1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 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 * $FreeBSD$ 30 * 31 */ 32 33 #ifndef __T4_TOM_H__ 34 #define __T4_TOM_H__ 35 #include <sys/vmem.h> 36 #include "common/t4_hw.h" 37 #include "common/t4_msg.h" 38 #include "tom/t4_tls.h" 39 40 #define LISTEN_HASH_SIZE 32 41 42 /* 43 * Min receive window. We want it to be large enough to accommodate receive 44 * coalescing, handle jumbo frames, and not trigger sender SWS avoidance. 45 */ 46 #define MIN_RCV_WND (24 * 1024U) 47 48 /* 49 * Max receive window supported by HW in bytes. Only a small part of it can 50 * be set through option0, the rest needs to be set through RX_DATA_ACK. 51 */ 52 #define MAX_RCV_WND ((1U << 27) - 1) 53 54 #define DDP_RSVD_WIN (16 * 1024U) 55 #define SB_DDP_INDICATE SB_IN_TOE /* soreceive must respond to indicate */ 56 57 #define USE_DDP_RX_FLOW_CONTROL 58 59 #define PPOD_SZ(n) ((n) * sizeof(struct pagepod)) 60 #define PPOD_SIZE (PPOD_SZ(1)) 61 62 /* TOE PCB flags */ 63 enum { 64 TPF_ATTACHED = (1 << 0), /* a tcpcb refers to this toepcb */ 65 TPF_FLOWC_WR_SENT = (1 << 1), /* firmware flow context WR sent */ 66 TPF_TX_DATA_SENT = (1 << 2), /* some data sent */ 67 TPF_TX_SUSPENDED = (1 << 3), /* tx suspended for lack of resources */ 68 TPF_SEND_FIN = (1 << 4), /* send FIN after all pending data */ 69 TPF_FIN_SENT = (1 << 5), /* FIN has been sent */ 70 TPF_ABORT_SHUTDOWN = (1 << 6), /* connection abort is in progress */ 71 TPF_CPL_PENDING = (1 << 7), /* haven't received the last CPL */ 72 TPF_SYNQE = (1 << 8), /* synq_entry, not really a toepcb */ 73 TPF_SYNQE_EXPANDED = (1 << 9), /* toepcb ready, tid context updated */ 74 TPF_TLS_STARTING = (1 << 10), /* starting TLS receive */ 75 TPF_KTLS = (1 << 11), /* send TLS records from KTLS */ 76 TPF_INITIALIZED = (1 << 12), /* init_toepcb has been called */ 77 TPF_TLS_RECEIVE = (1 << 13), /* should receive TLS records */ 78 TPF_TLS_RX_QUIESCED = (1 << 14), /* RX quiesced for TLS RX startup */ 79 TPF_WAITING_FOR_FINAL = (1<< 15), /* waiting for wakeup on final CPL */ 80 }; 81 82 enum { 83 DDP_OK = (1 << 0), /* OK to turn on DDP */ 84 DDP_SC_REQ = (1 << 1), /* state change (on/off) requested */ 85 DDP_ON = (1 << 2), /* DDP is turned on */ 86 DDP_BUF0_ACTIVE = (1 << 3), /* buffer 0 in use (not invalidated) */ 87 DDP_BUF1_ACTIVE = (1 << 4), /* buffer 1 in use (not invalidated) */ 88 DDP_TASK_ACTIVE = (1 << 5), /* requeue task is queued / running */ 89 DDP_DEAD = (1 << 6), /* toepcb is shutting down */ 90 }; 91 92 struct bio; 93 struct ctl_sg_entry; 94 struct sockopt; 95 struct offload_settings; 96 97 /* 98 * Connection parameters for an offloaded connection. These are mostly (but not 99 * all) hardware TOE parameters. 100 */ 101 struct conn_params { 102 int8_t rx_coalesce; 103 int8_t cong_algo; 104 int8_t tc_idx; 105 int8_t tstamp; 106 int8_t sack; 107 int8_t nagle; 108 int8_t keepalive; 109 int8_t wscale; 110 int8_t ecn; 111 int8_t mtu_idx; 112 int8_t ulp_mode; 113 int8_t tx_align; 114 int16_t txq_idx; /* ofld_txq = &sc->sge.ofld_txq[txq_idx] */ 115 int16_t rxq_idx; /* ofld_rxq = &sc->sge.ofld_rxq[rxq_idx] */ 116 int16_t l2t_idx; 117 uint16_t emss; 118 uint16_t opt0_bufsize; 119 u_int sndbuf; /* controls TP tx pages */ 120 }; 121 122 struct ofld_tx_sdesc { 123 uint32_t plen; /* payload length */ 124 uint8_t tx_credits; /* firmware tx credits (unit is 16B) */ 125 }; 126 127 struct ppod_region { 128 u_int pr_start; 129 u_int pr_len; 130 u_int pr_page_shift[4]; 131 uint32_t pr_tag_mask; /* hardware tagmask for this region. */ 132 uint32_t pr_invalid_bit; /* OR with this to invalidate tag. */ 133 uint32_t pr_alias_mask; /* AND with tag to get alias bits. */ 134 u_int pr_alias_shift; /* shift this much for first alias bit. */ 135 vmem_t *pr_arena; 136 }; 137 138 struct ppod_reservation { 139 struct ppod_region *prsv_pr; 140 uint32_t prsv_tag; /* Full tag: pgsz, alias, tag, color */ 141 u_int prsv_nppods; 142 }; 143 144 struct pageset { 145 TAILQ_ENTRY(pageset) link; 146 vm_page_t *pages; 147 int npages; 148 int flags; 149 int offset; /* offset in first page */ 150 int len; 151 struct ppod_reservation prsv; 152 struct vmspace *vm; 153 vm_offset_t start; 154 u_int vm_timestamp; 155 }; 156 157 TAILQ_HEAD(pagesetq, pageset); 158 159 #define PS_PPODS_WRITTEN 0x0001 /* Page pods written to the card. */ 160 161 struct ddp_buffer { 162 struct pageset *ps; 163 164 struct kaiocb *job; 165 int cancel_pending; 166 }; 167 168 struct ddp_pcb { 169 u_int flags; 170 struct ddp_buffer db[2]; 171 TAILQ_HEAD(, pageset) cached_pagesets; 172 TAILQ_HEAD(, kaiocb) aiojobq; 173 u_int waiting_count; 174 u_int active_count; 175 u_int cached_count; 176 int active_id; /* the currently active DDP buffer */ 177 struct task requeue_task; 178 struct kaiocb *queueing; 179 struct mtx lock; 180 }; 181 182 struct toepcb { 183 struct tom_data *td; 184 struct inpcb *inp; /* backpointer to host stack's PCB */ 185 u_int flags; /* miscellaneous flags */ 186 TAILQ_ENTRY(toepcb) link; /* toep_list */ 187 int refcount; 188 struct vnet *vnet; 189 struct vi_info *vi; /* virtual interface */ 190 struct sge_ofld_txq *ofld_txq; 191 struct sge_ofld_rxq *ofld_rxq; 192 struct sge_wrq *ctrlq; 193 struct l2t_entry *l2te; /* L2 table entry used by this connection */ 194 struct clip_entry *ce; /* CLIP table entry used by this tid */ 195 int tid; /* Connection identifier */ 196 197 /* tx credit handling */ 198 u_int tx_total; /* total tx WR credits (in 16B units) */ 199 u_int tx_credits; /* tx WR credits (in 16B units) available */ 200 u_int tx_nocompl; /* tx WR credits since last compl request */ 201 u_int plen_nocompl; /* payload since last compl request */ 202 203 struct conn_params params; 204 205 void *ulpcb; 206 void *ulpcb2; 207 struct mbufq ulp_pduq; /* PDUs waiting to be sent out. */ 208 struct mbufq ulp_pdu_reclaimq; 209 210 struct ddp_pcb ddp; 211 struct tls_ofld_info tls; 212 213 TAILQ_HEAD(, kaiocb) aiotx_jobq; 214 struct task aiotx_task; 215 struct socket *aiotx_so; 216 217 /* Tx software descriptor */ 218 uint8_t txsd_total; 219 uint8_t txsd_pidx; 220 uint8_t txsd_cidx; 221 uint8_t txsd_avail; 222 struct ofld_tx_sdesc txsd[]; 223 }; 224 225 static inline int 226 ulp_mode(struct toepcb *toep) 227 { 228 229 return (toep->params.ulp_mode); 230 } 231 232 #define DDP_LOCK(toep) mtx_lock(&(toep)->ddp.lock) 233 #define DDP_UNLOCK(toep) mtx_unlock(&(toep)->ddp.lock) 234 #define DDP_ASSERT_LOCKED(toep) mtx_assert(&(toep)->ddp.lock, MA_OWNED) 235 236 /* 237 * Compressed state for embryonic connections for a listener. 238 */ 239 struct synq_entry { 240 struct listen_ctx *lctx; /* backpointer to listen ctx */ 241 struct mbuf *syn; 242 int flags; /* same as toepcb's tp_flags */ 243 volatile int ok_to_respond; 244 volatile u_int refcnt; 245 int tid; 246 uint32_t iss; 247 uint32_t irs; 248 uint32_t ts; 249 uint32_t rss_hash; 250 __be16 tcp_opt; /* from cpl_pass_establish */ 251 struct toepcb *toep; 252 253 struct conn_params params; 254 }; 255 256 /* listen_ctx flags */ 257 #define LCTX_RPL_PENDING 1 /* waiting for a CPL_PASS_OPEN_RPL */ 258 259 struct listen_ctx { 260 LIST_ENTRY(listen_ctx) link; /* listen hash linkage */ 261 volatile int refcount; 262 int stid; 263 struct stid_region stid_region; 264 int flags; 265 struct inpcb *inp; /* listening socket's inp */ 266 struct vnet *vnet; 267 struct sge_wrq *ctrlq; 268 struct sge_ofld_rxq *ofld_rxq; 269 struct clip_entry *ce; 270 }; 271 272 /* tcb_histent flags */ 273 #define TE_RPL_PENDING 1 274 #define TE_ACTIVE 2 275 276 /* bits in one 8b tcb_histent sample. */ 277 #define TS_RTO (1 << 0) 278 #define TS_DUPACKS (1 << 1) 279 #define TS_FASTREXMT (1 << 2) 280 #define TS_SND_BACKLOGGED (1 << 3) 281 #define TS_CWND_LIMITED (1 << 4) 282 #define TS_ECN_ECE (1 << 5) 283 #define TS_ECN_CWR (1 << 6) 284 #define TS_RESERVED (1 << 7) /* Unused. */ 285 286 struct tcb_histent { 287 struct mtx te_lock; 288 struct callout te_callout; 289 uint64_t te_tcb[TCB_SIZE / sizeof(uint64_t)]; 290 struct adapter *te_adapter; 291 u_int te_flags; 292 u_int te_tid; 293 uint8_t te_pidx; 294 uint8_t te_sample[100]; 295 }; 296 297 struct tom_data { 298 struct toedev tod; 299 300 /* toepcb's associated with this TOE device */ 301 struct mtx toep_list_lock; 302 TAILQ_HEAD(, toepcb) toep_list; 303 304 struct mtx lctx_hash_lock; 305 LIST_HEAD(, listen_ctx) *listen_hash; 306 u_long listen_mask; 307 int lctx_count; /* # of lctx in the hash table */ 308 309 struct ppod_region pr; 310 311 struct rwlock tcb_history_lock __aligned(CACHE_LINE_SIZE); 312 struct tcb_histent **tcb_history; 313 int dupack_threshold; 314 315 /* WRs that will not be sent to the chip because L2 resolution failed */ 316 struct mtx unsent_wr_lock; 317 STAILQ_HEAD(, wrqe) unsent_wr_list; 318 struct task reclaim_wr_resources; 319 }; 320 321 static inline struct tom_data * 322 tod_td(struct toedev *tod) 323 { 324 325 return (__containerof(tod, struct tom_data, tod)); 326 } 327 328 static inline struct adapter * 329 td_adapter(struct tom_data *td) 330 { 331 332 return (td->tod.tod_softc); 333 } 334 335 static inline void 336 set_mbuf_raw_wr(struct mbuf *m, bool raw) 337 { 338 339 M_ASSERTPKTHDR(m); 340 m->m_pkthdr.PH_per.eight[6] = raw; 341 } 342 343 static inline bool 344 mbuf_raw_wr(struct mbuf *m) 345 { 346 347 M_ASSERTPKTHDR(m); 348 return (m->m_pkthdr.PH_per.eight[6]); 349 } 350 351 static inline void 352 set_mbuf_ulp_submode(struct mbuf *m, uint8_t ulp_submode) 353 { 354 355 M_ASSERTPKTHDR(m); 356 m->m_pkthdr.PH_per.eight[0] = ulp_submode; 357 } 358 359 static inline uint8_t 360 mbuf_ulp_submode(struct mbuf *m) 361 { 362 363 M_ASSERTPKTHDR(m); 364 return (m->m_pkthdr.PH_per.eight[0]); 365 } 366 367 static inline void 368 set_mbuf_iscsi_iso(struct mbuf *m, bool iso) 369 { 370 371 M_ASSERTPKTHDR(m); 372 m->m_pkthdr.PH_per.eight[1] = iso; 373 } 374 375 static inline bool 376 mbuf_iscsi_iso(struct mbuf *m) 377 { 378 379 M_ASSERTPKTHDR(m); 380 return (m->m_pkthdr.PH_per.eight[1]); 381 } 382 383 /* Flags for iSCSI segmentation offload. */ 384 #define CXGBE_ISO_TYPE(flags) ((flags) & 0x3) 385 #define CXGBE_ISO_F 0x4 386 387 static inline void 388 set_mbuf_iscsi_iso_flags(struct mbuf *m, uint8_t flags) 389 { 390 391 M_ASSERTPKTHDR(m); 392 m->m_pkthdr.PH_per.eight[2] = flags; 393 } 394 395 static inline uint8_t 396 mbuf_iscsi_iso_flags(struct mbuf *m) 397 { 398 399 M_ASSERTPKTHDR(m); 400 return (m->m_pkthdr.PH_per.eight[2]); 401 } 402 403 static inline void 404 set_mbuf_iscsi_iso_mss(struct mbuf *m, uint16_t mss) 405 { 406 407 M_ASSERTPKTHDR(m); 408 m->m_pkthdr.PH_per.sixteen[2] = mss; 409 } 410 411 static inline uint16_t 412 mbuf_iscsi_iso_mss(struct mbuf *m) 413 { 414 415 M_ASSERTPKTHDR(m); 416 return (m->m_pkthdr.PH_per.sixteen[2]); 417 } 418 419 /* t4_tom.c */ 420 struct toepcb *alloc_toepcb(struct vi_info *, int); 421 int init_toepcb(struct vi_info *, struct toepcb *); 422 struct toepcb *hold_toepcb(struct toepcb *); 423 void free_toepcb(struct toepcb *); 424 void offload_socket(struct socket *, struct toepcb *); 425 void restore_so_proto(struct socket *, bool); 426 void undo_offload_socket(struct socket *); 427 void final_cpl_received(struct toepcb *); 428 void insert_tid(struct adapter *, int, void *, int); 429 void *lookup_tid(struct adapter *, int); 430 void update_tid(struct adapter *, int, void *); 431 void remove_tid(struct adapter *, int, int); 432 u_long select_rcv_wnd(struct socket *); 433 int select_rcv_wscale(void); 434 void init_conn_params(struct vi_info *, struct offload_settings *, 435 struct in_conninfo *, struct socket *, const struct tcp_options *, int16_t, 436 struct conn_params *cp); 437 __be64 calc_options0(struct vi_info *, struct conn_params *); 438 __be32 calc_options2(struct vi_info *, struct conn_params *); 439 uint64_t select_ntuple(struct vi_info *, struct l2t_entry *); 440 int negative_advice(int); 441 int add_tid_to_history(struct adapter *, u_int); 442 443 /* t4_connect.c */ 444 void t4_init_connect_cpl_handlers(void); 445 void t4_uninit_connect_cpl_handlers(void); 446 int t4_connect(struct toedev *, struct socket *, struct nhop_object *, 447 struct sockaddr *); 448 void act_open_failure_cleanup(struct adapter *, u_int, u_int); 449 450 /* t4_listen.c */ 451 void t4_init_listen_cpl_handlers(void); 452 void t4_uninit_listen_cpl_handlers(void); 453 int t4_listen_start(struct toedev *, struct tcpcb *); 454 int t4_listen_stop(struct toedev *, struct tcpcb *); 455 void t4_syncache_added(struct toedev *, void *); 456 void t4_syncache_removed(struct toedev *, void *); 457 int t4_syncache_respond(struct toedev *, void *, struct mbuf *); 458 int do_abort_req_synqe(struct sge_iq *, const struct rss_header *, 459 struct mbuf *); 460 int do_abort_rpl_synqe(struct sge_iq *, const struct rss_header *, 461 struct mbuf *); 462 void t4_offload_socket(struct toedev *, void *, struct socket *); 463 void synack_failure_cleanup(struct adapter *, int); 464 465 /* t4_cpl_io.c */ 466 void aiotx_init_toep(struct toepcb *); 467 int t4_aio_queue_aiotx(struct socket *, struct kaiocb *); 468 void t4_init_cpl_io_handlers(void); 469 void t4_uninit_cpl_io_handlers(void); 470 void send_abort_rpl(struct adapter *, struct sge_ofld_txq *, int , int); 471 void send_flowc_wr(struct toepcb *, struct tcpcb *); 472 void send_reset(struct adapter *, struct toepcb *, uint32_t); 473 int send_rx_credits(struct adapter *, struct toepcb *, int); 474 void make_established(struct toepcb *, uint32_t, uint32_t, uint16_t); 475 int t4_close_conn(struct adapter *, struct toepcb *); 476 void t4_rcvd(struct toedev *, struct tcpcb *); 477 void t4_rcvd_locked(struct toedev *, struct tcpcb *); 478 int t4_tod_output(struct toedev *, struct tcpcb *); 479 int t4_send_fin(struct toedev *, struct tcpcb *); 480 int t4_send_rst(struct toedev *, struct tcpcb *); 481 void t4_set_tcb_field(struct adapter *, struct sge_wrq *, struct toepcb *, 482 uint16_t, uint64_t, uint64_t, int, int); 483 void t4_push_frames(struct adapter *, struct toepcb *, int); 484 void t4_push_pdus(struct adapter *, struct toepcb *, int); 485 486 /* t4_ddp.c */ 487 int t4_init_ppod_region(struct ppod_region *, struct t4_range *, u_int, 488 const char *); 489 void t4_free_ppod_region(struct ppod_region *); 490 int t4_alloc_page_pods_for_ps(struct ppod_region *, struct pageset *); 491 int t4_alloc_page_pods_for_bio(struct ppod_region *, struct bio *, 492 struct ppod_reservation *); 493 int t4_alloc_page_pods_for_buf(struct ppod_region *, vm_offset_t, int, 494 struct ppod_reservation *); 495 int t4_alloc_page_pods_for_sgl(struct ppod_region *, struct ctl_sg_entry *, int, 496 struct ppod_reservation *); 497 int t4_write_page_pods_for_ps(struct adapter *, struct sge_wrq *, int, 498 struct pageset *); 499 int t4_write_page_pods_for_bio(struct adapter *, struct toepcb *, 500 struct ppod_reservation *, struct bio *, struct mbufq *); 501 int t4_write_page_pods_for_buf(struct adapter *, struct toepcb *, 502 struct ppod_reservation *, vm_offset_t, int, struct mbufq *); 503 int t4_write_page_pods_for_sgl(struct adapter *, struct toepcb *, 504 struct ppod_reservation *, struct ctl_sg_entry *, int, int, struct mbufq *); 505 void t4_free_page_pods(struct ppod_reservation *); 506 int t4_soreceive_ddp(struct socket *, struct sockaddr **, struct uio *, 507 struct mbuf **, struct mbuf **, int *); 508 int t4_aio_queue_ddp(struct socket *, struct kaiocb *); 509 void t4_ddp_mod_load(void); 510 void t4_ddp_mod_unload(void); 511 void ddp_assert_empty(struct toepcb *); 512 void ddp_init_toep(struct toepcb *); 513 void ddp_uninit_toep(struct toepcb *); 514 void ddp_queue_toep(struct toepcb *); 515 void release_ddp_resources(struct toepcb *toep); 516 void handle_ddp_close(struct toepcb *, struct tcpcb *, uint32_t); 517 void handle_ddp_indicate(struct toepcb *); 518 void insert_ddp_data(struct toepcb *, uint32_t); 519 const struct offload_settings *lookup_offload_policy(struct adapter *, int, 520 struct mbuf *, uint16_t, struct inpcb *); 521 522 /* t4_tls.c */ 523 bool can_tls_offload(struct adapter *); 524 void do_rx_data_tls(const struct cpl_rx_data *, struct toepcb *, struct mbuf *); 525 void t4_push_ktls(struct adapter *, struct toepcb *, int); 526 void tls_received_starting_data(struct adapter *, struct toepcb *, 527 struct sockbuf *, int); 528 void t4_tls_mod_load(void); 529 void t4_tls_mod_unload(void); 530 void tls_init_toep(struct toepcb *); 531 int tls_tx_key(struct toepcb *); 532 void tls_uninit_toep(struct toepcb *); 533 int tls_alloc_ktls(struct toepcb *, struct ktls_session *, int); 534 535 #endif 536