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 <[email protected]>
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 };
76
77 enum {
78 DDP_OK = (1 << 0), /* OK to turn on DDP */
79 DDP_SC_REQ = (1 << 1), /* state change (on/off) requested */
80 DDP_ON = (1 << 2), /* DDP is turned on */
81 DDP_BUF0_ACTIVE = (1 << 3), /* buffer 0 in use (not invalidated) */
82 DDP_BUF1_ACTIVE = (1 << 4), /* buffer 1 in use (not invalidated) */
83 DDP_TASK_ACTIVE = (1 << 5), /* requeue task is queued / running */
84 DDP_DEAD = (1 << 6), /* toepcb is shutting down */
85 };
86
87 struct sockopt;
88 struct offload_settings;
89
90 /*
91 * Connection parameters for an offloaded connection. These are mostly (but not
92 * all) hardware TOE parameters.
93 */
94 struct conn_params {
95 int8_t rx_coalesce;
96 int8_t cong_algo;
97 int8_t tc_idx;
98 int8_t tstamp;
99 int8_t sack;
100 int8_t nagle;
101 int8_t keepalive;
102 int8_t wscale;
103 int8_t ecn;
104 int8_t mtu_idx;
105 int8_t ulp_mode;
106 int8_t tx_align;
107 int16_t txq_idx; /* ofld_txq = &sc->sge.ofld_txq[txq_idx] */
108 int16_t rxq_idx; /* ofld_rxq = &sc->sge.ofld_rxq[rxq_idx] */
109 int16_t l2t_idx;
110 uint16_t emss;
111 uint16_t opt0_bufsize;
112 u_int sndbuf; /* controls TP tx pages */
113 };
114
115 struct ofld_tx_sdesc {
116 uint32_t plen; /* payload length */
117 uint8_t tx_credits; /* firmware tx credits (unit is 16B) */
118 void *iv_buffer; /* optional buffer holding IVs for TLS */
119 };
120
121 struct ppod_region {
122 u_int pr_start;
123 u_int pr_len;
124 u_int pr_page_shift[4];
125 uint32_t pr_tag_mask; /* hardware tagmask for this region. */
126 uint32_t pr_invalid_bit; /* OR with this to invalidate tag. */
127 uint32_t pr_alias_mask; /* AND with tag to get alias bits. */
128 u_int pr_alias_shift; /* shift this much for first alias bit. */
129 vmem_t *pr_arena;
130 };
131
132 struct ppod_reservation {
133 struct ppod_region *prsv_pr;
134 uint32_t prsv_tag; /* Full tag: pgsz, alias, tag, color */
135 u_int prsv_nppods;
136 };
137
138 struct pageset {
139 TAILQ_ENTRY(pageset) link;
140 vm_page_t *pages;
141 int npages;
142 int flags;
143 int offset; /* offset in first page */
144 int len;
145 struct ppod_reservation prsv;
146 struct vmspace *vm;
147 vm_offset_t start;
148 u_int vm_timestamp;
149 };
150
151 TAILQ_HEAD(pagesetq, pageset);
152
153 #define PS_WIRED 0x0001 /* Pages wired rather than held. */
154 #define PS_PPODS_WRITTEN 0x0002 /* Page pods written to the card. */
155
156 #define EXT_FLAG_AIOTX EXT_FLAG_VENDOR1
157
158 #define IS_AIOTX_MBUF(m) \
159 ((m)->m_flags & M_EXT && (m)->m_ext.ext_flags & EXT_FLAG_AIOTX)
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 aiotx_buffer {
183 struct pageset ps;
184 struct kaiocb *job;
185 int refcount;
186 };
187
188 struct toepcb {
189 TAILQ_ENTRY(toepcb) link; /* toep_list */
190 u_int flags; /* miscellaneous flags */
191 int refcount;
192 struct tom_data *td;
193 struct inpcb *inp; /* backpointer to host stack's PCB */
194 struct vnet *vnet;
195 struct vi_info *vi; /* virtual interface */
196 struct sge_wrq *ofld_txq;
197 struct sge_ofld_rxq *ofld_rxq;
198 struct sge_wrq *ctrlq;
199 struct l2t_entry *l2te; /* L2 table entry used by this connection */
200 struct clip_entry *ce; /* CLIP table entry used by this tid */
201 int tid; /* Connection identifier */
202
203 /* tx credit handling */
204 u_int tx_total; /* total tx WR credits (in 16B units) */
205 u_int tx_credits; /* tx WR credits (in 16B units) available */
206 u_int tx_nocompl; /* tx WR credits since last compl request */
207 u_int plen_nocompl; /* payload since last compl request */
208
209 struct conn_params params;
210
211 void *ulpcb;
212 void *ulpcb2;
213 struct mbufq ulp_pduq; /* PDUs waiting to be sent out. */
214 struct mbufq ulp_pdu_reclaimq;
215
216 struct ddp_pcb ddp;
217 struct tls_ofld_info tls;
218
219 TAILQ_HEAD(, kaiocb) aiotx_jobq;
220 struct task aiotx_task;
221 struct socket *aiotx_so;
222
223 /* Tx software descriptor */
224 uint8_t txsd_total;
225 uint8_t txsd_pidx;
226 uint8_t txsd_cidx;
227 uint8_t txsd_avail;
228 struct ofld_tx_sdesc txsd[];
229 };
230
231 static inline int
ulp_mode(struct toepcb * toep)232 ulp_mode(struct toepcb *toep)
233 {
234
235 return (toep->params.ulp_mode);
236 }
237
238 #define DDP_LOCK(toep) mtx_lock(&(toep)->ddp.lock)
239 #define DDP_UNLOCK(toep) mtx_unlock(&(toep)->ddp.lock)
240 #define DDP_ASSERT_LOCKED(toep) mtx_assert(&(toep)->ddp.lock, MA_OWNED)
241
242 /*
243 * Compressed state for embryonic connections for a listener.
244 */
245 struct synq_entry {
246 struct listen_ctx *lctx; /* backpointer to listen ctx */
247 struct mbuf *syn;
248 int flags; /* same as toepcb's tp_flags */
249 volatile int ok_to_respond;
250 volatile u_int refcnt;
251 int tid;
252 uint32_t iss;
253 uint32_t irs;
254 uint32_t ts;
255 __be16 tcp_opt; /* from cpl_pass_establish */
256 struct toepcb *toep;
257
258 struct conn_params params;
259 };
260
261 /* listen_ctx flags */
262 #define LCTX_RPL_PENDING 1 /* waiting for a CPL_PASS_OPEN_RPL */
263
264 struct listen_ctx {
265 LIST_ENTRY(listen_ctx) link; /* listen hash linkage */
266 volatile int refcount;
267 int stid;
268 struct stid_region stid_region;
269 int flags;
270 struct inpcb *inp; /* listening socket's inp */
271 struct vnet *vnet;
272 struct sge_wrq *ctrlq;
273 struct sge_ofld_rxq *ofld_rxq;
274 struct clip_entry *ce;
275 };
276
277 /* tcb_histent flags */
278 #define TE_RPL_PENDING 1
279 #define TE_ACTIVE 2
280
281 /* bits in one 8b tcb_histent sample. */
282 #define TS_RTO (1 << 0)
283 #define TS_DUPACKS (1 << 1)
284 #define TS_FASTREXMT (1 << 2)
285 #define TS_SND_BACKLOGGED (1 << 3)
286 #define TS_CWND_LIMITED (1 << 4)
287 #define TS_ECN_ECE (1 << 5)
288 #define TS_ECN_CWR (1 << 6)
289 #define TS_RESERVED (1 << 7) /* Unused. */
290
291 struct tcb_histent {
292 struct mtx te_lock;
293 struct callout te_callout;
294 uint64_t te_tcb[TCB_SIZE / sizeof(uint64_t)];
295 struct adapter *te_adapter;
296 u_int te_flags;
297 u_int te_tid;
298 uint8_t te_pidx;
299 uint8_t te_sample[100];
300 };
301
302 struct tom_data {
303 struct toedev tod;
304
305 /* toepcb's associated with this TOE device */
306 struct mtx toep_list_lock;
307 TAILQ_HEAD(, toepcb) toep_list;
308
309 struct mtx lctx_hash_lock;
310 LIST_HEAD(, listen_ctx) *listen_hash;
311 u_long listen_mask;
312 int lctx_count; /* # of lctx in the hash table */
313
314 struct ppod_region pr;
315
316 struct rwlock tcb_history_lock __aligned(CACHE_LINE_SIZE);
317 struct tcb_histent **tcb_history;
318 int dupack_threshold;
319
320 /* WRs that will not be sent to the chip because L2 resolution failed */
321 struct mtx unsent_wr_lock;
322 STAILQ_HEAD(, wrqe) unsent_wr_list;
323 struct task reclaim_wr_resources;
324 };
325
326 static inline struct tom_data *
tod_td(struct toedev * tod)327 tod_td(struct toedev *tod)
328 {
329
330 return (__containerof(tod, struct tom_data, tod));
331 }
332
333 static inline struct adapter *
td_adapter(struct tom_data * td)334 td_adapter(struct tom_data *td)
335 {
336
337 return (td->tod.tod_softc);
338 }
339
340 static inline void
set_mbuf_ulp_submode(struct mbuf * m,uint8_t ulp_submode)341 set_mbuf_ulp_submode(struct mbuf *m, uint8_t ulp_submode)
342 {
343
344 M_ASSERTPKTHDR(m);
345 m->m_pkthdr.PH_per.eight[0] = ulp_submode;
346 }
347
348 static inline uint8_t
mbuf_ulp_submode(struct mbuf * m)349 mbuf_ulp_submode(struct mbuf *m)
350 {
351
352 M_ASSERTPKTHDR(m);
353 return (m->m_pkthdr.PH_per.eight[0]);
354 }
355
356 /* t4_tom.c */
357 struct toepcb *alloc_toepcb(struct vi_info *, int);
358 int init_toepcb(struct vi_info *, struct toepcb *);
359 struct toepcb *hold_toepcb(struct toepcb *);
360 void free_toepcb(struct toepcb *);
361 void offload_socket(struct socket *, struct toepcb *);
362 void undo_offload_socket(struct socket *);
363 void final_cpl_received(struct toepcb *);
364 void insert_tid(struct adapter *, int, void *, int);
365 void *lookup_tid(struct adapter *, int);
366 void update_tid(struct adapter *, int, void *);
367 void remove_tid(struct adapter *, int, int);
368 u_long select_rcv_wnd(struct socket *);
369 int select_rcv_wscale(void);
370 void init_conn_params(struct vi_info *, struct offload_settings *,
371 struct in_conninfo *, struct socket *, const struct tcp_options *, int16_t,
372 struct conn_params *cp);
373 __be64 calc_options0(struct vi_info *, struct conn_params *);
374 __be32 calc_options2(struct vi_info *, struct conn_params *);
375 uint64_t select_ntuple(struct vi_info *, struct l2t_entry *);
376 int negative_advice(int);
377 int add_tid_to_history(struct adapter *, u_int);
378
379 /* t4_connect.c */
380 void t4_init_connect_cpl_handlers(void);
381 void t4_uninit_connect_cpl_handlers(void);
382 int t4_connect(struct toedev *, struct socket *, struct rtentry *,
383 struct sockaddr *);
384 void act_open_failure_cleanup(struct adapter *, u_int, u_int);
385
386 /* t4_listen.c */
387 void t4_init_listen_cpl_handlers(void);
388 void t4_uninit_listen_cpl_handlers(void);
389 int t4_listen_start(struct toedev *, struct tcpcb *);
390 int t4_listen_stop(struct toedev *, struct tcpcb *);
391 void t4_syncache_added(struct toedev *, void *);
392 void t4_syncache_removed(struct toedev *, void *);
393 int t4_syncache_respond(struct toedev *, void *, struct mbuf *);
394 int do_abort_req_synqe(struct sge_iq *, const struct rss_header *,
395 struct mbuf *);
396 int do_abort_rpl_synqe(struct sge_iq *, const struct rss_header *,
397 struct mbuf *);
398 void t4_offload_socket(struct toedev *, void *, struct socket *);
399 void synack_failure_cleanup(struct adapter *, int);
400
401 /* t4_cpl_io.c */
402 void aiotx_init_toep(struct toepcb *);
403 int t4_aio_queue_aiotx(struct socket *, struct kaiocb *);
404 void t4_init_cpl_io_handlers(void);
405 void t4_uninit_cpl_io_handlers(void);
406 void send_abort_rpl(struct adapter *, struct sge_wrq *, int , int);
407 void send_flowc_wr(struct toepcb *, struct tcpcb *);
408 void send_reset(struct adapter *, struct toepcb *, uint32_t);
409 int send_rx_credits(struct adapter *, struct toepcb *, int);
410 void send_rx_modulate(struct adapter *, struct toepcb *);
411 void make_established(struct toepcb *, uint32_t, uint32_t, uint16_t);
412 int t4_close_conn(struct adapter *, struct toepcb *);
413 void t4_rcvd(struct toedev *, struct tcpcb *);
414 void t4_rcvd_locked(struct toedev *, struct tcpcb *);
415 int t4_tod_output(struct toedev *, struct tcpcb *);
416 int t4_send_fin(struct toedev *, struct tcpcb *);
417 int t4_send_rst(struct toedev *, struct tcpcb *);
418 void t4_set_tcb_field(struct adapter *, struct sge_wrq *, struct toepcb *,
419 uint16_t, uint64_t, uint64_t, int, int);
420 void t4_push_frames(struct adapter *, struct toepcb *, int);
421 void t4_push_pdus(struct adapter *, struct toepcb *, int);
422
423 /* t4_ddp.c */
424 int t4_init_ppod_region(struct ppod_region *, struct t4_range *, u_int,
425 const char *);
426 void t4_free_ppod_region(struct ppod_region *);
427 int t4_alloc_page_pods_for_ps(struct ppod_region *, struct pageset *);
428 int t4_alloc_page_pods_for_buf(struct ppod_region *, vm_offset_t, int,
429 struct ppod_reservation *);
430 int t4_write_page_pods_for_ps(struct adapter *, struct sge_wrq *, int,
431 struct pageset *);
432 int t4_write_page_pods_for_buf(struct adapter *, struct sge_wrq *, int tid,
433 struct ppod_reservation *, vm_offset_t, int);
434 void t4_free_page_pods(struct ppod_reservation *);
435 int t4_soreceive_ddp(struct socket *, struct sockaddr **, struct uio *,
436 struct mbuf **, struct mbuf **, int *);
437 int t4_aio_queue_ddp(struct socket *, struct kaiocb *);
438 void t4_ddp_mod_load(void);
439 void t4_ddp_mod_unload(void);
440 void ddp_assert_empty(struct toepcb *);
441 void ddp_init_toep(struct toepcb *);
442 void ddp_uninit_toep(struct toepcb *);
443 void ddp_queue_toep(struct toepcb *);
444 void release_ddp_resources(struct toepcb *toep);
445 void handle_ddp_close(struct toepcb *, struct tcpcb *, uint32_t);
446 void handle_ddp_indicate(struct toepcb *);
447 void insert_ddp_data(struct toepcb *, uint32_t);
448 const struct offload_settings *lookup_offload_policy(struct adapter *, int,
449 struct mbuf *, uint16_t, struct inpcb *);
450
451 /* t4_tls.c */
452 bool can_tls_offload(struct adapter *);
453 int t4_ctloutput_tls(struct socket *, struct sockopt *);
454 void t4_push_tls_records(struct adapter *, struct toepcb *, int);
455 void t4_tls_mod_load(void);
456 void t4_tls_mod_unload(void);
457 void tls_establish(struct toepcb *);
458 void tls_init_toep(struct toepcb *);
459 int tls_rx_key(struct toepcb *);
460 void tls_stop_handshake_timer(struct toepcb *);
461 int tls_tx_key(struct toepcb *);
462 void tls_uninit_toep(struct toepcb *);
463
464 #endif
465