xref: /illumos-gate/usr/src/uts/sun4v/sys/vnet.h (revision e557d412)
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 2010 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #ifndef _VNET_H
28 #define	_VNET_H
29 
30 #ifdef __cplusplus
31 extern "C" {
32 #endif
33 
34 #include <sys/vnet_res.h>
35 #include <sys/vnet_mailbox.h>
36 #include <sys/modhash.h>
37 #include <net/if.h>
38 #include <sys/mac_client.h>
39 
40 #define	VNET_SUCCESS		(0)	/* successful return */
41 #define	VNET_FAILURE		(-1)	/* unsuccessful return */
42 
43 #define	KMEM_FREE(_p)		kmem_free((_p), sizeof (*(_p)))
44 
45 #define	VNET_NTXDS		512		/* power of 2 tx descriptors */
46 #define	VNET_LDCWD_INTERVAL	1000		/* watchdog freq in msec */
47 #define	VNET_LDCWD_TXTIMEOUT	1000		/* tx timeout in msec */
48 #define	VNET_LDC_MTU		64		/* ldc mtu */
49 
50 
51 #define	IS_BROADCAST(ehp) \
52 		(ether_cmp(&ehp->ether_dhost, &etherbroadcastaddr) == 0)
53 #define	IS_MULTICAST(ehp) \
54 		((ehp->ether_dhost.ether_addr_octet[0] & 01) == 1)
55 
56 #define	VNET_MATCH_RES(vresp, vnetp)	\
57 	(ether_cmp(vresp->local_macaddr, vnetp->curr_macaddr) == 0)
58 
59 /*
60  * Flags used to indicate the state of the vnet device and its associated
61  * resources.
62  */
63 typedef enum vnet_flags {
64 	VNET_STOPPED = 0x0,
65 	VNET_STARTED = 0x1,
66 	VNET_STOPPING = 0x2
67 } vnet_flags_t;
68 
69 typedef struct vnet_hio_stats {
70 	/* Link Input/Output stats */
71 	uint64_t	ipackets;	/* # rx packets */
72 	uint64_t	ierrors;	/* # rx error */
73 	uint64_t	opackets;	/* # tx packets */
74 	uint64_t	oerrors;	/* # tx error */
75 
76 	/* MIB II variables */
77 	uint64_t	rbytes;		/* # bytes received */
78 	uint64_t	obytes;		/* # bytes transmitted */
79 	uint32_t	multircv;	/* # multicast packets received */
80 	uint32_t	multixmt;	/* # multicast packets for xmit */
81 	uint32_t	brdcstrcv;	/* # broadcast packets received */
82 	uint32_t	brdcstxmt;	/* # broadcast packets for xmit */
83 	uint32_t	norcvbuf;	/* # rcv packets discarded */
84 	uint32_t	noxmtbuf;	/* # xmit packets discarded */
85 } vnet_hio_stats_t;
86 
87 typedef struct vnet_hio_kstats {
88 	/* Link Input/Output stats */
89 	kstat_named_t	ipackets;
90 	kstat_named_t	ipackets64;
91 	kstat_named_t	ierrors;
92 	kstat_named_t	opackets;
93 	kstat_named_t	opackets64;
94 	kstat_named_t	oerrors;
95 
96 	/* required by kstat for MIB II objects(RFC 1213) */
97 	kstat_named_t	rbytes; 	/* MIB - ifInOctets */
98 	kstat_named_t	rbytes64;
99 	kstat_named_t	obytes; 	/* MIB - ifOutOctets */
100 	kstat_named_t	obytes64;
101 	kstat_named_t	multircv; 	/* MIB - ifInNUcastPkts */
102 	kstat_named_t	multixmt; 	/* MIB - ifOutNUcastPkts */
103 	kstat_named_t	brdcstrcv;	/* MIB - ifInNUcastPkts */
104 	kstat_named_t	brdcstxmt;	/* MIB - ifOutNUcastPkts */
105 	kstat_named_t	norcvbuf; 	/* MIB - ifInDiscards */
106 	kstat_named_t	noxmtbuf; 	/* MIB - ifOutDiscards */
107 } vnet_hio_kstats_t;
108 
109 typedef struct vnet_tx_ring_stats {
110 	uint64_t	opackets;	/* # tx packets */
111 	uint64_t	obytes;		/* # bytes transmitted */
112 } vnet_tx_ring_stats_t;
113 
114 /*
115  * A vnet resource structure.
116  */
117 typedef struct vnet_res {
118 	struct vnet_res		*nextp;		/* next resource in the list */
119 	mac_register_t		macreg;		/* resource's mac_reg */
120 	vio_net_res_type_t	type;		/* resource type */
121 	ether_addr_t		local_macaddr;	/* resource's macaddr */
122 	ether_addr_t		rem_macaddr;	/* resource's remote macaddr */
123 	vnet_flags_t		flags;		/* resource flags */
124 	uint32_t		refcnt;		/* reference count */
125 	struct	vnet		*vnetp;		/* back pointer to vnet */
126 	kstat_t			*ksp;		/* hio kstats */
127 	void			*rx_ringp;	/* assoc pseudo rx ring */
128 } vnet_res_t;
129 
130 #define	VNET_DDS_TASK_ADD_SHARE		0x01
131 #define	VNET_DDS_TASK_DEL_SHARE		0x02
132 #define	VNET_DDS_TASK_REL_SHARE		0x04
133 
134 /* An instance specific DDS structure */
135 typedef struct vnet_dds_info {
136 	kmutex_t	lock;		/* lock for this structure */
137 	uint8_t		task_flags;	/* flags for taskq */
138 	uint8_t		dds_req_id;	/* DDS message request id */
139 	vio_dds_msg_t	dmsg;		/* Pending DDS message */
140 	dev_info_t	*hio_dip;	/* Hybrid device's dip */
141 	uint64_t	hio_cookie;	/* Hybrid device's cookie */
142 	char		hio_ifname[LIFNAMSIZ];  /* Hybrid interface name */
143 	ddi_taskq_t	*dds_taskqp;	/* Taskq's used for DDS */
144 	struct vnet	*vnetp;		/* Back pointer to vnetp */
145 } vnet_dds_info_t;
146 
147 #define	VNET_NFDB_HASH	64
148 
149 #define	KEY_HASH(key, addr) \
150 	(key = (((uint64_t)(addr[0])) << 40) | \
151 	(((uint64_t)(addr[1])) << 32) | \
152 	(((uint64_t)(addr[2])) << 24) | \
153 	(((uint64_t)(addr[3])) << 16) | \
154 	(((uint64_t)(addr[4])) << 8) | \
155 	((uint64_t)(addr[5])));
156 
157 
158 /* rwlock macros */
159 #define	READ_ENTER(x)	rw_enter(x, RW_READER)
160 #define	WRITE_ENTER(x)	rw_enter(x, RW_WRITER)
161 #define	RW_EXIT(x)	rw_exit(x)
162 
163 #define	VLAN_ID_KEY(key)	((mod_hash_key_t)(uintptr_t)(key))
164 
165 typedef enum {
166 		AST_init = 0x0, AST_vnet_alloc = 0x1,
167 		AST_ring_init = 0x2, AST_vdds_init = 0x4,
168 		AST_read_macaddr = 0x8, AST_fdbh_alloc = 0x10,
169 		AST_taskq_create = 0x20, AST_vnet_list = 0x40,
170 		AST_vgen_init = 0x80, AST_macreg = 0x100,
171 		AST_init_mdeg = 0x200
172 } vnet_attach_progress_t;
173 
174 #define	VNET_NUM_PSEUDO_GROUPS		1	/* # of pseudo ring grps */
175 #define	VNET_NUM_HYBRID_RINGS		2	/* # of Hybrid tx/rx rings */
176 #define	VNET_HYBRID_RXRING_INDEX	1	/* Hybrid rx ring start index */
177 
178 /*
179  * # of Pseudo TX Rings is defined based on the possible
180  * # of TX Hardware Rings from a Hybrid resource.
181  */
182 #define	VNET_NUM_PSEUDO_TXRINGS		VNET_NUM_HYBRID_RINGS
183 
184 /*
185  * # of Pseudo RX Rings that are reserved and exposed by default.
186  * 1 for LDC resource to vsw + 2 for RX rings of Hybrid resource.
187  */
188 #define	VNET_NUM_PSEUDO_RXRINGS_DEFAULT	(VNET_NUM_HYBRID_RINGS + 1)
189 
190 /* Pseudo RX Ring States */
191 typedef enum {
192 	VNET_RXRING_FREE = 0x0,		/* Free */
193 	VNET_RXRING_INUSE = 0x1,	/* In use */
194 	VNET_RXRING_LDC_SERVICE = 0x2,	/* Mapped to vswitch */
195 	VNET_RXRING_LDC_GUEST = 0x4,	/* Mapped to a peer vnet */
196 	VNET_RXRING_HYBRID = 0x8,	/* Mapped to Hybrid resource */
197 	VNET_RXRING_STARTED = 0x10	/* Started */
198 } vnet_rxring_state_t;
199 
200 /* Pseudo TX Ring States */
201 typedef enum {
202 	VNET_TXRING_FREE = 0x0,		/* Free */
203 	VNET_TXRING_INUSE = 0x1,	/* In use */
204 	VNET_TXRING_SHARED = 0x2,	/* Shared among LDCs */
205 	VNET_TXRING_HYBRID = 0x4,	/* Shared among LDCs, Hybrid resource */
206 	VNET_TXRING_STARTED = 0x8	/* Started */
207 } vnet_txring_state_t;
208 
209 /*
210  * Psuedo TX Ring
211  */
212 typedef struct vnet_pseudo_tx_ring {
213 	uint_t			index;		/* ring index */
214 	vnet_txring_state_t	state;		/* ring state */
215 	void			*grp;		/* grp associated */
216 	void			*vnetp;		/* vnet associated */
217 	mac_ring_handle_t	handle;		/* ring handle in mac layer */
218 	mac_ring_handle_t	hw_rh;	/* Resource type dependent, internal */
219 					/* ring handle. Hybrid res: ring hdl */
220 					/* of hardware rx ring; LDC res: hdl */
221 					/* to the res itself (vnet_res_t)    */
222 	boolean_t		woken_up;
223 	vnet_tx_ring_stats_t	tx_ring_stats;	/* ring statistics */
224 } vnet_pseudo_tx_ring_t;
225 
226 /*
227  * Psuedo RX Ring
228  */
229 typedef struct vnet_pseudo_rx_ring {
230 	uint_t			index;		/* ring index */
231 	vnet_rxring_state_t	state;		/* ring state */
232 	void			*grp;		/* grp associated */
233 	void			*vnetp;		/* vnet associated */
234 	mac_ring_handle_t	handle;		/* ring handle in mac layer */
235 	mac_ring_handle_t	hw_rh;	/* Resource type dependent, internal */
236 					/* ring handle. Hybrid res: ring hdl */
237 					/* of hardware tx ring; otherwise    */
238 					/* NULL */
239 	uint64_t		gen_num;	/* Mac layer gen_num */
240 } vnet_pseudo_rx_ring_t;
241 
242 /*
243  * Psuedo TX Ring Group
244  */
245 typedef struct vnet_pseudo_tx_group {
246 	uint_t			index;		/* group index */
247 	void			*vnetp;		/* vnet associated */
248 	mac_group_handle_t	handle;		/* grp handle in mac layer */
249 	uint_t			ring_cnt;	/* total # of rings in grp */
250 	vnet_pseudo_tx_ring_t	*rings;		/* array of rings */
251 	kmutex_t		flowctl_lock;	/* flow control lock */
252 	kcondvar_t		flowctl_cv;
253 	kthread_t		*flowctl_thread;
254 	boolean_t		flowctl_done;
255 	void			*tx_notify_handle; /* Tx ring notification */
256 } vnet_pseudo_tx_group_t;
257 
258 /*
259  * Psuedo RX Ring Group
260  */
261 typedef struct vnet_pseudo_rx_group {
262 	krwlock_t		lock;		/* sync rings access in grp */
263 	int			index;		/* group index */
264 	void			*vnetp;		/* vnet this grp belongs to */
265 	mac_group_handle_t	handle;		/* grp handle in mac layer */
266 	uint_t			max_ring_cnt;	/* total # of rings in grp */
267 	uint_t			ring_cnt;	/* # of rings in use */
268 	vnet_pseudo_rx_ring_t	*rings;		/* array of rings */
269 } vnet_pseudo_rx_group_t;
270 
271 /*
272  * vnet instance state information
273  */
274 typedef struct vnet {
275 	int			instance;	/* instance # */
276 	dev_info_t		*dip;		/* dev_info */
277 	uint64_t		reg;		/* reg prop value */
278 	vnet_attach_progress_t	attach_progress; /* attach progress flags */
279 	struct vnet		*nextp;		/* next in list */
280 	mac_handle_t		mh;		/* handle to GLDv3 mac module */
281 	uchar_t			vendor_addr[ETHERADDRL]; /* orig macadr */
282 	uchar_t			curr_macaddr[ETHERADDRL]; /* current macadr */
283 	void			*vgenhdl;	/* Handle for vgen */
284 
285 	uint32_t		fdb_nchains;	/* # of hash chains in fdbtbl */
286 	mod_hash_t		*fdb_hashp;	/* forwarding database */
287 	vnet_res_t		*vsw_fp;	/* cached fdb entry of vsw */
288 	krwlock_t		vsw_fp_rw;	/* lock to protect vsw_fp */
289 	uint32_t		mtu;		/* mtu of the device */
290 
291 	uint16_t		default_vlan_id; /* default vlan id */
292 	uint16_t		pvid;		/* port vlan id (untagged) */
293 	uint16_t		*vids;		/* vlan ids (tagged) */
294 	uint16_t		nvids;		/* # of vids */
295 
296 	link_state_t		link_state;	/* link status */
297 	boolean_t		pls_update;	/* phys link state update ? */
298 	vnet_flags_t		flags;		/* interface flags */
299 	vnet_res_t		*hio_fp;	/* Hybrid IO resource */
300 	vnet_res_t		*vres_list;	/* Resource list */
301 	vnet_dds_info_t		vdds_info;	/* DDS related info */
302 	krwlock_t		vrwlock;	/* Resource list lock */
303 	ddi_taskq_t		*taskqp;	/* Resource taskq */
304 
305 	/* pseudo ring groups */
306 	vnet_pseudo_rx_group_t	rx_grp[VNET_NUM_PSEUDO_GROUPS];
307 	vnet_pseudo_tx_group_t	tx_grp[VNET_NUM_PSEUDO_GROUPS];
308 
309 	vio_net_handle_t	hio_vhp;	/* HIO resource hdl */
310 	mac_handle_t		hio_mh;		/* HIO mac hdl */
311 	mac_client_handle_t	hio_mch;	/* HIO mac client hdl */
312 	mac_unicast_handle_t	hio_muh;	/* HIO mac unicst hdl */
313 	mac_group_handle_t	rx_hwgh;	/* HIO rx ring-group hdl */
314 	mac_group_handle_t	tx_hwgh;	/* HIO tx ring-group hdl */
315 } vnet_t;
316 
317 #ifdef DEBUG
318 /*
319  * debug levels:
320  * DBG_LEVEL1:	Function entry/exit tracing
321  * DBG_LEVEL2:	Info messages
322  * DBG_LEVEL3:	Warning messages
323  * DBG_LEVEL4:	Error messages
324  */
325 
326 enum	{ DBG_LEVEL1 = 0x01, DBG_LEVEL2 = 0x02, DBG_WARN = 0x04,
327 	    DBG_ERR = 0x08 };
328 
329 #define	DBG1(...)	do {						\
330 			    if ((vnet_dbglevel & DBG_LEVEL1) != 0) {	\
331 				debug_printf(__func__, __VA_ARGS__);	\
332 			    }						\
333 			_NOTE(CONSTCOND) } while (0)
334 
335 #define	DBG2(...)	do {						\
336 			    if ((vnet_dbglevel & DBG_LEVEL2) != 0) {	\
337 				debug_printf(__func__, __VA_ARGS__);	\
338 			    }						\
339 			_NOTE(CONSTCOND) } while (0)
340 
341 #define	DWARN(...)	do {						\
342 			    if ((vnet_dbglevel & DBG_WARN) != 0) {	\
343 				debug_printf(__func__, __VA_ARGS__);	\
344 			    }						\
345 			_NOTE(CONSTCOND) } while (0)
346 
347 #define	DERR(...)	do {						\
348 			    if ((vnet_dbglevel & DBG_ERR) != 0) {	\
349 				debug_printf(__func__, __VA_ARGS__);	\
350 			    }						\
351 			_NOTE(CONSTCOND) } while (0)
352 
353 #else
354 
355 #define	DBG1(...)	if (0)	do { } while (0)
356 #define	DBG2(...)	if (0)	do { } while (0)
357 #define	DWARN(...)	if (0)	do { } while (0)
358 #define	DERR(...)	if (0)	do { } while (0)
359 
360 #endif
361 
362 #ifdef	VNET_IOC_DEBUG	/* Debug ioctls */
363 
364 #define	VNET_FORCE_LINK_DOWN	0x1
365 #define	VNET_FORCE_LINK_UP	0x2
366 
367 #endif
368 
369 #ifdef __cplusplus
370 }
371 #endif
372 
373 #endif	/* _VNET_H */
374