From 81658ad0d92306ceb271994b90cd49ffde10eeda Mon Sep 17 00:00:00 2001 From: "David S. Miller" Date: Tue, 28 Aug 2012 12:05:54 -0700 Subject: sparc64: Add CAMELLIA driver making use of the new camellia opcodes. Signed-off-by: David S. Miller --- arch/sparc/crypto/Makefile | 2 + arch/sparc/crypto/camellia_asm.S | 583 ++++++++++++++++++++++++++++++++++++++ arch/sparc/crypto/camellia_glue.c | 318 +++++++++++++++++++++ 3 files changed, 903 insertions(+) create mode 100644 arch/sparc/crypto/camellia_asm.S create mode 100644 arch/sparc/crypto/camellia_glue.c (limited to 'arch') diff --git a/arch/sparc/crypto/Makefile b/arch/sparc/crypto/Makefile index dd999c6c8609..5d469d81761f 100644 --- a/arch/sparc/crypto/Makefile +++ b/arch/sparc/crypto/Makefile @@ -9,6 +9,7 @@ obj-$(CONFIG_CRYPTO_MD5_SPARC64) += md5-sparc64.o obj-$(CONFIG_CRYPTO_AES_SPARC64) += aes-sparc64.o obj-$(CONFIG_CRYPTO_DES_SPARC64) += des-sparc64.o +obj-$(CONFIG_CRYPTO_DES_SPARC64) += camellia-sparc64.o obj-$(CONFIG_CRYPTO_CRC32C_SPARC64) += crc32c-sparc64.o @@ -19,5 +20,6 @@ md5-sparc64-y := md5_asm.o md5_glue.o aes-sparc64-y := aes_asm.o aes_glue.o des-sparc64-y := des_asm.o des_glue.o +camellia-sparc64-y := camellia_asm.o camellia_glue.o crc32c-sparc64-y := crc32c_asm.o crc32c_glue.o diff --git a/arch/sparc/crypto/camellia_asm.S b/arch/sparc/crypto/camellia_asm.S new file mode 100644 index 000000000000..b0ddb5bcfe5f --- /dev/null +++ b/arch/sparc/crypto/camellia_asm.S @@ -0,0 +1,583 @@ +#include +#include + +#define F3F(x,y,z) (((x)<<30)|((y)<<19)|((z)<<5)) + +#define FPD_ENCODE(x) (((x) >> 5) | ((x) & ~(0x20))) + +#define RS1(x) (FPD_ENCODE(x) << 14) +#define RS2(x) (FPD_ENCODE(x) << 0) +#define RS3(x) (FPD_ENCODE(x) << 9) +#define RD(x) (FPD_ENCODE(x) << 25) +#define IMM5(x) ((x) << 0) + +#define CAMELLIA_F(a,b,c,d) \ + .word (F3F(2, 0x19, 0x00c)|RS1(a)|RS2(b)|RS3(c)|RD(d)); +#define CAMELLIA_FL(a,b,c) \ + .word (F3F(2, 0x36, 0x13c)|RS1(a)|RS2(b)|RD(c)); +#define CAMELLIA_FLI(a,b,c) \ + .word (F3F(2, 0x36, 0x13d)|RS1(a)|RS2(b)|RD(c)); + +#define MOVDTOX_F0_O4 \ + .word 0x99b02200 +#define MOVDTOX_F2_O5 \ + .word 0x9bb02202 + +#define CAMELLIA_6ROUNDS(KEY_BASE, I0, I1) \ + CAMELLIA_F(KEY_BASE + 0, I1, I0, I1) \ + CAMELLIA_F(KEY_BASE + 2, I0, I1, I0) \ + CAMELLIA_F(KEY_BASE + 4, I1, I0, I1) \ + CAMELLIA_F(KEY_BASE + 6, I0, I1, I0) \ + CAMELLIA_F(KEY_BASE + 8, I1, I0, I1) \ + CAMELLIA_F(KEY_BASE + 10, I0, I1, I0) + +#define CAMELLIA_6ROUNDS_FL_FLI(KEY_BASE, I0, I1) \ + CAMELLIA_6ROUNDS(KEY_BASE, I0, I1) \ + CAMELLIA_FL(KEY_BASE + 12, I0, I0) \ + CAMELLIA_FLI(KEY_BASE + 14, I1, I1) + + .data + + .align 8 +SIGMA: .xword 0xA09E667F3BCC908B + .xword 0xB67AE8584CAA73B2 + .xword 0xC6EF372FE94F82BE + .xword 0x54FF53A5F1D36F1C + .xword 0x10E527FADE682D1D + .xword 0xB05688C2B3E6C1FD + + .text + + .align 32 +ENTRY(camellia_sparc64_key_expand) + /* %o0=in_key, %o1=out_key, %o2=key_len */ + VISEntry + ld [%o0 + 0x00], %f0 ! i0, k[0] + ld [%o0 + 0x04], %f1 ! i1, k[1] + ld [%o0 + 0x08], %f2 ! i2, k[2] + ld [%o0 + 0x0c], %f3 ! i3, k[3] + std %f0, [%o1 + 0x00] ! k[0, 1] + fsrc2 %f0, %f28 + std %f2, [%o1 + 0x08] ! k[2, 3] + cmp %o2, 16 + be 10f + fsrc2 %f2, %f30 + + ld [%o0 + 0x10], %f0 + ld [%o0 + 0x14], %f1 + std %f0, [%o1 + 0x20] ! k[8, 9] + cmp %o2, 24 + fone %f10 + be,a 1f + fxor %f10, %f0, %f2 + ld [%o0 + 0x18], %f2 + ld [%o0 + 0x1c], %f3 +1: + std %f2, [%o1 + 0x28] ! k[10, 11] + fxor %f28, %f0, %f0 + fxor %f30, %f2, %f2 + +10: + sethi %hi(SIGMA), %g3 + or %g3, %lo(SIGMA), %g3 + ldd [%g3 + 0x00], %f16 + ldd [%g3 + 0x08], %f18 + ldd [%g3 + 0x10], %f20 + ldd [%g3 + 0x18], %f22 + ldd [%g3 + 0x20], %f24 + ldd [%g3 + 0x28], %f26 + CAMELLIA_F(16, 2, 0, 2) + CAMELLIA_F(18, 0, 2, 0) + fxor %f28, %f0, %f0 + fxor %f30, %f2, %f2 + CAMELLIA_F(20, 2, 0, 2) + CAMELLIA_F(22, 0, 2, 0) + +#define ROTL128(S01, S23, TMP1, TMP2, N) \ + srlx S01, (64 - N), TMP1; \ + sllx S01, N, S01; \ + srlx S23, (64 - N), TMP2; \ + sllx S23, N, S23; \ + or S01, TMP2, S01; \ + or S23, TMP1, S23 + + cmp %o2, 16 + bne 1f + nop + /* 128-bit key */ + std %f0, [%o1 + 0x10] ! k[ 4, 5] + std %f2, [%o1 + 0x18] ! k[ 6, 7] + MOVDTOX_F0_O4 + MOVDTOX_F2_O5 + ROTL128(%o4, %o5, %g2, %g3, 15) + stx %o4, [%o1 + 0x30] ! k[12, 13] + stx %o5, [%o1 + 0x38] ! k[14, 15] + ROTL128(%o4, %o5, %g2, %g3, 15) + stx %o4, [%o1 + 0x40] ! k[16, 17] + stx %o5, [%o1 + 0x48] ! k[18, 19] + ROTL128(%o4, %o5, %g2, %g3, 15) + stx %o4, [%o1 + 0x60] ! k[24, 25] + ROTL128(%o4, %o5, %g2, %g3, 15) + stx %o4, [%o1 + 0x70] ! k[28, 29] + stx %o5, [%o1 + 0x78] ! k[30, 31] + ROTL128(%o4, %o5, %g2, %g3, 34) + stx %o4, [%o1 + 0xa0] ! k[40, 41] + stx %o5, [%o1 + 0xa8] ! k[42, 43] + ROTL128(%o4, %o5, %g2, %g3, 17) + stx %o4, [%o1 + 0xc0] ! k[48, 49] + stx %o5, [%o1 + 0xc8] ! k[50, 51] + + ldx [%o1 + 0x00], %o4 ! k[ 0, 1] + ldx [%o1 + 0x08], %o5 ! k[ 2, 3] + ROTL128(%o4, %o5, %g2, %g3, 15) + stx %o4, [%o1 + 0x20] ! k[ 8, 9] + stx %o5, [%o1 + 0x28] ! k[10, 11] + ROTL128(%o4, %o5, %g2, %g3, 30) + stx %o4, [%o1 + 0x50] ! k[20, 21] + stx %o5, [%o1 + 0x58] ! k[22, 23] + ROTL128(%o4, %o5, %g2, %g3, 15) + stx %o5, [%o1 + 0x68] ! k[26, 27] + ROTL128(%o4, %o5, %g2, %g3, 17) + stx %o4, [%o1 + 0x80] ! k[32, 33] + stx %o5, [%o1 + 0x88] ! k[34, 35] + ROTL128(%o4, %o5, %g2, %g3, 17) + stx %o4, [%o1 + 0x90] ! k[36, 37] + stx %o5, [%o1 + 0x98] ! k[38, 39] + ROTL128(%o4, %o5, %g2, %g3, 17) + stx %o4, [%o1 + 0xb0] ! k[44, 45] + stx %o5, [%o1 + 0xb8] ! k[46, 47] + + ba,pt %xcc, 2f + mov (3 * 16 * 4), %o0 + +1: + /* 192-bit or 256-bit key */ + std %f0, [%o1 + 0x30] ! k[12, 13] + std %f2, [%o1 + 0x38] ! k[14, 15] + ldd [%o1 + 0x20], %f4 ! k[ 8, 9] + ldd [%o1 + 0x28], %f6 ! k[10, 11] + fxor %f0, %f4, %f0 + fxor %f2, %f6, %f2 + CAMELLIA_F(24, 2, 0, 2) + CAMELLIA_F(26, 0, 2, 0) + std %f0, [%o1 + 0x10] ! k[ 4, 5] + std %f2, [%o1 + 0x18] ! k[ 6, 7] + MOVDTOX_F0_O4 + MOVDTOX_F2_O5 + ROTL128(%o4, %o5, %g2, %g3, 30) + stx %o4, [%o1 + 0x50] ! k[20, 21] + stx %o5, [%o1 + 0x58] ! k[22, 23] + ROTL128(%o4, %o5, %g2, %g3, 30) + stx %o4, [%o1 + 0xa0] ! k[40, 41] + stx %o5, [%o1 + 0xa8] ! k[42, 43] + ROTL128(%o4, %o5, %g2, %g3, 51) + stx %o4, [%o1 + 0x100] ! k[64, 65] + stx %o5, [%o1 + 0x108] ! k[66, 67] + ldx [%o1 + 0x20], %o4 ! k[ 8, 9] + ldx [%o1 + 0x28], %o5 ! k[10, 11] + ROTL128(%o4, %o5, %g2, %g3, 15) + stx %o4, [%o1 + 0x20] ! k[ 8, 9] + stx %o5, [%o1 + 0x28] ! k[10, 11] + ROTL128(%o4, %o5, %g2, %g3, 15) + stx %o4, [%o1 + 0x40] ! k[16, 17] + stx %o5, [%o1 + 0x48] ! k[18, 19] + ROTL128(%o4, %o5, %g2, %g3, 30) + stx %o4, [%o1 + 0x90] ! k[36, 37] + stx %o5, [%o1 + 0x98] ! k[38, 39] + ROTL128(%o4, %o5, %g2, %g3, 34) + stx %o4, [%o1 + 0xd0] ! k[52, 53] + stx %o5, [%o1 + 0xd8] ! k[54, 55] + ldx [%o1 + 0x30], %o4 ! k[12, 13] + ldx [%o1 + 0x38], %o5 ! k[14, 15] + ROTL128(%o4, %o5, %g2, %g3, 15) + stx %o4, [%o1 + 0x30] ! k[12, 13] + stx %o5, [%o1 + 0x38] ! k[14, 15] + ROTL128(%o4, %o5, %g2, %g3, 30) + stx %o4, [%o1 + 0x70] ! k[28, 29] + stx %o5, [%o1 + 0x78] ! k[30, 31] + srlx %o4, 32, %g2 + srlx %o5, 32, %g3 + stw %o4, [%o1 + 0xc0] ! k[48] + stw %g3, [%o1 + 0xc4] ! k[49] + stw %o5, [%o1 + 0xc8] ! k[50] + stw %g2, [%o1 + 0xcc] ! k[51] + ROTL128(%o4, %o5, %g2, %g3, 49) + stx %o4, [%o1 + 0xe0] ! k[56, 57] + stx %o5, [%o1 + 0xe8] ! k[58, 59] + ldx [%o1 + 0x00], %o4 ! k[ 0, 1] + ldx [%o1 + 0x08], %o5 ! k[ 2, 3] + ROTL128(%o4, %o5, %g2, %g3, 45) + stx %o4, [%o1 + 0x60] ! k[24, 25] + stx %o5, [%o1 + 0x68] ! k[26, 27] + ROTL128(%o4, %o5, %g2, %g3, 15) + stx %o4, [%o1 + 0x80] ! k[32, 33] + stx %o5, [%o1 + 0x88] ! k[34, 35] + ROTL128(%o4, %o5, %g2, %g3, 17) + stx %o4, [%o1 + 0xb0] ! k[44, 45] + stx %o5, [%o1 + 0xb8] ! k[46, 47] + ROTL128(%o4, %o5, %g2, %g3, 34) + stx %o4, [%o1 + 0xf0] ! k[60, 61] + stx %o5, [%o1 + 0xf8] ! k[62, 63] + mov (4 * 16 * 4), %o0 +2: + add %o1, %o0, %o1 + ldd [%o1 + 0x00], %f0 + ldd [%o1 + 0x08], %f2 + std %f0, [%o3 + 0x00] + std %f2, [%o3 + 0x08] + add %o3, 0x10, %o3 +1: + sub %o1, (16 * 4), %o1 + ldd [%o1 + 0x38], %f0 + ldd [%o1 + 0x30], %f2 + ldd [%o1 + 0x28], %f4 + ldd [%o1 + 0x20], %f6 + ldd [%o1 + 0x18], %f8 + ldd [%o1 + 0x10], %f10 + std %f0, [%o3 + 0x00] + std %f2, [%o3 + 0x08] + std %f4, [%o3 + 0x10] + std %f6, [%o3 + 0x18] + std %f8, [%o3 + 0x20] + std %f10, [%o3 + 0x28] + + ldd [%o1 + 0x08], %f0 + ldd [%o1 + 0x00], %f2 + std %f0, [%o3 + 0x30] + std %f2, [%o3 + 0x38] + subcc %o0, (16 * 4), %o0 + bne,pt %icc, 1b + add %o3, (16 * 4), %o3 + + std %f2, [%o3 - 0x10] + std %f0, [%o3 - 0x08] + + retl + VISExit +ENDPROC(camellia_sparc64_key_expand) + + .align 32 +ENTRY(camellia_sparc64_crypt) + /* %o0=key, %o1=input, %o2=output, %o3=key_len */ + VISEntry + + ld [%o1 + 0x00], %f0 + ld [%o1 + 0x04], %f1 + ld [%o1 + 0x08], %f2 + ld [%o1 + 0x0c], %f3 + + ldd [%o0 + 0x00], %f4 + ldd [%o0 + 0x08], %f6 + + cmp %o3, 16 + fxor %f4, %f0, %f0 + be 1f + fxor %f6, %f2, %f2 + + ldd [%o0 + 0x10], %f8 + ldd [%o0 + 0x18], %f10 + ldd [%o0 + 0x20], %f12 + ldd [%o0 + 0x28], %f14 + ldd [%o0 + 0x30], %f16 + ldd [%o0 + 0x38], %f18 + ldd [%o0 + 0x40], %f20 + ldd [%o0 + 0x48], %f22 + add %o0, 0x40, %o0 + + CAMELLIA_6ROUNDS_FL_FLI( 8, 0, 2) + +1: + ldd [%o0 + 0x10], %f8 + ldd [%o0 + 0x18], %f10 + ldd [%o0 + 0x20], %f12 + ldd [%o0 + 0x28], %f14 + ldd [%o0 + 0x30], %f16 + ldd [%o0 + 0x38], %f18 + ldd [%o0 + 0x40], %f20 + ldd [%o0 + 0x48], %f22 + ldd [%o0 + 0x50], %f24 + ldd [%o0 + 0x58], %f26 + ldd [%o0 + 0x60], %f28 + ldd [%o0 + 0x68], %f30 + ldd [%o0 + 0x70], %f32 + ldd [%o0 + 0x78], %f34 + ldd [%o0 + 0x80], %f36 + ldd [%o0 + 0x88], %f38 + ldd [%o0 + 0x90], %f40 + ldd [%o0 + 0x98], %f42 + ldd [%o0 + 0xa0], %f44 + ldd [%o0 + 0xa8], %f46 + ldd [%o0 + 0xb0], %f48 + ldd [%o0 + 0xb8], %f50 + ldd [%o0 + 0xc0], %f52 + ldd [%o0 + 0xc8], %f54 + + CAMELLIA_6ROUNDS_FL_FLI( 8, 0, 2) + CAMELLIA_6ROUNDS_FL_FLI(24, 0, 2) + CAMELLIA_6ROUNDS(40, 0, 2) + fxor %f52, %f2, %f2 + fxor %f54, %f0, %f0 + + st %f2, [%o2 + 0x00] + st %f3, [%o2 + 0x04] + st %f0, [%o2 + 0x08] + st %f1, [%o2 + 0x0c] + + retl + VISExit +ENDPROC(camellia_sparc64_crypt) + + .align 32 +ENTRY(camellia_sparc64_load_keys) + /* %o0=key, %o1=key_len */ + VISEntry + ldd [%o0 + 0x00], %f4 + ldd [%o0 + 0x08], %f6 + ldd [%o0 + 0x10], %f8 + ldd [%o0 + 0x18], %f10 + ldd [%o0 + 0x20], %f12 + ldd [%o0 + 0x28], %f14 + ldd [%o0 + 0x30], %f16 + ldd [%o0 + 0x38], %f18 + ldd [%o0 + 0x40], %f20 + ldd [%o0 + 0x48], %f22 + ldd [%o0 + 0x50], %f24 + ldd [%o0 + 0x58], %f26 + ldd [%o0 + 0x60], %f28 + ldd [%o0 + 0x68], %f30 + ldd [%o0 + 0x70], %f32 + ldd [%o0 + 0x78], %f34 + ldd [%o0 + 0x80], %f36 + ldd [%o0 + 0x88], %f38 + ldd [%o0 + 0x90], %f40 + ldd [%o0 + 0x98], %f42 + ldd [%o0 + 0xa0], %f44 + ldd [%o0 + 0xa8], %f46 + ldd [%o0 + 0xb0], %f48 + ldd [%o0 + 0xb8], %f50 + ldd [%o0 + 0xc0], %f52 + retl + ldd [%o0 + 0xc8], %f54 +ENDPROC(camellia_sparc64_load_keys) + + .align 32 +ENTRY(camellia_sparc64_ecb_crypt_3_grand_rounds) + /* %o0=input, %o1=output, %o2=len, %o3=key */ +1: ldd [%o0 + 0x00], %f0 + ldd [%o0 + 0x08], %f2 + add %o0, 0x10, %o0 + fxor %f4, %f0, %f0 + fxor %f6, %f2, %f2 + CAMELLIA_6ROUNDS_FL_FLI( 8, 0, 2) + CAMELLIA_6ROUNDS_FL_FLI(24, 0, 2) + CAMELLIA_6ROUNDS(40, 0, 2) + fxor %f52, %f2, %f2 + fxor %f54, %f0, %f0 + std %f2, [%o1 + 0x00] + std %f0, [%o1 + 0x08] + subcc %o2, 0x10, %o2 + bne,pt %icc, 1b + add %o1, 0x10, %o1 + retl + nop +ENDPROC(camellia_sparc64_ecb_crypt_3_grand_rounds) + + .align 32 +ENTRY(camellia_sparc64_ecb_crypt_4_grand_rounds) + /* %o0=input, %o1=output, %o2=len, %o3=key */ +1: ldd [%o0 + 0x00], %f0 + ldd [%o0 + 0x08], %f2 + add %o0, 0x10, %o0 + fxor %f4, %f0, %f0 + fxor %f6, %f2, %f2 + CAMELLIA_6ROUNDS_FL_FLI( 8, 0, 2) + ldd [%o3 + 0xd0], %f8 + ldd [%o3 + 0xd8], %f10 + ldd [%o3 + 0xe0], %f12 + ldd [%o3 + 0xe8], %f14 + ldd [%o3 + 0xf0], %f16 + ldd [%o3 + 0xf8], %f18 + ldd [%o3 + 0x100], %f20 + ldd [%o3 + 0x108], %f22 + CAMELLIA_6ROUNDS_FL_FLI(24, 0, 2) + CAMELLIA_6ROUNDS_FL_FLI(40, 0, 2) + CAMELLIA_F(8, 2, 0, 2) + CAMELLIA_F(10, 0, 2, 0) + ldd [%o3 + 0x10], %f8 + ldd [%o3 + 0x18], %f10 + CAMELLIA_F(12, 2, 0, 2) + CAMELLIA_F(14, 0, 2, 0) + ldd [%o3 + 0x20], %f12 + ldd [%o3 + 0x28], %f14 + CAMELLIA_F(16, 2, 0, 2) + CAMELLIA_F(18, 0, 2, 0) + ldd [%o3 + 0x30], %f16 + ldd [%o3 + 0x38], %f18 + fxor %f20, %f2, %f2 + fxor %f22, %f0, %f0 + ldd [%o3 + 0x40], %f20 + ldd [%o3 + 0x48], %f22 + std %f2, [%o1 + 0x00] + std %f0, [%o1 + 0x08] + subcc %o2, 0x10, %o2 + bne,pt %icc, 1b + add %o1, 0x10, %o1 + retl + nop +ENDPROC(camellia_sparc64_ecb_crypt_4_grand_rounds) + + .align 32 +ENTRY(camellia_sparc64_cbc_encrypt_3_grand_rounds) + /* %o0=input, %o1=output, %o2=len, %o3=key, %o4=IV */ + ldd [%o4 + 0x00], %f60 + ldd [%o4 + 0x08], %f62 +1: ldd [%o0 + 0x00], %f0 + ldd [%o0 + 0x08], %f2 + add %o0, 0x10, %o0 + fxor %f60, %f0, %f0 + fxor %f62, %f2, %f2 + fxor %f4, %f0, %f0 + fxor %f6, %f2, %f2 + CAMELLIA_6ROUNDS_FL_FLI( 8, 0, 2) + CAMELLIA_6ROUNDS_FL_FLI(24, 0, 2) + CAMELLIA_6ROUNDS(40, 0, 2) + fxor %f52, %f2, %f60 + fxor %f54, %f0, %f62 + std %f60, [%o1 + 0x00] + std %f62, [%o1 + 0x08] + subcc %o2, 0x10, %o2 + bne,pt %icc, 1b + add %o1, 0x10, %o1 + std %f60, [%o4 + 0x00] + retl + std %f62, [%o4 + 0x08] +ENDPROC(camellia_sparc64_cbc_encrypt_3_grand_rounds) + + .align 32 +ENTRY(camellia_sparc64_cbc_encrypt_4_grand_rounds) + /* %o0=input, %o1=output, %o2=len, %o3=key, %o4=IV */ + ldd [%o4 + 0x00], %f60 + ldd [%o4 + 0x08], %f62 +1: ldd [%o0 + 0x00], %f0 + ldd [%o0 + 0x08], %f2 + add %o0, 0x10, %o0 + fxor %f60, %f0, %f0 + fxor %f62, %f2, %f2 + fxor %f4, %f0, %f0 + fxor %f6, %f2, %f2 + CAMELLIA_6ROUNDS_FL_FLI( 8, 0, 2) + ldd [%o3 + 0xd0], %f8 + ldd [%o3 + 0xd8], %f10 + ldd [%o3 + 0xe0], %f12 + ldd [%o3 + 0xe8], %f14 + ldd [%o3 + 0xf0], %f16 + ldd [%o3 + 0xf8], %f18 + ldd [%o3 + 0x100], %f20 + ldd [%o3 + 0x108], %f22 + CAMELLIA_6ROUNDS_FL_FLI(24, 0, 2) + CAMELLIA_6ROUNDS_FL_FLI(40, 0, 2) + CAMELLIA_F(8, 2, 0, 2) + CAMELLIA_F(10, 0, 2, 0) + ldd [%o3 + 0x10], %f8 + ldd [%o3 + 0x18], %f10 + CAMELLIA_F(12, 2, 0, 2) + CAMELLIA_F(14, 0, 2, 0) + ldd [%o3 + 0x20], %f12 + ldd [%o3 + 0x28], %f14 + CAMELLIA_F(16, 2, 0, 2) + CAMELLIA_F(18, 0, 2, 0) + ldd [%o3 + 0x30], %f16 + ldd [%o3 + 0x38], %f18 + fxor %f20, %f2, %f60 + fxor %f22, %f0, %f62 + ldd [%o3 + 0x40], %f20 + ldd [%o3 + 0x48], %f22 + std %f60, [%o1 + 0x00] + std %f62, [%o1 + 0x08] + subcc %o2, 0x10, %o2 + bne,pt %icc, 1b + add %o1, 0x10, %o1 + std %f60, [%o4 + 0x00] + retl + std %f62, [%o4 + 0x08] +ENDPROC(camellia_sparc64_cbc_encrypt_4_grand_rounds) + + .align 32 +ENTRY(camellia_sparc64_cbc_decrypt_3_grand_rounds) + /* %o0=input, %o1=output, %o2=len, %o3=key, %o4=IV */ + ldd [%o4 + 0x00], %f60 + ldd [%o4 + 0x08], %f62 +1: ldd [%o0 + 0x00], %f56 + ldd [%o0 + 0x08], %f58 + add %o0, 0x10, %o0 + fxor %f4, %f56, %f0 + fxor %f6, %f58, %f2 + CAMELLIA_6ROUNDS_FL_FLI( 8, 0, 2) + CAMELLIA_6ROUNDS_FL_FLI(24, 0, 2) + CAMELLIA_6ROUNDS(40, 0, 2) + fxor %f52, %f2, %f2 + fxor %f54, %f0, %f0 + fxor %f60, %f2, %f2 + fxor %f62, %f0, %f0 + fsrc2 %f56, %f60 + fsrc2 %f58, %f62 + std %f2, [%o1 + 0x00] + std %f0, [%o1 + 0x08] + subcc %o2, 0x10, %o2 + bne,pt %icc, 1b + add %o1, 0x10, %o1 + std %f60, [%o4 + 0x00] + retl + std %f62, [%o4 + 0x08] +ENDPROC(camellia_sparc64_cbc_decrypt_3_grand_rounds) + + .align 32 +ENTRY(camellia_sparc64_cbc_decrypt_4_grand_rounds) + /* %o0=input, %o1=output, %o2=len, %o3=key, %o4=IV */ + ldd [%o4 + 0x00], %f60 + ldd [%o4 + 0x08], %f62 +1: ldd [%o0 + 0x00], %f56 + ldd [%o0 + 0x08], %f58 + add %o0, 0x10, %o0 + fxor %f4, %f56, %f0 + fxor %f6, %f58, %f2 + CAMELLIA_6ROUNDS_FL_FLI( 8, 0, 2) + ldd [%o3 + 0xd0], %f8 + ldd [%o3 + 0xd8], %f10 + ldd [%o3 + 0xe0], %f12 + ldd [%o3 + 0xe8], %f14 + ldd [%o3 + 0xf0], %f16 + ldd [%o3 + 0xf8], %f18 + ldd [%o3 + 0x100], %f20 + ldd [%o3 + 0x108], %f22 + CAMELLIA_6ROUNDS_FL_FLI(24, 0, 2) + CAMELLIA_6ROUNDS_FL_FLI(40, 0, 2) + CAMELLIA_F(8, 2, 0, 2) + CAMELLIA_F(10, 0, 2, 0) + ldd [%o3 + 0x10], %f8 + ldd [%o3 + 0x18], %f10 + CAMELLIA_F(12, 2, 0, 2) + CAMELLIA_F(14, 0, 2, 0) + ldd [%o3 + 0x20], %f12 + ldd [%o3 + 0x28], %f14 + CAMELLIA_F(16, 2, 0, 2) + CAMELLIA_F(18, 0, 2, 0) + ldd [%o3 + 0x30], %f16 + ldd [%o3 + 0x38], %f18 + fxor %f20, %f2, %f2 + fxor %f22, %f0, %f0 + ldd [%o3 + 0x40], %f20 + ldd [%o3 + 0x48], %f22 + fxor %f60, %f2, %f2 + fxor %f62, %f0, %f0 + fsrc2 %f56, %f60 + fsrc2 %f58, %f62 + std %f2, [%o1 + 0x00] + std %f0, [%o1 + 0x08] + subcc %o2, 0x10, %o2 + bne,pt %icc, 1b + add %o1, 0x10, %o1 + std %f60, [%o4 + 0x00] + retl + std %f62, [%o4 + 0x08] +ENDPROC(camellia_sparc64_cbc_decrypt_4_grand_rounds) diff --git a/arch/sparc/crypto/camellia_glue.c b/arch/sparc/crypto/camellia_glue.c new file mode 100644 index 000000000000..c258cc550a6b --- /dev/null +++ b/arch/sparc/crypto/camellia_glue.c @@ -0,0 +1,318 @@ +/* Glue code for CAMELLIA encryption optimized for sparc64 crypto opcodes. + * + * Copyright (C) 2012 David S. Miller + */ + +#include +#include +#include +#include +#include +#include + +#include +#include +#include + +#define CAMELLIA_MIN_KEY_SIZE 16 +#define CAMELLIA_MAX_KEY_SIZE 32 +#define CAMELLIA_BLOCK_SIZE 16 +#define CAMELLIA_TABLE_BYTE_LEN 272 + +struct camellia_sparc64_ctx { + u64 encrypt_key[CAMELLIA_TABLE_BYTE_LEN / sizeof(u64)]; + u64 decrypt_key[CAMELLIA_TABLE_BYTE_LEN / sizeof(u64)]; + int key_len; +}; + +extern void camellia_sparc64_key_expand(const u32 *in_key, u64 *encrypt_key, + unsigned int key_len, u64 *decrypt_key); + +static int camellia_set_key(struct crypto_tfm *tfm, const u8 *_in_key, + unsigned int key_len) +{ + struct camellia_sparc64_ctx *ctx = crypto_tfm_ctx(tfm); + const u32 *in_key = (const u32 *) _in_key; + u32 *flags = &tfm->crt_flags; + + if (key_len != 16 && key_len != 24 && key_len != 32) { + *flags |= CRYPTO_TFM_RES_BAD_KEY_LEN; + return -EINVAL; + } + + ctx->key_len = key_len; + + camellia_sparc64_key_expand(in_key, &ctx->encrypt_key[0], + key_len, &ctx->decrypt_key[0]); + return 0; +} + +extern void camellia_sparc64_crypt(const u64 *key, const u32 *input, + u32 *output, unsigned int key_len); + +static void camellia_encrypt(struct crypto_tfm *tfm, u8 *dst, const u8 *src) +{ + struct camellia_sparc64_ctx *ctx = crypto_tfm_ctx(tfm); + + camellia_sparc64_crypt(&ctx->encrypt_key[0], + (const u32 *) src, + (u32 *) dst, ctx->key_len); +} + +static void camellia_decrypt(struct crypto_tfm *tfm, u8 *dst, const u8 *src) +{ + struct camellia_sparc64_ctx *ctx = crypto_tfm_ctx(tfm); + + camellia_sparc64_crypt(&ctx->decrypt_key[0], + (const u32 *) src, + (u32 *) dst, ctx->key_len); +} + +extern void camellia_sparc64_load_keys(const u64 *key, unsigned int key_len); + +typedef void ecb_crypt_op(const u64 *input, u64 *output, unsigned int len, + const u64 *key); + +extern ecb_crypt_op camellia_sparc64_ecb_crypt_3_grand_rounds; +extern ecb_crypt_op camellia_sparc64_ecb_crypt_4_grand_rounds; + +#define CAMELLIA_BLOCK_MASK (~(CAMELLIA_BLOCK_SIZE - 1)) + +static int __ecb_crypt(struct blkcipher_desc *desc, + struct scatterlist *dst, struct scatterlist *src, + unsigned int nbytes, bool encrypt) +{ + struct camellia_sparc64_ctx *ctx = crypto_blkcipher_ctx(desc->tfm); + struct blkcipher_walk walk; + ecb_crypt_op *op; + const u64 *key; + int err; + + op = camellia_sparc64_ecb_crypt_3_grand_rounds; + if (ctx->key_len != 16) + op = camellia_sparc64_ecb_crypt_4_grand_rounds; + + blkcipher_walk_init(&walk, dst, src, nbytes); + err = blkcipher_walk_virt(desc, &walk); + + if (encrypt) + key = &ctx->encrypt_key[0]; + else + key = &ctx->decrypt_key[0]; + camellia_sparc64_load_keys(key, ctx->key_len); + while ((nbytes = walk.nbytes)) { + unsigned int block_len = nbytes & CAMELLIA_BLOCK_MASK; + + if (likely(block_len)) { + const u64 *src64; + u64 *dst64; + + src64 = (const u64 *)walk.src.virt.addr; + dst64 = (u64 *) walk.dst.virt.addr; + op(src64, dst64, block_len, key); + } + nbytes &= CAMELLIA_BLOCK_SIZE - 1; + err = blkcipher_walk_done(desc, &walk, nbytes); + } + fprs_write(0); + return err; +} + +static int ecb_encrypt(struct blkcipher_desc *desc, + struct scatterlist *dst, struct scatterlist *src, + unsigned int nbytes) +{ + return __ecb_crypt(desc, dst, src, nbytes, true); +} + +static int ecb_decrypt(struct blkcipher_desc *desc, + struct scatterlist *dst, struct scatterlist *src, + unsigned int nbytes) +{ + return __ecb_crypt(desc, dst, src, nbytes, false); +} + +typedef void cbc_crypt_op(const u64 *input, u64 *output, unsigned int len, + const u64 *key, u64 *iv); + +extern cbc_crypt_op camellia_sparc64_cbc_encrypt_3_grand_rounds; +extern cbc_crypt_op camellia_sparc64_cbc_encrypt_4_grand_rounds; +extern cbc_crypt_op camellia_sparc64_cbc_decrypt_3_grand_rounds; +extern cbc_crypt_op camellia_sparc64_cbc_decrypt_4_grand_rounds; + +static int cbc_encrypt(struct blkcipher_desc *desc, + struct scatterlist *dst, struct scatterlist *src, + unsigned int nbytes) +{ + struct camellia_sparc64_ctx *ctx = crypto_blkcipher_ctx(desc->tfm); + struct blkcipher_walk walk; + cbc_crypt_op *op; + const u64 *key; + int err; + + op = camellia_sparc64_cbc_encrypt_3_grand_rounds; + if (ctx->key_len != 16) + op = camellia_sparc64_cbc_encrypt_4_grand_rounds; + + blkcipher_walk_init(&walk, dst, src, nbytes); + err = blkcipher_walk_virt(desc, &walk); + + key = &ctx->encrypt_key[0]; + camellia_sparc64_load_keys(key, ctx->key_len); + while ((nbytes = walk.nbytes)) { + unsigned int block_len = nbytes & CAMELLIA_BLOCK_MASK; + + if (likely(block_len)) { + const u64 *src64; + u64 *dst64; + + src64 = (const u64 *)walk.src.virt.addr; + dst64 = (u64 *) walk.dst.virt.addr; + op(src64, dst64, block_len, key, + (u64 *) walk.iv); + } + nbytes &= CAMELLIA_BLOCK_SIZE - 1; + err = blkcipher_walk_done(desc, &walk, nbytes); + } + fprs_write(0); + return err; +} + +static int cbc_decrypt(struct blkcipher_desc *desc, + struct scatterlist *dst, struct scatterlist *src, + unsigned int nbytes) +{ + struct camellia_sparc64_ctx *ctx = crypto_blkcipher_ctx(desc->tfm); + struct blkcipher_walk walk; + cbc_crypt_op *op; + const u64 *key; + int err; + + op = camellia_sparc64_cbc_decrypt_3_grand_rounds; + if (ctx->key_len != 16) + op = camellia_sparc64_cbc_decrypt_4_grand_rounds; + + blkcipher_walk_init(&walk, dst, src, nbytes); + err = blkcipher_walk_virt(desc, &walk); + + key = &ctx->decrypt_key[0]; + camellia_sparc64_load_keys(key, ctx->key_len); + while ((nbytes = walk.nbytes)) { + unsigned int block_len = nbytes & CAMELLIA_BLOCK_MASK; + + if (likely(block_len)) { + const u64 *src64; + u64 *dst64; + + src64 = (const u64 *)walk.src.virt.addr; + dst64 = (u64 *) walk.dst.virt.addr; + op(src64, dst64, block_len, key, + (u64 *) walk.iv); + } + nbytes &= CAMELLIA_BLOCK_SIZE - 1; + err = blkcipher_walk_done(desc, &walk, nbytes); + } + fprs_write(0); + return err; +} + +static struct crypto_alg algs[] = { { + .cra_name = "camellia", + .cra_driver_name = "camellia-sparc64", + .cra_priority = 150, + .cra_flags = CRYPTO_ALG_TYPE_CIPHER, + .cra_blocksize = CAMELLIA_BLOCK_SIZE, + .cra_ctxsize = sizeof(struct camellia_sparc64_ctx), + .cra_alignmask = 3, + .cra_module = THIS_MODULE, + .cra_u = { + .cipher = { + .cia_min_keysize = CAMELLIA_MIN_KEY_SIZE, + .cia_max_keysize = CAMELLIA_MAX_KEY_SIZE, + .cia_setkey = camellia_set_key, + .cia_encrypt = camellia_encrypt, + .cia_decrypt = camellia_decrypt + } + } +}, { + .cra_name = "ecb(camellia)", + .cra_driver_name = "ecb-camellia-sparc64", + .cra_priority = 150, + .cra_flags = CRYPTO_ALG_TYPE_BLKCIPHER, + .cra_blocksize = CAMELLIA_BLOCK_SIZE, + .cra_ctxsize = sizeof(struct camellia_sparc64_ctx), + .cra_alignmask = 7, + .cra_type = &crypto_blkcipher_type, + .cra_module = THIS_MODULE, + .cra_u = { + .blkcipher = { + .min_keysize = CAMELLIA_MIN_KEY_SIZE, + .max_keysize = CAMELLIA_MAX_KEY_SIZE, + .setkey = camellia_set_key, + .encrypt = ecb_encrypt, + .decrypt = ecb_decrypt, + }, + }, +}, { + .cra_name = "cbc(camellia)", + .cra_driver_name = "cbc-camellia-sparc64", + .cra_priority = 150, + .cra_flags = CRYPTO_ALG_TYPE_BLKCIPHER, + .cra_blocksize = CAMELLIA_BLOCK_SIZE, + .cra_ctxsize = sizeof(struct camellia_sparc64_ctx), + .cra_alignmask = 7, + .cra_type = &crypto_blkcipher_type, + .cra_module = THIS_MODULE, + .cra_u = { + .blkcipher = { + .min_keysize = CAMELLIA_MIN_KEY_SIZE, + .max_keysize = CAMELLIA_MAX_KEY_SIZE, + .setkey = camellia_set_key, + .encrypt = cbc_encrypt, + .decrypt = cbc_decrypt, + }, + }, +} +}; + +static bool __init sparc64_has_camellia_opcode(void) +{ + unsigned long cfr; + + if (!(sparc64_elf_hwcap & HWCAP_SPARC_CRYPTO)) + return false; + + __asm__ __volatile__("rd %%asr26, %0" : "=r" (cfr)); + if (!(cfr & CFR_CAMELLIA)) + return false; + + return true; +} + +static int __init camellia_sparc64_mod_init(void) +{ + int i; + + for (i = 0; i < ARRAY_SIZE(algs); i++) + INIT_LIST_HEAD(&algs[i].cra_list); + + if (sparc64_has_camellia_opcode()) { + pr_info("Using sparc64 camellia opcodes optimized CAMELLIA implementation\n"); + return crypto_register_algs(algs, ARRAY_SIZE(algs)); + } + pr_info("sparc64 camellia opcodes not available.\n"); + return -ENODEV; +} + +static void __exit camellia_sparc64_mod_fini(void) +{ + crypto_unregister_algs(algs, ARRAY_SIZE(algs)); +} + +module_init(camellia_sparc64_mod_init); +module_exit(camellia_sparc64_mod_fini); + +MODULE_LICENSE("GPL"); +MODULE_DESCRIPTION("Camellia Cipher Algorithm, sparc64 camellia opcode accelerated"); + +MODULE_ALIAS("aes"); -- cgit v1.2.2