xref: /titanic_41/usr/src/uts/sun4v/io/vsw.c (revision 6a634c9dca3093f3922e4b7ab826d7bdf17bf78e)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 2006, 2010, Oracle and/or its affiliates. All rights reserved.
24  */
25 
26 #include <sys/types.h>
27 #include <sys/errno.h>
28 #include <sys/debug.h>
29 #include <sys/time.h>
30 #include <sys/sysmacros.h>
31 #include <sys/systm.h>
32 #include <sys/user.h>
33 #include <sys/stropts.h>
34 #include <sys/stream.h>
35 #include <sys/strlog.h>
36 #include <sys/strsubr.h>
37 #include <sys/cmn_err.h>
38 #include <sys/cpu.h>
39 #include <sys/kmem.h>
40 #include <sys/conf.h>
41 #include <sys/ddi.h>
42 #include <sys/sunddi.h>
43 #include <sys/ksynch.h>
44 #include <sys/stat.h>
45 #include <sys/kstat.h>
46 #include <sys/vtrace.h>
47 #include <sys/strsun.h>
48 #include <sys/dlpi.h>
49 #include <sys/ethernet.h>
50 #include <net/if.h>
51 #include <sys/varargs.h>
52 #include <sys/machsystm.h>
53 #include <sys/modctl.h>
54 #include <sys/modhash.h>
55 #include <sys/mac_provider.h>
56 #include <sys/mac_ether.h>
57 #include <sys/taskq.h>
58 #include <sys/note.h>
59 #include <sys/mach_descrip.h>
60 #include <sys/mac_provider.h>
61 #include <sys/mdeg.h>
62 #include <sys/ldc.h>
63 #include <sys/vsw_fdb.h>
64 #include <sys/vsw.h>
65 #include <sys/vio_mailbox.h>
66 #include <sys/vnet_mailbox.h>
67 #include <sys/vnet_common.h>
68 #include <sys/vio_util.h>
69 #include <sys/sdt.h>
70 #include <sys/atomic.h>
71 #include <sys/callb.h>
72 #include <sys/vlan.h>
73 
74 /*
75  * Function prototypes.
76  */
77 static	int vsw_attach(dev_info_t *, ddi_attach_cmd_t);
78 static	int vsw_detach(dev_info_t *, ddi_detach_cmd_t);
79 static	int vsw_unattach(vsw_t *vswp);
80 static	int vsw_get_md_physname(vsw_t *, md_t *, mde_cookie_t, char *);
81 static	int vsw_get_md_smodes(vsw_t *, md_t *, mde_cookie_t, uint8_t *);
82 void vsw_destroy_rxpools(void *);
83 
84 /* MDEG routines */
85 static	int vsw_mdeg_register(vsw_t *vswp);
86 static	void vsw_mdeg_unregister(vsw_t *vswp);
87 static	int vsw_mdeg_cb(void *cb_argp, mdeg_result_t *);
88 static	int vsw_port_mdeg_cb(void *cb_argp, mdeg_result_t *);
89 static	int vsw_get_initial_md_properties(vsw_t *vswp, md_t *, mde_cookie_t);
90 static	int vsw_read_mdprops(vsw_t *vswp);
91 static	void vsw_vlan_read_ids(void *arg, int type, md_t *mdp,
92 	mde_cookie_t node, uint16_t *pvidp, vsw_vlanid_t **vidspp,
93 	uint16_t *nvidsp, uint16_t *default_idp);
94 static	void vsw_port_read_bandwidth(vsw_port_t *portp, md_t *mdp,
95 	mde_cookie_t node, uint64_t *bw);
96 static	int vsw_port_read_props(vsw_port_t *portp, vsw_t *vswp,
97 	md_t *mdp, mde_cookie_t *node);
98 static	void vsw_read_pri_eth_types(vsw_t *vswp, md_t *mdp,
99 	mde_cookie_t node);
100 static	void vsw_mtu_read(vsw_t *vswp, md_t *mdp, mde_cookie_t node,
101 	uint32_t *mtu);
102 static	int vsw_mtu_update(vsw_t *vswp, uint32_t mtu);
103 static	void vsw_linkprop_read(vsw_t *vswp, md_t *mdp, mde_cookie_t node,
104 	boolean_t *pls);
105 static	void vsw_bandwidth_read(vsw_t *vswp, md_t *mdp, mde_cookie_t node,
106 	uint64_t *bw);
107 static	void vsw_update_md_prop(vsw_t *, md_t *, mde_cookie_t);
108 static void vsw_save_lmacaddr(vsw_t *vswp, uint64_t macaddr);
109 static boolean_t vsw_cmp_vids(vsw_vlanid_t *vids1,
110 	vsw_vlanid_t *vids2, int nvids);
111 
112 /* Mac driver related routines */
113 static int vsw_mac_register(vsw_t *);
114 static int vsw_mac_unregister(vsw_t *);
115 static int vsw_m_stat(void *, uint_t, uint64_t *);
116 static void vsw_m_stop(void *arg);
117 static int vsw_m_start(void *arg);
118 static int vsw_m_unicst(void *arg, const uint8_t *);
119 static int vsw_m_multicst(void *arg, boolean_t, const uint8_t *);
120 static int vsw_m_promisc(void *arg, boolean_t);
121 static mblk_t *vsw_m_tx(void *arg, mblk_t *);
122 void vsw_mac_link_update(vsw_t *vswp, link_state_t link_state);
123 void vsw_mac_rx(vsw_t *vswp, mac_resource_handle_t mrh,
124     mblk_t *mp, vsw_macrx_flags_t flags);
125 void vsw_physlink_state_update(vsw_t *vswp);
126 
127 /*
128  * Functions imported from other files.
129  */
130 extern void vsw_setup_switching_thread(void *arg);
131 extern int vsw_setup_switching_start(vsw_t *vswp);
132 extern void vsw_setup_switching_stop(vsw_t *vswp);
133 extern int vsw_setup_switching(vsw_t *);
134 extern void vsw_switch_frame_nop(vsw_t *vswp, mblk_t *mp, int caller,
135     vsw_port_t *port, mac_resource_handle_t mrh);
136 extern int vsw_add_mcst(vsw_t *, uint8_t, uint64_t, void *);
137 extern int vsw_del_mcst(vsw_t *, uint8_t, uint64_t, void *);
138 extern void vsw_del_mcst_vsw(vsw_t *);
139 extern mcst_addr_t *vsw_del_addr(uint8_t devtype, void *arg, uint64_t addr);
140 extern void vsw_detach_ports(vsw_t *vswp);
141 extern int vsw_port_add(vsw_t *vswp, md_t *mdp, mde_cookie_t *node);
142 extern int vsw_port_detach(vsw_t *vswp, int p_instance);
143 static int vsw_port_update(vsw_t *vswp, md_t *curr_mdp, mde_cookie_t curr_mdex,
144 	md_t *prev_mdp, mde_cookie_t prev_mdex);
145 extern	int vsw_port_attach(vsw_port_t *port);
146 extern vsw_port_t *vsw_lookup_port(vsw_t *vswp, int p_instance);
147 extern int vsw_mac_open(vsw_t *vswp);
148 extern void vsw_mac_close(vsw_t *vswp);
149 extern void vsw_mac_cleanup_ports(vsw_t *vswp);
150 extern void vsw_unset_addrs(vsw_t *vswp);
151 extern void vsw_setup_switching_post_process(vsw_t *vswp);
152 extern void vsw_create_vlans(void *arg, int type);
153 extern void vsw_destroy_vlans(void *arg, int type);
154 extern void vsw_vlan_add_ids(void *arg, int type);
155 extern void vsw_vlan_remove_ids(void *arg, int type);
156 extern void vsw_vlan_unaware_port_reset(vsw_port_t *portp);
157 extern uint32_t vsw_vlan_frame_untag(void *arg, int type, mblk_t **np,
158 	mblk_t **npt);
159 extern mblk_t *vsw_vlan_frame_pretag(void *arg, int type, mblk_t *mp);
160 extern void vsw_hio_cleanup(vsw_t *vswp);
161 extern void vsw_hio_start_ports(vsw_t *vswp);
162 extern void vsw_hio_port_update(vsw_port_t *portp, boolean_t hio_enabled);
163 extern int vsw_mac_multicast_add(vsw_t *, vsw_port_t *, mcst_addr_t *, int);
164 extern void vsw_mac_multicast_remove(vsw_t *, vsw_port_t *, mcst_addr_t *, int);
165 extern void vsw_mac_port_reconfig_vlans(vsw_port_t *portp, uint16_t new_pvid,
166     vsw_vlanid_t *new_vids, int new_nvids);
167 extern int vsw_mac_client_init(vsw_t *vswp, vsw_port_t *port, int type);
168 extern void vsw_mac_client_cleanup(vsw_t *vswp, vsw_port_t *port, int type);
169 extern void vsw_if_mac_reconfig(vsw_t *vswp, boolean_t update_vlans,
170     uint16_t new_pvid, vsw_vlanid_t *new_vids, int new_nvids);
171 extern void vsw_reset_ports(vsw_t *vswp);
172 extern void vsw_port_reset(vsw_port_t *portp);
173 extern void vsw_physlink_update_ports(vsw_t *vswp);
174 extern void vsw_update_bandwidth(vsw_t *vswp, vsw_port_t *port, int type,
175     uint64_t maxbw);
176 
177 /*
178  * Internal tunables.
179  */
180 int	vsw_num_handshakes = VNET_NUM_HANDSHAKES; /* # of handshake attempts */
181 int	vsw_wretries = 100;		/* # of write attempts */
182 int	vsw_setup_switching_delay = 3;	/* setup sw timeout interval in sec */
183 int	vsw_mac_open_retries = 300;	/* max # of mac_open() retries */
184 					/* 300*3 = 900sec(15min) of max tmout */
185 int	vsw_ldc_tx_delay = 5;		/* delay(ticks) for tx retries */
186 int	vsw_ldc_tx_retries = 10;	/* # of ldc tx retries */
187 int	vsw_ldc_retries = 5;		/* # of ldc_close() retries */
188 int	vsw_ldc_delay = 1000;		/* 1 ms delay for ldc_close() */
189 boolean_t vsw_ldc_rxthr_enabled = B_TRUE;	/* LDC Rx thread enabled */
190 boolean_t vsw_ldc_txthr_enabled = B_TRUE;	/* LDC Tx thread enabled */
191 int	vsw_rxpool_cleanup_delay = 100000;	/* 100ms */
192 
193 
194 uint32_t	vsw_fdb_nchains = 8;	/* # of chains in fdb hash table */
195 uint32_t	vsw_vlan_nchains = 4;	/* # of chains in vlan id hash table */
196 uint32_t	vsw_ethermtu = 1500;	/* mtu of the device */
197 
198 /* delay in usec to wait for all references on a fdb entry to be dropped */
199 uint32_t vsw_fdbe_refcnt_delay = 10;
200 
201 /*
202  * Default vlan id. This is only used internally when the "default-vlan-id"
203  * property is not present in the MD device node. Therefore, this should not be
204  * used as a tunable; if this value is changed, the corresponding variable
205  * should be updated to the same value in all vnets connected to this vsw.
206  */
207 uint16_t	vsw_default_vlan_id = 1;
208 
209 /*
210  * Workaround for a version handshake bug in obp's vnet.
211  * If vsw initiates version negotiation starting from the highest version,
212  * obp sends a nack and terminates version handshake. To workaround
213  * this, we do not initiate version handshake when the channel comes up.
214  * Instead, we wait for the peer to send its version info msg and go through
215  * the version protocol exchange. If we successfully negotiate a version,
216  * before sending the ack, we send our version info msg to the peer
217  * using the <major,minor> version that we are about to ack.
218  */
219 boolean_t vsw_obp_ver_proto_workaround = B_TRUE;
220 
221 /*
222  * In the absence of "priority-ether-types" property in MD, the following
223  * internal tunable can be set to specify a single priority ethertype.
224  */
225 uint64_t vsw_pri_eth_type = 0;
226 
227 /*
228  * Number of transmit priority buffers that are preallocated per device.
229  * This number is chosen to be a small value to throttle transmission
230  * of priority packets. Note: Must be a power of 2 for vio_create_mblks().
231  */
232 uint32_t vsw_pri_tx_nmblks = 64;
233 
234 /*
235  * Number of RARP packets sent to announce macaddr to the physical switch,
236  * after vsw's physical device is changed dynamically or after a guest (client
237  * vnet) is live migrated in.
238  */
239 uint32_t vsw_publish_macaddr_count = 3;
240 
241 /*
242  * Enable/disable HybridIO
243  */
244 boolean_t vsw_hio_enabled = B_TRUE;
245 
246 /*
247  * Max retries for HybridIO cleanup
248  */
249 int vsw_hio_max_cleanup_retries = 10;
250 
251 /*
252  * 10ms delay for HybridIO cleanup
253  */
254 int vsw_hio_cleanup_delay = 10000;
255 
256 /*
257  * Descriptor ring modes of LDC data transfer:
258  *
259  * 1) TxDring mode:
260  * In versions < v1.6 of VIO Protocol, we support only TxDring mode. In this
261  * mode, we create a transmit descriptor ring and export it to the peer through
262  * dring registration process of handshake. The descriptor ring is exported
263  * using LDC shared memory. Each descriptor is associated with a data buffer.
264  * The data buffer is also exported over LDC and the cookies for this data
265  * buffer are provided in the descriptor. The peer maps this ring as its
266  * receive ring. Similarly, the peer exports a transmit descriptor ring which
267  * is mapped by this device as its receive ring. In this mode, in a given data
268  * transfer direction, the transmitter copies the data to the exported data
269  * buffer (owned by itself), bound to the descriptor. The receiver uses the LDC
270  * cookies specified in the descriptor to copy the data into the receiving
271  * guest through the hypervisor (ldc_mem_copy()).
272  *
273  * 2) RxDringData mode:
274  * In versions >= v1.6 of VIO Protocol, we also support RxDringData mode. In
275  * this mode, we create a receive descriptor ring and export it to the peer
276  * through dring registration process of handshake. In addition, we export a
277  * receive buffer area and provide that information also in the dring
278  * registration message. The descriptor ring and the data buffer area are
279  * exported using LDC shared memory. Each descriptor is associated with a data
280  * buffer in the data buffer area and the offset of the specific data buffer
281  * within this area is specified in the descriptor. The peer maps this ring
282  * along with the data buffer area as its transmit ring. Similarly, the peer
283  * exports a receive ring which is mapped by this device as its transmit ring,
284  * along with its buffer area. In this mode, in a given data transfer
285  * direction, the transmitter copies the data to the data buffer offset
286  * specified in the descriptor. The receiver simply picks up the data buffer
287  * (owned by itself) without any copy operation into the receiving guest.
288  *
289  * We enable RxDringData mode during handshake negotiations if LDC supports
290  * mapping in large areas of shared memory(see ldc_is_viotsb_configured() API),
291  * which is required to support RxDringData mode.
292  */
293 
294 /*
295  * Number of descriptors;  must be power of 2.
296  */
297 uint32_t vsw_num_descriptors = VSW_NUM_DESCRIPTORS;
298 
299 /*
300  * In RxDringData mode, # of buffers is determined by multiplying the # of
301  * descriptors with the factor below. Note that the factor must be > 1; i.e,
302  * the # of buffers must always be > # of descriptors. This is needed because,
303  * while the shared memory buffers are sent up the stack on the receiver, the
304  * sender needs additional buffers that can be used for further transmits.
305  * See vsw_setup_rx_dring() for details.
306  */
307 uint32_t vsw_nrbufs_factor = 2;
308 
309 /*
310  * Delay when rx descr not ready; used in both dring modes.
311  */
312 int	vsw_recv_delay = 0;
313 
314 /*
315  * Retry when rx descr not ready; used in both dring modes.
316  */
317 int	vsw_recv_retries = 5;
318 
319 /*
320  * Max number of mblks received in one receive operation.
321  */
322 uint32_t vsw_chain_len = (VSW_NUM_MBLKS * 0.6);
323 
324 /*
325  * Internal tunables for receive buffer pools, that is,  the size and number of
326  * mblks for each pool. At least 3 sizes must be specified if these are used.
327  * The sizes must be specified in increasing order. Non-zero value of the first
328  * size will be used as a hint to use these values instead of the algorithm
329  * that determines the sizes based on MTU. Used in TxDring mode only.
330  */
331 uint32_t vsw_mblk_size1 = 0;
332 uint32_t vsw_mblk_size2 = 0;
333 uint32_t vsw_mblk_size3 = 0;
334 uint32_t vsw_mblk_size4 = 0;
335 uint32_t vsw_num_mblks1 = VSW_NUM_MBLKS;	/* number of mblks for pool1 */
336 uint32_t vsw_num_mblks2 = VSW_NUM_MBLKS;	/* number of mblks for pool2 */
337 uint32_t vsw_num_mblks3 = VSW_NUM_MBLKS;	/* number of mblks for pool3 */
338 uint32_t vsw_num_mblks4 = VSW_NUM_MBLKS;	/* number of mblks for pool4 */
339 
340 /*
341  * Set this to non-zero to enable additional internal receive buffer pools
342  * based on the MTU of the device for better performance at the cost of more
343  * memory consumption. This is turned off by default, to use allocb(9F) for
344  * receive buffer allocations of sizes > 2K.
345  */
346 boolean_t vsw_jumbo_rxpools = B_FALSE;
347 
348 /*
349  * vsw_max_tx_qcount is the maximum # of packets that can be queued
350  * before the tx worker thread begins processing the queue. Its value
351  * is chosen to be 4x the default length of tx descriptor ring.
352  */
353 uint32_t vsw_max_tx_qcount = 4 * VSW_NUM_DESCRIPTORS;
354 
355 /*
356  * MAC callbacks
357  */
358 static	mac_callbacks_t	vsw_m_callbacks = {
359 	0,
360 	vsw_m_stat,
361 	vsw_m_start,
362 	vsw_m_stop,
363 	vsw_m_promisc,
364 	vsw_m_multicst,
365 	vsw_m_unicst,
366 	vsw_m_tx
367 };
368 
369 static	struct	cb_ops	vsw_cb_ops = {
370 	nulldev,			/* cb_open */
371 	nulldev,			/* cb_close */
372 	nodev,				/* cb_strategy */
373 	nodev,				/* cb_print */
374 	nodev,				/* cb_dump */
375 	nodev,				/* cb_read */
376 	nodev,				/* cb_write */
377 	nodev,				/* cb_ioctl */
378 	nodev,				/* cb_devmap */
379 	nodev,				/* cb_mmap */
380 	nodev,				/* cb_segmap */
381 	nochpoll,			/* cb_chpoll */
382 	ddi_prop_op,			/* cb_prop_op */
383 	NULL,				/* cb_stream */
384 	D_MP,				/* cb_flag */
385 	CB_REV,				/* rev */
386 	nodev,				/* int (*cb_aread)() */
387 	nodev				/* int (*cb_awrite)() */
388 };
389 
390 static	struct	dev_ops	vsw_ops = {
391 	DEVO_REV,		/* devo_rev */
392 	0,			/* devo_refcnt */
393 	NULL,			/* devo_getinfo */
394 	nulldev,		/* devo_identify */
395 	nulldev,		/* devo_probe */
396 	vsw_attach,		/* devo_attach */
397 	vsw_detach,		/* devo_detach */
398 	nodev,			/* devo_reset */
399 	&vsw_cb_ops,		/* devo_cb_ops */
400 	(struct bus_ops *)NULL,	/* devo_bus_ops */
401 	ddi_power		/* devo_power */
402 };
403 
404 extern	struct	mod_ops	mod_driverops;
405 static struct modldrv vswmodldrv = {
406 	&mod_driverops,
407 	"sun4v Virtual Switch",
408 	&vsw_ops,
409 };
410 
411 #define	LDC_ENTER_LOCK(ldcp)	\
412 				mutex_enter(&((ldcp)->ldc_cblock));\
413 				mutex_enter(&((ldcp)->ldc_rxlock));\
414 				mutex_enter(&((ldcp)->ldc_txlock));
415 #define	LDC_EXIT_LOCK(ldcp)	\
416 				mutex_exit(&((ldcp)->ldc_txlock));\
417 				mutex_exit(&((ldcp)->ldc_rxlock));\
418 				mutex_exit(&((ldcp)->ldc_cblock));
419 
420 /* Driver soft state ptr  */
421 static void	*vsw_state;
422 
423 /*
424  * Linked list of "vsw_t" structures - one per instance.
425  */
426 vsw_t		*vsw_head = NULL;
427 krwlock_t	vsw_rw;
428 
429 /*
430  * Property names
431  */
432 static char vdev_propname[] = "virtual-device";
433 static char vsw_propname[] = "virtual-network-switch";
434 static char physdev_propname[] = "vsw-phys-dev";
435 static char smode_propname[] = "vsw-switch-mode";
436 static char macaddr_propname[] = "local-mac-address";
437 static char remaddr_propname[] = "remote-mac-address";
438 static char ldcids_propname[] = "ldc-ids";
439 static char chan_propname[] = "channel-endpoint";
440 static char id_propname[] = "id";
441 static char reg_propname[] = "reg";
442 static char pri_types_propname[] = "priority-ether-types";
443 static char vsw_pvid_propname[] = "port-vlan-id";
444 static char vsw_vid_propname[] = "vlan-id";
445 static char vsw_dvid_propname[] = "default-vlan-id";
446 static char port_pvid_propname[] = "remote-port-vlan-id";
447 static char port_vid_propname[] = "remote-vlan-id";
448 static char hybrid_propname[] = "hybrid";
449 static char vsw_mtu_propname[] = "mtu";
450 static char vsw_linkprop_propname[] = "linkprop";
451 static char vsw_maxbw_propname[] = "maxbw";
452 static char port_maxbw_propname[] = "maxbw";
453 
454 /*
455  * Matching criteria passed to the MDEG to register interest
456  * in changes to 'virtual-device-port' nodes identified by their
457  * 'id' property.
458  */
459 static md_prop_match_t vport_prop_match[] = {
460 	{ MDET_PROP_VAL,    "id"   },
461 	{ MDET_LIST_END,    NULL    }
462 };
463 
464 static mdeg_node_match_t vport_match = { "virtual-device-port",
465 						vport_prop_match };
466 
467 /*
468  * Matching criteria passed to the MDEG to register interest
469  * in changes to 'virtual-device' nodes (i.e. vsw nodes) identified
470  * by their 'name' and 'cfg-handle' properties.
471  */
472 static md_prop_match_t vdev_prop_match[] = {
473 	{ MDET_PROP_STR,    "name"   },
474 	{ MDET_PROP_VAL,    "cfg-handle" },
475 	{ MDET_LIST_END,    NULL    }
476 };
477 
478 static mdeg_node_match_t vdev_match = { "virtual-device",
479 						vdev_prop_match };
480 
481 
482 /*
483  * Specification of an MD node passed to the MDEG to filter any
484  * 'vport' nodes that do not belong to the specified node. This
485  * template is copied for each vsw instance and filled in with
486  * the appropriate 'cfg-handle' value before being passed to the MDEG.
487  */
488 static mdeg_prop_spec_t vsw_prop_template[] = {
489 	{ MDET_PROP_STR,    "name",		vsw_propname },
490 	{ MDET_PROP_VAL,    "cfg-handle",	NULL	},
491 	{ MDET_LIST_END,    NULL,		NULL	}
492 };
493 
494 #define	VSW_SET_MDEG_PROP_INST(specp, val)	(specp)[1].ps_val = (val);
495 
496 #ifdef	DEBUG
497 /*
498  * Print debug messages - set to 0x1f to enable all msgs
499  * or 0x0 to turn all off.
500  */
501 int vswdbg = 0x0;
502 
503 /*
504  * debug levels:
505  * 0x01:	Function entry/exit tracing
506  * 0x02:	Internal function messages
507  * 0x04:	Verbose internal messages
508  * 0x08:	Warning messages
509  * 0x10:	Error messages
510  */
511 
512 void
vswdebug(vsw_t * vswp,const char * fmt,...)513 vswdebug(vsw_t *vswp, const char *fmt, ...)
514 {
515 	char buf[512];
516 	va_list ap;
517 
518 	va_start(ap, fmt);
519 	(void) vsprintf(buf, fmt, ap);
520 	va_end(ap);
521 
522 	if (vswp == NULL)
523 		cmn_err(CE_CONT, "%s\n", buf);
524 	else
525 		cmn_err(CE_CONT, "vsw%d: %s\n", vswp->instance, buf);
526 }
527 
528 #endif	/* DEBUG */
529 
530 static struct modlinkage modlinkage = {
531 	MODREV_1,
532 	&vswmodldrv,
533 	NULL
534 };
535 
536 int
_init(void)537 _init(void)
538 {
539 	int status;
540 
541 	rw_init(&vsw_rw, NULL, RW_DRIVER, NULL);
542 
543 	status = ddi_soft_state_init(&vsw_state, sizeof (vsw_t), 1);
544 	if (status != 0) {
545 		return (status);
546 	}
547 
548 	mac_init_ops(&vsw_ops, DRV_NAME);
549 	status = mod_install(&modlinkage);
550 	if (status != 0) {
551 		ddi_soft_state_fini(&vsw_state);
552 	}
553 	return (status);
554 }
555 
556 int
_fini(void)557 _fini(void)
558 {
559 	int status;
560 
561 	status = mod_remove(&modlinkage);
562 	if (status != 0)
563 		return (status);
564 	mac_fini_ops(&vsw_ops);
565 	ddi_soft_state_fini(&vsw_state);
566 
567 	rw_destroy(&vsw_rw);
568 
569 	return (status);
570 }
571 
572 int
_info(struct modinfo * modinfop)573 _info(struct modinfo *modinfop)
574 {
575 	return (mod_info(&modlinkage, modinfop));
576 }
577 
578 static int
vsw_attach(dev_info_t * dip,ddi_attach_cmd_t cmd)579 vsw_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
580 {
581 	vsw_t			*vswp;
582 	int			instance;
583 	char			hashname[MAXNAMELEN];
584 	char			qname[TASKQ_NAMELEN];
585 	vsw_attach_progress_t	progress = PROG_init;
586 	int			rv;
587 
588 	switch (cmd) {
589 	case DDI_ATTACH:
590 		break;
591 	case DDI_RESUME:
592 		/* nothing to do for this non-device */
593 		return (DDI_SUCCESS);
594 	case DDI_PM_RESUME:
595 	default:
596 		return (DDI_FAILURE);
597 	}
598 
599 	instance = ddi_get_instance(dip);
600 	if (ddi_soft_state_zalloc(vsw_state, instance) != DDI_SUCCESS) {
601 		DERR(NULL, "vsw%d: ddi_soft_state_zalloc failed", instance);
602 		return (DDI_FAILURE);
603 	}
604 	vswp = ddi_get_soft_state(vsw_state, instance);
605 
606 	if (vswp == NULL) {
607 		DERR(NULL, "vsw%d: ddi_get_soft_state failed", instance);
608 		goto vsw_attach_fail;
609 	}
610 
611 	vswp->dip = dip;
612 	vswp->instance = instance;
613 	vswp->phys_link_state = LINK_STATE_UNKNOWN;
614 	ddi_set_driver_private(dip, (caddr_t)vswp);
615 
616 	mutex_init(&vswp->mac_lock, NULL, MUTEX_DRIVER, NULL);
617 	mutex_init(&vswp->mca_lock, NULL, MUTEX_DRIVER, NULL);
618 	mutex_init(&vswp->sw_thr_lock, NULL, MUTEX_DRIVER, NULL);
619 	cv_init(&vswp->sw_thr_cv, NULL, CV_DRIVER, NULL);
620 	rw_init(&vswp->maccl_rwlock, NULL, RW_DRIVER, NULL);
621 	rw_init(&vswp->if_lockrw, NULL, RW_DRIVER, NULL);
622 	rw_init(&vswp->mfdbrw, NULL, RW_DRIVER, NULL);
623 	rw_init(&vswp->plist.lockrw, NULL, RW_DRIVER, NULL);
624 
625 	progress |= PROG_locks;
626 
627 	rv = vsw_read_mdprops(vswp);
628 	if (rv != 0)
629 		goto vsw_attach_fail;
630 
631 	progress |= PROG_readmd;
632 
633 	/* setup the unicast forwarding database  */
634 	(void) snprintf(hashname, MAXNAMELEN, "vsw_unicst_table-%d",
635 	    vswp->instance);
636 	D2(vswp, "creating unicast hash table (%s)...", hashname);
637 	vswp->fdb_nchains = vsw_fdb_nchains;
638 	vswp->fdb_hashp = mod_hash_create_ptrhash(hashname, vswp->fdb_nchains,
639 	    mod_hash_null_valdtor, sizeof (void *));
640 	vsw_create_vlans((void *)vswp, VSW_LOCALDEV);
641 	progress |= PROG_fdb;
642 
643 	/* setup the multicast fowarding database */
644 	(void) snprintf(hashname, MAXNAMELEN, "vsw_mcst_table-%d",
645 	    vswp->instance);
646 	D2(vswp, "creating multicast hash table %s)...", hashname);
647 	vswp->mfdb = mod_hash_create_ptrhash(hashname, vsw_fdb_nchains,
648 	    mod_hash_null_valdtor, sizeof (void *));
649 
650 	progress |= PROG_mfdb;
651 
652 	/*
653 	 * Create the taskq which will process all the VIO
654 	 * control messages.
655 	 */
656 	(void) snprintf(qname, TASKQ_NAMELEN, "taskq%d", vswp->instance);
657 	if ((vswp->taskq_p = ddi_taskq_create(vswp->dip, qname, 1,
658 	    TASKQ_DEFAULTPRI, 0)) == NULL) {
659 		cmn_err(CE_WARN, "!vsw%d: Unable to create task queue",
660 		    vswp->instance);
661 		goto vsw_attach_fail;
662 	}
663 
664 	progress |= PROG_taskq;
665 
666 	(void) snprintf(qname, TASKQ_NAMELEN, "rxpool_taskq%d",
667 	    vswp->instance);
668 	if ((vswp->rxp_taskq = ddi_taskq_create(vswp->dip, qname, 1,
669 	    TASKQ_DEFAULTPRI, 0)) == NULL) {
670 		cmn_err(CE_WARN, "!vsw%d: Unable to create rxp task queue",
671 		    vswp->instance);
672 		goto vsw_attach_fail;
673 	}
674 
675 	progress |= PROG_rxp_taskq;
676 
677 	/* prevent auto-detaching */
678 	if (ddi_prop_update_int(DDI_DEV_T_NONE, vswp->dip,
679 	    DDI_NO_AUTODETACH, 1) != DDI_SUCCESS) {
680 		cmn_err(CE_NOTE, "!Unable to set \"%s\" property for "
681 		    "instance %u", DDI_NO_AUTODETACH, instance);
682 	}
683 
684 	/*
685 	 * The null switching function is set to avoid panic until
686 	 * switch mode is setup.
687 	 */
688 	vswp->vsw_switch_frame = vsw_switch_frame_nop;
689 
690 	/*
691 	 * Setup the required switching mode, based on the mdprops that we read
692 	 * earlier. We start a thread to do this, to avoid calling mac_open()
693 	 * directly from attach().
694 	 */
695 	rv = vsw_setup_switching_start(vswp);
696 	if (rv != 0) {
697 		goto vsw_attach_fail;
698 	}
699 
700 	progress |= PROG_swmode;
701 
702 	/* Register with mac layer as a provider */
703 	rv = vsw_mac_register(vswp);
704 	if (rv != 0)
705 		goto vsw_attach_fail;
706 
707 	progress |= PROG_macreg;
708 
709 	/*
710 	 * Now we have everything setup, register an interest in
711 	 * specific MD nodes.
712 	 *
713 	 * The callback is invoked in 2 cases, firstly if upon mdeg
714 	 * registration there are existing nodes which match our specified
715 	 * criteria, and secondly if the MD is changed (and again, there
716 	 * are nodes which we are interested in present within it. Note
717 	 * that our callback will be invoked even if our specified nodes
718 	 * have not actually changed).
719 	 *
720 	 */
721 	rv = vsw_mdeg_register(vswp);
722 	if (rv != 0)
723 		goto vsw_attach_fail;
724 
725 	progress |= PROG_mdreg;
726 
727 	vswp->attach_progress = progress;
728 
729 	WRITE_ENTER(&vsw_rw);
730 	vswp->next = vsw_head;
731 	vsw_head = vswp;
732 	RW_EXIT(&vsw_rw);
733 
734 	ddi_report_dev(vswp->dip);
735 	return (DDI_SUCCESS);
736 
737 vsw_attach_fail:
738 	DERR(NULL, "vsw_attach: failed");
739 
740 	vswp->attach_progress = progress;
741 	(void) vsw_unattach(vswp);
742 	ddi_soft_state_free(vsw_state, instance);
743 	return (DDI_FAILURE);
744 }
745 
746 static int
vsw_detach(dev_info_t * dip,ddi_detach_cmd_t cmd)747 vsw_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
748 {
749 	vsw_t			**vswpp, *vswp;
750 	int 			instance;
751 
752 	instance = ddi_get_instance(dip);
753 	vswp = ddi_get_soft_state(vsw_state, instance);
754 
755 	if (vswp == NULL) {
756 		return (DDI_FAILURE);
757 	}
758 
759 	switch (cmd) {
760 	case DDI_DETACH:
761 		break;
762 	case DDI_SUSPEND:
763 	case DDI_PM_SUSPEND:
764 	default:
765 		return (DDI_FAILURE);
766 	}
767 
768 	D2(vswp, "detaching instance %d", instance);
769 
770 	if (vsw_unattach(vswp) != 0) {
771 		return (DDI_FAILURE);
772 	}
773 
774 	ddi_remove_minor_node(dip, NULL);
775 
776 	WRITE_ENTER(&vsw_rw);
777 	for (vswpp = &vsw_head; *vswpp; vswpp = &(*vswpp)->next) {
778 		if (*vswpp == vswp) {
779 			*vswpp = vswp->next;
780 			break;
781 		}
782 	}
783 	RW_EXIT(&vsw_rw);
784 
785 	ddi_soft_state_free(vsw_state, instance);
786 
787 	return (DDI_SUCCESS);
788 }
789 
790 /*
791  * Common routine to handle vsw_attach() failure and vsw_detach(). Note that
792  * the only reason this function could fail is if mac_unregister() fails.
793  * Otherwise, this function must ensure that all resources are freed and return
794  * success.
795  */
796 static int
vsw_unattach(vsw_t * vswp)797 vsw_unattach(vsw_t *vswp)
798 {
799 	vsw_attach_progress_t	progress;
800 
801 	progress = vswp->attach_progress;
802 
803 	/*
804 	 * Unregister from the gldv3 subsystem. This can fail, in particular
805 	 * if there are still any open references to this mac device; in which
806 	 * case we just return failure without continuing to detach further.
807 	 */
808 	if (progress & PROG_macreg) {
809 		if (vsw_mac_unregister(vswp) != 0) {
810 			cmn_err(CE_WARN, "!vsw%d: Unable to detach from "
811 			    "MAC layer", vswp->instance);
812 			return (1);
813 		}
814 		progress &= ~PROG_macreg;
815 	}
816 
817 	/*
818 	 * Now that we have unregistered from gldv3, we must finish all other
819 	 * steps and successfully return from this function; otherwise we will
820 	 * end up leaving the device in a broken/unusable state.
821 	 *
822 	 * If we have registered with mdeg, unregister now to stop further
823 	 * callbacks to this vsw device and/or its ports. Then, detach any
824 	 * existing ports.
825 	 */
826 	if (progress & PROG_mdreg) {
827 		vsw_mdeg_unregister(vswp);
828 		vsw_detach_ports(vswp);
829 		progress &= ~PROG_mdreg;
830 	}
831 
832 	/*
833 	 * If we have started a thread to setup the switching mode, stop it, if
834 	 * it is still running. If it has finished setting up the switching
835 	 * mode, then we need to clean up some additional things if we are
836 	 * running in L2 mode: first free up any hybrid resources; then stop
837 	 * and close the underlying physical device. Note that we would have
838 	 * already released all per mac_client resources (ucast, mcast addrs,
839 	 * hio-shares etc) as all the ports are detached and if the vsw device
840 	 * itself was in use as an interface, it has been unplumbed (otherwise
841 	 * mac_unregister() above would fail).
842 	 */
843 	if (progress & PROG_swmode) {
844 
845 		vsw_setup_switching_stop(vswp);
846 
847 		if (vswp->hio_capable == B_TRUE) {
848 			vsw_hio_cleanup(vswp);
849 			vswp->hio_capable = B_FALSE;
850 		}
851 
852 		mutex_enter(&vswp->mac_lock);
853 		vsw_mac_close(vswp);
854 		mutex_exit(&vswp->mac_lock);
855 
856 		progress &= ~PROG_swmode;
857 	}
858 
859 	/*
860 	 * We now destroy the taskq used to clean up rx mblk pools that
861 	 * couldn't be destroyed when the ports/channels were detached.
862 	 * We implicitly wait for those tasks to complete in
863 	 * ddi_taskq_destroy().
864 	 */
865 	if (progress & PROG_rxp_taskq) {
866 		ddi_taskq_destroy(vswp->rxp_taskq);
867 		progress &= ~PROG_rxp_taskq;
868 	}
869 
870 	/*
871 	 * By now any pending tasks have finished and the underlying
872 	 * ldc's have been destroyed, so its safe to delete the control
873 	 * message taskq.
874 	 */
875 	if (progress & PROG_taskq) {
876 		ddi_taskq_destroy(vswp->taskq_p);
877 		progress &= ~PROG_taskq;
878 	}
879 
880 	/* Destroy the multicast hash table */
881 	if (progress & PROG_mfdb) {
882 		mod_hash_destroy_hash(vswp->mfdb);
883 		progress &= ~PROG_mfdb;
884 	}
885 
886 	/* Destroy the vlan hash table and fdb */
887 	if (progress & PROG_fdb) {
888 		vsw_destroy_vlans(vswp, VSW_LOCALDEV);
889 		mod_hash_destroy_hash(vswp->fdb_hashp);
890 		progress &= ~PROG_fdb;
891 	}
892 
893 	if (progress & PROG_readmd) {
894 		if (VSW_PRI_ETH_DEFINED(vswp)) {
895 			kmem_free(vswp->pri_types,
896 			    sizeof (uint16_t) * vswp->pri_num_types);
897 			(void) vio_destroy_mblks(vswp->pri_tx_vmp);
898 		}
899 		progress &= ~PROG_readmd;
900 	}
901 
902 	if (progress & PROG_locks) {
903 		rw_destroy(&vswp->plist.lockrw);
904 		rw_destroy(&vswp->mfdbrw);
905 		rw_destroy(&vswp->if_lockrw);
906 		rw_destroy(&vswp->maccl_rwlock);
907 		cv_destroy(&vswp->sw_thr_cv);
908 		mutex_destroy(&vswp->sw_thr_lock);
909 		mutex_destroy(&vswp->mca_lock);
910 		mutex_destroy(&vswp->mac_lock);
911 		progress &= ~PROG_locks;
912 	}
913 
914 	vswp->attach_progress = progress;
915 
916 	return (0);
917 }
918 
919 void
vsw_destroy_rxpools(void * arg)920 vsw_destroy_rxpools(void *arg)
921 {
922 	vio_mblk_pool_t	*poolp = (vio_mblk_pool_t *)arg;
923 	vio_mblk_pool_t	*npoolp;
924 
925 	while (poolp != NULL) {
926 		npoolp =  poolp->nextp;
927 		while (vio_destroy_mblks(poolp) != 0) {
928 			delay(drv_usectohz(vsw_rxpool_cleanup_delay));
929 		}
930 		poolp = npoolp;
931 	}
932 }
933 
934 /*
935  * Get the value of the "vsw-phys-dev" property in the specified
936  * node. This property is the name of the physical device that
937  * the virtual switch will use to talk to the outside world.
938  *
939  * Note it is valid for this property to be NULL (but the property
940  * itself must exist). Callers of this routine should verify that
941  * the value returned is what they expected (i.e. either NULL or non NULL).
942  *
943  * On success returns value of the property in region pointed to by
944  * the 'name' argument, and with return value of 0. Otherwise returns 1.
945  */
946 static int
vsw_get_md_physname(vsw_t * vswp,md_t * mdp,mde_cookie_t node,char * name)947 vsw_get_md_physname(vsw_t *vswp, md_t *mdp, mde_cookie_t node, char *name)
948 {
949 	int		len = 0;
950 	int		instance;
951 	char		*physname = NULL;
952 	char		*dev;
953 	const char	*dev_name;
954 	char		myname[MAXNAMELEN];
955 
956 	dev_name = ddi_driver_name(vswp->dip);
957 	instance = ddi_get_instance(vswp->dip);
958 	(void) snprintf(myname, MAXNAMELEN, "%s%d", dev_name, instance);
959 
960 	if (md_get_prop_data(mdp, node, physdev_propname,
961 	    (uint8_t **)(&physname), &len) != 0) {
962 		cmn_err(CE_WARN, "!vsw%d: Unable to get name(s) of physical "
963 		    "device(s) from MD", vswp->instance);
964 		return (1);
965 	} else if ((strlen(physname) + 1) > LIFNAMSIZ) {
966 		cmn_err(CE_WARN, "!vsw%d: %s is too long a device name",
967 		    vswp->instance, physname);
968 		return (1);
969 	} else if (strcmp(myname, physname) == 0) {
970 		/*
971 		 * Prevent the vswitch from opening itself as the
972 		 * network device.
973 		 */
974 		cmn_err(CE_WARN, "!vsw%d: %s is an invalid device name",
975 		    vswp->instance, physname);
976 		return (1);
977 	} else {
978 		(void) strncpy(name, physname, strlen(physname) + 1);
979 		D2(vswp, "%s: using first device specified (%s)",
980 		    __func__, physname);
981 	}
982 
983 #ifdef DEBUG
984 	/*
985 	 * As a temporary measure to aid testing we check to see if there
986 	 * is a vsw.conf file present. If there is we use the value of the
987 	 * vsw_physname property in the file as the name of the physical
988 	 * device, overriding the value from the MD.
989 	 *
990 	 * There may be multiple devices listed, but for the moment
991 	 * we just use the first one.
992 	 */
993 	if (ddi_prop_lookup_string(DDI_DEV_T_ANY, vswp->dip, 0,
994 	    "vsw_physname", &dev) == DDI_PROP_SUCCESS) {
995 		if ((strlen(dev) + 1) > LIFNAMSIZ) {
996 			cmn_err(CE_WARN, "vsw%d: %s is too long a device name",
997 			    vswp->instance, dev);
998 			ddi_prop_free(dev);
999 			return (1);
1000 		} else {
1001 			cmn_err(CE_NOTE, "vsw%d: Using device name (%s) from "
1002 			    "config file", vswp->instance, dev);
1003 
1004 			(void) strncpy(name, dev, strlen(dev) + 1);
1005 		}
1006 
1007 		ddi_prop_free(dev);
1008 	}
1009 #endif
1010 
1011 	return (0);
1012 }
1013 
1014 /*
1015  * Read the 'vsw-switch-mode' property from the specified MD node.
1016  *
1017  * Returns 0 on success, otherwise returns 1.
1018  */
1019 static int
vsw_get_md_smodes(vsw_t * vswp,md_t * mdp,mde_cookie_t node,uint8_t * mode)1020 vsw_get_md_smodes(vsw_t *vswp, md_t *mdp, mde_cookie_t node, uint8_t *mode)
1021 {
1022 	int		len = 0;
1023 	char		*smode = NULL;
1024 	char		*curr_mode = NULL;
1025 
1026 	D1(vswp, "%s: enter", __func__);
1027 
1028 	/*
1029 	 * Get the switch-mode property. The modes are listed in
1030 	 * decreasing order of preference, i.e. prefered mode is
1031 	 * first item in list.
1032 	 */
1033 	len = 0;
1034 	if (md_get_prop_data(mdp, node, smode_propname,
1035 	    (uint8_t **)(&smode), &len) != 0) {
1036 		/*
1037 		 * Unable to get switch-mode property from MD, nothing
1038 		 * more we can do.
1039 		 */
1040 		cmn_err(CE_WARN, "!vsw%d: Unable to get switch mode property"
1041 		    " from the MD", vswp->instance);
1042 		return (1);
1043 	}
1044 
1045 	curr_mode = smode;
1046 	/*
1047 	 * Modes of operation:
1048 	 * 'switched'	 - layer 2 switching, underlying HW in
1049 	 *			programmed mode.
1050 	 * 'promiscuous' - layer 2 switching, underlying HW in
1051 	 *			promiscuous mode.
1052 	 * 'routed'	 - layer 3 (i.e. IP) routing, underlying HW
1053 	 *			in non-promiscuous mode.
1054 	 */
1055 	while (curr_mode < (smode + len)) {
1056 		D2(vswp, "%s: curr_mode = [%s]", __func__, curr_mode);
1057 		if (strcmp(curr_mode, "switched") == 0) {
1058 			*mode = VSW_LAYER2;
1059 		} else if (strcmp(curr_mode, "promiscuous") == 0) {
1060 			*mode = VSW_LAYER2 | VSW_LAYER2_PROMISC;
1061 		} else if (strcmp(curr_mode, "routed") == 0) {
1062 			*mode = VSW_LAYER3;
1063 		} else {
1064 			cmn_err(CE_WARN, "!vsw%d: Unknown switch mode %s, "
1065 			    "setting to default switched mode",
1066 			    vswp->instance, curr_mode);
1067 			*mode = VSW_LAYER2;
1068 		}
1069 		curr_mode += strlen(curr_mode) + 1;
1070 	}
1071 
1072 	D2(vswp, "%s: %d mode", __func__, *mode);
1073 
1074 	D1(vswp, "%s: exit", __func__);
1075 
1076 	return (0);
1077 }
1078 
1079 /*
1080  * Register with the MAC layer as a network device, so we
1081  * can be plumbed if necessary.
1082  */
1083 static int
vsw_mac_register(vsw_t * vswp)1084 vsw_mac_register(vsw_t *vswp)
1085 {
1086 	mac_register_t	*macp;
1087 	int		rv;
1088 
1089 	D1(vswp, "%s: enter", __func__);
1090 
1091 	if ((macp = mac_alloc(MAC_VERSION)) == NULL)
1092 		return (EINVAL);
1093 	macp->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
1094 	macp->m_driver = vswp;
1095 	macp->m_dip = vswp->dip;
1096 	macp->m_src_addr = (uint8_t *)&vswp->if_addr;
1097 	macp->m_callbacks = &vsw_m_callbacks;
1098 	macp->m_min_sdu = 0;
1099 	macp->m_max_sdu = vswp->mtu;
1100 	macp->m_margin = VLAN_TAGSZ;
1101 	rv = mac_register(macp, &vswp->if_mh);
1102 	mac_free(macp);
1103 	if (rv != 0) {
1104 		/*
1105 		 * Treat this as a non-fatal error as we may be
1106 		 * able to operate in some other mode.
1107 		 */
1108 		cmn_err(CE_NOTE, "!vsw%d: Unable to register as "
1109 		    "a provider with MAC layer", vswp->instance);
1110 		return (rv);
1111 	}
1112 
1113 	vswp->if_state |= VSW_IF_REG;
1114 
1115 	D1(vswp, "%s: exit", __func__);
1116 
1117 	return (rv);
1118 }
1119 
1120 static int
vsw_mac_unregister(vsw_t * vswp)1121 vsw_mac_unregister(vsw_t *vswp)
1122 {
1123 	int		rv = 0;
1124 
1125 	D1(vswp, "%s: enter", __func__);
1126 
1127 	WRITE_ENTER(&vswp->if_lockrw);
1128 
1129 	if (vswp->if_state & VSW_IF_REG) {
1130 		rv = mac_unregister(vswp->if_mh);
1131 		if (rv != 0) {
1132 			DWARN(vswp, "%s: unable to unregister from MAC "
1133 			    "framework", __func__);
1134 
1135 			RW_EXIT(&vswp->if_lockrw);
1136 			D1(vswp, "%s: fail exit", __func__);
1137 			return (rv);
1138 		}
1139 
1140 		/* mark i/f as down and unregistered */
1141 		vswp->if_state &= ~(VSW_IF_UP | VSW_IF_REG);
1142 	}
1143 	RW_EXIT(&vswp->if_lockrw);
1144 
1145 	D1(vswp, "%s: exit", __func__);
1146 
1147 	return (rv);
1148 }
1149 
1150 static int
vsw_m_stat(void * arg,uint_t stat,uint64_t * val)1151 vsw_m_stat(void *arg, uint_t stat, uint64_t *val)
1152 {
1153 	vsw_t			*vswp = (vsw_t *)arg;
1154 
1155 	D1(vswp, "%s: enter", __func__);
1156 
1157 	mutex_enter(&vswp->mac_lock);
1158 	if (vswp->mh == NULL) {
1159 		mutex_exit(&vswp->mac_lock);
1160 		return (EINVAL);
1161 	}
1162 
1163 	/* return stats from underlying device */
1164 	*val = mac_stat_get(vswp->mh, stat);
1165 
1166 	mutex_exit(&vswp->mac_lock);
1167 
1168 	return (0);
1169 }
1170 
1171 static void
vsw_m_stop(void * arg)1172 vsw_m_stop(void *arg)
1173 {
1174 	vsw_t	*vswp = (vsw_t *)arg;
1175 
1176 	D1(vswp, "%s: enter", __func__);
1177 
1178 	WRITE_ENTER(&vswp->if_lockrw);
1179 	vswp->if_state &= ~VSW_IF_UP;
1180 	RW_EXIT(&vswp->if_lockrw);
1181 
1182 	/* Cleanup and close the mac client */
1183 	vsw_mac_client_cleanup(vswp, NULL, VSW_LOCALDEV);
1184 
1185 	D1(vswp, "%s: exit (state = %d)", __func__, vswp->if_state);
1186 }
1187 
1188 static int
vsw_m_start(void * arg)1189 vsw_m_start(void *arg)
1190 {
1191 	int		rv;
1192 	vsw_t		*vswp = (vsw_t *)arg;
1193 
1194 	D1(vswp, "%s: enter", __func__);
1195 
1196 	WRITE_ENTER(&vswp->if_lockrw);
1197 
1198 	vswp->if_state |= VSW_IF_UP;
1199 
1200 	if (vswp->switching_setup_done == B_FALSE) {
1201 		/*
1202 		 * If the switching mode has not been setup yet, just
1203 		 * return. The unicast address will be programmed
1204 		 * after the physical device is successfully setup by the
1205 		 * timeout handler.
1206 		 */
1207 		RW_EXIT(&vswp->if_lockrw);
1208 		return (0);
1209 	}
1210 
1211 	/* if in layer2 mode, program unicast address. */
1212 	if (vswp->mh != NULL) {
1213 		/* Init a mac client and program addresses */
1214 		rv = vsw_mac_client_init(vswp, NULL, VSW_LOCALDEV);
1215 		if (rv != 0) {
1216 			cmn_err(CE_NOTE,
1217 			    "!vsw%d: failed to program interface "
1218 			    "unicast address\n", vswp->instance);
1219 		}
1220 	}
1221 
1222 	RW_EXIT(&vswp->if_lockrw);
1223 
1224 	D1(vswp, "%s: exit (state = %d)", __func__, vswp->if_state);
1225 	return (0);
1226 }
1227 
1228 /*
1229  * Change the local interface address.
1230  *
1231  * Note: we don't support this entry point. The local
1232  * mac address of the switch can only be changed via its
1233  * MD node properties.
1234  */
1235 static int
vsw_m_unicst(void * arg,const uint8_t * macaddr)1236 vsw_m_unicst(void *arg, const uint8_t *macaddr)
1237 {
1238 	_NOTE(ARGUNUSED(arg, macaddr))
1239 
1240 	return (DDI_FAILURE);
1241 }
1242 
1243 static int
vsw_m_multicst(void * arg,boolean_t add,const uint8_t * mca)1244 vsw_m_multicst(void *arg, boolean_t add, const uint8_t *mca)
1245 {
1246 	vsw_t		*vswp = (vsw_t *)arg;
1247 	mcst_addr_t	*mcst_p = NULL;
1248 	uint64_t	addr = 0x0;
1249 	int		i, ret = 0;
1250 
1251 	D1(vswp, "%s: enter", __func__);
1252 
1253 	/*
1254 	 * Convert address into form that can be used
1255 	 * as hash table key.
1256 	 */
1257 	for (i = 0; i < ETHERADDRL; i++) {
1258 		addr = (addr << 8) | mca[i];
1259 	}
1260 
1261 	D2(vswp, "%s: addr = 0x%llx", __func__, addr);
1262 
1263 	if (add) {
1264 		D2(vswp, "%s: adding multicast", __func__);
1265 		if (vsw_add_mcst(vswp, VSW_LOCALDEV, addr, NULL) == 0) {
1266 			/*
1267 			 * Update the list of multicast addresses
1268 			 * contained within the vsw_t structure to
1269 			 * include this new one.
1270 			 */
1271 			mcst_p = kmem_zalloc(sizeof (mcst_addr_t), KM_NOSLEEP);
1272 			if (mcst_p == NULL) {
1273 				DERR(vswp, "%s unable to alloc mem", __func__);
1274 				(void) vsw_del_mcst(vswp,
1275 				    VSW_LOCALDEV, addr, NULL);
1276 				return (1);
1277 			}
1278 			mcst_p->addr = addr;
1279 			ether_copy(mca, &mcst_p->mca);
1280 
1281 			/*
1282 			 * Call into the underlying driver to program the
1283 			 * address into HW.
1284 			 */
1285 			ret = vsw_mac_multicast_add(vswp, NULL, mcst_p,
1286 			    VSW_LOCALDEV);
1287 			if (ret != 0) {
1288 				(void) vsw_del_mcst(vswp,
1289 				    VSW_LOCALDEV, addr, NULL);
1290 				kmem_free(mcst_p, sizeof (*mcst_p));
1291 				return (ret);
1292 			}
1293 
1294 			mutex_enter(&vswp->mca_lock);
1295 			mcst_p->nextp = vswp->mcap;
1296 			vswp->mcap = mcst_p;
1297 			mutex_exit(&vswp->mca_lock);
1298 		} else {
1299 			cmn_err(CE_WARN, "!vsw%d: unable to add multicast "
1300 			    "address", vswp->instance);
1301 		}
1302 		return (ret);
1303 	}
1304 
1305 	D2(vswp, "%s: removing multicast", __func__);
1306 	/*
1307 	 * Remove the address from the hash table..
1308 	 */
1309 	if (vsw_del_mcst(vswp, VSW_LOCALDEV, addr, NULL) == 0) {
1310 
1311 		/*
1312 		 * ..and then from the list maintained in the
1313 		 * vsw_t structure.
1314 		 */
1315 		mcst_p = vsw_del_addr(VSW_LOCALDEV, vswp, addr);
1316 		ASSERT(mcst_p != NULL);
1317 
1318 		vsw_mac_multicast_remove(vswp, NULL, mcst_p, VSW_LOCALDEV);
1319 		kmem_free(mcst_p, sizeof (*mcst_p));
1320 	}
1321 
1322 	D1(vswp, "%s: exit", __func__);
1323 
1324 	return (0);
1325 }
1326 
1327 static int
vsw_m_promisc(void * arg,boolean_t on)1328 vsw_m_promisc(void *arg, boolean_t on)
1329 {
1330 	vsw_t		*vswp = (vsw_t *)arg;
1331 
1332 	D1(vswp, "%s: enter", __func__);
1333 
1334 	WRITE_ENTER(&vswp->if_lockrw);
1335 	if (on)
1336 		vswp->if_state |= VSW_IF_PROMISC;
1337 	else
1338 		vswp->if_state &= ~VSW_IF_PROMISC;
1339 	RW_EXIT(&vswp->if_lockrw);
1340 
1341 	D1(vswp, "%s: exit", __func__);
1342 
1343 	return (0);
1344 }
1345 
1346 static mblk_t *
vsw_m_tx(void * arg,mblk_t * mp)1347 vsw_m_tx(void *arg, mblk_t *mp)
1348 {
1349 	vsw_t		*vswp = (vsw_t *)arg;
1350 
1351 	D1(vswp, "%s: enter", __func__);
1352 
1353 	mp = vsw_vlan_frame_pretag(vswp, VSW_LOCALDEV, mp);
1354 
1355 	if (mp == NULL) {
1356 		return (NULL);
1357 	}
1358 
1359 	vswp->vsw_switch_frame(vswp, mp, VSW_LOCALDEV, NULL, NULL);
1360 
1361 	D1(vswp, "%s: exit", __func__);
1362 
1363 	return (NULL);
1364 }
1365 
1366 /*
1367  * Register for machine description (MD) updates.
1368  *
1369  * Returns 0 on success, 1 on failure.
1370  */
1371 static int
vsw_mdeg_register(vsw_t * vswp)1372 vsw_mdeg_register(vsw_t *vswp)
1373 {
1374 	mdeg_prop_spec_t	*pspecp;
1375 	mdeg_node_spec_t	*inst_specp;
1376 	mdeg_handle_t		mdeg_hdl, mdeg_port_hdl;
1377 	size_t			templatesz;
1378 	int			rv;
1379 
1380 	D1(vswp, "%s: enter", __func__);
1381 
1382 	/*
1383 	 * Allocate and initialize a per-instance copy
1384 	 * of the global property spec array that will
1385 	 * uniquely identify this vsw instance.
1386 	 */
1387 	templatesz = sizeof (vsw_prop_template);
1388 	pspecp = kmem_zalloc(templatesz, KM_SLEEP);
1389 
1390 	bcopy(vsw_prop_template, pspecp, templatesz);
1391 
1392 	VSW_SET_MDEG_PROP_INST(pspecp, vswp->regprop);
1393 
1394 	/* initialize the complete prop spec structure */
1395 	inst_specp = kmem_zalloc(sizeof (mdeg_node_spec_t), KM_SLEEP);
1396 	inst_specp->namep = "virtual-device";
1397 	inst_specp->specp = pspecp;
1398 
1399 	D2(vswp, "%s: instance %d registering with mdeg", __func__,
1400 	    vswp->regprop);
1401 	/*
1402 	 * Register an interest in 'virtual-device' nodes with a
1403 	 * 'name' property of 'virtual-network-switch'
1404 	 */
1405 	rv = mdeg_register(inst_specp, &vdev_match, vsw_mdeg_cb,
1406 	    (void *)vswp, &mdeg_hdl);
1407 	if (rv != MDEG_SUCCESS) {
1408 		DERR(vswp, "%s: mdeg_register failed (%d) for vsw node",
1409 		    __func__, rv);
1410 		goto mdeg_reg_fail;
1411 	}
1412 
1413 	/*
1414 	 * Register an interest in 'vsw-port' nodes.
1415 	 */
1416 	rv = mdeg_register(inst_specp, &vport_match, vsw_port_mdeg_cb,
1417 	    (void *)vswp, &mdeg_port_hdl);
1418 	if (rv != MDEG_SUCCESS) {
1419 		DERR(vswp, "%s: mdeg_register failed (%d)\n", __func__, rv);
1420 		(void) mdeg_unregister(mdeg_hdl);
1421 		goto mdeg_reg_fail;
1422 	}
1423 
1424 	/* save off data that will be needed later */
1425 	vswp->inst_spec = inst_specp;
1426 	vswp->mdeg_hdl = mdeg_hdl;
1427 	vswp->mdeg_port_hdl = mdeg_port_hdl;
1428 
1429 	D1(vswp, "%s: exit", __func__);
1430 	return (0);
1431 
1432 mdeg_reg_fail:
1433 	cmn_err(CE_WARN, "!vsw%d: Unable to register MDEG callbacks",
1434 	    vswp->instance);
1435 	kmem_free(pspecp, templatesz);
1436 	kmem_free(inst_specp, sizeof (mdeg_node_spec_t));
1437 
1438 	vswp->mdeg_hdl = NULL;
1439 	vswp->mdeg_port_hdl = NULL;
1440 
1441 	return (1);
1442 }
1443 
1444 static void
vsw_mdeg_unregister(vsw_t * vswp)1445 vsw_mdeg_unregister(vsw_t *vswp)
1446 {
1447 	D1(vswp, "vsw_mdeg_unregister: enter");
1448 
1449 	if (vswp->mdeg_hdl != NULL)
1450 		(void) mdeg_unregister(vswp->mdeg_hdl);
1451 
1452 	if (vswp->mdeg_port_hdl != NULL)
1453 		(void) mdeg_unregister(vswp->mdeg_port_hdl);
1454 
1455 	if (vswp->inst_spec != NULL) {
1456 		if (vswp->inst_spec->specp != NULL) {
1457 			(void) kmem_free(vswp->inst_spec->specp,
1458 			    sizeof (vsw_prop_template));
1459 			vswp->inst_spec->specp = NULL;
1460 		}
1461 
1462 		(void) kmem_free(vswp->inst_spec, sizeof (mdeg_node_spec_t));
1463 		vswp->inst_spec = NULL;
1464 	}
1465 
1466 	D1(vswp, "vsw_mdeg_unregister: exit");
1467 }
1468 
1469 /*
1470  * Mdeg callback invoked for the vsw node itself.
1471  */
1472 static int
vsw_mdeg_cb(void * cb_argp,mdeg_result_t * resp)1473 vsw_mdeg_cb(void *cb_argp, mdeg_result_t *resp)
1474 {
1475 	vsw_t		*vswp;
1476 	md_t		*mdp;
1477 	mde_cookie_t	node;
1478 	uint64_t	inst;
1479 	char		*node_name = NULL;
1480 
1481 	if (resp == NULL)
1482 		return (MDEG_FAILURE);
1483 
1484 	vswp = (vsw_t *)cb_argp;
1485 
1486 	D1(vswp, "%s: added %d : removed %d : curr matched %d"
1487 	    " : prev matched %d", __func__, resp->added.nelem,
1488 	    resp->removed.nelem, resp->match_curr.nelem,
1489 	    resp->match_prev.nelem);
1490 
1491 	/*
1492 	 * We get an initial callback for this node as 'added'
1493 	 * after registering with mdeg. Note that we would have
1494 	 * already gathered information about this vsw node by
1495 	 * walking MD earlier during attach (in vsw_read_mdprops()).
1496 	 * So, there is a window where the properties of this
1497 	 * node might have changed when we get this initial 'added'
1498 	 * callback. We handle this as if an update occured
1499 	 * and invoke the same function which handles updates to
1500 	 * the properties of this vsw-node if any.
1501 	 *
1502 	 * A non-zero 'match' value indicates that the MD has been
1503 	 * updated and that a virtual-network-switch node is
1504 	 * present which may or may not have been updated. It is
1505 	 * up to the clients to examine their own nodes and
1506 	 * determine if they have changed.
1507 	 */
1508 	if (resp->added.nelem != 0) {
1509 
1510 		if (resp->added.nelem != 1) {
1511 			cmn_err(CE_NOTE, "!vsw%d: number of nodes added "
1512 			    "invalid: %d\n", vswp->instance, resp->added.nelem);
1513 			return (MDEG_FAILURE);
1514 		}
1515 
1516 		mdp = resp->added.mdp;
1517 		node = resp->added.mdep[0];
1518 
1519 	} else if (resp->match_curr.nelem != 0) {
1520 
1521 		if (resp->match_curr.nelem != 1) {
1522 			cmn_err(CE_NOTE, "!vsw%d: number of nodes updated "
1523 			    "invalid: %d\n", vswp->instance,
1524 			    resp->match_curr.nelem);
1525 			return (MDEG_FAILURE);
1526 		}
1527 
1528 		mdp = resp->match_curr.mdp;
1529 		node = resp->match_curr.mdep[0];
1530 
1531 	} else {
1532 		return (MDEG_FAILURE);
1533 	}
1534 
1535 	/* Validate name and instance */
1536 	if (md_get_prop_str(mdp, node, "name", &node_name) != 0) {
1537 		DERR(vswp, "%s: unable to get node name\n",  __func__);
1538 		return (MDEG_FAILURE);
1539 	}
1540 
1541 	/* is this a virtual-network-switch? */
1542 	if (strcmp(node_name, vsw_propname) != 0) {
1543 		DERR(vswp, "%s: Invalid node name: %s\n",
1544 		    __func__, node_name);
1545 		return (MDEG_FAILURE);
1546 	}
1547 
1548 	if (md_get_prop_val(mdp, node, "cfg-handle", &inst)) {
1549 		DERR(vswp, "%s: prop(cfg-handle) not found\n",
1550 		    __func__);
1551 		return (MDEG_FAILURE);
1552 	}
1553 
1554 	/* is this the right instance of vsw? */
1555 	if (inst != vswp->regprop) {
1556 		DERR(vswp, "%s: Invalid cfg-handle: %lx\n",
1557 		    __func__, inst);
1558 		return (MDEG_FAILURE);
1559 	}
1560 
1561 	vsw_update_md_prop(vswp, mdp, node);
1562 
1563 	return (MDEG_SUCCESS);
1564 }
1565 
1566 /*
1567  * Mdeg callback invoked for changes to the vsw-port nodes
1568  * under the vsw node.
1569  */
1570 static int
vsw_port_mdeg_cb(void * cb_argp,mdeg_result_t * resp)1571 vsw_port_mdeg_cb(void *cb_argp, mdeg_result_t *resp)
1572 {
1573 	vsw_t		*vswp;
1574 	int		idx;
1575 	md_t		*mdp;
1576 	mde_cookie_t	node;
1577 	uint64_t	inst;
1578 	int		rv;
1579 
1580 	if ((resp == NULL) || (cb_argp == NULL))
1581 		return (MDEG_FAILURE);
1582 
1583 	vswp = (vsw_t *)cb_argp;
1584 
1585 	D2(vswp, "%s: added %d : removed %d : curr matched %d"
1586 	    " : prev matched %d", __func__, resp->added.nelem,
1587 	    resp->removed.nelem, resp->match_curr.nelem,
1588 	    resp->match_prev.nelem);
1589 
1590 	/* process added ports */
1591 	for (idx = 0; idx < resp->added.nelem; idx++) {
1592 		mdp = resp->added.mdp;
1593 		node = resp->added.mdep[idx];
1594 
1595 		D2(vswp, "%s: adding node(%d) 0x%lx", __func__, idx, node);
1596 
1597 		if ((rv = vsw_port_add(vswp, mdp, &node)) != 0) {
1598 			cmn_err(CE_WARN, "!vsw%d: Unable to add new port "
1599 			    "(0x%lx), err=%d", vswp->instance, node, rv);
1600 		}
1601 	}
1602 
1603 	/* process removed ports */
1604 	for (idx = 0; idx < resp->removed.nelem; idx++) {
1605 		mdp = resp->removed.mdp;
1606 		node = resp->removed.mdep[idx];
1607 
1608 		if (md_get_prop_val(mdp, node, id_propname, &inst)) {
1609 			DERR(vswp, "%s: prop(%s) not found in port(%d)",
1610 			    __func__, id_propname, idx);
1611 			continue;
1612 		}
1613 
1614 		D2(vswp, "%s: removing node(%d) 0x%lx", __func__, idx, node);
1615 
1616 		if (vsw_port_detach(vswp, inst) != 0) {
1617 			cmn_err(CE_WARN, "!vsw%d: Unable to remove port %ld",
1618 			    vswp->instance, inst);
1619 		}
1620 	}
1621 
1622 	for (idx = 0; idx < resp->match_curr.nelem; idx++) {
1623 		(void) vsw_port_update(vswp, resp->match_curr.mdp,
1624 		    resp->match_curr.mdep[idx],
1625 		    resp->match_prev.mdp,
1626 		    resp->match_prev.mdep[idx]);
1627 	}
1628 
1629 	D1(vswp, "%s: exit", __func__);
1630 
1631 	return (MDEG_SUCCESS);
1632 }
1633 
1634 /*
1635  * Scan the machine description for this instance of vsw
1636  * and read its properties. Called only from vsw_attach().
1637  * Returns: 0 on success, 1 on failure.
1638  */
1639 static int
vsw_read_mdprops(vsw_t * vswp)1640 vsw_read_mdprops(vsw_t *vswp)
1641 {
1642 	md_t		*mdp = NULL;
1643 	mde_cookie_t	rootnode;
1644 	mde_cookie_t	*listp = NULL;
1645 	uint64_t	inst;
1646 	uint64_t	cfgh;
1647 	char		*name;
1648 	int		rv = 1;
1649 	int		num_nodes = 0;
1650 	int		num_devs = 0;
1651 	int		listsz = 0;
1652 	int		i;
1653 
1654 	/*
1655 	 * In each 'virtual-device' node in the MD there is a
1656 	 * 'cfg-handle' property which is the MD's concept of
1657 	 * an instance number (this may be completely different from
1658 	 * the device drivers instance #). OBP reads that value and
1659 	 * stores it in the 'reg' property of the appropriate node in
1660 	 * the device tree. We first read this reg property and use this
1661 	 * to compare against the 'cfg-handle' property of vsw nodes
1662 	 * in MD to get to this specific vsw instance and then read
1663 	 * other properties that we are interested in.
1664 	 * We also cache the value of 'reg' property and use it later
1665 	 * to register callbacks with mdeg (see vsw_mdeg_register())
1666 	 */
1667 	inst = ddi_prop_get_int(DDI_DEV_T_ANY, vswp->dip,
1668 	    DDI_PROP_DONTPASS, reg_propname, -1);
1669 	if (inst == -1) {
1670 		cmn_err(CE_NOTE, "!vsw%d: Unable to read %s property from "
1671 		    "OBP device tree", vswp->instance, reg_propname);
1672 		return (rv);
1673 	}
1674 
1675 	vswp->regprop = inst;
1676 
1677 	if ((mdp = md_get_handle()) == NULL) {
1678 		DWARN(vswp, "%s: cannot init MD\n", __func__);
1679 		return (rv);
1680 	}
1681 
1682 	num_nodes = md_node_count(mdp);
1683 	ASSERT(num_nodes > 0);
1684 
1685 	listsz = num_nodes * sizeof (mde_cookie_t);
1686 	listp = (mde_cookie_t *)kmem_zalloc(listsz, KM_SLEEP);
1687 
1688 	rootnode = md_root_node(mdp);
1689 
1690 	/* search for all "virtual_device" nodes */
1691 	num_devs = md_scan_dag(mdp, rootnode,
1692 	    md_find_name(mdp, vdev_propname),
1693 	    md_find_name(mdp, "fwd"), listp);
1694 	if (num_devs <= 0) {
1695 		DWARN(vswp, "%s: invalid num_devs:%d\n", __func__, num_devs);
1696 		goto vsw_readmd_exit;
1697 	}
1698 
1699 	/*
1700 	 * Now loop through the list of virtual-devices looking for
1701 	 * devices with name "virtual-network-switch" and for each
1702 	 * such device compare its instance with what we have from
1703 	 * the 'reg' property to find the right node in MD and then
1704 	 * read all its properties.
1705 	 */
1706 	for (i = 0; i < num_devs; i++) {
1707 
1708 		if (md_get_prop_str(mdp, listp[i], "name", &name) != 0) {
1709 			DWARN(vswp, "%s: name property not found\n",
1710 			    __func__);
1711 			goto vsw_readmd_exit;
1712 		}
1713 
1714 		/* is this a virtual-network-switch? */
1715 		if (strcmp(name, vsw_propname) != 0)
1716 			continue;
1717 
1718 		if (md_get_prop_val(mdp, listp[i], "cfg-handle", &cfgh) != 0) {
1719 			DWARN(vswp, "%s: cfg-handle property not found\n",
1720 			    __func__);
1721 			goto vsw_readmd_exit;
1722 		}
1723 
1724 		/* is this the required instance of vsw? */
1725 		if (inst != cfgh)
1726 			continue;
1727 
1728 		/* now read all properties of this vsw instance */
1729 		rv = vsw_get_initial_md_properties(vswp, mdp, listp[i]);
1730 		break;
1731 	}
1732 
1733 vsw_readmd_exit:
1734 
1735 	kmem_free(listp, listsz);
1736 	(void) md_fini_handle(mdp);
1737 	return (rv);
1738 }
1739 
1740 /*
1741  * Read the initial start-of-day values from the specified MD node.
1742  */
1743 static int
vsw_get_initial_md_properties(vsw_t * vswp,md_t * mdp,mde_cookie_t node)1744 vsw_get_initial_md_properties(vsw_t *vswp, md_t *mdp, mde_cookie_t node)
1745 {
1746 	uint64_t	macaddr = 0;
1747 
1748 	D1(vswp, "%s: enter", __func__);
1749 
1750 	if (vsw_get_md_physname(vswp, mdp, node, vswp->physname) != 0) {
1751 		return (1);
1752 	}
1753 
1754 	/* mac address for vswitch device itself */
1755 	if (md_get_prop_val(mdp, node, macaddr_propname, &macaddr) != 0) {
1756 		cmn_err(CE_WARN, "!vsw%d: Unable to get MAC address from MD",
1757 		    vswp->instance);
1758 		return (1);
1759 	}
1760 
1761 	vsw_save_lmacaddr(vswp, macaddr);
1762 
1763 	if (vsw_get_md_smodes(vswp, mdp, node, &vswp->smode)) {
1764 		DWARN(vswp, "%s: Unable to read %s property from MD, "
1765 		    "defaulting to 'switched' mode",
1766 		    __func__, smode_propname);
1767 
1768 		vswp->smode = VSW_LAYER2;
1769 	}
1770 
1771 	/*
1772 	 * Read the 'linkprop' property to know if this
1773 	 * vsw device wants to get physical link updates.
1774 	 */
1775 	vsw_linkprop_read(vswp, mdp, node, &vswp->pls_update);
1776 
1777 	/* read mtu */
1778 	vsw_mtu_read(vswp, mdp, node, &vswp->mtu);
1779 	if (vswp->mtu < ETHERMTU || vswp->mtu > VNET_MAX_MTU) {
1780 		vswp->mtu = ETHERMTU;
1781 	}
1782 	vswp->max_frame_size = vswp->mtu + sizeof (struct ether_header) +
1783 	    VLAN_TAGSZ;
1784 
1785 	/* read vlan id properties of this vsw instance */
1786 	vsw_vlan_read_ids(vswp, VSW_LOCALDEV, mdp, node, &vswp->pvid,
1787 	    &vswp->vids, &vswp->nvids, &vswp->default_vlan_id);
1788 
1789 	/* read priority-ether-types */
1790 	vsw_read_pri_eth_types(vswp, mdp, node);
1791 
1792 	/* read bandwidth property of this vsw instance */
1793 	vsw_bandwidth_read(vswp, mdp, node, &vswp->bandwidth);
1794 
1795 	D1(vswp, "%s: exit", __func__);
1796 	return (0);
1797 }
1798 
1799 /*
1800  * Read vlan id properties of the given MD node.
1801  * Arguments:
1802  *   arg:          device argument(vsw device or a port)
1803  *   type:         type of arg; VSW_LOCALDEV(vsw device) or VSW_VNETPORT(port)
1804  *   mdp:          machine description
1805  *   node:         md node cookie
1806  *
1807  * Returns:
1808  *   pvidp:        port-vlan-id of the node
1809  *   vidspp:       list of vlan-ids of the node
1810  *   nvidsp:       # of vlan-ids in the list
1811  *   default_idp:  default-vlan-id of the node(if node is vsw device)
1812  */
1813 static void
vsw_vlan_read_ids(void * arg,int type,md_t * mdp,mde_cookie_t node,uint16_t * pvidp,vsw_vlanid_t ** vidspp,uint16_t * nvidsp,uint16_t * default_idp)1814 vsw_vlan_read_ids(void *arg, int type, md_t *mdp, mde_cookie_t node,
1815 	uint16_t *pvidp, vsw_vlanid_t **vidspp, uint16_t *nvidsp,
1816 	uint16_t *default_idp)
1817 {
1818 	vsw_t		*vswp;
1819 	vsw_port_t	*portp;
1820 	char		*pvid_propname;
1821 	char		*vid_propname;
1822 	uint_t		nvids = 0;
1823 	uint32_t	vids_size;
1824 	int		rv;
1825 	int		i;
1826 	uint64_t	*data;
1827 	uint64_t	val;
1828 	int		size;
1829 	int		inst;
1830 
1831 	if (type == VSW_LOCALDEV) {
1832 
1833 		vswp = (vsw_t *)arg;
1834 		pvid_propname = vsw_pvid_propname;
1835 		vid_propname = vsw_vid_propname;
1836 		inst = vswp->instance;
1837 
1838 	} else if (type == VSW_VNETPORT) {
1839 
1840 		portp = (vsw_port_t *)arg;
1841 		vswp = portp->p_vswp;
1842 		pvid_propname = port_pvid_propname;
1843 		vid_propname = port_vid_propname;
1844 		inst = portp->p_instance;
1845 
1846 	} else {
1847 		return;
1848 	}
1849 
1850 	if (type == VSW_LOCALDEV && default_idp != NULL) {
1851 		rv = md_get_prop_val(mdp, node, vsw_dvid_propname, &val);
1852 		if (rv != 0) {
1853 			DWARN(vswp, "%s: prop(%s) not found", __func__,
1854 			    vsw_dvid_propname);
1855 
1856 			*default_idp = vsw_default_vlan_id;
1857 		} else {
1858 			*default_idp = val & 0xFFF;
1859 			D2(vswp, "%s: %s(%d): (%d)\n", __func__,
1860 			    vsw_dvid_propname, inst, *default_idp);
1861 		}
1862 	}
1863 
1864 	rv = md_get_prop_val(mdp, node, pvid_propname, &val);
1865 	if (rv != 0) {
1866 		DWARN(vswp, "%s: prop(%s) not found", __func__, pvid_propname);
1867 		*pvidp = vsw_default_vlan_id;
1868 	} else {
1869 
1870 		*pvidp = val & 0xFFF;
1871 		D2(vswp, "%s: %s(%d): (%d)\n", __func__,
1872 		    pvid_propname, inst, *pvidp);
1873 	}
1874 
1875 	rv = md_get_prop_data(mdp, node, vid_propname, (uint8_t **)&data,
1876 	    &size);
1877 	if (rv != 0) {
1878 		D2(vswp, "%s: prop(%s) not found", __func__, vid_propname);
1879 		size = 0;
1880 	} else {
1881 		size /= sizeof (uint64_t);
1882 	}
1883 	nvids = size;
1884 
1885 	if (nvids != 0) {
1886 		D2(vswp, "%s: %s(%d): ", __func__, vid_propname, inst);
1887 		vids_size = sizeof (vsw_vlanid_t) * nvids;
1888 		*vidspp = kmem_zalloc(vids_size, KM_SLEEP);
1889 		for (i = 0; i < nvids; i++) {
1890 			(*vidspp)[i].vl_vid = data[i] & 0xFFFF;
1891 			(*vidspp)[i].vl_set = B_FALSE;
1892 			D2(vswp, " %d ", (*vidspp)[i].vl_vid);
1893 		}
1894 		D2(vswp, "\n");
1895 	}
1896 
1897 	*nvidsp = nvids;
1898 }
1899 
1900 static void
vsw_port_read_bandwidth(vsw_port_t * portp,md_t * mdp,mde_cookie_t node,uint64_t * bw)1901 vsw_port_read_bandwidth(vsw_port_t *portp, md_t *mdp, mde_cookie_t node,
1902     uint64_t *bw)
1903 {
1904 	int		rv;
1905 	uint64_t	val;
1906 	vsw_t		*vswp;
1907 
1908 	vswp = portp->p_vswp;
1909 
1910 	rv = md_get_prop_val(mdp, node, port_maxbw_propname, &val);
1911 
1912 	if (rv != 0) {
1913 		*bw = 0;
1914 		D3(vswp, "%s: prop(%s) not found\n", __func__,
1915 		    port_maxbw_propname);
1916 	} else {
1917 		*bw = val;
1918 		D3(vswp, "%s: %s nodes found", __func__, port_maxbw_propname);
1919 	}
1920 }
1921 
1922 /*
1923  * This function reads "priority-ether-types" property from md. This property
1924  * is used to enable support for priority frames. Applications which need
1925  * guaranteed and timely delivery of certain high priority frames to/from
1926  * a vnet or vsw within ldoms, should configure this property by providing
1927  * the ether type(s) for which the priority facility is needed.
1928  * Normal data frames are delivered over a ldc channel using the descriptor
1929  * ring mechanism which is constrained by factors such as descriptor ring size,
1930  * the rate at which the ring is processed at the peer ldc end point, etc.
1931  * The priority mechanism provides an Out-Of-Band path to send/receive frames
1932  * as raw pkt data (VIO_PKT_DATA) messages over the channel, avoiding the
1933  * descriptor ring path and enables a more reliable and timely delivery of
1934  * frames to the peer.
1935  */
1936 static void
vsw_read_pri_eth_types(vsw_t * vswp,md_t * mdp,mde_cookie_t node)1937 vsw_read_pri_eth_types(vsw_t *vswp, md_t *mdp, mde_cookie_t node)
1938 {
1939 	int		rv;
1940 	uint16_t	*types;
1941 	uint64_t	*data;
1942 	int		size;
1943 	int		i;
1944 	size_t		mblk_sz;
1945 
1946 	rv = md_get_prop_data(mdp, node, pri_types_propname,
1947 	    (uint8_t **)&data, &size);
1948 	if (rv != 0) {
1949 		/*
1950 		 * Property may not exist if we are running pre-ldoms1.1 f/w.
1951 		 * Check if 'vsw_pri_eth_type' has been set in that case.
1952 		 */
1953 		if (vsw_pri_eth_type != 0) {
1954 			size = sizeof (vsw_pri_eth_type);
1955 			data = &vsw_pri_eth_type;
1956 		} else {
1957 			D3(vswp, "%s: prop(%s) not found", __func__,
1958 			    pri_types_propname);
1959 			size = 0;
1960 		}
1961 	}
1962 
1963 	if (size == 0) {
1964 		vswp->pri_num_types = 0;
1965 		return;
1966 	}
1967 
1968 	/*
1969 	 * we have some priority-ether-types defined;
1970 	 * allocate a table of these types and also
1971 	 * allocate a pool of mblks to transmit these
1972 	 * priority packets.
1973 	 */
1974 	size /= sizeof (uint64_t);
1975 	vswp->pri_num_types = size;
1976 	vswp->pri_types = kmem_zalloc(size * sizeof (uint16_t), KM_SLEEP);
1977 	for (i = 0, types = vswp->pri_types; i < size; i++) {
1978 		types[i] = data[i] & 0xFFFF;
1979 	}
1980 	mblk_sz = (VIO_PKT_DATA_HDRSIZE + ETHERMAX + 7) & ~7;
1981 	(void) vio_create_mblks(vsw_pri_tx_nmblks, mblk_sz, NULL,
1982 	    &vswp->pri_tx_vmp);
1983 }
1984 
1985 static void
vsw_mtu_read(vsw_t * vswp,md_t * mdp,mde_cookie_t node,uint32_t * mtu)1986 vsw_mtu_read(vsw_t *vswp, md_t *mdp, mde_cookie_t node, uint32_t *mtu)
1987 {
1988 	int		rv;
1989 	int		inst;
1990 	uint64_t	val;
1991 	char		*mtu_propname;
1992 
1993 	mtu_propname = vsw_mtu_propname;
1994 	inst = vswp->instance;
1995 
1996 	rv = md_get_prop_val(mdp, node, mtu_propname, &val);
1997 	if (rv != 0) {
1998 		D3(vswp, "%s: prop(%s) not found", __func__, mtu_propname);
1999 		*mtu = vsw_ethermtu;
2000 	} else {
2001 
2002 		*mtu = val & 0xFFFF;
2003 		D2(vswp, "%s: %s(%d): (%d)\n", __func__,
2004 		    mtu_propname, inst, *mtu);
2005 	}
2006 }
2007 
2008 /*
2009  * Update the mtu of the vsw device. We first check if the device has been
2010  * plumbed and if so fail the mtu update. Otherwise, we continue to update the
2011  * new mtu and reset all ports to initiate handshake re-negotiation with peers
2012  * using the new mtu.
2013  */
2014 static int
vsw_mtu_update(vsw_t * vswp,uint32_t mtu)2015 vsw_mtu_update(vsw_t *vswp, uint32_t mtu)
2016 {
2017 	int	rv;
2018 
2019 	WRITE_ENTER(&vswp->if_lockrw);
2020 
2021 	if (vswp->if_state & VSW_IF_UP) {
2022 
2023 		RW_EXIT(&vswp->if_lockrw);
2024 
2025 		cmn_err(CE_NOTE, "!vsw%d: Unable to process mtu update"
2026 		    " as the device is plumbed\n", vswp->instance);
2027 		return (EBUSY);
2028 
2029 	} else {
2030 
2031 		D2(vswp, "%s: curr_mtu(%d) new_mtu(%d)\n",
2032 		    __func__, vswp->mtu, mtu);
2033 
2034 		vswp->mtu = mtu;
2035 		vswp->max_frame_size = vswp->mtu +
2036 		    sizeof (struct ether_header) + VLAN_TAGSZ;
2037 
2038 		rv = mac_maxsdu_update(vswp->if_mh, mtu);
2039 		if (rv != 0) {
2040 			cmn_err(CE_NOTE,
2041 			    "!vsw%d: Unable to update mtu with mac"
2042 			    " layer\n", vswp->instance);
2043 		}
2044 
2045 		RW_EXIT(&vswp->if_lockrw);
2046 
2047 		/* Reset ports to renegotiate with the new mtu */
2048 		vsw_reset_ports(vswp);
2049 
2050 	}
2051 
2052 	return (0);
2053 }
2054 
2055 static void
vsw_linkprop_read(vsw_t * vswp,md_t * mdp,mde_cookie_t node,boolean_t * pls)2056 vsw_linkprop_read(vsw_t *vswp, md_t *mdp, mde_cookie_t node,
2057 	boolean_t *pls)
2058 {
2059 	int		rv;
2060 	uint64_t	val;
2061 	char		*linkpropname;
2062 
2063 	linkpropname = vsw_linkprop_propname;
2064 
2065 	rv = md_get_prop_val(mdp, node, linkpropname, &val);
2066 	if (rv != 0) {
2067 		D3(vswp, "%s: prop(%s) not found", __func__, linkpropname);
2068 		*pls = B_FALSE;
2069 	} else {
2070 
2071 		*pls = (val & 0x1) ? B_TRUE : B_FALSE;
2072 		D2(vswp, "%s: %s(%d): (%d)\n", __func__, linkpropname,
2073 		    vswp->instance, *pls);
2074 	}
2075 }
2076 
2077 void
vsw_mac_link_update(vsw_t * vswp,link_state_t link_state)2078 vsw_mac_link_update(vsw_t *vswp, link_state_t link_state)
2079 {
2080 	READ_ENTER(&vswp->if_lockrw);
2081 
2082 	if (vswp->if_state & VSW_IF_REG) {
2083 		mac_link_update(vswp->if_mh, link_state);
2084 	}
2085 
2086 	RW_EXIT(&vswp->if_lockrw);
2087 }
2088 
2089 void
vsw_physlink_state_update(vsw_t * vswp)2090 vsw_physlink_state_update(vsw_t *vswp)
2091 {
2092 	if (vswp->pls_update == B_TRUE) {
2093 		vsw_mac_link_update(vswp, vswp->phys_link_state);
2094 	}
2095 	vsw_physlink_update_ports(vswp);
2096 }
2097 
2098 static void
vsw_bandwidth_read(vsw_t * vswp,md_t * mdp,mde_cookie_t node,uint64_t * bw)2099 vsw_bandwidth_read(vsw_t *vswp, md_t *mdp, mde_cookie_t node, uint64_t *bw)
2100 {
2101 	/* read the vsw bandwidth from md */
2102 	int		rv;
2103 	uint64_t	val;
2104 
2105 	rv = md_get_prop_val(mdp, node, vsw_maxbw_propname, &val);
2106 	if (rv != 0) {
2107 		*bw = 0;
2108 		D3(vswp, "%s: prop(%s) not found", __func__,
2109 		    vsw_maxbw_propname);
2110 	} else {
2111 		*bw = val;
2112 		D3(vswp, "%s: %s(%d): (%ld)\n", __func__,
2113 		    vsw_maxbw_propname, vswp->instance, *bw);
2114 	}
2115 }
2116 
2117 /*
2118  * Check to see if the relevant properties in the specified node have
2119  * changed, and if so take the appropriate action.
2120  *
2121  * If any of the properties are missing or invalid we don't take
2122  * any action, as this function should only be invoked when modifications
2123  * have been made to what we assume is a working configuration, which
2124  * we leave active.
2125  *
2126  * Note it is legal for this routine to be invoked even if none of the
2127  * properties in the port node within the MD have actually changed.
2128  */
2129 static void
vsw_update_md_prop(vsw_t * vswp,md_t * mdp,mde_cookie_t node)2130 vsw_update_md_prop(vsw_t *vswp, md_t *mdp, mde_cookie_t node)
2131 {
2132 	char		physname[LIFNAMSIZ];
2133 	char		drv[LIFNAMSIZ];
2134 	uint_t		ddi_instance;
2135 	uint8_t		new_smode;
2136 	int		i;
2137 	uint64_t 	macaddr = 0;
2138 	enum		{MD_init = 0x1,
2139 				MD_physname = 0x2,
2140 				MD_macaddr = 0x4,
2141 				MD_smode = 0x8,
2142 				MD_vlans = 0x10,
2143 				MD_mtu = 0x20,
2144 				MD_pls = 0x40,
2145 				MD_bw = 0x80} updated;
2146 	int		rv;
2147 	uint16_t	pvid;
2148 	vsw_vlanid_t	*vids;
2149 	uint16_t	nvids;
2150 	uint32_t	mtu;
2151 	boolean_t	pls_update;
2152 	uint64_t	maxbw;
2153 
2154 	updated = MD_init;
2155 
2156 	D1(vswp, "%s: enter", __func__);
2157 
2158 	/*
2159 	 * Check if name of physical device in MD has changed.
2160 	 */
2161 	if (vsw_get_md_physname(vswp, mdp, node, (char *)&physname) == 0) {
2162 		/*
2163 		 * Do basic sanity check on new device name/instance,
2164 		 * if its non NULL. It is valid for the device name to
2165 		 * have changed from a non NULL to a NULL value, i.e.
2166 		 * the vsw is being changed to 'routed' mode.
2167 		 */
2168 		if ((strlen(physname) != 0) &&
2169 		    (ddi_parse(physname, drv,
2170 		    &ddi_instance) != DDI_SUCCESS)) {
2171 			cmn_err(CE_WARN, "!vsw%d: physical device %s is not"
2172 			    " a valid device name/instance",
2173 			    vswp->instance, physname);
2174 			goto fail_reconf;
2175 		}
2176 
2177 		if (strcmp(physname, vswp->physname)) {
2178 			D2(vswp, "%s: device name changed from %s to %s",
2179 			    __func__, vswp->physname, physname);
2180 
2181 			updated |= MD_physname;
2182 		} else {
2183 			D2(vswp, "%s: device name unchanged at %s",
2184 			    __func__, vswp->physname);
2185 		}
2186 	} else {
2187 		cmn_err(CE_WARN, "!vsw%d: Unable to read name of physical "
2188 		    "device from updated MD.", vswp->instance);
2189 		goto fail_reconf;
2190 	}
2191 
2192 	/*
2193 	 * Check if MAC address has changed.
2194 	 */
2195 	if (md_get_prop_val(mdp, node, macaddr_propname, &macaddr) != 0) {
2196 		cmn_err(CE_WARN, "!vsw%d: Unable to get MAC address from MD",
2197 		    vswp->instance);
2198 		goto fail_reconf;
2199 	} else {
2200 		uint64_t maddr = macaddr;
2201 		READ_ENTER(&vswp->if_lockrw);
2202 		for (i = ETHERADDRL - 1; i >= 0; i--) {
2203 			if (vswp->if_addr.ether_addr_octet[i]
2204 			    != (macaddr & 0xFF)) {
2205 				D2(vswp, "%s: octet[%d] 0x%x != 0x%x",
2206 				    __func__, i,
2207 				    vswp->if_addr.ether_addr_octet[i],
2208 				    (macaddr & 0xFF));
2209 				updated |= MD_macaddr;
2210 				macaddr = maddr;
2211 				break;
2212 			}
2213 			macaddr >>= 8;
2214 		}
2215 		RW_EXIT(&vswp->if_lockrw);
2216 		if (updated & MD_macaddr) {
2217 			vsw_save_lmacaddr(vswp, macaddr);
2218 		}
2219 	}
2220 
2221 	/*
2222 	 * Check if switching modes have changed.
2223 	 */
2224 	if (vsw_get_md_smodes(vswp, mdp, node, &new_smode)) {
2225 		cmn_err(CE_WARN, "!vsw%d: Unable to read %s property from MD",
2226 		    vswp->instance, smode_propname);
2227 		goto fail_reconf;
2228 	} else {
2229 		if (new_smode != vswp->smode) {
2230 			D2(vswp, "%s: switching mode changed from %d to %d",
2231 			    __func__, vswp->smode, new_smode);
2232 
2233 			updated |= MD_smode;
2234 		}
2235 	}
2236 
2237 	/* Read the vlan ids */
2238 	vsw_vlan_read_ids(vswp, VSW_LOCALDEV, mdp, node, &pvid, &vids,
2239 	    &nvids, NULL);
2240 
2241 	/* Determine if there are any vlan id updates */
2242 	if ((pvid != vswp->pvid) ||		/* pvid changed? */
2243 	    (nvids != vswp->nvids) ||		/* # of vids changed? */
2244 	    ((nvids != 0) && (vswp->nvids != 0) &&	/* vids changed? */
2245 	    !vsw_cmp_vids(vids, vswp->vids, nvids))) {
2246 		updated |= MD_vlans;
2247 	}
2248 
2249 	/* Read mtu */
2250 	vsw_mtu_read(vswp, mdp, node, &mtu);
2251 	if (mtu != vswp->mtu) {
2252 		if (mtu >= ETHERMTU && mtu <= VNET_MAX_MTU) {
2253 			updated |= MD_mtu;
2254 		} else {
2255 			cmn_err(CE_NOTE, "!vsw%d: Unable to process mtu update"
2256 			    " as the specified value:%d is invalid\n",
2257 			    vswp->instance, mtu);
2258 		}
2259 	}
2260 
2261 	/*
2262 	 * Read the 'linkprop' property.
2263 	 */
2264 	vsw_linkprop_read(vswp, mdp, node, &pls_update);
2265 	if (pls_update != vswp->pls_update) {
2266 		updated |= MD_pls;
2267 	}
2268 
2269 	/* Read bandwidth */
2270 	vsw_bandwidth_read(vswp, mdp, node, &maxbw);
2271 	if (maxbw != vswp->bandwidth) {
2272 		if (maxbw >= MRP_MAXBW_MINVAL || maxbw == 0) {
2273 			updated |= MD_bw;
2274 		} else {
2275 			cmn_err(CE_NOTE, "!vsw%d: Unable to process bandwidth"
2276 			    " update as the specified value:%ld is invalid\n",
2277 			    vswp->instance, maxbw);
2278 		}
2279 	}
2280 
2281 	/*
2282 	 * Now make any changes which are needed...
2283 	 */
2284 	if (updated & MD_pls) {
2285 
2286 		/* save the updated property. */
2287 		vswp->pls_update = pls_update;
2288 
2289 		if (pls_update == B_FALSE) {
2290 			/*
2291 			 * Phys link state update is now disabled for this vsw
2292 			 * interface. If we had previously reported a link-down
2293 			 * to the stack, undo that by sending a link-up.
2294 			 */
2295 			if (vswp->phys_link_state == LINK_STATE_DOWN) {
2296 				vsw_mac_link_update(vswp, LINK_STATE_UP);
2297 			}
2298 		} else {
2299 			/*
2300 			 * Phys link state update is now enabled. Send up an
2301 			 * update based on the current phys link state.
2302 			 */
2303 			if (vswp->smode & VSW_LAYER2) {
2304 				vsw_mac_link_update(vswp,
2305 				    vswp->phys_link_state);
2306 			}
2307 		}
2308 
2309 	}
2310 
2311 	if (updated & (MD_physname | MD_smode | MD_mtu)) {
2312 
2313 		/*
2314 		 * Stop any pending thread to setup switching mode.
2315 		 */
2316 		vsw_setup_switching_stop(vswp);
2317 
2318 		/* Cleanup HybridIO */
2319 		vsw_hio_cleanup(vswp);
2320 
2321 		/*
2322 		 * Remove unicst, mcst addrs of vsw interface
2323 		 * and ports from the physdev. This also closes
2324 		 * the corresponding mac clients.
2325 		 */
2326 		vsw_unset_addrs(vswp);
2327 
2328 		/*
2329 		 * Stop, detach and close the old device..
2330 		 */
2331 		mutex_enter(&vswp->mac_lock);
2332 		vsw_mac_close(vswp);
2333 		mutex_exit(&vswp->mac_lock);
2334 
2335 		/*
2336 		 * Update phys name.
2337 		 */
2338 		if (updated & MD_physname) {
2339 			cmn_err(CE_NOTE, "!vsw%d: changing from %s to %s",
2340 			    vswp->instance, vswp->physname, physname);
2341 			(void) strncpy(vswp->physname,
2342 			    physname, strlen(physname) + 1);
2343 		}
2344 
2345 		/*
2346 		 * Update array with the new switch mode values.
2347 		 */
2348 		if (updated & MD_smode) {
2349 			vswp->smode = new_smode;
2350 		}
2351 
2352 		/* Update mtu */
2353 		if (updated & MD_mtu) {
2354 			rv = vsw_mtu_update(vswp, mtu);
2355 			if (rv != 0) {
2356 				goto fail_update;
2357 			}
2358 		}
2359 
2360 		/*
2361 		 * ..and attach, start the new device.
2362 		 */
2363 		rv = vsw_setup_switching(vswp);
2364 		if (rv == EAGAIN) {
2365 			/*
2366 			 * Unable to setup switching mode.
2367 			 * As the error is EAGAIN, schedule a thread to retry
2368 			 * and return. Programming addresses of ports and
2369 			 * vsw interface will be done by the thread when the
2370 			 * switching setup completes successfully.
2371 			 */
2372 			if (vsw_setup_switching_start(vswp) != 0) {
2373 				goto fail_update;
2374 			}
2375 			return;
2376 
2377 		} else if (rv) {
2378 			goto fail_update;
2379 		}
2380 
2381 		vsw_setup_switching_post_process(vswp);
2382 	} else if (updated & MD_macaddr) {
2383 		/*
2384 		 * We enter here if only MD_macaddr is exclusively updated.
2385 		 * If MD_physname and/or MD_smode are also updated, then
2386 		 * as part of that, we would have implicitly processed
2387 		 * MD_macaddr update (above).
2388 		 */
2389 		cmn_err(CE_NOTE, "!vsw%d: changing mac address to 0x%lx",
2390 		    vswp->instance, macaddr);
2391 
2392 		READ_ENTER(&vswp->if_lockrw);
2393 		if (vswp->if_state & VSW_IF_UP) {
2394 			/* reconfigure with new address */
2395 			vsw_if_mac_reconfig(vswp, B_FALSE, 0, NULL, 0);
2396 
2397 			/*
2398 			 * Notify the MAC layer of the changed address.
2399 			 */
2400 			mac_unicst_update(vswp->if_mh,
2401 			    (uint8_t *)&vswp->if_addr);
2402 
2403 		}
2404 		RW_EXIT(&vswp->if_lockrw);
2405 
2406 	}
2407 
2408 	if (updated & MD_vlans) {
2409 		/* Remove existing vlan ids from the hash table. */
2410 		vsw_vlan_remove_ids(vswp, VSW_LOCALDEV);
2411 
2412 		if (vswp->if_state & VSW_IF_UP) {
2413 			vsw_if_mac_reconfig(vswp, B_TRUE, pvid, vids, nvids);
2414 		} else {
2415 			if (vswp->nvids != 0) {
2416 				kmem_free(vswp->vids,
2417 				    sizeof (vsw_vlanid_t) * vswp->nvids);
2418 			}
2419 			vswp->vids = vids;
2420 			vswp->nvids = nvids;
2421 			vswp->pvid = pvid;
2422 		}
2423 
2424 		/* add these new vlan ids into hash table */
2425 		vsw_vlan_add_ids(vswp, VSW_LOCALDEV);
2426 	} else {
2427 		if (nvids != 0) {
2428 			kmem_free(vids, sizeof (vsw_vlanid_t) * nvids);
2429 		}
2430 	}
2431 
2432 	if (updated & MD_bw) {
2433 		vsw_update_bandwidth(vswp, NULL, VSW_LOCALDEV, maxbw);
2434 	}
2435 
2436 	return;
2437 
2438 fail_reconf:
2439 	cmn_err(CE_WARN, "!vsw%d: configuration unchanged", vswp->instance);
2440 	return;
2441 
2442 fail_update:
2443 	cmn_err(CE_WARN, "!vsw%d: re-configuration failed",
2444 	    vswp->instance);
2445 }
2446 
2447 /*
2448  * Read the port's md properties.
2449  */
2450 static int
vsw_port_read_props(vsw_port_t * portp,vsw_t * vswp,md_t * mdp,mde_cookie_t * node)2451 vsw_port_read_props(vsw_port_t *portp, vsw_t *vswp,
2452 	md_t *mdp, mde_cookie_t *node)
2453 {
2454 	uint64_t		ldc_id;
2455 	uint8_t			*addrp;
2456 	int			i, addrsz;
2457 	int			num_nodes = 0, nchan = 0;
2458 	int			listsz = 0;
2459 	mde_cookie_t		*listp = NULL;
2460 	struct ether_addr	ea;
2461 	uint64_t		macaddr;
2462 	uint64_t		inst = 0;
2463 	uint64_t		val;
2464 
2465 	if (md_get_prop_val(mdp, *node, id_propname, &inst)) {
2466 		DWARN(vswp, "%s: prop(%s) not found", __func__,
2467 		    id_propname);
2468 		return (1);
2469 	}
2470 
2471 	/*
2472 	 * Find the channel endpoint node(s) (which should be under this
2473 	 * port node) which contain the channel id(s).
2474 	 */
2475 	if ((num_nodes = md_node_count(mdp)) <= 0) {
2476 		DERR(vswp, "%s: invalid number of nodes found (%d)",
2477 		    __func__, num_nodes);
2478 		return (1);
2479 	}
2480 
2481 	D2(vswp, "%s: %d nodes found", __func__, num_nodes);
2482 
2483 	/* allocate enough space for node list */
2484 	listsz = num_nodes * sizeof (mde_cookie_t);
2485 	listp = kmem_zalloc(listsz, KM_SLEEP);
2486 
2487 	nchan = md_scan_dag(mdp, *node, md_find_name(mdp, chan_propname),
2488 	    md_find_name(mdp, "fwd"), listp);
2489 
2490 	if (nchan <= 0) {
2491 		DWARN(vswp, "%s: no %s nodes found", __func__, chan_propname);
2492 		kmem_free(listp, listsz);
2493 		return (1);
2494 	}
2495 
2496 	D2(vswp, "%s: %d %s nodes found", __func__, nchan, chan_propname);
2497 
2498 	/* use property from first node found */
2499 	if (md_get_prop_val(mdp, listp[0], id_propname, &ldc_id)) {
2500 		DWARN(vswp, "%s: prop(%s) not found\n", __func__,
2501 		    id_propname);
2502 		kmem_free(listp, listsz);
2503 		return (1);
2504 	}
2505 
2506 	/* don't need list any more */
2507 	kmem_free(listp, listsz);
2508 
2509 	D2(vswp, "%s: ldc_id 0x%llx", __func__, ldc_id);
2510 
2511 	/* read mac-address property */
2512 	if (md_get_prop_data(mdp, *node, remaddr_propname,
2513 	    &addrp, &addrsz)) {
2514 		DWARN(vswp, "%s: prop(%s) not found",
2515 		    __func__, remaddr_propname);
2516 		return (1);
2517 	}
2518 
2519 	if (addrsz < ETHERADDRL) {
2520 		DWARN(vswp, "%s: invalid address size", __func__);
2521 		return (1);
2522 	}
2523 
2524 	macaddr = *((uint64_t *)addrp);
2525 	D2(vswp, "%s: remote mac address 0x%llx", __func__, macaddr);
2526 
2527 	for (i = ETHERADDRL - 1; i >= 0; i--) {
2528 		ea.ether_addr_octet[i] = macaddr & 0xFF;
2529 		macaddr >>= 8;
2530 	}
2531 
2532 	/* now update all properties into the port */
2533 	portp->p_vswp = vswp;
2534 	portp->p_instance = inst;
2535 	portp->addr_set = B_FALSE;
2536 	ether_copy(&ea, &portp->p_macaddr);
2537 	if (nchan > VSW_PORT_MAX_LDCS) {
2538 		D2(vswp, "%s: using first of %d ldc ids",
2539 		    __func__, nchan);
2540 		nchan = VSW_PORT_MAX_LDCS;
2541 	}
2542 	portp->num_ldcs = nchan;
2543 	portp->ldc_ids =
2544 	    kmem_zalloc(sizeof (uint64_t) * nchan, KM_SLEEP);
2545 	bcopy(&ldc_id, (portp->ldc_ids), sizeof (uint64_t) * nchan);
2546 
2547 	/* read vlan id properties of this port node */
2548 	vsw_vlan_read_ids(portp, VSW_VNETPORT, mdp, *node, &portp->pvid,
2549 	    &portp->vids, &portp->nvids, NULL);
2550 
2551 	/* Check if hybrid property is present */
2552 	if (md_get_prop_val(mdp, *node, hybrid_propname, &val) == 0) {
2553 		D1(vswp, "%s: prop(%s) found\n", __func__, hybrid_propname);
2554 		portp->p_hio_enabled = B_TRUE;
2555 	} else {
2556 		portp->p_hio_enabled = B_FALSE;
2557 	}
2558 	/*
2559 	 * Port hio capability determined after version
2560 	 * negotiation, i.e., when we know the peer is HybridIO capable.
2561 	 */
2562 	portp->p_hio_capable = B_FALSE;
2563 
2564 	/* Read bandwidth of this port */
2565 	vsw_port_read_bandwidth(portp, mdp, *node, &portp->p_bandwidth);
2566 
2567 	return (0);
2568 }
2569 
2570 /*
2571  * Add a new port to the system.
2572  *
2573  * Returns 0 on success, 1 on failure.
2574  */
2575 int
vsw_port_add(vsw_t * vswp,md_t * mdp,mde_cookie_t * node)2576 vsw_port_add(vsw_t *vswp, md_t *mdp, mde_cookie_t *node)
2577 {
2578 	vsw_port_t	*portp;
2579 	int		rv;
2580 
2581 	portp = kmem_zalloc(sizeof (vsw_port_t), KM_SLEEP);
2582 
2583 	rv = vsw_port_read_props(portp, vswp, mdp, node);
2584 	if (rv != 0) {
2585 		kmem_free(portp, sizeof (*portp));
2586 		return (1);
2587 	}
2588 
2589 	rv = vsw_port_attach(portp);
2590 	if (rv != 0) {
2591 		DERR(vswp, "%s: failed to attach port", __func__);
2592 		return (1);
2593 	}
2594 
2595 	return (0);
2596 }
2597 
2598 static int
vsw_port_update(vsw_t * vswp,md_t * curr_mdp,mde_cookie_t curr_mdex,md_t * prev_mdp,mde_cookie_t prev_mdex)2599 vsw_port_update(vsw_t *vswp, md_t *curr_mdp, mde_cookie_t curr_mdex,
2600 	md_t *prev_mdp, mde_cookie_t prev_mdex)
2601 {
2602 	uint64_t	cport_num;
2603 	uint64_t	pport_num;
2604 	vsw_port_list_t	*plistp;
2605 	vsw_port_t	*portp;
2606 	uint16_t	pvid;
2607 	vsw_vlanid_t	*vids;
2608 	uint16_t	nvids;
2609 	uint64_t	val;
2610 	boolean_t	hio_enabled = B_FALSE;
2611 	uint64_t	maxbw;
2612 	enum		{P_MD_init = 0x1,
2613 				P_MD_vlans = 0x2,
2614 				P_MD_hio = 0x4,
2615 				P_MD_maxbw = 0x8} updated;
2616 
2617 	updated = P_MD_init;
2618 
2619 	/*
2620 	 * For now, we get port updates only if vlan ids changed.
2621 	 * We read the port num and do some sanity check.
2622 	 */
2623 	if (md_get_prop_val(curr_mdp, curr_mdex, id_propname, &cport_num)) {
2624 		return (1);
2625 	}
2626 
2627 	if (md_get_prop_val(prev_mdp, prev_mdex, id_propname, &pport_num)) {
2628 		return (1);
2629 	}
2630 	if (cport_num != pport_num)
2631 		return (1);
2632 
2633 	plistp = &(vswp->plist);
2634 
2635 	READ_ENTER(&plistp->lockrw);
2636 
2637 	portp = vsw_lookup_port(vswp, cport_num);
2638 	if (portp == NULL) {
2639 		RW_EXIT(&plistp->lockrw);
2640 		return (1);
2641 	}
2642 
2643 	/* Read the vlan ids */
2644 	vsw_vlan_read_ids(portp, VSW_VNETPORT, curr_mdp, curr_mdex, &pvid,
2645 	    &vids, &nvids, NULL);
2646 
2647 	/* Determine if there are any vlan id updates */
2648 	if ((pvid != portp->pvid) ||		/* pvid changed? */
2649 	    (nvids != portp->nvids) ||		/* # of vids changed? */
2650 	    ((nvids != 0) && (portp->nvids != 0) &&	/* vids changed? */
2651 	    !vsw_cmp_vids(vids, portp->vids, nvids))) {
2652 		updated |= P_MD_vlans;
2653 	}
2654 
2655 	/* Check if hybrid property is present */
2656 	if (md_get_prop_val(curr_mdp, curr_mdex, hybrid_propname, &val) == 0) {
2657 		D1(vswp, "%s: prop(%s) found\n", __func__, hybrid_propname);
2658 		hio_enabled = B_TRUE;
2659 	}
2660 
2661 	if (portp->p_hio_enabled != hio_enabled) {
2662 		updated |= P_MD_hio;
2663 	}
2664 
2665 	/* Check if maxbw property is present */
2666 	vsw_port_read_bandwidth(portp, curr_mdp, curr_mdex, &maxbw);
2667 	if (maxbw != portp->p_bandwidth) {
2668 		if (maxbw >= MRP_MAXBW_MINVAL || maxbw == 0) {
2669 			updated |= P_MD_maxbw;
2670 		} else {
2671 			cmn_err(CE_NOTE, "!vsw%d: Unable to process bandwidth"
2672 			    " update for port %d as the specified value:%ld"
2673 			    " is invalid\n",
2674 			    vswp->instance, portp->p_instance, maxbw);
2675 		}
2676 	}
2677 
2678 	if (updated & P_MD_vlans) {
2679 		/* Remove existing vlan ids from the hash table. */
2680 		vsw_vlan_remove_ids(portp, VSW_VNETPORT);
2681 
2682 		/* Reconfigure vlans with network device */
2683 		vsw_mac_port_reconfig_vlans(portp, pvid, vids, nvids);
2684 
2685 		/* add these new vlan ids into hash table */
2686 		vsw_vlan_add_ids(portp, VSW_VNETPORT);
2687 
2688 		/* reset the port if it is vlan unaware (ver < 1.3) */
2689 		vsw_vlan_unaware_port_reset(portp);
2690 	}
2691 
2692 	if (updated & P_MD_hio) {
2693 		vsw_hio_port_update(portp, hio_enabled);
2694 	}
2695 
2696 	if (updated & P_MD_maxbw) {
2697 		vsw_update_bandwidth(NULL, portp, VSW_VNETPORT, maxbw);
2698 	}
2699 
2700 	RW_EXIT(&plistp->lockrw);
2701 
2702 	return (0);
2703 }
2704 
2705 /*
2706  * vsw_mac_rx -- A common function to send packets to the interface.
2707  * By default this function check if the interface is UP or not, the
2708  * rest of the behaviour depends on the flags as below:
2709  *
2710  *	VSW_MACRX_PROMISC -- Check if the promisc mode set or not.
2711  *	VSW_MACRX_COPYMSG -- Make a copy of the message(s).
2712  *	VSW_MACRX_FREEMSG -- Free if the messages cannot be sent up the stack.
2713  */
2714 void
vsw_mac_rx(vsw_t * vswp,mac_resource_handle_t mrh,mblk_t * mp,vsw_macrx_flags_t flags)2715 vsw_mac_rx(vsw_t *vswp, mac_resource_handle_t mrh,
2716     mblk_t *mp, vsw_macrx_flags_t flags)
2717 {
2718 	mblk_t		*mpt;
2719 
2720 	D1(vswp, "%s:enter\n", __func__);
2721 	READ_ENTER(&vswp->if_lockrw);
2722 	/* Check if the interface is up */
2723 	if (!(vswp->if_state & VSW_IF_UP)) {
2724 		RW_EXIT(&vswp->if_lockrw);
2725 		/* Free messages only if FREEMSG flag specified */
2726 		if (flags & VSW_MACRX_FREEMSG) {
2727 			freemsgchain(mp);
2728 		}
2729 		D1(vswp, "%s:exit\n", __func__);
2730 		return;
2731 	}
2732 	/*
2733 	 * If PROMISC flag is passed, then check if
2734 	 * the interface is in the PROMISC mode.
2735 	 * If not, drop the messages.
2736 	 */
2737 	if (flags & VSW_MACRX_PROMISC) {
2738 		if (!(vswp->if_state & VSW_IF_PROMISC)) {
2739 			RW_EXIT(&vswp->if_lockrw);
2740 			/* Free messages only if FREEMSG flag specified */
2741 			if (flags & VSW_MACRX_FREEMSG) {
2742 				freemsgchain(mp);
2743 			}
2744 			D1(vswp, "%s:exit\n", __func__);
2745 			return;
2746 		}
2747 	}
2748 	RW_EXIT(&vswp->if_lockrw);
2749 	/*
2750 	 * If COPYMSG flag is passed, then make a copy
2751 	 * of the message chain and send up the copy.
2752 	 */
2753 	if (flags & VSW_MACRX_COPYMSG) {
2754 		mp = copymsgchain(mp);
2755 		if (mp == NULL) {
2756 			D1(vswp, "%s:exit\n", __func__);
2757 			return;
2758 		}
2759 	}
2760 
2761 	D2(vswp, "%s: sending up stack", __func__);
2762 
2763 	mpt = NULL;
2764 	(void) vsw_vlan_frame_untag(vswp, VSW_LOCALDEV, &mp, &mpt);
2765 	if (mp != NULL) {
2766 		mac_rx(vswp->if_mh, mrh, mp);
2767 	}
2768 	D1(vswp, "%s:exit\n", __func__);
2769 }
2770 
2771 /* copy mac address of vsw into soft state structure */
2772 static void
vsw_save_lmacaddr(vsw_t * vswp,uint64_t macaddr)2773 vsw_save_lmacaddr(vsw_t *vswp, uint64_t macaddr)
2774 {
2775 	int	i;
2776 
2777 	WRITE_ENTER(&vswp->if_lockrw);
2778 	for (i = ETHERADDRL - 1; i >= 0; i--) {
2779 		vswp->if_addr.ether_addr_octet[i] = macaddr & 0xFF;
2780 		macaddr >>= 8;
2781 	}
2782 	RW_EXIT(&vswp->if_lockrw);
2783 }
2784 
2785 /* Compare VLAN ids, array size expected to be same. */
2786 static boolean_t
vsw_cmp_vids(vsw_vlanid_t * vids1,vsw_vlanid_t * vids2,int nvids)2787 vsw_cmp_vids(vsw_vlanid_t *vids1, vsw_vlanid_t *vids2, int nvids)
2788 {
2789 	int i, j;
2790 	uint16_t vid;
2791 
2792 	for (i = 0; i < nvids; i++) {
2793 		vid = vids1[i].vl_vid;
2794 		for (j = 0; j < nvids; j++) {
2795 			if (vid == vids2[i].vl_vid)
2796 				break;
2797 		}
2798 		if (j == nvids) {
2799 			return (B_FALSE);
2800 		}
2801 	}
2802 	return (B_TRUE);
2803 }
2804