diff options
Diffstat (limited to 'src/spdk/intel-ipsec-mb/sse/mb_mgr_hmac_submit_ni_sse.asm')
-rw-r--r-- | src/spdk/intel-ipsec-mb/sse/mb_mgr_hmac_submit_ni_sse.asm | 370 |
1 files changed, 370 insertions, 0 deletions
diff --git a/src/spdk/intel-ipsec-mb/sse/mb_mgr_hmac_submit_ni_sse.asm b/src/spdk/intel-ipsec-mb/sse/mb_mgr_hmac_submit_ni_sse.asm new file mode 100644 index 000000000..e0b0460f4 --- /dev/null +++ b/src/spdk/intel-ipsec-mb/sse/mb_mgr_hmac_submit_ni_sse.asm @@ -0,0 +1,370 @@ +;; +;; Copyright (c) 2012-2018, Intel Corporation +;; +;; Redistribution and use in source and binary forms, with or without +;; modification, are permitted provided that the following conditions are met: +;; +;; * Redistributions of source code must retain the above copyright notice, +;; this list of conditions and the following disclaimer. +;; * Redistributions in binary form must reproduce the above copyright +;; notice, this list of conditions and the following disclaimer in the +;; documentation and/or other materials provided with the distribution. +;; * Neither the name of Intel Corporation nor the names of its contributors +;; may be used to endorse or promote products derived from this software +;; without specific prior written permission. +;; +;; THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +;; AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +;; IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +;; DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE +;; FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +;; DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR +;; SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER +;; CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, +;; OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +;; OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +;; + +;; In System V AMD64 ABI +;; calle saves: RBX, RBP, R12-R15 +;; Windows x64 ABI +;; calle saves: RBX, RBP, RDI, RSI, RSP, R12-R15 +;; +;; Registers: RAX RBX RCX RDX RBP RSI RDI R8 R9 R10 R11 R12 R13 R14 R15 +;; ----------------------------------------------------------- +;; Windows clobbers: RAX RCX RDX R8 R9 R10 R11 +;; Windows preserves: RBX RBP RSI RDI R12 R13 R14 R15 +;; ----------------------------------------------------------- +;; Linux clobbers: RAX RCX RDX RSI RDI R8 R9 R10 R11 +;; Linux preserves: RBX RBP R12 R13 R14 R15 +;; ----------------------------------------------------------- +;; +;; Linux/Windows clobbers: xmm0 - xmm15 +;; + +%include "include/os.asm" +%include "job_aes_hmac.asm" +%include "mb_mgr_datastruct.asm" +%include "include/reg_sizes.asm" +%include "include/memcpy.asm" + +;%define DO_DBGPRINT +%include "include/dbgprint.asm" + +extern sha1_ni + +section .data +default rel + +align 16 +byteswap: + dq 0x0405060700010203 + dq 0x0c0d0e0f08090a0b + +section .text + +%ifdef LINUX +%define arg1 rdi +%define arg2 rsi +%define reg3 rcx +%define reg4 rdx +%else +%define arg1 rcx +%define arg2 rdx +%define reg3 rdi +%define reg4 rsi +%endif + +%define state arg1 +%define job arg2 +%define len2 arg2 + +; idx needs to be in rbx, rbp, r12-r15 +%define last_len rbp +%define idx rbp +%define p4 rbp + +%define p r11 +%define start_offset r11 + +%define unused_lanes rbx +%define tmp4 rbx +%define p3 rbx + +%define job_rax rax +%define len rax + +%define size_offset reg3 +%define tmp2 reg3 + +%define lane reg4 +%define tmp3 reg4 + +%define extra_blocks r8 + +%define tmp r9 +%define p2 r9 + +%define lane_data r10 + +struc STACK +_gpr_save: resq 4 +_rsp_save: resq 1 +endstruc + +; JOB* submit_job_hmac_ni_sse(MB_MGR_HMAC_SHA_1_OOO *state, JOB_AES_HMAC *job) +; arg 1 : rcx : state +; arg 2 : rdx : job +MKGLOBAL(submit_job_hmac_ni_sse,function,internal) +submit_job_hmac_ni_sse: + + mov rax, rsp + sub rsp, STACK_size + and rsp, -16 + + mov [rsp + _gpr_save + 8*0], rbx + mov [rsp + _gpr_save + 8*1], rbp +%ifndef LINUX + mov [rsp + _gpr_save + 8*2], rsi + mov [rsp + _gpr_save + 8*3], rdi +%endif + mov [rsp + _rsp_save], rax ; original SP + + DBGPRINTL "enter sha1-ni-sse submit" + mov unused_lanes, [state + _unused_lanes] + movzx lane, BYTE(unused_lanes) + DBGPRINTL64 "lane: ", lane + shr unused_lanes, 8 + imul lane_data, lane, _HMAC_SHA1_LANE_DATA_size + lea lane_data, [state + _ldata + lane_data] + mov [state + _unused_lanes], unused_lanes + mov len, [job + _msg_len_to_hash_in_bytes] + DBGPRINTL64 "length: ", len + mov tmp, len + shr tmp, 6 ; divide by 64, len in terms of blocks + + mov [lane_data + _job_in_lane], job + mov dword [lane_data + _outer_done], 0 + mov [state + _lens + 2*lane], WORD(tmp) + + mov last_len, len + and last_len, 63 + lea extra_blocks, [last_len + 9 + 63] + shr extra_blocks, 6 + mov [lane_data + _extra_blocks], DWORD(extra_blocks) + + mov p, [job + _src] + add p, [job + _hash_start_src_offset_in_bytes] + DBGPRINTL64 "src pointer + offset:", p + mov [state + _args_data_ptr + PTR_SZ*lane], p + cmp len, 64 + jb copy_lt64 + +fast_copy: + add p, len + movdqu xmm0, [p - 64 + 0*16] + movdqu xmm1, [p - 64 + 1*16] + movdqu xmm2, [p - 64 + 2*16] + movdqu xmm3, [p - 64 + 3*16] + movdqa [lane_data + _extra_block + 0*16], xmm0 + movdqa [lane_data + _extra_block + 1*16], xmm1 + movdqa [lane_data + _extra_block + 2*16], xmm2 + movdqa [lane_data + _extra_block + 3*16], xmm3 +end_fast_copy: + + mov size_offset, extra_blocks + shl size_offset, 6 + sub size_offset, last_len + add size_offset, 64-8 + mov [lane_data + _size_offset], DWORD(size_offset) + mov start_offset, 64 + sub start_offset, last_len + mov [lane_data + _start_offset], DWORD(start_offset) + + lea tmp, [8*64 + 8*len] + bswap tmp + mov [lane_data + _extra_block + size_offset], tmp + + mov tmp, [job + _auth_key_xor_ipad] + movdqu xmm0, [tmp] + mov DWORD(tmp), [tmp + 4*SHA1_DIGEST_WORD_SIZE] +%if SHA1NI_DIGEST_ROW_SIZE != 20 +%error "Below code has been optimized for SHA1NI_DIGEST_ROW_SIZE = 20!" +%endif + lea p4, [lane + lane*4] + movdqu [state + _args_digest + p4*4 + 0*SHA1_DIGEST_WORD_SIZE], xmm0 + mov [state + _args_digest + p4*4 + 4*SHA1_DIGEST_WORD_SIZE], DWORD(tmp) + test len, ~63 + jnz ge64_bytes + +lt64_bytes: + mov [state + _lens + 2*lane], WORD(extra_blocks) + lea tmp, [lane_data + _extra_block + start_offset] + mov [state + _args_data_ptr + PTR_SZ*lane], tmp + mov dword [lane_data + _extra_blocks], 0 + +ge64_bytes: + cmp unused_lanes, 0xff + jne return_null + jmp start_loop + + align 16 +start_loop: + ; Find min length - only two lanes available + xor len2, len2 + mov p3, 0x10000 + mov WORD(len2), word [state + _lens + 0*2] ; [0:15] - lane 0 length, [16:31] - lane index (0) + mov WORD(p3), word [state + _lens + 1*2] ; [0:15] - lane 1 length, [16:31] - lane index (1) + cmp WORD(len2), WORD(p3) + cmovg DWORD(len2), DWORD(p3) ; move if lane 0 length is greater than lane 1 length + + mov idx, len2 ; retrieve index & length from [16:31] and [0:15] bit fields + shr DWORD(idx), 16 + and DWORD(len2), 0xffff + je len_is_0 + + sub word [state + _lens + 0*2], WORD(len2) + sub word [state + _lens + 1*2], WORD(len2) + + ; "state" and "args" are the same address, arg1 + ; len is arg2 + call sha1_ni + ; state is intact + +len_is_0: + ; process completed job "idx" + imul lane_data, idx, _HMAC_SHA1_LANE_DATA_size + lea lane_data, [state + _ldata + lane_data] + mov DWORD(extra_blocks), [lane_data + _extra_blocks] + cmp extra_blocks, 0 + jne proc_extra_blocks + cmp dword [lane_data + _outer_done], 0 + jne end_loop + +proc_outer: + mov dword [lane_data + _outer_done], 1 + mov DWORD(size_offset), [lane_data + _size_offset] + mov qword [lane_data + _extra_block + size_offset], 0 + mov word [state + _lens + 2*idx], 1 + lea tmp, [lane_data + _outer_block] + mov job, [lane_data + _job_in_lane] + mov [state + _args_data_ptr + PTR_SZ*idx], tmp + +%if SHA1NI_DIGEST_ROW_SIZE != 20 +%error "Below code has been optimized for SHA1NI_DIGEST_ROW_SIZE = 20!" +%endif + lea p3, [idx + idx*4] + movdqu xmm0, [state + _args_digest + p3*4 + 0*SHA1_DIGEST_WORD_SIZE] + pshufb xmm0, [rel byteswap] + mov DWORD(tmp), [state + _args_digest + p3*4 + 4*SHA1_DIGEST_WORD_SIZE] + bswap DWORD(tmp) + movdqa [lane_data + _outer_block], xmm0 + mov [lane_data + _outer_block + 4*SHA1_DIGEST_WORD_SIZE], DWORD(tmp) + + mov tmp, [job + _auth_key_xor_opad] + movdqu xmm0, [tmp] + mov DWORD(tmp), [tmp + 4*SHA1_DIGEST_WORD_SIZE] + movdqu [state + _args_digest + p3*4 + 0*SHA1_DIGEST_WORD_SIZE], xmm0 + mov [state + _args_digest + p3*4 + 4*SHA1_DIGEST_WORD_SIZE], DWORD(tmp) + jmp start_loop + + align 16 +proc_extra_blocks: + mov DWORD(start_offset), [lane_data + _start_offset] + mov [state + _lens + 2*idx], WORD(extra_blocks) + lea tmp, [lane_data + _extra_block + start_offset] + mov [state + _args_data_ptr + PTR_SZ*idx], tmp + mov dword [lane_data + _extra_blocks], 0 + jmp start_loop + + align 16 +copy_lt64: + ;; less than one message block of data + ;; beginning of source block + ;; destination extrablock but backwards by len from where 0x80 pre-populated + lea p2, [lane_data + _extra_block + 64] + sub p2, len + memcpy_sse_64_1 p2, p, len, tmp4, tmp2, xmm0, xmm1, xmm2, xmm3 + mov unused_lanes, [state + _unused_lanes] + jmp end_fast_copy + +return_null: + xor job_rax, job_rax + jmp return + + align 16 +end_loop: + mov job_rax, [lane_data + _job_in_lane] + mov unused_lanes, [state + _unused_lanes] + mov qword [lane_data + _job_in_lane], 0 + or dword [job_rax + _status], STS_COMPLETED_HMAC + shl unused_lanes, 8 + or unused_lanes, idx + mov [state + _unused_lanes], unused_lanes + + mov p, [job_rax + _auth_tag_output] + + ; copy 12 bytes +%if SHA1NI_DIGEST_ROW_SIZE != 20 +%error "Below code has been optimized for SHA1NI_DIGEST_ROW_SIZE = 20!" +%endif + lea idx, [idx + 4*idx] + mov DWORD(tmp), [state + _args_digest + idx*4 + 0*SHA1_DIGEST_WORD_SIZE] + mov DWORD(tmp2), [state + _args_digest + idx*4 + 1*SHA1_DIGEST_WORD_SIZE] + mov DWORD(tmp3), [state + _args_digest + idx*4 + 2*SHA1_DIGEST_WORD_SIZE] + bswap DWORD(tmp) + bswap DWORD(tmp2) + bswap DWORD(tmp3) + mov [p + 0*SHA1_DIGEST_WORD_SIZE], DWORD(tmp) + mov [p + 1*SHA1_DIGEST_WORD_SIZE], DWORD(tmp2) + mov [p + 2*SHA1_DIGEST_WORD_SIZE], DWORD(tmp3) + + cmp qword [job_rax + _auth_tag_output_len_in_bytes], 12 + je clear_ret + + ;; copy remaining 8 bytes to return 20 byte digest + mov DWORD(tmp), [state + _args_digest + idx*4 + 3*SHA1_DIGEST_WORD_SIZE] + mov DWORD(tmp2), [state + _args_digest + idx*4 + 4*SHA1_DIGEST_WORD_SIZE] + bswap DWORD(tmp) + bswap DWORD(tmp2) + mov [p + 3*4], DWORD(tmp) + mov [p + 4*4], DWORD(tmp2) + +clear_ret: + +%ifdef SAFE_DATA + pxor xmm0, xmm0 + ;; Clear digest (20B), outer_block (20B) and extra_block (64B) + ;; idx = 0 or 5 (depending on lane) + movdqu [state + _args_digest + idx*4], xmm0 + mov dword [state + _args_digest + idx*4 + 16], 0 + + shr idx, 2 ;; idx == 5 ? 1 : 0 + imul lane_data, idx, _HMAC_SHA1_LANE_DATA_size + lea lane_data, [state + _ldata + lane_data] + ;; Clear first 64 bytes of extra_block +%assign offset 0 +%rep 4 + movdqa [lane_data + _extra_block + offset], xmm0 +%assign offset (offset + 16) +%endrep + + ;; Clear 20 bytes of outer_block + movdqa [lane_data + _outer_block], xmm0 + mov dword [lane_data + _outer_block + 16], 0 +%endif + +return: + mov rbx, [rsp + _gpr_save + 8*0] + mov rbp, [rsp + _gpr_save + 8*1] +%ifndef LINUX + mov rsi, [rsp + _gpr_save + 8*2] + mov rdi, [rsp + _gpr_save + 8*3] +%endif + mov rsp, [rsp + _rsp_save] ; original SP + + ret + +%ifdef LINUX +section .note.GNU-stack noalloc noexec nowrite progbits +%endif |