1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
24 */
25
26 /*
27 * Copyright 2019, Joyent, Inc.
28 */
29
30 #ifndef _SYS_IB_EOIB_ENX_IMPL_H
31 #define _SYS_IB_EOIB_ENX_IMPL_H
32
33 #ifdef __cplusplus
34 extern "C" {
35 #endif
36
37 #include <sys/ddi.h>
38 #include <sys/sunddi.h>
39 #include <sys/varargs.h>
40 #include <sys/ib/ibtl/ibti.h>
41 #include <sys/ib/ibtl/ibvti.h>
42 #include <sys/ib/ib_pkt_hdrs.h>
43 #include <sys/ib/ibtl/impl/ibtl_ibnex.h>
44 #include <sys/ib/mgt/sm_attr.h>
45
46 #include <sys/ib/clients/eoib/fip.h>
47 #include <sys/ib/clients/eoib/eib.h>
48
49 /*
50 * Driver specific constants
51 */
52 #define ENX_E_SUCCESS 0
53 #define ENX_E_FAILURE -1
54 #define ENX_MAX_LINE 128
55 #define ENX_GRH_SZ (sizeof (ib_grh_t))
56
57 /*
58 * Debug messages
59 */
60 #define ENX_MSGS_CRIT 0x01
61 #define ENX_MSGS_ERR 0x02
62 #define ENX_MSGS_WARN 0x04
63 #define ENX_MSGS_DEBUG 0x08
64 #define ENX_MSGS_ARGS 0x10
65 #define ENX_MSGS_VERBOSE 0x20
66 #define ENX_MSGS_DEFAULT (ENX_MSGS_CRIT | ENX_MSGS_ERR | ENX_MSGS_WARN)
67
68 #define ENX_LOGSZ_DEFAULT 0x20000
69
70 #define ENX_DPRINTF_CRIT eibnx_dprintf_crit
71 #define ENX_DPRINTF_ERR eibnx_dprintf_err
72 #define ENX_DPRINTF_WARN eibnx_dprintf_warn
73 #ifdef ENX_DEBUG
74 #define ENX_DPRINTF_DEBUG eibnx_dprintf_debug
75 #define ENX_DPRINTF_ARGS eibnx_dprintf_args
76 #define ENX_DPRINTF_VERBOSE eibnx_dprintf_verbose
77 #else
78 #define ENX_DPRINTF_DEBUG(...)
79 #define ENX_DPRINTF_ARGS(...)
80 #define ENX_DPRINTF_VERBOSE(...)
81 #endif
82
83 /*
84 * EoIB Nexus service threads
85 */
86 #define ENX_PORT_MONITOR "eibnx_port_%d_monitor"
87 #define ENX_NODE_CREATOR "eibnx_node_creator"
88
89 /*
90 * Default period (us) for unicast solicitations to discovered gateways.
91 * EoIB specification requires that hosts send solicitation atleast every
92 * 4 * GW_ADV_PERIOD.
93 */
94 #define ENX_DFL_SOLICIT_PERIOD_USEC 32000000
95
96 /*
97 * Portinfo list per HCA
98 */
99 typedef struct eibnx_port_s {
100 struct eibnx_port_s *po_next;
101 ibt_hca_portinfo_t *po_pi;
102 uint_t po_pi_size;
103 } eibnx_port_t;
104
105 /*
106 * HCA details
107 */
108 typedef struct eibnx_hca_s {
109 struct eibnx_hca_s *hc_next;
110 ib_guid_t hc_guid;
111 ibt_hca_hdl_t hc_hdl;
112 ibt_pd_hdl_t hc_pd;
113 eibnx_port_t *hc_port;
114 } eibnx_hca_t;
115
116 /*
117 * The port_monitor thread in EoIB nexus driver only sends two types of
118 * packets: multicast solicitation the first time around, and periodic
119 * unicast solicitations later to gateways that have been discovered. So
120 * we need a couple of send wqes for the multicast solicitation and
121 * probably as many send wqes as the number of gateways that may be
122 * discovered from each port, for sending the unicast solicitations.
123 * For unicast solicitations though, the UD destination needs to be set
124 * up at the time we receive the advertisement from the gateway, using
125 * ibt_modify_reply_ud_dest(), so we'll assign one send wqe for each
126 * gateway that we discover. This means that we need to acquire these
127 * send wqe entries during rx processing in the completion handler, which
128 * means we must avoid sleeping in trying to acquire the swqe. Therefore,
129 * we'll pre-allocate these unicast solication send wqes to be atleast
130 * twice the number of recv wqes.
131 *
132 * The receive packets expected by the EoIB nexus driver are the multicast
133 * and unicast messages on the SOLICIT and ADVERTISE groups. These
134 * shouldn't be too many, and should be tuned as we gain experience on
135 * the traffic pattern. We'll start with 16.
136 */
137 #define ENX_NUM_SWQE 46
138 #define ENX_NUM_RWQE 16
139 #define ENX_CQ_SIZE (ENX_NUM_SWQE + ENX_NUM_RWQE + 2)
140
141 /*
142 * qe_type values
143 */
144 #define ENX_QETYP_RWQE 0x1
145 #define ENX_QETYP_SWQE 0x2
146
147 /*
148 * qe_flags bitmasks (protected by qe_lock). None of the
149 * flag values may be zero.
150 */
151 #define ENX_QEFL_INUSE 0x01
152 #define ENX_QEFL_POSTED 0x02
153 #define ENX_QEFL_RELONCOMP 0x04
154
155 /*
156 * Recv and send workq entries
157 */
158 typedef struct eibnx_wqe_s {
159 uint_t qe_type;
160 uint_t qe_bufsz;
161 ibt_wr_ds_t qe_sgl;
162 ibt_all_wr_t qe_wr;
163 kmutex_t qe_lock;
164 uint_t qe_flags;
165 } eibnx_wqe_t;
166
167 /*
168 * Tx descriptor
169 */
170 typedef struct eibnx_tx_s {
171 ib_vaddr_t tx_vaddr;
172 ibt_mr_hdl_t tx_mr;
173 ibt_lkey_t tx_lkey;
174 eibnx_wqe_t tx_wqe[ENX_NUM_SWQE];
175 } eibnx_tx_t;
176
177 /*
178 * Rx descriptor
179 */
180 typedef struct eibnx_rx_s {
181 ib_vaddr_t rx_vaddr;
182 ibt_mr_hdl_t rx_mr;
183 ibt_lkey_t rx_lkey;
184 eibnx_wqe_t rx_wqe[ENX_NUM_RWQE];
185 } eibnx_rx_t;
186
187 /*
188 * Details about the address of each gateway we discover.
189 */
190 typedef struct eibnx_gw_addr_s {
191 ibt_adds_vect_t *ga_vect;
192 ib_gid_t ga_gid;
193 ib_qpn_t ga_qpn;
194 ib_qkey_t ga_qkey;
195 ib_pkey_t ga_pkey;
196 } eibnx_gw_addr_t;
197
198 /*
199 * States for each GW
200 */
201 #define ENX_GW_STATE_UNAVAILABLE 1 /* GW nackd availability */
202 #define ENX_GW_STATE_AVAILABLE 2 /* GW mcasted availability */
203 #define ENX_GW_STATE_READY_TO_LOGIN 3 /* GW ucasted availability */
204
205 typedef struct eibnx_gw_info_s {
206 struct eibnx_gw_info_s *gw_next;
207 eibnx_wqe_t *gw_swqe;
208 uint_t gw_state;
209
210 kmutex_t gw_adv_lock;
211 uint_t gw_adv_flag;
212 int64_t gw_adv_last_lbolt;
213 int64_t gw_adv_timeout_ticks;
214
215 eibnx_gw_addr_t gw_addr;
216
217 ib_guid_t gw_system_guid;
218 ib_guid_t gw_guid;
219
220 uint32_t gw_adv_period;
221 uint32_t gw_ka_period;
222 uint32_t gw_vnic_ka_period;
223 ib_qpn_t gw_ctrl_qpn;
224
225 ib_lid_t gw_lid;
226 uint16_t gw_portid;
227 uint16_t gw_num_net_vnics;
228
229 uint8_t gw_is_host_adm_vnics;
230 uint8_t gw_sl;
231 uint8_t gw_n_rss_qpn;
232 uint8_t gw_flag_ucast_advt;
233 uint8_t gw_flag_available;
234
235 uint8_t gw_system_name[EIB_GW_SYSNAME_LEN];
236 uint8_t gw_port_name[EIB_GW_PORTNAME_LEN];
237 uint8_t gw_vendor_id[EIB_GW_VENDOR_LEN];
238 } eibnx_gw_info_t;
239
240 /*
241 * Values for gw_adv_flag (non-zero only)
242 */
243 #define ENX_GW_DEAD 1
244 #define ENX_GW_ALIVE 2
245 #define ENX_GW_AWARE 3
246
247 /*
248 * Currently, we only expect the advertisement type of packets
249 * from the gw. But we do get login acks from the gateway also
250 * here in the nexus, so we'll need an identifier for that.
251 */
252 typedef enum {
253 FIP_GW_ADVERTISE_MCAST = 0,
254 FIP_GW_ADVERTISE_UCAST,
255 FIP_VNIC_LOGIN_ACK
256 } eibnx_gw_pkt_type_t;
257
258 /*
259 * Currently, the only gw response handled by the eibnx driver
260 * are the ucast/mcast advertisements. Information collected from
261 * both these responses may be packed into a eibnx_gw_info_t.
262 * In the future, if we decide to handle other types of responses
263 * from the gw, we could simply add the new types to the union.
264 */
265 typedef struct eibnx_gw_msg_s {
266 eibnx_gw_pkt_type_t gm_type;
267 union {
268 eibnx_gw_info_t gm_info;
269 } u;
270 } eibnx_gw_msg_t;
271
272 /*
273 * List to hold the devinfo nodes of eoib instances
274 */
275 typedef struct eibnx_child_s {
276 struct eibnx_child_s *ch_next;
277 dev_info_t *ch_dip;
278 eibnx_gw_info_t *ch_gwi;
279 char *ch_node_name;
280 } eibnx_child_t;
281
282 /*
283 * Event bitmasks for the port-monitor to wait on. None of these flags
284 * may be zero.
285 */
286 #define ENX_EVENT_LINK_UP 0x01
287 #define ENX_EVENT_MCGS_AVAILABLE 0x02
288 #define ENX_EVENT_TIMED_OUT 0x04
289 #define ENX_EVENT_DIE 0x08
290 #define ENX_EVENT_COMPLETION 0x10
291
292 /*
293 * MCG Query/Join status
294 */
295 #define ENX_MCGS_FOUND 0x1
296 #define ENX_MCGS_JOINED 0x2
297
298 /*
299 * Information that each port-monitor thread cares about
300 */
301 typedef struct eibnx_thr_info_s {
302 struct eibnx_thr_info_s *ti_next;
303 uint_t ti_progress;
304
305 /*
306 * Our kernel thread id
307 */
308 kt_did_t ti_kt_did;
309
310 /*
311 * HCA, port and protection domain information
312 */
313 ib_guid_t ti_hca_guid;
314 ibt_hca_hdl_t ti_hca;
315 ibt_pd_hdl_t ti_pd;
316 ibt_hca_portinfo_t *ti_pi;
317 char *ti_ident;
318
319 /*
320 * Well-known multicast groups for solicitations
321 * and advertisements.
322 */
323 kmutex_t ti_mcg_lock;
324 uint_t ti_mcg_status;
325 ibt_mcg_info_t *ti_advertise_mcg;
326 ibt_mcg_info_t *ti_solicit_mcg;
327 uint_t ti_mcast_done;
328
329 /*
330 * Completion queue stuff
331 */
332 ibt_cq_hdl_t ti_cq_hdl;
333 uint_t ti_cq_sz;
334 ibt_wc_t *ti_wc;
335 ddi_softint_handle_t ti_softint_hdl;
336
337 /*
338 * Channel related
339 */
340 ibt_channel_hdl_t ti_chan;
341 ib_qpn_t ti_qpn;
342
343 /*
344 * Transmit/Receive stuff
345 */
346 eibnx_tx_t ti_snd;
347 eibnx_rx_t ti_rcv;
348
349 /*
350 * GW related stuff
351 */
352 kmutex_t ti_gw_lock;
353 eibnx_gw_info_t *ti_gw;
354
355 /*
356 * Devinfo nodes for the eoib children
357 */
358 kmutex_t ti_child_lock;
359 eibnx_child_t *ti_child;
360
361 /*
362 * Events that we wait on and/or handle
363 */
364 kmutex_t ti_event_lock;
365 kcondvar_t ti_event_cv;
366 uint_t ti_event;
367 } eibnx_thr_info_t;
368
369 /*
370 * Workq entry for creation of eoib nodes
371 */
372 typedef struct eibnx_nodeq_s {
373 struct eibnx_nodeq_s *nc_next;
374 eibnx_thr_info_t *nc_info;
375 eibnx_gw_info_t *nc_gwi;
376 } eibnx_nodeq_t;
377
378 /*
379 * Bus config status flags. The in-prog is protected by
380 * nx_lock, and the rest of the flags (currently only
381 * buscfg-complete) is protected by the in-prog bit itself.
382 */
383 #define NX_FL_BUSOP_INPROG 0x1
384 #define NX_FL_BUSCFG_COMPLETE 0x2
385 #define NX_FL_BUSOP_MASK 0x3
386
387 /*
388 * EoIB nexus per-instance state
389 */
390 typedef struct eibnx_s {
391 dev_info_t *nx_dip;
392 ibt_clnt_hdl_t nx_ibt_hdl;
393
394 kmutex_t nx_lock;
395 eibnx_hca_t *nx_hca;
396 eibnx_thr_info_t *nx_thr_info;
397 boolean_t nx_monitors_up;
398
399 kmutex_t nx_nodeq_lock;
400 kcondvar_t nx_nodeq_cv;
401 eibnx_nodeq_t *nx_nodeq;
402 kt_did_t nx_nodeq_kt_did;
403 uint_t nx_nodeq_thr_die;
404
405 kmutex_t nx_busop_lock;
406 kcondvar_t nx_busop_cv;
407 uint_t nx_busop_flags;
408 } eibnx_t;
409
410
411 /*
412 * Event tags for EoIB Nexus events delivered to EoIB instances
413 */
414 #define ENX_EVENT_TAG_GW_INFO_UPDATE 0
415 #define ENX_EVENT_TAG_GW_AVAILABLE 1
416 #define ENX_EVENT_TAG_LOGIN_ACK 2
417
418 /*
419 * FUNCTION PROTOTYPES FOR CROSS-FILE LINKAGE
420 */
421
422 /*
423 * Threads and Event Handlers
424 */
425 void eibnx_port_monitor(eibnx_thr_info_t *);
426 void eibnx_subnet_notices_handler(void *, ib_gid_t, ibt_subnet_event_code_t,
427 ibt_subnet_event_t *);
428 void eibnx_async_handler(void *, ibt_hca_hdl_t, ibt_async_code_t,
429 ibt_async_event_t *);
430 boolean_t eibnx_is_gw_dead(eibnx_gw_info_t *);
431 void eibnx_create_eoib_node(void);
432 void eibnx_comp_intr(ibt_cq_hdl_t, void *);
433 uint_t eibnx_comp_handler(caddr_t, caddr_t);
434
435 /*
436 * IBT related functions
437 */
438 int eibnx_ibt_init(eibnx_t *);
439 int eibnx_find_mgroups(eibnx_thr_info_t *);
440 int eibnx_setup_cq(eibnx_thr_info_t *);
441 int eibnx_setup_ud_channel(eibnx_thr_info_t *);
442 int eibnx_setup_bufs(eibnx_thr_info_t *);
443 int eibnx_setup_cq_handler(eibnx_thr_info_t *);
444 int eibnx_join_mcgs(eibnx_thr_info_t *);
445 int eibnx_rejoin_mcgs(eibnx_thr_info_t *);
446 int eibnx_ibt_fini(eibnx_t *);
447
448 void eibnx_rb_find_mgroups(eibnx_thr_info_t *);
449 void eibnx_rb_setup_cq(eibnx_thr_info_t *);
450 void eibnx_rb_setup_ud_channel(eibnx_thr_info_t *);
451 void eibnx_rb_setup_bufs(eibnx_thr_info_t *);
452 void eibnx_rb_setup_cq_handler(eibnx_thr_info_t *);
453 void eibnx_rb_join_mcgs(eibnx_thr_info_t *);
454
455 eibnx_hca_t *eibnx_prepare_hca(ib_guid_t);
456 int eibnx_cleanup_hca(eibnx_hca_t *);
457
458 /*
459 * FIP packetizing related functions
460 */
461 int eibnx_fip_solicit_mcast(eibnx_thr_info_t *);
462 int eibnx_fip_solicit_ucast(eibnx_thr_info_t *, clock_t *);
463 int eibnx_fip_parse_pkt(uint8_t *, eibnx_gw_msg_t *);
464
465 /*
466 * Queue and List related routines
467 */
468 eibnx_wqe_t *eibnx_acquire_swqe(eibnx_thr_info_t *, int);
469 void eibnx_return_swqe(eibnx_wqe_t *);
470 void eibnx_return_rwqe(eibnx_thr_info_t *, eibnx_wqe_t *);
471 void eibnx_release_swqe(eibnx_wqe_t *);
472
473 void eibnx_enqueue_child(eibnx_thr_info_t *, eibnx_gw_info_t *, char *,
474 dev_info_t *);
475 int eibnx_update_child(eibnx_thr_info_t *, eibnx_gw_info_t *, dev_info_t *);
476 dev_info_t *eibnx_find_child_dip_by_inst(eibnx_thr_info_t *, int);
477 dev_info_t *eibnx_find_child_dip_by_gw(eibnx_thr_info_t *, uint16_t);
478
479 eibnx_gw_info_t *eibnx_find_gw_in_gwlist(eibnx_thr_info_t *, eibnx_gw_info_t *);
480 eibnx_gw_info_t *eibnx_add_gw_to_gwlist(eibnx_thr_info_t *, eibnx_gw_info_t *,
481 ibt_wc_t *, uint8_t *);
482 void eibnx_replace_gw_in_gwlist(eibnx_thr_info_t *, eibnx_gw_info_t *,
483 eibnx_gw_info_t *, ibt_wc_t *, uint8_t *, boolean_t *);
484 void eibnx_queue_for_creation(eibnx_thr_info_t *, eibnx_gw_info_t *);
485
486 /*
487 * Logging and Error reporting routines
488 */
489 void eibnx_debug_init(void);
490 void eibnx_debug_fini(void);
491 void eibnx_dprintf_crit(const char *fmt, ...);
492 void eibnx_dprintf_err(const char *fmt, ...);
493 void eibnx_dprintf_warn(const char *fmt, ...);
494 #ifdef ENX_DEBUG
495 void eibnx_dprintf_debug(const char *fmt, ...);
496 void eibnx_dprintf_args(const char *fmt, ...);
497 void eibnx_dprintf_verbose(const char *fmt, ...);
498 #endif
499
500 /*
501 * Miscellaneous
502 */
503 void eibnx_cleanup_port_nodes(eibnx_thr_info_t *);
504 void eibnx_create_node_props(dev_info_t *, eibnx_thr_info_t *,
505 eibnx_gw_info_t *);
506 int eibnx_name_child(dev_info_t *, char *, size_t);
507 void eibnx_busop_inprog_enter(eibnx_t *);
508 void eibnx_busop_inprog_exit(eibnx_t *);
509 eibnx_thr_info_t *eibnx_start_port_monitor(eibnx_hca_t *, eibnx_port_t *);
510 void eibnx_stop_port_monitor(eibnx_thr_info_t *);
511 void eibnx_terminate_monitors(void);
512 int eibnx_configure_node(eibnx_thr_info_t *, eibnx_gw_info_t *, dev_info_t **);
513 int eibnx_unconfigure_node(eibnx_thr_info_t *, eibnx_gw_info_t *);
514 int eibnx_locate_node_name(char *, eibnx_thr_info_t **, eibnx_gw_info_t **);
515 int eibnx_locate_unconfigured_node(eibnx_thr_info_t **, eibnx_gw_info_t **);
516
517 /*
518 * Devctl cbops (currently dummy)
519 */
520 int eibnx_devctl_open(dev_t *, int, int, cred_t *);
521 int eibnx_devctl_close(dev_t, int, int, cred_t *);
522 int eibnx_devctl_ioctl(dev_t, int, intptr_t, int, cred_t *, int *);
523
524 /*
525 * External variable references
526 */
527 extern pri_t minclsyspri;
528 extern eibnx_t *enx_global_ss;
529 extern ib_gid_t enx_solicit_mgid;
530 extern ib_gid_t enx_advertise_mgid;
531
532 #ifdef __cplusplus
533 }
534 #endif
535
536 #endif /* _SYS_IB_EOIB_ENX_IMPL_H */