1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2012, 2015 Chelsio Communications, Inc. 5 * All rights reserved. 6 * Written by: Navdeep Parhar <np@FreeBSD.org> 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 */ 30 31 #ifndef __T4_TOM_H__ 32 #define __T4_TOM_H__ 33 #include <sys/vmem.h> 34 #include "common/t4_hw.h" 35 #include "common/t4_msg.h" 36 #include "tom/t4_tls.h" 37 38 #define LISTEN_HASH_SIZE 32 39 40 /* 41 * Min receive window. We want it to be large enough to accommodate receive 42 * coalescing, handle jumbo frames, and not trigger sender SWS avoidance. 43 */ 44 #define MIN_RCV_WND (24 * 1024U) 45 46 /* 47 * Max receive window supported by HW in bytes. Only a small part of it can 48 * be set through option0, the rest needs to be set through RX_DATA_ACK. 49 */ 50 #define MAX_RCV_WND ((1U << 27) - 1) 51 52 #define DDP_RSVD_WIN (16 * 1024U) 53 #define SB_DDP_INDICATE SB_IN_TOE /* soreceive must respond to indicate */ 54 55 #define USE_DDP_RX_FLOW_CONTROL 56 57 #define PPOD_SZ(n) ((n) * sizeof(struct pagepod)) 58 #define PPOD_SIZE (PPOD_SZ(1)) 59 60 /* TOE PCB flags */ 61 enum { 62 TPF_ATTACHED = (1 << 0), /* a tcpcb refers to this toepcb */ 63 TPF_FLOWC_WR_SENT = (1 << 1), /* firmware flow context WR sent */ 64 TPF_TX_DATA_SENT = (1 << 2), /* some data sent */ 65 TPF_TX_SUSPENDED = (1 << 3), /* tx suspended for lack of resources */ 66 TPF_SEND_FIN = (1 << 4), /* send FIN after all pending data */ 67 TPF_FIN_SENT = (1 << 5), /* FIN has been sent */ 68 TPF_ABORT_SHUTDOWN = (1 << 6), /* connection abort is in progress */ 69 TPF_CPL_PENDING = (1 << 7), /* haven't received the last CPL */ 70 TPF_SYNQE = (1 << 8), /* synq_entry, not really a toepcb */ 71 TPF_SYNQE_EXPANDED = (1 << 9), /* toepcb ready, tid context updated */ 72 TPF_TLS_STARTING = (1 << 10), /* starting TLS receive */ 73 TPF_KTLS = (1 << 11), /* send TLS records from KTLS */ 74 TPF_INITIALIZED = (1 << 12), /* init_toepcb has been called */ 75 TPF_TLS_RECEIVE = (1 << 13), /* should receive TLS records */ 76 TPF_TLS_RX_QUIESCING = (1 << 14), /* RX quiesced for TLS RX startup */ 77 TPF_TLS_RX_QUIESCED = (1 << 15), /* RX quiesced for TLS RX startup */ 78 TPF_WAITING_FOR_FINAL = (1<< 16), /* waiting for wakeup on final CPL */ 79 }; 80 81 enum { 82 DDP_OK = (1 << 0), /* OK to turn on DDP */ 83 DDP_SC_REQ = (1 << 1), /* state change (on/off) requested */ 84 DDP_ON = (1 << 2), /* DDP is turned on */ 85 DDP_BUF0_ACTIVE = (1 << 3), /* buffer 0 in use (not invalidated) */ 86 DDP_BUF1_ACTIVE = (1 << 4), /* buffer 1 in use (not invalidated) */ 87 DDP_TASK_ACTIVE = (1 << 5), /* requeue task is queued / running */ 88 DDP_DEAD = (1 << 6), /* toepcb is shutting down */ 89 DDP_AIO = (1 << 7), /* DDP used for AIO, not so_rcv */ 90 DDP_RCVBUF = (1 << 8), /* DDP used for so_rcv, not AIO */ 91 }; 92 93 struct bio; 94 struct ctl_sg_entry; 95 struct sockopt; 96 struct offload_settings; 97 98 /* 99 * Connection parameters for an offloaded connection. These are mostly (but not 100 * all) hardware TOE parameters. 101 */ 102 struct conn_params { 103 int8_t rx_coalesce; 104 int8_t cong_algo; 105 int8_t tc_idx; 106 int8_t tstamp; 107 int8_t sack; 108 int8_t nagle; 109 int8_t keepalive; 110 int8_t wscale; 111 int8_t ecn; 112 int8_t mtu_idx; 113 int8_t ulp_mode; 114 int8_t tx_align; 115 int16_t txq_idx; /* ofld_txq = &sc->sge.ofld_txq[txq_idx] */ 116 int16_t rxq_idx; /* ofld_rxq = &sc->sge.ofld_rxq[rxq_idx] */ 117 int16_t l2t_idx; 118 uint16_t emss; 119 uint16_t opt0_bufsize; 120 u_int sndbuf; /* controls TP tx pages */ 121 }; 122 123 struct ofld_tx_sdesc { 124 uint32_t plen; /* payload length */ 125 uint8_t tx_credits; /* firmware tx credits (unit is 16B) */ 126 }; 127 128 struct ppod_region { 129 u_int pr_start; 130 u_int pr_len; 131 u_int pr_page_shift[4]; 132 uint32_t pr_tag_mask; /* hardware tagmask for this region. */ 133 uint32_t pr_invalid_bit; /* OR with this to invalidate tag. */ 134 uint32_t pr_alias_mask; /* AND with tag to get alias bits. */ 135 u_int pr_alias_shift; /* shift this much for first alias bit. */ 136 vmem_t *pr_arena; 137 }; 138 139 struct ppod_reservation { 140 struct ppod_region *prsv_pr; 141 uint32_t prsv_tag; /* Full tag: pgsz, alias, tag, color */ 142 u_int prsv_nppods; 143 }; 144 145 struct pageset { 146 TAILQ_ENTRY(pageset) link; 147 vm_page_t *pages; 148 int npages; 149 int flags; 150 int offset; /* offset in first page */ 151 int len; 152 struct ppod_reservation prsv; 153 struct vmspace *vm; 154 vm_offset_t start; 155 u_int vm_timestamp; 156 }; 157 158 TAILQ_HEAD(pagesetq, pageset); 159 160 #define PS_PPODS_WRITTEN 0x0001 /* Page pods written to the card. */ 161 162 struct ddp_rcv_buffer { 163 TAILQ_ENTRY(ddp_rcv_buffer) link; 164 void *buf; 165 struct ppod_reservation prsv; 166 size_t len; 167 u_int refs; 168 }; 169 170 struct ddp_buffer { 171 union { 172 /* DDP_AIO fields */ 173 struct { 174 struct pageset *ps; 175 struct kaiocb *job; 176 int cancel_pending; 177 }; 178 179 /* DDP_RCVBUF fields */ 180 struct { 181 struct ddp_rcv_buffer *drb; 182 uint32_t placed; 183 }; 184 }; 185 }; 186 187 /* 188 * (a) - DDP_AIO only 189 * (r) - DDP_RCVBUF only 190 */ 191 struct ddp_pcb { 192 struct mtx lock; 193 u_int flags; 194 int active_id; /* the currently active DDP buffer */ 195 struct ddp_buffer db[2]; 196 union { 197 TAILQ_HEAD(, pageset) cached_pagesets; /* (a) */ 198 TAILQ_HEAD(, ddp_rcv_buffer) cached_buffers; /* (r) */ 199 }; 200 TAILQ_HEAD(, kaiocb) aiojobq; /* (a) */ 201 u_int waiting_count; /* (a) */ 202 u_int active_count; 203 u_int cached_count; 204 struct task requeue_task; 205 struct kaiocb *queueing; /* (a) */ 206 struct mtx cache_lock; /* (r) */ 207 }; 208 209 struct toepcb { 210 struct tom_data *td; 211 struct inpcb *inp; /* backpointer to host stack's PCB */ 212 u_int flags; /* miscellaneous flags */ 213 TAILQ_ENTRY(toepcb) link; /* toep_list */ 214 int refcount; 215 struct vnet *vnet; 216 struct vi_info *vi; /* virtual interface */ 217 struct sge_ofld_txq *ofld_txq; 218 struct sge_ofld_rxq *ofld_rxq; 219 struct sge_wrq *ctrlq; 220 struct l2t_entry *l2te; /* L2 table entry used by this connection */ 221 struct clip_entry *ce; /* CLIP table entry used by this tid */ 222 int tid; /* Connection identifier */ 223 224 /* tx credit handling */ 225 u_int tx_total; /* total tx WR credits (in 16B units) */ 226 u_int tx_credits; /* tx WR credits (in 16B units) available */ 227 u_int tx_nocompl; /* tx WR credits since last compl request */ 228 u_int plen_nocompl; /* payload since last compl request */ 229 230 struct conn_params params; 231 232 void *ulpcb; 233 void *ulpcb2; 234 struct mbufq ulp_pduq; /* PDUs waiting to be sent out. */ 235 struct mbufq ulp_pdu_reclaimq; 236 237 struct ddp_pcb ddp; 238 struct tls_ofld_info tls; 239 240 TAILQ_HEAD(, kaiocb) aiotx_jobq; 241 struct task aiotx_task; 242 struct socket *aiotx_so; 243 244 /* Tx software descriptor */ 245 uint8_t txsd_total; 246 uint8_t txsd_pidx; 247 uint8_t txsd_cidx; 248 uint8_t txsd_avail; 249 struct ofld_tx_sdesc txsd[]; 250 }; 251 252 static inline int 253 ulp_mode(struct toepcb *toep) 254 { 255 256 return (toep->params.ulp_mode); 257 } 258 259 #define DDP_LOCK(toep) mtx_lock(&(toep)->ddp.lock) 260 #define DDP_UNLOCK(toep) mtx_unlock(&(toep)->ddp.lock) 261 #define DDP_ASSERT_LOCKED(toep) mtx_assert(&(toep)->ddp.lock, MA_OWNED) 262 #define DDP_CACHE_LOCK(toep) mtx_lock(&(toep)->ddp.cache_lock) 263 #define DDP_CACHE_UNLOCK(toep) mtx_unlock(&(toep)->ddp.cache_lock) 264 265 /* 266 * Compressed state for embryonic connections for a listener. 267 */ 268 struct synq_entry { 269 struct listen_ctx *lctx; /* backpointer to listen ctx */ 270 struct mbuf *syn; 271 int flags; /* same as toepcb's tp_flags */ 272 volatile int ok_to_respond; 273 volatile u_int refcnt; 274 int tid; 275 uint32_t iss; 276 uint32_t irs; 277 uint32_t ts; 278 uint32_t rss_hash; 279 __be16 tcp_opt; /* from cpl_pass_establish */ 280 struct toepcb *toep; 281 282 struct conn_params params; 283 }; 284 285 /* listen_ctx flags */ 286 #define LCTX_RPL_PENDING 1 /* waiting for a CPL_PASS_OPEN_RPL */ 287 288 struct listen_ctx { 289 LIST_ENTRY(listen_ctx) link; /* listen hash linkage */ 290 volatile int refcount; 291 int stid; 292 struct stid_region stid_region; 293 int flags; 294 struct inpcb *inp; /* listening socket's inp */ 295 struct vnet *vnet; 296 struct sge_wrq *ctrlq; 297 struct sge_ofld_rxq *ofld_rxq; 298 struct clip_entry *ce; 299 }; 300 301 /* tcb_histent flags */ 302 #define TE_RPL_PENDING 1 303 #define TE_ACTIVE 2 304 305 /* bits in one 8b tcb_histent sample. */ 306 #define TS_RTO (1 << 0) 307 #define TS_DUPACKS (1 << 1) 308 #define TS_FASTREXMT (1 << 2) 309 #define TS_SND_BACKLOGGED (1 << 3) 310 #define TS_CWND_LIMITED (1 << 4) 311 #define TS_ECN_ECE (1 << 5) 312 #define TS_ECN_CWR (1 << 6) 313 #define TS_RESERVED (1 << 7) /* Unused. */ 314 315 struct tcb_histent { 316 struct mtx te_lock; 317 struct callout te_callout; 318 uint64_t te_tcb[TCB_SIZE / sizeof(uint64_t)]; 319 struct adapter *te_adapter; 320 u_int te_flags; 321 u_int te_tid; 322 uint8_t te_pidx; 323 uint8_t te_sample[100]; 324 }; 325 326 struct tom_data { 327 struct toedev tod; 328 329 /* toepcb's associated with this TOE device */ 330 struct mtx toep_list_lock; 331 TAILQ_HEAD(, toepcb) toep_list; 332 333 struct mtx lctx_hash_lock; 334 LIST_HEAD(, listen_ctx) *listen_hash; 335 u_long listen_mask; 336 int lctx_count; /* # of lctx in the hash table */ 337 338 struct ppod_region pr; 339 340 struct rwlock tcb_history_lock __aligned(CACHE_LINE_SIZE); 341 struct tcb_histent **tcb_history; 342 int dupack_threshold; 343 344 /* WRs that will not be sent to the chip because L2 resolution failed */ 345 struct mtx unsent_wr_lock; 346 STAILQ_HEAD(, wrqe) unsent_wr_list; 347 struct task reclaim_wr_resources; 348 }; 349 350 static inline struct tom_data * 351 tod_td(struct toedev *tod) 352 { 353 354 return (__containerof(tod, struct tom_data, tod)); 355 } 356 357 static inline struct adapter * 358 td_adapter(struct tom_data *td) 359 { 360 361 return (td->tod.tod_softc); 362 } 363 364 static inline void 365 set_mbuf_raw_wr(struct mbuf *m, bool raw) 366 { 367 368 M_ASSERTPKTHDR(m); 369 m->m_pkthdr.PH_per.eight[6] = raw; 370 } 371 372 static inline bool 373 mbuf_raw_wr(struct mbuf *m) 374 { 375 376 M_ASSERTPKTHDR(m); 377 return (m->m_pkthdr.PH_per.eight[6]); 378 } 379 380 static inline void 381 set_mbuf_ulp_submode(struct mbuf *m, uint8_t ulp_submode) 382 { 383 384 M_ASSERTPKTHDR(m); 385 m->m_pkthdr.PH_per.eight[0] = ulp_submode; 386 } 387 388 static inline uint8_t 389 mbuf_ulp_submode(struct mbuf *m) 390 { 391 392 M_ASSERTPKTHDR(m); 393 return (m->m_pkthdr.PH_per.eight[0]); 394 } 395 396 static inline void 397 set_mbuf_iscsi_iso(struct mbuf *m, bool iso) 398 { 399 400 M_ASSERTPKTHDR(m); 401 m->m_pkthdr.PH_per.eight[1] = iso; 402 } 403 404 static inline bool 405 mbuf_iscsi_iso(struct mbuf *m) 406 { 407 408 M_ASSERTPKTHDR(m); 409 return (m->m_pkthdr.PH_per.eight[1]); 410 } 411 412 /* Flags for iSCSI segmentation offload. */ 413 #define CXGBE_ISO_TYPE(flags) ((flags) & 0x3) 414 #define CXGBE_ISO_F 0x4 415 416 static inline void 417 set_mbuf_iscsi_iso_flags(struct mbuf *m, uint8_t flags) 418 { 419 420 M_ASSERTPKTHDR(m); 421 m->m_pkthdr.PH_per.eight[2] = flags; 422 } 423 424 static inline uint8_t 425 mbuf_iscsi_iso_flags(struct mbuf *m) 426 { 427 428 M_ASSERTPKTHDR(m); 429 return (m->m_pkthdr.PH_per.eight[2]); 430 } 431 432 static inline void 433 set_mbuf_iscsi_iso_mss(struct mbuf *m, uint16_t mss) 434 { 435 436 M_ASSERTPKTHDR(m); 437 m->m_pkthdr.PH_per.sixteen[2] = mss; 438 } 439 440 static inline uint16_t 441 mbuf_iscsi_iso_mss(struct mbuf *m) 442 { 443 444 M_ASSERTPKTHDR(m); 445 return (m->m_pkthdr.PH_per.sixteen[2]); 446 } 447 448 /* t4_tom.c */ 449 struct toepcb *alloc_toepcb(struct vi_info *, int); 450 int init_toepcb(struct vi_info *, struct toepcb *); 451 struct toepcb *hold_toepcb(struct toepcb *); 452 void free_toepcb(struct toepcb *); 453 void offload_socket(struct socket *, struct toepcb *); 454 void restore_so_proto(struct socket *, bool); 455 void undo_offload_socket(struct socket *); 456 void final_cpl_received(struct toepcb *); 457 void insert_tid(struct adapter *, int, void *, int); 458 void *lookup_tid(struct adapter *, int); 459 void update_tid(struct adapter *, int, void *); 460 void remove_tid(struct adapter *, int, int); 461 u_long select_rcv_wnd(struct socket *); 462 int select_rcv_wscale(void); 463 void init_conn_params(struct vi_info *, struct offload_settings *, 464 struct in_conninfo *, struct socket *, const struct tcp_options *, int16_t, 465 struct conn_params *cp); 466 __be64 calc_options0(struct vi_info *, struct conn_params *); 467 __be32 calc_options2(struct vi_info *, struct conn_params *); 468 uint64_t select_ntuple(struct vi_info *, struct l2t_entry *); 469 int negative_advice(int); 470 int add_tid_to_history(struct adapter *, u_int); 471 472 /* t4_connect.c */ 473 void t4_init_connect_cpl_handlers(void); 474 void t4_uninit_connect_cpl_handlers(void); 475 int t4_connect(struct toedev *, struct socket *, struct nhop_object *, 476 struct sockaddr *); 477 void act_open_failure_cleanup(struct adapter *, u_int, u_int); 478 479 /* t4_listen.c */ 480 void t4_init_listen_cpl_handlers(void); 481 void t4_uninit_listen_cpl_handlers(void); 482 int t4_listen_start(struct toedev *, struct tcpcb *); 483 int t4_listen_stop(struct toedev *, struct tcpcb *); 484 void t4_syncache_added(struct toedev *, void *); 485 void t4_syncache_removed(struct toedev *, void *); 486 int t4_syncache_respond(struct toedev *, void *, struct mbuf *); 487 int do_abort_req_synqe(struct sge_iq *, const struct rss_header *, 488 struct mbuf *); 489 int do_abort_rpl_synqe(struct sge_iq *, const struct rss_header *, 490 struct mbuf *); 491 void t4_offload_socket(struct toedev *, void *, struct socket *); 492 void synack_failure_cleanup(struct adapter *, int); 493 494 /* t4_cpl_io.c */ 495 void aiotx_init_toep(struct toepcb *); 496 int t4_aio_queue_aiotx(struct socket *, struct kaiocb *); 497 void t4_init_cpl_io_handlers(void); 498 void t4_uninit_cpl_io_handlers(void); 499 void send_abort_rpl(struct adapter *, struct sge_ofld_txq *, int , int); 500 void send_flowc_wr(struct toepcb *, struct tcpcb *); 501 void send_reset(struct adapter *, struct toepcb *, uint32_t); 502 int send_rx_credits(struct adapter *, struct toepcb *, int); 503 void make_established(struct toepcb *, uint32_t, uint32_t, uint16_t); 504 int t4_close_conn(struct adapter *, struct toepcb *); 505 void t4_rcvd(struct toedev *, struct tcpcb *); 506 void t4_rcvd_locked(struct toedev *, struct tcpcb *); 507 int t4_tod_output(struct toedev *, struct tcpcb *); 508 int t4_send_fin(struct toedev *, struct tcpcb *); 509 int t4_send_rst(struct toedev *, struct tcpcb *); 510 void t4_set_tcb_field(struct adapter *, struct sge_wrq *, struct toepcb *, 511 uint16_t, uint64_t, uint64_t, int, int); 512 void t4_push_frames(struct adapter *, struct toepcb *, int); 513 void t4_push_pdus(struct adapter *, struct toepcb *, int); 514 515 /* t4_ddp.c */ 516 int t4_init_ppod_region(struct ppod_region *, struct t4_range *, u_int, 517 const char *); 518 void t4_free_ppod_region(struct ppod_region *); 519 int t4_alloc_page_pods_for_ps(struct ppod_region *, struct pageset *); 520 int t4_alloc_page_pods_for_bio(struct ppod_region *, struct bio *, 521 struct ppod_reservation *); 522 int t4_alloc_page_pods_for_buf(struct ppod_region *, vm_offset_t, int, 523 struct ppod_reservation *); 524 int t4_alloc_page_pods_for_sgl(struct ppod_region *, struct ctl_sg_entry *, int, 525 struct ppod_reservation *); 526 int t4_write_page_pods_for_ps(struct adapter *, struct sge_wrq *, int, 527 struct pageset *); 528 int t4_write_page_pods_for_bio(struct adapter *, struct toepcb *, 529 struct ppod_reservation *, struct bio *, struct mbufq *); 530 int t4_write_page_pods_for_buf(struct adapter *, struct toepcb *, 531 struct ppod_reservation *, vm_offset_t, int, struct mbufq *); 532 int t4_write_page_pods_for_sgl(struct adapter *, struct toepcb *, 533 struct ppod_reservation *, struct ctl_sg_entry *, int, int, struct mbufq *); 534 void t4_free_page_pods(struct ppod_reservation *); 535 int t4_aio_queue_ddp(struct socket *, struct kaiocb *); 536 int t4_enable_ddp_rcv(struct socket *, struct toepcb *); 537 void t4_ddp_mod_load(void); 538 void t4_ddp_mod_unload(void); 539 void ddp_assert_empty(struct toepcb *); 540 void ddp_uninit_toep(struct toepcb *); 541 void ddp_queue_toep(struct toepcb *); 542 void release_ddp_resources(struct toepcb *toep); 543 void handle_ddp_close(struct toepcb *, struct tcpcb *, uint32_t); 544 void handle_ddp_indicate(struct toepcb *); 545 void insert_ddp_data(struct toepcb *, uint32_t); 546 const struct offload_settings *lookup_offload_policy(struct adapter *, int, 547 struct mbuf *, uint16_t, struct inpcb *); 548 549 /* t4_tls.c */ 550 bool can_tls_offload(struct adapter *); 551 void do_rx_data_tls(const struct cpl_rx_data *, struct toepcb *, struct mbuf *); 552 void t4_push_ktls(struct adapter *, struct toepcb *, int); 553 void tls_received_starting_data(struct adapter *, struct toepcb *, 554 struct sockbuf *, int); 555 void t4_tls_mod_load(void); 556 void t4_tls_mod_unload(void); 557 void tls_init_toep(struct toepcb *); 558 int tls_tx_key(struct toepcb *); 559 void tls_uninit_toep(struct toepcb *); 560 int tls_alloc_ktls(struct toepcb *, struct ktls_session *, int); 561 562 #endif 563