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 48 #ifndef HFI1_VERBS_H 49 #define HFI1_VERBS_H 50 51 #include <linux/types.h> 52 #include <linux/seqlock.h> 53 #include <linux/kernel.h> 54 #include <linux/interrupt.h> 55 #include <linux/kref.h> 56 #include <linux/workqueue.h> 57 #include <linux/kthread.h> 58 #include <linux/completion.h> 59 #include <linux/slab.h> 60 #include <rdma/ib_pack.h> 61 #include <rdma/ib_user_verbs.h> 62 #include <rdma/ib_mad.h> 63 #include <rdma/ib_hdrs.h> 64 #include <rdma/rdma_vt.h> 65 #include <rdma/rdmavt_qp.h> 66 #include <rdma/rdmavt_cq.h> 67 68 struct hfi1_ctxtdata; 69 struct hfi1_pportdata; 70 struct hfi1_devdata; 71 struct hfi1_packet; 72 73 #include "iowait.h" 74 75 #define HFI1_MAX_RDMA_ATOMIC 16 76 77 /* 78 * Increment this value if any changes that break userspace ABI 79 * compatibility are made. 80 */ 81 #define HFI1_UVERBS_ABI_VERSION 2 82 83 /* IB Performance Manager status values */ 84 #define IB_PMA_SAMPLE_STATUS_DONE 0x00 85 #define IB_PMA_SAMPLE_STATUS_STARTED 0x01 86 #define IB_PMA_SAMPLE_STATUS_RUNNING 0x02 87 88 /* Mandatory IB performance counter select values. */ 89 #define IB_PMA_PORT_XMIT_DATA cpu_to_be16(0x0001) 90 #define IB_PMA_PORT_RCV_DATA cpu_to_be16(0x0002) 91 #define IB_PMA_PORT_XMIT_PKTS cpu_to_be16(0x0003) 92 #define IB_PMA_PORT_RCV_PKTS cpu_to_be16(0x0004) 93 #define IB_PMA_PORT_XMIT_WAIT cpu_to_be16(0x0005) 94 95 #define HFI1_VENDOR_IPG cpu_to_be16(0xFFA0) 96 97 #define IB_DEFAULT_GID_PREFIX cpu_to_be64(0xfe80000000000000ULL) 98 99 #define RC_OP(x) IB_OPCODE_RC_##x 100 #define UC_OP(x) IB_OPCODE_UC_##x 101 102 /* flags passed by hfi1_ib_rcv() */ 103 enum { 104 HFI1_HAS_GRH = (1 << 0), 105 }; 106 107 struct hfi1_ahg_info { 108 u32 ahgdesc[2]; 109 u16 tx_flags; 110 u8 ahgcount; 111 u8 ahgidx; 112 }; 113 114 struct hfi1_sdma_header { 115 __le64 pbc; 116 struct ib_header hdr; 117 } __packed; 118 119 /* 120 * hfi1 specific data structures that will be hidden from rvt after the queue 121 * pair is made common 122 */ 123 struct hfi1_qp_priv { 124 struct hfi1_ahg_info *s_ahg; /* ahg info for next header */ 125 struct sdma_engine *s_sde; /* current sde */ 126 struct send_context *s_sendcontext; /* current sendcontext */ 127 u8 s_sc; /* SC[0..4] for next packet */ 128 u8 r_adefered; /* number of acks defered */ 129 struct iowait s_iowait; 130 struct timer_list s_rnr_timer; 131 struct rvt_qp *owner; 132 }; 133 134 /* 135 * This structure is used to hold commonly lookedup and computed values during 136 * the send engine progress. 137 */ 138 struct hfi1_pkt_state { 139 struct hfi1_ibdev *dev; 140 struct hfi1_ibport *ibp; 141 struct hfi1_pportdata *ppd; 142 struct verbs_txreq *s_txreq; 143 unsigned long flags; 144 }; 145 146 #define HFI1_PSN_CREDIT 16 147 148 struct hfi1_opcode_stats { 149 u64 n_packets; /* number of packets */ 150 u64 n_bytes; /* total number of bytes */ 151 }; 152 153 struct hfi1_opcode_stats_perctx { 154 struct hfi1_opcode_stats stats[256]; 155 }; 156 157 static inline void inc_opstats( 158 u32 tlen, 159 struct hfi1_opcode_stats *stats) 160 { 161 #ifdef CONFIG_DEBUG_FS 162 stats->n_bytes += tlen; 163 stats->n_packets++; 164 #endif 165 } 166 167 struct hfi1_ibport { 168 struct rvt_qp __rcu *qp[2]; 169 struct rvt_ibport rvp; 170 171 /* the first 16 entries are sl_to_vl for !OPA */ 172 u8 sl_to_sc[32]; 173 u8 sc_to_sl[32]; 174 }; 175 176 struct hfi1_ibdev { 177 struct rvt_dev_info rdi; /* Must be first */ 178 179 /* QP numbers are shared by all IB ports */ 180 /* protect txwait list */ 181 seqlock_t txwait_lock ____cacheline_aligned_in_smp; 182 struct list_head txwait; /* list for wait verbs_txreq */ 183 struct list_head memwait; /* list for wait kernel memory */ 184 struct kmem_cache *verbs_txreq_cache; 185 u64 n_txwait; 186 u64 n_kmem_wait; 187 188 /* protect iowait lists */ 189 seqlock_t iowait_lock ____cacheline_aligned_in_smp; 190 u64 n_piowait; 191 u64 n_piodrain; 192 struct timer_list mem_timer; 193 194 #ifdef CONFIG_DEBUG_FS 195 /* per HFI debugfs */ 196 struct dentry *hfi1_ibdev_dbg; 197 /* per HFI symlinks to above */ 198 struct dentry *hfi1_ibdev_link; 199 #endif 200 }; 201 202 static inline struct hfi1_ibdev *to_idev(struct ib_device *ibdev) 203 { 204 struct rvt_dev_info *rdi; 205 206 rdi = container_of(ibdev, struct rvt_dev_info, ibdev); 207 return container_of(rdi, struct hfi1_ibdev, rdi); 208 } 209 210 static inline struct rvt_qp *iowait_to_qp(struct iowait *s_iowait) 211 { 212 struct hfi1_qp_priv *priv; 213 214 priv = container_of(s_iowait, struct hfi1_qp_priv, s_iowait); 215 return priv->owner; 216 } 217 218 /* 219 * Send if not busy or waiting for I/O and either 220 * a RC response is pending or we can process send work requests. 221 */ 222 static inline int hfi1_send_ok(struct rvt_qp *qp) 223 { 224 return !(qp->s_flags & (RVT_S_BUSY | RVT_S_ANY_WAIT_IO)) && 225 (qp->s_hdrwords || (qp->s_flags & RVT_S_RESP_PENDING) || 226 !(qp->s_flags & RVT_S_ANY_WAIT_SEND)); 227 } 228 229 /* 230 * This must be called with s_lock held. 231 */ 232 void hfi1_bad_pqkey(struct hfi1_ibport *ibp, __be16 trap_num, u32 key, u32 sl, 233 u32 qp1, u32 qp2, u16 lid1, u16 lid2); 234 void hfi1_cap_mask_chg(struct rvt_dev_info *rdi, u8 port_num); 235 void hfi1_sys_guid_chg(struct hfi1_ibport *ibp); 236 void hfi1_node_desc_chg(struct hfi1_ibport *ibp); 237 int hfi1_process_mad(struct ib_device *ibdev, int mad_flags, u8 port, 238 const struct ib_wc *in_wc, const struct ib_grh *in_grh, 239 const struct ib_mad_hdr *in_mad, size_t in_mad_size, 240 struct ib_mad_hdr *out_mad, size_t *out_mad_size, 241 u16 *out_mad_pkey_index); 242 243 /* 244 * The PSN_MASK and PSN_SHIFT allow for 245 * 1) comparing two PSNs 246 * 2) returning the PSN with any upper bits masked 247 * 3) returning the difference between to PSNs 248 * 249 * The number of significant bits in the PSN must 250 * necessarily be at least one bit less than 251 * the container holding the PSN. 252 */ 253 #ifndef CONFIG_HFI1_VERBS_31BIT_PSN 254 #define PSN_MASK 0xFFFFFF 255 #define PSN_SHIFT 8 256 #else 257 #define PSN_MASK 0x7FFFFFFF 258 #define PSN_SHIFT 1 259 #endif 260 #define PSN_MODIFY_MASK 0xFFFFFF 261 262 /* 263 * Compare the lower 24 bits of the msn values. 264 * Returns an integer <, ==, or > than zero. 265 */ 266 static inline int cmp_msn(u32 a, u32 b) 267 { 268 return (((int)a) - ((int)b)) << 8; 269 } 270 271 /* 272 * Compare two PSNs 273 * Returns an integer <, ==, or > than zero. 274 */ 275 static inline int cmp_psn(u32 a, u32 b) 276 { 277 return (((int)a) - ((int)b)) << PSN_SHIFT; 278 } 279 280 /* 281 * Return masked PSN 282 */ 283 static inline u32 mask_psn(u32 a) 284 { 285 return a & PSN_MASK; 286 } 287 288 /* 289 * Return delta between two PSNs 290 */ 291 static inline u32 delta_psn(u32 a, u32 b) 292 { 293 return (((int)a - (int)b) << PSN_SHIFT) >> PSN_SHIFT; 294 } 295 296 struct verbs_txreq; 297 void hfi1_put_txreq(struct verbs_txreq *tx); 298 299 int hfi1_verbs_send(struct rvt_qp *qp, struct hfi1_pkt_state *ps); 300 301 void hfi1_copy_sge(struct rvt_sge_state *ss, void *data, u32 length, 302 int release, int copy_last); 303 304 void hfi1_skip_sge(struct rvt_sge_state *ss, u32 length, int release); 305 306 void hfi1_cnp_rcv(struct hfi1_packet *packet); 307 308 void hfi1_uc_rcv(struct hfi1_packet *packet); 309 310 void hfi1_rc_rcv(struct hfi1_packet *packet); 311 312 void hfi1_rc_hdrerr( 313 struct hfi1_ctxtdata *rcd, 314 struct ib_header *hdr, 315 u32 rcv_flags, 316 struct rvt_qp *qp); 317 318 u8 ah_to_sc(struct ib_device *ibdev, struct ib_ah_attr *ah_attr); 319 320 struct ib_ah *hfi1_create_qp0_ah(struct hfi1_ibport *ibp, u16 dlid); 321 322 void hfi1_rc_rnr_retry(unsigned long arg); 323 void hfi1_add_rnr_timer(struct rvt_qp *qp, u32 to); 324 void hfi1_rc_timeout(unsigned long arg); 325 void hfi1_del_timers_sync(struct rvt_qp *qp); 326 void hfi1_stop_rc_timers(struct rvt_qp *qp); 327 328 void hfi1_rc_send_complete(struct rvt_qp *qp, struct ib_header *hdr); 329 330 void hfi1_rc_error(struct rvt_qp *qp, enum ib_wc_status err); 331 332 void hfi1_ud_rcv(struct hfi1_packet *packet); 333 334 int hfi1_lookup_pkey_idx(struct hfi1_ibport *ibp, u16 pkey); 335 336 int hfi1_rvt_get_rwqe(struct rvt_qp *qp, int wr_id_only); 337 338 void hfi1_migrate_qp(struct rvt_qp *qp); 339 340 int hfi1_check_modify_qp(struct rvt_qp *qp, struct ib_qp_attr *attr, 341 int attr_mask, struct ib_udata *udata); 342 343 void hfi1_modify_qp(struct rvt_qp *qp, struct ib_qp_attr *attr, 344 int attr_mask, struct ib_udata *udata); 345 346 int hfi1_check_send_wqe(struct rvt_qp *qp, struct rvt_swqe *wqe); 347 348 extern const u32 rc_only_opcode; 349 extern const u32 uc_only_opcode; 350 351 static inline u8 get_opcode(struct ib_header *h) 352 { 353 u16 lnh = be16_to_cpu(h->lrh[0]) & 3; 354 355 if (lnh == IB_LNH_IBA_LOCAL) 356 return be32_to_cpu(h->u.oth.bth[0]) >> 24; 357 else 358 return be32_to_cpu(h->u.l.oth.bth[0]) >> 24; 359 } 360 361 int hfi1_ruc_check_hdr(struct hfi1_ibport *ibp, struct ib_header *hdr, 362 int has_grh, struct rvt_qp *qp, u32 bth0); 363 364 u32 hfi1_make_grh(struct hfi1_ibport *ibp, struct ib_grh *hdr, 365 struct ib_global_route *grh, u32 hwords, u32 nwords); 366 367 void hfi1_make_ruc_header(struct rvt_qp *qp, struct ib_other_headers *ohdr, 368 u32 bth0, u32 bth2, int middle, 369 struct hfi1_pkt_state *ps); 370 371 void _hfi1_do_send(struct work_struct *work); 372 373 void hfi1_do_send(struct rvt_qp *qp); 374 375 void hfi1_send_complete(struct rvt_qp *qp, struct rvt_swqe *wqe, 376 enum ib_wc_status status); 377 378 void hfi1_send_rc_ack(struct hfi1_ctxtdata *, struct rvt_qp *qp, int is_fecn); 379 380 int hfi1_make_rc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps); 381 382 int hfi1_make_uc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps); 383 384 int hfi1_make_ud_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps); 385 386 int hfi1_register_ib_device(struct hfi1_devdata *); 387 388 void hfi1_unregister_ib_device(struct hfi1_devdata *); 389 390 void hfi1_ib_rcv(struct hfi1_packet *packet); 391 392 unsigned hfi1_get_npkeys(struct hfi1_devdata *); 393 394 int hfi1_verbs_send_dma(struct rvt_qp *qp, struct hfi1_pkt_state *ps, 395 u64 pbc); 396 397 int hfi1_verbs_send_pio(struct rvt_qp *qp, struct hfi1_pkt_state *ps, 398 u64 pbc); 399 400 int hfi1_wss_init(void); 401 void hfi1_wss_exit(void); 402 403 /* platform specific: return the lowest level cache (llc) size, in KiB */ 404 static inline int wss_llc_size(void) 405 { 406 /* assume that the boot CPU value is universal for all CPUs */ 407 return boot_cpu_data.x86_cache_size; 408 } 409 410 /* platform specific: cacheless copy */ 411 static inline void cacheless_memcpy(void *dst, void *src, size_t n) 412 { 413 /* 414 * Use the only available X64 cacheless copy. Add a __user cast 415 * to quiet sparse. The src agument is already in the kernel so 416 * there are no security issues. The extra fault recovery machinery 417 * is not invoked. 418 */ 419 __copy_user_nocache(dst, (void __user *)src, n, 0); 420 } 421 422 extern const enum ib_wc_opcode ib_hfi1_wc_opcode[]; 423 424 extern const u8 hdr_len_by_opcode[]; 425 426 extern const int ib_rvt_state_ops[]; 427 428 extern __be64 ib_hfi1_sys_image_guid; /* in network order */ 429 430 extern unsigned int hfi1_max_cqes; 431 432 extern unsigned int hfi1_max_cqs; 433 434 extern unsigned int hfi1_max_qp_wrs; 435 436 extern unsigned int hfi1_max_qps; 437 438 extern unsigned int hfi1_max_sges; 439 440 extern unsigned int hfi1_max_mcast_grps; 441 442 extern unsigned int hfi1_max_mcast_qp_attached; 443 444 extern unsigned int hfi1_max_srqs; 445 446 extern unsigned int hfi1_max_srq_sges; 447 448 extern unsigned int hfi1_max_srq_wrs; 449 450 extern unsigned short piothreshold; 451 452 extern const u32 ib_hfi1_rnr_table[]; 453 454 #endif /* HFI1_VERBS_H */ 455