ceph/src/seastar/dpdk/lib/librte_reorder/rte_reorder.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include <inttypes.h>
  35 #include <string.h>
  36
  37 #include <rte_log.h>
  38 #include <rte_mbuf.h>
  39 #include <rte_memzone.h>
  40 #include <rte_eal_memconfig.h>
  41 #include <rte_errno.h>
  42 #include <rte_malloc.h>
  43
  44 #include "rte_reorder.h"
  45
  46 TAILQ_HEAD(rte_reorder_list, rte_tailq_entry);
  47
  48 static struct rte_tailq_elem rte_reorder_tailq = {
  49         .name = "RTE_REORDER",
  50 };
  51 EAL_REGISTER_TAILQ(rte_reorder_tailq)
  52
  53 #define NO_FLAGS 0
  54 #define RTE_REORDER_PREFIX "RO_"
  55 #define RTE_REORDER_NAMESIZE 32
  56
  57 /* Macros for printing using RTE_LOG */
  58 #define RTE_LOGTYPE_REORDER     RTE_LOGTYPE_USER1
  59
  60 /* A generic circular buffer */
  61 struct cir_buffer {
  62         unsigned int size;   /**< Number of entries that can be stored */
  63         unsigned int mask;   /**< [buffer_size - 1]: used for wrap-around */
  64         unsigned int head;   /**< insertion point in buffer */
  65         unsigned int tail;   /**< extraction point in buffer */
  66         struct rte_mbuf **entries;
  67 } __rte_cache_aligned;
  68
  69 /* The reorder buffer data structure itself */
  70 struct rte_reorder_buffer {
  71         char name[RTE_REORDER_NAMESIZE];
  72         uint32_t min_seqn;  /**< Lowest seq. number that can be in the buffer */
  73         unsigned int memsize; /**< memory area size of reorder buffer */
  74         struct cir_buffer ready_buf; /**< temp buffer for dequeued entries */
  75         struct cir_buffer order_buf; /**< buffer used to reorder entries */
  76         int is_initialized;
  77 } __rte_cache_aligned;
  78
  79 static void
  80 rte_reorder_free_mbufs(struct rte_reorder_buffer *b);
  81
  82 struct rte_reorder_buffer *
  83 rte_reorder_init(struct rte_reorder_buffer *b, unsigned int bufsize,
  84                 const char *name, unsigned int size)
  85 {
  86         const unsigned int min_bufsize = sizeof(*b) +
  87                                         (2 * size * sizeof(struct rte_mbuf *));
  88
  89         if (b == NULL) {
  90                 RTE_LOG(ERR, REORDER, "Invalid reorder buffer parameter:"
  91                                         " NULL\n");
  92                 rte_errno = EINVAL;
  93                 return NULL;
  94         }
  95         if (!rte_is_power_of_2(size)) {
  96                 RTE_LOG(ERR, REORDER, "Invalid reorder buffer size"
  97                                 " - Not a power of 2\n");
  98                 rte_errno = EINVAL;
  99                 return NULL;
 100         }
 101         if (name == NULL) {
 102                 RTE_LOG(ERR, REORDER, "Invalid reorder buffer name ptr:"
 103                                         " NULL\n");
 104                 rte_errno = EINVAL;
 105                 return NULL;
 106         }
 107         if (bufsize < min_bufsize) {
 108                 RTE_LOG(ERR, REORDER, "Invalid reorder buffer memory size: %u, "
 109                         "minimum required: %u\n", bufsize, min_bufsize);
 110                 rte_errno = EINVAL;
 111                 return NULL;
 112         }
 113
 114         memset(b, 0, bufsize);
 115         snprintf(b->name, sizeof(b->name), "%s", name);
 116         b->memsize = bufsize;
 117         b->order_buf.size = b->ready_buf.size = size;
 118         b->order_buf.mask = b->ready_buf.mask = size - 1;
 119         b->ready_buf.entries = (void *)&b[1];
 120         b->order_buf.entries = RTE_PTR_ADD(&b[1],
 121                         size * sizeof(b->ready_buf.entries[0]));
 122
 123         return b;
 124 }
 125
 126 struct rte_reorder_buffer*
 127 rte_reorder_create(const char *name, unsigned socket_id, unsigned int size)
 128 {
 129         struct rte_reorder_buffer *b = NULL;
 130         struct rte_tailq_entry *te;
 131         struct rte_reorder_list *reorder_list;
 132         const unsigned int bufsize = sizeof(struct rte_reorder_buffer) +
 133                                         (2 * size * sizeof(struct rte_mbuf *));
 134
 135         reorder_list = RTE_TAILQ_CAST(rte_reorder_tailq.head, rte_reorder_list);
 136
 137         /* Check user arguments. */
 138         if (!rte_is_power_of_2(size)) {
 139                 RTE_LOG(ERR, REORDER, "Invalid reorder buffer size"
 140                                 " - Not a power of 2\n");
 141                 rte_errno = EINVAL;
 142                 return NULL;
 143         }
 144         if (name == NULL) {
 145                 RTE_LOG(ERR, REORDER, "Invalid reorder buffer name ptr:"
 146                                         " NULL\n");
 147                 rte_errno = EINVAL;
 148                 return NULL;
 149         }
 150
 151         rte_rwlock_write_lock(RTE_EAL_TAILQ_RWLOCK);
 152
 153         /* guarantee there's no existing */
 154         TAILQ_FOREACH(te, reorder_list, next) {
 155                 b = (struct rte_reorder_buffer *) te->data;
 156                 if (strncmp(name, b->name, RTE_REORDER_NAMESIZE) == 0)
 157                         break;
 158         }
 159         if (te != NULL)
 160                 goto exit;
 161
 162         /* allocate tailq entry */
 163         te = rte_zmalloc("REORDER_TAILQ_ENTRY", sizeof(*te), 0);
 164         if (te == NULL) {
 165                 RTE_LOG(ERR, REORDER, "Failed to allocate tailq entry\n");
 166                 rte_errno = ENOMEM;
 167                 b = NULL;
 168                 goto exit;
 169         }
 170
 171         /* Allocate memory to store the reorder buffer structure. */
 172         b = rte_zmalloc_socket("REORDER_BUFFER", bufsize, 0, socket_id);
 173         if (b == NULL) {
 174                 RTE_LOG(ERR, REORDER, "Memzone allocation failed\n");
 175                 rte_errno = ENOMEM;
 176                 rte_free(te);
 177         } else {
 178                 rte_reorder_init(b, bufsize, name, size);
 179                 te->data = (void *)b;
 180                 TAILQ_INSERT_TAIL(reorder_list, te, next);
 181         }
 182
 183 exit:
 184         rte_rwlock_write_unlock(RTE_EAL_TAILQ_RWLOCK);
 185         return b;
 186 }
 187
 188 void
 189 rte_reorder_reset(struct rte_reorder_buffer *b)
 190 {
 191         char name[RTE_REORDER_NAMESIZE];
 192
 193         rte_reorder_free_mbufs(b);
 194         snprintf(name, sizeof(name), "%s", b->name);
 195         /* No error checking as current values should be valid */
 196         rte_reorder_init(b, b->memsize, name, b->order_buf.size);
 197 }
 198
 199 static void
 200 rte_reorder_free_mbufs(struct rte_reorder_buffer *b)
 201 {
 202         unsigned i;
 203
 204         /* Free up the mbufs of order buffer & ready buffer */
 205         for (i = 0; i < b->order_buf.size; i++) {
 206                 if (b->order_buf.entries[i])
 207                         rte_pktmbuf_free(b->order_buf.entries[i]);
 208                 if (b->ready_buf.entries[i])
 209                         rte_pktmbuf_free(b->ready_buf.entries[i]);
 210         }
 211 }
 212
 213 void
 214 rte_reorder_free(struct rte_reorder_buffer *b)
 215 {
 216         struct rte_reorder_list *reorder_list;
 217         struct rte_tailq_entry *te;
 218
 219         /* Check user arguments. */
 220         if (b == NULL)
 221                 return;
 222
 223         reorder_list = RTE_TAILQ_CAST(rte_reorder_tailq.head, rte_reorder_list);
 224
 225         rte_rwlock_write_lock(RTE_EAL_TAILQ_RWLOCK);
 226
 227         /* find our tailq entry */
 228         TAILQ_FOREACH(te, reorder_list, next) {
 229                 if (te->data == (void *) b)
 230                         break;
 231         }
 232         if (te == NULL) {
 233                 rte_rwlock_write_unlock(RTE_EAL_TAILQ_RWLOCK);
 234                 return;
 235         }
 236
 237         TAILQ_REMOVE(reorder_list, te, next);
 238
 239         rte_rwlock_write_unlock(RTE_EAL_TAILQ_RWLOCK);
 240
 241         rte_reorder_free_mbufs(b);
 242
 243         rte_free(b);
 244         rte_free(te);
 245 }
 246
 247 struct rte_reorder_buffer *
 248 rte_reorder_find_existing(const char *name)
 249 {
 250         struct rte_reorder_buffer *b = NULL;
 251         struct rte_tailq_entry *te;
 252         struct rte_reorder_list *reorder_list;
 253
 254         reorder_list = RTE_TAILQ_CAST(rte_reorder_tailq.head, rte_reorder_list);
 255
 256         rte_rwlock_read_lock(RTE_EAL_TAILQ_RWLOCK);
 257         TAILQ_FOREACH(te, reorder_list, next) {
 258                 b = (struct rte_reorder_buffer *) te->data;
 259                 if (strncmp(name, b->name, RTE_REORDER_NAMESIZE) == 0)
 260                         break;
 261         }
 262         rte_rwlock_read_unlock(RTE_EAL_TAILQ_RWLOCK);
 263
 264         if (te == NULL) {
 265                 rte_errno = ENOENT;
 266                 return NULL;
 267         }
 268
 269         return b;
 270 }
 271
 272 static unsigned
 273 rte_reorder_fill_overflow(struct rte_reorder_buffer *b, unsigned n)
 274 {
 275         /*
 276          * 1. Move all ready entries that fit to the ready_buf
 277          * 2. check if we meet the minimum needed (n).
 278          * 3. If not, then skip any gaps and keep moving.
 279          * 4. If at any point the ready buffer is full, stop
 280          * 5. Return the number of positions the order_buf head has moved
 281          */
 282
 283         struct cir_buffer *order_buf = &b->order_buf,
 284                         *ready_buf = &b->ready_buf;
 285
 286         unsigned int order_head_adv = 0;
 287
 288         /*
 289          * move at least n packets to ready buffer, assuming ready buffer
 290          * has room for those packets.
 291          */
 292         while (order_head_adv < n &&
 293                         ((ready_buf->head + 1) & ready_buf->mask) != ready_buf->tail) {
 294
 295                 /* if we are blocked waiting on a packet, skip it */
 296                 if (order_buf->entries[order_buf->head] == NULL) {
 297                         order_buf->head = (order_buf->head + 1) & order_buf->mask;
 298                         order_head_adv++;
 299                 }
 300
 301                 /* Move all ready entries that fit to the ready_buf */
 302                 while (order_buf->entries[order_buf->head] != NULL) {
 303                         ready_buf->entries[ready_buf->head] =
 304                                         order_buf->entries[order_buf->head];
 305
 306                         order_buf->entries[order_buf->head] = NULL;
 307                         order_head_adv++;
 308
 309                         order_buf->head = (order_buf->head + 1) & order_buf->mask;
 310
 311                         if (((ready_buf->head + 1) & ready_buf->mask) == ready_buf->tail)
 312                                 break;
 313
 314                         ready_buf->head = (ready_buf->head + 1) & ready_buf->mask;
 315                 }
 316         }
 317
 318         b->min_seqn += order_head_adv;
 319         /* Return the number of positions the order_buf head has moved */
 320         return order_head_adv;
 321 }
 322
 323 int
 324 rte_reorder_insert(struct rte_reorder_buffer *b, struct rte_mbuf *mbuf)
 325 {
 326         uint32_t offset, position;
 327         struct cir_buffer *order_buf = &b->order_buf;
 328
 329         if (!b->is_initialized) {
 330                 b->min_seqn = mbuf->seqn;
 331                 b->is_initialized = 1;
 332         }
 333
 334         /*
 335          * calculate the offset from the head pointer we need to go.
 336          * The subtraction takes care of the sequence number wrapping.
 337          * For example (using 16-bit for brevity):
 338          *      min_seqn  = 0xFFFD
 339          *      mbuf_seqn = 0x0010
 340          *      offset    = 0x0010 - 0xFFFD = 0x13
 341          */
 342         offset = mbuf->seqn - b->min_seqn;
 343
 344         /*
 345          * action to take depends on offset.
 346          * offset < buffer->size: the mbuf fits within the current window of
 347          *    sequence numbers we can reorder. EXPECTED CASE.
 348          * offset > buffer->size: the mbuf is outside the current window. There
 349          *    are a number of cases to consider:
 350          *    1. The packet sequence is just outside the window, then we need
 351          *       to see about shifting the head pointer and taking any ready
 352          *       to return packets out of the ring. If there was a delayed
 353          *       or dropped packet preventing drains from shifting the window
 354          *       this case will skip over the dropped packet instead, and any
 355          *       packets dequeued here will be returned on the next drain call.
 356          *    2. The packet sequence number is vastly outside our window, taken
 357          *       here as having offset greater than twice the buffer size. In
 358          *       this case, the packet is probably an old or late packet that
 359          *       was previously skipped, so just enqueue the packet for
 360          *       immediate return on the next drain call, or else return error.
 361          */
 362         if (offset < b->order_buf.size) {
 363                 position = (order_buf->head + offset) & order_buf->mask;
 364                 order_buf->entries[position] = mbuf;
 365         } else if (offset < 2 * b->order_buf.size) {
 366                 if (rte_reorder_fill_overflow(b, offset + 1 - order_buf->size)
 367                                 < (offset + 1 - order_buf->size)) {
 368                         /* Put in handling for enqueue straight to output */
 369                         rte_errno = ENOSPC;
 370                         return -1;
 371                 }
 372                 offset = mbuf->seqn - b->min_seqn;
 373                 position = (order_buf->head + offset) & order_buf->mask;
 374                 order_buf->entries[position] = mbuf;
 375         } else {
 376                 /* Put in handling for enqueue straight to output */
 377                 rte_errno = ERANGE;
 378                 return -1;
 379         }
 380         return 0;
 381 }
 382
 383 unsigned int
 384 rte_reorder_drain(struct rte_reorder_buffer *b, struct rte_mbuf **mbufs,
 385                 unsigned max_mbufs)
 386 {
 387         unsigned int drain_cnt = 0;
 388
 389         struct cir_buffer *order_buf = &b->order_buf,
 390                         *ready_buf = &b->ready_buf;
 391
 392         /* Try to fetch requested number of mbufs from ready buffer */
 393         while ((drain_cnt < max_mbufs) && (ready_buf->tail != ready_buf->head)) {
 394                 mbufs[drain_cnt++] = ready_buf->entries[ready_buf->tail];
 395                 ready_buf->tail = (ready_buf->tail + 1) & ready_buf->mask;
 396         }
 397
 398         /*
 399          * If requested number of buffers not fetched from ready buffer, fetch
 400          * remaining buffers from order buffer
 401          */
 402         while ((drain_cnt < max_mbufs) &&
 403                         (order_buf->entries[order_buf->head] != NULL)) {
 404                 mbufs[drain_cnt++] = order_buf->entries[order_buf->head];
 405                 order_buf->entries[order_buf->head] = NULL;
 406                 b->min_seqn++;
 407                 order_buf->head = (order_buf->head + 1) & order_buf->mask;
 408         }
 409
 410         return drain_cnt;
 411 }