14 years agoMerge branch 'linus' into core/futexes
Thomas Gleixner [Mon, 24 Nov 2008 18:54:37 +0000 (19:54 +0100)]
Merge branch 'linus' into core/futexes

14 years agoLinux 2.6.28-rc6 v2.6.28-rc6
Linus Torvalds [Thu, 20 Nov 2008 23:19:22 +0000 (15:19 -0800)]
Linux 2.6.28-rc6

14 years agoMerge branch 'release' of git://
Linus Torvalds [Thu, 20 Nov 2008 23:07:40 +0000 (15:07 -0800)]
Merge branch 'release' of git://git./linux/kernel/git/aegl/linux-2.6

* 'release' of git://
  [IA64] xen: fix xen_get_eflags.
  [IA64] ia64/pv_ops/pv_cpu_ops: fix _IA64_REG_IP case.
  [IA64] remove duplicate include iommu.h
  [IA64] use mprintk instead of printk, in ia64_mca_modify_original_stack
  [IA64] Rationalize kernel mode alignment checking

14 years agoMerge git://
Linus Torvalds [Thu, 20 Nov 2008 21:53:21 +0000 (13:53 -0800)]
Merge git://git./linux/kernel/git/gregkh/usb-2.6

* git://
  USB: ACE1001 patch for cp2101.c
  USB: usbmon: fix read(2)
  USB: gadget rndis: send notifications
  USB: gadget rndis: stop windows self-immolation
  USB: storage: update unusual_devs entries for Nokia 5300 and 5310
  USB: storage: updates unusual_devs entry for the Nokia 6300
  usb: musb: fix bug in musb_schedule
  USB: fix SB700 usb subsystem hang bug

14 years ago[IA64] xen: fix xen_get_eflags.
Isaku Yamahata [Tue, 18 Nov 2008 10:20:51 +0000 (19:20 +0900)]
[IA64] xen: fix xen_get_eflags.

fix xen_get_eflags. It doesn't take any argument.

Signed-off-by: Isaku Yamahata <>
Signed-off-by: Tony Luck <>
14 years ago[IA64] ia64/pv_ops/pv_cpu_ops: fix _IA64_REG_IP case.
Isaku Yamahata [Tue, 18 Nov 2008 10:19:50 +0000 (19:19 +0900)]
[IA64] ia64/pv_ops/pv_cpu_ops: fix _IA64_REG_IP case.

pv_cpu_ops.getreg(_IA64_REG_IP) returned constant.
But the returned ip valued should be the one in the caller, not of the callee.
This patch fixes that.

Signed-off-by: Isaku Yamahata <>
Signed-off-by: Tony Luck <>
14 years ago[IA64] remove duplicate include iommu.h
Huang Weiyi [Thu, 20 Nov 2008 21:38:16 +0000 (13:38 -0800)]
[IA64] remove duplicate include iommu.h

arch/ia64/kernel/pci-dma.c only needs to include iommu once.

Signed-off-by: Huang Weiyi <>
Signed-off-by: Tony Luck <>
14 years ago[IA64] use mprintk instead of printk, in ia64_mca_modify_original_stack
Hidetoshi Seto [Mon, 17 Nov 2008 01:18:08 +0000 (10:18 +0900)]
[IA64] use mprintk instead of printk, in ia64_mca_modify_original_stack

Using printk from MCA/INIT context is unsafe since it can cause deadlock.
The ia64_mca_modify_original_stack is called from both of mca handler and
init handler, so it should use mprintk instead of printk.

Signed-off-by: Hidetoshi Seto <>
Signed-off-by: Tony Luck <>
14 years ago[IA64] Rationalize kernel mode alignment checking
Tony Luck [Thu, 20 Nov 2008 21:27:12 +0000 (13:27 -0800)]
[IA64] Rationalize kernel mode alignment checking

Itanium processors can handle some misaligned data accesses. They
also provide a mode where all such accesses are forced to trap. The
kernel was schizophrenic about use of this mode:

* Base kernel code ran in permissive mode where the only traps
  generated were from those cases that the h/w could not handle.
* Interrupt, syscall and trap code ran in strict mode where all
  unaligned accesses caused traps to the 0x5a00 unaligned reference

Use strict alignment checking throughout the kernel, but make
sure that we continue to let user mode use more relaxed mode
as the default.

Signed-off-by: Tony Luck <>
14 years agox86: Fix interrupt leak due to migration
Matthew Wilcox [Thu, 20 Nov 2008 21:09:33 +0000 (14:09 -0700)]
x86: Fix interrupt leak due to migration

When we migrate an interrupt from one CPU to another, we set the
move_in_progress flag and clean up the vectors later once they're not
being used.  If you're unlucky and call destroy_irq() before the vectors
become un-used, the move_in_progress flag is never cleared, which causes
the interrupt to become unusable.

This was discovered by Jesse Brandeburg for whom it manifested as an
MSI-X device refusing to use MSI-X mode when the driver was unloaded
and reloaded repeatedly.

Signed-off-by: Matthew Wilcox <>
Signed-off-by: Linus Torvalds <>
14 years agoSUNRPC: Fix a performance regression in the RPC authentication code
Trond Myklebust [Thu, 20 Nov 2008 21:06:21 +0000 (16:06 -0500)]
SUNRPC: Fix a performance regression in the RPC authentication code

Fix a regression reported by Max Kellermann whereby kernel profiling
showed that his clients were spending 45% of their time in

It turns out that although his processes had identical uid/gid/groups,
generic_match() was failing to detect this, because the task->group_info
pointers were not shared. This again lead to the creation of a huge number
of identical credentials at the RPC layer.

The regression is fixed by comparing the contents of task->group_info
if the actual pointers are not identical.

Signed-off-by: Trond Myklebust <>
Signed-off-by: Linus Torvalds <>
14 years agoMerge git://
Linus Torvalds [Thu, 20 Nov 2008 21:14:16 +0000 (13:14 -0800)]
Merge git://git./linux/kernel/git/sfrench/cifs-2.6

* git://
  [CIFS] Do not attempt to close invalidated file handles
  [CIFS] fix check for dead tcon in smb_init

14 years agoMerge branch 'upstream' of git://
Linus Torvalds [Thu, 20 Nov 2008 21:13:48 +0000 (13:13 -0800)]
Merge branch 'upstream' of git://

* 'upstream' of git://
  MIPS: csrc-r4k: Fix declaration depending on the wrong CONFIG_ symbol.
  MIPS: csrc-r4k: Fix spelling mistake.
  MIPS: RB532: Provide functions for gpio configuration
  MIPS: IP22: Make indy_sc_ops variable static
  MIPS: RB532: GPIO register offsets are relative to GPIOBASE
  MIPS: Malta: Fix include paths in malta-amon.c

14 years agoMerge branch 'core-fixes-for-linus' of git://
Linus Torvalds [Thu, 20 Nov 2008 21:13:03 +0000 (13:13 -0800)]
Merge branch 'core-fixes-for-linus' of git://git./linux/kernel/git/tip/linux-2.6-tip

* 'core-fixes-for-linus' of git://
  intel-iommu: fix compile warnings

14 years agoMerge git://
Linus Torvalds [Thu, 20 Nov 2008 21:12:14 +0000 (13:12 -0800)]
Merge git://git./linux/kernel/git/davem/net-2.6

* git:// (23 commits)
  net: fix tiny output corruption of /proc/net/snmp6
  atl2: don't request irq on resume if netif running
  ipv6: use seq_release_private for ip6mr.c /proc entries
  pkt_sched: fix missing check for packet overrun in qdisc_dump_stab()
  smc911x: Fix printf format typo in smc911x driver.
  asix: Fix asix-based cards connecting to 10/100Mbs LAN.
  mv643xx_eth: fix recycle check bound
  mv643xx_eth: fix the order of mdiobus_{unregister, free}() calls
  sh: sh_eth: Update to change of mii_bus
  TPROXY: supply a struct flowi->flags argument in inet_sk_rebuild_header()
  TPROXY: fill struct flowi->flags in udp_sendmsg()
  net: ipg.c fix bracing on endian swapping
  phylib: Fix auto-negotiation restart avoidance
  net: jme.c rxdesc.flags is __le16, other missing endian swaps
  phylib: fix phy name example in documentation
  net: Do not fire linkwatch events until the device is registered.
  phonet: fix compilation with gcc-3.4
  ixgbe: fix compilation with gcc-3.4
  pktgen: fix multiple queue warning
  net: fix ip_mr_init() error path

14 years agoMerge branch 'tracing-fixes-for-linus' of git://
Linus Torvalds [Thu, 20 Nov 2008 21:11:21 +0000 (13:11 -0800)]
Merge branch 'tracing-fixes-for-linus' of git://git./linux/kernel/git/tip/linux-2.6-tip

* 'tracing-fixes-for-linus' of git://
  ftrace: fix dyn ftrace filter selection
  ftrace: make filtered functions effective on setting
  ftrace: fix set_ftrace_filter
  trace: introduce missing mutex_unlock()
  tracing: kernel/trace/trace.c: introduce missing kfree()

14 years agoMerge branch 'x86-fixes-for-linus' of git://
Linus Torvalds [Thu, 20 Nov 2008 21:09:32 +0000 (13:09 -0800)]
Merge branch 'x86-fixes-for-linus' of git://git./linux/kernel/git/tip/linux-2.6-tip

* 'x86-fixes-for-linus' of git://
  x86: uaccess_64: fix return value in __copy_from_user()
  x86: quirk for reboot stalls on a Dell Optiplex 330

14 years agoparisc: fix bug in compat_arch_ptrace
Helge Deller [Thu, 20 Nov 2008 09:54:09 +0000 (10:54 +0100)]
parisc: fix bug in compat_arch_ptrace

Commit 81e192d6ce303b6792aa38ff35f41a1a7357f23a ("parisc: convert to
generic compat_sys_ptrace") introduced a bug which segfaults the parisc
64bit kernel when stracing 32bit applications:

  Kernel Fault: Code=15 regs=00000000bafa42b0 (Addr=00000001baf5ab57)
  PSW: 00001000000001101111111100001011 Tainted: G        W
  r00-03  000000ff0806ff0b 000000004068edc0 00000000401203f8 00000000fb3e2508
  r04-07  0000000040686dc0 00000000baf5a800 fffffffffffffffc fffffffffb3e2508
  r08-11  00000000baf5a800 000000000004b068 00000000000402b0 0000000000040d68
  r12-15  0000000000042a9c 0000000000040a9c 0000000000040d60 0000000000042e9c
  r16-19  000000000004b060 000000000004b058 0000000000042d9c ffffffffffffffff
  r20-23  000000000800000b 0000000000000000 000000000800000b fffffffffb3e2508
  r24-27  00000000fffffffc 0000000000000003 00000000fffffffc 0000000040686dc0
  r28-31  00000001baf5a7ff 00000000bafa4280 00000000bafa42b0 00000000000001d7
  sr00-03  0000000000fca000 0000000000000000 0000000000000000 0000000000fca000
  sr04-07  0000000000000000 0000000000000000 0000000000000000 0000000000000000

  IASQ: 0000000000000000 0000000000000000 IAOQ: 0000000040120400 0000000040120404
   IIR: 4b9a06b0    ISR: 0000000000000000  IOR: 00000001baf5ab57
   CPU:        0   CR30: 00000000bafa4000 CR31: 00000000d22344e0
   ORIG_R28: 00000000fb3e2248
   IAOQ[0]: compat_arch_ptrace+0xb8/0x160
   IAOQ[1]: compat_arch_ptrace+0xbc/0x160
   RP(r2): compat_arch_ptrace+0xb0/0x160
   [<00000000401612ac>] compat_sys_ptrace+0x15c/0x180
   [<0000000040104ef8>] syscall_exit+0x0/0x14

The problem is that compat_arch_ptrace() enters with an addr value of
type compat_ulong_t and calls translate_usr_offset() to translate the
address offset into a struct pt_regs offset like this:

addr = translate_usr_offset(addr)

this means that any return value of translate_usr_offset() is stored
back as compat_ulong_t type into the addr variable.

But since translate_usr_offset() returns -1 for invalid offsets, addr
can now get the value 0xffffffff which then fails the next return-value
sanity check and thus the kernel tries to access invalid memory:

if (addr < 0)

Fix this bug by modifying translate_usr_offset() to take and return
values of type compat_ulong_t, and by returning the value
"sizeof(struct pt_regs)" as an error indicator.

Additionally change the sanity check to check for return values
for >= sizeof(struct pt_regs).

This patch survived my compile and run-tests.

Signed-off-by: Helge Deller <>
Signed-off-by: Linus Torvalds <>
14 years ago[CIFS] Do not attempt to close invalidated file handles
Steve French [Thu, 20 Nov 2008 20:00:44 +0000 (20:00 +0000)]
[CIFS] Do not attempt to close invalidated file handles

If a connection with open file handles has gone down
and come back up and reconnected without reopening
the file handle yet, do not attempt to send an SMB close
request for this handle in cifs_close.  We were
checking for the connection being invalid in cifs_close
but since the connection may have been reconnected
we also need to check whether the file handle
was marked invalid (otherwise we could close the
wrong file handle by accident).

Acked-by: Jeff Layton <>
Signed-off-by: Steve French <>
14 years agoMIPS: csrc-r4k: Fix declaration depending on the wrong CONFIG_ symbol.
Ralf Baechle [Mon, 3 Nov 2008 11:32:34 +0000 (11:32 +0000)]
MIPS: csrc-r4k: Fix declaration depending on the wrong CONFIG_ symbol.

Signed-off-by: Ralf Baechle <>
14 years agoMIPS: csrc-r4k: Fix spelling mistake.
Ralf Baechle [Mon, 3 Nov 2008 11:31:54 +0000 (11:31 +0000)]
MIPS: csrc-r4k: Fix spelling mistake.

Signed-off-by: Ralf Baechle <>
14 years agoMIPS: RB532: Provide functions for gpio configuration
Phil Sutter [Sat, 1 Nov 2008 14:13:21 +0000 (15:13 +0100)]
MIPS: RB532: Provide functions for gpio configuration

As gpiolib doesn't support pin multiplexing, it provides no way to
access the GPIOFUNC register. Also there is no support for setting
interrupt status and level. These functions provide access to them and
are needed by the CompactFlash driver.

Signed-off-by: Phil Sutter <>
Signed-off-by: Ralf Baechle <>
14 years agoMIPS: IP22: Make indy_sc_ops variable static
Dmitri Vorobiev [Fri, 31 Oct 2008 17:54:11 +0000 (19:54 +0200)]
MIPS: IP22: Make indy_sc_ops variable static

The indy_sc_ops variable in arch/mips/mm/sc-ip22.c is needlessly defined
global, and this patch makes it static.

Signed-off-by: Dmitri Vorobiev <>
Signed-off-by: Ralf Baechle <>

14 years agoMIPS: RB532: GPIO register offsets are relative to GPIOBASE
Florian Fainelli [Fri, 31 Oct 2008 13:24:29 +0000 (14:24 +0100)]
MIPS: RB532: GPIO register offsets are relative to GPIOBASE

This patch fixes the wrong use of GPIO register offsets
in devices.c. To avoid further problems, use gpio_get_value
to return the NAND status instead of our own expanded code.

Also define the zero offset of the alternate function register to allow
consistent access.

Signed-off-by: Florian Fainelli <>
Signed-off-by: Phil Sutter <>
Signed-off-by: Ralf Baechle <>
14 years agoMIPS: Malta: Fix include paths in malta-amon.c
David Daney [Sat, 20 Sep 2008 17:16:36 +0000 (10:16 -0700)]
MIPS: Malta: Fix include paths in malta-amon.c

On linux-queue, malta doesn't build after the include file relocation.
This should fix it.

There some occurrences of 'asm-mips' in the comments of quite a few
files, but this is the only place I found it in any code.

Signed-off-by: David Daney <>
Signed-off-by: Ralf Baechle <>
14 years agonet: fix tiny output corruption of /proc/net/snmp6
Alexey Dobriyan [Thu, 20 Nov 2008 12:20:10 +0000 (04:20 -0800)]
net: fix tiny output corruption of /proc/net/snmp6

Because "name" is static, it can be occasionally be filled with
somewhat garbage if two processes read /proc/net/snmp6.

Also, remove useless casts and "-1" -- snprintf() correctly terminates it's

Signed-off-by: Alexey Dobriyan <>
Signed-off-by: David S. Miller <>
14 years agoatl2: don't request irq on resume if netif running
Alan Jenkins [Thu, 20 Nov 2008 12:18:25 +0000 (04:18 -0800)]
atl2: don't request irq on resume if netif running

If the device is suspended with the cable disconnected, then
resumed with the cable connected, dev->open is called before
resume. During resume, we request an IRQ, but the IRQ was
already assigned during dev->open, resulting in the warning
shown below.

Don't request an IRQ if the device is running.

Call Trace:
 [<c011b89a>] warn_on_slowpath+0x40/0x59
 [<c023df15>] raw_pci_read+0x4d/0x55
 [<c023dff3>] pci_read+0x1c/0x21
 [<c01bcd81>] __pci_find_next_cap_ttl+0x44/0x70
 [<c01bce86>] __pci_find_next_cap+0x1a/0x1f
 [<c01bcef9>] pci_find_capability+0x28/0x2c
 [<c01c4144>] pci_msi_check_device+0x53/0x62
 [<c01c49c2>] pci_enable_msi+0x3a/0x1cd
 [<e019f17b>] atl2_write_phy_reg+0x40/0x5f [atl2]
 [<c01061b1>] dma_generic_alloc_coherent+0x0/0xd7
 [<e019f107>] atl2_request_irq+0x15/0x49 [atl2]
 [<e01a1481>] atl2_open+0x20b/0x297 [atl2]
 [<c024a35c>] dev_open+0x62/0x91
 [<c0248b9a>] dev_change_flags+0x93/0x141
 [<c024f308>] do_setlink+0x238/0x2d5
 [<c02501b2>] rtnl_setlink+0xa9/0xbf
 [<c0297f0c>] mutex_lock+0xb/0x19
 [<c024ffa7>] rtnl_dump_ifinfo+0x0/0x69
 [<c0250109>] rtnl_setlink+0x0/0xbf
 [<c024fe42>] rtnetlink_rcv_msg+0x185/0x19f
 [<c0240fd1>] sock_rmalloc+0x23/0x57
 [<c024fcbd>] rtnetlink_rcv_msg+0x0/0x19f
 [<c0259457>] netlink_rcv_skb+0x2d/0x71
 [<c024fcb7>] rtnetlink_rcv+0x14/0x1a
 [<c025929e>] netlink_unicast+0x184/0x1e4
 [<c025992a>] netlink_sendmsg+0x233/0x240
 [<c023f405>] sock_sendmsg+0xb7/0xd0
 [<c0129131>] autoremove_wake_function+0x0/0x2b
 [<c0129131>] autoremove_wake_function+0x0/0x2b
 [<c0147796>] mempool_alloc+0x2d/0x9e
 [<c020c923>] scsi_pool_alloc_command+0x35/0x4f
 [<c0297f0c>] mutex_lock+0xb/0x19
 [<c028e867>] unix_stream_recvmsg+0x357/0x3e2
 [<c01b81c9>] copy_from_user+0x23/0x4f
 [<c02452ea>] verify_iovec+0x3e/0x6c
 [<c023f5ab>] sys_sendmsg+0x18d/0x1f0
 [<c023ffa8>] sys_recvmsg+0x146/0x1c8
 [<c0240016>] sys_recvmsg+0x1b4/0x1c8
 [<c0118f48>] __wake_up+0xf/0x15
 [<c02586cd>] netlink_table_ungrab+0x17/0x19
 [<c01b83ba>] copy_to_user+0x25/0x3b
 [<c023fe4a>] move_addr_to_user+0x50/0x68
 [<c0240266>] sys_getsockname+0x6f/0x9a
 [<c0240280>] sys_getsockname+0x89/0x9a
 [<c015046a>] do_wp_page+0x3ae/0x41a
 [<c0151525>] handle_mm_fault+0x4c5/0x540
 [<c02405d0>] sys_socketcall+0x176/0x1b0
 [<c010376d>] sysenter_do_call+0x12/0x21

Signed-off-by: Alan Jenkins <>
Signed-off-by: Jay Cliburn <>
Signed-off-by: David S. Miller <>
14 years agoipv6: use seq_release_private for ip6mr.c /proc entries
Benjamin Thery [Thu, 20 Nov 2008 12:16:12 +0000 (04:16 -0800)]
ipv6: use seq_release_private for ip6mr.c /proc entries

In ip6mr.c, /proc entries /proc/net/ip6_mr_cache and /proc/net/ip6_mr_vif
are opened with seq_open_private(), thus seq_release_private() should be
used to release them.
Should fix a small memory leak.

Signed-off-by: Benjamin Thery <>
Signed-off-by: David S. Miller <>
14 years agopkt_sched: fix missing check for packet overrun in qdisc_dump_stab()
Patrick McHardy [Thu, 20 Nov 2008 12:07:14 +0000 (04:07 -0800)]
pkt_sched: fix missing check for packet overrun in qdisc_dump_stab()

nla_nest_start() might return NULL, causing a NULL pointer dereference.

Signed-off-by: Patrick McHardy <>
Signed-off-by: David S. Miller <>
14 years agoMerge branch 'master' of git://
David S. Miller [Thu, 20 Nov 2008 12:01:29 +0000 (04:01 -0800)]
Merge branch 'master' of git://git./linux/kernel/git/linville/wireless-2.6

14 years agosmc911x: Fix printf format typo in smc911x driver.
Vernon Sauder [Thu, 20 Nov 2008 09:56:08 +0000 (01:56 -0800)]
smc911x: Fix printf format typo in smc911x driver.

Signed-off-by: Vernon Sauder <>
Signed-off-by: David S. Miller <>
14 years agoasix: Fix asix-based cards connecting to 10/100Mbs LAN.
Pantelis Koukousoulas [Thu, 20 Nov 2008 09:48:46 +0000 (01:48 -0800)]
asix: Fix asix-based cards connecting to 10/100Mbs LAN.

Add AX_MEDIUM_ENCK also when speed = 10/100Mbps. This allows my belkin
f5d5055 to work with my 100Mbps switch and with an old 10Mbps ISA card.
Without this patch, the card is recognized and the interface is brought
up fine, but no packets actually flow through the interface.

Signed-off-by: Pantelis Koukousoulas <>
Acked-by: David Hollis <>
Signed-off-by: David S. Miller <>
14 years agomv643xx_eth: fix recycle check bound
Lennert Buytenhek [Tue, 18 Nov 2008 04:28:58 +0000 (04:28 +0000)]
mv643xx_eth: fix recycle check bound

When mv643xx_eth allocates skbuffs, it adds
'dma_get_cache_alignment() - 1' to the length it needs, so that it can
align the skb's ->data pointer to a cache boundary.  When checking
whether a transmitted skbuff can be reused as a receive buffer, these
bytes needs to be included into the minimum bound for the recycle check.

Signed-off-by: Lennert Buytenhek <>
Signed-off-by: David S. Miller <>
14 years agomv643xx_eth: fix the order of mdiobus_{unregister, free}() calls
Lennert Buytenhek [Tue, 18 Nov 2008 04:28:35 +0000 (04:28 +0000)]
mv643xx_eth: fix the order of mdiobus_{unregister, free}() calls

Signed-off-by: Lennert Buytenhek <>
Signed-off-by: David S. Miller <>
14 years agosh: sh_eth: Update to change of mii_bus
Nobuhiro Iwamatsu [Mon, 17 Nov 2008 20:29:58 +0000 (20:29 +0000)]
sh: sh_eth: Update to change of mii_bus

Update to change of mii_bus interface and fix some warning.

Signed-off-by: Nobuhiro Iwamatsu <>
Signed-off-by: David S. Miller <>
14 years agoTPROXY: supply a struct flowi->flags argument in inet_sk_rebuild_header()
Balazs Scheidler [Thu, 20 Nov 2008 09:08:06 +0000 (01:08 -0800)]
TPROXY: supply a struct flowi->flags argument in inet_sk_rebuild_header()

    inet_sk_rebuild_header() does a new route lookup if the dst_entry
    associated with a socket becomes stale. However inet_sk_rebuild_header()
    didn't use struct flowi->flags, causing the route lookup to
    fail for foreign-bound IP_TRANSPARENT sockets, causing an error
    state to be set for the sockets in question.

Signed-off-by: Balazs Scheidler <>
Signed-off-by: David S. Miller <>
14 years agoTPROXY: fill struct flowi->flags in udp_sendmsg()
Balazs Scheidler [Thu, 20 Nov 2008 09:07:24 +0000 (01:07 -0800)]
TPROXY: fill struct flowi->flags in udp_sendmsg()

    udp_sendmsg() didn't fill struct flowi->flags, which means that
    the route lookup would fail for non-local IPs even if the
    IP_TRANSPARENT sockopt was set.

    This prevents sendto() to work properly for UDP sockets, whereas
    bind(foreign-ip) + connect() + send() worked fine.

Signed-off-by: Balazs Scheidler <>
Signed-off-by: David S. Miller <>
14 years agoUSB: ACE1001 patch for cp2101.c
M Kondrin [Fri, 14 Nov 2008 10:02:45 +0000 (13:02 +0300)]
USB: ACE1001 patch for cp2101.c

The patch which adds IDs for AKTAKOM USB->RS232 cable
( is attached.

From: M Kondrin <>
Signed-off-by: Greg Kroah-Hartman <>
14 years agoUSB: usbmon: fix read(2)
Pete Zaitcev [Fri, 14 Nov 2008 16:47:41 +0000 (09:47 -0700)]
USB: usbmon: fix read(2)

There's a bug in the usbmon binary reader: When using read() to fetch
the packets and a packet's data is partially read, the next read call
will once again return up to len_cap bytes of data. The b_read counter
is not regarded when determining the remaining chunk size.

So, when dumping USB data with "cat /dev/usbmon0 > usbmon.trace" while
reading from a USB storage device and analyzing the dump file
afterwards it will get out of sync after a couple of packets.

Signed-off-by: Ingo van Lil <>
Signed-off-by: Pete Zaitcev <>
Cc: stable <>
Signed-off-by: Greg Kroah-Hartman <>
14 years agoUSB: gadget rndis: send notifications
Richard Röjfors [Sun, 16 Nov 2008 03:53:24 +0000 (19:53 -0800)]
USB: gadget rndis: send notifications

It turns out that atomic_inc_return() returns the *new* value
not the original one, so the logic in rndis_response_available()
kept the first RNDIS response notification from getting out.
This prevented interoperation with MS-Windows (but not Linux).

Fix this to make RNDIS behave again.

Signed-off-by: Richard Röjfors <>
Signed-off-by: David Brownell <>
Cc: stable <>
Signed-off-by: Greg Kroah-Hartman <>
14 years agoUSB: gadget rndis: stop windows self-immolation
David Brownell [Sun, 16 Nov 2008 03:53:21 +0000 (19:53 -0800)]
USB: gadget rndis: stop windows self-immolation

Somewhere in the conversion of the RNDIS gadget code to the new
framework, the descriptor of its data interface seems to have
been copied from the CDC Ethernet driver.  Unfortunately that
means it got a nonzero altsetting ... which is incorrect.  Issue
uncovered by Richard Röjfors <>.

This patch fixes that problem, and resolves at least some cases
of Windows XP bluescreening itself.

Tested-by: Richard Röjfors <>.
Signed-off-by: David Brownell <>
Cc: stable <>
Signed-off-by: Greg Kroah-Hartman <>
14 years agoUSB: storage: update unusual_devs entries for Nokia 5300 and 5310
Alan Stern [Mon, 17 Nov 2008 19:32:16 +0000 (14:32 -0500)]
USB: storage: update unusual_devs entries for Nokia 5300 and 5310

This patch (as1168) updates the unusual_devs entry for the Nokia 5300.
According to Jorge Lucangeli Obes <>, some existing
models have a revision number lower than the lower limit of the
current entry.

The patch also moves the entry for the Nokia 5310 to its correct place
in the file.

Signed-off-by: Alan Stern <>
Signed-off-by: Greg Kroah-Hartman <>
14 years agoUSB: storage: updates unusual_devs entry for the Nokia 6300
Alan Stern [Mon, 17 Nov 2008 21:12:32 +0000 (16:12 -0500)]
USB: storage: updates unusual_devs entry for the Nokia 6300

This patch (as1169) modifies the unusual_devs entry for the Nokia
6300.  According to Maciej Gierok <> and David
McBride <>, the revision limits need to be wider.

This fixes Bugzilla #11768.

Signed-off-by: Alan Stern <>
Signed-off-by: Greg Kroah-Hartman <>
14 years agousb: musb: fix bug in musb_schedule
Felipe Balbi [Mon, 17 Nov 2008 07:08:16 +0000 (09:08 +0200)]
usb: musb: fix bug in musb_schedule

This bug was introduced recently. Fix it before bigger
problems appear.

Signed-off-by: Felipe Balbi <>
Cc: Sergei Shtylyov <>
Signed-off-by: Greg Kroah-Hartman <>
14 years agoUSB: fix SB700 usb subsystem hang bug
Andiry Xu [Fri, 14 Nov 2008 03:42:29 +0000 (11:42 +0800)]
USB: fix SB700 usb subsystem hang bug

This patch is required for AMD SB700 south bridge revision A12 and A13 to avoid
USB subsystem hang symptom. The USB subsystem hang symptom is observed when the
system has multiple USB devices connected to it. In some cases a USB hub may be
required to observe this symptom.

This patch works around the problem by correcting the internal register setting
that will help by changing the behavior of the internal logic to avoid the
USB subsystem hang issue. The change in the behavior of the logic does not
impact the normal operation of the USB subsystem.

Reported-by: Volker Armin Hemmann <>
Tested-by: Volker Armin Hemmann <>
Signed-off-by: Andiry Xu <>
Signed-off-by: Libin Yang <>
Cc: stable <>
Signed-off-by: Greg Kroah-Hartman <>
14 years agoMerge branch 'for-linus' of git://
Linus Torvalds [Thu, 20 Nov 2008 02:57:16 +0000 (18:57 -0800)]
Merge branch 'for-linus' of git://git./linux/kernel/git/jbarnes/pci-2.6

* 'for-linus' of git://
  PCI: Fix disable IRQ 0 in pci_reset_function()

14 years agoMerge branch 'x86/numa' of git://
Linus Torvalds [Thu, 20 Nov 2008 02:53:02 +0000 (18:53 -0800)]
Merge branch 'x86/numa' of git://git./linux/kernel/git/tip/linux-2.6-tip

* 'x86/numa' of git://
  x86: make NUMA on 32-bit depend on EXPERIMENTAL again
  x86, hibernate: fix breakage on x86_32 with CONFIG_NUMA set

14 years agoMerge branch 'x86-fixes-for-linus' of git://
Linus Torvalds [Thu, 20 Nov 2008 02:51:56 +0000 (18:51 -0800)]
Merge branch 'x86-fixes-for-linus' of git://git./linux/kernel/git/tip/linux-2.6-tip

* 'x86-fixes-for-linus' of git://
  x86: more general identifier for Phoenix BIOS
  AMD IOMMU: check for next_bit also in unmapped area
  AMD IOMMU: fix fullflush comparison length
  AMD IOMMU: enable device isolation per default
  AMD IOMMU: add parameter to disable device isolation
  x86, PEBS/DS: fix code flow in ds_request()
  x86: add rdtsc barrier to TSC sync check
  xen: fix scrub_page()
  x86: fix es7000 compiling
  x86, bts: fix unlock problem in ds.c
  x86, voyager: fix smp generic helper voyager breakage
  x86: move iomap.h to the new include location

14 years agocciss: fix DEBUG printk formats
Randy Dunlap [Wed, 19 Nov 2008 23:36:51 +0000 (15:36 -0800)]
cciss: fix DEBUG printk formats

Fix printk format warnings when CCISS_DEBUG is defined.

drivers/block/cciss.c:2856: warning: format '%d' expects type 'int', but argument 2 has type 'long unsigned int'
drivers/block/cciss.c:3205: warning: format '%x' expects type 'unsigned int', but argument 2 has type 'long unsigned int'
drivers/block/cciss.c:3236: warning: format '%x' expects type 'unsigned int', but argument 2 has type '__u64'
drivers/block/cciss.c:3246: warning: format '%x' expects type 'unsigned int', but argument 2 has type '__u64'

Signed-off-by: Randy Dunlap <>
Cc: Mike Miller <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agoW1 OMAP: Fix OMAP LDP boot crash
Stanley.Miao [Wed, 19 Nov 2008 23:36:50 +0000 (15:36 -0800)]
W1 OMAP: Fix OMAP LDP boot crash

OMAP LDP boot crash.  This is because w1 subsystem changed the search
interface, so update omap_hdq's search interface to follow the change.

Signed-off-by: Stanley.Miao <>
Signed-off-by: Evgeniy Polyakov <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agodrivers/video/backlight/da903x.c: introduce one more missing kfree
Mike Rapoport [Wed, 19 Nov 2008 23:36:49 +0000 (15:36 -0800)]
drivers/video/backlight/da903x.c: introduce one more missing kfree

One more error handling code should have kfree as well

Signed-off-by: Mike Rapoport <>
Acked-by: Eric Miao <>
Cc: Richard Purdie <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agocgroups: fix a serious bug in cgroupstats
Li Zefan [Wed, 19 Nov 2008 23:36:48 +0000 (15:36 -0800)]
cgroups: fix a serious bug in cgroupstats

Try this, and you'll get oops immediately:
 # cd Documentation/accounting/
 # gcc -o getdelays getdelays.c
 # mount -t cgroup -o debug xxx /mnt
 # ./getdelays -C /mnt/tasks

Because a normal file's dentry->d_fsdata is a pointer to struct cftype,
not struct cgroup.

After the patch, it returns EINVAL if we try to get cgroupstats
from a normal file.

Cc: Balbir Singh <>
Signed-off-by: Li Zefan <>
Acked-by: Paul Menage <>
Cc: <> [2.6.25.x, 2.6.26.x, 2.6.27.x]
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agohostfs: fix a duplicated global function name
WANG Cong [Wed, 19 Nov 2008 23:36:46 +0000 (15:36 -0800)]
hostfs: fix a duplicated global function name

fs/hostfs/hostfs_user.c defines do_readlink() as non-static, and so does
fs/xfs/linux-2.6/xfs_ioctl.c when CONFIG_XFS_DEBUG=y.  So rename
do_readlink() in hostfs to hostfs_do_readlink().

I think it's better if XFS guys will also rename their do_readlink(),
it's not necessary to use such a general name.

Signed-off-by: WANG Cong <>
Cc: Jeff Dike <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agocirrusfb: remove unused variables
Vlada Peric [Wed, 19 Nov 2008 23:36:45 +0000 (15:36 -0800)]
cirrusfb: remove unused variables

After commit a1d35a7a (cirrusfb: use modedb and add mode_option
parameter), these variables are no longer used, so remove them to fix
compilation warning.

Signed-off-by: Vlada Periæ <>
Acked-by: Krzysztof Helt <>
Cc: Stephen Rothwell <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agovmscan: fix get_scan_ratio() comment
Rik van Riel [Wed, 19 Nov 2008 23:36:44 +0000 (15:36 -0800)]
vmscan: fix get_scan_ratio() comment

Fix the old comment on the scan ratio calculations.

Signed-off-by: Rik van Riel <>
Cc: KOSAKI Motohiro <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agoMAINTAINERS: change email address for rostedt
Steven Rostedt [Wed, 19 Nov 2008 23:36:43 +0000 (15:36 -0800)]
MAINTAINERS: change email address for rostedt

I find that I answer my email quicker on my home email account, than I do
on my work email.  Not to mention that I never check my work email while
traveling.  Please change my email address in the MAINTAINERS file from to

Signed-off-by: Steven Rostedt <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agodrivers/hwmon/applesmc.c: add generic MacPro support
Henrik Rydberg [Wed, 19 Nov 2008 23:36:42 +0000 (15:36 -0800)]
drivers/hwmon/applesmc.c: add generic MacPro support

In order to analyze the SMC of the newer MacPros, applesmc needs to
recognize the machine.  This patch adds the missing generic dmi_match
entry for MacPro models.

Signed-off-by: Henrik Rydberg <>
Cc: Nicolas Boichat <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agodrivers/video/backlight/da903x.c: introduce missing kfree
Julia Lawall [Wed, 19 Nov 2008 23:36:41 +0000 (15:36 -0800)]
drivers/video/backlight/da903x.c: introduce missing kfree

Error handling code following a kzalloc should free the allocated data.

The semantic match that finds the problem is as follows:

// <smpl>
@r exists@
local idexpression x;
statement S;
expression E;
identifier f,l;
position p1,p2;
expression *ptr != NULL;

if ((x@p1 = \(kmalloc\|kzalloc\|kcalloc\)(...)) == NULL) S
x@p1 = \(kmalloc\|kzalloc\|kcalloc\)(...);
if (x == NULL) S
<... when != x
     when != if (...) { <+...x...+> }
x->f = E
 return \(0\|<+...x...+>\|ptr\);
 return@p2 ...;

p1 << r.p1;
p2 << r.p2;

print "* file: %s kmalloc %s return %s" % (p1[0].file,p1[0].line,p2[0].line)
// </smpl>

Signed-off-by: Julia Lawall <>
Cc: Mike Rapoport <>
Cc: Richard Purdie <>
Cc: Eric Miao <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agodon't unlink an active swapfile
Hugh Dickins [Wed, 19 Nov 2008 23:36:38 +0000 (15:36 -0800)]
don't unlink an active swapfile

Peter Cordes is sorry that he rm'ed his swapfiles while they were in use,
he then had no pathname to swapoff.  It's a curious little oversight, but
not one worth a lot of hackery.  Kudos to Willy Tarreau for turning this
around from a discussion of synthetic pathnames to how to prevent unlink.
Mimic immutable: prohibit unlinking an active swapfile in may_delete()
(and don't worry my little head over the tiny race window).

Signed-off-by: Hugh Dickins <>
Cc: Willy Tarreau <>
Acked-by: Christoph Hellwig <>
Cc: Peter Cordes <>
Cc: Bodo Eggert <>
Cc: David Newall <>
Cc: Peter Zijlstra <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agovmscan: let GFP_NOFS go to swap again
Hugh Dickins [Wed, 19 Nov 2008 23:36:37 +0000 (15:36 -0800)]
vmscan: let GFP_NOFS go to swap again

In the past, GFP_NOFS (but of course not GFP_NOIO) was allowed to reclaim
by writing to swap.  That got partially broken in 2.6.23, when may_enter_fs
initialization was moved up before the allocation of swap, so its
PageSwapCache test was failing the first time around,

Fix it by setting may_enter_fs when add_to_swap() succeeds with
__GFP_IO.  In fact, check __GFP_IO before calling add_to_swap():
allocating swap we're not ready to use just increases disk seeking.

Signed-off-by: Hugh Dickins <>
Cc: Rik van Riel <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agomigration: fix writepage error
Hugh Dickins [Wed, 19 Nov 2008 23:36:36 +0000 (15:36 -0800)]
migration: fix writepage error

Page migration's writeout() has got understandably confused by the nasty
AOP_WRITEPAGE_ACTIVATE case: as in normal success, a writepage() error has
unlocked the page, so writeout() then needs to relock it.

Signed-off-by: Hugh Dickins <>
Cc: KAMEZAWA Hiroyuki <>
Cc: Christoph Lameter <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agosprint_symbol(): use less stack
Hugh Dickins [Wed, 19 Nov 2008 23:36:36 +0000 (15:36 -0800)]
sprint_symbol(): use less stack

sprint_symbol(), itself used when dumping stacks, has been wasting 128
bytes of stack: lookup the symbol directly into the buffer supplied by the
caller, instead of using a locally declared namebuf.

I believe the name != buffer strcpy() is obsolete: the design here dates
from when module symbol lookup pointed into a supposedly const but sadly
volatile table; nowadays it copies, but an uncalled strcpy() looks better
here than the risk of a recursive BUG_ON().

Signed-off-by: Hugh Dickins <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agocgroup: fix potential deadlock in pre_destroy
KAMEZAWA Hiroyuki [Wed, 19 Nov 2008 23:36:34 +0000 (15:36 -0800)]
cgroup: fix potential deadlock in pre_destroy

As Balbir pointed out, memcg's pre_destroy handler has potential deadlock.

It has following lock sequence.

cgroup_mutex (cgroup_rmdir)
    -> pre_destroy -> mem_cgroup_pre_destroy-> force_empty
-> cpu_hotplug.lock. (lru_add_drain_all->

But, cpuset has following.
cpu_hotplug.lock (call notifier)
-> cgroup_mutex. (within notifier)

Then, this lock sequence should be fixed.

Considering how pre_destroy works, it's not necessary to holding
cgroup_mutex() while calling it.

As a side effect, we don't have to wait at this mutex while memcg's
force_empty works.(it can be long when there are tons of pages.)

Signed-off-by: KAMEZAWA Hiroyuki <>
Acked-by: Balbir Singh <>
Cc: Li Zefan <>
Cc: Paul Menage <>
Cc: Daisuke Nishimura <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agomm: vmalloc search restart fix
Glauber Costa [Wed, 19 Nov 2008 23:36:33 +0000 (15:36 -0800)]
mm: vmalloc search restart fix

Current vmalloc restart search for a free area in case we can't find one.
The reason is there are areas which are lazily freed, and could be
possibly freed now.  However, current implementation start searching the
tree from the last failing address, which is pretty much by definition at
the end of address space.  So, we fail.

The proposal of this patch is to restart the search from the beginning of
the requested vstart address.  This fixes the regression in running KVM
virtual machines for me, described in,
caused by commit db64fe02258f1507e13fe5212a989922323685ce.

Signed-off-by: Glauber Costa <>
Signed-off-by: Nick Piggin <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agomm: vmalloc failure flush fix
Nick Piggin [Wed, 19 Nov 2008 23:36:33 +0000 (15:36 -0800)]
mm: vmalloc failure flush fix

An initial vmalloc failure should start off a synchronous flush of lazy
areas, in case someone is in progress flushing them already, which could
cause us to return an allocation failure even if there is plenty of KVA

Signed-off-by: Nick Piggin <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agomm: vmalloc allocator off by one
Nick Piggin [Wed, 19 Nov 2008 23:36:32 +0000 (15:36 -0800)]
mm: vmalloc allocator off by one

Fix off by one bug in the KVA allocator that can leave gaps in the address

Signed-off-by: Nick Piggin <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agocpuset: update top cpuset's mems after adding a node
Miao Xie [Wed, 19 Nov 2008 23:36:30 +0000 (15:36 -0800)]
cpuset: update top cpuset's mems after adding a node

After adding a node into the machine, top cpuset's mems isn't updated.

By reviewing the code, we found that the update function


was invoked after node_states[N_ONLINE] changes.  It is wrong because
N_ONLINE just means node has pgdat, and if node has/added memory, we use
N_HIGH_MEMORY.  So, We should invoke the update function after
node_states[N_HIGH_MEMORY] changes, just like its commit says.

This patch fixes it.  And we use notifier of memory hotplug instead of
direct calling of cpuset_track_online_nodes().

Signed-off-by: Miao Xie <>
Acked-by: Yasunori Goto <>
Cc: David Rientjes <>
Cc: Paul Menage <
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agoeCryptfs: Allocate up to two scatterlists for crypto ops on keys
Michael Halcrow [Wed, 19 Nov 2008 23:36:28 +0000 (15:36 -0800)]
eCryptfs: Allocate up to two scatterlists for crypto ops on keys

I have received some reports of out-of-memory errors on some older AMD
architectures.  These errors are what I would expect to see if
crypt_stat->key were split between two separate pages.  eCryptfs should
not assume that any of the memory sent through virt_to_scatterlist() is
all contained in a single page, and so this patch allocates two
scatterlist structs instead of one when processing keys.  I have received
confirmation from one person affected by this bug that this patch resolves
the issue for him, and so I am submitting it for inclusion in a future
stable release.

Note that virt_to_scatterlist() runs sg_init_table() on the scatterlist
structs passed to it, so the calls to sg_init_table() in
decrypt_passphrase_encrypted_session_key() are redundant.

Signed-off-by: Michael Halcrow <>
Reported-by: Paulo J. S. Silva <>
Cc: "Leon Woestenberg" <>
Cc: Tim Gardner <>
Cc: <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agospi_imx: full duplex dma corruption bugfix
Andrea Paterniani [Wed, 19 Nov 2008 23:36:26 +0000 (15:36 -0800)]
spi_imx: full duplex dma corruption bugfix

Fix unsafe order in dma mapping operation: always flush data from the
cache *BEFORE* invalidating it, to allow full duplex transfers where the
same buffer may be used for both writes and reads.

Signed-off-by: Andrea Paterniani <>
Signed-off-by: David Brownell <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agolcd: fix oops if driver only interested in .set_power
Ben Dooks [Wed, 19 Nov 2008 23:36:25 +0000 (15:36 -0800)]
lcd: fix oops if driver only interested in .set_power

The LCD driver core calls LCD drivers when either the blanking state or
the display mode has changed, but does not make any check to see if the
called driver has a .set_mode method.

This means if a driver only has a .set_power method then the system will
OOPS on changing mode (and with the console semaphore held so you cannot
easily see the problem).

Fix the problem by ensuring that either callback is valid before use.

Signed-off-by: Ben Dooks <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agoviafb: fix releasing of /proc/viafb/ subtree
Bruno Prémont [Wed, 19 Nov 2008 23:36:23 +0000 (15:36 -0800)]
viafb: fix releasing of /proc/viafb/ subtree

When unloading viafb module it does not remove it's /proc/viafb/ subtree
which causes multiple viafb directories to appear below proc when
mobprobing viafb and also lets kernel WARN() on duplicate proc entries:

[  145.458387] WARNING: at /usr/src/linux-2.6.28-rc3-git6/fs/proc/generic.c:551 proc_register+0xe6/0x160()
[  145.458945] proc_dir_entry '/proc/viafb' already registered
[  145.459278] Modules linked in: viafb(+) i2c_algo_bit cfbcopyarea cfbimgblt cfbfillrect snd_hda_intel snd_pcm snd_timer snd soundcore snd_page_alloc sg via_agp agpgart [last unloaded: drm]
[  145.460647] Pid: 1904, comm: modprobe Tainted: G        W  2.6.28-rc3-git6 #4
[  145.461064] Call Trace:
[  145.461248]  [<c01066f1>] ? dump_stack+0x1/0x80
[  145.461533]  [<c01228a3>] warn_slowpath+0x63/0x80
[  145.461851]  [<c0253ec9>] ? idr_get_empty_slot+0xe9/0x250
[  145.462186]  [<c0254120>] ? ida_get_new_above+0xf0/0x150
[  145.462528]  [<c019fb86>] proc_register+0xe6/0x160
[  145.462827]  [<c019fdc6>] proc_mkdir_mode+0x36/0x50
[  145.463135]  [<c019fdef>] proc_mkdir+0xf/0x20
[  145.463457]  [<f807173c>] viafb_init+0x73c/0xc86 [viafb]
[  145.463823]  [<f8071000>] ? viafb_init+0x0/0xc86 [viafb]
[  145.464147]  [<c010111d>] do_one_initcall+0x2d/0x160
[  145.464460]  [<c01a6543>] ? sysfs_add_file+0x13/0x20
[  145.464786]  [<c015f031>] ? vfree+0x21/0x30
[  145.465049]  [<c01433b5>] ? load_module+0x1215/0x1500
[  145.465381]  [<c014e455>] ? __alloc_pages_internal+0x95/0x400
[  145.465755]  [<c0143723>] sys_init_module+0x83/0x1a0
[  145.466065]  [<c016ceed>] ? sys_read+0x3d/0x70
[  145.466354]  [<c0103bc1>] sysenter_do_call+0x12/0x25
[  145.466653] ---[ end trace c84b37826e16748c ]---

Signed-off-by: Bruno Prémont <>
Cc: <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agopxa2xx_spi: bugfix full duplex dma data corruption
Ned Forrester [Wed, 19 Nov 2008 23:36:21 +0000 (15:36 -0800)]
pxa2xx_spi: bugfix full duplex dma data corruption

Fixes a data corruption bug in pxa2xx_spi.c when operating in full duplex
mode with DMA and using buffers that overlap.

SPI transmit and receive buffers are allowed to be the same or to overlap.
 However, this driver fails if such overlap is attempted in DMA mode
because it maps the rx and tx buffers in the wrong order.  By mapping
DMA_FROM_DEVICE (read) before DMA_TO_DEVICE (write), it invalidates the
cache before flushing it, thus discarding data which should have been

The patch corrects the order of mapping.  This bug exists in all versions
of pxa2xx_spi.c; similar bugs are in the drivers for two other SPI
controllers (au1500, imx).

A version of this patch has been tested on kernel 2.6.20 using
verification of loopback data with: random transfer length, random
bits-per-word, random positive offsets (both larger and smaller than
transfer length) between the start of the rx and tx buffers, and varying
clock rates.

Signed-off-by: Ned Forrester <>
Cc: Vernon Sauder <>
Cc: J. Scott Merritt <>
Signed-off-by: David Brownell <>
Cc: <> [2.6.27.x]
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agolib/scatterlist.c: fix kunmap() argument in sg_miter_stop()
Arjan van de Ven [Wed, 19 Nov 2008 23:36:19 +0000 (15:36 -0800)]
lib/scatterlist.c: fix kunmap() argument in sg_miter_stop()

kunmap() takes as argument the struct page that orginally got kmap()'d,
however the sg_miter_stop() function passed it the kernel virtual address
instead, resulting in weird stuff.

Somehow I ended up fixing this bug by accident while looking for a bug in
the same area.

Acked-by: Tejun Heo <>
Signed-off-by: Arjan van de Ven <>
Cc: Hugh Dickins <>
Cc: <> [2.6.27.x]
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agotmiofb: fix compilation with ACCEL disabled
Dmitry Baryshkov [Wed, 19 Nov 2008 23:36:19 +0000 (15:36 -0800)]
tmiofb: fix compilation with ACCEL disabled

Restore support for compiling tmiofb with acceleration disabled.

Signed-off-by: Dmitry Baryshkov <>
Cc: Krzysztof Helt <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agoGRU: fix for debug option
Jack Steiner [Wed, 19 Nov 2008 23:36:18 +0000 (15:36 -0800)]
GRU: fix for debug option

Enable -D DEBUG in the GRU Makefile if CONFIG_SGI_GRU_DEBUG is selected.

Signed-off-by: Jack Steiner <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agogpiolib: extend gpio label column width in debugfs file
Jarkko Nikula [Wed, 19 Nov 2008 23:36:17 +0000 (15:36 -0800)]
gpiolib: extend gpio label column width in debugfs file

There are already various drivers having bigger label than 12 bytes.  Most
of them fit well under 20 bytes but make column width exact so that
oversized labels don't mess up output alignment.

Signed-off-by: Jarkko Nikula <>
Acked-by: David Brownell <>
Cc: <> [2.6.26.x, 2.6.26.x, 2.6.27.x]
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agoDocumentation/kernel-parameters.txt: add min_addr, fix max_addr
Randy Dunlap [Wed, 19 Nov 2008 23:36:16 +0000 (15:36 -0800)]
Documentation/kernel-parameters.txt: add min_addr, fix max_addr

Add "min_addr" documentation.
For "max_addr", add nn before [KMG] since a number is needed and this
is consistent with other uses of [KMG].

Signed-off-by: Randy Dunlap <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agosparc64: wire up accept4()
David Miller [Wed, 19 Nov 2008 23:36:15 +0000 (15:36 -0800)]
sparc64: wire up accept4()

This adds the sparc syscall hookups.

Signed-off-by: David S. Miller <>
Cc: Ulrich Drepper <>
Cc: Michael Kerrisk <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agoreintroduce accept4
Ulrich Drepper [Wed, 19 Nov 2008 23:36:14 +0000 (15:36 -0800)]
reintroduce accept4

Introduce a new accept4() system call.  The addition of this system call
matches analogous changes in 2.6.27 (dup3(), evenfd2(), signalfd4(),
inotify_init1(), epoll_create1(), pipe2()) which added new system calls
that differed from analogous traditional system calls in adding a flags
argument that can be used to access additional functionality.

The accept4() system call is exactly the same as accept(), except that
it adds a flags bit-mask argument.  Two flags are initially implemented.
(Most of the new system calls in 2.6.27 also had both of these flags.)

SOCK_CLOEXEC causes the close-on-exec (FD_CLOEXEC) flag to be enabled
for the new file descriptor returned by accept4().  This is a useful
security feature to avoid leaking information in a multithreaded
program where one thread is doing an accept() at the same time as
another thread is doing a fork() plus exec().  More details here: "Secure File Descriptor Handling",
Ulrich Drepper).

The other flag is SOCK_NONBLOCK, which causes the O_NONBLOCK flag
to be enabled on the new open file description created by accept4().
(This flag is merely a convenience, saving the use of additional calls
fcntl(F_GETFL) and fcntl (F_SETFL) to achieve the same result.

Here's a test program.  Works on x86-32.  Should work on x86-64, but
I (mtk) don't have a system to hand to test with.

It tests accept4() with each of the four possible combinations of
SOCK_CLOEXEC and SOCK_NONBLOCK set/clear in 'flags', and verifies
that the appropriate flags are set on the file descriptor/open file
description returned by accept4().

I tested Ulrich's patch in this thread by applying against 2.6.28-rc2,
and it passes according to my test program.

/* test_accept4.c

  Copyright (C) 2008, Linux Foundation, written by Michael Kerrisk

  Licensed under the GNU GPLv2 or later.
#define _GNU_SOURCE
#include <unistd.h>
#include <sys/syscall.h>
#include <sys/socket.h>
#include <netinet/in.h>
#include <stdlib.h>
#include <fcntl.h>
#include <stdio.h>
#include <string.h>

#define PORT_NUM 33333

#define die(msg) do { perror(msg); exit(EXIT_FAILURE); } while (0)


/* The following is what we need until glibc gets a wrapper for
  accept4() */

/* Flags for socket(), socketpair(), accept4() */

#ifdef __x86_64__
#define SYS_accept4 288
#elif __i386__
#define SYS_ACCEPT4 18
#error "Sorry -- don't know the syscall # on this architecture"

static int
accept4(int fd, struct sockaddr *sockaddr, socklen_t *addrlen, int flags)
   printf("Calling accept4(): flags = %x", flags);
   if (flags != 0) {
       printf(" (");
       if (flags & SOCK_CLOEXEC)
       if ((flags & SOCK_CLOEXEC) && (flags & SOCK_NONBLOCK))
           printf(" ");
       if (flags & SOCK_NONBLOCK)

   long args[6];

   args[0] = fd;
   args[1] = (long) sockaddr;
   args[2] = (long) addrlen;
   args[3] = flags;

   return syscall(SYS_socketcall, SYS_ACCEPT4, args);
   return syscall(SYS_accept4, fd, sockaddr, addrlen, flags);


static int
do_test(int lfd, struct sockaddr_in *conn_addr,
       int closeonexec_flag, int nonblock_flag)
   int connfd, acceptfd;
   int fdf, flf, fdf_pass, flf_pass;
   struct sockaddr_in claddr;
   socklen_t addrlen;


   connfd = socket(AF_INET, SOCK_STREAM, 0);
   if (connfd == -1)
   if (connect(connfd, (struct sockaddr *) conn_addr,
               sizeof(struct sockaddr_in)) == -1)

   addrlen = sizeof(struct sockaddr_in);
   acceptfd = accept4(lfd, (struct sockaddr *) &claddr, &addrlen,
                      closeonexec_flag | nonblock_flag);
   if (acceptfd == -1) {
       return 0;

   fdf = fcntl(acceptfd, F_GETFD);
   if (fdf == -1)
   fdf_pass = ((fdf & FD_CLOEXEC) != 0) ==
              ((closeonexec_flag & SOCK_CLOEXEC) != 0);
   printf("Close-on-exec flag is %sset (%s); ",
           (fdf & FD_CLOEXEC) ? "" : "not ",
           fdf_pass ? "OK" : "failed");

   flf = fcntl(acceptfd, F_GETFL);
   if (flf == -1)
   flf_pass = ((flf & O_NONBLOCK) != 0) ==
              ((nonblock_flag & SOCK_NONBLOCK) !=0);
   printf("nonblock flag is %sset (%s)\n",
           (flf & O_NONBLOCK) ? "" : "not ",
           flf_pass ? "OK" : "failed");


   printf("Test result: %s\n", (fdf_pass && flf_pass) ? "PASS" : "FAIL");
   return fdf_pass && flf_pass;

static int
create_listening_socket(int port_num)
   struct sockaddr_in svaddr;
   int lfd;
   int optval;

   memset(&svaddr, 0, sizeof(struct sockaddr_in));
   svaddr.sin_family = AF_INET;
   svaddr.sin_addr.s_addr = htonl(INADDR_ANY);
   svaddr.sin_port = htons(port_num);

   lfd = socket(AF_INET, SOCK_STREAM, 0);
   if (lfd == -1)

   optval = 1;
   if (setsockopt(lfd, SOL_SOCKET, SO_REUSEADDR, &optval,
                  sizeof(optval)) == -1)

   if (bind(lfd, (struct sockaddr *) &svaddr,
            sizeof(struct sockaddr_in)) == -1)

   if (listen(lfd, 5) == -1)

   return lfd;

main(int argc, char *argv[])
   struct sockaddr_in conn_addr;
   int lfd;
   int port_num;
   int passed;

   passed = 1;

   port_num = (argc > 1) ? atoi(argv[1]) : PORT_NUM;

   memset(&conn_addr, 0, sizeof(struct sockaddr_in));
   conn_addr.sin_family = AF_INET;
   conn_addr.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
   conn_addr.sin_port = htons(port_num);

   lfd = create_listening_socket(port_num);

   if (!do_test(lfd, &conn_addr, 0, 0))
       passed = 0;
   if (!do_test(lfd, &conn_addr, SOCK_CLOEXEC, 0))
       passed = 0;
   if (!do_test(lfd, &conn_addr, 0, SOCK_NONBLOCK))
       passed = 0;
   if (!do_test(lfd, &conn_addr, SOCK_CLOEXEC, SOCK_NONBLOCK))
       passed = 0;


   exit(passed ? EXIT_SUCCESS : EXIT_FAILURE);

[ rewrote changelog, updated test program]
Signed-off-by: Ulrich Drepper <>
Tested-by: Michael Kerrisk <>
Acked-by: Michael Kerrisk <>
Cc: <>
Cc: <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agofbdev: clean the penguin's dirty feet
Clemens Ladisch [Wed, 19 Nov 2008 23:36:10 +0000 (15:36 -0800)]
fbdev: clean the penguin's dirty feet

When booting in a direct color mode, the penguin has dirty feet, i.e.,
some pixels have the wrong color.  This is caused by
fb_set_logo_directpalette() which does not initialize the last 32 palette

Signed-off-by: Clemens Ladisch <>
Acked-by: Geert Uytterhoeven <>
Cc: Krzysztof Helt <>
Cc: <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agosysvipc: fix the ipc structures initialization
Nadia Derbey [Wed, 19 Nov 2008 23:36:08 +0000 (15:36 -0800)]
sysvipc: fix the ipc structures initialization

A problem was found while reviewing the code after Bugzilla bug

In ipc_addid(), the newly allocated ipc structure is inserted into the
ipcs tree (i.e made visible to readers) without locking it.  This is not
correct since its initialization continues after it has been inserted in
the tree.

This patch moves the ipc structure lock initialization + locking before
the actual insertion.

Signed-off-by: Nadia Derbey <>
Reported-by: Clement Calmels <>
Cc: Manfred Spraul <>
Cc: <> [2.6.27.x]
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agodrivers/video: bad error test before a dereference
Julien Brunel [Wed, 19 Nov 2008 23:36:07 +0000 (15:36 -0800)]
drivers/video: bad error test before a dereference

The error test that follows the call to backlight_device_register semms
not to concern the right variable.

A simplified version of the semantic match that finds this problem is
as follows:
// <smpl>
expression x;
position p0;
x@p0 = backlight_device_register(...)

expression def0.x,E;
position def0.p0;
position p;
statement S;
... when != x = E
if (!IS_ERR(x) && ...) {<... x@p ...>} else S

expression def0.x;
identifier fld;
position def0.p0;
position p != protected.p;
... when != x = E
* x@p->fld
// </smpl>

Signed-off-by: Julien Brunel <>
Signed-off-by: Julia Lawall <>
Acked-by: Nicolas Ferre <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agohwmon: applesmc: Add support for iMac 6
Henrik Rydberg [Wed, 19 Nov 2008 23:36:06 +0000 (15:36 -0800)]
hwmon: applesmc: Add support for iMac 6

Add temperature sensor support for iMac 6.

Signed-off-by: Henrik Rydberg <>
Tested-by: Caleb Hyde <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
14 years agonet: ipg.c fix bracing on endian swapping
Harvey Harrison [Wed, 19 Nov 2008 23:54:38 +0000 (15:54 -0800)]
net: ipg.c fix bracing on endian swapping

rxfd->frag_info is a __le64, IPG_RFI_FRAGLEN is a cpu-endian
constant and wants to be outside of the le64_to_cpu. Fixed
in multiple places.

Also an occurrence where le64_to_cpu was used instead of cpu_to_le64

Signed-off-by: Harvey Harrison <>
Signed-off-by: David S. Miller <>
14 years agophylib: Fix auto-negotiation restart avoidance
Trent Piepho [Wed, 19 Nov 2008 23:52:41 +0000 (15:52 -0800)]
phylib: Fix auto-negotiation restart avoidance

A previous patch, 51e2a3846eab18711f4eb59cd0a4c33054e2980a, made
genphy_config_aneg() not restart aneg by calling genphy_restart_aneg() if
the advertisement hadn't changed.

But, genphy_restart_aneg() doesn't just restart aneg, it may also *enable*
aneg or un-isolate the PHY from the MII (those functions are controlled by
the same register).  The code to avoid calling genphy_restart_aneg() didn't
consider this.

So, modify genphy_config_aneg() to also check if the PHY needs to have aneg
enabled or be un-isolated before deciding not to restart aneg.

This caused a problem with certain Davicom PHYs, as that driver isolates
the PHY (why?) before calling genphy_config_aneg() and expects the PHY to
be un-isolated by that function.

Signed-off-by: Trent Piepho <>
Reported-by: Scott Wood <>
Signed-off-by: David S. Miller <>
14 years agonet: jme.c rxdesc.flags is __le16, other missing endian swaps
Harvey Harrison [Wed, 19 Nov 2008 23:50:59 +0000 (15:50 -0800)]
net: jme.c rxdesc.flags is __le16, other missing endian swaps

This is the minimal patch to fix endian mismatches.  These are
probably bugs on big-endian arches, noops on little endian.

jme_rxsum_ok could be improved to directly take a __le16 and
change all of the masks/sets to be in little-endian, but
has not been done here to keep the patch small.

Signed-off-by: Harvey Harrison <>
Signed-off-by: David S. Miller <>
14 years agophylib: fix phy name example in documentation
Paulius Zaleckas [Wed, 19 Nov 2008 23:38:24 +0000 (15:38 -0800)]
phylib: fix phy name example in documentation

All MDIO bus drivers currently name bus with "%x" format.
There is one exception where mv643xx_eth driver is using "%d".
Phy address on the bus uses format "%02x".

Fixing phy name example to match all real life MDIO drivers.

Signed-off-by: Paulius Zaleckas <>
Signed-off-by: David S. Miller <>
14 years agonet: Do not fire linkwatch events until the device is registered.
David S. Miller [Wed, 19 Nov 2008 23:33:54 +0000 (15:33 -0800)]
net: Do not fire linkwatch events until the device is registered.

Several device drivers try to do things like netif_carrier_off()
before register_netdev() is invoked.  This is bogus, but too many
drivers do this to fix them all up in one go.

Reported-by: Folkert van Heusden <>
Signed-off-by: David S. Miller <>
14 years agophonet: fix compilation with gcc-3.4
Alexey Dobriyan [Wed, 19 Nov 2008 22:17:41 +0000 (14:17 -0800)]
phonet: fix compilation with gcc-3.4

  CC [M]  net/phonet/af_phonet.o
net/phonet/af_phonet.c: In function `pn_socket_create':
net/phonet/af_phonet.c:38: sorry, unimplemented: inlining failed in call to 'phonet_proto_put': function body not available
net/phonet/af_phonet.c:99: sorry, unimplemented: called from here
make[3]: *** [net/phonet/af_phonet.o] Error 1

Signed-off-by: Alexey Dobriyan <>
Signed-off-by: David S. Miller <>
14 years agoixgbe: fix compilation with gcc-3.4
Alexey Dobriyan [Wed, 19 Nov 2008 22:17:02 +0000 (14:17 -0800)]
ixgbe: fix compilation with gcc-3.4

  CC [M]  drivers/net/ixgbe/ixgbe_main.o
drivers/net/ixgbe/ixgbe_main.c: In function `ixgbe_intr':
drivers/net/ixgbe/ixgbe_main.c:1290: sorry, unimplemented: inlining failed in call to 'ixgbe_irq_enable': function body not available
drivers/net/ixgbe/ixgbe_main.c:1312: sorry, unimplemented: called from here
make[4]: *** [drivers/net/ixgbe/ixgbe_main.o] Error 1

Signed-off-by: Alexey Dobriyan <>
Acked-by: Peter P Waskiewicz Jr <>
Signed-off-by: David S. Miller <>
14 years agoPCI: Fix disable IRQ 0 in pci_reset_function()
Sheng Yang [Tue, 11 Nov 2008 09:17:45 +0000 (17:17 +0800)]
PCI: Fix disable IRQ 0 in pci_reset_function()

Before initialization, dev->irq may be zero. Make sure we don't disable
it at reset time in that case.

Reviewed-by: Matthew Wilcox <>
Signed-off-by: Sheng Yang <>
Signed-off-by: Jesse Barnes <>
14 years agopktgen: fix multiple queue warning
Robert Olsson [Wed, 19 Nov 2008 22:09:47 +0000 (14:09 -0800)]
pktgen: fix multiple queue warning

As number of TX queues in unrelated to number of CPU's we remove this test
and just make sure nxtq never gets exceeded.

Signed-off-by: Robert Olsson <>
Signed-off-by: David S. Miller <>
14 years agonet: fix ip_mr_init() error path
Benjamin Thery [Wed, 19 Nov 2008 22:07:41 +0000 (14:07 -0800)]
net: fix ip_mr_init() error path

Similarly to IPv6 ip6_mr_init() (fixed last week), the order of cleanup
operations in the error/exit section of ip_mr_init() is completely
inversed. It should be the other way around.
Also a del_timer() is missing in the error path.

I should have guessed last week that this same error existed in ipmr.c
too, as ip6mr.c is largely inspired by ipmr.c.

Signed-off-by: Benjamin Thery <>
Signed-off-by: David S. Miller <>
14 years agointel-iommu: fix compile warnings
FUJITA Tomonori [Wed, 19 Nov 2008 04:53:42 +0000 (13:53 +0900)]
intel-iommu: fix compile warnings

Impact: cleanup

I got the following warnings on IA64:

  linux-2.6/drivers/pci/intel-iommu.c: In function 'init_dmars':
  linux-2.6/drivers/pci/intel-iommu.c:1658: warning: format '%Lx' expects type 'long long unsigned int', but argument 2 has type 'u64'
  linux-2.6/drivers/pci/intel-iommu.c:1663: warning: format '%Lx' expects type 'long long unsigned int', but argument 2 has type 'u64'

Another victim of int-ll64.h versus int-l64.h confusion between platforms.

->reg_base_addr has a type of u64 - which can only be printed out
consistently if we cast its type up to LL.

[ Eventually reg_base_addr should be converted to phys_addr_t, for which
  we have the %pR printk helper - but that is out of the scope of late
  -rc's. ]

Signed-off-by: FUJITA Tomonori <>
Signed-off-by: Ingo Molnar <>
14 years agoMerge branch 'tip/urgent' of git://
Ingo Molnar [Wed, 19 Nov 2008 08:00:50 +0000 (09:00 +0100)]
Merge branch 'tip/urgent' of git://git./linux/kernel/git/rostedt/linux-2.6-trace into tracing/urgent

14 years agoftrace: fix dyn ftrace filter selection
Steven Rostedt [Wed, 19 Nov 2008 01:33:02 +0000 (20:33 -0500)]
ftrace: fix dyn ftrace filter selection

Impact: clean up and fix for dyn ftrace filter selection

The previous logic of the dynamic ftrace selection of enabling
or disabling functions was complex and incorrect. This patch simplifies
the code and corrects the usage. This simplification also makes the
code more robust.

Here is the correct logic:

  Given a function that can be traced by dynamic ftrace:

  If the function is not to be traced, disable it if it was enabled.
  (this is if the function is in the set_ftrace_notrace file)

  (filter is on if there exists any functions in set_ftrace_filter file)

  If the filter is on, and we are enabling functions:
    If the function is in set_ftrace_filter, enable it if it is not
      already enabled.
    If the function is not in set_ftrace_filter, disable it if it is not
      already disabled.

  Otherwise, if the filter is off and we are enabling function tracing:
    Enable the function if it is not already enabled.

  Otherwise, if we are disabling function tracing:
    Disable the function if it is not already disabled.

This code now sets or clears the ENABLED flag in the record, and at the
end it will enable the function if the flag is set, or disable the function
if the flag is cleared.

The parameters for the function that does the above logic is also
simplified. Instead of passing in confusing "new" and "old" where
they might be swapped if the "enabled" flag is not set. The old logic
even had one of the above always NULL and had to be filled in. The new
logic simply passes in one parameter called "nop". A "call" is calculated
in the code, and at the end of the logic, when we know we need to either
disable or enable the function, we can then use the "nop" and "call"

This code is more robust than the previous version.

Signed-off-by: Steven Rostedt <>
Signed-off-by: Ingo Molnar <>
14 years agoftrace: make filtered functions effective on setting
Steven Rostedt [Wed, 19 Nov 2008 04:57:14 +0000 (23:57 -0500)]
ftrace: make filtered functions effective on setting

Impact: fix filter selection to apply when set

It can be confusing when the set_filter_functions is set (or cleared)
and the functions being recorded by the dynamic tracer does not

This patch causes the code to be updated if the function tracer is
enabled and the filter is changed.

Signed-off-by: Steven Rostedt <>
Signed-off-by: Ingo Molnar <>
14 years agoftrace: fix set_ftrace_filter
Steven Rostedt [Sat, 8 Nov 2008 03:36:02 +0000 (22:36 -0500)]
ftrace: fix set_ftrace_filter

Impact: fix of output of set_ftrace_filter

The commit "ftrace: do not show freed records in

Removed a bit too much from the set_ftrace_filter code, where we now see
all functions in the set_ftrace_filter file even when we set a filter.

Signed-off-by: Steven Rostedt <>
Signed-off-by: Ingo Molnar <>
14 years agoiwlagn: fix RX skb alignment
Johannes Berg [Tue, 18 Nov 2008 00:47:21 +0000 (01:47 +0100)]
iwlagn: fix RX skb alignment

So I dug deeper into the DMA problems I had with iwlagn and a kind soul
helped me in that he said something about pci-e alignment and mentioned
the iwl_rx_allocate function to check for crossing 4KB boundaries. Since
there's 8KB A-MPDU support, crossing 4k boundaries didn't seem like
something the device would fail with, but when I looked into the
function for a minute anyway I stumbled over this little gem:

BUG_ON(rxb->dma_addr & (~DMA_BIT_MASK(36) & 0xff));

Clearly, that is a totally bogus check, one would hope the compiler
removes it entirely. (Think about it)

After fixing it, I obviously ran into it, nothing guarantees the
alignment the way you want it,  because of the way skbs and their
headroom are allocated. I won't explain that here nor double-check that
I'm right, that goes beyond what most of the CC'ed people care about.

So then I came up with the patch below, and so far my system has
survived minutes with 64K pages, when it would previously fail in
seconds. And I haven't seen a single instance of the TX bug either. But
when you see the patch it'll be pretty obvious to you why.

This should fix the following reported kernel bugs:

I haven't checked if there are any elsewhere, but I suppose RHBZ will
have a few instances too...

I'd like to ask anyone who is CC'ed (those are people I know ran into
the bug) to try this patch.

I am convinced that this patch is correct in spirit, but I haven't
understood why, for example, there are so many unmap calls. I'm not
entirely convinced that this is the only bug leading to the TX reply

Signed-off-by: Johannes Berg <>
Signed-off-by: John W. Linville <>
14 years agomac80211: remove ieee80211_notify_mac
Johannes Berg [Mon, 17 Nov 2008 09:59:59 +0000 (10:59 +0100)]
mac80211: remove ieee80211_notify_mac

Before ieee80211_notify_mac() was added, it was presented with the
use case of using it to tell mac80211 that the association may
have been lost because the firmware crashed/reset.

Since then, it has also been used by iwlwifi to (slightly) speed
up re-association after resume, a workaround around the fact that
mac80211 has no suspend/resume handling yet. It is also not used
by any other drivers, so clearly it cannot be necessary for "good
enough" suspend/resume.

Unfortunately, the callback suffers from a severe problem: It only
works for station mode. If suspend/resume happens while in IBSS or
any other mode (but station), then the callback is pointless.

Recently, it has created a number of locking issues, first because
it required rtnl locking rather than RCU due to calling sleeping
functions within the critical section, and now because it's called
by iwlwifi from the mac80211 workqueue that may not use the rtnl
because it is flushed under rtnl.

I think, therefore, that we should take a step back, remove it
entirely for now and add the small feature it provided properly.
For suspend and resume we will need to introduce new hooks, and for
the case where the firmware was reset the driver will probably
simply just pretend it has done a suspend/resume cycle to get
mac80211 to reprogram the hardware completely, not just try to
connect to the current AP again in station mode. When doing so, we
will need to take into account locking issues and possibly defer
to schedule_work from within mac80211 for the resume operation,
while the suspend operation must be done directly.

Proper suspend/resume should also not necessarily try to reconnect
to the current AP, the time spent in suspend may have been short
enough to not be disconnected from the AP, mac80211 will detect
that the AP went out of range quickly if it did, and if the
association is lost then the AP will disassoc as soon as a data
frame is sent. We might also take into account WWOL then, and
have mac80211 program the hardware into such a mode where it is
available and requested.

Signed-off-by: Johannes Berg <>
Signed-off-by: John W. Linville <>