1 /*- 2 * Copyright (c) 2012, 2015 Chelsio Communications, Inc. 3 * All rights reserved. 4 * Written by: Navdeep Parhar <np@FreeBSD.org> 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 25 * SUCH DAMAGE. 26 * 27 * $FreeBSD$ 28 * 29 */ 30 31 #ifndef __T4_TOM_H__ 32 #define __T4_TOM_H__ 33 #include <sys/vmem.h> 34 35 #define LISTEN_HASH_SIZE 32 36 37 /* 38 * Min receive window. We want it to be large enough to accommodate receive 39 * coalescing, handle jumbo frames, and not trigger sender SWS avoidance. 40 */ 41 #define MIN_RCV_WND (24 * 1024U) 42 43 /* 44 * Max receive window supported by HW in bytes. Only a small part of it can 45 * be set through option0, the rest needs to be set through RX_DATA_ACK. 46 */ 47 #define MAX_RCV_WND ((1U << 27) - 1) 48 49 #define DDP_RSVD_WIN (16 * 1024U) 50 #define SB_DDP_INDICATE SB_IN_TOE /* soreceive must respond to indicate */ 51 52 #define USE_DDP_RX_FLOW_CONTROL 53 54 /* TOE PCB flags */ 55 enum { 56 TPF_ATTACHED = (1 << 0), /* a tcpcb refers to this toepcb */ 57 TPF_FLOWC_WR_SENT = (1 << 1), /* firmware flow context WR sent */ 58 TPF_TX_DATA_SENT = (1 << 2), /* some data sent */ 59 TPF_TX_SUSPENDED = (1 << 3), /* tx suspended for lack of resources */ 60 TPF_SEND_FIN = (1 << 4), /* send FIN after all pending data */ 61 TPF_FIN_SENT = (1 << 5), /* FIN has been sent */ 62 TPF_ABORT_SHUTDOWN = (1 << 6), /* connection abort is in progress */ 63 TPF_CPL_PENDING = (1 << 7), /* haven't received the last CPL */ 64 TPF_SYNQE = (1 << 8), /* synq_entry, not really a toepcb */ 65 TPF_SYNQE_NEEDFREE = (1 << 9), /* synq_entry was malloc'd separately */ 66 TPF_SYNQE_TCPDDP = (1 << 10), /* ulp_mode TCPDDP in toepcb */ 67 TPF_SYNQE_EXPANDED = (1 << 11), /* toepcb ready, tid context updated */ 68 TPF_SYNQE_HAS_L2TE = (1 << 12), /* we've replied to PASS_ACCEPT_REQ */ 69 }; 70 71 enum { 72 DDP_OK = (1 << 0), /* OK to turn on DDP */ 73 DDP_SC_REQ = (1 << 1), /* state change (on/off) requested */ 74 DDP_ON = (1 << 2), /* DDP is turned on */ 75 DDP_BUF0_ACTIVE = (1 << 3), /* buffer 0 in use (not invalidated) */ 76 DDP_BUF1_ACTIVE = (1 << 4), /* buffer 1 in use (not invalidated) */ 77 DDP_TASK_ACTIVE = (1 << 5), /* requeue task is queued / running */ 78 DDP_DEAD = (1 << 6), /* toepcb is shutting down */ 79 }; 80 81 struct ofld_tx_sdesc { 82 uint32_t plen; /* payload length */ 83 uint8_t tx_credits; /* firmware tx credits (unit is 16B) */ 84 }; 85 86 struct pageset { 87 TAILQ_ENTRY(pageset) link; 88 vm_page_t *pages; 89 int npages; 90 int flags; 91 u_int ppod_addr; 92 int nppods; 93 uint32_t tag; /* includes color, page pod addr, and DDP page size */ 94 int offset; /* offset in first page */ 95 int len; 96 struct vmspace *vm; 97 u_int vm_timestamp; 98 }; 99 100 TAILQ_HEAD(pagesetq, pageset); 101 102 #define PS_WIRED 0x0001 /* Pages wired rather than held. */ 103 #define PS_PPODS_WRITTEN 0x0002 /* Page pods written to the card. */ 104 105 struct ddp_buffer { 106 struct pageset *ps; 107 108 struct kaiocb *job; 109 int cancel_pending; 110 }; 111 112 struct toepcb { 113 TAILQ_ENTRY(toepcb) link; /* toep_list */ 114 u_int flags; /* miscellaneous flags */ 115 int refcount; 116 struct tom_data *td; 117 struct inpcb *inp; /* backpointer to host stack's PCB */ 118 struct vi_info *vi; /* virtual interface */ 119 struct sge_wrq *ofld_txq; 120 struct sge_ofld_rxq *ofld_rxq; 121 struct sge_wrq *ctrlq; 122 struct l2t_entry *l2te; /* L2 table entry used by this connection */ 123 struct clip_entry *ce; /* CLIP table entry used by this tid */ 124 int tid; /* Connection identifier */ 125 126 /* tx credit handling */ 127 u_int tx_total; /* total tx WR credits (in 16B units) */ 128 u_int tx_credits; /* tx WR credits (in 16B units) available */ 129 u_int tx_nocompl; /* tx WR credits since last compl request */ 130 u_int plen_nocompl; /* payload since last compl request */ 131 132 /* rx credit handling */ 133 u_int sb_cc; /* last noted value of so_rcv->sb_cc */ 134 int rx_credits; /* rx credits (in bytes) to be returned to hw */ 135 136 u_int ulp_mode; /* ULP mode */ 137 void *ulpcb; 138 void *ulpcb2; 139 struct mbufq ulp_pduq; /* PDUs waiting to be sent out. */ 140 struct mbufq ulp_pdu_reclaimq; 141 142 u_int ddp_flags; 143 struct ddp_buffer db[2]; 144 TAILQ_HEAD(, pageset) ddp_cached_pagesets; 145 TAILQ_HEAD(, kaiocb) ddp_aiojobq; 146 u_int ddp_waiting_count; 147 u_int ddp_active_count; 148 u_int ddp_cached_count; 149 int ddp_active_id; /* the currently active DDP buffer */ 150 struct task ddp_requeue_task; 151 struct kaiocb *ddp_queueing; 152 struct mtx ddp_lock; 153 154 /* Tx software descriptor */ 155 uint8_t txsd_total; 156 uint8_t txsd_pidx; 157 uint8_t txsd_cidx; 158 uint8_t txsd_avail; 159 struct ofld_tx_sdesc txsd[]; 160 }; 161 162 #define DDP_LOCK(toep) mtx_lock(&(toep)->ddp_lock) 163 #define DDP_UNLOCK(toep) mtx_unlock(&(toep)->ddp_lock) 164 #define DDP_ASSERT_LOCKED(toep) mtx_assert(&(toep)->ddp_lock, MA_OWNED) 165 166 struct flowc_tx_params { 167 uint32_t snd_nxt; 168 uint32_t rcv_nxt; 169 unsigned int snd_space; 170 unsigned int mss; 171 }; 172 173 #define DDP_RETRY_WAIT 5 /* seconds to wait before re-enabling DDP */ 174 #define DDP_LOW_SCORE 1 175 #define DDP_HIGH_SCORE 3 176 177 /* 178 * Compressed state for embryonic connections for a listener. Barely fits in 179 * 64B, try not to grow it further. 180 */ 181 struct synq_entry { 182 TAILQ_ENTRY(synq_entry) link; /* listen_ctx's synq link */ 183 int flags; /* same as toepcb's tp_flags */ 184 int tid; 185 struct listen_ctx *lctx; /* backpointer to listen ctx */ 186 struct mbuf *syn; 187 uint32_t iss; 188 uint32_t ts; 189 volatile uintptr_t wr; 190 volatile u_int refcnt; 191 uint16_t l2e_idx; 192 uint16_t rcv_bufsize; 193 }; 194 195 /* listen_ctx flags */ 196 #define LCTX_RPL_PENDING 1 /* waiting for a CPL_PASS_OPEN_RPL */ 197 198 struct listen_ctx { 199 LIST_ENTRY(listen_ctx) link; /* listen hash linkage */ 200 volatile int refcount; 201 int stid; 202 struct stid_region stid_region; 203 int flags; 204 struct inpcb *inp; /* listening socket's inp */ 205 struct sge_wrq *ctrlq; 206 struct sge_ofld_rxq *ofld_rxq; 207 struct clip_entry *ce; 208 TAILQ_HEAD(, synq_entry) synq; 209 }; 210 211 struct clip_entry { 212 TAILQ_ENTRY(clip_entry) link; 213 struct in6_addr lip; /* local IPv6 address */ 214 u_int refcount; 215 }; 216 217 TAILQ_HEAD(clip_head, clip_entry); 218 struct tom_data { 219 struct toedev tod; 220 221 /* toepcb's associated with this TOE device */ 222 struct mtx toep_list_lock; 223 TAILQ_HEAD(, toepcb) toep_list; 224 225 struct mtx lctx_hash_lock; 226 LIST_HEAD(, listen_ctx) *listen_hash; 227 u_long listen_mask; 228 int lctx_count; /* # of lctx in the hash table */ 229 230 u_int ppod_start; 231 vmem_t *ppod_arena; 232 233 struct mtx clip_table_lock; 234 struct clip_head clip_table; 235 int clip_gen; 236 237 /* WRs that will not be sent to the chip because L2 resolution failed */ 238 struct mtx unsent_wr_lock; 239 STAILQ_HEAD(, wrqe) unsent_wr_list; 240 struct task reclaim_wr_resources; 241 }; 242 243 static inline struct tom_data * 244 tod_td(struct toedev *tod) 245 { 246 247 return (__containerof(tod, struct tom_data, tod)); 248 } 249 250 static inline struct adapter * 251 td_adapter(struct tom_data *td) 252 { 253 254 return (td->tod.tod_softc); 255 } 256 257 static inline void 258 set_mbuf_ulp_submode(struct mbuf *m, uint8_t ulp_submode) 259 { 260 261 M_ASSERTPKTHDR(m); 262 m->m_pkthdr.PH_per.eight[0] = ulp_submode; 263 } 264 265 static inline uint8_t 266 mbuf_ulp_submode(struct mbuf *m) 267 { 268 269 M_ASSERTPKTHDR(m); 270 return (m->m_pkthdr.PH_per.eight[0]); 271 } 272 273 /* t4_tom.c */ 274 struct toepcb *alloc_toepcb(struct vi_info *, int, int, int); 275 struct toepcb *hold_toepcb(struct toepcb *); 276 void free_toepcb(struct toepcb *); 277 void offload_socket(struct socket *, struct toepcb *); 278 void undo_offload_socket(struct socket *); 279 void final_cpl_received(struct toepcb *); 280 void insert_tid(struct adapter *, int, void *); 281 void *lookup_tid(struct adapter *, int); 282 void update_tid(struct adapter *, int, void *); 283 void remove_tid(struct adapter *, int); 284 void release_tid(struct adapter *, int, struct sge_wrq *); 285 int find_best_mtu_idx(struct adapter *, struct in_conninfo *, int); 286 u_long select_rcv_wnd(struct socket *); 287 int select_rcv_wscale(void); 288 uint64_t calc_opt0(struct socket *, struct vi_info *, struct l2t_entry *, 289 int, int, int, int); 290 uint64_t select_ntuple(struct vi_info *, struct l2t_entry *); 291 void set_tcpddp_ulp_mode(struct toepcb *); 292 int negative_advice(int); 293 struct clip_entry *hold_lip(struct tom_data *, struct in6_addr *); 294 void release_lip(struct tom_data *, struct clip_entry *); 295 296 /* t4_connect.c */ 297 void t4_init_connect_cpl_handlers(void); 298 int t4_connect(struct toedev *, struct socket *, struct rtentry *, 299 struct sockaddr *); 300 void act_open_failure_cleanup(struct adapter *, u_int, u_int); 301 302 /* t4_listen.c */ 303 void t4_init_listen_cpl_handlers(void); 304 int t4_listen_start(struct toedev *, struct tcpcb *); 305 int t4_listen_stop(struct toedev *, struct tcpcb *); 306 void t4_syncache_added(struct toedev *, void *); 307 void t4_syncache_removed(struct toedev *, void *); 308 int t4_syncache_respond(struct toedev *, void *, struct mbuf *); 309 int do_abort_req_synqe(struct sge_iq *, const struct rss_header *, 310 struct mbuf *); 311 int do_abort_rpl_synqe(struct sge_iq *, const struct rss_header *, 312 struct mbuf *); 313 void t4_offload_socket(struct toedev *, void *, struct socket *); 314 315 /* t4_cpl_io.c */ 316 void t4_init_cpl_io_handlers(void); 317 void t4_uninit_cpl_io_handlers(void); 318 void send_abort_rpl(struct adapter *, struct sge_wrq *, int , int); 319 void send_flowc_wr(struct toepcb *, struct flowc_tx_params *); 320 void send_reset(struct adapter *, struct toepcb *, uint32_t); 321 void make_established(struct toepcb *, uint32_t, uint32_t, uint16_t); 322 void t4_rcvd(struct toedev *, struct tcpcb *); 323 void t4_rcvd_locked(struct toedev *, struct tcpcb *); 324 int t4_tod_output(struct toedev *, struct tcpcb *); 325 int t4_send_fin(struct toedev *, struct tcpcb *); 326 int t4_send_rst(struct toedev *, struct tcpcb *); 327 void t4_set_tcb_field(struct adapter *, struct sge_wrq *, int, uint16_t, 328 uint64_t, uint64_t, int, int, int); 329 void t4_push_frames(struct adapter *sc, struct toepcb *toep, int drop); 330 void t4_push_pdus(struct adapter *sc, struct toepcb *toep, int drop); 331 int do_set_tcb_rpl(struct sge_iq *, const struct rss_header *, struct mbuf *); 332 333 /* t4_ddp.c */ 334 void t4_init_ddp(struct adapter *, struct tom_data *); 335 void t4_uninit_ddp(struct adapter *, struct tom_data *); 336 int t4_soreceive_ddp(struct socket *, struct sockaddr **, struct uio *, 337 struct mbuf **, struct mbuf **, int *); 338 int t4_aio_queue_ddp(struct socket *, struct kaiocb *); 339 int t4_ddp_mod_load(void); 340 void t4_ddp_mod_unload(void); 341 void ddp_assert_empty(struct toepcb *); 342 void ddp_init_toep(struct toepcb *); 343 void ddp_uninit_toep(struct toepcb *); 344 void ddp_queue_toep(struct toepcb *); 345 void release_ddp_resources(struct toepcb *toep); 346 void handle_ddp_close(struct toepcb *, struct tcpcb *, uint32_t); 347 void handle_ddp_indicate(struct toepcb *); 348 void handle_ddp_tcb_rpl(struct toepcb *, const struct cpl_set_tcb_rpl *); 349 void insert_ddp_data(struct toepcb *, uint32_t); 350 351 #endif 352