  [NET]: Do not check netif_running() and carrier state in ->poll()
  [NET]: Add NAPI_STATE_DISABLE.
  ...

16 years agoshow_task: real_parent
Roland McGrath [Wed, 9 Jan 2008 08:03:23 +0000 (00:03 -0800)]
show_task: real_parent

The show_task function invoked by sysrq-t et al displays the
pid and parent's pid of each task.  It seems more useful to
show the actual process hierarchy here than who is using
ptrace on each process.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years ago[ATM]: Check IP header validity in mpc_send_packet
Herbert Xu [Wed, 9 Jan 2008 11:51:59 +0000 (03:51 -0800)]
[ATM]: Check IP header validity in mpc_send_packet

Al went through the ip_fast_csum callers and found this piece of code
that did not validate the IP header.  While root crashing the machine
by sending bogus packets through raw or AF_PACKET sockets isn't that
serious, it is still nice to react gracefully.

This patch ensures that the skb has enough data for an IP header and
that the header length field is valid.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years ago[IPV6]: IPV6_MULTICAST_IF setting is ignored on link-local connect()
Brian Haley [Wed, 9 Jan 2008 07:52:21 +0000 (23:52 -0800)]
[IPV6]: IPV6_MULTICAST_IF setting is ignored on link-local connect()

Signed-off-by: Brian Haley <brian.haley@hp.com>
Acked-by: David L Stevens <dlstevens@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years ago[CONNECTOR]: Don't touch queue dev after decrement of ref count.
Li Zefan [Wed, 9 Jan 2008 07:44:44 +0000 (23:44 -0800)]
[CONNECTOR]: Don't touch queue dev after decrement of ref count.

cn_queue_free_callback() will touch 'dev'(i.e. cbq->pdev), so it
should be called before atomic_dec(&dev->refcnt).

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years ago[SOCK]: Adds a rcu_dereference() in sk_filter
Eric Dumazet [Wed, 9 Jan 2008 07:41:28 +0000 (23:41 -0800)]
[SOCK]: Adds a rcu_dereference() in sk_filter

It seems commit fda9ef5d679b07c9d9097aaf6ef7f069d794a8f9 introduced a RCU
protection for sk_filter(), without a rcu_dereference()

Either we need a rcu_dereference(), either a comment should explain why we
dont need it. I vote for the former.

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Acked-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years ago[XFRM]: xfrm_algo_clone() allocates too much memory
Eric Dumazet [Wed, 9 Jan 2008 07:39:06 +0000 (23:39 -0800)]
[XFRM]: xfrm_algo_clone() allocates too much memory

alg_key_len is the length in bits of the key, not in bytes.

Best way to fix this is to move alg_len() function from net/xfrm/xfrm_user.c
to include/net/xfrm.h, and to use it in xfrm_algo_clone()

alg_len() is renamed to xfrm_alg_len() because of its global exposition.

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years ago[FORCEDETH]: Fix reversing the MAC address on suspend.
Björn Steinbrink [Tue, 8 Jan 2008 07:22:53 +0000 (23:22 -0800)]
[FORCEDETH]: Fix reversing the MAC address on suspend.

For cards that initially have the MAC address stored in reverse order,
the forcedeth driver uses a flag to signal whether the address was
already corrected, so that it is not reversed again on a subsequent
probe.

Unfortunately this flag, which is stored in a register of the card,
seems to get lost during suspend, resulting in the MAC address being
reversed again. To fix that, the MAC address needs to be written back
in reversed order before we suspend and the flag needs to be reset.

The flag is still required because at least kexec will never write
back the reversed address and thus needs to know what state the card
is in.

Signed-off-by: Björn Steinbrink <B.Steinbrink@gmx.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years ago[NET]: mcs7830 passes msecs instead of jiffies to usb_control_msg
Russ Dill [Tue, 8 Jan 2008 07:15:41 +0000 (23:15 -0800)]
[NET]: mcs7830 passes msecs instead of jiffies to usb_control_msg

usb_control_msg was changed long ago (2.6.12-pre) to take milliseconds
instead of jiffies. Oddly, mcs7830 wasn't added until 2.6.19-rc3.

Signed-off-by: Russ Dill <Russ.Dill@asu.edu>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years ago[LRO] Fix lro_mgr->features checks
Brice Goglin [Tue, 8 Jan 2008 06:09:08 +0000 (22:09 -0800)]
[LRO] Fix lro_mgr->features checks

lro_mgr->features contains a bitmask of LRO_F_* values which are
defined as power of two, not as bit indexes.
They must be checked with x&LRO_F_FOO, not with test_bit(LRO_F_FOO,&x).

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
Acked-by: Andrew Gallatin <gallatin@myri.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years ago[NET]: Clone the sk_buff 'iif' field in __skb_clone()
Paul Moore [Tue, 8 Jan 2008 05:56:41 +0000 (21:56 -0800)]
[NET]: Clone the sk_buff 'iif' field in __skb_clone()

Both NetLabel and SELinux (other LSMs may grow to use it as well) rely
on the 'iif' field to determine the receiving network interface of
inbound packets.  Unfortunately, at present this field is not
preserved across a skb clone operation which can lead to garbage
values if the cloned skb is sent back through the network stack.  This
patch corrects this problem by properly copying the 'iif' field in
__skb_clone() and removing the 'iif' field assignment from
skb_act_clone() since it is no longer needed.

Also, while we are here, put the assignments in the same order as the
offsets to reduce cacheline bounces.

Signed-off-by: Paul Moore <paul.moore@hp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years ago[IPV4] ROUTE: ip_rt_dump() is unecessary slow
Eric Dumazet [Tue, 8 Jan 2008 05:52:14 +0000 (21:52 -0800)]
[IPV4] ROUTE: ip_rt_dump() is unecessary slow

I noticed "ip route list cache x.y.z.t" can be *very* slow.

While strace-ing -T it I also noticed that first part of route cache
is fetched quite fast :

recvmsg(3, {msg_name(12)={sa_family=AF_NETLINK, pid=0, groups=00000000}, msg_iov(1)=[{"p\0\0\0\30\0\2\0\254i\202
GXm\0\0\2  \0\376\0\0\2\0\2\0"..., 16384}], msg_controllen=0, msg_flags=0}, 0) = 3772 <0.000047>
recvmsg(3, {msg_name(12)={sa_family=AF_NETLINK, pid=0, groups=00000000}, msg_iov(1)=[{"\234\0\0\0\30\0\2\0\254i\
202GXm\0\0\2  \0\376\0\0\1\0\2"..., 16384}], msg_controllen=0, msg_flags=0}, 0) = 3736 <0.000042>
recvmsg(3, {msg_name(12)={sa_family=AF_NETLINK, pid=0, groups=00000000}, msg_iov(1)=[{"\204\0\0\0\30\0\2\0\254i\
202GXm\0\0\2  \0\376\0\0\1\0\2"..., 16384}], msg_controllen=0, msg_flags=0}, 0) = 3740 <0.000055>
recvmsg(3, {msg_name(12)={sa_family=AF_NETLINK, pid=0, groups=00000000}, msg_iov(1)=[{"\234\0\0\0\30\0\2\0\254i\
202GXm\0\0\2  \0\376\0\0\1\0\2"..., 16384}], msg_controllen=0, msg_flags=0}, 0) = 3712 <0.000043>
recvmsg(3, {msg_name(12)={sa_family=AF_NETLINK, pid=0, groups=00000000}, msg_iov(1)=[{"\204\0\0\0\30\0\2\0\254i\
202GXm\0\0\2  \0\376\0\0\1\0\2"..., 16384}], msg_controllen=0, msg_flags=0}, 0) = 3732 <0.000053>
recvmsg(3, {msg_name(12)={sa_family=AF_NETLINK, pid=0, groups=00000000}, msg_iov(1)=[{"p\0\0\0\30\0\2\0\254i\202
GXm\0\0\2  \0\376\0\0\2\0\2\0"..., 16384}], msg_controllen=0, msg_flags=0}, 0) = 3708 <0.000052>
recvmsg(3, {msg_name(12)={sa_family=AF_NETLINK, pid=0, groups=00000000}, msg_iov(1)=[{"p\0\0\0\30\0\2\0\254i\202
GXm\0\0\2  \0\376\0\0\2\0\2\0"..., 16384}], msg_controllen=0, msg_flags=0}, 0) = 3680 <0.000041>

while the part at the end of the table is more expensive:

recvmsg(3, {msg_name(12)={sa_family=AF_NETLINK, pid=0, groups=00000000}, msg_iov(1)=[{"\204\0\0\0\30\0\2\0\254i\202GXm\0\0\2  \0\376\0\0\1\0\2"..., 16384}], msg_controllen=0, msg_flags=0}, 0) = 3656 <0.003857>
recvmsg(3, {msg_name(12)={sa_family=AF_NETLINK, pid=0, groups=00000000}, msg_iov(1)=[{"\204\0\0\0\30\0\2\0\254i\202GXm\0\0\2  \0\376\0\0\1\0\2"..., 16384}], msg_controllen=0, msg_flags=0}, 0) = 3772 <0.003891>
recvmsg(3, {msg_name(12)={sa_family=AF_NETLINK, pid=0, groups=00000000}, msg_iov(1)=[{"p\0\0\0\30\0\2\0\254i\202GXm\0\0\2  \0\376\0\0\2\0\2\0"..., 16384}], msg_controllen=0, msg_flags=0}, 0) = 3712 <0.003765>
recvmsg(3, {msg_name(12)={sa_family=AF_NETLINK, pid=0, groups=00000000}, msg_iov(1)=[{"p\0\0\0\30\0\2\0\254i\202GXm\0\0\2  \0\376\0\0\2\0\2\0"..., 16384}], msg_controllen=0, msg_flags=0}, 0) = 3700 <0.003879>
recvmsg(3, {msg_name(12)={sa_family=AF_NETLINK, pid=0, groups=00000000}, msg_iov(1)=[{"p\0\0\0\30\0\2\0\254i\202GXm\0\0\2  \0\376\0\0\2\0\2\0"..., 16384}], msg_controllen=0, msg_flags=0}, 0) = 3676 <0.003797>
recvmsg(3, {msg_name(12)={sa_family=AF_NETLINK, pid=0, groups=00000000}, msg_iov(1)=[{"p\0\0\0\30\0\2\0\254i\202GXm\0\0\2  \0\376\0\0\2\0\2\0"..., 16384}], msg_controllen=0, msg_flags=0}, 0) = 3724 <0.003856>
recvmsg(3, {msg_name(12)={sa_family=AF_NETLINK, pid=0, groups=00000000}, msg_iov(1)=[{"\234\0\0\0\30\0\2\0\254i\202GXm\0\0\2  \0\376\0\0\1\0\2"..., 16384}], msg_controllen=0, msg_flags=0}, 0) = 3736 <0.003848>

The following patch corrects this performance/latency problem,
removing quadratic behavior.

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years ago[NET]: kaweth was forgotten in msec switchover of usb_start_wait_urb
Russ Dill [Tue, 8 Jan 2008 05:48:12 +0000 (21:48 -0800)]
[NET]: kaweth was forgotten in msec switchover of usb_start_wait_urb

Back in 2.6.12-pre, usb_start_wait_urb was switched over to take
milliseconds instead of jiffies. kaweth.c was never updated to match.

Signed-off-by: Russ Dill <Russ.Dill@asu.edu>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years ago[NET] Intel ethernet drivers: update MAINTAINERS
Auke Kok [Tue, 8 Jan 2008 05:47:25 +0000 (21:47 -0800)]
[NET] Intel ethernet drivers: update MAINTAINERS

Unfortunately Jeb decided to move away from our group. We wish Jeb
good luck with his new group!

Reordered people a bit so most active team members are on top.

Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years ago[NET]: Make ->poll() breakout consistent in Intel ethernet drivers.
David S. Miller [Tue, 8 Jan 2008 05:06:12 +0000 (21:06 -0800)]
[NET]: Make ->poll() breakout consistent in Intel ethernet drivers.

This makes the ->poll() routines of the E100, E1000, E1000E, IXGB, and
IXGBE drivers complete ->poll() consistently.

Now they will all break out when the amount of RX work done is less
than 'budget'.

At a later time, we may want put back code to include the TX work as
well (as at least one other NAPI driver does, but by in large NAPI
drivers do not do this).  But if so, it should be done consistently
across the board to all of these drivers.

Signed-off-by: David S. Miller <davem@davemloft.net>
Acked-by: Auke Kok <auke-jan.h.kok@intel.com>
16 years ago[NET]: Stop polling when napi_disable() is pending.
David S. Miller [Tue, 8 Jan 2008 05:00:40 +0000 (21:00 -0800)]
[NET]: Stop polling when napi_disable() is pending.

This finally adds the code in net_rx_action() to break out of the
->poll()'ing loop when a napi_disable() is found to be pending.

Now, even if a device is being flooded with packets it can be cleanly
brought down.

Signed-off-by: David S. Miller <davem@davemloft.net>
16 years ago[NET]: Fix drivers to handle napi_disable() disabling interrupts.
David S. Miller [Tue, 8 Jan 2008 04:53:33 +0000 (20:53 -0800)]
[NET]: Fix drivers to handle napi_disable() disabling interrupts.

When we add the generic napi_disable_pending() breakout
logic to net_rx_action() it means that napi_disable()
can cause NAPI poll interrupt events to be disabled.

And this is exactly what we want.  If a napi_disable()
is pending, and we are looping in the ->poll(), we want
->poll() event interrupts to stay disabled and we want
to complete the NAPI poll ASAP.

When ->poll() break out during device down was being handled on a
per-driver basis, often these drivers would turn interrupts back on
when '!netif_running()' was detected.

And this would just cause a reschedule of the NAPI ->poll() in the
interrupt handler before the napi_disable() could get in there and
grab the NAPI_STATE_SCHED bit.

The vast majority of drivers don't care if napi_disable() might have
the side effect of disabling NAPI ->poll() event interrupts.  In all
such cases, when a napi_disable() is performed, the driver just
disabled interrupts or is about to.

However there were three exceptions to this in PCNET32, R8169, and
SKY2.  To fix those cases, at the subsequent napi_enable() points, I
added code to ensure that the ->poll() interrupt events are enabled in
the hardware.

Signed-off-by: David S. Miller <davem@davemloft.net>
Acked-by: Don Fry <pcnet32@verizon.net>
16 years ago[NETXEN]: Fix ->poll() done logic.
David S. Miller [Tue, 8 Jan 2008 04:51:29 +0000 (20:51 -0800)]
[NETXEN]: Fix ->poll() done logic.

If work_done >= budget we should always elide the NAPI
completion.

Signed-off-by: David S. Miller <davem@davemloft.net>
16 years agomac80211: return an error when SIWRATE doesn't match any rate
Andrew Lutomirski [Fri, 4 Jan 2008 05:03:19 +0000 (21:03 -0800)]
mac80211: return an error when SIWRATE doesn't match any rate

Currently mac80211 fails silently when trying to set a nonexistent
rate.  Return an error instead.

Signed-Off-By: Andy Lutomirski <luto@myrealbox.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
16 years agossb: Fix probing of PCI cores if PCI and PCIE core is available
Michael Buesch [Thu, 3 Jan 2008 17:59:25 +0000 (18:59 +0100)]
ssb: Fix probing of PCI cores if PCI and PCIE core is available

This will make sure that always the correct core is selected, even if
there are both a PCI and PCI-E core on a PCI or PCI-E card.

Signed-off-by: Michael Buesch <mb@bu3sch.de>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
16 years ago[NET]: Do not check netif_running() and carrier state in ->poll()
David S. Miller [Tue, 8 Jan 2008 04:48:21 +0000 (20:48 -0800)]
[NET]: Do not check netif_running() and carrier state in ->poll()

Drivers do this to try to break out of the ->poll()'ing loop
when the device is being brought administratively down.

Now that we have a napi_disable() "pending" state we are going
to solve that problem generically.

Signed-off-by: David S. Miller <davem@davemloft.net>
16 years ago[NET]: Add NAPI_STATE_DISABLE.
David S. Miller [Tue, 8 Jan 2008 04:35:07 +0000 (20:35 -0800)]
[NET]: Add NAPI_STATE_DISABLE.

Create a bit to signal that a napi_disable() is in progress.

This sets up infrastructure such that net_rx_action() can generically
break out of the ->poll() loop on a NAPI context that has a pending
napi_disable() yet is being bombed with packets (and thus would
otherwise poll endlessly and not allow the napi_disable() to finish).

Now, what napi_disable() does is first set the NAPI_STATE_DISABLE bit
(to indicate that a disable is pending), then it polls for the
NAPI_STATE_SCHED bit, and once the NAPI_STATE_SCHED bit is acquired
the NAPI_STATE_DISABLE bit is cleared.  Here, the test_and_set_bit()
provides the necessary memory barrier between the various bitops.

napi_schedule_prep() now tests for a pending disable as it's first
action and won't try to obtain the NAPI_STATE_SCHED bit if a disable
is pending.

As a result, we can remove the netif_running() check in
netif_rx_schedule_prep() because the NAPI disable pending state serves
this purpose.  And, it does so in a NAPI centric manner which is what
we really want.

Signed-off-by: David S. Miller <davem@davemloft.net>
16 years ago[NET]: Do not grab device reference when scheduling a NAPI poll.
David S. Miller [Tue, 8 Jan 2008 04:26:59 +0000 (20:26 -0800)]
[NET]: Do not grab device reference when scheduling a NAPI poll.

It is pointless, because everything that can make a device go away
will do a napi_disable() first.

The main impetus behind this is that now we can legally do a NAPI
completion in generic code like net_rx_action() which a following
changeset needs to do.  net_rx_action() can only perform actions
in NAPI centric ways, because there may be a one to many mapping
between NAPI contexts and network devices (SKY2 is one example).

We also want to get rid of this because it's an extra atomic in the
NAPI paths, and also because it is one of the last instances where the
NAPI interfaces care about net devices.

The one remaining netdev detail the NAPI stuff cares about is the
netif_running() check which will be killed off in a subsequent
changeset.

Signed-off-by: David S. Miller <davem@davemloft.net>
16 years agob43: Fix rxheader channel parsing
Michael Buesch [Wed, 2 Jan 2008 17:55:53 +0000 (18:55 +0100)]
b43: Fix rxheader channel parsing

This patch fixes the parsing of the RX data header channel field.

The current code parses the header incorrectly and passes a wrong
channel number and frequency for each frame to mac80211.
The FIXMEs added by this patch don't matter for now as the code
where they live won't get executed anyway. They will be fixed later.

Signed-off-by: Michael Buesch <mb@bu3sch.de>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
16 years ago[IRDA]: irda_create() nuke user triggable printk
maximilian attems [Mon, 7 Jan 2008 08:30:57 +0000 (00:30 -0800)]
[IRDA]: irda_create() nuke user triggable printk

easy to trigger as user with sfuzz.

irda_create() is quiet on unknown sock->type,
match this behaviour for SOCK_DGRAM unknown protocol

Signed-off-by: maximilian attems <max@stro.at>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years ago[SCTP]: Add back the code that accounted for FORWARD_TSN parameter in INIT.
Vlad Yasevich [Mon, 7 Jan 2008 08:28:16 +0000 (00:28 -0800)]
[SCTP]: Add back the code that accounted for FORWARD_TSN parameter in INIT.

Some recent changes completely removed accounting for the FORWARD_TSN
parameter length in the INIT and INIT-ACK chunk.  This is wrong and
should be restored.

Signed-off-by: Vlad Yasevich <vladislav.yasevich@hp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years ago[SCTP]: Correctly handle AUTH parameters in unexpected INIT
Vlad Yasevich [Mon, 7 Jan 2008 08:27:45 +0000 (00:27 -0800)]
[SCTP]: Correctly handle AUTH parameters in unexpected INIT

When processing an unexpected INIT chunk, we do not need to
do any preservation of the old AUTH parameters.  In fact,
doing such preservations will nullify AUTH and allow connection
stealing.

Signed-off-by: Vlad Yasevich <vladislav.yasevich@hp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years ago[SCTP]: Fix the name of the authentication event.
Vlad Yasevich [Mon, 7 Jan 2008 08:27:16 +0000 (00:27 -0800)]
[SCTP]: Fix the name of the authentication event.

The even should be called SCTP_AUTHENTICATION_INDICATION.

Signed-off-by: Vlad Yasevich <vladislav.yasevich@hp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years ago[ATM]: [nicstar] delay irq setup until card is configured
Chas Williams [Mon, 7 Jan 2008 08:26:22 +0000 (00:26 -0800)]
[ATM]: [nicstar] delay irq setup until card is configured

Signed-off-by: Chas Williams <chas@cmf.nrl.navy.mil>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years ago[TULIP]: NAPI full quantum bug.
Stephen Hemminger [Mon, 7 Jan 2008 08:23:04 +0000 (00:23 -0800)]
[TULIP]: NAPI full quantum bug.

This should fix the kernel warn/oops reported while routing.

The tulip driver has a fencepost bug with new NAPI in 2.6.24
It has an off by one bug if a full quantum is reached.

Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years ago[METH]: Fix MAC address handling.
Thomas Bogendoerfer [Sun, 6 Jan 2008 08:21:47 +0000 (00:21 -0800)]
[METH]: Fix MAC address handling.

meth didn't set a valid mac address during probing, but later during
open. Newer kernel refuse to open device with 00:00:00:00:00:00 as mac
address -> dead ethernet. This patch sets the mac address in the probe
function and uses only the mac address from the netdevice struct when
setting up the hardware.

Signed-off-by: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years ago[NET]: Fix netx-eth.c compilation.
Adrian Bunk [Sun, 6 Jan 2008 07:55:13 +0000 (23:55 -0800)]
[NET]: Fix netx-eth.c compilation.

This was missed when commit e2ac455a18806b31c2d0da0a51d8740af5010b7a
fixed the compile errors in drivers/net/netx-eth.c caused by
commit 09f75cd7bf13720738e6a196cc0107ce9a5bd5a0.

Signed-off-by: Adrian Bunk <adrian.bunk@movial.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years ago[IPV4] ipconfig: Fix regression in ip command line processing
Amos Waterland [Sun, 6 Jan 2008 07:23:06 +0000 (23:23 -0800)]
[IPV4] ipconfig: Fix regression in ip command line processing

The recent changes for ip command line processing fixed some problems
but unfortunately broke some common usage scenarios.  In current
2.6.24-rc6 the following command line results in no IP address
assignment, which is surely a regression:

 ip=10.0.2.15::10.0.2.2:255.255.255.0::eth0:off

Please find below a patch that works for all cases I can find.

Signed-off-by: Amos Waterland <apw@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years ago[IPV4] raw: Strengthen check on validity of iph->ihl
Herbert Xu [Sun, 6 Jan 2008 07:14:44 +0000 (23:14 -0800)]
[IPV4] raw: Strengthen check on validity of iph->ihl

We currently check that iph->ihl is bounded by the real length and that
the real length is greater than the minimum IP header length.  However,
we did not check the caes where iph->ihl is less than the minimum IP
header length.

This breaks because some ip_fast_csum implementations assume that which
is quite reasonable.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years ago[NIU]: Update driver version and release date.
David S. Miller [Sat, 5 Jan 2008 08:02:59 +0000 (00:02 -0800)]
[NIU]: Update driver version and release date.

Signed-off-by: David S. Miller <davem@davemloft.net>
16 years ago[NIU]: Fix potentially stuck TCP socket send queues.
David S. Miller [Sat, 5 Jan 2008 07:54:06 +0000 (23:54 -0800)]
[NIU]: Fix potentially stuck TCP socket send queues.

It is possible for the TX ring to have packets sit in it for unbounded
amounts of time.

The only way to defer TX interrupts in the chip is to periodically set
"mark" bits, when processing of a TX descriptor with the mark bit set
is complete it triggers the interrupt for the TX queue's LDG.

A consequence of this kind of scheme is that if packet flow suddenly
stops, the remaining TX packets will just sit there.

If this happens, since those packets could be charged to TCP socket
send queues, such sockets could get stuck.

The simplest solution is to divorce the socket ownership of the packet
once the device takes the SKB, by using skb_orphan() in
niu_start_xmit().

In hindsight, it would have been much nicer if the chip provided two
interrupt sources for TX (like basically every other ethernet chip
does).  Namely, keep the "mark" bit, but also signal the LDG when the
TX queue becomes completely empty.  That way there is no need to have
a deadlock breaker like this.

Signed-off-by: David S. Miller <davem@davemloft.net>
16 years ago[NIU]: Missing ->last_rx update.
David S. Miller [Sat, 5 Jan 2008 07:52:06 +0000 (23:52 -0800)]
[NIU]: Missing ->last_rx update.

Noticed by Paul Lodridge.

Signed-off-by: David S. Miller <davem@davemloft.net>
16 years ago[NIU]: Fix slowpath interrupt handling.
Matheos Worku [Sat, 5 Jan 2008 07:48:26 +0000 (23:48 -0800)]
[NIU]: Fix slowpath interrupt handling.

niu_slowpath_interrupt() expects values to be setup in lp->{v0,v1,v2}
but they aren't.  That's only done by niu_schedule_napi() which is
done later in the interrupt path.

If niu_rx_error() returns zero, and v0 is clear, hit the
RX_DMA_CTL_STATE register with a RX_DMA_CTL_STAT_MEX.

Only emit verbose RX error logs if a fatal channel or port error is
signalled.  Other cases will be recorded into statistics by
niu_log_rxchan_errors().

Signed-off-by: David S. Miller <davem@davemloft.net>
16 years agofutex: Prevent stale futex owner when interrupted/timeout
Thomas Gleixner [Tue, 8 Jan 2008 18:47:38 +0000 (19:47 +0100)]
futex: Prevent stale futex owner when interrupted/timeout

Roland Westrelin did a great analysis of a long standing thinko in the
return path of futex_lock_pi.

While we fixed the lock steal case long ago, which was easy to trigger,
we never had a test case which exposed this problem and stupidly never
thought about the reverse lock stealing scenario and the return to user
space with a stale state.

When a blocked tasks returns from rt_mutex_timed_locked without holding
the rt_mutex (due to a signal or timeout) and at the same time the task
holding the futex is releasing the futex and assigning the ownership of
the futex to the returning task, then it might happen that a third task
acquires the rt_mutex before the final rt_mutex_trylock() of the
returning task happens under the futex hash bucket lock. The returning
task returns to user space with ETIMEOUT or EINTR, but the user space
futex value is assigned to this task. The task which acquired the
rt_mutex fixes the user space futex value right after the hash bucket
lock has been released by the returning task, but for a short period of
time the user space value is wrong.

Detailed description is available at:

   https://bugzilla.redhat.com/show_bug.cgi?id=400541

The fix for this is the same as we do when the rt_mutex was acquired by
a higher priority task via lock stealing from the designated new owner.
In that case we already fix the user space value and the internal
pi_state up before we return. This mechanism can be used to fixup the
above corner case as well. When the returning task, which failed to
acquire the rt_mutex, notices that it is the designated owner of the
futex, then it fixes up the stale user space value and the pi_state,
before returning to user space. This happens with the futex hash bucket
lock held, so the task which acquired the rt_mutex is guaranteed to be
blocked on the hash bucket lock. We can access the rt_mutex owner, which
gives us the pid of the new owner, safely here as the owner is not able
to modify (release) it while waiting on the hash bucket lock.

Rename the "curr" argument of fixup_pi_state_owner() to "newowner" to
avoid confusion with current and add the check for the stale state into
the failure path of rt_mutex_trylock() in the return path of
unlock_futex_pi(). If the situation is detected use
fixup_pi_state_owner() to assign everything to the owner of the
rt_mutex.

Pointed-out-and-tested-by: Roland Westrelin <roland.westrelin@sun.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agopl2303: Fix mode switching regression
Alan Cox [Tue, 8 Jan 2008 14:55:51 +0000 (14:55 +0000)]
pl2303: Fix mode switching regression

Cleaning out all the incorrect 'no change made' checks for termios
settings showed up a problem with the PL2303. The hardware here seems to
lose sync and bits if you tell it to make no changes. This shows up with
a real world application.

To fix this the driver check for meaningful hardware changes is restored
but doing the tests correctly and as a tty layer function so it doesn't
get duplicated wrongly everywhere if other drivers turn out to need it.

Signed-off-by: Alan Cox <alan@redhat.com>
Tested-by: Mirko Parthey <mirko.parthey@informatik.tu-chemnitz.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agohfs: handle more on-disk corruptions without oopsing
Eric Sandeen [Tue, 8 Jan 2008 23:33:20 +0000 (15:33 -0800)]
hfs: handle more on-disk corruptions without oopsing

hfs seems prone to bad things when it encounters on disk corruption.  Many
values are read from disk, and used as lengths to memcpy, as an example.
This patch fixes up several of these problematic cases.

o sanity check the on-disk maximum key lengths on mount
  (these are set to a defined value at mkfs time and shouldn't differ)
o check on-disk node keylens against the maximum key length for each tree
o fix hfs_btree_open so that going out via free_tree: doesn't wind
  up in hfs_releasepage, which wants to follow the very pointer
  we were trying to set up:
HFS_SB(sb)->cat_tree = hfs_btree_open()
...
failure gets to hfs_releasepage and tries
to follow HFS_SB(sb)->cat_tree

Tested with the fsfuzzer; it survives more than it used to.

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Cc: Roman Zippel <zippel@linux-m68k.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoFix crash with FLAT_MEMORY and ARCH_PFN_OFFSET != 0
Thomas Bogendoerfer [Tue, 8 Jan 2008 23:33:11 +0000 (15:33 -0800)]
Fix crash with FLAT_MEMORY and ARCH_PFN_OFFSET != 0

When using FLAT_MEMORY and ARCH_PFN_OFFSET is not 0, the kernel crashes in
memmap_init_zone().  This bug got introduced by commit
c713216deebd95d2b0ab38fef8bb2361c0180c2d

Signed-off-by: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Acked-by: Mel Gorman <mel@csn.ul.ie>
Cc: Bob Picco <bob.picco@hp.com>
Cc: Dave Hansen <haveblue@us.ibm.com>
Cc: Andy Whitcroft <apw@shadowen.org>
Cc: Andi Kleen <ak@muc.de>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: "Keith Mannthey" <kmannth@gmail.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Yasunori Goto <y-goto@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agosnd_mixer_oss_build_input(): fix for __you_cannot_kmalloc_that_much failure with...
Jean Delvare [Tue, 8 Jan 2008 23:33:10 +0000 (15:33 -0800)]
snd_mixer_oss_build_input(): fix for __you_cannot_kmalloc_that_much failure with gcc-3.2

Rework this functions so that gcc-3.2 can successfully perform
constant-folding.

Signed-off-by: Jean Delvare <khali@linux-fr.org>
Cc: Jaroslav Kysela <perex@suse.cz>
Cc: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agodmi-id: fix for __you_cannot_kmalloc_that_much failure
Jean Delvare [Tue, 8 Jan 2008 23:33:08 +0000 (15:33 -0800)]
dmi-id: fix for __you_cannot_kmalloc_that_much failure

gcc 3.2 has a hard time coping with the code in dmi_id_init():

drivers/built-in.o(.init.text+0x789e): In function `dmi_id_init':
: undefined reference to `__you_cannot_kmalloc_that_much'
make: *** [.tmp_vmlinux1] Error 1

Moving half of the code to a separate function seems to help.  This is a
no-op for gcc 4.1 which will successfully inline the code anyway.

Signed-off-by: Jean Delvare <khali@linux-fr.org>
Cc: Dave Airlie <airlied@linux.ie>
Tested-by: Kamalesh Babulal <kamalesh@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agovmcoreinfo: add the array length of "free_list" for filtering free pages
Ken'ichi Ohmichi [Tue, 8 Jan 2008 23:33:05 +0000 (15:33 -0800)]
vmcoreinfo: add the array length of "free_list" for filtering free pages

This patch adds the array length of "free_area.free_list" to the vmcoreinfo
data so that makedumpfile (dump filtering command) can exclude all free pages
in linux-2.6.24.

makedumpfile creates a small dumpfile by excluding unnecessary pages for the
analysis. To distinguish unnecessary pages, makedumpfile gets the vmcoreinfo
data which has the minimum debugging information only for dump filtering.

In 2.6.24-rc1 or later, the free_area.free_list is an array which has one list
for each migrate types instead of a single list. makedumpfile needs the array
length of "free_area.free_list" and the vmcoreinfo data should contain it.

Signed-off-by: Huang Ying <ying.huang@intel.com>
Tested-by: Ken'ichi Ohmichi <oomichi@mxs.nes.nec.co.jp>
Acked-by: Simon Horman <horms@verge.net.au>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoeCryptfs: fix dentry handling on create error, unlink, and inode destroy
Michael Halcrow [Tue, 8 Jan 2008 23:33:02 +0000 (15:33 -0800)]
eCryptfs: fix dentry handling on create error, unlink, and inode destroy

This patch corrects some erroneous dentry handling in eCryptfs.

If there is a problem creating the lower file, then there is nothing that
the persistent lower file can do to really help us.  This patch makes a
vfs_create() failure in the lower filesystem always lead to an
unconditional do_create failure in eCryptfs.

Under certain sequences of operations, the eCryptfs dentry can remain in
the dcache after an unlink.  This patch calls d_drop() on the eCryptfs
dentry to correct this.

eCryptfs has no business calling d_delete() directly on a lower
filesystem's dentry.  This patch removes the call to d_delete() on the
lower persistent file's dentry in ecryptfs_destroy_inode().

(Thanks to David Kleikamp, Eric Sandeen, and Jeff Moyer for helping
identify and resolve this issue)

Signed-off-by: Michael Halcrow <mhalcrow@us.ibm.com>
Cc: Dave Kleikamp <shaggy@austin.ibm.com>
Cc: Eric Sandeen <sandeen@redhat.com>
Cc: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoxip: fix get_zeroed_page with __GFP_HIGHMEM
Akinobu Mita [Tue, 8 Jan 2008 23:32:57 +0000 (15:32 -0800)]
xip: fix get_zeroed_page with __GFP_HIGHMEM

The use of get_zeroed_page() with __GFP_HIGHMEM is invalid.  Use
alloc_page() with __GFP_ZERO instead of invalid get_zeroed_page().

(This patch is only compile tested)

Cc: Carsten Otte <cotte@de.ibm.com>
Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Acked-by: Hugh Dickins <hugh@veritas.com>
Acked-by: Carsten Otte <cotte@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agomd: fix data corruption when a degraded raid5 array is reshaped
Dan Williams [Tue, 8 Jan 2008 23:32:53 +0000 (15:32 -0800)]
md: fix data corruption when a degraded raid5 array is reshaped

We currently do not wait for the block from the missing device to be
computed from parity before copying data to the new stripe layout.

The change in the raid6 code is not techincally needed as we don't delay
data block recovery in the same way for raid6 yet.  But making the change
now is safer long-term.

This bug exists in 2.6.23 and 2.6.24-rc

Cc: <stable@kernel.org>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Acked-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoKEYS: fix macro
Sebastian Siewior [Tue, 8 Jan 2008 23:32:47 +0000 (15:32 -0800)]
KEYS: fix macro

Commit 664cceb0093b755739e56572b836a99104ee8a75 changed the parameters of
the function make_key_ref().  The macros that are used in case CONFIG_KEY
is not defined did not change.

Cc: David Howells <dhowells@redhat.com>
Signed-off-by: Sebastian Siewior <sebastian@breakpoint.cc>
Acked-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agofat: optimize fat_count_free_clusters()
OGAWA Hirofumi [Tue, 8 Jan 2008 23:32:41 +0000 (15:32 -0800)]
fat: optimize fat_count_free_clusters()

On large partition, scanning the free clusters is very slow if users
doesn't use "usefree" option.

For optimizing it, this patch uses sb_breadahead() to read of FAT
sectors. On some user's 15GB partition, this patch improved it very
much (1min => 600ms).

The following is the result of 2GB partition on my machine.

without patch:
root@devron (/)# time df -h > /dev/null

real    0m1.202s
user    0m0.000s
sys     0m0.440s

with patch:
root@devron (/)# time df -h > /dev/null

real    0m0.378s
user    0m0.012s
sys     0m0.168s

Signed-off-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agospi_bitbang: always grab lock with irqs blocked
David Brownell [Tue, 8 Jan 2008 23:32:40 +0000 (15:32 -0800)]
spi_bitbang: always grab lock with irqs blocked

Fix a glitch reported by lockdep in the spi_bitbang code: it needs to
consistently block IRQs when holding that spinlock.

Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agox86: fix do_fork_idle section mismatch
Thomas Gleixner [Tue, 8 Jan 2008 23:18:28 +0000 (00:18 +0100)]
x86: fix do_fork_idle section mismatch

With CPU_HOTPLUG=n:

WARNING: vmlinux.o(.text+0x104f8): Section mismatch: reference to .init.text:fork_idle (between
'do_fork_idle' and 'lapic_timer_broadcast')

do_fork_idle() needs to be __cpuinit. It can be static as well.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
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 [Tue, 8 Jan 2008 20:29:52 +0000 (12:29 -0800)]
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/srp: Release transport before removing host
  IB/mlx4: Fix value of pkey_index in QP1 completions
  MAINTAINERS: Update Sean Hefty's email address

16 years agoIB/srp: Release transport before removing host
Dave Dillow [Fri, 4 Jan 2008 03:35:41 +0000 (22:35 -0500)]
IB/srp: Release transport before removing host

The documented call sequence for removing a host is to call the
transport xxx_remove_host() prior to scsi_remove_host(). The SRP
transport used to crash when that order was followed, but as it is now
fixed, use the documented order.

Signed-off-by: David Dillow <dillowda@ornl.gov>
Acked-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoIB/mlx4: Fix value of pkey_index in QP1 completions
Dotan Barak [Mon, 7 Jan 2008 07:01:25 +0000 (09:01 +0200)]
IB/mlx4: Fix value of pkey_index in QP1 completions

Fix the value of pkey_index in completions to get a valid value for
GSI QPs.  Without this fix, incoming GSI packets on port 2 get an
invalid P_Key index in the completion, which prevents the MAD layer
from sending back a response, which can make the second port of
ConnectX HCAs completely useless.

Signed-off-by: Dotan Barak <dotanb@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoRevert "hda_intel suspend latency: shorten codec read"
Linus Torvalds [Tue, 8 Jan 2008 19:46:37 +0000 (11:46 -0800)]
Revert "hda_intel suspend latency: shorten codec read"

This reverts commit 57a04513cb35086d54bcb2cb92e6627fc8fa0fae.

Harald Dunkel reports that it broke sound for him:
  "Alsa stopped working for me.  I still can access /dev/dsp, change the
   volume and so on, but the speakers are quiet."

Reverting it fixed things for him.

Reported-and-tested-by: Harald Dunkel <harald.dunkel@t-online.de>
Acked-by: Takashi Iwai <tiwai@suse.de>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoMAINTAINERS: Update Sean Hefty's email address
Sean Hefty [Wed, 2 Jan 2008 20:00:24 +0000 (12:00 -0800)]
MAINTAINERS: Update Sean Hefty's email address

My Unix email account is being discontinued at end of Q1 '08.

Signed-off-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
16 years agoacct: real_parent ppid
Roland McGrath [Mon, 7 Jan 2008 22:23:34 +0000 (14:23 -0800)]
acct: real_parent ppid

The ac_ppid field reported in process accounting records
should match what getppid() would have returned to that
process, regardless of whether a debugger is attached.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agocore dump: real_parent ppid
Roland McGrath [Mon, 7 Jan 2008 22:22:44 +0000 (14:22 -0800)]
core dump: real_parent ppid

The pr_ppid field reported in core dumps should match what
getppid() would have returned to that process, regardless of
whether a debugger is attached.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoMerge branch 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus
Linus Torvalds [Mon, 7 Jan 2008 21:31:03 +0000 (13:31 -0800)]
Merge branch 'upstream' of git://ftp.linux-mips.org/upstream-linus

* 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus:
  [MIPS] Fix CONFIG_BOOT_RAW.
  [MIPS] Assume R4000/R4400 newer than 3.0 don't have the mfc0 count bug
  [MIPS] Fix IP32 breakage
  [MIPS] Alchemy: Fix use of __init code bug exposed by modpost warning
  [MIPS] Move inclusing of kernel/time/Kconfig menu to appropriate place

16 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/mchehab/v4l-dvb
Linus Torvalds [Mon, 7 Jan 2008 21:29:39 +0000 (13:29 -0800)]
Merge git://git./linux/kernel/git/mchehab/v4l-dvb

* git://git.kernel.org/pub/scm/linux/kernel/git/mchehab/v4l-dvb:
  V4L/DVB (6916): ivtv: udelay has to be changed *after* the eeprom was read, not before
  V4L/DVB (6944a): Fix Regression VIDIOCGMBUF ioctl hangs on bttv driver

16 years ago[MIPS] Fix CONFIG_BOOT_RAW.
Ralf Baechle [Mon, 7 Jan 2008 15:09:50 +0000 (15:09 +0000)]
[MIPS] Fix CONFIG_BOOT_RAW.

This was broken by 017e3a492683b32d17dcd1b13b279745cc656073 (lmo) /
396a2ae08e5080b140330645743ab2567f6bc426 (kernel.org).

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
16 years ago[MIPS] Assume R4000/R4400 newer than 3.0 don't have the mfc0 count bug
Thomas Bogendoerfer [Fri, 4 Jan 2008 22:38:31 +0000 (23:38 +0100)]
[MIPS] Assume R4000/R4400 newer than 3.0 don't have the mfc0 count bug

This seems as reasonable assumption and gets some SNI machines to work
which currently must rely on the cp0 counter as clocksource.

Signed-off-by: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
16 years ago[MIPS] Fix IP32 breakage
Thomas Bogendoerfer [Sat, 5 Jan 2008 11:13:11 +0000 (12:13 +0100)]
[MIPS] Fix IP32 breakage

- suppress master aborts during config read
- set io_map_base
- only fixup end of iomem resource to avoid failing request_resource
  in serial driver
- killed useless setting of crime_int bit, which caused wrong interrupts
- use physcial address for serial port platform device and let 8250
  driver do the ioremap

Signed-off-by: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
16 years ago[MIPS] Alchemy: Fix use of __init code bug exposed by modpost warning
Sergei Shtylyov [Tue, 25 Dec 2007 18:00:45 +0000 (21:00 +0300)]
[MIPS] Alchemy: Fix use of __init code bug exposed by modpost warning

WARNING: vmlinux.o(.text+0x1ca608): Section mismatch: reference to
.init.text: add_wired_entry (between 'config_access' and 'config_read')

by refactoring the code calling add_wired_entry() from config_access() to
a separate function which is called from aau1x_pci_setup(). While at it:

- make some unnecassarily global variables 'static';

- fix the letter case, whitespace, etc. in the comments...

Signed-off-by: Sergei Shtylyov <sshtylyov@ru.mvista.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
16 years ago[MIPS] Move inclusing of kernel/time/Kconfig menu to appropriate place
Atsushi Nemoto [Sun, 11 Nov 2007 16:05:16 +0000 (01:05 +0900)]
[MIPS] Move inclusing of kernel/time/Kconfig menu to appropriate place

CONFIG_NO_HZ, CONFIG_HIGH_RES_TIMERS should be selected in "Kernel
type" menu, not in "CPU selection" menu.

Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
16 years agoV4L/DVB (6916): ivtv: udelay has to be changed *after* the eeprom was read, not before
Hans Verkuil [Mon, 7 Jan 2008 08:46:26 +0000 (06:46 -0200)]
V4L/DVB (6916): ivtv: udelay has to be changed *after* the eeprom was read, not before

The eeprom decides which Hauppauge model it is, so the decision whether to
use an udelay of 5 or 10 needs to be taken after reading the eeprom, not
before.

Signed-off-by: Hans Verkuil <hverkuil@xs4all.nl>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
16 years agoV4L/DVB (6944a): Fix Regression VIDIOCGMBUF ioctl hangs on bttv driver
Gregor Jasny [Sun, 6 Jan 2008 14:15:54 +0000 (11:15 -0300)]
V4L/DVB (6944a): Fix Regression VIDIOCGMBUF ioctl hangs on bttv driver

Fix bttv VIDIOCGMBUF locking like done in commit
820eacd84cff23b76693f4be1e28feb672f4488f.

Signed-off-by: Gregor Jasny <gjasny@web.de>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
16 years agoMerge master.kernel.org:/home/rmk/linux-2.6-arm
Linus Torvalds [Sun, 6 Jan 2008 22:35:19 +0000 (14:35 -0800)]
Merge master.kernel.org:/home/rmk/linux-2.6-arm

* master.kernel.org:/home/rmk/linux-2.6-arm:
  [ARM] 4691/1: add missing i2c_board_info struct for at91rm9200
  [ARM] 4735/1: Unbreak pxa25x suspend/resume

16 years agoLinux 2.6.24-rc7
Linus Torvalds [Sun, 6 Jan 2008 21:45:38 +0000 (13:45 -0800)]
Linux 2.6.24-rc7

16 years agoCPU hotplug: fix cpu_is_offline() on !CONFIG_HOTPLUG_CPU
Ingo Molnar [Sun, 30 Dec 2007 10:58:17 +0000 (11:58 +0100)]
CPU hotplug: fix cpu_is_offline() on !CONFIG_HOTPLUG_CPU

make randconfig bootup testing found that the cpufreq code
crashes on bootup, if the powernow-k8 driver is enabled and
if maxcpus=1 passed on the boot line to a !CONFIG_HOTPLUG_CPU
kernel.

First lockdep found out that there's an inconsistent unlock
sequence:

 =====================================
 [ BUG: bad unlock balance detected! ]
 -------------------------------------
 swapper/1 is trying to release lock (&per_cpu(cpu_policy_rwsem, cpu)) at:
 [<ffffffff806ffd8e>] unlock_policy_rwsem_write+0x3c/0x42
 but there are no more locks to release!

Call Trace:
 [<ffffffff806ffd8e>] unlock_policy_rwsem_write+0x3c/0x42
 [<ffffffff80251c29>] print_unlock_inbalance_bug+0x104/0x12c
 [<ffffffff80252f3a>] mark_held_locks+0x56/0x94
 [<ffffffff806ffd8e>] unlock_policy_rwsem_write+0x3c/0x42
 [<ffffffff807008b6>] cpufreq_add_dev+0x2a8/0x5c4
 ...

then shortly afterwards the cpufreq code crashed on an assert:

 ------------[ cut here ]------------
 kernel BUG at drivers/cpufreq/cpufreq.c:1068!
 invalid opcode: 0000 [1] SMP
 [...]
 Call Trace:
  [<ffffffff805145d6>] sysdev_driver_unregister+0x5b/0x91
  [<ffffffff806ff520>] cpufreq_register_driver+0x15d/0x1a2
  [<ffffffff80cc0596>] powernowk8_init+0x86/0x94
 [...]
 ---[ end trace 1e9219be2b4431de ]---

the bug was caused by maxcpus=1 bootup, which brought up the
secondary core as !cpu_online() but !cpu_is_offline() either,
which on on !CONFIG_HOTPLUG_CPU is always 0 (include/linux/cpu.h):

  /* CPUs don't go offline once they're online w/o CONFIG_HOTPLUG_CPU */
  static inline int cpu_is_offline(int cpu) { return 0; }

but the cpufreq code uses cpu_online() and cpu_is_offline() in
a mixed way - the low-level drivers use cpu_online(), while
the cpufreq core uses cpu_is_offline(). This opened up the
possibility to add the non-initialized sysdev device of the
secondary core:

 cpufreq-core: trying to register driver powernow-k8
 cpufreq-core: adding CPU 0
 powernow-k8: BIOS error - no PSB or ACPI _PSS objects
 cpufreq-core: initialization failed
 cpufreq-core: adding CPU 1
 cpufreq-core: initialization failed

which then blew up. The fix is to make cpu_is_offline() always
the negation of cpu_online(). With that fix applied the kernel
boots up fine without crashing:

 Calling initcall 0xffffffff80cc0510: powernowk8_init+0x0/0x94()
 powernow-k8: Found 1 AMD Athlon(tm) 64 X2 Dual Core Processor 3800+ processors (1 cpu cores) (version 2.20.00)
 powernow-k8: BIOS error - no PSB or ACPI _PSS objects
 initcall 0xffffffff80cc0510: powernowk8_init+0x0/0x94() returned -19.
 initcall 0xffffffff80cc0510 ran for 19 msecs: powernowk8_init+0x0/0x94()
 Calling initcall 0xffffffff80cc328f: init_lapic_nmi_sysfs+0x0/0x39()

We could fix this by making CPU enumeration aware of max_cpus, but that
would be more fragile IMO, and the cpu_online(cpu) != cpu_is_offline(cpu)
possibility was quite confusing and a continuous source of bugs too.

Most distributions have kernels with CPU hotplug enabled, so this bug
remained hidden for a long time.

Bug forensics:

The broken cpu_is_offline() API variant was introduced via:

 commit a59d2e4e6977e7b94e003c96a41f07e96cddc340
 Author: Rusty Russell <rusty@rustcorp.com.au>
 Date:   Mon Mar 8 06:06:03 2004 -0800

     [PATCH] minor cleanups for hotplug CPUs

( this predates linux-2.6.git, this commit is available from Thomas's
  historic git tree. )

Then 1.5 years later the cpufreq code made use of it:

 commit c32b6b8e524d2c337767d312814484d9289550cf
 Author: Ashok Raj <ashok.raj@intel.com>
 Date:   Sun Oct 30 14:59:54 2005 -0800

     [PATCH] create and destroy cpufreq sysfs entries based on cpu notifiers

 +       if (cpu_is_offline(cpu))
 +               return 0;

which is a correct use of the subtly broken new API. v2.6.15 then
shipped with this bug included.

then it took two more years for random-kernel qa to hit it.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agohda_intel suspend latency: shorten codec read
Ingo Molnar [Sun, 6 Jan 2008 20:09:53 +0000 (21:09 +0100)]
hda_intel suspend latency: shorten codec read

not sleeping for every codec read/write but doing a short udelay and
a conditional reschedule has cut suspend+resume latency by about 1
second on my T60.

The patch also fixes the unexpected codec-connection errors that
happen more often in the new power-save mode:
    http://lkml.org/lkml/2007/11/8/255
    http://bugzilla.kernel.org/show_bug.cgi?id=9332

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Acked-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years ago[ARM] 4691/1: add missing i2c_board_info struct for at91rm9200
Jan Altenberg [Mon, 10 Dec 2007 09:06:40 +0000 (10:06 +0100)]
[ARM] 4691/1: add missing i2c_board_info struct for at91rm9200

- Add missing i2c_board_info struct for at91rm9200

Signed-off-by: Jan Altenberg <jan.altenberg@linutronix.de>
Acked-by: Andrew Victor <avictor.za@gmail.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
16 years agofix: using joysticks in 32 bit applications on 64 bit systems
Akos Maroy [Sun, 6 Jan 2008 10:15:55 +0000 (11:15 +0100)]
fix: using joysticks in 32 bit applications on 64 bit systems

unfortunately 32 bit apps don't see the joysticks on a 64 bit system.
this prevents one playing X-Plane (http://www.x-plane.com/) or other
32-bit games with joysticks.

this is a known issue, and already raised several times:

 http://readlist.com/lists/vger.kernel.org/linux-kernel/28/144411.html

 http://www.brettcsmith.org/wiki/wiki.cgi?action=browse&diff=1&id=OzyComputer/Joystick

unfortunately this is still not fixed in the mainline kernel.

it would be nice to have this fixed, so that people can play these games
without having to patch their kernel.

the following patch solves the problem on 2.6.22.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoRevert "scsi: revert "[SCSI] Get rid of scsi_cmnd->done""
Linus Torvalds [Sun, 6 Jan 2008 18:17:12 +0000 (10:17 -0800)]
Revert "scsi: revert "[SCSI] Get rid of scsi_cmnd->done""

This reverts commit ac40532ef0b8649e6f7f83859ea0de1c4ed08a19, which gets
us back the original cleanup of 6f5391c283d7fdcf24bf40786ea79061919d1e1d.

It turns out that the bug that was triggered by that commit was
apparently not actually triggered by that commit at all, and just the
testing conditions had changed enough to make it appear to be due to it.

The real problem seems to have been found by Peter Osterlund:

  "pktcdvd sets it [block device size] when opening the /dev/pktcdvd
   device, but when the drive is later opened as /dev/scd0, there is
   nothing that sets it back.  (Btw, 40944 is possible if the disk is a
   CDRW that was formatted with "cdrwtool -m 10236".)

   The problem is that pktcdvd opens the cd device in non-blocking mode
   when pktsetup is run, and doesn't close it again until pktsetup -d is
   run.  The effect is that if you meanwhile open the cd device,
   blkdev.c:do_open() doesn't call bd_set_size() because
   bdev->bd_openers is non-zero."

In particular, to repeat the bug (regardless of whether commit
6f5391c283d7fdcf24bf40786ea79061919d1e1d is applied or not):

  " 1. Start with an empty drive.
    2. pktsetup 0 /dev/scd0
    3. Insert a CD containing an isofs filesystem.
    4. mount /dev/pktcdvd/0 /mnt/tmp
    5. umount /mnt/tmp
    6. Press the eject button.
    7. Insert a DVD containing a non-writable filesystem.
    8. mount /dev/scd0 /mnt/tmp
    9. find /mnt/tmp -type f -print0 | xargs -0 sha1sum >/dev/null
    10. If the DVD contains data beyond the physical size of a CD, you
        get I/O errors in the terminal, and dmesg reports lots of
        "attempt to access beyond end of device" errors."

which in turn is because the nested open after the media change won't
cause the size to be set properly (because the original open still holds
the block device, and we only do the bd_set_size() when we don't have
other people holding the device open).

The proper fix for that is probably to just do something like

bdev->bd_inode->i_size = (loff_t)get_capacity(disk)<<9;

in fs/block_dev.c:do_open() even for the cases where we're not the
original opener (but *not* call bd_set_size(), since that will also
change the block size of the device).

Cc: Peter Osterlund <petero2@telia.com>
Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
Cc: Matthew Wilcox <matthew@wil.cx>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years ago[SCSI] SRP transport: only remove our own entries
Dave Dillow [Fri, 4 Jan 2008 02:34:49 +0000 (21:34 -0500)]
[SCSI] SRP transport: only remove our own entries

The SCSI SRP transport class currently iterates over all children
devices of the host that is being removed in srp_remove_host(). However,
not all of those children were created by the SRP transport, and
removing them will cause corruption and an oops when their creator tries
to remove them.

Signed-off-by: David Dillow <dillowda@ornl.gov>
Acked-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>