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_FORCE_CREDITS = (1 << 10), /* always send credits */
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_ESTABLISHED = (1 << 14), /* TLS handshake timer initialized */
77 TPF_WAITING_FOR_FINAL = (1<< 15), /* waiting for wakeup on final CPL */
78 };
79
80 enum {
81 DDP_OK = (1 << 0), /* OK to turn on DDP */
82 DDP_SC_REQ = (1 << 1), /* state change (on/off) requested */
83 DDP_ON = (1 << 2), /* DDP is turned on */
84 DDP_BUF0_ACTIVE = (1 << 3), /* buffer 0 in use (not invalidated) */
85 DDP_BUF1_ACTIVE = (1 << 4), /* buffer 1 in use (not invalidated) */
86 DDP_TASK_ACTIVE = (1 << 5), /* requeue task is queued / running */
87 DDP_DEAD = (1 << 6), /* toepcb is shutting down */
88 };
89
90 struct bio;
91 struct ctl_sg_entry;
92 struct sockopt;
93 struct offload_settings;
94
95 /*
96 * Connection parameters for an offloaded connection. These are mostly (but not
97 * all) hardware TOE parameters.
98 */
99 struct conn_params {
100 int8_t rx_coalesce;
101 int8_t cong_algo;
102 int8_t tc_idx;
103 int8_t tstamp;
104 int8_t sack;
105 int8_t nagle;
106 int8_t keepalive;
107 int8_t wscale;
108 int8_t ecn;
109 int8_t mtu_idx;
110 int8_t ulp_mode;
111 int8_t tx_align;
112 int16_t txq_idx; /* ofld_txq = &sc->sge.ofld_txq[txq_idx] */
113 int16_t rxq_idx; /* ofld_rxq = &sc->sge.ofld_rxq[rxq_idx] */
114 int16_t l2t_idx;
115 uint16_t emss;
116 uint16_t opt0_bufsize;
117 u_int sndbuf; /* controls TP tx pages */
118 };
119
120 struct ofld_tx_sdesc {
121 uint32_t plen; /* payload length */
122 uint8_t tx_credits; /* firmware tx credits (unit is 16B) */
123 };
124
125 struct 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
136 struct 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
142 struct 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
155 TAILQ_HEAD(pagesetq, pageset);
156
157 #define PS_PPODS_WRITTEN 0x0001 /* Page pods written to the card. */
158
159 struct ddp_buffer {
160 struct pageset *ps;
161
162 struct kaiocb *job;
163 int cancel_pending;
164 };
165
166 struct 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
180 struct 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
223 static inline int
ulp_mode(struct toepcb * toep)224 ulp_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 */
237 struct 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
257 struct 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
284 struct 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
295 struct 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
319 static inline struct tom_data *
tod_td(struct toedev * tod)320 tod_td(struct toedev *tod)
321 {
322
323 return (__containerof(tod, struct tom_data, tod));
324 }
325
326 static inline struct adapter *
td_adapter(struct tom_data * td)327 td_adapter(struct tom_data *td)
328 {
329
330 return (td->tod.tod_softc);
331 }
332
333 static inline void
set_mbuf_raw_wr(struct mbuf * m,bool raw)334 set_mbuf_raw_wr(struct mbuf *m, bool raw)
335 {
336
337 M_ASSERTPKTHDR(m);
338 m->m_pkthdr.PH_per.eight[6] = raw;
339 }
340
341 static inline bool
mbuf_raw_wr(struct mbuf * m)342 mbuf_raw_wr(struct mbuf *m)
343 {
344
345 M_ASSERTPKTHDR(m);
346 return (m->m_pkthdr.PH_per.eight[6]);
347 }
348
349 static inline void
set_mbuf_ulp_submode(struct mbuf * m,uint8_t ulp_submode)350 set_mbuf_ulp_submode(struct mbuf *m, uint8_t ulp_submode)
351 {
352
353 M_ASSERTPKTHDR(m);
354 m->m_pkthdr.PH_per.eight[0] = ulp_submode;
355 }
356
357 static inline uint8_t
mbuf_ulp_submode(struct mbuf * m)358 mbuf_ulp_submode(struct mbuf *m)
359 {
360
361 M_ASSERTPKTHDR(m);
362 return (m->m_pkthdr.PH_per.eight[0]);
363 }
364
365 static inline void
set_mbuf_iscsi_iso(struct mbuf * m,bool iso)366 set_mbuf_iscsi_iso(struct mbuf *m, bool iso)
367 {
368
369 M_ASSERTPKTHDR(m);
370 m->m_pkthdr.PH_per.eight[1] = iso;
371 }
372
373 static inline bool
mbuf_iscsi_iso(struct mbuf * m)374 mbuf_iscsi_iso(struct mbuf *m)
375 {
376
377 M_ASSERTPKTHDR(m);
378 return (m->m_pkthdr.PH_per.eight[1]);
379 }
380
381 /* Flags for iSCSI segmentation offload. */
382 #define CXGBE_ISO_TYPE(flags) ((flags) & 0x3)
383 #define CXGBE_ISO_F 0x4
384
385 static inline void
set_mbuf_iscsi_iso_flags(struct mbuf * m,uint8_t flags)386 set_mbuf_iscsi_iso_flags(struct mbuf *m, uint8_t flags)
387 {
388
389 M_ASSERTPKTHDR(m);
390 m->m_pkthdr.PH_per.eight[2] = flags;
391 }
392
393 static inline uint8_t
mbuf_iscsi_iso_flags(struct mbuf * m)394 mbuf_iscsi_iso_flags(struct mbuf *m)
395 {
396
397 M_ASSERTPKTHDR(m);
398 return (m->m_pkthdr.PH_per.eight[2]);
399 }
400
401 static inline void
set_mbuf_iscsi_iso_mss(struct mbuf * m,uint16_t mss)402 set_mbuf_iscsi_iso_mss(struct mbuf *m, uint16_t mss)
403 {
404
405 M_ASSERTPKTHDR(m);
406 m->m_pkthdr.PH_per.sixteen[2] = mss;
407 }
408
409 static inline uint16_t
mbuf_iscsi_iso_mss(struct mbuf * m)410 mbuf_iscsi_iso_mss(struct mbuf *m)
411 {
412
413 M_ASSERTPKTHDR(m);
414 return (m->m_pkthdr.PH_per.sixteen[2]);
415 }
416
417 /* t4_tom.c */
418 struct toepcb *alloc_toepcb(struct vi_info *, int);
419 int init_toepcb(struct vi_info *, struct toepcb *);
420 struct toepcb *hold_toepcb(struct toepcb *);
421 void free_toepcb(struct toepcb *);
422 void offload_socket(struct socket *, struct toepcb *);
423 void restore_so_proto(struct socket *, bool);
424 void undo_offload_socket(struct socket *);
425 void final_cpl_received(struct toepcb *);
426 void insert_tid(struct adapter *, int, void *, int);
427 void *lookup_tid(struct adapter *, int);
428 void update_tid(struct adapter *, int, void *);
429 void remove_tid(struct adapter *, int, int);
430 u_long select_rcv_wnd(struct socket *);
431 int select_rcv_wscale(void);
432 void init_conn_params(struct vi_info *, struct offload_settings *,
433 struct in_conninfo *, struct socket *, const struct tcp_options *, int16_t,
434 struct conn_params *cp);
435 __be64 calc_options0(struct vi_info *, struct conn_params *);
436 __be32 calc_options2(struct vi_info *, struct conn_params *);
437 uint64_t select_ntuple(struct vi_info *, struct l2t_entry *);
438 int negative_advice(int);
439 int add_tid_to_history(struct adapter *, u_int);
440
441 /* t4_connect.c */
442 void t4_init_connect_cpl_handlers(void);
443 void t4_uninit_connect_cpl_handlers(void);
444 int t4_connect(struct toedev *, struct socket *, struct nhop_object *,
445 struct sockaddr *);
446 void act_open_failure_cleanup(struct adapter *, u_int, u_int);
447
448 /* t4_listen.c */
449 void t4_init_listen_cpl_handlers(void);
450 void t4_uninit_listen_cpl_handlers(void);
451 int t4_listen_start(struct toedev *, struct tcpcb *);
452 int t4_listen_stop(struct toedev *, struct tcpcb *);
453 void t4_syncache_added(struct toedev *, void *);
454 void t4_syncache_removed(struct toedev *, void *);
455 int t4_syncache_respond(struct toedev *, void *, struct mbuf *);
456 int do_abort_req_synqe(struct sge_iq *, const struct rss_header *,
457 struct mbuf *);
458 int do_abort_rpl_synqe(struct sge_iq *, const struct rss_header *,
459 struct mbuf *);
460 void t4_offload_socket(struct toedev *, void *, struct socket *);
461 void synack_failure_cleanup(struct adapter *, int);
462
463 /* t4_cpl_io.c */
464 void aiotx_init_toep(struct toepcb *);
465 int t4_aio_queue_aiotx(struct socket *, struct kaiocb *);
466 void t4_init_cpl_io_handlers(void);
467 void t4_uninit_cpl_io_handlers(void);
468 void send_abort_rpl(struct adapter *, struct sge_ofld_txq *, int , int);
469 void send_flowc_wr(struct toepcb *, struct tcpcb *);
470 void send_reset(struct adapter *, struct toepcb *, uint32_t);
471 int send_rx_credits(struct adapter *, struct toepcb *, int);
472 void send_rx_modulate(struct adapter *, struct toepcb *);
473 void make_established(struct toepcb *, uint32_t, uint32_t, uint16_t);
474 int t4_close_conn(struct adapter *, struct toepcb *);
475 void t4_rcvd(struct toedev *, struct tcpcb *);
476 void t4_rcvd_locked(struct toedev *, struct tcpcb *);
477 int t4_tod_output(struct toedev *, struct tcpcb *);
478 int t4_send_fin(struct toedev *, struct tcpcb *);
479 int t4_send_rst(struct toedev *, struct tcpcb *);
480 void t4_set_tcb_field(struct adapter *, struct sge_wrq *, struct toepcb *,
481 uint16_t, uint64_t, uint64_t, int, int);
482 void t4_push_frames(struct adapter *, struct toepcb *, int);
483 void t4_push_pdus(struct adapter *, struct toepcb *, int);
484
485 /* t4_ddp.c */
486 int t4_init_ppod_region(struct ppod_region *, struct t4_range *, u_int,
487 const char *);
488 void t4_free_ppod_region(struct ppod_region *);
489 int t4_alloc_page_pods_for_ps(struct ppod_region *, struct pageset *);
490 int t4_alloc_page_pods_for_bio(struct ppod_region *, struct bio *,
491 struct ppod_reservation *);
492 int t4_alloc_page_pods_for_buf(struct ppod_region *, vm_offset_t, int,
493 struct ppod_reservation *);
494 int t4_alloc_page_pods_for_sgl(struct ppod_region *, struct ctl_sg_entry *, int,
495 struct ppod_reservation *);
496 int t4_write_page_pods_for_ps(struct adapter *, struct sge_wrq *, int,
497 struct pageset *);
498 int t4_write_page_pods_for_bio(struct adapter *, struct toepcb *,
499 struct ppod_reservation *, struct bio *, struct mbufq *);
500 int t4_write_page_pods_for_buf(struct adapter *, struct toepcb *,
501 struct ppod_reservation *, vm_offset_t, int, struct mbufq *);
502 int t4_write_page_pods_for_sgl(struct adapter *, struct toepcb *,
503 struct ppod_reservation *, struct ctl_sg_entry *, int, int, struct mbufq *);
504 void t4_free_page_pods(struct ppod_reservation *);
505 int t4_aio_queue_ddp(struct socket *, struct kaiocb *);
506 void t4_ddp_mod_load(void);
507 void t4_ddp_mod_unload(void);
508 void ddp_assert_empty(struct toepcb *);
509 void ddp_init_toep(struct toepcb *);
510 void ddp_uninit_toep(struct toepcb *);
511 void ddp_queue_toep(struct toepcb *);
512 void release_ddp_resources(struct toepcb *toep);
513 void handle_ddp_close(struct toepcb *, struct tcpcb *, uint32_t);
514 void handle_ddp_indicate(struct toepcb *);
515 void insert_ddp_data(struct toepcb *, uint32_t);
516 const struct offload_settings *lookup_offload_policy(struct adapter *, int,
517 struct mbuf *, uint16_t, struct inpcb *);
518
519 /* t4_tls.c */
520 bool can_tls_offload(struct adapter *);
521 void do_rx_data_tls(const struct cpl_rx_data *, struct toepcb *, struct mbuf *);
522 void t4_push_ktls(struct adapter *, struct toepcb *, int);
523 void t4_tls_mod_load(void);
524 void t4_tls_mod_unload(void);
525 void tls_detach(struct toepcb *);
526 void tls_establish(struct toepcb *);
527 void tls_init_toep(struct toepcb *);
528 void tls_stop_handshake_timer(struct toepcb *);
529 int tls_tx_key(struct toepcb *);
530 void tls_uninit_toep(struct toepcb *);
531 int tls_alloc_ktls(struct toepcb *, struct ktls_session *, int);
532
533 #endif
534