diff options
| author | Linus Torvalds <torvalds@linux-foundation.org> | 2018-10-25 16:43:35 -0700 |
|---|---|---|
| committer | Linus Torvalds <torvalds@linux-foundation.org> | 2018-10-25 16:43:35 -0700 |
| commit | 62606c224d72a98c35d21a849f95cccf95b0a252 (patch) | |
| tree | 6f6f3466451edf9baa2ea8b5f9fc558aa555c69a /arch | |
| parent | 24ed334f33666f2ae929ccc08f72e7e72e353c64 (diff) | |
| parent | a1c6fd4308d37f072e939a2782f24214115fc7e8 (diff) | |
Merge branch 'linus' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6
Pull crypto updates from Herbert Xu:
"API:
- Remove VLA usage
- Add cryptostat user-space interface
- Add notifier for new crypto algorithms
Algorithms:
- Add OFB mode
- Remove speck
Drivers:
- Remove x86/sha*-mb as they are buggy
- Remove pcbc(aes) from x86/aesni
- Improve performance of arm/ghash-ce by up to 85%
- Implement CTS-CBC in arm64/aes-blk, faster by up to 50%
- Remove PMULL based arm64/crc32 driver
- Use PMULL in arm64/crct10dif
- Add aes-ctr support in s5p-sss
- Add caam/qi2 driver
Others:
- Pick better transform if one becomes available in crc-t10dif"
* 'linus' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6: (124 commits)
crypto: chelsio - Update ntx queue received from cxgb4
crypto: ccree - avoid implicit enum conversion
crypto: caam - add SPDX license identifier to all files
crypto: caam/qi - simplify CGR allocation, freeing
crypto: mxs-dcp - make symbols 'sha1_null_hash' and 'sha256_null_hash' static
crypto: arm64/aes-blk - ensure XTS mask is always loaded
crypto: testmgr - fix sizeof() on COMP_BUF_SIZE
crypto: chtls - remove set but not used variable 'csk'
crypto: axis - fix platform_no_drv_owner.cocci warnings
crypto: x86/aes-ni - fix build error following fpu template removal
crypto: arm64/aes - fix handling sub-block CTS-CBC inputs
crypto: caam/qi2 - avoid double export
crypto: mxs-dcp - Fix AES issues
crypto: mxs-dcp - Fix SHA null hashes and output length
crypto: mxs-dcp - Implement sha import/export
crypto: aegis/generic - fix for big endian systems
crypto: morus/generic - fix for big endian systems
crypto: lrw - fix rebase error after out of bounds fix
crypto: cavium/nitrox - use pci_alloc_irq_vectors() while enabling MSI-X.
crypto: cavium/nitrox - NITROX command queue changes.
...
Diffstat (limited to 'arch')
67 files changed, 974 insertions, 10747 deletions
diff --git a/arch/arm/crypto/Kconfig b/arch/arm/crypto/Kconfig index 925d1364727a..ef0c7feea6e2 100644 --- a/arch/arm/crypto/Kconfig +++ b/arch/arm/crypto/Kconfig @@ -99,6 +99,7 @@ config CRYPTO_GHASH_ARM_CE depends on KERNEL_MODE_NEON select CRYPTO_HASH select CRYPTO_CRYPTD + select CRYPTO_GF128MUL help Use an implementation of GHASH (used by the GCM AEAD chaining mode) that uses the 64x64 to 128 bit polynomial multiplication (vmull.p64) @@ -121,10 +122,4 @@ config CRYPTO_CHACHA20_NEON select CRYPTO_BLKCIPHER select CRYPTO_CHACHA20 -config CRYPTO_SPECK_NEON - tristate "NEON accelerated Speck cipher algorithms" - depends on KERNEL_MODE_NEON - select CRYPTO_BLKCIPHER - select CRYPTO_SPECK - endif diff --git a/arch/arm/crypto/Makefile b/arch/arm/crypto/Makefile index 8de542c48ade..bd5bceef0605 100644 --- a/arch/arm/crypto/Makefile +++ b/arch/arm/crypto/Makefile @@ -10,7 +10,6 @@ obj-$(CONFIG_CRYPTO_SHA1_ARM_NEON) += sha1-arm-neon.o obj-$(CONFIG_CRYPTO_SHA256_ARM) += sha256-arm.o obj-$(CONFIG_CRYPTO_SHA512_ARM) += sha512-arm.o obj-$(CONFIG_CRYPTO_CHACHA20_NEON) += chacha20-neon.o -obj-$(CONFIG_CRYPTO_SPECK_NEON) += speck-neon.o ce-obj-$(CONFIG_CRYPTO_AES_ARM_CE) += aes-arm-ce.o ce-obj-$(CONFIG_CRYPTO_SHA1_ARM_CE) += sha1-arm-ce.o @@ -54,7 +53,6 @@ ghash-arm-ce-y := ghash-ce-core.o ghash-ce-glue.o crct10dif-arm-ce-y := crct10dif-ce-core.o crct10dif-ce-glue.o crc32-arm-ce-y:= crc32-ce-core.o crc32-ce-glue.o chacha20-neon-y := chacha20-neon-core.o chacha20-neon-glue.o -speck-neon-y := speck-neon-core.o speck-neon-glue.o ifdef REGENERATE_ARM_CRYPTO quiet_cmd_perl = PERL $@ diff --git a/arch/arm/crypto/chacha20-neon-core.S b/arch/arm/crypto/chacha20-neon-core.S index 451a849ad518..50e7b9896818 100644 --- a/arch/arm/crypto/chacha20-neon-core.S +++ b/arch/arm/crypto/chacha20-neon-core.S @@ -18,6 +18,34 @@ * (at your option) any later version. */ + /* + * NEON doesn't have a rotate instruction. The alternatives are, more or less: + * + * (a) vshl.u32 + vsri.u32 (needs temporary register) + * (b) vshl.u32 + vshr.u32 + vorr (needs temporary register) + * (c) vrev32.16 (16-bit rotations only) + * (d) vtbl.8 + vtbl.8 (multiple of 8 bits rotations only, + * needs index vector) + * + * ChaCha20 has 16, 12, 8, and 7-bit rotations. For the 12 and 7-bit + * rotations, the only choices are (a) and (b). We use (a) since it takes + * two-thirds the cycles of (b) on both Cortex-A7 and Cortex-A53. + * + * For the 16-bit rotation, we use vrev32.16 since it's consistently fastest + * and doesn't need a temporary register. + * + * For the 8-bit rotation, we use vtbl.8 + vtbl.8. On Cortex-A7, this sequence + * is twice as fast as (a), even when doing (a) on multiple registers + * simultaneously to eliminate the stall between vshl and vsri. Also, it + * parallelizes better when temporary registers are scarce. + * + * A disadvantage is that on Cortex-A53, the vtbl sequence is the same speed as + * (a), so the need to load the rotation table actually makes the vtbl method + * slightly slower overall on that CPU (~1.3% slower ChaCha20). Still, it + * seems to be a good compromise to get a more significant speed boost on some + * CPUs, e.g. ~4.8% faster ChaCha20 on Cortex-A7. + */ + #include <linux/linkage.h> .text @@ -46,7 +74,9 @@ ENTRY(chacha20_block_xor_neon) vmov q10, q2 vmov q11, q3 + adr ip, .Lrol8_table mov r3, #10 + vld1.8 {d10}, [ip, :64] .Ldoubleround: // x0 += x1, x3 = rotl32(x3 ^ x0, 16) @@ -62,9 +92,9 @@ ENTRY(chacha20_block_xor_neon) // x0 += x1, x3 = rotl32(x3 ^ x0, 8) vadd.i32 q0, q0, q1 - veor q4, q3, q0 - vshl.u32 q3, q4, #8 - vsri.u32 q3, q4, #24 + veor q3, q3, q0 + vtbl.8 d6, {d6}, d10 + vtbl.8 d7, {d7}, d10 // x2 += x3, x1 = rotl32(x1 ^ x2, 7) vadd.i32 q2, q2, q3 @@ -92,9 +122,9 @@ ENTRY(chacha20_block_xor_neon) // x0 += x1, x3 = rotl32(x3 ^ x0, 8) vadd.i32 q0, q0, q1 - veor q4, q3, q0 - vshl.u32 q3, q4, #8 - vsri.u32 q3, q4, #24 + veor q3, q3, q0 + vtbl.8 d6, {d6}, d10 + vtbl.8 d7, {d7}, d10 // x2 += x3, x1 = rotl32(x1 ^ x2, 7) vadd.i32 q2, q2, q3 @@ -139,13 +169,17 @@ ENTRY(chacha20_block_xor_neon) bx lr ENDPROC(chacha20_block_xor_neon) + .align 4 +.Lctrinc: .word 0, 1, 2, 3 +.Lrol8_table: .byte 3, 0, 1, 2, 7, 4, 5, 6 + .align 5 ENTRY(chacha20_4block_xor_neon) - push {r4-r6, lr} - mov ip, sp // preserve the stack pointer - sub r3, sp, #0x20 // allocate a 32 byte buffer - bic r3, r3, #0x1f // aligned to 32 bytes - mov sp, r3 + push {r4-r5} + mov r4, sp // preserve the stack pointer + sub ip, sp, #0x20 // allocate a 32 byte buffer + bic ip, ip, #0x1f // aligned to 32 bytes + mov sp, ip // r0: Input state matrix, s // r1: 4 data blocks output, o @@ -155,25 +189,24 @@ ENTRY(chacha20_4block_xor_neon) // This function encrypts four consecutive ChaCha20 blocks by loading // the state matrix in NEON registers four times. The algorithm performs // each operation on the corresponding word of each state matrix, hence - // requires no word shuffling. For final XORing step we transpose the - // matrix by interleaving 32- and then 64-bit words, which allows us to - // do XOR in NEON registers. + // requires no word shuffling. The words are re-interleaved before the + // final addition of the original state and the XORing step. // - // x0..15[0-3] = s0..3[0..3] - add r3, r0, #0x20 + // x0..15[0-3] = s0..15[0-3] + add ip, r0, #0x20 vld1.32 {q0-q1}, [r0] - vld1.32 {q2-q3}, [r3] + vld1.32 {q2-q3}, [ip] - adr r3, CTRINC + adr r5, .Lctrinc vdup.32 q15, d7[1] vdup.32 q14, d7[0] - vld1.32 {q11}, [r3, :128] + vld1.32 {q4}, [r5, :128] vdup.32 q13, d6[1] vdup.32 q12, d6[0] - vadd.i32 q12, q12, q11 // x12 += counter values 0-3 vdup.32 q11, d5[1] vdup.32 q10, d5[0] + vadd.u32 q12, q12, q4 // x12 += counter values 0-3 vdup.32 q9, d4[1] vdup.32 q8, d4[0] vdup.32 q7, d3[1] @@ -185,9 +218,13 @@ ENTRY(chacha20_4block_xor_neon) vdup.32 q1, d0[1] vdup.32 q0, d0[0] + adr ip, .Lrol8_table mov r3, #10 + b 1f .Ldoubleround4: + vld1.32 {q8-q9}, [sp, :256] +1: // x0 += x4, x12 = rotl32(x12 ^ x0, 16) // x1 += x5, x13 = rotl32(x13 ^ x1, 16) // x2 += x6, x14 = rotl32(x14 ^ x2, 16) @@ -236,24 +273,25 @@ ENTRY(chacha20_4block_xor_neon) // x1 += x5, x13 = rotl32(x13 ^ x1, 8) // x2 += x6, x14 = rotl32(x14 ^ x2, 8) // x3 += x7, x15 = rotl32(x15 ^ x3, 8) + vld1.8 {d16}, [ip, :64] vadd.i32 q0, q0, q4 vadd.i32 q1, q1, q5 vadd.i32 q2, q2, q6 vadd.i32 q3, q3, q7 - veor q8, q12, q0 - veor q9, q13, q1 - vshl.u32 q12, q8, #8 - vshl.u32 q13, q9, #8 - vsri.u32 q12, q8, #24 - vsri.u32 q13, q9, #24 + veor q12, q12, q0 + veor q13, q13, q1 + veor q14, q14, q2 + veor q15, q15, q3 - veor q8, q14, q2 - veor q9, q15, q3 - vshl.u32 q14, q8, #8 - vshl.u32 q15, q9, #8 - vsri.u32 q14, q8, #24 - vsri.u32 q15, q9, #24 + vtbl.8 d24, {d24}, d16 + vtbl.8 d25, {d25}, d16 + vtbl.8 d26, {d26}, d16 + vtbl.8 d27, {d27}, d16 + vtbl.8 d28, {d28}, d16 + vtbl.8 d29, {d29}, d16 + vtbl.8 d30, {d30}, d16 + vtbl.8 d31, {d31}, d16 vld1.32 {q8-q9}, [sp, :256] @@ -332,24 +370,25 @@ ENTRY(chacha20_4block_xor_neon) // x1 += x6, x12 = rotl32(x12 ^ x1, 8) // x2 += x7, x13 = rotl32(x13 ^ x2, 8) // x3 += x4, x14 = rotl32(x14 ^ x3, 8) + vld1.8 {d16}, [ip, :64] vadd.i32 q0, q0, q5 vadd.i32 q1, q1, q6 vadd.i32 q2, q2, q7 vadd.i32 q3, q3, q4 - veor q8, q15, q0 - veor q9, q12, q1 - vshl.u32 q15, q8, #8 - vshl.u32 q12, q9, #8 - vsri.u32 q15, q8, #24 - vsri.u32 q12, q9, #24 + veor q15, q15, q0 + veor q12, q12, q1 + veor q13, q13, q2 + veor q14, q14, q3 - veor q8, q13, q2 - veor q9, q14, q3 - vshl.u32 q13, q8, #8 - vshl.u32 q14, q9, #8 - vsri.u32 q13, q8, #24 - vsri.u32 q14, q9, #24 + vtbl.8 d30, {d30}, d16 + vtbl.8 d31, {d31}, d16 + vtbl.8 d24, {d24}, d16 + vtbl.8 d25, {d25}, d16 + vtbl.8 d26, {d26}, d16 + vtbl.8 d27, {d27}, d16 + vtbl.8 d28, {d28}, d16 + vtbl.8 d29, {d29}, d16 vld1.32 {q8-q9}, [sp, :256] @@ -379,104 +418,76 @@ ENTRY(chacha20_4block_xor_neon) vsri.u32 q6, q9, #25 subs r3, r3, #1 - beq 0f - - vld1.32 {q8-q9}, [sp, :256] - b .Ldoubleround4 - - // x0[0-3] += s0[0] - // x1[0-3] += s0[1] - // x2[0-3] += s0[2] - // x3[0-3] += s0[3] -0: ldmia r0!, {r3-r6} - vdup.32 q8, r3 - vdup.32 q9, r4 - vadd.i32 q0, q0, q8 - vadd.i32 q1, q1, q9 - vdup.32 q8, r5 - vdup.32 q9, r6 - vadd.i32 q2, q2, q8 - vadd.i32 q3, q3, q9 - - // x4[0-3] += s1[0] - // x5[0-3] += s1[1] - // x6[0-3] += s1[2] - // x7[0-3] += s1[3] - ldmia r0!, {r3-r6} - vdup.32 q8, r3 - vdup.32 q9, r4 - vadd.i32 q4, q4, q8 - vadd.i32 q5, q5, q9 - vdup.32 q8, r5 - vdup.32 q9, r6 - vadd.i32 q6, q6, q8 - vadd.i32 q7, q7, q9 - - // interleave 32-bit words in state n, n+1 - vzip.32 q0, q1 - vzip.32 q2, q3 - vzip.32 q4, q5 - vzip.32 q6, q7 - - // interleave 64-bit words in state n, n+2 + bne .Ldoubleround4 + + // x0..7[0-3] are in q0-q7, x10..15[0-3] are in q10-q15. + // x8..9[0-3] are on the stack. + + // Re-interleave the words in the first two rows of each block (x0..7). + // Also add the counter values 0-3 to x12[0-3]. + vld1.32 {q8}, [r5, :128] // load counter values 0-3 + vzip.32 q0, q1 // => (0 1 0 1) (0 1 0 1) + vzip.32 q2, q3 // => (2 3 2 3) (2 3 2 3) + vzip.32 q4, q5 // => (4 5 4 5) (4 5 4 5) + vzip.32 q6, q7 // => (6 7 6 7) (6 7 6 7) + vadd.u32 q12, q8 // x12 += counter values 0-3 vswp d1, d4 vswp d3, d6 + vld1.32 {q8-q9}, [r0]! // load s0..7 vswp d9, d12 vswp d11, d14 - // xor with corresponding input, write to output + // Swap q1 and q4 so that we'll free up consecutive registers (q0-q1) + // after XORing the first 32 bytes. + vswp q1, q4 + + // First two rows of each block are (q0 q1) (q2 q6) (q4 q5) (q3 q7) + + // x0..3[0-3] += s0..3[0-3] (add orig state to 1st row of each block) + vadd.u32 q0, q0, q8 + vadd.u32 q2, q2, q8 + vadd.u32 q4, q4, q8 + vadd.u32 q3, q3, q8 + + // x4..7[0-3] += s4..7[0-3] (add orig state to 2nd row of each block) + vadd.u32 q1, q1, q9 + vadd.u32 q6, q6, q9 + vadd.u32 q5, q5, q9 + vadd.u32 q7, q7, q9 + + // XOR first 32 bytes using keystream from first two rows of first block vld1.8 {q8-q9}, [r2]! veor q8, q8, q0 - veor q9, q9, q4 + veor q9, q9, q1 vst1.8 {q8-q9}, [r1]! + // Re-interleave the words in the last two rows of each block (x8..15). vld1.32 {q8-q9}, [sp, :256] - - // x8[0-3] += s2[0] - // x9[0-3] += s2[1] - // x10[0-3] += s2[2] - // x11[0-3] += s2[3] - ldmia r0!, {r3-r6} - vdup.32 q0, r3 - vdup.32 q4, r4 - vadd.i32 q8, q8, q0 - vadd.i32 q9, q9, q4 - vdup.32 q0, r5 - vdup.32 q4, r6 - vadd.i32 q10, q10, q0 - vadd.i32 q11, q11, q4 - - // x12[0-3] += s3[0] - // x13[0-3] += s3[1] - // x14[0-3] += s3[2] - // x15[0-3] += s3[3] - ldmia r0!, {r3-r6} - vdup.32 q0, r3 - vdup.32 q4, r4 - adr r3, CTRINC - vadd.i32 q12, q12, q0 - vld1.32 {q0}, [r3, :128] - vadd.i32 q13, q13, q4 - vadd.i32 q12, q12, q0 // x12 += counter values 0-3 - - vdup.32 q0, r5 - vdup.32 q4, r6 - vadd.i32 q14, q14, q0 - vadd.i32 q15, q15, q4 - - // interleave 32-bit words in state n, n+1 - vzip.32 q8, q9 - vzip.32 q10, q11 - vzip.32 q12, q13 - vzip.32 q14, q15 - - // interleave 64-bit words in state n, n+2 - vswp d17, d20 - vswp d19, d22 + vzip.32 q12, q13 // => (12 13 12 13) (12 13 12 13) + vzip.32 q14, q15 // => (14 15 14 15) (14 15 14 15) + vzip.32 q8, q9 // => (8 9 8 9) (8 9 8 9) + vzip.32 q10, q11 // => (10 11 10 11) (10 11 10 11) + vld1.32 {q0-q1}, [r0] // load s8..15 vswp d25, d28 vswp d27, d30 + vswp d17, d20 + vswp d19, d22 + + // Last two rows of each block are (q8 q12) (q10 q14) (q9 q13) (q11 q15) + + // x8..11[0-3] += s8..11[0-3] (add orig state to 3rd row of each block) + vadd.u32 q8, q8, q0 + vadd.u32 q10, q10, q0 + vadd.u32 q9, q9, q0 + vadd.u32 q11, q11, q0 + + // x12..15[0-3] += s12..15[0-3] (add orig state to 4th row of each block) + vadd.u32 q12, q12, q1 + vadd.u32 q14, q14, q1 + vadd.u32 q13, q13, q1 + vadd.u32 q15, q15, q1 - vmov q4, q1 + // XOR the rest of the data with the keystream vld1.8 {q0-q1}, [r2]! veor q0, q0, q8 @@ -509,13 +520,11 @@ ENTRY(chacha20_4block_xor_neon) vst1.8 {q0-q1}, [r1]! vld1.8 {q0-q1}, [r2] + mov sp, r4 // restore original stack pointer veor q0, q0, q11 veor q1, q1, q15 vst1.8 {q0-q1}, [r1] - mov sp, ip - pop {r4-r6, pc} + pop {r4-r5} + bx lr ENDPROC(chacha20_4block_xor_neon) - - .align 4 -CTRINC: .word 0, 1, 2, 3 diff --git a/arch/arm/crypto/crc32-ce-glue.c b/arch/arm/crypto/crc32-ce-glue.c index 96e62ec105d0..cd9e93b46c2d 100644 --- a/arch/arm/crypto/crc32-ce-glue.c +++ b/arch/arm/crypto/crc32-ce-glue.c @@ -236,7 +236,7 @@ static void __exit crc32_pmull_mod_exit(void) ARRAY_SIZE(crc32_pmull_algs)); } -static const struct cpu_feature crc32_cpu_feature[] = { +static const struct cpu_feature __maybe_unused crc32_cpu_feature[] = { { cpu_feature(CRC32) }, { cpu_feature(PMULL) }, { } }; MODULE_DEVICE_TABLE(cpu, crc32_cpu_feature); diff --git a/arch/arm/crypto/ghash-ce-core.S b/arch/arm/crypto/ghash-ce-core.S index 2f78c10b1881..406009afa9cf 100644 --- a/arch/arm/crypto/ghash-ce-core.S +++ b/arch/arm/crypto/ghash-ce-core.S @@ -63,6 +63,33 @@ k48 .req d31 SHASH2_p64 .req d31 + HH .req q10 + HH3 .req q11 + HH4 .req q12 + HH34 .req q13 + + HH_L .req d20 + HH_H .req d21 + HH3_L .req d22 + HH3_H .req d23 + HH4_L .req d24 + HH4_H .req d25 + HH34_L .req d26 + HH34_H .req d27 + SHASH2_H .req d29 + + XL2 .req q5 + XM2 .req q6 + XH2 .req q7 + T3 .req q8 + + XL2_L .req d10 + XL2_H .req d11 + XM2_L .req d12 + XM2_H .req d13 + T3_L .req d16 + T3_H .req d17 + .text .fpu crypto-neon-fp-armv8 @@ -175,12 +202,77 @@ beq 0f vld1.64 {T1}, [ip] teq r0, #0 - b 1f + b 3f + +0: .ifc \pn, p64 + tst r0, #3 // skip until #blocks is a + bne 2f // round multiple of 4 + + vld1.8 {XL2-XM2}, [r2]! +1: vld1.8 {T3-T2}, [r2]! + vrev64.8 XL2, XL2 + vrev64.8 XM2, XM2 + + subs r0, r0, #4 + + vext.8 T1, XL2, XL2, #8 + veor XL2_H, XL2_H, XL_L + veor XL, XL, T1 + + vrev64.8 T3, T3 + vrev64.8 T1, T2 + + vmull.p64 XH, HH4_H, XL_H // a1 * b1 + veor XL2_H, XL2_H, XL_H + vmull.p64 XL, HH4_L, XL_L // a0 * b0 + vmull.p64 XM, HH34_H, XL2_H // (a1 + a0)(b1 + b0) + + vmull.p64 XH2, HH3_H, XM2_L // a1 * b1 + veor XM2_L, XM2_L, XM2_H + vmull.p64 XL2, HH3_L, XM2_H // a0 * b0 + vmull.p64 XM2, HH34_L, XM2_L // (a1 + a0)(b1 + b0) + + veor XH, XH, XH2 + veor XL, XL, XL2 + veor XM, XM, XM2 + + vmull.p64 XH2, HH_H, T3_L // a1 * b1 + veor T3_L, T3_L, T3_H + vmull.p64 XL2, HH_L, T3_H // a0 * b0 + vmull.p64 XM2, SHASH2_H, T3_L // (a1 + a0)(b1 + b0) + + veor XH, XH, XH2 + veor XL, XL, XL2 + veor XM, XM, XM2 + + vmull.p64 XH2, SHASH_H, T1_L // a1 * b1 + veor T1_L, T1_L, T1_H + vmull.p64 XL2, SHASH_L, T1_H // a0 * b0 + vmull.p64 XM2, SHASH2_p64, T1_L // (a1 + a0)(b1 + b0) + + veor XH, XH, XH2 + veor XL, XL, XL2 + veor XM, XM, XM2 -0: vld1.64 {T1}, [r2]! + beq 4f + + vld1.8 {XL2-XM2}, [r2]! + + veor T1, XL, XH + veor XM, XM, T1 + + __pmull_reduce_p64 + + veor T1, T1, XH + veor XL, XL, T1 + + b 1b + .endif + +2: vld1.64 {T1}, [r2]! subs r0, r0, #1 -1: /* multiply XL by SHASH in GF(2^128) */ +3: /* multiply XL by |