ceph/src/crypto/isa-l/isa-l_crypto/sm3_mb/sm3_ctx_avx2.c

   1 /**********************************************************************
   2   Copyright(c) 2011-2020 Intel Corporation All rights reserved.
   3
   4   Redistribution and use in source and binary forms, with or without
   5   modification, are permitted provided that the following conditions
   6   are met:
   7     * Redistributions of source code must retain the above copyright
   8       notice, this list of conditions and the following disclaimer.
   9     * Redistributions in binary form must reproduce the above copyright
  10       notice, this list of conditions and the following disclaimer in
  11       the documentation and/or other materials provided with the
  12       distribution.
  13     * Neither the name of Intel Corporation nor the names of its
  14       contributors may be used to endorse or promote products derived
  15       from this software without specific prior written permission.
  16
  17   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  18   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  19   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  20   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  21   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  22   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  23   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  24   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  25   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  26   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  27   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  28 **********************************************************************/
  29
  30 #if defined(__clang__)
  31 # pragma clang attribute push (__attribute__((target("avx2"))), apply_to=function)
  32 #elif defined(__ICC)
  33 # pragma intel optimization_parameter target_arch=AVX2
  34 #elif defined(__ICL)
  35 # pragma [intel] optimization_parameter target_arch=AVX2
  36 #elif (__GNUC__ >= 5)
  37 # pragma GCC target("avx2")
  38 #endif
  39
  40 #include "sm3_mb.h"
  41 #include "memcpy_inline.h"
  42 #include "endian_helper.h"
  43
  44 #ifdef _MSC_VER
  45 # include <intrin.h>
  46 # define inline __inline
  47 #endif
  48
  49 static inline void hash_init_digest(SM3_WORD_T * digest);
  50 static inline uint32_t hash_pad(uint8_t padblock[SM3_BLOCK_SIZE * 2], uint64_t total_len);
  51 static SM3_HASH_CTX *sm3_ctx_mgr_resubmit(SM3_HASH_CTX_MGR * mgr, SM3_HASH_CTX * ctx);
  52
  53 void sm3_mb_mgr_init_avx2(SM3_MB_JOB_MGR * state);
  54 SM3_JOB *sm3_mb_mgr_submit_avx2(SM3_MB_JOB_MGR * state, SM3_JOB * job);
  55 SM3_JOB *sm3_mb_mgr_flush_avx2(SM3_MB_JOB_MGR * state);
  56
  57 void sm3_mb_mgr_init_avx2(SM3_MB_JOB_MGR * state)
  58 {
  59         unsigned int j;
  60         state->unused_lanes = 0xF76543210;
  61         state->num_lanes_inuse = 0;
  62         for (j = 0; j < SM3_X8_LANES; j++) {
  63                 state->lens[j] = 0;
  64                 state->ldata[j].job_in_lane = 0;
  65         }
  66 }
  67
  68 void sm3_ctx_mgr_init_avx2(SM3_HASH_CTX_MGR * mgr)
  69 {
  70         sm3_mb_mgr_init_avx2(&mgr->mgr);
  71 }
  72
  73 SM3_HASH_CTX *sm3_ctx_mgr_submit_avx2(SM3_HASH_CTX_MGR * mgr, SM3_HASH_CTX * ctx,
  74                                       const void *buffer, uint32_t len, HASH_CTX_FLAG flags)
  75 {
  76         if (flags & (~HASH_ENTIRE)) {
  77                 // User should not pass anything other than FIRST, UPDATE, or LAST
  78                 ctx->error = HASH_CTX_ERROR_INVALID_FLAGS;
  79                 return ctx;
  80         }
  81
  82         if (ctx->status & HASH_CTX_STS_PROCESSING) {
  83                 // Cannot submit to a currently processing job.
  84                 ctx->error = HASH_CTX_ERROR_ALREADY_PROCESSING;
  85                 return ctx;
  86         }
  87
  88         if ((ctx->status & HASH_CTX_STS_COMPLETE) && !(flags & HASH_FIRST)) {
  89                 // Cannot update a finished job.
  90                 ctx->error = HASH_CTX_ERROR_ALREADY_COMPLETED;
  91                 return ctx;
  92         }
  93
  94         if (flags & HASH_FIRST) {
  95                 // Init digest
  96                 hash_init_digest(ctx->job.result_digest);
  97
  98                 // Reset byte counter
  99                 ctx->total_length = 0;
 100
 101                 // Clear extra blocks
 102                 ctx->partial_block_buffer_length = 0;
 103         }
 104         // If we made it here, there were no errors during this call to submit
 105         ctx->error = HASH_CTX_ERROR_NONE;
 106
 107         // Store buffer ptr info from user
 108         ctx->incoming_buffer = buffer;
 109         ctx->incoming_buffer_length = len;
 110
 111         // Store the user's request flags and mark this ctx as currently being processed.
 112         ctx->status = (flags & HASH_LAST) ?
 113             (HASH_CTX_STS) (HASH_CTX_STS_PROCESSING | HASH_CTX_STS_LAST) :
 114             HASH_CTX_STS_PROCESSING;
 115
 116         // Advance byte counter
 117         ctx->total_length += len;
 118
 119         // If there is anything currently buffered in the extra blocks, append to it until it contains a whole block.
 120         // Or if the user's buffer contains less than a whole block, append as much as possible to the extra block.
 121         if ((ctx->partial_block_buffer_length) | (len < SM3_BLOCK_SIZE)) {
 122                 // Compute how many bytes to copy from user buffer into extra block
 123                 uint32_t copy_len = SM3_BLOCK_SIZE - ctx->partial_block_buffer_length;
 124                 if (len < copy_len)
 125                         copy_len = len;
 126
 127                 if (copy_len) {
 128                         // Copy and update relevant pointers and counters
 129                         memcpy_varlen(&ctx->partial_block_buffer
 130                                       [ctx->partial_block_buffer_length], buffer, copy_len);
 131
 132                         ctx->partial_block_buffer_length += copy_len;
 133                         ctx->incoming_buffer = (const void *)((const char *)buffer + copy_len);
 134                         ctx->incoming_buffer_length = len - copy_len;
 135                 }
 136                 // The extra block should never contain more than 1 block here
 137                 assert(ctx->partial_block_buffer_length <= SM3_BLOCK_SIZE);
 138
 139                 // If the extra block buffer contains exactly 1 block, it can be hashed.
 140                 if (ctx->partial_block_buffer_length >= SM3_BLOCK_SIZE) {
 141                         ctx->partial_block_buffer_length = 0;
 142
 143                         ctx->job.buffer = ctx->partial_block_buffer;
 144                         ctx->job.len = 1;
 145                         ctx = (SM3_HASH_CTX *) sm3_mb_mgr_submit_avx2(&mgr->mgr, &ctx->job);
 146                 }
 147         }
 148
 149         return sm3_ctx_mgr_resubmit(mgr, ctx);
 150 }
 151
 152 SM3_HASH_CTX *sm3_ctx_mgr_flush_avx2(SM3_HASH_CTX_MGR * mgr)
 153 {
 154         SM3_HASH_CTX *ctx;
 155
 156         while (1) {
 157                 ctx = (SM3_HASH_CTX *) sm3_mb_mgr_flush_avx2(&mgr->mgr);
 158
 159                 // If flush returned 0, there are no more jobs in flight.
 160                 if (!ctx)
 161                         return NULL;
 162
 163                 // If flush returned a job, verify that it is safe to return to the user.
 164                 // If it is not ready, resubmit the job to finish processing.
 165                 ctx = sm3_ctx_mgr_resubmit(mgr, ctx);
 166
 167                 // If sm3_ctx_mgr_resubmit returned a job, it is ready to be returned.
 168                 if (ctx)
 169                         return ctx;
 170
 171                 // Otherwise, all jobs currently being managed by the SM3_HASH_CTX_MGR still need processing. Loop.
 172         }
 173 }
 174
 175 static SM3_HASH_CTX *sm3_ctx_mgr_resubmit(SM3_HASH_CTX_MGR * mgr, SM3_HASH_CTX * ctx)
 176 {
 177         while (ctx) {
 178                 if (ctx->status & HASH_CTX_STS_COMPLETE) {
 179                         unsigned int j;
 180                         ctx->status = HASH_CTX_STS_COMPLETE;    // Clear PROCESSING bit
 181                         for (j = 0; j < SM3_DIGEST_NWORDS; j++) {
 182                                 ctx->job.result_digest[j] =
 183                                     byteswap32(ctx->job.result_digest[j]);
 184                         }
 185                         return ctx;
 186                 }
 187                 // If the extra blocks are empty, begin hashing what remains in the user's buffer.
 188                 if (ctx->partial_block_buffer_length == 0 && ctx->incoming_buffer_length) {
 189                         const void *buffer = ctx->incoming_buffer;
 190                         uint32_t len = ctx->incoming_buffer_length;
 191
 192                         // Only entire blocks can be hashed. Copy remainder to extra blocks buffer.
 193                         uint32_t copy_len = len & (SM3_BLOCK_SIZE - 1);
 194
 195                         if (copy_len) {
 196                                 len -= copy_len;
 197                                 memcpy_varlen(ctx->partial_block_buffer,
 198                                               ((const char *)buffer + len), copy_len);
 199                                 ctx->partial_block_buffer_length = copy_len;
 200                         }
 201
 202                         ctx->incoming_buffer_length = 0;
 203
 204                         // len should be a multiple of the block size now
 205                         assert((len % SM3_BLOCK_SIZE) == 0);
 206
 207                         // Set len to the number of blocks to be hashed in the user's buffer
 208                         len >>= SM3_LOG2_BLOCK_SIZE;
 209
 210                         if (len) {
 211                                 ctx->job.buffer = (uint8_t *) buffer;
 212                                 ctx->job.len = len;
 213                                 ctx = (SM3_HASH_CTX *) sm3_mb_mgr_submit_avx2(&mgr->mgr,
 214                                                                               &ctx->job);
 215                                 continue;
 216                         }
 217                 }
 218                 // If the extra blocks are not empty, then we are either on the last block(s)
 219                 // or we need more user input before continuing.
 220                 if (ctx->status & HASH_CTX_STS_LAST) {
 221                         uint8_t *buf = ctx->partial_block_buffer;
 222                         uint32_t n_extra_blocks = hash_pad(buf, ctx->total_length);
 223
 224                         ctx->status =
 225                             (HASH_CTX_STS) (HASH_CTX_STS_PROCESSING | HASH_CTX_STS_COMPLETE);
 226                         ctx->job.buffer = buf;
 227                         ctx->job.len = (uint32_t) n_extra_blocks;
 228                         ctx = (SM3_HASH_CTX *) sm3_mb_mgr_submit_avx2(&mgr->mgr, &ctx->job);
 229                         continue;
 230                 }
 231
 232                 if (ctx)
 233                         ctx->status = HASH_CTX_STS_IDLE;
 234                 return ctx;
 235         }
 236
 237         return NULL;
 238 }
 239
 240 static inline void hash_init_digest(SM3_WORD_T * digest)
 241 {
 242         static const SM3_WORD_T hash_initial_digest[SM3_DIGEST_NWORDS] =
 243             { SM3_INITIAL_DIGEST };
 244         memcpy_fixedlen(digest, hash_initial_digest, sizeof(hash_initial_digest));
 245 }
 246
 247 static inline uint32_t hash_pad(uint8_t padblock[SM3_BLOCK_SIZE * 2], uint64_t total_len)
 248 {
 249         uint32_t i = (uint32_t) (total_len & (SM3_BLOCK_SIZE - 1));
 250
 251         memclr_fixedlen(&padblock[i], SM3_BLOCK_SIZE);
 252         padblock[i] = 0x80;
 253
 254         // Move i to the end of either 1st or 2nd extra block depending on length
 255         i += ((SM3_BLOCK_SIZE - 1) & (0 - (total_len + SM3_PADLENGTHFIELD_SIZE + 1))) +
 256             1 + SM3_PADLENGTHFIELD_SIZE;
 257
 258 #if SM3_PADLENGTHFIELD_SIZE == 16
 259         *((uint64_t *) & padblock[i - 16]) = 0;
 260 #endif
 261
 262         *((uint64_t *) & padblock[i - 8]) = to_be64((uint64_t) total_len << 3);
 263
 264         return i >> SM3_LOG2_BLOCK_SIZE;        // Number of extra blocks to hash
 265 }
 266
 267 struct slver {
 268         uint16_t snum;
 269         uint8_t ver;
 270         uint8_t core;
 271 };
 272
 273 struct slver sm3_ctx_mgr_init_avx2_slver_0000;
 274 struct slver sm3_ctx_mgr_init_avx2_slver = { 0x2309, 0x00, 0x00 };
 275
 276 struct slver sm3_ctx_mgr_submit_avx2_slver_0000;
 277 struct slver sm3_ctx_mgr_submit_avx2_slver = { 0x230a, 0x00, 0x00 };
 278
 279 struct slver sm3_ctx_mgr_flush_avx2_slver_0000;
 280 struct slver sm3_ctx_mgr_flush_avx2_slver = { 0x230b, 0x00, 0x00 };
 281
 282 #if defined(__clang__)
 283 # pragma clang attribute pop
 284 #endif