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) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 */
25
26 /*
27 * MAC Services Module
28 *
29 * The GLDv3 framework locking - The MAC layer
30 * --------------------------------------------
31 *
32 * The MAC layer is central to the GLD framework and can provide the locking
33 * framework needed for itself and for the use of MAC clients. MAC end points
34 * are fairly disjoint and don't share a lot of state. So a coarse grained
35 * multi-threading scheme is to single thread all create/modify/delete or set
36 * type of control operations on a per mac end point while allowing data threads
37 * concurrently.
38 *
39 * Control operations (set) that modify a mac end point are always serialized on
40 * a per mac end point basis, We have at most 1 such thread per mac end point
41 * at a time.
42 *
43 * All other operations that are not serialized are essentially multi-threaded.
44 * For example a control operation (get) like getting statistics which may not
45 * care about reading values atomically or data threads sending or receiving
46 * data. Mostly these type of operations don't modify the control state. Any
47 * state these operations care about are protected using traditional locks.
48 *
49 * The perimeter only serializes serial operations. It does not imply there
50 * aren't any other concurrent operations. However a serialized operation may
51 * sometimes need to make sure it is the only thread. In this case it needs
52 * to use reference counting mechanisms to cv_wait until any current data
53 * threads are done.
54 *
55 * The mac layer itself does not hold any locks across a call to another layer.
56 * The perimeter is however held across a down call to the driver to make the
57 * whole control operation atomic with respect to other control operations.
58 * Also the data path and get type control operations may proceed concurrently.
59 * These operations synchronize with the single serial operation on a given mac
60 * end point using regular locks. The perimeter ensures that conflicting
61 * operations like say a mac_multicast_add and a mac_multicast_remove on the
62 * same mac end point don't interfere with each other and also ensures that the
63 * changes in the mac layer and the call to the underlying driver to say add a
64 * multicast address are done atomically without interference from a thread
65 * trying to delete the same address.
66 *
67 * For example, consider
68 * mac_multicst_add()
69 * {
70 * mac_perimeter_enter(); serialize all control operations
71 *
72 * grab list lock protect against access by data threads
73 * add to list
74 * drop list lock
75 *
76 * call driver's mi_multicst
77 *
78 * mac_perimeter_exit();
79 * }
80 *
81 * To lessen the number of serialization locks and simplify the lock hierarchy,
82 * we serialize all the control operations on a per mac end point by using a
83 * single serialization lock called the perimeter. We allow recursive entry into
84 * the perimeter to facilitate use of this mechanism by both the mac client and
85 * the MAC layer itself.
86 *
87 * MAC client means an entity that does an operation on a mac handle
88 * obtained from a mac_open/mac_client_open. Similarly MAC driver means
89 * an entity that does an operation on a mac handle obtained from a
90 * mac_register. An entity could be both client and driver but on different
91 * handles eg. aggr. and should only make the corresponding mac interface calls
92 * i.e. mac driver interface or mac client interface as appropriate for that
93 * mac handle.
94 *
95 * General rules.
96 * -------------
97 *
98 * R1. The lock order of upcall threads is natually opposite to downcall
99 * threads. Hence upcalls must not hold any locks across layers for fear of
100 * recursive lock enter and lock order violation. This applies to all layers.
101 *
102 * R2. The perimeter is just another lock. Since it is held in the down
103 * direction, acquiring the perimeter in an upcall is prohibited as it would
104 * cause a deadlock. This applies to all layers.
105 *
106 * Note that upcalls that need to grab the mac perimeter (for example
107 * mac_notify upcalls) can still achieve that by posting the request to a
108 * thread, which can then grab all the required perimeters and locks in the
109 * right global order. Note that in the above example the mac layer iself
110 * won't grab the mac perimeter in the mac_notify upcall, instead the upcall
111 * to the client must do that. Please see the aggr code for an example.
112 *
113 * MAC client rules
114 * ----------------
115 *
116 * R3. A MAC client may use the MAC provided perimeter facility to serialize
117 * control operations on a per mac end point. It does this by by acquring
118 * and holding the perimeter across a sequence of calls to the mac layer.
119 * This ensures atomicity across the entire block of mac calls. In this
120 * model the MAC client must not hold any client locks across the calls to
121 * the mac layer. This model is the preferred solution.
122 *
123 * R4. However if a MAC client has a lot of global state across all mac end
124 * points the per mac end point serialization may not be sufficient. In this
125 * case the client may choose to use global locks or use its own serialization.
126 * To avoid deadlocks, these client layer locks held across the mac calls
127 * in the control path must never be acquired by the data path for the reason
128 * mentioned below.
129 *
130 * (Assume that a control operation that holds a client lock blocks in the
131 * mac layer waiting for upcall reference counts to drop to zero. If an upcall
132 * data thread that holds this reference count, tries to acquire the same
133 * client lock subsequently it will deadlock).
134 *
135 * A MAC client may follow either the R3 model or the R4 model, but can't
136 * mix both. In the former, the hierarchy is Perim -> client locks, but in
137 * the latter it is client locks -> Perim.
138 *
139 * R5. MAC clients must make MAC calls (excluding data calls) in a cv_wait'able
140 * context since they may block while trying to acquire the perimeter.
141 * In addition some calls may block waiting for upcall refcnts to come down to
142 * zero.
143 *
144 * R6. MAC clients must make sure that they are single threaded and all threads
145 * from the top (in particular data threads) have finished before calling
146 * mac_client_close. The MAC framework does not track the number of client
147 * threads using the mac client handle. Also mac clients must make sure
148 * they have undone all the control operations before calling mac_client_close.
149 * For example mac_unicast_remove/mac_multicast_remove to undo the corresponding
150 * mac_unicast_add/mac_multicast_add.
151 *
152 * MAC framework rules
153 * -------------------
154 *
155 * R7. The mac layer itself must not hold any mac layer locks (except the mac
156 * perimeter) across a call to any other layer from the mac layer. The call to
157 * any other layer could be via mi_* entry points, classifier entry points into
158 * the driver or via upcall pointers into layers above. The mac perimeter may
159 * be acquired or held only in the down direction, for e.g. when calling into
160 * a mi_* driver enty point to provide atomicity of the operation.
161 *
162 * R8. Since it is not guaranteed (see R14) that drivers won't hold locks across
163 * mac driver interfaces, the MAC layer must provide a cut out for control
164 * interfaces like upcall notifications and start them in a separate thread.
165 *
166 * R9. Note that locking order also implies a plumbing order. For example
167 * VNICs are allowed to be created over aggrs, but not vice-versa. An attempt
168 * to plumb in any other order must be failed at mac_open time, otherwise it
169 * could lead to deadlocks due to inverse locking order.
170 *
171 * R10. MAC driver interfaces must not block since the driver could call them
172 * in interrupt context.
173 *
174 * R11. Walkers must preferably not hold any locks while calling walker
175 * callbacks. Instead these can operate on reference counts. In simple
176 * callbacks it may be ok to hold a lock and call the callbacks, but this is
177 * harder to maintain in the general case of arbitrary callbacks.
178 *
179 * R12. The MAC layer must protect upcall notification callbacks using reference
180 * counts rather than holding locks across the callbacks.
181 *
182 * R13. Given the variety of drivers, it is preferable if the MAC layer can make
183 * sure that any pointers (such as mac ring pointers) it passes to the driver
184 * remain valid until mac unregister time. Currently the mac layer achieves
185 * this by using generation numbers for rings and freeing the mac rings only
186 * at unregister time. The MAC layer must provide a layer of indirection and
187 * must not expose underlying driver rings or driver data structures/pointers
188 * directly to MAC clients.
189 *
190 * MAC driver rules
191 * ----------------
192 *
193 * R14. It would be preferable if MAC drivers don't hold any locks across any
194 * mac call. However at a minimum they must not hold any locks across data
195 * upcalls. They must also make sure that all references to mac data structures
196 * are cleaned up and that it is single threaded at mac_unregister time.
197 *
198 * R15. MAC driver interfaces don't block and so the action may be done
199 * asynchronously in a separate thread as for example handling notifications.
200 * The driver must not assume that the action is complete when the call
201 * returns.
202 *
203 * R16. Drivers must maintain a generation number per Rx ring, and pass it
204 * back to mac_rx_ring(); They are expected to increment the generation
205 * number whenever the ring's stop routine is invoked.
206 * See comments in mac_rx_ring();
207 *
208 * R17 Similarly mi_stop is another synchronization point and the driver must
209 * ensure that all upcalls are done and there won't be any future upcall
210 * before returning from mi_stop.
211 *
212 * R18. The driver may assume that all set/modify control operations via
213 * the mi_* entry points are single threaded on a per mac end point.
214 *
215 * Lock and Perimeter hierarchy scenarios
216 * ---------------------------------------
217 *
218 * i_mac_impl_lock -> mi_rw_lock -> srs_lock -> s_ring_lock[i_mac_tx_srs_notify]
219 *
220 * ft_lock -> fe_lock [mac_flow_lookup]
221 *
222 * mi_rw_lock -> fe_lock [mac_bcast_send]
223 *
224 * srs_lock -> mac_bw_lock [mac_rx_srs_drain_bw]
225 *
226 * cpu_lock -> mac_srs_g_lock -> srs_lock -> s_ring_lock [mac_walk_srs_and_bind]
227 *
228 * i_dls_devnet_lock -> mac layer locks [dls_devnet_rename]
229 *
230 * Perimeters are ordered P1 -> P2 -> P3 from top to bottom in order of mac
231 * client to driver. In the case of clients that explictly use the mac provided
232 * perimeter mechanism for its serialization, the hierarchy is
233 * Perimeter -> mac layer locks, since the client never holds any locks across
234 * the mac calls. In the case of clients that use its own locks the hierarchy
235 * is Client locks -> Mac Perim -> Mac layer locks. The client never explicitly
236 * calls mac_perim_enter/exit in this case.
237 *
238 * Subflow creation rules
239 * ---------------------------
240 * o In case of a user specified cpulist present on underlying link and flows,
241 * the flows cpulist must be a subset of the underlying link.
242 * o In case of a user specified fanout mode present on link and flow, the
243 * subflow fanout count has to be less than or equal to that of the
244 * underlying link. The cpu-bindings for the subflows will be a subset of
245 * the underlying link.
246 * o In case if no cpulist specified on both underlying link and flow, the
247 * underlying link relies on a MAC tunable to provide out of box fanout.
248 * The subflow will have no cpulist (the subflow will be unbound)
249 * o In case if no cpulist is specified on the underlying link, a subflow can
250 * carry either a user-specified cpulist or fanout count. The cpu-bindings
251 * for the subflow will not adhere to restriction that they need to be subset
252 * of the underlying link.
253 * o In case where the underlying link is carrying either a user specified
254 * cpulist or fanout mode and for a unspecified subflow, the subflow will be
255 * created unbound.
256 * o While creating unbound subflows, bandwidth mode changes attempt to
257 * figure a right fanout count. In such cases the fanout count will override
258 * the unbound cpu-binding behavior.
259 * o In addition to this, while cycling between flow and link properties, we
260 * impose a restriction that if a link property has a subflow with
261 * user-specified attributes, we will not allow changing the link property.
262 * The administrator needs to reset all the user specified properties for the
263 * subflows before attempting a link property change.
264 * Some of the above rules can be overridden by specifying additional command
265 * line options while creating or modifying link or subflow properties.
266 */
267
268 #include <sys/types.h>
269 #include <sys/conf.h>
270 #include <sys/id_space.h>
271 #include <sys/esunddi.h>
272 #include <sys/stat.h>
273 #include <sys/mkdev.h>
274 #include <sys/stream.h>
275 #include <sys/strsun.h>
276 #include <sys/strsubr.h>
277 #include <sys/dlpi.h>
278 #include <sys/list.h>
279 #include <sys/modhash.h>
280 #include <sys/mac_provider.h>
281 #include <sys/mac_client_impl.h>
282 #include <sys/mac_soft_ring.h>
283 #include <sys/mac_stat.h>
284 #include <sys/mac_impl.h>
285 #include <sys/mac.h>
286 #include <sys/dls.h>
287 #include <sys/dld.h>
288 #include <sys/modctl.h>
289 #include <sys/fs/dv_node.h>
290 #include <sys/thread.h>
291 #include <sys/proc.h>
292 #include <sys/callb.h>
293 #include <sys/cpuvar.h>
294 #include <sys/atomic.h>
295 #include <sys/bitmap.h>
296 #include <sys/sdt.h>
297 #include <sys/mac_flow.h>
298 #include <sys/ddi_intr_impl.h>
299 #include <sys/disp.h>
300 #include <sys/sdt.h>
301 #include <sys/vnic.h>
302 #include <sys/vnic_impl.h>
303 #include <sys/vlan.h>
304 #include <inet/ip.h>
305 #include <inet/ip6.h>
306 #include <sys/exacct.h>
307 #include <sys/exacct_impl.h>
308 #include <inet/nd.h>
309 #include <sys/ethernet.h>
310 #include <sys/pool.h>
311 #include <sys/pool_pset.h>
312 #include <sys/cpupart.h>
313 #include <inet/wifi_ioctl.h>
314 #include <net/wpa.h>
315
316 #define IMPL_HASHSZ 67 /* prime */
317
318 kmem_cache_t *i_mac_impl_cachep;
319 mod_hash_t *i_mac_impl_hash;
320 krwlock_t i_mac_impl_lock;
321 uint_t i_mac_impl_count;
322 static kmem_cache_t *mac_ring_cache;
323 static id_space_t *minor_ids;
324 static uint32_t minor_count;
325 static pool_event_cb_t mac_pool_event_reg;
326
327 /*
328 * Logging stuff. Perhaps mac_logging_interval could be broken into
329 * mac_flow_log_interval and mac_link_log_interval if we want to be
330 * able to schedule them differently.
331 */
332 uint_t mac_logging_interval;
333 boolean_t mac_flow_log_enable;
334 boolean_t mac_link_log_enable;
335 timeout_id_t mac_logging_timer;
336
337 /* for debugging, see MAC_DBG_PRT() in mac_impl.h */
338 int mac_dbg = 0;
339
340 #define MACTYPE_KMODDIR "mac"
341 #define MACTYPE_HASHSZ 67
342 static mod_hash_t *i_mactype_hash;
343 /*
344 * i_mactype_lock synchronizes threads that obtain references to mactype_t
345 * structures through i_mactype_getplugin().
346 */
347 static kmutex_t i_mactype_lock;
348
349 /*
350 * mac_tx_percpu_cnt
351 *
352 * Number of per cpu locks per mac_client_impl_t. Used by the transmit side
353 * in mac_tx to reduce lock contention. This is sized at boot time in mac_init.
354 * mac_tx_percpu_cnt_max is settable in /etc/system and must be a power of 2.
355 * Per cpu locks may be disabled by setting mac_tx_percpu_cnt_max to 1.
356 */
357 int mac_tx_percpu_cnt;
358 int mac_tx_percpu_cnt_max = 128;
359
360 /*
361 * Call back functions for the bridge module. These are guaranteed to be valid
362 * when holding a reference on a link or when holding mip->mi_bridge_lock and
363 * mi_bridge_link is non-NULL.
364 */
365 mac_bridge_tx_t mac_bridge_tx_cb;
366 mac_bridge_rx_t mac_bridge_rx_cb;
367 mac_bridge_ref_t mac_bridge_ref_cb;
368 mac_bridge_ls_t mac_bridge_ls_cb;
369
370 static int i_mac_constructor(void *, void *, int);
371 static void i_mac_destructor(void *, void *);
372 static int i_mac_ring_ctor(void *, void *, int);
373 static void i_mac_ring_dtor(void *, void *);
374 static mblk_t *mac_rx_classify(mac_impl_t *, mac_resource_handle_t, mblk_t *);
375 void mac_tx_client_flush(mac_client_impl_t *);
376 void mac_tx_client_block(mac_client_impl_t *);
377 static void mac_rx_ring_quiesce(mac_ring_t *, uint_t);
378 static int mac_start_group_and_rings(mac_group_t *);
379 static void mac_stop_group_and_rings(mac_group_t *);
380 static void mac_pool_event_cb(pool_event_t, int, void *);
381
382 typedef struct netinfo_s {
383 list_node_t ni_link;
384 void *ni_record;
385 int ni_size;
386 int ni_type;
387 } netinfo_t;
388
389 /*
390 * Module initialization functions.
391 */
392
393 void
394 mac_init(void)
395 {
396 mac_tx_percpu_cnt = ((boot_max_ncpus == -1) ? max_ncpus :
397 boot_max_ncpus);
398
399 /* Upper bound is mac_tx_percpu_cnt_max */
400 if (mac_tx_percpu_cnt > mac_tx_percpu_cnt_max)
401 mac_tx_percpu_cnt = mac_tx_percpu_cnt_max;
402
403 if (mac_tx_percpu_cnt < 1) {
404 /* Someone set max_tx_percpu_cnt_max to 0 or less */
405 mac_tx_percpu_cnt = 1;
406 }
407
408 ASSERT(mac_tx_percpu_cnt >= 1);
409 mac_tx_percpu_cnt = (1 << highbit(mac_tx_percpu_cnt - 1));
410 /*
411 * Make it of the form 2**N - 1 in the range
412 * [0 .. mac_tx_percpu_cnt_max - 1]
413 */
414 mac_tx_percpu_cnt--;
415
416 i_mac_impl_cachep = kmem_cache_create("mac_impl_cache",
417 sizeof (mac_impl_t), 0, i_mac_constructor, i_mac_destructor,
418 NULL, NULL, NULL, 0);
419 ASSERT(i_mac_impl_cachep != NULL);
420
421 mac_ring_cache = kmem_cache_create("mac_ring_cache",
422 sizeof (mac_ring_t), 0, i_mac_ring_ctor, i_mac_ring_dtor, NULL,
423 NULL, NULL, 0);
424 ASSERT(mac_ring_cache != NULL);
425
426 i_mac_impl_hash = mod_hash_create_extended("mac_impl_hash",
427 IMPL_HASHSZ, mod_hash_null_keydtor, mod_hash_null_valdtor,
428 mod_hash_bystr, NULL, mod_hash_strkey_cmp, KM_SLEEP);
429 rw_init(&i_mac_impl_lock, NULL, RW_DEFAULT, NULL);
430
431 mac_flow_init();
432 mac_soft_ring_init();
433 mac_bcast_init();
434 mac_client_init();
435
436 i_mac_impl_count = 0;
437
438 i_mactype_hash = mod_hash_create_extended("mactype_hash",
439 MACTYPE_HASHSZ,
440 mod_hash_null_keydtor, mod_hash_null_valdtor,
441 mod_hash_bystr, NULL, mod_hash_strkey_cmp, KM_SLEEP);
442
443 /*
444 * Allocate an id space to manage minor numbers. The range of the
445 * space will be from MAC_MAX_MINOR+1 to MAC_PRIVATE_MINOR-1. This
446 * leaves half of the 32-bit minors available for driver private use.
447 */
448 minor_ids = id_space_create("mac_minor_ids", MAC_MAX_MINOR+1,
449 MAC_PRIVATE_MINOR-1);
450 ASSERT(minor_ids != NULL);
451 minor_count = 0;
452
453 /* Let's default to 20 seconds */
454 mac_logging_interval = 20;
455 mac_flow_log_enable = B_FALSE;
456 mac_link_log_enable = B_FALSE;
457 mac_logging_timer = 0;
458
459 /* Register to be notified of noteworthy pools events */
460 mac_pool_event_reg.pec_func = mac_pool_event_cb;
461 mac_pool_event_reg.pec_arg = NULL;
462 pool_event_cb_register(&mac_pool_event_reg);
463 }
464
465 int
466 mac_fini(void)
467 {
468
469 if (i_mac_impl_count > 0 || minor_count > 0)
470 return (EBUSY);
471
472 pool_event_cb_unregister(&mac_pool_event_reg);
473
474 id_space_destroy(minor_ids);
475 mac_flow_fini();
476
477 mod_hash_destroy_hash(i_mac_impl_hash);
478 rw_destroy(&i_mac_impl_lock);
479
480 mac_client_fini();
481 kmem_cache_destroy(mac_ring_cache);
482
483 mod_hash_destroy_hash(i_mactype_hash);
484 mac_soft_ring_finish();
485
486
487 return (0);
488 }
489
490 /*
491 * Initialize a GLDv3 driver's device ops. A driver that manages its own ops
492 * (e.g. softmac) may pass in a NULL ops argument.
493 */
494 void
495 mac_init_ops(struct dev_ops *ops, const char *name)
496 {
497 major_t major = ddi_name_to_major((char *)name);
498
499 /*
500 * By returning on error below, we are not letting the driver continue
501 * in an undefined context. The mac_register() function will faill if
502 * DN_GLDV3_DRIVER isn't set.
503 */
504 if (major == DDI_MAJOR_T_NONE)
505 return;
506 LOCK_DEV_OPS(&devnamesp[major].dn_lock);
507 devnamesp[major].dn_flags |= (DN_GLDV3_DRIVER | DN_NETWORK_DRIVER);
508 UNLOCK_DEV_OPS(&devnamesp[major].dn_lock);
509 if (ops != NULL)
510 dld_init_ops(ops, name);
511 }
512
513 void
514 mac_fini_ops(struct dev_ops *ops)
515 {
516 dld_fini_ops(ops);
517 }
518
519 /*ARGSUSED*/
520 static int
521 i_mac_constructor(void *buf, void *arg, int kmflag)
522 {
523 mac_impl_t *mip = buf;
524
525 bzero(buf, sizeof (mac_impl_t));
526
527 mip->mi_linkstate = LINK_STATE_UNKNOWN;
528
529 rw_init(&mip->mi_rw_lock, NULL, RW_DRIVER, NULL);
530 mutex_init(&mip->mi_notify_lock, NULL, MUTEX_DRIVER, NULL);
531 mutex_init(&mip->mi_promisc_lock, NULL, MUTEX_DRIVER, NULL);
532 mutex_init(&mip->mi_ring_lock, NULL, MUTEX_DEFAULT, NULL);
533
534 mip->mi_notify_cb_info.mcbi_lockp = &mip->mi_notify_lock;
535 cv_init(&mip->mi_notify_cb_info.mcbi_cv, NULL, CV_DRIVER, NULL);
536 mip->mi_promisc_cb_info.mcbi_lockp = &mip->mi_promisc_lock;
537 cv_init(&mip->mi_promisc_cb_info.mcbi_cv, NULL, CV_DRIVER, NULL);
538
539 mutex_init(&mip->mi_bridge_lock, NULL, MUTEX_DEFAULT, NULL);
540
541 return (0);
542 }
543
544 /*ARGSUSED*/
545 static void
546 i_mac_destructor(void *buf, void *arg)
547 {
548 mac_impl_t *mip = buf;
549 mac_cb_info_t *mcbi;
550
551 ASSERT(mip->mi_ref == 0);
552 ASSERT(mip->mi_active == 0);
553 ASSERT(mip->mi_linkstate == LINK_STATE_UNKNOWN);
554 ASSERT(mip->mi_devpromisc == 0);
555 ASSERT(mip->mi_ksp == NULL);
556 ASSERT(mip->mi_kstat_count == 0);
557 ASSERT(mip->mi_nclients == 0);
558 ASSERT(mip->mi_nactiveclients == 0);
559 ASSERT(mip->mi_single_active_client == NULL);
560 ASSERT(mip->mi_state_flags == 0);
561 ASSERT(mip->mi_factory_addr == NULL);
562 ASSERT(mip->mi_factory_addr_num == 0);
563 ASSERT(mip->mi_default_tx_ring == NULL);
564
565 mcbi = &mip->mi_notify_cb_info;
566 ASSERT(mcbi->mcbi_del_cnt == 0 && mcbi->mcbi_walker_cnt == 0);
567 ASSERT(mip->mi_notify_bits == 0);
568 ASSERT(mip->mi_notify_thread == NULL);
569 ASSERT(mcbi->mcbi_lockp == &mip->mi_notify_lock);
570 mcbi->mcbi_lockp = NULL;
571
572 mcbi = &mip->mi_promisc_cb_info;
573 ASSERT(mcbi->mcbi_del_cnt == 0 && mip->mi_promisc_list == NULL);
574 ASSERT(mip->mi_promisc_list == NULL);
575 ASSERT(mcbi->mcbi_lockp == &mip->mi_promisc_lock);
576 mcbi->mcbi_lockp = NULL;
577
578 ASSERT(mip->mi_bcast_ngrps == 0 && mip->mi_bcast_grp == NULL);
579 ASSERT(mip->mi_perim_owner == NULL && mip->mi_perim_ocnt == 0);
580
581 rw_destroy(&mip->mi_rw_lock);
582
583 mutex_destroy(&mip->mi_promisc_lock);
584 cv_destroy(&mip->mi_promisc_cb_info.mcbi_cv);
585 mutex_destroy(&mip->mi_notify_lock);
586 cv_destroy(&mip->mi_notify_cb_info.mcbi_cv);
587 mutex_destroy(&mip->mi_ring_lock);
588
589 ASSERT(mip->mi_bridge_link == NULL);
590 }
591
592 /* ARGSUSED */
593 static int
594 i_mac_ring_ctor(void *buf, void *arg, int kmflag)
595 {
596 mac_ring_t *ring = (mac_ring_t *)buf;
597
598 bzero(ring, sizeof (mac_ring_t));
599 cv_init(&ring->mr_cv, NULL, CV_DEFAULT, NULL);
600 mutex_init(&ring->mr_lock, NULL, MUTEX_DEFAULT, NULL);
601 ring->mr_state = MR_FREE;
602 return (0);
603 }
604
605 /* ARGSUSED */
606 static void
607 i_mac_ring_dtor(void *buf, void *arg)
608 {
609 mac_ring_t *ring = (mac_ring_t *)buf;
610
611 cv_destroy(&ring->mr_cv);
612 mutex_destroy(&ring->mr_lock);
613 }
614
615 /*
616 * Common functions to do mac callback addition and deletion. Currently this is
617 * used by promisc callbacks and notify callbacks. List addition and deletion
618 * need to take care of list walkers. List walkers in general, can't hold list
619 * locks and make upcall callbacks due to potential lock order and recursive
620 * reentry issues. Instead list walkers increment the list walker count to mark
621 * the presence of a walker thread. Addition can be carefully done to ensure
622 * that the list walker always sees either the old list or the new list.
623 * However the deletion can't be done while the walker is active, instead the
624 * deleting thread simply marks the entry as logically deleted. The last walker
625 * physically deletes and frees up the logically deleted entries when the walk
626 * is complete.
627 */
628 void
629 mac_callback_add(mac_cb_info_t *mcbi, mac_cb_t **mcb_head,
630 mac_cb_t *mcb_elem)
631 {
632 mac_cb_t *p;
633 mac_cb_t **pp;
634
635 /* Verify it is not already in the list */
636 for (pp = mcb_head; (p = *pp) != NULL; pp = &p->mcb_nextp) {
637 if (p == mcb_elem)
638 break;
639 }
640 VERIFY(p == NULL);
641
642 /*
643 * Add it to the head of the callback list. The membar ensures that
644 * the following list pointer manipulations reach global visibility
645 * in exactly the program order below.
646 */
647 ASSERT(MUTEX_HELD(mcbi->mcbi_lockp));
648
649 mcb_elem->mcb_nextp = *mcb_head;
650 membar_producer();
651 *mcb_head = mcb_elem;
652 }
653
654 /*
655 * Mark the entry as logically deleted. If there aren't any walkers unlink
656 * from the list. In either case return the corresponding status.
657 */
658 boolean_t
659 mac_callback_remove(mac_cb_info_t *mcbi, mac_cb_t **mcb_head,
660 mac_cb_t *mcb_elem)
661 {
662 mac_cb_t *p;
663 mac_cb_t **pp;
664
665 ASSERT(MUTEX_HELD(mcbi->mcbi_lockp));
666 /*
667 * Search the callback list for the entry to be removed
668 */
669 for (pp = mcb_head; (p = *pp) != NULL; pp = &p->mcb_nextp) {
670 if (p == mcb_elem)
671 break;
672 }
673 VERIFY(p != NULL);
674
675 /*
676 * If there are walkers just mark it as deleted and the last walker
677 * will remove from the list and free it.
678 */
679 if (mcbi->mcbi_walker_cnt != 0) {
680 p->mcb_flags |= MCB_CONDEMNED;
681 mcbi->mcbi_del_cnt++;
682 return (B_FALSE);
683 }
684
685 ASSERT(mcbi->mcbi_del_cnt == 0);
686 *pp = p->mcb_nextp;
687 p->mcb_nextp = NULL;
688 return (B_TRUE);
689 }
690
691 /*
692 * Wait for all pending callback removals to be completed
693 */
694 void
695 mac_callback_remove_wait(mac_cb_info_t *mcbi)
696 {
697 ASSERT(MUTEX_HELD(mcbi->mcbi_lockp));
698 while (mcbi->mcbi_del_cnt != 0) {
699 DTRACE_PROBE1(need_wait, mac_cb_info_t *, mcbi);
700 cv_wait(&mcbi->mcbi_cv, mcbi->mcbi_lockp);
701 }
702 }
703
704 /*
705 * The last mac callback walker does the cleanup. Walk the list and unlik
706 * all the logically deleted entries and construct a temporary list of
707 * removed entries. Return the list of removed entries to the caller.
708 */
709 mac_cb_t *
710 mac_callback_walker_cleanup(mac_cb_info_t *mcbi, mac_cb_t **mcb_head)
711 {
712 mac_cb_t *p;
713 mac_cb_t **pp;
714 mac_cb_t *rmlist = NULL; /* List of removed elements */
715 int cnt = 0;
716
717 ASSERT(MUTEX_HELD(mcbi->mcbi_lockp));
718 ASSERT(mcbi->mcbi_del_cnt != 0 && mcbi->mcbi_walker_cnt == 0);
719
720 pp = mcb_head;
721 while (*pp != NULL) {
722 if ((*pp)->mcb_flags & MCB_CONDEMNED) {
723 p = *pp;
724 *pp = p->mcb_nextp;
725 p->mcb_nextp = rmlist;
726 rmlist = p;
727 cnt++;
728 continue;
729 }
730 pp = &(*pp)->mcb_nextp;
731 }
732
733 ASSERT(mcbi->mcbi_del_cnt == cnt);
734 mcbi->mcbi_del_cnt = 0;
735 return (rmlist);
736 }
737
738 boolean_t
739 mac_callback_lookup(mac_cb_t **mcb_headp, mac_cb_t *mcb_elem)
740 {
741 mac_cb_t *mcb;
742
743 /* Verify it is not already in the list */
744 for (mcb = *mcb_headp; mcb != NULL; mcb = mcb->mcb_nextp) {
745 if (mcb == mcb_elem)
746 return (B_TRUE);
747 }
748
749 return (B_FALSE);
750 }
751
752 boolean_t
753 mac_callback_find(mac_cb_info_t *mcbi, mac_cb_t **mcb_headp, mac_cb_t *mcb_elem)
754 {
755 boolean_t found;
756
757 mutex_enter(mcbi->mcbi_lockp);
758 found = mac_callback_lookup(mcb_headp, mcb_elem);
759 mutex_exit(mcbi->mcbi_lockp);
760
761 return (found);
762 }
763
764 /* Free the list of removed callbacks */
765 void
766 mac_callback_free(mac_cb_t *rmlist)
767 {
768 mac_cb_t *mcb;
769 mac_cb_t *mcb_next;
770
771 for (mcb = rmlist; mcb != NULL; mcb = mcb_next) {
772 mcb_next = mcb->mcb_nextp;
773 kmem_free(mcb->mcb_objp, mcb->mcb_objsize);
774 }
775 }
776
777 /*
778 * The promisc callbacks are in 2 lists, one off the 'mip' and another off the
779 * 'mcip' threaded by mpi_mi_link and mpi_mci_link respectively. However there
780 * is only a single shared total walker count, and an entry can't be physically
781 * unlinked if a walker is active on either list. The last walker does this
782 * cleanup of logically deleted entries.
783 */
784 void
785 i_mac_promisc_walker_cleanup(mac_impl_t *mip)
786 {
787 mac_cb_t *rmlist;
788 mac_cb_t *mcb;
789 mac_cb_t *mcb_next;
790 mac_promisc_impl_t *mpip;
791
792 /*
793 * Construct a temporary list of deleted callbacks by walking the
794 * the mi_promisc_list. Then for each entry in the temporary list,
795 * remove it from the mci_promisc_list and free the entry.
796 */
797 rmlist = mac_callback_walker_cleanup(&mip->mi_promisc_cb_info,
798 &mip->mi_promisc_list);
799
800 for (mcb = rmlist; mcb != NULL; mcb = mcb_next) {
801 mcb_next = mcb->mcb_nextp;
802 mpip = (mac_promisc_impl_t *)mcb->mcb_objp;
803 VERIFY(mac_callback_remove(&mip->mi_promisc_cb_info,
804 &mpip->mpi_mcip->mci_promisc_list, &mpip->mpi_mci_link));
805 mcb->mcb_flags = 0;
806 mcb->mcb_nextp = NULL;
807 kmem_cache_free(mac_promisc_impl_cache, mpip);
808 }
809 }
810
811 void
812 i_mac_notify(mac_impl_t *mip, mac_notify_type_t type)
813 {
814 mac_cb_info_t *mcbi;
815
816 /*
817 * Signal the notify thread even after mi_ref has become zero and
818 * mi_disabled is set. The synchronization with the notify thread
819 * happens in mac_unregister and that implies the driver must make
820 * sure it is single-threaded (with respect to mac calls) and that
821 * all pending mac calls have returned before it calls mac_unregister
822 */
823 rw_enter(&i_mac_impl_lock, RW_READER);
824 if (mip->mi_state_flags & MIS_DISABLED)
825 goto exit;
826
827 /*
828 * Guard against incorrect notifications. (Running a newer
829 * mac client against an older implementation?)
830 */
831 if (type >= MAC_NNOTE)
832 goto exit;
833
834 mcbi = &mip->mi_notify_cb_info;
835 mutex_enter(mcbi->mcbi_lockp);
836 mip->mi_notify_bits |= (1 << type);
837 cv_broadcast(&mcbi->mcbi_cv);
838 mutex_exit(mcbi->mcbi_lockp);
839
840 exit:
841 rw_exit(&i_mac_impl_lock);
842 }
843
844 /*
845 * Mac serialization primitives. Please see the block comment at the
846 * top of the file.
847 */
848 void
849 i_mac_perim_enter(mac_impl_t *mip)
850 {
851 mac_client_impl_t *mcip;
852
853 if (mip->mi_state_flags & MIS_IS_VNIC) {
854 /*
855 * This is a VNIC. Return the lower mac since that is what
856 * we want to serialize on.
857 */
858 mcip = mac_vnic_lower(mip);
859 mip = mcip->mci_mip;
860 }
861
862 mutex_enter(&mip->mi_perim_lock);
863 if (mip->mi_perim_owner == curthread) {
864 mip->mi_perim_ocnt++;
865 mutex_exit(&mip->mi_perim_lock);
866 return;
867 }
868
869 while (mip->mi_perim_owner != NULL)
870 cv_wait(&mip->mi_perim_cv, &mip->mi_perim_lock);
871
872 mip->mi_perim_owner = curthread;
873 ASSERT(mip->mi_perim_ocnt == 0);
874 mip->mi_perim_ocnt++;
875 #ifdef DEBUG
876 mip->mi_perim_stack_depth = getpcstack(mip->mi_perim_stack,
877 MAC_PERIM_STACK_DEPTH);
878 #endif
879 mutex_exit(&mip->mi_perim_lock);
880 }
881
882 int
883 i_mac_perim_enter_nowait(mac_impl_t *mip)
884 {
885 /*
886 * The vnic is a special case, since the serialization is done based
887 * on the lower mac. If the lower mac is busy, it does not imply the
888 * vnic can't be unregistered. But in the case of other drivers,
889 * a busy perimeter or open mac handles implies that the mac is busy
890 * and can't be unregistered.
891 */
892 if (mip->mi_state_flags & MIS_IS_VNIC) {
893 i_mac_perim_enter(mip);
894 return (0);
895 }
896
897 mutex_enter(&mip->mi_perim_lock);
898 if (mip->mi_perim_owner != NULL) {
899 mutex_exit(&mip->mi_perim_lock);
900 return (EBUSY);
901 }
902 ASSERT(mip->mi_perim_ocnt == 0);
903 mip->mi_perim_owner = curthread;
904 mip->mi_perim_ocnt++;
905 mutex_exit(&mip->mi_perim_lock);
906
907 return (0);
908 }
909
910 void
911 i_mac_perim_exit(mac_impl_t *mip)
912 {
913 mac_client_impl_t *mcip;
914
915 if (mip->mi_state_flags & MIS_IS_VNIC) {
916 /*
917 * This is a VNIC. Return the lower mac since that is what
918 * we want to serialize on.
919 */
920 mcip = mac_vnic_lower(mip);
921 mip = mcip->mci_mip;
922 }
923
924 ASSERT(mip->mi_perim_owner == curthread && mip->mi_perim_ocnt != 0);
925
926 mutex_enter(&mip->mi_perim_lock);
927 if (--mip->mi_perim_ocnt == 0) {
928 mip->mi_perim_owner = NULL;
929 cv_signal(&mip->mi_perim_cv);
930 }
931 mutex_exit(&mip->mi_perim_lock);
932 }
933
934 /*
935 * Returns whether the current thread holds the mac perimeter. Used in making
936 * assertions.
937 */
938 boolean_t
939 mac_perim_held(mac_handle_t mh)
940 {
941 mac_impl_t *mip = (mac_impl_t *)mh;
942 mac_client_impl_t *mcip;
943
944 if (mip->mi_state_flags & MIS_IS_VNIC) {
945 /*
946 * This is a VNIC. Return the lower mac since that is what
947 * we want to serialize on.
948 */
949 mcip = mac_vnic_lower(mip);
950 mip = mcip->mci_mip;
951 }
952 return (mip->mi_perim_owner == curthread);
953 }
954
955 /*
956 * mac client interfaces to enter the mac perimeter of a mac end point, given
957 * its mac handle, or macname or linkid.
958 */
959 void
960 mac_perim_enter_by_mh(mac_handle_t mh, mac_perim_handle_t *mphp)
961 {
962 mac_impl_t *mip = (mac_impl_t *)mh;
963
964 i_mac_perim_enter(mip);
965 /*
966 * The mac_perim_handle_t returned encodes the 'mip' and whether a
967 * mac_open has been done internally while entering the perimeter.
968 * This information is used in mac_perim_exit
969 */
970 MAC_ENCODE_MPH(*mphp, mip, 0);
971 }
972
973 int
974 mac_perim_enter_by_macname(const char *name, mac_perim_handle_t *mphp)
975 {
976 int err;
977 mac_handle_t mh;
978
979 if ((err = mac_open(name, &mh)) != 0)
980 return (err);
981
982 mac_perim_enter_by_mh(mh, mphp);
983 MAC_ENCODE_MPH(*mphp, mh, 1);
984 return (0);
985 }
986
987 int
988 mac_perim_enter_by_linkid(datalink_id_t linkid, mac_perim_handle_t *mphp)
989 {
990 int err;
991 mac_handle_t mh;
992
993 if ((err = mac_open_by_linkid(linkid, &mh)) != 0)
994 return (err);
995
996 mac_perim_enter_by_mh(mh, mphp);
997 MAC_ENCODE_MPH(*mphp, mh, 1);
998 return (0);
999 }
1000
1001 void
1002 mac_perim_exit(mac_perim_handle_t mph)
1003 {
1004 mac_impl_t *mip;
1005 boolean_t need_close;
1006
1007 MAC_DECODE_MPH(mph, mip, need_close);
1008 i_mac_perim_exit(mip);
1009 if (need_close)
1010 mac_close((mac_handle_t)mip);
1011 }
1012
1013 int
1014 mac_hold(const char *macname, mac_impl_t **pmip)
1015 {
1016 mac_impl_t *mip;
1017 int err;
1018
1019 /*
1020 * Check the device name length to make sure it won't overflow our
1021 * buffer.
1022 */
1023 if (strlen(macname) >= MAXNAMELEN)
1024 return (EINVAL);
1025
1026 /*
1027 * Look up its entry in the global hash table.
1028 */
1029 rw_enter(&i_mac_impl_lock, RW_WRITER);
1030 err = mod_hash_find(i_mac_impl_hash, (mod_hash_key_t)macname,
1031 (mod_hash_val_t *)&mip);
1032
1033 if (err != 0) {
1034 rw_exit(&i_mac_impl_lock);
1035 return (ENOENT);
1036 }
1037
1038 if (mip->mi_state_flags & MIS_DISABLED) {
1039 rw_exit(&i_mac_impl_lock);
1040 return (ENOENT);
1041 }
1042
1043 if (mip->mi_state_flags & MIS_EXCLUSIVE_HELD) {
1044 rw_exit(&i_mac_impl_lock);
1045 return (EBUSY);
1046 }
1047
1048 mip->mi_ref++;
1049 rw_exit(&i_mac_impl_lock);
1050
1051 *pmip = mip;
1052 return (0);
1053 }
1054
1055 void
1056 mac_rele(mac_impl_t *mip)
1057 {
1058 rw_enter(&i_mac_impl_lock, RW_WRITER);
1059 ASSERT(mip->mi_ref != 0);
1060 if (--mip->mi_ref == 0) {
1061 ASSERT(mip->mi_nactiveclients == 0 &&
1062 !(mip->mi_state_flags & MIS_EXCLUSIVE));
1063 }
1064 rw_exit(&i_mac_impl_lock);
1065 }
1066
1067 /*
1068 * Private GLDv3 function to start a MAC instance.
1069 */
1070 int
1071 mac_start(mac_handle_t mh)
1072 {
1073 mac_impl_t *mip = (mac_impl_t *)mh;
1074 int err = 0;
1075 mac_group_t *defgrp;
1076
1077 ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1078 ASSERT(mip->mi_start != NULL);
1079
1080 /*
1081 * Check whether the device is already started.
1082 */
1083 if (mip->mi_active++ == 0) {
1084 mac_ring_t *ring = NULL;
1085
1086 /*
1087 * Start the device.
1088 */
1089 err = mip->mi_start(mip->mi_driver);
1090 if (err != 0) {
1091 mip->mi_active--;
1092 return (err);
1093 }
1094
1095 /*
1096 * Start the default tx ring.
1097 */
1098 if (mip->mi_default_tx_ring != NULL) {
1099
1100 ring = (mac_ring_t *)mip->mi_default_tx_ring;
1101 if (ring->mr_state != MR_INUSE) {
1102 err = mac_start_ring(ring);
1103 if (err != 0) {
1104 mip->mi_active--;
1105 return (err);
1106 }
1107 }
1108 }
1109
1110 if ((defgrp = MAC_DEFAULT_RX_GROUP(mip)) != NULL) {
1111 /*
1112 * Start the default ring, since it will be needed
1113 * to receive broadcast and multicast traffic for
1114 * both primary and non-primary MAC clients.
1115 */
1116 ASSERT(defgrp->mrg_state == MAC_GROUP_STATE_REGISTERED);
1117 err = mac_start_group_and_rings(defgrp);
1118 if (err != 0) {
1119 mip->mi_active--;
1120 if ((ring != NULL) &&
1121 (ring->mr_state == MR_INUSE))
1122 mac_stop_ring(ring);
1123 return (err);
1124 }
1125 mac_set_group_state(defgrp, MAC_GROUP_STATE_SHARED);
1126 }
1127 }
1128
1129 return (err);
1130 }
1131
1132 /*
1133 * Private GLDv3 function to stop a MAC instance.
1134 */
1135 void
1136 mac_stop(mac_handle_t mh)
1137 {
1138 mac_impl_t *mip = (mac_impl_t *)mh;
1139 mac_group_t *grp;
1140
1141 ASSERT(mip->mi_stop != NULL);
1142 ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1143
1144 /*
1145 * Check whether the device is still needed.
1146 */
1147 ASSERT(mip->mi_active != 0);
1148 if (--mip->mi_active == 0) {
1149 if ((grp = MAC_DEFAULT_RX_GROUP(mip)) != NULL) {
1150 /*
1151 * There should be no more active clients since the
1152 * MAC is being stopped. Stop the default RX group
1153 * and transition it back to registered state.
1154 *
1155 * When clients are torn down, the groups
1156 * are release via mac_release_rx_group which
1157 * knows the the default group is always in
1158 * started mode since broadcast uses it. So
1159 * we can assert that their are no clients
1160 * (since mac_bcast_add doesn't register itself
1161 * as a client) and group is in SHARED state.
1162 */
1163 ASSERT(grp->mrg_state == MAC_GROUP_STATE_SHARED);
1164 ASSERT(MAC_GROUP_NO_CLIENT(grp) &&
1165 mip->mi_nactiveclients == 0);
1166 mac_stop_group_and_rings(grp);
1167 mac_set_group_state(grp, MAC_GROUP_STATE_REGISTERED);
1168 }
1169
1170 if (mip->mi_default_tx_ring != NULL) {
1171 mac_ring_t *ring;
1172
1173 ring = (mac_ring_t *)mip->mi_default_tx_ring;
1174 if (ring->mr_state == MR_INUSE) {
1175 mac_stop_ring(ring);
1176 ring->mr_flag = 0;
1177 }
1178 }
1179
1180 /*
1181 * Stop the device.
1182 */
1183 mip->mi_stop(mip->mi_driver);
1184 }
1185 }
1186
1187 int
1188 i_mac_promisc_set(mac_impl_t *mip, boolean_t on)
1189 {
1190 int err = 0;
1191
1192 ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
1193 ASSERT(mip->mi_setpromisc != NULL);
1194
1195 if (on) {
1196 /*
1197 * Enable promiscuous mode on the device if not yet enabled.
1198 */
1199 if (mip->mi_devpromisc++ == 0) {
1200 err = mip->mi_setpromisc(mip->mi_driver, B_TRUE);
1201 if (err != 0) {
1202 mip->mi_devpromisc--;
1203 return (err);
1204 }
1205 i_mac_notify(mip, MAC_NOTE_DEVPROMISC);
1206 }
1207 } else {
1208 if (mip->mi_devpromisc == 0)
1209 return (EPROTO);
1210
1211 /*
1212 * Disable promiscuous mode on the device if this is the last
1213 * enabling.
1214 */
1215 if (--mip->mi_devpromisc == 0) {
1216 err = mip->mi_setpromisc(mip->mi_driver, B_FALSE);
1217 if (err != 0) {
1218 mip->mi_devpromisc++;
1219 return (err);
1220 }
1221 i_mac_notify(mip, MAC_NOTE_DEVPROMISC);
1222 }
1223 }
1224
1225 return (0);
1226 }
1227
1228 /*
1229 * The promiscuity state can change any time. If the caller needs to take
1230 * actions that are atomic with the promiscuity state, then the caller needs
1231 * to bracket the entire sequence with mac_perim_enter/exit
1232 */
1233 boolean_t
1234 mac_promisc_get(mac_handle_t mh)
1235 {
1236 mac_impl_t *mip = (mac_impl_t *)mh;
1237
1238 /*
1239 * Return the current promiscuity.
1240 */
1241 return (mip->mi_devpromisc != 0);
1242 }
1243
1244 /*
1245 * Invoked at MAC instance attach time to initialize the list
1246 * of factory MAC addresses supported by a MAC instance. This function
1247 * builds a local cache in the mac_impl_t for the MAC addresses
1248 * supported by the underlying hardware. The MAC clients themselves
1249 * use the mac_addr_factory*() functions to query and reserve
1250 * factory MAC addresses.
1251 */
1252 void
1253 mac_addr_factory_init(mac_impl_t *mip)
1254 {
1255 mac_capab_multifactaddr_t capab;
1256 uint8_t *addr;
1257 int i;
1258
1259 /*
1260 * First round to see how many factory MAC addresses are available.
1261 */
1262 bzero(&capab, sizeof (capab));
1263 if (!i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_MULTIFACTADDR,
1264 &capab) || (capab.mcm_naddr == 0)) {
1265 /*
1266 * The MAC instance doesn't support multiple factory
1267 * MAC addresses, we're done here.
1268 */
1269 return;
1270 }
1271
1272 /*
1273 * Allocate the space and get all the factory addresses.
1274 */
1275 addr = kmem_alloc(capab.mcm_naddr * MAXMACADDRLEN, KM_SLEEP);
1276 capab.mcm_getaddr(mip->mi_driver, capab.mcm_naddr, addr);
1277
1278 mip->mi_factory_addr_num = capab.mcm_naddr;
1279 mip->mi_factory_addr = kmem_zalloc(mip->mi_factory_addr_num *
1280 sizeof (mac_factory_addr_t), KM_SLEEP);
1281
1282 for (i = 0; i < capab.mcm_naddr; i++) {
1283 bcopy(addr + i * MAXMACADDRLEN,
1284 mip->mi_factory_addr[i].mfa_addr,
1285 mip->mi_type->mt_addr_length);
1286 mip->mi_factory_addr[i].mfa_in_use = B_FALSE;
1287 }
1288
1289 kmem_free(addr, capab.mcm_naddr * MAXMACADDRLEN);
1290 }
1291
1292 void
1293 mac_addr_factory_fini(mac_impl_t *mip)
1294 {
1295 if (mip->mi_factory_addr == NULL) {
1296 ASSERT(mip->mi_factory_addr_num == 0);
1297 return;
1298 }
1299
1300 kmem_free(mip->mi_factory_addr, mip->mi_factory_addr_num *
1301 sizeof (mac_factory_addr_t));
1302
1303 mip->mi_factory_addr = NULL;
1304 mip->mi_factory_addr_num = 0;
1305 }
1306
1307 /*
1308 * Reserve a factory MAC address. If *slot is set to -1, the function
1309 * attempts to reserve any of the available factory MAC addresses and
1310 * returns the reserved slot id. If no slots are available, the function
1311 * returns ENOSPC. If *slot is not set to -1, the function reserves
1312 * the specified slot if it is available, or returns EBUSY is the slot
1313 * is already used. Returns ENOTSUP if the underlying MAC does not
1314 * support multiple factory addresses. If the slot number is not -1 but
1315 * is invalid, returns EINVAL.
1316 */
1317 int
1318 mac_addr_factory_reserve(mac_client_handle_t mch, int *slot)
1319 {
1320 mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
1321 mac_impl_t *mip = mcip->mci_mip;
1322 int i, ret = 0;
1323
1324 i_mac_perim_enter(mip);
1325 /*
1326 * Protect against concurrent readers that may need a self-consistent
1327 * view of the factory addresses
1328 */
1329 rw_enter(&mip->mi_rw_lock, RW_WRITER);
1330
1331 if (mip->mi_factory_addr_num == 0) {
1332 ret = ENOTSUP;
1333 goto bail;
1334 }
1335
1336 if (*slot != -1) {
1337 /* check the specified slot */
1338 if (*slot < 1 || *slot > mip->mi_factory_addr_num) {
1339 ret = EINVAL;
1340 goto bail;
1341 }
1342 if (mip->mi_factory_addr[*slot-1].mfa_in_use) {
1343 ret = EBUSY;
1344 goto bail;
1345 }
1346 } else {
1347 /* pick the next available slot */
1348 for (i = 0; i < mip->mi_factory_addr_num; i++) {
1349 if (!mip->mi_factory_addr[i].mfa_in_use)
1350 break;
1351 }
1352
1353 if (i == mip->mi_factory_addr_num) {
1354 ret = ENOSPC;
1355 goto bail;
1356 }
1357 *slot = i+1;
1358 }
1359
1360 mip->mi_factory_addr[*slot-1].mfa_in_use = B_TRUE;
1361 mip->mi_factory_addr[*slot-1].mfa_client = mcip;
1362
1363 bail:
1364 rw_exit(&mip->mi_rw_lock);
1365 i_mac_perim_exit(mip);
1366 return (ret);
1367 }
1368
1369 /*
1370 * Release the specified factory MAC address slot.
1371 */
1372 void
1373 mac_addr_factory_release(mac_client_handle_t mch, uint_t slot)
1374 {
1375 mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
1376 mac_impl_t *mip = mcip->mci_mip;
1377
1378 i_mac_perim_enter(mip);
1379 /*
1380 * Protect against concurrent readers that may need a self-consistent
1381 * view of the factory addresses
1382 */
1383 rw_enter(&mip->mi_rw_lock, RW_WRITER);
1384
1385 ASSERT(slot > 0 && slot <= mip->mi_factory_addr_num);
1386 ASSERT(mip->mi_factory_addr[slot-1].mfa_in_use);
1387
1388 mip->mi_factory_addr[slot-1].mfa_in_use = B_FALSE;
1389
1390 rw_exit(&mip->mi_rw_lock);
1391 i_mac_perim_exit(mip);
1392 }
1393
1394 /*
1395 * Stores in mac_addr the value of the specified MAC address. Returns
1396 * 0 on success, or EINVAL if the slot number is not valid for the MAC.
1397 * The caller must provide a string of at least MAXNAMELEN bytes.
1398 */
1399 void
1400 mac_addr_factory_value(mac_handle_t mh, int slot, uchar_t *mac_addr,
1401 uint_t *addr_len, char *client_name, boolean_t *in_use_arg)
1402 {
1403 mac_impl_t *mip = (mac_impl_t *)mh;
1404 boolean_t in_use;
1405
1406 ASSERT(slot > 0 && slot <= mip->mi_factory_addr_num);
1407
1408 /*
1409 * Readers need to hold mi_rw_lock. Writers need to hold mac perimeter
1410 * and mi_rw_lock
1411 */
1412 rw_enter(&mip->mi_rw_lock, RW_READER);
1413 bcopy(mip->mi_factory_addr[slot-1].mfa_addr, mac_addr, MAXMACADDRLEN);
1414 *addr_len = mip->mi_type->mt_addr_length;
1415 in_use = mip->mi_factory_addr[slot-1].mfa_in_use;
1416 if (in_use && client_name != NULL) {
1417 bcopy(mip->mi_factory_addr[slot-1].mfa_client->mci_name,
1418 client_name, MAXNAMELEN);
1419 }
1420 if (in_use_arg != NULL)
1421 *in_use_arg = in_use;
1422 rw_exit(&mip->mi_rw_lock);
1423 }
1424
1425 /*
1426 * Returns the number of factory MAC addresses (in addition to the
1427 * primary MAC address), 0 if the underlying MAC doesn't support
1428 * that feature.
1429 */
1430 uint_t
1431 mac_addr_factory_num(mac_handle_t mh)
1432 {
1433 mac_impl_t *mip = (mac_impl_t *)mh;
1434
1435 return (mip->mi_factory_addr_num);
1436 }
1437
1438
1439 void
1440 mac_rx_group_unmark(mac_group_t *grp, uint_t flag)
1441 {
1442 mac_ring_t *ring;
1443
1444 for (ring = grp->mrg_rings; ring != NULL; ring = ring->mr_next)
1445 ring->mr_flag &= ~flag;
1446 }
1447
1448 /*
1449 * The following mac_hwrings_xxx() functions are private mac client functions
1450 * used by the aggr driver to access and control the underlying HW Rx group
1451 * and rings. In this case, the aggr driver has exclusive control of the
1452 * underlying HW Rx group/rings, it calls the following functions to
1453 * start/stop the HW Rx rings, disable/enable polling, add/remove mac'
1454 * addresses, or set up the Rx callback.
1455 */
1456 /* ARGSUSED */
1457 static void
1458 mac_hwrings_rx_process(void *arg, mac_resource_handle_t srs,
1459 mblk_t *mp_chain, boolean_t loopback)
1460 {
1461 mac_soft_ring_set_t *mac_srs = (mac_soft_ring_set_t *)srs;
1462 mac_srs_rx_t *srs_rx = &mac_srs->srs_rx;
1463 mac_direct_rx_t proc;
1464 void *arg1;
1465 mac_resource_handle_t arg2;
1466
1467 proc = srs_rx->sr_func;
1468 arg1 = srs_rx->sr_arg1;
1469 arg2 = mac_srs->srs_mrh;
1470
1471 proc(arg1, arg2, mp_chain, NULL);
1472 }
1473
1474 /*
1475 * This function is called to get the list of HW rings that are reserved by
1476 * an exclusive mac client.
1477 *
1478 * Return value: the number of HW rings.
1479 */
1480 int
1481 mac_hwrings_get(mac_client_handle_t mch, mac_group_handle_t *hwgh,
1482 mac_ring_handle_t *hwrh, mac_ring_type_t rtype)
1483 {
1484 mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
1485 flow_entry_t *flent = mcip->mci_flent;
1486 mac_group_t *grp;
1487 mac_ring_t *ring;
1488 int cnt = 0;
1489
1490 if (rtype == MAC_RING_TYPE_RX) {
1491 grp = flent->fe_rx_ring_group;
1492 } else if (rtype == MAC_RING_TYPE_TX) {
1493 grp = flent->fe_tx_ring_group;
1494 } else {
1495 ASSERT(B_FALSE);
1496 return (-1);
1497 }
1498 /*
1499 * The mac client did not reserve any RX group, return directly.
1500 * This is probably because the underlying MAC does not support
1501 * any groups.
1502 */
1503 if (hwgh != NULL)
1504 *hwgh = NULL;
1505 if (grp == NULL)
1506 return (0);
1507 /*
1508 * This group must be reserved by this mac client.
1509 */
1510 ASSERT((grp->mrg_state == MAC_GROUP_STATE_RESERVED) &&
1511 (mcip == MAC_GROUP_ONLY_CLIENT(grp)));
1512
1513 for (ring = grp->mrg_rings; ring != NULL; ring = ring->mr_next, cnt++) {
1514 ASSERT(cnt < MAX_RINGS_PER_GROUP);
1515 hwrh[cnt] = (mac_ring_handle_t)ring;
1516 }
1517 if (hwgh != NULL)
1518 *hwgh = (mac_group_handle_t)grp;
1519
1520 return (cnt);
1521 }
1522
1523 /*
1524 * This function is called to get info about Tx/Rx rings.
1525 *
1526 * Return value: returns uint_t which will have various bits set
1527 * that indicates different properties of the ring.
1528 */
1529 uint_t
1530 mac_hwring_getinfo(mac_ring_handle_t rh)
1531 {
1532 mac_ring_t *ring = (mac_ring_t *)rh;
1533 mac_ring_info_t *info = &ring->mr_info;
1534
1535 return (info->mri_flags);
1536 }
1537
1538 /*
1539 * Export ddi interrupt handles from the HW ring to the pseudo ring and
1540 * setup the RX callback of the mac client which exclusively controls
1541 * HW ring.
1542 */
1543 void
1544 mac_hwring_setup(mac_ring_handle_t hwrh, mac_resource_handle_t prh,
1545 mac_ring_handle_t pseudo_rh)
1546 {
1547 mac_ring_t *hw_ring = (mac_ring_t *)hwrh;
1548 mac_ring_t *pseudo_ring;
1549 mac_soft_ring_set_t *mac_srs = hw_ring->mr_srs;
1550
1551 if (pseudo_rh != NULL) {
1552 pseudo_ring = (mac_ring_t *)pseudo_rh;
1553 /* Export the ddi handles to pseudo ring */
1554 pseudo_ring->mr_info.mri_intr.mi_ddi_handle =
1555 hw_ring->mr_info.mri_intr.mi_ddi_handle;
1556 pseudo_ring->mr_info.mri_intr.mi_ddi_shared =
1557 hw_ring->mr_info.mri_intr.mi_ddi_shared;
1558 /*
1559 * Save a pointer to pseudo ring in the hw ring. If
1560 * interrupt handle changes, the hw ring will be
1561 * notified of the change (see mac_ring_intr_set())
1562 * and the appropriate change has to be made to
1563 * the pseudo ring that has exported the ddi handle.
1564 */
1565 hw_ring->mr_prh = pseudo_rh;
1566 }
1567
1568 if (hw_ring->mr_type == MAC_RING_TYPE_RX) {
1569 ASSERT(!(mac_srs->srs_type & SRST_TX));
1570 mac_srs->srs_mrh = prh;
1571 mac_srs->srs_rx.sr_lower_proc = mac_hwrings_rx_process;
1572 }
1573 }
1574
1575 void
1576 mac_hwring_teardown(mac_ring_handle_t hwrh)
1577 {
1578 mac_ring_t *hw_ring = (mac_ring_t *)hwrh;
1579 mac_soft_ring_set_t *mac_srs;
1580
1581 if (hw_ring == NULL)
1582 return;
1583 hw_ring->mr_prh = NULL;
1584 if (hw_ring->mr_type == MAC_RING_TYPE_RX) {
1585 mac_srs = hw_ring->mr_srs;
1586 ASSERT(!(mac_srs->srs_type & SRST_TX));
1587 mac_srs->srs_rx.sr_lower_proc = mac_rx_srs_process;
1588 mac_srs->srs_mrh = NULL;
1589 }
1590 }
1591
1592 int
1593 mac_hwring_disable_intr(mac_ring_handle_t rh)
1594 {
1595 mac_ring_t *rr_ring = (mac_ring_t *)rh;
1596 mac_intr_t *intr = &rr_ring->mr_info.mri_intr;
1597
1598 return (intr->mi_disable(intr->mi_handle));
1599 }
1600
1601 int
1602 mac_hwring_enable_intr(mac_ring_handle_t rh)
1603 {
1604 mac_ring_t *rr_ring = (mac_ring_t *)rh;
1605 mac_intr_t *intr = &rr_ring->mr_info.mri_intr;
1606
1607 return (intr->mi_enable(intr->mi_handle));
1608 }
1609
1610 int
1611 mac_hwring_start(mac_ring_handle_t rh)
1612 {
1613 mac_ring_t *rr_ring = (mac_ring_t *)rh;
1614
1615 MAC_RING_UNMARK(rr_ring, MR_QUIESCE);
1616 return (0);
1617 }
1618
1619 void
1620 mac_hwring_stop(mac_ring_handle_t rh)
1621 {
1622 mac_ring_t *rr_ring = (mac_ring_t *)rh;
1623
1624 mac_rx_ring_quiesce(rr_ring, MR_QUIESCE);
1625 }
1626
1627 mblk_t *
1628 mac_hwring_poll(mac_ring_handle_t rh, int bytes_to_pickup)
1629 {
1630 mac_ring_t *rr_ring = (mac_ring_t *)rh;
1631 mac_ring_info_t *info = &rr_ring->mr_info;
1632
1633 return (info->mri_poll(info->mri_driver, bytes_to_pickup));
1634 }
1635
1636 /*
1637 * Send packets through a selected tx ring.
1638 */
1639 mblk_t *
1640 mac_hwring_tx(mac_ring_handle_t rh, mblk_t *mp)
1641 {
1642 mac_ring_t *ring = (mac_ring_t *)rh;
1643 mac_ring_info_t *info = &ring->mr_info;
1644
1645 ASSERT(ring->mr_type == MAC_RING_TYPE_TX &&
1646 ring->mr_state >= MR_INUSE);
1647 return (info->mri_tx(info->mri_driver, mp));
1648 }
1649
1650 /*
1651 * Query stats for a particular rx/tx ring
1652 */
1653 int
1654 mac_hwring_getstat(mac_ring_handle_t rh, uint_t stat, uint64_t *val)
1655 {
1656 mac_ring_t *ring = (mac_ring_t *)rh;
1657 mac_ring_info_t *info = &ring->mr_info;
1658
1659 return (info->mri_stat(info->mri_driver, stat, val));
1660 }
1661
1662 /*
1663 * Private function that is only used by aggr to send packets through
1664 * a port/Tx ring. Since aggr exposes a pseudo Tx ring even for ports
1665 * that does not expose Tx rings, aggr_ring_tx() entry point needs
1666 * access to mac_impl_t to send packets through m_tx() entry point.
1667 * It accomplishes this by calling mac_hwring_send_priv() function.
1668 */
1669 mblk_t *
1670 mac_hwring_send_priv(mac_client_handle_t mch, mac_ring_handle_t rh, mblk_t *mp)
1671 {
1672 mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
1673 mac_impl_t *mip = mcip->mci_mip;
1674
1675 MAC_TX(mip, rh, mp, mcip);
1676 return (mp);
1677 }
1678
1679 int
1680 mac_hwgroup_addmac(mac_group_handle_t gh, const uint8_t *addr)
1681 {
1682 mac_group_t *group = (mac_group_t *)gh;
1683
1684 return (mac_group_addmac(group, addr));
1685 }
1686
1687 int
1688 mac_hwgroup_remmac(mac_group_handle_t gh, const uint8_t *addr)
1689 {
1690 mac_group_t *group = (mac_group_t *)gh;
1691
1692 return (mac_group_remmac(group, addr));
1693 }
1694
1695 /*
1696 * Set the RX group to be shared/reserved. Note that the group must be
1697 * started/stopped outside of this function.
1698 */
1699 void
1700 mac_set_group_state(mac_group_t *grp, mac_group_state_t state)
1701 {
1702 /*
1703 * If there is no change in the group state, just return.
1704 */
1705 if (grp->mrg_state == state)
1706 return;
1707
1708 switch (state) {
1709 case MAC_GROUP_STATE_RESERVED:
1710 /*
1711 * Successfully reserved the group.
1712 *
1713 * Given that there is an exclusive client controlling this
1714 * group, we enable the group level polling when available,
1715 * so that SRSs get to turn on/off individual rings they's
1716 * assigned to.
1717 */
1718 ASSERT(MAC_PERIM_HELD(grp->mrg_mh));
1719
1720 if (grp->mrg_type == MAC_RING_TYPE_RX &&
1721 GROUP_INTR_DISABLE_FUNC(grp) != NULL) {
1722 GROUP_INTR_DISABLE_FUNC(grp)(GROUP_INTR_HANDLE(grp));
1723 }
1724 break;
1725
1726 case MAC_GROUP_STATE_SHARED:
1727 /*
1728 * Set all rings of this group to software classified.
1729 * If the group has an overriding interrupt, then re-enable it.
1730 */
1731 ASSERT(MAC_PERIM_HELD(grp->mrg_mh));
1732
1733 if (grp->mrg_type == MAC_RING_TYPE_RX &&
1734 GROUP_INTR_ENABLE_FUNC(grp) != NULL) {
1735 GROUP_INTR_ENABLE_FUNC(grp)(GROUP_INTR_HANDLE(grp));
1736 }
1737 /* The ring is not available for reservations any more */
1738 break;
1739
1740 case MAC_GROUP_STATE_REGISTERED:
1741 /* Also callable from mac_register, perim is not held */
1742 break;
1743
1744 default:
1745 ASSERT(B_FALSE);
1746 break;
1747 }
1748
1749 grp->mrg_state = state;
1750 }
1751
1752 /*
1753 * Quiesce future hardware classified packets for the specified Rx ring
1754 */
1755 static void
1756 mac_rx_ring_quiesce(mac_ring_t *rx_ring, uint_t ring_flag)
1757 {
1758 ASSERT(rx_ring->mr_classify_type == MAC_HW_CLASSIFIER);
1759 ASSERT(ring_flag == MR_CONDEMNED || ring_flag == MR_QUIESCE);
1760
1761 mutex_enter(&rx_ring->mr_lock);
1762 rx_ring->mr_flag |= ring_flag;
1763 while (rx_ring->mr_refcnt != 0)
1764 cv_wait(&rx_ring->mr_cv, &rx_ring->mr_lock);
1765 mutex_exit(&rx_ring->mr_lock);
1766 }
1767
1768 /*
1769 * Please see mac_tx for details about the per cpu locking scheme
1770 */
1771 static void
1772 mac_tx_lock_all(mac_client_impl_t *mcip)
1773 {
1774 int i;
1775
1776 for (i = 0; i <= mac_tx_percpu_cnt; i++)
1777 mutex_enter(&mcip->mci_tx_pcpu[i].pcpu_tx_lock);
1778 }
1779
1780 static void
1781 mac_tx_unlock_all(mac_client_impl_t *mcip)
1782 {
1783 int i;
1784
1785 for (i = mac_tx_percpu_cnt; i >= 0; i--)
1786 mutex_exit(&mcip->mci_tx_pcpu[i].pcpu_tx_lock);
1787 }
1788
1789 static void
1790 mac_tx_unlock_allbutzero(mac_client_impl_t *mcip)
1791 {
1792 int i;
1793
1794 for (i = mac_tx_percpu_cnt; i > 0; i--)
1795 mutex_exit(&mcip->mci_tx_pcpu[i].pcpu_tx_lock);
1796 }
1797
1798 static int
1799 mac_tx_sum_refcnt(mac_client_impl_t *mcip)
1800 {
1801 int i;
1802 int refcnt = 0;
1803
1804 for (i = 0; i <= mac_tx_percpu_cnt; i++)
1805 refcnt += mcip->mci_tx_pcpu[i].pcpu_tx_refcnt;
1806
1807 return (refcnt);
1808 }
1809
1810 /*
1811 * Stop future Tx packets coming down from the client in preparation for
1812 * quiescing the Tx side. This is needed for dynamic reclaim and reassignment
1813 * of rings between clients
1814 */
1815 void
1816 mac_tx_client_block(mac_client_impl_t *mcip)
1817 {
1818 mac_tx_lock_all(mcip);
1819 mcip->mci_tx_flag |= MCI_TX_QUIESCE;
1820 while (mac_tx_sum_refcnt(mcip) != 0) {
1821 mac_tx_unlock_allbutzero(mcip);
1822 cv_wait(&mcip->mci_tx_cv, &mcip->mci_tx_pcpu[0].pcpu_tx_lock);
1823 mutex_exit(&mcip->mci_tx_pcpu[0].pcpu_tx_lock);
1824 mac_tx_lock_all(mcip);
1825 }
1826 mac_tx_unlock_all(mcip);
1827 }
1828
1829 void
1830 mac_tx_client_unblock(mac_client_impl_t *mcip)
1831 {
1832 mac_tx_lock_all(mcip);
1833 mcip->mci_tx_flag &= ~MCI_TX_QUIESCE;
1834 mac_tx_unlock_all(mcip);
1835 /*
1836 * We may fail to disable flow control for the last MAC_NOTE_TX
1837 * notification because the MAC client is quiesced. Send the
1838 * notification again.
1839 */
1840 i_mac_notify(mcip->mci_mip, MAC_NOTE_TX);
1841 }
1842
1843 /*
1844 * Wait for an SRS to quiesce. The SRS worker will signal us when the
1845 * quiesce is done.
1846 */
1847 static void
1848 mac_srs_quiesce_wait(mac_soft_ring_set_t *srs, uint_t srs_flag)
1849 {
1850 mutex_enter(&srs->srs_lock);
1851 while (!(srs->srs_state & srs_flag))
1852 cv_wait(&srs->srs_quiesce_done_cv, &srs->srs_lock);
1853 mutex_exit(&srs->srs_lock);
1854 }
1855
1856 /*
1857 * Quiescing an Rx SRS is achieved by the following sequence. The protocol
1858 * works bottom up by cutting off packet flow from the bottommost point in the
1859 * mac, then the SRS, and then the soft rings. There are 2 use cases of this
1860 * mechanism. One is a temporary quiesce of the SRS, such as say while changing
1861 * the Rx callbacks. Another use case is Rx SRS teardown. In the former case
1862 * the QUIESCE prefix/suffix is used and in the latter the CONDEMNED is used
1863 * for the SRS and MR flags. In the former case the threads pause waiting for
1864 * a restart, while in the latter case the threads exit. The Tx SRS teardown
1865 * is also mostly similar to the above.
1866 *
1867 * 1. Stop future hardware classified packets at the lowest level in the mac.
1868 * Remove any hardware classification rule (CONDEMNED case) and mark the
1869 * rings as CONDEMNED or QUIESCE as appropriate. This prevents the mr_refcnt
1870 * from increasing. Upcalls from the driver that come through hardware
1871 * classification will be dropped in mac_rx from now on. Then we wait for
1872 * the mr_refcnt to drop to zero. When the mr_refcnt reaches zero we are
1873 * sure there aren't any upcall threads from the driver through hardware
1874 * classification. In the case of SRS teardown we also remove the
1875 * classification rule in the driver.
1876 *
1877 * 2. Stop future software classified packets by marking the flow entry with
1878 * FE_QUIESCE or FE_CONDEMNED as appropriate which prevents the refcnt from
1879 * increasing. We also remove the flow entry from the table in the latter
1880 * case. Then wait for the fe_refcnt to reach an appropriate quiescent value
1881 * that indicates there aren't any active threads using that flow entry.
1882 *
1883 * 3. Quiesce the SRS and softrings by signaling the SRS. The SRS poll thread,
1884 * SRS worker thread, and the soft ring threads are quiesced in sequence
1885 * with the SRS worker thread serving as a master controller. This
1886 * mechansim is explained in mac_srs_worker_quiesce().
1887 *
1888 * The restart mechanism to reactivate the SRS and softrings is explained
1889 * in mac_srs_worker_restart(). Here we just signal the SRS worker to start the
1890 * restart sequence.
1891 */
1892 void
1893 mac_rx_srs_quiesce(mac_soft_ring_set_t *srs, uint_t srs_quiesce_flag)
1894 {
1895 flow_entry_t *flent = srs->srs_flent;
1896 uint_t mr_flag, srs_done_flag;
1897
1898 ASSERT(MAC_PERIM_HELD((mac_handle_t)FLENT_TO_MIP(flent)));
1899 ASSERT(!(srs->srs_type & SRST_TX));
1900
1901 if (srs_quiesce_flag == SRS_CONDEMNED) {
1902 mr_flag = MR_CONDEMNED;
1903 srs_done_flag = SRS_CONDEMNED_DONE;
1904 if (srs->srs_type & SRST_CLIENT_POLL_ENABLED)
1905 mac_srs_client_poll_disable(srs->srs_mcip, srs);
1906 } else {
1907 ASSERT(srs_quiesce_flag == SRS_QUIESCE);
1908 mr_flag = MR_QUIESCE;
1909 srs_done_flag = SRS_QUIESCE_DONE;
1910 if (srs->srs_type & SRST_CLIENT_POLL_ENABLED)
1911 mac_srs_client_poll_quiesce(srs->srs_mcip, srs);
1912 }
1913
1914 if (srs->srs_ring != NULL) {
1915 mac_rx_ring_quiesce(srs->srs_ring, mr_flag);
1916 } else {
1917 /*
1918 * SRS is driven by software classification. In case
1919 * of CONDEMNED, the top level teardown functions will
1920 * deal with flow removal.
1921 */
1922 if (srs_quiesce_flag != SRS_CONDEMNED) {
1923 FLOW_MARK(flent, FE_QUIESCE);
1924 mac_flow_wait(flent, FLOW_DRIVER_UPCALL);
1925 }
1926 }
1927
1928 /*
1929 * Signal the SRS to quiesce itself, and then cv_wait for the
1930 * SRS quiesce to complete. The SRS worker thread will wake us
1931 * up when the quiesce is complete
1932 */
1933 mac_srs_signal(srs, srs_quiesce_flag);
1934 mac_srs_quiesce_wait(srs, srs_done_flag);
1935 }
1936
1937 /*
1938 * Remove an SRS.
1939 */
1940 void
1941 mac_rx_srs_remove(mac_soft_ring_set_t *srs)
1942 {
1943 flow_entry_t *flent = srs->srs_flent;
1944 int i;
1945
1946 mac_rx_srs_quiesce(srs, SRS_CONDEMNED);
1947 /*
1948 * Locate and remove our entry in the fe_rx_srs[] array, and
1949 * adjust the fe_rx_srs array entries and array count by
1950 * moving the last entry into the vacated spot.
1951 */
1952 mutex_enter(&flent->fe_lock);
1953 for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
1954 if (flent->fe_rx_srs[i] == srs)
1955 break;
1956 }
1957
1958 ASSERT(i != 0 && i < flent->fe_rx_srs_cnt);
1959 if (i != flent->fe_rx_srs_cnt - 1) {
1960 flent->fe_rx_srs[i] =
1961 flent->fe_rx_srs[flent->fe_rx_srs_cnt - 1];
1962 i = flent->fe_rx_srs_cnt - 1;
1963 }
1964
1965 flent->fe_rx_srs[i] = NULL;
1966 flent->fe_rx_srs_cnt--;
1967 mutex_exit(&flent->fe_lock);
1968
1969 mac_srs_free(srs);
1970 }
1971
1972 static void
1973 mac_srs_clear_flag(mac_soft_ring_set_t *srs, uint_t flag)
1974 {
1975 mutex_enter(&srs->srs_lock);
1976 srs->srs_state &= ~flag;
1977 mutex_exit(&srs->srs_lock);
1978 }
1979
1980 void
1981 mac_rx_srs_restart(mac_soft_ring_set_t *srs)
1982 {
1983 flow_entry_t *flent = srs->srs_flent;
1984 mac_ring_t *mr;
1985
1986 ASSERT(MAC_PERIM_HELD((mac_handle_t)FLENT_TO_MIP(flent)));
1987 ASSERT((srs->srs_type & SRST_TX) == 0);
1988
1989 /*
1990 * This handles a change in the number of SRSs between the quiesce and
1991 * and restart operation of a flow.
1992 */
1993 if (!SRS_QUIESCED(srs))
1994 return;
1995
1996 /*
1997 * Signal the SRS to restart itself. Wait for the restart to complete
1998 * Note that we only restart the SRS if it is not marked as
1999 * permanently quiesced.
2000 */
2001 if (!SRS_QUIESCED_PERMANENT(srs)) {
2002 mac_srs_signal(srs, SRS_RESTART);
2003 mac_srs_quiesce_wait(srs, SRS_RESTART_DONE);
2004 mac_srs_clear_flag(srs, SRS_RESTART_DONE);
2005
2006 mac_srs_client_poll_restart(srs->srs_mcip, srs);
2007 }
2008
2009 /* Finally clear the flags to let the packets in */
2010 mr = srs->srs_ring;
2011 if (mr != NULL) {
2012 MAC_RING_UNMARK(mr, MR_QUIESCE);
2013 /* In case the ring was stopped, safely restart it */
2014 if (mr->mr_state != MR_INUSE)
2015 (void) mac_start_ring(mr);
2016 } else {
2017 FLOW_UNMARK(flent, FE_QUIESCE);
2018 }
2019 }
2020
2021 /*
2022 * Temporary quiesce of a flow and associated Rx SRS.
2023 * Please see block comment above mac_rx_classify_flow_rem.
2024 */
2025 /* ARGSUSED */
2026 int
2027 mac_rx_classify_flow_quiesce(flow_entry_t *flent, void *arg)
2028 {
2029 int i;
2030
2031 for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
2032 mac_rx_srs_quiesce((mac_soft_ring_set_t *)flent->fe_rx_srs[i],
2033 SRS_QUIESCE);
2034 }
2035 return (0);
2036 }
2037
2038 /*
2039 * Restart a flow and associated Rx SRS that has been quiesced temporarily
2040 * Please see block comment above mac_rx_classify_flow_rem
2041 */
2042 /* ARGSUSED */
2043 int
2044 mac_rx_classify_flow_restart(flow_entry_t *flent, void *arg)
2045 {
2046 int i;
2047
2048 for (i = 0; i < flent->fe_rx_srs_cnt; i++)
2049 mac_rx_srs_restart((mac_soft_ring_set_t *)flent->fe_rx_srs[i]);
2050
2051 return (0);
2052 }
2053
2054 void
2055 mac_srs_perm_quiesce(mac_client_handle_t mch, boolean_t on)
2056 {
2057 mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
2058 flow_entry_t *flent = mcip->mci_flent;
2059 mac_impl_t *mip = mcip->mci_mip;
2060 mac_soft_ring_set_t *mac_srs;
2061 int i;
2062
2063 ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
2064
2065 if (flent == NULL)
2066 return;
2067
2068 for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
2069 mac_srs = flent->fe_rx_srs[i];
2070 mutex_enter(&mac_srs->srs_lock);
2071 if (on)
2072 mac_srs->srs_state |= SRS_QUIESCE_PERM;
2073 else
2074 mac_srs->srs_state &= ~SRS_QUIESCE_PERM;
2075 mutex_exit(&mac_srs->srs_lock);
2076 }
2077 }
2078
2079 void
2080 mac_rx_client_quiesce(mac_client_handle_t mch)
2081 {
2082 mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
2083 mac_impl_t *mip = mcip->mci_mip;
2084
2085 ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
2086
2087 if (MCIP_DATAPATH_SETUP(mcip)) {
2088 (void) mac_rx_classify_flow_quiesce(mcip->mci_flent,
2089 NULL);
2090 (void) mac_flow_walk_nolock(mcip->mci_subflow_tab,
2091 mac_rx_classify_flow_quiesce, NULL);
2092 }
2093 }
2094
2095 void
2096 mac_rx_client_restart(mac_client_handle_t mch)
2097 {
2098 mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
2099 mac_impl_t *mip = mcip->mci_mip;
2100
2101 ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
2102
2103 if (MCIP_DATAPATH_SETUP(mcip)) {
2104 (void) mac_rx_classify_flow_restart(mcip->mci_flent, NULL);
2105 (void) mac_flow_walk_nolock(mcip->mci_subflow_tab,
2106 mac_rx_classify_flow_restart, NULL);
2107 }
2108 }
2109
2110 /*
2111 * This function only quiesces the Tx SRS and softring worker threads. Callers
2112 * need to make sure that there aren't any mac client threads doing current or
2113 * future transmits in the mac before calling this function.
2114 */
2115 void
2116 mac_tx_srs_quiesce(mac_soft_ring_set_t *srs, uint_t srs_quiesce_flag)
2117 {
2118 mac_client_impl_t *mcip = srs->srs_mcip;
2119
2120 ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
2121
2122 ASSERT(srs->srs_type & SRST_TX);
2123 ASSERT(srs_quiesce_flag == SRS_CONDEMNED ||
2124 srs_quiesce_flag == SRS_QUIESCE);
2125
2126 /*
2127 * Signal the SRS to quiesce itself, and then cv_wait for the
2128 * SRS quiesce to complete. The SRS worker thread will wake us
2129 * up when the quiesce is complete
2130 */
2131 mac_srs_signal(srs, srs_quiesce_flag);
2132 mac_srs_quiesce_wait(srs, srs_quiesce_flag == SRS_QUIESCE ?
2133 SRS_QUIESCE_DONE : SRS_CONDEMNED_DONE);
2134 }
2135
2136 void
2137 mac_tx_srs_restart(mac_soft_ring_set_t *srs)
2138 {
2139 /*
2140 * Resizing the fanout could result in creation of new SRSs.
2141 * They may not necessarily be in the quiesced state in which
2142 * case it need be restarted
2143 */
2144 if (!SRS_QUIESCED(srs))
2145 return;
2146
2147 mac_srs_signal(srs, SRS_RESTART);
2148 mac_srs_quiesce_wait(srs, SRS_RESTART_DONE);
2149 mac_srs_clear_flag(srs, SRS_RESTART_DONE);
2150 }
2151
2152 /*
2153 * Temporary quiesce of a flow and associated Rx SRS.
2154 * Please see block comment above mac_rx_srs_quiesce
2155 */
2156 /* ARGSUSED */
2157 int
2158 mac_tx_flow_quiesce(flow_entry_t *flent, void *arg)
2159 {
2160 /*
2161 * The fe_tx_srs is null for a subflow on an interface that is
2162 * not plumbed
2163 */
2164 if (flent->fe_tx_srs != NULL)
2165 mac_tx_srs_quiesce(flent->fe_tx_srs, SRS_QUIESCE);
2166 return (0);
2167 }
2168
2169 /* ARGSUSED */
2170 int
2171 mac_tx_flow_restart(flow_entry_t *flent, void *arg)
2172 {
2173 /*
2174 * The fe_tx_srs is null for a subflow on an interface that is
2175 * not plumbed
2176 */
2177 if (flent->fe_tx_srs != NULL)
2178 mac_tx_srs_restart(flent->fe_tx_srs);
2179 return (0);
2180 }
2181
2182 static void
2183 i_mac_tx_client_quiesce(mac_client_handle_t mch, uint_t srs_quiesce_flag)
2184 {
2185 mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
2186
2187 ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
2188
2189 mac_tx_client_block(mcip);
2190 if (MCIP_TX_SRS(mcip) != NULL) {
2191 mac_tx_srs_quiesce(MCIP_TX_SRS(mcip), srs_quiesce_flag);
2192 (void) mac_flow_walk_nolock(mcip->mci_subflow_tab,
2193 mac_tx_flow_quiesce, NULL);
2194 }
2195 }
2196
2197 void
2198 mac_tx_client_quiesce(mac_client_handle_t mch)
2199 {
2200 i_mac_tx_client_quiesce(mch, SRS_QUIESCE);
2201 }
2202
2203 void
2204 mac_tx_client_condemn(mac_client_handle_t mch)
2205 {
2206 i_mac_tx_client_quiesce(mch, SRS_CONDEMNED);
2207 }
2208
2209 void
2210 mac_tx_client_restart(mac_client_handle_t mch)
2211 {
2212 mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
2213
2214 ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
2215
2216 mac_tx_client_unblock(mcip);
2217 if (MCIP_TX_SRS(mcip) != NULL) {
2218 mac_tx_srs_restart(MCIP_TX_SRS(mcip));
2219 (void) mac_flow_walk_nolock(mcip->mci_subflow_tab,
2220 mac_tx_flow_restart, NULL);
2221 }
2222 }
2223
2224 void
2225 mac_tx_client_flush(mac_client_impl_t *mcip)
2226 {
2227 ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
2228
2229 mac_tx_client_quiesce((mac_client_handle_t)mcip);
2230 mac_tx_client_restart((mac_client_handle_t)mcip);
2231 }
2232
2233 void
2234 mac_client_quiesce(mac_client_impl_t *mcip)
2235 {
2236 mac_rx_client_quiesce((mac_client_handle_t)mcip);
2237 mac_tx_client_quiesce((mac_client_handle_t)mcip);
2238 }
2239
2240 void
2241 mac_client_restart(mac_client_impl_t *mcip)
2242 {
2243 mac_rx_client_restart((mac_client_handle_t)mcip);
2244 mac_tx_client_restart((mac_client_handle_t)mcip);
2245 }
2246
2247 /*
2248 * Allocate a minor number.
2249 */
2250 minor_t
2251 mac_minor_hold(boolean_t sleep)
2252 {
2253 minor_t minor;
2254
2255 /*
2256 * Grab a value from the arena.
2257 */
2258 atomic_add_32(&minor_count, 1);
2259
2260 if (sleep)
2261 minor = (uint_t)id_alloc(minor_ids);
2262 else
2263 minor = (uint_t)id_alloc_nosleep(minor_ids);
2264
2265 if (minor == 0) {
2266 atomic_add_32(&minor_count, -1);
2267 return (0);
2268 }
2269
2270 return (minor);
2271 }
2272
2273 /*
2274 * Release a previously allocated minor number.
2275 */
2276 void
2277 mac_minor_rele(minor_t minor)
2278 {
2279 /*
2280 * Return the value to the arena.
2281 */
2282 id_free(minor_ids, minor);
2283 atomic_add_32(&minor_count, -1);
2284 }
2285
2286 uint32_t
2287 mac_no_notification(mac_handle_t mh)
2288 {
2289 mac_impl_t *mip = (mac_impl_t *)mh;
2290
2291 return (((mip->mi_state_flags & MIS_LEGACY) != 0) ?
2292 mip->mi_capab_legacy.ml_unsup_note : 0);
2293 }
2294
2295 /*
2296 * Prevent any new opens of this mac in preparation for unregister
2297 */
2298 int
2299 i_mac_disable(mac_impl_t *mip)
2300 {
2301 mac_client_impl_t *mcip;
2302
2303 rw_enter(&i_mac_impl_lock, RW_WRITER);
2304 if (mip->mi_state_flags & MIS_DISABLED) {
2305 /* Already disabled, return success */
2306 rw_exit(&i_mac_impl_lock);
2307 return (0);
2308 }
2309 /*
2310 * See if there are any other references to this mac_t (e.g., VLAN's).
2311 * If so return failure. If all the other checks below pass, then
2312 * set mi_disabled atomically under the i_mac_impl_lock to prevent
2313 * any new VLAN's from being created or new mac client opens of this
2314 * mac end point.
2315 */
2316 if (mip->mi_ref > 0) {
2317 rw_exit(&i_mac_impl_lock);
2318 return (EBUSY);
2319 }
2320
2321 /*
2322 * mac clients must delete all multicast groups they join before
2323 * closing. bcast groups are reference counted, the last client
2324 * to delete the group will wait till the group is physically
2325 * deleted. Since all clients have closed this mac end point
2326 * mi_bcast_ngrps must be zero at this point
2327 */
2328 ASSERT(mip->mi_bcast_ngrps == 0);
2329
2330 /*
2331 * Don't let go of this if it has some flows.
2332 * All other code guarantees no flows are added to a disabled
2333 * mac, therefore it is sufficient to check for the flow table
2334 * only here.
2335 */
2336 mcip = mac_primary_client_handle(mip);
2337 if ((mcip != NULL) && mac_link_has_flows((mac_client_handle_t)mcip)) {
2338 rw_exit(&i_mac_impl_lock);
2339 return (ENOTEMPTY);
2340 }
2341
2342 mip->mi_state_flags |= MIS_DISABLED;
2343 rw_exit(&i_mac_impl_lock);
2344 return (0);
2345 }
2346
2347 int
2348 mac_disable_nowait(mac_handle_t mh)
2349 {
2350 mac_impl_t *mip = (mac_impl_t *)mh;
2351 int err;
2352
2353 if ((err = i_mac_perim_enter_nowait(mip)) != 0)
2354 return (err);
2355 err = i_mac_disable(mip);
2356 i_mac_perim_exit(mip);
2357 return (err);
2358 }
2359
2360 int
2361 mac_disable(mac_handle_t mh)
2362 {
2363 mac_impl_t *mip = (mac_impl_t *)mh;
2364 int err;
2365
2366 i_mac_perim_enter(mip);
2367 err = i_mac_disable(mip);
2368 i_mac_perim_exit(mip);
2369
2370 /*
2371 * Clean up notification thread and wait for it to exit.
2372 */
2373 if (err == 0)
2374 i_mac_notify_exit(mip);
2375
2376 return (err);
2377 }
2378
2379 /*
2380 * Called when the MAC instance has a non empty flow table, to de-multiplex
2381 * incoming packets to the right flow.
2382 * The MAC's rw lock is assumed held as a READER.
2383 */
2384 /* ARGSUSED */
2385 static mblk_t *
2386 mac_rx_classify(mac_impl_t *mip, mac_resource_handle_t mrh, mblk_t *mp)
2387 {
2388 flow_entry_t *flent = NULL;
2389 uint_t flags = FLOW_INBOUND;
2390 int err;
2391
2392 /*
2393 * If the mac is a port of an aggregation, pass FLOW_IGNORE_VLAN
2394 * to mac_flow_lookup() so that the VLAN packets can be successfully
2395 * passed to the non-VLAN aggregation flows.
2396 *
2397 * Note that there is possibly a race between this and
2398 * mac_unicast_remove/add() and VLAN packets could be incorrectly
2399 * classified to non-VLAN flows of non-aggregation mac clients. These
2400 * VLAN packets will be then filtered out by the mac module.
2401 */
2402 if ((mip->mi_state_flags & MIS_EXCLUSIVE) != 0)
2403 flags |= FLOW_IGNORE_VLAN;
2404
2405 err = mac_flow_lookup(mip->mi_flow_tab, mp, flags, &flent);
2406 if (err != 0) {
2407 /* no registered receive function */
2408 return (mp);
2409 } else {
2410 mac_client_impl_t *mcip;
2411
2412 /*
2413 * This flent might just be an additional one on the MAC client,
2414 * i.e. for classification purposes (different fdesc), however
2415 * the resources, SRS et. al., are in the mci_flent, so if
2416 * this isn't the mci_flent, we need to get it.
2417 */
2418 if ((mcip = flent->fe_mcip) != NULL &&
2419 mcip->mci_flent != flent) {
2420 FLOW_REFRELE(flent);
2421 flent = mcip->mci_flent;
2422 FLOW_TRY_REFHOLD(flent, err);
2423 if (err != 0)
2424 return (mp);
2425 }
2426 (flent->fe_cb_fn)(flent->fe_cb_arg1, flent->fe_cb_arg2, mp,
2427 B_FALSE);
2428 FLOW_REFRELE(flent);
2429 }
2430 return (NULL);
2431 }
2432
2433 mblk_t *
2434 mac_rx_flow(mac_handle_t mh, mac_resource_handle_t mrh, mblk_t *mp_chain)
2435 {
2436 mac_impl_t *mip = (mac_impl_t *)mh;
2437 mblk_t *bp, *bp1, **bpp, *list = NULL;
2438
2439 /*
2440 * We walk the chain and attempt to classify each packet.
2441 * The packets that couldn't be classified will be returned
2442 * back to the caller.
2443 */
2444 bp = mp_chain;
2445 bpp = &list;
2446 while (bp != NULL) {
2447 bp1 = bp;
2448 bp = bp->b_next;
2449 bp1->b_next = NULL;
2450
2451 if (mac_rx_classify(mip, mrh, bp1) != NULL) {
2452 *bpp = bp1;
2453 bpp = &bp1->b_next;
2454 }
2455 }
2456 return (list);
2457 }
2458
2459 static int
2460 mac_tx_flow_srs_wakeup(flow_entry_t *flent, void *arg)
2461 {
2462 mac_ring_handle_t ring = arg;
2463
2464 if (flent->fe_tx_srs)
2465 mac_tx_srs_wakeup(flent->fe_tx_srs, ring);
2466 return (0);
2467 }
2468
2469 void
2470 i_mac_tx_srs_notify(mac_impl_t *mip, mac_ring_handle_t ring)
2471 {
2472 mac_client_impl_t *cclient;
2473 mac_soft_ring_set_t *mac_srs;
2474
2475 /*
2476 * After grabbing the mi_rw_lock, the list of clients can't change.
2477 * If there are any clients mi_disabled must be B_FALSE and can't
2478 * get set since there are clients. If there aren't any clients we
2479 * don't do anything. In any case the mip has to be valid. The driver
2480 * must make sure that it goes single threaded (with respect to mac
2481 * calls) and wait for all pending mac calls to finish before calling
2482 * mac_unregister.
2483 */
2484 rw_enter(&i_mac_impl_lock, RW_READER);
2485 if (mip->mi_state_flags & MIS_DISABLED) {
2486 rw_exit(&i_mac_impl_lock);
2487 return;
2488 }
2489
2490 /*
2491 * Get MAC tx srs from walking mac_client_handle list.
2492 */
2493 rw_enter(&mip->mi_rw_lock, RW_READER);
2494 for (cclient = mip->mi_clients_list; cclient != NULL;
2495 cclient = cclient->mci_client_next) {
2496 if ((mac_srs = MCIP_TX_SRS(cclient)) != NULL) {
2497 mac_tx_srs_wakeup(mac_srs, ring);
2498 } else {
2499 /*
2500 * Aggr opens underlying ports in exclusive mode
2501 * and registers flow control callbacks using
2502 * mac_tx_client_notify(). When opened in
2503 * exclusive mode, Tx SRS won't be created
2504 * during mac_unicast_add().
2505 */
2506 if (cclient->mci_state_flags & MCIS_EXCLUSIVE) {
2507 mac_tx_invoke_callbacks(cclient,
2508 (mac_tx_cookie_t)ring);
2509 }
2510 }
2511 (void) mac_flow_walk(cclient->mci_subflow_tab,
2512 mac_tx_flow_srs_wakeup, ring);
2513 }
2514 rw_exit(&mip->mi_rw_lock);
2515 rw_exit(&i_mac_impl_lock);
2516 }
2517
2518 /* ARGSUSED */
2519 void
2520 mac_multicast_refresh(mac_handle_t mh, mac_multicst_t refresh, void *arg,
2521 boolean_t add)
2522 {
2523 mac_impl_t *mip = (mac_impl_t *)mh;
2524
2525 i_mac_perim_enter((mac_impl_t *)mh);
2526 /*
2527 * If no specific refresh function was given then default to the
2528 * driver's m_multicst entry point.
2529 */
2530 if (refresh == NULL) {
2531 refresh = mip->mi_multicst;
2532 arg = mip->mi_driver;
2533 }
2534
2535 mac_bcast_refresh(mip, refresh, arg, add);
2536 i_mac_perim_exit((mac_impl_t *)mh);
2537 }
2538
2539 void
2540 mac_promisc_refresh(mac_handle_t mh, mac_setpromisc_t refresh, void *arg)
2541 {
2542 mac_impl_t *mip = (mac_impl_t *)mh;
2543
2544 /*
2545 * If no specific refresh function was given then default to the
2546 * driver's m_promisc entry point.
2547 */
2548 if (refresh == NULL) {
2549 refresh = mip->mi_setpromisc;
2550 arg = mip->mi_driver;
2551 }
2552 ASSERT(refresh != NULL);
2553
2554 /*
2555 * Call the refresh function with the current promiscuity.
2556 */
2557 refresh(arg, (mip->mi_devpromisc != 0));
2558 }
2559
2560 /*
2561 * The mac client requests that the mac not to change its margin size to
2562 * be less than the specified value. If "current" is B_TRUE, then the client
2563 * requests the mac not to change its margin size to be smaller than the
2564 * current size. Further, return the current margin size value in this case.
2565 *
2566 * We keep every requested size in an ordered list from largest to smallest.
2567 */
2568 int
2569 mac_margin_add(mac_handle_t mh, uint32_t *marginp, boolean_t current)
2570 {
2571 mac_impl_t *mip = (mac_impl_t *)mh;
2572 mac_margin_req_t **pp, *p;
2573 int err = 0;
2574
2575 rw_enter(&(mip->mi_rw_lock), RW_WRITER);
2576 if (current)
2577 *marginp = mip->mi_margin;
2578
2579 /*
2580 * If the current margin value cannot satisfy the margin requested,
2581 * return ENOTSUP directly.
2582 */
2583 if (*marginp > mip->mi_margin) {
2584 err = ENOTSUP;
2585 goto done;
2586 }
2587
2588 /*
2589 * Check whether the given margin is already in the list. If so,
2590 * bump the reference count.
2591 */
2592 for (pp = &mip->mi_mmrp; (p = *pp) != NULL; pp = &p->mmr_nextp) {
2593 if (p->mmr_margin == *marginp) {
2594 /*
2595 * The margin requested is already in the list,
2596 * so just bump the reference count.
2597 */
2598 p->mmr_ref++;
2599 goto done;
2600 }
2601 if (p->mmr_margin < *marginp)
2602 break;
2603 }
2604
2605
2606 p = kmem_zalloc(sizeof (mac_margin_req_t), KM_SLEEP);
2607 p->mmr_margin = *marginp;
2608 p->mmr_ref++;
2609 p->mmr_nextp = *pp;
2610 *pp = p;
2611
2612 done:
2613 rw_exit(&(mip->mi_rw_lock));
2614 return (err);
2615 }
2616
2617 /*
2618 * The mac client requests to cancel its previous mac_margin_add() request.
2619 * We remove the requested margin size from the list.
2620 */
2621 int
2622 mac_margin_remove(mac_handle_t mh, uint32_t margin)
2623 {
2624 mac_impl_t *mip = (mac_impl_t *)mh;
2625 mac_margin_req_t **pp, *p;
2626 int err = 0;
2627
2628 rw_enter(&(mip->mi_rw_lock), RW_WRITER);
2629 /*
2630 * Find the entry in the list for the given margin.
2631 */
2632 for (pp = &(mip->mi_mmrp); (p = *pp) != NULL; pp = &(p->mmr_nextp)) {
2633 if (p->mmr_margin == margin) {
2634 if (--p->mmr_ref == 0)
2635 break;
2636
2637 /*
2638 * There is still a reference to this address so
2639 * there's nothing more to do.
2640 */
2641 goto done;
2642 }
2643 }
2644
2645 /*
2646 * We did not find an entry for the given margin.
2647 */
2648 if (p == NULL) {
2649 err = ENOENT;
2650 goto done;
2651 }
2652
2653 ASSERT(p->mmr_ref == 0);
2654
2655 /*
2656 * Remove it from the list.
2657 */
2658 *pp = p->mmr_nextp;
2659 kmem_free(p, sizeof (mac_margin_req_t));
2660 done:
2661 rw_exit(&(mip->mi_rw_lock));
2662 return (err);
2663 }
2664
2665 boolean_t
2666 mac_margin_update(mac_handle_t mh, uint32_t margin)
2667 {
2668 mac_impl_t *mip = (mac_impl_t *)mh;
2669 uint32_t margin_needed = 0;
2670
2671 rw_enter(&(mip->mi_rw_lock), RW_WRITER);
2672
2673 if (mip->mi_mmrp != NULL)
2674 margin_needed = mip->mi_mmrp->mmr_margin;
2675
2676 if (margin_needed <= margin)
2677 mip->mi_margin = margin;
2678
2679 rw_exit(&(mip->mi_rw_lock));
2680
2681 if (margin_needed <= margin)
2682 i_mac_notify(mip, MAC_NOTE_MARGIN);
2683
2684 return (margin_needed <= margin);
2685 }
2686
2687 /*
2688 * MAC Type Plugin functions.
2689 */
2690
2691 mactype_t *
2692 mactype_getplugin(const char *pname)
2693 {
2694 mactype_t *mtype = NULL;
2695 boolean_t tried_modload = B_FALSE;
2696
2697 mutex_enter(&i_mactype_lock);
2698
2699 find_registered_mactype:
2700 if (mod_hash_find(i_mactype_hash, (mod_hash_key_t)pname,
2701 (mod_hash_val_t *)&mtype) != 0) {
2702 if (!tried_modload) {
2703 /*
2704 * If the plugin has not yet been loaded, then
2705 * attempt to load it now. If modload() succeeds,
2706 * the plugin should have registered using
2707 * mactype_register(), in which case we can go back
2708 * and attempt to find it again.
2709 */
2710 if (modload(MACTYPE_KMODDIR, (char *)pname) != -1) {
2711 tried_modload = B_TRUE;
2712 goto find_registered_mactype;
2713 }
2714 }
2715 } else {
2716 /*
2717 * Note that there's no danger that the plugin we've loaded
2718 * could be unloaded between the modload() step and the
2719 * reference count bump here, as we're holding
2720 * i_mactype_lock, which mactype_unregister() also holds.
2721 */
2722 atomic_inc_32(&mtype->mt_ref);
2723 }
2724
2725 mutex_exit(&i_mactype_lock);
2726 return (mtype);
2727 }
2728
2729 mactype_register_t *
2730 mactype_alloc(uint_t mactype_version)
2731 {
2732 mactype_register_t *mtrp;
2733
2734 /*
2735 * Make sure there isn't a version mismatch between the plugin and
2736 * the framework. In the future, if multiple versions are
2737 * supported, this check could become more sophisticated.
2738 */
2739 if (mactype_version != MACTYPE_VERSION)
2740 return (NULL);
2741
2742 mtrp = kmem_zalloc(sizeof (mactype_register_t), KM_SLEEP);
2743 mtrp->mtr_version = mactype_version;
2744 return (mtrp);
2745 }
2746
2747 void
2748 mactype_free(mactype_register_t *mtrp)
2749 {
2750 kmem_free(mtrp, sizeof (mactype_register_t));
2751 }
2752
2753 int
2754 mactype_register(mactype_register_t *mtrp)
2755 {
2756 mactype_t *mtp;
2757 mactype_ops_t *ops = mtrp->mtr_ops;
2758
2759 /* Do some sanity checking before we register this MAC type. */
2760 if (mtrp->mtr_ident == NULL || ops == NULL)
2761 return (EINVAL);
2762
2763 /*
2764 * Verify that all mandatory callbacks are set in the ops
2765 * vector.
2766 */
2767 if (ops->mtops_unicst_verify == NULL ||
2768 ops->mtops_multicst_verify == NULL ||
2769 ops->mtops_sap_verify == NULL ||
2770 ops->mtops_header == NULL ||
2771 ops->mtops_header_info == NULL) {
2772 return (EINVAL);
2773 }
2774
2775 mtp = kmem_zalloc(sizeof (*mtp), KM_SLEEP);
2776 mtp->mt_ident = mtrp->mtr_ident;
2777 mtp->mt_ops = *ops;
2778 mtp->mt_type = mtrp->mtr_mactype;
2779 mtp->mt_nativetype = mtrp->mtr_nativetype;
2780 mtp->mt_addr_length = mtrp->mtr_addrlen;
2781 if (mtrp->mtr_brdcst_addr != NULL) {
2782 mtp->mt_brdcst_addr = kmem_alloc(mtrp->mtr_addrlen, KM_SLEEP);
2783 bcopy(mtrp->mtr_brdcst_addr, mtp->mt_brdcst_addr,
2784 mtrp->mtr_addrlen);
2785 }
2786
2787 mtp->mt_stats = mtrp->mtr_stats;
2788 mtp->mt_statcount = mtrp->mtr_statcount;
2789
2790 mtp->mt_mapping = mtrp->mtr_mapping;
2791 mtp->mt_mappingcount = mtrp->mtr_mappingcount;
2792
2793 if (mod_hash_insert(i_mactype_hash,
2794 (mod_hash_key_t)mtp->mt_ident, (mod_hash_val_t)mtp) != 0) {
2795 kmem_free(mtp->mt_brdcst_addr, mtp->mt_addr_length);
2796 kmem_free(mtp, sizeof (*mtp));
2797 return (EEXIST);
2798 }
2799 return (0);
2800 }
2801
2802 int
2803 mactype_unregister(const char *ident)
2804 {
2805 mactype_t *mtp;
2806 mod_hash_val_t val;
2807 int err;
2808
2809 /*
2810 * Let's not allow MAC drivers to use this plugin while we're
2811 * trying to unregister it. Holding i_mactype_lock also prevents a
2812 * plugin from unregistering while a MAC driver is attempting to
2813 * hold a reference to it in i_mactype_getplugin().
2814 */
2815 mutex_enter(&i_mactype_lock);
2816
2817 if ((err = mod_hash_find(i_mactype_hash, (mod_hash_key_t)ident,
2818 (mod_hash_val_t *)&mtp)) != 0) {
2819 /* A plugin is trying to unregister, but it never registered. */
2820 err = ENXIO;
2821 goto done;
2822 }
2823
2824 if (mtp->mt_ref != 0) {
2825 err = EBUSY;
2826 goto done;
2827 }
2828
2829 err = mod_hash_remove(i_mactype_hash, (mod_hash_key_t)ident, &val);
2830 ASSERT(err == 0);
2831 if (err != 0) {
2832 /* This should never happen, thus the ASSERT() above. */
2833 err = EINVAL;
2834 goto done;
2835 }
2836 ASSERT(mtp == (mactype_t *)val);
2837
2838 if (mtp->mt_brdcst_addr != NULL)
2839 kmem_free(mtp->mt_brdcst_addr, mtp->mt_addr_length);
2840 kmem_free(mtp, sizeof (mactype_t));
2841 done:
2842 mutex_exit(&i_mactype_lock);
2843 return (err);
2844 }
2845
2846 /*
2847 * Checks the size of the value size specified for a property as
2848 * part of a property operation. Returns B_TRUE if the size is
2849 * correct, B_FALSE otherwise.
2850 */
2851 boolean_t
2852 mac_prop_check_size(mac_prop_id_t id, uint_t valsize, boolean_t is_range)
2853 {
2854 uint_t minsize = 0;
2855
2856 if (is_range)
2857 return (valsize >= sizeof (mac_propval_range_t));
2858
2859 switch (id) {
2860 case MAC_PROP_ZONE:
2861 minsize = sizeof (dld_ioc_zid_t);
2862 break;
2863 case MAC_PROP_AUTOPUSH:
2864 if (valsize != 0)
2865 minsize = sizeof (struct dlautopush);
2866 break;
2867 case MAC_PROP_TAGMODE:
2868 minsize = sizeof (link_tagmode_t);
2869 break;
2870 case MAC_PROP_RESOURCE:
2871 case MAC_PROP_RESOURCE_EFF:
2872 minsize = sizeof (mac_resource_props_t);
2873 break;
2874 case MAC_PROP_DUPLEX:
2875 minsize = sizeof (link_duplex_t);
2876 break;
2877 case MAC_PROP_SPEED:
2878 minsize = sizeof (uint64_t);
2879 break;
2880 case MAC_PROP_STATUS:
2881 minsize = sizeof (link_state_t);
2882 break;
2883 case MAC_PROP_AUTONEG:
2884 case MAC_PROP_EN_AUTONEG:
2885 minsize = sizeof (uint8_t);
2886 break;
2887 case MAC_PROP_MTU:
2888 case MAC_PROP_LLIMIT:
2889 case MAC_PROP_LDECAY:
2890 minsize = sizeof (uint32_t);
2891 break;
2892 case MAC_PROP_FLOWCTRL:
2893 minsize = sizeof (link_flowctrl_t);
2894 break;
2895 case MAC_PROP_ADV_10GFDX_CAP:
2896 case MAC_PROP_EN_10GFDX_CAP:
2897 case MAC_PROP_ADV_1000HDX_CAP:
2898 case MAC_PROP_EN_1000HDX_CAP:
2899 case MAC_PROP_ADV_100FDX_CAP:
2900 case MAC_PROP_EN_100FDX_CAP:
2901 case MAC_PROP_ADV_100HDX_CAP:
2902 case MAC_PROP_EN_100HDX_CAP:
2903 case MAC_PROP_ADV_10FDX_CAP:
2904 case MAC_PROP_EN_10FDX_CAP:
2905 case MAC_PROP_ADV_10HDX_CAP:
2906 case MAC_PROP_EN_10HDX_CAP:
2907 case MAC_PROP_ADV_100T4_CAP:
2908 case MAC_PROP_EN_100T4_CAP:
2909 minsize = sizeof (uint8_t);
2910 break;
2911 case MAC_PROP_PVID:
2912 minsize = sizeof (uint16_t);
2913 break;
2914 case MAC_PROP_IPTUN_HOPLIMIT:
2915 minsize = sizeof (uint32_t);
2916 break;
2917 case MAC_PROP_IPTUN_ENCAPLIMIT:
2918 minsize = sizeof (uint32_t);
2919 break;
2920 case MAC_PROP_MAX_TX_RINGS_AVAIL:
2921 case MAC_PROP_MAX_RX_RINGS_AVAIL:
2922 case MAC_PROP_MAX_RXHWCLNT_AVAIL:
2923 case MAC_PROP_MAX_TXHWCLNT_AVAIL:
2924 minsize = sizeof (uint_t);
2925 break;
2926 case MAC_PROP_WL_ESSID:
2927 minsize = sizeof (wl_linkstatus_t);
2928 break;
2929 case MAC_PROP_WL_BSSID:
2930 minsize = sizeof (wl_bssid_t);
2931 break;
2932 case MAC_PROP_WL_BSSTYPE:
2933 minsize = sizeof (wl_bss_type_t);
2934 break;
2935 case MAC_PROP_WL_LINKSTATUS:
2936 minsize = sizeof (wl_linkstatus_t);
2937 break;
2938 case MAC_PROP_WL_DESIRED_RATES:
2939 minsize = sizeof (wl_rates_t);
2940 break;
2941 case MAC_PROP_WL_SUPPORTED_RATES:
2942 minsize = sizeof (wl_rates_t);
2943 break;
2944 case MAC_PROP_WL_AUTH_MODE:
2945 minsize = sizeof (wl_authmode_t);
2946 break;
2947 case MAC_PROP_WL_ENCRYPTION:
2948 minsize = sizeof (wl_encryption_t);
2949 break;
2950 case MAC_PROP_WL_RSSI:
2951 minsize = sizeof (wl_rssi_t);
2952 break;
2953 case MAC_PROP_WL_PHY_CONFIG:
2954 minsize = sizeof (wl_phy_conf_t);
2955 break;
2956 case MAC_PROP_WL_CAPABILITY:
2957 minsize = sizeof (wl_capability_t);
2958 break;
2959 case MAC_PROP_WL_WPA:
2960 minsize = sizeof (wl_wpa_t);
2961 break;
2962 case MAC_PROP_WL_SCANRESULTS:
2963 minsize = sizeof (wl_wpa_ess_t);
2964 break;
2965 case MAC_PROP_WL_POWER_MODE:
2966 minsize = sizeof (wl_ps_mode_t);
2967 break;
2968 case MAC_PROP_WL_RADIO:
2969 minsize = sizeof (wl_radio_t);
2970 break;
2971 case MAC_PROP_WL_ESS_LIST:
2972 minsize = sizeof (wl_ess_list_t);
2973 break;
2974 case MAC_PROP_WL_KEY_TAB:
2975 minsize = sizeof (wl_wep_key_tab_t);
2976 break;
2977 case MAC_PROP_WL_CREATE_IBSS:
2978 minsize = sizeof (wl_create_ibss_t);
2979 break;
2980 case MAC_PROP_WL_SETOPTIE:
2981 minsize = sizeof (wl_wpa_ie_t);
2982 break;
2983 case MAC_PROP_WL_DELKEY:
2984 minsize = sizeof (wl_del_key_t);
2985 break;
2986 case MAC_PROP_WL_KEY:
2987 minsize = sizeof (wl_key_t);
2988 break;
2989 case MAC_PROP_WL_MLME:
2990 minsize = sizeof (wl_mlme_t);
2991 break;
2992 }
2993
2994 return (valsize >= minsize);
2995 }
2996
2997 /*
2998 * mac_set_prop() sets MAC or hardware driver properties:
2999 *
3000 * - MAC-managed properties such as resource properties include maxbw,
3001 * priority, and cpu binding list, as well as the default port VID
3002 * used by bridging. These properties are consumed by the MAC layer
3003 * itself and not passed down to the driver. For resource control
3004 * properties, this function invokes mac_set_resources() which will
3005 * cache the property value in mac_impl_t and may call
3006 * mac_client_set_resource() to update property value of the primary
3007 * mac client, if it exists.
3008 *
3009 * - Properties which act on the hardware and must be passed to the
3010 * driver, such as MTU, through the driver's mc_setprop() entry point.
3011 */
3012 int
3013 mac_set_prop(mac_handle_t mh, mac_prop_id_t id, char *name, void *val,
3014 uint_t valsize)
3015 {
3016 int err = ENOTSUP;
3017 mac_impl_t *mip = (mac_impl_t *)mh;
3018
3019 ASSERT(MAC_PERIM_HELD(mh));
3020
3021 switch (id) {
3022 case MAC_PROP_RESOURCE: {
3023 mac_resource_props_t *mrp;
3024
3025 /* call mac_set_resources() for MAC properties */
3026 ASSERT(valsize >= sizeof (mac_resource_props_t));
3027 mrp = kmem_zalloc(sizeof (*mrp), KM_SLEEP);
3028 bcopy(val, mrp, sizeof (*mrp));
3029 err = mac_set_resources(mh, mrp);
3030 kmem_free(mrp, sizeof (*mrp));
3031 break;
3032 }
3033
3034 case MAC_PROP_PVID:
3035 ASSERT(valsize >= sizeof (uint16_t));
3036 if (mip->mi_state_flags & MIS_IS_VNIC)
3037 return (EINVAL);
3038 err = mac_set_pvid(mh, *(uint16_t *)val);
3039 break;
3040
3041 case MAC_PROP_MTU: {
3042 uint32_t mtu;
3043
3044 ASSERT(valsize >= sizeof (uint32_t));
3045 bcopy(val, &mtu, sizeof (mtu));
3046 err = mac_set_mtu(mh, mtu, NULL);
3047 break;
3048 }
3049
3050 case MAC_PROP_LLIMIT:
3051 case MAC_PROP_LDECAY: {
3052 uint32_t learnval;
3053
3054 if (valsize < sizeof (learnval) ||
3055 (mip->mi_state_flags & MIS_IS_VNIC))
3056 return (EINVAL);
3057 bcopy(val, &learnval, sizeof (learnval));
3058 if (learnval == 0 && id == MAC_PROP_LDECAY)
3059 return (EINVAL);
3060 if (id == MAC_PROP_LLIMIT)
3061 mip->mi_llimit = learnval;
3062 else
3063 mip->mi_ldecay = learnval;
3064 err = 0;
3065 break;
3066 }
3067
3068 default:
3069 /* For other driver properties, call driver's callback */
3070 if (mip->mi_callbacks->mc_callbacks & MC_SETPROP) {
3071 err = mip->mi_callbacks->mc_setprop(mip->mi_driver,
3072 name, id, valsize, val);
3073 }
3074 }
3075 return (err);
3076 }
3077
3078 /*
3079 * mac_get_prop() gets MAC or device driver properties.
3080 *
3081 * If the property is a driver property, mac_get_prop() calls driver's callback
3082 * entry point to get it.
3083 * If the property is a MAC property, mac_get_prop() invokes mac_get_resources()
3084 * which returns the cached value in mac_impl_t.
3085 */
3086 int
3087 mac_get_prop(mac_handle_t mh, mac_prop_id_t id, char *name, void *val,
3088 uint_t valsize)
3089 {
3090 int err = ENOTSUP;
3091 mac_impl_t *mip = (mac_impl_t *)mh;
3092 uint_t rings;
3093 uint_t vlinks;
3094
3095 bzero(val, valsize);
3096
3097 switch (id) {
3098 case MAC_PROP_RESOURCE: {
3099 mac_resource_props_t *mrp;
3100
3101 /* If mac property, read from cache */
3102 ASSERT(valsize >= sizeof (mac_resource_props_t));
3103 mrp = kmem_zalloc(sizeof (*mrp), KM_SLEEP);
3104 mac_get_resources(mh, mrp);
3105 bcopy(mrp, val, sizeof (*mrp));
3106 kmem_free(mrp, sizeof (*mrp));
3107 return (0);
3108 }
3109 case MAC_PROP_RESOURCE_EFF: {
3110 mac_resource_props_t *mrp;
3111
3112 /* If mac effective property, read from client */
3113 ASSERT(valsize >= sizeof (mac_resource_props_t));
3114 mrp = kmem_zalloc(sizeof (*mrp), KM_SLEEP);
3115 mac_get_effective_resources(mh, mrp);
3116 bcopy(mrp, val, sizeof (*mrp));
3117 kmem_free(mrp, sizeof (*mrp));
3118 return (0);
3119 }
3120
3121 case MAC_PROP_PVID:
3122 ASSERT(valsize >= sizeof (uint16_t));
3123 if (mip->mi_state_flags & MIS_IS_VNIC)
3124 return (EINVAL);
3125 *(uint16_t *)val = mac_get_pvid(mh);
3126 return (0);
3127
3128 case MAC_PROP_LLIMIT:
3129 case MAC_PROP_LDECAY:
3130 ASSERT(valsize >= sizeof (uint32_t));
3131 if (mip->mi_state_flags & MIS_IS_VNIC)
3132 return (EINVAL);
3133 if (id == MAC_PROP_LLIMIT)
3134 bcopy(&mip->mi_llimit, val, sizeof (mip->mi_llimit));
3135 else
3136 bcopy(&mip->mi_ldecay, val, sizeof (mip->mi_ldecay));
3137 return (0);
3138
3139 case MAC_PROP_MTU: {
3140 uint32_t sdu;
3141
3142 ASSERT(valsize >= sizeof (uint32_t));
3143 mac_sdu_get2(mh, NULL, &sdu, NULL);
3144 bcopy(&sdu, val, sizeof (sdu));
3145
3146 return (0);
3147 }
3148 case MAC_PROP_STATUS: {
3149 link_state_t link_state;
3150
3151 if (valsize < sizeof (link_state))
3152 return (EINVAL);
3153 link_state = mac_link_get(mh);
3154 bcopy(&link_state, val, sizeof (link_state));
3155
3156 return (0);
3157 }
3158
3159 case MAC_PROP_MAX_RX_RINGS_AVAIL:
3160 case MAC_PROP_MAX_TX_RINGS_AVAIL:
3161 ASSERT(valsize >= sizeof (uint_t));
3162 rings = id == MAC_PROP_MAX_RX_RINGS_AVAIL ?
3163 mac_rxavail_get(mh) : mac_txavail_get(mh);
3164 bcopy(&rings, val, sizeof (uint_t));
3165 return (0);
3166
3167 case MAC_PROP_MAX_RXHWCLNT_AVAIL:
3168 case MAC_PROP_MAX_TXHWCLNT_AVAIL:
3169 ASSERT(valsize >= sizeof (uint_t));
3170 vlinks = id == MAC_PROP_MAX_RXHWCLNT_AVAIL ?
3171 mac_rxhwlnksavail_get(mh) : mac_txhwlnksavail_get(mh);
3172 bcopy(&vlinks, val, sizeof (uint_t));
3173 return (0);
3174
3175 case MAC_PROP_RXRINGSRANGE:
3176 case MAC_PROP_TXRINGSRANGE:
3177 /*
3178 * The value for these properties are returned through
3179 * the MAC_PROP_RESOURCE property.
3180 */
3181 return (0);
3182
3183 default:
3184 break;
3185
3186 }
3187
3188 /* If driver property, request from driver */
3189 if (mip->mi_callbacks->mc_callbacks & MC_GETPROP) {
3190 err = mip->mi_callbacks->mc_getprop(mip->mi_driver, name, id,
3191 valsize, val);
3192 }
3193
3194 return (err);
3195 }
3196
3197 /*
3198 * Helper function to initialize the range structure for use in
3199 * mac_get_prop. If the type can be other than uint32, we can
3200 * pass that as an arg.
3201 */
3202 static void
3203 _mac_set_range(mac_propval_range_t *range, uint32_t min, uint32_t max)
3204 {
3205 range->mpr_count = 1;
3206 range->mpr_type = MAC_PROPVAL_UINT32;
3207 range->mpr_range_uint32[0].mpur_min = min;
3208 range->mpr_range_uint32[0].mpur_max = max;
3209 }
3210
3211 /*
3212 * Returns information about the specified property, such as default
3213 * values or permissions.
3214 */
3215 int
3216 mac_prop_info(mac_handle_t mh, mac_prop_id_t id, char *name,
3217 void *default_val, uint_t default_size, mac_propval_range_t *range,
3218 uint_t *perm)
3219 {
3220 mac_prop_info_state_t state;
3221 mac_impl_t *mip = (mac_impl_t *)mh;
3222 uint_t max;
3223
3224 /*
3225 * A property is read/write by default unless the driver says
3226 * otherwise.
3227 */
3228 if (perm != NULL)
3229 *perm = MAC_PROP_PERM_RW;
3230
3231 if (default_val != NULL)
3232 bzero(default_val, default_size);
3233
3234 /*
3235 * First, handle framework properties for which we don't need to
3236 * involve the driver.
3237 */
3238 switch (id) {
3239 case MAC_PROP_RESOURCE:
3240 case MAC_PROP_PVID:
3241 case MAC_PROP_LLIMIT:
3242 case MAC_PROP_LDECAY:
3243 return (0);
3244
3245 case MAC_PROP_MAX_RX_RINGS_AVAIL:
3246 case MAC_PROP_MAX_TX_RINGS_AVAIL:
3247 case MAC_PROP_MAX_RXHWCLNT_AVAIL:
3248 case MAC_PROP_MAX_TXHWCLNT_AVAIL:
3249 if (perm != NULL)
3250 *perm = MAC_PROP_PERM_READ;
3251 return (0);
3252
3253 case MAC_PROP_RXRINGSRANGE:
3254 case MAC_PROP_TXRINGSRANGE:
3255 /*
3256 * Currently, we support range for RX and TX rings properties.
3257 * When we extend this support to maxbw, cpus and priority,
3258 * we should move this to mac_get_resources.
3259 * There is no default value for RX or TX rings.
3260 */
3261 if ((mip->mi_state_flags & MIS_IS_VNIC) &&
3262 mac_is_vnic_primary(mh)) {
3263 /*
3264 * We don't support setting rings for a VLAN
3265 * data link because it shares its ring with the
3266 * primary MAC client.
3267 */
3268 if (perm != NULL)
3269 *perm = MAC_PROP_PERM_READ;
3270 if (range != NULL)
3271 range->mpr_count = 0;
3272 } else if (range != NULL) {
3273 if (mip->mi_state_flags & MIS_IS_VNIC)
3274 mh = mac_get_lower_mac_handle(mh);
3275 mip = (mac_impl_t *)mh;
3276 if ((id == MAC_PROP_RXRINGSRANGE &&
3277 mip->mi_rx_group_type == MAC_GROUP_TYPE_STATIC) ||
3278 (id == MAC_PROP_TXRINGSRANGE &&
3279 mip->mi_tx_group_type == MAC_GROUP_TYPE_STATIC)) {
3280 if (id == MAC_PROP_RXRINGSRANGE) {
3281 if ((mac_rxhwlnksavail_get(mh) +
3282 mac_rxhwlnksrsvd_get(mh)) <= 1) {
3283 /*
3284 * doesn't support groups or
3285 * rings
3286 */
3287 range->mpr_count = 0;
3288 } else {
3289 /*
3290 * supports specifying groups,
3291 * but not rings
3292 */
3293 _mac_set_range(range, 0, 0);
3294 }
3295 } else {
3296 if ((mac_txhwlnksavail_get(mh) +
3297 mac_txhwlnksrsvd_get(mh)) <= 1) {
3298 /*
3299 * doesn't support groups or
3300 * rings
3301 */
3302 range->mpr_count = 0;
3303 } else {
3304 /*
3305 * supports specifying groups,
3306 * but not rings
3307 */
3308 _mac_set_range(range, 0, 0);
3309 }
3310 }
3311 } else {
3312 max = id == MAC_PROP_RXRINGSRANGE ?
3313 mac_rxavail_get(mh) + mac_rxrsvd_get(mh) :
3314 mac_txavail_get(mh) + mac_txrsvd_get(mh);
3315 if (max <= 1) {
3316 /*
3317 * doesn't support groups or
3318 * rings
3319 */
3320 range->mpr_count = 0;
3321 } else {
3322 /*
3323 * -1 because we have to leave out the
3324 * default ring.
3325 */
3326 _mac_set_range(range, 1, max - 1);
3327 }
3328 }
3329 }
3330 return (0);
3331
3332 case MAC_PROP_STATUS:
3333 if (perm != NULL)
3334 *perm = MAC_PROP_PERM_READ;
3335 return (0);
3336 }
3337
3338 /*
3339 * Get the property info from the driver if it implements the
3340 * property info entry point.
3341 */
3342 bzero(&state, sizeof (state));
3343
3344 if (mip->mi_callbacks->mc_callbacks & MC_PROPINFO) {
3345 state.pr_default = default_val;
3346 state.pr_default_size = default_size;
3347
3348 /*
3349 * The caller specifies the maximum number of ranges
3350 * it can accomodate using mpr_count. We don't touch
3351 * this value until the driver returns from its
3352 * mc_propinfo() callback, and ensure we don't exceed
3353 * this number of range as the driver defines
3354 * supported range from its mc_propinfo().
3355 *
3356 * pr_range_cur_count keeps track of how many ranges
3357 * were defined by the driver from its mc_propinfo()
3358 * entry point.
3359 *
3360 * On exit, the user-specified range mpr_count returns
3361 * the number of ranges specified by the driver on
3362 * success, or the number of ranges it wanted to
3363 * define if that number of ranges could not be
3364 * accomodated by the specified range structure. In
3365 * the latter case, the caller will be able to
3366 * allocate a larger range structure, and query the
3367 * property again.
3368 */
3369 state.pr_range_cur_count = 0;
3370 state.pr_range = range;
3371
3372 mip->mi_callbacks->mc_propinfo(mip->mi_driver, name, id,
3373 (mac_prop_info_handle_t)&state);
3374
3375 if (state.pr_flags & MAC_PROP_INFO_RANGE)
3376 range->mpr_count = state.pr_range_cur_count;
3377
3378 /*
3379 * The operation could fail if the buffer supplied by
3380 * the user was too small for the range or default
3381 * value of the property.
3382 */
3383 if (state.pr_errno != 0)
3384 return (state.pr_errno);
3385
3386 if (perm != NULL && state.pr_flags & MAC_PROP_INFO_PERM)
3387 *perm = state.pr_perm;
3388 }
3389
3390 /*
3391 * The MAC layer may want to provide default values or allowed
3392 * ranges for properties if the driver does not provide a
3393 * property info entry point, or that entry point exists, but
3394 * it did not provide a default value or allowed ranges for
3395 * that property.
3396 */
3397 switch (id) {
3398 case MAC_PROP_MTU: {
3399 uint32_t sdu;
3400
3401 mac_sdu_get2(mh, NULL, &sdu, NULL);
3402
3403 if (range != NULL && !(state.pr_flags &
3404 MAC_PROP_INFO_RANGE)) {
3405 /* MTU range */
3406 _mac_set_range(range, sdu, sdu);
3407 }
3408
3409 if (default_val != NULL && !(state.pr_flags &
3410 MAC_PROP_INFO_DEFAULT)) {
3411 if (mip->mi_info.mi_media == DL_ETHER)
3412 sdu = ETHERMTU;
3413 /* default MTU value */
3414 bcopy(&sdu, default_val, sizeof (sdu));
3415 }
3416 }
3417 }
3418
3419 return (0);
3420 }
3421
3422 int
3423 mac_fastpath_disable(mac_handle_t mh)
3424 {
3425 mac_impl_t *mip = (mac_impl_t *)mh;
3426
3427 if ((mip->mi_state_flags & MIS_LEGACY) == 0)
3428 return (0);
3429
3430 return (mip->mi_capab_legacy.ml_fastpath_disable(mip->mi_driver));
3431 }
3432
3433 void
3434 mac_fastpath_enable(mac_handle_t mh)
3435 {
3436 mac_impl_t *mip = (mac_impl_t *)mh;
3437
3438 if ((mip->mi_state_flags & MIS_LEGACY) == 0)
3439 return;
3440
3441 mip->mi_capab_legacy.ml_fastpath_enable(mip->mi_driver);
3442 }
3443
3444 void
3445 mac_register_priv_prop(mac_impl_t *mip, char **priv_props)
3446 {
3447 uint_t nprops, i;
3448
3449 if (priv_props == NULL)
3450 return;
3451
3452 nprops = 0;
3453 while (priv_props[nprops] != NULL)
3454 nprops++;
3455 if (nprops == 0)
3456 return;
3457
3458
3459 mip->mi_priv_prop = kmem_zalloc(nprops * sizeof (char *), KM_SLEEP);
3460
3461 for (i = 0; i < nprops; i++) {
3462 mip->mi_priv_prop[i] = kmem_zalloc(MAXLINKPROPNAME, KM_SLEEP);
3463 (void) strlcpy(mip->mi_priv_prop[i], priv_props[i],
3464 MAXLINKPROPNAME);
3465 }
3466
3467 mip->mi_priv_prop_count = nprops;
3468 }
3469
3470 void
3471 mac_unregister_priv_prop(mac_impl_t *mip)
3472 {
3473 uint_t i;
3474
3475 if (mip->mi_priv_prop_count == 0) {
3476 ASSERT(mip->mi_priv_prop == NULL);
3477 return;
3478 }
3479
3480 for (i = 0; i < mip->mi_priv_prop_count; i++)
3481 kmem_free(mip->mi_priv_prop[i], MAXLINKPROPNAME);
3482 kmem_free(mip->mi_priv_prop, mip->mi_priv_prop_count *
3483 sizeof (char *));
3484
3485 mip->mi_priv_prop = NULL;
3486 mip->mi_priv_prop_count = 0;
3487 }
3488
3489 /*
3490 * mac_ring_t 'mr' macros. Some rogue drivers may access ring structure
3491 * (by invoking mac_rx()) even after processing mac_stop_ring(). In such
3492 * cases if MAC free's the ring structure after mac_stop_ring(), any
3493 * illegal access to the ring structure coming from the driver will panic
3494 * the system. In order to protect the system from such inadverent access,
3495 * we maintain a cache of rings in the mac_impl_t after they get free'd up.
3496 * When packets are received on free'd up rings, MAC (through the generation
3497 * count mechanism) will drop such packets.
3498 */
3499 static mac_ring_t *
3500 mac_ring_alloc(mac_impl_t *mip)
3501 {
3502 mac_ring_t *ring;
3503
3504 mutex_enter(&mip->mi_ring_lock);
3505 if (mip->mi_ring_freelist != NULL) {
3506 ring = mip->mi_ring_freelist;
3507 mip->mi_ring_freelist = ring->mr_next;
3508 bzero(ring, sizeof (mac_ring_t));
3509 mutex_exit(&mip->mi_ring_lock);
3510 } else {
3511 mutex_exit(&mip->mi_ring_lock);
3512 ring = kmem_cache_alloc(mac_ring_cache, KM_SLEEP);
3513 }
3514 ASSERT((ring != NULL) && (ring->mr_state == MR_FREE));
3515 return (ring);
3516 }
3517
3518 static void
3519 mac_ring_free(mac_impl_t *mip, mac_ring_t *ring)
3520 {
3521 ASSERT(ring->mr_state == MR_FREE);
3522
3523 mutex_enter(&mip->mi_ring_lock);
3524 ring->mr_state = MR_FREE;
3525 ring->mr_flag = 0;
3526 ring->mr_next = mip->mi_ring_freelist;
3527 ring->mr_mip = NULL;
3528 mip->mi_ring_freelist = ring;
3529 mac_ring_stat_delete(ring);
3530 mutex_exit(&mip->mi_ring_lock);
3531 }
3532
3533 static void
3534 mac_ring_freeall(mac_impl_t *mip)
3535 {
3536 mac_ring_t *ring_next;
3537 mutex_enter(&mip->mi_ring_lock);
3538 mac_ring_t *ring = mip->mi_ring_freelist;
3539 while (ring != NULL) {
3540 ring_next = ring->mr_next;
3541 kmem_cache_free(mac_ring_cache, ring);
3542 ring = ring_next;
3543 }
3544 mip->mi_ring_freelist = NULL;
3545 mutex_exit(&mip->mi_ring_lock);
3546 }
3547
3548 int
3549 mac_start_ring(mac_ring_t *ring)
3550 {
3551 int rv = 0;
3552
3553 ASSERT(ring->mr_state == MR_FREE);
3554
3555 if (ring->mr_start != NULL) {
3556 rv = ring->mr_start(ring->mr_driver, ring->mr_gen_num);
3557 if (rv != 0)
3558 return (rv);
3559 }
3560
3561 ring->mr_state = MR_INUSE;
3562 return (rv);
3563 }
3564
3565 void
3566 mac_stop_ring(mac_ring_t *ring)
3567 {
3568 ASSERT(ring->mr_state == MR_INUSE);
3569
3570 if (ring->mr_stop != NULL)
3571 ring->mr_stop(ring->mr_driver);
3572
3573 ring->mr_state = MR_FREE;
3574
3575 /*
3576 * Increment the ring generation number for this ring.
3577 */
3578 ring->mr_gen_num++;
3579 }
3580
3581 int
3582 mac_start_group(mac_group_t *group)
3583 {
3584 int rv = 0;
3585
3586 if (group->mrg_start != NULL)
3587 rv = group->mrg_start(group->mrg_driver);
3588
3589 return (rv);
3590 }
3591
3592 void
3593 mac_stop_group(mac_group_t *group)
3594 {
3595 if (group->mrg_stop != NULL)
3596 group->mrg_stop(group->mrg_driver);
3597 }
3598
3599 /*
3600 * Called from mac_start() on the default Rx group. Broadcast and multicast
3601 * packets are received only on the default group. Hence the default group
3602 * needs to be up even if the primary client is not up, for the other groups
3603 * to be functional. We do this by calling this function at mac_start time
3604 * itself. However the broadcast packets that are received can't make their
3605 * way beyond mac_rx until a mac client creates a broadcast flow.
3606 */
3607 static int
3608 mac_start_group_and_rings(mac_group_t *group)
3609 {
3610 mac_ring_t *ring;
3611 int rv = 0;
3612
3613 ASSERT(group->mrg_state == MAC_GROUP_STATE_REGISTERED);
3614 if ((rv = mac_start_group(group)) != 0)
3615 return (rv);
3616
3617 for (ring = group->mrg_rings; ring != NULL; ring = ring->mr_next) {
3618 ASSERT(ring->mr_state == MR_FREE);
3619 if ((rv = mac_start_ring(ring)) != 0)
3620 goto error;
3621 ring->mr_classify_type = MAC_SW_CLASSIFIER;
3622 }
3623 return (0);
3624
3625 error:
3626 mac_stop_group_and_rings(group);
3627 return (rv);
3628 }
3629
3630 /* Called from mac_stop on the default Rx group */
3631 static void
3632 mac_stop_group_and_rings(mac_group_t *group)
3633 {
3634 mac_ring_t *ring;
3635
3636 for (ring = group->mrg_rings; ring != NULL; ring = ring->mr_next) {
3637 if (ring->mr_state != MR_FREE) {
3638 mac_stop_ring(ring);
3639 ring->mr_flag = 0;
3640 ring->mr_classify_type = MAC_NO_CLASSIFIER;
3641 }
3642 }
3643 mac_stop_group(group);
3644 }
3645
3646
3647 static mac_ring_t *
3648 mac_init_ring(mac_impl_t *mip, mac_group_t *group, int index,
3649 mac_capab_rings_t *cap_rings)
3650 {
3651 mac_ring_t *ring, *rnext;
3652 mac_ring_info_t ring_info;
3653 ddi_intr_handle_t ddi_handle;
3654
3655 ring = mac_ring_alloc(mip);
3656
3657 /* Prepare basic information of ring */
3658
3659 /*
3660 * Ring index is numbered to be unique across a particular device.
3661 * Ring index computation makes following assumptions:
3662 * - For drivers with static grouping (e.g. ixgbe, bge),
3663 * ring index exchanged with the driver (e.g. during mr_rget)
3664 * is unique only across the group the ring belongs to.
3665 * - Drivers with dynamic grouping (e.g. nxge), start
3666 * with single group (mrg_index = 0).
3667 */
3668 ring->mr_index = group->mrg_index * group->mrg_info.mgi_count + index;
3669 ring->mr_type = group->mrg_type;
3670 ring->mr_gh = (mac_group_handle_t)group;
3671
3672 /* Insert the new ring to the list. */
3673 ring->mr_next = group->mrg_rings;
3674 group->mrg_rings = ring;
3675
3676 /* Zero to reuse the info data structure */
3677 bzero(&ring_info, sizeof (ring_info));
3678
3679 /* Query ring information from driver */
3680 cap_rings->mr_rget(mip->mi_driver, group->mrg_type, group->mrg_index,
3681 index, &ring_info, (mac_ring_handle_t)ring);
3682
3683 ring->mr_info = ring_info;
3684
3685 /*
3686 * The interrupt handle could be shared among multiple rings.
3687 * Thus if there is a bunch of rings that are sharing an
3688 * interrupt, then only one ring among the bunch will be made
3689 * available for interrupt re-targeting; the rest will have
3690 * ddi_shared flag set to TRUE and would not be available for
3691 * be interrupt re-targeting.
3692 */
3693 if ((ddi_handle = ring_info.mri_intr.mi_ddi_handle) != NULL) {
3694 rnext = ring->mr_next;
3695 while (rnext != NULL) {
3696 if (rnext->mr_info.mri_intr.mi_ddi_handle ==
3697 ddi_handle) {
3698 /*
3699 * If default ring (mr_index == 0) is part
3700 * of a group of rings sharing an
3701 * interrupt, then set ddi_shared flag for
3702 * the default ring and give another ring
3703 * the chance to be re-targeted.
3704 */
3705 if (rnext->mr_index == 0 &&
3706 !rnext->mr_info.mri_intr.mi_ddi_shared) {
3707 rnext->mr_info.mri_intr.mi_ddi_shared =
3708 B_TRUE;
3709 } else {
3710 ring->mr_info.mri_intr.mi_ddi_shared =
3711 B_TRUE;
3712 }
3713 break;
3714 }
3715 rnext = rnext->mr_next;
3716 }
3717 /*
3718 * If rnext is NULL, then no matching ddi_handle was found.
3719 * Rx rings get registered first. So if this is a Tx ring,
3720 * then go through all the Rx rings and see if there is a
3721 * matching ddi handle.
3722 */
3723 if (rnext == NULL && ring->mr_type == MAC_RING_TYPE_TX) {
3724 mac_compare_ddi_handle(mip->mi_rx_groups,
3725 mip->mi_rx_group_count, ring);
3726 }
3727 }
3728
3729 /* Update ring's status */
3730 ring->mr_state = MR_FREE;
3731 ring->mr_flag = 0;
3732
3733 /* Update the ring count of the group */
3734 group->mrg_cur_count++;
3735
3736 /* Create per ring kstats */
3737 if (ring->mr_stat != NULL) {
3738 ring->mr_mip = mip;
3739 mac_ring_stat_create(ring);
3740 }
3741
3742 return (ring);
3743 }
3744
3745 /*
3746 * Rings are chained together for easy regrouping.
3747 */
3748 static void
3749 mac_init_group(mac_impl_t *mip, mac_group_t *group, int size,
3750 mac_capab_rings_t *cap_rings)
3751 {
3752 int index;
3753
3754 /*
3755 * Initialize all ring members of this group. Size of zero will not
3756 * enter the loop, so it's safe for initializing an empty group.
3757 */
3758 for (index = size - 1; index >= 0; index--)
3759 (void) mac_init_ring(mip, group, index, cap_rings);
3760 }
3761
3762 int
3763 mac_init_rings(mac_impl_t *mip, mac_ring_type_t rtype)
3764 {
3765 mac_capab_rings_t *cap_rings;
3766 mac_group_t *group;
3767 mac_group_t *groups;
3768 mac_group_info_t group_info;
3769 uint_t group_free = 0;
3770 uint_t ring_left;
3771 mac_ring_t *ring;
3772 int g;
3773 int err = 0;
3774 uint_t grpcnt;
3775 boolean_t pseudo_txgrp = B_FALSE;
3776
3777 switch (rtype) {
3778 case MAC_RING_TYPE_RX:
3779 ASSERT(mip->mi_rx_groups == NULL);
3780
3781 cap_rings = &mip->mi_rx_rings_cap;
3782 cap_rings->mr_type = MAC_RING_TYPE_RX;
3783 break;
3784 case MAC_RING_TYPE_TX:
3785 ASSERT(mip->mi_tx_groups == NULL);
3786
3787 cap_rings = &mip->mi_tx_rings_cap;
3788 cap_rings->mr_type = MAC_RING_TYPE_TX;
3789 break;
3790 default:
3791 ASSERT(B_FALSE);
3792 }
3793
3794 if (!i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_RINGS, cap_rings))
3795 return (0);
3796 grpcnt = cap_rings->mr_gnum;
3797
3798 /*
3799 * If we have multiple TX rings, but only one TX group, we can
3800 * create pseudo TX groups (one per TX ring) in the MAC layer,
3801 * except for an aggr. For an aggr currently we maintain only
3802 * one group with all the rings (for all its ports), going
3803 * forwards we might change this.
3804 */
3805 if (rtype == MAC_RING_TYPE_TX &&
3806 cap_rings->mr_gnum == 0 && cap_rings->mr_rnum > 0 &&
3807 (mip->mi_state_flags & MIS_IS_AGGR) == 0) {
3808 /*
3809 * The -1 here is because we create a default TX group
3810 * with all the rings in it.
3811 */
3812 grpcnt = cap_rings->mr_rnum - 1;
3813 pseudo_txgrp = B_TRUE;
3814 }
3815
3816 /*
3817 * Allocate a contiguous buffer for all groups.
3818 */
3819 groups = kmem_zalloc(sizeof (mac_group_t) * (grpcnt+ 1), KM_SLEEP);
3820
3821 ring_left = cap_rings->mr_rnum;
3822
3823 /*
3824 * Get all ring groups if any, and get their ring members
3825 * if any.
3826 */
3827 for (g = 0; g < grpcnt; g++) {
3828 group = groups + g;
3829
3830 /* Prepare basic information of the group */
3831 group->mrg_index = g;
3832 group->mrg_type = rtype;
3833 group->mrg_state = MAC_GROUP_STATE_UNINIT;
3834 group->mrg_mh = (mac_handle_t)mip;
3835 group->mrg_next = group + 1;
3836
3837 /* Zero to reuse the info data structure */
3838 bzero(&group_info, sizeof (group_info));
3839
3840 if (pseudo_txgrp) {
3841 /*
3842 * This is a pseudo group that we created, apart
3843 * from setting the state there is nothing to be
3844 * done.
3845 */
3846 group->mrg_state = MAC_GROUP_STATE_REGISTERED;
3847 group_free++;
3848 continue;
3849 }
3850 /* Query group information from driver */
3851 cap_rings->mr_gget(mip->mi_driver, rtype, g, &group_info,
3852 (mac_group_handle_t)group);
3853
3854 switch (cap_rings->mr_group_type) {
3855 case MAC_GROUP_TYPE_DYNAMIC:
3856 if (cap_rings->mr_gaddring == NULL ||
3857 cap_rings->mr_gremring == NULL) {
3858 DTRACE_PROBE3(
3859 mac__init__rings_no_addremring,
3860 char *, mip->mi_name,
3861 mac_group_add_ring_t,
3862 cap_rings->mr_gaddring,
3863 mac_group_add_ring_t,
3864 cap_rings->mr_gremring);
3865 err = EINVAL;
3866 goto bail;
3867 }
3868
3869 switch (rtype) {
3870 case MAC_RING_TYPE_RX:
3871 /*
3872 * The first RX group must have non-zero
3873 * rings, and the following groups must
3874 * have zero rings.
3875 */
3876 if (g == 0 && group_info.mgi_count == 0) {
3877 DTRACE_PROBE1(
3878 mac__init__rings__rx__def__zero,
3879 char *, mip->mi_name);
3880 err = EINVAL;
3881 goto bail;
3882 }
3883 if (g > 0 && group_info.mgi_count != 0) {
3884 DTRACE_PROBE3(
3885 mac__init__rings__rx__nonzero,
3886 char *, mip->mi_name,
3887 int, g, int, group_info.mgi_count);
3888 err = EINVAL;
3889 goto bail;
3890 }
3891 break;
3892 case MAC_RING_TYPE_TX:
3893 /*
3894 * All TX ring groups must have zero rings.
3895 */
3896 if (group_info.mgi_count != 0) {
3897 DTRACE_PROBE3(
3898 mac__init__rings__tx__nonzero,
3899 char *, mip->mi_name,
3900 int, g, int, group_info.mgi_count);
3901 err = EINVAL;
3902 goto bail;
3903 }
3904 break;
3905 }
3906 break;
3907 case MAC_GROUP_TYPE_STATIC:
3908 /*
3909 * Note that an empty group is allowed, e.g., an aggr
3910 * would start with an empty group.
3911 */
3912 break;
3913 default:
3914 /* unknown group type */
3915 DTRACE_PROBE2(mac__init__rings__unknown__type,
3916 char *, mip->mi_name,
3917 int, cap_rings->mr_group_type);
3918 err = EINVAL;
3919 goto bail;
3920 }
3921
3922
3923 /*
3924 * Driver must register group->mgi_addmac/remmac() for rx groups
3925 * to support multiple MAC addresses.
3926 */
3927 if (rtype == MAC_RING_TYPE_RX) {
3928 if ((group_info.mgi_addmac == NULL) ||
3929 (group_info.mgi_addmac == NULL)) {
3930 goto bail;
3931 }
3932 }
3933
3934 /* Cache driver-supplied information */
3935 group->mrg_info = group_info;
3936
3937 /* Update the group's status and group count. */
3938 mac_set_group_state(group, MAC_GROUP_STATE_REGISTERED);
3939 group_free++;
3940
3941 group->mrg_rings = NULL;
3942 group->mrg_cur_count = 0;
3943 mac_init_group(mip, group, group_info.mgi_count, cap_rings);
3944 ring_left -= group_info.mgi_count;
3945
3946 /* The current group size should be equal to default value */
3947 ASSERT(group->mrg_cur_count == group_info.mgi_count);
3948 }
3949
3950 /* Build up a dummy group for free resources as a pool */
3951 group = groups + grpcnt;
3952
3953 /* Prepare basic information of the group */
3954 group->mrg_index = -1;
3955 group->mrg_type = rtype;
3956 group->mrg_state = MAC_GROUP_STATE_UNINIT;
3957 group->mrg_mh = (mac_handle_t)mip;
3958 group->mrg_next = NULL;
3959
3960 /*
3961 * If there are ungrouped rings, allocate a continuous buffer for
3962 * remaining resources.
3963 */
3964 if (ring_left != 0) {
3965 group->mrg_rings = NULL;
3966 group->mrg_cur_count = 0;
3967 mac_init_group(mip, group, ring_left, cap_rings);
3968
3969 /* The current group size should be equal to ring_left */
3970 ASSERT(group->mrg_cur_count == ring_left);
3971
3972 ring_left = 0;
3973
3974 /* Update this group's status */
3975 mac_set_group_state(group, MAC_GROUP_STATE_REGISTERED);
3976 } else
3977 group->mrg_rings = NULL;
3978
3979 ASSERT(ring_left == 0);
3980
3981 bail:
3982
3983 /* Cache other important information to finalize the initialization */
3984 switch (rtype) {
3985 case MAC_RING_TYPE_RX:
3986 mip->mi_rx_group_type = cap_rings->mr_group_type;
3987 mip->mi_rx_group_count = cap_rings->mr_gnum;
3988 mip->mi_rx_groups = groups;
3989 mip->mi_rx_donor_grp = groups;
3990 if (mip->mi_rx_group_type == MAC_GROUP_TYPE_DYNAMIC) {
3991 /*
3992 * The default ring is reserved since it is
3993 * used for sending the broadcast etc. packets.
3994 */
3995 mip->mi_rxrings_avail =
3996 mip->mi_rx_groups->mrg_cur_count - 1;
3997 mip->mi_rxrings_rsvd = 1;
3998 }
3999 /*
4000 * The default group cannot be reserved. It is used by
4001 * all the clients that do not have an exclusive group.
4002 */
4003 mip->mi_rxhwclnt_avail = mip->mi_rx_group_count - 1;
4004 mip->mi_rxhwclnt_used = 1;
4005 break;
4006 case MAC_RING_TYPE_TX:
4007 mip->mi_tx_group_type = pseudo_txgrp ? MAC_GROUP_TYPE_DYNAMIC :
4008 cap_rings->mr_group_type;
4009 mip->mi_tx_group_count = grpcnt;
4010 mip->mi_tx_group_free = group_free;
4011 mip->mi_tx_groups = groups;
4012
4013 group = groups + grpcnt;
4014 ring = group->mrg_rings;
4015 /*
4016 * The ring can be NULL in the case of aggr. Aggr will
4017 * have an empty Tx group which will get populated
4018 * later when pseudo Tx rings are added after
4019 * mac_register() is done.
4020 */
4021 if (ring == NULL) {
4022 ASSERT(mip->mi_state_flags & MIS_IS_AGGR);
4023 /*
4024 * pass the group to aggr so it can add Tx
4025 * rings to the group later.
4026 */
4027 cap_rings->mr_gget(mip->mi_driver, rtype, 0, NULL,
4028 (mac_group_handle_t)group);
4029 /*
4030 * Even though there are no rings at this time
4031 * (rings will come later), set the group
4032 * state to registered.
4033 */
4034 group->mrg_state = MAC_GROUP_STATE_REGISTERED;
4035 } else {
4036 /*
4037 * Ring 0 is used as the default one and it could be
4038 * assigned to a client as well.
4039 */
4040 while ((ring->mr_index != 0) && (ring->mr_next != NULL))
4041 ring = ring->mr_next;
4042 ASSERT(ring->mr_index == 0);
4043 mip->mi_default_tx_ring = (mac_ring_handle_t)ring;
4044 }
4045 if (mip->mi_tx_group_type == MAC_GROUP_TYPE_DYNAMIC)
4046 mip->mi_txrings_avail = group->mrg_cur_count - 1;
4047 /*
4048 * The default ring cannot be reserved.
4049 */
4050 mip->mi_txrings_rsvd = 1;
4051 /*
4052 * The default group cannot be reserved. It will be shared
4053 * by clients that do not have an exclusive group.
4054 */
4055 mip->mi_txhwclnt_avail = mip->mi_tx_group_count;
4056 mip->mi_txhwclnt_used = 1;
4057 break;
4058 default:
4059 ASSERT(B_FALSE);
4060 }
4061
4062 if (err != 0)
4063 mac_free_rings(mip, rtype);
4064
4065 return (err);
4066 }
4067
4068 /*
4069 * The ddi interrupt handle could be shared amoung rings. If so, compare
4070 * the new ring's ddi handle with the existing ones and set ddi_shared
4071 * flag.
4072 */
4073 void
4074 mac_compare_ddi_handle(mac_group_t *groups, uint_t grpcnt, mac_ring_t *cring)
4075 {
4076 mac_group_t *group;
4077 mac_ring_t *ring;
4078 ddi_intr_handle_t ddi_handle;
4079 int g;
4080
4081 ddi_handle = cring->mr_info.mri_intr.mi_ddi_handle;
4082 for (g = 0; g < grpcnt; g++) {
4083 group = groups + g;
4084 for (ring = group->mrg_rings; ring != NULL;
4085 ring = ring->mr_next) {
4086 if (ring == cring)
4087 continue;
4088 if (ring->mr_info.mri_intr.mi_ddi_handle ==
4089 ddi_handle) {
4090 if (cring->mr_type == MAC_RING_TYPE_RX &&
4091 ring->mr_index == 0 &&
4092 !ring->mr_info.mri_intr.mi_ddi_shared) {
4093 ring->mr_info.mri_intr.mi_ddi_shared =
4094 B_TRUE;
4095 } else {
4096 cring->mr_info.mri_intr.mi_ddi_shared =
4097 B_TRUE;
4098 }
4099 return;
4100 }
4101 }
4102 }
4103 }
4104
4105 /*
4106 * Called to free all groups of particular type (RX or TX). It's assumed that
4107 * no clients are using these groups.
4108 */
4109 void
4110 mac_free_rings(mac_impl_t *mip, mac_ring_type_t rtype)
4111 {
4112 mac_group_t *group, *groups;
4113 uint_t group_count;
4114
4115 switch (rtype) {
4116 case MAC_RING_TYPE_RX:
4117 if (mip->mi_rx_groups == NULL)
4118 return;
4119
4120 groups = mip->mi_rx_groups;
4121 group_count = mip->mi_rx_group_count;
4122
4123 mip->mi_rx_groups = NULL;
4124 mip->mi_rx_donor_grp = NULL;
4125 mip->mi_rx_group_count = 0;
4126 break;
4127 case MAC_RING_TYPE_TX:
4128 ASSERT(mip->mi_tx_group_count == mip->mi_tx_group_free);
4129
4130 if (mip->mi_tx_groups == NULL)
4131 return;
4132
4133 groups = mip->mi_tx_groups;
4134 group_count = mip->mi_tx_group_count;
4135
4136 mip->mi_tx_groups = NULL;
4137 mip->mi_tx_group_count = 0;
4138 mip->mi_tx_group_free = 0;
4139 mip->mi_default_tx_ring = NULL;
4140 break;
4141 default:
4142 ASSERT(B_FALSE);
4143 }
4144
4145 for (group = groups; group != NULL; group = group->mrg_next) {
4146 mac_ring_t *ring;
4147
4148 if (group->mrg_cur_count == 0)
4149 continue;
4150
4151 ASSERT(group->mrg_rings != NULL);
4152
4153 while ((ring = group->mrg_rings) != NULL) {
4154 group->mrg_rings = ring->mr_next;
4155 mac_ring_free(mip, ring);
4156 }
4157 }
4158
4159 /* Free all the cached rings */
4160 mac_ring_freeall(mip);
4161 /* Free the block of group data strutures */
4162 kmem_free(groups, sizeof (mac_group_t) * (group_count + 1));
4163 }
4164
4165 /*
4166 * Associate a MAC address with a receive group.
4167 *
4168 * The return value of this function should always be checked properly, because
4169 * any type of failure could cause unexpected results. A group can be added
4170 * or removed with a MAC address only after it has been reserved. Ideally,
4171 * a successful reservation always leads to calling mac_group_addmac() to
4172 * steer desired traffic. Failure of adding an unicast MAC address doesn't
4173 * always imply that the group is functioning abnormally.
4174 *
4175 * Currently this function is called everywhere, and it reflects assumptions
4176 * about MAC addresses in the implementation. CR 6735196.
4177 */
4178 int
4179 mac_group_addmac(mac_group_t *group, const uint8_t *addr)
4180 {
4181 ASSERT(group->mrg_type == MAC_RING_TYPE_RX);
4182 ASSERT(group->mrg_info.mgi_addmac != NULL);
4183
4184 return (group->mrg_info.mgi_addmac(group->mrg_info.mgi_driver, addr));
4185 }
4186
4187 /*
4188 * Remove the association between MAC address and receive group.
4189 */
4190 int
4191 mac_group_remmac(mac_group_t *group, const uint8_t *addr)
4192 {
4193 ASSERT(group->mrg_type == MAC_RING_TYPE_RX);
4194 ASSERT(group->mrg_info.mgi_remmac != NULL);
4195
4196 return (group->mrg_info.mgi_remmac(group->mrg_info.mgi_driver, addr));
4197 }
4198
4199 /*
4200 * This is the entry point for packets transmitted through the bridging code.
4201 * If no bridge is in place, MAC_RING_TX transmits using tx ring. The 'rh'
4202 * pointer may be NULL to select the default ring.
4203 */
4204 mblk_t *
4205 mac_bridge_tx(mac_impl_t *mip, mac_ring_handle_t rh, mblk_t *mp)
4206 {
4207 mac_handle_t mh;
4208
4209 /*
4210 * Once we take a reference on the bridge link, the bridge
4211 * module itself can't unload, so the callback pointers are
4212 * stable.
4213 */
4214 mutex_enter(&mip->mi_bridge_lock);
4215 if ((mh = mip->mi_bridge_link) != NULL)
4216 mac_bridge_ref_cb(mh, B_TRUE);
4217 mutex_exit(&mip->mi_bridge_lock);
4218 if (mh == NULL) {
4219 MAC_RING_TX(mip, rh, mp, mp);
4220 } else {
4221 mp = mac_bridge_tx_cb(mh, rh, mp);
4222 mac_bridge_ref_cb(mh, B_FALSE);
4223 }
4224
4225 return (mp);
4226 }
4227
4228 /*
4229 * Find a ring from its index.
4230 */
4231 mac_ring_handle_t
4232 mac_find_ring(mac_group_handle_t gh, int index)
4233 {
4234 mac_group_t *group = (mac_group_t *)gh;
4235 mac_ring_t *ring = group->mrg_rings;
4236
4237 for (ring = group->mrg_rings; ring != NULL; ring = ring->mr_next)
4238 if (ring->mr_index == index)
4239 break;
4240
4241 return ((mac_ring_handle_t)ring);
4242 }
4243 /*
4244 * Add a ring to an existing group.
4245 *
4246 * The ring must be either passed directly (for example if the ring
4247 * movement is initiated by the framework), or specified through a driver
4248 * index (for example when the ring is added by the driver.
4249 *
4250 * The caller needs to call mac_perim_enter() before calling this function.
4251 */
4252 int
4253 i_mac_group_add_ring(mac_group_t *group, mac_ring_t *ring, int index)
4254 {
4255 mac_impl_t *mip = (mac_impl_t *)group->mrg_mh;
4256 mac_capab_rings_t *cap_rings;
4257 boolean_t driver_call = (ring == NULL);
4258 mac_group_type_t group_type;
4259 int ret = 0;
4260 flow_entry_t *flent;
4261
4262 ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
4263
4264 switch (group->mrg_type) {
4265 case MAC_RING_TYPE_RX:
4266 cap_rings = &mip->mi_rx_rings_cap;
4267 group_type = mip->mi_rx_group_type;
4268 break;
4269 case MAC_RING_TYPE_TX:
4270 cap_rings = &mip->mi_tx_rings_cap;
4271 group_type = mip->mi_tx_group_type;
4272 break;
4273 default:
4274 ASSERT(B_FALSE);
4275 }
4276
4277 /*
4278 * There should be no ring with the same ring index in the target
4279 * group.
4280 */
4281 ASSERT(mac_find_ring((mac_group_handle_t)group,
4282 driver_call ? index : ring->mr_index) == NULL);
4283
4284 if (driver_call) {
4285 /*
4286 * The function is called as a result of a request from
4287 * a driver to add a ring to an existing group, for example
4288 * from the aggregation driver. Allocate a new mac_ring_t
4289 * for that ring.
4290 */
4291 ring = mac_init_ring(mip, group, index, cap_rings);
4292 ASSERT(group->mrg_state > MAC_GROUP_STATE_UNINIT);
4293 } else {
4294 /*
4295 * The function is called as a result of a MAC layer request
4296 * to add a ring to an existing group. In this case the
4297 * ring is being moved between groups, which requires
4298 * the underlying driver to support dynamic grouping,
4299 * and the mac_ring_t already exists.
4300 */
4301 ASSERT(group_type == MAC_GROUP_TYPE_DYNAMIC);
4302 ASSERT(group->mrg_driver == NULL ||
4303 cap_rings->mr_gaddring != NULL);
4304 ASSERT(ring->mr_gh == NULL);
4305 }
4306
4307 /*
4308 * At this point the ring should not be in use, and it should be
4309 * of the right for the target group.
4310 */
4311 ASSERT(ring->mr_state < MR_INUSE);
4312 ASSERT(ring->mr_srs == NULL);
4313 ASSERT(ring->mr_type == group->mrg_type);
4314
4315 if (!driver_call) {
4316 /*
4317 * Add the driver level hardware ring if the process was not
4318 * initiated by the driver, and the target group is not the
4319 * group.
4320 */
4321 if (group->mrg_driver != NULL) {
4322 cap_rings->mr_gaddring(group->mrg_driver,
4323 ring->mr_driver, ring->mr_type);
4324 }
4325
4326 /*
4327 * Insert the ring ahead existing rings.
4328 */
4329 ring->mr_next = group->mrg_rings;
4330 group->mrg_rings = ring;
4331 ring->mr_gh = (mac_group_handle_t)group;
4332 group->mrg_cur_count++;
4333 }
4334
4335 /*
4336 * If the group has not been actively used, we're done.
4337 */
4338 if (group->mrg_index != -1 &&
4339 group->mrg_state < MAC_GROUP_STATE_RESERVED)
4340 return (0);
4341
4342 /*
4343 * Start the ring if needed. Failure causes to undo the grouping action.
4344 */
4345 if (ring->mr_state != MR_INUSE) {
4346 if ((ret = mac_start_ring(ring)) != 0) {
4347 if (!driver_call) {
4348 cap_rings->mr_gremring(group->mrg_driver,
4349 ring->mr_driver, ring->mr_type);
4350 }
4351 group->mrg_cur_count--;
4352 group->mrg_rings = ring->mr_next;
4353
4354 ring->mr_gh = NULL;
4355
4356 if (driver_call)
4357 mac_ring_free(mip, ring);
4358
4359 return (ret);
4360 }
4361 }
4362
4363 /*
4364 * Set up SRS/SR according to the ring type.
4365 */
4366 switch (ring->mr_type) {
4367 case MAC_RING_TYPE_RX:
4368 /*
4369 * Setup SRS on top of the new ring if the group is
4370 * reserved for someones exclusive use.
4371 */
4372 if (group->mrg_state == MAC_GROUP_STATE_RESERVED) {
4373 mac_client_impl_t *mcip;
4374
4375 mcip = MAC_GROUP_ONLY_CLIENT(group);
4376 /*
4377 * Even though this group is reserved we migth still
4378 * have multiple clients, i.e a VLAN shares the
4379 * group with the primary mac client.
4380 */
4381 if (mcip != NULL) {
4382 flent = mcip->mci_flent;
4383 ASSERT(flent->fe_rx_srs_cnt > 0);
4384 mac_rx_srs_group_setup(mcip, flent, SRST_LINK);
4385 mac_fanout_setup(mcip, flent,
4386 MCIP_RESOURCE_PROPS(mcip), mac_rx_deliver,
4387 mcip, NULL, NULL);
4388 } else {
4389 ring->mr_classify_type = MAC_SW_CLASSIFIER;
4390 }
4391 }
4392 break;
4393 case MAC_RING_TYPE_TX:
4394 {
4395 mac_grp_client_t *mgcp = group->mrg_clients;
4396 mac_client_impl_t *mcip;
4397 mac_soft_ring_set_t *mac_srs;
4398 mac_srs_tx_t *tx;
4399
4400 if (MAC_GROUP_NO_CLIENT(group)) {
4401 if (ring->mr_state == MR_INUSE)
4402 mac_stop_ring(ring);
4403 ring->mr_flag = 0;
4404 break;
4405 }
4406 /*
4407 * If the rings are being moved to a group that has
4408 * clients using it, then add the new rings to the
4409 * clients SRS.
4410 */
4411 while (mgcp != NULL) {
4412 boolean_t is_aggr;
4413
4414 mcip = mgcp->mgc_client;
4415 flent = mcip->mci_flent;
4416 is_aggr = (mcip->mci_state_flags & MCIS_IS_AGGR);
4417 mac_srs = MCIP_TX_SRS(mcip);
4418 tx = &mac_srs->srs_tx;
4419 mac_tx_client_quiesce((mac_client_handle_t)mcip);
4420 /*
4421 * If we are growing from 1 to multiple rings.
4422 */
4423 if (tx->st_mode == SRS_TX_BW ||
4424 tx->st_mode == SRS_TX_SERIALIZE ||
4425 tx->st_mode == SRS_TX_DEFAULT) {
4426 mac_ring_t *tx_ring = tx->st_arg2;
4427
4428 tx->st_arg2 = NULL;
4429 mac_tx_srs_stat_recreate(mac_srs, B_TRUE);
4430 mac_tx_srs_add_ring(mac_srs, tx_ring);
4431 if (mac_srs->srs_type & SRST_BW_CONTROL) {
4432 tx->st_mode = is_aggr ? SRS_TX_BW_AGGR :
4433 SRS_TX_BW_FANOUT;
4434 } else {
4435 tx->st_mode = is_aggr ? SRS_TX_AGGR :
4436 SRS_TX_FANOUT;
4437 }
4438 tx->st_func = mac_tx_get_func(tx->st_mode);
4439 }
4440 mac_tx_srs_add_ring(mac_srs, ring);
4441 mac_fanout_setup(mcip, flent, MCIP_RESOURCE_PROPS(mcip),
4442 mac_rx_deliver, mcip, NULL, NULL);
4443 mac_tx_client_restart((mac_client_handle_t)mcip);
4444 mgcp = mgcp->mgc_next;
4445 }
4446 break;
4447 }
4448 default:
4449 ASSERT(B_FALSE);
4450 }
4451 /*
4452 * For aggr, the default ring will be NULL to begin with. If it
4453 * is NULL, then pick the first ring that gets added as the
4454 * default ring. Any ring in an aggregation can be removed at
4455 * any time (by the user action of removing a link) and if the
4456 * current default ring gets removed, then a new one gets
4457 * picked (see i_mac_group_rem_ring()).
4458 */
4459 if (mip->mi_state_flags & MIS_IS_AGGR &&
4460 mip->mi_default_tx_ring == NULL &&
4461 ring->mr_type == MAC_RING_TYPE_TX) {
4462 mip->mi_default_tx_ring = (mac_ring_handle_t)ring;
4463 }
4464
4465 MAC_RING_UNMARK(ring, MR_INCIPIENT);
4466 return (0);
4467 }
4468
4469 /*
4470 * Remove a ring from it's current group. MAC internal function for dynamic
4471 * grouping.
4472 *
4473 * The caller needs to call mac_perim_enter() before calling this function.
4474 */
4475 void
4476 i_mac_group_rem_ring(mac_group_t *group, mac_ring_t *ring,
4477 boolean_t driver_call)
4478 {
4479 mac_impl_t *mip = (mac_impl_t *)group->mrg_mh;
4480 mac_capab_rings_t *cap_rings = NULL;
4481 mac_group_type_t group_type;
4482
4483 ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
4484
4485 ASSERT(mac_find_ring((mac_group_handle_t)group,
4486 ring->mr_index) == (mac_ring_handle_t)ring);
4487 ASSERT((mac_group_t *)ring->mr_gh == group);
4488 ASSERT(ring->mr_type == group->mrg_type);
4489
4490 if (ring->mr_state == MR_INUSE)
4491 mac_stop_ring(ring);
4492 switch (ring->mr_type) {
4493 case MAC_RING_TYPE_RX:
4494 group_type = mip->mi_rx_group_type;
4495 cap_rings = &mip->mi_rx_rings_cap;
4496
4497 /*
4498 * Only hardware classified packets hold a reference to the
4499 * ring all the way up the Rx path. mac_rx_srs_remove()
4500 * will take care of quiescing the Rx path and removing the
4501 * SRS. The software classified path neither holds a reference
4502 * nor any association with the ring in mac_rx.
4503 */
4504 if (ring->mr_srs != NULL) {
4505 mac_rx_srs_remove(ring->mr_srs);
4506 ring->mr_srs = NULL;
4507 }
4508
4509 break;
4510 case MAC_RING_TYPE_TX:
4511 {
4512 mac_grp_client_t *mgcp;
4513 mac_client_impl_t *mcip;
4514 mac_soft_ring_set_t *mac_srs;
4515 mac_srs_tx_t *tx;
4516 mac_ring_t *rem_ring;
4517 mac_group_t *defgrp;
4518 uint_t ring_info = 0;
4519
4520 /*
4521 * For TX this function is invoked in three
4522 * cases:
4523 *
4524 * 1) In the case of a failure during the
4525 * initial creation of a group when a share is
4526 * associated with a MAC client. So the SRS is not
4527 * yet setup, and will be setup later after the
4528 * group has been reserved and populated.
4529 *
4530 * 2) From mac_release_tx_group() when freeing
4531 * a TX SRS.
4532 *
4533 * 3) In the case of aggr, when a port gets removed,
4534 * the pseudo Tx rings that it exposed gets removed.
4535 *
4536 * In the first two cases the SRS and its soft
4537 * rings are already quiesced.
4538 */
4539 if (driver_call) {
4540 mac_client_impl_t *mcip;
4541 mac_soft_ring_set_t *mac_srs;
4542 mac_soft_ring_t *sringp;
4543 mac_srs_tx_t *srs_tx;
4544
4545 if (mip->mi_state_flags & MIS_IS_AGGR &&
4546 mip->mi_default_tx_ring ==
4547 (mac_ring_handle_t)ring) {
4548 /* pick a new default Tx ring */
4549 mip->mi_default_tx_ring =
4550 (group->mrg_rings != ring) ?
4551 (mac_ring_handle_t)group->mrg_rings :
4552 (mac_ring_handle_t)(ring->mr_next);
4553 }
4554 /* Presently only aggr case comes here */
4555 if (group->mrg_state != MAC_GROUP_STATE_RESERVED)
4556 break;
4557
4558 mcip = MAC_GROUP_ONLY_CLIENT(group);
4559 ASSERT(mcip != NULL);
4560 ASSERT(mcip->mci_state_flags & MCIS_IS_AGGR);
4561 mac_srs = MCIP_TX_SRS(mcip);
4562 ASSERT(mac_srs->srs_tx.st_mode == SRS_TX_AGGR ||
4563 mac_srs->srs_tx.st_mode == SRS_TX_BW_AGGR);
4564 srs_tx = &mac_srs->srs_tx;
4565 /*
4566 * Wakeup any callers blocked on this
4567 * Tx ring due to flow control.
4568 */
4569 sringp = srs_tx->st_soft_rings[ring->mr_index];
4570 ASSERT(sringp != NULL);
4571 mac_tx_invoke_callbacks(mcip, (mac_tx_cookie_t)sringp);
4572 mac_tx_client_quiesce((mac_client_handle_t)mcip);
4573 mac_tx_srs_del_ring(mac_srs, ring);
4574 mac_tx_client_restart((mac_client_handle_t)mcip);
4575 break;
4576 }
4577 ASSERT(ring != (mac_ring_t *)mip->mi_default_tx_ring);
4578 group_type = mip->mi_tx_group_type;
4579 cap_rings = &mip->mi_tx_rings_cap;
4580 /*
4581 * See if we need to take it out of the MAC clients using
4582 * this group
4583 */
4584 if (MAC_GROUP_NO_CLIENT(group))
4585 break;
4586 mgcp = group->mrg_clients;
4587 defgrp = MAC_DEFAULT_TX_GROUP(mip);
4588 while (mgcp != NULL) {
4589 mcip = mgcp->mgc_client;
4590 mac_srs = MCIP_TX_SRS(mcip);
4591 tx = &mac_srs->srs_tx;
4592 mac_tx_client_quiesce((mac_client_handle_t)mcip);
4593 /*
4594 * If we are here when removing rings from the
4595 * defgroup, mac_reserve_tx_ring would have
4596 * already deleted the ring from the MAC
4597 * clients in the group.
4598 */
4599 if (group != defgrp) {
4600 mac_tx_invoke_callbacks(mcip,
4601 (mac_tx_cookie_t)
4602 mac_tx_srs_get_soft_ring(mac_srs, ring));
4603 mac_tx_srs_del_ring(mac_srs, ring);
4604 }
4605 /*
4606 * Additionally, if we are left with only
4607 * one ring in the group after this, we need
4608 * to modify the mode etc. to. (We haven't
4609 * yet taken the ring out, so we check with 2).
4610 */
4611 if (group->mrg_cur_count == 2) {
4612 if (ring->mr_next == NULL)
4613 rem_ring = group->mrg_rings;
4614 else
4615 rem_ring = ring->mr_next;
4616 mac_tx_invoke_callbacks(mcip,
4617 (mac_tx_cookie_t)
4618 mac_tx_srs_get_soft_ring(mac_srs,
4619 rem_ring));
4620 mac_tx_srs_del_ring(mac_srs, rem_ring);
4621 if (rem_ring->mr_state != MR_INUSE) {
4622 (void) mac_start_ring(rem_ring);
4623 }
4624 tx->st_arg2 = (void *)rem_ring;
4625 mac_tx_srs_stat_recreate(mac_srs, B_FALSE);
4626 ring_info = mac_hwring_getinfo(
4627 (mac_ring_handle_t)rem_ring);
4628 /*
4629 * We are shrinking from multiple
4630 * to 1 ring.
4631 */
4632 if (mac_srs->srs_type & SRST_BW_CONTROL) {
4633 tx->st_mode = SRS_TX_BW;
4634 } else if (mac_tx_serialize ||
4635 (ring_info & MAC_RING_TX_SERIALIZE)) {
4636 tx->st_mode = SRS_TX_SERIALIZE;
4637 } else {
4638 tx->st_mode = SRS_TX_DEFAULT;
4639 }
4640 tx->st_func = mac_tx_get_func(tx->st_mode);
4641 }
4642 mac_tx_client_restart((mac_client_handle_t)mcip);
4643 mgcp = mgcp->mgc_next;
4644 }
4645 break;
4646 }
4647 default:
4648 ASSERT(B_FALSE);
4649 }
4650
4651 /*
4652 * Remove the ring from the group.
4653 */
4654 if (ring == group->mrg_rings)
4655 group->mrg_rings = ring->mr_next;
4656 else {
4657 mac_ring_t *pre;
4658
4659 pre = group->mrg_rings;
4660 while (pre->mr_next != ring)
4661 pre = pre->mr_next;
4662 pre->mr_next = ring->mr_next;
4663 }
4664 group->mrg_cur_count--;
4665
4666 if (!driver_call) {
4667 ASSERT(group_type == MAC_GROUP_TYPE_DYNAMIC);
4668 ASSERT(group->mrg_driver == NULL ||
4669 cap_rings->mr_gremring != NULL);
4670
4671 /*
4672 * Remove the driver level hardware ring.
4673 */
4674 if (group->mrg_driver != NULL) {
4675 cap_rings->mr_gremring(group->mrg_driver,
4676 ring->mr_driver, ring->mr_type);
4677 }
4678 }
4679
4680 ring->mr_gh = NULL;
4681 if (driver_call)
4682 mac_ring_free(mip, ring);
4683 else
4684 ring->mr_flag = 0;
4685 }
4686
4687 /*
4688 * Move a ring to the target group. If needed, remove the ring from the group
4689 * that it currently belongs to.
4690 *
4691 * The caller need to enter MAC's perimeter by calling mac_perim_enter().
4692 */
4693 static int
4694 mac_group_mov_ring(mac_impl_t *mip, mac_group_t *d_group, mac_ring_t *ring)
4695 {
4696 mac_group_t *s_group = (mac_group_t *)ring->mr_gh;
4697 int rv;
4698
4699 ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
4700 ASSERT(d_group != NULL);
4701 ASSERT(s_group->mrg_mh == d_group->mrg_mh);
4702
4703 if (s_group == d_group)
4704 return (0);
4705
4706 /*
4707 * Remove it from current group first.
4708 */
4709 if (s_group != NULL)
4710 i_mac_group_rem_ring(s_group, ring, B_FALSE);
4711
4712 /*
4713 * Add it to the new group.
4714 */
4715 rv = i_mac_group_add_ring(d_group, ring, 0);
4716 if (rv != 0) {
4717 /*
4718 * Failed to add ring back to source group. If
4719 * that fails, the ring is stuck in limbo, log message.
4720 */
4721 if (i_mac_group_add_ring(s_group, ring, 0)) {
4722 cmn_err(CE_WARN, "%s: failed to move ring %p\n",
4723 mip->mi_name, (void *)ring);
4724 }
4725 }
4726
4727 return (rv);
4728 }
4729
4730 /*
4731 * Find a MAC address according to its value.
4732 */
4733 mac_address_t *
4734 mac_find_macaddr(mac_impl_t *mip, uint8_t *mac_addr)
4735 {
4736 mac_address_t *map;
4737
4738 ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
4739
4740 for (map = mip->mi_addresses; map != NULL; map = map->ma_next) {
4741 if (bcmp(mac_addr, map->ma_addr, map->ma_len) == 0)
4742 break;
4743 }
4744
4745 return (map);
4746 }
4747
4748 /*
4749 * Check whether the MAC address is shared by multiple clients.
4750 */
4751 boolean_t
4752 mac_check_macaddr_shared(mac_address_t *map)
4753 {
4754 ASSERT(MAC_PERIM_HELD((mac_handle_t)map->ma_mip));
4755
4756 return (map->ma_nusers > 1);
4757 }
4758
4759 /*
4760 * Remove the specified MAC address from the MAC address list and free it.
4761 */
4762 static void
4763 mac_free_macaddr(mac_address_t *map)
4764 {
4765 mac_impl_t *mip = map->ma_mip;
4766
4767 ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
4768 ASSERT(mip->mi_addresses != NULL);
4769
4770 map = mac_find_macaddr(mip, map->ma_addr);
4771
4772 ASSERT(map != NULL);
4773 ASSERT(map->ma_nusers == 0);
4774
4775 if (map == mip->mi_addresses) {
4776 mip->mi_addresses = map->ma_next;
4777 } else {
4778 mac_address_t *pre;
4779
4780 pre = mip->mi_addresses;
4781 while (pre->ma_next != map)
4782 pre = pre->ma_next;
4783 pre->ma_next = map->ma_next;
4784 }
4785
4786 kmem_free(map, sizeof (mac_address_t));
4787 }
4788
4789 /*
4790 * Add a MAC address reference for a client. If the desired MAC address
4791 * exists, add a reference to it. Otherwise, add the new address by adding
4792 * it to a reserved group or setting promiscuous mode. Won't try different
4793 * group is the group is non-NULL, so the caller must explictly share
4794 * default group when needed.
4795 *
4796 * Note, the primary MAC address is initialized at registration time, so
4797 * to add it to default group only need to activate it if its reference
4798 * count is still zero. Also, some drivers may not have advertised RINGS
4799 * capability.
4800 */
4801 int
4802 mac_add_macaddr(mac_impl_t *mip, mac_group_t *group, uint8_t *mac_addr,
4803 boolean_t use_hw)
4804 {
4805 mac_address_t *map;
4806 int err = 0;
4807 boolean_t allocated_map = B_FALSE;
4808
4809 ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
4810
4811 map = mac_find_macaddr(mip, mac_addr);
4812
4813 /*
4814 * If the new MAC address has not been added. Allocate a new one
4815 * and set it up.
4816 */
4817 if (map == NULL) {
4818 map = kmem_zalloc(sizeof (mac_address_t), KM_SLEEP);
4819 map->ma_len = mip->mi_type->mt_addr_length;
4820 bcopy(mac_addr, map->ma_addr, map->ma_len);
4821 map->ma_nusers = 0;
4822 map->ma_group = group;
4823 map->ma_mip = mip;
4824
4825 /* add the new MAC address to the head of the address list */
4826 map->ma_next = mip->mi_addresses;
4827 mip->mi_addresses = map;
4828
4829 allocated_map = B_TRUE;
4830 }
4831
4832 ASSERT(map->ma_group == NULL || map->ma_group == group);
4833 if (map->ma_group == NULL)
4834 map->ma_group = group;
4835
4836 /*
4837 * If the MAC address is already in use, simply account for the
4838 * new client.
4839 */
4840 if (map->ma_nusers++ > 0)
4841 return (0);
4842
4843 /*
4844 * Activate this MAC address by adding it to the reserved group.
4845 */
4846 if (group != NULL) {
4847 err = mac_group_addmac(group, (const uint8_t *)mac_addr);
4848 if (err == 0) {
4849 map->ma_type = MAC_ADDRESS_TYPE_UNICAST_CLASSIFIED;
4850 return (0);
4851 }
4852 }
4853
4854 /*
4855 * The MAC address addition failed. If the client requires a
4856 * hardware classified MAC address, fail the operation.
4857 */
4858 if (use_hw) {
4859 err = ENOSPC;
4860 goto bail;
4861 }
4862
4863 /*
4864 * Try promiscuous mode.
4865 *
4866 * For drivers that don't advertise RINGS capability, do
4867 * nothing for the primary address.
4868 */
4869 if ((group == NULL) &&
4870 (bcmp(map->ma_addr, mip->mi_addr, map->ma_len) == 0)) {
4871 map->ma_type = MAC_ADDRESS_TYPE_UNICAST_CLASSIFIED;
4872 return (0);
4873 }
4874
4875 /*
4876 * Enable promiscuous mode in order to receive traffic
4877 * to the new MAC address.
4878 */
4879 if ((err = i_mac_promisc_set(mip, B_TRUE)) == 0) {
4880 map->ma_type = MAC_ADDRESS_TYPE_UNICAST_PROMISC;
4881 return (0);
4882 }
4883
4884 /*
4885 * Free the MAC address that could not be added. Don't free
4886 * a pre-existing address, it could have been the entry
4887 * for the primary MAC address which was pre-allocated by
4888 * mac_init_macaddr(), and which must remain on the list.
4889 */
4890 bail:
4891 map->ma_nusers--;
4892 if (allocated_map)
4893 mac_free_macaddr(map);
4894 return (err);
4895 }
4896
4897 /*
4898 * Remove a reference to a MAC address. This may cause to remove the MAC
4899 * address from an associated group or to turn off promiscuous mode.
4900 * The caller needs to handle the failure properly.
4901 */
4902 int
4903 mac_remove_macaddr(mac_address_t *map)
4904 {
4905 mac_impl_t *mip = map->ma_mip;
4906 int err = 0;
4907
4908 ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
4909
4910 ASSERT(map == mac_find_macaddr(mip, map->ma_addr));
4911
4912 /*
4913 * If it's not the last client using this MAC address, only update
4914 * the MAC clients count.
4915 */
4916 if (--map->ma_nusers > 0)
4917 return (0);
4918
4919 /*
4920 * The MAC address is no longer used by any MAC client, so remove
4921 * it from its associated group, or turn off promiscuous mode
4922 * if it was enabled for the MAC address.
4923 */
4924 switch (map->ma_type) {
4925 case MAC_ADDRESS_TYPE_UNICAST_CLASSIFIED:
4926 /*
4927 * Don't free the preset primary address for drivers that
4928 * don't advertise RINGS capability.
4929 */
4930 if (map->ma_group == NULL)
4931 return (0);
4932
4933 err = mac_group_remmac(map->ma_group, map->ma_addr);
4934 if (err == 0)
4935 map->ma_group = NULL;
4936 break;
4937 case MAC_ADDRESS_TYPE_UNICAST_PROMISC:
4938 err = i_mac_promisc_set(mip, B_FALSE);
4939 break;
4940 default:
4941 ASSERT(B_FALSE);
4942 }
4943
4944 if (err != 0)
4945 return (err);
4946
4947 /*
4948 * We created MAC address for the primary one at registration, so we
4949 * won't free it here. mac_fini_macaddr() will take care of it.
4950 */
4951 if (bcmp(map->ma_addr, mip->mi_addr, map->ma_len) != 0)
4952 mac_free_macaddr(map);
4953
4954 return (0);
4955 }
4956
4957 /*
4958 * Update an existing MAC address. The caller need to make sure that the new
4959 * value has not been used.
4960 */
4961 int
4962 mac_update_macaddr(mac_address_t *map, uint8_t *mac_addr)
4963 {
4964 mac_impl_t *mip = map->ma_mip;
4965 int err = 0;
4966
4967 ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
4968 ASSERT(mac_find_macaddr(mip, mac_addr) == NULL);
4969
4970 switch (map->ma_type) {
4971 case MAC_ADDRESS_TYPE_UNICAST_CLASSIFIED:
4972 /*
4973 * Update the primary address for drivers that are not
4974 * RINGS capable.
4975 */
4976 if (mip->mi_rx_groups == NULL) {
4977 err = mip->mi_unicst(mip->mi_driver, (const uint8_t *)
4978 mac_addr);
4979 if (err != 0)
4980 return (err);
4981 break;
4982 }
4983
4984 /*
4985 * If this MAC address is not currently in use,
4986 * simply break out and update the value.
4987 */
4988 if (map->ma_nusers == 0)
4989 break;
4990
4991 /*
4992 * Need to replace the MAC address associated with a group.
4993 */
4994 err = mac_group_remmac(map->ma_group, map->ma_addr);
4995 if (err != 0)
4996 return (err);
4997
4998 err = mac_group_addmac(map->ma_group, mac_addr);
4999
5000 /*
5001 * Failure hints hardware error. The MAC layer needs to
5002 * have error notification facility to handle this.
5003 * Now, simply try to restore the value.
5004 */
5005 if (err != 0)
5006 (void) mac_group_addmac(map->ma_group, map->ma_addr);
5007
5008 break;
5009 case MAC_ADDRESS_TYPE_UNICAST_PROMISC:
5010 /*
5011 * Need to do nothing more if in promiscuous mode.
5012 */
5013 break;
5014 default:
5015 ASSERT(B_FALSE);
5016 }
5017
5018 /*
5019 * Successfully replaced the MAC address.
5020 */
5021 if (err == 0)
5022 bcopy(mac_addr, map->ma_addr, map->ma_len);
5023
5024 return (err);
5025 }
5026
5027 /*
5028 * Freshen the MAC address with new value. Its caller must have updated the
5029 * hardware MAC address before calling this function.
5030 * This funcitons is supposed to be used to handle the MAC address change
5031 * notification from underlying drivers.
5032 */
5033 void
5034 mac_freshen_macaddr(mac_address_t *map, uint8_t *mac_addr)
5035 {
5036 mac_impl_t *mip = map->ma_mip;
5037
5038 ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
5039 ASSERT(mac_find_macaddr(mip, mac_addr) == NULL);
5040
5041 /*
5042 * Freshen the MAC address with new value.
5043 */
5044 bcopy(mac_addr, map->ma_addr, map->ma_len);
5045 bcopy(mac_addr, mip->mi_addr, map->ma_len);
5046
5047 /*
5048 * Update all MAC clients that share this MAC address.
5049 */
5050 mac_unicast_update_clients(mip, map);
5051 }
5052
5053 /*
5054 * Set up the primary MAC address.
5055 */
5056 void
5057 mac_init_macaddr(mac_impl_t *mip)
5058 {
5059 mac_address_t *map;
5060
5061 /*
5062 * The reference count is initialized to zero, until it's really
5063 * activated.
5064 */
5065 map = kmem_zalloc(sizeof (mac_address_t), KM_SLEEP);
5066 map->ma_len = mip->mi_type->mt_addr_length;
5067 bcopy(mip->mi_addr, map->ma_addr, map->ma_len);
5068
5069 /*
5070 * If driver advertises RINGS capability, it shouldn't have initialized
5071 * its primary MAC address. For other drivers, including VNIC, the
5072 * primary address must work after registration.
5073 */
5074 if (mip->mi_rx_groups == NULL)
5075 map->ma_type = MAC_ADDRESS_TYPE_UNICAST_CLASSIFIED;
5076
5077 map->ma_mip = mip;
5078
5079 mip->mi_addresses = map;
5080 }
5081
5082 /*
5083 * Clean up the primary MAC address. Note, only one primary MAC address
5084 * is allowed. All other MAC addresses must have been freed appropriately.
5085 */
5086 void
5087 mac_fini_macaddr(mac_impl_t *mip)
5088 {
5089 mac_address_t *map = mip->mi_addresses;
5090
5091 if (map == NULL)
5092 return;
5093
5094 /*
5095 * If mi_addresses is initialized, there should be exactly one
5096 * entry left on the list with no users.
5097 */
5098 ASSERT(map->ma_nusers == 0);
5099 ASSERT(map->ma_next == NULL);
5100
5101 kmem_free(map, sizeof (mac_address_t));
5102 mip->mi_addresses = NULL;
5103 }
5104
5105 /*
5106 * Logging related functions.
5107 *
5108 * Note that Kernel statistics have been extended to maintain fine
5109 * granularity of statistics viz. hardware lane, software lane, fanout
5110 * stats etc. However, extended accounting continues to support only
5111 * aggregate statistics like before.
5112 */
5113
5114 /* Write the flow description to a netinfo_t record */
5115 static netinfo_t *
5116 mac_write_flow_desc(flow_entry_t *flent, mac_client_impl_t *mcip)
5117 {
5118 netinfo_t *ninfo;
5119 net_desc_t *ndesc;
5120 flow_desc_t *fdesc;
5121 mac_resource_props_t *mrp;
5122
5123 ninfo = kmem_zalloc(sizeof (netinfo_t), KM_NOSLEEP);
5124 if (ninfo == NULL)
5125 return (NULL);
5126 ndesc = kmem_zalloc(sizeof (net_desc_t), KM_NOSLEEP);
5127 if (ndesc == NULL) {
5128 kmem_free(ninfo, sizeof (netinfo_t));
5129 return (NULL);
5130 }
5131
5132 /*
5133 * Grab the fe_lock to see a self-consistent fe_flow_desc.
5134 * Updates to the fe_flow_desc are done under the fe_lock
5135 */
5136 mutex_enter(&flent->fe_lock);
5137 fdesc = &flent->fe_flow_desc;
5138 mrp = &flent->fe_resource_props;
5139
5140 ndesc->nd_name = flent->fe_flow_name;
5141 ndesc->nd_devname = mcip->mci_name;
5142 bcopy(fdesc->fd_src_mac, ndesc->nd_ehost, ETHERADDRL);
5143 bcopy(fdesc->fd_dst_mac, ndesc->nd_edest, ETHERADDRL);
5144 ndesc->nd_sap = htonl(fdesc->fd_sap);
5145 ndesc->nd_isv4 = (uint8_t)fdesc->fd_ipversion == IPV4_VERSION;
5146 ndesc->nd_bw_limit = mrp->mrp_maxbw;
5147 if (ndesc->nd_isv4) {
5148 ndesc->nd_saddr[3] = htonl(fdesc->fd_local_addr.s6_addr32[3]);
5149 ndesc->nd_daddr[3] = htonl(fdesc->fd_remote_addr.s6_addr32[3]);
5150 } else {
5151 bcopy(&fdesc->fd_local_addr, ndesc->nd_saddr, IPV6_ADDR_LEN);
5152 bcopy(&fdesc->fd_remote_addr, ndesc->nd_daddr, IPV6_ADDR_LEN);
5153 }
5154 ndesc->nd_sport = htons(fdesc->fd_local_port);
5155 ndesc->nd_dport = htons(fdesc->fd_remote_port);
5156 ndesc->nd_protocol = (uint8_t)fdesc->fd_protocol;
5157 mutex_exit(&flent->fe_lock);
5158
5159 ninfo->ni_record = ndesc;
5160 ninfo->ni_size = sizeof (net_desc_t);
5161 ninfo->ni_type = EX_NET_FLDESC_REC;
5162
5163 return (ninfo);
5164 }
5165
5166 /* Write the flow statistics to a netinfo_t record */
5167 static netinfo_t *
5168 mac_write_flow_stats(flow_entry_t *flent)
5169 {
5170 netinfo_t *ninfo;
5171 net_stat_t *nstat;
5172 mac_soft_ring_set_t *mac_srs;
5173 mac_rx_stats_t *mac_rx_stat;
5174 mac_tx_stats_t *mac_tx_stat;
5175 int i;
5176
5177 ninfo = kmem_zalloc(sizeof (netinfo_t), KM_NOSLEEP);
5178 if (ninfo == NULL)
5179 return (NULL);
5180 nstat = kmem_zalloc(sizeof (net_stat_t), KM_NOSLEEP);
5181 if (nstat == NULL) {
5182 kmem_free(ninfo, sizeof (netinfo_t));
5183 return (NULL);
5184 }
5185
5186 nstat->ns_name = flent->fe_flow_name;
5187 for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
5188 mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i];
5189 mac_rx_stat = &mac_srs->srs_rx.sr_stat;
5190
5191 nstat->ns_ibytes += mac_rx_stat->mrs_intrbytes +
5192 mac_rx_stat->mrs_pollbytes + mac_rx_stat->mrs_lclbytes;
5193 nstat->ns_ipackets += mac_rx_stat->mrs_intrcnt +
5194 mac_rx_stat->mrs_pollcnt + mac_rx_stat->mrs_lclcnt;
5195 nstat->ns_oerrors += mac_rx_stat->mrs_ierrors;
5196 }
5197
5198 mac_srs = (mac_soft_ring_set_t *)(flent->fe_tx_srs);
5199 if (mac_srs != NULL) {
5200 mac_tx_stat = &mac_srs->srs_tx.st_stat;
5201
5202 nstat->ns_obytes = mac_tx_stat->mts_obytes;
5203 nstat->ns_opackets = mac_tx_stat->mts_opackets;
5204 nstat->ns_oerrors = mac_tx_stat->mts_oerrors;
5205 }
5206
5207 ninfo->ni_record = nstat;
5208 ninfo->ni_size = sizeof (net_stat_t);
5209 ninfo->ni_type = EX_NET_FLSTAT_REC;
5210
5211 return (ninfo);
5212 }
5213
5214 /* Write the link description to a netinfo_t record */
5215 static netinfo_t *
5216 mac_write_link_desc(mac_client_impl_t *mcip)
5217 {
5218 netinfo_t *ninfo;
5219 net_desc_t *ndesc;
5220 flow_entry_t *flent = mcip->mci_flent;
5221
5222 ninfo = kmem_zalloc(sizeof (netinfo_t), KM_NOSLEEP);
5223 if (ninfo == NULL)
5224 return (NULL);
5225 ndesc = kmem_zalloc(sizeof (net_desc_t), KM_NOSLEEP);
5226 if (ndesc == NULL) {
5227 kmem_free(ninfo, sizeof (netinfo_t));
5228 return (NULL);
5229 }
5230
5231 ndesc->nd_name = mcip->mci_name;
5232 ndesc->nd_devname = mcip->mci_name;
5233 ndesc->nd_isv4 = B_TRUE;
5234 /*
5235 * Grab the fe_lock to see a self-consistent fe_flow_desc.
5236 * Updates to the fe_flow_desc are done under the fe_lock
5237 * after removing the flent from the flow table.
5238 */
5239 mutex_enter(&flent->fe_lock);
5240 bcopy(flent->fe_flow_desc.fd_src_mac, ndesc->nd_ehost, ETHERADDRL);
5241 mutex_exit(&flent->fe_lock);
5242
5243 ninfo->ni_record = ndesc;
5244 ninfo->ni_size = sizeof (net_desc_t);
5245 ninfo->ni_type = EX_NET_LNDESC_REC;
5246
5247 return (ninfo);
5248 }
5249
5250 /* Write the link statistics to a netinfo_t record */
5251 static netinfo_t *
5252 mac_write_link_stats(mac_client_impl_t *mcip)
5253 {
5254 netinfo_t *ninfo;
5255 net_stat_t *nstat;
5256 flow_entry_t *flent;
5257 mac_soft_ring_set_t *mac_srs;
5258 mac_rx_stats_t *mac_rx_stat;
5259 mac_tx_stats_t *mac_tx_stat;
5260 int i;
5261
5262 ninfo = kmem_zalloc(sizeof (netinfo_t), KM_NOSLEEP);
5263 if (ninfo == NULL)
5264 return (NULL);
5265 nstat = kmem_zalloc(sizeof (net_stat_t), KM_NOSLEEP);
5266 if (nstat == NULL) {
5267 kmem_free(ninfo, sizeof (netinfo_t));
5268 return (NULL);
5269 }
5270
5271 nstat->ns_name = mcip->mci_name;
5272 flent = mcip->mci_flent;
5273 if (flent != NULL) {
5274 for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
5275 mac_srs = (mac_soft_ring_set_t *)flent->fe_rx_srs[i];
5276 mac_rx_stat = &mac_srs->srs_rx.sr_stat;
5277
5278 nstat->ns_ibytes += mac_rx_stat->mrs_intrbytes +
5279 mac_rx_stat->mrs_pollbytes +
5280 mac_rx_stat->mrs_lclbytes;
5281 nstat->ns_ipackets += mac_rx_stat->mrs_intrcnt +
5282 mac_rx_stat->mrs_pollcnt + mac_rx_stat->mrs_lclcnt;
5283 nstat->ns_oerrors += mac_rx_stat->mrs_ierrors;
5284 }
5285 }
5286
5287 mac_srs = (mac_soft_ring_set_t *)(mcip->mci_flent->fe_tx_srs);
5288 if (mac_srs != NULL) {
5289 mac_tx_stat = &mac_srs->srs_tx.st_stat;
5290
5291 nstat->ns_obytes = mac_tx_stat->mts_obytes;
5292 nstat->ns_opackets = mac_tx_stat->mts_opackets;
5293 nstat->ns_oerrors = mac_tx_stat->mts_oerrors;
5294 }
5295
5296 ninfo->ni_record = nstat;
5297 ninfo->ni_size = sizeof (net_stat_t);
5298 ninfo->ni_type = EX_NET_LNSTAT_REC;
5299
5300 return (ninfo);
5301 }
5302
5303 typedef struct i_mac_log_state_s {
5304 boolean_t mi_last;
5305 int mi_fenable;
5306 int mi_lenable;
5307 list_t *mi_list;
5308 } i_mac_log_state_t;
5309
5310 /*
5311 * For a given flow, if the description has not been logged before, do it now.
5312 * If it is a VNIC, then we have collected information about it from the MAC
5313 * table, so skip it.
5314 *
5315 * Called through mac_flow_walk_nolock()
5316 *
5317 * Return 0 if successful.
5318 */
5319 static int
5320 mac_log_flowinfo(flow_entry_t *flent, void *arg)
5321 {
5322 mac_client_impl_t *mcip = flent->fe_mcip;
5323 i_mac_log_state_t *lstate = arg;
5324 netinfo_t *ninfo;
5325
5326 if (mcip == NULL)
5327 return (0);
5328
5329 /*
5330 * If the name starts with "vnic", and fe_user_generated is true (to
5331 * exclude the mcast and active flow entries created implicitly for
5332 * a vnic, it is a VNIC flow. i.e. vnic1 is a vnic flow,
5333 * vnic/bge1/mcast1 is not and neither is vnic/bge1/active.
5334 */
5335 if (strncasecmp(flent->fe_flow_name, "vnic", 4) == 0 &&
5336 (flent->fe_type & FLOW_USER) != 0) {
5337 return (0);
5338 }
5339
5340 if (!flent->fe_desc_logged) {
5341 /*
5342 * We don't return error because we want to continue the
5343 * walk in case this is the last walk which means we
5344 * need to reset fe_desc_logged in all the flows.
5345 */
5346 if ((ninfo = mac_write_flow_desc(flent, mcip)) == NULL)
5347 return (0);
5348 list_insert_tail(lstate->mi_list, ninfo);
5349 flent->fe_desc_logged = B_TRUE;
5350 }
5351
5352 /*
5353 * Regardless of the error, we want to proceed in case we have to
5354 * reset fe_desc_logged.
5355 */
5356 ninfo = mac_write_flow_stats(flent);
5357 if (ninfo == NULL)
5358 return (-1);
5359
5360 list_insert_tail(lstate->mi_list, ninfo);
5361
5362 if (mcip != NULL && !(mcip->mci_state_flags & MCIS_DESC_LOGGED))
5363 flent->fe_desc_logged = B_FALSE;
5364
5365 return (0);
5366 }
5367
5368 /*
5369 * Log the description for each mac client of this mac_impl_t, if it
5370 * hasn't already been done. Additionally, log statistics for the link as
5371 * well. Walk the flow table and log information for each flow as well.
5372 * If it is the last walk (mci_last), then we turn off mci_desc_logged (and
5373 * also fe_desc_logged, if flow logging is on) since we want to log the
5374 * description if and when logging is restarted.
5375 *
5376 * Return 0 upon success or -1 upon failure
5377 */
5378 static int
5379 i_mac_impl_log(mac_impl_t *mip, i_mac_log_state_t *lstate)
5380 {
5381 mac_client_impl_t *mcip;
5382 netinfo_t *ninfo;
5383
5384 i_mac_perim_enter(mip);
5385 /*
5386 * Only walk the client list for NIC and etherstub
5387 */
5388 if ((mip->mi_state_flags & MIS_DISABLED) ||
5389 ((mip->mi_state_flags & MIS_IS_VNIC) &&
5390 (mac_get_lower_mac_handle((mac_handle_t)mip) != NULL))) {
5391 i_mac_perim_exit(mip);
5392 return (0);
5393 }
5394
5395 for (mcip = mip->mi_clients_list; mcip != NULL;
5396 mcip = mcip->mci_client_next) {
5397 if (!MCIP_DATAPATH_SETUP(mcip))
5398 continue;
5399 if (lstate->mi_lenable) {
5400 if (!(mcip->mci_state_flags & MCIS_DESC_LOGGED)) {
5401 ninfo = mac_write_link_desc(mcip);
5402 if (ninfo == NULL) {
5403 /*
5404 * We can't terminate it if this is the last
5405 * walk, else there might be some links with
5406 * mi_desc_logged set to true, which means
5407 * their description won't be logged the next
5408 * time logging is started (similarly for the
5409 * flows within such links). We can continue
5410 * without walking the flow table (i.e. to
5411 * set fe_desc_logged to false) because we
5412 * won't have written any flow stuff for this
5413 * link as we haven't logged the link itself.
5414 */
5415 i_mac_perim_exit(mip);
5416 if (lstate->mi_last)
5417 return (0);
5418 else
5419 return (-1);
5420 }
5421 mcip->mci_state_flags |= MCIS_DESC_LOGGED;
5422 list_insert_tail(lstate->mi_list, ninfo);
5423 }
5424 }
5425
5426 ninfo = mac_write_link_stats(mcip);
5427 if (ninfo == NULL && !lstate->mi_last) {
5428 i_mac_perim_exit(mip);
5429 return (-1);
5430 }
5431 list_insert_tail(lstate->mi_list, ninfo);
5432
5433 if (lstate->mi_last)
5434 mcip->mci_state_flags &= ~MCIS_DESC_LOGGED;
5435
5436 if (lstate->mi_fenable) {
5437 if (mcip->mci_subflow_tab != NULL) {
5438 (void) mac_flow_walk_nolock(
5439 mcip->mci_subflow_tab, mac_log_flowinfo,
5440 lstate);
5441 }
5442 }
5443 }
5444 i_mac_perim_exit(mip);
5445 return (0);
5446 }
5447
5448 /*
5449 * modhash walker function to add a mac_impl_t to a list
5450 */
5451 /*ARGSUSED*/
5452 static uint_t
5453 i_mac_impl_list_walker(mod_hash_key_t key, mod_hash_val_t *val, void *arg)
5454 {
5455 list_t *list = (list_t *)arg;
5456 mac_impl_t *mip = (mac_impl_t *)val;
5457
5458 if ((mip->mi_state_flags & MIS_DISABLED) == 0) {
5459 list_insert_tail(list, mip);
5460 mip->mi_ref++;
5461 }
5462
5463 return (MH_WALK_CONTINUE);
5464 }
5465
5466 void
5467 i_mac_log_info(list_t *net_log_list, i_mac_log_state_t *lstate)
5468 {
5469 list_t mac_impl_list;
5470 mac_impl_t *mip;
5471 netinfo_t *ninfo;
5472
5473 /* Create list of mac_impls */
5474 ASSERT(RW_LOCK_HELD(&i_mac_impl_lock));
5475 list_create(&mac_impl_list, sizeof (mac_impl_t), offsetof(mac_impl_t,
5476 mi_node));
5477 mod_hash_walk(i_mac_impl_hash, i_mac_impl_list_walker, &mac_impl_list);
5478 rw_exit(&i_mac_impl_lock);
5479
5480 /* Create log entries for each mac_impl */
5481 for (mip = list_head(&mac_impl_list); mip != NULL;
5482 mip = list_next(&mac_impl_list, mip)) {
5483 if (i_mac_impl_log(mip, lstate) != 0)
5484 continue;
5485 }
5486
5487 /* Remove elements and destroy list of mac_impls */
5488 rw_enter(&i_mac_impl_lock, RW_WRITER);
5489 while ((mip = list_remove_tail(&mac_impl_list)) != NULL) {
5490 mip->mi_ref--;
5491 }
5492 rw_exit(&i_mac_impl_lock);
5493 list_destroy(&mac_impl_list);
5494
5495 /*
5496 * Write log entries to files outside of locks, free associated
5497 * structures, and remove entries from the list.
5498 */
5499 while ((ninfo = list_head(net_log_list)) != NULL) {
5500 (void) exacct_commit_netinfo(ninfo->ni_record, ninfo->ni_type);
5501 list_remove(net_log_list, ninfo);
5502 kmem_free(ninfo->ni_record, ninfo->ni_size);
5503 kmem_free(ninfo, sizeof (*ninfo));
5504 }
5505 list_destroy(net_log_list);
5506 }
5507
5508 /*
5509 * The timer thread that runs every mac_logging_interval seconds and logs
5510 * link and/or flow information.
5511 */
5512 /* ARGSUSED */
5513 void
5514 mac_log_linkinfo(void *arg)
5515 {
5516 i_mac_log_state_t lstate;
5517 list_t net_log_list;
5518
5519 list_create(&net_log_list, sizeof (netinfo_t),
5520 offsetof(netinfo_t, ni_link));
5521
5522 rw_enter(&i_mac_impl_lock, RW_READER);
5523 if (!mac_flow_log_enable && !mac_link_log_enable) {
5524 rw_exit(&i_mac_impl_lock);
5525 return;
5526 }
5527 lstate.mi_fenable = mac_flow_log_enable;
5528 lstate.mi_lenable = mac_link_log_enable;
5529 lstate.mi_last = B_FALSE;
5530 lstate.mi_list = &net_log_list;
5531
5532 /* Write log entries for each mac_impl in the list */
5533 i_mac_log_info(&net_log_list, &lstate);
5534
5535 if (mac_flow_log_enable || mac_link_log_enable) {
5536 mac_logging_timer = timeout(mac_log_linkinfo, NULL,
5537 SEC_TO_TICK(mac_logging_interval));
5538 }
5539 }
5540
5541 typedef struct i_mac_fastpath_state_s {
5542 boolean_t mf_disable;
5543 int mf_err;
5544 } i_mac_fastpath_state_t;
5545
5546 /* modhash walker function to enable or disable fastpath */
5547 /*ARGSUSED*/
5548 static uint_t
5549 i_mac_fastpath_walker(mod_hash_key_t key, mod_hash_val_t *val,
5550 void *arg)
5551 {
5552 i_mac_fastpath_state_t *state = arg;
5553 mac_handle_t mh = (mac_handle_t)val;
5554
5555 if (state->mf_disable)
5556 state->mf_err = mac_fastpath_disable(mh);
5557 else
5558 mac_fastpath_enable(mh);
5559
5560 return (state->mf_err == 0 ? MH_WALK_CONTINUE : MH_WALK_TERMINATE);
5561 }
5562
5563 /*
5564 * Start the logging timer.
5565 */
5566 int
5567 mac_start_logusage(mac_logtype_t type, uint_t interval)
5568 {
5569 i_mac_fastpath_state_t dstate = {B_TRUE, 0};
5570 i_mac_fastpath_state_t estate = {B_FALSE, 0};
5571 int err;
5572
5573 rw_enter(&i_mac_impl_lock, RW_WRITER);
5574 switch (type) {
5575 case MAC_LOGTYPE_FLOW:
5576 if (mac_flow_log_enable) {
5577 rw_exit(&i_mac_impl_lock);
5578 return (0);
5579 }
5580 /* FALLTHRU */
5581 case MAC_LOGTYPE_LINK:
5582 if (mac_link_log_enable) {
5583 rw_exit(&i_mac_impl_lock);
5584 return (0);
5585 }
5586 break;
5587 default:
5588 ASSERT(0);
5589 }
5590
5591 /* Disable fastpath */
5592 mod_hash_walk(i_mac_impl_hash, i_mac_fastpath_walker, &dstate);
5593 if ((err = dstate.mf_err) != 0) {
5594 /* Reenable fastpath */
5595 mod_hash_walk(i_mac_impl_hash, i_mac_fastpath_walker, &estate);
5596 rw_exit(&i_mac_impl_lock);
5597 return (err);
5598 }
5599
5600 switch (type) {
5601 case MAC_LOGTYPE_FLOW:
5602 mac_flow_log_enable = B_TRUE;
5603 /* FALLTHRU */
5604 case MAC_LOGTYPE_LINK:
5605 mac_link_log_enable = B_TRUE;
5606 break;
5607 }
5608
5609 mac_logging_interval = interval;
5610 rw_exit(&i_mac_impl_lock);
5611 mac_log_linkinfo(NULL);
5612 return (0);
5613 }
5614
5615 /*
5616 * Stop the logging timer if both link and flow logging are turned off.
5617 */
5618 void
5619 mac_stop_logusage(mac_logtype_t type)
5620 {
5621 i_mac_log_state_t lstate;
5622 i_mac_fastpath_state_t estate = {B_FALSE, 0};
5623 list_t net_log_list;
5624
5625 list_create(&net_log_list, sizeof (netinfo_t),
5626 offsetof(netinfo_t, ni_link));
5627
5628 rw_enter(&i_mac_impl_lock, RW_WRITER);
5629
5630 lstate.mi_fenable = mac_flow_log_enable;
5631 lstate.mi_lenable = mac_link_log_enable;
5632 lstate.mi_list = &net_log_list;
5633
5634 /* Last walk */
5635 lstate.mi_last = B_TRUE;
5636
5637 switch (type) {
5638 case MAC_LOGTYPE_FLOW:
5639 if (lstate.mi_fenable) {
5640 ASSERT(mac_link_log_enable);
5641 mac_flow_log_enable = B_FALSE;
5642 mac_link_log_enable = B_FALSE;
5643 break;
5644 }
5645 /* FALLTHRU */
5646 case MAC_LOGTYPE_LINK:
5647 if (!lstate.mi_lenable || mac_flow_log_enable) {
5648 rw_exit(&i_mac_impl_lock);
5649 return;
5650 }
5651 mac_link_log_enable = B_FALSE;
5652 break;
5653 default:
5654 ASSERT(0);
5655 }
5656
5657 /* Reenable fastpath */
5658 mod_hash_walk(i_mac_impl_hash, i_mac_fastpath_walker, &estate);
5659
5660 (void) untimeout(mac_logging_timer);
5661 mac_logging_timer = 0;
5662
5663 /* Write log entries for each mac_impl in the list */
5664 i_mac_log_info(&net_log_list, &lstate);
5665 }
5666
5667 /*
5668 * Walk the rx and tx SRS/SRs for a flow and update the priority value.
5669 */
5670 void
5671 mac_flow_update_priority(mac_client_impl_t *mcip, flow_entry_t *flent)
5672 {
5673 pri_t pri;
5674 int count;
5675 mac_soft_ring_set_t *mac_srs;
5676
5677 if (flent->fe_rx_srs_cnt <= 0)
5678 return;
5679
5680 if (((mac_soft_ring_set_t *)flent->fe_rx_srs[0])->srs_type ==
5681 SRST_FLOW) {
5682 pri = FLOW_PRIORITY(mcip->mci_min_pri,
5683 mcip->mci_max_pri,
5684 flent->fe_resource_props.mrp_priority);
5685 } else {
5686 pri = mcip->mci_max_pri;
5687 }
5688
5689 for (count = 0; count < flent->fe_rx_srs_cnt; count++) {
5690 mac_srs = flent->fe_rx_srs[count];
5691 mac_update_srs_priority(mac_srs, pri);
5692 }
5693 /*
5694 * If we have a Tx SRS, we need to modify all the threads associated
5695 * with it.
5696 */
5697 if (flent->fe_tx_srs != NULL)
5698 mac_update_srs_priority(flent->fe_tx_srs, pri);
5699 }
5700
5701 /*
5702 * RX and TX rings are reserved according to different semantics depending
5703 * on the requests from the MAC clients and type of rings:
5704 *
5705 * On the Tx side, by default we reserve individual rings, independently from
5706 * the groups.
5707 *
5708 * On the Rx side, the reservation is at the granularity of the group
5709 * of rings, and used for v12n level 1 only. It has a special case for the
5710 * primary client.
5711 *
5712 * If a share is allocated to a MAC client, we allocate a TX group and an
5713 * RX group to the client, and assign TX rings and RX rings to these
5714 * groups according to information gathered from the driver through
5715 * the share capability.
5716 *
5717 * The foreseable evolution of Rx rings will handle v12n level 2 and higher
5718 * to allocate individual rings out of a group and program the hw classifier
5719 * based on IP address or higher level criteria.
5720 */
5721
5722 /*
5723 * mac_reserve_tx_ring()
5724 * Reserve a unused ring by marking it with MR_INUSE state.
5725 * As reserved, the ring is ready to function.
5726 *
5727 * Notes for Hybrid I/O:
5728 *
5729 * If a specific ring is needed, it is specified through the desired_ring
5730 * argument. Otherwise that argument is set to NULL.
5731 * If the desired ring was previous allocated to another client, this
5732 * function swaps it with a new ring from the group of unassigned rings.
5733 */
5734 mac_ring_t *
5735 mac_reserve_tx_ring(mac_impl_t *mip, mac_ring_t *desired_ring)
5736 {
5737 mac_group_t *group;
5738 mac_grp_client_t *mgcp;
5739 mac_client_impl_t *mcip;
5740 mac_soft_ring_set_t *srs;
5741
5742 ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
5743
5744 /*
5745 * Find an available ring and start it before changing its status.
5746 * The unassigned rings are at the end of the mi_tx_groups
5747 * array.
5748 */
5749 group = MAC_DEFAULT_TX_GROUP(mip);
5750
5751 /* Can't take the default ring out of the default group */
5752 ASSERT(desired_ring != (mac_ring_t *)mip->mi_default_tx_ring);
5753
5754 if (desired_ring->mr_state == MR_FREE) {
5755 ASSERT(MAC_GROUP_NO_CLIENT(group));
5756 if (mac_start_ring(desired_ring) != 0)
5757 return (NULL);
5758 return (desired_ring);
5759 }
5760 /*
5761 * There are clients using this ring, so let's move the clients
5762 * away from using this ring.
5763 */
5764 for (mgcp = group->mrg_clients; mgcp != NULL; mgcp = mgcp->mgc_next) {
5765 mcip = mgcp->mgc_client;
5766 mac_tx_client_quiesce((mac_client_handle_t)mcip);
5767 srs = MCIP_TX_SRS(mcip);
5768 ASSERT(mac_tx_srs_ring_present(srs, desired_ring));
5769 mac_tx_invoke_callbacks(mcip,
5770 (mac_tx_cookie_t)mac_tx_srs_get_soft_ring(srs,
5771 desired_ring));
5772 mac_tx_srs_del_ring(srs, desired_ring);
5773 mac_tx_client_restart((mac_client_handle_t)mcip);
5774 }
5775 return (desired_ring);
5776 }
5777
5778 /*
5779 * For a reserved group with multiple clients, return the primary client.
5780 */
5781 static mac_client_impl_t *
5782 mac_get_grp_primary(mac_group_t *grp)
5783 {
5784 mac_grp_client_t *mgcp = grp->mrg_clients;
5785 mac_client_impl_t *mcip;
5786
5787 while (mgcp != NULL) {
5788 mcip = mgcp->mgc_client;
5789 if (mcip->mci_flent->fe_type & FLOW_PRIMARY_MAC)
5790 return (mcip);
5791 mgcp = mgcp->mgc_next;
5792 }
5793 return (NULL);
5794 }
5795
5796 /*
5797 * Hybrid I/O specifies the ring that should be given to a share.
5798 * If the ring is already used by clients, then we need to release
5799 * the ring back to the default group so that we can give it to
5800 * the share. This means the clients using this ring now get a
5801 * replacement ring. If there aren't any replacement rings, this
5802 * function returns a failure.
5803 */
5804 static int
5805 mac_reclaim_ring_from_grp(mac_impl_t *mip, mac_ring_type_t ring_type,
5806 mac_ring_t *ring, mac_ring_t **rings, int nrings)
5807 {
5808 mac_group_t *group = (mac_group_t *)ring->mr_gh;
5809 mac_resource_props_t *mrp;
5810 mac_client_impl_t *mcip;
5811 mac_group_t *defgrp;
5812 mac_ring_t *tring;
5813 mac_group_t *tgrp;
5814 int i;
5815 int j;
5816
5817 mcip = MAC_GROUP_ONLY_CLIENT(group);
5818 if (mcip == NULL)
5819 mcip = mac_get_grp_primary(group);
5820 ASSERT(mcip != NULL);
5821 ASSERT(mcip->mci_share == NULL);
5822
5823 mrp = MCIP_RESOURCE_PROPS(mcip);
5824 if (ring_type == MAC_RING_TYPE_RX) {
5825 defgrp = mip->mi_rx_donor_grp;
5826 if ((mrp->mrp_mask & MRP_RX_RINGS) == 0) {
5827 /* Need to put this mac client in the default group */
5828 if (mac_rx_switch_group(mcip, group, defgrp) != 0)
5829 return (ENOSPC);
5830 } else {
5831 /*
5832 * Switch this ring with some other ring from
5833 * the default group.
5834 */
5835 for (tring = defgrp->mrg_rings; tring != NULL;
5836 tring = tring->mr_next) {
5837 if (tring->mr_index == 0)
5838 continue;
5839 for (j = 0; j < nrings; j++) {
5840 if (rings[j] == tring)
5841 break;
5842 }
5843 if (j >= nrings)
5844 break;
5845 }
5846 if (tring == NULL)
5847 return (ENOSPC);
5848 if (mac_group_mov_ring(mip, group, tring) != 0)
5849 return (ENOSPC);
5850 if (mac_group_mov_ring(mip, defgrp, ring) != 0) {
5851 (void) mac_group_mov_ring(mip, defgrp, tring);
5852 return (ENOSPC);
5853 }
5854 }
5855 ASSERT(ring->mr_gh == (mac_group_handle_t)defgrp);
5856 return (0);
5857 }
5858
5859 defgrp = MAC_DEFAULT_TX_GROUP(mip);
5860 if (ring == (mac_ring_t *)mip->mi_default_tx_ring) {
5861 /*
5862 * See if we can get a spare ring to replace the default
5863 * ring.
5864 */
5865 if (defgrp->mrg_cur_count == 1) {
5866 /*
5867 * Need to get a ring from another client, see if
5868 * there are any clients that can be moved to
5869 * the default group, thereby freeing some rings.
5870 */
5871 for (i = 0; i < mip->mi_tx_group_count; i++) {
5872 tgrp = &mip->mi_tx_groups[i];
5873 if (tgrp->mrg_state ==
5874 MAC_GROUP_STATE_REGISTERED) {
5875 continue;
5876 }
5877 mcip = MAC_GROUP_ONLY_CLIENT(tgrp);
5878 if (mcip == NULL)
5879 mcip = mac_get_grp_primary(tgrp);
5880 ASSERT(mcip != NULL);
5881 mrp = MCIP_RESOURCE_PROPS(mcip);
5882 if ((mrp->mrp_mask & MRP_TX_RINGS) == 0) {
5883 ASSERT(tgrp->mrg_cur_count == 1);
5884 /*
5885 * If this ring is part of the
5886 * rings asked by the share we cannot
5887 * use it as the default ring.
5888 */
5889 for (j = 0; j < nrings; j++) {
5890 if (rings[j] == tgrp->mrg_rings)
5891 break;
5892 }
5893 if (j < nrings)
5894 continue;
5895 mac_tx_client_quiesce(
5896 (mac_client_handle_t)mcip);
5897 mac_tx_switch_group(mcip, tgrp,
5898 defgrp);
5899 mac_tx_client_restart(
5900 (mac_client_handle_t)mcip);
5901 break;
5902 }
5903 }
5904 /*
5905 * All the rings are reserved, can't give up the
5906 * default ring.
5907 */
5908 if (defgrp->mrg_cur_count <= 1)
5909 return (ENOSPC);
5910 }
5911 /*
5912 * Swap the default ring with another.
5913 */
5914 for (tring = defgrp->mrg_rings; tring != NULL;
5915 tring = tring->mr_next) {
5916 /*
5917 * If this ring is part of the rings asked by the
5918 * share we cannot use it as the default ring.
5919 */
5920 for (j = 0; j < nrings; j++) {
5921 if (rings[j] == tring)
5922 break;
5923 }
5924 if (j >= nrings)
5925 break;
5926 }
5927 ASSERT(tring != NULL);
5928 mip->mi_default_tx_ring = (mac_ring_handle_t)tring;
5929 return (0);
5930 }
5931 /*
5932 * The Tx ring is with a group reserved by a MAC client. See if
5933 * we can swap it.
5934 */
5935 ASSERT(group->mrg_state == MAC_GROUP_STATE_RESERVED);
5936 mcip = MAC_GROUP_ONLY_CLIENT(group);
5937 if (mcip == NULL)
5938 mcip = mac_get_grp_primary(group);
5939 ASSERT(mcip != NULL);
5940 mrp = MCIP_RESOURCE_PROPS(mcip);
5941 mac_tx_client_quiesce((mac_client_handle_t)mcip);
5942 if ((mrp->mrp_mask & MRP_TX_RINGS) == 0) {
5943 ASSERT(group->mrg_cur_count == 1);
5944 /* Put this mac client in the default group */
5945 mac_tx_switch_group(mcip, group, defgrp);
5946 } else {
5947 /*
5948 * Switch this ring with some other ring from
5949 * the default group.
5950 */
5951 for (tring = defgrp->mrg_rings; tring != NULL;
5952 tring = tring->mr_next) {
5953 if (tring == (mac_ring_t *)mip->mi_default_tx_ring)
5954 continue;
5955 /*
5956 * If this ring is part of the rings asked by the
5957 * share we cannot use it for swapping.
5958 */
5959 for (j = 0; j < nrings; j++) {
5960 if (rings[j] == tring)
5961 break;
5962 }
5963 if (j >= nrings)
5964 break;
5965 }
5966 if (tring == NULL) {
5967 mac_tx_client_restart((mac_client_handle_t)mcip);
5968 return (ENOSPC);
5969 }
5970 if (mac_group_mov_ring(mip, group, tring) != 0) {
5971 mac_tx_client_restart((mac_client_handle_t)mcip);
5972 return (ENOSPC);
5973 }
5974 if (mac_group_mov_ring(mip, defgrp, ring) != 0) {
5975 (void) mac_group_mov_ring(mip, defgrp, tring);
5976 mac_tx_client_restart((mac_client_handle_t)mcip);
5977 return (ENOSPC);
5978 }
5979 }
5980 mac_tx_client_restart((mac_client_handle_t)mcip);
5981 ASSERT(ring->mr_gh == (mac_group_handle_t)defgrp);
5982 return (0);
5983 }
5984
5985 /*
5986 * Populate a zero-ring group with rings. If the share is non-NULL,
5987 * the rings are chosen according to that share.
5988 * Invoked after allocating a new RX or TX group through
5989 * mac_reserve_rx_group() or mac_reserve_tx_group(), respectively.
5990 * Returns zero on success, an errno otherwise.
5991 */
5992 int
5993 i_mac_group_allocate_rings(mac_impl_t *mip, mac_ring_type_t ring_type,
5994 mac_group_t *src_group, mac_group_t *new_group, mac_share_handle_t share,
5995 uint32_t ringcnt)
5996 {
5997 mac_ring_t **rings, *ring;
5998 uint_t nrings;
5999 int rv = 0, i = 0, j;
6000
6001 ASSERT((ring_type == MAC_RING_TYPE_RX &&
6002 mip->mi_rx_group_type == MAC_GROUP_TYPE_DYNAMIC) ||
6003 (ring_type == MAC_RING_TYPE_TX &&
6004 mip->mi_tx_group_type == MAC_GROUP_TYPE_DYNAMIC));
6005
6006 /*
6007 * First find the rings to allocate to the group.
6008 */
6009 if (share != NULL) {
6010 /* get rings through ms_squery() */
6011 mip->mi_share_capab.ms_squery(share, ring_type, NULL, &nrings);
6012 ASSERT(nrings != 0);
6013 rings = kmem_alloc(nrings * sizeof (mac_ring_handle_t),
6014 KM_SLEEP);
6015 mip->mi_share_capab.ms_squery(share, ring_type,
6016 (mac_ring_handle_t *)rings, &nrings);
6017 for (i = 0; i < nrings; i++) {
6018 /*
6019 * If we have given this ring to a non-default
6020 * group, we need to check if we can get this
6021 * ring.
6022 */
6023 ring = rings[i];
6024 if (ring->mr_gh != (mac_group_handle_t)src_group ||
6025 ring == (mac_ring_t *)mip->mi_default_tx_ring) {
6026 if (mac_reclaim_ring_from_grp(mip, ring_type,
6027 ring, rings, nrings) != 0) {
6028 rv = ENOSPC;
6029 goto bail;
6030 }
6031 }
6032 }
6033 } else {
6034 /*
6035 * Pick one ring from default group.
6036 *
6037 * for now pick the second ring which requires the first ring
6038 * at index 0 to stay in the default group, since it is the
6039 * ring which carries the multicast traffic.
6040 * We need a better way for a driver to indicate this,
6041 * for example a per-ring flag.
6042 */
6043 rings = kmem_alloc(ringcnt * sizeof (mac_ring_handle_t),
6044 KM_SLEEP);
6045 for (ring = src_group->mrg_rings; ring != NULL;
6046 ring = ring->mr_next) {
6047 if (ring_type == MAC_RING_TYPE_RX &&
6048 ring->mr_index == 0) {
6049 continue;
6050 }
6051 if (ring_type == MAC_RING_TYPE_TX &&
6052 ring == (mac_ring_t *)mip->mi_default_tx_ring) {
6053 continue;
6054 }
6055 rings[i++] = ring;
6056 if (i == ringcnt)
6057 break;
6058 }
6059 ASSERT(ring != NULL);
6060 nrings = i;
6061 /* Not enough rings as required */
6062 if (nrings != ringcnt) {
6063 rv = ENOSPC;
6064 goto bail;
6065 }
6066 }
6067
6068 switch (ring_type) {
6069 case MAC_RING_TYPE_RX:
6070 if (src_group->mrg_cur_count - nrings < 1) {
6071 /* we ran out of rings */
6072 rv = ENOSPC;
6073 goto bail;
6074 }
6075
6076 /* move receive rings to new group */
6077 for (i = 0; i < nrings; i++) {
6078 rv = mac_group_mov_ring(mip, new_group, rings[i]);
6079 if (rv != 0) {
6080 /* move rings back on failure */
6081 for (j = 0; j < i; j++) {
6082 (void) mac_group_mov_ring(mip,
6083 src_group, rings[j]);
6084 }
6085 goto bail;
6086 }
6087 }
6088 break;
6089
6090 case MAC_RING_TYPE_TX: {
6091 mac_ring_t *tmp_ring;
6092
6093 /* move the TX rings to the new group */
6094 for (i = 0; i < nrings; i++) {
6095 /* get the desired ring */
6096 tmp_ring = mac_reserve_tx_ring(mip, rings[i]);
6097 if (tmp_ring == NULL) {
6098 rv = ENOSPC;
6099 goto bail;
6100 }
6101 ASSERT(tmp_ring == rings[i]);
6102 rv = mac_group_mov_ring(mip, new_group, rings[i]);
6103 if (rv != 0) {
6104 /* cleanup on failure */
6105 for (j = 0; j < i; j++) {
6106 (void) mac_group_mov_ring(mip,
6107 MAC_DEFAULT_TX_GROUP(mip),
6108 rings[j]);
6109 }
6110 goto bail;
6111 }
6112 }
6113 break;
6114 }
6115 }
6116
6117 /* add group to share */
6118 if (share != NULL)
6119 mip->mi_share_capab.ms_sadd(share, new_group->mrg_driver);
6120
6121 bail:
6122 /* free temporary array of rings */
6123 kmem_free(rings, nrings * sizeof (mac_ring_handle_t));
6124
6125 return (rv);
6126 }
6127
6128 void
6129 mac_group_add_client(mac_group_t *grp, mac_client_impl_t *mcip)
6130 {
6131 mac_grp_client_t *mgcp;
6132
6133 for (mgcp = grp->mrg_clients; mgcp != NULL; mgcp = mgcp->mgc_next) {
6134 if (mgcp->mgc_client == mcip)
6135 break;
6136 }
6137
6138 VERIFY(mgcp == NULL);
6139
6140 mgcp = kmem_zalloc(sizeof (mac_grp_client_t), KM_SLEEP);
6141 mgcp->mgc_client = mcip;
6142 mgcp->mgc_next = grp->mrg_clients;
6143 grp->mrg_clients = mgcp;
6144
6145 }
6146
6147 void
6148 mac_group_remove_client(mac_group_t *grp, mac_client_impl_t *mcip)
6149 {
6150 mac_grp_client_t *mgcp, **pprev;
6151
6152 for (pprev = &grp->mrg_clients, mgcp = *pprev; mgcp != NULL;
6153 pprev = &mgcp->mgc_next, mgcp = *pprev) {
6154 if (mgcp->mgc_client == mcip)
6155 break;
6156 }
6157
6158 ASSERT(mgcp != NULL);
6159
6160 *pprev = mgcp->mgc_next;
6161 kmem_free(mgcp, sizeof (mac_grp_client_t));
6162 }
6163
6164 /*
6165 * mac_reserve_rx_group()
6166 *
6167 * Finds an available group and exclusively reserves it for a client.
6168 * The group is chosen to suit the flow's resource controls (bandwidth and
6169 * fanout requirements) and the address type.
6170 * If the requestor is the pimary MAC then return the group with the
6171 * largest number of rings, otherwise the default ring when available.
6172 */
6173 mac_group_t *
6174 mac_reserve_rx_group(mac_client_impl_t *mcip, uint8_t *mac_addr, boolean_t move)
6175 {
6176 mac_share_handle_t share = mcip->mci_share;
6177 mac_impl_t *mip = mcip->mci_mip;
6178 mac_group_t *grp = NULL;
6179 int i;
6180 int err = 0;
6181 mac_address_t *map;
6182 mac_resource_props_t *mrp = MCIP_RESOURCE_PROPS(mcip);
6183 int nrings;
6184 int donor_grp_rcnt;
6185 boolean_t need_exclgrp = B_FALSE;
6186 int need_rings = 0;
6187 mac_group_t *candidate_grp = NULL;
6188 mac_client_impl_t *gclient;
6189 mac_resource_props_t *gmrp;
6190 mac_group_t *donorgrp = NULL;
6191 boolean_t rxhw = mrp->mrp_mask & MRP_RX_RINGS;
6192 boolean_t unspec = mrp->mrp_mask & MRP_RXRINGS_UNSPEC;
6193 boolean_t isprimary;
6194
6195 ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
6196
6197 isprimary = mcip->mci_flent->fe_type & FLOW_PRIMARY_MAC;
6198
6199 /*
6200 * Check if a group already has this mac address (case of VLANs)
6201 * unless we are moving this MAC client from one group to another.
6202 */
6203 if (!move && (map = mac_find_macaddr(mip, mac_addr)) != NULL) {
6204 if (map->ma_group != NULL)
6205 return (map->ma_group);
6206 }
6207 if (mip->mi_rx_groups == NULL || mip->mi_rx_group_count == 0)
6208 return (NULL);
6209 /*
6210 * If exclusive open, return NULL which will enable the
6211 * caller to use the default group.
6212 */
6213 if (mcip->mci_state_flags & MCIS_EXCLUSIVE)
6214 return (NULL);
6215
6216 /* For dynamic groups default unspecified to 1 */
6217 if (rxhw && unspec &&
6218 mip->mi_rx_group_type == MAC_GROUP_TYPE_DYNAMIC) {
6219 mrp->mrp_nrxrings = 1;
6220 }
6221 /*
6222 * For static grouping we allow only specifying rings=0 and
6223 * unspecified
6224 */
6225 if (rxhw && mrp->mrp_nrxrings > 0 &&
6226 mip->mi_rx_group_type == MAC_GROUP_TYPE_STATIC) {
6227 return (NULL);
6228 }
6229 if (rxhw) {
6230 /*
6231 * We have explicitly asked for a group (with nrxrings,
6232 * if unspec).
6233 */
6234 if (unspec || mrp->mrp_nrxrings > 0) {
6235 need_exclgrp = B_TRUE;
6236 need_rings = mrp->mrp_nrxrings;
6237 } else if (mrp->mrp_nrxrings == 0) {
6238 /*
6239 * We have asked for a software group.
6240 */
6241 return (NULL);
6242 }
6243 } else if (isprimary && mip->mi_nactiveclients == 1 &&
6244 mip->mi_rx_group_type == MAC_GROUP_TYPE_DYNAMIC) {
6245 /*
6246 * If the primary is the only active client on this
6247 * mip and we have not asked for any rings, we give
6248 * it the default group so that the primary gets to
6249 * use all the rings.
6250 */
6251 return (NULL);
6252 }
6253
6254 /* The group that can donate rings */
6255 donorgrp = mip->mi_rx_donor_grp;
6256
6257 /*
6258 * The number of rings that the default group can donate.
6259 * We need to leave at least one ring.
6260 */
6261 donor_grp_rcnt = donorgrp->mrg_cur_count - 1;
6262
6263 /*
6264 * Try to exclusively reserve a RX group.
6265 *
6266 * For flows requiring HW_DEFAULT_RING (unicast flow of the primary
6267 * client), try to reserve the a non-default RX group and give
6268 * it all the rings from the donor group, except the default ring
6269 *
6270 * For flows requiring HW_RING (unicast flow of other clients), try
6271 * to reserve non-default RX group with the specified number of
6272 * rings, if available.
6273 *
6274 * For flows that have not asked for software or hardware ring,
6275 * try to reserve a non-default group with 1 ring, if available.
6276 */
6277 for (i = 1; i < mip->mi_rx_group_count; i++) {
6278 grp = &mip->mi_rx_groups[i];
6279
6280 DTRACE_PROBE3(rx__group__trying, char *, mip->mi_name,
6281 int, grp->mrg_index, mac_group_state_t, grp->mrg_state);
6282
6283 /*
6284 * Check if this group could be a candidate group for
6285 * eviction if we need a group for this MAC client,
6286 * but there aren't any. A candidate group is one
6287 * that didn't ask for an exclusive group, but got
6288 * one and it has enough rings (combined with what
6289 * the donor group can donate) for the new MAC
6290 * client
6291 */
6292 if (grp->mrg_state >= MAC_GROUP_STATE_RESERVED) {
6293 /*
6294 * If the primary/donor group is not the default
6295 * group, don't bother looking for a candidate group.
6296 * If we don't have enough rings we will check
6297 * if the primary group can be vacated.
6298 */
6299 if (candidate_grp == NULL &&
6300 donorgrp == MAC_DEFAULT_RX_GROUP(mip)) {
6301 ASSERT(!MAC_GROUP_NO_CLIENT(grp));
6302 gclient = MAC_GROUP_ONLY_CLIENT(grp);
6303 if (gclient == NULL)
6304 gclient = mac_get_grp_primary(grp);
6305 ASSERT(gclient != NULL);
6306 gmrp = MCIP_RESOURCE_PROPS(gclient);
6307 if (gclient->mci_share == NULL &&
6308 (gmrp->mrp_mask & MRP_RX_RINGS) == 0 &&
6309 (unspec ||
6310 (grp->mrg_cur_count + donor_grp_rcnt >=
6311 need_rings))) {
6312 candidate_grp = grp;
6313 }
6314 }
6315 continue;
6316 }
6317 /*
6318 * This group could already be SHARED by other multicast
6319 * flows on this client. In that case, the group would
6320 * be shared and has already been started.
6321 */
6322 ASSERT(grp->mrg_state != MAC_GROUP_STATE_UNINIT);
6323
6324 if ((grp->mrg_state == MAC_GROUP_STATE_REGISTERED) &&
6325 (mac_start_group(grp) != 0)) {
6326 continue;
6327 }
6328
6329 if (mip->mi_rx_group_type != MAC_GROUP_TYPE_DYNAMIC)
6330 break;
6331 ASSERT(grp->mrg_cur_count == 0);
6332
6333 /*
6334 * Populate the group. Rings should be taken
6335 * from the donor group.
6336 */
6337 nrings = rxhw ? need_rings : isprimary ? donor_grp_rcnt: 1;
6338
6339 /*
6340 * If the donor group can't donate, let's just walk and
6341 * see if someone can vacate a group, so that we have
6342 * enough rings for this, unless we already have
6343 * identified a candiate group..
6344 */
6345 if (nrings <= donor_grp_rcnt) {
6346 err = i_mac_group_allocate_rings(mip, MAC_RING_TYPE_RX,
6347 donorgrp, grp, share, nrings);
6348 if (err == 0) {
6349 /*
6350 * For a share i_mac_group_allocate_rings gets
6351 * the rings from the driver, let's populate
6352 * the property for the client now.
6353 */
6354 if (share != NULL) {
6355 mac_client_set_rings(
6356 (mac_client_handle_t)mcip,
6357 grp->mrg_cur_count, -1);
6358 }
6359 if (mac_is_primary_client(mcip) && !rxhw)
6360 mip->mi_rx_donor_grp = grp;
6361 break;
6362 }
6363 }
6364
6365 DTRACE_PROBE3(rx__group__reserve__alloc__rings, char *,
6366 mip->mi_name, int, grp->mrg_index, int, err);
6367
6368 /*
6369 * It's a dynamic group but the grouping operation
6370 * failed.
6371 */
6372 mac_stop_group(grp);
6373 }
6374 /* We didn't find an exclusive group for this MAC client */
6375 if (i >= mip->mi_rx_group_count) {
6376
6377 if (!need_exclgrp)
6378 return (NULL);
6379
6380 /*
6381 * If we found a candidate group then we switch the
6382 * MAC client from the candidate_group to the default
6383 * group and give the group to this MAC client. If
6384 * we didn't find a candidate_group, check if the
6385 * primary is in its own group and if it can make way
6386 * for this MAC client.
6387 */
6388 if (candidate_grp == NULL &&
6389 donorgrp != MAC_DEFAULT_RX_GROUP(mip) &&
6390 donorgrp->mrg_cur_count >= need_rings) {
6391 candidate_grp = donorgrp;
6392 }
6393 if (candidate_grp != NULL) {
6394 boolean_t prim_grp = B_FALSE;
6395
6396 /*
6397 * Switch the MAC client from the candidate group
6398 * to the default group.. If this group was the
6399 * donor group, then after the switch we need
6400 * to update the donor group too.
6401 */
6402 grp = candidate_grp;
6403 gclient = MAC_GROUP_ONLY_CLIENT(grp);
6404 if (gclient == NULL)
6405 gclient = mac_get_grp_primary(grp);
6406 if (grp == mip->mi_rx_donor_grp)
6407 prim_grp = B_TRUE;
6408 if (mac_rx_switch_group(gclient, grp,
6409 MAC_DEFAULT_RX_GROUP(mip)) != 0) {
6410 return (NULL);
6411 }
6412 if (prim_grp) {
6413 mip->mi_rx_donor_grp =
6414 MAC_DEFAULT_RX_GROUP(mip);
6415 donorgrp = MAC_DEFAULT_RX_GROUP(mip);
6416 }
6417
6418
6419 /*
6420 * Now give this group with the required rings
6421 * to this MAC client.
6422 */
6423 ASSERT(grp->mrg_state == MAC_GROUP_STATE_REGISTERED);
6424 if (mac_start_group(grp) != 0)
6425 return (NULL);
6426
6427 if (mip->mi_rx_group_type != MAC_GROUP_TYPE_DYNAMIC)
6428 return (grp);
6429
6430 donor_grp_rcnt = donorgrp->mrg_cur_count - 1;
6431 ASSERT(grp->mrg_cur_count == 0);
6432 ASSERT(donor_grp_rcnt >= need_rings);
6433 err = i_mac_group_allocate_rings(mip, MAC_RING_TYPE_RX,
6434 donorgrp, grp, share, need_rings);
6435 if (err == 0) {
6436 /*
6437 * For a share i_mac_group_allocate_rings gets
6438 * the rings from the driver, let's populate
6439 * the property for the client now.
6440 */
6441 if (share != NULL) {
6442 mac_client_set_rings(
6443 (mac_client_handle_t)mcip,
6444 grp->mrg_cur_count, -1);
6445 }
6446 DTRACE_PROBE2(rx__group__reserved,
6447 char *, mip->mi_name, int, grp->mrg_index);
6448 return (grp);
6449 }
6450 DTRACE_PROBE3(rx__group__reserve__alloc__rings, char *,
6451 mip->mi_name, int, grp->mrg_index, int, err);
6452 mac_stop_group(grp);
6453 }
6454 return (NULL);
6455 }
6456 ASSERT(grp != NULL);
6457
6458 DTRACE_PROBE2(rx__group__reserved,
6459 char *, mip->mi_name, int, grp->mrg_index);
6460 return (grp);
6461 }
6462
6463 /*
6464 * mac_rx_release_group()
6465 *
6466 * This is called when there are no clients left for the group.
6467 * The group is stopped and marked MAC_GROUP_STATE_REGISTERED,
6468 * and if it is a non default group, the shares are removed and
6469 * all rings are assigned back to default group.
6470 */
6471 void
6472 mac_release_rx_group(mac_client_impl_t *mcip, mac_group_t *group)
6473 {
6474 mac_impl_t *mip = mcip->mci_mip;
6475 mac_ring_t *ring;
6476
6477 ASSERT(group != MAC_DEFAULT_RX_GROUP(mip));
6478
6479 if (mip->mi_rx_donor_grp == group)
6480 mip->mi_rx_donor_grp = MAC_DEFAULT_RX_GROUP(mip);
6481
6482 /*
6483 * This is the case where there are no clients left. Any
6484 * SRS etc on this group have also be quiesced.
6485 */
6486 for (ring = group->mrg_rings; ring != NULL; ring = ring->mr_next) {
6487 if (ring->mr_classify_type == MAC_HW_CLASSIFIER) {
6488 ASSERT(group->mrg_state == MAC_GROUP_STATE_RESERVED);
6489 /*
6490 * Remove the SRS associated with the HW ring.
6491 * As a result, polling will be disabled.
6492 */
6493 ring->mr_srs = NULL;
6494 }
6495 ASSERT(group->mrg_state < MAC_GROUP_STATE_RESERVED ||
6496 ring->mr_state == MR_INUSE);
6497 if (ring->mr_state == MR_INUSE) {
6498 mac_stop_ring(ring);
6499 ring->mr_flag = 0;
6500 }
6501 }
6502
6503 /* remove group from share */
6504 if (mcip->mci_share != NULL) {
6505 mip->mi_share_capab.ms_sremove(mcip->mci_share,
6506 group->mrg_driver);
6507 }
6508
6509 if (mip->mi_rx_group_type == MAC_GROUP_TYPE_DYNAMIC) {
6510 mac_ring_t *ring;
6511
6512 /*
6513 * Rings were dynamically allocated to group.
6514 * Move rings back to default group.
6515 */
6516 while ((ring = group->mrg_rings) != NULL) {
6517 (void) mac_group_mov_ring(mip, mip->mi_rx_donor_grp,
6518 ring);
6519 }
6520 }
6521 mac_stop_group(group);
6522 /*
6523 * Possible improvement: See if we can assign the group just released
6524 * to a another client of the mip
6525 */
6526 }
6527
6528 /*
6529 * When we move the primary's mac address between groups, we need to also
6530 * take all the clients sharing the same mac address along with it (VLANs)
6531 * We remove the mac address for such clients from the group after quiescing
6532 * them. When we add the mac address we restart the client. Note that
6533 * the primary's mac address is removed from the group after all the
6534 * other clients sharing the address are removed. Similarly, the primary's
6535 * mac address is added before all the other client's mac address are
6536 * added. While grp is the group where the clients reside, tgrp is
6537 * the group where the addresses have to be added.
6538 */
6539 static void
6540 mac_rx_move_macaddr_prim(mac_client_impl_t *mcip, mac_group_t *grp,
6541 mac_group_t *tgrp, uint8_t *maddr, boolean_t add)
6542 {
6543 mac_impl_t *mip = mcip->mci_mip;
6544 mac_grp_client_t *mgcp = grp->mrg_clients;
6545 mac_client_impl_t *gmcip;
6546 boolean_t prim;
6547
6548 prim = (mcip->mci_state_flags & MCIS_UNICAST_HW) != 0;
6549
6550 /*
6551 * If the clients are in a non-default group, we just have to
6552 * walk the group's client list. If it is in the default group
6553 * (which will be shared by other clients as well, we need to
6554 * check if the unicast address matches mcip's unicast.
6555 */
6556 while (mgcp != NULL) {
6557 gmcip = mgcp->mgc_client;
6558 if (gmcip != mcip &&
6559 (grp != MAC_DEFAULT_RX_GROUP(mip) ||
6560 mcip->mci_unicast == gmcip->mci_unicast)) {
6561 if (!add) {
6562 mac_rx_client_quiesce(
6563 (mac_client_handle_t)gmcip);
6564 (void) mac_remove_macaddr(mcip->mci_unicast);
6565 } else {
6566 (void) mac_add_macaddr(mip, tgrp, maddr, prim);
6567 mac_rx_client_restart(
6568 (mac_client_handle_t)gmcip);
6569 }
6570 }
6571 mgcp = mgcp->mgc_next;
6572 }
6573 }
6574
6575
6576 /*
6577 * Move the MAC address from fgrp to tgrp. If this is the primary client,
6578 * we need to take any VLANs etc. together too.
6579 */
6580 static int
6581 mac_rx_move_macaddr(mac_client_impl_t *mcip, mac_group_t *fgrp,
6582 mac_group_t *tgrp)
6583 {
6584 mac_impl_t *mip = mcip->mci_mip;
6585 uint8_t maddr[MAXMACADDRLEN];
6586 int err = 0;
6587 boolean_t prim;
6588 boolean_t multiclnt = B_FALSE;
6589
6590 mac_rx_client_quiesce((mac_client_handle_t)mcip);
6591 ASSERT(mcip->mci_unicast != NULL);
6592 bcopy(mcip->mci_unicast->ma_addr, maddr, mcip->mci_unicast->ma_len);
6593
6594 prim = (mcip->mci_state_flags & MCIS_UNICAST_HW) != 0;
6595 if (mcip->mci_unicast->ma_nusers > 1) {
6596 mac_rx_move_macaddr_prim(mcip, fgrp, NULL, maddr, B_FALSE);
6597 multiclnt = B_TRUE;
6598 }
6599 ASSERT(mcip->mci_unicast->ma_nusers == 1);
6600 err = mac_remove_macaddr(mcip->mci_unicast);
6601 if (err != 0) {
6602 mac_rx_client_restart((mac_client_handle_t)mcip);
6603 if (multiclnt) {
6604 mac_rx_move_macaddr_prim(mcip, fgrp, fgrp, maddr,
6605 B_TRUE);
6606 }
6607 return (err);
6608 }
6609 /*
6610 * Program the H/W Classifier first, if this fails we need
6611 * not proceed with the other stuff.
6612 */
6613 if ((err = mac_add_macaddr(mip, tgrp, maddr, prim)) != 0) {
6614 /* Revert back the H/W Classifier */
6615 if ((err = mac_add_macaddr(mip, fgrp, maddr, prim)) != 0) {
6616 /*
6617 * This should not fail now since it worked earlier,
6618 * should we panic?
6619 */
6620 cmn_err(CE_WARN,
6621 "mac_rx_switch_group: switching %p back"
6622 " to group %p failed!!", (void *)mcip,
6623 (void *)fgrp);
6624 }
6625 mac_rx_client_restart((mac_client_handle_t)mcip);
6626 if (multiclnt) {
6627 mac_rx_move_macaddr_prim(mcip, fgrp, fgrp, maddr,
6628 B_TRUE);
6629 }
6630 return (err);
6631 }
6632 mcip->mci_unicast = mac_find_macaddr(mip, maddr);
6633 mac_rx_client_restart((mac_client_handle_t)mcip);
6634 if (multiclnt)
6635 mac_rx_move_macaddr_prim(mcip, fgrp, tgrp, maddr, B_TRUE);
6636 return (err);
6637 }
6638
6639 /*
6640 * Switch the MAC client from one group to another. This means we need
6641 * to remove the MAC address from the group, remove the MAC client,
6642 * teardown the SRSs and revert the group state. Then, we add the client
6643 * to the destination group, set the SRSs, and add the MAC address to the
6644 * group.
6645 */
6646 int
6647 mac_rx_switch_group(mac_client_impl_t *mcip, mac_group_t *fgrp,
6648 mac_group_t *tgrp)
6649 {
6650 int err;
6651 mac_group_state_t next_state;
6652 mac_client_impl_t *group_only_mcip;
6653 mac_client_impl_t *gmcip;
6654 mac_impl_t *mip = mcip->mci_mip;
6655 mac_grp_client_t *mgcp;
6656
6657 ASSERT(fgrp == mcip->mci_flent->fe_rx_ring_group);
6658
6659 if ((err = mac_rx_move_macaddr(mcip, fgrp, tgrp)) != 0)
6660 return (err);
6661
6662 /*
6663 * The group might be reserved, but SRSs may not be set up, e.g.
6664 * primary and its vlans using a reserved group.
6665 */
6666 if (fgrp->mrg_state == MAC_GROUP_STATE_RESERVED &&
6667 MAC_GROUP_ONLY_CLIENT(fgrp) != NULL) {
6668 mac_rx_srs_group_teardown(mcip->mci_flent, B_TRUE);
6669 }
6670 if (fgrp != MAC_DEFAULT_RX_GROUP(mip)) {
6671 mgcp = fgrp->mrg_clients;
6672 while (mgcp != NULL) {
6673 gmcip = mgcp->mgc_client;
6674 mgcp = mgcp->mgc_next;
6675 mac_group_remove_client(fgrp, gmcip);
6676 mac_group_add_client(tgrp, gmcip);
6677 gmcip->mci_flent->fe_rx_ring_group = tgrp;
6678 }
6679 mac_release_rx_group(mcip, fgrp);
6680 ASSERT(MAC_GROUP_NO_CLIENT(fgrp));
6681 mac_set_group_state(fgrp, MAC_GROUP_STATE_REGISTERED);
6682 } else {
6683 mac_group_remove_client(fgrp, mcip);
6684 mac_group_add_client(tgrp, mcip);
6685 mcip->mci_flent->fe_rx_ring_group = tgrp;
6686 /*
6687 * If there are other clients (VLANs) sharing this address
6688 * we should be here only for the primary.
6689 */
6690 if (mcip->mci_unicast->ma_nusers > 1) {
6691 /*
6692 * We need to move all the clients that are using
6693 * this h/w address.
6694 */
6695 mgcp = fgrp->mrg_clients;
6696 while (mgcp != NULL) {
6697 gmcip = mgcp->mgc_client;
6698 mgcp = mgcp->mgc_next;
6699 if (mcip->mci_unicast == gmcip->mci_unicast) {
6700 mac_group_remove_client(fgrp, gmcip);
6701 mac_group_add_client(tgrp, gmcip);
6702 gmcip->mci_flent->fe_rx_ring_group =
6703 tgrp;
6704 }
6705 }
6706 }
6707 /*
6708 * The default group will still take the multicast,
6709 * broadcast traffic etc., so it won't go to
6710 * MAC_GROUP_STATE_REGISTERED.
6711 */
6712 if (fgrp->mrg_state == MAC_GROUP_STATE_RESERVED)
6713 mac_rx_group_unmark(fgrp, MR_CONDEMNED);
6714 mac_set_group_state(fgrp, MAC_GROUP_STATE_SHARED);
6715 }
6716 next_state = mac_group_next_state(tgrp, &group_only_mcip,
6717 MAC_DEFAULT_RX_GROUP(mip), B_TRUE);
6718 mac_set_group_state(tgrp, next_state);
6719 /*
6720 * If the destination group is reserved, setup the SRSs etc.
6721 */
6722 if (tgrp->mrg_state == MAC_GROUP_STATE_RESERVED) {
6723 mac_rx_srs_group_setup(mcip, mcip->mci_flent, SRST_LINK);
6724 mac_fanout_setup(mcip, mcip->mci_flent,
6725 MCIP_RESOURCE_PROPS(mcip), mac_rx_deliver, mcip, NULL,
6726 NULL);
6727 mac_rx_group_unmark(tgrp, MR_INCIPIENT);
6728 } else {
6729 mac_rx_switch_grp_to_sw(tgrp);
6730 }
6731 return (0);
6732 }
6733
6734 /*
6735 * Reserves a TX group for the specified share. Invoked by mac_tx_srs_setup()
6736 * when a share was allocated to the client.
6737 */
6738 mac_group_t *
6739 mac_reserve_tx_group(mac_client_impl_t *mcip, boolean_t move)
6740 {
6741 mac_impl_t *mip = mcip->mci_mip;
6742 mac_group_t *grp = NULL;
6743 int rv;
6744 int i;
6745 int err;
6746 mac_group_t *defgrp;
6747 mac_share_handle_t share = mcip->mci_share;
6748 mac_resource_props_t *mrp = MCIP_RESOURCE_PROPS(mcip);
6749 int nrings;
6750 int defnrings;
6751 boolean_t need_exclgrp = B_FALSE;
6752 int need_rings = 0;
6753 mac_group_t *candidate_grp = NULL;
6754 mac_client_impl_t *gclient;
6755 mac_resource_props_t *gmrp;
6756 boolean_t txhw = mrp->mrp_mask & MRP_TX_RINGS;
6757 boolean_t unspec = mrp->mrp_mask & MRP_TXRINGS_UNSPEC;
6758 boolean_t isprimary;
6759
6760 isprimary = mcip->mci_flent->fe_type & FLOW_PRIMARY_MAC;
6761 /*
6762 * When we come here for a VLAN on the primary (dladm create-vlan),
6763 * we need to pair it along with the primary (to keep it consistent
6764 * with the RX side). So, we check if the primary is already assigned
6765 * to a group and return the group if so. The other way is also
6766 * true, i.e. the VLAN is already created and now we are plumbing
6767 * the primary.
6768 */
6769 if (!move && isprimary) {
6770 for (gclient = mip->mi_clients_list; gclient != NULL;
6771 gclient = gclient->mci_client_next) {
6772 if (gclient->mci_flent->fe_type & FLOW_PRIMARY_MAC &&
6773 gclient->mci_flent->fe_tx_ring_group != NULL) {
6774 return (gclient->mci_flent->fe_tx_ring_group);
6775 }
6776 }
6777 }
6778
6779 if (mip->mi_tx_groups == NULL || mip->mi_tx_group_count == 0)
6780 return (NULL);
6781
6782 /* For dynamic groups, default unspec to 1 */
6783 if (txhw && unspec &&
6784 mip->mi_tx_group_type == MAC_GROUP_TYPE_DYNAMIC) {
6785 mrp->mrp_ntxrings = 1;
6786 }
6787 /*
6788 * For static grouping we allow only specifying rings=0 and
6789 * unspecified
6790 */
6791 if (txhw && mrp->mrp_ntxrings > 0 &&
6792 mip->mi_tx_group_type == MAC_GROUP_TYPE_STATIC) {
6793 return (NULL);
6794 }
6795
6796 if (txhw) {
6797 /*
6798 * We have explicitly asked for a group (with ntxrings,
6799 * if unspec).
6800 */
6801 if (unspec || mrp->mrp_ntxrings > 0) {
6802 need_exclgrp = B_TRUE;
6803 need_rings = mrp->mrp_ntxrings;
6804 } else if (mrp->mrp_ntxrings == 0) {
6805 /*
6806 * We have asked for a software group.
6807 */
6808 return (NULL);
6809 }
6810 }
6811 defgrp = MAC_DEFAULT_TX_GROUP(mip);
6812 /*
6813 * The number of rings that the default group can donate.
6814 * We need to leave at least one ring - the default ring - in
6815 * this group.
6816 */
6817 defnrings = defgrp->mrg_cur_count - 1;
6818
6819 /*
6820 * Primary gets default group unless explicitly told not
6821 * to (i.e. rings > 0).
6822 */
6823 if (isprimary && !need_exclgrp)
6824 return (NULL);
6825
6826 nrings = (mrp->mrp_mask & MRP_TX_RINGS) != 0 ? mrp->mrp_ntxrings : 1;
6827 for (i = 0; i < mip->mi_tx_group_count; i++) {
6828 grp = &mip->mi_tx_groups[i];
6829 if ((grp->mrg_state == MAC_GROUP_STATE_RESERVED) ||
6830 (grp->mrg_state == MAC_GROUP_STATE_UNINIT)) {
6831 /*
6832 * Select a candidate for replacement if we don't
6833 * get an exclusive group. A candidate group is one
6834 * that didn't ask for an exclusive group, but got
6835 * one and it has enough rings (combined with what
6836 * the default group can donate) for the new MAC
6837 * client.
6838 */
6839 if (grp->mrg_state == MAC_GROUP_STATE_RESERVED &&
6840 candidate_grp == NULL) {
6841 gclient = MAC_GROUP_ONLY_CLIENT(grp);
6842 if (gclient == NULL)
6843 gclient = mac_get_grp_primary(grp);
6844 gmrp = MCIP_RESOURCE_PROPS(gclient);
6845 if (gclient->mci_share == NULL &&
6846 (gmrp->mrp_mask & MRP_TX_RINGS) == 0 &&
6847 (unspec ||
6848 (grp->mrg_cur_count + defnrings) >=
6849 need_rings)) {
6850 candidate_grp = grp;
6851 }
6852 }
6853 continue;
6854 }
6855 /*
6856 * If the default can't donate let's just walk and
6857 * see if someone can vacate a group, so that we have
6858 * enough rings for this.
6859 */
6860 if (mip->mi_tx_group_type != MAC_GROUP_TYPE_DYNAMIC ||
6861 nrings <= defnrings) {
6862 if (grp->mrg_state == MAC_GROUP_STATE_REGISTERED) {
6863 rv = mac_start_group(grp);
6864 ASSERT(rv == 0);
6865 }
6866 break;
6867 }
6868 }
6869
6870 /* The default group */
6871 if (i >= mip->mi_tx_group_count) {
6872 /*
6873 * If we need an exclusive group and have identified a
6874 * candidate group we switch the MAC client from the
6875 * candidate group to the default group and give the
6876 * candidate group to this client.
6877 */
6878 if (need_exclgrp && candidate_grp != NULL) {
6879 /*
6880 * Switch the MAC client from the candidate group
6881 * to the default group.
6882 */
6883 grp = candidate_grp;
6884 gclient = MAC_GROUP_ONLY_CLIENT(grp);
6885 if (gclient == NULL)
6886 gclient = mac_get_grp_primary(grp);
6887 mac_tx_client_quiesce((mac_client_handle_t)gclient);
6888 mac_tx_switch_group(gclient, grp, defgrp);
6889 mac_tx_client_restart((mac_client_handle_t)gclient);
6890
6891 /*
6892 * Give the candidate group with the specified number
6893 * of rings to this MAC client.
6894 */
6895 ASSERT(grp->mrg_state == MAC_GROUP_STATE_REGISTERED);
6896 rv = mac_start_group(grp);
6897 ASSERT(rv == 0);
6898
6899 if (mip->mi_tx_group_type != MAC_GROUP_TYPE_DYNAMIC)
6900 return (grp);
6901
6902 ASSERT(grp->mrg_cur_count == 0);
6903 ASSERT(defgrp->mrg_cur_count > need_rings);
6904
6905 err = i_mac_group_allocate_rings(mip, MAC_RING_TYPE_TX,
6906 defgrp, grp, share, need_rings);
6907 if (err == 0) {
6908 /*
6909 * For a share i_mac_group_allocate_rings gets
6910 * the rings from the driver, let's populate
6911 * the property for the client now.
6912 */
6913 if (share != NULL) {
6914 mac_client_set_rings(
6915 (mac_client_handle_t)mcip, -1,
6916 grp->mrg_cur_count);
6917 }
6918 mip->mi_tx_group_free--;
6919 return (grp);
6920 }
6921 DTRACE_PROBE3(tx__group__reserve__alloc__rings, char *,
6922 mip->mi_name, int, grp->mrg_index, int, err);
6923 mac_stop_group(grp);
6924 }
6925 return (NULL);
6926 }
6927 /*
6928 * We got an exclusive group, but it is not dynamic.
6929 */
6930 if (mip->mi_tx_group_type != MAC_GROUP_TYPE_DYNAMIC) {
6931 mip->mi_tx_group_free--;
6932 return (grp);
6933 }
6934
6935 rv = i_mac_group_allocate_rings(mip, MAC_RING_TYPE_TX, defgrp, grp,
6936 share, nrings);
6937 if (rv != 0) {
6938 DTRACE_PROBE3(tx__group__reserve__alloc__rings,
6939 char *, mip->mi_name, int, grp->mrg_index, int, rv);
6940 mac_stop_group(grp);
6941 return (NULL);
6942 }
6943 /*
6944 * For a share i_mac_group_allocate_rings gets the rings from the
6945 * driver, let's populate the property for the client now.
6946 */
6947 if (share != NULL) {
6948 mac_client_set_rings((mac_client_handle_t)mcip, -1,
6949 grp->mrg_cur_count);
6950 }
6951 mip->mi_tx_group_free--;
6952 return (grp);
6953 }
6954
6955 void
6956 mac_release_tx_group(mac_client_impl_t *mcip, mac_group_t *grp)
6957 {
6958 mac_impl_t *mip = mcip->mci_mip;
6959 mac_share_handle_t share = mcip->mci_share;
6960 mac_ring_t *ring;
6961 mac_soft_ring_set_t *srs = MCIP_TX_SRS(mcip);
6962 mac_group_t *defgrp;
6963
6964 defgrp = MAC_DEFAULT_TX_GROUP(mip);
6965 if (srs != NULL) {
6966 if (srs->srs_soft_ring_count > 0) {
6967 for (ring = grp->mrg_rings; ring != NULL;
6968 ring = ring->mr_next) {
6969 ASSERT(mac_tx_srs_ring_present(srs, ring));
6970 mac_tx_invoke_callbacks(mcip,
6971 (mac_tx_cookie_t)
6972 mac_tx_srs_get_soft_ring(srs, ring));
6973 mac_tx_srs_del_ring(srs, ring);
6974 }
6975 } else {
6976 ASSERT(srs->srs_tx.st_arg2 != NULL);
6977 srs->srs_tx.st_arg2 = NULL;
6978 mac_srs_stat_delete(srs);
6979 }
6980 }
6981 if (share != NULL)
6982 mip->mi_share_capab.ms_sremove(share, grp->mrg_driver);
6983
6984 /* move the ring back to the pool */
6985 if (mip->mi_tx_group_type == MAC_GROUP_TYPE_DYNAMIC) {
6986 while ((ring = grp->mrg_rings) != NULL)
6987 (void) mac_group_mov_ring(mip, defgrp, ring);
6988 }
6989 mac_stop_group(grp);
6990 mip->mi_tx_group_free++;
6991 }
6992
6993 /*
6994 * Disassociate a MAC client from a group, i.e go through the rings in the
6995 * group and delete all the soft rings tied to them.
6996 */
6997 static void
6998 mac_tx_dismantle_soft_rings(mac_group_t *fgrp, flow_entry_t *flent)
6999 {
7000 mac_client_impl_t *mcip = flent->fe_mcip;
7001 mac_soft_ring_set_t *tx_srs;
7002 mac_srs_tx_t *tx;
7003 mac_ring_t *ring;
7004
7005 tx_srs = flent->fe_tx_srs;
7006 tx = &tx_srs->srs_tx;
7007
7008 /* Single ring case we haven't created any soft rings */
7009 if (tx->st_mode == SRS_TX_BW || tx->st_mode == SRS_TX_SERIALIZE ||
7010 tx->st_mode == SRS_TX_DEFAULT) {
7011 tx->st_arg2 = NULL;
7012 mac_srs_stat_delete(tx_srs);
7013 /* Fanout case, where we have to dismantle the soft rings */
7014 } else {
7015 for (ring = fgrp->mrg_rings; ring != NULL;
7016 ring = ring->mr_next) {
7017 ASSERT(mac_tx_srs_ring_present(tx_srs, ring));
7018 mac_tx_invoke_callbacks(mcip,
7019 (mac_tx_cookie_t)mac_tx_srs_get_soft_ring(tx_srs,
7020 ring));
7021 mac_tx_srs_del_ring(tx_srs, ring);
7022 }
7023 ASSERT(tx->st_arg2 == NULL);
7024 }
7025 }
7026
7027 /*
7028 * Switch the MAC client from one group to another. This means we need
7029 * to remove the MAC client, teardown the SRSs and revert the group state.
7030 * Then, we add the client to the destination roup, set the SRSs etc.
7031 */
7032 void
7033 mac_tx_switch_group(mac_client_impl_t *mcip, mac_group_t *fgrp,
7034 mac_group_t *tgrp)
7035 {
7036 mac_client_impl_t *group_only_mcip;
7037 mac_impl_t *mip = mcip->mci_mip;
7038 flow_entry_t *flent = mcip->mci_flent;
7039 mac_group_t *defgrp;
7040 mac_grp_client_t *mgcp;
7041 mac_client_impl_t *gmcip;
7042 flow_entry_t *gflent;
7043
7044 defgrp = MAC_DEFAULT_TX_GROUP(mip);
7045 ASSERT(fgrp == flent->fe_tx_ring_group);
7046
7047 if (fgrp == defgrp) {
7048 /*
7049 * If this is the primary we need to find any VLANs on
7050 * the primary and move them too.
7051 */
7052 mac_group_remove_client(fgrp, mcip);
7053 mac_tx_dismantle_soft_rings(fgrp, flent);
7054 if (mcip->mci_unicast->ma_nusers > 1) {
7055 mgcp = fgrp->mrg_clients;
7056 while (mgcp != NULL) {
7057 gmcip = mgcp->mgc_client;
7058 mgcp = mgcp->mgc_next;
7059 if (mcip->mci_unicast != gmcip->mci_unicast)
7060 continue;
7061 mac_tx_client_quiesce(
7062 (mac_client_handle_t)gmcip);
7063
7064 gflent = gmcip->mci_flent;
7065 mac_group_remove_client(fgrp, gmcip);
7066 mac_tx_dismantle_soft_rings(fgrp, gflent);
7067
7068 mac_group_add_client(tgrp, gmcip);
7069 gflent->fe_tx_ring_group = tgrp;
7070 /* We could directly set this to SHARED */
7071 tgrp->mrg_state = mac_group_next_state(tgrp,
7072 &group_only_mcip, defgrp, B_FALSE);
7073
7074 mac_tx_srs_group_setup(gmcip, gflent,
7075 SRST_LINK);
7076 mac_fanout_setup(gmcip, gflent,
7077 MCIP_RESOURCE_PROPS(gmcip), mac_rx_deliver,
7078 gmcip, NULL, NULL);
7079
7080 mac_tx_client_restart(
7081 (mac_client_handle_t)gmcip);
7082 }
7083 }
7084 if (MAC_GROUP_NO_CLIENT(fgrp)) {
7085 mac_ring_t *ring;
7086 int cnt;
7087 int ringcnt;
7088
7089 fgrp->mrg_state = MAC_GROUP_STATE_REGISTERED;
7090 /*
7091 * Additionally, we also need to stop all
7092 * the rings in the default group, except
7093 * the default ring. The reason being
7094 * this group won't be released since it is
7095 * the default group, so the rings won't
7096 * be stopped otherwise.
7097 */
7098 ringcnt = fgrp->mrg_cur_count;
7099 ring = fgrp->mrg_rings;
7100 for (cnt = 0; cnt < ringcnt; cnt++) {
7101 if (ring->mr_state == MR_INUSE &&
7102 ring !=
7103 (mac_ring_t *)mip->mi_default_tx_ring) {
7104 mac_stop_ring(ring);
7105 ring->mr_flag = 0;
7106 }
7107 ring = ring->mr_next;
7108 }
7109 } else if (MAC_GROUP_ONLY_CLIENT(fgrp) != NULL) {
7110 fgrp->mrg_state = MAC_GROUP_STATE_RESERVED;
7111 } else {
7112 ASSERT(fgrp->mrg_state == MAC_GROUP_STATE_SHARED);
7113 }
7114 } else {
7115 /*
7116 * We could have VLANs sharing the non-default group with
7117 * the primary.
7118 */
7119 mgcp = fgrp->mrg_clients;
7120 while (mgcp != NULL) {
7121 gmcip = mgcp->mgc_client;
7122 mgcp = mgcp->mgc_next;
7123 if (gmcip == mcip)
7124 continue;
7125 mac_tx_client_quiesce((mac_client_handle_t)gmcip);
7126 gflent = gmcip->mci_flent;
7127
7128 mac_group_remove_client(fgrp, gmcip);
7129 mac_tx_dismantle_soft_rings(fgrp, gflent);
7130
7131 mac_group_add_client(tgrp, gmcip);
7132 gflent->fe_tx_ring_group = tgrp;
7133 /* We could directly set this to SHARED */
7134 tgrp->mrg_state = mac_group_next_state(tgrp,
7135 &group_only_mcip, defgrp, B_FALSE);
7136 mac_tx_srs_group_setup(gmcip, gflent, SRST_LINK);
7137 mac_fanout_setup(gmcip, gflent,
7138 MCIP_RESOURCE_PROPS(gmcip), mac_rx_deliver,
7139 gmcip, NULL, NULL);
7140
7141 mac_tx_client_restart((mac_client_handle_t)gmcip);
7142 }
7143 mac_group_remove_client(fgrp, mcip);
7144 mac_release_tx_group(mcip, fgrp);
7145 fgrp->mrg_state = MAC_GROUP_STATE_REGISTERED;
7146 }
7147
7148 /* Add it to the tgroup */
7149 mac_group_add_client(tgrp, mcip);
7150 flent->fe_tx_ring_group = tgrp;
7151 tgrp->mrg_state = mac_group_next_state(tgrp, &group_only_mcip,
7152 defgrp, B_FALSE);
7153
7154 mac_tx_srs_group_setup(mcip, flent, SRST_LINK);
7155 mac_fanout_setup(mcip, flent, MCIP_RESOURCE_PROPS(mcip),
7156 mac_rx_deliver, mcip, NULL, NULL);
7157 }
7158
7159 /*
7160 * This is a 1-time control path activity initiated by the client (IP).
7161 * The mac perimeter protects against other simultaneous control activities,
7162 * for example an ioctl that attempts to change the degree of fanout and
7163 * increase or decrease the number of softrings associated with this Tx SRS.
7164 */
7165 static mac_tx_notify_cb_t *
7166 mac_client_tx_notify_add(mac_client_impl_t *mcip,
7167 mac_tx_notify_t notify, void *arg)
7168 {
7169 mac_cb_info_t *mcbi;
7170 mac_tx_notify_cb_t *mtnfp;
7171
7172 ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
7173
7174 mtnfp = kmem_zalloc(sizeof (mac_tx_notify_cb_t), KM_SLEEP);
7175 mtnfp->mtnf_fn = notify;
7176 mtnfp->mtnf_arg = arg;
7177 mtnfp->mtnf_link.mcb_objp = mtnfp;
7178 mtnfp->mtnf_link.mcb_objsize = sizeof (mac_tx_notify_cb_t);
7179 mtnfp->mtnf_link.mcb_flags = MCB_TX_NOTIFY_CB_T;
7180
7181 mcbi = &mcip->mci_tx_notify_cb_info;
7182 mutex_enter(mcbi->mcbi_lockp);
7183 mac_callback_add(mcbi, &mcip->mci_tx_notify_cb_list, &mtnfp->mtnf_link);
7184 mutex_exit(mcbi->mcbi_lockp);
7185 return (mtnfp);
7186 }
7187
7188 static void
7189 mac_client_tx_notify_remove(mac_client_impl_t *mcip, mac_tx_notify_cb_t *mtnfp)
7190 {
7191 mac_cb_info_t *mcbi;
7192 mac_cb_t **cblist;
7193
7194 ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
7195
7196 if (!mac_callback_find(&mcip->mci_tx_notify_cb_info,
7197 &mcip->mci_tx_notify_cb_list, &mtnfp->mtnf_link)) {
7198 cmn_err(CE_WARN,
7199 "mac_client_tx_notify_remove: callback not "
7200 "found, mcip 0x%p mtnfp 0x%p", (void *)mcip, (void *)mtnfp);
7201 return;
7202 }
7203
7204 mcbi = &mcip->mci_tx_notify_cb_info;
7205 cblist = &mcip->mci_tx_notify_cb_list;
7206 mutex_enter(mcbi->mcbi_lockp);
7207 if (mac_callback_remove(mcbi, cblist, &mtnfp->mtnf_link))
7208 kmem_free(mtnfp, sizeof (mac_tx_notify_cb_t));
7209 else
7210 mac_callback_remove_wait(&mcip->mci_tx_notify_cb_info);
7211 mutex_exit(mcbi->mcbi_lockp);
7212 }
7213
7214 /*
7215 * mac_client_tx_notify():
7216 * call to add and remove flow control callback routine.
7217 */
7218 mac_tx_notify_handle_t
7219 mac_client_tx_notify(mac_client_handle_t mch, mac_tx_notify_t callb_func,
7220 void *ptr)
7221 {
7222 mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
7223 mac_tx_notify_cb_t *mtnfp = NULL;
7224
7225 i_mac_perim_enter(mcip->mci_mip);
7226
7227 if (callb_func != NULL) {
7228 /* Add a notify callback */
7229 mtnfp = mac_client_tx_notify_add(mcip, callb_func, ptr);
7230 } else {
7231 mac_client_tx_notify_remove(mcip, (mac_tx_notify_cb_t *)ptr);
7232 }
7233 i_mac_perim_exit(mcip->mci_mip);
7234
7235 return ((mac_tx_notify_handle_t)mtnfp);
7236 }
7237
7238 void
7239 mac_bridge_vectors(mac_bridge_tx_t txf, mac_bridge_rx_t rxf,
7240 mac_bridge_ref_t reff, mac_bridge_ls_t lsf)
7241 {
7242 mac_bridge_tx_cb = txf;
7243 mac_bridge_rx_cb = rxf;
7244 mac_bridge_ref_cb = reff;
7245 mac_bridge_ls_cb = lsf;
7246 }
7247
7248 int
7249 mac_bridge_set(mac_handle_t mh, mac_handle_t link)
7250 {
7251 mac_impl_t *mip = (mac_impl_t *)mh;
7252 int retv;
7253
7254 mutex_enter(&mip->mi_bridge_lock);
7255 if (mip->mi_bridge_link == NULL) {
7256 mip->mi_bridge_link = link;
7257 retv = 0;
7258 } else {
7259 retv = EBUSY;
7260 }
7261 mutex_exit(&mip->mi_bridge_lock);
7262 if (retv == 0) {
7263 mac_poll_state_change(mh, B_FALSE);
7264 mac_capab_update(mh);
7265 }
7266 return (retv);
7267 }
7268
7269 /*
7270 * Disable bridging on the indicated link.
7271 */
7272 void
7273 mac_bridge_clear(mac_handle_t mh, mac_handle_t link)
7274 {
7275 mac_impl_t *mip = (mac_impl_t *)mh;
7276
7277 mutex_enter(&mip->mi_bridge_lock);
7278 ASSERT(mip->mi_bridge_link == link);
7279 mip->mi_bridge_link = NULL;
7280 mutex_exit(&mip->mi_bridge_lock);
7281 mac_poll_state_change(mh, B_TRUE);
7282 mac_capab_update(mh);
7283 }
7284
7285 void
7286 mac_no_active(mac_handle_t mh)
7287 {
7288 mac_impl_t *mip = (mac_impl_t *)mh;
7289
7290 i_mac_perim_enter(mip);
7291 mip->mi_state_flags |= MIS_NO_ACTIVE;
7292 i_mac_perim_exit(mip);
7293 }
7294
7295 /*
7296 * Walk the primary VLAN clients whenever the primary's rings property
7297 * changes and update the mac_resource_props_t for the VLAN's client.
7298 * We need to do this since we don't support setting these properties
7299 * on the primary's VLAN clients, but the VLAN clients have to
7300 * follow the primary w.r.t the rings property;
7301 */
7302 void
7303 mac_set_prim_vlan_rings(mac_impl_t *mip, mac_resource_props_t *mrp)
7304 {
7305 mac_client_impl_t *vmcip;
7306 mac_resource_props_t *vmrp;
7307
7308 for (vmcip = mip->mi_clients_list; vmcip != NULL;
7309 vmcip = vmcip->mci_client_next) {
7310 if (!(vmcip->mci_flent->fe_type & FLOW_PRIMARY_MAC) ||
7311 mac_client_vid((mac_client_handle_t)vmcip) ==
7312 VLAN_ID_NONE) {
7313 continue;
7314 }
7315 vmrp = MCIP_RESOURCE_PROPS(vmcip);
7316
7317 vmrp->mrp_nrxrings = mrp->mrp_nrxrings;
7318 if (mrp->mrp_mask & MRP_RX_RINGS)
7319 vmrp->mrp_mask |= MRP_RX_RINGS;
7320 else if (vmrp->mrp_mask & MRP_RX_RINGS)
7321 vmrp->mrp_mask &= ~MRP_RX_RINGS;
7322
7323 vmrp->mrp_ntxrings = mrp->mrp_ntxrings;
7324 if (mrp->mrp_mask & MRP_TX_RINGS)
7325 vmrp->mrp_mask |= MRP_TX_RINGS;
7326 else if (vmrp->mrp_mask & MRP_TX_RINGS)
7327 vmrp->mrp_mask &= ~MRP_TX_RINGS;
7328
7329 if (mrp->mrp_mask & MRP_RXRINGS_UNSPEC)
7330 vmrp->mrp_mask |= MRP_RXRINGS_UNSPEC;
7331 else
7332 vmrp->mrp_mask &= ~MRP_RXRINGS_UNSPEC;
7333
7334 if (mrp->mrp_mask & MRP_TXRINGS_UNSPEC)
7335 vmrp->mrp_mask |= MRP_TXRINGS_UNSPEC;
7336 else
7337 vmrp->mrp_mask &= ~MRP_TXRINGS_UNSPEC;
7338 }
7339 }
7340
7341 /*
7342 * We are adding or removing ring(s) from a group. The source for taking
7343 * rings is the default group. The destination for giving rings back is
7344 * the default group.
7345 */
7346 int
7347 mac_group_ring_modify(mac_client_impl_t *mcip, mac_group_t *group,
7348 mac_group_t *defgrp)
7349 {
7350 mac_resource_props_t *mrp = MCIP_RESOURCE_PROPS(mcip);
7351 uint_t modify;
7352 int count;
7353 mac_ring_t *ring;
7354 mac_ring_t *next;
7355 mac_impl_t *mip = mcip->mci_mip;
7356 mac_ring_t **rings;
7357 uint_t ringcnt;
7358 int i = 0;
7359 boolean_t rx_group = group->mrg_type == MAC_RING_TYPE_RX;
7360 int start;
7361 int end;
7362 mac_group_t *tgrp;
7363 int j;
7364 int rv = 0;
7365
7366 /*
7367 * If we are asked for just a group, we give 1 ring, else
7368 * the specified number of rings.
7369 */
7370 if (rx_group) {
7371 ringcnt = (mrp->mrp_mask & MRP_RXRINGS_UNSPEC) ? 1:
7372 mrp->mrp_nrxrings;
7373 } else {
7374 ringcnt = (mrp->mrp_mask & MRP_TXRINGS_UNSPEC) ? 1:
7375 mrp->mrp_ntxrings;
7376 }
7377
7378 /* don't allow modifying rings for a share for now. */
7379 ASSERT(mcip->mci_share == NULL);
7380
7381 if (ringcnt == group->mrg_cur_count)
7382 return (0);
7383
7384 if (group->mrg_cur_count > ringcnt) {
7385 modify = group->mrg_cur_count - ringcnt;
7386 if (rx_group) {
7387 if (mip->mi_rx_donor_grp == group) {
7388 ASSERT(mac_is_primary_client(mcip));
7389 mip->mi_rx_donor_grp = defgrp;
7390 } else {
7391 defgrp = mip->mi_rx_donor_grp;
7392 }
7393 }
7394 ring = group->mrg_rings;
7395 rings = kmem_alloc(modify * sizeof (mac_ring_handle_t),
7396 KM_SLEEP);
7397 j = 0;
7398 for (count = 0; count < modify; count++) {
7399 next = ring->mr_next;
7400 rv = mac_group_mov_ring(mip, defgrp, ring);
7401 if (rv != 0) {
7402 /* cleanup on failure */
7403 for (j = 0; j < count; j++) {
7404 (void) mac_group_mov_ring(mip, group,
7405 rings[j]);
7406 }
7407 break;
7408 }
7409 rings[j++] = ring;
7410 ring = next;
7411 }
7412 kmem_free(rings, modify * sizeof (mac_ring_handle_t));
7413 return (rv);
7414 }
7415 if (ringcnt >= MAX_RINGS_PER_GROUP)
7416 return (EINVAL);
7417
7418 modify = ringcnt - group->mrg_cur_count;
7419
7420 if (rx_group) {
7421 if (group != mip->mi_rx_donor_grp)
7422 defgrp = mip->mi_rx_donor_grp;
7423 else
7424 /*
7425 * This is the donor group with all the remaining
7426 * rings. Default group now gets to be the donor
7427 */
7428 mip->mi_rx_donor_grp = defgrp;
7429 start = 1;
7430 end = mip->mi_rx_group_count;
7431 } else {
7432 start = 0;
7433 end = mip->mi_tx_group_count - 1;
7434 }
7435 /*
7436 * If the default doesn't have any rings, lets see if we can
7437 * take rings given to an h/w client that doesn't need it.
7438 * For now, we just see if there is any one client that can donate
7439 * all the required rings.
7440 */
7441 if (defgrp->mrg_cur_count < (modify + 1)) {
7442 for (i = start; i < end; i++) {
7443 if (rx_group) {
7444 tgrp = &mip->mi_rx_groups[i];
7445 if (tgrp == group || tgrp->mrg_state <
7446 MAC_GROUP_STATE_RESERVED) {
7447 continue;
7448 }
7449 mcip = MAC_GROUP_ONLY_CLIENT(tgrp);
7450 if (mcip == NULL)
7451 mcip = mac_get_grp_primary(tgrp);
7452 ASSERT(mcip != NULL);
7453 mrp = MCIP_RESOURCE_PROPS(mcip);
7454 if ((mrp->mrp_mask & MRP_RX_RINGS) != 0)
7455 continue;
7456 if ((tgrp->mrg_cur_count +
7457 defgrp->mrg_cur_count) < (modify + 1)) {
7458 continue;
7459 }
7460 if (mac_rx_switch_group(mcip, tgrp,
7461 defgrp) != 0) {
7462 return (ENOSPC);
7463 }
7464 } else {
7465 tgrp = &mip->mi_tx_groups[i];
7466 if (tgrp == group || tgrp->mrg_state <
7467 MAC_GROUP_STATE_RESERVED) {
7468 continue;
7469 }
7470 mcip = MAC_GROUP_ONLY_CLIENT(tgrp);
7471 if (mcip == NULL)
7472 mcip = mac_get_grp_primary(tgrp);
7473 mrp = MCIP_RESOURCE_PROPS(mcip);
7474 if ((mrp->mrp_mask & MRP_TX_RINGS) != 0)
7475 continue;
7476 if ((tgrp->mrg_cur_count +
7477 defgrp->mrg_cur_count) < (modify + 1)) {
7478 continue;
7479 }
7480 /* OK, we can switch this to s/w */
7481 mac_tx_client_quiesce(
7482 (mac_client_handle_t)mcip);
7483 mac_tx_switch_group(mcip, tgrp, defgrp);
7484 mac_tx_client_restart(
7485 (mac_client_handle_t)mcip);
7486 }
7487 }
7488 if (defgrp->mrg_cur_count < (modify + 1))
7489 return (ENOSPC);
7490 }
7491 if ((rv = i_mac_group_allocate_rings(mip, group->mrg_type, defgrp,
7492 group, mcip->mci_share, modify)) != 0) {
7493 return (rv);
7494 }
7495 return (0);
7496 }
7497
7498 /*
7499 * Given the poolname in mac_resource_props, find the cpupart
7500 * that is associated with this pool. The cpupart will be used
7501 * later for finding the cpus to be bound to the networking threads.
7502 *
7503 * use_default is set B_TRUE if pools are enabled and pool_default
7504 * is returned. This avoids a 2nd lookup to set the poolname
7505 * for pool-effective.
7506 *
7507 * returns:
7508 *
7509 * NULL - pools are disabled or if the 'cpus' property is set.
7510 * cpupart of pool_default - pools are enabled and the pool
7511 * is not available or poolname is blank
7512 * cpupart of named pool - pools are enabled and the pool
7513 * is available.
7514 */
7515 cpupart_t *
7516 mac_pset_find(mac_resource_props_t *mrp, boolean_t *use_default)
7517 {
7518 pool_t *pool;
7519 cpupart_t *cpupart;
7520
7521 *use_default = B_FALSE;
7522
7523 /* CPUs property is set */
7524 if (mrp->mrp_mask & MRP_CPUS)
7525 return (NULL);
7526
7527 ASSERT(pool_lock_held());
7528
7529 /* Pools are disabled, no pset */
7530 if (pool_state == POOL_DISABLED)
7531 return (NULL);
7532
7533 /* Pools property is set */
7534 if (mrp->mrp_mask & MRP_POOL) {
7535 if ((pool = pool_lookup_pool_by_name(mrp->mrp_pool)) == NULL) {
7536 /* Pool not found */
7537 DTRACE_PROBE1(mac_pset_find_no_pool, char *,
7538 mrp->mrp_pool);
7539 *use_default = B_TRUE;
7540 pool = pool_default;
7541 }
7542 /* Pools property is not set */
7543 } else {
7544 *use_default = B_TRUE;
7545 pool = pool_default;
7546 }
7547
7548 /* Find the CPU pset that corresponds to the pool */
7549 mutex_enter(&cpu_lock);
7550 if ((cpupart = cpupart_find(pool->pool_pset->pset_id)) == NULL) {
7551 DTRACE_PROBE1(mac_find_pset_no_pset, psetid_t,
7552 pool->pool_pset->pset_id);
7553 }
7554 mutex_exit(&cpu_lock);
7555
7556 return (cpupart);
7557 }
7558
7559 void
7560 mac_set_pool_effective(boolean_t use_default, cpupart_t *cpupart,
7561 mac_resource_props_t *mrp, mac_resource_props_t *emrp)
7562 {
7563 ASSERT(pool_lock_held());
7564
7565 if (cpupart != NULL) {
7566 emrp->mrp_mask |= MRP_POOL;
7567 if (use_default) {
7568 (void) strcpy(emrp->mrp_pool,
7569 "pool_default");
7570 } else {
7571 ASSERT(strlen(mrp->mrp_pool) != 0);
7572 (void) strcpy(emrp->mrp_pool,
7573 mrp->mrp_pool);
7574 }
7575 } else {
7576 emrp->mrp_mask &= ~MRP_POOL;
7577 bzero(emrp->mrp_pool, MAXPATHLEN);
7578 }
7579 }
7580
7581 struct mac_pool_arg {
7582 char mpa_poolname[MAXPATHLEN];
7583 pool_event_t mpa_what;
7584 };
7585
7586 /*ARGSUSED*/
7587 static uint_t
7588 mac_pool_link_update(mod_hash_key_t key, mod_hash_val_t *val, void *arg)
7589 {
7590 struct mac_pool_arg *mpa = arg;
7591 mac_impl_t *mip = (mac_impl_t *)val;
7592 mac_client_impl_t *mcip;
7593 mac_resource_props_t *mrp, *emrp;
7594 boolean_t pool_update = B_FALSE;
7595 boolean_t pool_clear = B_FALSE;
7596 boolean_t use_default = B_FALSE;
7597 cpupart_t *cpupart = NULL;
7598
7599 mrp = kmem_zalloc(sizeof (*mrp), KM_SLEEP);
7600 i_mac_perim_enter(mip);
7601 for (mcip = mip->mi_clients_list; mcip != NULL;
7602 mcip = mcip->mci_client_next) {
7603 pool_update = B_FALSE;
7604 pool_clear = B_FALSE;
7605 use_default = B_FALSE;
7606 mac_client_get_resources((mac_client_handle_t)mcip, mrp);
7607 emrp = MCIP_EFFECTIVE_PROPS(mcip);
7608
7609 /*
7610 * When pools are enabled
7611 */
7612 if ((mpa->mpa_what == POOL_E_ENABLE) &&
7613 ((mrp->mrp_mask & MRP_CPUS) == 0)) {
7614 mrp->mrp_mask |= MRP_POOL;
7615 pool_update = B_TRUE;
7616 }
7617
7618 /*
7619 * When pools are disabled
7620 */
7621 if ((mpa->mpa_what == POOL_E_DISABLE) &&
7622 ((mrp->mrp_mask & MRP_CPUS) == 0)) {
7623 mrp->mrp_mask |= MRP_POOL;
7624 pool_clear = B_TRUE;
7625 }
7626
7627 /*
7628 * Look for links with the pool property set and the poolname
7629 * matching the one which is changing.
7630 */
7631 if (strcmp(mrp->mrp_pool, mpa->mpa_poolname) == 0) {
7632 /*
7633 * The pool associated with the link has changed.
7634 */
7635 if (mpa->mpa_what == POOL_E_CHANGE) {
7636 mrp->mrp_mask |= MRP_POOL;
7637 pool_update = B_TRUE;
7638 }
7639 }
7640
7641 /*
7642 * This link is associated with pool_default and
7643 * pool_default has changed.
7644 */
7645 if ((mpa->mpa_what == POOL_E_CHANGE) &&
7646 (strcmp(emrp->mrp_pool, "pool_default") == 0) &&
7647 (strcmp(mpa->mpa_poolname, "pool_default") == 0)) {
7648 mrp->mrp_mask |= MRP_POOL;
7649 pool_update = B_TRUE;
7650 }
7651
7652 /*
7653 * Get new list of cpus for the pool, bind network
7654 * threads to new list of cpus and update resources.
7655 */
7656 if (pool_update) {
7657 if (MCIP_DATAPATH_SETUP(mcip)) {
7658 pool_lock();
7659 cpupart = mac_pset_find(mrp, &use_default);
7660 mac_fanout_setup(mcip, mcip->mci_flent, mrp,
7661 mac_rx_deliver, mcip, NULL, cpupart);
7662 mac_set_pool_effective(use_default, cpupart,
7663 mrp, emrp);
7664 pool_unlock();
7665 }
7666 mac_update_resources(mrp, MCIP_RESOURCE_PROPS(mcip),
7667 B_FALSE);
7668 }
7669
7670 /*
7671 * Clear the effective pool and bind network threads
7672 * to any available CPU.
7673 */
7674 if (pool_clear) {
7675 if (MCIP_DATAPATH_SETUP(mcip)) {
7676 emrp->mrp_mask &= ~MRP_POOL;
7677 bzero(emrp->mrp_pool, MAXPATHLEN);
7678 mac_fanout_setup(mcip, mcip->mci_flent, mrp,
7679 mac_rx_deliver, mcip, NULL, NULL);
7680 }
7681 mac_update_resources(mrp, MCIP_RESOURCE_PROPS(mcip),
7682 B_FALSE);
7683 }
7684 }
7685 i_mac_perim_exit(mip);
7686 kmem_free(mrp, sizeof (*mrp));
7687 return (MH_WALK_CONTINUE);
7688 }
7689
7690 static void
7691 mac_pool_update(void *arg)
7692 {
7693 mod_hash_walk(i_mac_impl_hash, mac_pool_link_update, arg);
7694 kmem_free(arg, sizeof (struct mac_pool_arg));
7695 }
7696
7697 /*
7698 * Callback function to be executed when a noteworthy pool event
7699 * takes place.
7700 */
7701 /* ARGSUSED */
7702 static void
7703 mac_pool_event_cb(pool_event_t what, poolid_t id, void *arg)
7704 {
7705 pool_t *pool;
7706 char *poolname = NULL;
7707 struct mac_pool_arg *mpa;
7708
7709 pool_lock();
7710 mpa = kmem_zalloc(sizeof (struct mac_pool_arg), KM_SLEEP);
7711
7712 switch (what) {
7713 case POOL_E_ENABLE:
7714 case POOL_E_DISABLE:
7715 break;
7716
7717 case POOL_E_CHANGE:
7718 pool = pool_lookup_pool_by_id(id);
7719 if (pool == NULL) {
7720 kmem_free(mpa, sizeof (struct mac_pool_arg));
7721 pool_unlock();
7722 return;
7723 }
7724 pool_get_name(pool, &poolname);
7725 (void) strlcpy(mpa->mpa_poolname, poolname,
7726 sizeof (mpa->mpa_poolname));
7727 break;
7728
7729 default:
7730 kmem_free(mpa, sizeof (struct mac_pool_arg));
7731 pool_unlock();
7732 return;
7733 }
7734 pool_unlock();
7735
7736 mpa->mpa_what = what;
7737
7738 mac_pool_update(mpa);
7739 }
7740
7741 /*
7742 * Set effective rings property. This could be called from datapath_setup/
7743 * datapath_teardown or set-linkprop.
7744 * If the group is reserved we just go ahead and set the effective rings.
7745 * Additionally, for TX this could mean the default group has lost/gained
7746 * some rings, so if the default group is reserved, we need to adjust the
7747 * effective rings for the default group clients. For RX, if we are working
7748 * with the non-default group, we just need * to reset the effective props
7749 * for the default group clients.
7750 */
7751 void
7752 mac_set_rings_effective(mac_client_impl_t *mcip)
7753 {
7754 mac_impl_t *mip = mcip->mci_mip;
7755 mac_group_t *grp;
7756 mac_group_t *defgrp;
7757 flow_entry_t *flent = mcip->mci_flent;
7758 mac_resource_props_t *emrp = MCIP_EFFECTIVE_PROPS(mcip);
7759 mac_grp_client_t *mgcp;
7760 mac_client_impl_t *gmcip;
7761
7762 grp = flent->fe_rx_ring_group;
7763 if (grp != NULL) {
7764 defgrp = MAC_DEFAULT_RX_GROUP(mip);
7765 /*
7766 * If we have reserved a group, set the effective rings
7767 * to the ring count in the group.
7768 */
7769 if (grp->mrg_state == MAC_GROUP_STATE_RESERVED) {
7770 emrp->mrp_mask |= MRP_RX_RINGS;
7771 emrp->mrp_nrxrings = grp->mrg_cur_count;
7772 }
7773
7774 /*
7775 * We go through the clients in the shared group and
7776 * reset the effective properties. It is possible this
7777 * might have already been done for some client (i.e.
7778 * if some client is being moved to a group that is
7779 * already shared). The case where the default group is
7780 * RESERVED is taken care of above (note in the RX side if
7781 * there is a non-default group, the default group is always
7782 * SHARED).
7783 */
7784 if (grp != defgrp || grp->mrg_state == MAC_GROUP_STATE_SHARED) {
7785 if (grp->mrg_state == MAC_GROUP_STATE_SHARED)
7786 mgcp = grp->mrg_clients;
7787 else
7788 mgcp = defgrp->mrg_clients;
7789 while (mgcp != NULL) {
7790 gmcip = mgcp->mgc_client;
7791 emrp = MCIP_EFFECTIVE_PROPS(gmcip);
7792 if (emrp->mrp_mask & MRP_RX_RINGS) {
7793 emrp->mrp_mask &= ~MRP_RX_RINGS;
7794 emrp->mrp_nrxrings = 0;
7795 }
7796 mgcp = mgcp->mgc_next;
7797 }
7798 }
7799 }
7800
7801 /* Now the TX side */
7802 grp = flent->fe_tx_ring_group;
7803 if (grp != NULL) {
7804 defgrp = MAC_DEFAULT_TX_GROUP(mip);
7805
7806 if (grp->mrg_state == MAC_GROUP_STATE_RESERVED) {
7807 emrp->mrp_mask |= MRP_TX_RINGS;
7808 emrp->mrp_ntxrings = grp->mrg_cur_count;
7809 } else if (grp->mrg_state == MAC_GROUP_STATE_SHARED) {
7810 mgcp = grp->mrg_clients;
7811 while (mgcp != NULL) {
7812 gmcip = mgcp->mgc_client;
7813 emrp = MCIP_EFFECTIVE_PROPS(gmcip);
7814 if (emrp->mrp_mask & MRP_TX_RINGS) {
7815 emrp->mrp_mask &= ~MRP_TX_RINGS;
7816 emrp->mrp_ntxrings = 0;
7817 }
7818 mgcp = mgcp->mgc_next;
7819 }
7820 }
7821
7822 /*
7823 * If the group is not the default group and the default
7824 * group is reserved, the ring count in the default group
7825 * might have changed, update it.
7826 */
7827 if (grp != defgrp &&
7828 defgrp->mrg_state == MAC_GROUP_STATE_RESERVED) {
7829 gmcip = MAC_GROUP_ONLY_CLIENT(defgrp);
7830 emrp = MCIP_EFFECTIVE_PROPS(gmcip);
7831 emrp->mrp_ntxrings = defgrp->mrg_cur_count;
7832 }
7833 }
7834 emrp = MCIP_EFFECTIVE_PROPS(mcip);
7835 }
7836
7837 /*
7838 * Check if the primary is in the default group. If so, see if we
7839 * can give it a an exclusive group now that another client is
7840 * being configured. We take the primary out of the default group
7841 * because the multicast/broadcast packets for the all the clients
7842 * will land in the default ring in the default group which means
7843 * any client in the default group, even if it is the only on in
7844 * the group, will lose exclusive access to the rings, hence
7845 * polling.
7846 */
7847 mac_client_impl_t *
7848 mac_check_primary_relocation(mac_client_impl_t *mcip, boolean_t rxhw)
7849 {
7850 mac_impl_t *mip = mcip->mci_mip;
7851 mac_group_t *defgrp = MAC_DEFAULT_RX_GROUP(mip);
7852 flow_entry_t *flent = mcip->mci_flent;
7853 mac_resource_props_t *mrp = MCIP_RESOURCE_PROPS(mcip);
7854 uint8_t *mac_addr;
7855 mac_group_t *ngrp;
7856
7857 /*
7858 * Check if the primary is in the default group, if not
7859 * or if it is explicitly configured to be in the default
7860 * group OR set the RX rings property, return.
7861 */
7862 if (flent->fe_rx_ring_group != defgrp || mrp->mrp_mask & MRP_RX_RINGS)
7863 return (NULL);
7864
7865 /*
7866 * If the new client needs an exclusive group and we
7867 * don't have another for the primary, return.
7868 */
7869 if (rxhw && mip->mi_rxhwclnt_avail < 2)
7870 return (NULL);
7871
7872 mac_addr = flent->fe_flow_desc.fd_dst_mac;
7873 /*
7874 * We call this when we are setting up the datapath for
7875 * the first non-primary.
7876 */
7877 ASSERT(mip->mi_nactiveclients == 2);
7878 /*
7879 * OK, now we have the primary that needs to be relocated.
7880 */
7881 ngrp = mac_reserve_rx_group(mcip, mac_addr, B_TRUE);
7882 if (ngrp == NULL)
7883 return (NULL);
7884 if (mac_rx_switch_group(mcip, defgrp, ngrp) != 0) {
7885 mac_stop_group(ngrp);
7886 return (NULL);
7887 }
7888 return (mcip);
7889 }