platform/kernel/linux-rpi.git
13 years agocrypto: user - Fix rwsem leak in crypto_user
Jonathan Corbet [Tue, 1 Nov 2011 22:15:16 +0000 (09:15 +1100)]
crypto: user - Fix rwsem leak in crypto_user

The list_empty case in crypto_alg_match() will return without calling
up_read() on crypto_alg_sem.  We could do the "goto out" routine, but the
function will clearly do the right thing with that test simply removed.

Signed-off-by: Jonathan Corbet <corbet@lwn.net>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: user - Depend on NET instead of selecting it
Herbert Xu [Tue, 1 Nov 2011 01:12:43 +0000 (12:12 +1100)]
crypto: user - Depend on NET instead of selecting it

Selecting NET causes all sorts of issues, including a dependency
loop involving bluetooth.  This patch makes it a dependency instead.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: user - Add dependency on NET
Herbert Xu [Wed, 26 Oct 2011 15:15:10 +0000 (17:15 +0200)]
crypto: user - Add dependency on NET

Since the configuration interface relies on netlink we need to
select NET.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: talitos - handle descriptor not found in error path
Kim Phillips [Fri, 21 Oct 2011 13:20:28 +0000 (15:20 +0200)]
crypto: talitos - handle descriptor not found in error path

The CDPR (Current Descriptor Pointer Register) can be unreliable
when trying to locate an offending descriptor.  Handle that case by
(a) not OOPSing, and (b) reverting to the machine internal copy of
the descriptor header in order to report the correct execution unit
error.

Note: printing all execution units' ISRs is not effective because it
results in an internal time out (ITO) error and the EU resetting its
ISR value (at least when specifying an invalid key length on an SEC
2.2/MPC8313E).

Reported-by: Sven Schnelle <svens@stackframe.org>
Signed-off-by: Kim Phillips <kim.phillips@freescale.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: user - Initialise match in crypto_alg_match
Herbert Xu [Fri, 21 Oct 2011 12:37:10 +0000 (14:37 +0200)]
crypto: user - Initialise match in crypto_alg_match

We need to default match to 0 as otherwise it may lead to a false
positive.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: testmgr - add twofish tests
Jussi Kivilinna [Mon, 10 Oct 2011 20:03:12 +0000 (23:03 +0300)]
crypto: testmgr - add twofish tests

Add tests for parallel twofish-x86_64-3way code paths.

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: testmgr - add blowfish test-vectors
Jussi Kivilinna [Mon, 10 Oct 2011 20:03:03 +0000 (23:03 +0300)]
crypto: testmgr - add blowfish test-vectors

Add tests for parallel blowfish-x86_64 code paths.

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: Make hifn_795x build depend on !ARCH_DMA_ADDR_T_64BIT
Richard Weinberger [Mon, 10 Oct 2011 10:55:41 +0000 (12:55 +0200)]
crypto: Make hifn_795x build depend on !ARCH_DMA_ADDR_T_64BIT

hifn_795x works only on 32 bit, remove the detection while loading
the module and catch non-32 bit systems at build time.

Signed-off-by: Richard Weinberger <richard@nod.at>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: twofish-x86_64-3way - fix ctr blocksize to 1
Jussi Kivilinna [Mon, 10 Oct 2011 09:33:02 +0000 (12:33 +0300)]
crypto: twofish-x86_64-3way - fix ctr blocksize to 1

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: blowfish-x86_64 - fix ctr blocksize to 1
Jussi Kivilinna [Mon, 10 Oct 2011 09:32:15 +0000 (12:32 +0300)]
crypto: blowfish-x86_64 - fix ctr blocksize to 1

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: whirlpool - count rounds from 0
Alexey Dobriyan [Tue, 27 Sep 2011 05:26:24 +0000 (08:26 +0300)]
crypto: whirlpool - count rounds from 0

rc[0] is unused because rounds are counted from 1.
Save an u64!

Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: Add userspace report for compress type algorithms
Steffen Klassert [Tue, 27 Sep 2011 05:48:48 +0000 (07:48 +0200)]
crypto: Add userspace report for compress type algorithms

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: Add userspace report for cipher type algorithms
Steffen Klassert [Tue, 27 Sep 2011 05:48:01 +0000 (07:48 +0200)]
crypto: Add userspace report for cipher type algorithms

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: Add userspace report for rng type algorithms
Steffen Klassert [Tue, 27 Sep 2011 05:47:11 +0000 (07:47 +0200)]
crypto: Add userspace report for rng type algorithms

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: Add userspace report for pcompress type algorithms
Steffen Klassert [Tue, 27 Sep 2011 05:46:32 +0000 (07:46 +0200)]
crypto: Add userspace report for pcompress type algorithms

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: Add userspace report for nivaead type algorithms
Steffen Klassert [Tue, 27 Sep 2011 05:45:44 +0000 (07:45 +0200)]
crypto: Add userspace report for nivaead type algorithms

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: Add userspace report for aead type algorithms
Steffen Klassert [Tue, 27 Sep 2011 05:44:27 +0000 (07:44 +0200)]
crypto: Add userspace report for aead type algorithms

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: Add userspace report for givcipher type algorithms
Steffen Klassert [Tue, 27 Sep 2011 05:43:24 +0000 (07:43 +0200)]
crypto: Add userspace report for givcipher type algorithms

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: Add userspace report for ablkcipher type algorithms
Steffen Klassert [Tue, 27 Sep 2011 05:42:32 +0000 (07:42 +0200)]
crypto: Add userspace report for ablkcipher type algorithms

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: Add userspace report for blkcipher type algorithms
Steffen Klassert [Tue, 27 Sep 2011 05:41:54 +0000 (07:41 +0200)]
crypto: Add userspace report for blkcipher type algorithms

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: Add userspace report for ahash type algorithms
Steffen Klassert [Tue, 27 Sep 2011 05:41:07 +0000 (07:41 +0200)]
crypto: Add userspace report for ahash type algorithms

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: Add userspace report for shash type algorithms
Steffen Klassert [Tue, 27 Sep 2011 05:26:10 +0000 (07:26 +0200)]
crypto: Add userspace report for shash type algorithms

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: Add userspace report for larval type algorithms
Steffen Klassert [Tue, 27 Sep 2011 05:25:05 +0000 (07:25 +0200)]
crypto: Add userspace report for larval type algorithms

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: Add a report function pointer to crypto_type
Steffen Klassert [Tue, 27 Sep 2011 05:24:29 +0000 (07:24 +0200)]
crypto: Add a report function pointer to crypto_type

We add a report function pointer to struct crypto_type. This function
pointer is used from the crypto userspace configuration API to report
crypto algorithms to userspace.

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: Add userspace configuration API
Steffen Klassert [Tue, 27 Sep 2011 05:23:50 +0000 (07:23 +0200)]
crypto: Add userspace configuration API

This patch adds a basic userspace configuration API for the crypto layer.
With this it is possible to instantiate, remove and to show crypto
algorithms from userspace.

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: Export crypto_remove_final
Steffen Klassert [Tue, 27 Sep 2011 05:23:07 +0000 (07:23 +0200)]
crypto: Export crypto_remove_final

The upcomming crypto usrerspace configuration api needs
to remove the spawns on top on an algorithm, so export
crypto_remove_final.

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: Export crypto_remove_spawns
Steffen Klassert [Tue, 27 Sep 2011 05:22:08 +0000 (07:22 +0200)]
crypto: Export crypto_remove_spawns

The upcomming crypto usrerspace configuration api needs
to remove the spawns on top on an algorithm, so export
crypto_remove_spawns.

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: Add a flag to identify crypto instances
Steffen Klassert [Tue, 27 Sep 2011 05:21:26 +0000 (07:21 +0200)]
crypto: Add a flag to identify crypto instances

The upcomming crypto user configuration api needs to identify
crypto instances. This patch adds a flag that is set if the
algorithm is an instance that is build from templates.

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: twofish - add 3-way parallel x86_64 assembler implemention
Jussi Kivilinna [Mon, 26 Sep 2011 13:47:25 +0000 (16:47 +0300)]
crypto: twofish - add 3-way parallel x86_64 assembler implemention

Patch adds 3-way parallel x86_64 assembly implementation of twofish as new
module. New assembler functions crypt data in three blocks chunks, improving
cipher performance on out-of-order CPUs.

Patch has been tested with tcrypt and automated filesystem tests.

Summary of the tcrypt benchmarks:

Twofish 3-way-asm vs twofish asm (128bit 8kb block ECB)
 encrypt: 1.3x speed
 decrypt: 1.3x speed

Twofish 3-way-asm vs twofish asm (128bit 8kb block CBC)
 encrypt: 1.07x speed
 decrypt: 1.4x speed

Twofish 3-way-asm vs twofish asm (128bit 8kb block CTR)
 encrypt: 1.4x speed

Twofish 3-way-asm vs AES asm (128bit 8kb block ECB)
 encrypt: 1.0x speed
 decrypt: 1.0x speed

Twofish 3-way-asm vs AES asm (128bit 8kb block CBC)
 encrypt: 0.84x speed
 decrypt: 1.09x speed

Twofish 3-way-asm vs AES asm (128bit 8kb block CTR)
 encrypt: 1.15x speed

Full output:
 http://koti.mbnet.fi/axh/kernel/crypto/tcrypt-speed-twofish-3way-asm-x86_64.txt
 http://koti.mbnet.fi/axh/kernel/crypto/tcrypt-speed-twofish-asm-x86_64.txt
 http://koti.mbnet.fi/axh/kernel/crypto/tcrypt-speed-aes-asm-x86_64.txt

Tests were run on:
 vendor_id  : AuthenticAMD
 cpu family : 16
 model      : 10
 model name : AMD Phenom(tm) II X6 1055T Processor

Also userspace test were run on:
 vendor_id  : GenuineIntel
 cpu family : 6
 model      : 15
 model name : Intel(R) Xeon(R) CPU           E7330  @ 2.40GHz
 stepping   : 11

Userspace test results:

Encryption/decryption of twofish 3-way vs x86_64-asm on AMD Phenom II:
 encrypt: 1.27x
 decrypt: 1.25x

Encryption/decryption of twofish 3-way vs x86_64-asm on Intel Xeon E7330:
 encrypt: 1.36x
 decrypt: 1.36x

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: twofish-x86-asm - make assembler functions use twofish_ctx instead of crypto_tfm
Jussi Kivilinna [Mon, 26 Sep 2011 13:47:20 +0000 (16:47 +0300)]
crypto: twofish-x86-asm - make assembler functions use twofish_ctx instead of crypto_tfm

This needed by 3-way twofish patch to be able to easily use one block
assembler functions. As glue code is shared between i586/x86_64 apply
change to i586 assembler too. Also export assembler functions for
3-way parallel twofish module.

CC: Joachim Fritschi <jfritschi@freenet.de>
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: tcrypt - add ctr(twofish) speed test
Jussi Kivilinna [Mon, 26 Sep 2011 13:47:15 +0000 (16:47 +0300)]
crypto: tcrypt - add ctr(twofish) speed test

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: blowfish-x86_64 - add credits
Jussi Kivilinna [Fri, 23 Sep 2011 16:51:00 +0000 (19:51 +0300)]
crypto: blowfish-x86_64 - add credits

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: blowfish-x86_64 - improve x86_64 blowfish 4-way performance
Jussi Kivilinna [Fri, 23 Sep 2011 16:50:55 +0000 (19:50 +0300)]
crypto: blowfish-x86_64 - improve x86_64 blowfish 4-way performance

This patch adds improved F-macro for 4-way parallel functions. With new
F-macro for 4-way parallel functions, blowfish sees ~15% improvement in
speed tests on AMD Phenom II (~5% on Intel Xeon E7330).

However when used in 1-way blowfish function new macro would be ~10%
slower than original, so old F-macro is kept for 1-way functions.
Patch cleans up old F-macro as it is no longer needed in 4-way part.

Patch also does register macro renaming to reduce stack usage.

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: picoxcell - add dependency on HAVE_CLK
Jamie Iles [Thu, 20 Oct 2011 12:10:26 +0000 (14:10 +0200)]
crypto: picoxcell - add dependency on HAVE_CLK

The picoxcell crypto driver requires the clk API, but the platform in
mainline does not currently support it.  Add an explicit dependency on
HAVE_CLK to avoid build breakage.

Signed-off-by: Jamie Iles <jamie@jamieiles.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: aes-x86 - quiet sparse noise about symbol not declared
H Hartley Sweeten [Thu, 22 Sep 2011 11:33:01 +0000 (21:33 +1000)]
crypto: aes-x86 - quiet sparse noise about symbol not declared

Include <asm/aes.h> to pick up the declarations for crypto_aes_encrypt_x86
and crypto_aes_decrypt_x86 to quiet the sparse noise:

warning: symbol 'crypto_aes_encrypt_x86' was not declared. Should it be static?
warning: symbol 'crypto_aes_decrypt_x86' was not declared. Should it be static?

Signed-off-by: H Hartley Sweeten <hsweeten@visionengravers.com>
Acked-by: Mandeep Singh Baines <msb@chromium.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: blowfish - add x86_64 assembly implementation
Jussi Kivilinna [Thu, 1 Sep 2011 22:45:22 +0000 (01:45 +0300)]
crypto: blowfish - add x86_64 assembly implementation

Patch adds x86_64 assembly implementation of blowfish. Two set of assembler
functions are provided. First set is regular 'one-block at time'
encrypt/decrypt functions. Second is 'four-block at time' functions that
gain performance increase on out-of-order CPUs. Performance of 4-way
functions should be equal to 1-way functions with in-order CPUs.

Summary of the tcrypt benchmarks:

Blowfish assembler vs blowfish C (256bit 8kb block ECB)
encrypt: 2.2x speed
decrypt: 2.3x speed

Blowfish assembler vs blowfish C (256bit 8kb block CBC)
encrypt: 1.12x speed
decrypt: 2.5x speed

Blowfish assembler vs blowfish C (256bit 8kb block CTR)
encrypt: 2.5x speed

Full output:
http://koti.mbnet.fi/axh/kernel/crypto/tcrypt-speed-blowfish-asm-x86_64.txt
http://koti.mbnet.fi/axh/kernel/crypto/tcrypt-speed-blowfish-c-x86_64.txt

Tests were run on:
 vendor_id : AuthenticAMD
 cpu family : 16
 model : 10
 model name : AMD Phenom(tm) II X6 1055T Processor
 stepping : 0

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: tcrypt - add ctr(blowfish) speed test
Jussi Kivilinna [Thu, 1 Sep 2011 22:45:17 +0000 (01:45 +0300)]
crypto: tcrypt - add ctr(blowfish) speed test

Add ctr(blowfish) speed test to receive results for blowfish x86_64 assembly
patch.

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: blowfish - rename C-version to blowfish_generic
Jussi Kivilinna [Thu, 1 Sep 2011 22:45:12 +0000 (01:45 +0300)]
crypto: blowfish - rename C-version to blowfish_generic

Rename blowfish to blowfish_generic so that assembler versions of blowfish
cipher can autoload. Module alias 'blowfish' is added.

Also fix checkpatch warnings.

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: blowfish - split generic and common c code
Jussi Kivilinna [Thu, 1 Sep 2011 22:45:07 +0000 (01:45 +0300)]
crypto: blowfish - split generic and common c code

Patch splits up the blowfish crypto routine into a common part (key setup)
which will be used by blowfish crypto modules (x86_64 assembly and generic-c).

Also fixes errors/warnings reported by checkpatch.

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: cryptd - Use subsys_initcall to prevent races with aesni
Herbert Xu [Fri, 19 Aug 2011 08:11:23 +0000 (16:11 +0800)]
crypto: cryptd - Use subsys_initcall to prevent races with aesni

As cryptd is depeneded on by other algorithms such as aesni-intel,
it needs to be registered before them.  When everything is built
as modules, this occurs naturally.  However, for this to work when
they are built-in, we need to use subsys_initcall in cryptd.

Tested-by: Josh Boyer <jwboyer@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: sha - Fix build error due to crypto_sha1_update
Herbert Xu [Tue, 16 Aug 2011 06:03:05 +0000 (14:03 +0800)]
crypto: sha - Fix build error due to crypto_sha1_update

On Tue, Aug 16, 2011 at 03:22:34PM +1000, Stephen Rothwell wrote:
>
> After merging the final tree, today's linux-next build (powerpc
> allyesconfig) produced this warning:
>
> In file included from security/integrity/ima/../integrity.h:16:0,
>                  from security/integrity/ima/ima.h:27,
>                  from security/integrity/ima/ima_policy.c:20:
> include/crypto/sha.h:86:10: warning: 'struct shash_desc' declared inside parameter list
> include/crypto/sha.h:86:10: warning: its scope is only this definition or declaration, which is probably not what you want
>
> Introduced by commit 7c390170b493 ("crypto: sha1 - export sha1_update for
> reuse").  I guess you need to include crypto/hash.h in crypto/sha.h.

This patch fixes this by providing a declaration for struct shash_desc.

Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: n2 - Fix a get/put_cpu() imbalance
Thomas Meyer [Mon, 15 Aug 2011 07:20:19 +0000 (15:20 +0800)]
crypto: n2 - Fix a get/put_cpu() imbalance

Fix a get/put_cpu() imbalance in the error case when qp == NULL

Signed-off-by: Thomas Meyer <thomas@m3y3r.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: sha1 - SSSE3 based SHA1 implementation for x86-64
Mathias Krause [Thu, 4 Aug 2011 18:19:25 +0000 (20:19 +0200)]
crypto: sha1 - SSSE3 based SHA1 implementation for x86-64

This is an assembler implementation of the SHA1 algorithm using the
Supplemental SSE3 (SSSE3) instructions or, when available, the
Advanced Vector Extensions (AVX).

Testing with the tcrypt module shows the raw hash performance is up to
2.3 times faster than the C implementation, using 8k data blocks on a
Core 2 Duo T5500. For the smalest data set (16 byte) it is still 25%
faster.

Since this implementation uses SSE/YMM registers it cannot safely be
used in every situation, e.g. while an IRQ interrupts a kernel thread.
The implementation falls back to the generic SHA1 variant, if using
the SSE/YMM registers is not possible.

With this algorithm I was able to increase the throughput of a single
IPsec link from 344 Mbit/s to 464 Mbit/s on a Core 2 Quad CPU using
the SSSE3 variant -- a speedup of +34.8%.

Saving and restoring SSE/YMM state might make the actual throughput
fluctuate when there are FPU intensive userland applications running.
For example, meassuring the performance using iperf2 directly on the
machine under test gives wobbling numbers because iperf2 uses the FPU
for each packet to check if the reporting interval has expired (in the
above test I got min/max/avg: 402/484/464 MBit/s).

Using this algorithm on a IPsec gateway gives much more reasonable and
stable numbers, albeit not as high as in the directly connected case.
Here is the result from an RFC 2544 test run with a EXFO Packet Blazer
FTB-8510:

 frame size    sha1-generic     sha1-ssse3    delta
    64 byte     37.5 MBit/s    37.5 MBit/s     0.0%
   128 byte     56.3 MBit/s    62.5 MBit/s   +11.0%
   256 byte     87.5 MBit/s   100.0 MBit/s   +14.3%
   512 byte    131.3 MBit/s   150.0 MBit/s   +14.2%
  1024 byte    162.5 MBit/s   193.8 MBit/s   +19.3%
  1280 byte    175.0 MBit/s   212.5 MBit/s   +21.4%
  1420 byte    175.0 MBit/s   218.7 MBit/s   +25.0%
  1518 byte    150.0 MBit/s   181.2 MBit/s   +20.8%

The throughput for the largest frame size is lower than for the
previous size because the IP packets need to be fragmented in this
case to make there way through the IPsec tunnel.

Signed-off-by: Mathias Krause <minipli@googlemail.com>
Cc: Maxim Locktyukhin <maxim.locktyukhin@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: sha1 - export sha1_update for reuse
Mathias Krause [Thu, 4 Aug 2011 18:19:24 +0000 (20:19 +0200)]
crypto: sha1 - export sha1_update for reuse

Export the update function as crypto_sha1_update() to not have the need
to reimplement the same algorithm for each SHA-1 implementation. This
way the generic SHA-1 implementation can be used as fallback for other
implementations that fail to run under certain circumstances, like the
need for an FPU context while executing in IRQ context.

Signed-off-by: Mathias Krause <minipli@googlemail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: picoxcell - fix possible invalid pointer dereference
Jamie Iles [Tue, 2 Aug 2011 10:29:06 +0000 (11:29 +0100)]
crypto: picoxcell - fix possible invalid pointer dereference

The completion callback will free the request so we must remove it from
the completion list before calling the callback.

Cc: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: Jamie Iles <jamie@jamieiles.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: picoxcell - support for device tree matching
Jamie Iles [Mon, 1 Aug 2011 16:25:19 +0000 (17:25 +0100)]
crypto: picoxcell - support for device tree matching

Allow the crypto engines to be matched from device tree bindings.

Cc: devicetree-discuss@lists.ozlabs.org
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: Jamie Iles <jamie@jamieiles.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: picoxcell - add connection ID to the clock name
Jamie Iles [Mon, 1 Aug 2011 16:25:18 +0000 (17:25 +0100)]
crypto: picoxcell - add connection ID to the clock name

For using the device tree probing we use a connection ID for the
clk_get() operation.

Cc: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: Jamie Iles <jamie@jamieiles.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: picoxcell - convert to platform ID table
Jamie Iles [Mon, 1 Aug 2011 16:25:17 +0000 (17:25 +0100)]
crypto: picoxcell - convert to platform ID table

Use a platform ID table and a single platform_driver.  It's neater and
makes the device tree addition easier and more consistent.  Rename the
match values to be inline with what they'll be in the device tree
bindings.  There aren't any current in-tree users of the existing device
names.

Cc: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: Jamie Iles <jamie@jamieiles.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agocrypto: padlock-aes - Make module loading even quieter when hardware is missing
Jonathan Nieder [Wed, 3 Aug 2011 06:07:11 +0000 (14:07 +0800)]
crypto: padlock-aes - Make module loading even quieter when hardware is missing

When loading aes via the module alias, a padlock module failing to
load due to missing hardware is not particularly notable.  With
v2.6.27-rc1~1107^2~14 (crypto: padlock - Make module loading quieter
when hardware isn't available, 2008-07-03), the padlock-aes module
suppresses the relevant messages when the "quiet" flag is in use; but
better to suppress this particular message completely, since the
administrator can already distinguish such errors by the absence of a
message indicating initialization failing or succeeding.

This avoids occasional messages in syslog of the form

padlock_aes: VIA PadLock not detected.

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
13 years agooom: task->mm == NULL doesn't mean the memory was freed
Oleg Nesterov [Sat, 30 Jul 2011 14:35:02 +0000 (16:35 +0200)]
oom: task->mm == NULL doesn't mean the memory was freed

exit_mm() sets ->mm == NULL then it does mmput()->exit_mmap() which
frees the memory.

However select_bad_process() checks ->mm != NULL before TIF_MEMDIE,
so it continues to kill other tasks even if we have the oom-killed
task freeing its memory.

Change select_bad_process() to check ->mm after TIF_MEMDIE, but skip
the tasks which have already passed exit_notify() to ensure a zombie
with TIF_MEMDIE set can't block oom-killer. Alternatively we could
probably clear TIF_MEMDIE after exit_mmap().

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/lrg/voltage-2.6
Linus Torvalds [Tue, 2 Aug 2011 00:05:46 +0000 (14:05 -1000)]
Merge branch 'for-linus' of git://git./linux/kernel/git/lrg/voltage-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/lrg/voltage-2.6: (23 commits)
  regulator: Improve WM831x DVS VSEL selection algorithm
  regulator: Bootstrap wm831x DVS VSEL value from ON VSEL if not already set
  regulator: Set up GPIO for WM831x VSEL before enabling VSEL mode
  regulator: Add EPEs to the MODULE_ALIAS() for wm831x-dcdc
  regulator: Fix WM831x DCDC DVS VSEL bootstrapping
  regulator: Fix WM831x regulator ID lookups for multiple WM831xs
  regulator: Fix argument format type errors in error prints
  regulator: Fix memory leak in set_machine_constraints() error paths
  regulator: Make core more chatty about some errors
  regulator: tps65910: Fix array access out of bounds bug
  regulator: tps65910: Add missing breaks in switch/case
  regulator: tps65910: Fix a memory leak in tps65910_probe error path
  regulator: TWL: Remove entry of RES_ID for 6030 macros
  ASoC: tlv320aic3x: Add correct hw registers to Line1 cross connect muxes
  regulator: Add basic per consumer debugfs
  regulator: Add rdev_crit() macro
  regulator: Refactor supply implementation to work as regular consumers
  regulator: Include the device name in the microamps_requested_ file
  regulator: Increase the limit on sysfs file names
  regulator: Properly register dummy regulator driver
  ...

13 years agoMerge branch 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4
Linus Torvalds [Mon, 1 Aug 2011 23:56:03 +0000 (13:56 -1000)]
Merge branch 'for_linus' of git://git./linux/kernel/git/tytso/ext4

* 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4: (60 commits)
  ext4: prevent memory leaks from ext4_mb_init_backend() on error path
  ext4: use EXT4_BAD_INO for buddy cache to avoid colliding with valid inode #
  ext4: use ext4_msg() instead of printk in mballoc
  ext4: use ext4_kvzalloc()/ext4_kvmalloc() for s_group_desc and s_group_info
  ext4: introduce ext4_kvmalloc(), ext4_kzalloc(), and ext4_kvfree()
  ext4: use the correct error exit path in ext4_init_inode_table()
  ext4: add missing kfree() on error return path in add_new_gdb()
  ext4: change umode_t in tracepoint headers to be an explicit __u16
  ext4: fix races in ext4_sync_parent()
  ext4: Fix overflow caused by missing cast in ext4_fallocate()
  ext4: add action of moving index in ext4_ext_rm_idx for Punch Hole
  ext4: simplify parameters of reserve_backup_gdb()
  ext4: simplify parameters of add_new_gdb()
  ext4: remove lock_buffer in bclean() and setup_new_group_blocks()
  ext4: simplify journal handling in setup_new_group_blocks()
  ext4: let setup_new_group_blocks() set multiple bits at a time
  ext4: fix a typo in ext4_group_extend()
  ext4: let ext4_group_add_blocks() handle 0 blocks quickly
  ext4: let ext4_group_add_blocks() return an error code
  ext4: rename ext4_add_groupblocks() to ext4_group_add_blocks()
  ...

Fix up conflict in fs/ext4/inode.c: commit aacfc19c626e ("fs: simplify
the blockdev_direct_IO prototype") had changed the ext4_ind_direct_IO()
function for the new simplified calling convention, while commit
dae1e52cb126 ("ext4: move ext4_ind_* functions from inode.c to
indirect.c") moved the function to another file.

13 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs-2.6
Linus Torvalds [Mon, 1 Aug 2011 23:48:31 +0000 (13:48 -1000)]
Merge branch 'for-linus' of git://git./linux/kernel/git/viro/vfs-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs-2.6:
  xfs: Fix build breakage in xfs_iops.c when CONFIG_FS_POSIX_ACL is not set
  VFS: Reorganise shrink_dcache_for_umount_subtree() after demise of dcache_lock
  VFS: Remove dentry->d_lock locking from shrink_dcache_for_umount_subtree()
  VFS: Remove detached-dentry counter from shrink_dcache_for_umount_subtree()
  switch posix_acl_chmod() to umode_t
  switch posix_acl_from_mode() to umode_t
  switch posix_acl_equiv_mode() to umode_t *
  switch posix_acl_create() to umode_t *
  block: initialise bd_super in bdget()
  vfs: avoid call to inode_lru_list_del() if possible
  vfs: avoid taking inode_hash_lock on pipes and sockets
  vfs: conditionally call inode_wb_list_del()
  VFS: Fix automount for negative autofs dentries
  Btrfs: load the key from the dir item in readdir into a fake dentry
  devtmpfs: missing initialialization in never-hit case
  hppfs: missing include

13 years agoMerge branch 'for-linus' of git://git.infradead.org/users/vkoul/slave-dma
Linus Torvalds [Mon, 1 Aug 2011 23:46:37 +0000 (13:46 -1000)]
Merge branch 'for-linus' of git://git.infradead.org/users/vkoul/slave-dma

* 'for-linus' of git://git.infradead.org/users/vkoul/slave-dma: (37 commits)
  Improve slave/cyclic DMA engine documentation
  dmaengine: pl08x: handle the rest of enums in pl08x_width
  DMA: PL08x: cleanup selection of burst size
  DMA: PL08x: avoid recalculating cctl at each prepare
  DMA: PL08x: cleanup selection of buswidth
  DMA: PL08x: constify plchan->cd and plat->slave_channels
  DMA: PL08x: separately store source/destination cctl
  DMA: PL08x: separately store source/destination slave address
  DMA: PL08x: clean up LLI debugging
  DMA: PL08x: select LLI bus only once per LLI setup
  DMA: PL08x: remove unused constants
  ARM: mxs-dma: reset after disable channel
  dma: intel_mid_dma: remove redundant pci_set_drvdata calls
  dma: mxs-dma: fix unterminated platform_device_id table
  dmaengine: pl330: make platform data optional
  dmaengine: imx-sdma: return proper error if kzalloc fails
  pch_dma: Fix CTL register access issue
  dmaengine: mxs-dma: skip request_irq for NO_IRQ
  dmaengine/coh901318: fix slave submission semantics
  dmaengine/ste_dma40: allow memory buswidth/burst to be configured
  ...

Fix trivial whitespace conflict in drivers/dma/mv_xor.c

13 years agoMerge branch 'gpiolib' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6
Linus Torvalds [Mon, 1 Aug 2011 23:41:43 +0000 (13:41 -1000)]
Merge branch 'gpiolib' of git://git./linux/kernel/git/aegl/linux-2.6

* 'gpiolib' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6:
  [IA64] Hook up gpiolib support

13 years agoMerge branch 'pstore-efi' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6
Linus Torvalds [Mon, 1 Aug 2011 23:40:51 +0000 (13:40 -1000)]
Merge branch 'pstore-efi' of git://git./linux/kernel/git/aegl/linux-2.6

* 'pstore-efi' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6:
  efivars: Introduce PSTORE_EFI_ATTRIBUTES
  efivars: Use string functions in pstore_write
  efivars: introduce utf16_strncmp
  efivars: String functions
  efi: Add support for using efivars as a pstore backend
  pstore: Allow the user to explicitly choose a backend
  pstore: Make "part" unsigned
  pstore: Add extra context for writes and erases
  pstore: Extend API for more flexibility in new backends

13 years agoMerge branch 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jwessel...
Linus Torvalds [Mon, 1 Aug 2011 23:39:40 +0000 (13:39 -1000)]
Merge branch 'for_linus' of git://git./linux/kernel/git/jwessel/linux-2.6-kgdb

* 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jwessel/linux-2.6-kgdb:
  kdb,kgdb: Allow arbitrary kgdb magic knock sequences
  kdb: Remove all references to DOING_KGDB2
  kdb,kgdb: Implement switch and pass buffer from kdb -> gdb
  kdb: cleanup unused variables missed in the original kdb merge

13 years agoext4: prevent memory leaks from ext4_mb_init_backend() on error path
Yu Jian [Mon, 1 Aug 2011 21:41:46 +0000 (17:41 -0400)]
ext4: prevent memory leaks from ext4_mb_init_backend() on error path

In ext4_mb_init(), if the s_locality_group allocation fails it will
currently cause the allocations made in ext4_mb_init_backend() to
be leaked.  Moving the ext4_mb_init_backend() allocation after the
s_locality_group allocation avoids that problem.

Signed-off-by: Yu Jian <yujian@whamcloud.com>
Signed-off-by: Andreas Dilger <adilger@whamcloud.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
13 years agoext4: use EXT4_BAD_INO for buddy cache to avoid colliding with valid inode #
Yu Jian [Mon, 1 Aug 2011 21:41:39 +0000 (17:41 -0400)]
ext4: use EXT4_BAD_INO for buddy cache to avoid colliding with valid inode #

Signed-off-by: Yu Jian <yujian@whamcloud.com>
Signed-off-by: Andreas Dilger <adilger@whamcloud.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
13 years agoext4: use ext4_msg() instead of printk in mballoc
Theodore Ts'o [Mon, 1 Aug 2011 21:41:35 +0000 (17:41 -0400)]
ext4: use ext4_msg() instead of printk in mballoc

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
13 years agokdb,kgdb: Allow arbitrary kgdb magic knock sequences
Jason Wessel [Tue, 24 May 2011 15:43:06 +0000 (10:43 -0500)]
kdb,kgdb: Allow arbitrary kgdb magic knock sequences

The first packet that gdb sends when the kernel is in kdb mode seems
to change with every release of gdb.  Instead of continuing to add
many different gdb packets, change kdb to automatically look for any
thing that looks like a gdb packet.

Example 1 cold start test:
echo g > /proc/sysrq-trigger
$D#44+

Example 2 cold start test:
echo g > /proc/sysrq-trigger
$3#33

The second one should re-enter kdb's shell right away and is purely a
test.

Signed-off-by: Jason Wessel <jason.wessel@windriver.com>
13 years agokdb: Remove all references to DOING_KGDB2
Jason Wessel [Mon, 23 May 2011 18:22:54 +0000 (13:22 -0500)]
kdb: Remove all references to DOING_KGDB2

The DOING_KGDB2 was originally a state variable for one of the two
ways to automatically transition from kdb to kgdb.  Purge all these
variables and just use one single state for the transition.

Signed-off-by: Jason Wessel <jason.wessel@windriver.com>
13 years agokdb,kgdb: Implement switch and pass buffer from kdb -> gdb
Jason Wessel [Mon, 23 May 2011 18:17:41 +0000 (13:17 -0500)]
kdb,kgdb: Implement switch and pass buffer from kdb -> gdb

When switching from kdb mode to kgdb mode packets were getting lost
depending on the size of the fifo queue of the serial chip.  When gdb
initially connects if it is in kdb mode it should entirely send any
character buffer over to the gdbstub when switching connections.

Previously kdb was zero'ing out the character buffer and this could
lead to gdb failing to connect at all, or a lengthy pause could occur
on the initial connect.

Signed-off-by: Jason Wessel <jason.wessel@windriver.com>
13 years agokdb: cleanup unused variables missed in the original kdb merge
Jason Wessel [Thu, 30 Jun 2011 19:12:00 +0000 (14:12 -0500)]
kdb: cleanup unused variables missed in the original kdb merge

The BTARGS and BTSYMARG variables do not have any function in the
mainline version of kdb.

Reported-by: Tim Bird <tim.bird@am.sony.com>
Signed-off-by: Jason Wessel <jason.wessel@windriver.com>
13 years agoMerge branch 'for-next' into for-linus
Liam Girdwood [Mon, 1 Aug 2011 17:25:06 +0000 (18:25 +0100)]
Merge branch 'for-next' into for-linus

13 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/sfrench/cifs-2.6
Linus Torvalds [Mon, 1 Aug 2011 16:14:25 +0000 (06:14 -1000)]
Merge git://git./linux/kernel/git/sfrench/cifs-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/sfrench/cifs-2.6:
  CIFS: Cleanup demupltiplex thread exiting code
  CIFS: Move mid search to a separate function
  CIFS: Move RFC1002 check to a separate function
  CIFS: Simplify socket reading in demultiplex thread
  CIFS: Move buffer allocation to a separate function
  cifs: remove unneeded variable initialization in cifs_reconnect_tcon
  cifs: simplify refcounting for oplock breaks
  cifs: fix compiler warning in CIFSSMBQAllEAs
  cifs: fix name parsing in CIFSSMBQAllEAs
  cifs: don't start signing too early
  cifs: trivial: goto out here is unnecessary
  cifs: advertise the right receive buffer size to the server

13 years agoMerge branch 'gpio/next' of git://git.secretlab.ca/git/linux-2.6
Linus Torvalds [Mon, 1 Aug 2011 16:13:48 +0000 (06:13 -1000)]
Merge branch 'gpio/next' of git://git.secretlab.ca/git/linux-2.6

* 'gpio/next' of git://git.secretlab.ca/git/linux-2.6:
  gpio_msm: Move Qualcomm MSM v2 gpio driver into drivers
  gpio_msm: Move Qualcomm v6 MSM driver into drivers
  msm: gpio: Fold register defs into C file
  msm: gpiomux: Move public API to public header
  msm: gpio: Remove ifdefs on gpio chip registers
  msm: gpio: Remove chip-specific register definitions
  msm: Remove chip-ifdefs for GPIO io mappings
  msm: gpio: Remove unsupported devices
  gpio: ab8500: fix MODULE_ALIAS for ab8500
  of/gpio: export of_gpio_simple_xlate

13 years agoMerge branch 'for-3.1-rc1' of git://gitorious.org/linux-omap-dss2/linux
Linus Torvalds [Mon, 1 Aug 2011 16:13:05 +0000 (06:13 -1000)]
Merge branch 'for-3.1-rc1' of git://gitorious.org/linux-omap-dss2/linux

* 'for-3.1-rc1' of git://gitorious.org/linux-omap-dss2/linux: (31 commits)
  OMAP: DSS2: HDMI: fix hdmi clock name
  HACK: OMAP: DSS2: clk hack for OMAP2/3
  OMAP: DSS2: DSS: Fix context save/restore
  OMAP: DSS2: DISPC: Fix context save/restore
  OMAP: DSS2: Remove ctx loss count from dss.c
  OMAP: DSS2: Remove unused code from display.c
  OMAP: DSS2: DISPC: remove finegrained clk enables/disables
  OMAP: DSS2: Remove unused opt_clock_available
  OMAP: DSS2: Use PM runtime & HWMOD support
  OMAP: DSS2: Remove CONFIG_OMAP2_DSS_SLEEP_BEFORE_RESET
  OMAP: DSS2: Remove core_dump_clocks
  OMAP: DSS2: DPI: remove unneeded SYSCK enable/disable
  OMAP: DSS2: Use omap_pm_get_dev_context_loss_count to get ctx loss count
  OMAP: DSS2: rewrite use of context_loss_count
  OMAP: DSS2: Remove clk optimization at dss init
  OMAP: DSS2: Fix init and unit sequence
  OMAP: DSS2: Clean up probe for DSS & DSI
  OMAP: DSS2: Handle dpll4_m4_ck in dss_get/put_clocks
  OMAP: DSS2: Fix FIFO threshold and burst size for OMAP4
  OMAP: DSS2: DSI: sync when disabling a display
  ...

13 years agoMerge branch 'sh-latest' of git://git.kernel.org/pub/scm/linux/kernel/git/lethal...
Linus Torvalds [Mon, 1 Aug 2011 16:10:16 +0000 (06:10 -1000)]
Merge branch 'sh-latest' of git://git./linux/kernel/git/lethal/sh-3.x

* 'sh-latest' of git://git.kernel.org/pub/scm/linux/kernel/git/lethal/sh-3.x: (39 commits)
  SH: static should be at beginning of declaration
  sh: move CLKDEV_xxx_ID macro to sh_clk.h
  sh: clock-shx3: add CLKDEV_ICK_ID for cleanup
  sh: clock-sh7786: add CLKDEV_ICK_ID for cleanup
  sh: clock-sh7785: add CLKDEV_ICK_ID for cleanup
  sh: clock-sh7757: add CLKDEV_ICK_ID for cleanup
  sh: clock-sh7366: add CLKDEV_ICK_ID for cleanup
  sh: clock-sh7343: add CLKDEV_ICK_ID for cleanup
  sh: clock-sh7722: add CLKDEV_ICK_ID for cleanup
  sh: clock-sh7724: add CLKDEV_ICK_ID for cleanup
  sh: clock-sh7366: modify I2C clock settings
  sh: clock-sh7343: modify I2C clock settings
  sh: clock-sh7723: modify I2C clock settings
  sh: clock-sh7722: modify I2C clock settings
  sh: clock-sh7724: modify I2C clock settings
  serial: sh-sci: Fix up pretty name printing for port IRQs.
  serial: sh-sci: Kill off per-port enable/disable callbacks.
  serial: sh-sci: Add missing module description/author bits.
  serial: sh-sci: Regtype probing doesn't need to be fatal.
  sh: Tidy up pre-clkdev clk_get() error handling.
  ...

13 years agoMerge branch 'msm-move-gpio' of git://codeaurora.org/quic/kernel/davidb/linux-msm...
Grant Likely [Mon, 1 Aug 2011 14:16:05 +0000 (15:16 +0100)]
Merge branch 'msm-move-gpio' of git://codeaurora.org/quic/kernel/davidb/linux-msm into gpio/next

Conflicts:
drivers/gpio/Kconfig
drivers/gpio/Makefile

13 years agoCIFS: Cleanup demupltiplex thread exiting code
Pavel Shilovsky [Mon, 1 Aug 2011 09:19:44 +0000 (13:19 +0400)]
CIFS: Cleanup demupltiplex thread exiting code

Reviewed-and-Tested-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Pavel Shilovsky <piastryyy@gmail.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>
13 years agoCIFS: Move mid search to a separate function
Pavel Shilovsky [Mon, 1 Aug 2011 09:19:43 +0000 (13:19 +0400)]
CIFS: Move mid search to a separate function

Reviewed-and-Tested-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Pavel Shilovsky <piastryyy@gmail.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>
13 years agoCIFS: Move RFC1002 check to a separate function
Pavel Shilovsky [Mon, 1 Aug 2011 09:19:42 +0000 (13:19 +0400)]
CIFS: Move RFC1002 check to a separate function

Reviewed-and-Tested-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Pavel Shilovsky <piastryyy@gmail.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>
13 years agoCIFS: Simplify socket reading in demultiplex thread
Pavel Shilovsky [Mon, 1 Aug 2011 09:19:41 +0000 (13:19 +0400)]
CIFS: Simplify socket reading in demultiplex thread

Move reading to separate function and remove csocket variable.

Also change semantic in a little: goto incomplete_rcv only when
we get -EAGAIN (or a familiar error) while reading rfc1002 header.
In this case we don't check for echo timeout when we don't get whole
header at once, as it was before.

Reviewed-and-Tested-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Pavel Shilovsky <piastryyy@gmail.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>
13 years agoext4: use ext4_kvzalloc()/ext4_kvmalloc() for s_group_desc and s_group_info
Theodore Ts'o [Mon, 1 Aug 2011 12:45:38 +0000 (08:45 -0400)]
ext4: use ext4_kvzalloc()/ext4_kvmalloc() for s_group_desc and s_group_info

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
13 years agoext4: introduce ext4_kvmalloc(), ext4_kzalloc(), and ext4_kvfree()
Theodore Ts'o [Mon, 1 Aug 2011 12:45:02 +0000 (08:45 -0400)]
ext4: introduce ext4_kvmalloc(), ext4_kzalloc(), and ext4_kvfree()

Introduce new helper functions which try kmalloc, and then fall back
to vmalloc if necessary, and use them for allocating and deallocating
s_flex_groups.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
13 years agoCIFS: Move buffer allocation to a separate function
Pavel Shilovsky [Mon, 1 Aug 2011 09:19:40 +0000 (13:19 +0400)]
CIFS: Move buffer allocation to a separate function

Reviewed-and-Tested-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Pavel Shilovsky <piastryyy@gmail.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>
13 years agogpio_msm: Move Qualcomm MSM v2 gpio driver into drivers
David Brown [Mon, 16 May 2011 22:53:38 +0000 (15:53 -0700)]
gpio_msm: Move Qualcomm MSM v2 gpio driver into drivers

Migrate the driver for the v7-based MSM chips into drivers/gpio.  The
driver is unchanged, only moved.

Change-Id: I810db5b50b71cdca4e869aa0d0310f7f48781a55
Signed-off-by: David Brown <davidb@codeaurora.org>
Acked-by: Linus Walleij <linus.walleij@linaro.org>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
13 years agogpio_msm: Move Qualcomm v6 MSM driver into drivers
David Brown [Mon, 16 May 2011 20:57:39 +0000 (13:57 -0700)]
gpio_msm: Move Qualcomm v6 MSM driver into drivers

Migrate the driver for the v6-based MSM chips into drivers/gpio.  The
driver is unchanged, only moved.

Change-Id: I03ba597b95b4d62b42da112a8efac88d67aa40f9
Signed-off-by: David Brown <davidb@codeaurora.org>
Acked-by: Linus Walleij <linus.walleij@linaro.org>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
13 years agomsm: gpio: Fold register defs into C file
David Brown [Mon, 6 Jun 2011 21:09:45 +0000 (14:09 -0700)]
msm: gpio: Fold register defs into C file

No need to have a separate header file containing only register
definitions that are used by a single driver.  Fold these into the
gpio driver.

Signed-off-by: David Brown <davidb@codeaurora.org>
13 years agomsm: gpiomux: Move public API to public header
David Brown [Mon, 16 May 2011 20:32:15 +0000 (13:32 -0700)]
msm: gpiomux: Move public API to public header

The gpiomux.h header contains some SOC ifdefs.  However, the API that
is actually used by the GPIO driver only uses two functions that are
general.  Move these general definitions into a public header file.

Change-Id: Ia5df8af87dba268225598d56908e523bcfc24ef6
Signed-off-by: David Brown <davidb@codeaurora.org>
Acked-by: Linus Walleij <linus.walleij@linaro.org>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
13 years agomsm: gpio: Remove ifdefs on gpio chip registers
David Brown [Thu, 12 May 2011 08:28:01 +0000 (01:28 -0700)]
msm: gpio: Remove ifdefs on gpio chip registers

Select the GPIO register configuration at runtime rather than through
idefs.

Change-Id: I02ea0a3d61bc81669f32097c32420f0688552231
Signed-off-by: David Brown <davidb@codeaurora.org>
Acked-by: Linus Walleij <linus.walleij@linaro.org>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
13 years agomsm: gpio: Remove chip-specific register definitions
David Brown [Thu, 12 May 2011 08:16:46 +0000 (01:16 -0700)]
msm: gpio: Remove chip-specific register definitions

Put an SOC prefix on each GPIO register definition, eliminating the
need to have SOC ifdefs around the definitions.

Change-Id: I5a01fd328a89ce1be610847934d6e118f5465e42
Signed-off-by: David Brown <davidb@codeaurora.org>
Acked-by: Linus Walleij <linus.walleij@linaro.org>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
13 years agomsm: Remove chip-ifdefs for GPIO io mappings
David Brown [Thu, 12 May 2011 07:54:36 +0000 (00:54 -0700)]
msm: Remove chip-ifdefs for GPIO io mappings

The two GPIO controllers are always mapped to the same virtual address
across all MSM devices.  Instead of selecting this at compile time,
determine the physical address at runtime, eliminating yet something
else preventing multiple MSM targets from being compiled into the same
kernel.

Change-Id: I1672219d978ab6243526adeda6badf49472baa27
Signed-off-by: David Brown <davidb@codeaurora.org>
Acked-by: Linus Walleij <linus.walleij@linaro.org>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
13 years agomsm: gpio: Remove unsupported devices
David Brown [Thu, 12 May 2011 07:33:38 +0000 (00:33 -0700)]
msm: gpio: Remove unsupported devices

The MSM7x25 and MSM7x27 devices are not yet supported in the kernel.
Remove #ifdef-based tables supporting these chips for now.

Change-Id: I4d9f5abc4cc0942ce75a067097b072489493c1b8
Signed-off-by: David Brown <davidb@codeaurora.org>
Acked-by: Linus Walleij <linus.walleij@linaro.org>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
13 years agoOMAP: DSS2: HDMI: fix hdmi clock name
Tomi Valkeinen [Mon, 1 Aug 2011 11:54:44 +0000 (14:54 +0300)]
OMAP: DSS2: HDMI: fix hdmi clock name

The HDMI clock (hdmi_clk) is missing in the current OMAP4 HWMOD
database. Fix this in the DSS driver by using the old clock name
(dss_48mhz_clk).

Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
13 years agoHACK: OMAP: DSS2: clk hack for OMAP2/3
Tomi Valkeinen [Mon, 1 Aug 2011 11:32:23 +0000 (14:32 +0300)]
HACK: OMAP: DSS2: clk hack for OMAP2/3

The HWMOD data for OMAP2 and 3 are currently not up to date regarding
DSS (OMAP4 HWMOD data is fine). This patch makes the DSS driver to get
the opt clocks needed for OMAP2/3 with the old clock names, thus
allowing DSS driver to use runtime PM.

The HWMOD databases should be fixes ASAP, and this patch can be reverted
after that.

Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
13 years agoext4: use the correct error exit path in ext4_init_inode_table()
Yongqiang Yang [Mon, 1 Aug 2011 10:32:19 +0000 (06:32 -0400)]
ext4: use the correct error exit path in ext4_init_inode_table()

This patch lets ext4_init_inode_table() handle errors right.
ext4_init_inode_table() should down_write() alloc_sem which
has been up_write()ed and stop the started journal handle.

Signed-off-by: Yongqiang Yang <xiaoqiangnk@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
13 years agoxfs: Fix build breakage in xfs_iops.c when CONFIG_FS_POSIX_ACL is not set
Markus Trippelsdorf [Tue, 26 Jul 2011 09:15:20 +0000 (11:15 +0200)]
xfs: Fix build breakage in xfs_iops.c when CONFIG_FS_POSIX_ACL is not set

commit 4e34e719e45, that takes the ACL checks to common code,
accidentely broke the build when CONFIG_FS_POSIX_ACL is not set:

  CC      fs/xfs/linux-2.6/xfs_iops.o
fs/xfs/linux-2.6/xfs_iops.c:1025:14: error: ‘xfs_get_acl’ undeclared here (not in a function)

Fix this by declaring xfs_get_acl a static inline function.

Signed-off-by: Markus Trippelsdorf <markus@trippelsdorf.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
13 years agoVFS: Reorganise shrink_dcache_for_umount_subtree() after demise of dcache_lock
David Howells [Tue, 7 Jun 2011 13:09:30 +0000 (14:09 +0100)]
VFS: Reorganise shrink_dcache_for_umount_subtree() after demise of dcache_lock

Reorganise shrink_dcache_for_umount_subtree() in light of the demise of
dcache_lock.  Without that dcache_lock, there is no need for the batching of
removal of dentries from the system under it (we wanted to make intensive use
of the locked data whilst we held it, but didn't want to hold it for long at a
time).

This works, provided the preceding patch is correct in its removal of locking
on dentry->d_lock on the basis that no one should be locking these dentries any
more as the whole superblock is defunct.

With this patch, the calls to dentry_lru_del() and __d_shrink() are placed at
the point where each dentry is detached handled.

It is possible that, as an alternative, the batching should still be done -
but only for dentry_lru_del() of all a dentry's children in one go.  In such a
case, the batching would be done under dcache_lru_lock.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
13 years agoVFS: Remove dentry->d_lock locking from shrink_dcache_for_umount_subtree()
David Howells [Tue, 7 Jun 2011 13:09:20 +0000 (14:09 +0100)]
VFS: Remove dentry->d_lock locking from shrink_dcache_for_umount_subtree()

Locks of the dcache_lock were replaced by locks of dentry->d_lock in commits
such as:

2304450783dfde7b0b94ae234edd0dbffa865073
2fd6b7f50797f2e993eea59e0a0b8c6399c811dc

as part of the RCU-based pathwalk changes, despite the fact that the caller
(shrink_dcache_for_umount()) notes in the banner comment the reasons that
d_lock is not necessary in these functions:

/*
 * destroy the dentries attached to a superblock on unmounting
 * - we don't need to use dentry->d_lock because:
 *   - the superblock is detached from all mountings and open files, so the
 *     dentry trees will not be rearranged by the VFS
 *   - s_umount is write-locked, so the memory pressure shrinker will ignore
 *     any dentries belonging to this superblock that it comes across
 *   - the filesystem itself is no longer permitted to rearrange the dentries
 *     in this superblock
 */

So remove these locks.  If the locks are actually necessary, then this banner
comment should be altered instead.

The hash table chains are protected by 1-bit locks in the hash table heads, so
those shouldn't be a problem.

Note that to make this work, __d_drop() has to be split so that the RCUwalk
barrier can be avoided.  This causes problems otherwise as it has an assertion
that dentry->d_lock is locked - but there is no need for that as no one else
can be trying to access this dentry, except to step over it (and that should
be handled by d_free(), I think).

Signed-off-by: David Howells <dhowells@redhat.com>
Cc: Nick Piggin <npiggin@kernel.dk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
13 years agoVFS: Remove detached-dentry counter from shrink_dcache_for_umount_subtree()
David Howells [Tue, 7 Jun 2011 13:09:10 +0000 (14:09 +0100)]
VFS: Remove detached-dentry counter from shrink_dcache_for_umount_subtree()

Remove the detached-dentry counter from shrink_dcache_for_umount_subtree() as
the value it computes is no longer used as of commit
312d3ca856d369bb04d0443846b85b4cdde6fa8a which made the nr_dentry counters
summed per-CPU rather than global atomic.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
13 years agoswitch posix_acl_chmod() to umode_t
Al Viro [Sat, 23 Jul 2011 23:03:11 +0000 (19:03 -0400)]
switch posix_acl_chmod() to umode_t

again, that's what all callers pass to it

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
13 years agoswitch posix_acl_from_mode() to umode_t
Al Viro [Sat, 23 Jul 2011 23:01:48 +0000 (19:01 -0400)]
switch posix_acl_from_mode() to umode_t

... seeing that this is what all callers pass to it anyway.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
13 years agoswitch posix_acl_equiv_mode() to umode_t *
Al Viro [Sat, 23 Jul 2011 22:56:36 +0000 (18:56 -0400)]
switch posix_acl_equiv_mode() to umode_t *

... so that &inode->i_mode could be passed to it

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
13 years agoswitch posix_acl_create() to umode_t *
Al Viro [Sat, 23 Jul 2011 22:37:50 +0000 (18:37 -0400)]
switch posix_acl_create() to umode_t *

so we can pass &inode->i_mode to it

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
13 years agoblock: initialise bd_super in bdget()
Lachlan McIlroy [Thu, 30 Jun 2011 01:01:45 +0000 (11:01 +1000)]
block: initialise bd_super in bdget()

bd_super is currently reset to NULL in kill_block_super() so we rely on previous
users of the block_device object to initialise this value for the next user.
This quirk was exposed on RHEL5 when a third party filesystem did not always use
kill_block_super() and therefore bd_super wasn't being reset when a block_device
object was recycled within the cache.  This may not be a problem upstream but
makes sense to be defensive.

Signed-off-by: Lachlan McIlroy <lmcilroy@redhat.com>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
13 years agovfs: avoid call to inode_lru_list_del() if possible
Eric Dumazet [Thu, 28 Jul 2011 04:55:13 +0000 (06:55 +0200)]
vfs: avoid call to inode_lru_list_del() if possible

inode_lru_list_del() is expensive because of per superblock lru locking,
while some inodes are not in lru list.

Adding a check in iput_final() can speedup pipe/sockets workloads on
SMP.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
13 years agovfs: avoid taking inode_hash_lock on pipes and sockets
Eric Dumazet [Thu, 28 Jul 2011 04:41:09 +0000 (06:41 +0200)]
vfs: avoid taking inode_hash_lock on pipes and sockets

Some inodes (pipes, sockets, ...) are not hashed, no need to take
contended inode_hash_lock at dismantle time.

nice speedup on SMP machines on socket intensive workloads.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
13 years agovfs: conditionally call inode_wb_list_del()
Eric Dumazet [Thu, 28 Jul 2011 04:11:47 +0000 (06:11 +0200)]
vfs: conditionally call inode_wb_list_del()

Some inodes (pipes, sockets, ...) are not in bdi writeback list.

evict() can avoid calling inode_wb_list_del() and its expensive spinlock
by checking inode i_wb_list being empty or not.

At this point, no other cpu/user can concurrently manipulate this inode
i_wb_list

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>