xref: /illumos-gate/usr/src/uts/sun4v/sys/vdc.h (revision bbf21555)
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 #ifndef	_VDC_H
27 #define	_VDC_H
28 
29 /*
30  * Virtual disk client implementation definitions
31  */
32 
33 #include <sys/sysmacros.h>
34 #include <sys/note.h>
35 
36 #include <sys/ldc.h>
37 #include <sys/vio_mailbox.h>
38 #include <sys/vdsk_mailbox.h>
39 #include <sys/vdsk_common.h>
40 
41 #ifdef	__cplusplus
42 extern "C" {
43 #endif
44 
45 #define	VDC_DRIVER_NAME		"vdc"
46 
47 /*
48  * Bit-field values to indicate if parts of the vdc driver are initialised.
49  */
50 #define	VDC_SOFT_STATE	0x0001
51 #define	VDC_LOCKS	0x0002
52 #define	VDC_MINOR	0x0004
53 #define	VDC_THREAD	0x0008
54 #define	VDC_DRING_INIT	0x0010	/* The DRing was created */
55 #define	VDC_DRING_BOUND	0x0020	/* The DRing was bound to an LDC channel */
56 #define	VDC_DRING_LOCAL	0x0040	/* The local private DRing was allocated */
57 #define	VDC_DRING_ENTRY	0x0080	/* At least one DRing entry was initialised */
58 #define	VDC_DRING	(VDC_DRING_INIT | VDC_DRING_BOUND |	\
59 				VDC_DRING_LOCAL | VDC_DRING_ENTRY)
60 #define	VDC_HANDSHAKE	0x0100	/* Indicates if a handshake is in progress */
61 #define	VDC_HANDSHAKE_STOP	0x0200	/* stop further handshakes */
62 
63 /*
64  * Definitions of MD nodes/properties.
65  */
66 #define	VDC_MD_CHAN_NAME		"channel-endpoint"
67 #define	VDC_MD_VDEV_NAME		"virtual-device"
68 #define	VDC_MD_PORT_NAME		"virtual-device-port"
69 #define	VDC_MD_DISK_NAME		"disk"
70 #define	VDC_MD_CFG_HDL			"cfg-handle"
71 #define	VDC_MD_TIMEOUT			"vdc-timeout"
72 #define	VDC_MD_ID			"id"
73 
74 /*
75  * Definition of actions to be carried out when processing the sequence ID
76  * of a message received from the vDisk server. The function verifying the
77  * sequence number checks the 'seq_num_xxx' fields in the soft state and
78  * returns whether the message should be processed (VDC_SEQ_NUM_TODO) or
79  * whether it was it was previously processed (VDC_SEQ_NUM_SKIP).
80  */
81 #define	VDC_SEQ_NUM_INVALID		-1	/* Error */
82 #define	VDC_SEQ_NUM_SKIP		0	/* Request already processed */
83 #define	VDC_SEQ_NUM_TODO		1	/* Request needs processing */
84 
85 /*
86  * DRing reserved entries. Entry 0 is reserved and only used for error
87  * checking. This is done so that error checking can be done even if the
88  * DRing is full. All other entries are available for regular I/Os.
89  */
90 #define	VDC_DRING_NUM_RESV		1	/* #reserved entries */
91 #define	VDC_DRING_FIRST_RESV		0	/* 1st reserved entry */
92 #define	VDC_DRING_FIRST_ENTRY		\
93 	    (VDC_DRING_FIRST_RESV + VDC_DRING_NUM_RESV)	/* 1st non-resv entry */
94 
95 /*
96  * Flags for virtual disk operations.
97  */
98 #define	VDC_OP_STATE_RUNNING	0x01	/* do operation in running state */
99 #define	VDC_OP_ERRCHK_BACKEND	0x02	/* check backend on error */
100 #define	VDC_OP_ERRCHK_CONFLICT	0x04	/* check resv conflict on error */
101 #define	VDC_OP_DRING_RESERVED	0x08	/* use dring reserved entry */
102 #define	VDC_OP_RESUBMIT		0x10	/* I/O is being resubmitted */
103 
104 #define	VDC_OP_ERRCHK	(VDC_OP_ERRCHK_BACKEND | VDC_OP_ERRCHK_CONFLICT)
105 #define	VDC_OP_NORMAL	(VDC_OP_STATE_RUNNING | VDC_OP_ERRCHK)
106 
107 /*
108  * Macros to get UNIT and PART number
109  */
110 #define	VDCUNIT_SHIFT	3
111 #define	VDCPART_MASK	7
112 
113 #define	VDCUNIT(dev)	(getminor((dev)) >> VDCUNIT_SHIFT)
114 #define	VDCPART(dev)	(getminor((dev)) &  VDCPART_MASK)
115 
116 /*
117  * Scheme to store the instance number and the slice number in the minor number.
118  * (NOTE: Uses the same format and definitions as the sd(4D) driver)
119  */
120 #define	VD_MAKE_DEV(instance, minor)	((instance << VDCUNIT_SHIFT) | minor)
121 
122 #define	VDC_EFI_DEV_SET(dev, vdsk, ioctl)	\
123 	VDSK_EFI_DEV_SET(dev, vdsk, ioctl,	\
124 	    (vdsk)->vdisk_bsize, (vdsk)->vdisk_size)
125 
126 /* max number of handshake retries per server */
127 #define	VDC_HSHAKE_RETRIES	3
128 
129 /* minimum number of attribute negotiations before handshake failure */
130 #define	VDC_HATTR_MIN_INITIAL	3
131 #define	VDC_HATTR_MIN		1
132 
133 /*
134  * This macro returns the number of Hz that the vdc driver should wait before
135  * a timeout is triggered. The 'timeout' parameter specifiecs the wait
136  * time in Hz. The 'mul' parameter allows for a multiplier to be
137  * specified allowing for a backoff to be implemented (e.g. using the
138  * retry number as a multiplier) where the wait time will get longer if
139  * there is no response on the previous retry.
140  */
141 #define	VD_GET_TIMEOUT_HZ(timeout, mul)	\
142 	(ddi_get_lbolt() + ((timeout) * MAX(1, (mul))))
143 
144 /*
145  * Macros to manipulate Descriptor Ring variables in the soft state
146  * structure.
147  */
148 #define	VDC_GET_NEXT_REQ_ID(vdc)	((vdc)->req_id++)
149 
150 #define	VDC_GET_DRING_ENTRY_PTR(vdc, idx)	\
151 		(vd_dring_entry_t *)(uintptr_t)((vdc)->dring_mem_info.vaddr + \
152 			(idx * (vdc)->dring_entry_size))
153 
154 #define	VDC_MARK_DRING_ENTRY_FREE(vdc, idx)			\
155 	{ \
156 		vd_dring_entry_t *dep = NULL;				\
157 		ASSERT(vdc != NULL);					\
158 		ASSERT(idx < vdc->dring_len);		\
159 		ASSERT(vdc->dring_mem_info.vaddr != NULL);		\
160 		dep = (vd_dring_entry_t *)(uintptr_t)			\
161 			(vdc->dring_mem_info.vaddr +	\
162 			(idx * vdc->dring_entry_size));			\
163 		ASSERT(dep != NULL);					\
164 		dep->hdr.dstate = VIO_DESC_FREE;			\
165 	}
166 
167 /* Initialise the Session ID and Sequence Num in the DRing msg */
168 #define	VDC_INIT_DRING_DATA_MSG_IDS(dmsg, vdc)		\
169 		ASSERT(vdc != NULL);			\
170 		dmsg.tag.vio_sid = vdc->session_id;	\
171 		dmsg.seq_num = vdc->seq_num;
172 
173 /*
174  * The states that the read thread can be in.
175  */
176 typedef enum vdc_rd_state {
177 	VDC_READ_IDLE,			/* idling - conn is not up */
178 	VDC_READ_WAITING,		/* waiting for data */
179 	VDC_READ_PENDING,		/* pending data avail for read */
180 	VDC_READ_RESET			/* channel was reset - stop reads */
181 } vdc_rd_state_t;
182 
183 /*
184  * The states that the vdc-vds connection can be in.
185  */
186 typedef enum vdc_state {
187 	VDC_STATE_INIT,			/* device is initialized */
188 	VDC_STATE_INIT_WAITING,		/* waiting for ldc connection */
189 	VDC_STATE_NEGOTIATE,		/* doing handshake negotiation */
190 	VDC_STATE_HANDLE_PENDING,	/* handle requests in backup dring */
191 	VDC_STATE_FAULTED,		/* multipath backend is inaccessible */
192 	VDC_STATE_FAILED,		/* device is not usable */
193 	VDC_STATE_RUNNING,		/* running and accepting requests */
194 	VDC_STATE_DETACH,		/* detaching */
195 	VDC_STATE_RESETTING		/* resetting connection with vds */
196 } vdc_state_t;
197 
198 /*
199  * States of the service provided by a vds server
200  */
201 typedef enum vdc_service_state {
202 	VDC_SERVICE_NONE = -1, 		/* no state define */
203 	VDC_SERVICE_OFFLINE,		/* no connection with the service */
204 	VDC_SERVICE_CONNECTED,		/* connection established */
205 	VDC_SERVICE_ONLINE,		/* connection and backend available */
206 	VDC_SERVICE_FAILED,		/* connection failed */
207 	VDC_SERVICE_FAULTED		/* connection but backend unavailable */
208 } vdc_service_state_t;
209 
210 /*
211  * The states that the vdc instance can be in.
212  */
213 typedef enum vdc_lc_state {
214 	VDC_LC_ATTACHING,	/* driver is attaching */
215 	VDC_LC_ONLINE_PENDING,	/* driver is attached, handshake pending */
216 	VDC_LC_ONLINE,		/* driver is attached and online */
217 	VDC_LC_DETACHING	/* driver is detaching */
218 } vdc_lc_state_t;
219 
220 /*
221  * Local Descriptor Ring entry
222  *
223  * vdc creates a Local (private) descriptor ring the same size as the
224  * public descriptor ring it exports to vds.
225  */
226 
227 typedef enum {
228 	VIO_read_dir,		/* read data from server */
229 	VIO_write_dir,		/* write data to server */
230 	VIO_both_dir		/* transfer both in and out in same buffer */
231 } vio_desc_direction_t;
232 
233 typedef struct vdc_local_desc {
234 	boolean_t		is_free;	/* local state - inuse or not */
235 
236 	int			operation;	/* VD_OP_xxx to be performed */
237 	caddr_t			addr;		/* addr passed in by consumer */
238 	int			slice;
239 	diskaddr_t		offset;		/* disk offset */
240 	size_t			nbytes;
241 	struct buf		*buf;		/* buf of operation */
242 	vio_desc_direction_t	dir;		/* direction of transfer */
243 	int			flags;		/* flags of operation */
244 
245 	caddr_t			align_addr;	/* used if addr non-aligned */
246 	ldc_mem_handle_t	desc_mhdl;	/* Mem handle of buf */
247 	vd_dring_entry_t	*dep;		/* public Dring Entry Pointer */
248 
249 } vdc_local_desc_t;
250 
251 /*
252  * I/O queue used for checking backend or failfast
253  */
254 typedef struct vdc_io {
255 	struct vdc_io	*vio_next;	/* next pending I/O in the queue */
256 	int		vio_index;	/* descriptor index */
257 	clock_t		vio_qtime;	/* time the I/O was queued */
258 } vdc_io_t;
259 
260 /*
261  * Per vDisk server channel states
262  */
263 #define	VDC_LDC_INIT	0x0001
264 #define	VDC_LDC_CB	0x0002
265 #define	VDC_LDC_OPEN	0x0004
266 #define	VDC_LDC		(VDC_LDC_INIT | VDC_LDC_CB | VDC_LDC_OPEN)
267 
268 /*
269  * vDisk server information
270  */
271 typedef struct vdc_server {
272 	struct vdc_server	*next;			/* Next server */
273 	struct vdc		*vdcp;			/* Ptr to vdc struct */
274 	uint64_t		id;			/* Server port id */
275 	uint64_t		state;			/* Server state */
276 	vdc_service_state_t	svc_state;		/* Service state */
277 	vdc_service_state_t	log_state;		/* Last state logged */
278 	uint64_t		ldc_id;			/* Server LDC id */
279 	ldc_handle_t		ldc_handle;		/* Server LDC handle */
280 	ldc_status_t		ldc_state;		/* Server LDC state */
281 	uint64_t		ctimeout;		/* conn tmout (secs) */
282 	uint_t			hshake_cnt;		/* handshakes count */
283 	uint_t			hattr_cnt;		/* attr. neg. count */
284 	uint_t			hattr_total;		/* attr. neg. total */
285 } vdc_server_t;
286 
287 /*
288  * vdc soft state structure
289  */
290 typedef struct vdc {
291 
292 	kmutex_t	lock;		/* protects next 2 sections of vars */
293 	kcondvar_t	running_cv;	/* signal when upper layers can send */
294 	kcondvar_t	initwait_cv;	/* signal when ldc conn is up */
295 	kcondvar_t	dring_free_cv;	/* signal when desc is avail */
296 	kcondvar_t	membind_cv;	/* signal when mem can be bound */
297 	boolean_t	self_reset;	/* self initiated reset */
298 	kcondvar_t	io_pending_cv;	/* signal on pending I/O */
299 	boolean_t	io_pending;	/* pending I/O */
300 
301 	int		initialized;	/* keeps track of what's init'ed */
302 	vdc_lc_state_t	lifecycle;	/* Current state of the vdc instance */
303 	uint_t		hattr_min;	/* min. # attribute negotiations */
304 
305 	uint8_t		open[OTYPCNT];	/* mask of opened slices */
306 	uint8_t		open_excl;	/* mask of exclusively opened slices */
307 	ulong_t		open_lyr[V_NUMPAR]; /* number of layered opens */
308 	int		dkio_flush_pending; /* # outstanding DKIO flushes */
309 	int		validate_pending; /* # outstanding validate request */
310 	vd_disk_label_t vdisk_label; 	/* label type of device/disk imported */
311 	struct extvtoc	*vtoc;		/* structure to store VTOC data */
312 	struct dk_geom	*geom;		/* structure to store geometry data */
313 	vd_slice_t	slice[V_NUMPAR]; /* logical partitions */
314 
315 	kthread_t	*msg_proc_thr;	/* main msg processing thread */
316 
317 	kmutex_t	read_lock;	/* lock to protect read */
318 	kcondvar_t	read_cv;	/* cv to wait for READ events */
319 	vdc_rd_state_t	read_state;	/* current read state */
320 
321 	uint32_t	sync_op_cnt;	/* num of active sync operations */
322 	boolean_t	sync_op_blocked; /* blocked waiting to do sync op */
323 	kcondvar_t	sync_blocked_cv; /* cv wait for other syncs to finish */
324 
325 	uint64_t	session_id;	/* common ID sent with all messages */
326 	uint64_t	seq_num;	/* most recent sequence num generated */
327 	uint64_t	seq_num_reply;	/* Last seq num ACK/NACK'ed by vds */
328 	uint64_t	req_id;		/* Most recent Request ID generated */
329 	uint64_t	req_id_proc;	/* Last request ID processed by vdc */
330 	vdc_state_t	state;		/* Current disk client-server state */
331 
332 	dev_info_t	*dip;		/* device info pointer */
333 	int		instance;	/* driver instance number */
334 
335 	vio_ver_t	ver;		/* version number agreed with server */
336 	vd_disk_type_t	vdisk_type;	/* type of device/disk being imported */
337 	uint32_t	vdisk_media;	/* physical media type of vDisk */
338 	uint64_t	vdisk_size;	/* device size in blocks */
339 	uint64_t	max_xfer_sz;	/* maximum block size of a descriptor */
340 	uint64_t	vdisk_bsize;	/* blk size for the virtual disk */
341 	uint32_t	vio_bmask;	/* mask to check vio blk alignment */
342 	int		vio_bshift;	/* shift for vio blk conversion */
343 	uint64_t	operations;	/* bitmask of ops. server supports */
344 	struct dk_cinfo	*cinfo;		/* structure to store DKIOCINFO data */
345 	struct dk_minfo	*minfo;		/* structure for DKIOCGMEDIAINFO data */
346 	ddi_devid_t	devid;		/* device id */
347 	boolean_t	ctimeout_reached; /* connection timeout has expired */
348 
349 	/*
350 	 * The ownership fields are protected by the lock mutex. The
351 	 * ownership_lock mutex is used to serialize ownership operations;
352 	 * it should be acquired before the lock mutex.
353 	 */
354 	kmutex_t	ownership_lock;		/* serialize ownership ops */
355 	int		ownership;		/* ownership status flags */
356 	kthread_t	*ownership_thread;	/* ownership thread */
357 	kcondvar_t	ownership_cv;		/* cv for ownership update */
358 
359 	/*
360 	 * The eio and failfast fields are protected by the lock mutex.
361 	 */
362 	kthread_t	*eio_thread;		/* error io thread */
363 	kcondvar_t	eio_cv;			/* cv for eio thread update */
364 	vdc_io_t	*eio_queue;		/* error io queue */
365 	clock_t		failfast_interval;	/* interval in microsecs */
366 
367 	/*
368 	 * kstats used to store I/O statistics consumed by iostat(8).
369 	 * These are protected by the lock mutex.
370 	 */
371 	kstat_t		*io_stats;
372 	kstat_t		*err_stats;
373 
374 	ldc_dring_handle_t	dring_hdl;		/* dring handle */
375 	ldc_mem_info_t		dring_mem_info;		/* dring information */
376 	uint_t			dring_curr_idx;		/* current index */
377 	uint32_t		dring_len;		/* dring length */
378 	uint32_t		dring_max_cookies;	/* dring max cookies */
379 	uint32_t		dring_cookie_count;	/* num cookies */
380 	uint32_t		dring_entry_size;	/* descriptor size */
381 	ldc_mem_cookie_t 	*dring_cookie;		/* dring cookies */
382 	uint64_t		dring_ident;		/* dring ident */
383 
384 	uint64_t		threads_pending; 	/* num of threads */
385 
386 	vdc_local_desc_t	*local_dring;		/* local dring */
387 	vdc_local_desc_t	*local_dring_backup;	/* local dring backup */
388 	int			local_dring_backup_tail; /* backup dring tail */
389 	int			local_dring_backup_len;	/* backup dring len */
390 
391 	int			num_servers;		/* no. of servers */
392 	vdc_server_t		*server_list;		/* vdisk server list */
393 	vdc_server_t		*curr_server;		/* curr vdisk server */
394 } vdc_t;
395 
396 /*
397  * Ownership status flags
398  */
399 #define	VDC_OWNERSHIP_NONE	0x00 /* no ownership wanted */
400 #define	VDC_OWNERSHIP_WANTED	0x01 /* ownership is wanted */
401 #define	VDC_OWNERSHIP_GRANTED	0x02 /* ownership has been granted */
402 #define	VDC_OWNERSHIP_RESET	0x04 /* ownership has been reset */
403 
404 /*
405  * Reservation conflict panic message
406  */
407 #define	VDC_RESV_CONFLICT_FMT_STR	"Reservation Conflict\nDisk: "
408 #define	VDC_RESV_CONFLICT_FMT_LEN 	(sizeof (VDC_RESV_CONFLICT_FMT_STR))
409 
410 /*
411  * Debugging macros
412  */
413 #ifdef DEBUG
414 extern int	vdc_msglevel;
415 extern uint64_t	vdc_matchinst;
416 
417 #define	DMSG(_vdc, err_level, format, ...)				\
418 	do {								\
419 		if (vdc_msglevel > err_level &&				\
420 		(vdc_matchinst & (1ull << (_vdc)->instance)))		\
421 			cmn_err(CE_CONT, "?[%d,t@%p] %s: "format,	\
422 			(_vdc)->instance, (void *)curthread,		\
423 			__func__, __VA_ARGS__);				\
424 		_NOTE(CONSTANTCONDITION)				\
425 	} while (0);
426 
427 #define	DMSGX(err_level, format, ...)					\
428 	do {								\
429 		if (vdc_msglevel > err_level)				\
430 			cmn_err(CE_CONT, "?%s: "format, __func__, __VA_ARGS__);\
431 		_NOTE(CONSTANTCONDITION)				\
432 	} while (0);
433 
434 #define	VDC_DUMP_DRING_MSG(dmsgp)					\
435 		DMSGX(0, "sq:%lu start:%d end:%d ident:%lu\n",		\
436 			dmsgp->seq_num, dmsgp->start_idx,		\
437 			dmsgp->end_idx, dmsgp->dring_ident);
438 
439 #else	/* !DEBUG */
440 #define	DMSG(err_level, ...)
441 #define	DMSGX(err_level, format, ...)
442 #define	VDC_DUMP_DRING_MSG(dmsgp)
443 
444 #endif	/* !DEBUG */
445 
446 #ifdef	__cplusplus
447 }
448 #endif
449 
450 #endif	/* _VDC_H */
451