xref: /freebsd/sys/dev/cxgbe/tom/t4_tom.h (revision 5c831a5bd61576cacb48b39f8eeb47b92707a355)
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 #define PPOD_SZ(n)	((n) * sizeof(struct pagepod))
55 #define PPOD_SIZE	(PPOD_SZ(1))
56 
57 /* TOE PCB flags */
58 enum {
59 	TPF_ATTACHED	   = (1 << 0),	/* a tcpcb refers to this toepcb */
60 	TPF_FLOWC_WR_SENT  = (1 << 1),	/* firmware flow context WR sent */
61 	TPF_TX_DATA_SENT   = (1 << 2),	/* some data sent */
62 	TPF_TX_SUSPENDED   = (1 << 3),	/* tx suspended for lack of resources */
63 	TPF_SEND_FIN	   = (1 << 4),	/* send FIN after all pending data */
64 	TPF_FIN_SENT	   = (1 << 5),	/* FIN has been sent */
65 	TPF_ABORT_SHUTDOWN = (1 << 6),	/* connection abort is in progress */
66 	TPF_CPL_PENDING    = (1 << 7),	/* haven't received the last CPL */
67 	TPF_SYNQE	   = (1 << 8),	/* synq_entry, not really a toepcb */
68 	TPF_SYNQE_NEEDFREE = (1 << 9),	/* synq_entry was malloc'd separately */
69 	TPF_SYNQE_TCPDDP   = (1 << 10),	/* ulp_mode TCPDDP in toepcb */
70 	TPF_SYNQE_EXPANDED = (1 << 11),	/* toepcb ready, tid context updated */
71 	TPF_SYNQE_HAS_L2TE = (1 << 12),	/* we've replied to PASS_ACCEPT_REQ */
72 };
73 
74 enum {
75 	DDP_OK		= (1 << 0),	/* OK to turn on DDP */
76 	DDP_SC_REQ	= (1 << 1),	/* state change (on/off) requested */
77 	DDP_ON		= (1 << 2),	/* DDP is turned on */
78 	DDP_BUF0_ACTIVE	= (1 << 3),	/* buffer 0 in use (not invalidated) */
79 	DDP_BUF1_ACTIVE	= (1 << 4),	/* buffer 1 in use (not invalidated) */
80 	DDP_TASK_ACTIVE = (1 << 5),	/* requeue task is queued / running */
81 	DDP_DEAD	= (1 << 6),	/* toepcb is shutting down */
82 };
83 
84 struct ofld_tx_sdesc {
85 	uint32_t plen;		/* payload length */
86 	uint8_t tx_credits;	/* firmware tx credits (unit is 16B) */
87 };
88 
89 struct ppod_region {
90 	u_int pr_start;
91 	u_int pr_len;
92 	u_int pr_page_shift[4];
93 	uint32_t pr_tag_mask;		/* hardware tagmask for this region. */
94 	uint32_t pr_invalid_bit;	/* OR with this to invalidate tag. */
95 	uint32_t pr_alias_mask;		/* AND with tag to get alias bits. */
96 	u_int pr_alias_shift;		/* shift this much for first alias bit. */
97 	vmem_t *pr_arena;
98 };
99 
100 struct ppod_reservation {
101 	struct ppod_region *prsv_pr;
102 	uint32_t prsv_tag;		/* Full tag: pgsz, alias, tag, color */
103 	u_int prsv_nppods;
104 };
105 
106 struct pageset {
107 	TAILQ_ENTRY(pageset) link;
108 	vm_page_t *pages;
109 	int npages;
110 	int flags;
111 	int offset;		/* offset in first page */
112 	int len;
113 	struct ppod_reservation prsv;
114 	struct vmspace *vm;
115 	vm_offset_t start;
116 	u_int vm_timestamp;
117 };
118 
119 TAILQ_HEAD(pagesetq, pageset);
120 
121 #define	PS_WIRED		0x0001	/* Pages wired rather than held. */
122 #define	PS_PPODS_WRITTEN	0x0002	/* Page pods written to the card. */
123 
124 #define	EXT_FLAG_AIOTX		EXT_FLAG_VENDOR1
125 
126 struct ddp_buffer {
127 	struct pageset *ps;
128 
129 	struct kaiocb *job;
130 	int cancel_pending;
131 };
132 
133 struct aiotx_buffer {
134 	struct pageset ps;
135 	struct kaiocb *job;
136 	int refcount;
137 };
138 
139 struct toepcb {
140 	TAILQ_ENTRY(toepcb) link; /* toep_list */
141 	u_int flags;		/* miscellaneous flags */
142 	int refcount;
143 	struct tom_data *td;
144 	struct inpcb *inp;	/* backpointer to host stack's PCB */
145 	struct vnet *vnet;
146 	struct vi_info *vi;	/* virtual interface */
147 	struct sge_wrq *ofld_txq;
148 	struct sge_ofld_rxq *ofld_rxq;
149 	struct sge_wrq *ctrlq;
150 	struct l2t_entry *l2te;	/* L2 table entry used by this connection */
151 	struct clip_entry *ce;	/* CLIP table entry used by this tid */
152 	int tid;		/* Connection identifier */
153 	int tc_idx;		/* traffic class that this tid is bound to */
154 
155 	/* tx credit handling */
156 	u_int tx_total;		/* total tx WR credits (in 16B units) */
157 	u_int tx_credits;	/* tx WR credits (in 16B units) available */
158 	u_int tx_nocompl;	/* tx WR credits since last compl request */
159 	u_int plen_nocompl;	/* payload since last compl request */
160 
161 	/* rx credit handling */
162 	u_int sb_cc;		/* last noted value of so_rcv->sb_cc */
163 	int rx_credits;		/* rx credits (in bytes) to be returned to hw */
164 
165 	u_int ulp_mode;	/* ULP mode */
166 	void *ulpcb;
167 	void *ulpcb2;
168 	struct mbufq ulp_pduq;	/* PDUs waiting to be sent out. */
169 	struct mbufq ulp_pdu_reclaimq;
170 
171 	u_int ddp_flags;
172 	struct ddp_buffer db[2];
173 	TAILQ_HEAD(, pageset) ddp_cached_pagesets;
174 	TAILQ_HEAD(, kaiocb) ddp_aiojobq;
175 	u_int ddp_waiting_count;
176 	u_int ddp_active_count;
177 	u_int ddp_cached_count;
178 	int ddp_active_id;	/* the currently active DDP buffer */
179 	struct task ddp_requeue_task;
180 	struct kaiocb *ddp_queueing;
181 	struct mtx ddp_lock;
182 
183 	TAILQ_HEAD(, kaiocb) aiotx_jobq;
184 	struct task aiotx_task;
185 	bool aiotx_task_active;
186 
187 	/* Tx software descriptor */
188 	uint8_t txsd_total;
189 	uint8_t txsd_pidx;
190 	uint8_t txsd_cidx;
191 	uint8_t txsd_avail;
192 	struct ofld_tx_sdesc txsd[];
193 };
194 
195 #define	DDP_LOCK(toep)		mtx_lock(&(toep)->ddp_lock)
196 #define	DDP_UNLOCK(toep)	mtx_unlock(&(toep)->ddp_lock)
197 #define	DDP_ASSERT_LOCKED(toep)	mtx_assert(&(toep)->ddp_lock, MA_OWNED)
198 
199 struct flowc_tx_params {
200 	uint32_t snd_nxt;
201 	uint32_t rcv_nxt;
202 	unsigned int snd_space;
203 	unsigned int mss;
204 };
205 
206 #define	DDP_RETRY_WAIT	5	/* seconds to wait before re-enabling DDP */
207 #define	DDP_LOW_SCORE	1
208 #define	DDP_HIGH_SCORE	3
209 
210 /*
211  * Compressed state for embryonic connections for a listener.  Barely fits in
212  * 64B, try not to grow it further.
213  */
214 struct synq_entry {
215 	TAILQ_ENTRY(synq_entry) link;	/* listen_ctx's synq link */
216 	int flags;			/* same as toepcb's tp_flags */
217 	int tid;
218 	struct listen_ctx *lctx;	/* backpointer to listen ctx */
219 	struct mbuf *syn;
220 	uint32_t iss;
221 	uint32_t ts;
222 	volatile uintptr_t wr;
223 	volatile u_int refcnt;
224 	uint16_t l2e_idx;
225 	uint16_t rcv_bufsize;
226 };
227 
228 /* listen_ctx flags */
229 #define LCTX_RPL_PENDING 1	/* waiting for a CPL_PASS_OPEN_RPL */
230 
231 struct listen_ctx {
232 	LIST_ENTRY(listen_ctx) link;	/* listen hash linkage */
233 	volatile int refcount;
234 	int stid;
235 	struct stid_region stid_region;
236 	int flags;
237 	struct inpcb *inp;		/* listening socket's inp */
238 	struct vnet *vnet;
239 	struct sge_wrq *ctrlq;
240 	struct sge_ofld_rxq *ofld_rxq;
241 	struct clip_entry *ce;
242 	TAILQ_HEAD(, synq_entry) synq;
243 };
244 
245 struct clip_entry {
246 	TAILQ_ENTRY(clip_entry) link;
247 	struct in6_addr lip;	/* local IPv6 address */
248 	u_int refcount;
249 };
250 
251 TAILQ_HEAD(clip_head, clip_entry);
252 struct tom_data {
253 	struct toedev tod;
254 
255 	/* toepcb's associated with this TOE device */
256 	struct mtx toep_list_lock;
257 	TAILQ_HEAD(, toepcb) toep_list;
258 
259 	struct mtx lctx_hash_lock;
260 	LIST_HEAD(, listen_ctx) *listen_hash;
261 	u_long listen_mask;
262 	int lctx_count;		/* # of lctx in the hash table */
263 
264 	struct ppod_region pr;
265 
266 	struct mtx clip_table_lock;
267 	struct clip_head clip_table;
268 	int clip_gen;
269 
270 	/* WRs that will not be sent to the chip because L2 resolution failed */
271 	struct mtx unsent_wr_lock;
272 	STAILQ_HEAD(, wrqe) unsent_wr_list;
273 	struct task reclaim_wr_resources;
274 };
275 
276 static inline struct tom_data *
277 tod_td(struct toedev *tod)
278 {
279 
280 	return (__containerof(tod, struct tom_data, tod));
281 }
282 
283 static inline struct adapter *
284 td_adapter(struct tom_data *td)
285 {
286 
287 	return (td->tod.tod_softc);
288 }
289 
290 static inline void
291 set_mbuf_ulp_submode(struct mbuf *m, uint8_t ulp_submode)
292 {
293 
294 	M_ASSERTPKTHDR(m);
295 	m->m_pkthdr.PH_per.eight[0] = ulp_submode;
296 }
297 
298 static inline uint8_t
299 mbuf_ulp_submode(struct mbuf *m)
300 {
301 
302 	M_ASSERTPKTHDR(m);
303 	return (m->m_pkthdr.PH_per.eight[0]);
304 }
305 
306 /* t4_tom.c */
307 struct toepcb *alloc_toepcb(struct vi_info *, int, int, int);
308 struct toepcb *hold_toepcb(struct toepcb *);
309 void free_toepcb(struct toepcb *);
310 void offload_socket(struct socket *, struct toepcb *);
311 void undo_offload_socket(struct socket *);
312 void final_cpl_received(struct toepcb *);
313 void insert_tid(struct adapter *, int, void *, int);
314 void *lookup_tid(struct adapter *, int);
315 void update_tid(struct adapter *, int, void *);
316 void remove_tid(struct adapter *, int, int);
317 void release_tid(struct adapter *, int, struct sge_wrq *);
318 int find_best_mtu_idx(struct adapter *, struct in_conninfo *, int);
319 u_long select_rcv_wnd(struct socket *);
320 int select_rcv_wscale(void);
321 uint64_t calc_opt0(struct socket *, struct vi_info *, struct l2t_entry *,
322     int, int, int, int);
323 uint64_t select_ntuple(struct vi_info *, struct l2t_entry *);
324 void set_tcpddp_ulp_mode(struct toepcb *);
325 int negative_advice(int);
326 struct clip_entry *hold_lip(struct tom_data *, struct in6_addr *,
327     struct clip_entry *);
328 void release_lip(struct tom_data *, struct clip_entry *);
329 
330 /* t4_connect.c */
331 void t4_init_connect_cpl_handlers(void);
332 void t4_uninit_connect_cpl_handlers(void);
333 int t4_connect(struct toedev *, struct socket *, struct rtentry *,
334     struct sockaddr *);
335 void act_open_failure_cleanup(struct adapter *, u_int, u_int);
336 
337 /* t4_listen.c */
338 void t4_init_listen_cpl_handlers(void);
339 void t4_uninit_listen_cpl_handlers(void);
340 int t4_listen_start(struct toedev *, struct tcpcb *);
341 int t4_listen_stop(struct toedev *, struct tcpcb *);
342 void t4_syncache_added(struct toedev *, void *);
343 void t4_syncache_removed(struct toedev *, void *);
344 int t4_syncache_respond(struct toedev *, void *, struct mbuf *);
345 int do_abort_req_synqe(struct sge_iq *, const struct rss_header *,
346     struct mbuf *);
347 int do_abort_rpl_synqe(struct sge_iq *, const struct rss_header *,
348     struct mbuf *);
349 void t4_offload_socket(struct toedev *, void *, struct socket *);
350 
351 /* t4_cpl_io.c */
352 void aiotx_init_toep(struct toepcb *);
353 int t4_aio_queue_aiotx(struct socket *, struct kaiocb *);
354 void t4_init_cpl_io_handlers(void);
355 void t4_uninit_cpl_io_handlers(void);
356 void send_abort_rpl(struct adapter *, struct sge_wrq *, int , int);
357 void send_flowc_wr(struct toepcb *, struct flowc_tx_params *);
358 void send_reset(struct adapter *, struct toepcb *, uint32_t);
359 void make_established(struct toepcb *, uint32_t, uint32_t, uint16_t);
360 void t4_rcvd(struct toedev *, struct tcpcb *);
361 void t4_rcvd_locked(struct toedev *, struct tcpcb *);
362 int t4_tod_output(struct toedev *, struct tcpcb *);
363 int t4_send_fin(struct toedev *, struct tcpcb *);
364 int t4_send_rst(struct toedev *, struct tcpcb *);
365 void t4_set_tcb_field(struct adapter *, struct sge_wrq *, int, uint16_t,
366     uint64_t, uint64_t, int, int, int);
367 void t4_push_frames(struct adapter *sc, struct toepcb *toep, int drop);
368 void t4_push_pdus(struct adapter *sc, struct toepcb *toep, int drop);
369 int do_set_tcb_rpl(struct sge_iq *, const struct rss_header *, struct mbuf *);
370 
371 /* t4_ddp.c */
372 int t4_init_ppod_region(struct ppod_region *, struct t4_range *, u_int,
373     const char *);
374 void t4_free_ppod_region(struct ppod_region *);
375 int t4_alloc_page_pods_for_ps(struct ppod_region *, struct pageset *);
376 int t4_alloc_page_pods_for_buf(struct ppod_region *, vm_offset_t, int,
377     struct ppod_reservation *);
378 int t4_write_page_pods_for_ps(struct adapter *, struct sge_wrq *, int,
379     struct pageset *);
380 int t4_write_page_pods_for_buf(struct adapter *, struct sge_wrq *, int tid,
381     struct ppod_reservation *, vm_offset_t, int);
382 void t4_free_page_pods(struct ppod_reservation *);
383 int t4_soreceive_ddp(struct socket *, struct sockaddr **, struct uio *,
384     struct mbuf **, struct mbuf **, int *);
385 int t4_aio_queue_ddp(struct socket *, struct kaiocb *);
386 int t4_ddp_mod_load(void);
387 void t4_ddp_mod_unload(void);
388 void ddp_assert_empty(struct toepcb *);
389 void ddp_init_toep(struct toepcb *);
390 void ddp_uninit_toep(struct toepcb *);
391 void ddp_queue_toep(struct toepcb *);
392 void release_ddp_resources(struct toepcb *toep);
393 void handle_ddp_close(struct toepcb *, struct tcpcb *, uint32_t);
394 void handle_ddp_indicate(struct toepcb *);
395 void handle_ddp_tcb_rpl(struct toepcb *, const struct cpl_set_tcb_rpl *);
396 void insert_ddp_data(struct toepcb *, uint32_t);
397 
398 #endif
399