1 /* SPDX-License-Identifier: GPL-2.0 */ 2 /* XDP user-space ring structure 3 * Copyright(c) 2018 Intel Corporation. 4 */ 5 6 #ifndef _LINUX_XSK_QUEUE_H 7 #define _LINUX_XSK_QUEUE_H 8 9 #include <linux/types.h> 10 #include <linux/if_xdp.h> 11 #include <net/xdp_sock.h> 12 #include <net/xsk_buff_pool.h> 13 14 #include "xsk.h" 15 16 struct xdp_ring { 17 u32 producer ____cacheline_aligned_in_smp; 18 /* Hinder the adjacent cache prefetcher to prefetch the consumer 19 * pointer if the producer pointer is touched and vice versa. 20 */ 21 u32 pad1 ____cacheline_aligned_in_smp; 22 u32 consumer ____cacheline_aligned_in_smp; 23 u32 pad2 ____cacheline_aligned_in_smp; 24 u32 flags; 25 u32 pad3 ____cacheline_aligned_in_smp; 26 }; 27 28 /* Used for the RX and TX queues for packets */ 29 struct xdp_rxtx_ring { 30 struct xdp_ring ptrs; 31 struct xdp_desc desc[] ____cacheline_aligned_in_smp; 32 }; 33 34 /* Used for the fill and completion queues for buffers */ 35 struct xdp_umem_ring { 36 struct xdp_ring ptrs; 37 u64 desc[] ____cacheline_aligned_in_smp; 38 }; 39 40 struct xsk_queue { 41 u32 ring_mask; 42 u32 nentries; 43 u32 cached_prod; 44 u32 cached_cons; 45 struct xdp_ring *ring; 46 u64 invalid_descs; 47 u64 queue_empty_descs; 48 size_t ring_vmalloc_size; 49 }; 50 51 /* The structure of the shared state of the rings are a simple 52 * circular buffer, as outlined in 53 * Documentation/core-api/circular-buffers.rst. For the Rx and 54 * completion ring, the kernel is the producer and user space is the 55 * consumer. For the Tx and fill rings, the kernel is the consumer and 56 * user space is the producer. 57 * 58 * producer consumer 59 * 60 * if (LOAD ->consumer) { (A) LOAD.acq ->producer (C) 61 * STORE $data LOAD $data 62 * STORE.rel ->producer (B) STORE.rel ->consumer (D) 63 * } 64 * 65 * (A) pairs with (D), and (B) pairs with (C). 66 * 67 * Starting with (B), it protects the data from being written after 68 * the producer pointer. If this barrier was missing, the consumer 69 * could observe the producer pointer being set and thus load the data 70 * before the producer has written the new data. The consumer would in 71 * this case load the old data. 72 * 73 * (C) protects the consumer from speculatively loading the data before 74 * the producer pointer actually has been read. If we do not have this 75 * barrier, some architectures could load old data as speculative loads 76 * are not discarded as the CPU does not know there is a dependency 77 * between ->producer and data. 78 * 79 * (A) is a control dependency that separates the load of ->consumer 80 * from the stores of $data. In case ->consumer indicates there is no 81 * room in the buffer to store $data we do not. The dependency will 82 * order both of the stores after the loads. So no barrier is needed. 83 * 84 * (D) protects the load of the data to be observed to happen after the 85 * store of the consumer pointer. If we did not have this memory 86 * barrier, the producer could observe the consumer pointer being set 87 * and overwrite the data with a new value before the consumer got the 88 * chance to read the old value. The consumer would thus miss reading 89 * the old entry and very likely read the new entry twice, once right 90 * now and again after circling through the ring. 91 */ 92 93 /* The operations on the rings are the following: 94 * 95 * producer consumer 96 * 97 * RESERVE entries PEEK in the ring for entries 98 * WRITE data into the ring READ data from the ring 99 * SUBMIT entries RELEASE entries 100 * 101 * The producer reserves one or more entries in the ring. It can then 102 * fill in these entries and finally submit them so that they can be 103 * seen and read by the consumer. 104 * 105 * The consumer peeks into the ring to see if the producer has written 106 * any new entries. If so, the consumer can then read these entries 107 * and when it is done reading them release them back to the producer 108 * so that the producer can use these slots to fill in new entries. 109 * 110 * The function names below reflect these operations. 111 */ 112 113 /* Functions that read and validate content from consumer rings. */ 114 115 static inline void __xskq_cons_read_addr_unchecked(struct xsk_queue *q, u32 cached_cons, u64 *addr) 116 { 117 struct xdp_umem_ring *ring = (struct xdp_umem_ring *)q->ring; 118 u32 idx = cached_cons & q->ring_mask; 119 120 *addr = ring->desc[idx]; 121 } 122 123 static inline bool xskq_cons_read_addr_unchecked(struct xsk_queue *q, u64 *addr) 124 { 125 if (q->cached_cons != q->cached_prod) { 126 __xskq_cons_read_addr_unchecked(q, q->cached_cons, addr); 127 return true; 128 } 129 130 return false; 131 } 132 133 static inline bool xp_aligned_validate_desc(struct xsk_buff_pool *pool, 134 struct xdp_desc *desc) 135 { 136 u64 offset = desc->addr & (pool->chunk_size - 1); 137 138 if (offset + desc->len > pool->chunk_size) 139 return false; 140 141 if (desc->addr >= pool->addrs_cnt) 142 return false; 143 144 if (desc->options) 145 return false; 146 return true; 147 } 148 149 static inline bool xp_unaligned_validate_desc(struct xsk_buff_pool *pool, 150 struct xdp_desc *desc) 151 { 152 u64 addr = xp_unaligned_add_offset_to_addr(desc->addr); 153 154 if (desc->len > pool->chunk_size) 155 return false; 156 157 if (addr >= pool->addrs_cnt || addr + desc->len > pool->addrs_cnt || 158 xp_desc_crosses_non_contig_pg(pool, addr, desc->len)) 159 return false; 160 161 if (desc->options) 162 return false; 163 return true; 164 } 165 166 static inline bool xp_validate_desc(struct xsk_buff_pool *pool, 167 struct xdp_desc *desc) 168 { 169 return pool->unaligned ? xp_unaligned_validate_desc(pool, desc) : 170 xp_aligned_validate_desc(pool, desc); 171 } 172 173 static inline bool xskq_cons_is_valid_desc(struct xsk_queue *q, 174 struct xdp_desc *d, 175 struct xsk_buff_pool *pool) 176 { 177 if (!xp_validate_desc(pool, d)) { 178 q->invalid_descs++; 179 return false; 180 } 181 return true; 182 } 183 184 static inline bool xskq_cons_read_desc(struct xsk_queue *q, 185 struct xdp_desc *desc, 186 struct xsk_buff_pool *pool) 187 { 188 while (q->cached_cons != q->cached_prod) { 189 struct xdp_rxtx_ring *ring = (struct xdp_rxtx_ring *)q->ring; 190 u32 idx = q->cached_cons & q->ring_mask; 191 192 *desc = ring->desc[idx]; 193 if (xskq_cons_is_valid_desc(q, desc, pool)) 194 return true; 195 196 q->cached_cons++; 197 } 198 199 return false; 200 } 201 202 static inline void xskq_cons_release_n(struct xsk_queue *q, u32 cnt) 203 { 204 q->cached_cons += cnt; 205 } 206 207 static inline u32 xskq_cons_read_desc_batch(struct xsk_queue *q, struct xsk_buff_pool *pool, 208 u32 max) 209 { 210 u32 cached_cons = q->cached_cons, nb_entries = 0; 211 struct xdp_desc *descs = pool->tx_descs; 212 213 while (cached_cons != q->cached_prod && nb_entries < max) { 214 struct xdp_rxtx_ring *ring = (struct xdp_rxtx_ring *)q->ring; 215 u32 idx = cached_cons & q->ring_mask; 216 217 descs[nb_entries] = ring->desc[idx]; 218 if (unlikely(!xskq_cons_is_valid_desc(q, &descs[nb_entries], pool))) { 219 /* Skip the entry */ 220 cached_cons++; 221 continue; 222 } 223 224 nb_entries++; 225 cached_cons++; 226 } 227 228 /* Release valid plus any invalid entries */ 229 xskq_cons_release_n(q, cached_cons - q->cached_cons); 230 return nb_entries; 231 } 232 233 /* Functions for consumers */ 234 235 static inline void __xskq_cons_release(struct xsk_queue *q) 236 { 237 smp_store_release(&q->ring->consumer, q->cached_cons); /* D, matchees A */ 238 } 239 240 static inline void __xskq_cons_peek(struct xsk_queue *q) 241 { 242 /* Refresh the local pointer */ 243 q->cached_prod = smp_load_acquire(&q->ring->producer); /* C, matches B */ 244 } 245 246 static inline void xskq_cons_get_entries(struct xsk_queue *q) 247 { 248 __xskq_cons_release(q); 249 __xskq_cons_peek(q); 250 } 251 252 static inline u32 xskq_cons_nb_entries(struct xsk_queue *q, u32 max) 253 { 254 u32 entries = q->cached_prod - q->cached_cons; 255 256 if (entries >= max) 257 return max; 258 259 __xskq_cons_peek(q); 260 entries = q->cached_prod - q->cached_cons; 261 262 return entries >= max ? max : entries; 263 } 264 265 static inline bool xskq_cons_has_entries(struct xsk_queue *q, u32 cnt) 266 { 267 return xskq_cons_nb_entries(q, cnt) >= cnt; 268 } 269 270 static inline bool xskq_cons_peek_addr_unchecked(struct xsk_queue *q, u64 *addr) 271 { 272 if (q->cached_prod == q->cached_cons) 273 xskq_cons_get_entries(q); 274 return xskq_cons_read_addr_unchecked(q, addr); 275 } 276 277 static inline bool xskq_cons_peek_desc(struct xsk_queue *q, 278 struct xdp_desc *desc, 279 struct xsk_buff_pool *pool) 280 { 281 if (q->cached_prod == q->cached_cons) 282 xskq_cons_get_entries(q); 283 return xskq_cons_read_desc(q, desc, pool); 284 } 285 286 /* To improve performance in the xskq_cons_release functions, only update local state here. 287 * Reflect this to global state when we get new entries from the ring in 288 * xskq_cons_get_entries() and whenever Rx or Tx processing are completed in the NAPI loop. 289 */ 290 static inline void xskq_cons_release(struct xsk_queue *q) 291 { 292 q->cached_cons++; 293 } 294 295 static inline u32 xskq_cons_present_entries(struct xsk_queue *q) 296 { 297 /* No barriers needed since data is not accessed */ 298 return READ_ONCE(q->ring->producer) - READ_ONCE(q->ring->consumer); 299 } 300 301 /* Functions for producers */ 302 303 static inline u32 xskq_prod_nb_free(struct xsk_queue *q, u32 max) 304 { 305 u32 free_entries = q->nentries - (q->cached_prod - q->cached_cons); 306 307 if (free_entries >= max) 308 return max; 309 310 /* Refresh the local tail pointer */ 311 q->cached_cons = READ_ONCE(q->ring->consumer); 312 free_entries = q->nentries - (q->cached_prod - q->cached_cons); 313 314 return free_entries >= max ? max : free_entries; 315 } 316 317 static inline bool xskq_prod_is_full(struct xsk_queue *q) 318 { 319 return xskq_prod_nb_free(q, 1) ? false : true; 320 } 321 322 static inline void xskq_prod_cancel(struct xsk_queue *q) 323 { 324 q->cached_prod--; 325 } 326 327 static inline int xskq_prod_reserve(struct xsk_queue *q) 328 { 329 if (xskq_prod_is_full(q)) 330 return -ENOSPC; 331 332 /* A, matches D */ 333 q->cached_prod++; 334 return 0; 335 } 336 337 static inline int xskq_prod_reserve_addr(struct xsk_queue *q, u64 addr) 338 { 339 struct xdp_umem_ring *ring = (struct xdp_umem_ring *)q->ring; 340 341 if (xskq_prod_is_full(q)) 342 return -ENOSPC; 343 344 /* A, matches D */ 345 ring->desc[q->cached_prod++ & q->ring_mask] = addr; 346 return 0; 347 } 348 349 static inline void xskq_prod_write_addr_batch(struct xsk_queue *q, struct xdp_desc *descs, 350 u32 nb_entries) 351 { 352 struct xdp_umem_ring *ring = (struct xdp_umem_ring *)q->ring; 353 u32 i, cached_prod; 354 355 /* A, matches D */ 356 cached_prod = q->cached_prod; 357 for (i = 0; i < nb_entries; i++) 358 ring->desc[cached_prod++ & q->ring_mask] = descs[i].addr; 359 q->cached_prod = cached_prod; 360 } 361 362 static inline int xskq_prod_reserve_desc(struct xsk_queue *q, 363 u64 addr, u32 len) 364 { 365 struct xdp_rxtx_ring *ring = (struct xdp_rxtx_ring *)q->ring; 366 u32 idx; 367 368 if (xskq_prod_is_full(q)) 369 return -ENOBUFS; 370 371 /* A, matches D */ 372 idx = q->cached_prod++ & q->ring_mask; 373 ring->desc[idx].addr = addr; 374 ring->desc[idx].len = len; 375 376 return 0; 377 } 378 379 static inline void __xskq_prod_submit(struct xsk_queue *q, u32 idx) 380 { 381 smp_store_release(&q->ring->producer, idx); /* B, matches C */ 382 } 383 384 static inline void xskq_prod_submit(struct xsk_queue *q) 385 { 386 __xskq_prod_submit(q, q->cached_prod); 387 } 388 389 static inline void xskq_prod_submit_addr(struct xsk_queue *q, u64 addr) 390 { 391 struct xdp_umem_ring *ring = (struct xdp_umem_ring *)q->ring; 392 u32 idx = q->ring->producer; 393 394 ring->desc[idx++ & q->ring_mask] = addr; 395 396 __xskq_prod_submit(q, idx); 397 } 398 399 static inline void xskq_prod_submit_n(struct xsk_queue *q, u32 nb_entries) 400 { 401 __xskq_prod_submit(q, q->ring->producer + nb_entries); 402 } 403 404 static inline bool xskq_prod_is_empty(struct xsk_queue *q) 405 { 406 /* No barriers needed since data is not accessed */ 407 return READ_ONCE(q->ring->consumer) == READ_ONCE(q->ring->producer); 408 } 409 410 /* For both producers and consumers */ 411 412 static inline u64 xskq_nb_invalid_descs(struct xsk_queue *q) 413 { 414 return q ? q->invalid_descs : 0; 415 } 416 417 static inline u64 xskq_nb_queue_empty_descs(struct xsk_queue *q) 418 { 419 return q ? q->queue_empty_descs : 0; 420 } 421 422 struct xsk_queue *xskq_create(u32 nentries, bool umem_queue); 423 void xskq_destroy(struct xsk_queue *q_ops); 424 425 #endif /* _LINUX_XSK_QUEUE_H */ 426