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_FORCE_CREDITS = (1 << 10), /* always send credits */ 75 TPF_KTLS = (1 << 11), /* send TLS records from KTLS */ 76 }; 77 78 enum { 79 DDP_OK = (1 << 0), /* OK to turn on DDP */ 80 DDP_SC_REQ = (1 << 1), /* state change (on/off) requested */ 81 DDP_ON = (1 << 2), /* DDP is turned on */ 82 DDP_BUF0_ACTIVE = (1 << 3), /* buffer 0 in use (not invalidated) */ 83 DDP_BUF1_ACTIVE = (1 << 4), /* buffer 1 in use (not invalidated) */ 84 DDP_TASK_ACTIVE = (1 << 5), /* requeue task is queued / running */ 85 DDP_DEAD = (1 << 6), /* toepcb is shutting down */ 86 }; 87 88 struct sockopt; 89 struct offload_settings; 90 91 /* 92 * Connection parameters for an offloaded connection. These are mostly (but not 93 * all) hardware TOE parameters. 94 */ 95 struct conn_params { 96 int8_t rx_coalesce; 97 int8_t cong_algo; 98 int8_t tc_idx; 99 int8_t tstamp; 100 int8_t sack; 101 int8_t nagle; 102 int8_t keepalive; 103 int8_t wscale; 104 int8_t ecn; 105 int8_t mtu_idx; 106 int8_t ulp_mode; 107 int8_t tx_align; 108 int16_t txq_idx; /* ofld_txq = &sc->sge.ofld_txq[txq_idx] */ 109 int16_t rxq_idx; /* ofld_rxq = &sc->sge.ofld_rxq[rxq_idx] */ 110 int16_t l2t_idx; 111 uint16_t emss; 112 uint16_t opt0_bufsize; 113 u_int sndbuf; /* controls TP tx pages */ 114 }; 115 116 struct ofld_tx_sdesc { 117 uint32_t plen; /* payload length */ 118 uint8_t tx_credits; /* firmware tx credits (unit is 16B) */ 119 void *iv_buffer; /* optional buffer holding IVs for TLS */ 120 }; 121 122 struct ppod_region { 123 u_int pr_start; 124 u_int pr_len; 125 u_int pr_page_shift[4]; 126 uint32_t pr_tag_mask; /* hardware tagmask for this region. */ 127 uint32_t pr_invalid_bit; /* OR with this to invalidate tag. */ 128 uint32_t pr_alias_mask; /* AND with tag to get alias bits. */ 129 u_int pr_alias_shift; /* shift this much for first alias bit. */ 130 vmem_t *pr_arena; 131 }; 132 133 struct ppod_reservation { 134 struct ppod_region *prsv_pr; 135 uint32_t prsv_tag; /* Full tag: pgsz, alias, tag, color */ 136 u_int prsv_nppods; 137 }; 138 139 struct pageset { 140 TAILQ_ENTRY(pageset) link; 141 vm_page_t *pages; 142 int npages; 143 int flags; 144 int offset; /* offset in first page */ 145 int len; 146 struct ppod_reservation prsv; 147 struct vmspace *vm; 148 vm_offset_t start; 149 u_int vm_timestamp; 150 }; 151 152 TAILQ_HEAD(pagesetq, pageset); 153 154 #define PS_PPODS_WRITTEN 0x0001 /* Page pods written to the card. */ 155 156 struct ddp_buffer { 157 struct pageset *ps; 158 159 struct kaiocb *job; 160 int cancel_pending; 161 }; 162 163 struct ddp_pcb { 164 u_int flags; 165 struct ddp_buffer db[2]; 166 TAILQ_HEAD(, pageset) cached_pagesets; 167 TAILQ_HEAD(, kaiocb) aiojobq; 168 u_int waiting_count; 169 u_int active_count; 170 u_int cached_count; 171 int active_id; /* the currently active DDP buffer */ 172 struct task requeue_task; 173 struct kaiocb *queueing; 174 struct mtx lock; 175 }; 176 177 struct toepcb { 178 TAILQ_ENTRY(toepcb) link; /* toep_list */ 179 u_int flags; /* miscellaneous flags */ 180 int refcount; 181 struct tom_data *td; 182 struct inpcb *inp; /* backpointer to host stack's PCB */ 183 struct vnet *vnet; 184 struct vi_info *vi; /* virtual interface */ 185 struct sge_wrq *ofld_txq; 186 struct sge_ofld_rxq *ofld_rxq; 187 struct sge_wrq *ctrlq; 188 struct l2t_entry *l2te; /* L2 table entry used by this connection */ 189 struct clip_entry *ce; /* CLIP table entry used by this tid */ 190 int tid; /* Connection identifier */ 191 192 /* tx credit handling */ 193 u_int tx_total; /* total tx WR credits (in 16B units) */ 194 u_int tx_credits; /* tx WR credits (in 16B units) available */ 195 u_int tx_nocompl; /* tx WR credits since last compl request */ 196 u_int plen_nocompl; /* payload since last compl request */ 197 198 struct conn_params params; 199 200 void *ulpcb; 201 void *ulpcb2; 202 struct mbufq ulp_pduq; /* PDUs waiting to be sent out. */ 203 struct mbufq ulp_pdu_reclaimq; 204 205 struct ddp_pcb ddp; 206 struct tls_ofld_info tls; 207 208 TAILQ_HEAD(, kaiocb) aiotx_jobq; 209 struct task aiotx_task; 210 struct socket *aiotx_so; 211 212 /* Tx software descriptor */ 213 uint8_t txsd_total; 214 uint8_t txsd_pidx; 215 uint8_t txsd_cidx; 216 uint8_t txsd_avail; 217 struct ofld_tx_sdesc txsd[]; 218 }; 219 220 static inline int 221 ulp_mode(struct toepcb *toep) 222 { 223 224 return (toep->params.ulp_mode); 225 } 226 227 #define DDP_LOCK(toep) mtx_lock(&(toep)->ddp.lock) 228 #define DDP_UNLOCK(toep) mtx_unlock(&(toep)->ddp.lock) 229 #define DDP_ASSERT_LOCKED(toep) mtx_assert(&(toep)->ddp.lock, MA_OWNED) 230 231 /* 232 * Compressed state for embryonic connections for a listener. 233 */ 234 struct synq_entry { 235 struct listen_ctx *lctx; /* backpointer to listen ctx */ 236 struct mbuf *syn; 237 int flags; /* same as toepcb's tp_flags */ 238 volatile int ok_to_respond; 239 volatile u_int refcnt; 240 int tid; 241 uint32_t iss; 242 uint32_t irs; 243 uint32_t ts; 244 __be16 tcp_opt; /* from cpl_pass_establish */ 245 struct toepcb *toep; 246 247 struct conn_params params; 248 }; 249 250 /* listen_ctx flags */ 251 #define LCTX_RPL_PENDING 1 /* waiting for a CPL_PASS_OPEN_RPL */ 252 253 struct listen_ctx { 254 LIST_ENTRY(listen_ctx) link; /* listen hash linkage */ 255 volatile int refcount; 256 int stid; 257 struct stid_region stid_region; 258 int flags; 259 struct inpcb *inp; /* listening socket's inp */ 260 struct vnet *vnet; 261 struct sge_wrq *ctrlq; 262 struct sge_ofld_rxq *ofld_rxq; 263 struct clip_entry *ce; 264 }; 265 266 /* tcb_histent flags */ 267 #define TE_RPL_PENDING 1 268 #define TE_ACTIVE 2 269 270 /* bits in one 8b tcb_histent sample. */ 271 #define TS_RTO (1 << 0) 272 #define TS_DUPACKS (1 << 1) 273 #define TS_FASTREXMT (1 << 2) 274 #define TS_SND_BACKLOGGED (1 << 3) 275 #define TS_CWND_LIMITED (1 << 4) 276 #define TS_ECN_ECE (1 << 5) 277 #define TS_ECN_CWR (1 << 6) 278 #define TS_RESERVED (1 << 7) /* Unused. */ 279 280 struct tcb_histent { 281 struct mtx te_lock; 282 struct callout te_callout; 283 uint64_t te_tcb[TCB_SIZE / sizeof(uint64_t)]; 284 struct adapter *te_adapter; 285 u_int te_flags; 286 u_int te_tid; 287 uint8_t te_pidx; 288 uint8_t te_sample[100]; 289 }; 290 291 struct tom_data { 292 struct toedev tod; 293 294 /* toepcb's associated with this TOE device */ 295 struct mtx toep_list_lock; 296 TAILQ_HEAD(, toepcb) toep_list; 297 298 struct mtx lctx_hash_lock; 299 LIST_HEAD(, listen_ctx) *listen_hash; 300 u_long listen_mask; 301 int lctx_count; /* # of lctx in the hash table */ 302 303 struct ppod_region pr; 304 305 struct rwlock tcb_history_lock __aligned(CACHE_LINE_SIZE); 306 struct tcb_histent **tcb_history; 307 int dupack_threshold; 308 309 /* WRs that will not be sent to the chip because L2 resolution failed */ 310 struct mtx unsent_wr_lock; 311 STAILQ_HEAD(, wrqe) unsent_wr_list; 312 struct task reclaim_wr_resources; 313 }; 314 315 static inline struct tom_data * 316 tod_td(struct toedev *tod) 317 { 318 319 return (__containerof(tod, struct tom_data, tod)); 320 } 321 322 static inline struct adapter * 323 td_adapter(struct tom_data *td) 324 { 325 326 return (td->tod.tod_softc); 327 } 328 329 static inline void 330 set_mbuf_ulp_submode(struct mbuf *m, uint8_t ulp_submode) 331 { 332 333 M_ASSERTPKTHDR(m); 334 m->m_pkthdr.PH_per.eight[0] = ulp_submode; 335 } 336 337 static inline uint8_t 338 mbuf_ulp_submode(struct mbuf *m) 339 { 340 341 M_ASSERTPKTHDR(m); 342 return (m->m_pkthdr.PH_per.eight[0]); 343 } 344 345 /* t4_tom.c */ 346 struct toepcb *alloc_toepcb(struct vi_info *, int); 347 int init_toepcb(struct vi_info *, struct toepcb *); 348 struct toepcb *hold_toepcb(struct toepcb *); 349 void free_toepcb(struct toepcb *); 350 void offload_socket(struct socket *, struct toepcb *); 351 void undo_offload_socket(struct socket *); 352 void final_cpl_received(struct toepcb *); 353 void insert_tid(struct adapter *, int, void *, int); 354 void *lookup_tid(struct adapter *, int); 355 void update_tid(struct adapter *, int, void *); 356 void remove_tid(struct adapter *, int, int); 357 u_long select_rcv_wnd(struct socket *); 358 int select_rcv_wscale(void); 359 void init_conn_params(struct vi_info *, struct offload_settings *, 360 struct in_conninfo *, struct socket *, const struct tcp_options *, int16_t, 361 struct conn_params *cp); 362 __be64 calc_options0(struct vi_info *, struct conn_params *); 363 __be32 calc_options2(struct vi_info *, struct conn_params *); 364 uint64_t select_ntuple(struct vi_info *, struct l2t_entry *); 365 int negative_advice(int); 366 int add_tid_to_history(struct adapter *, u_int); 367 368 /* t4_connect.c */ 369 void t4_init_connect_cpl_handlers(void); 370 void t4_uninit_connect_cpl_handlers(void); 371 int t4_connect(struct toedev *, struct socket *, struct rtentry *, 372 struct sockaddr *); 373 void act_open_failure_cleanup(struct adapter *, u_int, u_int); 374 375 /* t4_listen.c */ 376 void t4_init_listen_cpl_handlers(void); 377 void t4_uninit_listen_cpl_handlers(void); 378 int t4_listen_start(struct toedev *, struct tcpcb *); 379 int t4_listen_stop(struct toedev *, struct tcpcb *); 380 void t4_syncache_added(struct toedev *, void *); 381 void t4_syncache_removed(struct toedev *, void *); 382 int t4_syncache_respond(struct toedev *, void *, struct mbuf *); 383 int do_abort_req_synqe(struct sge_iq *, const struct rss_header *, 384 struct mbuf *); 385 int do_abort_rpl_synqe(struct sge_iq *, const struct rss_header *, 386 struct mbuf *); 387 void t4_offload_socket(struct toedev *, void *, struct socket *); 388 void synack_failure_cleanup(struct adapter *, int); 389 390 /* t4_cpl_io.c */ 391 void aiotx_init_toep(struct toepcb *); 392 int t4_aio_queue_aiotx(struct socket *, struct kaiocb *); 393 void t4_init_cpl_io_handlers(void); 394 void t4_uninit_cpl_io_handlers(void); 395 void send_abort_rpl(struct adapter *, struct sge_wrq *, int , int); 396 void send_flowc_wr(struct toepcb *, struct tcpcb *); 397 void send_reset(struct adapter *, struct toepcb *, uint32_t); 398 int send_rx_credits(struct adapter *, struct toepcb *, int); 399 void send_rx_modulate(struct adapter *, struct toepcb *); 400 void make_established(struct toepcb *, uint32_t, uint32_t, uint16_t); 401 int t4_close_conn(struct adapter *, struct toepcb *); 402 void t4_rcvd(struct toedev *, struct tcpcb *); 403 void t4_rcvd_locked(struct toedev *, struct tcpcb *); 404 int t4_tod_output(struct toedev *, struct tcpcb *); 405 int t4_send_fin(struct toedev *, struct tcpcb *); 406 int t4_send_rst(struct toedev *, struct tcpcb *); 407 void t4_set_tcb_field(struct adapter *, struct sge_wrq *, struct toepcb *, 408 uint16_t, uint64_t, uint64_t, int, int); 409 void t4_push_frames(struct adapter *, struct toepcb *, int); 410 void t4_push_pdus(struct adapter *, struct toepcb *, int); 411 412 /* t4_ddp.c */ 413 int t4_init_ppod_region(struct ppod_region *, struct t4_range *, u_int, 414 const char *); 415 void t4_free_ppod_region(struct ppod_region *); 416 int t4_alloc_page_pods_for_ps(struct ppod_region *, struct pageset *); 417 int t4_alloc_page_pods_for_buf(struct ppod_region *, vm_offset_t, int, 418 struct ppod_reservation *); 419 int t4_write_page_pods_for_ps(struct adapter *, struct sge_wrq *, int, 420 struct pageset *); 421 int t4_write_page_pods_for_buf(struct adapter *, struct sge_wrq *, int tid, 422 struct ppod_reservation *, vm_offset_t, int); 423 void t4_free_page_pods(struct ppod_reservation *); 424 int t4_soreceive_ddp(struct socket *, struct sockaddr **, struct uio *, 425 struct mbuf **, struct mbuf **, int *); 426 int t4_aio_queue_ddp(struct socket *, struct kaiocb *); 427 void t4_ddp_mod_load(void); 428 void t4_ddp_mod_unload(void); 429 void ddp_assert_empty(struct toepcb *); 430 void ddp_init_toep(struct toepcb *); 431 void ddp_uninit_toep(struct toepcb *); 432 void ddp_queue_toep(struct toepcb *); 433 void release_ddp_resources(struct toepcb *toep); 434 void handle_ddp_close(struct toepcb *, struct tcpcb *, uint32_t); 435 void handle_ddp_indicate(struct toepcb *); 436 void insert_ddp_data(struct toepcb *, uint32_t); 437 const struct offload_settings *lookup_offload_policy(struct adapter *, int, 438 struct mbuf *, uint16_t, struct inpcb *); 439 440 /* t4_tls.c */ 441 bool can_tls_offload(struct adapter *); 442 int t4_ctloutput_tls(struct socket *, struct sockopt *); 443 void t4_push_tls_records(struct adapter *, struct toepcb *, int); 444 void t4_push_ktls(struct adapter *, struct toepcb *, int); 445 void t4_tls_mod_load(void); 446 void t4_tls_mod_unload(void); 447 void tls_establish(struct toepcb *); 448 void tls_init_toep(struct toepcb *); 449 int tls_rx_key(struct toepcb *); 450 void tls_stop_handshake_timer(struct toepcb *); 451 int tls_tx_key(struct toepcb *); 452 void tls_uninit_toep(struct toepcb *); 453 int tls_alloc_ktls(struct toepcb *, struct ktls_session *); 454 455 #endif 456