Linus Torvalds [Mon, 21 May 2012 23:14:04 +0000 (16:14 -0700)]
vfs: be even more careful about dentry RCU name lookups
Miklos Szeredi points out that we need to also worry about memory
odering when doing the dentry name comparison asynchronously with RCU.
In particular, doing a rename can do a memcpy() of one dentry name over
another, and we want to make sure that any unlocked reader will always
see the proper terminating NUL character, so that it won't ever run off
the allocation.
Rather than having to be extra careful with the name copy or at lookup
time for each character, this resolves the issue by making sure that all
names that are inlined in the dentry always have a NUL character at the
end of the name allocation. If we do that at dentry allocation time, we
know that no future name copy will ever change that final NUL to
anything else, so there are no memory ordering issues.
So even if a concurrent rename ends up overwriting the NUL character
that terminates the original name, we always know that there is one
final NUL at the end, and there is no worry about the lockless RCU
lookup traversing the name too far.
The out-of-line allocations are never copied over, so we can just make
sure that we write the name (with terminating NULL) and do a write
barrier before we expose the name to anything else by setting it in the
dentry.
Reported-by: Miklos Szeredi <mszeredi@suse.cz> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Nick Piggin <npiggin@gmail.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Linus Torvalds [Mon, 21 May 2012 23:06:20 +0000 (16:06 -0700)]
vfs: make AIO use the proper rw_verify_area() area helpers
We had for some reason overlooked the AIO interface, and it didn't use
the proper rw_verify_area() helper function that checks (for example)
mandatory locking on the file, and that the size of the access doesn't
cause us to overflow the provided offset limits etc.
Instead, AIO did just the security_file_permission() thing (that
rw_verify_area() also does) directly.
This fixes it to do all the proper helper functions, which not only
means that now mandatory file locking works with AIO too, we can
actually remove lines of code.
Linus Torvalds [Mon, 21 May 2012 23:01:50 +0000 (16:01 -0700)]
Merge branch 'for-linus' of git://git.linaro.org/people/rmk/linux-arm
Pull core ARM updates from Russell King:
"This is the bulk of the core ARM updates for this merge window.
Included in here is a different way to handle the VIVT cache flushing
on context switch, which should allow scheduler folk to remove a
special case in their core code.
We have architectured timer support here, which is a set of timers
specified by the ARM architecture for future SoCs. So we should see
less variability in timer design going forward.
The last big thing here is my cleanup to the way we handle PCI across
ARM, fixing some oddities in some platforms which hadn't realised
there was a way to deal with their private data already built in to
our PCI backend.
I've also removed support for the ARMv3 architecture; it hasn't worked
properly for years so it seems pointless to keep it around."
* 'for-linus' of git://git.linaro.org/people/rmk/linux-arm: (47 commits)
ARM: PCI: remove per-pci_hw list of buses
ARM: PCI: dove/kirkwood/mv78xx0: use sys->private_data
ARM: PCI: provide a default bus scan implementation
ARM: PCI: get rid of pci_std_swizzle()
ARM: PCI: versatile: fix PCI interrupt setup
ARM: PCI: integrator: use common PCI swizzle
ARM: 7416/1: LPAE: Remove unused L_PTE_(BUFFERABLE|CACHEABLE) macros
ARM: 7415/1: vfp: convert printk's to pr_*'s
ARM: decompressor: avoid speculative prefetch from non-RAM areas
ARM: Remove ARMv3 support from decompressor
ARM: 7413/1: move read_{boot,persistent}_clock to the architecture level
ARM: Remove support for ARMv3 ARM610 and ARM710 CPUs
ARM: 7363/1: DEBUG_LL: limit early mapping to the minimum
ARM: 7391/1: versatile: add some auxdata for device trees
ARM: 7389/2: plat-versatile: modernize FPGA IRQ controller
AMBA: get rid of last two uses of NO_IRQ
ARM: 7408/1: cacheflush: return error to userspace when flushing syscall fails
ARM: 7409/1: Do not call flush_cache_user_range with mmap_sem held
ARM: 7404/1: cmpxchg64: use atomic64 and local64 routines for cmpxchg64
ARM: 7347/1: SCU: use cpu_logical_map for per-CPU low power mode
...
Linus Torvalds [Mon, 21 May 2012 22:59:46 +0000 (15:59 -0700)]
Merge branch 'clkdev' of git://git.linaro.org/people/rmk/linux-arm
Pull clkdev updates from Russell King:
"This supplements clkdev with a device-managed API, allowing drivers
cleanup paths to be simplified. We also optimize clk_find() so that
it exits as soon as it finds a perfect match, and we provide a way to
minimise the amount of code platforms need to register clkdev entries.
Some of the code in arm-soc depends on these changes."
* 'clkdev' of git://git.linaro.org/people/rmk/linux-arm:
CLKDEV: provide helpers for common clock framework
ARM: 7392/1: CLKDEV: Optimize clk_find()
ARM: 7376/1: clkdev: Implement managed clk_get()
Linus Torvalds [Mon, 21 May 2012 19:52:42 +0000 (12:52 -0700)]
Fix blocking allocations called very early during bootup
During early boot, when the scheduler hasn't really been fully set up,
we really can't do blocking allocations because with certain (dubious)
configurations the "might_resched()" calls can actually result in
scheduling events.
We could just make such users always use GFP_ATOMIC, but quite often the
code that does the allocation isn't really aware of the fact that the
scheduler isn't up yet, and forcing that kind of random knowledge on the
initialization code is just annoying and not good for anybody.
And we actually have a the 'gfp_allowed_mask' exactly for this reason:
it's just that the kernel init sequence happens to set it to allow
blocking allocations much too early.
So move the 'gfp_allowed_mask' initialization from 'start_kernel()'
(which is some of the earliest init code, and runs with preemption
disabled for good reasons) into 'kernel_init()'. kernel_init() is run
in the newly created thread that will become the 'init' process, as
opposed to the early startup code that runs within the context of what
will be the first idle thread.
So by the time we reach 'kernel_init()', we know that the scheduler must
be at least limping along, because we've already scheduled from the idle
thread into the init thread.
Reported-by: Steven Rostedt <rostedt@goodmis.org> Cc: David Rientjes <rientjes@google.com> Cc: stable@vger.kernel.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Sam Ravnborg [Mon, 21 May 2012 18:45:37 +0000 (20:45 +0200)]
net: drop NET dependency from HAVE_BPF_JIT
There is no point having the NET dependency on the select target, as it
forces all users to depend on NET to tell they support BPF_JIT. Move
the config option to the bottom of the file - this could be a nice place
also for future "selectable" config symbols.
Fix up all users to drop the dependency on NET now that it is not
required to supress warnings for non-NET builds.
Reported-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Sam Ravnborg <sam@ravnborg.org> Acked-by: David Miller <davem@davemloft.net> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Linus Torvalds [Mon, 21 May 2012 19:47:53 +0000 (12:47 -0700)]
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/lliubbo/blackfin
Pull blackfin changes from Bob Liu:
"The biggest change was added an new processor(bf60x series).
Bf60x series processor of blackfin can up to 1GHz with Hardware
Support for HD Video Analytics, it use the same blackfin ISA but with
some changes on system buses, interrupt controller and peripheral
devices.
Added dir arch/blackfin/mach-bf609/ and did some changes to the
framework made linux working fine on the reference board bf609-ezkit
now."
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/lliubbo/blackfin: (41 commits)
blackfin: fix build after add bf60x mach/pm.h
blackfin: twi: include linux/i2c.h
blackfin: bf60x: add head file for crc controller
blackfin: bf60x: twi: work around temporary anomaly 0501001
blackfin: twi: Move TWI MMR access macro to twi head file
blackfin: twi: Move TWI peripheral pin request array to platform data
blackfin: bf60x: anomaly: Add a temporary anomaly 0501001
blackfin: bf60x: Rename the DDR controller macro
blackfin: mach-bf609: pm: cleanup bfin_deepsleep
blackfin: bf60x: cleanup get clock code
blackfin: bf60x: pm: Add a debug option to calculate kernel wakeup time.
blackfin: bf60x: add wakeup source select
blackfin: bf60x: make clock changeable in kernel menuconfig
blackfin:mach-bf609: fix norflash for bf609-ezkit
blackfin: mach-bf609: add can_wakeup to ethernet device
blackfin: remove redundant CONFIG_BF60x macro
blackfin: rotary: Add pm_wakeup flag to platform data structure.
bfin_gpio: fix bf548-ezkit kernel fail to boot
bfin_dma: fix initcall return error in proc_dma_init()
Blackfin: delete fork func
...
Linus Torvalds [Mon, 21 May 2012 19:46:48 +0000 (12:46 -0700)]
Merge tag 'for-linus' of git://linux-c6x.org/git/projects/linux-c6x-upstreaming
Pull c6x updates from Mark Salter:
"Clean up some c6x Kconfig items and add support for Elf FDPIC loader."
* tag 'for-linus' of git://linux-c6x.org/git/projects/linux-c6x-upstreaming:
C6X: remove unused config items
C6X: add support to build with BINFMT_ELF_FDPIC
C6X: change main arch kbuild symbol
Linus Torvalds [Mon, 21 May 2012 19:43:54 +0000 (12:43 -0700)]
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/geert/linux-m68k
Pull m68k updates from Geert Uytterhoeven.
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/geert/linux-m68k:
m68k: Setup CROSS_COMPILE at the top
m68k: Correct the Atari ALLOWINT definition
m68k/video: Create <asm/vga.h>
m68k: Make sure {read,write}s[bwl]() are always defined
m68k/mm: Port OOM changes to do_page_fault()
scsi/atari: Make more functions static
scsi/atari: Revive "atascsi=" setup option
net/ariadne: Improve debug prints
m68k/atari: Change VME irq numbers from unsigned long to unsigned int
m68k/amiga: Use arch_initcall() for registering platform devices
m68k/amiga: Add error checks when registering platform devices
m68k/amiga: Mark z_dev_present() __init
m68k: Remove unused MAX_NOINT_IPL definition
1) Kill off support for sun4c and Cypress sun4m chips.
And as a result we were able to also kill off that ugly btfixup thing
that required multi-stage links of the final vmlinux image in the
Kbuild system. This should make the kbuild maintainers really happy.
Thanks a lot to Sam Ravnborg for his tireless efforts to get this
going.
2) Convert sparc64 to nobootmem. I suspect now with sparc32 being a lot
cleaner, it should be able to fall in line and modernize in this area
too.
3) Make sparc32 use generic clockevents, from Tkhai Kirill.
[ I fixed up the BPF rules, and tried to clean up the build rules too.
But I don't have - or want - a sparc cross-build environment, so the
BPF rule bug and the related build cleanup was all done with just a
bare "make -n" pseudo-test. - Linus ]
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc-next: (110 commits)
sparc32: use flushi when run-time patching in per_cpu_patch
sparc32: fix cpuid_patch run-time patching
sparc32: drop unused inline functions in srmmu.c
sparc32: drop unused functions in pgtsrmmu.h
sparc32,leon: move leon mmu functions to leon_mm.c
sparc32,leon: remove duplicate definitions in leon.h
sparc32,leon: remove duplicate UART register definitions
sparc32,leon: move leon ASI definitions to asi.h
sparc32: move trap table to a separate file
sparc64: renamed ttable.S to ttable_64.S
sparc32: Remove asm/sysen.h header.
sparc32: Delete asm/smpprim.h
sparc32: Remove unused empty_bad_page{,_table} declarations.
sparc32: Kill boot_cpu_id4
sparc32: Move GET_PROCESSOR*_ID() out of asm/asmmacro.h
sparc32: Remove completely unused code from asm/cache.h
sparc32: Add ucmpdi2.o to obj-y instead of lib-y.
sparc32: add ucmpdi2
sparc: introduce arch/sparc/Kbuild
sparc: remove obsolete documentation
...
1) Get rid of the error prone NLA_PUT*() macros that used an embedded
goto.
2) Kill off the token-ring and MCA networking drivers, from Paul
Gortmaker.
3) Reduce high-order allocations made by datagram AF_UNIX sockets, from
Eric Dumazet.
4) Add PTP hardware clock support to IGB and IXGBE, from Richard
Cochran and Jacob Keller.
5) Allow users to query timestamping capabilities of a card via
ethtool, from Richard Cochran.
6) Add loadbalance mode to the teaming driver, from Jiri Pirko. Part
of this is that we can now have BPF filters not attached to sockets,
and the loadbalancing function is calculated using one.
7) Francois Romieu went through the network drivers removing gratuitous
uses of netdev->base_addr, perhaps some day we can remove it
completely but it's used for ISA probing still.
8) Add a BPF JIT for sparc. I know, who cares, right? :-)
9) Move networking sysctl registry away from using the compatability
mode interfaces in the sysctl code. From Eric W Biederman.
10) Pavel Emelyanov added a way to save and restore TCP socket state via
TCP_REPAIR, TCP_REPAIR_QUEUE, and TCP_QUEUE_SEQ socket options as
well as a way to forcefully bind a socket to a port via the
sk->sk_reuse value SK_FORCE_REUSE. There is also a
TCP_REPAIR_OPTIONS which allows to reinstante the TCP options
enabled on the connection.
11) Several enhancements from Eric Dumazet that, in particular, can
enhance splice performance on TCP sockets significantly.
a) Reset the offset of the per-socket sendmsg page when we know
we're the only use of the page in linear_to_page().
b) Add facilities such that skb->data can be backed a page rather
than SLAB kmalloc'd memory. In particular devices which were
receiving into linear RX buffers can now end up providing paged
data.
The big result is that code like splice and GRO do not have to copy
any more.
12) Allow a pure sender to more gracefully handle ACK backlogs in TCP.
What can happen at high rates is that the sender hasn't grown his
receive buffer limits at all (he's not receiving data so really
doesn't need to), but the non-data ACKs consume receive buffer
space.
sk_add_backlog() is too aggressive in dropping frames in this case,
so relax it's requirements by using the receive buffer plus the send
buffer limit as the backlog limit instead of just the former.
Also from Eric Dumazet.
13) Add ipv6 support to L2TP, from Benjamin LaHaise, James Chapman, and
Chris Elston.
14) Implement TCP early retransmit (RFC 5827), from Yuchung Cheng.
Basically, we can start fast retransmit before hiting the dupack
threshold under certain conditions.
15) New CODEL active queue management packet scheduler, from Eric
Dumazet based upon initial work by Dave Taht.
Basically, the big feature is that packets are dropped (or ECN bits
are set) based upon how long packets live in the queue, rather than
the queue length (which is what RED uses).
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next: (1341 commits)
drivers/net/stmmac: seq_file fix memory leak
ipv6/exthdrs: strict Pad1 and PadN check
USB: qmi_wwan: Add ZTE (Vodafone) K3520-Z
USB: qmi_wwan: Add ZTE (Vodafone) K3765-Z
USB: qmi_wwan: Make forced int 4 whitelist generic
net/ipv4: replace simple_strtoul with kstrtoul
net/ipv4/ipconfig: neaten __setup placement
net: qmi_wwan: Add Vodafone/Huawei K5005 support
net: cdc_ether: Add ZTE WWAN matches before generic Ethernet
ipv6: use skb coalescing in reassembly
ipv4: use skb coalescing in defragmentation
net: introduce skb_try_coalesce()
net:ipv6:fixed space issues relating to operators.
net:ipv6:fixed a trailing white space issue.
ipv6: disable GSO on sockets hitting dst_allfrag
tg3: use netdev_alloc_frag() API
net: napi_frags_skb() is static
ppp: avoid false drop_monitor false positives
ipv6: bool/const conversions phase2
ipx: Remove spurious NULL checking in ipx_ioctl().
...
Linus Torvalds [Mon, 21 May 2012 15:50:57 +0000 (08:50 -0700)]
Merge branch 'dentry-cleanups' (dcache access cleanups and optimizations)
This branch simplifies and clarifies the dcache lookup, and allows us to
do certain nice optimizations when comparing dentries. It also cleans
up the interface to __d_lookup_rcu(), especially around passing the
inode information around.
* dentry-cleanups:
vfs: make it possible to access the dentry hash/len as one 64-bit entry
vfs: move dentry name length comparison from dentry_cmp() into callers
vfs: do the careful dentry name access for all dentry_cmp cases
vfs: remove unnecessary d_unhashed() check from __d_lookup_rcu
vfs: clean up __d_lookup_rcu() and dentry_cmp() interfaces
Linus Torvalds [Mon, 21 May 2012 15:46:08 +0000 (08:46 -0700)]
Merge branch 'vfs-cleanups' (random vfs cleanups)
This teaches vfs_fstat() to use the appropriate f[get|put]_light
functions, allowing it to avoid some unnecessary locking for the common
case.
More noticeably, it also cleans up and simplifies the "getname_flags()"
function, which now relies on the architecture strncpy_from_user() doing
all the user access checks properly, instead of hacking around the fact
that on x86 it didn't use to do it right (see commit 92ae03f2ef99: "x86:
merge 32/64-bit versions of 'strncpy_from_user()' and speed it up").
* vfs-cleanups:
VFS: make vfs_fstat() use f[get|put]_light()
VFS: clean up and simplify getname_flags()
x86: make word-at-a-time strncpy_from_user clear bytes at the end
Linus Torvalds [Mon, 21 May 2012 15:41:38 +0000 (08:41 -0700)]
Merge branch 'stat-cleanups' (clean up copying of stat info to user space)
This makes cp_new_stat() a bit more readable, and avoids having to
memset() the whole structure just to fill in a couple of padding fields.
This is another result of me looking at code generation of functions
that show up high on certain kernel profiles, and just going "Oh, let's
just clean that up".
Architectures that don't supply the #define to fill just the padding
fields will still fall back to memset().
* stat-cleanups:
vfs: don't force a big memset of stat data just to clear padding fields
vfs: de-crapify "cp_new_stat()" function
This series sanitizes the interface to unmap_vma(). The crazy interface
annoyed me no end when I was looking at unmap_single_vma(), which we can
spend quite a lot of time in (especially with loads that have a lot of
small fork/exec's: shell scripts etc).
Moving the nr_accounted calculations to where they belong at least
clarifies things a little. I hope to come back to look at the
performance of this later, but if/when I get back to it I at least don't
have to see the crazy interfaces any more.
* vm-cleanups:
vm: remove 'nr_accounted' calculations from the unmap_vmas() interfaces
vm: simplify unmap_vmas() calling convention
Bob Liu [Sat, 28 Apr 2012 09:12:32 +0000 (17:12 +0800)]
blackfin: asm: fix blackfin.h broken
do{...}while makes no sense in __ASSEMBLY__ code paths. now
kernels fail to build:
arch/blackfin/mach-bf561/atomic.S: Assembler messages:
arch/blackfin/mach-bf561/atomic.S:48: Error: syntax error. Input text
was do.
arch/blackfin/mach-bf561/atomic.S:48: Error:
arch/blackfin/mach-bf561/atomic.S:48: Error: syntax error. Input text
was }.
arch/blackfin/mach-bf561/atomic.S:48: Error:
arch/blackfin/mach-bf561/atomic.S:53: Error: syntax error. Input text
was do.
arch/blackfin/mach-bf561/atomic.S:53: Error:
arch/blackfin/mach-bf561/atomic.S:53: Error: syntax error. Input text
was }.
* Take into account combinations of consecutive Pad1 and PadN.
* Catch the corner case of when only padding is present in the
header, when the extention header length is 0 (i.e., 8 bytes).
In this case, the header would have exactly 6 bytes of padding:
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
: Next Header : Hdr Ext Len=0 : :
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ +
: Padding (Pad1 or PadN) :
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Signed-off-by: Eldad Zack <eldad@fogrefinery.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Add the ZTE (Vodafone) K3765-Z to the whitelist. This requires the
previous patch to make the whitelist with forced interface 4 generic
or the device fails to initialise. After applying this patch and
loading the Option driver without usb-modeswitch's bind all
interfaces trick, a wwan0 net interface and /dev/cdc-wdm0 device
file were created. Using Bjorn Mork's perl connection script a
connection was made to a mobile network using QMI and the network
interface's IPv4 address was configured OK.
Signed-off-by: Andrew Bird <ajb@spheresystems.co.uk> Acked-by: Bjørn Mork <bjorn@mork.no> Signed-off-by: David S. Miller <davem@davemloft.net>
USB: qmi_wwan: Make forced int 4 whitelist generic
Change the forced interface 4 whitelist to use the generic shared
binder instead of the Gobi specific one. Certain ZTE devices
(K3520-Z & K3765-Z) don't work with the Gobi version, but function
quite happily with the generic. This has been tested with the following
devices:
K3520-Z
K3565-Z
K3765-Z
K4505-Z
It hasn't been tested with the ZTE MF820D, which is the only other
device that uses this whitelist at present. Although Bjorn doesn't
expect any problems, any testing with that device would be appreciated.
Signed-off-by: Andrew Bird <ajb@spheresystems.co.uk> Acked-by: Bjørn Mork <bjorn@mork.no> Signed-off-by: David S. Miller <davem@davemloft.net>
Sam Ravnborg [Sun, 20 May 2012 15:27:53 +0000 (17:27 +0200)]
sparc32: use flushi when run-time patching in per_cpu_patch
Davis S. Miller wrote:
"
The way we do that now is overkill. We only needed to use the MMU
cache ops when we had sun4c around because sun4c lacked support for
the "flush" instruction.
But all sun4m and later chips have it so we can use it
unconditionally.
So in the per_cpu_patch() code, get rid of the cache ops invocation,
and instead execute a "flush %reg" after each of the instruction patch
assignments, where %reg is set to the address of the instruction that
was stored into.
Perhaps take the flushi() definition from asm/cacheflush_64.h and
place it into asm/cacheflush.h, then you can simply use that.
"
Implemented as per suggestion.
Moved run-time patching before we call paging_init(),
so helper methods in paging_init() may utilise run-time patching too.
Signed-off-by: Sam Ravnborg <sam@ravnborg.org> Signed-off-by: David S. Miller <davem@davemloft.net>
Eldad Zack [Sat, 19 May 2012 14:13:18 +0000 (14:13 +0000)]
net/ipv4: replace simple_strtoul with kstrtoul
Replace simple_strtoul with kstrtoul in three similar occurrences, all setup
handlers:
* route.c: set_rhash_entries
* tcp.c: set_thash_entries
* udp.c: set_uhash_entries
Also check if the conversion failed.
Signed-off-by: Eldad Zack <eldad@fogrefinery.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Eric Dumazet [Sat, 19 May 2012 03:02:35 +0000 (03:02 +0000)]
ipv6: use skb coalescing in reassembly
ip6_frag_reasm() can use skb_try_coalesce() to build optimized skb,
reducing memory used by them (truesize), and reducing number of cache
line misses and overhead for the consumer.
Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Alexander Duyck <alexander.h.duyck@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Eric Dumazet [Sat, 19 May 2012 03:02:20 +0000 (03:02 +0000)]
ipv4: use skb coalescing in defragmentation
ip_frag_reasm() can use skb_try_coalesce() to build optimized skb,
reducing memory used by them (truesize), and reducing number of cache
line misses and overhead for the consumer.
Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Alexander Duyck <alexander.h.duyck@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Eric Dumazet [Sat, 19 May 2012 03:02:02 +0000 (03:02 +0000)]
net: introduce skb_try_coalesce()
Move tcp_try_coalesce() protocol independent part to
skb_try_coalesce().
skb_try_coalesce() can be used in IPv4 defrag and IPv6 reassembly,
to build optimized skbs (less sk_buff, and possibly less 'headers')
skb_try_coalesce() is zero copy, unless the copy can fit in destination
header (its a rare case)
kfree_skb_partial() is also moved to net/core/skbuff.c and exported,
because IPv6 will need it in patch (ipv6: use skb coalescing in
reassembly).
Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Alexander Duyck <alexander.h.duyck@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Linus Torvalds [Sat, 19 May 2012 22:30:15 +0000 (15:30 -0700)]
Merge tag 'parisc-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/parisc-2.6
Pull PA-RISC fixes from James Bottomley:
"This is a set of three bug fixes that gets parisc running again on
systems with PA1.1 processors.
Two fix regressions introduced in 2.6.39 and one fixes a prefetch bug
that only affects PA7300LC processors. We also have another pending
fix to do with the sectional arrangement of vmlinux.lds, but there's a
query on it during testing on one particular system type, so I'll hold
off sending it in for now."
* tag 'parisc-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/parisc-2.6:
[PARISC] fix panic on prefetch(NULL) on PA7300LC
[PARISC] fix crash in flush_icache_page_asm on PA1.1
[PARISC] fix PA1.1 oops on boot
Linus Torvalds [Sat, 19 May 2012 22:28:22 +0000 (15:28 -0700)]
Merge branch 'x86/ld-fix' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
Pull x86 linker bug workarounds from Peter Anvin.
GNU ld-2.22.52.0.[12] (*) has an unfortunate bug where it incorrectly
turns certain relocation entries absolute. Section-relative symbols
that are part of otherwise empty sections are silently changed them to
absolute. We rely on section-relative symbols staying section-relative,
and actually have several sections in the linker script solely for this
purpose.
We could just black-list the buggy linker, but it appears that it got
shipped in at least F17, and possibly other distros too, so it's sadly
not some rare unusual case.
This backports the workaround from the x86/trampoline branch, and as
Peter says: "This is not a minimal fix, not at all, but it is a tested
code base."
* 'x86/ld-fix' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86, relocs: When printing an error, say relative or absolute
x86, relocs: Workaround for binutils 2.22.52.0.1 section bug
x86, realmode: 16-bit real-mode code support for relocs tool
(*) That's a manly release numbering system. Stupid, sure. But manly.
Linus Torvalds [Sat, 19 May 2012 17:12:17 +0000 (10:12 -0700)]
Merge branch 'for-linus' of git://git.kernel.dk/linux-block
Pull block layer fixes from Jens Axboe:
"A few small, but important fixes. Most of them are marked for stable
as well
- Fix failure to release a semaphore on error path in mtip32xx.
- Fix crashable condition in bio_get_nr_vecs().
- Don't mark end-of-disk buffers as mapped, limit it to i_size.
- Fix for build problem with CONFIG_BLOCK=n on arm at least.
- Fix for a buffer overlow on UUID partition printing.
- Trivial removal of unused variables in dac960."
* 'for-linus' of git://git.kernel.dk/linux-block:
block: fix buffer overflow when printing partition UUIDs
Fix blkdev.h build errors when BLOCK=n
bio allocation failure due to bio_get_nr_vecs()
block: don't mark buffers beyond end of disk as mapped
mtip32xx: release the semaphore on an error path
dac960: Remove unused variables from DAC960_CreateProcEntries()
Hugh Dickins [Fri, 18 May 2012 18:28:34 +0000 (11:28 -0700)]
memcg,thp: fix res_counter:96 regression
Occasionally, testing memcg's move_charge_at_immigrate on rc7 shows
a flurry of hundreds of warnings at kernel/res_counter.c:96, where
res_counter_uncharge_locked() does WARN_ON(counter->usage < val).
The first trace of each flurry implicates __mem_cgroup_cancel_charge()
of mc.precharge, and an audit of mc.precharge handling points to
mem_cgroup_move_charge_pte_range()'s THP handling in commit 12724850e806
("memcg: avoid THP split in task migration").
Checking !mc.precharge is good everywhere else, when a single page is to
be charged; but here the "mc.precharge -= HPAGE_PMD_NR" likely to
follow, is liable to result in underflow (a lot can change since the
precharge was estimated).
Simply check against HPAGE_PMD_NR: there's probably a better
alternative, trying precharge for more, splitting if unsuccessful; but
this one-liner is safer for now - no kernel/res_counter.c:96 warnings
seen in 26 hours.
Eric Dumazet [Fri, 18 May 2012 21:51:44 +0000 (21:51 +0000)]
ipv6: disable GSO on sockets hitting dst_allfrag
If the allfrag feature has been set on a host route (due to an ICMPv6
Packet Too Big received indicating a MTU of less than 1280), we hit a
very slow behavior in TCP stack, because all big packets are dropped and
only a retransmit timer is able to push one MSS frame every 200 ms.
One way to handle this is to disable GSO on the socket the first time a
super packet is dropped. Adding a specific dst_allfrag() in the fast
path is probably overkill since the dst_allfrag() case almost never
happen.
Result on netperf TCP_STREAM, one flow :
Before : 60 kbit/sec
After : 1.6 Gbit/sec
Reported-by: Tore Anderson <tore@fud.no> Signed-off-by: Eric Dumazet <edumazet@google.com> Tested-by: Tore Anderson <tore@fud.no> Signed-off-by: David S. Miller <davem@davemloft.net>