verbs.h 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444
  1. /*
  2. * Copyright(c) 2015 - 2017 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. struct iowait s_iowait;
  114. struct rvt_qp *owner;
  115. };
  116. /*
  117. * This structure is used to hold commonly lookedup and computed values during
  118. * the send engine progress.
  119. */
  120. struct hfi1_pkt_state {
  121. struct hfi1_ibdev *dev;
  122. struct hfi1_ibport *ibp;
  123. struct hfi1_pportdata *ppd;
  124. struct verbs_txreq *s_txreq;
  125. unsigned long flags;
  126. unsigned long timeout;
  127. unsigned long timeout_int;
  128. int cpu;
  129. bool in_thread;
  130. };
  131. #define HFI1_PSN_CREDIT 16
  132. struct hfi1_opcode_stats {
  133. u64 n_packets; /* number of packets */
  134. u64 n_bytes; /* total number of bytes */
  135. };
  136. struct hfi1_opcode_stats_perctx {
  137. struct hfi1_opcode_stats stats[256];
  138. };
  139. static inline void inc_opstats(
  140. u32 tlen,
  141. struct hfi1_opcode_stats *stats)
  142. {
  143. #ifdef CONFIG_DEBUG_FS
  144. stats->n_bytes += tlen;
  145. stats->n_packets++;
  146. #endif
  147. }
  148. struct hfi1_ibport {
  149. struct rvt_qp __rcu *qp[2];
  150. struct rvt_ibport rvp;
  151. /* the first 16 entries are sl_to_vl for !OPA */
  152. u8 sl_to_sc[32];
  153. u8 sc_to_sl[32];
  154. };
  155. struct hfi1_ibdev {
  156. struct rvt_dev_info rdi; /* Must be first */
  157. /* QP numbers are shared by all IB ports */
  158. /* protect txwait list */
  159. seqlock_t txwait_lock ____cacheline_aligned_in_smp;
  160. struct list_head txwait; /* list for wait verbs_txreq */
  161. struct list_head memwait; /* list for wait kernel memory */
  162. struct kmem_cache *verbs_txreq_cache;
  163. u64 n_txwait;
  164. u64 n_kmem_wait;
  165. /* protect iowait lists */
  166. seqlock_t iowait_lock ____cacheline_aligned_in_smp;
  167. u64 n_piowait;
  168. u64 n_piodrain;
  169. struct timer_list mem_timer;
  170. #ifdef CONFIG_DEBUG_FS
  171. /* per HFI debugfs */
  172. struct dentry *hfi1_ibdev_dbg;
  173. /* per HFI symlinks to above */
  174. struct dentry *hfi1_ibdev_link;
  175. #ifdef CONFIG_FAULT_INJECTION
  176. struct fault_opcode *fault_opcode;
  177. struct fault_packet *fault_packet;
  178. bool fault_suppress_err;
  179. #endif
  180. #endif
  181. };
  182. static inline struct hfi1_ibdev *to_idev(struct ib_device *ibdev)
  183. {
  184. struct rvt_dev_info *rdi;
  185. rdi = container_of(ibdev, struct rvt_dev_info, ibdev);
  186. return container_of(rdi, struct hfi1_ibdev, rdi);
  187. }
  188. static inline struct rvt_qp *iowait_to_qp(struct iowait *s_iowait)
  189. {
  190. struct hfi1_qp_priv *priv;
  191. priv = container_of(s_iowait, struct hfi1_qp_priv, s_iowait);
  192. return priv->owner;
  193. }
  194. /*
  195. * Send if not busy or waiting for I/O and either
  196. * a RC response is pending or we can process send work requests.
  197. */
  198. static inline int hfi1_send_ok(struct rvt_qp *qp)
  199. {
  200. return !(qp->s_flags & (RVT_S_BUSY | RVT_S_ANY_WAIT_IO)) &&
  201. (qp->s_hdrwords || (qp->s_flags & RVT_S_RESP_PENDING) ||
  202. !(qp->s_flags & RVT_S_ANY_WAIT_SEND));
  203. }
  204. /*
  205. * This must be called with s_lock held.
  206. */
  207. void hfi1_bad_pqkey(struct hfi1_ibport *ibp, __be16 trap_num, u32 key, u32 sl,
  208. u32 qp1, u32 qp2, u16 lid1, u16 lid2);
  209. void hfi1_cap_mask_chg(struct rvt_dev_info *rdi, u8 port_num);
  210. void hfi1_sys_guid_chg(struct hfi1_ibport *ibp);
  211. void hfi1_node_desc_chg(struct hfi1_ibport *ibp);
  212. int hfi1_process_mad(struct ib_device *ibdev, int mad_flags, u8 port,
  213. const struct ib_wc *in_wc, const struct ib_grh *in_grh,
  214. const struct ib_mad_hdr *in_mad, size_t in_mad_size,
  215. struct ib_mad_hdr *out_mad, size_t *out_mad_size,
  216. u16 *out_mad_pkey_index);
  217. /*
  218. * The PSN_MASK and PSN_SHIFT allow for
  219. * 1) comparing two PSNs
  220. * 2) returning the PSN with any upper bits masked
  221. * 3) returning the difference between to PSNs
  222. *
  223. * The number of significant bits in the PSN must
  224. * necessarily be at least one bit less than
  225. * the container holding the PSN.
  226. */
  227. #ifndef CONFIG_HFI1_VERBS_31BIT_PSN
  228. #define PSN_MASK 0xFFFFFF
  229. #define PSN_SHIFT 8
  230. #else
  231. #define PSN_MASK 0x7FFFFFFF
  232. #define PSN_SHIFT 1
  233. #endif
  234. #define PSN_MODIFY_MASK 0xFFFFFF
  235. /*
  236. * Compare two PSNs
  237. * Returns an integer <, ==, or > than zero.
  238. */
  239. static inline int cmp_psn(u32 a, u32 b)
  240. {
  241. return (((int)a) - ((int)b)) << PSN_SHIFT;
  242. }
  243. /*
  244. * Return masked PSN
  245. */
  246. static inline u32 mask_psn(u32 a)
  247. {
  248. return a & PSN_MASK;
  249. }
  250. /*
  251. * Return delta between two PSNs
  252. */
  253. static inline u32 delta_psn(u32 a, u32 b)
  254. {
  255. return (((int)a - (int)b) << PSN_SHIFT) >> PSN_SHIFT;
  256. }
  257. struct verbs_txreq;
  258. void hfi1_put_txreq(struct verbs_txreq *tx);
  259. int hfi1_verbs_send(struct rvt_qp *qp, struct hfi1_pkt_state *ps);
  260. void hfi1_copy_sge(struct rvt_sge_state *ss, void *data, u32 length,
  261. bool release, bool copy_last);
  262. void hfi1_cnp_rcv(struct hfi1_packet *packet);
  263. void hfi1_uc_rcv(struct hfi1_packet *packet);
  264. void hfi1_rc_rcv(struct hfi1_packet *packet);
  265. void hfi1_rc_hdrerr(
  266. struct hfi1_ctxtdata *rcd,
  267. struct ib_header *hdr,
  268. u32 rcv_flags,
  269. struct rvt_qp *qp);
  270. u8 ah_to_sc(struct ib_device *ibdev, struct rdma_ah_attr *ah_attr);
  271. struct ib_ah *hfi1_create_qp0_ah(struct hfi1_ibport *ibp, u16 dlid);
  272. void hfi1_rc_send_complete(struct rvt_qp *qp, struct ib_header *hdr);
  273. void hfi1_ud_rcv(struct hfi1_packet *packet);
  274. int hfi1_lookup_pkey_idx(struct hfi1_ibport *ibp, u16 pkey);
  275. int hfi1_rvt_get_rwqe(struct rvt_qp *qp, int wr_id_only);
  276. void hfi1_migrate_qp(struct rvt_qp *qp);
  277. int hfi1_check_modify_qp(struct rvt_qp *qp, struct ib_qp_attr *attr,
  278. int attr_mask, struct ib_udata *udata);
  279. void hfi1_modify_qp(struct rvt_qp *qp, struct ib_qp_attr *attr,
  280. int attr_mask, struct ib_udata *udata);
  281. void hfi1_restart_rc(struct rvt_qp *qp, u32 psn, int wait);
  282. int hfi1_check_send_wqe(struct rvt_qp *qp, struct rvt_swqe *wqe);
  283. extern const u32 rc_only_opcode;
  284. extern const u32 uc_only_opcode;
  285. static inline u8 get_opcode(struct ib_header *h)
  286. {
  287. u16 lnh = be16_to_cpu(h->lrh[0]) & 3;
  288. if (lnh == IB_LNH_IBA_LOCAL)
  289. return be32_to_cpu(h->u.oth.bth[0]) >> 24;
  290. else
  291. return be32_to_cpu(h->u.l.oth.bth[0]) >> 24;
  292. }
  293. int hfi1_ruc_check_hdr(struct hfi1_ibport *ibp, struct ib_header *hdr,
  294. int has_grh, struct rvt_qp *qp, u32 bth0);
  295. u32 hfi1_make_grh(struct hfi1_ibport *ibp, struct ib_grh *hdr,
  296. const struct ib_global_route *grh, u32 hwords, u32 nwords);
  297. void hfi1_make_ruc_header(struct rvt_qp *qp, struct ib_other_headers *ohdr,
  298. u32 bth0, u32 bth2, int middle,
  299. struct hfi1_pkt_state *ps);
  300. void _hfi1_do_send(struct work_struct *work);
  301. void hfi1_do_send_from_rvt(struct rvt_qp *qp);
  302. void hfi1_do_send(struct rvt_qp *qp, bool in_thread);
  303. void hfi1_send_complete(struct rvt_qp *qp, struct rvt_swqe *wqe,
  304. enum ib_wc_status status);
  305. void hfi1_send_rc_ack(struct hfi1_ctxtdata *, struct rvt_qp *qp, int is_fecn);
  306. int hfi1_make_rc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps);
  307. int hfi1_make_uc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps);
  308. int hfi1_make_ud_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps);
  309. int hfi1_register_ib_device(struct hfi1_devdata *);
  310. void hfi1_unregister_ib_device(struct hfi1_devdata *);
  311. void hfi1_ib_rcv(struct hfi1_packet *packet);
  312. unsigned hfi1_get_npkeys(struct hfi1_devdata *);
  313. int hfi1_verbs_send_dma(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
  314. u64 pbc);
  315. int hfi1_verbs_send_pio(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
  316. u64 pbc);
  317. int hfi1_wss_init(void);
  318. void hfi1_wss_exit(void);
  319. /* platform specific: return the lowest level cache (llc) size, in KiB */
  320. static inline int wss_llc_size(void)
  321. {
  322. /* assume that the boot CPU value is universal for all CPUs */
  323. return boot_cpu_data.x86_cache_size;
  324. }
  325. /* platform specific: cacheless copy */
  326. static inline void cacheless_memcpy(void *dst, void *src, size_t n)
  327. {
  328. /*
  329. * Use the only available X64 cacheless copy. Add a __user cast
  330. * to quiet sparse. The src agument is already in the kernel so
  331. * there are no security issues. The extra fault recovery machinery
  332. * is not invoked.
  333. */
  334. __copy_user_nocache(dst, (void __user *)src, n, 0);
  335. }
  336. extern const enum ib_wc_opcode ib_hfi1_wc_opcode[];
  337. extern const u8 hdr_len_by_opcode[];
  338. extern const int ib_rvt_state_ops[];
  339. extern __be64 ib_hfi1_sys_image_guid; /* in network order */
  340. extern unsigned int hfi1_max_cqes;
  341. extern unsigned int hfi1_max_cqs;
  342. extern unsigned int hfi1_max_qp_wrs;
  343. extern unsigned int hfi1_max_qps;
  344. extern unsigned int hfi1_max_sges;
  345. extern unsigned int hfi1_max_mcast_grps;
  346. extern unsigned int hfi1_max_mcast_qp_attached;
  347. extern unsigned int hfi1_max_srqs;
  348. extern unsigned int hfi1_max_srq_sges;
  349. extern unsigned int hfi1_max_srq_wrs;
  350. extern unsigned short piothreshold;
  351. extern const u32 ib_hfi1_rnr_table[];
  352. #endif /* HFI1_VERBS_H */