xsk_buff_pool.h 6.4 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224
  1. /* SPDX-License-Identifier: GPL-2.0 */
  2. /* Copyright(c) 2020 Intel Corporation. */
  3. #ifndef XSK_BUFF_POOL_H_
  4. #define XSK_BUFF_POOL_H_
  5. #include <linux/if_xdp.h>
  6. #include <linux/types.h>
  7. #include <linux/dma-mapping.h>
  8. #include <linux/bpf.h>
  9. #include <net/xdp.h>
  10. struct xsk_buff_pool;
  11. struct xdp_rxq_info;
  12. struct xsk_queue;
  13. struct xdp_desc;
  14. struct xdp_umem;
  15. struct xdp_sock;
  16. struct device;
  17. struct page;
  18. struct xdp_buff_xsk {
  19. struct xdp_buff xdp;
  20. dma_addr_t dma;
  21. dma_addr_t frame_dma;
  22. struct xsk_buff_pool *pool;
  23. u64 orig_addr;
  24. struct list_head free_list_node;
  25. };
  26. struct xsk_dma_map {
  27. dma_addr_t *dma_pages;
  28. struct device *dev;
  29. struct net_device *netdev;
  30. refcount_t users;
  31. struct list_head list; /* Protected by the RTNL_LOCK */
  32. u32 dma_pages_cnt;
  33. bool dma_need_sync;
  34. };
  35. struct xsk_buff_pool {
  36. /* Members only used in the control path first. */
  37. struct device *dev;
  38. struct net_device *netdev;
  39. struct list_head xsk_tx_list;
  40. /* Protects modifications to the xsk_tx_list */
  41. spinlock_t xsk_tx_list_lock;
  42. refcount_t users;
  43. struct xdp_umem *umem;
  44. struct work_struct work;
  45. struct list_head free_list;
  46. u32 heads_cnt;
  47. u16 queue_id;
  48. /* Data path members as close to free_heads at the end as possible. */
  49. struct xsk_queue *fq ____cacheline_aligned_in_smp;
  50. struct xsk_queue *cq;
  51. /* For performance reasons, each buff pool has its own array of dma_pages
  52. * even when they are identical.
  53. */
  54. dma_addr_t *dma_pages;
  55. struct xdp_buff_xsk *heads;
  56. struct xdp_desc *tx_descs;
  57. u64 chunk_mask;
  58. u64 addrs_cnt;
  59. u32 free_list_cnt;
  60. u32 dma_pages_cnt;
  61. u32 free_heads_cnt;
  62. u32 headroom;
  63. u32 chunk_size;
  64. u32 chunk_shift;
  65. u32 frame_len;
  66. u8 cached_need_wakeup;
  67. bool uses_need_wakeup;
  68. bool dma_need_sync;
  69. bool unaligned;
  70. void *addrs;
  71. /* Mutual exclusion of the completion ring in the SKB mode. Two cases to protect:
  72. * NAPI TX thread and sendmsg error paths in the SKB destructor callback and when
  73. * sockets share a single cq when the same netdev and queue id is shared.
  74. */
  75. spinlock_t cq_lock;
  76. struct xdp_buff_xsk *free_heads[];
  77. };
  78. /* Masks for xdp_umem_page flags.
  79. * The low 12-bits of the addr will be 0 since this is the page address, so we
  80. * can use them for flags.
  81. */
  82. #define XSK_NEXT_PG_CONTIG_SHIFT 0
  83. #define XSK_NEXT_PG_CONTIG_MASK BIT_ULL(XSK_NEXT_PG_CONTIG_SHIFT)
  84. /* AF_XDP core. */
  85. struct xsk_buff_pool *xp_create_and_assign_umem(struct xdp_sock *xs,
  86. struct xdp_umem *umem);
  87. int xp_assign_dev(struct xsk_buff_pool *pool, struct net_device *dev,
  88. u16 queue_id, u16 flags);
  89. int xp_assign_dev_shared(struct xsk_buff_pool *pool, struct xdp_sock *umem_xs,
  90. struct net_device *dev, u16 queue_id);
  91. int xp_alloc_tx_descs(struct xsk_buff_pool *pool, struct xdp_sock *xs);
  92. void xp_destroy(struct xsk_buff_pool *pool);
  93. void xp_get_pool(struct xsk_buff_pool *pool);
  94. bool xp_put_pool(struct xsk_buff_pool *pool);
  95. void xp_clear_dev(struct xsk_buff_pool *pool);
  96. void xp_add_xsk(struct xsk_buff_pool *pool, struct xdp_sock *xs);
  97. void xp_del_xsk(struct xsk_buff_pool *pool, struct xdp_sock *xs);
  98. /* AF_XDP, and XDP core. */
  99. void xp_free(struct xdp_buff_xsk *xskb);
  100. static inline void xp_init_xskb_addr(struct xdp_buff_xsk *xskb, struct xsk_buff_pool *pool,
  101. u64 addr)
  102. {
  103. xskb->orig_addr = addr;
  104. xskb->xdp.data_hard_start = pool->addrs + addr + pool->headroom;
  105. }
  106. static inline void xp_init_xskb_dma(struct xdp_buff_xsk *xskb, struct xsk_buff_pool *pool,
  107. dma_addr_t *dma_pages, u64 addr)
  108. {
  109. xskb->frame_dma = (dma_pages[addr >> PAGE_SHIFT] & ~XSK_NEXT_PG_CONTIG_MASK) +
  110. (addr & ~PAGE_MASK);
  111. xskb->dma = xskb->frame_dma + pool->headroom + XDP_PACKET_HEADROOM;
  112. }
  113. /* AF_XDP ZC drivers, via xdp_sock_buff.h */
  114. void xp_set_rxq_info(struct xsk_buff_pool *pool, struct xdp_rxq_info *rxq);
  115. int xp_dma_map(struct xsk_buff_pool *pool, struct device *dev,
  116. unsigned long attrs, struct page **pages, u32 nr_pages);
  117. void xp_dma_unmap(struct xsk_buff_pool *pool, unsigned long attrs);
  118. struct xdp_buff *xp_alloc(struct xsk_buff_pool *pool);
  119. u32 xp_alloc_batch(struct xsk_buff_pool *pool, struct xdp_buff **xdp, u32 max);
  120. bool xp_can_alloc(struct xsk_buff_pool *pool, u32 count);
  121. void *xp_raw_get_data(struct xsk_buff_pool *pool, u64 addr);
  122. dma_addr_t xp_raw_get_dma(struct xsk_buff_pool *pool, u64 addr);
  123. static inline dma_addr_t xp_get_dma(struct xdp_buff_xsk *xskb)
  124. {
  125. return xskb->dma;
  126. }
  127. static inline dma_addr_t xp_get_frame_dma(struct xdp_buff_xsk *xskb)
  128. {
  129. return xskb->frame_dma;
  130. }
  131. void xp_dma_sync_for_cpu_slow(struct xdp_buff_xsk *xskb);
  132. static inline void xp_dma_sync_for_cpu(struct xdp_buff_xsk *xskb)
  133. {
  134. xp_dma_sync_for_cpu_slow(xskb);
  135. }
  136. void xp_dma_sync_for_device_slow(struct xsk_buff_pool *pool, dma_addr_t dma,
  137. size_t size);
  138. static inline void xp_dma_sync_for_device(struct xsk_buff_pool *pool,
  139. dma_addr_t dma, size_t size)
  140. {
  141. if (!pool->dma_need_sync)
  142. return;
  143. xp_dma_sync_for_device_slow(pool, dma, size);
  144. }
  145. /* Masks for xdp_umem_page flags.
  146. * The low 12-bits of the addr will be 0 since this is the page address, so we
  147. * can use them for flags.
  148. */
  149. #define XSK_NEXT_PG_CONTIG_SHIFT 0
  150. #define XSK_NEXT_PG_CONTIG_MASK BIT_ULL(XSK_NEXT_PG_CONTIG_SHIFT)
  151. static inline bool xp_desc_crosses_non_contig_pg(struct xsk_buff_pool *pool,
  152. u64 addr, u32 len)
  153. {
  154. bool cross_pg = (addr & (PAGE_SIZE - 1)) + len > PAGE_SIZE;
  155. if (likely(!cross_pg))
  156. return false;
  157. return pool->dma_pages_cnt &&
  158. !(pool->dma_pages[addr >> PAGE_SHIFT] & XSK_NEXT_PG_CONTIG_MASK);
  159. }
  160. static inline u64 xp_aligned_extract_addr(struct xsk_buff_pool *pool, u64 addr)
  161. {
  162. return addr & pool->chunk_mask;
  163. }
  164. static inline u64 xp_unaligned_extract_addr(u64 addr)
  165. {
  166. return addr & XSK_UNALIGNED_BUF_ADDR_MASK;
  167. }
  168. static inline u64 xp_unaligned_extract_offset(u64 addr)
  169. {
  170. return addr >> XSK_UNALIGNED_BUF_OFFSET_SHIFT;
  171. }
  172. static inline u64 xp_unaligned_add_offset_to_addr(u64 addr)
  173. {
  174. return xp_unaligned_extract_addr(addr) +
  175. xp_unaligned_extract_offset(addr);
  176. }
  177. static inline u32 xp_aligned_extract_idx(struct xsk_buff_pool *pool, u64 addr)
  178. {
  179. return xp_aligned_extract_addr(pool, addr) >> pool->chunk_shift;
  180. }
  181. static inline void xp_release(struct xdp_buff_xsk *xskb)
  182. {
  183. if (xskb->pool->unaligned)
  184. xskb->pool->free_heads[xskb->pool->free_heads_cnt++] = xskb;
  185. }
  186. static inline u64 xp_get_handle(struct xdp_buff_xsk *xskb)
  187. {
  188. u64 offset = xskb->xdp.data - xskb->xdp.data_hard_start;
  189. offset += xskb->pool->headroom;
  190. if (!xskb->pool->unaligned)
  191. return xskb->orig_addr + offset;
  192. return xskb->orig_addr + (offset << XSK_UNALIGNED_BUF_OFFSET_SHIFT);
  193. }
  194. #endif /* XSK_BUFF_POOL_H_ */