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 */ 63enum { 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 TPF_INITIALIZED = (1 << 12), /* init_toepcb has been called */ 77 TPF_TLS_RECEIVE = (1 << 13), /* should receive TLS records */ 78 TPF_TLS_ESTABLISHED = (1 << 14), /* TLS handshake timer initialized */ 79}; 80 81enum { 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}; 90 91struct sockopt; 92struct offload_settings; 93 94/* 95 * Connection parameters for an offloaded connection. These are mostly (but not 96 * all) hardware TOE parameters. 97 */ 98struct conn_params { 99 int8_t rx_coalesce; 100 int8_t cong_algo; 101 int8_t tc_idx; 102 int8_t tstamp; 103 int8_t sack; 104 int8_t nagle; 105 int8_t keepalive; 106 int8_t wscale; 107 int8_t ecn; 108 int8_t mtu_idx; 109 int8_t ulp_mode; 110 int8_t tx_align; 111 int16_t txq_idx; /* ofld_txq = &sc->sge.ofld_txq[txq_idx] */ 112 int16_t rxq_idx; /* ofld_rxq = &sc->sge.ofld_rxq[rxq_idx] */ 113 int16_t l2t_idx; 114 uint16_t emss; 115 uint16_t opt0_bufsize; 116 u_int sndbuf; /* controls TP tx pages */ 117}; 118 119struct ofld_tx_sdesc { 120 uint32_t plen; /* payload length */ 121 uint8_t tx_credits; /* firmware tx credits (unit is 16B) */ 122 void *iv_buffer; /* optional buffer holding IVs for TLS */ 123}; 124 125struct ppod_region { 126 u_int pr_start; 127 u_int pr_len; 128 u_int pr_page_shift[4]; 129 uint32_t pr_tag_mask; /* hardware tagmask for this region. */ 130 uint32_t pr_invalid_bit; /* OR with this to invalidate tag. */ 131 uint32_t pr_alias_mask; /* AND with tag to get alias bits. */ 132 u_int pr_alias_shift; /* shift this much for first alias bit. */ 133 vmem_t *pr_arena; 134}; 135 136struct ppod_reservation { 137 struct ppod_region *prsv_pr; 138 uint32_t prsv_tag; /* Full tag: pgsz, alias, tag, color */ 139 u_int prsv_nppods; 140}; 141 142struct pageset { 143 TAILQ_ENTRY(pageset) link; 144 vm_page_t *pages; 145 int npages; 146 int flags; 147 int offset; /* offset in first page */ 148 int len; 149 struct ppod_reservation prsv; 150 struct vmspace *vm; 151 vm_offset_t start; 152 u_int vm_timestamp; 153}; 154 155TAILQ_HEAD(pagesetq, pageset); 156 157#define PS_PPODS_WRITTEN 0x0001 /* Page pods written to the card. */ 158 159struct ddp_buffer { 160 struct pageset *ps; 161 162 struct kaiocb *job; 163 int cancel_pending; 164}; 165 166struct ddp_pcb { 167 u_int flags; 168 struct ddp_buffer db[2]; 169 TAILQ_HEAD(, pageset) cached_pagesets; 170 TAILQ_HEAD(, kaiocb) aiojobq; 171 u_int waiting_count; 172 u_int active_count; 173 u_int cached_count; 174 int active_id; /* the currently active DDP buffer */ 175 struct task requeue_task; 176 struct kaiocb *queueing; 177 struct mtx lock; 178}; 179 180struct toepcb { 181 struct tom_data *td; 182 struct inpcb *inp; /* backpointer to host stack's PCB */ 183 u_int flags; /* miscellaneous flags */ 184 TAILQ_ENTRY(toepcb) link; /* toep_list */ 185 int refcount; 186 struct vnet *vnet; 187 struct vi_info *vi; /* virtual interface */ 188 struct sge_ofld_txq *ofld_txq; 189 struct sge_ofld_rxq *ofld_rxq; 190 struct sge_wrq *ctrlq; 191 struct l2t_entry *l2te; /* L2 table entry used by this connection */ 192 struct clip_entry *ce; /* CLIP table entry used by this tid */ 193 int tid; /* Connection identifier */ 194 195 /* tx credit handling */ 196 u_int tx_total; /* total tx WR credits (in 16B units) */ 197 u_int tx_credits; /* tx WR credits (in 16B units) available */ 198 u_int tx_nocompl; /* tx WR credits since last compl request */ 199 u_int plen_nocompl; /* payload since last compl request */ 200 201 struct conn_params params; 202 203 void *ulpcb; 204 void *ulpcb2; 205 struct mbufq ulp_pduq; /* PDUs waiting to be sent out. */ 206 struct mbufq ulp_pdu_reclaimq; 207 208 struct ddp_pcb ddp; 209 struct tls_ofld_info tls; 210 211 TAILQ_HEAD(, kaiocb) aiotx_jobq; 212 struct task aiotx_task; 213 struct socket *aiotx_so; 214 215 /* Tx software descriptor */ 216 uint8_t txsd_total; 217 uint8_t txsd_pidx; 218 uint8_t txsd_cidx; 219 uint8_t txsd_avail; 220 struct ofld_tx_sdesc txsd[]; 221}; 222 223static inline int 224ulp_mode(struct toepcb *toep) 225{ 226 227 return (toep->params.ulp_mode); 228} 229 230#define DDP_LOCK(toep) mtx_lock(&(toep)->ddp.lock) 231#define DDP_UNLOCK(toep) mtx_unlock(&(toep)->ddp.lock) 232#define DDP_ASSERT_LOCKED(toep) mtx_assert(&(toep)->ddp.lock, MA_OWNED) 233 234/* 235 * Compressed state for embryonic connections for a listener. 236 */ 237struct synq_entry { 238 struct listen_ctx *lctx; /* backpointer to listen ctx */ 239 struct mbuf *syn; 240 int flags; /* same as toepcb's tp_flags */ 241 volatile int ok_to_respond; 242 volatile u_int refcnt; 243 int tid; 244 uint32_t iss; 245 uint32_t irs; 246 uint32_t ts; 247 uint32_t rss_hash; 248 __be16 tcp_opt; /* from cpl_pass_establish */ 249 struct toepcb *toep; 250 251 struct conn_params params; 252}; 253 254/* listen_ctx flags */ 255#define LCTX_RPL_PENDING 1 /* waiting for a CPL_PASS_OPEN_RPL */ 256 257struct listen_ctx { 258 LIST_ENTRY(listen_ctx) link; /* listen hash linkage */ 259 volatile int refcount; 260 int stid; 261 struct stid_region stid_region; 262 int flags; 263 struct inpcb *inp; /* listening socket's inp */ 264 struct vnet *vnet; 265 struct sge_wrq *ctrlq; 266 struct sge_ofld_rxq *ofld_rxq; 267 struct clip_entry *ce; 268}; 269 270/* tcb_histent flags */ 271#define TE_RPL_PENDING 1 272#define TE_ACTIVE 2 273 274/* bits in one 8b tcb_histent sample. */ 275#define TS_RTO (1 << 0) 276#define TS_DUPACKS (1 << 1) 277#define TS_FASTREXMT (1 << 2) 278#define TS_SND_BACKLOGGED (1 << 3) 279#define TS_CWND_LIMITED (1 << 4) 280#define TS_ECN_ECE (1 << 5) 281#define TS_ECN_CWR (1 << 6) 282#define TS_RESERVED (1 << 7) /* Unused. */ 283 284struct tcb_histent { 285 struct mtx te_lock; 286 struct callout te_callout; 287 uint64_t te_tcb[TCB_SIZE / sizeof(uint64_t)]; 288 struct adapter *te_adapter; 289 u_int te_flags; 290 u_int te_tid; 291 uint8_t te_pidx; 292 uint8_t te_sample[100]; 293}; 294 295struct tom_data { 296 struct toedev tod; 297 298 /* toepcb's associated with this TOE device */ 299 struct mtx toep_list_lock; 300 TAILQ_HEAD(, toepcb) toep_list; 301 302 struct mtx lctx_hash_lock; 303 LIST_HEAD(, listen_ctx) *listen_hash; 304 u_long listen_mask; 305 int lctx_count; /* # of lctx in the hash table */ 306 307 struct ppod_region pr; 308 309 struct rwlock tcb_history_lock __aligned(CACHE_LINE_SIZE); 310 struct tcb_histent **tcb_history; 311 int dupack_threshold; 312 313 /* WRs that will not be sent to the chip because L2 resolution failed */ 314 struct mtx unsent_wr_lock; 315 STAILQ_HEAD(, wrqe) unsent_wr_list; 316 struct task reclaim_wr_resources; 317}; 318 319static inline struct tom_data * 320tod_td(struct toedev *tod) 321{ 322 323 return (__containerof(tod, struct tom_data, tod)); 324} 325 326static inline struct adapter * 327td_adapter(struct tom_data *td) 328{ 329 330 return (td->tod.tod_softc); 331} 332 333static inline void 334set_mbuf_ulp_submode(struct mbuf *m, uint8_t ulp_submode) 335{ 336 337 M_ASSERTPKTHDR(m); 338 m->m_pkthdr.PH_per.eight[0] = ulp_submode; 339} 340 341static inline uint8_t 342mbuf_ulp_submode(struct mbuf *m) 343{ 344 345 M_ASSERTPKTHDR(m); 346 return (m->m_pkthdr.PH_per.eight[0]); 347} 348 349/* t4_tom.c */ 350struct toepcb *alloc_toepcb(struct vi_info *, int); 351int init_toepcb(struct vi_info *, struct toepcb *); 352struct toepcb *hold_toepcb(struct toepcb *); 353void free_toepcb(struct toepcb *); 354void offload_socket(struct socket *, struct toepcb *); 355void restore_so_proto(struct socket *, bool); 356void undo_offload_socket(struct socket *); 357void final_cpl_received(struct toepcb *); 358void insert_tid(struct adapter *, int, void *, int); 359void *lookup_tid(struct adapter *, int); 360void update_tid(struct adapter *, int, void *); 361void remove_tid(struct adapter *, int, int); 362u_long select_rcv_wnd(struct socket *); 363int select_rcv_wscale(void); 364void init_conn_params(struct vi_info *, struct offload_settings *, 365 struct in_conninfo *, struct socket *, const struct tcp_options *, int16_t, 366 struct conn_params *cp); 367__be64 calc_options0(struct vi_info *, struct conn_params *); 368__be32 calc_options2(struct vi_info *, struct conn_params *); 369uint64_t select_ntuple(struct vi_info *, struct l2t_entry *); 370int negative_advice(int); 371int add_tid_to_history(struct adapter *, u_int); 372 373/* t4_connect.c */ 374void t4_init_connect_cpl_handlers(void); 375void t4_uninit_connect_cpl_handlers(void); 376int t4_connect(struct toedev *, struct socket *, struct nhop_object *, 377 struct sockaddr *); 378void act_open_failure_cleanup(struct adapter *, u_int, u_int); 379 380/* t4_listen.c */ 381void t4_init_listen_cpl_handlers(void); 382void t4_uninit_listen_cpl_handlers(void); 383int t4_listen_start(struct toedev *, struct tcpcb *); 384int t4_listen_stop(struct toedev *, struct tcpcb *); 385void t4_syncache_added(struct toedev *, void *); 386void t4_syncache_removed(struct toedev *, void *); 387int t4_syncache_respond(struct toedev *, void *, struct mbuf *); 388int do_abort_req_synqe(struct sge_iq *, const struct rss_header *, 389 struct mbuf *); 390int do_abort_rpl_synqe(struct sge_iq *, const struct rss_header *, 391 struct mbuf *); 392void t4_offload_socket(struct toedev *, void *, struct socket *); 393void synack_failure_cleanup(struct adapter *, int); 394 395/* t4_cpl_io.c */ 396void aiotx_init_toep(struct toepcb *); 397int t4_aio_queue_aiotx(struct socket *, struct kaiocb *); 398void t4_init_cpl_io_handlers(void); 399void t4_uninit_cpl_io_handlers(void); 400void send_abort_rpl(struct adapter *, struct sge_ofld_txq *, int , int); 401void send_flowc_wr(struct toepcb *, struct tcpcb *); 402void send_reset(struct adapter *, struct toepcb *, uint32_t); 403int send_rx_credits(struct adapter *, struct toepcb *, int); 404void send_rx_modulate(struct adapter *, struct toepcb *); 405void make_established(struct toepcb *, uint32_t, uint32_t, uint16_t); 406int t4_close_conn(struct adapter *, struct toepcb *); 407void t4_rcvd(struct toedev *, struct tcpcb *); 408void t4_rcvd_locked(struct toedev *, struct tcpcb *); 409int t4_tod_output(struct toedev *, struct tcpcb *); 410int t4_send_fin(struct toedev *, struct tcpcb *); 411int t4_send_rst(struct toedev *, struct tcpcb *); 412void t4_set_tcb_field(struct adapter *, struct sge_wrq *, struct toepcb *, 413 uint16_t, uint64_t, uint64_t, int, int); 414void t4_push_frames(struct adapter *, struct toepcb *, int); 415void t4_push_pdus(struct adapter *, struct toepcb *, int); 416 417/* t4_ddp.c */ 418int t4_init_ppod_region(struct ppod_region *, struct t4_range *, u_int, 419 const char *); 420void t4_free_ppod_region(struct ppod_region *); 421int t4_alloc_page_pods_for_ps(struct ppod_region *, struct pageset *); 422int t4_alloc_page_pods_for_buf(struct ppod_region *, vm_offset_t, int, 423 struct ppod_reservation *); 424int t4_write_page_pods_for_ps(struct adapter *, struct sge_wrq *, int, 425 struct pageset *); 426int t4_write_page_pods_for_buf(struct adapter *, struct sge_wrq *, int, 427 struct ppod_reservation *, vm_offset_t, int); 428void t4_free_page_pods(struct ppod_reservation *); 429int t4_soreceive_ddp(struct socket *, struct sockaddr **, struct uio *, 430 struct mbuf **, struct mbuf **, int *); 431int t4_aio_queue_ddp(struct socket *, struct kaiocb *); 432void t4_ddp_mod_load(void); 433void t4_ddp_mod_unload(void); 434void ddp_assert_empty(struct toepcb *); 435void ddp_init_toep(struct toepcb *); 436void ddp_uninit_toep(struct toepcb *); 437void ddp_queue_toep(struct toepcb *); 438void release_ddp_resources(struct toepcb *toep); 439void handle_ddp_close(struct toepcb *, struct tcpcb *, uint32_t); 440void handle_ddp_indicate(struct toepcb *); 441void insert_ddp_data(struct toepcb *, uint32_t); 442const struct offload_settings *lookup_offload_policy(struct adapter *, int, 443 struct mbuf *, uint16_t, struct inpcb *); 444 445/* t4_tls.c */ 446bool can_tls_offload(struct adapter *); 447void do_rx_data_tls(const struct cpl_rx_data *, struct toepcb *, struct mbuf *); 448int t4_ctloutput_tls(struct socket *, struct sockopt *); 449void t4_push_tls_records(struct adapter *, struct toepcb *, int); 450void t4_push_ktls(struct adapter *, struct toepcb *, int); 451void t4_tls_mod_load(void); 452void t4_tls_mod_unload(void); 453void tls_detach(struct toepcb *); 454void tls_establish(struct toepcb *); 455void tls_init_toep(struct toepcb *); 456int tls_rx_key(struct toepcb *); 457void tls_stop_handshake_timer(struct toepcb *); 458int tls_tx_key(struct toepcb *); 459void tls_uninit_toep(struct toepcb *); 460int tls_alloc_ktls(struct toepcb *, struct ktls_session *, int); 461 462#endif 463