xref: /openbmc/linux/drivers/block/sunvdc.c (revision be58f710)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* sunvdc.c: Sun LDOM Virtual Disk Client.
3  *
4  * Copyright (C) 2007, 2008 David S. Miller <davem@davemloft.net>
5  */
6 
7 #include <linux/module.h>
8 #include <linux/kernel.h>
9 #include <linux/types.h>
10 #include <linux/blk-mq.h>
11 #include <linux/hdreg.h>
12 #include <linux/genhd.h>
13 #include <linux/cdrom.h>
14 #include <linux/slab.h>
15 #include <linux/spinlock.h>
16 #include <linux/completion.h>
17 #include <linux/delay.h>
18 #include <linux/init.h>
19 #include <linux/list.h>
20 #include <linux/scatterlist.h>
21 
22 #include <asm/vio.h>
23 #include <asm/ldc.h>
24 
25 #define DRV_MODULE_NAME		"sunvdc"
26 #define PFX DRV_MODULE_NAME	": "
27 #define DRV_MODULE_VERSION	"1.2"
28 #define DRV_MODULE_RELDATE	"November 24, 2014"
29 
30 static char version[] =
31 	DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
32 MODULE_AUTHOR("David S. Miller (davem@davemloft.net)");
33 MODULE_DESCRIPTION("Sun LDOM virtual disk client driver");
34 MODULE_LICENSE("GPL");
35 MODULE_VERSION(DRV_MODULE_VERSION);
36 
37 #define VDC_TX_RING_SIZE	512
38 #define VDC_DEFAULT_BLK_SIZE	512
39 
40 #define MAX_XFER_BLKS		(128 * 1024)
41 #define MAX_XFER_SIZE		(MAX_XFER_BLKS / VDC_DEFAULT_BLK_SIZE)
42 #define MAX_RING_COOKIES	((MAX_XFER_BLKS / PAGE_SIZE) + 2)
43 
44 #define WAITING_FOR_LINK_UP	0x01
45 #define WAITING_FOR_TX_SPACE	0x02
46 #define WAITING_FOR_GEN_CMD	0x04
47 #define WAITING_FOR_ANY		-1
48 
49 #define	VDC_MAX_RETRIES	10
50 
51 static struct workqueue_struct *sunvdc_wq;
52 
53 struct vdc_req_entry {
54 	struct request		*req;
55 };
56 
57 struct vdc_port {
58 	struct vio_driver_state	vio;
59 
60 	struct gendisk		*disk;
61 
62 	struct vdc_completion	*cmp;
63 
64 	u64			req_id;
65 	u64			seq;
66 	struct vdc_req_entry	rq_arr[VDC_TX_RING_SIZE];
67 
68 	unsigned long		ring_cookies;
69 
70 	u64			max_xfer_size;
71 	u32			vdisk_block_size;
72 	u32			drain;
73 
74 	u64			ldc_timeout;
75 	struct delayed_work	ldc_reset_timer_work;
76 	struct work_struct	ldc_reset_work;
77 
78 	/* The server fills these in for us in the disk attribute
79 	 * ACK packet.
80 	 */
81 	u64			operations;
82 	u32			vdisk_size;
83 	u8			vdisk_type;
84 	u8			vdisk_mtype;
85 	u32			vdisk_phys_blksz;
86 
87 	struct blk_mq_tag_set	tag_set;
88 
89 	char			disk_name[32];
90 };
91 
92 static void vdc_ldc_reset(struct vdc_port *port);
93 static void vdc_ldc_reset_work(struct work_struct *work);
94 static void vdc_ldc_reset_timer_work(struct work_struct *work);
95 
96 static inline struct vdc_port *to_vdc_port(struct vio_driver_state *vio)
97 {
98 	return container_of(vio, struct vdc_port, vio);
99 }
100 
101 /* Ordered from largest major to lowest */
102 static struct vio_version vdc_versions[] = {
103 	{ .major = 1, .minor = 2 },
104 	{ .major = 1, .minor = 1 },
105 	{ .major = 1, .minor = 0 },
106 };
107 
108 static inline int vdc_version_supported(struct vdc_port *port,
109 					u16 major, u16 minor)
110 {
111 	return port->vio.ver.major == major && port->vio.ver.minor >= minor;
112 }
113 
114 #define VDCBLK_NAME	"vdisk"
115 static int vdc_major;
116 #define PARTITION_SHIFT	3
117 
118 static inline u32 vdc_tx_dring_avail(struct vio_dring_state *dr)
119 {
120 	return vio_dring_avail(dr, VDC_TX_RING_SIZE);
121 }
122 
123 static int vdc_getgeo(struct block_device *bdev, struct hd_geometry *geo)
124 {
125 	struct gendisk *disk = bdev->bd_disk;
126 	sector_t nsect = get_capacity(disk);
127 	sector_t cylinders = nsect;
128 
129 	geo->heads = 0xff;
130 	geo->sectors = 0x3f;
131 	sector_div(cylinders, geo->heads * geo->sectors);
132 	geo->cylinders = cylinders;
133 	if ((sector_t)(geo->cylinders + 1) * geo->heads * geo->sectors < nsect)
134 		geo->cylinders = 0xffff;
135 
136 	return 0;
137 }
138 
139 /* Add ioctl/CDROM_GET_CAPABILITY to support cdrom_id in udev
140  * when vdisk_mtype is VD_MEDIA_TYPE_CD or VD_MEDIA_TYPE_DVD.
141  * Needed to be able to install inside an ldom from an iso image.
142  */
143 static int vdc_ioctl(struct block_device *bdev, fmode_t mode,
144 		     unsigned command, unsigned long argument)
145 {
146 	int i;
147 	struct gendisk *disk;
148 
149 	switch (command) {
150 	case CDROMMULTISESSION:
151 		pr_debug(PFX "Multisession CDs not supported\n");
152 		for (i = 0; i < sizeof(struct cdrom_multisession); i++)
153 			if (put_user(0, (char __user *)(argument + i)))
154 				return -EFAULT;
155 		return 0;
156 
157 	case CDROM_GET_CAPABILITY:
158 		disk = bdev->bd_disk;
159 
160 		if (bdev->bd_disk && (disk->flags & GENHD_FL_CD))
161 			return 0;
162 		return -EINVAL;
163 
164 	default:
165 		pr_debug(PFX "ioctl %08x not supported\n", command);
166 		return -EINVAL;
167 	}
168 }
169 
170 static const struct block_device_operations vdc_fops = {
171 	.owner		= THIS_MODULE,
172 	.getgeo		= vdc_getgeo,
173 	.ioctl		= vdc_ioctl,
174 	.compat_ioctl	= blkdev_compat_ptr_ioctl,
175 };
176 
177 static void vdc_blk_queue_start(struct vdc_port *port)
178 {
179 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
180 
181 	/* restart blk queue when ring is half emptied. also called after
182 	 * handshake completes, so check for initial handshake before we've
183 	 * allocated a disk.
184 	 */
185 	if (port->disk && vdc_tx_dring_avail(dr) * 100 / VDC_TX_RING_SIZE >= 50)
186 		blk_mq_start_stopped_hw_queues(port->disk->queue, true);
187 }
188 
189 static void vdc_finish(struct vio_driver_state *vio, int err, int waiting_for)
190 {
191 	if (vio->cmp &&
192 	    (waiting_for == -1 ||
193 	     vio->cmp->waiting_for == waiting_for)) {
194 		vio->cmp->err = err;
195 		complete(&vio->cmp->com);
196 		vio->cmp = NULL;
197 	}
198 }
199 
200 static void vdc_handshake_complete(struct vio_driver_state *vio)
201 {
202 	struct vdc_port *port = to_vdc_port(vio);
203 
204 	cancel_delayed_work(&port->ldc_reset_timer_work);
205 	vdc_finish(vio, 0, WAITING_FOR_LINK_UP);
206 	vdc_blk_queue_start(port);
207 }
208 
209 static int vdc_handle_unknown(struct vdc_port *port, void *arg)
210 {
211 	struct vio_msg_tag *pkt = arg;
212 
213 	printk(KERN_ERR PFX "Received unknown msg [%02x:%02x:%04x:%08x]\n",
214 	       pkt->type, pkt->stype, pkt->stype_env, pkt->sid);
215 	printk(KERN_ERR PFX "Resetting connection.\n");
216 
217 	ldc_disconnect(port->vio.lp);
218 
219 	return -ECONNRESET;
220 }
221 
222 static int vdc_send_attr(struct vio_driver_state *vio)
223 {
224 	struct vdc_port *port = to_vdc_port(vio);
225 	struct vio_disk_attr_info pkt;
226 
227 	memset(&pkt, 0, sizeof(pkt));
228 
229 	pkt.tag.type = VIO_TYPE_CTRL;
230 	pkt.tag.stype = VIO_SUBTYPE_INFO;
231 	pkt.tag.stype_env = VIO_ATTR_INFO;
232 	pkt.tag.sid = vio_send_sid(vio);
233 
234 	pkt.xfer_mode = VIO_DRING_MODE;
235 	pkt.vdisk_block_size = port->vdisk_block_size;
236 	pkt.max_xfer_size = port->max_xfer_size;
237 
238 	viodbg(HS, "SEND ATTR xfer_mode[0x%x] blksz[%u] max_xfer[%llu]\n",
239 	       pkt.xfer_mode, pkt.vdisk_block_size, pkt.max_xfer_size);
240 
241 	return vio_ldc_send(&port->vio, &pkt, sizeof(pkt));
242 }
243 
244 static int vdc_handle_attr(struct vio_driver_state *vio, void *arg)
245 {
246 	struct vdc_port *port = to_vdc_port(vio);
247 	struct vio_disk_attr_info *pkt = arg;
248 
249 	viodbg(HS, "GOT ATTR stype[0x%x] ops[%llx] disk_size[%llu] disk_type[%x] "
250 	       "mtype[0x%x] xfer_mode[0x%x] blksz[%u] max_xfer[%llu]\n",
251 	       pkt->tag.stype, pkt->operations,
252 	       pkt->vdisk_size, pkt->vdisk_type, pkt->vdisk_mtype,
253 	       pkt->xfer_mode, pkt->vdisk_block_size,
254 	       pkt->max_xfer_size);
255 
256 	if (pkt->tag.stype == VIO_SUBTYPE_ACK) {
257 		switch (pkt->vdisk_type) {
258 		case VD_DISK_TYPE_DISK:
259 		case VD_DISK_TYPE_SLICE:
260 			break;
261 
262 		default:
263 			printk(KERN_ERR PFX "%s: Bogus vdisk_type 0x%x\n",
264 			       vio->name, pkt->vdisk_type);
265 			return -ECONNRESET;
266 		}
267 
268 		if (pkt->vdisk_block_size > port->vdisk_block_size) {
269 			printk(KERN_ERR PFX "%s: BLOCK size increased "
270 			       "%u --> %u\n",
271 			       vio->name,
272 			       port->vdisk_block_size, pkt->vdisk_block_size);
273 			return -ECONNRESET;
274 		}
275 
276 		port->operations = pkt->operations;
277 		port->vdisk_type = pkt->vdisk_type;
278 		if (vdc_version_supported(port, 1, 1)) {
279 			port->vdisk_size = pkt->vdisk_size;
280 			port->vdisk_mtype = pkt->vdisk_mtype;
281 		}
282 		if (pkt->max_xfer_size < port->max_xfer_size)
283 			port->max_xfer_size = pkt->max_xfer_size;
284 		port->vdisk_block_size = pkt->vdisk_block_size;
285 
286 		port->vdisk_phys_blksz = VDC_DEFAULT_BLK_SIZE;
287 		if (vdc_version_supported(port, 1, 2))
288 			port->vdisk_phys_blksz = pkt->phys_block_size;
289 
290 		return 0;
291 	} else {
292 		printk(KERN_ERR PFX "%s: Attribute NACK\n", vio->name);
293 
294 		return -ECONNRESET;
295 	}
296 }
297 
298 static void vdc_end_special(struct vdc_port *port, struct vio_disk_desc *desc)
299 {
300 	int err = desc->status;
301 
302 	vdc_finish(&port->vio, -err, WAITING_FOR_GEN_CMD);
303 }
304 
305 static void vdc_end_one(struct vdc_port *port, struct vio_dring_state *dr,
306 			unsigned int index)
307 {
308 	struct vio_disk_desc *desc = vio_dring_entry(dr, index);
309 	struct vdc_req_entry *rqe = &port->rq_arr[index];
310 	struct request *req;
311 
312 	if (unlikely(desc->hdr.state != VIO_DESC_DONE))
313 		return;
314 
315 	ldc_unmap(port->vio.lp, desc->cookies, desc->ncookies);
316 	desc->hdr.state = VIO_DESC_FREE;
317 	dr->cons = vio_dring_next(dr, index);
318 
319 	req = rqe->req;
320 	if (req == NULL) {
321 		vdc_end_special(port, desc);
322 		return;
323 	}
324 
325 	rqe->req = NULL;
326 
327 	blk_mq_end_request(req, desc->status ? BLK_STS_IOERR : 0);
328 
329 	vdc_blk_queue_start(port);
330 }
331 
332 static int vdc_ack(struct vdc_port *port, void *msgbuf)
333 {
334 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
335 	struct vio_dring_data *pkt = msgbuf;
336 
337 	if (unlikely(pkt->dring_ident != dr->ident ||
338 		     pkt->start_idx != pkt->end_idx ||
339 		     pkt->start_idx >= VDC_TX_RING_SIZE))
340 		return 0;
341 
342 	vdc_end_one(port, dr, pkt->start_idx);
343 
344 	return 0;
345 }
346 
347 static int vdc_nack(struct vdc_port *port, void *msgbuf)
348 {
349 	/* XXX Implement me XXX */
350 	return 0;
351 }
352 
353 static void vdc_event(void *arg, int event)
354 {
355 	struct vdc_port *port = arg;
356 	struct vio_driver_state *vio = &port->vio;
357 	unsigned long flags;
358 	int err;
359 
360 	spin_lock_irqsave(&vio->lock, flags);
361 
362 	if (unlikely(event == LDC_EVENT_RESET)) {
363 		vio_link_state_change(vio, event);
364 		queue_work(sunvdc_wq, &port->ldc_reset_work);
365 		goto out;
366 	}
367 
368 	if (unlikely(event == LDC_EVENT_UP)) {
369 		vio_link_state_change(vio, event);
370 		goto out;
371 	}
372 
373 	if (unlikely(event != LDC_EVENT_DATA_READY)) {
374 		pr_warn(PFX "Unexpected LDC event %d\n", event);
375 		goto out;
376 	}
377 
378 	err = 0;
379 	while (1) {
380 		union {
381 			struct vio_msg_tag tag;
382 			u64 raw[8];
383 		} msgbuf;
384 
385 		err = ldc_read(vio->lp, &msgbuf, sizeof(msgbuf));
386 		if (unlikely(err < 0)) {
387 			if (err == -ECONNRESET)
388 				vio_conn_reset(vio);
389 			break;
390 		}
391 		if (err == 0)
392 			break;
393 		viodbg(DATA, "TAG [%02x:%02x:%04x:%08x]\n",
394 		       msgbuf.tag.type,
395 		       msgbuf.tag.stype,
396 		       msgbuf.tag.stype_env,
397 		       msgbuf.tag.sid);
398 		err = vio_validate_sid(vio, &msgbuf.tag);
399 		if (err < 0)
400 			break;
401 
402 		if (likely(msgbuf.tag.type == VIO_TYPE_DATA)) {
403 			if (msgbuf.tag.stype == VIO_SUBTYPE_ACK)
404 				err = vdc_ack(port, &msgbuf);
405 			else if (msgbuf.tag.stype == VIO_SUBTYPE_NACK)
406 				err = vdc_nack(port, &msgbuf);
407 			else
408 				err = vdc_handle_unknown(port, &msgbuf);
409 		} else if (msgbuf.tag.type == VIO_TYPE_CTRL) {
410 			err = vio_control_pkt_engine(vio, &msgbuf);
411 		} else {
412 			err = vdc_handle_unknown(port, &msgbuf);
413 		}
414 		if (err < 0)
415 			break;
416 	}
417 	if (err < 0)
418 		vdc_finish(&port->vio, err, WAITING_FOR_ANY);
419 out:
420 	spin_unlock_irqrestore(&vio->lock, flags);
421 }
422 
423 static int __vdc_tx_trigger(struct vdc_port *port)
424 {
425 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
426 	struct vio_dring_data hdr = {
427 		.tag = {
428 			.type		= VIO_TYPE_DATA,
429 			.stype		= VIO_SUBTYPE_INFO,
430 			.stype_env	= VIO_DRING_DATA,
431 			.sid		= vio_send_sid(&port->vio),
432 		},
433 		.dring_ident		= dr->ident,
434 		.start_idx		= dr->prod,
435 		.end_idx		= dr->prod,
436 	};
437 	int err, delay;
438 	int retries = 0;
439 
440 	hdr.seq = dr->snd_nxt;
441 	delay = 1;
442 	do {
443 		err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
444 		if (err > 0) {
445 			dr->snd_nxt++;
446 			break;
447 		}
448 		udelay(delay);
449 		if ((delay <<= 1) > 128)
450 			delay = 128;
451 		if (retries++ > VDC_MAX_RETRIES)
452 			break;
453 	} while (err == -EAGAIN);
454 
455 	if (err == -ENOTCONN)
456 		vdc_ldc_reset(port);
457 	return err;
458 }
459 
460 static int __send_request(struct request *req)
461 {
462 	struct vdc_port *port = req->rq_disk->private_data;
463 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
464 	struct scatterlist sg[MAX_RING_COOKIES];
465 	struct vdc_req_entry *rqe;
466 	struct vio_disk_desc *desc;
467 	unsigned int map_perm;
468 	int nsg, err, i;
469 	u64 len;
470 	u8 op;
471 
472 	if (WARN_ON(port->ring_cookies > MAX_RING_COOKIES))
473 		return -EINVAL;
474 
475 	map_perm = LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_IO;
476 
477 	if (rq_data_dir(req) == READ) {
478 		map_perm |= LDC_MAP_W;
479 		op = VD_OP_BREAD;
480 	} else {
481 		map_perm |= LDC_MAP_R;
482 		op = VD_OP_BWRITE;
483 	}
484 
485 	sg_init_table(sg, port->ring_cookies);
486 	nsg = blk_rq_map_sg(req->q, req, sg);
487 
488 	len = 0;
489 	for (i = 0; i < nsg; i++)
490 		len += sg[i].length;
491 
492 	desc = vio_dring_cur(dr);
493 
494 	err = ldc_map_sg(port->vio.lp, sg, nsg,
495 			 desc->cookies, port->ring_cookies,
496 			 map_perm);
497 	if (err < 0) {
498 		printk(KERN_ERR PFX "ldc_map_sg() failure, err=%d.\n", err);
499 		return err;
500 	}
501 
502 	rqe = &port->rq_arr[dr->prod];
503 	rqe->req = req;
504 
505 	desc->hdr.ack = VIO_ACK_ENABLE;
506 	desc->req_id = port->req_id;
507 	desc->operation = op;
508 	if (port->vdisk_type == VD_DISK_TYPE_DISK) {
509 		desc->slice = 0xff;
510 	} else {
511 		desc->slice = 0;
512 	}
513 	desc->status = ~0;
514 	desc->offset = (blk_rq_pos(req) << 9) / port->vdisk_block_size;
515 	desc->size = len;
516 	desc->ncookies = err;
517 
518 	/* This has to be a non-SMP write barrier because we are writing
519 	 * to memory which is shared with the peer LDOM.
520 	 */
521 	wmb();
522 	desc->hdr.state = VIO_DESC_READY;
523 
524 	err = __vdc_tx_trigger(port);
525 	if (err < 0) {
526 		printk(KERN_ERR PFX "vdc_tx_trigger() failure, err=%d\n", err);
527 	} else {
528 		port->req_id++;
529 		dr->prod = vio_dring_next(dr, dr->prod);
530 	}
531 
532 	return err;
533 }
534 
535 static blk_status_t vdc_queue_rq(struct blk_mq_hw_ctx *hctx,
536 				 const struct blk_mq_queue_data *bd)
537 {
538 	struct vdc_port *port = hctx->queue->queuedata;
539 	struct vio_dring_state *dr;
540 	unsigned long flags;
541 
542 	dr = &port->vio.drings[VIO_DRIVER_TX_RING];
543 
544 	blk_mq_start_request(bd->rq);
545 
546 	spin_lock_irqsave(&port->vio.lock, flags);
547 
548 	/*
549 	 * Doing drain, just end the request in error
550 	 */
551 	if (unlikely(port->drain)) {
552 		spin_unlock_irqrestore(&port->vio.lock, flags);
553 		return BLK_STS_IOERR;
554 	}
555 
556 	if (unlikely(vdc_tx_dring_avail(dr) < 1)) {
557 		spin_unlock_irqrestore(&port->vio.lock, flags);
558 		blk_mq_stop_hw_queue(hctx);
559 		return BLK_STS_DEV_RESOURCE;
560 	}
561 
562 	if (__send_request(bd->rq) < 0) {
563 		spin_unlock_irqrestore(&port->vio.lock, flags);
564 		return BLK_STS_IOERR;
565 	}
566 
567 	spin_unlock_irqrestore(&port->vio.lock, flags);
568 	return BLK_STS_OK;
569 }
570 
571 static int generic_request(struct vdc_port *port, u8 op, void *buf, int len)
572 {
573 	struct vio_dring_state *dr;
574 	struct vio_completion comp;
575 	struct vio_disk_desc *desc;
576 	unsigned int map_perm;
577 	unsigned long flags;
578 	int op_len, err;
579 	void *req_buf;
580 
581 	if (!(((u64)1 << (u64)op) & port->operations))
582 		return -EOPNOTSUPP;
583 
584 	switch (op) {
585 	case VD_OP_BREAD:
586 	case VD_OP_BWRITE:
587 	default:
588 		return -EINVAL;
589 
590 	case VD_OP_FLUSH:
591 		op_len = 0;
592 		map_perm = 0;
593 		break;
594 
595 	case VD_OP_GET_WCE:
596 		op_len = sizeof(u32);
597 		map_perm = LDC_MAP_W;
598 		break;
599 
600 	case VD_OP_SET_WCE:
601 		op_len = sizeof(u32);
602 		map_perm = LDC_MAP_R;
603 		break;
604 
605 	case VD_OP_GET_VTOC:
606 		op_len = sizeof(struct vio_disk_vtoc);
607 		map_perm = LDC_MAP_W;
608 		break;
609 
610 	case VD_OP_SET_VTOC:
611 		op_len = sizeof(struct vio_disk_vtoc);
612 		map_perm = LDC_MAP_R;
613 		break;
614 
615 	case VD_OP_GET_DISKGEOM:
616 		op_len = sizeof(struct vio_disk_geom);
617 		map_perm = LDC_MAP_W;
618 		break;
619 
620 	case VD_OP_SET_DISKGEOM:
621 		op_len = sizeof(struct vio_disk_geom);
622 		map_perm = LDC_MAP_R;
623 		break;
624 
625 	case VD_OP_SCSICMD:
626 		op_len = 16;
627 		map_perm = LDC_MAP_RW;
628 		break;
629 
630 	case VD_OP_GET_DEVID:
631 		op_len = sizeof(struct vio_disk_devid);
632 		map_perm = LDC_MAP_W;
633 		break;
634 
635 	case VD_OP_GET_EFI:
636 	case VD_OP_SET_EFI:
637 		return -EOPNOTSUPP;
638 	}
639 
640 	map_perm |= LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_IO;
641 
642 	op_len = (op_len + 7) & ~7;
643 	req_buf = kzalloc(op_len, GFP_KERNEL);
644 	if (!req_buf)
645 		return -ENOMEM;
646 
647 	if (len > op_len)
648 		len = op_len;
649 
650 	if (map_perm & LDC_MAP_R)
651 		memcpy(req_buf, buf, len);
652 
653 	spin_lock_irqsave(&port->vio.lock, flags);
654 
655 	dr = &port->vio.drings[VIO_DRIVER_TX_RING];
656 
657 	/* XXX If we want to use this code generically we have to
658 	 * XXX handle TX ring exhaustion etc.
659 	 */
660 	desc = vio_dring_cur(dr);
661 
662 	err = ldc_map_single(port->vio.lp, req_buf, op_len,
663 			     desc->cookies, port->ring_cookies,
664 			     map_perm);
665 	if (err < 0) {
666 		spin_unlock_irqrestore(&port->vio.lock, flags);
667 		kfree(req_buf);
668 		return err;
669 	}
670 
671 	init_completion(&comp.com);
672 	comp.waiting_for = WAITING_FOR_GEN_CMD;
673 	port->vio.cmp = &comp;
674 
675 	desc->hdr.ack = VIO_ACK_ENABLE;
676 	desc->req_id = port->req_id;
677 	desc->operation = op;
678 	desc->slice = 0;
679 	desc->status = ~0;
680 	desc->offset = 0;
681 	desc->size = op_len;
682 	desc->ncookies = err;
683 
684 	/* This has to be a non-SMP write barrier because we are writing
685 	 * to memory which is shared with the peer LDOM.
686 	 */
687 	wmb();
688 	desc->hdr.state = VIO_DESC_READY;
689 
690 	err = __vdc_tx_trigger(port);
691 	if (err >= 0) {
692 		port->req_id++;
693 		dr->prod = vio_dring_next(dr, dr->prod);
694 		spin_unlock_irqrestore(&port->vio.lock, flags);
695 
696 		wait_for_completion(&comp.com);
697 		err = comp.err;
698 	} else {
699 		port->vio.cmp = NULL;
700 		spin_unlock_irqrestore(&port->vio.lock, flags);
701 	}
702 
703 	if (map_perm & LDC_MAP_W)
704 		memcpy(buf, req_buf, len);
705 
706 	kfree(req_buf);
707 
708 	return err;
709 }
710 
711 static int vdc_alloc_tx_ring(struct vdc_port *port)
712 {
713 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
714 	unsigned long len, entry_size;
715 	int ncookies;
716 	void *dring;
717 
718 	entry_size = sizeof(struct vio_disk_desc) +
719 		(sizeof(struct ldc_trans_cookie) * port->ring_cookies);
720 	len = (VDC_TX_RING_SIZE * entry_size);
721 
722 	ncookies = VIO_MAX_RING_COOKIES;
723 	dring = ldc_alloc_exp_dring(port->vio.lp, len,
724 				    dr->cookies, &ncookies,
725 				    (LDC_MAP_SHADOW |
726 				     LDC_MAP_DIRECT |
727 				     LDC_MAP_RW));
728 	if (IS_ERR(dring))
729 		return PTR_ERR(dring);
730 
731 	dr->base = dring;
732 	dr->entry_size = entry_size;
733 	dr->num_entries = VDC_TX_RING_SIZE;
734 	dr->prod = dr->cons = 0;
735 	dr->pending = VDC_TX_RING_SIZE;
736 	dr->ncookies = ncookies;
737 
738 	return 0;
739 }
740 
741 static void vdc_free_tx_ring(struct vdc_port *port)
742 {
743 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
744 
745 	if (dr->base) {
746 		ldc_free_exp_dring(port->vio.lp, dr->base,
747 				   (dr->entry_size * dr->num_entries),
748 				   dr->cookies, dr->ncookies);
749 		dr->base = NULL;
750 		dr->entry_size = 0;
751 		dr->num_entries = 0;
752 		dr->pending = 0;
753 		dr->ncookies = 0;
754 	}
755 }
756 
757 static int vdc_port_up(struct vdc_port *port)
758 {
759 	struct vio_completion comp;
760 
761 	init_completion(&comp.com);
762 	comp.err = 0;
763 	comp.waiting_for = WAITING_FOR_LINK_UP;
764 	port->vio.cmp = &comp;
765 
766 	vio_port_up(&port->vio);
767 	wait_for_completion(&comp.com);
768 	return comp.err;
769 }
770 
771 static void vdc_port_down(struct vdc_port *port)
772 {
773 	ldc_disconnect(port->vio.lp);
774 	ldc_unbind(port->vio.lp);
775 	vdc_free_tx_ring(port);
776 	vio_ldc_free(&port->vio);
777 }
778 
779 static const struct blk_mq_ops vdc_mq_ops = {
780 	.queue_rq	= vdc_queue_rq,
781 };
782 
783 static int probe_disk(struct vdc_port *port)
784 {
785 	struct request_queue *q;
786 	struct gendisk *g;
787 	int err;
788 
789 	err = vdc_port_up(port);
790 	if (err)
791 		return err;
792 
793 	/* Using version 1.2 means vdisk_phys_blksz should be set unless the
794 	 * disk is reserved by another system.
795 	 */
796 	if (vdc_version_supported(port, 1, 2) && !port->vdisk_phys_blksz)
797 		return -ENODEV;
798 
799 	if (vdc_version_supported(port, 1, 1)) {
800 		/* vdisk_size should be set during the handshake, if it wasn't
801 		 * then the underlying disk is reserved by another system
802 		 */
803 		if (port->vdisk_size == -1)
804 			return -ENODEV;
805 	} else {
806 		struct vio_disk_geom geom;
807 
808 		err = generic_request(port, VD_OP_GET_DISKGEOM,
809 				      &geom, sizeof(geom));
810 		if (err < 0) {
811 			printk(KERN_ERR PFX "VD_OP_GET_DISKGEOM returns "
812 			       "error %d\n", err);
813 			return err;
814 		}
815 		port->vdisk_size = ((u64)geom.num_cyl *
816 				    (u64)geom.num_hd *
817 				    (u64)geom.num_sec);
818 	}
819 
820 	err = blk_mq_alloc_sq_tag_set(&port->tag_set, &vdc_mq_ops,
821 			VDC_TX_RING_SIZE, BLK_MQ_F_SHOULD_MERGE);
822 	if (err)
823 		return err;
824 
825 	g = blk_mq_alloc_disk(&port->tag_set, port);
826 	if (IS_ERR(g)) {
827 		printk(KERN_ERR PFX "%s: Could not allocate gendisk.\n",
828 		       port->vio.name);
829 		blk_mq_free_tag_set(&port->tag_set);
830 		return PTR_ERR(g);
831 	}
832 
833 	port->disk = g;
834 	q = g->queue;
835 
836 	/* Each segment in a request is up to an aligned page in size. */
837 	blk_queue_segment_boundary(q, PAGE_SIZE - 1);
838 	blk_queue_max_segment_size(q, PAGE_SIZE);
839 
840 	blk_queue_max_segments(q, port->ring_cookies);
841 	blk_queue_max_hw_sectors(q, port->max_xfer_size);
842 	g->major = vdc_major;
843 	g->first_minor = port->vio.vdev->dev_no << PARTITION_SHIFT;
844 	g->minors = 1 << PARTITION_SHIFT;
845 	strcpy(g->disk_name, port->disk_name);
846 
847 	g->fops = &vdc_fops;
848 	g->queue = q;
849 	g->private_data = port;
850 
851 	set_capacity(g, port->vdisk_size);
852 
853 	if (vdc_version_supported(port, 1, 1)) {
854 		switch (port->vdisk_mtype) {
855 		case VD_MEDIA_TYPE_CD:
856 			pr_info(PFX "Virtual CDROM %s\n", port->disk_name);
857 			g->flags |= GENHD_FL_CD;
858 			g->flags |= GENHD_FL_REMOVABLE;
859 			set_disk_ro(g, 1);
860 			break;
861 
862 		case VD_MEDIA_TYPE_DVD:
863 			pr_info(PFX "Virtual DVD %s\n", port->disk_name);
864 			g->flags |= GENHD_FL_CD;
865 			g->flags |= GENHD_FL_REMOVABLE;
866 			set_disk_ro(g, 1);
867 			break;
868 
869 		case VD_MEDIA_TYPE_FIXED:
870 			pr_info(PFX "Virtual Hard disk %s\n", port->disk_name);
871 			break;
872 		}
873 	}
874 
875 	blk_queue_physical_block_size(q, port->vdisk_phys_blksz);
876 
877 	pr_info(PFX "%s: %u sectors (%u MB) protocol %d.%d\n",
878 	       g->disk_name,
879 	       port->vdisk_size, (port->vdisk_size >> (20 - 9)),
880 	       port->vio.ver.major, port->vio.ver.minor);
881 
882 	device_add_disk(&port->vio.vdev->dev, g, NULL);
883 
884 	return 0;
885 }
886 
887 static struct ldc_channel_config vdc_ldc_cfg = {
888 	.event		= vdc_event,
889 	.mtu		= 64,
890 	.mode		= LDC_MODE_UNRELIABLE,
891 };
892 
893 static struct vio_driver_ops vdc_vio_ops = {
894 	.send_attr		= vdc_send_attr,
895 	.handle_attr		= vdc_handle_attr,
896 	.handshake_complete	= vdc_handshake_complete,
897 };
898 
899 static void print_version(void)
900 {
901 	static int version_printed;
902 
903 	if (version_printed++ == 0)
904 		printk(KERN_INFO "%s", version);
905 }
906 
907 struct vdc_check_port_data {
908 	int	dev_no;
909 	char	*type;
910 };
911 
912 static int vdc_device_probed(struct device *dev, void *arg)
913 {
914 	struct vio_dev *vdev = to_vio_dev(dev);
915 	struct vdc_check_port_data *port_data;
916 
917 	port_data = (struct vdc_check_port_data *)arg;
918 
919 	if ((vdev->dev_no == port_data->dev_no) &&
920 	    (!(strcmp((char *)&vdev->type, port_data->type))) &&
921 		dev_get_drvdata(dev)) {
922 		/* This device has already been configured
923 		 * by vdc_port_probe()
924 		 */
925 		return 1;
926 	} else {
927 		return 0;
928 	}
929 }
930 
931 /* Determine whether the VIO device is part of an mpgroup
932  * by locating all the virtual-device-port nodes associated
933  * with the parent virtual-device node for the VIO device
934  * and checking whether any of these nodes are vdc-ports
935  * which have already been configured.
936  *
937  * Returns true if this device is part of an mpgroup and has
938  * already been probed.
939  */
940 static bool vdc_port_mpgroup_check(struct vio_dev *vdev)
941 {
942 	struct vdc_check_port_data port_data;
943 	struct device *dev;
944 
945 	port_data.dev_no = vdev->dev_no;
946 	port_data.type = (char *)&vdev->type;
947 
948 	dev = device_find_child(vdev->dev.parent, &port_data,
949 				vdc_device_probed);
950 
951 	if (dev)
952 		return true;
953 
954 	return false;
955 }
956 
957 static int vdc_port_probe(struct vio_dev *vdev, const struct vio_device_id *id)
958 {
959 	struct mdesc_handle *hp;
960 	struct vdc_port *port;
961 	int err;
962 	const u64 *ldc_timeout;
963 
964 	print_version();
965 
966 	hp = mdesc_grab();
967 
968 	err = -ENODEV;
969 	if ((vdev->dev_no << PARTITION_SHIFT) & ~(u64)MINORMASK) {
970 		printk(KERN_ERR PFX "Port id [%llu] too large.\n",
971 		       vdev->dev_no);
972 		goto err_out_release_mdesc;
973 	}
974 
975 	/* Check if this device is part of an mpgroup */
976 	if (vdc_port_mpgroup_check(vdev)) {
977 		printk(KERN_WARNING
978 			"VIO: Ignoring extra vdisk port %s",
979 			dev_name(&vdev->dev));
980 		goto err_out_release_mdesc;
981 	}
982 
983 	port = kzalloc(sizeof(*port), GFP_KERNEL);
984 	if (!port) {
985 		err = -ENOMEM;
986 		goto err_out_release_mdesc;
987 	}
988 
989 	if (vdev->dev_no >= 26)
990 		snprintf(port->disk_name, sizeof(port->disk_name),
991 			 VDCBLK_NAME "%c%c",
992 			 'a' + ((int)vdev->dev_no / 26) - 1,
993 			 'a' + ((int)vdev->dev_no % 26));
994 	else
995 		snprintf(port->disk_name, sizeof(port->disk_name),
996 			 VDCBLK_NAME "%c", 'a' + ((int)vdev->dev_no % 26));
997 	port->vdisk_size = -1;
998 
999 	/* Actual wall time may be double due to do_generic_file_read() doing
1000 	 * a readahead I/O first, and once that fails it will try to read a
1001 	 * single page.
1002 	 */
1003 	ldc_timeout = mdesc_get_property(hp, vdev->mp, "vdc-timeout", NULL);
1004 	port->ldc_timeout = ldc_timeout ? *ldc_timeout : 0;
1005 	INIT_DELAYED_WORK(&port->ldc_reset_timer_work, vdc_ldc_reset_timer_work);
1006 	INIT_WORK(&port->ldc_reset_work, vdc_ldc_reset_work);
1007 
1008 	err = vio_driver_init(&port->vio, vdev, VDEV_DISK,
1009 			      vdc_versions, ARRAY_SIZE(vdc_versions),
1010 			      &vdc_vio_ops, port->disk_name);
1011 	if (err)
1012 		goto err_out_free_port;
1013 
1014 	port->vdisk_block_size = VDC_DEFAULT_BLK_SIZE;
1015 	port->max_xfer_size = MAX_XFER_SIZE;
1016 	port->ring_cookies = MAX_RING_COOKIES;
1017 
1018 	err = vio_ldc_alloc(&port->vio, &vdc_ldc_cfg, port);
1019 	if (err)
1020 		goto err_out_free_port;
1021 
1022 	err = vdc_alloc_tx_ring(port);
1023 	if (err)
1024 		goto err_out_free_ldc;
1025 
1026 	err = probe_disk(port);
1027 	if (err)
1028 		goto err_out_free_tx_ring;
1029 
1030 	/* Note that the device driver_data is used to determine
1031 	 * whether the port has been probed.
1032 	 */
1033 	dev_set_drvdata(&vdev->dev, port);
1034 
1035 	mdesc_release(hp);
1036 
1037 	return 0;
1038 
1039 err_out_free_tx_ring:
1040 	vdc_free_tx_ring(port);
1041 
1042 err_out_free_ldc:
1043 	vio_ldc_free(&port->vio);
1044 
1045 err_out_free_port:
1046 	kfree(port);
1047 
1048 err_out_release_mdesc:
1049 	mdesc_release(hp);
1050 	return err;
1051 }
1052 
1053 static void vdc_port_remove(struct vio_dev *vdev)
1054 {
1055 	struct vdc_port *port = dev_get_drvdata(&vdev->dev);
1056 
1057 	if (port) {
1058 		blk_mq_stop_hw_queues(port->disk->queue);
1059 
1060 		flush_work(&port->ldc_reset_work);
1061 		cancel_delayed_work_sync(&port->ldc_reset_timer_work);
1062 		del_timer_sync(&port->vio.timer);
1063 
1064 		del_gendisk(port->disk);
1065 		blk_cleanup_disk(port->disk);
1066 		blk_mq_free_tag_set(&port->tag_set);
1067 
1068 		vdc_free_tx_ring(port);
1069 		vio_ldc_free(&port->vio);
1070 
1071 		dev_set_drvdata(&vdev->dev, NULL);
1072 
1073 		kfree(port);
1074 	}
1075 }
1076 
1077 static void vdc_requeue_inflight(struct vdc_port *port)
1078 {
1079 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1080 	u32 idx;
1081 
1082 	for (idx = dr->cons; idx != dr->prod; idx = vio_dring_next(dr, idx)) {
1083 		struct vio_disk_desc *desc = vio_dring_entry(dr, idx);
1084 		struct vdc_req_entry *rqe = &port->rq_arr[idx];
1085 		struct request *req;
1086 
1087 		ldc_unmap(port->vio.lp, desc->cookies, desc->ncookies);
1088 		desc->hdr.state = VIO_DESC_FREE;
1089 		dr->cons = vio_dring_next(dr, idx);
1090 
1091 		req = rqe->req;
1092 		if (req == NULL) {
1093 			vdc_end_special(port, desc);
1094 			continue;
1095 		}
1096 
1097 		rqe->req = NULL;
1098 		blk_mq_requeue_request(req, false);
1099 	}
1100 }
1101 
1102 static void vdc_queue_drain(struct vdc_port *port)
1103 {
1104 	struct request_queue *q = port->disk->queue;
1105 
1106 	/*
1107 	 * Mark the queue as draining, then freeze/quiesce to ensure
1108 	 * that all existing requests are seen in ->queue_rq() and killed
1109 	 */
1110 	port->drain = 1;
1111 	spin_unlock_irq(&port->vio.lock);
1112 
1113 	blk_mq_freeze_queue(q);
1114 	blk_mq_quiesce_queue(q);
1115 
1116 	spin_lock_irq(&port->vio.lock);
1117 	port->drain = 0;
1118 	blk_mq_unquiesce_queue(q);
1119 	blk_mq_unfreeze_queue(q);
1120 }
1121 
1122 static void vdc_ldc_reset_timer_work(struct work_struct *work)
1123 {
1124 	struct vdc_port *port;
1125 	struct vio_driver_state *vio;
1126 
1127 	port = container_of(work, struct vdc_port, ldc_reset_timer_work.work);
1128 	vio = &port->vio;
1129 
1130 	spin_lock_irq(&vio->lock);
1131 	if (!(port->vio.hs_state & VIO_HS_COMPLETE)) {
1132 		pr_warn(PFX "%s ldc down %llu seconds, draining queue\n",
1133 			port->disk_name, port->ldc_timeout);
1134 		vdc_queue_drain(port);
1135 		vdc_blk_queue_start(port);
1136 	}
1137 	spin_unlock_irq(&vio->lock);
1138 }
1139 
1140 static void vdc_ldc_reset_work(struct work_struct *work)
1141 {
1142 	struct vdc_port *port;
1143 	struct vio_driver_state *vio;
1144 	unsigned long flags;
1145 
1146 	port = container_of(work, struct vdc_port, ldc_reset_work);
1147 	vio = &port->vio;
1148 
1149 	spin_lock_irqsave(&vio->lock, flags);
1150 	vdc_ldc_reset(port);
1151 	spin_unlock_irqrestore(&vio->lock, flags);
1152 }
1153 
1154 static void vdc_ldc_reset(struct vdc_port *port)
1155 {
1156 	int err;
1157 
1158 	assert_spin_locked(&port->vio.lock);
1159 
1160 	pr_warn(PFX "%s ldc link reset\n", port->disk_name);
1161 	blk_mq_stop_hw_queues(port->disk->queue);
1162 	vdc_requeue_inflight(port);
1163 	vdc_port_down(port);
1164 
1165 	err = vio_ldc_alloc(&port->vio, &vdc_ldc_cfg, port);
1166 	if (err) {
1167 		pr_err(PFX "%s vio_ldc_alloc:%d\n", port->disk_name, err);
1168 		return;
1169 	}
1170 
1171 	err = vdc_alloc_tx_ring(port);
1172 	if (err) {
1173 		pr_err(PFX "%s vio_alloc_tx_ring:%d\n", port->disk_name, err);
1174 		goto err_free_ldc;
1175 	}
1176 
1177 	if (port->ldc_timeout)
1178 		mod_delayed_work(system_wq, &port->ldc_reset_timer_work,
1179 			  round_jiffies(jiffies + HZ * port->ldc_timeout));
1180 	mod_timer(&port->vio.timer, round_jiffies(jiffies + HZ));
1181 	return;
1182 
1183 err_free_ldc:
1184 	vio_ldc_free(&port->vio);
1185 }
1186 
1187 static const struct vio_device_id vdc_port_match[] = {
1188 	{
1189 		.type = "vdc-port",
1190 	},
1191 	{},
1192 };
1193 MODULE_DEVICE_TABLE(vio, vdc_port_match);
1194 
1195 static struct vio_driver vdc_port_driver = {
1196 	.id_table	= vdc_port_match,
1197 	.probe		= vdc_port_probe,
1198 	.remove		= vdc_port_remove,
1199 	.name		= "vdc_port",
1200 };
1201 
1202 static int __init vdc_init(void)
1203 {
1204 	int err;
1205 
1206 	sunvdc_wq = alloc_workqueue("sunvdc", 0, 0);
1207 	if (!sunvdc_wq)
1208 		return -ENOMEM;
1209 
1210 	err = register_blkdev(0, VDCBLK_NAME);
1211 	if (err < 0)
1212 		goto out_free_wq;
1213 
1214 	vdc_major = err;
1215 
1216 	err = vio_register_driver(&vdc_port_driver);
1217 	if (err)
1218 		goto out_unregister_blkdev;
1219 
1220 	return 0;
1221 
1222 out_unregister_blkdev:
1223 	unregister_blkdev(vdc_major, VDCBLK_NAME);
1224 	vdc_major = 0;
1225 
1226 out_free_wq:
1227 	destroy_workqueue(sunvdc_wq);
1228 	return err;
1229 }
1230 
1231 static void __exit vdc_exit(void)
1232 {
1233 	vio_unregister_driver(&vdc_port_driver);
1234 	unregister_blkdev(vdc_major, VDCBLK_NAME);
1235 	destroy_workqueue(sunvdc_wq);
1236 }
1237 
1238 module_init(vdc_init);
1239 module_exit(vdc_exit);
1240