Print this page
11493 aggr needs support for multiple pseudo rx groups
Portions contributed by: Dan McDonald <danmcd@joyent.com>
Reviewed by: Patrick Mooney <patrick.mooney@joyent.com>
Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com>
Reviewed by: Robert Mustacchi <rm@joyent.com>
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/sys/aggr_impl.h
+++ new/usr/src/uts/common/sys/aggr_impl.h
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21 /*
22 22 * Copyright 2010 Sun Microsystems, Inc. All rights reserved.
23 23 * Use is subject to license terms.
24 24 * Copyright 2012 OmniTI Computer Consulting, Inc All rights reserved.
25 25 * Copyright 2018 Joyent, Inc.
26 26 */
27 27
28 28 #ifndef _SYS_AGGR_IMPL_H
29 29 #define _SYS_AGGR_IMPL_H
30 30
31 31 #include <sys/types.h>
32 32 #include <sys/cred.h>
33 33 #include <sys/mac_ether.h>
34 34 #include <sys/mac_provider.h>
35 35 #include <sys/mac_client.h>
36 36 #include <sys/mac_client_priv.h>
37 37 #include <sys/aggr_lacp.h>
38 38
39 39 #ifdef __cplusplus
40 40 extern "C" {
41 41 #endif
42 42
43 43 #ifdef _KERNEL
44 44
45 45 #define AGGR_MINOR_CTL 1 /* control interface minor */
46 46
47 47 /* flags for aggr_grp_modify() */
48 48 #define AGGR_MODIFY_POLICY 0x01
|
↓ open down ↓ |
48 lines elided |
↑ open up ↑ |
49 49 #define AGGR_MODIFY_MAC 0x02
50 50 #define AGGR_MODIFY_LACP_MODE 0x04
51 51 #define AGGR_MODIFY_LACP_TIMER 0x08
52 52
53 53 /*
54 54 * Possible value of aggr_rseudo_rx_ring_t.arr_flags. Set when the ring entry
55 55 * in the pseudo RX group is used.
56 56 */
57 57 #define MAC_PSEUDO_RING_INUSE 0x01
58 58
59 +#define MAX_GROUPS_PER_PORT 128
60 +
59 61 /*
60 62 * VLAN filters placed on the Rx pseudo group.
61 63 */
62 64 typedef struct aggr_vlan {
63 65 list_node_t av_link;
64 66 uint16_t av_vid; /* VLAN ID */
65 67 uint_t av_refs; /* num aggr clients using this VID */
66 68 } aggr_vlan_t;
67 69
68 70 typedef struct aggr_unicst_addr_s {
69 71 uint8_t aua_addr[ETHERADDRL];
70 72 struct aggr_unicst_addr_s *aua_next;
71 73 } aggr_unicst_addr_t;
72 74
73 75 typedef struct aggr_pseudo_rx_ring_s {
74 - mac_ring_handle_t arr_rh; /* filled in by aggr_fill_ring() */
75 - struct aggr_port_s *arr_port;
76 - mac_ring_handle_t arr_hw_rh;
77 - uint_t arr_flags;
78 - uint64_t arr_gen;
76 + mac_ring_handle_t arr_rh; /* set by aggr_fill_ring() */
77 + struct aggr_port_s *arr_port;
78 + struct aggr_pseudo_rx_group_s *arr_grp;
79 + mac_ring_handle_t arr_hw_rh;
80 + uint_t arr_flags;
81 + uint64_t arr_gen;
79 82 } aggr_pseudo_rx_ring_t;
80 83
84 +/*
85 + * An aggr pseudo group abstracts the underlying ports' HW groups. For
86 + * example, if each port has 8 groups (mac_group_t), then the aggr
87 + * will create 8 pseudo groups. Each pseudo group represents a
88 + * collection of HW groups: one group from each port. If you have
89 + * three ports then the pseudo group stands in for three HW groups.
90 + */
81 91 typedef struct aggr_pseudo_rx_group_s {
92 + uint_t arg_index;
82 93 struct aggr_grp_s *arg_grp; /* filled in by aggr_fill_group() */
83 94 mac_group_handle_t arg_gh; /* filled in by aggr_fill_group() */
84 95 aggr_unicst_addr_t *arg_macaddr;
85 96 aggr_pseudo_rx_ring_t arg_rings[MAX_RINGS_PER_GROUP];
86 97 uint_t arg_ring_cnt;
87 98 uint_t arg_untagged; /* num clients untagged */
88 99 list_t arg_vlans; /* VLANs on this group */
89 100 } aggr_pseudo_rx_group_t;
90 101
91 102 typedef struct aggr_pseudo_tx_ring_s {
92 103 mac_ring_handle_t atr_rh; /* filled in by aggr_fill_ring() */
93 104 struct aggr_port_s *atr_port;
94 105 mac_ring_handle_t atr_hw_rh;
95 106 uint_t atr_flags;
96 107 } aggr_pseudo_tx_ring_t;
97 108
98 109 typedef struct aggr_pseudo_tx_group_s {
99 110 mac_group_handle_t atg_gh; /* filled in by aggr_fill_group() */
100 111 uint_t atg_ring_cnt;
101 112 aggr_pseudo_tx_ring_t atg_rings[MAX_RINGS_PER_GROUP];
102 113 } aggr_pseudo_tx_group_t;
103 114
104 115 /*
105 116 * A link aggregation MAC port.
106 117 * Note that lp_next is protected by the lg_lock of the group the
107 118 * port is part of.
108 119 */
109 120 typedef struct aggr_port_s {
110 121 struct aggr_port_s *lp_next;
111 122 struct aggr_grp_s *lp_grp; /* back ptr to group */
|
↓ open down ↓ |
20 lines elided |
↑ open up ↑ |
112 123 datalink_id_t lp_linkid;
113 124 uint16_t lp_portid;
114 125 uint8_t lp_addr[ETHERADDRL]; /* port MAC address */
115 126 uint32_t lp_refs; /* refcount */
116 127 aggr_port_state_t lp_state;
117 128 uint32_t lp_started : 1,
118 129 lp_tx_enabled : 1,
119 130 lp_collector_enabled : 1,
120 131 lp_promisc_on : 1,
121 132 lp_no_link_update : 1,
122 - lp_rx_grp_added : 1,
123 133 lp_tx_grp_added : 1,
124 134 lp_closing : 1,
125 - lp_pad_bits : 24;
135 + lp_pad_bits : 25;
126 136 mac_handle_t lp_mh;
127 - mac_client_handle_t lp_mch;
137 +
138 + mac_client_handle_t lp_mch;
139 +
128 140 const mac_info_t *lp_mip;
129 141 mac_notify_handle_t lp_mnh;
130 142 uint_t lp_tx_idx; /* idx in group's tx array */
131 143 uint64_t lp_ifspeed;
132 144 link_state_t lp_link_state;
133 145 link_duplex_t lp_link_duplex;
134 146 uint64_t lp_stat[MAC_NSTAT];
135 147 uint64_t lp_ether_stat[ETHER_NSTAT];
136 148 aggr_lacp_port_t lp_lacp; /* LACP state */
137 149 lacp_stats_t lp_lacp_stats;
138 150 uint32_t lp_margin;
139 - mac_promisc_handle_t lp_mphp;
151 +
140 152 mac_unicast_handle_t lp_mah;
141 153
142 154 /* List of non-primary addresses that requires promiscous mode set */
143 155 aggr_unicst_addr_t *lp_prom_addr;
144 - /* handle of the underlying HW RX group */
145 - mac_group_handle_t lp_hwgh;
156 +
157 + /*
158 + * References to the underlying HW Rx groups of this port.
159 + * Used by aggr to program HW classification for the pseudo
160 + * groups.
161 + */
162 + mac_group_handle_t lp_hwghs[MAX_GROUPS_PER_PORT];
163 +
146 164 int lp_tx_ring_cnt;
147 165 /* handles of the underlying HW TX rings */
148 166 mac_ring_handle_t *lp_tx_rings;
149 167 /*
150 168 * Handles of the pseudo TX rings. Each of them maps to
151 169 * corresponding hardware TX ring in lp_tx_rings[]. A
152 170 * pseudo TX ring is presented to aggr primary mac
153 171 * client even when underlying NIC has no TX ring.
154 172 */
155 173 mac_ring_handle_t *lp_pseudo_tx_rings;
156 174 void *lp_tx_notify_mh;
157 175 } aggr_port_t;
158 176
159 177 /*
160 178 * A link aggregation group.
161 179 *
162 180 * The following per-group flags are defined:
163 181 *
164 182 * - lg_addr_fixed: set when the MAC address has been explicitely set
165 183 * when the group was created, or by a m_unicst_set() request.
166 184 * If this flag is not set, the MAC address of the group will be
167 185 * set to the first port that is added to the group.
168 186 *
169 187 * - lg_add_set: used only when lg_addr_fixed is not set. Captures whether
170 188 * the MAC address was initialized according to the members of the group.
171 189 * When set, the lg_port field points to the port from which the
172 190 * MAC address was initialized.
173 191 *
174 192 */
175 193 typedef struct aggr_grp_s {
176 194 datalink_id_t lg_linkid;
177 195 uint16_t lg_key; /* key (group port number) */
178 196 uint32_t lg_refs; /* refcount */
179 197 uint16_t lg_nports; /* number of MAC ports */
180 198 uint8_t lg_addr[ETHERADDRL]; /* group MAC address */
181 199 uint16_t
|
↓ open down ↓ |
26 lines elided |
↑ open up ↑ |
182 200 lg_closing : 1,
183 201 lg_addr_fixed : 1, /* fixed MAC address? */
184 202 lg_started : 1, /* group started? */
185 203 lg_promisc : 1, /* in promiscuous mode? */
186 204 lg_zcopy : 1,
187 205 lg_vlan : 1,
188 206 lg_force : 1,
189 207 lg_lso : 1,
190 208 lg_pad_bits : 8;
191 209 aggr_port_t *lg_ports; /* list of configured ports */
192 - aggr_port_t *lg_mac_addr_port;
210 + aggr_port_t *lg_mac_addr_port; /* using address of this port */
193 211 mac_handle_t lg_mh;
194 212 zoneid_t lg_zoneid;
195 213 uint_t lg_nattached_ports;
196 214 krwlock_t lg_tx_lock;
197 215 uint_t lg_ntx_ports;
198 216 aggr_port_t **lg_tx_ports; /* array of tx ports */
199 217 uint_t lg_tx_ports_size; /* size of lg_tx_ports */
200 218 uint32_t lg_tx_policy; /* outbound policy */
201 219 uint8_t lg_mac_tx_policy;
202 220 link_state_t lg_link_state;
203 221
204 222
205 223 /*
206 224 * The lg_stat_lock must be held when accessing these fields.
207 225 */
208 226 kmutex_t lg_stat_lock;
209 227 uint64_t lg_ifspeed;
210 228 link_duplex_t lg_link_duplex;
211 229 uint64_t lg_stat[MAC_NSTAT];
212 230 uint64_t lg_ether_stat[ETHER_NSTAT];
213 231
214 232 aggr_lacp_mode_t lg_lacp_mode; /* off, active, or passive */
215 233 Agg_t aggr; /* 802.3ad data */
216 234 uint32_t lg_hcksum_txflags;
217 235 uint_t lg_max_sdu;
218 236 uint32_t lg_margin;
219 237 mac_capab_lso_t lg_cap_lso;
220 238
221 239 /*
222 240 * The following fields are used by the LACP packets processing.
223 241 * Specifically, as the LACP packets processing is not performance
224 242 * critical, all LACP packets will be handled by a dedicated thread
225 243 * instead of in the mac_rx() call. This is to avoid the dead lock
|
↓ open down ↓ |
23 lines elided |
↑ open up ↑ |
226 244 * with mac_unicast_remove(), which holding the mac perimeter of the
227 245 * aggr, and wait for the mr_refcnt of the RX ring to drop to zero.
228 246 */
229 247 kmutex_t lg_lacp_lock;
230 248 kcondvar_t lg_lacp_cv;
231 249 mblk_t *lg_lacp_head;
232 250 mblk_t *lg_lacp_tail;
233 251 kthread_t *lg_lacp_rx_thread;
234 252 boolean_t lg_lacp_done;
235 253
236 - aggr_pseudo_rx_group_t lg_rx_group;
254 + uint_t lg_rx_group_count;
255 + aggr_pseudo_rx_group_t lg_rx_groups[MAX_GROUPS_PER_PORT];
256 +
237 257 aggr_pseudo_tx_group_t lg_tx_group;
238 258
239 259 kmutex_t lg_tx_flowctl_lock;
240 260 kcondvar_t lg_tx_flowctl_cv;
241 261 uint_t lg_tx_blocked_cnt;
242 262 mac_ring_handle_t *lg_tx_blocked_rings;
243 263 kthread_t *lg_tx_notify_thread;
244 264 boolean_t lg_tx_notify_done;
245 265
246 266 /*
247 267 * The following fields are used by aggr to wait for all the
248 268 * aggr_port_notify_cb() and aggr_port_timer_thread() to finish
249 269 * before it calls mac_unregister() when the aggr is deleted.
250 270 */
251 271 kmutex_t lg_port_lock;
252 272 kcondvar_t lg_port_cv;
253 273 int lg_port_ref;
254 274 } aggr_grp_t;
255 275
256 276 #define AGGR_GRP_REFHOLD(grp) { \
257 277 atomic_inc_32(&(grp)->lg_refs); \
258 278 ASSERT((grp)->lg_refs != 0); \
259 279 }
260 280
261 281 #define AGGR_GRP_REFRELE(grp) { \
262 282 ASSERT((grp)->lg_refs != 0); \
263 283 membar_exit(); \
264 284 if (atomic_dec_32_nv(&(grp)->lg_refs) == 0) \
265 285 aggr_grp_free(grp); \
266 286 }
267 287
268 288 #define AGGR_PORT_REFHOLD(port) { \
269 289 atomic_inc_32(&(port)->lp_refs); \
270 290 ASSERT((port)->lp_refs != 0); \
271 291 }
272 292
273 293 #define AGGR_PORT_REFRELE(port) { \
274 294 ASSERT((port)->lp_refs != 0); \
275 295 membar_exit(); \
276 296 if (atomic_dec_32_nv(&(port)->lp_refs) == 0) \
277 297 aggr_port_free(port); \
278 298 }
279 299
280 300 extern dev_info_t *aggr_dip;
281 301 extern int aggr_ioc_init(void);
282 302 extern void aggr_ioc_fini(void);
283 303
284 304 typedef int (*aggr_grp_info_new_grp_fn_t)(void *, datalink_id_t, uint32_t,
285 305 uchar_t *, boolean_t, boolean_t, uint32_t, uint32_t, aggr_lacp_mode_t,
286 306 aggr_lacp_timer_t);
287 307 typedef int (*aggr_grp_info_new_port_fn_t)(void *, datalink_id_t, uchar_t *,
288 308 aggr_port_state_t, aggr_lacp_state_t *);
289 309
290 310 extern void aggr_grp_init(void);
291 311 extern void aggr_grp_fini(void);
292 312 extern int aggr_grp_create(datalink_id_t, uint32_t, uint_t, laioc_port_t *,
293 313 uint32_t, boolean_t, boolean_t, uchar_t *, aggr_lacp_mode_t,
294 314 aggr_lacp_timer_t, cred_t *);
295 315 extern int aggr_grp_delete(datalink_id_t, cred_t *);
296 316 extern void aggr_grp_free(aggr_grp_t *);
297 317
298 318 extern int aggr_grp_info(datalink_id_t, void *, aggr_grp_info_new_grp_fn_t,
299 319 aggr_grp_info_new_port_fn_t, cred_t *);
300 320 extern void aggr_grp_notify(aggr_grp_t *, uint32_t);
301 321 extern boolean_t aggr_grp_attach_port(aggr_grp_t *, aggr_port_t *);
302 322 extern boolean_t aggr_grp_detach_port(aggr_grp_t *, aggr_port_t *);
303 323 extern void aggr_grp_port_mac_changed(aggr_grp_t *, aggr_port_t *,
304 324 boolean_t *, boolean_t *);
305 325 extern int aggr_grp_add_ports(datalink_id_t, uint_t, boolean_t,
306 326 laioc_port_t *);
307 327 extern int aggr_grp_rem_ports(datalink_id_t, uint_t, laioc_port_t *);
308 328 extern boolean_t aggr_grp_update_ports_mac(aggr_grp_t *);
309 329 extern int aggr_grp_modify(datalink_id_t, uint8_t, uint32_t, boolean_t,
310 330 const uchar_t *, aggr_lacp_mode_t, aggr_lacp_timer_t);
311 331 extern void aggr_grp_multicst_port(aggr_port_t *, boolean_t);
312 332 extern uint_t aggr_grp_count(void);
313 333 extern void aggr_grp_update_default(aggr_grp_t *);
314 334
315 335 extern void aggr_port_init(void);
316 336 extern void aggr_port_fini(void);
317 337 extern int aggr_port_create(aggr_grp_t *, const datalink_id_t, boolean_t,
318 338 aggr_port_t **);
319 339 extern void aggr_port_delete(aggr_port_t *);
320 340 extern void aggr_port_free(aggr_port_t *);
|
↓ open down ↓ |
74 lines elided |
↑ open up ↑ |
321 341 extern int aggr_port_start(aggr_port_t *);
322 342 extern void aggr_port_stop(aggr_port_t *);
323 343 extern int aggr_port_promisc(aggr_port_t *, boolean_t);
324 344 extern int aggr_port_unicst(aggr_port_t *);
325 345 extern int aggr_port_multicst(void *, boolean_t, const uint8_t *);
326 346 extern uint64_t aggr_port_stat(aggr_port_t *, uint_t);
327 347 extern boolean_t aggr_port_notify_link(aggr_grp_t *, aggr_port_t *);
328 348 extern void aggr_port_init_callbacks(aggr_port_t *);
329 349
330 350 extern void aggr_recv_cb(void *, mac_resource_handle_t, mblk_t *, boolean_t);
331 -extern void aggr_recv_promisc_cb(void *, mac_resource_handle_t, mblk_t *,
332 - boolean_t);
333 351
334 352 extern void aggr_tx_ring_update(void *, uintptr_t);
335 353 extern void aggr_tx_notify_thread(void *);
336 354 extern void aggr_send_port_enable(aggr_port_t *);
337 355 extern void aggr_send_port_disable(aggr_port_t *);
338 356 extern void aggr_send_update_policy(aggr_grp_t *, uint32_t);
339 357
340 358 extern void aggr_lacp_init(void);
341 359 extern void aggr_lacp_fini(void);
342 360 extern void aggr_lacp_init_port(aggr_port_t *);
343 361 extern void aggr_lacp_init_grp(aggr_grp_t *);
344 362 extern void aggr_lacp_set_mode(aggr_grp_t *, aggr_lacp_mode_t,
345 363 aggr_lacp_timer_t);
346 364 extern void aggr_lacp_update_mode(aggr_grp_t *, aggr_lacp_mode_t);
347 365 extern void aggr_lacp_update_timer(aggr_grp_t *, aggr_lacp_timer_t);
348 366 extern void aggr_lacp_rx_enqueue(aggr_port_t *, mblk_t *);
349 367 extern void aggr_lacp_port_attached(aggr_port_t *);
|
↓ open down ↓ |
7 lines elided |
↑ open up ↑ |
350 368 extern void aggr_lacp_port_detached(aggr_port_t *);
351 369 extern void aggr_port_lacp_set_mode(aggr_grp_t *, aggr_port_t *);
352 370
353 371 extern void aggr_lacp_rx_thread(void *);
354 372 extern void aggr_recv_lacp(aggr_port_t *, mac_resource_handle_t, mblk_t *);
355 373
356 374 extern void aggr_grp_port_hold(aggr_port_t *);
357 375 extern void aggr_grp_port_rele(aggr_port_t *);
358 376 extern void aggr_grp_port_wait(aggr_grp_t *);
359 377
360 -extern int aggr_port_addmac(aggr_port_t *, const uint8_t *);
361 -extern void aggr_port_remmac(aggr_port_t *, const uint8_t *);
378 +extern int aggr_port_addmac(aggr_port_t *, uint_t, const uint8_t *);
379 +extern void aggr_port_remmac(aggr_port_t *, uint_t, const uint8_t *);
362 380
363 -extern int aggr_port_addvlan(aggr_port_t *, uint16_t);
364 -extern int aggr_port_remvlan(aggr_port_t *, uint16_t);
381 +extern int aggr_port_addvlan(aggr_port_t *, uint_t, uint16_t);
382 +extern int aggr_port_remvlan(aggr_port_t *, uint_t, uint16_t);
365 383
366 384 extern mblk_t *aggr_ring_tx(void *, mblk_t *);
367 385 extern mblk_t *aggr_find_tx_ring(void *, mblk_t *,
368 386 uintptr_t, mac_ring_handle_t *);
369 387
370 388 #endif /* _KERNEL */
371 389
372 390 #ifdef __cplusplus
373 391 }
374 392 #endif
375 393
376 394 #endif /* _SYS_AGGR_IMPL_H */
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX