android_kernel_samsung_msm8976/crypto/Makefile

107 lines
3.8 KiB
Makefile
Raw Normal View History

#
# Cryptographic API
#
obj-$(CONFIG_CRYPTO) += crypto.o
crypto: crypto_memneq - add equality testing of memory regions w/o timing leaks commit 6bf37e5aa90f18baf5acf4874bca505dd667c37f upstream. When comparing MAC hashes, AEAD authentication tags, or other hash values in the context of authentication or integrity checking, it is important not to leak timing information to a potential attacker, i.e. when communication happens over a network. Bytewise memory comparisons (such as memcmp) are usually optimized so that they return a nonzero value as soon as a mismatch is found. E.g, on x86_64/i5 for 512 bytes this can be ~50 cyc for a full mismatch and up to ~850 cyc for a full match (cold). This early-return behavior can leak timing information as a side channel, allowing an attacker to iteratively guess the correct result. This patch adds a new method crypto_memneq ("memory not equal to each other") to the crypto API that compares memory areas of the same length in roughly "constant time" (cache misses could change the timing, but since they don't reveal information about the content of the strings being compared, they are effectively benign). Iow, best and worst case behaviour take the same amount of time to complete (in contrast to memcmp). Note that crypto_memneq (unlike memcmp) can only be used to test for equality or inequality, NOT for lexicographical order. This, however, is not an issue for its use-cases within the crypto API. We tried to locate all of the places in the crypto API where memcmp was being used for authentication or integrity checking, and convert them over to crypto_memneq. crypto_memneq is declared noinline, placed in its own source file, and compiled with optimizations that might increase code size disabled ("Os") because a smart compiler (or LTO) might notice that the return value is always compared against zero/nonzero, and might then reintroduce the same early-return optimization that we are trying to avoid. Using #pragma or __attribute__ optimization annotations of the code for disabling optimization was avoided as it seems to be considered broken or unmaintained for long time in GCC [1]. Therefore, we work around that by specifying the compile flag for memneq.o directly in the Makefile. We found that this seems to be most appropriate. As we use ("Os"), this patch also provides a loop-free "fast-path" for frequently used 16 byte digests. Similarly to kernel library string functions, leave an option for future even further optimized architecture specific assembler implementations. This was a joint work of James Yonan and Daniel Borkmann. Also thanks for feedback from Florian Weimer on this and earlier proposals [2]. [1] http://gcc.gnu.org/ml/gcc/2012-07/msg00211.html [2] https://lkml.org/lkml/2013/2/10/131 Signed-off-by: James Yonan <james@openvpn.net> Signed-off-by: Daniel Borkmann <dborkman@redhat.com> Cc: Florian Weimer <fw@deneb.enyo.de> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Cc: Jason A. Donenfeld <Jason@zx2c4.com> Signed-off-by: Willy Tarreau <w@1wt.eu>
2013-09-26 08:20:39 +00:00
crypto-y := api.o cipher.o compress.o memneq.o
obj-$(CONFIG_CRYPTO_WORKQUEUE) += crypto_wq.o
obj-$(CONFIG_CRYPTO_FIPS) += fips.o
crypto_algapi-$(CONFIG_PROC_FS) += proc.o
crypto_algapi-y := algapi.o scatterwalk.o $(crypto_algapi-y)
obj-$(CONFIG_CRYPTO_ALGAPI2) += crypto_algapi.o
obj-$(CONFIG_CRYPTO_AEAD2) += aead.o
crypto_blkcipher-y := ablkcipher.o
crypto_blkcipher-y += blkcipher.o
obj-$(CONFIG_CRYPTO_BLKCIPHER2) += crypto_blkcipher.o
obj-$(CONFIG_CRYPTO_BLKCIPHER2) += chainiv.o
obj-$(CONFIG_CRYPTO_BLKCIPHER2) += eseqiv.o
obj-$(CONFIG_CRYPTO_SEQIV) += seqiv.o
crypto_hash-y += ahash.o
crypto_hash-y += shash.o
obj-$(CONFIG_CRYPTO_HASH2) += crypto_hash.o
obj-$(CONFIG_CRYPTO_PCOMP2) += pcompress.o
cryptomgr-y := algboss.o testmgr.o
obj-$(CONFIG_CRYPTO_MANAGER2) += cryptomgr.o
obj-$(CONFIG_CRYPTO_USER) += crypto_user.o
obj-$(CONFIG_CRYPTO_CMAC) += cmac.o
obj-$(CONFIG_CRYPTO_HMAC) += hmac.o
obj-$(CONFIG_CRYPTO_VMAC) += vmac.o
obj-$(CONFIG_CRYPTO_XCBC) += xcbc.o
obj-$(CONFIG_CRYPTO_NULL) += crypto_null.o
obj-$(CONFIG_CRYPTO_MD4) += md4.o
obj-$(CONFIG_CRYPTO_MD5) += md5.o
obj-$(CONFIG_CRYPTO_RMD128) += rmd128.o
obj-$(CONFIG_CRYPTO_RMD160) += rmd160.o
obj-$(CONFIG_CRYPTO_RMD256) += rmd256.o
obj-$(CONFIG_CRYPTO_RMD320) += rmd320.o
obj-$(CONFIG_CRYPTO_SHA1) += sha1_generic.o
obj-$(CONFIG_CRYPTO_SHA256) += sha256_generic.o
obj-$(CONFIG_CRYPTO_SHA512) += sha512_generic.o
obj-$(CONFIG_CRYPTO_WP512) += wp512.o
crypto: improve gcc optimization flags for serpent and wp512 commit 7d6e9105026788c497f0ab32fa16c82f4ab5ff61 upstream. An ancient gcc bug (first reported in 2003) has apparently resurfaced on MIPS, where kernelci.org reports an overly large stack frame in the whirlpool hash algorithm: crypto/wp512.c:987:1: warning: the frame size of 1112 bytes is larger than 1024 bytes [-Wframe-larger-than=] With some testing in different configurations, I'm seeing large variations in stack frames size up to 1500 bytes for what should have around 300 bytes at most. I also checked the reference implementation, which is essentially the same code but also comes with some test and benchmarking infrastructure. It seems that recent compiler versions on at least arm, arm64 and powerpc have a partial fix for this problem, but enabling "-fsched-pressure", but even with that fix they suffer from the issue to a certain degree. Some testing on arm64 shows that the time needed to hash a given amount of data is roughly proportional to the stack frame size here, which makes sense given that the wp512 implementation is doing lots of loads for table lookups, and the problem with the overly large stack is a result of doing a lot more loads and stores for spilled registers (as seen from inspecting the object code). Disabling -fschedule-insns consistently fixes the problem for wp512, in my collection of cross-compilers, the results are consistently better or identical when comparing the stack sizes in this function, though some architectures (notable x86) have schedule-insns disabled by default. The four columns are: default: -O2 press: -O2 -fsched-pressure nopress: -O2 -fschedule-insns -fno-sched-pressure nosched: -O2 -no-schedule-insns (disables sched-pressure) default press nopress nosched alpha-linux-gcc-4.9.3 1136 848 1136 176 am33_2.0-linux-gcc-4.9.3 2100 2076 2100 2104 arm-linux-gnueabi-gcc-4.9.3 848 848 1048 352 cris-linux-gcc-4.9.3 272 272 272 272 frv-linux-gcc-4.9.3 1128 1000 1128 280 hppa64-linux-gcc-4.9.3 1128 336 1128 184 hppa-linux-gcc-4.9.3 644 308 644 276 i386-linux-gcc-4.9.3 352 352 352 352 m32r-linux-gcc-4.9.3 720 656 720 268 microblaze-linux-gcc-4.9.3 1108 604 1108 256 mips64-linux-gcc-4.9.3 1328 592 1328 208 mips-linux-gcc-4.9.3 1096 624 1096 240 powerpc64-linux-gcc-4.9.3 1088 432 1088 160 powerpc-linux-gcc-4.9.3 1080 584 1080 224 s390-linux-gcc-4.9.3 456 456 624 360 sh3-linux-gcc-4.9.3 292 292 292 292 sparc64-linux-gcc-4.9.3 992 240 992 208 sparc-linux-gcc-4.9.3 680 592 680 312 x86_64-linux-gcc-4.9.3 224 240 272 224 xtensa-linux-gcc-4.9.3 1152 704 1152 304 aarch64-linux-gcc-7.0.0 224 224 1104 208 arm-linux-gnueabi-gcc-7.0.1 824 824 1048 352 mips-linux-gcc-7.0.0 1120 648 1120 272 x86_64-linux-gcc-7.0.1 240 240 304 240 arm-linux-gnueabi-gcc-4.4.7 840 392 arm-linux-gnueabi-gcc-4.5.4 784 728 784 320 arm-linux-gnueabi-gcc-4.6.4 736 728 736 304 arm-linux-gnueabi-gcc-4.7.4 944 784 944 352 arm-linux-gnueabi-gcc-4.8.5 464 464 760 352 arm-linux-gnueabi-gcc-4.9.3 848 848 1048 352 arm-linux-gnueabi-gcc-5.3.1 824 824 1064 336 arm-linux-gnueabi-gcc-6.1.1 808 808 1056 344 arm-linux-gnueabi-gcc-7.0.1 824 824 1048 352 Trying the same test for serpent-generic, the picture is a bit different, and while -fno-schedule-insns is generally better here than the default, -fsched-pressure wins overall, so I picked that instead. default press nopress nosched alpha-linux-gcc-4.9.3 1392 864 1392 960 am33_2.0-linux-gcc-4.9.3 536 524 536 528 arm-linux-gnueabi-gcc-4.9.3 552 552 776 536 cris-linux-gcc-4.9.3 528 528 528 528 frv-linux-gcc-4.9.3 536 400 536 504 hppa64-linux-gcc-4.9.3 524 208 524 480 hppa-linux-gcc-4.9.3 768 472 768 508 i386-linux-gcc-4.9.3 564 564 564 564 m32r-linux-gcc-4.9.3 712 576 712 532 microblaze-linux-gcc-4.9.3 724 392 724 512 mips64-linux-gcc-4.9.3 720 384 720 496 mips-linux-gcc-4.9.3 728 384 728 496 powerpc64-linux-gcc-4.9.3 704 304 704 480 powerpc-linux-gcc-4.9.3 704 296 704 480 s390-linux-gcc-4.9.3 560 560 592 536 sh3-linux-gcc-4.9.3 540 540 540 540 sparc64-linux-gcc-4.9.3 544 352 544 496 sparc-linux-gcc-4.9.3 544 344 544 496 x86_64-linux-gcc-4.9.3 528 536 576 528 xtensa-linux-gcc-4.9.3 752 544 752 544 aarch64-linux-gcc-7.0.0 432 432 656 480 arm-linux-gnueabi-gcc-7.0.1 616 616 808 536 mips-linux-gcc-7.0.0 720 464 720 488 x86_64-linux-gcc-7.0.1 536 528 600 536 arm-linux-gnueabi-gcc-4.4.7 592 440 arm-linux-gnueabi-gcc-4.5.4 776 448 776 544 arm-linux-gnueabi-gcc-4.6.4 776 448 776 544 arm-linux-gnueabi-gcc-4.7.4 768 448 768 544 arm-linux-gnueabi-gcc-4.8.5 488 488 776 544 arm-linux-gnueabi-gcc-4.9.3 552 552 776 536 arm-linux-gnueabi-gcc-5.3.1 552 552 776 536 arm-linux-gnueabi-gcc-6.1.1 560 560 776 536 arm-linux-gnueabi-gcc-7.0.1 616 616 808 536 I did not do any runtime tests with serpent, so it is possible that stack frame size does not directly correlate with runtime performance here and it actually makes things worse, but it's more likely to help here, and the reduced stack frame size is probably enough reason to apply the patch, especially given that the crypto code is often used in deep call chains. Link: https://kernelci.org/build/id/58797d7559b5149efdf6c3a9/logs/ Link: http://www.larc.usp.br/~pbarreto/WhirlpoolPage.html Link: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=11488 Link: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=79149 Cc: Ralf Baechle <ralf@linux-mips.org> Signed-off-by: Arnd Bergmann <arnd@arndb.de> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: Willy Tarreau <w@1wt.eu>
2017-02-03 22:33:23 +00:00
CFLAGS_wp512.o := $(call cc-option,-fno-schedule-insns) # https://gcc.gnu.org/bugzilla/show_bug.cgi?id=79149
obj-$(CONFIG_CRYPTO_TGR192) += tgr192.o
obj-$(CONFIG_CRYPTO_GF128MUL) += gf128mul.o
obj-$(CONFIG_CRYPTO_ECB) += ecb.o
obj-$(CONFIG_CRYPTO_CBC) += cbc.o
obj-$(CONFIG_CRYPTO_PCBC) += pcbc.o
obj-$(CONFIG_CRYPTO_CTS) += cts.o
obj-$(CONFIG_CRYPTO_LRW) += lrw.o
obj-$(CONFIG_CRYPTO_XTS) += xts.o
obj-$(CONFIG_CRYPTO_CTR) += ctr.o
obj-$(CONFIG_CRYPTO_GCM) += gcm.o
obj-$(CONFIG_CRYPTO_CCM) += ccm.o
obj-$(CONFIG_CRYPTO_PCRYPT) += pcrypt.o
obj-$(CONFIG_CRYPTO_CRYPTD) += cryptd.o
obj-$(CONFIG_CRYPTO_DES) += des_generic.o
obj-$(CONFIG_CRYPTO_FCRYPT) += fcrypt.o
obj-$(CONFIG_CRYPTO_BLOWFISH) += blowfish_generic.o
obj-$(CONFIG_CRYPTO_BLOWFISH_COMMON) += blowfish_common.o
obj-$(CONFIG_CRYPTO_TWOFISH) += twofish_generic.o
obj-$(CONFIG_CRYPTO_TWOFISH_COMMON) += twofish_common.o
obj-$(CONFIG_CRYPTO_SERPENT) += serpent_generic.o
crypto: improve gcc optimization flags for serpent and wp512 commit 7d6e9105026788c497f0ab32fa16c82f4ab5ff61 upstream. An ancient gcc bug (first reported in 2003) has apparently resurfaced on MIPS, where kernelci.org reports an overly large stack frame in the whirlpool hash algorithm: crypto/wp512.c:987:1: warning: the frame size of 1112 bytes is larger than 1024 bytes [-Wframe-larger-than=] With some testing in different configurations, I'm seeing large variations in stack frames size up to 1500 bytes for what should have around 300 bytes at most. I also checked the reference implementation, which is essentially the same code but also comes with some test and benchmarking infrastructure. It seems that recent compiler versions on at least arm, arm64 and powerpc have a partial fix for this problem, but enabling "-fsched-pressure", but even with that fix they suffer from the issue to a certain degree. Some testing on arm64 shows that the time needed to hash a given amount of data is roughly proportional to the stack frame size here, which makes sense given that the wp512 implementation is doing lots of loads for table lookups, and the problem with the overly large stack is a result of doing a lot more loads and stores for spilled registers (as seen from inspecting the object code). Disabling -fschedule-insns consistently fixes the problem for wp512, in my collection of cross-compilers, the results are consistently better or identical when comparing the stack sizes in this function, though some architectures (notable x86) have schedule-insns disabled by default. The four columns are: default: -O2 press: -O2 -fsched-pressure nopress: -O2 -fschedule-insns -fno-sched-pressure nosched: -O2 -no-schedule-insns (disables sched-pressure) default press nopress nosched alpha-linux-gcc-4.9.3 1136 848 1136 176 am33_2.0-linux-gcc-4.9.3 2100 2076 2100 2104 arm-linux-gnueabi-gcc-4.9.3 848 848 1048 352 cris-linux-gcc-4.9.3 272 272 272 272 frv-linux-gcc-4.9.3 1128 1000 1128 280 hppa64-linux-gcc-4.9.3 1128 336 1128 184 hppa-linux-gcc-4.9.3 644 308 644 276 i386-linux-gcc-4.9.3 352 352 352 352 m32r-linux-gcc-4.9.3 720 656 720 268 microblaze-linux-gcc-4.9.3 1108 604 1108 256 mips64-linux-gcc-4.9.3 1328 592 1328 208 mips-linux-gcc-4.9.3 1096 624 1096 240 powerpc64-linux-gcc-4.9.3 1088 432 1088 160 powerpc-linux-gcc-4.9.3 1080 584 1080 224 s390-linux-gcc-4.9.3 456 456 624 360 sh3-linux-gcc-4.9.3 292 292 292 292 sparc64-linux-gcc-4.9.3 992 240 992 208 sparc-linux-gcc-4.9.3 680 592 680 312 x86_64-linux-gcc-4.9.3 224 240 272 224 xtensa-linux-gcc-4.9.3 1152 704 1152 304 aarch64-linux-gcc-7.0.0 224 224 1104 208 arm-linux-gnueabi-gcc-7.0.1 824 824 1048 352 mips-linux-gcc-7.0.0 1120 648 1120 272 x86_64-linux-gcc-7.0.1 240 240 304 240 arm-linux-gnueabi-gcc-4.4.7 840 392 arm-linux-gnueabi-gcc-4.5.4 784 728 784 320 arm-linux-gnueabi-gcc-4.6.4 736 728 736 304 arm-linux-gnueabi-gcc-4.7.4 944 784 944 352 arm-linux-gnueabi-gcc-4.8.5 464 464 760 352 arm-linux-gnueabi-gcc-4.9.3 848 848 1048 352 arm-linux-gnueabi-gcc-5.3.1 824 824 1064 336 arm-linux-gnueabi-gcc-6.1.1 808 808 1056 344 arm-linux-gnueabi-gcc-7.0.1 824 824 1048 352 Trying the same test for serpent-generic, the picture is a bit different, and while -fno-schedule-insns is generally better here than the default, -fsched-pressure wins overall, so I picked that instead. default press nopress nosched alpha-linux-gcc-4.9.3 1392 864 1392 960 am33_2.0-linux-gcc-4.9.3 536 524 536 528 arm-linux-gnueabi-gcc-4.9.3 552 552 776 536 cris-linux-gcc-4.9.3 528 528 528 528 frv-linux-gcc-4.9.3 536 400 536 504 hppa64-linux-gcc-4.9.3 524 208 524 480 hppa-linux-gcc-4.9.3 768 472 768 508 i386-linux-gcc-4.9.3 564 564 564 564 m32r-linux-gcc-4.9.3 712 576 712 532 microblaze-linux-gcc-4.9.3 724 392 724 512 mips64-linux-gcc-4.9.3 720 384 720 496 mips-linux-gcc-4.9.3 728 384 728 496 powerpc64-linux-gcc-4.9.3 704 304 704 480 powerpc-linux-gcc-4.9.3 704 296 704 480 s390-linux-gcc-4.9.3 560 560 592 536 sh3-linux-gcc-4.9.3 540 540 540 540 sparc64-linux-gcc-4.9.3 544 352 544 496 sparc-linux-gcc-4.9.3 544 344 544 496 x86_64-linux-gcc-4.9.3 528 536 576 528 xtensa-linux-gcc-4.9.3 752 544 752 544 aarch64-linux-gcc-7.0.0 432 432 656 480 arm-linux-gnueabi-gcc-7.0.1 616 616 808 536 mips-linux-gcc-7.0.0 720 464 720 488 x86_64-linux-gcc-7.0.1 536 528 600 536 arm-linux-gnueabi-gcc-4.4.7 592 440 arm-linux-gnueabi-gcc-4.5.4 776 448 776 544 arm-linux-gnueabi-gcc-4.6.4 776 448 776 544 arm-linux-gnueabi-gcc-4.7.4 768 448 768 544 arm-linux-gnueabi-gcc-4.8.5 488 488 776 544 arm-linux-gnueabi-gcc-4.9.3 552 552 776 536 arm-linux-gnueabi-gcc-5.3.1 552 552 776 536 arm-linux-gnueabi-gcc-6.1.1 560 560 776 536 arm-linux-gnueabi-gcc-7.0.1 616 616 808 536 I did not do any runtime tests with serpent, so it is possible that stack frame size does not directly correlate with runtime performance here and it actually makes things worse, but it's more likely to help here, and the reduced stack frame size is probably enough reason to apply the patch, especially given that the crypto code is often used in deep call chains. Link: https://kernelci.org/build/id/58797d7559b5149efdf6c3a9/logs/ Link: http://www.larc.usp.br/~pbarreto/WhirlpoolPage.html Link: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=11488 Link: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=79149 Cc: Ralf Baechle <ralf@linux-mips.org> Signed-off-by: Arnd Bergmann <arnd@arndb.de> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: Willy Tarreau <w@1wt.eu>
2017-02-03 22:33:23 +00:00
CFLAGS_serpent_generic.o := $(call cc-option,-fsched-pressure) # https://gcc.gnu.org/bugzilla/show_bug.cgi?id=79149
obj-$(CONFIG_CRYPTO_AES) += aes_generic.o
obj-$(CONFIG_CRYPTO_CAMELLIA) += camellia_generic.o
obj-$(CONFIG_CRYPTO_CAST_COMMON) += cast_common.o
obj-$(CONFIG_CRYPTO_CAST5) += cast5_generic.o
obj-$(CONFIG_CRYPTO_CAST6) += cast6_generic.o
obj-$(CONFIG_CRYPTO_ARC4) += arc4.o
obj-$(CONFIG_CRYPTO_TEA) += tea.o
obj-$(CONFIG_CRYPTO_KHAZAD) += khazad.o
obj-$(CONFIG_CRYPTO_ANUBIS) += anubis.o
obj-$(CONFIG_CRYPTO_SEED) += seed.o
obj-$(CONFIG_CRYPTO_SALSA20) += salsa20_generic.o
obj-$(CONFIG_CRYPTO_DEFLATE) += deflate.o
obj-$(CONFIG_CRYPTO_ZLIB) += zlib.o
obj-$(CONFIG_CRYPTO_MICHAEL_MIC) += michael_mic.o
obj-$(CONFIG_CRYPTO_CRC32C) += crc32c.o
obj-$(CONFIG_CRYPTO_CRC32) += crc32.o
obj-$(CONFIG_CRYPTO_AUTHENC) += authenc.o authencesn.o
obj-$(CONFIG_CRYPTO_LZO) += lzo.o
obj-$(CONFIG_CRYPTO_842) += 842.o
obj-$(CONFIG_CRYPTO_RNG2) += rng.o
obj-$(CONFIG_CRYPTO_RNG2) += krng.o
obj-$(CONFIG_CRYPTO_ANSI_CPRNG) += ansi_cprng.o
obj-$(CONFIG_CRYPTO_TEST) += tcrypt.o
obj-$(CONFIG_CRYPTO_GHASH) += ghash-generic.o
obj-$(CONFIG_CRYPTO_USER_API) += af_alg.o
obj-$(CONFIG_CRYPTO_USER_API_HASH) += algif_hash.o
obj-$(CONFIG_CRYPTO_USER_API_SKCIPHER) += algif_skcipher.o
#
# generic algorithms and the async_tx api
#
obj-$(CONFIG_XOR_BLOCKS) += xor.o
async_tx: add the async_tx api The async_tx api provides methods for describing a chain of asynchronous bulk memory transfers/transforms with support for inter-transactional dependencies. It is implemented as a dmaengine client that smooths over the details of different hardware offload engine implementations. Code that is written to the api can optimize for asynchronous operation and the api will fit the chain of operations to the available offload resources. I imagine that any piece of ADMA hardware would register with the 'async_*' subsystem, and a call to async_X would be routed as appropriate, or be run in-line. - Neil Brown async_tx exploits the capabilities of struct dma_async_tx_descriptor to provide an api of the following general format: struct dma_async_tx_descriptor * async_<operation>(..., struct dma_async_tx_descriptor *depend_tx, dma_async_tx_callback cb_fn, void *cb_param) { struct dma_chan *chan = async_tx_find_channel(depend_tx, <operation>); struct dma_device *device = chan ? chan->device : NULL; int int_en = cb_fn ? 1 : 0; struct dma_async_tx_descriptor *tx = device ? device->device_prep_dma_<operation>(chan, len, int_en) : NULL; if (tx) { /* run <operation> asynchronously */ ... tx->tx_set_dest(addr, tx, index); ... tx->tx_set_src(addr, tx, index); ... async_tx_submit(chan, tx, flags, depend_tx, cb_fn, cb_param); } else { /* run <operation> synchronously */ ... <operation> ... async_tx_sync_epilog(flags, depend_tx, cb_fn, cb_param); } return tx; } async_tx_find_channel() returns a capable channel from its pool. The channel pool is organized as a per-cpu array of channel pointers. The async_tx_rebalance() routine is tasked with managing these arrays. In the uniprocessor case async_tx_rebalance() tries to spread responsibility evenly over channels of similar capabilities. For example if there are two copy+xor channels, one will handle copy operations and the other will handle xor. In the SMP case async_tx_rebalance() attempts to spread the operations evenly over the cpus, e.g. cpu0 gets copy channel0 and xor channel0 while cpu1 gets copy channel 1 and xor channel 1. When a dependency is specified async_tx_find_channel defaults to keeping the operation on the same channel. A xor->copy->xor chain will stay on one channel if it supports both operation types, otherwise the transaction will transition between a copy and a xor resource. Currently the raid5 implementation in the MD raid456 driver has been converted to the async_tx api. A driver for the offload engines on the Intel Xscale series of I/O processors, iop-adma, is provided in a later commit. With the iop-adma driver and async_tx, raid456 is able to offload copy, xor, and xor-zero-sum operations to hardware engines. On iop342 tiobench showed higher throughput for sequential writes (20 - 30% improvement) and sequential reads to a degraded array (40 - 55% improvement). For the other cases performance was roughly equal, +/- a few percentage points. On a x86-smp platform the performance of the async_tx implementation (in synchronous mode) was also +/- a few percentage points of the original implementation. According to 'top' on iop342 CPU utilization drops from ~50% to ~15% during a 'resync' while the speed according to /proc/mdstat doubles from ~25 MB/s to ~50 MB/s. The tiobench command line used for testing was: tiobench --size 2048 --block 4096 --block 131072 --dir /mnt/raid --numruns 5 * iop342 had 1GB of memory available Details: * if CONFIG_DMA_ENGINE=n the asynchronous path is compiled away by making async_tx_find_channel a static inline routine that always returns NULL * when a callback is specified for a given transaction an interrupt will fire at operation completion time and the callback will occur in a tasklet. if the the channel does not support interrupts then a live polling wait will be performed * the api is written as a dmaengine client that requests all available channels * In support of dependencies the api implicitly schedules channel-switch interrupts. The interrupt triggers the cleanup tasklet which causes pending operations to be scheduled on the next channel * Xor engines treat an xor destination address differently than a software xor routine. To the software routine the destination address is an implied source, whereas engines treat it as a write-only destination. This patch modifies the xor_blocks routine to take a an explicit destination address to mirror the hardware. Changelog: * fixed a leftover debug print * don't allow callbacks in async_interrupt_cond * fixed xor_block changes * fixed usage of ASYNC_TX_XOR_DROP_DEST * drop dma mapping methods, suggested by Chris Leech * printk warning fixups from Andrew Morton * don't use inline in C files, Adrian Bunk * select the API when MD is enabled * BUG_ON xor source counts <= 1 * implicitly handle hardware concerns like channel switching and interrupts, Neil Brown * remove the per operation type list, and distribute operation capabilities evenly amongst the available channels * simplify async_tx_find_channel to optimize the fast path * introduce the channel_table_initialized flag to prevent early calls to the api * reorganize the code to mimic crypto * include mm.h as not all archs include it in dma-mapping.h * make the Kconfig options non-user visible, Adrian Bunk * move async_tx under crypto since it is meant as 'core' functionality, and the two may share algorithms in the future * move large inline functions into c files * checkpatch.pl fixes * gpl v2 only correction Cc: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Acked-By: NeilBrown <neilb@suse.de>
2007-01-02 18:10:44 +00:00
obj-$(CONFIG_ASYNC_CORE) += async_tx/
obj-$(CONFIG_ASYMMETRIC_KEY_TYPE) += asymmetric_keys/
obj-$(CONFIG_CRYPTO_ABLK_HELPER) += ablk_helper.o