xref: /illumos-gate/usr/src/uts/sun4v/sys/vdc.h (revision aedf2b3bb56b025fcaf87b49ec6c8aeea07f16d7)
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 2009 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #ifndef	_VDC_H
28 #define	_VDC_H
29 
30 /*
31  * Virtual disk client implementation definitions
32  */
33 
34 #include <sys/sysmacros.h>
35 #include <sys/note.h>
36 
37 #include <sys/ldc.h>
38 #include <sys/vio_mailbox.h>
39 #include <sys/vdsk_mailbox.h>
40 #include <sys/vdsk_common.h>
41 
42 #ifdef	__cplusplus
43 extern "C" {
44 #endif
45 
46 #define	VDC_DRIVER_NAME		"vdc"
47 
48 /*
49  * Bit-field values to indicate if parts of the vdc driver are initialised.
50  */
51 #define	VDC_SOFT_STATE	0x0001
52 #define	VDC_LOCKS	0x0002
53 #define	VDC_MINOR	0x0004
54 #define	VDC_THREAD	0x0008
55 #define	VDC_DRING_INIT	0x0010	/* The DRing was created */
56 #define	VDC_DRING_BOUND	0x0020	/* The DRing was bound to an LDC channel */
57 #define	VDC_DRING_LOCAL	0x0040	/* The local private DRing was allocated */
58 #define	VDC_DRING_ENTRY	0x0080	/* At least one DRing entry was initialised */
59 #define	VDC_DRING	(VDC_DRING_INIT | VDC_DRING_BOUND |	\
60 				VDC_DRING_LOCAL | VDC_DRING_ENTRY)
61 #define	VDC_HANDSHAKE	0x0100	/* Indicates if a handshake is in progress */
62 #define	VDC_HANDSHAKE_STOP	0x0200	/* stop further handshakes */
63 
64 /*
65  * Definitions of MD nodes/properties.
66  */
67 #define	VDC_MD_CHAN_NAME		"channel-endpoint"
68 #define	VDC_MD_VDEV_NAME		"virtual-device"
69 #define	VDC_MD_PORT_NAME		"virtual-device-port"
70 #define	VDC_MD_DISK_NAME		"disk"
71 #define	VDC_MD_CFG_HDL			"cfg-handle"
72 #define	VDC_MD_TIMEOUT			"vdc-timeout"
73 #define	VDC_MD_ID			"id"
74 
75 /*
76  * Definition of actions to be carried out when processing the sequence ID
77  * of a message received from the vDisk server. The function verifying the
78  * sequence number checks the 'seq_num_xxx' fields in the soft state and
79  * returns whether the message should be processed (VDC_SEQ_NUM_TODO) or
80  * whether it was it was previously processed (VDC_SEQ_NUM_SKIP).
81  */
82 #define	VDC_SEQ_NUM_INVALID		-1	/* Error */
83 #define	VDC_SEQ_NUM_SKIP		0	/* Request already processed */
84 #define	VDC_SEQ_NUM_TODO		1	/* Request needs processing */
85 
86 /*
87  * Flags for virtual disk operations.
88  */
89 #define	VDC_OP_STATE_RUNNING	0x01	/* do operation in running state */
90 #define	VDC_OP_ERRCHK_BACKEND	0x02	/* check backend on error */
91 #define	VDC_OP_ERRCHK_CONFLICT	0x04	/* check resv conflict on error */
92 
93 #define	VDC_OP_ERRCHK	(VDC_OP_ERRCHK_BACKEND | VDC_OP_ERRCHK_CONFLICT)
94 #define	VDC_OP_NORMAL	(VDC_OP_STATE_RUNNING | VDC_OP_ERRCHK)
95 
96 /*
97  * Macros to get UNIT and PART number
98  */
99 #define	VDCUNIT_SHIFT	3
100 #define	VDCPART_MASK	7
101 
102 #define	VDCUNIT(dev)	(getminor((dev)) >> VDCUNIT_SHIFT)
103 #define	VDCPART(dev)	(getminor((dev)) &  VDCPART_MASK)
104 
105 /*
106  * Scheme to store the instance number and the slice number in the minor number.
107  * (NOTE: Uses the same format and definitions as the sd(7D) driver)
108  */
109 #define	VD_MAKE_DEV(instance, minor)	((instance << VDCUNIT_SHIFT) | minor)
110 
111 #define	VDC_EFI_DEV_SET(dev, vdsk, ioctl)	\
112 	VDSK_EFI_DEV_SET(dev, vdsk, ioctl,	\
113 	    (vdsk)->vdisk_bsize, (vdsk)->vdisk_size)
114 
115 /*
116  * variables controlling how long to wait before timing out and how many
117  * retries to attempt before giving up when communicating with vds.
118  *
119  * These values need to be sufficiently large so that a guest can survive
120  * the reboot of the service domain.
121  */
122 #define	VDC_RETRIES	10
123 
124 #define	VDC_USEC_TIMEOUT_MIN	(30 * MICROSEC)		/* 30 sec */
125 
126 /*
127  * This macro returns the number of Hz that the vdc driver should wait before
128  * a timeout is triggered. The 'timeout' parameter specifiecs the wait
129  * time in Hz. The 'mul' parameter allows for a multiplier to be
130  * specified allowing for a backoff to be implemented (e.g. using the
131  * retry number as a multiplier) where the wait time will get longer if
132  * there is no response on the previous retry.
133  */
134 #define	VD_GET_TIMEOUT_HZ(timeout, mul)	\
135 	(ddi_get_lbolt() + ((timeout) * MAX(1, (mul))))
136 
137 /*
138  * Macros to manipulate Descriptor Ring variables in the soft state
139  * structure.
140  */
141 #define	VDC_GET_NEXT_REQ_ID(vdc)	((vdc)->req_id++)
142 
143 #define	VDC_GET_DRING_ENTRY_PTR(vdc, idx)	\
144 		(vd_dring_entry_t *)(uintptr_t)((vdc)->dring_mem_info.vaddr + \
145 			(idx * (vdc)->dring_entry_size))
146 
147 #define	VDC_MARK_DRING_ENTRY_FREE(vdc, idx)			\
148 	{ \
149 		vd_dring_entry_t *dep = NULL;				\
150 		ASSERT(vdc != NULL);					\
151 		ASSERT(idx < vdc->dring_len);		\
152 		ASSERT(vdc->dring_mem_info.vaddr != NULL);		\
153 		dep = (vd_dring_entry_t *)(uintptr_t)			\
154 			(vdc->dring_mem_info.vaddr +	\
155 			(idx * vdc->dring_entry_size));			\
156 		ASSERT(dep != NULL);					\
157 		dep->hdr.dstate = VIO_DESC_FREE;			\
158 	}
159 
160 /* Initialise the Session ID and Sequence Num in the DRing msg */
161 #define	VDC_INIT_DRING_DATA_MSG_IDS(dmsg, vdc)		\
162 		ASSERT(vdc != NULL);			\
163 		dmsg.tag.vio_sid = vdc->session_id;	\
164 		dmsg.seq_num = vdc->seq_num;
165 
166 /*
167  * The states that the read thread can be in.
168  */
169 typedef enum vdc_rd_state {
170 	VDC_READ_IDLE,			/* idling - conn is not up */
171 	VDC_READ_WAITING,		/* waiting for data */
172 	VDC_READ_PENDING,		/* pending data avail for read */
173 	VDC_READ_RESET			/* channel was reset - stop reads */
174 } vdc_rd_state_t;
175 
176 /*
177  * The states that the vdc-vds connection can be in.
178  */
179 typedef enum vdc_state {
180 	VDC_STATE_INIT,			/* device is initialized */
181 	VDC_STATE_INIT_WAITING,		/* waiting for ldc connection */
182 	VDC_STATE_NEGOTIATE,		/* doing handshake negotiation */
183 	VDC_STATE_HANDLE_PENDING,	/* handle requests in backup dring */
184 	VDC_STATE_FAULTED,		/* multipath backend is inaccessible */
185 	VDC_STATE_FAILED,		/* device is not usable */
186 	VDC_STATE_RUNNING,		/* running and accepting requests */
187 	VDC_STATE_DETACH,		/* detaching */
188 	VDC_STATE_RESETTING		/* resetting connection with vds */
189 } vdc_state_t;
190 
191 /*
192  * States of the service provided by a vds server
193  */
194 typedef enum vdc_service_state {
195 	VDC_SERVICE_NONE = -1, 		/* no state define */
196 	VDC_SERVICE_OFFLINE,		/* no connection with the service */
197 	VDC_SERVICE_CONNECTED,		/* connection established */
198 	VDC_SERVICE_ONLINE,		/* connection and backend available */
199 	VDC_SERVICE_FAILED,		/* connection failed */
200 	VDC_SERVICE_FAULTED		/* connection but backend unavailable */
201 } vdc_service_state_t;
202 
203 /*
204  * The states that the vdc instance can be in.
205  */
206 typedef enum vdc_lc_state {
207 	VDC_LC_ATTACHING,	/* driver is attaching */
208 	VDC_LC_ONLINE,		/* driver is attached and online */
209 	VDC_LC_DETACHING	/* driver is detaching */
210 } vdc_lc_state_t;
211 
212 /*
213  * Local Descriptor Ring entry
214  *
215  * vdc creates a Local (private) descriptor ring the same size as the
216  * public descriptor ring it exports to vds.
217  */
218 
219 typedef enum {
220 	VIO_read_dir,		/* read data from server */
221 	VIO_write_dir,		/* write data to server */
222 	VIO_both_dir		/* transfer both in and out in same buffer */
223 } vio_desc_direction_t;
224 
225 typedef struct vdc_local_desc {
226 	boolean_t		is_free;	/* local state - inuse or not */
227 
228 	int			operation;	/* VD_OP_xxx to be performed */
229 	caddr_t			addr;		/* addr passed in by consumer */
230 	int			slice;
231 	diskaddr_t		offset;		/* disk offset */
232 	size_t			nbytes;
233 	struct buf		*buf;		/* buf of operation */
234 	vio_desc_direction_t	dir;		/* direction of transfer */
235 	int			flags;		/* flags of operation */
236 
237 	caddr_t			align_addr;	/* used if addr non-aligned */
238 	ldc_mem_handle_t	desc_mhdl;	/* Mem handle of buf */
239 	vd_dring_entry_t	*dep;		/* public Dring Entry Pointer */
240 
241 } vdc_local_desc_t;
242 
243 /*
244  * I/O queue used for checking backend or failfast
245  */
246 typedef struct vdc_io {
247 	struct vdc_io	*vio_next;	/* next pending I/O in the queue */
248 	int		vio_index;	/* descriptor index */
249 	clock_t		vio_qtime;	/* time the I/O was queued */
250 } vdc_io_t;
251 
252 /*
253  * Per vDisk server channel states
254  */
255 #define	VDC_LDC_INIT	0x0001
256 #define	VDC_LDC_CB	0x0002
257 #define	VDC_LDC_OPEN	0x0004
258 #define	VDC_LDC		(VDC_LDC_INIT | VDC_LDC_CB | VDC_LDC_OPEN)
259 
260 /*
261  * vDisk server information
262  */
263 typedef struct vdc_server {
264 	struct vdc_server	*next;			/* Next server */
265 	struct vdc		*vdcp;			/* Ptr to vdc struct */
266 	uint64_t		id;			/* Server port id */
267 	uint64_t		state;			/* Server state */
268 	vdc_service_state_t	svc_state;		/* Service state */
269 	vdc_service_state_t	log_state;		/* Last state logged */
270 	uint64_t		ldc_id;			/* Server LDC id */
271 	ldc_handle_t		ldc_handle;		/* Server LDC handle */
272 	ldc_status_t		ldc_state;		/* Server LDC state */
273 	uint64_t		ctimeout;		/* conn tmout (secs) */
274 } vdc_server_t;
275 
276 /*
277  * vdc soft state structure
278  */
279 typedef struct vdc {
280 
281 	kmutex_t	lock;		/* protects next 2 sections of vars */
282 	kcondvar_t	running_cv;	/* signal when upper layers can send */
283 	kcondvar_t	initwait_cv;	/* signal when ldc conn is up */
284 	kcondvar_t	dring_free_cv;	/* signal when desc is avail */
285 	kcondvar_t	membind_cv;	/* signal when mem can be bound */
286 	boolean_t	self_reset;	/* self initiated reset */
287 	kcondvar_t	io_pending_cv;	/* signal on pending I/O */
288 	boolean_t	io_pending;	/* pending I/O */
289 
290 	int		initialized;	/* keeps track of what's init'ed */
291 	vdc_lc_state_t	lifecycle;	/* Current state of the vdc instance */
292 
293 	int		hshake_cnt;	/* number of failed handshakes */
294 	uint8_t		open[OTYPCNT];	/* mask of opened slices */
295 	uint8_t		open_excl;	/* mask of exclusively opened slices */
296 	ulong_t		open_lyr[V_NUMPAR]; /* number of layered opens */
297 	int		dkio_flush_pending; /* # outstanding DKIO flushes */
298 	int		validate_pending; /* # outstanding validate request */
299 	vd_disk_label_t vdisk_label; 	/* label type of device/disk imported */
300 	struct extvtoc	*vtoc;		/* structure to store VTOC data */
301 	struct dk_geom	*geom;		/* structure to store geometry data */
302 	vd_slice_t	slice[V_NUMPAR]; /* logical partitions */
303 
304 	kthread_t	*msg_proc_thr;	/* main msg processing thread */
305 
306 	kmutex_t	read_lock;	/* lock to protect read */
307 	kcondvar_t	read_cv;	/* cv to wait for READ events */
308 	vdc_rd_state_t	read_state;	/* current read state */
309 
310 	uint32_t	sync_op_cnt;	/* num of active sync operations */
311 	boolean_t	sync_op_blocked; /* blocked waiting to do sync op */
312 	kcondvar_t	sync_blocked_cv; /* cv wait for other syncs to finish */
313 
314 	uint64_t	session_id;	/* common ID sent with all messages */
315 	uint64_t	seq_num;	/* most recent sequence num generated */
316 	uint64_t	seq_num_reply;	/* Last seq num ACK/NACK'ed by vds */
317 	uint64_t	req_id;		/* Most recent Request ID generated */
318 	uint64_t	req_id_proc;	/* Last request ID processed by vdc */
319 	vdc_state_t	state;		/* Current disk client-server state */
320 
321 	dev_info_t	*dip;		/* device info pointer */
322 	int		instance;	/* driver instance number */
323 
324 	vio_ver_t	ver;		/* version number agreed with server */
325 	vd_disk_type_t	vdisk_type;	/* type of device/disk being imported */
326 	uint32_t	vdisk_media;	/* physical media type of vDisk */
327 	uint64_t	vdisk_size;	/* device size in blocks */
328 	uint64_t	max_xfer_sz;	/* maximum block size of a descriptor */
329 	uint64_t	vdisk_bsize;	/* blk size for the virtual disk */
330 	uint32_t	vio_bmask;	/* mask to check vio blk alignment */
331 	int		vio_bshift;	/* shift for vio blk conversion */
332 	uint64_t	operations;	/* bitmask of ops. server supports */
333 	struct dk_cinfo	*cinfo;		/* structure to store DKIOCINFO data */
334 	struct dk_minfo	*minfo;		/* structure for DKIOCGMEDIAINFO data */
335 	ddi_devid_t	devid;		/* device id */
336 	boolean_t	ctimeout_reached; /* connection timeout has expired */
337 
338 	/*
339 	 * The ownership fields are protected by the lock mutex. The
340 	 * ownership_lock mutex is used to serialize ownership operations;
341 	 * it should be acquired before the lock mutex.
342 	 */
343 	kmutex_t	ownership_lock;		/* serialize ownership ops */
344 	int		ownership;		/* ownership status flags */
345 	kthread_t	*ownership_thread;	/* ownership thread */
346 	kcondvar_t	ownership_cv;		/* cv for ownership update */
347 
348 	/*
349 	 * The eio and failfast fields are protected by the lock mutex.
350 	 */
351 	kthread_t	*eio_thread;		/* error io thread */
352 	kcondvar_t	eio_cv;			/* cv for eio thread update */
353 	vdc_io_t	*eio_queue;		/* error io queue */
354 	clock_t		failfast_interval;	/* interval in microsecs */
355 
356 	/*
357 	 * kstats used to store I/O statistics consumed by iostat(1M).
358 	 * These are protected by the lock mutex.
359 	 */
360 	kstat_t		*io_stats;
361 	kstat_t		*err_stats;
362 
363 	ldc_dring_handle_t	dring_hdl;		/* dring handle */
364 	ldc_mem_info_t		dring_mem_info;		/* dring information */
365 	uint_t			dring_curr_idx;		/* current index */
366 	uint32_t		dring_len;		/* dring length */
367 	uint32_t		dring_max_cookies;	/* dring max cookies */
368 	uint32_t		dring_cookie_count;	/* num cookies */
369 	uint32_t		dring_entry_size;	/* descriptor size */
370 	ldc_mem_cookie_t 	*dring_cookie;		/* dring cookies */
371 	uint64_t		dring_ident;		/* dring ident */
372 
373 	uint64_t		threads_pending; 	/* num of threads */
374 
375 	vdc_local_desc_t	*local_dring;		/* local dring */
376 	vdc_local_desc_t	*local_dring_backup;	/* local dring backup */
377 	int			local_dring_backup_tail; /* backup dring tail */
378 	int			local_dring_backup_len;	/* backup dring len */
379 
380 	int			num_servers;		/* no. of servers */
381 	vdc_server_t		*server_list;		/* vdisk server list */
382 	vdc_server_t		*curr_server;		/* curr vdisk server */
383 } vdc_t;
384 
385 /*
386  * Ownership status flags
387  */
388 #define	VDC_OWNERSHIP_NONE	0x00 /* no ownership wanted */
389 #define	VDC_OWNERSHIP_WANTED	0x01 /* ownership is wanted */
390 #define	VDC_OWNERSHIP_GRANTED	0x02 /* ownership has been granted */
391 #define	VDC_OWNERSHIP_RESET	0x04 /* ownership has been reset */
392 
393 /*
394  * Reservation conflict panic message
395  */
396 #define	VDC_RESV_CONFLICT_FMT_STR	"Reservation Conflict\nDisk: "
397 #define	VDC_RESV_CONFLICT_FMT_LEN 	(sizeof (VDC_RESV_CONFLICT_FMT_STR))
398 
399 /*
400  * Debugging macros
401  */
402 #ifdef DEBUG
403 extern int	vdc_msglevel;
404 extern uint64_t	vdc_matchinst;
405 
406 #define	DMSG(_vdc, err_level, format, ...)				\
407 	do {								\
408 		if (vdc_msglevel > err_level &&				\
409 		(vdc_matchinst & (1ull << (_vdc)->instance)))		\
410 			cmn_err(CE_CONT, "?[%d,t@%p] %s: "format,	\
411 			(_vdc)->instance, (void *)curthread,		\
412 			__func__, __VA_ARGS__);				\
413 		_NOTE(CONSTANTCONDITION)				\
414 	} while (0);
415 
416 #define	DMSGX(err_level, format, ...)					\
417 	do {								\
418 		if (vdc_msglevel > err_level)				\
419 			cmn_err(CE_CONT, "?%s: "format, __func__, __VA_ARGS__);\
420 		_NOTE(CONSTANTCONDITION)				\
421 	} while (0);
422 
423 #define	VDC_DUMP_DRING_MSG(dmsgp)					\
424 		DMSGX(0, "sq:%lu start:%d end:%d ident:%lu\n",		\
425 			dmsgp->seq_num, dmsgp->start_idx,		\
426 			dmsgp->end_idx, dmsgp->dring_ident);
427 
428 #else	/* !DEBUG */
429 #define	DMSG(err_level, ...)
430 #define	DMSGX(err_level, format, ...)
431 #define	VDC_DUMP_DRING_MSG(dmsgp)
432 
433 #endif	/* !DEBUG */
434 
435 #ifdef	__cplusplus
436 }
437 #endif
438 
439 #endif	/* _VDC_H */
440