From: Eric Biggers <ebiggers@xxxxxxxxxx> Remove the AEGIS assembly code paths that were "optimized" to operate on 16-byte aligned data using movdqa, and instead just use the code paths that use movdqu and can handle data with any alignment. This does not reduce performance. movdqa is basically a historical artifact; on aligned data, movdqu and movdqa have had the same performance since Intel Nehalem (2008) and AMD Bulldozer (2011). And code that requires AES-NI cannot run on CPUs older than those anyway. Signed-off-by: Eric Biggers <ebiggers@xxxxxxxxxx> --- arch/x86/crypto/aegis128-aesni-asm.S | 122 +++++---------------------- 1 file changed, 22 insertions(+), 100 deletions(-) diff --git a/arch/x86/crypto/aegis128-aesni-asm.S b/arch/x86/crypto/aegis128-aesni-asm.S index 1b57558548c7..5541aca2fd0d 100644 --- a/arch/x86/crypto/aegis128-aesni-asm.S +++ b/arch/x86/crypto/aegis128-aesni-asm.S @@ -243,56 +243,12 @@ SYM_FUNC_START(crypto_aegis128_aesni_ad) movdqu 0x10(STATEP), STATE1 movdqu 0x20(STATEP), STATE2 movdqu 0x30(STATEP), STATE3 movdqu 0x40(STATEP), STATE4 - mov SRC, %r8 - and $0xF, %r8 - jnz .Lad_u_loop - -.align 8 -.Lad_a_loop: - movdqa 0x00(SRC), MSG - aegis128_update - pxor MSG, STATE4 - sub $0x10, LEN - cmp $0x10, LEN - jl .Lad_out_1 - - movdqa 0x10(SRC), MSG - aegis128_update - pxor MSG, STATE3 - sub $0x10, LEN - cmp $0x10, LEN - jl .Lad_out_2 - - movdqa 0x20(SRC), MSG - aegis128_update - pxor MSG, STATE2 - sub $0x10, LEN - cmp $0x10, LEN - jl .Lad_out_3 - - movdqa 0x30(SRC), MSG - aegis128_update - pxor MSG, STATE1 - sub $0x10, LEN - cmp $0x10, LEN - jl .Lad_out_4 - - movdqa 0x40(SRC), MSG - aegis128_update - pxor MSG, STATE0 - sub $0x10, LEN - cmp $0x10, LEN - jl .Lad_out_0 - - add $0x50, SRC - jmp .Lad_a_loop - .align 8 -.Lad_u_loop: +.Lad_loop: movdqu 0x00(SRC), MSG aegis128_update pxor MSG, STATE4 sub $0x10, LEN cmp $0x10, LEN @@ -325,11 +281,11 @@ SYM_FUNC_START(crypto_aegis128_aesni_ad) sub $0x10, LEN cmp $0x10, LEN jl .Lad_out_0 add $0x50, SRC - jmp .Lad_u_loop + jmp .Lad_loop /* store the state: */ .Lad_out_0: movdqu STATE0, 0x00(STATEP) movdqu STATE1, 0x10(STATEP) @@ -378,19 +334,19 @@ SYM_FUNC_START(crypto_aegis128_aesni_ad) .Lad_out: FRAME_END RET SYM_FUNC_END(crypto_aegis128_aesni_ad) -.macro encrypt_block a s0 s1 s2 s3 s4 i - movdq\a (\i * 0x10)(SRC), MSG +.macro encrypt_block s0 s1 s2 s3 s4 i + movdqu (\i * 0x10)(SRC), MSG movdqa MSG, T0 pxor \s1, T0 pxor \s4, T0 movdqa \s2, T1 pand \s3, T1 pxor T1, T0 - movdq\a T0, (\i * 0x10)(DST) + movdqu T0, (\i * 0x10)(DST) aegis128_update pxor MSG, \s4 sub $0x10, LEN @@ -413,38 +369,21 @@ SYM_FUNC_START(crypto_aegis128_aesni_enc) movdqu 0x10(STATEP), STATE1 movdqu 0x20(STATEP), STATE2 movdqu 0x30(STATEP), STATE3 movdqu 0x40(STATEP), STATE4 - mov SRC, %r8 - or DST, %r8 - and $0xF, %r8 - jnz .Lenc_u_loop - .align 8 -.Lenc_a_loop: - encrypt_block a STATE0 STATE1 STATE2 STATE3 STATE4 0 - encrypt_block a STATE4 STATE0 STATE1 STATE2 STATE3 1 - encrypt_block a STATE3 STATE4 STATE0 STATE1 STATE2 2 - encrypt_block a STATE2 STATE3 STATE4 STATE0 STATE1 3 - encrypt_block a STATE1 STATE2 STATE3 STATE4 STATE0 4 +.Lenc_loop: + encrypt_block STATE0 STATE1 STATE2 STATE3 STATE4 0 + encrypt_block STATE4 STATE0 STATE1 STATE2 STATE3 1 + encrypt_block STATE3 STATE4 STATE0 STATE1 STATE2 2 + encrypt_block STATE2 STATE3 STATE4 STATE0 STATE1 3 + encrypt_block STATE1 STATE2 STATE3 STATE4 STATE0 4 add $0x50, SRC add $0x50, DST - jmp .Lenc_a_loop - -.align 8 -.Lenc_u_loop: - encrypt_block u STATE0 STATE1 STATE2 STATE3 STATE4 0 - encrypt_block u STATE4 STATE0 STATE1 STATE2 STATE3 1 - encrypt_block u STATE3 STATE4 STATE0 STATE1 STATE2 2 - encrypt_block u STATE2 STATE3 STATE4 STATE0 STATE1 3 - encrypt_block u STATE1 STATE2 STATE3 STATE4 STATE0 4 - - add $0x50, SRC - add $0x50, DST - jmp .Lenc_u_loop + jmp .Lenc_loop /* store the state: */ .Lenc_out_0: movdqu STATE4, 0x00(STATEP) movdqu STATE0, 0x10(STATEP) @@ -533,18 +472,18 @@ SYM_FUNC_START(crypto_aegis128_aesni_enc_tail) FRAME_END RET SYM_FUNC_END(crypto_aegis128_aesni_enc_tail) -.macro decrypt_block a s0 s1 s2 s3 s4 i - movdq\a (\i * 0x10)(SRC), MSG +.macro decrypt_block s0 s1 s2 s3 s4 i + movdqu (\i * 0x10)(SRC), MSG pxor \s1, MSG pxor \s4, MSG movdqa \s2, T1 pand \s3, T1 pxor T1, MSG - movdq\a MSG, (\i * 0x10)(DST) + movdqu MSG, (\i * 0x10)(DST) aegis128_update pxor MSG, \s4 sub $0x10, LEN @@ -567,38 +506,21 @@ SYM_FUNC_START(crypto_aegis128_aesni_dec) movdqu 0x10(STATEP), STATE1 movdqu 0x20(STATEP), STATE2 movdqu 0x30(STATEP), STATE3 movdqu 0x40(STATEP), STATE4 - mov SRC, %r8 - or DST, %r8 - and $0xF, %r8 - jnz .Ldec_u_loop - -.align 8 -.Ldec_a_loop: - decrypt_block a STATE0 STATE1 STATE2 STATE3 STATE4 0 - decrypt_block a STATE4 STATE0 STATE1 STATE2 STATE3 1 - decrypt_block a STATE3 STATE4 STATE0 STATE1 STATE2 2 - decrypt_block a STATE2 STATE3 STATE4 STATE0 STATE1 3 - decrypt_block a STATE1 STATE2 STATE3 STATE4 STATE0 4 - - add $0x50, SRC - add $0x50, DST - jmp .Ldec_a_loop - .align 8 -.Ldec_u_loop: - decrypt_block u STATE0 STATE1 STATE2 STATE3 STATE4 0 - decrypt_block u STATE4 STATE0 STATE1 STATE2 STATE3 1 - decrypt_block u STATE3 STATE4 STATE0 STATE1 STATE2 2 - decrypt_block u STATE2 STATE3 STATE4 STATE0 STATE1 3 - decrypt_block u STATE1 STATE2 STATE3 STATE4 STATE0 4 +.Ldec_loop: + decrypt_block STATE0 STATE1 STATE2 STATE3 STATE4 0 + decrypt_block STATE4 STATE0 STATE1 STATE2 STATE3 1 + decrypt_block STATE3 STATE4 STATE0 STATE1 STATE2 2 + decrypt_block STATE2 STATE3 STATE4 STATE0 STATE1 3 + decrypt_block STATE1 STATE2 STATE3 STATE4 STATE0 4 add $0x50, SRC add $0x50, DST - jmp .Ldec_u_loop + jmp .Ldec_loop /* store the state: */ .Ldec_out_0: movdqu STATE4, 0x00(STATEP) movdqu STATE0, 0x10(STATEP) -- 2.46.2
