xref: /freebsd/sys/dev/cxgbe/tom/t4_tom.h (revision a7623790fb345e6dc986dfd31df0ace115e6f2e4)
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 	TPF_INITIALIZED    = (1 << 12), /* init_toepcb has been called */
77 	TPF_TLS_RECEIVE	   = (1 << 13), /* should receive TLS records */
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 sockopt;
91 struct offload_settings;
92 
93 /*
94  * Connection parameters for an offloaded connection.  These are mostly (but not
95  * all) hardware TOE parameters.
96  */
97 struct conn_params {
98 	int8_t rx_coalesce;
99 	int8_t cong_algo;
100 	int8_t tc_idx;
101 	int8_t tstamp;
102 	int8_t sack;
103 	int8_t nagle;
104 	int8_t keepalive;
105 	int8_t wscale;
106 	int8_t ecn;
107 	int8_t mtu_idx;
108 	int8_t ulp_mode;
109 	int8_t tx_align;
110 	int16_t txq_idx;	/* ofld_txq = &sc->sge.ofld_txq[txq_idx] */
111 	int16_t rxq_idx;	/* ofld_rxq = &sc->sge.ofld_rxq[rxq_idx] */
112 	int16_t l2t_idx;
113 	uint16_t emss;
114 	uint16_t opt0_bufsize;
115 	u_int sndbuf;		/* controls TP tx pages */
116 };
117 
118 struct ofld_tx_sdesc {
119 	uint32_t plen;		/* payload length */
120 	uint8_t tx_credits;	/* firmware tx credits (unit is 16B) */
121 	void *iv_buffer;	/* optional buffer holding IVs for TLS */
122 };
123 
124 struct ppod_region {
125 	u_int pr_start;
126 	u_int pr_len;
127 	u_int pr_page_shift[4];
128 	uint32_t pr_tag_mask;		/* hardware tagmask for this region. */
129 	uint32_t pr_invalid_bit;	/* OR with this to invalidate tag. */
130 	uint32_t pr_alias_mask;		/* AND with tag to get alias bits. */
131 	u_int pr_alias_shift;		/* shift this much for first alias bit. */
132 	vmem_t *pr_arena;
133 };
134 
135 struct ppod_reservation {
136 	struct ppod_region *prsv_pr;
137 	uint32_t prsv_tag;		/* Full tag: pgsz, alias, tag, color */
138 	u_int prsv_nppods;
139 };
140 
141 struct pageset {
142 	TAILQ_ENTRY(pageset) link;
143 	vm_page_t *pages;
144 	int npages;
145 	int flags;
146 	int offset;		/* offset in first page */
147 	int len;
148 	struct ppod_reservation prsv;
149 	struct vmspace *vm;
150 	vm_offset_t start;
151 	u_int vm_timestamp;
152 };
153 
154 TAILQ_HEAD(pagesetq, pageset);
155 
156 #define	PS_PPODS_WRITTEN	0x0001	/* Page pods written to the card. */
157 
158 struct ddp_buffer {
159 	struct pageset *ps;
160 
161 	struct kaiocb *job;
162 	int cancel_pending;
163 };
164 
165 struct ddp_pcb {
166 	u_int flags;
167 	struct ddp_buffer db[2];
168 	TAILQ_HEAD(, pageset) cached_pagesets;
169 	TAILQ_HEAD(, kaiocb) aiojobq;
170 	u_int waiting_count;
171 	u_int active_count;
172 	u_int cached_count;
173 	int active_id;	/* the currently active DDP buffer */
174 	struct task requeue_task;
175 	struct kaiocb *queueing;
176 	struct mtx lock;
177 };
178 
179 struct toepcb {
180 	struct tom_data *td;
181 	struct inpcb *inp;	/* backpointer to host stack's PCB */
182 	u_int flags;		/* miscellaneous flags */
183 	TAILQ_ENTRY(toepcb) link; /* toep_list */
184 	int refcount;
185 	struct vnet *vnet;
186 	struct vi_info *vi;	/* virtual interface */
187 	struct sge_wrq *ofld_txq;
188 	struct sge_ofld_rxq *ofld_rxq;
189 	struct sge_wrq *ctrlq;
190 	struct l2t_entry *l2te;	/* L2 table entry used by this connection */
191 	struct clip_entry *ce;	/* CLIP table entry used by this tid */
192 	int tid;		/* Connection identifier */
193 
194 	/* tx credit handling */
195 	u_int tx_total;		/* total tx WR credits (in 16B units) */
196 	u_int tx_credits;	/* tx WR credits (in 16B units) available */
197 	u_int tx_nocompl;	/* tx WR credits since last compl request */
198 	u_int plen_nocompl;	/* payload since last compl request */
199 
200 	struct conn_params params;
201 
202 	void *ulpcb;
203 	void *ulpcb2;
204 	struct mbufq ulp_pduq;	/* PDUs waiting to be sent out. */
205 	struct mbufq ulp_pdu_reclaimq;
206 
207 	struct ddp_pcb ddp;
208 	struct tls_ofld_info tls;
209 
210 	TAILQ_HEAD(, kaiocb) aiotx_jobq;
211 	struct task aiotx_task;
212 	struct socket *aiotx_so;
213 
214 	/* Tx software descriptor */
215 	uint8_t txsd_total;
216 	uint8_t txsd_pidx;
217 	uint8_t txsd_cidx;
218 	uint8_t txsd_avail;
219 	struct ofld_tx_sdesc txsd[];
220 };
221 
222 static inline int
223 ulp_mode(struct toepcb *toep)
224 {
225 
226 	return (toep->params.ulp_mode);
227 }
228 
229 #define	DDP_LOCK(toep)		mtx_lock(&(toep)->ddp.lock)
230 #define	DDP_UNLOCK(toep)	mtx_unlock(&(toep)->ddp.lock)
231 #define	DDP_ASSERT_LOCKED(toep)	mtx_assert(&(toep)->ddp.lock, MA_OWNED)
232 
233 /*
234  * Compressed state for embryonic connections for a listener.
235  */
236 struct synq_entry {
237 	struct listen_ctx *lctx;	/* backpointer to listen ctx */
238 	struct mbuf *syn;
239 	int flags;			/* same as toepcb's tp_flags */
240 	volatile int ok_to_respond;
241 	volatile u_int refcnt;
242 	int tid;
243 	uint32_t iss;
244 	uint32_t irs;
245 	uint32_t ts;
246 	uint32_t rss_hash;
247 	__be16 tcp_opt; /* from cpl_pass_establish */
248 	struct toepcb *toep;
249 
250 	struct conn_params params;
251 };
252 
253 /* listen_ctx flags */
254 #define LCTX_RPL_PENDING 1	/* waiting for a CPL_PASS_OPEN_RPL */
255 
256 struct listen_ctx {
257 	LIST_ENTRY(listen_ctx) link;	/* listen hash linkage */
258 	volatile int refcount;
259 	int stid;
260 	struct stid_region stid_region;
261 	int flags;
262 	struct inpcb *inp;		/* listening socket's inp */
263 	struct vnet *vnet;
264 	struct sge_wrq *ctrlq;
265 	struct sge_ofld_rxq *ofld_rxq;
266 	struct clip_entry *ce;
267 };
268 
269 /* tcb_histent flags */
270 #define TE_RPL_PENDING	1
271 #define TE_ACTIVE	2
272 
273 /* bits in one 8b tcb_histent sample. */
274 #define TS_RTO			(1 << 0)
275 #define TS_DUPACKS		(1 << 1)
276 #define TS_FASTREXMT		(1 << 2)
277 #define TS_SND_BACKLOGGED	(1 << 3)
278 #define TS_CWND_LIMITED		(1 << 4)
279 #define TS_ECN_ECE		(1 << 5)
280 #define TS_ECN_CWR		(1 << 6)
281 #define TS_RESERVED		(1 << 7)	/* Unused. */
282 
283 struct tcb_histent {
284 	struct mtx te_lock;
285 	struct callout te_callout;
286 	uint64_t te_tcb[TCB_SIZE / sizeof(uint64_t)];
287 	struct adapter *te_adapter;
288 	u_int te_flags;
289 	u_int te_tid;
290 	uint8_t te_pidx;
291 	uint8_t te_sample[100];
292 };
293 
294 struct tom_data {
295 	struct toedev tod;
296 
297 	/* toepcb's associated with this TOE device */
298 	struct mtx toep_list_lock;
299 	TAILQ_HEAD(, toepcb) toep_list;
300 
301 	struct mtx lctx_hash_lock;
302 	LIST_HEAD(, listen_ctx) *listen_hash;
303 	u_long listen_mask;
304 	int lctx_count;		/* # of lctx in the hash table */
305 
306 	struct ppod_region pr;
307 
308 	struct rwlock tcb_history_lock __aligned(CACHE_LINE_SIZE);
309 	struct tcb_histent **tcb_history;
310 	int dupack_threshold;
311 
312 	/* WRs that will not be sent to the chip because L2 resolution failed */
313 	struct mtx unsent_wr_lock;
314 	STAILQ_HEAD(, wrqe) unsent_wr_list;
315 	struct task reclaim_wr_resources;
316 };
317 
318 static inline struct tom_data *
319 tod_td(struct toedev *tod)
320 {
321 
322 	return (__containerof(tod, struct tom_data, tod));
323 }
324 
325 static inline struct adapter *
326 td_adapter(struct tom_data *td)
327 {
328 
329 	return (td->tod.tod_softc);
330 }
331 
332 static inline void
333 set_mbuf_ulp_submode(struct mbuf *m, uint8_t ulp_submode)
334 {
335 
336 	M_ASSERTPKTHDR(m);
337 	m->m_pkthdr.PH_per.eight[0] = ulp_submode;
338 }
339 
340 static inline uint8_t
341 mbuf_ulp_submode(struct mbuf *m)
342 {
343 
344 	M_ASSERTPKTHDR(m);
345 	return (m->m_pkthdr.PH_per.eight[0]);
346 }
347 
348 /* t4_tom.c */
349 struct toepcb *alloc_toepcb(struct vi_info *, int);
350 int init_toepcb(struct vi_info *, struct toepcb *);
351 struct toepcb *hold_toepcb(struct toepcb *);
352 void free_toepcb(struct toepcb *);
353 void offload_socket(struct socket *, struct toepcb *);
354 void undo_offload_socket(struct socket *);
355 void final_cpl_received(struct toepcb *);
356 void insert_tid(struct adapter *, int, void *, int);
357 void *lookup_tid(struct adapter *, int);
358 void update_tid(struct adapter *, int, void *);
359 void remove_tid(struct adapter *, int, int);
360 u_long select_rcv_wnd(struct socket *);
361 int select_rcv_wscale(void);
362 void init_conn_params(struct vi_info *, struct offload_settings *,
363     struct in_conninfo *, struct socket *, const struct tcp_options *, int16_t,
364     struct conn_params *cp);
365 __be64 calc_options0(struct vi_info *, struct conn_params *);
366 __be32 calc_options2(struct vi_info *, struct conn_params *);
367 uint64_t select_ntuple(struct vi_info *, struct l2t_entry *);
368 int negative_advice(int);
369 int add_tid_to_history(struct adapter *, u_int);
370 
371 /* t4_connect.c */
372 void t4_init_connect_cpl_handlers(void);
373 void t4_uninit_connect_cpl_handlers(void);
374 int t4_connect(struct toedev *, struct socket *, struct nhop_object *,
375     struct sockaddr *);
376 void act_open_failure_cleanup(struct adapter *, u_int, u_int);
377 
378 /* t4_listen.c */
379 void t4_init_listen_cpl_handlers(void);
380 void t4_uninit_listen_cpl_handlers(void);
381 int t4_listen_start(struct toedev *, struct tcpcb *);
382 int t4_listen_stop(struct toedev *, struct tcpcb *);
383 void t4_syncache_added(struct toedev *, void *);
384 void t4_syncache_removed(struct toedev *, void *);
385 int t4_syncache_respond(struct toedev *, void *, struct mbuf *);
386 int do_abort_req_synqe(struct sge_iq *, const struct rss_header *,
387     struct mbuf *);
388 int do_abort_rpl_synqe(struct sge_iq *, const struct rss_header *,
389     struct mbuf *);
390 void t4_offload_socket(struct toedev *, void *, struct socket *);
391 void synack_failure_cleanup(struct adapter *, int);
392 
393 /* t4_cpl_io.c */
394 void aiotx_init_toep(struct toepcb *);
395 int t4_aio_queue_aiotx(struct socket *, struct kaiocb *);
396 void t4_init_cpl_io_handlers(void);
397 void t4_uninit_cpl_io_handlers(void);
398 void send_abort_rpl(struct adapter *, struct sge_wrq *, int , int);
399 void send_flowc_wr(struct toepcb *, struct tcpcb *);
400 void send_reset(struct adapter *, struct toepcb *, uint32_t);
401 int send_rx_credits(struct adapter *, struct toepcb *, int);
402 void send_rx_modulate(struct adapter *, struct toepcb *);
403 void make_established(struct toepcb *, uint32_t, uint32_t, uint16_t);
404 int t4_close_conn(struct adapter *, struct toepcb *);
405 void t4_rcvd(struct toedev *, struct tcpcb *);
406 void t4_rcvd_locked(struct toedev *, struct tcpcb *);
407 int t4_tod_output(struct toedev *, struct tcpcb *);
408 int t4_send_fin(struct toedev *, struct tcpcb *);
409 int t4_send_rst(struct toedev *, struct tcpcb *);
410 void t4_set_tcb_field(struct adapter *, struct sge_wrq *, struct toepcb *,
411     uint16_t, uint64_t, uint64_t, int, int);
412 void t4_push_frames(struct adapter *, struct toepcb *, int);
413 void t4_push_pdus(struct adapter *, struct toepcb *, int);
414 
415 /* t4_ddp.c */
416 int t4_init_ppod_region(struct ppod_region *, struct t4_range *, u_int,
417     const char *);
418 void t4_free_ppod_region(struct ppod_region *);
419 int t4_alloc_page_pods_for_ps(struct ppod_region *, struct pageset *);
420 int t4_alloc_page_pods_for_buf(struct ppod_region *, vm_offset_t, int,
421     struct ppod_reservation *);
422 int t4_write_page_pods_for_ps(struct adapter *, struct sge_wrq *, int,
423     struct pageset *);
424 int t4_write_page_pods_for_buf(struct adapter *, struct sge_wrq *, int tid,
425     struct ppod_reservation *, vm_offset_t, int);
426 void t4_free_page_pods(struct ppod_reservation *);
427 int t4_soreceive_ddp(struct socket *, struct sockaddr **, struct uio *,
428     struct mbuf **, struct mbuf **, int *);
429 int t4_aio_queue_ddp(struct socket *, struct kaiocb *);
430 void t4_ddp_mod_load(void);
431 void t4_ddp_mod_unload(void);
432 void ddp_assert_empty(struct toepcb *);
433 void ddp_init_toep(struct toepcb *);
434 void ddp_uninit_toep(struct toepcb *);
435 void ddp_queue_toep(struct toepcb *);
436 void release_ddp_resources(struct toepcb *toep);
437 void handle_ddp_close(struct toepcb *, struct tcpcb *, uint32_t);
438 void handle_ddp_indicate(struct toepcb *);
439 void insert_ddp_data(struct toepcb *, uint32_t);
440 const struct offload_settings *lookup_offload_policy(struct adapter *, int,
441     struct mbuf *, uint16_t, struct inpcb *);
442 
443 /* t4_tls.c */
444 bool can_tls_offload(struct adapter *);
445 void do_rx_data_tls(const struct cpl_rx_data *, struct toepcb *, struct mbuf *);
446 int t4_ctloutput_tls(struct socket *, struct sockopt *);
447 void t4_push_tls_records(struct adapter *, struct toepcb *, int);
448 void t4_push_ktls(struct adapter *, struct toepcb *, int);
449 void t4_tls_mod_load(void);
450 void t4_tls_mod_unload(void);
451 void tls_establish(struct toepcb *);
452 void tls_init_toep(struct toepcb *);
453 int tls_rx_key(struct toepcb *);
454 void tls_stop_handshake_timer(struct toepcb *);
455 int tls_tx_key(struct toepcb *);
456 void tls_uninit_toep(struct toepcb *);
457 int tls_alloc_ktls(struct toepcb *, struct ktls_session *, int);
458 
459 #endif
460