xref: /titanic_41/usr/src/uts/sun4v/sys/vsw_ldc.h (revision cdc7c7fe7b52f0c955073224bbd5fd2002db45d0)
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 /*
28  * This header file contains the basic data structures which the
29  * virtual switch (vsw) uses to communicate with vnet clients.
30  *
31  * The virtual switch reads the machine description (MD) to
32  * determine how many port_t structures to create (each port_t
33  * can support communications to a single network device). The
34  * port_t's are maintained in a linked list.
35  *
36  * Each port in turn contains a number of logical domain channels
37  * (ldc's) which are inter domain communications channels which
38  * are used for passing small messages between the domains. There
39  * may be any number of channels associated with each port, though
40  * currently most devices only have a single channel. The current
41  * implementation provides support for only one channel per port.
42  *
43  * The ldc is a bi-directional channel, which is divided up into
44  * two directional 'lanes', one outbound from the switch to the
45  * virtual network device, the other inbound to the switch.
46  * Depending on the type of device each lane may have seperate
47  * communication paramaters (such as mtu etc).
48  *
49  * For those network clients which use descriptor rings the
50  * rings are associated with the appropriate lane. I.e. rings
51  * which the switch exports are associated with the outbound lanes
52  * while those which the network clients are exporting to the switch
53  * are associated with the inbound lane.
54  *
55  * In diagram form the data structures look as follows:
56  *
57  * vsw instance
58  *     |
59  *     +----->port_t----->port_t----->port_t----->
60  *		|
61  *		+--->ldc_t
62  *		       |
63  *		       +--->lane_t (inbound)
64  *		       |       |
65  *		       |       +--->dring
66  *		       |
67  *		       +--->lane_t (outbound)
68  *			       |
69  *			       +--->dring
70  *
71  */
72 
73 #ifndef	_VSW_LDC_H
74 #define	_VSW_LDC_H
75 
76 #ifdef	__cplusplus
77 extern "C" {
78 #endif
79 
80 /*
81  * LDC pkt tranfer MTU - largest msg size used
82  */
83 #define	VSW_LDC_MTU		64
84 
85 #define	VSW_DEF_MSG_WORDS	\
86 	(VNET_DRING_REG_EXT_MSG_SIZE_MAX / sizeof (uint64_t))
87 
88 /*
89  * Default message type.
90  */
91 typedef struct def_msg {
92 	uint64_t	data[VSW_DEF_MSG_WORDS];
93 } def_msg_t;
94 
95 /*
96  * Currently only support one major/minor pair.
97  */
98 #define	VSW_NUM_VER	1
99 
100 typedef struct ver_sup {
101 	uint16_t	ver_major;	/* major version number */
102 	uint16_t	ver_minor;	/* minor version number */
103 } ver_sup_t;
104 
105 /*
106  * Lane states.
107  */
108 #define	VSW_LANE_INACTIV	0x0	/* No params set for lane */
109 
110 #define	VSW_VER_INFO_SENT	0x1	/* Version # sent to peer */
111 #define	VSW_VER_INFO_RECV	0x2	/* Version # recv from peer */
112 #define	VSW_VER_ACK_RECV	0x4
113 #define	VSW_VER_ACK_SENT	0x8
114 #define	VSW_VER_NACK_RECV	0x10
115 #define	VSW_VER_NACK_SENT	0x20
116 
117 #define	VSW_ATTR_INFO_SENT	0x40	/* Attributes sent to peer */
118 #define	VSW_ATTR_INFO_RECV	0x80	/* Peer attributes received */
119 #define	VSW_ATTR_ACK_SENT	0x100
120 #define	VSW_ATTR_ACK_RECV	0x200
121 #define	VSW_ATTR_NACK_SENT	0x400
122 #define	VSW_ATTR_NACK_RECV	0x800
123 
124 #define	VSW_DRING_INFO_SENT	0x1000	/* Dring info sent to peer */
125 #define	VSW_DRING_INFO_RECV	0x2000	/* Dring info received */
126 #define	VSW_DRING_ACK_SENT	0x4000
127 #define	VSW_DRING_ACK_RECV	0x8000
128 #define	VSW_DRING_NACK_SENT	0x10000
129 #define	VSW_DRING_NACK_RECV	0x20000
130 
131 #define	VSW_RDX_INFO_SENT	0x40000	/* RDX sent to peer */
132 #define	VSW_RDX_INFO_RECV	0x80000	/* RDX received from peer */
133 #define	VSW_RDX_ACK_SENT	0x100000
134 #define	VSW_RDX_ACK_RECV	0x200000
135 #define	VSW_RDX_NACK_SENT	0x400000
136 #define	VSW_RDX_NACK_RECV	0x800000
137 
138 #define	VSW_MCST_INFO_SENT	0x1000000
139 #define	VSW_MCST_INFO_RECV	0x2000000
140 #define	VSW_MCST_ACK_SENT	0x4000000
141 #define	VSW_MCST_ACK_RECV	0x8000000
142 #define	VSW_MCST_NACK_SENT	0x10000000
143 #define	VSW_MCST_NACK_RECV	0x20000000
144 
145 #define	VSW_LANE_ACTIVE		0x40000000	/* Lane open to xmit data */
146 
147 /* Handshake milestones */
148 #define	VSW_MILESTONE0		0x1	/* ver info exchanged */
149 #define	VSW_MILESTONE1		0x2	/* attribute exchanged */
150 #define	VSW_MILESTONE2		0x4	/* dring info exchanged */
151 #define	VSW_MILESTONE3		0x8	/* rdx exchanged */
152 #define	VSW_MILESTONE4		0x10	/* handshake complete */
153 
154 /*
155  * Lane direction (relative to ourselves).
156  */
157 #define	INBOUND			0x1
158 #define	OUTBOUND		0x2
159 
160 /* Peer session id received */
161 #define	VSW_PEER_SESSION	0x1
162 
163 /*
164  * Maximum number of consecutive reads of data from channel
165  */
166 #define	VSW_MAX_CHAN_READ	50
167 
168 /*
169  * Currently only support one ldc per port.
170  */
171 #define	VSW_PORT_MAX_LDCS	1	/* max # of ldcs per port */
172 
173 /*
174  * Used for port add/deletion.
175  */
176 #define	VSW_PORT_UPDATED	0x1
177 
178 #define	LDC_TX_SUCCESS		0	/* ldc transmit success */
179 #define	LDC_TX_FAILURE		1	/* ldc transmit failure */
180 #define	LDC_TX_NORESOURCES	2	/* out of descriptors */
181 
182 /*
183  * Descriptor ring info
184  *
185  * Each descriptor element has a pre-allocated data buffer
186  * associated with it, into which data being transmitted is
187  * copied. By pre-allocating we speed up the copying process.
188  * The buffer is re-used once the peer has indicated that it is
189  * finished with the descriptor.
190  */
191 #define	VSW_RING_EL_DATA_SZ	2048	/* Size of data section (bytes) */
192 #define	VSW_PRIV_SIZE	sizeof (vnet_private_desc_t)
193 
194 #define	VSW_MAX_COOKIES		((ETHERMTU >> MMU_PAGESHIFT) + 2)
195 
196 /*
197  * Size of the mblk in each mblk pool.
198  */
199 #define	VSW_MBLK_SZ_128		128
200 #define	VSW_MBLK_SZ_256		256
201 #define	VSW_MBLK_SZ_2048	2048
202 
203 /*
204  * Number of mblks in each mblk pool.
205  */
206 #define	VSW_NUM_MBLKS	1024
207 
208 /* increment recv index */
209 #define	INCR_DESC_INDEX(dp, i)	\
210 		((i) = (((i) + 1) & ((dp)->num_descriptors - 1)))
211 
212 /* decrement recv index */
213 #define	DECR_DESC_INDEX(dp, i)	\
214 		((i) = (((i) - 1) & ((dp)->num_descriptors - 1)))
215 
216 #define	INCR_TXI	INCR_DESC_INDEX
217 #define	DECR_TXI	DECR_DESC_INDEX
218 #define	INCR_RXI	INCR_DESC_INDEX
219 #define	DECR_RXI	DECR_DESC_INDEX
220 
221 /* bounds check rx index */
222 #define	CHECK_DESC_INDEX(dp, i)	\
223 		(((i) >= 0) && ((i) < (dp)->num_descriptors))
224 
225 #define	CHECK_RXI	CHECK_DESC_INDEX
226 #define	CHECK_TXI	CHECK_DESC_INDEX
227 
228 /*
229  * Private descriptor
230  */
231 typedef struct vsw_private_desc {
232 	/*
233 	 * Below lock must be held when accessing the state of
234 	 * a descriptor on either the private or public sections
235 	 * of the ring.
236 	 */
237 	kmutex_t		dstate_lock;
238 	uint64_t		dstate;
239 	vnet_public_desc_t	*descp;
240 	ldc_mem_handle_t	memhandle;
241 	void			*datap;
242 	uint64_t		datalen;
243 	uint64_t		ncookies;
244 	ldc_mem_cookie_t	memcookie[VSW_MAX_COOKIES];
245 	int			bound;
246 } vsw_private_desc_t;
247 
248 /*
249  * Descriptor ring structure
250  */
251 typedef struct dring_info {
252 	kmutex_t		dlock;		/* sync access */
253 	uint32_t		num_descriptors; /* # of descriptors */
254 	uint32_t		descriptor_size; /* size of descriptor */
255 	uint32_t		options;	/* dring options (mode) */
256 	ldc_dring_handle_t	dring_handle;	/* dring LDC handle */
257 	uint32_t		dring_ncookies;	/* # of dring cookies */
258 	ldc_mem_cookie_t	dring_cookie[1]; /* LDC cookie of dring */
259 	ldc_mem_handle_t	data_handle;	/* data area  LDC handle */
260 	uint32_t		data_ncookies;	/* # of data area cookies */
261 	ldc_mem_cookie_t	*data_cookie;	/* data area LDC cookies */
262 	uint64_t		ident;		/* identifier sent to peer */
263 	uint64_t		end_idx;	/* last idx processed */
264 	int64_t			last_ack_recv;	/* last ack received */
265 	kmutex_t		txlock;		/* protect tx desc alloc */
266 	uint32_t		next_txi;	/* next tx descriptor index */
267 	uint32_t		next_rxi;	/* next expected recv index */
268 	kmutex_t		restart_lock;	/* protect restart_reqd */
269 	boolean_t		restart_reqd;	/* send restart msg */
270 	void			*pub_addr;	/* base of public section */
271 	void			*priv_addr;	/* base of private section */
272 	void			*data_addr;	/* base of data section */
273 	size_t			data_sz;	/* size of data section */
274 	size_t			desc_data_sz;	/* size of descr data blk */
275 	uint8_t			dring_mtype;	/* dring mem map type */
276 	uint32_t		num_bufs;	/* # of buffers */
277 	vio_mblk_pool_t		*rx_vmp;	/* rx mblk pool */
278 	vio_mblk_t		**rxdp_to_vmp;	/* descr to buf map tbl */
279 } dring_info_t;
280 
281 /*
282  * Each ldc connection is comprised of two lanes, incoming
283  * from a peer, and outgoing to that peer. Each lane shares
284  * common ldc parameters and also has private lane-specific
285  * parameters.
286  */
287 typedef struct lane {
288 	uint64_t	lstate;		/* Lane state */
289 	uint16_t	ver_major;	/* Version major number */
290 	uint16_t	ver_minor;	/* Version minor number */
291 	uint64_t	seq_num;	/* Sequence number */
292 	uint64_t	mtu;		/* ETHERMTU */
293 	uint64_t	addr;		/* Unique physical address */
294 	uint8_t		addr_type;	/* Only MAC address at moment */
295 	uint8_t		xfer_mode;	/* Dring or Pkt based */
296 	uint8_t		ack_freq;	/* Only non zero for Pkt based xfer */
297 	uint32_t	physlink_update;	/* physlink updates */
298 	uint8_t		dring_mode;	/* Descriptor ring mode */
299 	dring_info_t	*dringp;	/* List of drings for this lane */
300 } lane_t;
301 
302 /* channel drain states */
303 #define	VSW_LDC_INIT		0x1	/* Initial non-drain state */
304 #define	VSW_LDC_DRAINING	0x2	/* Channel draining */
305 
306 /*
307  * vnet-protocol-version dependent function prototypes.
308  */
309 typedef int	(*vsw_ldctx_t) (void *, mblk_t *, mblk_t *, uint32_t);
310 typedef void	(*vsw_ldcrx_pktdata_t) (void *, void *, uint32_t);
311 typedef void	(*vsw_ldcrx_dringdata_t) (void *, void *);
312 
313 /* ldc information associated with a vsw-port */
314 typedef struct vsw_ldc {
315 	struct vsw_ldc		*ldc_next;	/* next ldc in the list */
316 	struct vsw_port		*ldc_port;	/* associated port */
317 	struct vsw		*ldc_vswp;	/* associated vsw */
318 	kmutex_t		ldc_cblock;	/* sync callback processing */
319 	kmutex_t		ldc_txlock;	/* sync transmits */
320 	kmutex_t		ldc_rxlock;	/* sync rx */
321 	uint64_t		ldc_id;		/* channel number */
322 	ldc_handle_t		ldc_handle;	/* channel handle */
323 	kmutex_t		drain_cv_lock;
324 	kcondvar_t		drain_cv;	/* channel draining */
325 	int			drain_state;
326 	uint32_t		hphase;		/* handshake phase */
327 	int			hcnt;		/* # handshake attempts */
328 	kmutex_t		status_lock;
329 	ldc_status_t		ldc_status;	/* channel status */
330 	uint8_t			reset_active;	/* reset flag */
331 	uint64_t		local_session;	/* Our session id */
332 	uint64_t		peer_session;	/* Our peers session id */
333 	uint8_t			session_status;	/* Session recv'd, sent */
334 	uint32_t		hss_id;		/* Handshake session id */
335 	uint64_t		next_ident;	/* Next dring ident # to use */
336 	lane_t			lane_in;	/* Inbound lane */
337 	lane_t			lane_out;	/* Outbound lane */
338 	uint8_t			dev_class;	/* Peer device class */
339 	boolean_t		pls_negotiated;	/* phys link state update ? */
340 	vio_multi_pool_t	vmp;		/* Receive mblk pools */
341 	uint32_t		max_rxpool_size; /* max size of rxpool in use */
342 	uint64_t		*ldcmsg;	/* msg buffer for ldc_read() */
343 	uint64_t		msglen;		/* size of ldcmsg */
344 	uint32_t		dringdata_msgid; /* msgid in RxDringData mode */
345 
346 	/* tx thread fields */
347 	kthread_t		*tx_thread;	/* tx thread */
348 	uint32_t		tx_thr_flags;	/* tx thread flags */
349 	kmutex_t		tx_thr_lock;	/* lock for tx thread */
350 	kcondvar_t		tx_thr_cv;	/* cond.var for tx thread */
351 	mblk_t			*tx_mhead;	/* tx mblks head */
352 	mblk_t			*tx_mtail;	/* tx mblks tail */
353 	uint32_t		tx_cnt;		/* # of pkts queued for tx */
354 
355 	/* message thread fields */
356 	kthread_t		*msg_thread;	/* message thread */
357 	uint32_t		msg_thr_flags;	/* message thread flags */
358 	kmutex_t		msg_thr_lock;	/* lock for message thread */
359 	kcondvar_t		msg_thr_cv;	/* cond.var for msg thread */
360 
361 	/* receive thread fields */
362 	kthread_t		*rcv_thread;	/* receive thread */
363 	uint32_t		rcv_thr_flags;	/* receive thread flags */
364 	kmutex_t		rcv_thr_lock;	/* lock for receive thread */
365 	kcondvar_t		rcv_thr_cv;	/* cond.var for recv thread */
366 
367 	vsw_ldctx_t		tx;		/* transmit function */
368 	vsw_ldcrx_pktdata_t	rx_pktdata;	/* process raw data msg */
369 	vsw_ldcrx_dringdata_t	rx_dringdata;	/* process dring data msg */
370 
371 	/* channel statistics */
372 	vgen_stats_t		ldc_stats;	/* channel statistics */
373 	kstat_t			*ksp;		/* channel kstats */
374 } vsw_ldc_t;
375 
376 /* worker thread flags */
377 #define	VSW_WTHR_DATARCVD 	0x01	/* data received */
378 #define	VSW_WTHR_STOP 		0x02	/* stop worker thread request */
379 
380 /* multicast addresses port is interested in */
381 typedef struct mcst_addr {
382 	struct mcst_addr	*nextp;
383 	struct ether_addr	mca;	/* multicast address */
384 	uint64_t		addr;	/* mcast addr converted to hash key */
385 	boolean_t		mac_added; /* added into physical device */
386 } mcst_addr_t;
387 
388 /* Port detach states */
389 #define	VSW_PORT_INIT		0x1	/* Initial non-detach state */
390 #define	VSW_PORT_DETACHING	0x2	/* In process of being detached */
391 #define	VSW_PORT_DETACHABLE	0x4	/* Safe to detach */
392 
393 /* port information associated with a vsw */
394 typedef struct vsw_port {
395 	int			p_instance;	/* port instance */
396 	struct vsw_port		*p_next;	/* next port in the list */
397 	struct vsw		*p_vswp;	/* associated vsw */
398 	int			num_ldcs;	/* # of ldcs in the port */
399 	uint64_t		*ldc_ids;	/* ldc ids */
400 	vsw_ldc_t		*ldcp;		/* ldc for this port */
401 
402 	kmutex_t		tx_lock;	/* transmit lock */
403 	int			(*transmit)(vsw_ldc_t *, mblk_t *);
404 
405 	int			state;		/* port state */
406 	kmutex_t		state_lock;
407 	kcondvar_t		state_cv;
408 
409 	krwlock_t		maccl_rwlock;	/* protect fields below */
410 	mac_client_handle_t	p_mch;		/* mac client handle */
411 	mac_unicast_handle_t	p_muh;		/* mac unicast handle */
412 
413 	kmutex_t		mca_lock;	/* multicast lock */
414 	mcst_addr_t		*mcap;		/* list of multicast addrs */
415 
416 	boolean_t		addr_set;	/* Addr set where */
417 
418 	/*
419 	 * mac address of the port & connected device
420 	 */
421 	struct ether_addr	p_macaddr;
422 	uint16_t		pvid;	/* port vlan id (untagged) */
423 	struct vsw_vlanid	*vids;	/* vlan ids (tagged) */
424 	uint16_t		nvids;	/* # of vids */
425 	mod_hash_t		*vlan_hashp;	/* vlan hash table */
426 	uint32_t		vlan_nchains;	/* # of vlan hash chains */
427 
428 	/* HybridIO related info */
429 	uint32_t		p_hio_enabled;	/* Hybrid mode enabled? */
430 	uint32_t		p_hio_capable;	/* Port capable of HIO */
431 
432 	/* bandwidth limit */
433 	uint64_t		p_bandwidth;	/* bandwidth limit */
434 } vsw_port_t;
435 
436 /* list of ports per vsw */
437 typedef struct vsw_port_list {
438 	vsw_port_t	*head;		/* head of the list */
439 	krwlock_t	lockrw;		/* sync access(rw) to the list */
440 	int		num_ports;	/* number of ports in the list */
441 } vsw_port_list_t;
442 
443 /*
444  * Taskq control message
445  */
446 typedef struct vsw_ctrl_task {
447 	vsw_ldc_t	*ldcp;
448 	def_msg_t	pktp;
449 	uint32_t	hss_id;
450 } vsw_ctrl_task_t;
451 
452 /*
453  * State of connection to peer. Some of these states
454  * can be mapped to LDC events as follows:
455  *
456  * VSW_CONN_RESET -> LDC_RESET_EVT
457  * VSW_CONN_UP    -> LDC_UP_EVT
458  */
459 #define	VSW_CONN_UP		0x1	/* Connection come up */
460 #define	VSW_CONN_RESET		0x2	/* Connection reset */
461 #define	VSW_CONN_RESTART	0x4	/* Restarting handshake on connection */
462 
463 typedef struct vsw_conn_evt {
464 	uint16_t	evt;		/* Connection event */
465 	vsw_ldc_t	*ldcp;
466 } vsw_conn_evt_t;
467 
468 /*
469  * Ethernet broadcast address definition.
470  */
471 static	struct	ether_addr	etherbroadcastaddr = {
472 	0xff, 0xff, 0xff, 0xff, 0xff, 0xff
473 };
474 
475 #define	IS_BROADCAST(ehp) \
476 	(bcmp(&ehp->ether_dhost, &etherbroadcastaddr, ETHERADDRL) == 0)
477 #define	IS_MULTICAST(ehp) \
478 	((ehp->ether_dhost.ether_addr_octet[0] & 01) == 1)
479 
480 #define	READ_ENTER(x)	rw_enter(x, RW_READER)
481 #define	WRITE_ENTER(x)	rw_enter(x, RW_WRITER)
482 #define	RW_EXIT(x)	rw_exit(x)
483 
484 #define	VSW_PORT_REFHOLD(portp)	atomic_inc_32(&((portp)->ref_cnt))
485 #define	VSW_PORT_REFRELE(portp)	atomic_dec_32(&((portp)->ref_cnt))
486 
487 #ifdef	__cplusplus
488 }
489 #endif
490 
491 #endif	/* _VSW_LDC_H */
492