verbs.h 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434
  1. /*
  2. * Copyright(c) 2015, 2016 Intel Corporation.
  3. *
  4. * This file is provided under a dual BSD/GPLv2 license. When using or
  5. * redistributing this file, you may do so under either license.
  6. *
  7. * GPL LICENSE SUMMARY
  8. *
  9. * This program is free software; you can redistribute it and/or modify
  10. * it under the terms of version 2 of the GNU General Public License as
  11. * published by the Free Software Foundation.
  12. *
  13. * This program is distributed in the hope that it will be useful, but
  14. * WITHOUT ANY WARRANTY; without even the implied warranty of
  15. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  16. * General Public License for more details.
  17. *
  18. * BSD LICENSE
  19. *
  20. * Redistribution and use in source and binary forms, with or without
  21. * modification, are permitted provided that the following conditions
  22. * are met:
  23. *
  24. * - Redistributions of source code must retain the above copyright
  25. * notice, this list of conditions and the following disclaimer.
  26. * - Redistributions in binary form must reproduce the above copyright
  27. * notice, this list of conditions and the following disclaimer in
  28. * the documentation and/or other materials provided with the
  29. * distribution.
  30. * - Neither the name of Intel Corporation nor the names of its
  31. * contributors may be used to endorse or promote products derived
  32. * from this software without specific prior written permission.
  33. *
  34. * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  35. * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  36. * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  37. * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  38. * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  39. * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  40. * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  41. * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  42. * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  43. * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  44. * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  45. *
  46. */
  47. #ifndef HFI1_VERBS_H
  48. #define HFI1_VERBS_H
  49. #include <linux/types.h>
  50. #include <linux/seqlock.h>
  51. #include <linux/kernel.h>
  52. #include <linux/interrupt.h>
  53. #include <linux/kref.h>
  54. #include <linux/workqueue.h>
  55. #include <linux/kthread.h>
  56. #include <linux/completion.h>
  57. #include <linux/slab.h>
  58. #include <rdma/ib_pack.h>
  59. #include <rdma/ib_user_verbs.h>
  60. #include <rdma/ib_mad.h>
  61. #include <rdma/ib_hdrs.h>
  62. #include <rdma/rdma_vt.h>
  63. #include <rdma/rdmavt_qp.h>
  64. #include <rdma/rdmavt_cq.h>
  65. struct hfi1_ctxtdata;
  66. struct hfi1_pportdata;
  67. struct hfi1_devdata;
  68. struct hfi1_packet;
  69. #include "iowait.h"
  70. #define HFI1_MAX_RDMA_ATOMIC 16
  71. /*
  72. * Increment this value if any changes that break userspace ABI
  73. * compatibility are made.
  74. */
  75. #define HFI1_UVERBS_ABI_VERSION 2
  76. /* IB Performance Manager status values */
  77. #define IB_PMA_SAMPLE_STATUS_DONE 0x00
  78. #define IB_PMA_SAMPLE_STATUS_STARTED 0x01
  79. #define IB_PMA_SAMPLE_STATUS_RUNNING 0x02
  80. /* Mandatory IB performance counter select values. */
  81. #define IB_PMA_PORT_XMIT_DATA cpu_to_be16(0x0001)
  82. #define IB_PMA_PORT_RCV_DATA cpu_to_be16(0x0002)
  83. #define IB_PMA_PORT_XMIT_PKTS cpu_to_be16(0x0003)
  84. #define IB_PMA_PORT_RCV_PKTS cpu_to_be16(0x0004)
  85. #define IB_PMA_PORT_XMIT_WAIT cpu_to_be16(0x0005)
  86. #define HFI1_VENDOR_IPG cpu_to_be16(0xFFA0)
  87. #define IB_DEFAULT_GID_PREFIX cpu_to_be64(0xfe80000000000000ULL)
  88. #define RC_OP(x) IB_OPCODE_RC_##x
  89. #define UC_OP(x) IB_OPCODE_UC_##x
  90. /* flags passed by hfi1_ib_rcv() */
  91. enum {
  92. HFI1_HAS_GRH = (1 << 0),
  93. };
  94. struct hfi1_ahg_info {
  95. u32 ahgdesc[2];
  96. u16 tx_flags;
  97. u8 ahgcount;
  98. u8 ahgidx;
  99. };
  100. struct hfi1_sdma_header {
  101. __le64 pbc;
  102. struct ib_header hdr;
  103. } __packed;
  104. /*
  105. * hfi1 specific data structures that will be hidden from rvt after the queue
  106. * pair is made common
  107. */
  108. struct hfi1_qp_priv {
  109. struct hfi1_ahg_info *s_ahg; /* ahg info for next header */
  110. struct sdma_engine *s_sde; /* current sde */
  111. struct send_context *s_sendcontext; /* current sendcontext */
  112. u8 s_sc; /* SC[0..4] for next packet */
  113. u8 r_adefered; /* number of acks defered */
  114. struct iowait s_iowait;
  115. struct rvt_qp *owner;
  116. };
  117. /*
  118. * This structure is used to hold commonly lookedup and computed values during
  119. * the send engine progress.
  120. */
  121. struct hfi1_pkt_state {
  122. struct hfi1_ibdev *dev;
  123. struct hfi1_ibport *ibp;
  124. struct hfi1_pportdata *ppd;
  125. struct verbs_txreq *s_txreq;
  126. unsigned long flags;
  127. };
  128. #define HFI1_PSN_CREDIT 16
  129. struct hfi1_opcode_stats {
  130. u64 n_packets; /* number of packets */
  131. u64 n_bytes; /* total number of bytes */
  132. };
  133. struct hfi1_opcode_stats_perctx {
  134. struct hfi1_opcode_stats stats[256];
  135. };
  136. static inline void inc_opstats(
  137. u32 tlen,
  138. struct hfi1_opcode_stats *stats)
  139. {
  140. #ifdef CONFIG_DEBUG_FS
  141. stats->n_bytes += tlen;
  142. stats->n_packets++;
  143. #endif
  144. }
  145. struct hfi1_ibport {
  146. struct rvt_qp __rcu *qp[2];
  147. struct rvt_ibport rvp;
  148. /* the first 16 entries are sl_to_vl for !OPA */
  149. u8 sl_to_sc[32];
  150. u8 sc_to_sl[32];
  151. };
  152. struct hfi1_ibdev {
  153. struct rvt_dev_info rdi; /* Must be first */
  154. /* QP numbers are shared by all IB ports */
  155. /* protect txwait list */
  156. seqlock_t txwait_lock ____cacheline_aligned_in_smp;
  157. struct list_head txwait; /* list for wait verbs_txreq */
  158. struct list_head memwait; /* list for wait kernel memory */
  159. struct kmem_cache *verbs_txreq_cache;
  160. u64 n_txwait;
  161. u64 n_kmem_wait;
  162. /* protect iowait lists */
  163. seqlock_t iowait_lock ____cacheline_aligned_in_smp;
  164. u64 n_piowait;
  165. u64 n_piodrain;
  166. struct timer_list mem_timer;
  167. #ifdef CONFIG_DEBUG_FS
  168. /* per HFI debugfs */
  169. struct dentry *hfi1_ibdev_dbg;
  170. /* per HFI symlinks to above */
  171. struct dentry *hfi1_ibdev_link;
  172. #endif
  173. };
  174. static inline struct hfi1_ibdev *to_idev(struct ib_device *ibdev)
  175. {
  176. struct rvt_dev_info *rdi;
  177. rdi = container_of(ibdev, struct rvt_dev_info, ibdev);
  178. return container_of(rdi, struct hfi1_ibdev, rdi);
  179. }
  180. static inline struct rvt_qp *iowait_to_qp(struct iowait *s_iowait)
  181. {
  182. struct hfi1_qp_priv *priv;
  183. priv = container_of(s_iowait, struct hfi1_qp_priv, s_iowait);
  184. return priv->owner;
  185. }
  186. /*
  187. * Send if not busy or waiting for I/O and either
  188. * a RC response is pending or we can process send work requests.
  189. */
  190. static inline int hfi1_send_ok(struct rvt_qp *qp)
  191. {
  192. return !(qp->s_flags & (RVT_S_BUSY | RVT_S_ANY_WAIT_IO)) &&
  193. (qp->s_hdrwords || (qp->s_flags & RVT_S_RESP_PENDING) ||
  194. !(qp->s_flags & RVT_S_ANY_WAIT_SEND));
  195. }
  196. /*
  197. * This must be called with s_lock held.
  198. */
  199. void hfi1_bad_pqkey(struct hfi1_ibport *ibp, __be16 trap_num, u32 key, u32 sl,
  200. u32 qp1, u32 qp2, u16 lid1, u16 lid2);
  201. void hfi1_cap_mask_chg(struct rvt_dev_info *rdi, u8 port_num);
  202. void hfi1_sys_guid_chg(struct hfi1_ibport *ibp);
  203. void hfi1_node_desc_chg(struct hfi1_ibport *ibp);
  204. int hfi1_process_mad(struct ib_device *ibdev, int mad_flags, u8 port,
  205. const struct ib_wc *in_wc, const struct ib_grh *in_grh,
  206. const struct ib_mad_hdr *in_mad, size_t in_mad_size,
  207. struct ib_mad_hdr *out_mad, size_t *out_mad_size,
  208. u16 *out_mad_pkey_index);
  209. /*
  210. * The PSN_MASK and PSN_SHIFT allow for
  211. * 1) comparing two PSNs
  212. * 2) returning the PSN with any upper bits masked
  213. * 3) returning the difference between to PSNs
  214. *
  215. * The number of significant bits in the PSN must
  216. * necessarily be at least one bit less than
  217. * the container holding the PSN.
  218. */
  219. #ifndef CONFIG_HFI1_VERBS_31BIT_PSN
  220. #define PSN_MASK 0xFFFFFF
  221. #define PSN_SHIFT 8
  222. #else
  223. #define PSN_MASK 0x7FFFFFFF
  224. #define PSN_SHIFT 1
  225. #endif
  226. #define PSN_MODIFY_MASK 0xFFFFFF
  227. /*
  228. * Compare two PSNs
  229. * Returns an integer <, ==, or > than zero.
  230. */
  231. static inline int cmp_psn(u32 a, u32 b)
  232. {
  233. return (((int)a) - ((int)b)) << PSN_SHIFT;
  234. }
  235. /*
  236. * Return masked PSN
  237. */
  238. static inline u32 mask_psn(u32 a)
  239. {
  240. return a & PSN_MASK;
  241. }
  242. /*
  243. * Return delta between two PSNs
  244. */
  245. static inline u32 delta_psn(u32 a, u32 b)
  246. {
  247. return (((int)a - (int)b) << PSN_SHIFT) >> PSN_SHIFT;
  248. }
  249. struct verbs_txreq;
  250. void hfi1_put_txreq(struct verbs_txreq *tx);
  251. int hfi1_verbs_send(struct rvt_qp *qp, struct hfi1_pkt_state *ps);
  252. void hfi1_copy_sge(struct rvt_sge_state *ss, void *data, u32 length,
  253. bool release, bool copy_last);
  254. void hfi1_cnp_rcv(struct hfi1_packet *packet);
  255. void hfi1_uc_rcv(struct hfi1_packet *packet);
  256. void hfi1_rc_rcv(struct hfi1_packet *packet);
  257. void hfi1_rc_hdrerr(
  258. struct hfi1_ctxtdata *rcd,
  259. struct ib_header *hdr,
  260. u32 rcv_flags,
  261. struct rvt_qp *qp);
  262. u8 ah_to_sc(struct ib_device *ibdev, struct ib_ah_attr *ah_attr);
  263. struct ib_ah *hfi1_create_qp0_ah(struct hfi1_ibport *ibp, u16 dlid);
  264. void hfi1_rc_send_complete(struct rvt_qp *qp, struct ib_header *hdr);
  265. void hfi1_ud_rcv(struct hfi1_packet *packet);
  266. int hfi1_lookup_pkey_idx(struct hfi1_ibport *ibp, u16 pkey);
  267. int hfi1_rvt_get_rwqe(struct rvt_qp *qp, int wr_id_only);
  268. void hfi1_migrate_qp(struct rvt_qp *qp);
  269. int hfi1_check_modify_qp(struct rvt_qp *qp, struct ib_qp_attr *attr,
  270. int attr_mask, struct ib_udata *udata);
  271. void hfi1_modify_qp(struct rvt_qp *qp, struct ib_qp_attr *attr,
  272. int attr_mask, struct ib_udata *udata);
  273. void hfi1_restart_rc(struct rvt_qp *qp, u32 psn, int wait);
  274. int hfi1_check_send_wqe(struct rvt_qp *qp, struct rvt_swqe *wqe);
  275. extern const u32 rc_only_opcode;
  276. extern const u32 uc_only_opcode;
  277. static inline u8 get_opcode(struct ib_header *h)
  278. {
  279. u16 lnh = be16_to_cpu(h->lrh[0]) & 3;
  280. if (lnh == IB_LNH_IBA_LOCAL)
  281. return be32_to_cpu(h->u.oth.bth[0]) >> 24;
  282. else
  283. return be32_to_cpu(h->u.l.oth.bth[0]) >> 24;
  284. }
  285. int hfi1_ruc_check_hdr(struct hfi1_ibport *ibp, struct ib_header *hdr,
  286. int has_grh, struct rvt_qp *qp, u32 bth0);
  287. u32 hfi1_make_grh(struct hfi1_ibport *ibp, struct ib_grh *hdr,
  288. struct ib_global_route *grh, u32 hwords, u32 nwords);
  289. void hfi1_make_ruc_header(struct rvt_qp *qp, struct ib_other_headers *ohdr,
  290. u32 bth0, u32 bth2, int middle,
  291. struct hfi1_pkt_state *ps);
  292. void _hfi1_do_send(struct work_struct *work);
  293. void hfi1_do_send(struct rvt_qp *qp);
  294. void hfi1_send_complete(struct rvt_qp *qp, struct rvt_swqe *wqe,
  295. enum ib_wc_status status);
  296. void hfi1_send_rc_ack(struct hfi1_ctxtdata *, struct rvt_qp *qp, int is_fecn);
  297. int hfi1_make_rc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps);
  298. int hfi1_make_uc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps);
  299. int hfi1_make_ud_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps);
  300. int hfi1_register_ib_device(struct hfi1_devdata *);
  301. void hfi1_unregister_ib_device(struct hfi1_devdata *);
  302. void hfi1_ib_rcv(struct hfi1_packet *packet);
  303. unsigned hfi1_get_npkeys(struct hfi1_devdata *);
  304. int hfi1_verbs_send_dma(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
  305. u64 pbc);
  306. int hfi1_verbs_send_pio(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
  307. u64 pbc);
  308. int hfi1_wss_init(void);
  309. void hfi1_wss_exit(void);
  310. /* platform specific: return the lowest level cache (llc) size, in KiB */
  311. static inline int wss_llc_size(void)
  312. {
  313. /* assume that the boot CPU value is universal for all CPUs */
  314. return boot_cpu_data.x86_cache_size;
  315. }
  316. /* platform specific: cacheless copy */
  317. static inline void cacheless_memcpy(void *dst, void *src, size_t n)
  318. {
  319. /*
  320. * Use the only available X64 cacheless copy. Add a __user cast
  321. * to quiet sparse. The src agument is already in the kernel so
  322. * there are no security issues. The extra fault recovery machinery
  323. * is not invoked.
  324. */
  325. __copy_user_nocache(dst, (void __user *)src, n, 0);
  326. }
  327. extern const enum ib_wc_opcode ib_hfi1_wc_opcode[];
  328. extern const u8 hdr_len_by_opcode[];
  329. extern const int ib_rvt_state_ops[];
  330. extern __be64 ib_hfi1_sys_image_guid; /* in network order */
  331. extern unsigned int hfi1_max_cqes;
  332. extern unsigned int hfi1_max_cqs;
  333. extern unsigned int hfi1_max_qp_wrs;
  334. extern unsigned int hfi1_max_qps;
  335. extern unsigned int hfi1_max_sges;
  336. extern unsigned int hfi1_max_mcast_grps;
  337. extern unsigned int hfi1_max_mcast_qp_attached;
  338. extern unsigned int hfi1_max_srqs;
  339. extern unsigned int hfi1_max_srq_sges;
  340. extern unsigned int hfi1_max_srq_wrs;
  341. extern unsigned short piothreshold;
  342. extern const u32 ib_hfi1_rnr_table[];
  343. #endif /* HFI1_VERBS_H */