xref: /linux/drivers/block/sunvdc.c (revision 19f2e267a5d0d26282a64f8f788c482852c95324)
1 /* sunvdc.c: Sun LDOM Virtual Disk Client.
2  *
3  * Copyright (C) 2007, 2008 David S. Miller <davem@davemloft.net>
4  */
5 
6 #include <linux/module.h>
7 #include <linux/kernel.h>
8 #include <linux/types.h>
9 #include <linux/blk-mq.h>
10 #include <linux/hdreg.h>
11 #include <linux/genhd.h>
12 #include <linux/cdrom.h>
13 #include <linux/slab.h>
14 #include <linux/spinlock.h>
15 #include <linux/completion.h>
16 #include <linux/delay.h>
17 #include <linux/init.h>
18 #include <linux/list.h>
19 #include <linux/scatterlist.h>
20 
21 #include <asm/vio.h>
22 #include <asm/ldc.h>
23 
24 #define DRV_MODULE_NAME		"sunvdc"
25 #define PFX DRV_MODULE_NAME	": "
26 #define DRV_MODULE_VERSION	"1.2"
27 #define DRV_MODULE_RELDATE	"November 24, 2014"
28 
29 static char version[] =
30 	DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
31 MODULE_AUTHOR("David S. Miller (davem@davemloft.net)");
32 MODULE_DESCRIPTION("Sun LDOM virtual disk client driver");
33 MODULE_LICENSE("GPL");
34 MODULE_VERSION(DRV_MODULE_VERSION);
35 
36 #define VDC_TX_RING_SIZE	512
37 #define VDC_DEFAULT_BLK_SIZE	512
38 
39 #define MAX_XFER_BLKS		(128 * 1024)
40 #define MAX_XFER_SIZE		(MAX_XFER_BLKS / VDC_DEFAULT_BLK_SIZE)
41 #define MAX_RING_COOKIES	((MAX_XFER_BLKS / PAGE_SIZE) + 2)
42 
43 #define WAITING_FOR_LINK_UP	0x01
44 #define WAITING_FOR_TX_SPACE	0x02
45 #define WAITING_FOR_GEN_CMD	0x04
46 #define WAITING_FOR_ANY		-1
47 
48 #define	VDC_MAX_RETRIES	10
49 
50 static struct workqueue_struct *sunvdc_wq;
51 
52 struct vdc_req_entry {
53 	struct request		*req;
54 };
55 
56 struct vdc_port {
57 	struct vio_driver_state	vio;
58 
59 	struct gendisk		*disk;
60 
61 	struct vdc_completion	*cmp;
62 
63 	u64			req_id;
64 	u64			seq;
65 	struct vdc_req_entry	rq_arr[VDC_TX_RING_SIZE];
66 
67 	unsigned long		ring_cookies;
68 
69 	u64			max_xfer_size;
70 	u32			vdisk_block_size;
71 	u32			drain;
72 
73 	u64			ldc_timeout;
74 	struct delayed_work	ldc_reset_timer_work;
75 	struct work_struct	ldc_reset_work;
76 
77 	/* The server fills these in for us in the disk attribute
78 	 * ACK packet.
79 	 */
80 	u64			operations;
81 	u32			vdisk_size;
82 	u8			vdisk_type;
83 	u8			vdisk_mtype;
84 	u32			vdisk_phys_blksz;
85 
86 	struct blk_mq_tag_set	tag_set;
87 
88 	char			disk_name[32];
89 };
90 
91 static void vdc_ldc_reset(struct vdc_port *port);
92 static void vdc_ldc_reset_work(struct work_struct *work);
93 static void vdc_ldc_reset_timer_work(struct work_struct *work);
94 
95 static inline struct vdc_port *to_vdc_port(struct vio_driver_state *vio)
96 {
97 	return container_of(vio, struct vdc_port, vio);
98 }
99 
100 /* Ordered from largest major to lowest */
101 static struct vio_version vdc_versions[] = {
102 	{ .major = 1, .minor = 2 },
103 	{ .major = 1, .minor = 1 },
104 	{ .major = 1, .minor = 0 },
105 };
106 
107 static inline int vdc_version_supported(struct vdc_port *port,
108 					u16 major, u16 minor)
109 {
110 	return port->vio.ver.major == major && port->vio.ver.minor >= minor;
111 }
112 
113 #define VDCBLK_NAME	"vdisk"
114 static int vdc_major;
115 #define PARTITION_SHIFT	3
116 
117 static inline u32 vdc_tx_dring_avail(struct vio_dring_state *dr)
118 {
119 	return vio_dring_avail(dr, VDC_TX_RING_SIZE);
120 }
121 
122 static int vdc_getgeo(struct block_device *bdev, struct hd_geometry *geo)
123 {
124 	struct gendisk *disk = bdev->bd_disk;
125 	sector_t nsect = get_capacity(disk);
126 	sector_t cylinders = nsect;
127 
128 	geo->heads = 0xff;
129 	geo->sectors = 0x3f;
130 	sector_div(cylinders, geo->heads * geo->sectors);
131 	geo->cylinders = cylinders;
132 	if ((sector_t)(geo->cylinders + 1) * geo->heads * geo->sectors < nsect)
133 		geo->cylinders = 0xffff;
134 
135 	return 0;
136 }
137 
138 /* Add ioctl/CDROM_GET_CAPABILITY to support cdrom_id in udev
139  * when vdisk_mtype is VD_MEDIA_TYPE_CD or VD_MEDIA_TYPE_DVD.
140  * Needed to be able to install inside an ldom from an iso image.
141  */
142 static int vdc_ioctl(struct block_device *bdev, fmode_t mode,
143 		     unsigned command, unsigned long argument)
144 {
145 	int i;
146 	struct gendisk *disk;
147 
148 	switch (command) {
149 	case CDROMMULTISESSION:
150 		pr_debug(PFX "Multisession CDs not supported\n");
151 		for (i = 0; i < sizeof(struct cdrom_multisession); i++)
152 			if (put_user(0, (char __user *)(argument + i)))
153 				return -EFAULT;
154 		return 0;
155 
156 	case CDROM_GET_CAPABILITY:
157 		disk = bdev->bd_disk;
158 
159 		if (bdev->bd_disk && (disk->flags & GENHD_FL_CD))
160 			return 0;
161 		return -EINVAL;
162 
163 	default:
164 		pr_debug(PFX "ioctl %08x not supported\n", command);
165 		return -EINVAL;
166 	}
167 }
168 
169 static const struct block_device_operations vdc_fops = {
170 	.owner		= THIS_MODULE,
171 	.getgeo		= vdc_getgeo,
172 	.ioctl		= vdc_ioctl,
173 };
174 
175 static void vdc_blk_queue_start(struct vdc_port *port)
176 {
177 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
178 
179 	/* restart blk queue when ring is half emptied. also called after
180 	 * handshake completes, so check for initial handshake before we've
181 	 * allocated a disk.
182 	 */
183 	if (port->disk && vdc_tx_dring_avail(dr) * 100 / VDC_TX_RING_SIZE >= 50)
184 		blk_mq_start_hw_queues(port->disk->queue);
185 }
186 
187 static void vdc_finish(struct vio_driver_state *vio, int err, int waiting_for)
188 {
189 	if (vio->cmp &&
190 	    (waiting_for == -1 ||
191 	     vio->cmp->waiting_for == waiting_for)) {
192 		vio->cmp->err = err;
193 		complete(&vio->cmp->com);
194 		vio->cmp = NULL;
195 	}
196 }
197 
198 static void vdc_handshake_complete(struct vio_driver_state *vio)
199 {
200 	struct vdc_port *port = to_vdc_port(vio);
201 
202 	cancel_delayed_work(&port->ldc_reset_timer_work);
203 	vdc_finish(vio, 0, WAITING_FOR_LINK_UP);
204 	vdc_blk_queue_start(port);
205 }
206 
207 static int vdc_handle_unknown(struct vdc_port *port, void *arg)
208 {
209 	struct vio_msg_tag *pkt = arg;
210 
211 	printk(KERN_ERR PFX "Received unknown msg [%02x:%02x:%04x:%08x]\n",
212 	       pkt->type, pkt->stype, pkt->stype_env, pkt->sid);
213 	printk(KERN_ERR PFX "Resetting connection.\n");
214 
215 	ldc_disconnect(port->vio.lp);
216 
217 	return -ECONNRESET;
218 }
219 
220 static int vdc_send_attr(struct vio_driver_state *vio)
221 {
222 	struct vdc_port *port = to_vdc_port(vio);
223 	struct vio_disk_attr_info pkt;
224 
225 	memset(&pkt, 0, sizeof(pkt));
226 
227 	pkt.tag.type = VIO_TYPE_CTRL;
228 	pkt.tag.stype = VIO_SUBTYPE_INFO;
229 	pkt.tag.stype_env = VIO_ATTR_INFO;
230 	pkt.tag.sid = vio_send_sid(vio);
231 
232 	pkt.xfer_mode = VIO_DRING_MODE;
233 	pkt.vdisk_block_size = port->vdisk_block_size;
234 	pkt.max_xfer_size = port->max_xfer_size;
235 
236 	viodbg(HS, "SEND ATTR xfer_mode[0x%x] blksz[%u] max_xfer[%llu]\n",
237 	       pkt.xfer_mode, pkt.vdisk_block_size, pkt.max_xfer_size);
238 
239 	return vio_ldc_send(&port->vio, &pkt, sizeof(pkt));
240 }
241 
242 static int vdc_handle_attr(struct vio_driver_state *vio, void *arg)
243 {
244 	struct vdc_port *port = to_vdc_port(vio);
245 	struct vio_disk_attr_info *pkt = arg;
246 
247 	viodbg(HS, "GOT ATTR stype[0x%x] ops[%llx] disk_size[%llu] disk_type[%x] "
248 	       "mtype[0x%x] xfer_mode[0x%x] blksz[%u] max_xfer[%llu]\n",
249 	       pkt->tag.stype, pkt->operations,
250 	       pkt->vdisk_size, pkt->vdisk_type, pkt->vdisk_mtype,
251 	       pkt->xfer_mode, pkt->vdisk_block_size,
252 	       pkt->max_xfer_size);
253 
254 	if (pkt->tag.stype == VIO_SUBTYPE_ACK) {
255 		switch (pkt->vdisk_type) {
256 		case VD_DISK_TYPE_DISK:
257 		case VD_DISK_TYPE_SLICE:
258 			break;
259 
260 		default:
261 			printk(KERN_ERR PFX "%s: Bogus vdisk_type 0x%x\n",
262 			       vio->name, pkt->vdisk_type);
263 			return -ECONNRESET;
264 		}
265 
266 		if (pkt->vdisk_block_size > port->vdisk_block_size) {
267 			printk(KERN_ERR PFX "%s: BLOCK size increased "
268 			       "%u --> %u\n",
269 			       vio->name,
270 			       port->vdisk_block_size, pkt->vdisk_block_size);
271 			return -ECONNRESET;
272 		}
273 
274 		port->operations = pkt->operations;
275 		port->vdisk_type = pkt->vdisk_type;
276 		if (vdc_version_supported(port, 1, 1)) {
277 			port->vdisk_size = pkt->vdisk_size;
278 			port->vdisk_mtype = pkt->vdisk_mtype;
279 		}
280 		if (pkt->max_xfer_size < port->max_xfer_size)
281 			port->max_xfer_size = pkt->max_xfer_size;
282 		port->vdisk_block_size = pkt->vdisk_block_size;
283 
284 		port->vdisk_phys_blksz = VDC_DEFAULT_BLK_SIZE;
285 		if (vdc_version_supported(port, 1, 2))
286 			port->vdisk_phys_blksz = pkt->phys_block_size;
287 
288 		return 0;
289 	} else {
290 		printk(KERN_ERR PFX "%s: Attribute NACK\n", vio->name);
291 
292 		return -ECONNRESET;
293 	}
294 }
295 
296 static void vdc_end_special(struct vdc_port *port, struct vio_disk_desc *desc)
297 {
298 	int err = desc->status;
299 
300 	vdc_finish(&port->vio, -err, WAITING_FOR_GEN_CMD);
301 }
302 
303 static void vdc_end_one(struct vdc_port *port, struct vio_dring_state *dr,
304 			unsigned int index)
305 {
306 	struct vio_disk_desc *desc = vio_dring_entry(dr, index);
307 	struct vdc_req_entry *rqe = &port->rq_arr[index];
308 	struct request *req;
309 
310 	if (unlikely(desc->hdr.state != VIO_DESC_DONE))
311 		return;
312 
313 	ldc_unmap(port->vio.lp, desc->cookies, desc->ncookies);
314 	desc->hdr.state = VIO_DESC_FREE;
315 	dr->cons = vio_dring_next(dr, index);
316 
317 	req = rqe->req;
318 	if (req == NULL) {
319 		vdc_end_special(port, desc);
320 		return;
321 	}
322 
323 	rqe->req = NULL;
324 
325 	blk_mq_end_request(req, desc->status ? BLK_STS_IOERR : 0);
326 
327 	vdc_blk_queue_start(port);
328 }
329 
330 static int vdc_ack(struct vdc_port *port, void *msgbuf)
331 {
332 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
333 	struct vio_dring_data *pkt = msgbuf;
334 
335 	if (unlikely(pkt->dring_ident != dr->ident ||
336 		     pkt->start_idx != pkt->end_idx ||
337 		     pkt->start_idx >= VDC_TX_RING_SIZE))
338 		return 0;
339 
340 	vdc_end_one(port, dr, pkt->start_idx);
341 
342 	return 0;
343 }
344 
345 static int vdc_nack(struct vdc_port *port, void *msgbuf)
346 {
347 	/* XXX Implement me XXX */
348 	return 0;
349 }
350 
351 static void vdc_event(void *arg, int event)
352 {
353 	struct vdc_port *port = arg;
354 	struct vio_driver_state *vio = &port->vio;
355 	unsigned long flags;
356 	int err;
357 
358 	spin_lock_irqsave(&vio->lock, flags);
359 
360 	if (unlikely(event == LDC_EVENT_RESET)) {
361 		vio_link_state_change(vio, event);
362 		queue_work(sunvdc_wq, &port->ldc_reset_work);
363 		goto out;
364 	}
365 
366 	if (unlikely(event == LDC_EVENT_UP)) {
367 		vio_link_state_change(vio, event);
368 		goto out;
369 	}
370 
371 	if (unlikely(event != LDC_EVENT_DATA_READY)) {
372 		pr_warn(PFX "Unexpected LDC event %d\n", event);
373 		goto out;
374 	}
375 
376 	err = 0;
377 	while (1) {
378 		union {
379 			struct vio_msg_tag tag;
380 			u64 raw[8];
381 		} msgbuf;
382 
383 		err = ldc_read(vio->lp, &msgbuf, sizeof(msgbuf));
384 		if (unlikely(err < 0)) {
385 			if (err == -ECONNRESET)
386 				vio_conn_reset(vio);
387 			break;
388 		}
389 		if (err == 0)
390 			break;
391 		viodbg(DATA, "TAG [%02x:%02x:%04x:%08x]\n",
392 		       msgbuf.tag.type,
393 		       msgbuf.tag.stype,
394 		       msgbuf.tag.stype_env,
395 		       msgbuf.tag.sid);
396 		err = vio_validate_sid(vio, &msgbuf.tag);
397 		if (err < 0)
398 			break;
399 
400 		if (likely(msgbuf.tag.type == VIO_TYPE_DATA)) {
401 			if (msgbuf.tag.stype == VIO_SUBTYPE_ACK)
402 				err = vdc_ack(port, &msgbuf);
403 			else if (msgbuf.tag.stype == VIO_SUBTYPE_NACK)
404 				err = vdc_nack(port, &msgbuf);
405 			else
406 				err = vdc_handle_unknown(port, &msgbuf);
407 		} else if (msgbuf.tag.type == VIO_TYPE_CTRL) {
408 			err = vio_control_pkt_engine(vio, &msgbuf);
409 		} else {
410 			err = vdc_handle_unknown(port, &msgbuf);
411 		}
412 		if (err < 0)
413 			break;
414 	}
415 	if (err < 0)
416 		vdc_finish(&port->vio, err, WAITING_FOR_ANY);
417 out:
418 	spin_unlock_irqrestore(&vio->lock, flags);
419 }
420 
421 static int __vdc_tx_trigger(struct vdc_port *port)
422 {
423 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
424 	struct vio_dring_data hdr = {
425 		.tag = {
426 			.type		= VIO_TYPE_DATA,
427 			.stype		= VIO_SUBTYPE_INFO,
428 			.stype_env	= VIO_DRING_DATA,
429 			.sid		= vio_send_sid(&port->vio),
430 		},
431 		.dring_ident		= dr->ident,
432 		.start_idx		= dr->prod,
433 		.end_idx		= dr->prod,
434 	};
435 	int err, delay;
436 	int retries = 0;
437 
438 	hdr.seq = dr->snd_nxt;
439 	delay = 1;
440 	do {
441 		err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
442 		if (err > 0) {
443 			dr->snd_nxt++;
444 			break;
445 		}
446 		udelay(delay);
447 		if ((delay <<= 1) > 128)
448 			delay = 128;
449 		if (retries++ > VDC_MAX_RETRIES)
450 			break;
451 	} while (err == -EAGAIN);
452 
453 	if (err == -ENOTCONN)
454 		vdc_ldc_reset(port);
455 	return err;
456 }
457 
458 static int __send_request(struct request *req)
459 {
460 	struct vdc_port *port = req->rq_disk->private_data;
461 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
462 	struct scatterlist sg[MAX_RING_COOKIES];
463 	struct vdc_req_entry *rqe;
464 	struct vio_disk_desc *desc;
465 	unsigned int map_perm;
466 	int nsg, err, i;
467 	u64 len;
468 	u8 op;
469 
470 	if (WARN_ON(port->ring_cookies > MAX_RING_COOKIES))
471 		return -EINVAL;
472 
473 	map_perm = LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_IO;
474 
475 	if (rq_data_dir(req) == READ) {
476 		map_perm |= LDC_MAP_W;
477 		op = VD_OP_BREAD;
478 	} else {
479 		map_perm |= LDC_MAP_R;
480 		op = VD_OP_BWRITE;
481 	}
482 
483 	sg_init_table(sg, port->ring_cookies);
484 	nsg = blk_rq_map_sg(req->q, req, sg);
485 
486 	len = 0;
487 	for (i = 0; i < nsg; i++)
488 		len += sg[i].length;
489 
490 	desc = vio_dring_cur(dr);
491 
492 	err = ldc_map_sg(port->vio.lp, sg, nsg,
493 			 desc->cookies, port->ring_cookies,
494 			 map_perm);
495 	if (err < 0) {
496 		printk(KERN_ERR PFX "ldc_map_sg() failure, err=%d.\n", err);
497 		return err;
498 	}
499 
500 	rqe = &port->rq_arr[dr->prod];
501 	rqe->req = req;
502 
503 	desc->hdr.ack = VIO_ACK_ENABLE;
504 	desc->req_id = port->req_id;
505 	desc->operation = op;
506 	if (port->vdisk_type == VD_DISK_TYPE_DISK) {
507 		desc->slice = 0xff;
508 	} else {
509 		desc->slice = 0;
510 	}
511 	desc->status = ~0;
512 	desc->offset = (blk_rq_pos(req) << 9) / port->vdisk_block_size;
513 	desc->size = len;
514 	desc->ncookies = err;
515 
516 	/* This has to be a non-SMP write barrier because we are writing
517 	 * to memory which is shared with the peer LDOM.
518 	 */
519 	wmb();
520 	desc->hdr.state = VIO_DESC_READY;
521 
522 	err = __vdc_tx_trigger(port);
523 	if (err < 0) {
524 		printk(KERN_ERR PFX "vdc_tx_trigger() failure, err=%d\n", err);
525 	} else {
526 		port->req_id++;
527 		dr->prod = vio_dring_next(dr, dr->prod);
528 	}
529 
530 	return err;
531 }
532 
533 static blk_status_t vdc_queue_rq(struct blk_mq_hw_ctx *hctx,
534 				 const struct blk_mq_queue_data *bd)
535 {
536 	struct vdc_port *port = hctx->queue->queuedata;
537 	struct vio_dring_state *dr;
538 	unsigned long flags;
539 
540 	dr = &port->vio.drings[VIO_DRIVER_TX_RING];
541 
542 	blk_mq_start_request(bd->rq);
543 
544 	spin_lock_irqsave(&port->vio.lock, flags);
545 
546 	/*
547 	 * Doing drain, just end the request in error
548 	 */
549 	if (unlikely(port->drain)) {
550 		spin_unlock_irqrestore(&port->vio.lock, flags);
551 		return BLK_STS_IOERR;
552 	}
553 
554 	if (unlikely(vdc_tx_dring_avail(dr) < 1)) {
555 		spin_unlock_irqrestore(&port->vio.lock, flags);
556 		blk_mq_stop_hw_queue(hctx);
557 		return BLK_STS_DEV_RESOURCE;
558 	}
559 
560 	if (__send_request(bd->rq) < 0) {
561 		spin_unlock_irqrestore(&port->vio.lock, flags);
562 		return BLK_STS_IOERR;
563 	}
564 
565 	spin_unlock_irqrestore(&port->vio.lock, flags);
566 	return BLK_STS_OK;
567 }
568 
569 static int generic_request(struct vdc_port *port, u8 op, void *buf, int len)
570 {
571 	struct vio_dring_state *dr;
572 	struct vio_completion comp;
573 	struct vio_disk_desc *desc;
574 	unsigned int map_perm;
575 	unsigned long flags;
576 	int op_len, err;
577 	void *req_buf;
578 
579 	if (!(((u64)1 << (u64)op) & port->operations))
580 		return -EOPNOTSUPP;
581 
582 	switch (op) {
583 	case VD_OP_BREAD:
584 	case VD_OP_BWRITE:
585 	default:
586 		return -EINVAL;
587 
588 	case VD_OP_FLUSH:
589 		op_len = 0;
590 		map_perm = 0;
591 		break;
592 
593 	case VD_OP_GET_WCE:
594 		op_len = sizeof(u32);
595 		map_perm = LDC_MAP_W;
596 		break;
597 
598 	case VD_OP_SET_WCE:
599 		op_len = sizeof(u32);
600 		map_perm = LDC_MAP_R;
601 		break;
602 
603 	case VD_OP_GET_VTOC:
604 		op_len = sizeof(struct vio_disk_vtoc);
605 		map_perm = LDC_MAP_W;
606 		break;
607 
608 	case VD_OP_SET_VTOC:
609 		op_len = sizeof(struct vio_disk_vtoc);
610 		map_perm = LDC_MAP_R;
611 		break;
612 
613 	case VD_OP_GET_DISKGEOM:
614 		op_len = sizeof(struct vio_disk_geom);
615 		map_perm = LDC_MAP_W;
616 		break;
617 
618 	case VD_OP_SET_DISKGEOM:
619 		op_len = sizeof(struct vio_disk_geom);
620 		map_perm = LDC_MAP_R;
621 		break;
622 
623 	case VD_OP_SCSICMD:
624 		op_len = 16;
625 		map_perm = LDC_MAP_RW;
626 		break;
627 
628 	case VD_OP_GET_DEVID:
629 		op_len = sizeof(struct vio_disk_devid);
630 		map_perm = LDC_MAP_W;
631 		break;
632 
633 	case VD_OP_GET_EFI:
634 	case VD_OP_SET_EFI:
635 		return -EOPNOTSUPP;
636 		break;
637 	};
638 
639 	map_perm |= LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_IO;
640 
641 	op_len = (op_len + 7) & ~7;
642 	req_buf = kzalloc(op_len, GFP_KERNEL);
643 	if (!req_buf)
644 		return -ENOMEM;
645 
646 	if (len > op_len)
647 		len = op_len;
648 
649 	if (map_perm & LDC_MAP_R)
650 		memcpy(req_buf, buf, len);
651 
652 	spin_lock_irqsave(&port->vio.lock, flags);
653 
654 	dr = &port->vio.drings[VIO_DRIVER_TX_RING];
655 
656 	/* XXX If we want to use this code generically we have to
657 	 * XXX handle TX ring exhaustion etc.
658 	 */
659 	desc = vio_dring_cur(dr);
660 
661 	err = ldc_map_single(port->vio.lp, req_buf, op_len,
662 			     desc->cookies, port->ring_cookies,
663 			     map_perm);
664 	if (err < 0) {
665 		spin_unlock_irqrestore(&port->vio.lock, flags);
666 		kfree(req_buf);
667 		return err;
668 	}
669 
670 	init_completion(&comp.com);
671 	comp.waiting_for = WAITING_FOR_GEN_CMD;
672 	port->vio.cmp = &comp;
673 
674 	desc->hdr.ack = VIO_ACK_ENABLE;
675 	desc->req_id = port->req_id;
676 	desc->operation = op;
677 	desc->slice = 0;
678 	desc->status = ~0;
679 	desc->offset = 0;
680 	desc->size = op_len;
681 	desc->ncookies = err;
682 
683 	/* This has to be a non-SMP write barrier because we are writing
684 	 * to memory which is shared with the peer LDOM.
685 	 */
686 	wmb();
687 	desc->hdr.state = VIO_DESC_READY;
688 
689 	err = __vdc_tx_trigger(port);
690 	if (err >= 0) {
691 		port->req_id++;
692 		dr->prod = vio_dring_next(dr, dr->prod);
693 		spin_unlock_irqrestore(&port->vio.lock, flags);
694 
695 		wait_for_completion(&comp.com);
696 		err = comp.err;
697 	} else {
698 		port->vio.cmp = NULL;
699 		spin_unlock_irqrestore(&port->vio.lock, flags);
700 	}
701 
702 	if (map_perm & LDC_MAP_W)
703 		memcpy(buf, req_buf, len);
704 
705 	kfree(req_buf);
706 
707 	return err;
708 }
709 
710 static int vdc_alloc_tx_ring(struct vdc_port *port)
711 {
712 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
713 	unsigned long len, entry_size;
714 	int ncookies;
715 	void *dring;
716 
717 	entry_size = sizeof(struct vio_disk_desc) +
718 		(sizeof(struct ldc_trans_cookie) * port->ring_cookies);
719 	len = (VDC_TX_RING_SIZE * entry_size);
720 
721 	ncookies = VIO_MAX_RING_COOKIES;
722 	dring = ldc_alloc_exp_dring(port->vio.lp, len,
723 				    dr->cookies, &ncookies,
724 				    (LDC_MAP_SHADOW |
725 				     LDC_MAP_DIRECT |
726 				     LDC_MAP_RW));
727 	if (IS_ERR(dring))
728 		return PTR_ERR(dring);
729 
730 	dr->base = dring;
731 	dr->entry_size = entry_size;
732 	dr->num_entries = VDC_TX_RING_SIZE;
733 	dr->prod = dr->cons = 0;
734 	dr->pending = VDC_TX_RING_SIZE;
735 	dr->ncookies = ncookies;
736 
737 	return 0;
738 }
739 
740 static void vdc_free_tx_ring(struct vdc_port *port)
741 {
742 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
743 
744 	if (dr->base) {
745 		ldc_free_exp_dring(port->vio.lp, dr->base,
746 				   (dr->entry_size * dr->num_entries),
747 				   dr->cookies, dr->ncookies);
748 		dr->base = NULL;
749 		dr->entry_size = 0;
750 		dr->num_entries = 0;
751 		dr->pending = 0;
752 		dr->ncookies = 0;
753 	}
754 }
755 
756 static int vdc_port_up(struct vdc_port *port)
757 {
758 	struct vio_completion comp;
759 
760 	init_completion(&comp.com);
761 	comp.err = 0;
762 	comp.waiting_for = WAITING_FOR_LINK_UP;
763 	port->vio.cmp = &comp;
764 
765 	vio_port_up(&port->vio);
766 	wait_for_completion(&comp.com);
767 	return comp.err;
768 }
769 
770 static void vdc_port_down(struct vdc_port *port)
771 {
772 	ldc_disconnect(port->vio.lp);
773 	ldc_unbind(port->vio.lp);
774 	vdc_free_tx_ring(port);
775 	vio_ldc_free(&port->vio);
776 }
777 
778 static const struct blk_mq_ops vdc_mq_ops = {
779 	.queue_rq	= vdc_queue_rq,
780 };
781 
782 static void cleanup_queue(struct request_queue *q)
783 {
784 	struct vdc_port *port = q->queuedata;
785 
786 	blk_cleanup_queue(q);
787 	blk_mq_free_tag_set(&port->tag_set);
788 }
789 
790 static struct request_queue *init_queue(struct vdc_port *port)
791 {
792 	struct request_queue *q;
793 
794 	q = blk_mq_init_sq_queue(&port->tag_set, &vdc_mq_ops, VDC_TX_RING_SIZE,
795 					BLK_MQ_F_SHOULD_MERGE);
796 	if (IS_ERR(q))
797 		return q;
798 
799 	q->queuedata = port;
800 	return q;
801 }
802 
803 static int probe_disk(struct vdc_port *port)
804 {
805 	struct request_queue *q;
806 	struct gendisk *g;
807 	int err;
808 
809 	err = vdc_port_up(port);
810 	if (err)
811 		return err;
812 
813 	/* Using version 1.2 means vdisk_phys_blksz should be set unless the
814 	 * disk is reserved by another system.
815 	 */
816 	if (vdc_version_supported(port, 1, 2) && !port->vdisk_phys_blksz)
817 		return -ENODEV;
818 
819 	if (vdc_version_supported(port, 1, 1)) {
820 		/* vdisk_size should be set during the handshake, if it wasn't
821 		 * then the underlying disk is reserved by another system
822 		 */
823 		if (port->vdisk_size == -1)
824 			return -ENODEV;
825 	} else {
826 		struct vio_disk_geom geom;
827 
828 		err = generic_request(port, VD_OP_GET_DISKGEOM,
829 				      &geom, sizeof(geom));
830 		if (err < 0) {
831 			printk(KERN_ERR PFX "VD_OP_GET_DISKGEOM returns "
832 			       "error %d\n", err);
833 			return err;
834 		}
835 		port->vdisk_size = ((u64)geom.num_cyl *
836 				    (u64)geom.num_hd *
837 				    (u64)geom.num_sec);
838 	}
839 
840 	q = init_queue(port);
841 	if (IS_ERR(q)) {
842 		printk(KERN_ERR PFX "%s: Could not allocate queue.\n",
843 		       port->vio.name);
844 		return PTR_ERR(q);
845 	}
846 	g = alloc_disk(1 << PARTITION_SHIFT);
847 	if (!g) {
848 		printk(KERN_ERR PFX "%s: Could not allocate gendisk.\n",
849 		       port->vio.name);
850 		cleanup_queue(q);
851 		return -ENOMEM;
852 	}
853 
854 	port->disk = g;
855 
856 	/* Each segment in a request is up to an aligned page in size. */
857 	blk_queue_segment_boundary(q, PAGE_SIZE - 1);
858 	blk_queue_max_segment_size(q, PAGE_SIZE);
859 
860 	blk_queue_max_segments(q, port->ring_cookies);
861 	blk_queue_max_hw_sectors(q, port->max_xfer_size);
862 	g->major = vdc_major;
863 	g->first_minor = port->vio.vdev->dev_no << PARTITION_SHIFT;
864 	strcpy(g->disk_name, port->disk_name);
865 
866 	g->fops = &vdc_fops;
867 	g->queue = q;
868 	g->private_data = port;
869 
870 	set_capacity(g, port->vdisk_size);
871 
872 	if (vdc_version_supported(port, 1, 1)) {
873 		switch (port->vdisk_mtype) {
874 		case VD_MEDIA_TYPE_CD:
875 			pr_info(PFX "Virtual CDROM %s\n", port->disk_name);
876 			g->flags |= GENHD_FL_CD;
877 			g->flags |= GENHD_FL_REMOVABLE;
878 			set_disk_ro(g, 1);
879 			break;
880 
881 		case VD_MEDIA_TYPE_DVD:
882 			pr_info(PFX "Virtual DVD %s\n", port->disk_name);
883 			g->flags |= GENHD_FL_CD;
884 			g->flags |= GENHD_FL_REMOVABLE;
885 			set_disk_ro(g, 1);
886 			break;
887 
888 		case VD_MEDIA_TYPE_FIXED:
889 			pr_info(PFX "Virtual Hard disk %s\n", port->disk_name);
890 			break;
891 		}
892 	}
893 
894 	blk_queue_physical_block_size(q, port->vdisk_phys_blksz);
895 
896 	pr_info(PFX "%s: %u sectors (%u MB) protocol %d.%d\n",
897 	       g->disk_name,
898 	       port->vdisk_size, (port->vdisk_size >> (20 - 9)),
899 	       port->vio.ver.major, port->vio.ver.minor);
900 
901 	device_add_disk(&port->vio.vdev->dev, g, NULL);
902 
903 	return 0;
904 }
905 
906 static struct ldc_channel_config vdc_ldc_cfg = {
907 	.event		= vdc_event,
908 	.mtu		= 64,
909 	.mode		= LDC_MODE_UNRELIABLE,
910 };
911 
912 static struct vio_driver_ops vdc_vio_ops = {
913 	.send_attr		= vdc_send_attr,
914 	.handle_attr		= vdc_handle_attr,
915 	.handshake_complete	= vdc_handshake_complete,
916 };
917 
918 static void print_version(void)
919 {
920 	static int version_printed;
921 
922 	if (version_printed++ == 0)
923 		printk(KERN_INFO "%s", version);
924 }
925 
926 struct vdc_check_port_data {
927 	int	dev_no;
928 	char	*type;
929 };
930 
931 static int vdc_device_probed(struct device *dev, void *arg)
932 {
933 	struct vio_dev *vdev = to_vio_dev(dev);
934 	struct vdc_check_port_data *port_data;
935 
936 	port_data = (struct vdc_check_port_data *)arg;
937 
938 	if ((vdev->dev_no == port_data->dev_no) &&
939 	    (!(strcmp((char *)&vdev->type, port_data->type))) &&
940 		dev_get_drvdata(dev)) {
941 		/* This device has already been configured
942 		 * by vdc_port_probe()
943 		 */
944 		return 1;
945 	} else {
946 		return 0;
947 	}
948 }
949 
950 /* Determine whether the VIO device is part of an mpgroup
951  * by locating all the virtual-device-port nodes associated
952  * with the parent virtual-device node for the VIO device
953  * and checking whether any of these nodes are vdc-ports
954  * which have already been configured.
955  *
956  * Returns true if this device is part of an mpgroup and has
957  * already been probed.
958  */
959 static bool vdc_port_mpgroup_check(struct vio_dev *vdev)
960 {
961 	struct vdc_check_port_data port_data;
962 	struct device *dev;
963 
964 	port_data.dev_no = vdev->dev_no;
965 	port_data.type = (char *)&vdev->type;
966 
967 	dev = device_find_child(vdev->dev.parent, &port_data,
968 				vdc_device_probed);
969 
970 	if (dev)
971 		return true;
972 
973 	return false;
974 }
975 
976 static int vdc_port_probe(struct vio_dev *vdev, const struct vio_device_id *id)
977 {
978 	struct mdesc_handle *hp;
979 	struct vdc_port *port;
980 	int err;
981 	const u64 *ldc_timeout;
982 
983 	print_version();
984 
985 	hp = mdesc_grab();
986 
987 	err = -ENODEV;
988 	if ((vdev->dev_no << PARTITION_SHIFT) & ~(u64)MINORMASK) {
989 		printk(KERN_ERR PFX "Port id [%llu] too large.\n",
990 		       vdev->dev_no);
991 		goto err_out_release_mdesc;
992 	}
993 
994 	/* Check if this device is part of an mpgroup */
995 	if (vdc_port_mpgroup_check(vdev)) {
996 		printk(KERN_WARNING
997 			"VIO: Ignoring extra vdisk port %s",
998 			dev_name(&vdev->dev));
999 		goto err_out_release_mdesc;
1000 	}
1001 
1002 	port = kzalloc(sizeof(*port), GFP_KERNEL);
1003 	err = -ENOMEM;
1004 	if (!port) {
1005 		printk(KERN_ERR PFX "Cannot allocate vdc_port.\n");
1006 		goto err_out_release_mdesc;
1007 	}
1008 
1009 	if (vdev->dev_no >= 26)
1010 		snprintf(port->disk_name, sizeof(port->disk_name),
1011 			 VDCBLK_NAME "%c%c",
1012 			 'a' + ((int)vdev->dev_no / 26) - 1,
1013 			 'a' + ((int)vdev->dev_no % 26));
1014 	else
1015 		snprintf(port->disk_name, sizeof(port->disk_name),
1016 			 VDCBLK_NAME "%c", 'a' + ((int)vdev->dev_no % 26));
1017 	port->vdisk_size = -1;
1018 
1019 	/* Actual wall time may be double due to do_generic_file_read() doing
1020 	 * a readahead I/O first, and once that fails it will try to read a
1021 	 * single page.
1022 	 */
1023 	ldc_timeout = mdesc_get_property(hp, vdev->mp, "vdc-timeout", NULL);
1024 	port->ldc_timeout = ldc_timeout ? *ldc_timeout : 0;
1025 	INIT_DELAYED_WORK(&port->ldc_reset_timer_work, vdc_ldc_reset_timer_work);
1026 	INIT_WORK(&port->ldc_reset_work, vdc_ldc_reset_work);
1027 
1028 	err = vio_driver_init(&port->vio, vdev, VDEV_DISK,
1029 			      vdc_versions, ARRAY_SIZE(vdc_versions),
1030 			      &vdc_vio_ops, port->disk_name);
1031 	if (err)
1032 		goto err_out_free_port;
1033 
1034 	port->vdisk_block_size = VDC_DEFAULT_BLK_SIZE;
1035 	port->max_xfer_size = MAX_XFER_SIZE;
1036 	port->ring_cookies = MAX_RING_COOKIES;
1037 
1038 	err = vio_ldc_alloc(&port->vio, &vdc_ldc_cfg, port);
1039 	if (err)
1040 		goto err_out_free_port;
1041 
1042 	err = vdc_alloc_tx_ring(port);
1043 	if (err)
1044 		goto err_out_free_ldc;
1045 
1046 	err = probe_disk(port);
1047 	if (err)
1048 		goto err_out_free_tx_ring;
1049 
1050 	/* Note that the device driver_data is used to determine
1051 	 * whether the port has been probed.
1052 	 */
1053 	dev_set_drvdata(&vdev->dev, port);
1054 
1055 	mdesc_release(hp);
1056 
1057 	return 0;
1058 
1059 err_out_free_tx_ring:
1060 	vdc_free_tx_ring(port);
1061 
1062 err_out_free_ldc:
1063 	vio_ldc_free(&port->vio);
1064 
1065 err_out_free_port:
1066 	kfree(port);
1067 
1068 err_out_release_mdesc:
1069 	mdesc_release(hp);
1070 	return err;
1071 }
1072 
1073 static int vdc_port_remove(struct vio_dev *vdev)
1074 {
1075 	struct vdc_port *port = dev_get_drvdata(&vdev->dev);
1076 
1077 	if (port) {
1078 		blk_mq_stop_hw_queues(port->disk->queue);
1079 
1080 		flush_work(&port->ldc_reset_work);
1081 		cancel_delayed_work_sync(&port->ldc_reset_timer_work);
1082 		del_timer_sync(&port->vio.timer);
1083 
1084 		del_gendisk(port->disk);
1085 		cleanup_queue(port->disk->queue);
1086 		put_disk(port->disk);
1087 		port->disk = NULL;
1088 
1089 		vdc_free_tx_ring(port);
1090 		vio_ldc_free(&port->vio);
1091 
1092 		dev_set_drvdata(&vdev->dev, NULL);
1093 
1094 		kfree(port);
1095 	}
1096 	return 0;
1097 }
1098 
1099 static void vdc_requeue_inflight(struct vdc_port *port)
1100 {
1101 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1102 	u32 idx;
1103 
1104 	for (idx = dr->cons; idx != dr->prod; idx = vio_dring_next(dr, idx)) {
1105 		struct vio_disk_desc *desc = vio_dring_entry(dr, idx);
1106 		struct vdc_req_entry *rqe = &port->rq_arr[idx];
1107 		struct request *req;
1108 
1109 		ldc_unmap(port->vio.lp, desc->cookies, desc->ncookies);
1110 		desc->hdr.state = VIO_DESC_FREE;
1111 		dr->cons = vio_dring_next(dr, idx);
1112 
1113 		req = rqe->req;
1114 		if (req == NULL) {
1115 			vdc_end_special(port, desc);
1116 			continue;
1117 		}
1118 
1119 		rqe->req = NULL;
1120 		blk_mq_requeue_request(req, false);
1121 	}
1122 }
1123 
1124 static void vdc_queue_drain(struct vdc_port *port)
1125 {
1126 	struct request_queue *q = port->disk->queue;
1127 
1128 	/*
1129 	 * Mark the queue as draining, then freeze/quiesce to ensure
1130 	 * that all existing requests are seen in ->queue_rq() and killed
1131 	 */
1132 	port->drain = 1;
1133 	spin_unlock_irq(&port->vio.lock);
1134 
1135 	blk_mq_freeze_queue(q);
1136 	blk_mq_quiesce_queue(q);
1137 
1138 	spin_lock_irq(&port->vio.lock);
1139 	port->drain = 0;
1140 	blk_mq_unquiesce_queue(q);
1141 	blk_mq_unfreeze_queue(q);
1142 }
1143 
1144 static void vdc_ldc_reset_timer_work(struct work_struct *work)
1145 {
1146 	struct vdc_port *port;
1147 	struct vio_driver_state *vio;
1148 
1149 	port = container_of(work, struct vdc_port, ldc_reset_timer_work.work);
1150 	vio = &port->vio;
1151 
1152 	spin_lock_irq(&vio->lock);
1153 	if (!(port->vio.hs_state & VIO_HS_COMPLETE)) {
1154 		pr_warn(PFX "%s ldc down %llu seconds, draining queue\n",
1155 			port->disk_name, port->ldc_timeout);
1156 		vdc_queue_drain(port);
1157 		vdc_blk_queue_start(port);
1158 	}
1159 	spin_unlock_irq(&vio->lock);
1160 }
1161 
1162 static void vdc_ldc_reset_work(struct work_struct *work)
1163 {
1164 	struct vdc_port *port;
1165 	struct vio_driver_state *vio;
1166 	unsigned long flags;
1167 
1168 	port = container_of(work, struct vdc_port, ldc_reset_work);
1169 	vio = &port->vio;
1170 
1171 	spin_lock_irqsave(&vio->lock, flags);
1172 	vdc_ldc_reset(port);
1173 	spin_unlock_irqrestore(&vio->lock, flags);
1174 }
1175 
1176 static void vdc_ldc_reset(struct vdc_port *port)
1177 {
1178 	int err;
1179 
1180 	assert_spin_locked(&port->vio.lock);
1181 
1182 	pr_warn(PFX "%s ldc link reset\n", port->disk_name);
1183 	blk_mq_stop_hw_queues(port->disk->queue);
1184 	vdc_requeue_inflight(port);
1185 	vdc_port_down(port);
1186 
1187 	err = vio_ldc_alloc(&port->vio, &vdc_ldc_cfg, port);
1188 	if (err) {
1189 		pr_err(PFX "%s vio_ldc_alloc:%d\n", port->disk_name, err);
1190 		return;
1191 	}
1192 
1193 	err = vdc_alloc_tx_ring(port);
1194 	if (err) {
1195 		pr_err(PFX "%s vio_alloc_tx_ring:%d\n", port->disk_name, err);
1196 		goto err_free_ldc;
1197 	}
1198 
1199 	if (port->ldc_timeout)
1200 		mod_delayed_work(system_wq, &port->ldc_reset_timer_work,
1201 			  round_jiffies(jiffies + HZ * port->ldc_timeout));
1202 	mod_timer(&port->vio.timer, round_jiffies(jiffies + HZ));
1203 	return;
1204 
1205 err_free_ldc:
1206 	vio_ldc_free(&port->vio);
1207 }
1208 
1209 static const struct vio_device_id vdc_port_match[] = {
1210 	{
1211 		.type = "vdc-port",
1212 	},
1213 	{},
1214 };
1215 MODULE_DEVICE_TABLE(vio, vdc_port_match);
1216 
1217 static struct vio_driver vdc_port_driver = {
1218 	.id_table	= vdc_port_match,
1219 	.probe		= vdc_port_probe,
1220 	.remove		= vdc_port_remove,
1221 	.name		= "vdc_port",
1222 };
1223 
1224 static int __init vdc_init(void)
1225 {
1226 	int err;
1227 
1228 	sunvdc_wq = alloc_workqueue("sunvdc", 0, 0);
1229 	if (!sunvdc_wq)
1230 		return -ENOMEM;
1231 
1232 	err = register_blkdev(0, VDCBLK_NAME);
1233 	if (err < 0)
1234 		goto out_free_wq;
1235 
1236 	vdc_major = err;
1237 
1238 	err = vio_register_driver(&vdc_port_driver);
1239 	if (err)
1240 		goto out_unregister_blkdev;
1241 
1242 	return 0;
1243 
1244 out_unregister_blkdev:
1245 	unregister_blkdev(vdc_major, VDCBLK_NAME);
1246 	vdc_major = 0;
1247 
1248 out_free_wq:
1249 	destroy_workqueue(sunvdc_wq);
1250 	return err;
1251 }
1252 
1253 static void __exit vdc_exit(void)
1254 {
1255 	vio_unregister_driver(&vdc_port_driver);
1256 	unregister_blkdev(vdc_major, VDCBLK_NAME);
1257 	destroy_workqueue(sunvdc_wq);
1258 }
1259 
1260 module_init(vdc_init);
1261 module_exit(vdc_exit);
1262