xref: /onnv-gate/usr/src/uts/common/io/mac/mac.c (revision 8400:773afdd54b44)
10Sstevel@tonic-gate /*
20Sstevel@tonic-gate  * CDDL HEADER START
30Sstevel@tonic-gate  *
40Sstevel@tonic-gate  * The contents of this file are subject to the terms of the
51852Syz147064  * Common Development and Distribution License (the "License").
61852Syz147064  * You may not use this file except in compliance with the License.
70Sstevel@tonic-gate  *
80Sstevel@tonic-gate  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
90Sstevel@tonic-gate  * or http://www.opensolaris.org/os/licensing.
100Sstevel@tonic-gate  * See the License for the specific language governing permissions
110Sstevel@tonic-gate  * and limitations under the License.
120Sstevel@tonic-gate  *
130Sstevel@tonic-gate  * When distributing Covered Code, include this CDDL HEADER in each
140Sstevel@tonic-gate  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
150Sstevel@tonic-gate  * If applicable, add the following below this CDDL HEADER, with the
160Sstevel@tonic-gate  * fields enclosed by brackets "[]" replaced with your own identifying
170Sstevel@tonic-gate  * information: Portions Copyright [yyyy] [name of copyright owner]
180Sstevel@tonic-gate  *
190Sstevel@tonic-gate  * CDDL HEADER END
200Sstevel@tonic-gate  */
215084Sjohnlev 
220Sstevel@tonic-gate /*
235895Syz147064  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
240Sstevel@tonic-gate  * Use is subject to license terms.
250Sstevel@tonic-gate  */
260Sstevel@tonic-gate 
270Sstevel@tonic-gate /*
280Sstevel@tonic-gate  * MAC Services Module
298275SEric Cheng  *
308275SEric Cheng  * The GLDv3 framework locking -  The MAC layer
318275SEric Cheng  * --------------------------------------------
328275SEric Cheng  *
338275SEric Cheng  * The MAC layer is central to the GLD framework and can provide the locking
348275SEric Cheng  * framework needed for itself and for the use of MAC clients. MAC end points
358275SEric Cheng  * are fairly disjoint and don't share a lot of state. So a coarse grained
368275SEric Cheng  * multi-threading scheme is to single thread all create/modify/delete or set
378275SEric Cheng  * type of control operations on a per mac end point while allowing data threads
388275SEric Cheng  * concurrently.
398275SEric Cheng  *
408275SEric Cheng  * Control operations (set) that modify a mac end point are always serialized on
418275SEric Cheng  * a per mac end point basis, We have at most 1 such thread per mac end point
428275SEric Cheng  * at a time.
438275SEric Cheng  *
448275SEric Cheng  * All other operations that are not serialized are essentially multi-threaded.
458275SEric Cheng  * For example a control operation (get) like getting statistics which may not
468275SEric Cheng  * care about reading values atomically or data threads sending or receiving
478275SEric Cheng  * data. Mostly these type of operations don't modify the control state. Any
488275SEric Cheng  * state these operations care about are protected using traditional locks.
498275SEric Cheng  *
508275SEric Cheng  * The perimeter only serializes serial operations. It does not imply there
518275SEric Cheng  * aren't any other concurrent operations. However a serialized operation may
528275SEric Cheng  * sometimes need to make sure it is the only thread. In this case it needs
538275SEric Cheng  * to use reference counting mechanisms to cv_wait until any current data
548275SEric Cheng  * threads are done.
558275SEric Cheng  *
568275SEric Cheng  * The mac layer itself does not hold any locks across a call to another layer.
578275SEric Cheng  * The perimeter is however held across a down call to the driver to make the
588275SEric Cheng  * whole control operation atomic with respect to other control operations.
598275SEric Cheng  * Also the data path and get type control operations may proceed concurrently.
608275SEric Cheng  * These operations synchronize with the single serial operation on a given mac
618275SEric Cheng  * end point using regular locks. The perimeter ensures that conflicting
628275SEric Cheng  * operations like say a mac_multicast_add and a mac_multicast_remove on the
638275SEric Cheng  * same mac end point don't interfere with each other and also ensures that the
648275SEric Cheng  * changes in the mac layer and the call to the underlying driver to say add a
658275SEric Cheng  * multicast address are done atomically without interference from a thread
668275SEric Cheng  * trying to delete the same address.
678275SEric Cheng  *
688275SEric Cheng  * For example, consider
698275SEric Cheng  * mac_multicst_add()
708275SEric Cheng  * {
718275SEric Cheng  *	mac_perimeter_enter();	serialize all control operations
728275SEric Cheng  *
738275SEric Cheng  *	grab list lock		protect against access by data threads
748275SEric Cheng  *	add to list
758275SEric Cheng  *	drop list lock
768275SEric Cheng  *
778275SEric Cheng  *	call driver's mi_multicst
788275SEric Cheng  *
798275SEric Cheng  *	mac_perimeter_exit();
808275SEric Cheng  * }
818275SEric Cheng  *
828275SEric Cheng  * To lessen the number of serialization locks and simplify the lock hierarchy,
838275SEric Cheng  * we serialize all the control operations on a per mac end point by using a
848275SEric Cheng  * single serialization lock called the perimeter. We allow recursive entry into
858275SEric Cheng  * the perimeter to facilitate use of this mechanism by both the mac client and
868275SEric Cheng  * the MAC layer itself.
878275SEric Cheng  *
888275SEric Cheng  * MAC client means an entity that does an operation on a mac handle
898275SEric Cheng  * obtained from a mac_open/mac_client_open. Similarly MAC driver means
908275SEric Cheng  * an entity that does an operation on a mac handle obtained from a
918275SEric Cheng  * mac_register. An entity could be both client and driver but on different
928275SEric Cheng  * handles eg. aggr. and should only make the corresponding mac interface calls
938275SEric Cheng  * i.e. mac driver interface or mac client interface as appropriate for that
948275SEric Cheng  * mac handle.
958275SEric Cheng  *
968275SEric Cheng  * General rules.
978275SEric Cheng  * -------------
988275SEric Cheng  *
998275SEric Cheng  * R1. The lock order of upcall threads is natually opposite to downcall
1008275SEric Cheng  * threads. Hence upcalls must not hold any locks across layers for fear of
1018275SEric Cheng  * recursive lock enter and lock order violation. This applies to all layers.
1028275SEric Cheng  *
1038275SEric Cheng  * R2. The perimeter is just another lock. Since it is held in the down
1048275SEric Cheng  * direction, acquiring the perimeter in an upcall is prohibited as it would
1058275SEric Cheng  * cause a deadlock. This applies to all layers.
1068275SEric Cheng  *
1078275SEric Cheng  * Note that upcalls that need to grab the mac perimeter (for example
1088275SEric Cheng  * mac_notify upcalls) can still achieve that by posting the request to a
1098275SEric Cheng  * thread, which can then grab all the required perimeters and locks in the
1108275SEric Cheng  * right global order. Note that in the above example the mac layer iself
1118275SEric Cheng  * won't grab the mac perimeter in the mac_notify upcall, instead the upcall
1128275SEric Cheng  * to the client must do that. Please see the aggr code for an example.
1138275SEric Cheng  *
1148275SEric Cheng  * MAC client rules
1158275SEric Cheng  * ----------------
1168275SEric Cheng  *
1178275SEric Cheng  * R3. A MAC client may use the MAC provided perimeter facility to serialize
1188275SEric Cheng  * control operations on a per mac end point. It does this by by acquring
1198275SEric Cheng  * and holding the perimeter across a sequence of calls to the mac layer.
1208275SEric Cheng  * This ensures atomicity across the entire block of mac calls. In this
1218275SEric Cheng  * model the MAC client must not hold any client locks across the calls to
1228275SEric Cheng  * the mac layer. This model is the preferred solution.
1238275SEric Cheng  *
1248275SEric Cheng  * R4. However if a MAC client has a lot of global state across all mac end
1258275SEric Cheng  * points the per mac end point serialization may not be sufficient. In this
1268275SEric Cheng  * case the client may choose to use global locks or use its own serialization.
1278275SEric Cheng  * To avoid deadlocks, these client layer locks held across the mac calls
1288275SEric Cheng  * in the control path must never be acquired by the data path for the reason
1298275SEric Cheng  * mentioned below.
1308275SEric Cheng  *
1318275SEric Cheng  * (Assume that a control operation that holds a client lock blocks in the
1328275SEric Cheng  * mac layer waiting for upcall reference counts to drop to zero. If an upcall
1338275SEric Cheng  * data thread that holds this reference count, tries to acquire the same
1348275SEric Cheng  * client lock subsequently it will deadlock).
1358275SEric Cheng  *
1368275SEric Cheng  * A MAC client may follow either the R3 model or the R4 model, but can't
1378275SEric Cheng  * mix both. In the former, the hierarchy is Perim -> client locks, but in
1388275SEric Cheng  * the latter it is client locks -> Perim.
1398275SEric Cheng  *
1408275SEric Cheng  * R5. MAC clients must make MAC calls (excluding data calls) in a cv_wait'able
1418275SEric Cheng  * context since they may block while trying to acquire the perimeter.
1428275SEric Cheng  * In addition some calls may block waiting for upcall refcnts to come down to
1438275SEric Cheng  * zero.
1448275SEric Cheng  *
1458275SEric Cheng  * R6. MAC clients must make sure that they are single threaded and all threads
1468275SEric Cheng  * from the top (in particular data threads) have finished before calling
1478275SEric Cheng  * mac_client_close. The MAC framework does not track the number of client
1488275SEric Cheng  * threads using the mac client handle. Also mac clients must make sure
1498275SEric Cheng  * they have undone all the control operations before calling mac_client_close.
1508275SEric Cheng  * For example mac_unicast_remove/mac_multicast_remove to undo the corresponding
1518275SEric Cheng  * mac_unicast_add/mac_multicast_add.
1528275SEric Cheng  *
1538275SEric Cheng  * MAC framework rules
1548275SEric Cheng  * -------------------
1558275SEric Cheng  *
1568275SEric Cheng  * R7. The mac layer itself must not hold any mac layer locks (except the mac
1578275SEric Cheng  * perimeter) across a call to any other layer from the mac layer. The call to
1588275SEric Cheng  * any other layer could be via mi_* entry points, classifier entry points into
1598275SEric Cheng  * the driver or via upcall pointers into layers above. The mac perimeter may
1608275SEric Cheng  * be acquired or held only in the down direction, for e.g. when calling into
1618275SEric Cheng  * a mi_* driver enty point to provide atomicity of the operation.
1628275SEric Cheng  *
1638275SEric Cheng  * R8. Since it is not guaranteed (see R14) that drivers won't hold locks across
1648275SEric Cheng  * mac driver interfaces, the MAC layer must provide a cut out for control
1658275SEric Cheng  * interfaces like upcall notifications and start them in a separate thread.
1668275SEric Cheng  *
1678275SEric Cheng  * R9. Note that locking order also implies a plumbing order. For example
1688275SEric Cheng  * VNICs are allowed to be created over aggrs, but not vice-versa. An attempt
1698275SEric Cheng  * to plumb in any other order must be failed at mac_open time, otherwise it
1708275SEric Cheng  * could lead to deadlocks due to inverse locking order.
1718275SEric Cheng  *
1728275SEric Cheng  * R10. MAC driver interfaces must not block since the driver could call them
1738275SEric Cheng  * in interrupt context.
1748275SEric Cheng  *
1758275SEric Cheng  * R11. Walkers must preferably not hold any locks while calling walker
1768275SEric Cheng  * callbacks. Instead these can operate on reference counts. In simple
1778275SEric Cheng  * callbacks it may be ok to hold a lock and call the callbacks, but this is
1788275SEric Cheng  * harder to maintain in the general case of arbitrary callbacks.
1798275SEric Cheng  *
1808275SEric Cheng  * R12. The MAC layer must protect upcall notification callbacks using reference
1818275SEric Cheng  * counts rather than holding locks across the callbacks.
1828275SEric Cheng  *
1838275SEric Cheng  * R13. Given the variety of drivers, it is preferable if the MAC layer can make
1848275SEric Cheng  * sure that any pointers (such as mac ring pointers) it passes to the driver
1858275SEric Cheng  * remain valid until mac unregister time. Currently the mac layer achieves
1868275SEric Cheng  * this by using generation numbers for rings and freeing the mac rings only
1878275SEric Cheng  * at unregister time.  The MAC layer must provide a layer of indirection and
1888275SEric Cheng  * must not expose underlying driver rings or driver data structures/pointers
1898275SEric Cheng  * directly to MAC clients.
1908275SEric Cheng  *
1918275SEric Cheng  * MAC driver rules
1928275SEric Cheng  * ----------------
1938275SEric Cheng  *
1948275SEric Cheng  * R14. It would be preferable if MAC drivers don't hold any locks across any
1958275SEric Cheng  * mac call. However at a minimum they must not hold any locks across data
1968275SEric Cheng  * upcalls. They must also make sure that all references to mac data structures
1978275SEric Cheng  * are cleaned up and that it is single threaded at mac_unregister time.
1988275SEric Cheng  *
1998275SEric Cheng  * R15. MAC driver interfaces don't block and so the action may be done
2008275SEric Cheng  * asynchronously in a separate thread as for example handling notifications.
2018275SEric Cheng  * The driver must not assume that the action is complete when the call
2028275SEric Cheng  * returns.
2038275SEric Cheng  *
2048275SEric Cheng  * R16. Drivers must maintain a generation number per Rx ring, and pass it
2058275SEric Cheng  * back to mac_rx_ring(); They are expected to increment the generation
2068275SEric Cheng  * number whenever the ring's stop routine is invoked.
2078275SEric Cheng  * See comments in mac_rx_ring();
2088275SEric Cheng  *
2098275SEric Cheng  * R17 Similarly mi_stop is another synchronization point and the driver must
2108275SEric Cheng  * ensure that all upcalls are done and there won't be any future upcall
2118275SEric Cheng  * before returning from mi_stop.
2128275SEric Cheng  *
2138275SEric Cheng  * R18. The driver may assume that all set/modify control operations via
2148275SEric Cheng  * the mi_* entry points are single threaded on a per mac end point.
2158275SEric Cheng  *
2168275SEric Cheng  * Lock and Perimeter hierarchy scenarios
2178275SEric Cheng  * ---------------------------------------
2188275SEric Cheng  *
2198275SEric Cheng  * i_mac_impl_lock -> mi_rw_lock -> srs_lock -> s_ring_lock[i_mac_tx_srs_notify]
2208275SEric Cheng  *
2218275SEric Cheng  * ft_lock -> fe_lock [mac_flow_lookup]
2228275SEric Cheng  *
2238275SEric Cheng  * mi_rw_lock -> fe_lock [mac_bcast_send]
2248275SEric Cheng  *
2258275SEric Cheng  * srs_lock -> mac_bw_lock [mac_rx_srs_drain_bw]
2268275SEric Cheng  *
2278275SEric Cheng  * cpu_lock -> mac_srs_g_lock -> srs_lock -> s_ring_lock [mac_walk_srs_and_bind]
2288275SEric Cheng  *
2298275SEric Cheng  * i_dls_devnet_lock -> mac layer locks [dls_devnet_rename]
2308275SEric Cheng  *
2318275SEric Cheng  * Perimeters are ordered P1 -> P2 -> P3 from top to bottom in order of mac
2328275SEric Cheng  * client to driver. In the case of clients that explictly use the mac provided
2338275SEric Cheng  * perimeter mechanism for its serialization, the hierarchy is
2348275SEric Cheng  * Perimeter -> mac layer locks, since the client never holds any locks across
2358275SEric Cheng  * the mac calls. In the case of clients that use its own locks the hierarchy
2368275SEric Cheng  * is Client locks -> Mac Perim -> Mac layer locks. The client never explicitly
2378275SEric Cheng  * calls mac_perim_enter/exit in this case.
2388275SEric Cheng  *
2398275SEric Cheng  * Subflow creation rules
2408275SEric Cheng  * ---------------------------
2418275SEric Cheng  * o In case of a user specified cpulist present on underlying link and flows,
2428275SEric Cheng  * the flows cpulist must be a subset of the underlying link.
2438275SEric Cheng  * o In case of a user specified fanout mode present on link and flow, the
2448275SEric Cheng  * subflow fanout count has to be less than or equal to that of the
2458275SEric Cheng  * underlying link. The cpu-bindings for the subflows will be a subset of
2468275SEric Cheng  * the underlying link.
2478275SEric Cheng  * o In case if no cpulist specified on both underlying link and flow, the
2488275SEric Cheng  * underlying link relies on a  MAC tunable to provide out of box fanout.
2498275SEric Cheng  * The subflow will have no cpulist (the subflow will be unbound)
2508275SEric Cheng  * o In case if no cpulist is specified on the underlying link, a subflow can
2518275SEric Cheng  * carry  either a user-specified cpulist or fanout count. The cpu-bindings
2528275SEric Cheng  * for the subflow will not adhere to restriction that they need to be subset
2538275SEric Cheng  * of the underlying link.
2548275SEric Cheng  * o In case where the underlying link is carrying either a user specified
2558275SEric Cheng  * cpulist or fanout mode and for a unspecified subflow, the subflow will be
2568275SEric Cheng  * created unbound.
2578275SEric Cheng  * o While creating unbound subflows, bandwidth mode changes attempt to
2588275SEric Cheng  * figure a right fanout count. In such cases the fanout count will override
2598275SEric Cheng  * the unbound cpu-binding behavior.
2608275SEric Cheng  * o In addition to this, while cycling between flow and link properties, we
2618275SEric Cheng  * impose a restriction that if a link property has a subflow with
2628275SEric Cheng  * user-specified attributes, we will not allow changing the link property.
2638275SEric Cheng  * The administrator needs to reset all the user specified properties for the
2648275SEric Cheng  * subflows before attempting a link property change.
2658275SEric Cheng  * Some of the above rules can be overridden by specifying additional command
2668275SEric Cheng  * line options while creating or modifying link or subflow properties.
2670Sstevel@tonic-gate  */
2680Sstevel@tonic-gate 
2690Sstevel@tonic-gate #include <sys/types.h>
2700Sstevel@tonic-gate #include <sys/conf.h>
2715895Syz147064 #include <sys/id_space.h>
2726077Syz147064 #include <sys/esunddi.h>
2730Sstevel@tonic-gate #include <sys/stat.h>
2745895Syz147064 #include <sys/mkdev.h>
2750Sstevel@tonic-gate #include <sys/stream.h>
2760Sstevel@tonic-gate #include <sys/strsun.h>
2770Sstevel@tonic-gate #include <sys/strsubr.h>
2780Sstevel@tonic-gate #include <sys/dlpi.h>
2798275SEric Cheng #include <sys/modhash.h>
2808275SEric Cheng #include <sys/mac_provider.h>
2818275SEric Cheng #include <sys/mac_client_impl.h>
2828275SEric Cheng #include <sys/mac_soft_ring.h>
2838275SEric Cheng #include <sys/mac_impl.h>
2848275SEric Cheng #include <sys/mac.h>
2855895Syz147064 #include <sys/dls.h>
286269Sericheng #include <sys/dld.h>
2872311Sseb #include <sys/modctl.h>
2883448Sdh155122 #include <sys/fs/dv_node.h>
2895009Sgd78059 #include <sys/thread.h>
2905009Sgd78059 #include <sys/proc.h>
2915009Sgd78059 #include <sys/callb.h>
2925009Sgd78059 #include <sys/cpuvar.h>
2933288Sseb #include <sys/atomic.h>
2948275SEric Cheng #include <sys/bitmap.h>
2954913Sethindra #include <sys/sdt.h>
2968275SEric Cheng #include <sys/mac_flow.h>
2978275SEric Cheng #include <sys/ddi_intr_impl.h>
2988275SEric Cheng #include <sys/disp.h>
2998275SEric Cheng #include <sys/sdt.h>
3008275SEric Cheng #include <sys/vnic.h>
3018275SEric Cheng #include <sys/vnic_impl.h>
3028275SEric Cheng #include <sys/vlan.h>
3038275SEric Cheng #include <inet/ip.h>
3048275SEric Cheng #include <inet/ip6.h>
3058275SEric Cheng #include <sys/exacct.h>
3068275SEric Cheng #include <sys/exacct_impl.h>
3075903Ssowmini #include <inet/nd.h>
3086512Ssowmini #include <sys/ethernet.h>
3090Sstevel@tonic-gate 
3100Sstevel@tonic-gate #define	IMPL_HASHSZ	67	/* prime */
3110Sstevel@tonic-gate 
3128275SEric Cheng kmem_cache_t	*i_mac_impl_cachep;
3138275SEric Cheng mod_hash_t		*i_mac_impl_hash;
314269Sericheng krwlock_t		i_mac_impl_lock;
315269Sericheng uint_t			i_mac_impl_count;
3168275SEric Cheng static kmem_cache_t	*mac_ring_cache;
3175895Syz147064 static id_space_t	*minor_ids;
3185895Syz147064 static uint32_t		minor_count;
3190Sstevel@tonic-gate 
3208275SEric Cheng /*
3218275SEric Cheng  * Logging stuff. Perhaps mac_logging_interval could be broken into
3228275SEric Cheng  * mac_flow_log_interval and mac_link_log_interval if we want to be
3238275SEric Cheng  * able to schedule them differently.
3248275SEric Cheng  */
3258275SEric Cheng uint_t			mac_logging_interval;
3268275SEric Cheng boolean_t		mac_flow_log_enable;
3278275SEric Cheng boolean_t		mac_link_log_enable;
3288275SEric Cheng timeout_id_t		mac_logging_timer;
3298275SEric Cheng 
3308275SEric Cheng /* for debugging, see MAC_DBG_PRT() in mac_impl.h */
3318275SEric Cheng int mac_dbg = 0;
3328275SEric Cheng 
3332311Sseb #define	MACTYPE_KMODDIR	"mac"
3342311Sseb #define	MACTYPE_HASHSZ	67
3352311Sseb static mod_hash_t	*i_mactype_hash;
3363288Sseb /*
3373288Sseb  * i_mactype_lock synchronizes threads that obtain references to mactype_t
3383288Sseb  * structures through i_mactype_getplugin().
3393288Sseb  */
3403288Sseb static kmutex_t		i_mactype_lock;
3412311Sseb 
3420Sstevel@tonic-gate /*
3438275SEric Cheng  * mac_tx_percpu_cnt
3448275SEric Cheng  *
3458275SEric Cheng  * Number of per cpu locks per mac_client_impl_t. Used by the transmit side
3468275SEric Cheng  * in mac_tx to reduce lock contention. This is sized at boot time in mac_init.
3478275SEric Cheng  * mac_tx_percpu_cnt_max is settable in /etc/system and must be a power of 2.
3488275SEric Cheng  * Per cpu locks may be disabled by setting mac_tx_percpu_cnt_max to 1.
3495084Sjohnlev  */
3508275SEric Cheng int mac_tx_percpu_cnt;
3518275SEric Cheng int mac_tx_percpu_cnt_max = 128;
3528275SEric Cheng 
3538275SEric Cheng static int i_mac_constructor(void *, void *, int);
3548275SEric Cheng static void i_mac_destructor(void *, void *);
3558275SEric Cheng static int i_mac_ring_ctor(void *, void *, int);
3568275SEric Cheng static void i_mac_ring_dtor(void *, void *);
3578275SEric Cheng static mblk_t *mac_rx_classify(mac_impl_t *, mac_resource_handle_t, mblk_t *);
3588275SEric Cheng void mac_tx_client_flush(mac_client_impl_t *);
3598275SEric Cheng void mac_tx_client_block(mac_client_impl_t *);
3608275SEric Cheng static void mac_rx_ring_quiesce(mac_ring_t *, uint_t);
3618275SEric Cheng static int mac_start_group_and_rings(mac_group_t *);
3628275SEric Cheng static void mac_stop_group_and_rings(mac_group_t *);
3632311Sseb 
3640Sstevel@tonic-gate /*
3650Sstevel@tonic-gate  * Module initialization functions.
3660Sstevel@tonic-gate  */
3670Sstevel@tonic-gate 
3680Sstevel@tonic-gate void
3690Sstevel@tonic-gate mac_init(void)
3700Sstevel@tonic-gate {
3718275SEric Cheng 	mac_tx_percpu_cnt = ((boot_max_ncpus == -1) ? max_ncpus :
3728275SEric Cheng 	    boot_max_ncpus);
3738275SEric Cheng 
3748275SEric Cheng 	/* Upper bound is mac_tx_percpu_cnt_max */
3758275SEric Cheng 	if (mac_tx_percpu_cnt > mac_tx_percpu_cnt_max)
3768275SEric Cheng 		mac_tx_percpu_cnt = mac_tx_percpu_cnt_max;
3778275SEric Cheng 
3788275SEric Cheng 	if (mac_tx_percpu_cnt < 1) {
3798275SEric Cheng 		/* Someone set max_tx_percpu_cnt_max to 0 or less */
3808275SEric Cheng 		mac_tx_percpu_cnt = 1;
3818275SEric Cheng 	}
3828275SEric Cheng 
3838275SEric Cheng 	ASSERT(mac_tx_percpu_cnt >= 1);
3848275SEric Cheng 	mac_tx_percpu_cnt = (1 << highbit(mac_tx_percpu_cnt - 1));
3858275SEric Cheng 	/*
3868275SEric Cheng 	 * Make it of the form 2**N - 1 in the range
3878275SEric Cheng 	 * [0 .. mac_tx_percpu_cnt_max - 1]
3888275SEric Cheng 	 */
3898275SEric Cheng 	mac_tx_percpu_cnt--;
3908275SEric Cheng 
3910Sstevel@tonic-gate 	i_mac_impl_cachep = kmem_cache_create("mac_impl_cache",
3922311Sseb 	    sizeof (mac_impl_t), 0, i_mac_constructor, i_mac_destructor,
3932311Sseb 	    NULL, NULL, NULL, 0);
3940Sstevel@tonic-gate 	ASSERT(i_mac_impl_cachep != NULL);
3950Sstevel@tonic-gate 
3968275SEric Cheng 	mac_ring_cache = kmem_cache_create("mac_ring_cache",
3978275SEric Cheng 	    sizeof (mac_ring_t), 0, i_mac_ring_ctor, i_mac_ring_dtor, NULL,
3988275SEric Cheng 	    NULL, NULL, 0);
3998275SEric Cheng 	ASSERT(mac_ring_cache != NULL);
4005084Sjohnlev 
401269Sericheng 	i_mac_impl_hash = mod_hash_create_extended("mac_impl_hash",
402269Sericheng 	    IMPL_HASHSZ, mod_hash_null_keydtor, mod_hash_null_valdtor,
403269Sericheng 	    mod_hash_bystr, NULL, mod_hash_strkey_cmp, KM_SLEEP);
404269Sericheng 	rw_init(&i_mac_impl_lock, NULL, RW_DEFAULT, NULL);
4058275SEric Cheng 
4068275SEric Cheng 	mac_flow_init();
4078275SEric Cheng 	mac_soft_ring_init();
4088275SEric Cheng 	mac_bcast_init();
4098275SEric Cheng 	mac_client_init();
4108275SEric Cheng 
411269Sericheng 	i_mac_impl_count = 0;
4122311Sseb 
4132311Sseb 	i_mactype_hash = mod_hash_create_extended("mactype_hash",
4142311Sseb 	    MACTYPE_HASHSZ,
4152311Sseb 	    mod_hash_null_keydtor, mod_hash_null_valdtor,
4162311Sseb 	    mod_hash_bystr, NULL, mod_hash_strkey_cmp, KM_SLEEP);
4175895Syz147064 
4185895Syz147064 	/*
4195895Syz147064 	 * Allocate an id space to manage minor numbers. The range of the
4205895Syz147064 	 * space will be from MAC_MAX_MINOR+1 to MAXMIN32 (maximum legal
4215895Syz147064 	 * minor number is MAXMIN, but id_t is type of integer and does not
4225895Syz147064 	 * allow MAXMIN).
4235895Syz147064 	 */
4245895Syz147064 	minor_ids = id_space_create("mac_minor_ids", MAC_MAX_MINOR+1, MAXMIN32);
4255895Syz147064 	ASSERT(minor_ids != NULL);
4265895Syz147064 	minor_count = 0;
4278275SEric Cheng 
4288275SEric Cheng 	/* Let's default to 20 seconds */
4298275SEric Cheng 	mac_logging_interval = 20;
4308275SEric Cheng 	mac_flow_log_enable = B_FALSE;
4318275SEric Cheng 	mac_link_log_enable = B_FALSE;
4328275SEric Cheng 	mac_logging_timer = 0;
4330Sstevel@tonic-gate }
4340Sstevel@tonic-gate 
4350Sstevel@tonic-gate int
4360Sstevel@tonic-gate mac_fini(void)
4370Sstevel@tonic-gate {
4385895Syz147064 	if (i_mac_impl_count > 0 || minor_count > 0)
439269Sericheng 		return (EBUSY);
4400Sstevel@tonic-gate 
4415895Syz147064 	id_space_destroy(minor_ids);
4428275SEric Cheng 	mac_flow_fini();
4435895Syz147064 
444269Sericheng 	mod_hash_destroy_hash(i_mac_impl_hash);
445269Sericheng 	rw_destroy(&i_mac_impl_lock);
4460Sstevel@tonic-gate 
4478275SEric Cheng 	mac_client_fini();
4488275SEric Cheng 	kmem_cache_destroy(mac_ring_cache);
4492311Sseb 
4502311Sseb 	mod_hash_destroy_hash(i_mactype_hash);
4518275SEric Cheng 	mac_soft_ring_finish();
4520Sstevel@tonic-gate 	return (0);
4530Sstevel@tonic-gate }
4540Sstevel@tonic-gate 
4558275SEric Cheng void
4568275SEric Cheng mac_init_ops(struct dev_ops *ops, const char *name)
4578275SEric Cheng {
4588275SEric Cheng 	dld_init_ops(ops, name);
4598275SEric Cheng }
4608275SEric Cheng 
4618275SEric Cheng void
4628275SEric Cheng mac_fini_ops(struct dev_ops *ops)
4638275SEric Cheng {
4648275SEric Cheng 	dld_fini_ops(ops);
4658275SEric Cheng }
4668275SEric Cheng 
4678275SEric Cheng /*ARGSUSED*/
4688275SEric Cheng static int
4698275SEric Cheng i_mac_constructor(void *buf, void *arg, int kmflag)
4708275SEric Cheng {
4718275SEric Cheng 	mac_impl_t	*mip = buf;
4728275SEric Cheng 
4738275SEric Cheng 	bzero(buf, sizeof (mac_impl_t));
4748275SEric Cheng 
4758275SEric Cheng 	mip->mi_linkstate = LINK_STATE_UNKNOWN;
4768275SEric Cheng 	mip->mi_nclients = 0;
4778275SEric Cheng 
4788275SEric Cheng 	mutex_init(&mip->mi_lock, NULL, MUTEX_DRIVER, NULL);
4798275SEric Cheng 	rw_init(&mip->mi_rw_lock, NULL, RW_DRIVER, NULL);
4808275SEric Cheng 	mutex_init(&mip->mi_notify_lock, NULL, MUTEX_DRIVER, NULL);
4818275SEric Cheng 	mutex_init(&mip->mi_promisc_lock, NULL, MUTEX_DRIVER, NULL);
4828275SEric Cheng 	mutex_init(&mip->mi_ring_lock, NULL, MUTEX_DEFAULT, NULL);
4838275SEric Cheng 
4848275SEric Cheng 	mip->mi_notify_cb_info.mcbi_lockp = &mip->mi_notify_lock;
4858275SEric Cheng 	cv_init(&mip->mi_notify_cb_info.mcbi_cv, NULL, CV_DRIVER, NULL);
4868275SEric Cheng 	mip->mi_promisc_cb_info.mcbi_lockp = &mip->mi_promisc_lock;
4878275SEric Cheng 	cv_init(&mip->mi_promisc_cb_info.mcbi_cv, NULL, CV_DRIVER, NULL);
4888275SEric Cheng 	return (0);
4898275SEric Cheng }
4908275SEric Cheng 
4918275SEric Cheng /*ARGSUSED*/
4928275SEric Cheng static void
4938275SEric Cheng i_mac_destructor(void *buf, void *arg)
4948275SEric Cheng {
4958275SEric Cheng 	mac_impl_t	*mip = buf;
4968275SEric Cheng 	mac_cb_info_t	*mcbi;
4978275SEric Cheng 
4988275SEric Cheng 	ASSERT(mip->mi_ref == 0);
4998275SEric Cheng 	ASSERT(mip->mi_active == 0);
5008275SEric Cheng 	ASSERT(mip->mi_linkstate == LINK_STATE_UNKNOWN);
5018275SEric Cheng 	ASSERT(mip->mi_devpromisc == 0);
5028275SEric Cheng 	ASSERT(mip->mi_promisc == 0);
5038275SEric Cheng 	ASSERT(mip->mi_ksp == NULL);
5048275SEric Cheng 	ASSERT(mip->mi_kstat_count == 0);
5058275SEric Cheng 	ASSERT(mip->mi_nclients == 0);
5068275SEric Cheng 	ASSERT(mip->mi_nactiveclients == 0);
5078275SEric Cheng 	ASSERT(mip->mi_state_flags == 0);
5088275SEric Cheng 	ASSERT(mip->mi_factory_addr == NULL);
5098275SEric Cheng 	ASSERT(mip->mi_factory_addr_num == 0);
5108275SEric Cheng 	ASSERT(mip->mi_default_tx_ring == NULL);
5118275SEric Cheng 
5128275SEric Cheng 	mcbi = &mip->mi_notify_cb_info;
5138275SEric Cheng 	ASSERT(mcbi->mcbi_del_cnt == 0 && mcbi->mcbi_walker_cnt == 0);
5148275SEric Cheng 	ASSERT(mip->mi_notify_bits == 0);
5158275SEric Cheng 	ASSERT(mip->mi_notify_thread == NULL);
5168275SEric Cheng 	ASSERT(mcbi->mcbi_lockp == &mip->mi_notify_lock);
5178275SEric Cheng 	mcbi->mcbi_lockp = NULL;
5188275SEric Cheng 
5198275SEric Cheng 	mcbi = &mip->mi_promisc_cb_info;
5208275SEric Cheng 	ASSERT(mcbi->mcbi_del_cnt == 0 && mip->mi_promisc_list == NULL);
5218275SEric Cheng 	ASSERT(mip->mi_promisc_list == NULL);
5228275SEric Cheng 	ASSERT(mcbi->mcbi_lockp == &mip->mi_promisc_lock);
5238275SEric Cheng 	mcbi->mcbi_lockp = NULL;
5248275SEric Cheng 
5258275SEric Cheng 	ASSERT(mip->mi_bcast_ngrps == 0 && mip->mi_bcast_grp == NULL);
5268275SEric Cheng 	ASSERT(mip->mi_perim_owner == NULL && mip->mi_perim_ocnt == 0);
5278275SEric Cheng 
5288275SEric Cheng 	mutex_destroy(&mip->mi_lock);
5298275SEric Cheng 	rw_destroy(&mip->mi_rw_lock);
5308275SEric Cheng 
5318275SEric Cheng 	mutex_destroy(&mip->mi_promisc_lock);
5328275SEric Cheng 	cv_destroy(&mip->mi_promisc_cb_info.mcbi_cv);
5338275SEric Cheng 	mutex_destroy(&mip->mi_notify_lock);
5348275SEric Cheng 	cv_destroy(&mip->mi_notify_cb_info.mcbi_cv);
5358275SEric Cheng 	mutex_destroy(&mip->mi_ring_lock);
5368275SEric Cheng }
5378275SEric Cheng 
5388275SEric Cheng /* ARGSUSED */
5398275SEric Cheng static int
5408275SEric Cheng i_mac_ring_ctor(void *buf, void *arg, int kmflag)
5418275SEric Cheng {
5428275SEric Cheng 	mac_ring_t *ring = (mac_ring_t *)buf;
5438275SEric Cheng 
5448275SEric Cheng 	bzero(ring, sizeof (mac_ring_t));
5458275SEric Cheng 	cv_init(&ring->mr_cv, NULL, CV_DEFAULT, NULL);
5468275SEric Cheng 	mutex_init(&ring->mr_lock, NULL, MUTEX_DEFAULT, NULL);
5478275SEric Cheng 	ring->mr_state = MR_FREE;
5488275SEric Cheng 	return (0);
5498275SEric Cheng }
5508275SEric Cheng 
5518275SEric Cheng /* ARGSUSED */
5528275SEric Cheng static void
5538275SEric Cheng i_mac_ring_dtor(void *buf, void *arg)
5548275SEric Cheng {
5558275SEric Cheng 	mac_ring_t *ring = (mac_ring_t *)buf;
5568275SEric Cheng 
5578275SEric Cheng 	cv_destroy(&ring->mr_cv);
5588275SEric Cheng 	mutex_destroy(&ring->mr_lock);
5598275SEric Cheng }
5608275SEric Cheng 
5618275SEric Cheng /*
5628275SEric Cheng  * Common functions to do mac callback addition and deletion. Currently this is
5638275SEric Cheng  * used by promisc callbacks and notify callbacks. List addition and deletion
5648275SEric Cheng  * need to take care of list walkers. List walkers in general, can't hold list
5658275SEric Cheng  * locks and make upcall callbacks due to potential lock order and recursive
5668275SEric Cheng  * reentry issues. Instead list walkers increment the list walker count to mark
5678275SEric Cheng  * the presence of a walker thread. Addition can be carefully done to ensure
5688275SEric Cheng  * that the list walker always sees either the old list or the new list.
5698275SEric Cheng  * However the deletion can't be done while the walker is active, instead the
5708275SEric Cheng  * deleting thread simply marks the entry as logically deleted. The last walker
5718275SEric Cheng  * physically deletes and frees up the logically deleted entries when the walk
5728275SEric Cheng  * is complete.
5738275SEric Cheng  */
5748275SEric Cheng void
5758275SEric Cheng mac_callback_add(mac_cb_info_t *mcbi, mac_cb_t **mcb_head,
5768275SEric Cheng     mac_cb_t *mcb_elem)
5778275SEric Cheng {
5788275SEric Cheng 	mac_cb_t	*p;
5798275SEric Cheng 	mac_cb_t	**pp;
5808275SEric Cheng 
5818275SEric Cheng 	/* Verify it is not already in the list */
5828275SEric Cheng 	for (pp = mcb_head; (p = *pp) != NULL; pp = &p->mcb_nextp) {
5838275SEric Cheng 		if (p == mcb_elem)
5848275SEric Cheng 			break;
5858275SEric Cheng 	}
5868275SEric Cheng 	VERIFY(p == NULL);
5878275SEric Cheng 
5888275SEric Cheng 	/*
5898275SEric Cheng 	 * Add it to the head of the callback list. The membar ensures that
5908275SEric Cheng 	 * the following list pointer manipulations reach global visibility
5918275SEric Cheng 	 * in exactly the program order below.
5928275SEric Cheng 	 */
5938275SEric Cheng 	ASSERT(MUTEX_HELD(mcbi->mcbi_lockp));
5948275SEric Cheng 
5958275SEric Cheng 	mcb_elem->mcb_nextp = *mcb_head;
5968275SEric Cheng 	membar_producer();
5978275SEric Cheng 	*mcb_head = mcb_elem;
5988275SEric Cheng }
5998275SEric Cheng 
6008275SEric Cheng /*
6018275SEric Cheng  * Mark the entry as logically deleted. If there aren't any walkers unlink
6028275SEric Cheng  * from the list. In either case return the corresponding status.
6038275SEric Cheng  */
6048275SEric Cheng boolean_t
6058275SEric Cheng mac_callback_remove(mac_cb_info_t *mcbi, mac_cb_t **mcb_head,
6068275SEric Cheng     mac_cb_t *mcb_elem)
6078275SEric Cheng {
6088275SEric Cheng 	mac_cb_t	*p;
6098275SEric Cheng 	mac_cb_t	**pp;
6108275SEric Cheng 
6118275SEric Cheng 	ASSERT(MUTEX_HELD(mcbi->mcbi_lockp));
6128275SEric Cheng 	/*
6138275SEric Cheng 	 * Search the callback list for the entry to be removed
6148275SEric Cheng 	 */
6158275SEric Cheng 	for (pp = mcb_head; (p = *pp) != NULL; pp = &p->mcb_nextp) {
6168275SEric Cheng 		if (p == mcb_elem)
6178275SEric Cheng 			break;
6188275SEric Cheng 	}
6198275SEric Cheng 	VERIFY(p != NULL);
6208275SEric Cheng 
6218275SEric Cheng 	/*
6228275SEric Cheng 	 * If there are walkers just mark it as deleted and the last walker
6238275SEric Cheng 	 * will remove from the list and free it.
6248275SEric Cheng 	 */
6258275SEric Cheng 	if (mcbi->mcbi_walker_cnt != 0) {
6268275SEric Cheng 		p->mcb_flags |= MCB_CONDEMNED;
6278275SEric Cheng 		mcbi->mcbi_del_cnt++;
6288275SEric Cheng 		return (B_FALSE);
6298275SEric Cheng 	}
6308275SEric Cheng 
6318275SEric Cheng 	ASSERT(mcbi->mcbi_del_cnt == 0);
6328275SEric Cheng 	*pp = p->mcb_nextp;
6338275SEric Cheng 	p->mcb_nextp = NULL;
6348275SEric Cheng 	return (B_TRUE);
6358275SEric Cheng }
6368275SEric Cheng 
6378275SEric Cheng /*
6388275SEric Cheng  * Wait for all pending callback removals to be completed
6398275SEric Cheng  */
6408275SEric Cheng void
6418275SEric Cheng mac_callback_remove_wait(mac_cb_info_t *mcbi)
6428275SEric Cheng {
6438275SEric Cheng 	ASSERT(MUTEX_HELD(mcbi->mcbi_lockp));
6448275SEric Cheng 	while (mcbi->mcbi_del_cnt != 0) {
6458275SEric Cheng 		DTRACE_PROBE1(need_wait, mac_cb_info_t *, mcbi);
6468275SEric Cheng 		cv_wait(&mcbi->mcbi_cv, mcbi->mcbi_lockp);
6478275SEric Cheng 	}
6488275SEric Cheng }
6498275SEric Cheng 
6500Sstevel@tonic-gate /*
6518275SEric Cheng  * The last mac callback walker does the cleanup. Walk the list and unlik
6528275SEric Cheng  * all the logically deleted entries and construct a temporary list of
6538275SEric Cheng  * removed entries. Return the list of removed entries to the caller.
6548275SEric Cheng  */
6558275SEric Cheng mac_cb_t *
6568275SEric Cheng mac_callback_walker_cleanup(mac_cb_info_t *mcbi, mac_cb_t **mcb_head)
6578275SEric Cheng {
6588275SEric Cheng 	mac_cb_t	*p;
6598275SEric Cheng 	mac_cb_t	**pp;
6608275SEric Cheng 	mac_cb_t	*rmlist = NULL;		/* List of removed elements */
6618275SEric Cheng 	int	cnt = 0;
6628275SEric Cheng 
6638275SEric Cheng 	ASSERT(MUTEX_HELD(mcbi->mcbi_lockp));
6648275SEric Cheng 	ASSERT(mcbi->mcbi_del_cnt != 0 && mcbi->mcbi_walker_cnt == 0);
6658275SEric Cheng 
6668275SEric Cheng 	pp = mcb_head;
6678275SEric Cheng 	while (*pp != NULL) {
6688275SEric Cheng 		if ((*pp)->mcb_flags & MCB_CONDEMNED) {
6698275SEric Cheng 			p = *pp;
6708275SEric Cheng 			*pp = p->mcb_nextp;
6718275SEric Cheng 			p->mcb_nextp = rmlist;
6728275SEric Cheng 			rmlist = p;
6738275SEric Cheng 			cnt++;
6748275SEric Cheng 			continue;
6758275SEric Cheng 		}
6768275SEric Cheng 		pp = &(*pp)->mcb_nextp;
6778275SEric Cheng 	}
6788275SEric Cheng 
6798275SEric Cheng 	ASSERT(mcbi->mcbi_del_cnt == cnt);
6808275SEric Cheng 	mcbi->mcbi_del_cnt = 0;
6818275SEric Cheng 	return (rmlist);
6828275SEric Cheng }
6838275SEric Cheng 
6848275SEric Cheng boolean_t
6858275SEric Cheng mac_callback_lookup(mac_cb_t **mcb_headp, mac_cb_t *mcb_elem)
6868275SEric Cheng {
6878275SEric Cheng 	mac_cb_t	*mcb;
6888275SEric Cheng 
6898275SEric Cheng 	/* Verify it is not already in the list */
6908275SEric Cheng 	for (mcb = *mcb_headp; mcb != NULL; mcb = mcb->mcb_nextp) {
6918275SEric Cheng 		if (mcb == mcb_elem)
6928275SEric Cheng 			return (B_TRUE);
6938275SEric Cheng 	}
6948275SEric Cheng 
6958275SEric Cheng 	return (B_FALSE);
6968275SEric Cheng }
6978275SEric Cheng 
6988275SEric Cheng boolean_t
6998275SEric Cheng mac_callback_find(mac_cb_info_t *mcbi, mac_cb_t **mcb_headp, mac_cb_t *mcb_elem)
7008275SEric Cheng {
7018275SEric Cheng 	boolean_t	found;
7028275SEric Cheng 
7038275SEric Cheng 	mutex_enter(mcbi->mcbi_lockp);
7048275SEric Cheng 	found = mac_callback_lookup(mcb_headp, mcb_elem);
7058275SEric Cheng 	mutex_exit(mcbi->mcbi_lockp);
7068275SEric Cheng 
7078275SEric Cheng 	return (found);
7088275SEric Cheng }
7098275SEric Cheng 
7108275SEric Cheng /* Free the list of removed callbacks */
7118275SEric Cheng void
7128275SEric Cheng mac_callback_free(mac_cb_t *rmlist)
7138275SEric Cheng {
7148275SEric Cheng 	mac_cb_t	*mcb;
7158275SEric Cheng 	mac_cb_t	*mcb_next;
7168275SEric Cheng 
7178275SEric Cheng 	for (mcb = rmlist; mcb != NULL; mcb = mcb_next) {
7188275SEric Cheng 		mcb_next = mcb->mcb_nextp;
7198275SEric Cheng 		kmem_free(mcb->mcb_objp, mcb->mcb_objsize);
7208275SEric Cheng 	}
7218275SEric Cheng }
7228275SEric Cheng 
7238275SEric Cheng /*
7248275SEric Cheng  * The promisc callbacks are in 2 lists, one off the 'mip' and another off the
7258275SEric Cheng  * 'mcip' threaded by mpi_mi_link and mpi_mci_link respectively. However there
7268275SEric Cheng  * is only a single shared total walker count, and an entry can't be physically
7278275SEric Cheng  * unlinked if a walker is active on either list. The last walker does this
7288275SEric Cheng  * cleanup of logically deleted entries.
7298275SEric Cheng  */
7308275SEric Cheng void
7318275SEric Cheng i_mac_promisc_walker_cleanup(mac_impl_t *mip)
7328275SEric Cheng {
7338275SEric Cheng 	mac_cb_t	*rmlist;
7348275SEric Cheng 	mac_cb_t	*mcb;
7358275SEric Cheng 	mac_cb_t	*mcb_next;
7368275SEric Cheng 	mac_promisc_impl_t	*mpip;
7378275SEric Cheng 
7388275SEric Cheng 	/*
7398275SEric Cheng 	 * Construct a temporary list of deleted callbacks by walking the
7408275SEric Cheng 	 * the mi_promisc_list. Then for each entry in the temporary list,
7418275SEric Cheng 	 * remove it from the mci_promisc_list and free the entry.
7428275SEric Cheng 	 */
7438275SEric Cheng 	rmlist = mac_callback_walker_cleanup(&mip->mi_promisc_cb_info,
7448275SEric Cheng 	    &mip->mi_promisc_list);
7458275SEric Cheng 
7468275SEric Cheng 	for (mcb = rmlist; mcb != NULL; mcb = mcb_next) {
7478275SEric Cheng 		mcb_next = mcb->mcb_nextp;
7488275SEric Cheng 		mpip = (mac_promisc_impl_t *)mcb->mcb_objp;
7498275SEric Cheng 		VERIFY(mac_callback_remove(&mip->mi_promisc_cb_info,
7508275SEric Cheng 		    &mpip->mpi_mcip->mci_promisc_list, &mpip->mpi_mci_link));
7518275SEric Cheng 		mcb->mcb_flags = 0;
7528275SEric Cheng 		mcb->mcb_nextp = NULL;
7538275SEric Cheng 		kmem_cache_free(mac_promisc_impl_cache, mpip);
7548275SEric Cheng 	}
7558275SEric Cheng }
7568275SEric Cheng 
7578275SEric Cheng void
7588275SEric Cheng i_mac_notify(mac_impl_t *mip, mac_notify_type_t type)
7598275SEric Cheng {
7608275SEric Cheng 	mac_cb_info_t	*mcbi;
7618275SEric Cheng 
7628275SEric Cheng 	/*
7638275SEric Cheng 	 * Signal the notify thread even after mi_ref has become zero and
7648275SEric Cheng 	 * mi_disabled is set. The synchronization with the notify thread
7658275SEric Cheng 	 * happens in mac_unregister and that implies the driver must make
7668275SEric Cheng 	 * sure it is single-threaded (with respect to mac calls) and that
7678275SEric Cheng 	 * all pending mac calls have returned before it calls mac_unregister
7688275SEric Cheng 	 */
7698275SEric Cheng 	rw_enter(&i_mac_impl_lock, RW_READER);
7708275SEric Cheng 	if (mip->mi_state_flags & MIS_DISABLED)
7718275SEric Cheng 		goto exit;
7728275SEric Cheng 
7738275SEric Cheng 	/*
7748275SEric Cheng 	 * Guard against incorrect notifications.  (Running a newer
7758275SEric Cheng 	 * mac client against an older implementation?)
7768275SEric Cheng 	 */
7778275SEric Cheng 	if (type >= MAC_NNOTE)
7788275SEric Cheng 		goto exit;
7798275SEric Cheng 
7808275SEric Cheng 	mcbi = &mip->mi_notify_cb_info;
7818275SEric Cheng 	mutex_enter(mcbi->mcbi_lockp);
7828275SEric Cheng 	mip->mi_notify_bits |= (1 << type);
7838275SEric Cheng 	cv_broadcast(&mcbi->mcbi_cv);
7848275SEric Cheng 	mutex_exit(mcbi->mcbi_lockp);
7858275SEric Cheng 
7868275SEric Cheng exit:
7878275SEric Cheng 	rw_exit(&i_mac_impl_lock);
7888275SEric Cheng }
7898275SEric Cheng 
7908275SEric Cheng /*
7918275SEric Cheng  * Mac serialization primitives. Please see the block comment at the
7928275SEric Cheng  * top of the file.
7930Sstevel@tonic-gate  */
7948275SEric Cheng void
7958275SEric Cheng i_mac_perim_enter(mac_impl_t *mip)
7968275SEric Cheng {
7978275SEric Cheng 	mac_client_impl_t	*mcip;
7988275SEric Cheng 
7998275SEric Cheng 	if (mip->mi_state_flags & MIS_IS_VNIC) {
8008275SEric Cheng 		/*
8018275SEric Cheng 		 * This is a VNIC. Return the lower mac since that is what
8028275SEric Cheng 		 * we want to serialize on.
8038275SEric Cheng 		 */
8048275SEric Cheng 		mcip = mac_vnic_lower(mip);
8058275SEric Cheng 		mip = mcip->mci_mip;
8068275SEric Cheng 	}
8078275SEric Cheng 
8088275SEric Cheng 	mutex_enter(&mip->mi_perim_lock);
8098275SEric Cheng 	if (mip->mi_perim_owner == curthread) {
8108275SEric Cheng 		mip->mi_perim_ocnt++;
8118275SEric Cheng 		mutex_exit(&mip->mi_perim_lock);
8128275SEric Cheng 		return;
8138275SEric Cheng 	}
8148275SEric Cheng 
8158275SEric Cheng 	while (mip->mi_perim_owner != NULL)
8168275SEric Cheng 		cv_wait(&mip->mi_perim_cv, &mip->mi_perim_lock);
8178275SEric Cheng 
8188275SEric Cheng 	mip->mi_perim_owner = curthread;
8198275SEric Cheng 	ASSERT(mip->mi_perim_ocnt == 0);
8208275SEric Cheng 	mip->mi_perim_ocnt++;
8218275SEric Cheng #ifdef DEBUG
8228275SEric Cheng 	mip->mi_perim_stack_depth = getpcstack(mip->mi_perim_stack,
8238275SEric Cheng 	    MAC_PERIM_STACK_DEPTH);
8248275SEric Cheng #endif
8258275SEric Cheng 	mutex_exit(&mip->mi_perim_lock);
8268275SEric Cheng }
8278275SEric Cheng 
8288275SEric Cheng int
8298275SEric Cheng i_mac_perim_enter_nowait(mac_impl_t *mip)
8308275SEric Cheng {
8318275SEric Cheng 	/*
8328275SEric Cheng 	 * The vnic is a special case, since the serialization is done based
8338275SEric Cheng 	 * on the lower mac. If the lower mac is busy, it does not imply the
8348275SEric Cheng 	 * vnic can't be unregistered. But in the case of other drivers,
8358275SEric Cheng 	 * a busy perimeter or open mac handles implies that the mac is busy
8368275SEric Cheng 	 * and can't be unregistered.
8378275SEric Cheng 	 */
8388275SEric Cheng 	if (mip->mi_state_flags & MIS_IS_VNIC) {
8398275SEric Cheng 		i_mac_perim_enter(mip);
8408275SEric Cheng 		return (0);
8418275SEric Cheng 	}
8428275SEric Cheng 
8438275SEric Cheng 	mutex_enter(&mip->mi_perim_lock);
8448275SEric Cheng 	if (mip->mi_perim_owner != NULL) {
8458275SEric Cheng 		mutex_exit(&mip->mi_perim_lock);
8468275SEric Cheng 		return (EBUSY);
8478275SEric Cheng 	}
8488275SEric Cheng 	ASSERT(mip->mi_perim_ocnt == 0);
8498275SEric Cheng 	mip->mi_perim_owner = curthread;
8508275SEric Cheng 	mip->mi_perim_ocnt++;
8518275SEric Cheng 	mutex_exit(&mip->mi_perim_lock);
8528275SEric Cheng 
8538275SEric Cheng 	return (0);
8548275SEric Cheng }
8558275SEric Cheng 
8568275SEric Cheng void
8578275SEric Cheng i_mac_perim_exit(mac_impl_t *mip)
8588275SEric Cheng {
8598275SEric Cheng 	mac_client_impl_t *mcip;
8608275SEric Cheng 
8618275SEric Cheng 	if (mip->mi_state_flags & MIS_IS_VNIC) {
8628275SEric Cheng 		/*
8638275SEric Cheng 		 * This is a VNIC. Return the lower mac since that is what
8648275SEric Cheng 		 * we want to serialize on.
8658275SEric Cheng 		 */
8668275SEric Cheng 		mcip = mac_vnic_lower(mip);
8678275SEric Cheng 		mip = mcip->mci_mip;
8688275SEric Cheng 	}
8698275SEric Cheng 
8708275SEric Cheng 	ASSERT(mip->mi_perim_owner == curthread && mip->mi_perim_ocnt != 0);
8718275SEric Cheng 
8728275SEric Cheng 	mutex_enter(&mip->mi_perim_lock);
8738275SEric Cheng 	if (--mip->mi_perim_ocnt == 0) {
8748275SEric Cheng 		mip->mi_perim_owner = NULL;
8758275SEric Cheng 		cv_signal(&mip->mi_perim_cv);
8768275SEric Cheng 	}
8778275SEric Cheng 	mutex_exit(&mip->mi_perim_lock);
8788275SEric Cheng }
8798275SEric Cheng 
8808275SEric Cheng /*
8818275SEric Cheng  * Returns whether the current thread holds the mac perimeter. Used in making
8828275SEric Cheng  * assertions.
8838275SEric Cheng  */
8848275SEric Cheng boolean_t
8858275SEric Cheng mac_perim_held(mac_handle_t mh)
8868275SEric Cheng {
8878275SEric Cheng 	mac_impl_t	*mip = (mac_impl_t *)mh;
8888275SEric Cheng 	mac_client_impl_t *mcip;
8898275SEric Cheng 
8908275SEric Cheng 	if (mip->mi_state_flags & MIS_IS_VNIC) {
8918275SEric Cheng 		/*
8928275SEric Cheng 		 * This is a VNIC. Return the lower mac since that is what
8938275SEric Cheng 		 * we want to serialize on.
8948275SEric Cheng 		 */
8958275SEric Cheng 		mcip = mac_vnic_lower(mip);
8968275SEric Cheng 		mip = mcip->mci_mip;
8978275SEric Cheng 	}
8988275SEric Cheng 	return (mip->mi_perim_owner == curthread);
8998275SEric Cheng }
9008275SEric Cheng 
9018275SEric Cheng /*
9028275SEric Cheng  * mac client interfaces to enter the mac perimeter of a mac end point, given
9038275SEric Cheng  * its mac handle, or macname or linkid.
9048275SEric Cheng  */
9058275SEric Cheng void
9068275SEric Cheng mac_perim_enter_by_mh(mac_handle_t mh, mac_perim_handle_t *mphp)
9078275SEric Cheng {
9088275SEric Cheng 	mac_impl_t	*mip = (mac_impl_t *)mh;
9098275SEric Cheng 
9108275SEric Cheng 	i_mac_perim_enter(mip);
9118275SEric Cheng 	/*
9128275SEric Cheng 	 * The mac_perim_handle_t returned encodes the 'mip' and whether a
9138275SEric Cheng 	 * mac_open has been done internally while entering the perimeter.
9148275SEric Cheng 	 * This information is used in mac_perim_exit
9158275SEric Cheng 	 */
9168275SEric Cheng 	MAC_ENCODE_MPH(*mphp, mip, 0);
9178275SEric Cheng }
9188275SEric Cheng 
9198275SEric Cheng int
9208275SEric Cheng mac_perim_enter_by_macname(const char *name, mac_perim_handle_t *mphp)
9218275SEric Cheng {
9228275SEric Cheng 	int	err;
9238275SEric Cheng 	mac_handle_t	mh;
9248275SEric Cheng 
9258275SEric Cheng 	if ((err = mac_open(name, &mh)) != 0)
9268275SEric Cheng 		return (err);
9278275SEric Cheng 
9288275SEric Cheng 	mac_perim_enter_by_mh(mh, mphp);
9298275SEric Cheng 	MAC_ENCODE_MPH(*mphp, mh, 1);
9308275SEric Cheng 	return (0);
9318275SEric Cheng }
9328275SEric Cheng 
9338275SEric Cheng int
9348275SEric Cheng mac_perim_enter_by_linkid(datalink_id_t linkid, mac_perim_handle_t *mphp)
9358275SEric Cheng {
9368275SEric Cheng 	int	err;
9378275SEric Cheng 	mac_handle_t	mh;
9388275SEric Cheng 
9398275SEric Cheng 	if ((err = mac_open_by_linkid(linkid, &mh)) != 0)
9408275SEric Cheng 		return (err);
9418275SEric Cheng 
9428275SEric Cheng 	mac_perim_enter_by_mh(mh, mphp);
9438275SEric Cheng 	MAC_ENCODE_MPH(*mphp, mh, 1);
9448275SEric Cheng 	return (0);
9458275SEric Cheng }
9468275SEric Cheng 
9478275SEric Cheng void
9488275SEric Cheng mac_perim_exit(mac_perim_handle_t mph)
9498275SEric Cheng {
9508275SEric Cheng 	mac_impl_t	*mip;
9518275SEric Cheng 	boolean_t	need_close;
9528275SEric Cheng 
9538275SEric Cheng 	MAC_DECODE_MPH(mph, mip, need_close);
9548275SEric Cheng 	i_mac_perim_exit(mip);
9558275SEric Cheng 	if (need_close)
9568275SEric Cheng 		mac_close((mac_handle_t)mip);
9578275SEric Cheng }
9588275SEric Cheng 
9598275SEric Cheng int
9605895Syz147064 mac_hold(const char *macname, mac_impl_t **pmip)
9610Sstevel@tonic-gate {
9620Sstevel@tonic-gate 	mac_impl_t	*mip;
9630Sstevel@tonic-gate 	int		err;
9640Sstevel@tonic-gate 
9650Sstevel@tonic-gate 	/*
9660Sstevel@tonic-gate 	 * Check the device name length to make sure it won't overflow our
9670Sstevel@tonic-gate 	 * buffer.
9680Sstevel@tonic-gate 	 */
9692311Sseb 	if (strlen(macname) >= MAXNAMELEN)
9700Sstevel@tonic-gate 		return (EINVAL);
9710Sstevel@tonic-gate 
9720Sstevel@tonic-gate 	/*
9735895Syz147064 	 * Look up its entry in the global hash table.
9740Sstevel@tonic-gate 	 */
9755895Syz147064 	rw_enter(&i_mac_impl_lock, RW_WRITER);
9765895Syz147064 	err = mod_hash_find(i_mac_impl_hash, (mod_hash_key_t)macname,
9775895Syz147064 	    (mod_hash_val_t *)&mip);
9785895Syz147064 
9795895Syz147064 	if (err != 0) {
9805895Syz147064 		rw_exit(&i_mac_impl_lock);
9815895Syz147064 		return (ENOENT);
9825895Syz147064 	}
9835895Syz147064 
9848275SEric Cheng 	if (mip->mi_state_flags & MIS_DISABLED) {
9855895Syz147064 		rw_exit(&i_mac_impl_lock);
9865895Syz147064 		return (ENOENT);
9875895Syz147064 	}
9885895Syz147064 
9898275SEric Cheng 	if (mip->mi_state_flags & MIS_EXCLUSIVE_HELD) {
9905895Syz147064 		rw_exit(&i_mac_impl_lock);
9915895Syz147064 		return (EBUSY);
9925895Syz147064 	}
9935895Syz147064 
9945895Syz147064 	mip->mi_ref++;
9955895Syz147064 	rw_exit(&i_mac_impl_lock);
9965895Syz147064 
9975895Syz147064 	*pmip = mip;
9985895Syz147064 	return (0);
9995895Syz147064 }
10005895Syz147064 
10018275SEric Cheng void
10025895Syz147064 mac_rele(mac_impl_t *mip)
10035895Syz147064 {
10045895Syz147064 	rw_enter(&i_mac_impl_lock, RW_WRITER);
10055895Syz147064 	ASSERT(mip->mi_ref != 0);
10068275SEric Cheng 	if (--mip->mi_ref == 0) {
10078275SEric Cheng 		ASSERT(mip->mi_nactiveclients == 0 &&
10088275SEric Cheng 		    !(mip->mi_state_flags & MIS_EXCLUSIVE));
10095895Syz147064 	}
10105895Syz147064 	rw_exit(&i_mac_impl_lock);
10115895Syz147064 }
10125895Syz147064 
10138275SEric Cheng /*
10148275SEric Cheng  * This function is called only by mac_client_open.
10158275SEric Cheng  */
10165895Syz147064 int
10178275SEric Cheng mac_start(mac_impl_t *mip)
10180Sstevel@tonic-gate {
10198275SEric Cheng 	int		err = 0;
10208275SEric Cheng 
10218275SEric Cheng 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
10222311Sseb 	ASSERT(mip->mi_start != NULL);
10230Sstevel@tonic-gate 
10240Sstevel@tonic-gate 	/*
10250Sstevel@tonic-gate 	 * Check whether the device is already started.
10260Sstevel@tonic-gate 	 */
10278275SEric Cheng 	if (mip->mi_active++ == 0) {
10288275SEric Cheng 		mac_ring_t *ring = NULL;
10298275SEric Cheng 
10308275SEric Cheng 		/*
10318275SEric Cheng 		 * Start the device.
10328275SEric Cheng 		 */
10338275SEric Cheng 		err = mip->mi_start(mip->mi_driver);
10348275SEric Cheng 		if (err != 0) {
10358275SEric Cheng 			mip->mi_active--;
10368275SEric Cheng 			return (err);
10378275SEric Cheng 		}
10388275SEric Cheng 
10390Sstevel@tonic-gate 		/*
10408275SEric Cheng 		 * Start the default tx ring.
10410Sstevel@tonic-gate 		 */
10428275SEric Cheng 		if (mip->mi_default_tx_ring != NULL) {
10438275SEric Cheng 
10448275SEric Cheng 			ring = (mac_ring_t *)mip->mi_default_tx_ring;
10458275SEric Cheng 			err = mac_start_ring(ring);
10468275SEric Cheng 			if (err != 0) {
10478275SEric Cheng 				mip->mi_active--;
10488275SEric Cheng 				return (err);
10498275SEric Cheng 			}
10508275SEric Cheng 			ring->mr_state = MR_INUSE;
10518275SEric Cheng 		}
10528275SEric Cheng 
10538275SEric Cheng 		if (mip->mi_rx_groups != NULL) {
10548275SEric Cheng 			/*
10558275SEric Cheng 			 * Start the default ring, since it will be needed
10568275SEric Cheng 			 * to receive broadcast and multicast traffic for
10578275SEric Cheng 			 * both primary and non-primary MAC clients.
10588275SEric Cheng 			 */
10598275SEric Cheng 			mac_group_t *grp = &mip->mi_rx_groups[0];
10608275SEric Cheng 
10618275SEric Cheng 			ASSERT(grp->mrg_state == MAC_GROUP_STATE_REGISTERED);
10628275SEric Cheng 			err = mac_start_group_and_rings(grp);
10638275SEric Cheng 			if (err != 0) {
10648275SEric Cheng 				mip->mi_active--;
10658275SEric Cheng 				if (ring != NULL) {
10668275SEric Cheng 					mac_stop_ring(ring);
10678275SEric Cheng 					ring->mr_state = MR_FREE;
10688275SEric Cheng 				}
10698275SEric Cheng 				return (err);
10708275SEric Cheng 			}
10718275SEric Cheng 			mac_set_rx_group_state(grp, MAC_GROUP_STATE_SHARED);
10728275SEric Cheng 		}
10730Sstevel@tonic-gate 	}
10740Sstevel@tonic-gate 
10750Sstevel@tonic-gate 	return (err);
10760Sstevel@tonic-gate }
10770Sstevel@tonic-gate 
10788275SEric Cheng /*
10798275SEric Cheng  * This function is called only by mac_client_close.
10808275SEric Cheng  */
10810Sstevel@tonic-gate void
10828275SEric Cheng mac_stop(mac_impl_t *mip)
10830Sstevel@tonic-gate {
10842311Sseb 	ASSERT(mip->mi_stop != NULL);
10858275SEric Cheng 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
10860Sstevel@tonic-gate 
10870Sstevel@tonic-gate 	/*
10880Sstevel@tonic-gate 	 * Check whether the device is still needed.
10890Sstevel@tonic-gate 	 */
10900Sstevel@tonic-gate 	ASSERT(mip->mi_active != 0);
10918275SEric Cheng 	if (--mip->mi_active == 0) {
10928275SEric Cheng 		if (mip->mi_rx_groups != NULL) {
10930Sstevel@tonic-gate 			/*
10948275SEric Cheng 			 * There should be no more active clients since the
10958275SEric Cheng 			 * MAC is being stopped. Stop the default RX group
10968275SEric Cheng 			 * and transition it back to registered state.
10970Sstevel@tonic-gate 			 */
10988275SEric Cheng 			mac_group_t *grp = &mip->mi_rx_groups[0];
10990Sstevel@tonic-gate 
11000Sstevel@tonic-gate 			/*
11018275SEric Cheng 			 * When clients are torn down, the groups
11028275SEric Cheng 			 * are release via mac_release_rx_group which
11038275SEric Cheng 			 * knows the the default group is always in
11048275SEric Cheng 			 * started mode since broadcast uses it. So
11058275SEric Cheng 			 * we can assert that their are no clients
11068275SEric Cheng 			 * (since mac_bcast_add doesn't register itself
11078275SEric Cheng 			 * as a client) and group is in SHARED state.
11080Sstevel@tonic-gate 			 */
11098275SEric Cheng 			ASSERT(grp->mrg_state == MAC_GROUP_STATE_SHARED);
11108275SEric Cheng 			ASSERT(MAC_RX_GROUP_NO_CLIENT(grp) &&
11118275SEric Cheng 			    mip->mi_nactiveclients == 0);
11128275SEric Cheng 			mac_stop_group_and_rings(grp);
11138275SEric Cheng 			mac_set_rx_group_state(grp, MAC_GROUP_STATE_REGISTERED);
11140Sstevel@tonic-gate 		}
11158275SEric Cheng 
11168275SEric Cheng 		if (mip->mi_default_tx_ring != NULL) {
11178275SEric Cheng 			mac_ring_t *ring;
11188275SEric Cheng 
11198275SEric Cheng 			ring = (mac_ring_t *)mip->mi_default_tx_ring;
11208275SEric Cheng 			mac_stop_ring(ring);
11218275SEric Cheng 			ring->mr_state = MR_FREE;
11228275SEric Cheng 		}
11238275SEric Cheng 
11248275SEric Cheng 		/*
11258275SEric Cheng 		 * Stop the device.
11268275SEric Cheng 		 */
11278275SEric Cheng 		mip->mi_stop(mip->mi_driver);
11282331Skrgopi 	}
11292331Skrgopi }
11302331Skrgopi 
11310Sstevel@tonic-gate int
11328275SEric Cheng i_mac_promisc_set(mac_impl_t *mip, boolean_t on, mac_promisc_type_t ptype)
11330Sstevel@tonic-gate {
11340Sstevel@tonic-gate 	int		err = 0;
11350Sstevel@tonic-gate 
11368275SEric Cheng 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
11372311Sseb 	ASSERT(mip->mi_setpromisc != NULL);
11380Sstevel@tonic-gate 	ASSERT(ptype == MAC_DEVPROMISC || ptype == MAC_PROMISC);
11390Sstevel@tonic-gate 
11400Sstevel@tonic-gate 	/*
11410Sstevel@tonic-gate 	 * Determine whether we should enable or disable promiscuous mode.
11420Sstevel@tonic-gate 	 * For details on the distinction between "device promiscuous mode"
11430Sstevel@tonic-gate 	 * and "MAC promiscuous mode", see PSARC/2005/289.
11440Sstevel@tonic-gate 	 */
11450Sstevel@tonic-gate 	if (on) {
11460Sstevel@tonic-gate 		/*
11470Sstevel@tonic-gate 		 * Enable promiscuous mode on the device if not yet enabled.
11480Sstevel@tonic-gate 		 */
11490Sstevel@tonic-gate 		if (mip->mi_devpromisc++ == 0) {
11502311Sseb 			err = mip->mi_setpromisc(mip->mi_driver, B_TRUE);
11512311Sseb 			if (err != 0) {
11520Sstevel@tonic-gate 				mip->mi_devpromisc--;
11538275SEric Cheng 				return (err);
11540Sstevel@tonic-gate 			}
11550Sstevel@tonic-gate 			i_mac_notify(mip, MAC_NOTE_DEVPROMISC);
11560Sstevel@tonic-gate 		}
11570Sstevel@tonic-gate 
11580Sstevel@tonic-gate 		/*
11590Sstevel@tonic-gate 		 * Enable promiscuous mode on the MAC if not yet enabled.
11600Sstevel@tonic-gate 		 */
11610Sstevel@tonic-gate 		if (ptype == MAC_PROMISC && mip->mi_promisc++ == 0)
11620Sstevel@tonic-gate 			i_mac_notify(mip, MAC_NOTE_PROMISC);
11630Sstevel@tonic-gate 	} else {
11648275SEric Cheng 		if (mip->mi_devpromisc == 0)
11658275SEric Cheng 			return (EPROTO);
11668275SEric Cheng 
11670Sstevel@tonic-gate 		/*
11680Sstevel@tonic-gate 		 * Disable promiscuous mode on the device if this is the last
11690Sstevel@tonic-gate 		 * enabling.
11700Sstevel@tonic-gate 		 */
11710Sstevel@tonic-gate 		if (--mip->mi_devpromisc == 0) {
11722311Sseb 			err = mip->mi_setpromisc(mip->mi_driver, B_FALSE);
11732311Sseb 			if (err != 0) {
11740Sstevel@tonic-gate 				mip->mi_devpromisc++;
11758275SEric Cheng 				return (err);
11760Sstevel@tonic-gate 			}
11770Sstevel@tonic-gate 			i_mac_notify(mip, MAC_NOTE_DEVPROMISC);
11780Sstevel@tonic-gate 		}
11790Sstevel@tonic-gate 
11800Sstevel@tonic-gate 		/*
11810Sstevel@tonic-gate 		 * Disable promiscuous mode on the MAC if this is the last
11820Sstevel@tonic-gate 		 * enabling.
11830Sstevel@tonic-gate 		 */
11840Sstevel@tonic-gate 		if (ptype == MAC_PROMISC && --mip->mi_promisc == 0)
11850Sstevel@tonic-gate 			i_mac_notify(mip, MAC_NOTE_PROMISC);
11860Sstevel@tonic-gate 	}
11870Sstevel@tonic-gate 
11888275SEric Cheng 	return (0);
11890Sstevel@tonic-gate }
11900Sstevel@tonic-gate 
11918275SEric Cheng int
11928275SEric Cheng mac_promisc_set(mac_handle_t mh, boolean_t on, mac_promisc_type_t ptype)
11938275SEric Cheng {
11948275SEric Cheng 	mac_impl_t	*mip = (mac_impl_t *)mh;
11958275SEric Cheng 	int		rv;
11968275SEric Cheng 
11978275SEric Cheng 	i_mac_perim_enter(mip);
11988275SEric Cheng 	rv = i_mac_promisc_set(mip, on, ptype);
11998275SEric Cheng 	i_mac_perim_exit(mip);
12008275SEric Cheng 
12018275SEric Cheng 	return (rv);
12028275SEric Cheng }
12038275SEric Cheng 
12048275SEric Cheng /*
12058275SEric Cheng  * The promiscuity state can change any time. If the caller needs to take
12068275SEric Cheng  * actions that are atomic with the promiscuity state, then the caller needs
12078275SEric Cheng  * to bracket the entire sequence with mac_perim_enter/exit
12088275SEric Cheng  */
12090Sstevel@tonic-gate boolean_t
12100Sstevel@tonic-gate mac_promisc_get(mac_handle_t mh, mac_promisc_type_t ptype)
12110Sstevel@tonic-gate {
12120Sstevel@tonic-gate 	mac_impl_t		*mip = (mac_impl_t *)mh;
12130Sstevel@tonic-gate 
12140Sstevel@tonic-gate 	ASSERT(ptype == MAC_DEVPROMISC || ptype == MAC_PROMISC);
12150Sstevel@tonic-gate 
12160Sstevel@tonic-gate 	/*
12170Sstevel@tonic-gate 	 * Return the current promiscuity.
12180Sstevel@tonic-gate 	 */
12190Sstevel@tonic-gate 	if (ptype == MAC_DEVPROMISC)
12200Sstevel@tonic-gate 		return (mip->mi_devpromisc != 0);
12210Sstevel@tonic-gate 	else
12220Sstevel@tonic-gate 		return (mip->mi_promisc != 0);
12230Sstevel@tonic-gate }
12240Sstevel@tonic-gate 
12258275SEric Cheng /*
12268275SEric Cheng  * Invoked at MAC instance attach time to initialize the list
12278275SEric Cheng  * of factory MAC addresses supported by a MAC instance. This function
12288275SEric Cheng  * builds a local cache in the mac_impl_t for the MAC addresses
12298275SEric Cheng  * supported by the underlying hardware. The MAC clients themselves
12308275SEric Cheng  * use the mac_addr_factory*() functions to query and reserve
12318275SEric Cheng  * factory MAC addresses.
12328275SEric Cheng  */
12330Sstevel@tonic-gate void
12348275SEric Cheng mac_addr_factory_init(mac_impl_t *mip)
12355903Ssowmini {
12368275SEric Cheng 	mac_capab_multifactaddr_t capab;
12378275SEric Cheng 	uint8_t *addr;
12388275SEric Cheng 	int i;
12390Sstevel@tonic-gate 
12400Sstevel@tonic-gate 	/*
12418275SEric Cheng 	 * First round to see how many factory MAC addresses are available.
12420Sstevel@tonic-gate 	 */
12438275SEric Cheng 	bzero(&capab, sizeof (capab));
12448275SEric Cheng 	if (!i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_MULTIFACTADDR,
12458275SEric Cheng 	    &capab) || (capab.mcm_naddr == 0)) {
12466512Ssowmini 		/*
12478275SEric Cheng 		 * The MAC instance doesn't support multiple factory
12488275SEric Cheng 		 * MAC addresses, we're done here.
12496512Ssowmini 		 */
12506512Ssowmini 		return;
12515903Ssowmini 	}
12526512Ssowmini 
12530Sstevel@tonic-gate 	/*
12548275SEric Cheng 	 * Allocate the space and get all the factory addresses.
125556Smeem 	 */
12568275SEric Cheng 	addr = kmem_alloc(capab.mcm_naddr * MAXMACADDRLEN, KM_SLEEP);
12578275SEric Cheng 	capab.mcm_getaddr(mip->mi_driver, capab.mcm_naddr, addr);
12588275SEric Cheng 
12598275SEric Cheng 	mip->mi_factory_addr_num = capab.mcm_naddr;
12608275SEric Cheng 	mip->mi_factory_addr = kmem_zalloc(mip->mi_factory_addr_num *
12618275SEric Cheng 	    sizeof (mac_factory_addr_t), KM_SLEEP);
12628275SEric Cheng 
12638275SEric Cheng 	for (i = 0; i < capab.mcm_naddr; i++) {
12648275SEric Cheng 		bcopy(addr + i * MAXMACADDRLEN,
12658275SEric Cheng 		    mip->mi_factory_addr[i].mfa_addr,
12668275SEric Cheng 		    mip->mi_type->mt_addr_length);
12678275SEric Cheng 		mip->mi_factory_addr[i].mfa_in_use = B_FALSE;
126856Smeem 	}
126956Smeem 
12708275SEric Cheng 	kmem_free(addr, capab.mcm_naddr * MAXMACADDRLEN);
12718275SEric Cheng }
12728275SEric Cheng 
12738275SEric Cheng void
12748275SEric Cheng mac_addr_factory_fini(mac_impl_t *mip)
12758275SEric Cheng {
12768275SEric Cheng 	if (mip->mi_factory_addr == NULL) {
12778275SEric Cheng 		ASSERT(mip->mi_factory_addr_num == 0);
12788275SEric Cheng 		return;
12798275SEric Cheng 	}
12808275SEric Cheng 
12818275SEric Cheng 	kmem_free(mip->mi_factory_addr, mip->mi_factory_addr_num *
12828275SEric Cheng 	    sizeof (mac_factory_addr_t));
12838275SEric Cheng 
12848275SEric Cheng 	mip->mi_factory_addr = NULL;
12858275SEric Cheng 	mip->mi_factory_addr_num = 0;
12860Sstevel@tonic-gate }
12870Sstevel@tonic-gate 
12885084Sjohnlev /*
12898275SEric Cheng  * Reserve a factory MAC address. If *slot is set to -1, the function
12908275SEric Cheng  * attempts to reserve any of the available factory MAC addresses and
12918275SEric Cheng  * returns the reserved slot id. If no slots are available, the function
12928275SEric Cheng  * returns ENOSPC. If *slot is not set to -1, the function reserves
12938275SEric Cheng  * the specified slot if it is available, or returns EBUSY is the slot
12948275SEric Cheng  * is already used. Returns ENOTSUP if the underlying MAC does not
12958275SEric Cheng  * support multiple factory addresses. If the slot number is not -1 but
12968275SEric Cheng  * is invalid, returns EINVAL.
12978275SEric Cheng  */
12988275SEric Cheng int
12998275SEric Cheng mac_addr_factory_reserve(mac_client_handle_t mch, int *slot)
13008275SEric Cheng {
13018275SEric Cheng 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
13028275SEric Cheng 	mac_impl_t *mip = mcip->mci_mip;
13038275SEric Cheng 	int i, ret = 0;
13048275SEric Cheng 
13058275SEric Cheng 	i_mac_perim_enter(mip);
13068275SEric Cheng 	/*
13078275SEric Cheng 	 * Protect against concurrent readers that may need a self-consistent
13088275SEric Cheng 	 * view of the factory addresses
13098275SEric Cheng 	 */
13108275SEric Cheng 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
13118275SEric Cheng 
13128275SEric Cheng 	if (mip->mi_factory_addr_num == 0) {
13138275SEric Cheng 		ret = ENOTSUP;
13148275SEric Cheng 		goto bail;
13158275SEric Cheng 	}
13168275SEric Cheng 
13178275SEric Cheng 	if (*slot != -1) {
13188275SEric Cheng 		/* check the specified slot */
13198275SEric Cheng 		if (*slot < 1 || *slot > mip->mi_factory_addr_num) {
13208275SEric Cheng 			ret = EINVAL;
13218275SEric Cheng 			goto bail;
13228275SEric Cheng 		}
13238275SEric Cheng 		if (mip->mi_factory_addr[*slot-1].mfa_in_use) {
13248275SEric Cheng 			ret = EBUSY;
13258275SEric Cheng 			goto bail;
13268275SEric Cheng 		}
13278275SEric Cheng 	} else {
13288275SEric Cheng 		/* pick the next available slot */
13298275SEric Cheng 		for (i = 0; i < mip->mi_factory_addr_num; i++) {
13308275SEric Cheng 			if (!mip->mi_factory_addr[i].mfa_in_use)
13318275SEric Cheng 				break;
13328275SEric Cheng 		}
13338275SEric Cheng 
13348275SEric Cheng 		if (i == mip->mi_factory_addr_num) {
13358275SEric Cheng 			ret = ENOSPC;
13368275SEric Cheng 			goto bail;
13378275SEric Cheng 		}
13388275SEric Cheng 		*slot = i+1;
13398275SEric Cheng 	}
13408275SEric Cheng 
13418275SEric Cheng 	mip->mi_factory_addr[*slot-1].mfa_in_use = B_TRUE;
13428275SEric Cheng 	mip->mi_factory_addr[*slot-1].mfa_client = mcip;
13438275SEric Cheng 
13448275SEric Cheng bail:
13458275SEric Cheng 	rw_exit(&mip->mi_rw_lock);
13468275SEric Cheng 	i_mac_perim_exit(mip);
13478275SEric Cheng 	return (ret);
13488275SEric Cheng }
13498275SEric Cheng 
13508275SEric Cheng /*
13518275SEric Cheng  * Release the specified factory MAC address slot.
13525084Sjohnlev  */
13538275SEric Cheng void
13548275SEric Cheng mac_addr_factory_release(mac_client_handle_t mch, uint_t slot)
13558275SEric Cheng {
13568275SEric Cheng 	mac_client_impl_t *mcip = (mac_client_impl_t *)mch;
13578275SEric Cheng 	mac_impl_t *mip = mcip->mci_mip;
13588275SEric Cheng 
13598275SEric Cheng 	i_mac_perim_enter(mip);
13608275SEric Cheng 	/*
13618275SEric Cheng 	 * Protect against concurrent readers that may need a self-consistent
13628275SEric Cheng 	 * view of the factory addresses
13638275SEric Cheng 	 */
13648275SEric Cheng 	rw_enter(&mip->mi_rw_lock, RW_WRITER);
13658275SEric Cheng 
13668275SEric Cheng 	ASSERT(slot > 0 && slot <= mip->mi_factory_addr_num);
13678275SEric Cheng 	ASSERT(mip->mi_factory_addr[slot-1].mfa_in_use);
13688275SEric Cheng 
13698275SEric Cheng 	mip->mi_factory_addr[slot-1].mfa_in_use = B_FALSE;
13708275SEric Cheng 
13718275SEric Cheng 	rw_exit(&mip->mi_rw_lock);
13728275SEric Cheng 	i_mac_perim_exit(mip);
13738275SEric Cheng }
13748275SEric Cheng 
13758275SEric Cheng /*
13768275SEric Cheng  * Stores in mac_addr the value of the specified MAC address. Returns
13778275SEric Cheng  * 0 on success, or EINVAL if the slot number is not valid for the MAC.
13788275SEric Cheng  * The caller must provide a string of at least MAXNAMELEN bytes.
13798275SEric Cheng  */
13808275SEric Cheng void
13818275SEric Cheng mac_addr_factory_value(mac_handle_t mh, int slot, uchar_t *mac_addr,
13828275SEric Cheng     uint_t *addr_len, char *client_name, boolean_t *in_use_arg)
13835084Sjohnlev {
13848275SEric Cheng 	mac_impl_t *mip = (mac_impl_t *)mh;
13858275SEric Cheng 	boolean_t in_use;
13868275SEric Cheng 
13878275SEric Cheng 	ASSERT(slot > 0 && slot <= mip->mi_factory_addr_num);
13888275SEric Cheng 
13898275SEric Cheng 	/*
13908275SEric Cheng 	 * Readers need to hold mi_rw_lock. Writers need to hold mac perimeter
13918275SEric Cheng 	 * and mi_rw_lock
13928275SEric Cheng 	 */
13938275SEric Cheng 	rw_enter(&mip->mi_rw_lock, RW_READER);
13948275SEric Cheng 	bcopy(mip->mi_factory_addr[slot-1].mfa_addr, mac_addr, MAXMACADDRLEN);
13958275SEric Cheng 	*addr_len = mip->mi_type->mt_addr_length;
13968275SEric Cheng 	in_use = mip->mi_factory_addr[slot-1].mfa_in_use;
13978275SEric Cheng 	if (in_use && client_name != NULL) {
13988275SEric Cheng 		bcopy(mip->mi_factory_addr[slot-1].mfa_client->mci_name,
13998275SEric Cheng 		    client_name, MAXNAMELEN);
14008275SEric Cheng 	}
14018275SEric Cheng 	if (in_use_arg != NULL)
14028275SEric Cheng 		*in_use_arg = in_use;
14038275SEric Cheng 	rw_exit(&mip->mi_rw_lock);
14048275SEric Cheng }
14058275SEric Cheng 
14068275SEric Cheng /*
14078275SEric Cheng  * Returns the number of factory MAC addresses (in addition to the
14088275SEric Cheng  * primary MAC address), 0 if the underlying MAC doesn't support
14098275SEric Cheng  * that feature.
14108275SEric Cheng  */
14118275SEric Cheng uint_t
14128275SEric Cheng mac_addr_factory_num(mac_handle_t mh)
14138275SEric Cheng {
14148275SEric Cheng 	mac_impl_t *mip = (mac_impl_t *)mh;
14158275SEric Cheng 
14168275SEric Cheng 	return (mip->mi_factory_addr_num);
14178275SEric Cheng }
14188275SEric Cheng 
14198275SEric Cheng 
14208275SEric Cheng void
14218275SEric Cheng mac_rx_group_unmark(mac_group_t *grp, uint_t flag)
14228275SEric Cheng {
14238275SEric Cheng 	mac_ring_t	*ring;
14248275SEric Cheng 
14258275SEric Cheng 	for (ring = grp->mrg_rings; ring != NULL; ring = ring->mr_next)
14268275SEric Cheng 		ring->mr_flag &= ~flag;
14275084Sjohnlev }
14285084Sjohnlev 
14295084Sjohnlev /*
14308275SEric Cheng  * The following mac_hwrings_xxx() functions are private mac client functions
14318275SEric Cheng  * used by the aggr driver to access and control the underlying HW Rx group
14328275SEric Cheng  * and rings. In this case, the aggr driver has exclusive control of the
14338275SEric Cheng  * underlying HW Rx group/rings, it calls the following functions to
14348275SEric Cheng  * start/stop the HW Rx rings, disable/enable polling, add/remove mac'
14358275SEric Cheng  * addresses, or set up the Rx callback.
14365084Sjohnlev  */
14378275SEric Cheng /* ARGSUSED */
14388275SEric Cheng static void
14398275SEric Cheng mac_hwrings_rx_process(void *arg, mac_resource_handle_t srs,
14408275SEric Cheng     mblk_t *mp_chain, boolean_t loopback)
14410Sstevel@tonic-gate {
14428275SEric Cheng 	mac_soft_ring_set_t	*mac_srs = (mac_soft_ring_set_t *)srs;
14438275SEric Cheng 	mac_srs_rx_t		*srs_rx = &mac_srs->srs_rx;
14448275SEric Cheng 	mac_direct_rx_t		proc;
14458275SEric Cheng 	void			*arg1;
14468275SEric Cheng 	mac_resource_handle_t	arg2;
14478275SEric Cheng 
14488275SEric Cheng 	proc = srs_rx->sr_func;
14498275SEric Cheng 	arg1 = srs_rx->sr_arg1;
14508275SEric Cheng 	arg2 = mac_srs->srs_mrh;
14518275SEric Cheng 
14528275SEric Cheng 	proc(arg1, arg2, mp_chain, NULL);
14530Sstevel@tonic-gate }
14540Sstevel@tonic-gate 
14558275SEric Cheng /*
14568275SEric Cheng  * This function is called to get the list of HW rings that are reserved by
14578275SEric Cheng  * an exclusive mac client.
14588275SEric Cheng  *
14598275SEric Cheng  * Return value: the number of HW rings.
14608275SEric Cheng  */
14618275SEric Cheng int
14628275SEric Cheng mac_hwrings_get(mac_client_handle_t mch, mac_group_handle_t *hwgh,
14638275SEric Cheng     mac_ring_handle_t *hwrh)
14640Sstevel@tonic-gate {
14658275SEric Cheng 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
14668275SEric Cheng 	flow_entry_t		*flent = mcip->mci_flent;
14678275SEric Cheng 	mac_group_t		*grp = flent->fe_rx_ring_group;
14688275SEric Cheng 	mac_ring_t		*ring;
14698275SEric Cheng 	int			cnt = 0;
14700Sstevel@tonic-gate 
14710Sstevel@tonic-gate 	/*
14728275SEric Cheng 	 * The mac client did not reserve any RX group, return directly.
14738275SEric Cheng 	 * This is probably because the underlying MAC does not support
14748275SEric Cheng 	 * any RX groups.
14758275SEric Cheng 	 */
14768275SEric Cheng 	*hwgh = NULL;
14778275SEric Cheng 	if (grp == NULL)
14788275SEric Cheng 		return (0);
14798275SEric Cheng 
14808275SEric Cheng 	/*
14818275SEric Cheng 	 * This RX group must be reserved by this mac client.
14820Sstevel@tonic-gate 	 */
14838275SEric Cheng 	ASSERT((grp->mrg_state == MAC_GROUP_STATE_RESERVED) &&
14848275SEric Cheng 	    (mch == (mac_client_handle_t)(MAC_RX_GROUP_ONLY_CLIENT(grp))));
14858275SEric Cheng 
14868275SEric Cheng 	for (ring = grp->mrg_rings; ring != NULL; ring = ring->mr_next) {
14878275SEric Cheng 		ASSERT(cnt < MAX_RINGS_PER_GROUP);
14888275SEric Cheng 		hwrh[cnt++] = (mac_ring_handle_t)ring;
14898275SEric Cheng 	}
14908275SEric Cheng 	*hwgh = (mac_group_handle_t)grp;
14918275SEric Cheng 	return (cnt);
14928275SEric Cheng }
14938275SEric Cheng 
14948275SEric Cheng /*
14958275SEric Cheng  * Setup the RX callback of the mac client which exclusively controls HW ring.
14968275SEric Cheng  */
14978275SEric Cheng void
14988275SEric Cheng mac_hwring_setup(mac_ring_handle_t hwrh, mac_resource_handle_t prh)
14998275SEric Cheng {
15008275SEric Cheng 	mac_ring_t		*hw_ring = (mac_ring_t *)hwrh;
15018275SEric Cheng 	mac_soft_ring_set_t	*mac_srs = hw_ring->mr_srs;
15028275SEric Cheng 
15038275SEric Cheng 	mac_srs->srs_mrh = prh;
15048275SEric Cheng 	mac_srs->srs_rx.sr_lower_proc = mac_hwrings_rx_process;
15050Sstevel@tonic-gate }
15060Sstevel@tonic-gate 
15070Sstevel@tonic-gate void
15088275SEric Cheng mac_hwring_teardown(mac_ring_handle_t hwrh)
15098275SEric Cheng {
15108275SEric Cheng 	mac_ring_t		*hw_ring = (mac_ring_t *)hwrh;
15118275SEric Cheng 	mac_soft_ring_set_t	*mac_srs = hw_ring->mr_srs;
15128275SEric Cheng 
15138275SEric Cheng 	mac_srs->srs_rx.sr_lower_proc = mac_rx_srs_process;
15148275SEric Cheng 	mac_srs->srs_mrh = NULL;
15158275SEric Cheng }
15168275SEric Cheng 
15178275SEric Cheng int
15188275SEric Cheng mac_hwring_disable_intr(mac_ring_handle_t rh)
15190Sstevel@tonic-gate {
15208275SEric Cheng 	mac_ring_t *rr_ring = (mac_ring_t *)rh;
15218275SEric Cheng 	mac_intr_t *intr = &rr_ring->mr_info.mri_intr;
15228275SEric Cheng 
15238275SEric Cheng 	return (intr->mi_disable(intr->mi_handle));
15248275SEric Cheng }
15258275SEric Cheng 
15268275SEric Cheng int
15278275SEric Cheng mac_hwring_enable_intr(mac_ring_handle_t rh)
15288275SEric Cheng {
15298275SEric Cheng 	mac_ring_t *rr_ring = (mac_ring_t *)rh;
15308275SEric Cheng 	mac_intr_t *intr = &rr_ring->mr_info.mri_intr;
15318275SEric Cheng 
15328275SEric Cheng 	return (intr->mi_enable(intr->mi_handle));
15338275SEric Cheng }
15348275SEric Cheng 
15358275SEric Cheng int
15368275SEric Cheng mac_hwring_start(mac_ring_handle_t rh)
15378275SEric Cheng {
15388275SEric Cheng 	mac_ring_t *rr_ring = (mac_ring_t *)rh;
15398275SEric Cheng 
15408275SEric Cheng 	MAC_RING_UNMARK(rr_ring, MR_QUIESCE);
15418275SEric Cheng 	return (0);
15420Sstevel@tonic-gate }
15430Sstevel@tonic-gate 
15440Sstevel@tonic-gate void
15458275SEric Cheng mac_hwring_stop(mac_ring_handle_t rh)
15468275SEric Cheng {
15478275SEric Cheng 	mac_ring_t *rr_ring = (mac_ring_t *)rh;
15488275SEric Cheng 
15498275SEric Cheng 	mac_rx_ring_quiesce(rr_ring, MR_QUIESCE);
15508275SEric Cheng }
15518275SEric Cheng 
15528275SEric Cheng mblk_t *
15538275SEric Cheng mac_hwring_poll(mac_ring_handle_t rh, int bytes_to_pickup)
15548275SEric Cheng {
15558275SEric Cheng 	mac_ring_t *rr_ring = (mac_ring_t *)rh;
15568275SEric Cheng 	mac_ring_info_t *info = &rr_ring->mr_info;
15578275SEric Cheng 
15588275SEric Cheng 	return (info->mri_poll(info->mri_driver, bytes_to_pickup));
15598275SEric Cheng }
15608275SEric Cheng 
15618275SEric Cheng int
15628275SEric Cheng mac_hwgroup_addmac(mac_group_handle_t gh, const uint8_t *addr)
15638275SEric Cheng {
15648275SEric Cheng 	mac_group_t *group = (mac_group_t *)gh;
15658275SEric Cheng 
15668275SEric Cheng 	return (mac_group_addmac(group, addr));
15678275SEric Cheng }
15688275SEric Cheng 
15698275SEric Cheng int
15708275SEric Cheng mac_hwgroup_remmac(mac_group_handle_t gh, const uint8_t *addr)
15718275SEric Cheng {
15728275SEric Cheng 	mac_group_t *group = (mac_group_t *)gh;
15738275SEric Cheng 
15748275SEric Cheng 	return (mac_group_remmac(group, addr));
15758275SEric Cheng }
15768275SEric Cheng 
15778275SEric Cheng /*
15788275SEric Cheng  * Set the RX group to be shared/reserved. Note that the group must be
15798275SEric Cheng  * started/stopped outside of this function.
15808275SEric Cheng  */
15818275SEric Cheng void
15828275SEric Cheng mac_set_rx_group_state(mac_group_t *grp, mac_group_state_t state)
15830Sstevel@tonic-gate {
15848275SEric Cheng 	/*
15858275SEric Cheng 	 * If there is no change in the group state, just return.
15868275SEric Cheng 	 */
15878275SEric Cheng 	if (grp->mrg_state == state)
15888275SEric Cheng 		return;
15898275SEric Cheng 
15908275SEric Cheng 	switch (state) {
15918275SEric Cheng 	case MAC_GROUP_STATE_RESERVED:
15928275SEric Cheng 		/*
15938275SEric Cheng 		 * Successfully reserved the group.
15948275SEric Cheng 		 *
15958275SEric Cheng 		 * Given that there is an exclusive client controlling this
15968275SEric Cheng 		 * group, we enable the group level polling when available,
15978275SEric Cheng 		 * so that SRSs get to turn on/off individual rings they's
15988275SEric Cheng 		 * assigned to.
15998275SEric Cheng 		 */
16008275SEric Cheng 		ASSERT(MAC_PERIM_HELD(grp->mrg_mh));
16018275SEric Cheng 
16028275SEric Cheng 		if (GROUP_INTR_DISABLE_FUNC(grp) != NULL)
16038275SEric Cheng 			GROUP_INTR_DISABLE_FUNC(grp)(GROUP_INTR_HANDLE(grp));
16048275SEric Cheng 
16058275SEric Cheng 		break;
16068275SEric Cheng 
16078275SEric Cheng 	case MAC_GROUP_STATE_SHARED:
16088275SEric Cheng 		/*
16098275SEric Cheng 		 * Set all rings of this group to software classified.
16108275SEric Cheng 		 * If the group has an overriding interrupt, then re-enable it.
16118275SEric Cheng 		 */
16128275SEric Cheng 		ASSERT(MAC_PERIM_HELD(grp->mrg_mh));
16138275SEric Cheng 
16148275SEric Cheng 		if (GROUP_INTR_ENABLE_FUNC(grp) != NULL)
16158275SEric Cheng 			GROUP_INTR_ENABLE_FUNC(grp)(GROUP_INTR_HANDLE(grp));
16168275SEric Cheng 
16178275SEric Cheng 		/* The ring is not available for reservations any more */
16188275SEric Cheng 		break;
16198275SEric Cheng 
16208275SEric Cheng 	case MAC_GROUP_STATE_REGISTERED:
16218275SEric Cheng 		/* Also callable from mac_register, perim is not held */
16228275SEric Cheng 		break;
16238275SEric Cheng 
16248275SEric Cheng 	default:
16258275SEric Cheng 		ASSERT(B_FALSE);
16268275SEric Cheng 		break;
16278275SEric Cheng 	}
16288275SEric Cheng 
16298275SEric Cheng 	grp->mrg_state = state;
16308275SEric Cheng }
16318275SEric Cheng 
16328275SEric Cheng /*
16338275SEric Cheng  * Quiesce future hardware classified packets for the specified Rx ring
16348275SEric Cheng  */
16358275SEric Cheng static void
16368275SEric Cheng mac_rx_ring_quiesce(mac_ring_t *rx_ring, uint_t ring_flag)
16378275SEric Cheng {
16388275SEric Cheng 	ASSERT(rx_ring->mr_classify_type == MAC_HW_CLASSIFIER);
16398275SEric Cheng 	ASSERT(ring_flag == MR_CONDEMNED || ring_flag  == MR_QUIESCE);
16408275SEric Cheng 
16418275SEric Cheng 	mutex_enter(&rx_ring->mr_lock);
16428275SEric Cheng 	rx_ring->mr_flag |= ring_flag;
16438275SEric Cheng 	while (rx_ring->mr_refcnt != 0)
16448275SEric Cheng 		cv_wait(&rx_ring->mr_cv, &rx_ring->mr_lock);
16458275SEric Cheng 	mutex_exit(&rx_ring->mr_lock);
16460Sstevel@tonic-gate }
16470Sstevel@tonic-gate 
16484913Sethindra /*
16498275SEric Cheng  * Please see mac_tx for details about the per cpu locking scheme
16504913Sethindra  */
16518275SEric Cheng static void
16528275SEric Cheng mac_tx_lock_all(mac_client_impl_t *mcip)
16538275SEric Cheng {
16548275SEric Cheng 	int	i;
16558275SEric Cheng 
16568275SEric Cheng 	for (i = 0; i <= mac_tx_percpu_cnt; i++)
16578275SEric Cheng 		mutex_enter(&mcip->mci_tx_pcpu[i].pcpu_tx_lock);
16588275SEric Cheng }
16598275SEric Cheng 
16608275SEric Cheng static void
16618275SEric Cheng mac_tx_unlock_all(mac_client_impl_t *mcip)
16628275SEric Cheng {
16638275SEric Cheng 	int	i;
16648275SEric Cheng 
16658275SEric Cheng 	for (i = mac_tx_percpu_cnt; i >= 0; i--)
16668275SEric Cheng 		mutex_exit(&mcip->mci_tx_pcpu[i].pcpu_tx_lock);
16678275SEric Cheng }
16688275SEric Cheng 
16698275SEric Cheng static void
16708275SEric Cheng mac_tx_unlock_allbutzero(mac_client_impl_t *mcip)
16718275SEric Cheng {
16728275SEric Cheng 	int	i;
16738275SEric Cheng 
16748275SEric Cheng 	for (i = mac_tx_percpu_cnt; i > 0; i--)
16758275SEric Cheng 		mutex_exit(&mcip->mci_tx_pcpu[i].pcpu_tx_lock);
16768275SEric Cheng }
16778275SEric Cheng 
16788275SEric Cheng static int
16798275SEric Cheng mac_tx_sum_refcnt(mac_client_impl_t *mcip)
16800Sstevel@tonic-gate {
16818275SEric Cheng 	int	i;
16828275SEric Cheng 	int	refcnt = 0;
16838275SEric Cheng 
16848275SEric Cheng 	for (i = 0; i <= mac_tx_percpu_cnt; i++)
16858275SEric Cheng 		refcnt += mcip->mci_tx_pcpu[i].pcpu_tx_refcnt;
16868275SEric Cheng 
16878275SEric Cheng 	return (refcnt);
16880Sstevel@tonic-gate }
16890Sstevel@tonic-gate 
16908275SEric Cheng /*
16918275SEric Cheng  * Stop future Tx packets coming down from the client in preparation for
16928275SEric Cheng  * quiescing the Tx side. This is needed for dynamic reclaim and reassignment
16938275SEric Cheng  * of rings between clients
16948275SEric Cheng  */
16958275SEric Cheng void
16968275SEric Cheng mac_tx_client_block(mac_client_impl_t *mcip)
16975084Sjohnlev {
16988275SEric Cheng 	mac_tx_lock_all(mcip);
16998275SEric Cheng 	mcip->mci_tx_flag |= MCI_TX_QUIESCE;
17008275SEric Cheng 	while (mac_tx_sum_refcnt(mcip) != 0) {
17018275SEric Cheng 		mac_tx_unlock_allbutzero(mcip);
17028275SEric Cheng 		cv_wait(&mcip->mci_tx_cv, &mcip->mci_tx_pcpu[0].pcpu_tx_lock);
17038275SEric Cheng 		mutex_exit(&mcip->mci_tx_pcpu[0].pcpu_tx_lock);
17048275SEric Cheng 		mac_tx_lock_all(mcip);
17058275SEric Cheng 	}
17068275SEric Cheng 	mac_tx_unlock_all(mcip);
17075084Sjohnlev }
17085084Sjohnlev 
17098275SEric Cheng void
17108275SEric Cheng mac_tx_client_unblock(mac_client_impl_t *mcip)
17115084Sjohnlev {
17128275SEric Cheng 	mac_tx_lock_all(mcip);
17138275SEric Cheng 	mcip->mci_tx_flag &= ~MCI_TX_QUIESCE;
17148275SEric Cheng 	mac_tx_unlock_all(mcip);
17155084Sjohnlev }
17165084Sjohnlev 
17170Sstevel@tonic-gate /*
17188275SEric Cheng  * Wait for an SRS to quiesce. The SRS worker will signal us when the
17198275SEric Cheng  * quiesce is done.
17208275SEric Cheng  */
17218275SEric Cheng static void
17228275SEric Cheng mac_srs_quiesce_wait(mac_soft_ring_set_t *srs, uint_t srs_flag)
17238275SEric Cheng {
17248275SEric Cheng 	mutex_enter(&srs->srs_lock);
17258275SEric Cheng 	while (!(srs->srs_state & srs_flag))
17268275SEric Cheng 		cv_wait(&srs->srs_quiesce_done_cv, &srs->srs_lock);
17278275SEric Cheng 	mutex_exit(&srs->srs_lock);
17288275SEric Cheng }
17298275SEric Cheng 
17308275SEric Cheng /*
17318275SEric Cheng  * Quiescing an Rx SRS is achieved by the following sequence. The protocol
17328275SEric Cheng  * works bottom up by cutting off packet flow from the bottommost point in the
17338275SEric Cheng  * mac, then the SRS, and then the soft rings. There are 2 use cases of this
17348275SEric Cheng  * mechanism. One is a temporary quiesce of the SRS, such as say while changing
17358275SEric Cheng  * the Rx callbacks. Another use case is Rx SRS teardown. In the former case
17368275SEric Cheng  * the QUIESCE prefix/suffix is used and in the latter the CONDEMNED is used
17378275SEric Cheng  * for the SRS and MR flags. In the former case the threads pause waiting for
17388275SEric Cheng  * a restart, while in the latter case the threads exit. The Tx SRS teardown
17398275SEric Cheng  * is also mostly similar to the above.
17408275SEric Cheng  *
17418275SEric Cheng  * 1. Stop future hardware classified packets at the lowest level in the mac.
17428275SEric Cheng  *    Remove any hardware classification rule (CONDEMNED case) and mark the
17438275SEric Cheng  *    rings as CONDEMNED or QUIESCE as appropriate. This prevents the mr_refcnt
17448275SEric Cheng  *    from increasing. Upcalls from the driver that come through hardware
17458275SEric Cheng  *    classification will be dropped in mac_rx from now on. Then we wait for
17468275SEric Cheng  *    the mr_refcnt to drop to zero. When the mr_refcnt reaches zero we are
17478275SEric Cheng  *    sure there aren't any upcall threads from the driver through hardware
17488275SEric Cheng  *    classification. In the case of SRS teardown we also remove the
17498275SEric Cheng  *    classification rule in the driver.
17508275SEric Cheng  *
17518275SEric Cheng  * 2. Stop future software classified packets by marking the flow entry with
17528275SEric Cheng  *    FE_QUIESCE or FE_CONDEMNED as appropriate which prevents the refcnt from
17538275SEric Cheng  *    increasing. We also remove the flow entry from the table in the latter
17548275SEric Cheng  *    case. Then wait for the fe_refcnt to reach an appropriate quiescent value
17558275SEric Cheng  *    that indicates there aren't any active threads using that flow entry.
17568275SEric Cheng  *
17578275SEric Cheng  * 3. Quiesce the SRS and softrings by signaling the SRS. The SRS poll thread,
17588275SEric Cheng  *    SRS worker thread, and the soft ring threads are quiesced in sequence
17598275SEric Cheng  *    with the SRS worker thread serving as a master controller. This
17608275SEric Cheng  *    mechansim is explained in mac_srs_worker_quiesce().
17618275SEric Cheng  *
17628275SEric Cheng  * The restart mechanism to reactivate the SRS and softrings is explained
17638275SEric Cheng  * in mac_srs_worker_restart(). Here we just signal the SRS worker to start the
17648275SEric Cheng  * restart sequence.
17650Sstevel@tonic-gate  */
17660Sstevel@tonic-gate void
17678275SEric Cheng mac_rx_srs_quiesce(mac_soft_ring_set_t *srs, uint_t srs_quiesce_flag)
17680Sstevel@tonic-gate {
17698275SEric Cheng 	flow_entry_t	*flent = srs->srs_flent;
17708275SEric Cheng 	uint_t	mr_flag, srs_done_flag;
17718275SEric Cheng 
17728275SEric Cheng 	ASSERT(MAC_PERIM_HELD((mac_handle_t)FLENT_TO_MIP(flent)));
17738275SEric Cheng 	ASSERT(!(srs->srs_type & SRST_TX));
17748275SEric Cheng 
17758275SEric Cheng 	if (srs_quiesce_flag == SRS_CONDEMNED) {
17768275SEric Cheng 		mr_flag = MR_CONDEMNED;
17778275SEric Cheng 		srs_done_flag = SRS_CONDEMNED_DONE;
17788275SEric Cheng 		if (srs->srs_type & SRST_CLIENT_POLL_ENABLED)
17798275SEric Cheng 			mac_srs_client_poll_disable(srs->srs_mcip, srs);
17808275SEric Cheng 	} else {
17818275SEric Cheng 		ASSERT(srs_quiesce_flag == SRS_QUIESCE);
17828275SEric Cheng 		mr_flag = MR_QUIESCE;
17838275SEric Cheng 		srs_done_flag = SRS_QUIESCE_DONE;
17848275SEric Cheng 		if (srs->srs_type & SRST_CLIENT_POLL_ENABLED)
17858275SEric Cheng 			mac_srs_client_poll_quiesce(srs->srs_mcip, srs);
17868275SEric Cheng 	}
17878275SEric Cheng 
17888275SEric Cheng 	if (srs->srs_ring != NULL) {
17898275SEric Cheng 		mac_rx_ring_quiesce(srs->srs_ring, mr_flag);
17908275SEric Cheng 	} else {
17918275SEric Cheng 		/*
17928275SEric Cheng 		 * SRS is driven by software classification. In case
17938275SEric Cheng 		 * of CONDEMNED, the top level teardown functions will
17948275SEric Cheng 		 * deal with flow removal.
17958275SEric Cheng 		 */
17968275SEric Cheng 		if (srs_quiesce_flag != SRS_CONDEMNED) {
17978275SEric Cheng 			FLOW_MARK(flent, FE_QUIESCE);
17988275SEric Cheng 			mac_flow_wait(flent, FLOW_DRIVER_UPCALL);
17998275SEric Cheng 		}
18008275SEric Cheng 	}
18010Sstevel@tonic-gate 
18020Sstevel@tonic-gate 	/*
18038275SEric Cheng 	 * Signal the SRS to quiesce itself, and then cv_wait for the
18048275SEric Cheng 	 * SRS quiesce to complete. The SRS worker thread will wake us
18058275SEric Cheng 	 * up when the quiesce is complete
18064913Sethindra 	 */
18078275SEric Cheng 	mac_srs_signal(srs, srs_quiesce_flag);
18088275SEric Cheng 	mac_srs_quiesce_wait(srs, srs_done_flag);
18094913Sethindra }
18104913Sethindra 
18114913Sethindra /*
18128275SEric Cheng  * Remove an SRS.
18134913Sethindra  */
18144913Sethindra void
18158275SEric Cheng mac_rx_srs_remove(mac_soft_ring_set_t *srs)
18164913Sethindra {
18178275SEric Cheng 	flow_entry_t *flent = srs->srs_flent;
18188275SEric Cheng 	int i;
18198275SEric Cheng 
18208275SEric Cheng 	mac_rx_srs_quiesce(srs, SRS_CONDEMNED);
18218275SEric Cheng 	/*
18228275SEric Cheng 	 * Locate and remove our entry in the fe_rx_srs[] array, and
18238275SEric Cheng 	 * adjust the fe_rx_srs array entries and array count by
18248275SEric Cheng 	 * moving the last entry into the vacated spot.
18258275SEric Cheng 	 */
18268275SEric Cheng 	mutex_enter(&flent->fe_lock);
18278275SEric Cheng 	for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
18288275SEric Cheng 		if (flent->fe_rx_srs[i] == srs)
18298275SEric Cheng 			break;
18304913Sethindra 	}
18318275SEric Cheng 
18328275SEric Cheng 	ASSERT(i != 0 && i < flent->fe_rx_srs_cnt);
18338275SEric Cheng 	if (i != flent->fe_rx_srs_cnt - 1) {
18348275SEric Cheng 		flent->fe_rx_srs[i] =
18358275SEric Cheng 		    flent->fe_rx_srs[flent->fe_rx_srs_cnt - 1];
18368275SEric Cheng 		i = flent->fe_rx_srs_cnt - 1;
18378275SEric Cheng 	}
18388275SEric Cheng 
18398275SEric Cheng 	flent->fe_rx_srs[i] = NULL;
18408275SEric Cheng 	flent->fe_rx_srs_cnt--;
18418275SEric Cheng 	mutex_exit(&flent->fe_lock);
18428275SEric Cheng 
18438275SEric Cheng 	mac_srs_free(srs);
18440Sstevel@tonic-gate }
18450Sstevel@tonic-gate 
18468275SEric Cheng static void
18478275SEric Cheng mac_srs_clear_flag(mac_soft_ring_set_t *srs, uint_t flag)
18480Sstevel@tonic-gate {
18498275SEric Cheng 	mutex_enter(&srs->srs_lock);
18508275SEric Cheng 	srs->srs_state &= ~flag;
18518275SEric Cheng 	mutex_exit(&srs->srs_lock);
18528275SEric Cheng }
18538275SEric Cheng 
18548275SEric Cheng void
18558275SEric Cheng mac_rx_srs_restart(mac_soft_ring_set_t *srs)
18568275SEric Cheng {
18578275SEric Cheng 	flow_entry_t	*flent = srs->srs_flent;
18588275SEric Cheng 	mac_ring_t	*mr;
18598275SEric Cheng 
18608275SEric Cheng 	ASSERT(MAC_PERIM_HELD((mac_handle_t)FLENT_TO_MIP(flent)));
18618275SEric Cheng 	ASSERT((srs->srs_type & SRST_TX) == 0);
18620Sstevel@tonic-gate 
18630Sstevel@tonic-gate 	/*
18648275SEric Cheng 	 * This handles a change in the number of SRSs between the quiesce and
18658275SEric Cheng 	 * and restart operation of a flow.
18668275SEric Cheng 	 */
18678275SEric Cheng 	if (!SRS_QUIESCED(srs))
18688275SEric Cheng 		return;
18698275SEric Cheng 
18708275SEric Cheng 	/*
18718275SEric Cheng 	 * Signal the SRS to restart itself. Wait for the restart to complete
18728275SEric Cheng 	 * Note that we only restart the SRS if it is not marked as
18738275SEric Cheng 	 * permanently quiesced.
18740Sstevel@tonic-gate 	 */
18758275SEric Cheng 	if (!SRS_QUIESCED_PERMANENT(srs)) {
18768275SEric Cheng 		mac_srs_signal(srs, SRS_RESTART);
18778275SEric Cheng 		mac_srs_quiesce_wait(srs, SRS_RESTART_DONE);
18788275SEric Cheng 		mac_srs_clear_flag(srs, SRS_RESTART_DONE);
18798275SEric Cheng 
18808275SEric Cheng 		mac_srs_client_poll_restart(srs->srs_mcip, srs);
18818275SEric Cheng 	}
18828275SEric Cheng 
18838275SEric Cheng 	/* Finally clear the flags to let the packets in */
18848275SEric Cheng 	mr = srs->srs_ring;
18858275SEric Cheng 	if (mr != NULL) {
18868275SEric Cheng 		MAC_RING_UNMARK(mr, MR_QUIESCE);
18878275SEric Cheng 		/* In case the ring was stopped, safely restart it */
18888275SEric Cheng 		(void) mac_start_ring(mr);
18898275SEric Cheng 	} else {
18908275SEric Cheng 		FLOW_UNMARK(flent, FE_QUIESCE);
18918275SEric Cheng 	}
18928275SEric Cheng }
18938275SEric Cheng 
18948275SEric Cheng /*
18958275SEric Cheng  * Temporary quiesce of a flow and associated Rx SRS.
18968275SEric Cheng  * Please see block comment above mac_rx_classify_flow_rem.
18978275SEric Cheng  */
18988275SEric Cheng /* ARGSUSED */
18998275SEric Cheng int
19008275SEric Cheng mac_rx_classify_flow_quiesce(flow_entry_t *flent, void *arg)
19018275SEric Cheng {
19028275SEric Cheng 	int		i;
19038275SEric Cheng 
19048275SEric Cheng 	for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
19058275SEric Cheng 		mac_rx_srs_quiesce((mac_soft_ring_set_t *)flent->fe_rx_srs[i],
19068275SEric Cheng 		    SRS_QUIESCE);
19078275SEric Cheng 	}
19088275SEric Cheng 	return (0);
19090Sstevel@tonic-gate }
19100Sstevel@tonic-gate 
19110Sstevel@tonic-gate /*
19128275SEric Cheng  * Restart a flow and associated Rx SRS that has been quiesced temporarily
19138275SEric Cheng  * Please see block comment above mac_rx_classify_flow_rem
19140Sstevel@tonic-gate  */
19158275SEric Cheng /* ARGSUSED */
19168275SEric Cheng int
19178275SEric Cheng mac_rx_classify_flow_restart(flow_entry_t *flent, void *arg)
19188275SEric Cheng {
19198275SEric Cheng 	int		i;
19208275SEric Cheng 
19218275SEric Cheng 	for (i = 0; i < flent->fe_rx_srs_cnt; i++)
19228275SEric Cheng 		mac_rx_srs_restart((mac_soft_ring_set_t *)flent->fe_rx_srs[i]);
19238275SEric Cheng 
19248275SEric Cheng 	return (0);
19258275SEric Cheng }
19268275SEric Cheng 
19270Sstevel@tonic-gate void
19288275SEric Cheng mac_srs_perm_quiesce(mac_client_handle_t mch, boolean_t on)
19290Sstevel@tonic-gate {
19308275SEric Cheng 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
19318275SEric Cheng 	flow_entry_t		*flent = mcip->mci_flent;
19328275SEric Cheng 	mac_impl_t		*mip = mcip->mci_mip;
19338275SEric Cheng 	mac_soft_ring_set_t	*mac_srs;
19348275SEric Cheng 	int			i;
19358275SEric Cheng 
19368275SEric Cheng 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
19378275SEric Cheng 
19388275SEric Cheng 	if (flent == NULL)
19398275SEric Cheng 		return;
19408275SEric Cheng 
19418275SEric Cheng 	for (i = 0; i < flent->fe_rx_srs_cnt; i++) {
19428275SEric Cheng 		mac_srs = flent->fe_rx_srs[i];
19438275SEric Cheng 		mutex_enter(&mac_srs->srs_lock);
19448275SEric Cheng 		if (on)
19458275SEric Cheng 			mac_srs->srs_state |= SRS_QUIESCE_PERM;
19468275SEric Cheng 		else
19478275SEric Cheng 			mac_srs->srs_state &= ~SRS_QUIESCE_PERM;
19488275SEric Cheng 		mutex_exit(&mac_srs->srs_lock);
19490Sstevel@tonic-gate 	}
19508275SEric Cheng }
19518275SEric Cheng 
19528275SEric Cheng void
19538275SEric Cheng mac_rx_client_quiesce(mac_client_handle_t mch)
19548275SEric Cheng {
19558275SEric Cheng 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
19568275SEric Cheng 	mac_impl_t		*mip = mcip->mci_mip;
19578275SEric Cheng 
19588275SEric Cheng 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
19598275SEric Cheng 
19608275SEric Cheng 	if (MCIP_DATAPATH_SETUP(mcip)) {
19618275SEric Cheng 		(void) mac_rx_classify_flow_quiesce(mcip->mci_flent,
19628275SEric Cheng 		    NULL);
19638275SEric Cheng 		(void) mac_flow_walk_nolock(mcip->mci_subflow_tab,
19648275SEric Cheng 		    mac_rx_classify_flow_quiesce, NULL);
19658275SEric Cheng 	}
19660Sstevel@tonic-gate }
19670Sstevel@tonic-gate 
19680Sstevel@tonic-gate void
19698275SEric Cheng mac_rx_client_restart(mac_client_handle_t mch)
19700Sstevel@tonic-gate {
19718275SEric Cheng 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
19728275SEric Cheng 	mac_impl_t		*mip = mcip->mci_mip;
19738275SEric Cheng 
19748275SEric Cheng 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
19758275SEric Cheng 
19768275SEric Cheng 	if (MCIP_DATAPATH_SETUP(mcip)) {
19778275SEric Cheng 		(void) mac_rx_classify_flow_restart(mcip->mci_flent, NULL);
19788275SEric Cheng 		(void) mac_flow_walk_nolock(mcip->mci_subflow_tab,
19798275SEric Cheng 		    mac_rx_classify_flow_restart, NULL);
19808275SEric Cheng 	}
19818275SEric Cheng }
19828275SEric Cheng 
19838275SEric Cheng /*
19848275SEric Cheng  * This function only quiesces the Tx SRS and softring worker threads. Callers
19858275SEric Cheng  * need to make sure that there aren't any mac client threads doing current or
19868275SEric Cheng  * future transmits in the mac before calling this function.
19878275SEric Cheng  */
19888275SEric Cheng void
19898275SEric Cheng mac_tx_srs_quiesce(mac_soft_ring_set_t *srs, uint_t srs_quiesce_flag)
19908275SEric Cheng {
19918275SEric Cheng 	mac_client_impl_t	*mcip = srs->srs_mcip;
19928275SEric Cheng 
19938275SEric Cheng 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
19948275SEric Cheng 
19958275SEric Cheng 	ASSERT(srs->srs_type & SRST_TX);
19968275SEric Cheng 	ASSERT(srs_quiesce_flag == SRS_CONDEMNED ||
19978275SEric Cheng 	    srs_quiesce_flag == SRS_QUIESCE);
19980Sstevel@tonic-gate 
19990Sstevel@tonic-gate 	/*
20008275SEric Cheng 	 * Signal the SRS to quiesce itself, and then cv_wait for the
20018275SEric Cheng 	 * SRS quiesce to complete. The SRS worker thread will wake us
20028275SEric Cheng 	 * up when the quiesce is complete
20030Sstevel@tonic-gate 	 */
20048275SEric Cheng 	mac_srs_signal(srs, srs_quiesce_flag);
20058275SEric Cheng 	mac_srs_quiesce_wait(srs, srs_quiesce_flag == SRS_QUIESCE ?
20068275SEric Cheng 	    SRS_QUIESCE_DONE : SRS_CONDEMNED_DONE);
20078275SEric Cheng }
20088275SEric Cheng 
20098275SEric Cheng void
20108275SEric Cheng mac_tx_srs_restart(mac_soft_ring_set_t *srs)
20118275SEric Cheng {
20128275SEric Cheng 	/*
20138275SEric Cheng 	 * Resizing the fanout could result in creation of new SRSs.
20148275SEric Cheng 	 * They may not necessarily be in the quiesced state in which
20158275SEric Cheng 	 * case it need be restarted
20168275SEric Cheng 	 */
20178275SEric Cheng 	if (!SRS_QUIESCED(srs))
20188275SEric Cheng 		return;
20198275SEric Cheng 
20208275SEric Cheng 	mac_srs_signal(srs, SRS_RESTART);
20218275SEric Cheng 	mac_srs_quiesce_wait(srs, SRS_RESTART_DONE);
20228275SEric Cheng 	mac_srs_clear_flag(srs, SRS_RESTART_DONE);
20230Sstevel@tonic-gate }
20240Sstevel@tonic-gate 
20250Sstevel@tonic-gate /*
20268275SEric Cheng  * Temporary quiesce of a flow and associated Rx SRS.
20278275SEric Cheng  * Please see block comment above mac_rx_srs_quiesce
20280Sstevel@tonic-gate  */
20298275SEric Cheng /* ARGSUSED */
20308275SEric Cheng int
20318275SEric Cheng mac_tx_flow_quiesce(flow_entry_t *flent, void *arg)
20320Sstevel@tonic-gate {
20332311Sseb 	/*
20348275SEric Cheng 	 * The fe_tx_srs is null for a subflow on an interface that is
20358275SEric Cheng 	 * not plumbed
20362311Sseb 	 */
20378275SEric Cheng 	if (flent->fe_tx_srs != NULL)
20388275SEric Cheng 		mac_tx_srs_quiesce(flent->fe_tx_srs, SRS_QUIESCE);
20398275SEric Cheng 	return (0);
20408275SEric Cheng }
20418275SEric Cheng 
20428275SEric Cheng /* ARGSUSED */
20438275SEric Cheng int
20448275SEric Cheng mac_tx_flow_restart(flow_entry_t *flent, void *arg)
20458275SEric Cheng {
20468275SEric Cheng 	/*
20478275SEric Cheng 	 * The fe_tx_srs is null for a subflow on an interface that is
20488275SEric Cheng 	 * not plumbed
20498275SEric Cheng 	 */
20508275SEric Cheng 	if (flent->fe_tx_srs != NULL)
20518275SEric Cheng 		mac_tx_srs_restart(flent->fe_tx_srs);
20528275SEric Cheng 	return (0);
20532311Sseb }
20542311Sseb 
20552311Sseb void
20568275SEric Cheng mac_tx_client_quiesce(mac_client_impl_t *mcip, uint_t srs_quiesce_flag)
20578275SEric Cheng {
20588275SEric Cheng 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
20598275SEric Cheng 
20608275SEric Cheng 	mac_tx_client_block(mcip);
20618275SEric Cheng 	if (MCIP_TX_SRS(mcip) != NULL) {
20628275SEric Cheng 		mac_tx_srs_quiesce(MCIP_TX_SRS(mcip), srs_quiesce_flag);
20638275SEric Cheng 		(void) mac_flow_walk_nolock(mcip->mci_subflow_tab,
20648275SEric Cheng 		    mac_tx_flow_quiesce, NULL);
20658275SEric Cheng 	}
20668275SEric Cheng }
20678275SEric Cheng 
20688275SEric Cheng void
20698275SEric Cheng mac_tx_client_restart(mac_client_impl_t *mcip)
20702311Sseb {
20718275SEric Cheng 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
20728275SEric Cheng 
20738275SEric Cheng 	mac_tx_client_unblock(mcip);
20748275SEric Cheng 	if (MCIP_TX_SRS(mcip) != NULL) {
20758275SEric Cheng 		mac_tx_srs_restart(MCIP_TX_SRS(mcip));
20768275SEric Cheng 		(void) mac_flow_walk_nolock(mcip->mci_subflow_tab,
20778275SEric Cheng 		    mac_tx_flow_restart, NULL);
20788275SEric Cheng 	}
20798275SEric Cheng }
20808275SEric Cheng 
20818275SEric Cheng void
20828275SEric Cheng mac_tx_client_flush(mac_client_impl_t *mcip)
20838275SEric Cheng {
20848275SEric Cheng 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
20858275SEric Cheng 
20868275SEric Cheng 	mac_tx_client_quiesce(mcip, SRS_QUIESCE);
20878275SEric Cheng 	mac_tx_client_restart(mcip);
20888275SEric Cheng }
20898275SEric Cheng 
20908275SEric Cheng void
20918275SEric Cheng mac_client_quiesce(mac_client_impl_t *mcip)
20928275SEric Cheng {
20938275SEric Cheng 	mac_rx_client_quiesce((mac_client_handle_t)mcip);
20948275SEric Cheng 	mac_tx_client_quiesce(mcip, SRS_QUIESCE);
20958275SEric Cheng }
20968275SEric Cheng 
20978275SEric Cheng void
20988275SEric Cheng mac_client_restart(mac_client_impl_t *mcip)
20998275SEric Cheng {
21008275SEric Cheng 	mac_rx_client_restart((mac_client_handle_t)mcip);
21018275SEric Cheng 	mac_tx_client_restart(mcip);
21022311Sseb }
21032311Sseb 
21042311Sseb /*
21055895Syz147064  * Allocate a minor number.
21065895Syz147064  */
21075895Syz147064 minor_t
21085895Syz147064 mac_minor_hold(boolean_t sleep)
21095895Syz147064 {
21105895Syz147064 	minor_t	minor;
21115895Syz147064 
21125895Syz147064 	/*
21135895Syz147064 	 * Grab a value from the arena.
21145895Syz147064 	 */
21155895Syz147064 	atomic_add_32(&minor_count, 1);
21165895Syz147064 
21175895Syz147064 	if (sleep)
21185895Syz147064 		minor = (uint_t)id_alloc(minor_ids);
21195895Syz147064 	else
21205895Syz147064 		minor = (uint_t)id_alloc_nosleep(minor_ids);
21215895Syz147064 
21225895Syz147064 	if (minor == 0) {
21235895Syz147064 		atomic_add_32(&minor_count, -1);
21245895Syz147064 		return (0);
21255895Syz147064 	}
21265895Syz147064 
21275895Syz147064 	return (minor);
21285895Syz147064 }
21295895Syz147064 
21305895Syz147064 /*
21315895Syz147064  * Release a previously allocated minor number.
21325895Syz147064  */
21335895Syz147064 void
21345895Syz147064 mac_minor_rele(minor_t minor)
21355895Syz147064 {
21365895Syz147064 	/*
21375895Syz147064 	 * Return the value to the arena.
21385895Syz147064 	 */
21395895Syz147064 	id_free(minor_ids, minor);
21405895Syz147064 	atomic_add_32(&minor_count, -1);
21415895Syz147064 }
21425895Syz147064 
21435895Syz147064 uint32_t
21445895Syz147064 mac_no_notification(mac_handle_t mh)
21455895Syz147064 {
21465895Syz147064 	mac_impl_t *mip = (mac_impl_t *)mh;
21475895Syz147064 	return (mip->mi_unsup_note);
21485895Syz147064 }
21495895Syz147064 
21505895Syz147064 /*
21518275SEric Cheng  * Prevent any new opens of this mac in preparation for unregister
21522311Sseb  */
21532311Sseb int
21548275SEric Cheng i_mac_disable(mac_impl_t *mip)
21552311Sseb {
21568275SEric Cheng 	mac_client_impl_t	*mcip;
21578275SEric Cheng 
21588275SEric Cheng 	rw_enter(&i_mac_impl_lock, RW_WRITER);
21598275SEric Cheng 	if (mip->mi_state_flags & MIS_DISABLED) {
21608275SEric Cheng 		/* Already disabled, return success */
21618275SEric Cheng 		rw_exit(&i_mac_impl_lock);
21628275SEric Cheng 		return (0);
21635895Syz147064 	}
21642311Sseb 	/*
21658275SEric Cheng 	 * See if there are any other references to this mac_t (e.g., VLAN's).
21668275SEric Cheng 	 * If so return failure. If all the other checks below pass, then
21678275SEric Cheng 	 * set mi_disabled atomically under the i_mac_impl_lock to prevent
21688275SEric Cheng 	 * any new VLAN's from being created or new mac client opens of this
21698275SEric Cheng 	 * mac end point.
21702311Sseb 	 */
21718275SEric Cheng 	if (mip->mi_ref > 0) {
21728275SEric Cheng 		rw_exit(&i_mac_impl_lock);
21738275SEric Cheng 		return (EBUSY);
21742311Sseb 	}
21752311Sseb 
21762311Sseb 	/*
21778275SEric Cheng 	 * mac clients must delete all multicast groups they join before
21788275SEric Cheng 	 * closing. bcast groups are reference counted, the last client
21798275SEric Cheng 	 * to delete the group will wait till the group is physically
21808275SEric Cheng 	 * deleted. Since all clients have closed this mac end point
21818275SEric Cheng 	 * mi_bcast_ngrps must be zero at this point
21822311Sseb 	 */
21838275SEric Cheng 	ASSERT(mip->mi_bcast_ngrps == 0);
21845009Sgd78059 
21855009Sgd78059 	/*
21868275SEric Cheng 	 * Don't let go of this if it has some flows.
21878275SEric Cheng 	 * All other code guarantees no flows are added to a disabled
21888275SEric Cheng 	 * mac, therefore it is sufficient to check for the flow table
21898275SEric Cheng 	 * only here.
21902311Sseb 	 */
21918275SEric Cheng 	mcip = mac_primary_client_handle(mip);
21928275SEric Cheng 	if ((mcip != NULL) && mac_link_has_flows((mac_client_handle_t)mcip)) {
21938275SEric Cheng 		rw_exit(&i_mac_impl_lock);
21948275SEric Cheng 		return (ENOTEMPTY);
21955895Syz147064 	}
21965895Syz147064 
21978275SEric Cheng 	mip->mi_state_flags |= MIS_DISABLED;
21981852Syz147064 	rw_exit(&i_mac_impl_lock);
2199269Sericheng 	return (0);
22008275SEric Cheng }
22018275SEric Cheng 
22028275SEric Cheng int
22038275SEric Cheng mac_disable_nowait(mac_handle_t mh)
22048275SEric Cheng {
22058275SEric Cheng 	mac_impl_t	*mip = (mac_impl_t *)mh;
22068275SEric Cheng 	int err;
22078275SEric Cheng 
22088275SEric Cheng 	if ((err = i_mac_perim_enter_nowait(mip)) != 0)
22098275SEric Cheng 		return (err);
22108275SEric Cheng 	err = i_mac_disable(mip);
22118275SEric Cheng 	i_mac_perim_exit(mip);
2212269Sericheng 	return (err);
22130Sstevel@tonic-gate }
22140Sstevel@tonic-gate 
22150Sstevel@tonic-gate int
22165084Sjohnlev mac_disable(mac_handle_t mh)
22170Sstevel@tonic-gate {
22188275SEric Cheng 	mac_impl_t	*mip = (mac_impl_t *)mh;
22198275SEric Cheng 	int err;
22208275SEric Cheng 
22218275SEric Cheng 	i_mac_perim_enter(mip);
22228275SEric Cheng 	err = i_mac_disable(mip);
22238275SEric Cheng 	i_mac_perim_exit(mip);
22245084Sjohnlev 
22250Sstevel@tonic-gate 	/*
22268275SEric Cheng 	 * Clean up notification thread and wait for it to exit.
22275009Sgd78059 	 */
22288275SEric Cheng 	if (err == 0)
22298275SEric Cheng 		i_mac_notify_exit(mip);
22308275SEric Cheng 
22318275SEric Cheng 	return (err);
22320Sstevel@tonic-gate }
22330Sstevel@tonic-gate 
22344913Sethindra /*
22358275SEric Cheng  * Called when the MAC instance has a non empty flow table, to de-multiplex
22368275SEric Cheng  * incoming packets to the right flow.
22378275SEric Cheng  * The MAC's rw lock is assumed held as a READER.
22384913Sethindra  */
22398275SEric Cheng /* ARGSUSED */
22408275SEric Cheng static mblk_t *
22418275SEric Cheng mac_rx_classify(mac_impl_t *mip, mac_resource_handle_t mrh, mblk_t *mp)
22420Sstevel@tonic-gate {
22438275SEric Cheng 	flow_entry_t	*flent = NULL;
22448275SEric Cheng 	uint_t		flags = FLOW_INBOUND;
22458275SEric Cheng 	int		err;
22464913Sethindra 
22474913Sethindra 	/*
22488275SEric Cheng 	 * If the mac is a port of an aggregation, pass FLOW_IGNORE_VLAN
22498275SEric Cheng 	 * to mac_flow_lookup() so that the VLAN packets can be successfully
22508275SEric Cheng 	 * passed to the non-VLAN aggregation flows.
22518275SEric Cheng 	 *
22528275SEric Cheng 	 * Note that there is possibly a race between this and
22538275SEric Cheng 	 * mac_unicast_remove/add() and VLAN packets could be incorrectly
22548275SEric Cheng 	 * classified to non-VLAN flows of non-aggregation mac clients. These
22558275SEric Cheng 	 * VLAN packets will be then filtered out by the mac module.
22564913Sethindra 	 */
22578275SEric Cheng 	if ((mip->mi_state_flags & MIS_EXCLUSIVE) != 0)
22588275SEric Cheng 		flags |= FLOW_IGNORE_VLAN;
22598275SEric Cheng 
22608275SEric Cheng 	err = mac_flow_lookup(mip->mi_flow_tab, mp, flags, &flent);
22618275SEric Cheng 	if (err != 0) {
22628275SEric Cheng 		/* no registered receive function */
22638275SEric Cheng 		return (mp);
22648275SEric Cheng 	} else {
22658275SEric Cheng 		mac_client_impl_t	*mcip;
22664913Sethindra 
22674913Sethindra 		/*
22688275SEric Cheng 		 * This flent might just be an additional one on the MAC client,
22698275SEric Cheng 		 * i.e. for classification purposes (different fdesc), however
22708275SEric Cheng 		 * the resources, SRS et. al., are in the mci_flent, so if
22718275SEric Cheng 		 * this isn't the mci_flent, we need to get it.
22724913Sethindra 		 */
22738275SEric Cheng 		if ((mcip = flent->fe_mcip) != NULL &&
22748275SEric Cheng 		    mcip->mci_flent != flent) {
22758275SEric Cheng 			FLOW_REFRELE(flent);
22768275SEric Cheng 			flent = mcip->mci_flent;
22778275SEric Cheng 			FLOW_TRY_REFHOLD(flent, err);
22788275SEric Cheng 			if (err != 0)
22798275SEric Cheng 				return (mp);
22808275SEric Cheng 		}
22818275SEric Cheng 		(flent->fe_cb_fn)(flent->fe_cb_arg1, flent->fe_cb_arg2, mp,
22828275SEric Cheng 		    B_FALSE);
22838275SEric Cheng 		FLOW_REFRELE(flent);
22845084Sjohnlev 	}
22855084Sjohnlev 	return (NULL);
22865084Sjohnlev }
22875084Sjohnlev 
22885084Sjohnlev mblk_t *
22898275SEric Cheng mac_rx_flow(mac_handle_t mh, mac_resource_handle_t mrh, mblk_t *mp_chain)
22900Sstevel@tonic-gate {
22912311Sseb 	mac_impl_t	*mip = (mac_impl_t *)mh;
22928275SEric Cheng 	mblk_t		*bp, *bp1, **bpp, *list = NULL;
22930Sstevel@tonic-gate 
22940Sstevel@tonic-gate 	/*
22958275SEric Cheng 	 * We walk the chain and attempt to classify each packet.
22968275SEric Cheng 	 * The packets that couldn't be classified will be returned
22978275SEric Cheng 	 * back to the caller.
22980Sstevel@tonic-gate 	 */
22998275SEric Cheng 	bp = mp_chain;
23008275SEric Cheng 	bpp = &list;
23018275SEric Cheng 	while (bp != NULL) {
23028275SEric Cheng 		bp1 = bp;
23038275SEric Cheng 		bp = bp->b_next;
23048275SEric Cheng 		bp1->b_next = NULL;
23058275SEric Cheng 
23068275SEric Cheng 		if (mac_rx_classify(mip, mrh, bp1) != NULL) {
23078275SEric Cheng 			*bpp = bp1;
23088275SEric Cheng 			bpp = &bp1->b_next;
23098275SEric Cheng 		}
23108275SEric Cheng 	}
23118275SEric Cheng 	return (list);
23120Sstevel@tonic-gate }
23130Sstevel@tonic-gate 
23148275SEric Cheng static int
23158275SEric Cheng mac_tx_flow_srs_wakeup(flow_entry_t *flent, void *arg)
23160Sstevel@tonic-gate {
23178275SEric Cheng 	mac_ring_handle_t ring = arg;
23188275SEric Cheng 
23198275SEric Cheng 	if (flent->fe_tx_srs)
23208275SEric Cheng 		mac_tx_srs_wakeup(flent->fe_tx_srs, ring);
23212311Sseb 	return (0);
23222311Sseb }
23232311Sseb 
23240Sstevel@tonic-gate void
23258275SEric Cheng i_mac_tx_srs_notify(mac_impl_t *mip, mac_ring_handle_t ring)
23268275SEric Cheng {
23278275SEric Cheng 	mac_client_impl_t	*cclient;
23288275SEric Cheng 	mac_soft_ring_set_t	*mac_srs;
23298275SEric Cheng 
23308275SEric Cheng 	/*
23318275SEric Cheng 	 * After grabbing the mi_rw_lock, the list of clients can't change.
23328275SEric Cheng 	 * If there are any clients mi_disabled must be B_FALSE and can't
23338275SEric Cheng 	 * get set since there are clients. If there aren't any clients we
23348275SEric Cheng 	 * don't do anything. In any case the mip has to be valid. The driver
23358275SEric Cheng 	 * must make sure that it goes single threaded (with respect to mac
23368275SEric Cheng 	 * calls) and wait for all pending mac calls to finish before calling
23378275SEric Cheng 	 * mac_unregister.
23388275SEric Cheng 	 */
23398275SEric Cheng 	rw_enter(&i_mac_impl_lock, RW_READER);
23408275SEric Cheng 	if (mip->mi_state_flags & MIS_DISABLED) {
23418275SEric Cheng 		rw_exit(&i_mac_impl_lock);
23428275SEric Cheng 		return;
23438275SEric Cheng 	}
23448275SEric Cheng 
23458275SEric Cheng 	/*
23468275SEric Cheng 	 * Get MAC tx srs from walking mac_client_handle list.
23478275SEric Cheng 	 */
23488275SEric Cheng 	rw_enter(&mip->mi_rw_lock, RW_READER);
23498275SEric Cheng 	for (cclient = mip->mi_clients_list; cclient != NULL;
23508275SEric Cheng 	    cclient = cclient->mci_client_next) {
23518275SEric Cheng 		if ((mac_srs = MCIP_TX_SRS(cclient)) != NULL)
23528275SEric Cheng 			mac_tx_srs_wakeup(mac_srs, ring);
23538275SEric Cheng 		if (!FLOW_TAB_EMPTY(cclient->mci_subflow_tab)) {
23548275SEric Cheng 			(void) mac_flow_walk_nolock(cclient->mci_subflow_tab,
23558275SEric Cheng 			    mac_tx_flow_srs_wakeup, ring);
23568275SEric Cheng 		}
23578275SEric Cheng 	}
23588275SEric Cheng 	rw_exit(&mip->mi_rw_lock);
23598275SEric Cheng 	rw_exit(&i_mac_impl_lock);
23608275SEric Cheng }
23618275SEric Cheng 
23628275SEric Cheng /* ARGSUSED */
23638275SEric Cheng void
23648275SEric Cheng mac_multicast_refresh(mac_handle_t mh, mac_multicst_t refresh, void *arg,
23650Sstevel@tonic-gate     boolean_t add)
23660Sstevel@tonic-gate {
23678275SEric Cheng 	mac_impl_t *mip = (mac_impl_t *)mh;
23688275SEric Cheng 
23698275SEric Cheng 	i_mac_perim_enter((mac_impl_t *)mh);
23700Sstevel@tonic-gate 	/*
23710Sstevel@tonic-gate 	 * If no specific refresh function was given then default to the
23720Sstevel@tonic-gate 	 * driver's m_multicst entry point.
23730Sstevel@tonic-gate 	 */
23740Sstevel@tonic-gate 	if (refresh == NULL) {
23752311Sseb 		refresh = mip->mi_multicst;
23762311Sseb 		arg = mip->mi_driver;
23770Sstevel@tonic-gate 	}
23788275SEric Cheng 
23798275SEric Cheng 	mac_bcast_refresh(mip, refresh, arg, add);
23808275SEric Cheng 	i_mac_perim_exit((mac_impl_t *)mh);
23810Sstevel@tonic-gate }
23820Sstevel@tonic-gate 
23830Sstevel@tonic-gate void
23842311Sseb mac_promisc_refresh(mac_handle_t mh, mac_setpromisc_t refresh, void *arg)
23850Sstevel@tonic-gate {
23862311Sseb 	mac_impl_t	*mip = (mac_impl_t *)mh;
23870Sstevel@tonic-gate 
23880Sstevel@tonic-gate 	/*
23890Sstevel@tonic-gate 	 * If no specific refresh function was given then default to the
23900Sstevel@tonic-gate 	 * driver's m_promisc entry point.
23910Sstevel@tonic-gate 	 */
23920Sstevel@tonic-gate 	if (refresh == NULL) {
23932311Sseb 		refresh = mip->mi_setpromisc;
23942311Sseb 		arg = mip->mi_driver;
23950Sstevel@tonic-gate 	}
23960Sstevel@tonic-gate 	ASSERT(refresh != NULL);
23970Sstevel@tonic-gate 
23980Sstevel@tonic-gate 	/*
23990Sstevel@tonic-gate 	 * Call the refresh function with the current promiscuity.
24000Sstevel@tonic-gate 	 */
24010Sstevel@tonic-gate 	refresh(arg, (mip->mi_devpromisc != 0));
24020Sstevel@tonic-gate }
24030Sstevel@tonic-gate 
24045895Syz147064 /*
24055895Syz147064  * The mac client requests that the mac not to change its margin size to
24065895Syz147064  * be less than the specified value.  If "current" is B_TRUE, then the client
24075895Syz147064  * requests the mac not to change its margin size to be smaller than the
24085895Syz147064  * current size. Further, return the current margin size value in this case.
24095895Syz147064  *
24105895Syz147064  * We keep every requested size in an ordered list from largest to smallest.
24115895Syz147064  */
24125895Syz147064 int
24135895Syz147064 mac_margin_add(mac_handle_t mh, uint32_t *marginp, boolean_t current)
24145895Syz147064 {
24155895Syz147064 	mac_impl_t		*mip = (mac_impl_t *)mh;
24165895Syz147064 	mac_margin_req_t	**pp, *p;
24175895Syz147064 	int			err = 0;
24185895Syz147064 
24198275SEric Cheng 	rw_enter(&(mip->mi_rw_lock), RW_WRITER);
24205895Syz147064 	if (current)
24215895Syz147064 		*marginp = mip->mi_margin;
24225895Syz147064 
24235895Syz147064 	/*
24245895Syz147064 	 * If the current margin value cannot satisfy the margin requested,
24255895Syz147064 	 * return ENOTSUP directly.
24265895Syz147064 	 */
24275895Syz147064 	if (*marginp > mip->mi_margin) {
24285895Syz147064 		err = ENOTSUP;
24295895Syz147064 		goto done;
24305895Syz147064 	}
24315895Syz147064 
24325895Syz147064 	/*
24335895Syz147064 	 * Check whether the given margin is already in the list. If so,
24345895Syz147064 	 * bump the reference count.
24355895Syz147064 	 */
24368275SEric Cheng 	for (pp = &mip->mi_mmrp; (p = *pp) != NULL; pp = &p->mmr_nextp) {
24375895Syz147064 		if (p->mmr_margin == *marginp) {
24385895Syz147064 			/*
24395895Syz147064 			 * The margin requested is already in the list,
24405895Syz147064 			 * so just bump the reference count.
24415895Syz147064 			 */
24425895Syz147064 			p->mmr_ref++;
24435895Syz147064 			goto done;
24445895Syz147064 		}
24455895Syz147064 		if (p->mmr_margin < *marginp)
24465895Syz147064 			break;
24475895Syz147064 	}
24485895Syz147064 
24495895Syz147064 
24508275SEric Cheng 	p = kmem_zalloc(sizeof (mac_margin_req_t), KM_SLEEP);
24515895Syz147064 	p->mmr_margin = *marginp;
24525895Syz147064 	p->mmr_ref++;
24535895Syz147064 	p->mmr_nextp = *pp;
24545895Syz147064 	*pp = p;
24555895Syz147064 
24565895Syz147064 done:
24578275SEric Cheng 	rw_exit(&(mip->mi_rw_lock));
24585895Syz147064 	return (err);
24595895Syz147064 }
24605895Syz147064 
24615895Syz147064 /*
24625895Syz147064  * The mac client requests to cancel its previous mac_margin_add() request.
24635895Syz147064  * We remove the requested margin size from the list.
24645895Syz147064  */
24655895Syz147064 int
24665895Syz147064 mac_margin_remove(mac_handle_t mh, uint32_t margin)
24675895Syz147064 {
24685895Syz147064 	mac_impl_t		*mip = (mac_impl_t *)mh;
24695895Syz147064 	mac_margin_req_t	**pp, *p;
24705895Syz147064 	int			err = 0;
24715895Syz147064 
24728275SEric Cheng 	rw_enter(&(mip->mi_rw_lock), RW_WRITER);
24735895Syz147064 	/*
24745895Syz147064 	 * Find the entry in the list for the given margin.
24755895Syz147064 	 */
24765895Syz147064 	for (pp = &(mip->mi_mmrp); (p = *pp) != NULL; pp = &(p->mmr_nextp)) {
24775895Syz147064 		if (p->mmr_margin == margin) {
24785895Syz147064 			if (--p->mmr_ref == 0)
24795895Syz147064 				break;
24805895Syz147064 
24815895Syz147064 			/*
24825895Syz147064 			 * There is still a reference to this address so
24835895Syz147064 			 * there's nothing more to do.
24845895Syz147064 			 */
24855895Syz147064 			goto done;
24865895Syz147064 		}
24875895Syz147064 	}
24885895Syz147064 
24895895Syz147064 	/*
24905895Syz147064 	 * We did not find an entry for the given margin.
24915895Syz147064 	 */
24925895Syz147064 	if (p == NULL) {
24935895Syz147064 		err = ENOENT;
24945895Syz147064 		goto done;
24955895Syz147064 	}
24965895Syz147064 
24975895Syz147064 	ASSERT(p->mmr_ref == 0);
24985895Syz147064 
24995895Syz147064 	/*
25005895Syz147064 	 * Remove it from the list.
25015895Syz147064 	 */
25025895Syz147064 	*pp = p->mmr_nextp;
25035895Syz147064 	kmem_free(p, sizeof (mac_margin_req_t));
25045895Syz147064 done:
25058275SEric Cheng 	rw_exit(&(mip->mi_rw_lock));
25065895Syz147064 	return (err);
25075895Syz147064 }
25085895Syz147064 
25095895Syz147064 boolean_t
25105895Syz147064 mac_margin_update(mac_handle_t mh, uint32_t margin)
25115895Syz147064 {
25125895Syz147064 	mac_impl_t	*mip = (mac_impl_t *)mh;
25135895Syz147064 	uint32_t	margin_needed = 0;
25145895Syz147064 
25158275SEric Cheng 	rw_enter(&(mip->mi_rw_lock), RW_WRITER);
25165895Syz147064 
25175895Syz147064 	if (mip->mi_mmrp != NULL)
25185895Syz147064 		margin_needed = mip->mi_mmrp->mmr_margin;
25195895Syz147064 
25205895Syz147064 	if (margin_needed <= margin)
25215895Syz147064 		mip->mi_margin = margin;
25225895Syz147064 
25238275SEric Cheng 	rw_exit(&(mip->mi_rw_lock));
25245895Syz147064 
25255895Syz147064 	if (margin_needed <= margin)
25265895Syz147064 		i_mac_notify(mip, MAC_NOTE_MARGIN);
25275895Syz147064 
25285895Syz147064 	return (margin_needed <= margin);
25295895Syz147064 }
25305895Syz147064 
25312311Sseb /*
25322311Sseb  * MAC Type Plugin functions.
25332311Sseb  */
25342311Sseb 
25358275SEric Cheng mactype_t *
25368275SEric Cheng mactype_getplugin(const char *pname)
25378275SEric Cheng {
25388275SEric Cheng 	mactype_t	*mtype = NULL;
25398275SEric Cheng 	boolean_t	tried_modload = B_FALSE;
25408275SEric Cheng 
25418275SEric Cheng 	mutex_enter(&i_mactype_lock);
25428275SEric Cheng 
25438275SEric Cheng find_registered_mactype:
25448275SEric Cheng 	if (mod_hash_find(i_mactype_hash, (mod_hash_key_t)pname,
25458275SEric Cheng 	    (mod_hash_val_t *)&mtype) != 0) {
25468275SEric Cheng 		if (!tried_modload) {
25478275SEric Cheng 			/*
25488275SEric Cheng 			 * If the plugin has not yet been loaded, then
25498275SEric Cheng 			 * attempt to load it now.  If modload() succeeds,
25508275SEric Cheng 			 * the plugin should have registered using
25518275SEric Cheng 			 * mactype_register(), in which case we can go back
25528275SEric Cheng 			 * and attempt to find it again.
25538275SEric Cheng 			 */
25548275SEric Cheng 			if (modload(MACTYPE_KMODDIR, (char *)pname) != -1) {
25558275SEric Cheng 				tried_modload = B_TRUE;
25568275SEric Cheng 				goto find_registered_mactype;
25578275SEric Cheng 			}
25588275SEric Cheng 		}
25598275SEric Cheng 	} else {
25608275SEric Cheng 		/*
25618275SEric Cheng 		 * Note that there's no danger that the plugin we've loaded
25628275SEric Cheng 		 * could be unloaded between the modload() step and the
25638275SEric Cheng 		 * reference count bump here, as we're holding
25648275SEric Cheng 		 * i_mactype_lock, which mactype_unregister() also holds.
25658275SEric Cheng 		 */
25668275SEric Cheng 		atomic_inc_32(&mtype->mt_ref);
25678275SEric Cheng 	}
25688275SEric Cheng 
25698275SEric Cheng 	mutex_exit(&i_mactype_lock);
25708275SEric Cheng 	return (mtype);
25718275SEric Cheng }
25728275SEric Cheng 
25732311Sseb mactype_register_t *
25742311Sseb mactype_alloc(uint_t mactype_version)
25752311Sseb {
25762311Sseb 	mactype_register_t *mtrp;
25772311Sseb 
25782311Sseb 	/*
25792311Sseb 	 * Make sure there isn't a version mismatch between the plugin and
25802311Sseb 	 * the framework.  In the future, if multiple versions are
25812311Sseb 	 * supported, this check could become more sophisticated.
25822311Sseb 	 */
25832311Sseb 	if (mactype_version != MACTYPE_VERSION)
25842311Sseb 		return (NULL);
25852311Sseb 
25862311Sseb 	mtrp = kmem_zalloc(sizeof (mactype_register_t), KM_SLEEP);
25872311Sseb 	mtrp->mtr_version = mactype_version;
25882311Sseb 	return (mtrp);
25892311Sseb }
25902311Sseb 
25912311Sseb void
25922311Sseb mactype_free(mactype_register_t *mtrp)
25932311Sseb {
25942311Sseb 	kmem_free(mtrp, sizeof (mactype_register_t));
25952311Sseb }
25962311Sseb 
25972311Sseb int
25982311Sseb mactype_register(mactype_register_t *mtrp)
25992311Sseb {
26002311Sseb 	mactype_t	*mtp;
26012311Sseb 	mactype_ops_t	*ops = mtrp->mtr_ops;
26022311Sseb 
26032311Sseb 	/* Do some sanity checking before we register this MAC type. */
26046353Sdr146992 	if (mtrp->mtr_ident == NULL || ops == NULL)
26052311Sseb 		return (EINVAL);
26062311Sseb 
26072311Sseb 	/*
26082311Sseb 	 * Verify that all mandatory callbacks are set in the ops
26092311Sseb 	 * vector.
26102311Sseb 	 */
26112311Sseb 	if (ops->mtops_unicst_verify == NULL ||
26122311Sseb 	    ops->mtops_multicst_verify == NULL ||
26132311Sseb 	    ops->mtops_sap_verify == NULL ||
26142311Sseb 	    ops->mtops_header == NULL ||
26152311Sseb 	    ops->mtops_header_info == NULL) {
26162311Sseb 		return (EINVAL);
26172311Sseb 	}
26182311Sseb 
26192311Sseb 	mtp = kmem_zalloc(sizeof (*mtp), KM_SLEEP);
26202311Sseb 	mtp->mt_ident = mtrp->mtr_ident;
26212311Sseb 	mtp->mt_ops = *ops;
26222311Sseb 	mtp->mt_type = mtrp->mtr_mactype;
26233147Sxc151355 	mtp->mt_nativetype = mtrp->mtr_nativetype;
26242311Sseb 	mtp->mt_addr_length = mtrp->mtr_addrlen;
26252311Sseb 	if (mtrp->mtr_brdcst_addr != NULL) {
26262311Sseb 		mtp->mt_brdcst_addr = kmem_alloc(mtrp->mtr_addrlen, KM_SLEEP);
26272311Sseb 		bcopy(mtrp->mtr_brdcst_addr, mtp->mt_brdcst_addr,
26282311Sseb 		    mtrp->mtr_addrlen);
26292311Sseb 	}
26302311Sseb 
26312311Sseb 	mtp->mt_stats = mtrp->mtr_stats;
26322311Sseb 	mtp->mt_statcount = mtrp->mtr_statcount;
26332311Sseb 
26346512Ssowmini 	mtp->mt_mapping = mtrp->mtr_mapping;
26356512Ssowmini 	mtp->mt_mappingcount = mtrp->mtr_mappingcount;
26366512Ssowmini 
26372311Sseb 	if (mod_hash_insert(i_mactype_hash,
26382311Sseb 	    (mod_hash_key_t)mtp->mt_ident, (mod_hash_val_t)mtp) != 0) {
26392311Sseb 		kmem_free(mtp->mt_brdcst_addr, mtp->mt_addr_length);
26402311Sseb 		kmem_free(mtp, sizeof (*mtp));
26412311Sseb 		return (EEXIST);
26422311Sseb 	}
26432311Sseb 	return (0);
26442311Sseb }
26452311Sseb 
26462311Sseb int
26472311Sseb mactype_unregister(const char *ident)
26482311Sseb {
26492311Sseb 	mactype_t	*mtp;
26502311Sseb 	mod_hash_val_t	val;
26512311Sseb 	int 		err;
26522311Sseb 
26532311Sseb 	/*
26542311Sseb 	 * Let's not allow MAC drivers to use this plugin while we're
26553288Sseb 	 * trying to unregister it.  Holding i_mactype_lock also prevents a
26563288Sseb 	 * plugin from unregistering while a MAC driver is attempting to
26573288Sseb 	 * hold a reference to it in i_mactype_getplugin().
26582311Sseb 	 */
26593288Sseb 	mutex_enter(&i_mactype_lock);
26602311Sseb 
26612311Sseb 	if ((err = mod_hash_find(i_mactype_hash, (mod_hash_key_t)ident,
26622311Sseb 	    (mod_hash_val_t *)&mtp)) != 0) {
26632311Sseb 		/* A plugin is trying to unregister, but it never registered. */
26643288Sseb 		err = ENXIO;
26653288Sseb 		goto done;
26662311Sseb 	}
26672311Sseb 
26683288Sseb 	if (mtp->mt_ref != 0) {
26693288Sseb 		err = EBUSY;
26703288Sseb 		goto done;
26712311Sseb 	}
26722311Sseb 
26732311Sseb 	err = mod_hash_remove(i_mactype_hash, (mod_hash_key_t)ident, &val);
26742311Sseb 	ASSERT(err == 0);
26752311Sseb 	if (err != 0) {
26762311Sseb 		/* This should never happen, thus the ASSERT() above. */
26773288Sseb 		err = EINVAL;
26783288Sseb 		goto done;
26792311Sseb 	}
26802311Sseb 	ASSERT(mtp == (mactype_t *)val);
26812311Sseb 
26822311Sseb 	kmem_free(mtp->mt_brdcst_addr, mtp->mt_addr_length);
26832311Sseb 	kmem_free(mtp, sizeof (mactype_t));
26843288Sseb done:
26853288Sseb 	mutex_exit(&i_mactype_lock);
26863288Sseb 	return (err);
26872311Sseb }
26885903Ssowmini 
26898275SEric Cheng /*
26908275SEric Cheng  * Returns TRUE when the specified property is intended for the MAC framework,
26918275SEric Cheng  * as opposed to driver defined properties.
26928275SEric Cheng  */
26938275SEric Cheng static boolean_t
26948275SEric Cheng mac_is_macprop(mac_prop_t *macprop)
26958275SEric Cheng {
26968275SEric Cheng 	switch (macprop->mp_id) {
26978275SEric Cheng 	case MAC_PROP_MAXBW:
26988275SEric Cheng 	case MAC_PROP_PRIO:
26998275SEric Cheng 	case MAC_PROP_BIND_CPU:
27008275SEric Cheng 		return (B_TRUE);
27018275SEric Cheng 	default:
27028275SEric Cheng 		return (B_FALSE);
27038275SEric Cheng 	}
27048275SEric Cheng }
27058275SEric Cheng 
27068275SEric Cheng /*
27078275SEric Cheng  * mac_set_prop() sets mac or hardware driver properties:
27088275SEric Cheng  * 	mac properties include maxbw, priority, and cpu binding list. Driver
27098275SEric Cheng  *	properties are private properties to the hardware, such as mtu, speed
27108275SEric Cheng  *	etc.
27118275SEric Cheng  * If the property is a driver property, mac_set_prop() calls driver's callback
27128275SEric Cheng  * function to set it.
27138275SEric Cheng  * If the property is a mac property, mac_set_prop() invokes mac_set_resources()
27148275SEric Cheng  * which will cache the property value in mac_impl_t and may call
27158275SEric Cheng  * mac_client_set_resource() to update property value of the primary mac client,
27168275SEric Cheng  * if it exists.
27178275SEric Cheng  */
27185903Ssowmini int
27195903Ssowmini mac_set_prop(mac_handle_t mh, mac_prop_t *macprop, void *val, uint_t valsize)
27205903Ssowmini {
27215903Ssowmini 	int err = ENOTSUP;
27225903Ssowmini 	mac_impl_t *mip = (mac_impl_t *)mh;
27235903Ssowmini 
27248275SEric Cheng 	ASSERT(MAC_PERIM_HELD(mh));
27258275SEric Cheng 
27268275SEric Cheng 	/* If it is mac property, call mac_set_resources() */
27278275SEric Cheng 	if (mac_is_macprop(macprop)) {
27288275SEric Cheng 		mac_resource_props_t mrp;
27298275SEric Cheng 
27308275SEric Cheng 		if (valsize < sizeof (mac_resource_props_t))
27318275SEric Cheng 			return (EINVAL);
27328275SEric Cheng 		bzero(&mrp, sizeof (mac_resource_props_t));
27338275SEric Cheng 		bcopy(val, &mrp, sizeof (mrp));
27348275SEric Cheng 		return (mac_set_resources(mh, &mrp));
27358275SEric Cheng 	}
27368275SEric Cheng 	/* For driver properties, call driver's callback */
27375903Ssowmini 	if (mip->mi_callbacks->mc_callbacks & MC_SETPROP) {
27385903Ssowmini 		err = mip->mi_callbacks->mc_setprop(mip->mi_driver,
27395903Ssowmini 		    macprop->mp_name, macprop->mp_id, valsize, val);
27405903Ssowmini 	}
27418275SEric Cheng 
27425903Ssowmini 	return (err);
27435903Ssowmini }
27445903Ssowmini 
27458275SEric Cheng /*
27468275SEric Cheng  * mac_get_prop() gets mac or hardware driver properties.
27478275SEric Cheng  *
27488275SEric Cheng  * If the property is a driver property, mac_get_prop() calls driver's callback
27498275SEric Cheng  * function to get it.
27508275SEric Cheng  * If the property is a mac property, mac_get_prop() invokes mac_get_resources()
27518275SEric Cheng  * which returns the cached value in mac_impl_t.
27528275SEric Cheng  */
27535903Ssowmini int
27548118SVasumathi.Sundaram@Sun.COM mac_get_prop(mac_handle_t mh, mac_prop_t *macprop, void *val, uint_t valsize,
27558118SVasumathi.Sundaram@Sun.COM     uint_t *perm)
27565903Ssowmini {
27575903Ssowmini 	int err = ENOTSUP;
27585903Ssowmini 	mac_impl_t *mip = (mac_impl_t *)mh;
27596512Ssowmini 	uint32_t sdu;
27606512Ssowmini 	link_state_t link_state;
27616512Ssowmini 
27628275SEric Cheng 	/* If mac property, read from cache */
27638275SEric Cheng 	if (mac_is_macprop(macprop)) {
27648275SEric Cheng 		mac_resource_props_t mrp;
27658275SEric Cheng 
27668275SEric Cheng 		if (valsize < sizeof (mac_resource_props_t))
27678275SEric Cheng 			return (EINVAL);
27688275SEric Cheng 		bzero(&mrp, sizeof (mac_resource_props_t));
27698275SEric Cheng 		mac_get_resources(mh, &mrp);
27708275SEric Cheng 		bcopy(&mrp, val, sizeof (mac_resource_props_t));
27718275SEric Cheng 		return (0);
27728275SEric Cheng 	}
27738275SEric Cheng 
27746512Ssowmini 	switch (macprop->mp_id) {
27756789Sam223141 	case MAC_PROP_MTU:
27766512Ssowmini 		if (valsize < sizeof (sdu))
27776512Ssowmini 			return (EINVAL);
27786789Sam223141 		if ((macprop->mp_flags & MAC_PROP_DEFAULT) == 0) {
27796512Ssowmini 			mac_sdu_get(mh, NULL, &sdu);
27806512Ssowmini 			bcopy(&sdu, val, sizeof (sdu));
27818118SVasumathi.Sundaram@Sun.COM 			if (mac_set_prop(mh, macprop, val, sizeof (sdu)) != 0)
27828118SVasumathi.Sundaram@Sun.COM 				*perm = MAC_PROP_PERM_READ;
27838118SVasumathi.Sundaram@Sun.COM 			else
27848118SVasumathi.Sundaram@Sun.COM 				*perm = MAC_PROP_PERM_RW;
27856512Ssowmini 			return (0);
27866512Ssowmini 		} else {
27876512Ssowmini 			if (mip->mi_info.mi_media == DL_ETHER) {
27886512Ssowmini 				sdu = ETHERMTU;
27896512Ssowmini 				bcopy(&sdu, val, sizeof (sdu));
27906512Ssowmini 				return (0);
27916512Ssowmini 			}
27926512Ssowmini 			/*
27936512Ssowmini 			 * ask driver for its default.
27946512Ssowmini 			 */
27956512Ssowmini 			break;
27966512Ssowmini 		}
27976789Sam223141 	case MAC_PROP_STATUS:
27986512Ssowmini 		if (valsize < sizeof (link_state))
27996512Ssowmini 			return (EINVAL);
28008118SVasumathi.Sundaram@Sun.COM 		*perm = MAC_PROP_PERM_READ;
28016512Ssowmini 		link_state = mac_link_get(mh);
28026512Ssowmini 		bcopy(&link_state, val, sizeof (link_state));
28036512Ssowmini 		return (0);
28046512Ssowmini 	default:
28056512Ssowmini 		break;
28068275SEric Cheng 
28076512Ssowmini 	}
28088275SEric Cheng 	/* If driver property, request from driver */
28095903Ssowmini 	if (mip->mi_callbacks->mc_callbacks & MC_GETPROP) {
28105903Ssowmini 		err = mip->mi_callbacks->mc_getprop(mip->mi_driver,
28116512Ssowmini 		    macprop->mp_name, macprop->mp_id, macprop->mp_flags,
28128118SVasumathi.Sundaram@Sun.COM 		    valsize, val, perm);
28135903Ssowmini 	}
28145903Ssowmini 	return (err);
28155903Ssowmini }
28165903Ssowmini 
28178275SEric Cheng void
28186512Ssowmini mac_register_priv_prop(mac_impl_t *mip, mac_priv_prop_t *mpp, uint_t nprop)
28196512Ssowmini {
28206512Ssowmini 	mac_priv_prop_t *mpriv;
28216512Ssowmini 
28226512Ssowmini 	if (mpp == NULL)
28236512Ssowmini 		return;
28246512Ssowmini 
28256512Ssowmini 	mpriv = kmem_zalloc(nprop * sizeof (*mpriv), KM_SLEEP);
28266512Ssowmini 	(void) memcpy(mpriv, mpp, nprop * sizeof (*mpriv));
28276512Ssowmini 	mip->mi_priv_prop = mpriv;
28286512Ssowmini 	mip->mi_priv_prop_count = nprop;
28296512Ssowmini }
28307406SSowmini.Varadhan@Sun.COM 
28318275SEric Cheng void
28327406SSowmini.Varadhan@Sun.COM mac_unregister_priv_prop(mac_impl_t *mip)
28337406SSowmini.Varadhan@Sun.COM {
28347406SSowmini.Varadhan@Sun.COM 	mac_priv_prop_t	*mpriv;
28357406SSowmini.Varadhan@Sun.COM 
28367406SSowmini.Varadhan@Sun.COM 	mpriv = mip->mi_priv_prop;
28377406SSowmini.Varadhan@Sun.COM 	if (mpriv != NULL) {
28387406SSowmini.Varadhan@Sun.COM 		kmem_free(mpriv, mip->mi_priv_prop_count * sizeof (*mpriv));
28397406SSowmini.Varadhan@Sun.COM 		mip->mi_priv_prop = NULL;
28407406SSowmini.Varadhan@Sun.COM 	}
28417406SSowmini.Varadhan@Sun.COM 	mip->mi_priv_prop_count = 0;
28427406SSowmini.Varadhan@Sun.COM }
28438275SEric Cheng 
28448275SEric Cheng /*
28458275SEric Cheng  * mac_ring_t 'mr' macros. Some rogue drivers may access ring structure
28468275SEric Cheng  * (by invoking mac_rx()) even after processing mac_stop_ring(). In such
28478275SEric Cheng  * cases if MAC free's the ring structure after mac_stop_ring(), any
28488275SEric Cheng  * illegal access to the ring structure coming from the driver will panic
28498275SEric Cheng  * the system. In order to protect the system from such inadverent access,
28508275SEric Cheng  * we maintain a cache of rings in the mac_impl_t after they get free'd up.
28518275SEric Cheng  * When packets are received on free'd up rings, MAC (through the generation
28528275SEric Cheng  * count mechanism) will drop such packets.
28538275SEric Cheng  */
28548275SEric Cheng static mac_ring_t *
28558275SEric Cheng mac_ring_alloc(mac_impl_t *mip, mac_capab_rings_t *cap_rings)
28568275SEric Cheng {
28578275SEric Cheng 	mac_ring_t *ring;
28588275SEric Cheng 
28598275SEric Cheng 	if (cap_rings->mr_type == MAC_RING_TYPE_RX) {
28608275SEric Cheng 		mutex_enter(&mip->mi_ring_lock);
28618275SEric Cheng 		if (mip->mi_ring_freelist != NULL) {
28628275SEric Cheng 			ring = mip->mi_ring_freelist;
28638275SEric Cheng 			mip->mi_ring_freelist = ring->mr_next;
28648275SEric Cheng 			bzero(ring, sizeof (mac_ring_t));
28658275SEric Cheng 		} else {
28668275SEric Cheng 			ring = kmem_cache_alloc(mac_ring_cache, KM_SLEEP);
28678275SEric Cheng 		}
28688275SEric Cheng 		mutex_exit(&mip->mi_ring_lock);
28698275SEric Cheng 	} else {
28708275SEric Cheng 		ring = kmem_zalloc(sizeof (mac_ring_t), KM_SLEEP);
28718275SEric Cheng 	}
28728275SEric Cheng 	ASSERT((ring != NULL) && (ring->mr_state == MR_FREE));
28738275SEric Cheng 	return (ring);
28748275SEric Cheng }
28758275SEric Cheng 
28768275SEric Cheng static void
28778275SEric Cheng mac_ring_free(mac_impl_t *mip, mac_ring_t *ring)
28788275SEric Cheng {
28798275SEric Cheng 	if (ring->mr_type == MAC_RING_TYPE_RX) {
28808275SEric Cheng 		mutex_enter(&mip->mi_ring_lock);
28818275SEric Cheng 		ring->mr_state = MR_FREE;
28828275SEric Cheng 		ring->mr_flag = 0;
28838275SEric Cheng 		ring->mr_next = mip->mi_ring_freelist;
28848275SEric Cheng 		mip->mi_ring_freelist = ring;
28858275SEric Cheng 		mutex_exit(&mip->mi_ring_lock);
28868275SEric Cheng 	} else {
28878275SEric Cheng 		kmem_free(ring, sizeof (mac_ring_t));
28888275SEric Cheng 	}
28898275SEric Cheng }
28908275SEric Cheng 
28918275SEric Cheng static void
28928275SEric Cheng mac_ring_freeall(mac_impl_t *mip)
28938275SEric Cheng {
28948275SEric Cheng 	mac_ring_t *ring_next;
28958275SEric Cheng 	mutex_enter(&mip->mi_ring_lock);
28968275SEric Cheng 	mac_ring_t *ring = mip->mi_ring_freelist;
28978275SEric Cheng 	while (ring != NULL) {
28988275SEric Cheng 		ring_next = ring->mr_next;
28998275SEric Cheng 		kmem_cache_free(mac_ring_cache, ring);
29008275SEric Cheng 		ring = ring_next;
29018275SEric Cheng 	}
29028275SEric Cheng 	mip->mi_ring_freelist = NULL;
29038275SEric Cheng 	mutex_exit(&mip->mi_ring_lock);
29048275SEric Cheng }
29058275SEric Cheng 
29068275SEric Cheng int
29078275SEric Cheng mac_start_ring(mac_ring_t *ring)
29088275SEric Cheng {
29098275SEric Cheng 	int rv = 0;
29108275SEric Cheng 
29118275SEric Cheng 	if (ring->mr_start != NULL)
29128275SEric Cheng 		rv = ring->mr_start(ring->mr_driver, ring->mr_gen_num);
29138275SEric Cheng 
29148275SEric Cheng 	return (rv);
29158275SEric Cheng }
29168275SEric Cheng 
29178275SEric Cheng void
29188275SEric Cheng mac_stop_ring(mac_ring_t *ring)
29198275SEric Cheng {
29208275SEric Cheng 	if (ring->mr_stop != NULL)
29218275SEric Cheng 		ring->mr_stop(ring->mr_driver);
29228275SEric Cheng 
29238275SEric Cheng 	/*
29248275SEric Cheng 	 * Increment the ring generation number for this ring.
29258275SEric Cheng 	 */
29268275SEric Cheng 	ring->mr_gen_num++;
29278275SEric Cheng }
29288275SEric Cheng 
29298275SEric Cheng int
29308275SEric Cheng mac_start_group(mac_group_t *group)
29318275SEric Cheng {
29328275SEric Cheng 	int rv = 0;
29338275SEric Cheng 
29348275SEric Cheng 	if (group->mrg_start != NULL)
29358275SEric Cheng 		rv = group->mrg_start(group->mrg_driver);
29368275SEric Cheng 
29378275SEric Cheng 	return (rv);
29388275SEric Cheng }
29398275SEric Cheng 
29408275SEric Cheng void
29418275SEric Cheng mac_stop_group(mac_group_t *group)
29428275SEric Cheng {
29438275SEric Cheng 	if (group->mrg_stop != NULL)
29448275SEric Cheng 		group->mrg_stop(group->mrg_driver);
29458275SEric Cheng }
29468275SEric Cheng 
29478275SEric Cheng /*
29488275SEric Cheng  * Called from mac_start() on the default Rx group. Broadcast and multicast
29498275SEric Cheng  * packets are received only on the default group. Hence the default group
29508275SEric Cheng  * needs to be up even if the primary client is not up, for the other groups
29518275SEric Cheng  * to be functional. We do this by calling this function at mac_start time
29528275SEric Cheng  * itself. However the broadcast packets that are received can't make their
29538275SEric Cheng  * way beyond mac_rx until a mac client creates a broadcast flow.
29548275SEric Cheng  */
29558275SEric Cheng static int
29568275SEric Cheng mac_start_group_and_rings(mac_group_t *group)
29578275SEric Cheng {
29588275SEric Cheng 	mac_ring_t	*ring;
29598275SEric Cheng 	int		rv = 0;
29608275SEric Cheng 
29618275SEric Cheng 	ASSERT(group->mrg_state == MAC_GROUP_STATE_REGISTERED);
29628275SEric Cheng 	if ((rv = mac_start_group(group)) != 0)
29638275SEric Cheng 		return (rv);
29648275SEric Cheng 
29658275SEric Cheng 	for (ring = group->mrg_rings; ring != NULL; ring = ring->mr_next) {
29668275SEric Cheng 		ASSERT(ring->mr_state == MR_FREE);
29678275SEric Cheng 		if ((rv = mac_start_ring(ring)) != 0)
29688275SEric Cheng 			goto error;
29698275SEric Cheng 		ring->mr_state = MR_INUSE;
29708275SEric Cheng 		ring->mr_classify_type = MAC_SW_CLASSIFIER;
29718275SEric Cheng 	}
29728275SEric Cheng 	return (0);
29738275SEric Cheng 
29748275SEric Cheng error:
29758275SEric Cheng 	mac_stop_group_and_rings(group);
29768275SEric Cheng 	return (rv);
29778275SEric Cheng }
29788275SEric Cheng 
29798275SEric Cheng /* Called from mac_stop on the default Rx group */
29808275SEric Cheng static void
29818275SEric Cheng mac_stop_group_and_rings(mac_group_t *group)
29828275SEric Cheng {
29838275SEric Cheng 	mac_ring_t	*ring;
29848275SEric Cheng 
29858275SEric Cheng 	for (ring = group->mrg_rings; ring != NULL; ring = ring->mr_next) {
29868275SEric Cheng 		if (ring->mr_state != MR_FREE) {
29878275SEric Cheng 			mac_stop_ring(ring);
29888275SEric Cheng 			ring->mr_state = MR_FREE;
29898275SEric Cheng 			ring->mr_flag = 0;
29908275SEric Cheng 			ring->mr_classify_type = MAC_NO_CLASSIFIER;
29918275SEric Cheng 		}
29928275SEric Cheng 	}
29938275SEric Cheng 	mac_stop_group(group);
29948275SEric Cheng }
29958275SEric Cheng 
29968275SEric Cheng 
29978275SEric Cheng static mac_ring_t *
29988275SEric Cheng mac_init_ring(mac_impl_t *mip, mac_group_t *group, int index,
29998275SEric Cheng     mac_capab_rings_t *cap_rings)
30008275SEric Cheng {
30018275SEric Cheng 	mac_ring_t *ring;
30028275SEric Cheng 	mac_ring_info_t ring_info;
30038275SEric Cheng 
30048275SEric Cheng 	ring = mac_ring_alloc(mip, cap_rings);
30058275SEric Cheng 
30068275SEric Cheng 	/* Prepare basic information of ring */
30078275SEric Cheng 	ring->mr_index = index;
30088275SEric Cheng 	ring->mr_type = group->mrg_type;
30098275SEric Cheng 	ring->mr_gh = (mac_group_handle_t)group;
30108275SEric Cheng 
30118275SEric Cheng 	/* Insert the new ring to the list. */
30128275SEric Cheng 	ring->mr_next = group->mrg_rings;
30138275SEric Cheng 	group->mrg_rings = ring;
30148275SEric Cheng 
30158275SEric Cheng 	/* Zero to reuse the info data structure */
30168275SEric Cheng 	bzero(&ring_info, sizeof (ring_info));
30178275SEric Cheng 
30188275SEric Cheng 	/* Query ring information from driver */
30198275SEric Cheng 	cap_rings->mr_rget(mip->mi_driver, group->mrg_type, group->mrg_index,
30208275SEric Cheng 	    index, &ring_info, (mac_ring_handle_t)ring);
30218275SEric Cheng 
30228275SEric Cheng 	ring->mr_info = ring_info;
30238275SEric Cheng 
30248275SEric Cheng 	/* Update ring's status */
30258275SEric Cheng 	ring->mr_state = MR_FREE;
30268275SEric Cheng 	ring->mr_flag = 0;
30278275SEric Cheng 
30288275SEric Cheng 	/* Update the ring count of the group */
30298275SEric Cheng 	group->mrg_cur_count++;
30308275SEric Cheng 	return (ring);
30318275SEric Cheng }
30328275SEric Cheng 
30338275SEric Cheng /*
30348275SEric Cheng  * Rings are chained together for easy regrouping.
30358275SEric Cheng  */
30368275SEric Cheng static void
30378275SEric Cheng mac_init_group(mac_impl_t *mip, mac_group_t *group, int size,
30388275SEric Cheng     mac_capab_rings_t *cap_rings)
30398275SEric Cheng {
30408275SEric Cheng 	int index;
30418275SEric Cheng 
30428275SEric Cheng 	/*
30438275SEric Cheng 	 * Initialize all ring members of this group. Size of zero will not
30448275SEric Cheng 	 * enter the loop, so it's safe for initializing an empty group.
30458275SEric Cheng 	 */
30468275SEric Cheng 	for (index = size - 1; index >= 0; index--)
30478275SEric Cheng 		(void) mac_init_ring(mip, group, index, cap_rings);
30488275SEric Cheng }
30498275SEric Cheng 
30508275SEric Cheng int
30518275SEric Cheng mac_init_rings(mac_impl_t *mip, mac_ring_type_t rtype)
30528275SEric Cheng {
30538275SEric Cheng 	mac_capab_rings_t *cap_rings;
30548275SEric Cheng 	mac_group_t *group, *groups;
30558275SEric Cheng 	mac_group_info_t group_info;
30568275SEric Cheng 	uint_t group_free = 0;
30578275SEric Cheng 	uint_t ring_left;
30588275SEric Cheng 	mac_ring_t *ring;
30598275SEric Cheng 	int g, err = 0;
30608275SEric Cheng 
30618275SEric Cheng 	switch (rtype) {
30628275SEric Cheng 	case MAC_RING_TYPE_RX:
30638275SEric Cheng 		ASSERT(mip->mi_rx_groups == NULL);
30648275SEric Cheng 
30658275SEric Cheng 		cap_rings = &mip->mi_rx_rings_cap;
30668275SEric Cheng 		cap_rings->mr_type = MAC_RING_TYPE_RX;
30678275SEric Cheng 		break;
30688275SEric Cheng 	case MAC_RING_TYPE_TX:
30698275SEric Cheng 		ASSERT(mip->mi_tx_groups == NULL);
30708275SEric Cheng 
30718275SEric Cheng 		cap_rings = &mip->mi_tx_rings_cap;
30728275SEric Cheng 		cap_rings->mr_type = MAC_RING_TYPE_TX;
30738275SEric Cheng 		break;
30748275SEric Cheng 	default:
30758275SEric Cheng 		ASSERT(B_FALSE);
30768275SEric Cheng 	}
30778275SEric Cheng 
30788275SEric Cheng 	if (!i_mac_capab_get((mac_handle_t)mip, MAC_CAPAB_RINGS,
30798275SEric Cheng 	    cap_rings))
30808275SEric Cheng 		return (0);
30818275SEric Cheng 
30828275SEric Cheng 	/*
30838275SEric Cheng 	 * Allocate a contiguous buffer for all groups.
30848275SEric Cheng 	 */
30858275SEric Cheng 	groups = kmem_zalloc(sizeof (mac_group_t) * (cap_rings->mr_gnum + 1),
30868275SEric Cheng 	    KM_SLEEP);
30878275SEric Cheng 
30888275SEric Cheng 	ring_left = cap_rings->mr_rnum;
30898275SEric Cheng 
30908275SEric Cheng 	/*
30918275SEric Cheng 	 * Get all ring groups if any, and get their ring members
30928275SEric Cheng 	 * if any.
30938275SEric Cheng 	 */
30948275SEric Cheng 	for (g = 0; g < cap_rings->mr_gnum; g++) {
30958275SEric Cheng 		group = groups + g;
30968275SEric Cheng 
30978275SEric Cheng 		/* Prepare basic information of the group */
30988275SEric Cheng 		group->mrg_index = g;
30998275SEric Cheng 		group->mrg_type = rtype;
31008275SEric Cheng 		group->mrg_state = MAC_GROUP_STATE_UNINIT;
31018275SEric Cheng 		group->mrg_mh = (mac_handle_t)mip;
31028275SEric Cheng 		group->mrg_next = group + 1;
31038275SEric Cheng 
31048275SEric Cheng 		/* Zero to reuse the info data structure */
31058275SEric Cheng 		bzero(&group_info, sizeof (group_info));
31068275SEric Cheng 
31078275SEric Cheng 		/* Query group information from driver */
31088275SEric Cheng 		cap_rings->mr_gget(mip->mi_driver, rtype, g, &group_info,
31098275SEric Cheng 		    (mac_group_handle_t)group);
31108275SEric Cheng 
31118275SEric Cheng 		switch (cap_rings->mr_group_type) {
31128275SEric Cheng 		case MAC_GROUP_TYPE_DYNAMIC:
31138275SEric Cheng 			if (cap_rings->mr_gaddring == NULL ||
31148275SEric Cheng 			    cap_rings->mr_gremring == NULL) {
31158275SEric Cheng 				DTRACE_PROBE3(
31168275SEric Cheng 				    mac__init__rings_no_addremring,
31178275SEric Cheng 				    char *, mip->mi_name,
31188275SEric Cheng 				    mac_group_add_ring_t,
31198275SEric Cheng 				    cap_rings->mr_gaddring,
31208275SEric Cheng 				    mac_group_add_ring_t,
31218275SEric Cheng 				    cap_rings->mr_gremring);
31228275SEric Cheng 				err = EINVAL;
31238275SEric Cheng 				goto bail;
31248275SEric Cheng 			}
31258275SEric Cheng 
31268275SEric Cheng 			switch (rtype) {
31278275SEric Cheng 			case MAC_RING_TYPE_RX:
31288275SEric Cheng 				/*
31298275SEric Cheng 				 * The first RX group must have non-zero
31308275SEric Cheng 				 * rings, and the following groups must
31318275SEric Cheng 				 * have zero rings.
31328275SEric Cheng 				 */
31338275SEric Cheng 				if (g == 0 && group_info.mgi_count == 0) {
31348275SEric Cheng 					DTRACE_PROBE1(
31358275SEric Cheng 					    mac__init__rings__rx__def__zero,
31368275SEric Cheng 					    char *, mip->mi_name);
31378275SEric Cheng 					err = EINVAL;
31388275SEric Cheng 					goto bail;
31398275SEric Cheng 				}
31408275SEric Cheng 				if (g > 0 && group_info.mgi_count != 0) {
31418275SEric Cheng 					DTRACE_PROBE3(
31428275SEric Cheng 					    mac__init__rings__rx__nonzero,
31438275SEric Cheng 					    char *, mip->mi_name,
31448275SEric Cheng 					    int, g, int, group_info.mgi_count);
31458275SEric Cheng 					err = EINVAL;
31468275SEric Cheng 					goto bail;
31478275SEric Cheng 				}
31488275SEric Cheng 				break;
31498275SEric Cheng 			case MAC_RING_TYPE_TX:
31508275SEric Cheng 				/*
31518275SEric Cheng 				 * All TX ring groups must have zero rings.
31528275SEric Cheng 				 */
31538275SEric Cheng 				if (group_info.mgi_count != 0) {
31548275SEric Cheng 					DTRACE_PROBE3(
31558275SEric Cheng 					    mac__init__rings__tx__nonzero,
31568275SEric Cheng 					    char *, mip->mi_name,
31578275SEric Cheng 					    int, g, int, group_info.mgi_count);
31588275SEric Cheng 					err = EINVAL;
31598275SEric Cheng 					goto bail;
31608275SEric Cheng 				}
31618275SEric Cheng 				break;
31628275SEric Cheng 			}
31638275SEric Cheng 			break;
31648275SEric Cheng 		case MAC_GROUP_TYPE_STATIC:
31658275SEric Cheng 			/*
31668275SEric Cheng 			 * Note that an empty group is allowed, e.g., an aggr
31678275SEric Cheng 			 * would start with an empty group.
31688275SEric Cheng 			 */
31698275SEric Cheng 			break;
31708275SEric Cheng 		default:
31718275SEric Cheng 			/* unknown group type */
31728275SEric Cheng 			DTRACE_PROBE2(mac__init__rings__unknown__type,
31738275SEric Cheng 			    char *, mip->mi_name,
31748275SEric Cheng 			    int, cap_rings->mr_group_type);
31758275SEric Cheng 			err = EINVAL;
31768275SEric Cheng 			goto bail;
31778275SEric Cheng 		}
31788275SEric Cheng 
31798275SEric Cheng 
31808275SEric Cheng 		/*
31818275SEric Cheng 		 * Driver must register group->mgi_addmac/remmac() for rx groups
31828275SEric Cheng 		 * to support multiple MAC addresses.
31838275SEric Cheng 		 */
31848275SEric Cheng 		if (rtype == MAC_RING_TYPE_RX) {
31858275SEric Cheng 			if ((group_info.mgi_addmac == NULL) ||
31868275SEric Cheng 			    (group_info.mgi_addmac == NULL))
31878275SEric Cheng 				goto bail;
31888275SEric Cheng 		}
31898275SEric Cheng 
31908275SEric Cheng 		/* Cache driver-supplied information */
31918275SEric Cheng 		group->mrg_info = group_info;
31928275SEric Cheng 
31938275SEric Cheng 		/* Update the group's status and group count. */
31948275SEric Cheng 		mac_set_rx_group_state(group, MAC_GROUP_STATE_REGISTERED);
31958275SEric Cheng 		group_free++;
31968275SEric Cheng 
31978275SEric Cheng 		group->mrg_rings = NULL;
31988275SEric Cheng 		group->mrg_cur_count = 0;
31998275SEric Cheng 		mac_init_group(mip, group, group_info.mgi_count, cap_rings);
32008275SEric Cheng 		ring_left -= group_info.mgi_count;
32018275SEric Cheng 
32028275SEric Cheng 		/* The current group size should be equal to default value */
32038275SEric Cheng 		ASSERT(group->mrg_cur_count == group_info.mgi_count);
32048275SEric Cheng 	}
32058275SEric Cheng 
32068275SEric Cheng 	/* Build up a dummy group for free resources as a pool */
32078275SEric Cheng 	group = groups + cap_rings->mr_gnum;
32088275SEric Cheng 
32098275SEric Cheng 	/* Prepare basic information of the group */
32108275SEric Cheng 	group->mrg_index = -1;
32118275SEric Cheng 	group->mrg_type = rtype;
32128275SEric Cheng 	group->mrg_state = MAC_GROUP_STATE_UNINIT;
32138275SEric Cheng 	group->mrg_mh = (mac_handle_t)mip;
32148275SEric Cheng 	group->mrg_next = NULL;
32158275SEric Cheng 
32168275SEric Cheng 	/*
32178275SEric Cheng 	 * If there are ungrouped rings, allocate a continuous buffer for
32188275SEric Cheng 	 * remaining resources.
32198275SEric Cheng 	 */
32208275SEric Cheng 	if (ring_left != 0) {
32218275SEric Cheng 		group->mrg_rings = NULL;
32228275SEric Cheng 		group->mrg_cur_count = 0;
32238275SEric Cheng 		mac_init_group(mip, group, ring_left, cap_rings);
32248275SEric Cheng 
32258275SEric Cheng 		/* The current group size should be equal to ring_left */
32268275SEric Cheng 		ASSERT(group->mrg_cur_count == ring_left);
32278275SEric Cheng 
32288275SEric Cheng 		ring_left = 0;
32298275SEric Cheng 
32308275SEric Cheng 		/* Update this group's status */
32318275SEric Cheng 		mac_set_rx_group_state(group, MAC_GROUP_STATE_REGISTERED);
32328275SEric Cheng 	} else
32338275SEric Cheng 		group->mrg_rings = NULL;
32348275SEric Cheng 
32358275SEric Cheng 	ASSERT(ring_left == 0);
32368275SEric Cheng 
32378275SEric Cheng bail:
32388275SEric Cheng 	/* Cache other important information to finalize the initialization */
32398275SEric Cheng 	switch (rtype) {
32408275SEric Cheng 	case MAC_RING_TYPE_RX:
32418275SEric Cheng 		mip->mi_rx_group_type = cap_rings->mr_group_type;
32428275SEric Cheng 		mip->mi_rx_group_count = cap_rings->mr_gnum;
32438275SEric Cheng 		mip->mi_rx_groups = groups;
32448275SEric Cheng 		break;
32458275SEric Cheng 	case MAC_RING_TYPE_TX:
32468275SEric Cheng 		mip->mi_tx_group_type = cap_rings->mr_group_type;
32478275SEric Cheng 		mip->mi_tx_group_count = cap_rings->mr_gnum;
32488275SEric Cheng 		mip->mi_tx_group_free = group_free;
32498275SEric Cheng 		mip->mi_tx_groups = groups;
32508275SEric Cheng 
32518275SEric Cheng 		/*
32528275SEric Cheng 		 * Ring 0 is used as the default one and it could be assigned
32538275SEric Cheng 		 * to a client as well.
32548275SEric Cheng 		 */
32558275SEric Cheng 		group = groups + cap_rings->mr_gnum;
32568275SEric Cheng 		ring = group->mrg_rings;
32578275SEric Cheng 		while ((ring->mr_index != 0) && (ring->mr_next != NULL))
32588275SEric Cheng 			ring = ring->mr_next;
32598275SEric Cheng 		ASSERT(ring->mr_index == 0);
32608275SEric Cheng 		mip->mi_default_tx_ring = (mac_ring_handle_t)ring;
32618275SEric Cheng 		break;
32628275SEric Cheng 	default:
32638275SEric Cheng 		ASSERT(B_FALSE);
32648275SEric Cheng 	}
32658275SEric Cheng 
32668275SEric Cheng 	if (err != 0)
32678275SEric Cheng 		mac_free_rings(mip, rtype);
32688275SEric Cheng 
32698275SEric Cheng 	return (err);
32708275SEric Cheng }
32718275SEric Cheng 
32728275SEric Cheng /*
32738275SEric Cheng  * Called to free all ring groups with particular type. It's supposed all groups
32748275SEric Cheng  * have been released by clinet.
32758275SEric Cheng  */
32768275SEric Cheng void
32778275SEric Cheng mac_free_rings(mac_impl_t *mip, mac_ring_type_t rtype)
32788275SEric Cheng {
32798275SEric Cheng 	mac_group_t *group, *groups;
32808275SEric Cheng 	uint_t group_count;
32818275SEric Cheng 
32828275SEric Cheng 	switch (rtype) {
32838275SEric Cheng 	case MAC_RING_TYPE_RX:
32848275SEric Cheng 		if (mip->mi_rx_groups == NULL)
32858275SEric Cheng 			return;
32868275SEric Cheng 
32878275SEric Cheng 		groups = mip->mi_rx_groups;
32888275SEric Cheng 		group_count = mip->mi_rx_group_count;
32898275SEric Cheng 
32908275SEric Cheng 		mip->mi_rx_groups = NULL;
32918275SEric Cheng 		mip->mi_rx_group_count = 0;
32928275SEric Cheng 		break;
32938275SEric Cheng 	case MAC_RING_TYPE_TX:
32948275SEric Cheng 		ASSERT(mip->mi_tx_group_count == mip->mi_tx_group_free);
32958275SEric Cheng 
32968275SEric Cheng 		if (mip->mi_tx_groups == NULL)
32978275SEric Cheng 			return;
32988275SEric Cheng 
32998275SEric Cheng 		groups = mip->mi_tx_groups;
33008275SEric Cheng 		group_count = mip->mi_tx_group_count;
33018275SEric Cheng 
33028275SEric Cheng 		mip->mi_tx_groups = NULL;
33038275SEric Cheng 		mip->mi_tx_group_count = 0;
33048275SEric Cheng 		mip->mi_tx_group_free = 0;
33058275SEric Cheng 		mip->mi_default_tx_ring = NULL;
33068275SEric Cheng 		break;
33078275SEric Cheng 	default:
33088275SEric Cheng 		ASSERT(B_FALSE);
33098275SEric Cheng 	}
33108275SEric Cheng 
33118275SEric Cheng 	for (group = groups; group != NULL; group = group->mrg_next) {
33128275SEric Cheng 		mac_ring_t *ring;
33138275SEric Cheng 
33148275SEric Cheng 		if (group->mrg_cur_count == 0)
33158275SEric Cheng 			continue;
33168275SEric Cheng 
33178275SEric Cheng 		ASSERT(group->mrg_rings != NULL);
33188275SEric Cheng 
33198275SEric Cheng 		while ((ring = group->mrg_rings) != NULL) {
33208275SEric Cheng 			group->mrg_rings = ring->mr_next;
33218275SEric Cheng 			mac_ring_free(mip, ring);
33228275SEric Cheng 		}
33238275SEric Cheng 	}
33248275SEric Cheng 
33258275SEric Cheng 	/* Free all the cached rings */
33268275SEric Cheng 	mac_ring_freeall(mip);
33278275SEric Cheng 	/* Free the block of group data strutures */
33288275SEric Cheng 	kmem_free(groups, sizeof (mac_group_t) * (group_count + 1));
33298275SEric Cheng }
33308275SEric Cheng 
33318275SEric Cheng /*
33328275SEric Cheng  * Associate a MAC address with a receive group.
33338275SEric Cheng  *
33348275SEric Cheng  * The return value of this function should always be checked properly, because
33358275SEric Cheng  * any type of failure could cause unexpected results. A group can be added
33368275SEric Cheng  * or removed with a MAC address only after it has been reserved. Ideally,
33378275SEric Cheng  * a successful reservation always leads to calling mac_group_addmac() to
33388275SEric Cheng  * steer desired traffic. Failure of adding an unicast MAC address doesn't
33398275SEric Cheng  * always imply that the group is functioning abnormally.
33408275SEric Cheng  *
33418275SEric Cheng  * Currently this function is called everywhere, and it reflects assumptions
33428275SEric Cheng  * about MAC addresses in the implementation. CR 6735196.
33438275SEric Cheng  */
33448275SEric Cheng int
33458275SEric Cheng mac_group_addmac(mac_group_t *group, const uint8_t *addr)
33468275SEric Cheng {
33478275SEric Cheng 	ASSERT(group->mrg_type == MAC_RING_TYPE_RX);
33488275SEric Cheng 	ASSERT(group->mrg_info.mgi_addmac != NULL);
33498275SEric Cheng 
33508275SEric Cheng 	return (group->mrg_info.mgi_addmac(group->mrg_info.mgi_driver, addr));
33518275SEric Cheng }
33528275SEric Cheng 
33538275SEric Cheng /*
33548275SEric Cheng  * Remove the association between MAC address and receive group.
33558275SEric Cheng  */
33568275SEric Cheng int
33578275SEric Cheng mac_group_remmac(mac_group_t *group, const uint8_t *addr)
33588275SEric Cheng {
33598275SEric Cheng 	ASSERT(group->mrg_type == MAC_RING_TYPE_RX);
33608275SEric Cheng 	ASSERT(group->mrg_info.mgi_remmac != NULL);
33618275SEric Cheng 
33628275SEric Cheng 	return (group->mrg_info.mgi_remmac(group->mrg_info.mgi_driver, addr));
33638275SEric Cheng }
33648275SEric Cheng 
33658275SEric Cheng /*
33668275SEric Cheng  * Release a ring in use by marking it MR_FREE.
33678275SEric Cheng  * Any other client may reserve it for its use.
33688275SEric Cheng  */
33698275SEric Cheng void
33708275SEric Cheng mac_release_tx_ring(mac_ring_handle_t rh)
33718275SEric Cheng {
33728275SEric Cheng 	mac_ring_t *ring = (mac_ring_t *)rh;
33738275SEric Cheng 	mac_group_t *group = (mac_group_t *)ring->mr_gh;
33748275SEric Cheng 	mac_impl_t *mip = (mac_impl_t *)group->mrg_mh;
33758275SEric Cheng 
33768275SEric Cheng 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
33778275SEric Cheng 	ASSERT(ring->mr_state != MR_FREE);
33788275SEric Cheng 
33798275SEric Cheng 	/*
33808275SEric Cheng 	 * Default tx ring will be released by mac_stop().
33818275SEric Cheng 	 */
33828275SEric Cheng 	if (rh == mip->mi_default_tx_ring)
33838275SEric Cheng 		return;
33848275SEric Cheng 
33858275SEric Cheng 	mac_stop_ring(ring);
33868275SEric Cheng 
33878275SEric Cheng 	ring->mr_state = MR_FREE;
33888275SEric Cheng 	ring->mr_flag = 0;
33898275SEric Cheng }
33908275SEric Cheng 
33918275SEric Cheng /*
33928275SEric Cheng  * Send packets through a selected tx ring.
33938275SEric Cheng  */
33948275SEric Cheng mblk_t *
33958275SEric Cheng mac_ring_tx(mac_ring_handle_t rh, mblk_t *mp)
33968275SEric Cheng {
33978275SEric Cheng 	mac_ring_t *ring = (mac_ring_t *)rh;
33988275SEric Cheng 	mac_ring_info_t *info = &ring->mr_info;
33998275SEric Cheng 
34008275SEric Cheng 	ASSERT(ring->mr_type == MAC_RING_TYPE_TX);
34018275SEric Cheng 	ASSERT(ring->mr_state >= MR_INUSE);
34028275SEric Cheng 	ASSERT(info->mri_tx != NULL);
34038275SEric Cheng 
34048275SEric Cheng 	return (info->mri_tx(info->mri_driver, mp));
34058275SEric Cheng }
34068275SEric Cheng 
34078275SEric Cheng /*
34088275SEric Cheng  * Find a ring from its index.
34098275SEric Cheng  */
34108275SEric Cheng mac_ring_t *
34118275SEric Cheng mac_find_ring(mac_group_t *group, int index)
34128275SEric Cheng {
34138275SEric Cheng 	mac_ring_t *ring = group->mrg_rings;
34148275SEric Cheng 
34158275SEric Cheng 	for (ring = group->mrg_rings; ring != NULL; ring = ring->mr_next)
34168275SEric Cheng 		if (ring->mr_index == index)
34178275SEric Cheng 			break;
34188275SEric Cheng 
34198275SEric Cheng 	return (ring);
34208275SEric Cheng }
34218275SEric Cheng /*
34228275SEric Cheng  * Add a ring to an existing group.
34238275SEric Cheng  *
34248275SEric Cheng  * The ring must be either passed directly (for example if the ring
34258275SEric Cheng  * movement is initiated by the framework), or specified through a driver
34268275SEric Cheng  * index (for example when the ring is added by the driver.
34278275SEric Cheng  *
34288275SEric Cheng  * The caller needs to call mac_perim_enter() before calling this function.
34298275SEric Cheng  */
34308275SEric Cheng int
34318275SEric Cheng i_mac_group_add_ring(mac_group_t *group, mac_ring_t *ring, int index)
34328275SEric Cheng {
34338275SEric Cheng 	mac_impl_t *mip = (mac_impl_t *)group->mrg_mh;
34348275SEric Cheng 	mac_capab_rings_t *cap_rings;
34358275SEric Cheng 	boolean_t driver_call = (ring == NULL);
34368275SEric Cheng 	mac_group_type_t group_type;
34378275SEric Cheng 	int ret = 0;
34388275SEric Cheng 
34398275SEric Cheng 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
34408275SEric Cheng 
34418275SEric Cheng 	switch (group->mrg_type) {
34428275SEric Cheng 	case MAC_RING_TYPE_RX:
34438275SEric Cheng 		cap_rings = &mip->mi_rx_rings_cap;
34448275SEric Cheng 		group_type = mip->mi_rx_group_type;
34458275SEric Cheng 		break;
34468275SEric Cheng 	case MAC_RING_TYPE_TX:
34478275SEric Cheng 		cap_rings = &mip->mi_tx_rings_cap;
34488275SEric Cheng 		group_type = mip->mi_tx_group_type;
34498275SEric Cheng 		break;
34508275SEric Cheng 	default:
34518275SEric Cheng 		ASSERT(B_FALSE);
34528275SEric Cheng 	}
34538275SEric Cheng 
34548275SEric Cheng 	/*
34558275SEric Cheng 	 * There should be no ring with the same ring index in the target
34568275SEric Cheng 	 * group.
34578275SEric Cheng 	 */
34588275SEric Cheng 	ASSERT(mac_find_ring(group, driver_call ? index : ring->mr_index) ==
34598275SEric Cheng 	    NULL);
34608275SEric Cheng 
34618275SEric Cheng 	if (driver_call) {
34628275SEric Cheng 		/*
34638275SEric Cheng 		 * The function is called as a result of a request from
34648275SEric Cheng 		 * a driver to add a ring to an existing group, for example
34658275SEric Cheng 		 * from the aggregation driver. Allocate a new mac_ring_t
34668275SEric Cheng 		 * for that ring.
34678275SEric Cheng 		 */
34688275SEric Cheng 		ring = mac_init_ring(mip, group, index, cap_rings);
34698275SEric Cheng 		ASSERT(group->mrg_state > MAC_GROUP_STATE_UNINIT);
34708275SEric Cheng 	} else {
34718275SEric Cheng 		/*
34728275SEric Cheng 		 * The function is called as a result of a MAC layer request
34738275SEric Cheng 		 * to add a ring to an existing group. In this case the
34748275SEric Cheng 		 * ring is being moved between groups, which requires
34758275SEric Cheng 		 * the underlying driver to support dynamic grouping,
34768275SEric Cheng 		 * and the mac_ring_t already exists.
34778275SEric Cheng 		 */
34788275SEric Cheng 		ASSERT(group_type == MAC_GROUP_TYPE_DYNAMIC);
34798275SEric Cheng 		ASSERT(cap_rings->mr_gaddring != NULL);
34808275SEric Cheng 		ASSERT(ring->mr_gh == NULL);
34818275SEric Cheng 	}
34828275SEric Cheng 
34838275SEric Cheng 	/*
34848275SEric Cheng 	 * At this point the ring should not be in use, and it should be
34858275SEric Cheng 	 * of the right for the target group.
34868275SEric Cheng 	 */
34878275SEric Cheng 	ASSERT(ring->mr_state < MR_INUSE);
34888275SEric Cheng 	ASSERT(ring->mr_srs == NULL);
34898275SEric Cheng 	ASSERT(ring->mr_type == group->mrg_type);
34908275SEric Cheng 
34918275SEric Cheng 	if (!driver_call) {
34928275SEric Cheng 		/*
34938275SEric Cheng 		 * Add the driver level hardware ring if the process was not
34948275SEric Cheng 		 * initiated by the driver, and the target group is not the
34958275SEric Cheng 		 * group.
34968275SEric Cheng 		 */
34978275SEric Cheng 		if (group->mrg_driver != NULL) {
34988275SEric Cheng 			cap_rings->mr_gaddring(group->mrg_driver,
34998275SEric Cheng 			    ring->mr_driver, ring->mr_type);
35008275SEric Cheng 		}
35018275SEric Cheng 
35028275SEric Cheng 		/*
35038275SEric Cheng 		 * Insert the ring ahead existing rings.
35048275SEric Cheng 		 */
35058275SEric Cheng 		ring->mr_next = group->mrg_rings;
35068275SEric Cheng 		group->mrg_rings = ring;
35078275SEric Cheng 		ring->mr_gh = (mac_group_handle_t)group;
35088275SEric Cheng 		group->mrg_cur_count++;
35098275SEric Cheng 	}
35108275SEric Cheng 
35118275SEric Cheng 	/*
35128275SEric Cheng 	 * If the group has not been actively used, we're done.
35138275SEric Cheng 	 */
35148275SEric Cheng 	if (group->mrg_index != -1 &&
35158275SEric Cheng 	    group->mrg_state < MAC_GROUP_STATE_RESERVED)
35168275SEric Cheng 		return (0);
35178275SEric Cheng 
35188275SEric Cheng 	/*
35198275SEric Cheng 	 * Set up SRS/SR according to the ring type.
35208275SEric Cheng 	 */
35218275SEric Cheng 	switch (ring->mr_type) {
35228275SEric Cheng 	case MAC_RING_TYPE_RX:
35238275SEric Cheng 		/*
35248275SEric Cheng 		 * Setup SRS on top of the new ring if the group is
35258275SEric Cheng 		 * reserved for someones exclusive use.
35268275SEric Cheng 		 */
35278275SEric Cheng 		if (group->mrg_state == MAC_GROUP_STATE_RESERVED) {
35288275SEric Cheng 			flow_entry_t *flent;
35298275SEric Cheng 			mac_client_impl_t *mcip;
35308275SEric Cheng 
35318275SEric Cheng 			mcip = MAC_RX_GROUP_ONLY_CLIENT(group);
35328275SEric Cheng 			ASSERT(mcip != NULL);
35338275SEric Cheng 			flent = mcip->mci_flent;
35348275SEric Cheng 			ASSERT(flent->fe_rx_srs_cnt > 0);
35358275SEric Cheng 			mac_srs_group_setup(mcip, flent, group, SRST_LINK);
35368275SEric Cheng 		}
35378275SEric Cheng 		break;
35388275SEric Cheng 	case MAC_RING_TYPE_TX:
35398275SEric Cheng 		/*
35408275SEric Cheng 		 * For TX this function is only invoked during the
35418275SEric Cheng 		 * initial creation of a group when a share is
35428275SEric Cheng 		 * associated with a MAC client. So the datapath is not
35438275SEric Cheng 		 * yet setup, and will be setup later after the
35448275SEric Cheng 		 * group has been reserved and populated.
35458275SEric Cheng 		 */
35468275SEric Cheng 		break;
35478275SEric Cheng 	default:
35488275SEric Cheng 		ASSERT(B_FALSE);
35498275SEric Cheng 	}
35508275SEric Cheng 
35518275SEric Cheng 	/*
35528275SEric Cheng 	 * Start the ring if needed. Failure causes to undo the grouping action.
35538275SEric Cheng 	 */
35548275SEric Cheng 	if ((ret = mac_start_ring(ring)) != 0) {
35558275SEric Cheng 		if (ring->mr_type == MAC_RING_TYPE_RX) {
35568275SEric Cheng 			if (ring->mr_srs != NULL) {
35578275SEric Cheng 				mac_rx_srs_remove(ring->mr_srs);
35588275SEric Cheng 				ring->mr_srs = NULL;
35598275SEric Cheng 			}
35608275SEric Cheng 		}
35618275SEric Cheng 		if (!driver_call) {
35628275SEric Cheng 			cap_rings->mr_gremring(group->mrg_driver,
35638275SEric Cheng 			    ring->mr_driver, ring->mr_type);
35648275SEric Cheng 		}
35658275SEric Cheng 		group->mrg_cur_count--;
35668275SEric Cheng 		group->mrg_rings = ring->mr_next;
35678275SEric Cheng 
35688275SEric Cheng 		ring->mr_gh = NULL;
35698275SEric Cheng 
35708275SEric Cheng 		if (driver_call)
35718275SEric Cheng 			mac_ring_free(mip, ring);
35728275SEric Cheng 
35738275SEric Cheng 		return (ret);
35748275SEric Cheng 	}
35758275SEric Cheng 
35768275SEric Cheng 	/*
35778275SEric Cheng 	 * Update the ring's state.
35788275SEric Cheng 	 */
35798275SEric Cheng 	ring->mr_state = MR_INUSE;
35808275SEric Cheng 	MAC_RING_UNMARK(ring, MR_INCIPIENT);
35818275SEric Cheng 	return (0);
35828275SEric Cheng }
35838275SEric Cheng 
35848275SEric Cheng /*
35858275SEric Cheng  * Remove a ring from it's current group. MAC internal function for dynamic
35868275SEric Cheng  * grouping.
35878275SEric Cheng  *
35888275SEric Cheng  * The caller needs to call mac_perim_enter() before calling this function.
35898275SEric Cheng  */
35908275SEric Cheng void
35918275SEric Cheng i_mac_group_rem_ring(mac_group_t *group, mac_ring_t *ring,
35928275SEric Cheng     boolean_t driver_call)
35938275SEric Cheng {
35948275SEric Cheng 	mac_impl_t *mip = (mac_impl_t *)group->mrg_mh;
35958275SEric Cheng 	mac_capab_rings_t *cap_rings = NULL;
35968275SEric Cheng 	mac_group_type_t group_type;
35978275SEric Cheng 
35988275SEric Cheng 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
35998275SEric Cheng 
36008275SEric Cheng 	ASSERT(mac_find_ring(group, ring->mr_index) == ring);
36018275SEric Cheng 	ASSERT((mac_group_t *)ring->mr_gh == group);
36028275SEric Cheng 	ASSERT(ring->mr_type == group->mrg_type);
36038275SEric Cheng 
36048275SEric Cheng 	switch (ring->mr_type) {
36058275SEric Cheng 	case MAC_RING_TYPE_RX:
36068275SEric Cheng 		group_type = mip->mi_rx_group_type;
36078275SEric Cheng 		cap_rings = &mip->mi_rx_rings_cap;
36088275SEric Cheng 
36098275SEric Cheng 		if (group->mrg_state >= MAC_GROUP_STATE_RESERVED)
36108275SEric Cheng 			mac_stop_ring(ring);
36118275SEric Cheng 
36128275SEric Cheng 		/*
36138275SEric Cheng 		 * Only hardware classified packets hold a reference to the
36148275SEric Cheng 		 * ring all the way up the Rx path. mac_rx_srs_remove()
36158275SEric Cheng 		 * will take care of quiescing the Rx path and removing the
36168275SEric Cheng 		 * SRS. The software classified path neither holds a reference
36178275SEric Cheng 		 * nor any association with the ring in mac_rx.
36188275SEric Cheng 		 */
36198275SEric Cheng 		if (ring->mr_srs != NULL) {
36208275SEric Cheng 			mac_rx_srs_remove(ring->mr_srs);
36218275SEric Cheng 			ring->mr_srs = NULL;
36228275SEric Cheng 		}
36238275SEric Cheng 		ring->mr_state = MR_FREE;
36248275SEric Cheng 		ring->mr_flag = 0;
36258275SEric Cheng 
36268275SEric Cheng 		break;
36278275SEric Cheng 	case MAC_RING_TYPE_TX:
36288275SEric Cheng 		/*
36298275SEric Cheng 		 * For TX this function is only invoked in two
36308275SEric Cheng 		 * cases:
36318275SEric Cheng 		 *
36328275SEric Cheng 		 * 1) In the case of a failure during the
36338275SEric Cheng 		 * initial creation of a group when a share is
36348275SEric Cheng 		 * associated with a MAC client. So the SRS is not
36358275SEric Cheng 		 * yet setup, and will be setup later after the
36368275SEric Cheng 		 * group has been reserved and populated.
36378275SEric Cheng 		 *
36388275SEric Cheng 		 * 2) From mac_release_tx_group() when freeing
36398275SEric Cheng 		 * a TX SRS.
36408275SEric Cheng 		 *
36418275SEric Cheng 		 * In both cases the SRS and its soft rings are
36428275SEric Cheng 		 * already quiesced.
36438275SEric Cheng 		 */
36448275SEric Cheng 		ASSERT(!driver_call);
36458275SEric Cheng 		group_type = mip->mi_tx_group_type;
36468275SEric Cheng 		cap_rings = &mip->mi_tx_rings_cap;
36478275SEric Cheng 		break;
36488275SEric Cheng 	default:
36498275SEric Cheng 		ASSERT(B_FALSE);
36508275SEric Cheng 	}
36518275SEric Cheng 
36528275SEric Cheng 	/*
36538275SEric Cheng 	 * Remove the ring from the group.
36548275SEric Cheng 	 */
36558275SEric Cheng 	if (ring == group->mrg_rings)
36568275SEric Cheng 		group->mrg_rings = ring->mr_next;
36578275SEric Cheng 	else {
36588275SEric Cheng 		mac_ring_t *pre;
36598275SEric Cheng 
36608275SEric Cheng 		pre = group->mrg_rings;
36618275SEric Cheng 		while (pre->mr_next != ring)
36628275SEric Cheng 			pre = pre->mr_next;
36638275SEric Cheng 		pre->mr_next = ring->mr_next;
36648275SEric Cheng 	}
36658275SEric Cheng 	group->mrg_cur_count--;
36668275SEric Cheng 
36678275SEric Cheng 	if (!driver_call) {
36688275SEric Cheng 		ASSERT(group_type == MAC_GROUP_TYPE_DYNAMIC);
36698275SEric Cheng 		ASSERT(cap_rings->mr_gremring != NULL);
36708275SEric Cheng 
36718275SEric Cheng 		/*
36728275SEric Cheng 		 * Remove the driver level hardware ring.
36738275SEric Cheng 		 */
36748275SEric Cheng 		if (group->mrg_driver != NULL) {
36758275SEric Cheng 			cap_rings->mr_gremring(group->mrg_driver,
36768275SEric Cheng 			    ring->mr_driver, ring->mr_type);
36778275SEric Cheng 		}
36788275SEric Cheng 	}
36798275SEric Cheng 
36808275SEric Cheng 	ring->mr_gh = NULL;
36818275SEric Cheng 	if (driver_call) {
36828275SEric Cheng 		mac_ring_free(mip, ring);
36838275SEric Cheng 	} else {
36848275SEric Cheng 		ring->mr_state = MR_FREE;
36858275SEric Cheng 		ring->mr_flag = 0;
36868275SEric Cheng 	}
36878275SEric Cheng }
36888275SEric Cheng 
36898275SEric Cheng /*
36908275SEric Cheng  * Move a ring to the target group. If needed, remove the ring from the group
36918275SEric Cheng  * that it currently belongs to.
36928275SEric Cheng  *
36938275SEric Cheng  * The caller need to enter MAC's perimeter by calling mac_perim_enter().
36948275SEric Cheng  */
36958275SEric Cheng static int
36968275SEric Cheng mac_group_mov_ring(mac_impl_t *mip, mac_group_t *d_group, mac_ring_t *ring)
36978275SEric Cheng {
36988275SEric Cheng 	mac_group_t *s_group = (mac_group_t *)ring->mr_gh;
36998275SEric Cheng 	int rv;
37008275SEric Cheng 
37018275SEric Cheng 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
37028275SEric Cheng 	ASSERT(d_group != NULL);
37038275SEric Cheng 	ASSERT(s_group->mrg_mh == d_group->mrg_mh);
37048275SEric Cheng 
37058275SEric Cheng 	if (s_group == d_group)
37068275SEric Cheng 		return (0);
37078275SEric Cheng 
37088275SEric Cheng 	/*
37098275SEric Cheng 	 * Remove it from current group first.
37108275SEric Cheng 	 */
37118275SEric Cheng 	if (s_group != NULL)
37128275SEric Cheng 		i_mac_group_rem_ring(s_group, ring, B_FALSE);
37138275SEric Cheng 
37148275SEric Cheng 	/*
37158275SEric Cheng 	 * Add it to the new group.
37168275SEric Cheng 	 */
37178275SEric Cheng 	rv = i_mac_group_add_ring(d_group, ring, 0);
37188275SEric Cheng 	if (rv != 0) {
37198275SEric Cheng 		/*
37208275SEric Cheng 		 * Failed to add ring back to source group. If
37218275SEric Cheng 		 * that fails, the ring is stuck in limbo, log message.
37228275SEric Cheng 		 */
37238275SEric Cheng 		if (i_mac_group_add_ring(s_group, ring, 0)) {
37248275SEric Cheng 			cmn_err(CE_WARN, "%s: failed to move ring %p\n",
37258275SEric Cheng 			    mip->mi_name, (void *)ring);
37268275SEric Cheng 		}
37278275SEric Cheng 	}
37288275SEric Cheng 
37298275SEric Cheng 	return (rv);
37308275SEric Cheng }
37318275SEric Cheng 
37328275SEric Cheng /*
37338275SEric Cheng  * Find a MAC address according to its value.
37348275SEric Cheng  */
37358275SEric Cheng mac_address_t *
37368275SEric Cheng mac_find_macaddr(mac_impl_t *mip, uint8_t *mac_addr)
37378275SEric Cheng {
37388275SEric Cheng 	mac_address_t *map;
37398275SEric Cheng 
37408275SEric Cheng 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
37418275SEric Cheng 
37428275SEric Cheng 	for (map = mip->mi_addresses; map != NULL; map = map->ma_next) {
37438275SEric Cheng 		if (bcmp(mac_addr, map->ma_addr, map->ma_len) == 0)
37448275SEric Cheng 			break;
37458275SEric Cheng 	}
37468275SEric Cheng 
37478275SEric Cheng 	return (map);
37488275SEric Cheng }
37498275SEric Cheng 
37508275SEric Cheng /*
37518275SEric Cheng  * Check whether the MAC address is shared by multiple clients.
37528275SEric Cheng  */
37538275SEric Cheng boolean_t
37548275SEric Cheng mac_check_macaddr_shared(mac_address_t *map)
37558275SEric Cheng {
37568275SEric Cheng 	ASSERT(MAC_PERIM_HELD((mac_handle_t)map->ma_mip));
37578275SEric Cheng 
37588275SEric Cheng 	return (map->ma_nusers > 1);
37598275SEric Cheng }
37608275SEric Cheng 
37618275SEric Cheng /*
37628275SEric Cheng  * Remove the specified MAC address from the MAC address list and free it.
37638275SEric Cheng  */
37648275SEric Cheng static void
37658275SEric Cheng mac_free_macaddr(mac_address_t *map)
37668275SEric Cheng {
37678275SEric Cheng 	mac_impl_t *mip = map->ma_mip;
37688275SEric Cheng 
37698275SEric Cheng 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
37708275SEric Cheng 	ASSERT(mip->mi_addresses != NULL);
37718275SEric Cheng 
37728275SEric Cheng 	map = mac_find_macaddr(mip, map->ma_addr);
37738275SEric Cheng 
37748275SEric Cheng 	ASSERT(map != NULL);
37758275SEric Cheng 	ASSERT(map->ma_nusers == 0);
37768275SEric Cheng 
37778275SEric Cheng 	if (map == mip->mi_addresses) {
37788275SEric Cheng 		mip->mi_addresses = map->ma_next;
37798275SEric Cheng 	} else {
37808275SEric Cheng 		mac_address_t *pre;
37818275SEric Cheng 
37828275SEric Cheng 		pre = mip->mi_addresses;
37838275SEric Cheng 		while (pre->ma_next != map)
37848275SEric Cheng 			pre = pre->ma_next;
37858275SEric Cheng 		pre->ma_next = map->ma_next;
37868275SEric Cheng 	}
37878275SEric Cheng 
37888275SEric Cheng 	kmem_free(map, sizeof (mac_address_t));
37898275SEric Cheng }
37908275SEric Cheng 
37918275SEric Cheng /*
37928275SEric Cheng  * Add a MAC address reference for a client. If the desired MAC address
37938275SEric Cheng  * exists, add a reference to it. Otherwise, add the new address by adding
37948275SEric Cheng  * it to a reserved group or setting promiscuous mode. Won't try different
37958275SEric Cheng  * group is the group is non-NULL, so the caller must explictly share
37968275SEric Cheng  * default group when needed.
37978275SEric Cheng  *
37988275SEric Cheng  * Note, the primary MAC address is initialized at registration time, so
37998275SEric Cheng  * to add it to default group only need to activate it if its reference
38008275SEric Cheng  * count is still zero. Also, some drivers may not have advertised RINGS
38018275SEric Cheng  * capability.
38028275SEric Cheng  */
38038275SEric Cheng int
3804*8400SNicolas.Droux@Sun.COM mac_add_macaddr(mac_impl_t *mip, mac_group_t *group, uint8_t *mac_addr,
3805*8400SNicolas.Droux@Sun.COM     boolean_t use_hw)
38068275SEric Cheng {
38078275SEric Cheng 	mac_address_t *map;
38088275SEric Cheng 	int err = 0;
38098275SEric Cheng 	boolean_t allocated_map = B_FALSE;
38108275SEric Cheng 
38118275SEric Cheng 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
38128275SEric Cheng 
38138275SEric Cheng 	map = mac_find_macaddr(mip, mac_addr);
38148275SEric Cheng 
38158275SEric Cheng 	/*
38168275SEric Cheng 	 * If the new MAC address has not been added. Allocate a new one
38178275SEric Cheng 	 * and set it up.
38188275SEric Cheng 	 */
38198275SEric Cheng 	if (map == NULL) {
38208275SEric Cheng 		map = kmem_zalloc(sizeof (mac_address_t), KM_SLEEP);
38218275SEric Cheng 		map->ma_len = mip->mi_type->mt_addr_length;
38228275SEric Cheng 		bcopy(mac_addr, map->ma_addr, map->ma_len);
38238275SEric Cheng 		map->ma_nusers = 0;
38248275SEric Cheng 		map->ma_group = group;
38258275SEric Cheng 		map->ma_mip = mip;
38268275SEric Cheng 
38278275SEric Cheng 		/* add the new MAC address to the head of the address list */
38288275SEric Cheng 		map->ma_next = mip->mi_addresses;
38298275SEric Cheng 		mip->mi_addresses = map;
38308275SEric Cheng 
38318275SEric Cheng 		allocated_map = B_TRUE;
38328275SEric Cheng 	}
38338275SEric Cheng 
38348275SEric Cheng 	ASSERT(map->ma_group == group);
38358275SEric Cheng 
38368275SEric Cheng 	/*
38378275SEric Cheng 	 * If the MAC address is already in use, simply account for the
38388275SEric Cheng 	 * new client.
38398275SEric Cheng 	 */
38408275SEric Cheng 	if (map->ma_nusers++ > 0)
38418275SEric Cheng 		return (0);
38428275SEric Cheng 
38438275SEric Cheng 	/*
38448275SEric Cheng 	 * Activate this MAC address by adding it to the reserved group.
38458275SEric Cheng 	 */
38468275SEric Cheng 	if (group != NULL) {
38478275SEric Cheng 		err = mac_group_addmac(group, (const uint8_t *)mac_addr);
38488275SEric Cheng 		if (err == 0) {
38498275SEric Cheng 			map->ma_type = MAC_ADDRESS_TYPE_UNICAST_CLASSIFIED;
38508275SEric Cheng 			return (0);
38518275SEric Cheng 		}
38528275SEric Cheng 	}
38538275SEric Cheng 
38548275SEric Cheng 	/*
3855*8400SNicolas.Droux@Sun.COM 	 * The MAC address addition failed. If the client requires a
3856*8400SNicolas.Droux@Sun.COM 	 * hardware classified MAC address, fail the operation.
3857*8400SNicolas.Droux@Sun.COM 	 */
3858*8400SNicolas.Droux@Sun.COM 	if (use_hw) {
3859*8400SNicolas.Droux@Sun.COM 		err = ENOSPC;
3860*8400SNicolas.Droux@Sun.COM 		goto bail;
3861*8400SNicolas.Droux@Sun.COM 	}
3862*8400SNicolas.Droux@Sun.COM 
3863*8400SNicolas.Droux@Sun.COM 	/*
3864*8400SNicolas.Droux@Sun.COM 	 * Try promiscuous mode.
3865*8400SNicolas.Droux@Sun.COM 	 *
3866*8400SNicolas.Droux@Sun.COM 	 * For drivers that don't advertise RINGS capability, do
3867*8400SNicolas.Droux@Sun.COM 	 * nothing for the primary address.
38688275SEric Cheng 	 */
3869*8400SNicolas.Droux@Sun.COM 	if ((group == NULL) &&
3870*8400SNicolas.Droux@Sun.COM 	    (bcmp(map->ma_addr, mip->mi_addr, map->ma_len) == 0)) {
3871*8400SNicolas.Droux@Sun.COM 		map->ma_type = MAC_ADDRESS_TYPE_UNICAST_CLASSIFIED;
3872*8400SNicolas.Droux@Sun.COM 		return (0);
3873*8400SNicolas.Droux@Sun.COM 	}
3874*8400SNicolas.Droux@Sun.COM 
3875*8400SNicolas.Droux@Sun.COM 	/*
3876*8400SNicolas.Droux@Sun.COM 	 * Enable promiscuous mode in order to receive traffic
3877*8400SNicolas.Droux@Sun.COM 	 * to the new MAC address.
3878*8400SNicolas.Droux@Sun.COM 	 */
3879*8400SNicolas.Droux@Sun.COM 	if ((err = i_mac_promisc_set(mip, B_TRUE, MAC_DEVPROMISC)) == 0) {
3880*8400SNicolas.Droux@Sun.COM 		map->ma_type = MAC_ADDRESS_TYPE_UNICAST_PROMISC;
3881*8400SNicolas.Droux@Sun.COM 		return (0);
38828275SEric Cheng 	}
38838275SEric Cheng 
38848275SEric Cheng 	/*
38858275SEric Cheng 	 * Free the MAC address that could not be added. Don't free
38868275SEric Cheng 	 * a pre-existing address, it could have been the entry
38878275SEric Cheng 	 * for the primary MAC address which was pre-allocated by
38888275SEric Cheng 	 * mac_init_macaddr(), and which must remain on the list.
38898275SEric Cheng 	 */
3890*8400SNicolas.Droux@Sun.COM bail:
38918275SEric Cheng 	map->ma_nusers--;
38928275SEric Cheng 	if (allocated_map)
38938275SEric Cheng 		mac_free_macaddr(map);
38948275SEric Cheng 	return (err);
38958275SEric Cheng }
38968275SEric Cheng 
38978275SEric Cheng /*
38988275SEric Cheng  * Remove a reference to a MAC address. This may cause to remove the MAC
38998275SEric Cheng  * address from an associated group or to turn off promiscuous mode.
39008275SEric Cheng  * The caller needs to handle the failure properly.
39018275SEric Cheng  */
39028275SEric Cheng int
39038275SEric Cheng mac_remove_macaddr(mac_address_t *map)
39048275SEric Cheng {
39058275SEric Cheng 	mac_impl_t *mip = map->ma_mip;
39068275SEric Cheng 	int err = 0;
39078275SEric Cheng 
39088275SEric Cheng 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
39098275SEric Cheng 
39108275SEric Cheng 	ASSERT(map == mac_find_macaddr(mip, map->ma_addr));
39118275SEric Cheng 
39128275SEric Cheng 	/*
39138275SEric Cheng 	 * If it's not the last client using this MAC address, only update
39148275SEric Cheng 	 * the MAC clients count.
39158275SEric Cheng 	 */
39168275SEric Cheng 	if (--map->ma_nusers > 0)
39178275SEric Cheng 		return (0);
39188275SEric Cheng 
39198275SEric Cheng 	/*
39208275SEric Cheng 	 * The MAC address is no longer used by any MAC client, so remove
39218275SEric Cheng 	 * it from its associated group, or turn off promiscuous mode
39228275SEric Cheng 	 * if it was enabled for the MAC address.
39238275SEric Cheng 	 */
39248275SEric Cheng 	switch (map->ma_type) {
39258275SEric Cheng 	case MAC_ADDRESS_TYPE_UNICAST_CLASSIFIED:
39268275SEric Cheng 		/*
39278275SEric Cheng 		 * Don't free the preset primary address for drivers that
39288275SEric Cheng 		 * don't advertise RINGS capability.
39298275SEric Cheng 		 */
39308275SEric Cheng 		if (map->ma_group == NULL)
39318275SEric Cheng 			return (0);
39328275SEric Cheng 
39338275SEric Cheng 		err = mac_group_remmac(map->ma_group, map->ma_addr);
39348275SEric Cheng 		break;
39358275SEric Cheng 	case MAC_ADDRESS_TYPE_UNICAST_PROMISC:
3936*8400SNicolas.Droux@Sun.COM 		err = i_mac_promisc_set(mip, B_FALSE, MAC_DEVPROMISC);
39378275SEric Cheng 		break;
39388275SEric Cheng 	default:
39398275SEric Cheng 		ASSERT(B_FALSE);
39408275SEric Cheng 	}
39418275SEric Cheng 
39428275SEric Cheng 	if (err != 0)
39438275SEric Cheng 		return (err);
39448275SEric Cheng 
39458275SEric Cheng 	/*
39468275SEric Cheng 	 * We created MAC address for the primary one at registration, so we
39478275SEric Cheng 	 * won't free it here. mac_fini_macaddr() will take care of it.
39488275SEric Cheng 	 */
39498275SEric Cheng 	if (bcmp(map->ma_addr, mip->mi_addr, map->ma_len) != 0)
39508275SEric Cheng 		mac_free_macaddr(map);
39518275SEric Cheng 
39528275SEric Cheng 	return (0);
39538275SEric Cheng }
39548275SEric Cheng 
39558275SEric Cheng /*
39568275SEric Cheng  * Update an existing MAC address. The caller need to make sure that the new
39578275SEric Cheng  * value has not been used.
39588275SEric Cheng  */
39598275SEric Cheng int
39608275SEric Cheng mac_update_macaddr(mac_address_t *map, uint8_t *mac_addr)
39618275SEric Cheng {
39628275SEric Cheng 	mac_impl_t *mip = map->ma_mip;
39638275SEric Cheng 	int err = 0;
39648275SEric Cheng 
39658275SEric Cheng 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
39668275SEric Cheng 	ASSERT(mac_find_macaddr(mip, mac_addr) == NULL);
39678275SEric Cheng 
39688275SEric Cheng 	switch (map->ma_type) {
39698275SEric Cheng 	case MAC_ADDRESS_TYPE_UNICAST_CLASSIFIED:
39708275SEric Cheng 		/*
39718275SEric Cheng 		 * Update the primary address for drivers that are not
39728275SEric Cheng 		 * RINGS capable.
39738275SEric Cheng 		 */
39748275SEric Cheng 		if (map->ma_group == NULL) {
39758275SEric Cheng 			err = mip->mi_unicst(mip->mi_driver, (const uint8_t *)
39768275SEric Cheng 			    mac_addr);
39778275SEric Cheng 			if (err != 0)
39788275SEric Cheng 				return (err);
39798275SEric Cheng 			break;
39808275SEric Cheng 		}
39818275SEric Cheng 
39828275SEric Cheng 		/*
39838275SEric Cheng 		 * If this MAC address is not currently in use,
39848275SEric Cheng 		 * simply break out and update the value.
39858275SEric Cheng 		 */
39868275SEric Cheng 		if (map->ma_nusers == 0)
39878275SEric Cheng 			break;
39888275SEric Cheng 
39898275SEric Cheng 		/*
39908275SEric Cheng 		 * Need to replace the MAC address associated with a group.
39918275SEric Cheng 		 */
39928275SEric Cheng 		err = mac_group_remmac(map->ma_group, map->ma_addr);
39938275SEric Cheng 		if (err != 0)
39948275SEric Cheng 			return (err);
39958275SEric Cheng 
39968275SEric Cheng 		err = mac_group_addmac(map->ma_group, mac_addr);
39978275SEric Cheng 
39988275SEric Cheng 		/*
39998275SEric Cheng 		 * Failure hints hardware error. The MAC layer needs to
40008275SEric Cheng 		 * have error notification facility to handle this.
40018275SEric Cheng 		 * Now, simply try to restore the value.
40028275SEric Cheng 		 */
40038275SEric Cheng 		if (err != 0)
40048275SEric Cheng 			(void) mac_group_addmac(map->ma_group, map->ma_addr);
40058275SEric Cheng 
40068275SEric Cheng 		break;
40078275SEric Cheng 	case MAC_ADDRESS_TYPE_UNICAST_PROMISC:
40088275SEric Cheng 		/*
40098275SEric Cheng 		 * Need to do nothing more if in promiscuous mode.
40108275SEric Cheng 		 */
40118275SEric Cheng 		break;
40128275SEric Cheng 	default:
40138275SEric Cheng 		ASSERT(B_FALSE);
40148275SEric Cheng 	}
40158275SEric Cheng 
40168275SEric Cheng 	/*
40178275SEric Cheng 	 * Successfully replaced the MAC address.
40188275SEric Cheng 	 */
40198275SEric Cheng 	if (err == 0)
40208275SEric Cheng 		bcopy(mac_addr, map->ma_addr, map->ma_len);
40218275SEric Cheng 
40228275SEric Cheng 	return (err);
40238275SEric Cheng }
40248275SEric Cheng 
40258275SEric Cheng /*
40268275SEric Cheng  * Freshen the MAC address with new value. Its caller must have updated the
40278275SEric Cheng  * hardware MAC address before calling this function.
40288275SEric Cheng  * This funcitons is supposed to be used to handle the MAC address change
40298275SEric Cheng  * notification from underlying drivers.
40308275SEric Cheng  */
40318275SEric Cheng void
40328275SEric Cheng mac_freshen_macaddr(mac_address_t *map, uint8_t *mac_addr)
40338275SEric Cheng {
40348275SEric Cheng 	mac_impl_t *mip = map->ma_mip;
40358275SEric Cheng 
40368275SEric Cheng 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
40378275SEric Cheng 	ASSERT(mac_find_macaddr(mip, mac_addr) == NULL);
40388275SEric Cheng 
40398275SEric Cheng 	/*
40408275SEric Cheng 	 * Freshen the MAC address with new value.
40418275SEric Cheng 	 */
40428275SEric Cheng 	bcopy(mac_addr, map->ma_addr, map->ma_len);
40438275SEric Cheng 	bcopy(mac_addr, mip->mi_addr, map->ma_len);
40448275SEric Cheng 
40458275SEric Cheng 	/*
40468275SEric Cheng 	 * Update all MAC clients that share this MAC address.
40478275SEric Cheng 	 */
40488275SEric Cheng 	mac_unicast_update_clients(mip, map);
40498275SEric Cheng }
40508275SEric Cheng 
40518275SEric Cheng /*
40528275SEric Cheng  * Set up the primary MAC address.
40538275SEric Cheng  */
40548275SEric Cheng void
40558275SEric Cheng mac_init_macaddr(mac_impl_t *mip)
40568275SEric Cheng {
40578275SEric Cheng 	mac_address_t *map;
40588275SEric Cheng 
40598275SEric Cheng 	/*
40608275SEric Cheng 	 * The reference count is initialized to zero, until it's really
40618275SEric Cheng 	 * activated.
40628275SEric Cheng 	 */
40638275SEric Cheng 	map = kmem_zalloc(sizeof (mac_address_t), KM_SLEEP);
40648275SEric Cheng 	map->ma_len = mip->mi_type->mt_addr_length;
40658275SEric Cheng 	bcopy(mip->mi_addr, map->ma_addr, map->ma_len);
40668275SEric Cheng 
40678275SEric Cheng 	/*
40688275SEric Cheng 	 * If driver advertises RINGS capability, it shouldn't have initialized
40698275SEric Cheng 	 * its primary MAC address. For other drivers, including VNIC, the
40708275SEric Cheng 	 * primary address must work after registration.
40718275SEric Cheng 	 */
40728275SEric Cheng 	if (mip->mi_rx_groups == NULL)
40738275SEric Cheng 		map->ma_type = MAC_ADDRESS_TYPE_UNICAST_CLASSIFIED;
40748275SEric Cheng 
40758275SEric Cheng 	/*
40768275SEric Cheng 	 * The primary MAC address is reserved for default group according
40778275SEric Cheng 	 * to current design.
40788275SEric Cheng 	 */
40798275SEric Cheng 	map->ma_group = mip->mi_rx_groups;
40808275SEric Cheng 	map->ma_mip = mip;
40818275SEric Cheng 
40828275SEric Cheng 	mip->mi_addresses = map;
40838275SEric Cheng }
40848275SEric Cheng 
40858275SEric Cheng /*
40868275SEric Cheng  * Clean up the primary MAC address. Note, only one primary MAC address
40878275SEric Cheng  * is allowed. All other MAC addresses must have been freed appropriately.
40888275SEric Cheng  */
40898275SEric Cheng void
40908275SEric Cheng mac_fini_macaddr(mac_impl_t *mip)
40918275SEric Cheng {
40928275SEric Cheng 	mac_address_t *map = mip->mi_addresses;
40938275SEric Cheng 
40948275SEric Cheng 	/* there should be exactly one entry left on the list */
40958275SEric Cheng 	ASSERT(map != NULL);
40968275SEric Cheng 	ASSERT(map->ma_nusers == 0);
40978275SEric Cheng 	ASSERT(map->ma_next == NULL);
40988275SEric Cheng 
40998275SEric Cheng 	kmem_free(map, sizeof (mac_address_t));
41008275SEric Cheng 	mip->mi_addresses = NULL;
41018275SEric Cheng }
41028275SEric Cheng 
41038275SEric Cheng /*
41048275SEric Cheng  * Logging related functions.
41058275SEric Cheng  */
41068275SEric Cheng 
41078275SEric Cheng /* Write the Flow description to the log file */
41088275SEric Cheng int
41098275SEric Cheng mac_write_flow_desc(flow_entry_t *flent, mac_client_impl_t *mcip)
41108275SEric Cheng {
41118275SEric Cheng 	flow_desc_t		*fdesc;
41128275SEric Cheng 	mac_resource_props_t	*mrp;
41138275SEric Cheng 	net_desc_t		ndesc;
41148275SEric Cheng 
41158275SEric Cheng 	bzero(&ndesc, sizeof (net_desc_t));
41168275SEric Cheng 
41178275SEric Cheng 	/*
41188275SEric Cheng 	 * Grab the fe_lock to see a self-consistent fe_flow_desc.
41198275SEric Cheng 	 * Updates to the fe_flow_desc are done under the fe_lock
41208275SEric Cheng 	 */
41218275SEric Cheng 	mutex_enter(&flent->fe_lock);
41228275SEric Cheng 	fdesc = &flent->fe_flow_desc;
41238275SEric Cheng 	mrp = &flent->fe_resource_props;
41248275SEric Cheng 
41258275SEric Cheng 	ndesc.nd_name = flent->fe_flow_name;
41268275SEric Cheng 	ndesc.nd_devname = mcip->mci_name;
41278275SEric Cheng 	bcopy(fdesc->fd_src_mac, ndesc.nd_ehost, ETHERADDRL);
41288275SEric Cheng 	bcopy(fdesc->fd_dst_mac, ndesc.nd_edest, ETHERADDRL);
41298275SEric Cheng 	ndesc.nd_sap = htonl(fdesc->fd_sap);
41308275SEric Cheng 	ndesc.nd_isv4 = (uint8_t)fdesc->fd_ipversion == IPV4_VERSION;
41318275SEric Cheng 	ndesc.nd_bw_limit = mrp->mrp_maxbw;
41328275SEric Cheng 	if (ndesc.nd_isv4) {
41338275SEric Cheng 		ndesc.nd_saddr[3] = htonl(fdesc->fd_local_addr.s6_addr32[3]);
41348275SEric Cheng 		ndesc.nd_daddr[3] = htonl(fdesc->fd_remote_addr.s6_addr32[3]);
41358275SEric Cheng 	} else {
41368275SEric Cheng 		bcopy(&fdesc->fd_local_addr, ndesc.nd_saddr, IPV6_ADDR_LEN);
41378275SEric Cheng 		bcopy(&fdesc->fd_remote_addr, ndesc.nd_daddr, IPV6_ADDR_LEN);
41388275SEric Cheng 	}
41398275SEric Cheng 	ndesc.nd_sport = htons(fdesc->fd_local_port);
41408275SEric Cheng 	ndesc.nd_dport = htons(fdesc->fd_remote_port);
41418275SEric Cheng 	ndesc.nd_protocol = (uint8_t)fdesc->fd_protocol;
41428275SEric Cheng 	mutex_exit(&flent->fe_lock);
41438275SEric Cheng 
41448275SEric Cheng 	return (exacct_commit_netinfo((void *)&ndesc, EX_NET_FLDESC_REC));
41458275SEric Cheng }
41468275SEric Cheng 
41478275SEric Cheng /* Write the Flow statistics to the log file */
41488275SEric Cheng int
41498275SEric Cheng mac_write_flow_stats(flow_entry_t *flent)
41508275SEric Cheng {
41518275SEric Cheng 	flow_stats_t	*fl_stats;
41528275SEric Cheng 	net_stat_t	nstat;
41538275SEric Cheng 
41548275SEric Cheng 	fl_stats = &flent->fe_flowstats;
41558275SEric Cheng 	nstat.ns_name = flent->fe_flow_name;
41568275SEric Cheng 	nstat.ns_ibytes = fl_stats->fs_rbytes;
41578275SEric Cheng 	nstat.ns_obytes = fl_stats->fs_obytes;
41588275SEric Cheng 	nstat.ns_ipackets = fl_stats->fs_ipackets;
41598275SEric Cheng 	nstat.ns_opackets = fl_stats->fs_opackets;
41608275SEric Cheng 	nstat.ns_ierrors = fl_stats->fs_ierrors;
41618275SEric Cheng 	nstat.ns_oerrors = fl_stats->fs_oerrors;
41628275SEric Cheng 
41638275SEric Cheng 	return (exacct_commit_netinfo((void *)&nstat, EX_NET_FLSTAT_REC));
41648275SEric Cheng }
41658275SEric Cheng 
41668275SEric Cheng /* Write the Link Description to the log file */
41678275SEric Cheng int
41688275SEric Cheng mac_write_link_desc(mac_client_impl_t *mcip)
41698275SEric Cheng {
41708275SEric Cheng 	net_desc_t		ndesc;
41718275SEric Cheng 	flow_entry_t		*flent = mcip->mci_flent;
41728275SEric Cheng 
41738275SEric Cheng 	bzero(&ndesc, sizeof (net_desc_t));
41748275SEric Cheng 
41758275SEric Cheng 	ndesc.nd_name = mcip->mci_name;
41768275SEric Cheng 	ndesc.nd_devname = mcip->mci_name;
41778275SEric Cheng 	ndesc.nd_isv4 = B_TRUE;
41788275SEric Cheng 	/*
41798275SEric Cheng 	 * Grab the fe_lock to see a self-consistent fe_flow_desc.
41808275SEric Cheng 	 * Updates to the fe_flow_desc are done under the fe_lock
41818275SEric Cheng 	 * after removing the flent from the flow table.
41828275SEric Cheng 	 */
41838275SEric Cheng 	mutex_enter(&flent->fe_lock);
41848275SEric Cheng 	bcopy(flent->fe_flow_desc.fd_src_mac, ndesc.nd_ehost, ETHERADDRL);
41858275SEric Cheng 	mutex_exit(&flent->fe_lock);
41868275SEric Cheng 
41878275SEric Cheng 	return (exacct_commit_netinfo((void *)&ndesc, EX_NET_LNDESC_REC));
41888275SEric Cheng }
41898275SEric Cheng 
41908275SEric Cheng /* Write the Link statistics to the log file */
41918275SEric Cheng int
41928275SEric Cheng mac_write_link_stats(mac_client_impl_t *mcip)
41938275SEric Cheng {
41948275SEric Cheng 	net_stat_t	nstat;
41958275SEric Cheng 
41968275SEric Cheng 	nstat.ns_name = mcip->mci_name;
41978275SEric Cheng 	nstat.ns_ibytes = mcip->mci_stat_ibytes;
41988275SEric Cheng 	nstat.ns_obytes = mcip->mci_stat_obytes;
41998275SEric Cheng 	nstat.ns_ipackets = mcip->mci_stat_ipackets;
42008275SEric Cheng 	nstat.ns_opackets = mcip->mci_stat_opackets;
42018275SEric Cheng 	nstat.ns_ierrors = mcip->mci_stat_ierrors;
42028275SEric Cheng 	nstat.ns_oerrors = mcip->mci_stat_oerrors;
42038275SEric Cheng 
42048275SEric Cheng 	return (exacct_commit_netinfo((void *)&nstat, EX_NET_LNSTAT_REC));
42058275SEric Cheng }
42068275SEric Cheng 
42078275SEric Cheng /*
42088275SEric Cheng  * For a given flow, if the descrition has not been logged before, do it now.
42098275SEric Cheng  * If it is a VNIC, then we have collected information about it from the MAC
42108275SEric Cheng  * table, so skip it.
42118275SEric Cheng  */
42128275SEric Cheng /*ARGSUSED*/
42138275SEric Cheng static int
42148275SEric Cheng mac_log_flowinfo(flow_entry_t *flent, void *args)
42158275SEric Cheng {
42168275SEric Cheng 	mac_client_impl_t	*mcip = flent->fe_mcip;
42178275SEric Cheng 
42188275SEric Cheng 	if (mcip == NULL)
42198275SEric Cheng 		return (0);
42208275SEric Cheng 
42218275SEric Cheng 	/*
42228275SEric Cheng 	 * If the name starts with "vnic", and fe_user_generated is true (to
42238275SEric Cheng 	 * exclude the mcast and active flow entries created implicitly for
42248275SEric Cheng 	 * a vnic, it is a VNIC flow.  i.e. vnic1 is a vnic flow,
42258275SEric Cheng 	 * vnic/bge1/mcast1 is not and neither is vnic/bge1/active.
42268275SEric Cheng 	 */
42278275SEric Cheng 	if (strncasecmp(flent->fe_flow_name, "vnic", 4) == 0 &&
42288275SEric Cheng 	    (flent->fe_type & FLOW_USER) != 0) {
42298275SEric Cheng 		return (0);
42308275SEric Cheng 	}
42318275SEric Cheng 
42328275SEric Cheng 	if (!flent->fe_desc_logged) {
42338275SEric Cheng 		/*
42348275SEric Cheng 		 * We don't return error because we want to continu the
42358275SEric Cheng 		 * walk in case this is the last walk which means we
42368275SEric Cheng 		 * need to reset fe_desc_logged in all the flows.
42378275SEric Cheng 		 */
42388275SEric Cheng 		if (mac_write_flow_desc(flent, mcip) != 0)
42398275SEric Cheng 			return (0);
42408275SEric Cheng 		flent->fe_desc_logged = B_TRUE;
42418275SEric Cheng 	}
42428275SEric Cheng 
42438275SEric Cheng 	/*
42448275SEric Cheng 	 * Regardless of the error, we want to proceed in case we have to
42458275SEric Cheng 	 * reset fe_desc_logged.
42468275SEric Cheng 	 */
42478275SEric Cheng 	(void) mac_write_flow_stats(flent);
42488275SEric Cheng 
42498275SEric Cheng 	if (mcip != NULL && !(mcip->mci_state_flags & MCIS_DESC_LOGGED))
42508275SEric Cheng 		flent->fe_desc_logged = B_FALSE;
42518275SEric Cheng 
42528275SEric Cheng 	return (0);
42538275SEric Cheng }
42548275SEric Cheng 
42558275SEric Cheng typedef struct i_mac_log_state_s {
42568275SEric Cheng 	boolean_t	mi_last;
42578275SEric Cheng 	int		mi_fenable;
42588275SEric Cheng 	int		mi_lenable;
42598275SEric Cheng } i_mac_log_state_t;
42608275SEric Cheng 
42618275SEric Cheng /*
42628275SEric Cheng  * Walk the mac_impl_ts and log the description for each mac client of this mac,
42638275SEric Cheng  * if it hasn't already been done. Additionally, log statistics for the link as
42648275SEric Cheng  * well. Walk the flow table and log information for each flow as well.
42658275SEric Cheng  * If it is the last walk (mci_last), then we turn off mci_desc_logged (and
42668275SEric Cheng  * also fe_desc_logged, if flow logging is on) since we want to log the
42678275SEric Cheng  * description if and when logging is restarted.
42688275SEric Cheng  */
42698275SEric Cheng /*ARGSUSED*/
42708275SEric Cheng static uint_t
42718275SEric Cheng i_mac_log_walker(mod_hash_key_t key, mod_hash_val_t *val, void *arg)
42728275SEric Cheng {
42738275SEric Cheng 	mac_impl_t		*mip = (mac_impl_t *)val;
42748275SEric Cheng 	i_mac_log_state_t	*lstate = (i_mac_log_state_t *)arg;
42758275SEric Cheng 	int			ret;
42768275SEric Cheng 	mac_client_impl_t	*mcip;
42778275SEric Cheng 
42788275SEric Cheng 	/*
42798275SEric Cheng 	 * Only walk the client list for NIC and etherstub
42808275SEric Cheng 	 */
42818275SEric Cheng 	if ((mip->mi_state_flags & MIS_DISABLED) ||
42828275SEric Cheng 	    ((mip->mi_state_flags & MIS_IS_VNIC) &&
42838275SEric Cheng 	    (mac_get_lower_mac_handle((mac_handle_t)mip) != NULL)))
42848275SEric Cheng 		return (MH_WALK_CONTINUE);
42858275SEric Cheng 
42868275SEric Cheng 	for (mcip = mip->mi_clients_list; mcip != NULL;
42878275SEric Cheng 	    mcip = mcip->mci_client_next) {
42888275SEric Cheng 		if (!MCIP_DATAPATH_SETUP(mcip))
42898275SEric Cheng 			continue;
42908275SEric Cheng 		if (lstate->mi_lenable) {
42918275SEric Cheng 			if (!(mcip->mci_state_flags & MCIS_DESC_LOGGED)) {
42928275SEric Cheng 				ret = mac_write_link_desc(mcip);
42938275SEric Cheng 				if (ret != 0) {
42948275SEric Cheng 				/*
42958275SEric Cheng 				 * We can't terminate it if this is the last
42968275SEric Cheng 				 * walk, else there might be some links with
42978275SEric Cheng 				 * mi_desc_logged set to true, which means
42988275SEric Cheng 				 * their description won't be logged the next
42998275SEric Cheng 				 * time logging is started (similarly for the
43008275SEric Cheng 				 * flows within such links). We can continue
43018275SEric Cheng 				 * without walking the flow table (i.e. to
43028275SEric Cheng 				 * set fe_desc_logged to false) because we
43038275SEric Cheng 				 * won't have written any flow stuff for this
43048275SEric Cheng 				 * link as we haven't logged the link itself.
43058275SEric Cheng 				 */
43068275SEric Cheng 					if (lstate->mi_last)
43078275SEric Cheng 						return (MH_WALK_CONTINUE);
43088275SEric Cheng 					else
43098275SEric Cheng 						return (MH_WALK_TERMINATE);
43108275SEric Cheng 				}
43118275SEric Cheng 				mcip->mci_state_flags |= MCIS_DESC_LOGGED;
43128275SEric Cheng 			}
43138275SEric Cheng 		}
43148275SEric Cheng 
43158275SEric Cheng 		if (mac_write_link_stats(mcip) != 0 && !lstate->mi_last)
43168275SEric Cheng 			return (MH_WALK_TERMINATE);
43178275SEric Cheng 
43188275SEric Cheng 		if (lstate->mi_last)
43198275SEric Cheng 			mcip->mci_state_flags &= ~MCIS_DESC_LOGGED;
43208275SEric Cheng 
43218275SEric Cheng 		if (lstate->mi_fenable) {
43228275SEric Cheng 			if (mcip->mci_subflow_tab != NULL) {
43238275SEric Cheng 				(void) mac_flow_walk(mcip->mci_subflow_tab,
43248275SEric Cheng 				    mac_log_flowinfo, mip);
43258275SEric Cheng 			}
43268275SEric Cheng 		}
43278275SEric Cheng 	}
43288275SEric Cheng 	return (MH_WALK_CONTINUE);
43298275SEric Cheng }
43308275SEric Cheng 
43318275SEric Cheng /*
43328275SEric Cheng  * The timer thread that runs every mac_logging_interval seconds and logs
43338275SEric Cheng  * link and/or flow information.
43348275SEric Cheng  */
43358275SEric Cheng /* ARGSUSED */
43368275SEric Cheng void
43378275SEric Cheng mac_log_linkinfo(void *arg)
43388275SEric Cheng {
43398275SEric Cheng 	i_mac_log_state_t	lstate;
43408275SEric Cheng 
43418275SEric Cheng 	rw_enter(&i_mac_impl_lock, RW_READER);
43428275SEric Cheng 	if (!mac_flow_log_enable && !mac_link_log_enable) {
43438275SEric Cheng 		rw_exit(&i_mac_impl_lock);
43448275SEric Cheng 		return;
43458275SEric Cheng 	}
43468275SEric Cheng 	lstate.mi_fenable = mac_flow_log_enable;
43478275SEric Cheng 	lstate.mi_lenable = mac_link_log_enable;
43488275SEric Cheng 	lstate.mi_last = B_FALSE;
43498275SEric Cheng 	rw_exit(&i_mac_impl_lock);
43508275SEric Cheng 
43518275SEric Cheng 	mod_hash_walk(i_mac_impl_hash, i_mac_log_walker, &lstate);
43528275SEric Cheng 
43538275SEric Cheng 	rw_enter(&i_mac_impl_lock, RW_WRITER);
43548275SEric Cheng 	if (mac_flow_log_enable || mac_link_log_enable) {
43558275SEric Cheng 		mac_logging_timer = timeout(mac_log_linkinfo, NULL,
43568275SEric Cheng 		    SEC_TO_TICK(mac_logging_interval));
43578275SEric Cheng 	}
43588275SEric Cheng 	rw_exit(&i_mac_impl_lock);
43598275SEric Cheng }
43608275SEric Cheng 
43618275SEric Cheng /*
43628275SEric Cheng  * Start the logging timer.
43638275SEric Cheng  */
43648275SEric Cheng void
43658275SEric Cheng mac_start_logusage(mac_logtype_t type, uint_t interval)
43668275SEric Cheng {
43678275SEric Cheng 	rw_enter(&i_mac_impl_lock, RW_WRITER);
43688275SEric Cheng 	switch (type) {
43698275SEric Cheng 	case MAC_LOGTYPE_FLOW:
43708275SEric Cheng 		if (mac_flow_log_enable) {
43718275SEric Cheng 			rw_exit(&i_mac_impl_lock);
43728275SEric Cheng 			return;
43738275SEric Cheng 		}
43748275SEric Cheng 		mac_flow_log_enable = B_TRUE;
43758275SEric Cheng 		/* FALLTHRU */
43768275SEric Cheng 	case MAC_LOGTYPE_LINK:
43778275SEric Cheng 		if (mac_link_log_enable) {
43788275SEric Cheng 			rw_exit(&i_mac_impl_lock);
43798275SEric Cheng 			return;
43808275SEric Cheng 		}
43818275SEric Cheng 		mac_link_log_enable = B_TRUE;
43828275SEric Cheng 		break;
43838275SEric Cheng 	default:
43848275SEric Cheng 		ASSERT(0);
43858275SEric Cheng 	}
43868275SEric Cheng 	mac_logging_interval = interval;
43878275SEric Cheng 	rw_exit(&i_mac_impl_lock);
43888275SEric Cheng 	mac_log_linkinfo(NULL);
43898275SEric Cheng }
43908275SEric Cheng 
43918275SEric Cheng /*
43928275SEric Cheng  * Stop the logging timer if both Link and Flow logging are turned off.
43938275SEric Cheng  */
43948275SEric Cheng void
43958275SEric Cheng mac_stop_logusage(mac_logtype_t type)
43968275SEric Cheng {
43978275SEric Cheng 	i_mac_log_state_t	lstate;
43988275SEric Cheng 
43998275SEric Cheng 	rw_enter(&i_mac_impl_lock, RW_WRITER);
44008275SEric Cheng 	lstate.mi_fenable = mac_flow_log_enable;
44018275SEric Cheng 	lstate.mi_lenable = mac_link_log_enable;
44028275SEric Cheng 
44038275SEric Cheng 	/* Last walk */
44048275SEric Cheng 	lstate.mi_last = B_TRUE;
44058275SEric Cheng 
44068275SEric Cheng 	switch (type) {
44078275SEric Cheng 	case MAC_LOGTYPE_FLOW:
44088275SEric Cheng 		if (lstate.mi_fenable) {
44098275SEric Cheng 			ASSERT(mac_link_log_enable);
44108275SEric Cheng 			mac_flow_log_enable = B_FALSE;
44118275SEric Cheng 			mac_link_log_enable = B_FALSE;
44128275SEric Cheng 			break;
44138275SEric Cheng 		}
44148275SEric Cheng 		/* FALLTHRU */
44158275SEric Cheng 	case MAC_LOGTYPE_LINK:
44168275SEric Cheng 		if (!lstate.mi_lenable || mac_flow_log_enable) {
44178275SEric Cheng 			rw_exit(&i_mac_impl_lock);
44188275SEric Cheng 			return;
44198275SEric Cheng 		}
44208275SEric Cheng 		mac_link_log_enable = B_FALSE;
44218275SEric Cheng 		break;
44228275SEric Cheng 	default:
44238275SEric Cheng 		ASSERT(0);
44248275SEric Cheng 	}
44258275SEric Cheng 	rw_exit(&i_mac_impl_lock);
44268275SEric Cheng 	(void) untimeout(mac_logging_timer);
44278275SEric Cheng 	mac_logging_timer = 0;
44288275SEric Cheng 
44298275SEric Cheng 	/* Last walk */
44308275SEric Cheng 	mod_hash_walk(i_mac_impl_hash, i_mac_log_walker, &lstate);
44318275SEric Cheng }
44328275SEric Cheng 
44338275SEric Cheng /*
44348275SEric Cheng  * Walk the rx and tx SRS/SRs for a flow and update the priority value.
44358275SEric Cheng  */
44368275SEric Cheng void
44378275SEric Cheng mac_flow_update_priority(mac_client_impl_t *mcip, flow_entry_t *flent)
44388275SEric Cheng {
44398275SEric Cheng 	pri_t			pri;
44408275SEric Cheng 	int			count;
44418275SEric Cheng 	mac_soft_ring_set_t	*mac_srs;
44428275SEric Cheng 
44438275SEric Cheng 	if (flent->fe_rx_srs_cnt <= 0)
44448275SEric Cheng 		return;
44458275SEric Cheng 
44468275SEric Cheng 	if (((mac_soft_ring_set_t *)flent->fe_rx_srs[0])->srs_type ==
44478275SEric Cheng 	    SRST_FLOW) {
44488275SEric Cheng 		pri = FLOW_PRIORITY(mcip->mci_min_pri,
44498275SEric Cheng 		    mcip->mci_max_pri,
44508275SEric Cheng 		    flent->fe_resource_props.mrp_priority);
44518275SEric Cheng 	} else {
44528275SEric Cheng 		pri = mcip->mci_max_pri;
44538275SEric Cheng 	}
44548275SEric Cheng 
44558275SEric Cheng 	for (count = 0; count < flent->fe_rx_srs_cnt; count++) {
44568275SEric Cheng 		mac_srs = flent->fe_rx_srs[count];
44578275SEric Cheng 		mac_update_srs_priority(mac_srs, pri);
44588275SEric Cheng 	}
44598275SEric Cheng 	/*
44608275SEric Cheng 	 * If we have a Tx SRS, we need to modify all the threads associated
44618275SEric Cheng 	 * with it.
44628275SEric Cheng 	 */
44638275SEric Cheng 	if (flent->fe_tx_srs != NULL)
44648275SEric Cheng 		mac_update_srs_priority(flent->fe_tx_srs, pri);
44658275SEric Cheng }
44668275SEric Cheng 
44678275SEric Cheng /*
44688275SEric Cheng  * RX and TX rings are reserved according to different semantics depending
44698275SEric Cheng  * on the requests from the MAC clients and type of rings:
44708275SEric Cheng  *
44718275SEric Cheng  * On the Tx side, by default we reserve individual rings, independently from
44728275SEric Cheng  * the groups.
44738275SEric Cheng  *
44748275SEric Cheng  * On the Rx side, the reservation is at the granularity of the group
44758275SEric Cheng  * of rings, and used for v12n level 1 only. It has a special case for the
44768275SEric Cheng  * primary client.
44778275SEric Cheng  *
44788275SEric Cheng  * If a share is allocated to a MAC client, we allocate a TX group and an
44798275SEric Cheng  * RX group to the client, and assign TX rings and RX rings to these
44808275SEric Cheng  * groups according to information gathered from the driver through
44818275SEric Cheng  * the share capability.
44828275SEric Cheng  *
44838275SEric Cheng  * The foreseable evolution of Rx rings will handle v12n level 2 and higher
44848275SEric Cheng  * to allocate individual rings out of a group and program the hw classifier
44858275SEric Cheng  * based on IP address or higher level criteria.
44868275SEric Cheng  */
44878275SEric Cheng 
44888275SEric Cheng /*
44898275SEric Cheng  * mac_reserve_tx_ring()
44908275SEric Cheng  * Reserve a unused ring by marking it with MR_INUSE state.
44918275SEric Cheng  * As reserved, the ring is ready to function.
44928275SEric Cheng  *
44938275SEric Cheng  * Notes for Hybrid I/O:
44948275SEric Cheng  *
44958275SEric Cheng  * If a specific ring is needed, it is specified through the desired_ring
44968275SEric Cheng  * argument. Otherwise that argument is set to NULL.
44978275SEric Cheng  * If the desired ring was previous allocated to another client, this
44988275SEric Cheng  * function swaps it with a new ring from the group of unassigned rings.
44998275SEric Cheng  */
45008275SEric Cheng mac_ring_t *
45018275SEric Cheng mac_reserve_tx_ring(mac_impl_t *mip, mac_ring_t *desired_ring)
45028275SEric Cheng {
45038275SEric Cheng 	mac_group_t *group;
45048275SEric Cheng 	mac_ring_t *ring;
45058275SEric Cheng 
45068275SEric Cheng 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
45078275SEric Cheng 
45088275SEric Cheng 	if (mip->mi_tx_groups == NULL)
45098275SEric Cheng 		return (NULL);
45108275SEric Cheng 
45118275SEric Cheng 	/*
45128275SEric Cheng 	 * Find an available ring and start it before changing its status.
45138275SEric Cheng 	 * The unassigned rings are at the end of the mi_tx_groups
45148275SEric Cheng 	 * array.
45158275SEric Cheng 	 */
45168275SEric Cheng 	group = mip->mi_tx_groups + mip->mi_tx_group_count;
45178275SEric Cheng 
45188275SEric Cheng 	for (ring = group->mrg_rings; ring != NULL;
45198275SEric Cheng 	    ring = ring->mr_next) {
45208275SEric Cheng 		if (desired_ring == NULL) {
45218275SEric Cheng 			if (ring->mr_state == MR_FREE)
45228275SEric Cheng 				/* wanted any free ring and found one */
45238275SEric Cheng 				break;
45248275SEric Cheng 		} else {
45258275SEric Cheng 			mac_ring_t *sring;
45268275SEric Cheng 			mac_client_impl_t *client;
45278275SEric Cheng 			mac_soft_ring_set_t *srs;
45288275SEric Cheng 
45298275SEric Cheng 			if (ring != desired_ring)
45308275SEric Cheng 				/* wants a desired ring but this one ain't it */
45318275SEric Cheng 				continue;
45328275SEric Cheng 
45338275SEric Cheng 			if (ring->mr_state == MR_FREE)
45348275SEric Cheng 				break;
45358275SEric Cheng 
45368275SEric Cheng 			/*
45378275SEric Cheng 			 * Found the desired ring but it's already in use.
45388275SEric Cheng 			 * Swap it with a new ring.
45398275SEric Cheng 			 */
45408275SEric Cheng 
45418275SEric Cheng 			/* find the client which owns that ring */
45428275SEric Cheng 			for (client = mip->mi_clients_list; client != NULL;
45438275SEric Cheng 			    client = client->mci_client_next) {
45448275SEric Cheng 				srs = MCIP_TX_SRS(client);
45458275SEric Cheng 				if (srs != NULL && mac_tx_srs_ring_present(srs,
45468275SEric Cheng 				    desired_ring)) {
45478275SEric Cheng 					/* found our ring */
45488275SEric Cheng 					break;
45498275SEric Cheng 				}
45508275SEric Cheng 			}
4551*8400SNicolas.Droux@Sun.COM 			if (client == NULL) {
4552*8400SNicolas.Droux@Sun.COM 				/*
4553*8400SNicolas.Droux@Sun.COM 				 * The TX ring is in use, but it's not
4554*8400SNicolas.Droux@Sun.COM 				 * associated with any clients, so it
4555*8400SNicolas.Droux@Sun.COM 				 * has to be the default ring. In that
4556*8400SNicolas.Droux@Sun.COM 				 * case we can simply assign a new ring
4557*8400SNicolas.Droux@Sun.COM 				 * as the default ring, and we're done.
4558*8400SNicolas.Droux@Sun.COM 				 */
4559*8400SNicolas.Droux@Sun.COM 				ASSERT(mip->mi_default_tx_ring ==
4560*8400SNicolas.Droux@Sun.COM 				    (mac_ring_handle_t)desired_ring);
4561*8400SNicolas.Droux@Sun.COM 
4562*8400SNicolas.Droux@Sun.COM 				/*
4563*8400SNicolas.Droux@Sun.COM 				 * Quiesce all clients on top of
4564*8400SNicolas.Droux@Sun.COM 				 * the NIC to make sure there are no
4565*8400SNicolas.Droux@Sun.COM 				 * pending threads still relying on
4566*8400SNicolas.Droux@Sun.COM 				 * that default ring, for example
4567*8400SNicolas.Droux@Sun.COM 				 * the multicast path.
4568*8400SNicolas.Droux@Sun.COM 				 */
4569*8400SNicolas.Droux@Sun.COM 				for (client = mip->mi_clients_list;
4570*8400SNicolas.Droux@Sun.COM 				    client != NULL;
4571*8400SNicolas.Droux@Sun.COM 				    client = client->mci_client_next) {
4572*8400SNicolas.Droux@Sun.COM 					mac_tx_client_quiesce(client,
4573*8400SNicolas.Droux@Sun.COM 					    SRS_QUIESCE);
4574*8400SNicolas.Droux@Sun.COM 				}
4575*8400SNicolas.Droux@Sun.COM 
4576*8400SNicolas.Droux@Sun.COM 				mip->mi_default_tx_ring = (mac_ring_handle_t)
4577*8400SNicolas.Droux@Sun.COM 				    mac_reserve_tx_ring(mip, NULL);
4578*8400SNicolas.Droux@Sun.COM 
4579*8400SNicolas.Droux@Sun.COM 				/* resume the clients */
4580*8400SNicolas.Droux@Sun.COM 				for (client = mip->mi_clients_list;
4581*8400SNicolas.Droux@Sun.COM 				    client != NULL;
4582*8400SNicolas.Droux@Sun.COM 				    client = client->mci_client_next)
4583*8400SNicolas.Droux@Sun.COM 					mac_tx_client_restart(client);
4584*8400SNicolas.Droux@Sun.COM 
4585*8400SNicolas.Droux@Sun.COM 				break;
4586*8400SNicolas.Droux@Sun.COM 			}
45878275SEric Cheng 
45888275SEric Cheng 			/*
45898275SEric Cheng 			 * Note that we cannot simply invoke the group
45908275SEric Cheng 			 * add/rem routines since the client doesn't have a
45918275SEric Cheng 			 * TX group. So we need to instead add/remove
45928275SEric Cheng 			 * the rings from the SRS.
45938275SEric Cheng 			 */
45948275SEric Cheng 			ASSERT(client->mci_share == NULL);
45958275SEric Cheng 
45968275SEric Cheng 			/* first quiece the client */
45978275SEric Cheng 			mac_tx_client_quiesce(client, SRS_QUIESCE);
45988275SEric Cheng 
45998275SEric Cheng 			/* give a new ring to the client... */
46008275SEric Cheng 			sring = mac_reserve_tx_ring(mip, NULL);
46018275SEric Cheng 			if (sring != NULL) {
46028275SEric Cheng 				/*
46038275SEric Cheng 				 * There are no other available ring
46048275SEric Cheng 				 * on that MAC instance. The client
46058275SEric Cheng 				 * will fallback to the shared TX
46068275SEric Cheng 				 * ring.
46078275SEric Cheng 				 */
46088275SEric Cheng 				mac_tx_srs_add_ring(srs, sring);
46098275SEric Cheng 			}
46108275SEric Cheng 
46118275SEric Cheng 			/* ... in exchange for our desired ring */
46128275SEric Cheng 			mac_tx_srs_del_ring(srs, desired_ring);
46138275SEric Cheng 
46148275SEric Cheng 			/* restart the client */
46158275SEric Cheng 			mac_tx_client_restart(client);
46168275SEric Cheng 
4617*8400SNicolas.Droux@Sun.COM 			if (mip->mi_default_tx_ring ==
4618*8400SNicolas.Droux@Sun.COM 			    (mac_ring_handle_t)desired_ring) {
4619*8400SNicolas.Droux@Sun.COM 				/*
4620*8400SNicolas.Droux@Sun.COM 				 * The desired ring is the default ring,
4621*8400SNicolas.Droux@Sun.COM 				 * and there are one or more clients
4622*8400SNicolas.Droux@Sun.COM 				 * using that default ring directly.
4623*8400SNicolas.Droux@Sun.COM 				 */
4624*8400SNicolas.Droux@Sun.COM 				mip->mi_default_tx_ring =
4625*8400SNicolas.Droux@Sun.COM 				    (mac_ring_handle_t)sring;
4626*8400SNicolas.Droux@Sun.COM 				/*
4627*8400SNicolas.Droux@Sun.COM 				 * Find clients using default ring and
4628*8400SNicolas.Droux@Sun.COM 				 * swap it with the new default ring.
4629*8400SNicolas.Droux@Sun.COM 				 */
4630*8400SNicolas.Droux@Sun.COM 				for (client = mip->mi_clients_list;
4631*8400SNicolas.Droux@Sun.COM 				    client != NULL;
4632*8400SNicolas.Droux@Sun.COM 				    client = client->mci_client_next) {
4633*8400SNicolas.Droux@Sun.COM 					srs = MCIP_TX_SRS(client);
4634*8400SNicolas.Droux@Sun.COM 					if (srs != NULL &&
4635*8400SNicolas.Droux@Sun.COM 					    mac_tx_srs_ring_present(srs,
4636*8400SNicolas.Droux@Sun.COM 					    desired_ring)) {
4637*8400SNicolas.Droux@Sun.COM 						/* first quiece the client */
4638*8400SNicolas.Droux@Sun.COM 						mac_tx_client_quiesce(client,
4639*8400SNicolas.Droux@Sun.COM 						    SRS_QUIESCE);
4640*8400SNicolas.Droux@Sun.COM 
4641*8400SNicolas.Droux@Sun.COM 						/*
4642*8400SNicolas.Droux@Sun.COM 						 * Give it the new default
4643*8400SNicolas.Droux@Sun.COM 						 * ring, and remove the old
4644*8400SNicolas.Droux@Sun.COM 						 * one.
4645*8400SNicolas.Droux@Sun.COM 						 */
4646*8400SNicolas.Droux@Sun.COM 						if (sring != NULL) {
4647*8400SNicolas.Droux@Sun.COM 							mac_tx_srs_add_ring(srs,
4648*8400SNicolas.Droux@Sun.COM 							    sring);
4649*8400SNicolas.Droux@Sun.COM 						}
4650*8400SNicolas.Droux@Sun.COM 						mac_tx_srs_del_ring(srs,
4651*8400SNicolas.Droux@Sun.COM 						    desired_ring);
4652*8400SNicolas.Droux@Sun.COM 
4653*8400SNicolas.Droux@Sun.COM 						/* restart the client */
4654*8400SNicolas.Droux@Sun.COM 						mac_tx_client_restart(client);
4655*8400SNicolas.Droux@Sun.COM 					}
4656*8400SNicolas.Droux@Sun.COM 				}
4657*8400SNicolas.Droux@Sun.COM 			}
46588275SEric Cheng 			break;
46598275SEric Cheng 		}
46608275SEric Cheng 	}
46618275SEric Cheng 
46628275SEric Cheng 	if (ring != NULL) {
46638275SEric Cheng 		if (mac_start_ring(ring) != 0)
46648275SEric Cheng 			return (NULL);
46658275SEric Cheng 		ring->mr_state = MR_INUSE;
46668275SEric Cheng 	}
46678275SEric Cheng 
46688275SEric Cheng 	return (ring);
46698275SEric Cheng }
46708275SEric Cheng 
46718275SEric Cheng /*
46728275SEric Cheng  * Minimum number of rings to leave in the default TX group when allocating
46738275SEric Cheng  * rings to new clients.
46748275SEric Cheng  */
46758275SEric Cheng static uint_t mac_min_rx_default_rings = 1;
46768275SEric Cheng 
46778275SEric Cheng /*
46788275SEric Cheng  * Populate a zero-ring group with rings. If the share is non-NULL,
46798275SEric Cheng  * the rings are chosen according to that share.
46808275SEric Cheng  * Invoked after allocating a new RX or TX group through
46818275SEric Cheng  * mac_reserve_rx_group() or mac_reserve_tx_group(), respectively.
46828275SEric Cheng  * Returns zero on success, an errno otherwise.
46838275SEric Cheng  */
46848275SEric Cheng int
46858275SEric Cheng i_mac_group_allocate_rings(mac_impl_t *mip, mac_ring_type_t ring_type,
46868275SEric Cheng     mac_group_t *src_group, mac_group_t *new_group, mac_share_handle_t share)
46878275SEric Cheng {
46888275SEric Cheng 	mac_ring_t **rings, *tmp_ring[1], *ring;
46898275SEric Cheng 	uint_t nrings;
46908275SEric Cheng 	int rv, i, j;
46918275SEric Cheng 
46928275SEric Cheng 	ASSERT(mip->mi_rx_group_type == MAC_GROUP_TYPE_DYNAMIC &&
46938275SEric Cheng 	    mip->mi_tx_group_type == MAC_GROUP_TYPE_DYNAMIC);
46948275SEric Cheng 	ASSERT(new_group->mrg_cur_count == 0);
46958275SEric Cheng 
46968275SEric Cheng 	/*
46978275SEric Cheng 	 * First find the rings to allocate to the group.
46988275SEric Cheng 	 */
46998275SEric Cheng 	if (share != NULL) {
47008275SEric Cheng 		/* get rings through ms_squery() */
47018275SEric Cheng 		mip->mi_share_capab.ms_squery(share, ring_type, NULL, &nrings);
47028275SEric Cheng 		ASSERT(nrings != 0);
47038275SEric Cheng 		rings = kmem_alloc(nrings * sizeof (mac_ring_handle_t),
47048275SEric Cheng 		    KM_SLEEP);
47058275SEric Cheng 		mip->mi_share_capab.ms_squery(share, ring_type,
47068275SEric Cheng 		    (mac_ring_handle_t *)rings, &nrings);
47078275SEric Cheng 	} else {
47088275SEric Cheng 		/* this function is called for TX only with a share */
47098275SEric Cheng 		ASSERT(ring_type == MAC_RING_TYPE_RX);
47108275SEric Cheng 		/*
47118275SEric Cheng 		 * Pick one ring from default group.
47128275SEric Cheng 		 *
47138275SEric Cheng 		 * for now pick the second ring which requires the first ring
47148275SEric Cheng 		 * at index 0 to stay in the default group, since it is the
47158275SEric Cheng 		 * ring which carries the multicast traffic.
47168275SEric Cheng 		 * We need a better way for a driver to indicate this,
47178275SEric Cheng 		 * for example a per-ring flag.
47188275SEric Cheng 		 */
47198275SEric Cheng 		for (ring = src_group->mrg_rings; ring != NULL;
47208275SEric Cheng 		    ring = ring->mr_next) {
47218275SEric Cheng 			if (ring->mr_index != 0)
47228275SEric Cheng 				break;
47238275SEric Cheng 		}
47248275SEric Cheng 		ASSERT(ring != NULL);
47258275SEric Cheng 		nrings = 1;
47268275SEric Cheng 		tmp_ring[0] = ring;
47278275SEric Cheng 		rings = tmp_ring;
47288275SEric Cheng 	}
47298275SEric Cheng 
47308275SEric Cheng 	switch (ring_type) {
47318275SEric Cheng 	case MAC_RING_TYPE_RX:
47328275SEric Cheng 		if (src_group->mrg_cur_count - nrings <
47338275SEric Cheng 		    mac_min_rx_default_rings) {
47348275SEric Cheng 			/* we ran out of rings */
47358275SEric Cheng 			return (ENOSPC);
47368275SEric Cheng 		}
47378275SEric Cheng 
47388275SEric Cheng 		/* move receive rings to new group */
47398275SEric Cheng 		for (i = 0; i < nrings; i++) {
47408275SEric Cheng 			rv = mac_group_mov_ring(mip, new_group, rings[i]);
47418275SEric Cheng 			if (rv != 0) {
47428275SEric Cheng 				/* move rings back on failure */
47438275SEric Cheng 				for (j = 0; j < i; j++) {
47448275SEric Cheng 					(void) mac_group_mov_ring(mip,
47458275SEric Cheng 					    src_group, rings[j]);
47468275SEric Cheng 				}
47478275SEric Cheng 				return (rv);
47488275SEric Cheng 			}
47498275SEric Cheng 		}
47508275SEric Cheng 		break;
47518275SEric Cheng 
47528275SEric Cheng 	case MAC_RING_TYPE_TX: {
47538275SEric Cheng 		mac_ring_t *tmp_ring;
47548275SEric Cheng 
47558275SEric Cheng 		/* move the TX rings to the new group */
47568275SEric Cheng 		ASSERT(src_group == NULL);
47578275SEric Cheng 		for (i = 0; i < nrings; i++) {
47588275SEric Cheng 			/* get the desired ring */
47598275SEric Cheng 			tmp_ring = mac_reserve_tx_ring(mip, rings[i]);
47608275SEric Cheng 			ASSERT(tmp_ring == rings[i]);
47618275SEric Cheng 			rv = mac_group_mov_ring(mip, new_group, rings[i]);
47628275SEric Cheng 			if (rv != 0) {
47638275SEric Cheng 				/* cleanup on failure */
47648275SEric Cheng 				for (j = 0; j < i; j++) {
47658275SEric Cheng 					(void) mac_group_mov_ring(mip,
47668275SEric Cheng 					    mip->mi_tx_groups +
47678275SEric Cheng 					    mip->mi_tx_group_count, rings[j]);
47688275SEric Cheng 				}
47698275SEric Cheng 			}
47708275SEric Cheng 		}
47718275SEric Cheng 		break;
47728275SEric Cheng 	}
47738275SEric Cheng 	}
47748275SEric Cheng 
47758275SEric Cheng 	if (share != NULL) {
47768275SEric Cheng 		/* add group to share */
47778275SEric Cheng 		mip->mi_share_capab.ms_sadd(share, new_group->mrg_driver);
47788275SEric Cheng 		/* free temporary array of rings */
47798275SEric Cheng 		kmem_free(rings, nrings * sizeof (mac_ring_handle_t));
47808275SEric Cheng 	}
47818275SEric Cheng 
47828275SEric Cheng 	return (0);
47838275SEric Cheng }
47848275SEric Cheng 
47858275SEric Cheng void
47868275SEric Cheng mac_rx_group_add_client(mac_group_t *grp, mac_client_impl_t *mcip)
47878275SEric Cheng {
47888275SEric Cheng 	mac_grp_client_t *mgcp;
47898275SEric Cheng 
47908275SEric Cheng 	for (mgcp = grp->mrg_clients; mgcp != NULL; mgcp = mgcp->mgc_next) {
47918275SEric Cheng 		if (mgcp->mgc_client == mcip)
47928275SEric Cheng 			break;
47938275SEric Cheng 	}
47948275SEric Cheng 
47958275SEric Cheng 	VERIFY(mgcp == NULL);
47968275SEric Cheng 
47978275SEric Cheng 	mgcp = kmem_zalloc(sizeof (mac_grp_client_t), KM_SLEEP);
47988275SEric Cheng 	mgcp->mgc_client = mcip;
47998275SEric Cheng 	mgcp->mgc_next = grp->mrg_clients;
48008275SEric Cheng 	grp->mrg_clients = mgcp;
48018275SEric Cheng 
48028275SEric Cheng }
48038275SEric Cheng 
48048275SEric Cheng void
48058275SEric Cheng mac_rx_group_remove_client(mac_group_t *grp, mac_client_impl_t *mcip)
48068275SEric Cheng {
48078275SEric Cheng 	mac_grp_client_t *mgcp, **pprev;
48088275SEric Cheng 
48098275SEric Cheng 	for (pprev = &grp->mrg_clients, mgcp = *pprev; mgcp != NULL;
48108275SEric Cheng 	    pprev = &mgcp->mgc_next, mgcp = *pprev) {
48118275SEric Cheng 		if (mgcp->mgc_client == mcip)
48128275SEric Cheng 			break;
48138275SEric Cheng 	}
48148275SEric Cheng 
48158275SEric Cheng 	ASSERT(mgcp != NULL);
48168275SEric Cheng 
48178275SEric Cheng 	*pprev = mgcp->mgc_next;
48188275SEric Cheng 	kmem_free(mgcp, sizeof (mac_grp_client_t));
48198275SEric Cheng }
48208275SEric Cheng 
48218275SEric Cheng /*
48228275SEric Cheng  * mac_reserve_rx_group()
48238275SEric Cheng  *
48248275SEric Cheng  * Finds an available group and exclusively reserves it for a client.
48258275SEric Cheng  * The group is chosen to suit the flow's resource controls (bandwidth and
48268275SEric Cheng  * fanout requirements) and the address type.
48278275SEric Cheng  * If the requestor is the pimary MAC then return the group with the
48288275SEric Cheng  * largest number of rings, otherwise the default ring when available.
48298275SEric Cheng  */
48308275SEric Cheng mac_group_t *
48318275SEric Cheng mac_reserve_rx_group(mac_client_impl_t *mcip, uint8_t *mac_addr,
48328275SEric Cheng     mac_rx_group_reserve_type_t rtype)
48338275SEric Cheng {
48348275SEric Cheng 	mac_share_handle_t	share = mcip->mci_share;
48358275SEric Cheng 	mac_impl_t		*mip = mcip->mci_mip;
48368275SEric Cheng 	mac_group_t		*grp = NULL;
48378275SEric Cheng 	int			i, start, loopcount;
48388275SEric Cheng 	int			err;
48398275SEric Cheng 	mac_address_t		*map;
48408275SEric Cheng 
48418275SEric Cheng 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mip));
48428275SEric Cheng 
48438275SEric Cheng 	/* Check if a group already has this mac address (case of VLANs) */
48448275SEric Cheng 	if ((map = mac_find_macaddr(mip, mac_addr)) != NULL)
48458275SEric Cheng 		return (map->ma_group);
48468275SEric Cheng 
48478275SEric Cheng 	if (mip->mi_rx_groups == NULL || mip->mi_rx_group_count == 0 ||
48488275SEric Cheng 	    rtype == MAC_RX_NO_RESERVE)
48498275SEric Cheng 		return (NULL);
48508275SEric Cheng 
48518275SEric Cheng 	/*
48528275SEric Cheng 	 * Try to exclusively reserve a RX group.
48538275SEric Cheng 	 *
48548275SEric Cheng 	 * For flows requires SW_RING it always goes to the default group
48558275SEric Cheng 	 * (Until we can explicitely call out default groups (CR 6695600),
48568275SEric Cheng 	 * we assume that the default group is always at position zero);
48578275SEric Cheng 	 *
48588275SEric Cheng 	 * For flows requires HW_DEFAULT_RING (unicast flow of the primary
48598275SEric Cheng 	 * client), try to reserve the default RX group only.
48608275SEric Cheng 	 *
48618275SEric Cheng 	 * For flows requires HW_RING (unicast flow of other clients), try
48628275SEric Cheng 	 * to reserve non-default RX group then the default group.
48638275SEric Cheng 	 */
48648275SEric Cheng 	switch (rtype) {
48658275SEric Cheng 	case MAC_RX_RESERVE_DEFAULT:
48668275SEric Cheng 		start = 0;
48678275SEric Cheng 		loopcount = 1;
48688275SEric Cheng 		break;
48698275SEric Cheng 	case MAC_RX_RESERVE_NONDEFAULT:
48708275SEric Cheng 		start = 1;
48718275SEric Cheng 		loopcount = mip->mi_rx_group_count;
48728275SEric Cheng 	}
48738275SEric Cheng 
48748275SEric Cheng 	for (i = start; i < start + loopcount; i++) {
48758275SEric Cheng 		grp = &mip->mi_rx_groups[i % mip->mi_rx_group_count];
48768275SEric Cheng 
48778275SEric Cheng 		DTRACE_PROBE3(rx__group__trying, char *, mip->mi_name,
48788275SEric Cheng 		    int, grp->mrg_index, mac_group_state_t, grp->mrg_state);
48798275SEric Cheng 
48808275SEric Cheng 		/*
48818275SEric Cheng 		 * Check to see whether this mac client is the only client
48828275SEric Cheng 		 * on this RX group. If not, we cannot exclusively reserve
48838275SEric Cheng 		 * this RX group.
48848275SEric Cheng 		 */
48858275SEric Cheng 		if (!MAC_RX_GROUP_NO_CLIENT(grp) &&
48868275SEric Cheng 		    (MAC_RX_GROUP_ONLY_CLIENT(grp) != mcip)) {
48878275SEric Cheng 			continue;
48888275SEric Cheng 		}
48898275SEric Cheng 
48908275SEric Cheng 		/*
48918275SEric Cheng 		 * This group could already be SHARED by other multicast
48928275SEric Cheng 		 * flows on this client. In that case, the group would
48938275SEric Cheng 		 * be shared and has already been started.
48948275SEric Cheng 		 */
48958275SEric Cheng 		ASSERT(grp->mrg_state != MAC_GROUP_STATE_UNINIT);
48968275SEric Cheng 
48978275SEric Cheng 		if ((grp->mrg_state == MAC_GROUP_STATE_REGISTERED) &&
48988275SEric Cheng 		    (mac_start_group(grp) != 0)) {
48998275SEric Cheng 			continue;
49008275SEric Cheng 		}
49018275SEric Cheng 
49028275SEric Cheng 		if ((i % mip->mi_rx_group_count) == 0 ||
49038275SEric Cheng 		    mip->mi_rx_group_type != MAC_GROUP_TYPE_DYNAMIC) {
49048275SEric Cheng 			break;
49058275SEric Cheng 		}
49068275SEric Cheng 
49078275SEric Cheng 		ASSERT(grp->mrg_cur_count == 0);
49088275SEric Cheng 
49098275SEric Cheng 		/*
49108275SEric Cheng 		 * Populate the group. Rings should be taken
49118275SEric Cheng 		 * from the default group at position 0 for now.
49128275SEric Cheng 		 */
49138275SEric Cheng 
49148275SEric Cheng 		err = i_mac_group_allocate_rings(mip, MAC_RING_TYPE_RX,
49158275SEric Cheng 		    &mip->mi_rx_groups[0], grp, share);
49168275SEric Cheng 		if (err == 0)
49178275SEric Cheng 			break;
49188275SEric Cheng 
49198275SEric Cheng 		DTRACE_PROBE3(rx__group__reserve__alloc__rings, char *,
49208275SEric Cheng 		    mip->mi_name, int, grp->mrg_index, int, err);
49218275SEric Cheng 
49228275SEric Cheng 		/*
49238275SEric Cheng 		 * It's a dynamic group but the grouping operation failed.
49248275SEric Cheng 		 */
49258275SEric Cheng 		mac_stop_group(grp);
49268275SEric Cheng 	}
49278275SEric Cheng 
49288275SEric Cheng 	if (i == start + loopcount)
49298275SEric Cheng 		return (NULL);
49308275SEric Cheng 
49318275SEric Cheng 	ASSERT(grp != NULL);
49328275SEric Cheng 
49338275SEric Cheng 	DTRACE_PROBE2(rx__group__reserved,
49348275SEric Cheng 	    char *, mip->mi_name, int, grp->mrg_index);
49358275SEric Cheng 	return (grp);
49368275SEric Cheng }
49378275SEric Cheng 
49388275SEric Cheng /*
49398275SEric Cheng  * mac_rx_release_group()
49408275SEric Cheng  *
49418275SEric Cheng  * This is called when there are no clients left for the group.
49428275SEric Cheng  * The group is stopped and marked MAC_GROUP_STATE_REGISTERED,
49438275SEric Cheng  * and if it is a non default group, the shares are removed and
49448275SEric Cheng  * all rings are assigned back to default group.
49458275SEric Cheng  */
49468275SEric Cheng void
49478275SEric Cheng mac_release_rx_group(mac_client_impl_t *mcip, mac_group_t *group)
49488275SEric Cheng {
49498275SEric Cheng 	mac_impl_t	*mip = mcip->mci_mip;
49508275SEric Cheng 	mac_ring_t	*ring;
49518275SEric Cheng 
49528275SEric Cheng 	ASSERT(group != &mip->mi_rx_groups[0]);
49538275SEric Cheng 
49548275SEric Cheng 	/*
49558275SEric Cheng 	 * This is the case where there are no clients left. Any
49568275SEric Cheng 	 * SRS etc on this group have also be quiesced.
49578275SEric Cheng 	 */
49588275SEric Cheng 	for (ring = group->mrg_rings; ring != NULL; ring = ring->mr_next) {
49598275SEric Cheng 		if (ring->mr_classify_type == MAC_HW_CLASSIFIER) {
49608275SEric Cheng 			ASSERT(group->mrg_state == MAC_GROUP_STATE_RESERVED);
49618275SEric Cheng 			/*
49628275SEric Cheng 			 * Remove the SRS associated with the HW ring.
49638275SEric Cheng 			 * As a result, polling will be disabled.
49648275SEric Cheng 			 */
49658275SEric Cheng 			ring->mr_srs = NULL;
49668275SEric Cheng 		}
49678275SEric Cheng 		ASSERT(ring->mr_state == MR_INUSE);
49688275SEric Cheng 		mac_stop_ring(ring);
49698275SEric Cheng 		ring->mr_state = MR_FREE;
49708275SEric Cheng 		ring->mr_flag = 0;
49718275SEric Cheng 	}
49728275SEric Cheng 
49738275SEric Cheng 	/* remove group from share */
49748275SEric Cheng 	if (mcip->mci_share != NULL) {
49758275SEric Cheng 		mip->mi_share_capab.ms_sremove(mcip->mci_share,
49768275SEric Cheng 		    group->mrg_driver);
49778275SEric Cheng 	}
49788275SEric Cheng 
49798275SEric Cheng 	if (mip->mi_rx_group_type == MAC_GROUP_TYPE_DYNAMIC) {
49808275SEric Cheng 		mac_ring_t *ring;
49818275SEric Cheng 
49828275SEric Cheng 		/*
49838275SEric Cheng 		 * Rings were dynamically allocated to group.
49848275SEric Cheng 		 * Move rings back to default group.
49858275SEric Cheng 		 */
49868275SEric Cheng 		while ((ring = group->mrg_rings) != NULL) {
49878275SEric Cheng 			(void) mac_group_mov_ring(mip,
49888275SEric Cheng 			    &mip->mi_rx_groups[0], ring);
49898275SEric Cheng 		}
49908275SEric Cheng 	}
49918275SEric Cheng 	mac_stop_group(group);
49928275SEric Cheng 	/*
49938275SEric Cheng 	 * Possible improvement: See if we can assign the group just released
49948275SEric Cheng 	 * to a another client of the mip
49958275SEric Cheng 	 */
49968275SEric Cheng }
49978275SEric Cheng 
49988275SEric Cheng /*
49998275SEric Cheng  * Reserves a TX group for the specified share. Invoked by mac_tx_srs_setup()
50008275SEric Cheng  * when a share was allocated to the client.
50018275SEric Cheng  */
50028275SEric Cheng mac_group_t *
50038275SEric Cheng mac_reserve_tx_group(mac_impl_t *mip, mac_share_handle_t share)
50048275SEric Cheng {
50058275SEric Cheng 	mac_group_t *grp;
50068275SEric Cheng 	int rv, i;
50078275SEric Cheng 
50088275SEric Cheng 	/*
50098275SEric Cheng 	 * TX groups are currently allocated only to MAC clients
50108275SEric Cheng 	 * which are associated with a share. Since we have a fixed
50118275SEric Cheng 	 * number of share and groups, and we already successfully
50128275SEric Cheng 	 * allocated a share, find an available TX group.
50138275SEric Cheng 	 */
50148275SEric Cheng 	ASSERT(share != NULL);
50158275SEric Cheng 	ASSERT(mip->mi_tx_group_free > 0);
50168275SEric Cheng 
50178275SEric Cheng 	for (i = 0; i <  mip->mi_tx_group_count; i++) {
50188275SEric Cheng 		grp = &mip->mi_tx_groups[i];
50198275SEric Cheng 
50208275SEric Cheng 		if ((grp->mrg_state == MAC_GROUP_STATE_RESERVED) ||
50218275SEric Cheng 		    (grp->mrg_state == MAC_GROUP_STATE_UNINIT))
50228275SEric Cheng 			continue;
50238275SEric Cheng 
50248275SEric Cheng 		rv = mac_start_group(grp);
50258275SEric Cheng 		ASSERT(rv == 0);
50268275SEric Cheng 
50278275SEric Cheng 		grp->mrg_state = MAC_GROUP_STATE_RESERVED;
50288275SEric Cheng 		break;
50298275SEric Cheng 	}
50308275SEric Cheng 
50318275SEric Cheng 	ASSERT(grp != NULL);
50328275SEric Cheng 
50338275SEric Cheng 	/*
50348275SEric Cheng 	 * Populate the group. Rings should be taken from the group
50358275SEric Cheng 	 * of unassigned rings, which is past the array of TX
50368275SEric Cheng 	 * groups adversized by the driver.
50378275SEric Cheng 	 */
50388275SEric Cheng 	rv = i_mac_group_allocate_rings(mip, MAC_RING_TYPE_TX, NULL,
50398275SEric Cheng 	    grp, share);
50408275SEric Cheng 	if (rv != 0) {
50418275SEric Cheng 		DTRACE_PROBE3(tx__group__reserve__alloc__rings,
50428275SEric Cheng 		    char *, mip->mi_name, int, grp->mrg_index, int, rv);
50438275SEric Cheng 
50448275SEric Cheng 		mac_stop_group(grp);
50458275SEric Cheng 		grp->mrg_state = MAC_GROUP_STATE_UNINIT;
50468275SEric Cheng 
50478275SEric Cheng 		return (NULL);
50488275SEric Cheng 	}
50498275SEric Cheng 
50508275SEric Cheng 	mip->mi_tx_group_free--;
50518275SEric Cheng 
50528275SEric Cheng 	return (grp);
50538275SEric Cheng }
50548275SEric Cheng 
50558275SEric Cheng void
50568275SEric Cheng mac_release_tx_group(mac_impl_t *mip, mac_group_t *grp)
50578275SEric Cheng {
50588275SEric Cheng 	mac_client_impl_t *mcip = grp->mrg_tx_client;
50598275SEric Cheng 	mac_share_handle_t share = mcip->mci_share;
50608275SEric Cheng 	mac_ring_t *ring;
50618275SEric Cheng 
50628275SEric Cheng 	ASSERT(mip->mi_tx_group_type == MAC_GROUP_TYPE_DYNAMIC);
50638275SEric Cheng 	ASSERT(share != NULL);
50648275SEric Cheng 	ASSERT(grp->mrg_state == MAC_GROUP_STATE_RESERVED);
50658275SEric Cheng 
50668275SEric Cheng 	mip->mi_share_capab.ms_sremove(share, grp->mrg_driver);
50678275SEric Cheng 	while ((ring = grp->mrg_rings) != NULL) {
50688275SEric Cheng 		/* move the ring back to the pool */
50698275SEric Cheng 		(void) mac_group_mov_ring(mip, mip->mi_tx_groups +
50708275SEric Cheng 		    mip->mi_tx_group_count, ring);
50718275SEric Cheng 	}
50728275SEric Cheng 	mac_stop_group(grp);
50738275SEric Cheng 	mac_set_rx_group_state(grp, MAC_GROUP_STATE_REGISTERED);
50748275SEric Cheng 	grp->mrg_tx_client = NULL;
50758275SEric Cheng 	mip->mi_tx_group_free++;
50768275SEric Cheng }
50778275SEric Cheng 
50788275SEric Cheng /*
50798275SEric Cheng  * This is a 1-time control path activity initiated by the client (IP).
50808275SEric Cheng  * The mac perimeter protects against other simultaneous control activities,
50818275SEric Cheng  * for example an ioctl that attempts to change the degree of fanout and
50828275SEric Cheng  * increase or decrease the number of softrings associated with this Tx SRS.
50838275SEric Cheng  */
50848275SEric Cheng static mac_tx_notify_cb_t *
50858275SEric Cheng mac_client_tx_notify_add(mac_client_impl_t *mcip,
50868275SEric Cheng     mac_tx_notify_t notify, void *arg)
50878275SEric Cheng {
50888275SEric Cheng 	mac_cb_info_t *mcbi;
50898275SEric Cheng 	mac_tx_notify_cb_t *mtnfp;
50908275SEric Cheng 
50918275SEric Cheng 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
50928275SEric Cheng 
50938275SEric Cheng 	mtnfp = kmem_zalloc(sizeof (mac_tx_notify_cb_t), KM_SLEEP);
50948275SEric Cheng 	mtnfp->mtnf_fn = notify;
50958275SEric Cheng 	mtnfp->mtnf_arg = arg;
50968275SEric Cheng 	mtnfp->mtnf_link.mcb_objp = mtnfp;
50978275SEric Cheng 	mtnfp->mtnf_link.mcb_objsize = sizeof (mac_tx_notify_cb_t);
50988275SEric Cheng 	mtnfp->mtnf_link.mcb_flags = MCB_TX_NOTIFY_CB_T;
50998275SEric Cheng 
51008275SEric Cheng 	mcbi = &mcip->mci_tx_notify_cb_info;
51018275SEric Cheng 	mutex_enter(mcbi->mcbi_lockp);
51028275SEric Cheng 	mac_callback_add(mcbi, &mcip->mci_tx_notify_cb_list, &mtnfp->mtnf_link);
51038275SEric Cheng 	mutex_exit(mcbi->mcbi_lockp);
51048275SEric Cheng 	return (mtnfp);
51058275SEric Cheng }
51068275SEric Cheng 
51078275SEric Cheng static void
51088275SEric Cheng mac_client_tx_notify_remove(mac_client_impl_t *mcip, mac_tx_notify_cb_t *mtnfp)
51098275SEric Cheng {
51108275SEric Cheng 	mac_cb_info_t	*mcbi;
51118275SEric Cheng 	mac_cb_t	**cblist;
51128275SEric Cheng 
51138275SEric Cheng 	ASSERT(MAC_PERIM_HELD((mac_handle_t)mcip->mci_mip));
51148275SEric Cheng 
51158275SEric Cheng 	if (!mac_callback_find(&mcip->mci_tx_notify_cb_info,
51168275SEric Cheng 	    &mcip->mci_tx_notify_cb_list, &mtnfp->mtnf_link)) {
51178275SEric Cheng 		cmn_err(CE_WARN,
51188275SEric Cheng 		    "mac_client_tx_notify_remove: callback not "
51198275SEric Cheng 		    "found, mcip 0x%p mtnfp 0x%p", (void *)mcip, (void *)mtnfp);
51208275SEric Cheng 		return;
51218275SEric Cheng 	}
51228275SEric Cheng 
51238275SEric Cheng 	mcbi = &mcip->mci_tx_notify_cb_info;
51248275SEric Cheng 	cblist = &mcip->mci_tx_notify_cb_list;
51258275SEric Cheng 	mutex_enter(mcbi->mcbi_lockp);
51268275SEric Cheng 	if (mac_callback_remove(mcbi, cblist, &mtnfp->mtnf_link))
51278275SEric Cheng 		kmem_free(mtnfp, sizeof (mac_tx_notify_cb_t));
51288275SEric Cheng 	else
51298275SEric Cheng 		mac_callback_remove_wait(&mcip->mci_tx_notify_cb_info);
51308275SEric Cheng 	mutex_exit(mcbi->mcbi_lockp);
51318275SEric Cheng }
51328275SEric Cheng 
51338275SEric Cheng /*
51348275SEric Cheng  * mac_client_tx_notify():
51358275SEric Cheng  * call to add and remove flow control callback routine.
51368275SEric Cheng  */
51378275SEric Cheng mac_tx_notify_handle_t
51388275SEric Cheng mac_client_tx_notify(mac_client_handle_t mch, mac_tx_notify_t callb_func,
51398275SEric Cheng     void *ptr)
51408275SEric Cheng {
51418275SEric Cheng 	mac_client_impl_t	*mcip = (mac_client_impl_t *)mch;
51428275SEric Cheng 	mac_tx_notify_cb_t	*mtnfp = NULL;
51438275SEric Cheng 
51448275SEric Cheng 	i_mac_perim_enter(mcip->mci_mip);
51458275SEric Cheng 
51468275SEric Cheng 	if (callb_func != NULL) {
51478275SEric Cheng 		/* Add a notify callback */
51488275SEric Cheng 		mtnfp = mac_client_tx_notify_add(mcip, callb_func, ptr);
51498275SEric Cheng 	} else {
51508275SEric Cheng 		mac_client_tx_notify_remove(mcip, (mac_tx_notify_cb_t *)ptr);
51518275SEric Cheng 	}
51528275SEric Cheng 	i_mac_perim_exit(mcip->mci_mip);
51538275SEric Cheng 
51548275SEric Cheng 	return ((mac_tx_notify_handle_t)mtnfp);
51558275SEric Cheng }
5156