xref: /openbmc/linux/drivers/block/sunvdc.c (revision f71a261a)
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/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 	struct vdc_port *port = bdev->bd_disk->private_data;
146 	int i;
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 		if (!vdc_version_supported(port, 1, 1))
158 			return -EINVAL;
159 		switch (port->vdisk_mtype) {
160 		case VD_MEDIA_TYPE_CD:
161 		case VD_MEDIA_TYPE_DVD:
162 			return 0;
163 		default:
164 			return -EINVAL;
165 		}
166 	default:
167 		pr_debug(PFX "ioctl %08x not supported\n", command);
168 		return -EINVAL;
169 	}
170 }
171 
172 static const struct block_device_operations vdc_fops = {
173 	.owner		= THIS_MODULE,
174 	.getgeo		= vdc_getgeo,
175 	.ioctl		= vdc_ioctl,
176 	.compat_ioctl	= blkdev_compat_ptr_ioctl,
177 };
178 
179 static void vdc_blk_queue_start(struct vdc_port *port)
180 {
181 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
182 
183 	/* restart blk queue when ring is half emptied. also called after
184 	 * handshake completes, so check for initial handshake before we've
185 	 * allocated a disk.
186 	 */
187 	if (port->disk && vdc_tx_dring_avail(dr) * 100 / VDC_TX_RING_SIZE >= 50)
188 		blk_mq_start_stopped_hw_queues(port->disk->queue, true);
189 }
190 
191 static void vdc_finish(struct vio_driver_state *vio, int err, int waiting_for)
192 {
193 	if (vio->cmp &&
194 	    (waiting_for == -1 ||
195 	     vio->cmp->waiting_for == waiting_for)) {
196 		vio->cmp->err = err;
197 		complete(&vio->cmp->com);
198 		vio->cmp = NULL;
199 	}
200 }
201 
202 static void vdc_handshake_complete(struct vio_driver_state *vio)
203 {
204 	struct vdc_port *port = to_vdc_port(vio);
205 
206 	cancel_delayed_work(&port->ldc_reset_timer_work);
207 	vdc_finish(vio, 0, WAITING_FOR_LINK_UP);
208 	vdc_blk_queue_start(port);
209 }
210 
211 static int vdc_handle_unknown(struct vdc_port *port, void *arg)
212 {
213 	struct vio_msg_tag *pkt = arg;
214 
215 	printk(KERN_ERR PFX "Received unknown msg [%02x:%02x:%04x:%08x]\n",
216 	       pkt->type, pkt->stype, pkt->stype_env, pkt->sid);
217 	printk(KERN_ERR PFX "Resetting connection.\n");
218 
219 	ldc_disconnect(port->vio.lp);
220 
221 	return -ECONNRESET;
222 }
223 
224 static int vdc_send_attr(struct vio_driver_state *vio)
225 {
226 	struct vdc_port *port = to_vdc_port(vio);
227 	struct vio_disk_attr_info pkt;
228 
229 	memset(&pkt, 0, sizeof(pkt));
230 
231 	pkt.tag.type = VIO_TYPE_CTRL;
232 	pkt.tag.stype = VIO_SUBTYPE_INFO;
233 	pkt.tag.stype_env = VIO_ATTR_INFO;
234 	pkt.tag.sid = vio_send_sid(vio);
235 
236 	pkt.xfer_mode = VIO_DRING_MODE;
237 	pkt.vdisk_block_size = port->vdisk_block_size;
238 	pkt.max_xfer_size = port->max_xfer_size;
239 
240 	viodbg(HS, "SEND ATTR xfer_mode[0x%x] blksz[%u] max_xfer[%llu]\n",
241 	       pkt.xfer_mode, pkt.vdisk_block_size, pkt.max_xfer_size);
242 
243 	return vio_ldc_send(&port->vio, &pkt, sizeof(pkt));
244 }
245 
246 static int vdc_handle_attr(struct vio_driver_state *vio, void *arg)
247 {
248 	struct vdc_port *port = to_vdc_port(vio);
249 	struct vio_disk_attr_info *pkt = arg;
250 
251 	viodbg(HS, "GOT ATTR stype[0x%x] ops[%llx] disk_size[%llu] disk_type[%x] "
252 	       "mtype[0x%x] xfer_mode[0x%x] blksz[%u] max_xfer[%llu]\n",
253 	       pkt->tag.stype, pkt->operations,
254 	       pkt->vdisk_size, pkt->vdisk_type, pkt->vdisk_mtype,
255 	       pkt->xfer_mode, pkt->vdisk_block_size,
256 	       pkt->max_xfer_size);
257 
258 	if (pkt->tag.stype == VIO_SUBTYPE_ACK) {
259 		switch (pkt->vdisk_type) {
260 		case VD_DISK_TYPE_DISK:
261 		case VD_DISK_TYPE_SLICE:
262 			break;
263 
264 		default:
265 			printk(KERN_ERR PFX "%s: Bogus vdisk_type 0x%x\n",
266 			       vio->name, pkt->vdisk_type);
267 			return -ECONNRESET;
268 		}
269 
270 		if (pkt->vdisk_block_size > port->vdisk_block_size) {
271 			printk(KERN_ERR PFX "%s: BLOCK size increased "
272 			       "%u --> %u\n",
273 			       vio->name,
274 			       port->vdisk_block_size, pkt->vdisk_block_size);
275 			return -ECONNRESET;
276 		}
277 
278 		port->operations = pkt->operations;
279 		port->vdisk_type = pkt->vdisk_type;
280 		if (vdc_version_supported(port, 1, 1)) {
281 			port->vdisk_size = pkt->vdisk_size;
282 			port->vdisk_mtype = pkt->vdisk_mtype;
283 		}
284 		if (pkt->max_xfer_size < port->max_xfer_size)
285 			port->max_xfer_size = pkt->max_xfer_size;
286 		port->vdisk_block_size = pkt->vdisk_block_size;
287 
288 		port->vdisk_phys_blksz = VDC_DEFAULT_BLK_SIZE;
289 		if (vdc_version_supported(port, 1, 2))
290 			port->vdisk_phys_blksz = pkt->phys_block_size;
291 
292 		return 0;
293 	} else {
294 		printk(KERN_ERR PFX "%s: Attribute NACK\n", vio->name);
295 
296 		return -ECONNRESET;
297 	}
298 }
299 
300 static void vdc_end_special(struct vdc_port *port, struct vio_disk_desc *desc)
301 {
302 	int err = desc->status;
303 
304 	vdc_finish(&port->vio, -err, WAITING_FOR_GEN_CMD);
305 }
306 
307 static void vdc_end_one(struct vdc_port *port, struct vio_dring_state *dr,
308 			unsigned int index)
309 {
310 	struct vio_disk_desc *desc = vio_dring_entry(dr, index);
311 	struct vdc_req_entry *rqe = &port->rq_arr[index];
312 	struct request *req;
313 
314 	if (unlikely(desc->hdr.state != VIO_DESC_DONE))
315 		return;
316 
317 	ldc_unmap(port->vio.lp, desc->cookies, desc->ncookies);
318 	desc->hdr.state = VIO_DESC_FREE;
319 	dr->cons = vio_dring_next(dr, index);
320 
321 	req = rqe->req;
322 	if (req == NULL) {
323 		vdc_end_special(port, desc);
324 		return;
325 	}
326 
327 	rqe->req = NULL;
328 
329 	blk_mq_end_request(req, desc->status ? BLK_STS_IOERR : 0);
330 
331 	vdc_blk_queue_start(port);
332 }
333 
334 static int vdc_ack(struct vdc_port *port, void *msgbuf)
335 {
336 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
337 	struct vio_dring_data *pkt = msgbuf;
338 
339 	if (unlikely(pkt->dring_ident != dr->ident ||
340 		     pkt->start_idx != pkt->end_idx ||
341 		     pkt->start_idx >= VDC_TX_RING_SIZE))
342 		return 0;
343 
344 	vdc_end_one(port, dr, pkt->start_idx);
345 
346 	return 0;
347 }
348 
349 static int vdc_nack(struct vdc_port *port, void *msgbuf)
350 {
351 	/* XXX Implement me XXX */
352 	return 0;
353 }
354 
355 static void vdc_event(void *arg, int event)
356 {
357 	struct vdc_port *port = arg;
358 	struct vio_driver_state *vio = &port->vio;
359 	unsigned long flags;
360 	int err;
361 
362 	spin_lock_irqsave(&vio->lock, flags);
363 
364 	if (unlikely(event == LDC_EVENT_RESET)) {
365 		vio_link_state_change(vio, event);
366 		queue_work(sunvdc_wq, &port->ldc_reset_work);
367 		goto out;
368 	}
369 
370 	if (unlikely(event == LDC_EVENT_UP)) {
371 		vio_link_state_change(vio, event);
372 		goto out;
373 	}
374 
375 	if (unlikely(event != LDC_EVENT_DATA_READY)) {
376 		pr_warn(PFX "Unexpected LDC event %d\n", event);
377 		goto out;
378 	}
379 
380 	err = 0;
381 	while (1) {
382 		union {
383 			struct vio_msg_tag tag;
384 			u64 raw[8];
385 		} msgbuf;
386 
387 		err = ldc_read(vio->lp, &msgbuf, sizeof(msgbuf));
388 		if (unlikely(err < 0)) {
389 			if (err == -ECONNRESET)
390 				vio_conn_reset(vio);
391 			break;
392 		}
393 		if (err == 0)
394 			break;
395 		viodbg(DATA, "TAG [%02x:%02x:%04x:%08x]\n",
396 		       msgbuf.tag.type,
397 		       msgbuf.tag.stype,
398 		       msgbuf.tag.stype_env,
399 		       msgbuf.tag.sid);
400 		err = vio_validate_sid(vio, &msgbuf.tag);
401 		if (err < 0)
402 			break;
403 
404 		if (likely(msgbuf.tag.type == VIO_TYPE_DATA)) {
405 			if (msgbuf.tag.stype == VIO_SUBTYPE_ACK)
406 				err = vdc_ack(port, &msgbuf);
407 			else if (msgbuf.tag.stype == VIO_SUBTYPE_NACK)
408 				err = vdc_nack(port, &msgbuf);
409 			else
410 				err = vdc_handle_unknown(port, &msgbuf);
411 		} else if (msgbuf.tag.type == VIO_TYPE_CTRL) {
412 			err = vio_control_pkt_engine(vio, &msgbuf);
413 		} else {
414 			err = vdc_handle_unknown(port, &msgbuf);
415 		}
416 		if (err < 0)
417 			break;
418 	}
419 	if (err < 0)
420 		vdc_finish(&port->vio, err, WAITING_FOR_ANY);
421 out:
422 	spin_unlock_irqrestore(&vio->lock, flags);
423 }
424 
425 static int __vdc_tx_trigger(struct vdc_port *port)
426 {
427 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
428 	struct vio_dring_data hdr = {
429 		.tag = {
430 			.type		= VIO_TYPE_DATA,
431 			.stype		= VIO_SUBTYPE_INFO,
432 			.stype_env	= VIO_DRING_DATA,
433 			.sid		= vio_send_sid(&port->vio),
434 		},
435 		.dring_ident		= dr->ident,
436 		.start_idx		= dr->prod,
437 		.end_idx		= dr->prod,
438 	};
439 	int err, delay;
440 	int retries = 0;
441 
442 	hdr.seq = dr->snd_nxt;
443 	delay = 1;
444 	do {
445 		err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
446 		if (err > 0) {
447 			dr->snd_nxt++;
448 			break;
449 		}
450 		udelay(delay);
451 		if ((delay <<= 1) > 128)
452 			delay = 128;
453 		if (retries++ > VDC_MAX_RETRIES)
454 			break;
455 	} while (err == -EAGAIN);
456 
457 	if (err == -ENOTCONN)
458 		vdc_ldc_reset(port);
459 	return err;
460 }
461 
462 static int __send_request(struct request *req)
463 {
464 	struct vdc_port *port = req->q->disk->private_data;
465 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
466 	struct scatterlist sg[MAX_RING_COOKIES];
467 	struct vdc_req_entry *rqe;
468 	struct vio_disk_desc *desc;
469 	unsigned int map_perm;
470 	int nsg, err, i;
471 	u64 len;
472 	u8 op;
473 
474 	if (WARN_ON(port->ring_cookies > MAX_RING_COOKIES))
475 		return -EINVAL;
476 
477 	map_perm = LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_IO;
478 
479 	if (rq_data_dir(req) == READ) {
480 		map_perm |= LDC_MAP_W;
481 		op = VD_OP_BREAD;
482 	} else {
483 		map_perm |= LDC_MAP_R;
484 		op = VD_OP_BWRITE;
485 	}
486 
487 	sg_init_table(sg, port->ring_cookies);
488 	nsg = blk_rq_map_sg(req->q, req, sg);
489 
490 	len = 0;
491 	for (i = 0; i < nsg; i++)
492 		len += sg[i].length;
493 
494 	desc = vio_dring_cur(dr);
495 
496 	err = ldc_map_sg(port->vio.lp, sg, nsg,
497 			 desc->cookies, port->ring_cookies,
498 			 map_perm);
499 	if (err < 0) {
500 		printk(KERN_ERR PFX "ldc_map_sg() failure, err=%d.\n", err);
501 		return err;
502 	}
503 
504 	rqe = &port->rq_arr[dr->prod];
505 	rqe->req = req;
506 
507 	desc->hdr.ack = VIO_ACK_ENABLE;
508 	desc->req_id = port->req_id;
509 	desc->operation = op;
510 	if (port->vdisk_type == VD_DISK_TYPE_DISK) {
511 		desc->slice = 0xff;
512 	} else {
513 		desc->slice = 0;
514 	}
515 	desc->status = ~0;
516 	desc->offset = (blk_rq_pos(req) << 9) / port->vdisk_block_size;
517 	desc->size = len;
518 	desc->ncookies = err;
519 
520 	/* This has to be a non-SMP write barrier because we are writing
521 	 * to memory which is shared with the peer LDOM.
522 	 */
523 	wmb();
524 	desc->hdr.state = VIO_DESC_READY;
525 
526 	err = __vdc_tx_trigger(port);
527 	if (err < 0) {
528 		printk(KERN_ERR PFX "vdc_tx_trigger() failure, err=%d\n", err);
529 	} else {
530 		port->req_id++;
531 		dr->prod = vio_dring_next(dr, dr->prod);
532 	}
533 
534 	return err;
535 }
536 
537 static blk_status_t vdc_queue_rq(struct blk_mq_hw_ctx *hctx,
538 				 const struct blk_mq_queue_data *bd)
539 {
540 	struct vdc_port *port = hctx->queue->queuedata;
541 	struct vio_dring_state *dr;
542 	unsigned long flags;
543 
544 	dr = &port->vio.drings[VIO_DRIVER_TX_RING];
545 
546 	blk_mq_start_request(bd->rq);
547 
548 	spin_lock_irqsave(&port->vio.lock, flags);
549 
550 	/*
551 	 * Doing drain, just end the request in error
552 	 */
553 	if (unlikely(port->drain)) {
554 		spin_unlock_irqrestore(&port->vio.lock, flags);
555 		return BLK_STS_IOERR;
556 	}
557 
558 	if (unlikely(vdc_tx_dring_avail(dr) < 1)) {
559 		spin_unlock_irqrestore(&port->vio.lock, flags);
560 		blk_mq_stop_hw_queue(hctx);
561 		return BLK_STS_DEV_RESOURCE;
562 	}
563 
564 	if (__send_request(bd->rq) < 0) {
565 		spin_unlock_irqrestore(&port->vio.lock, flags);
566 		return BLK_STS_IOERR;
567 	}
568 
569 	spin_unlock_irqrestore(&port->vio.lock, flags);
570 	return BLK_STS_OK;
571 }
572 
573 static int generic_request(struct vdc_port *port, u8 op, void *buf, int len)
574 {
575 	struct vio_dring_state *dr;
576 	struct vio_completion comp;
577 	struct vio_disk_desc *desc;
578 	unsigned int map_perm;
579 	unsigned long flags;
580 	int op_len, err;
581 	void *req_buf;
582 
583 	if (!(((u64)1 << (u64)op) & port->operations))
584 		return -EOPNOTSUPP;
585 
586 	switch (op) {
587 	case VD_OP_BREAD:
588 	case VD_OP_BWRITE:
589 	default:
590 		return -EINVAL;
591 
592 	case VD_OP_FLUSH:
593 		op_len = 0;
594 		map_perm = 0;
595 		break;
596 
597 	case VD_OP_GET_WCE:
598 		op_len = sizeof(u32);
599 		map_perm = LDC_MAP_W;
600 		break;
601 
602 	case VD_OP_SET_WCE:
603 		op_len = sizeof(u32);
604 		map_perm = LDC_MAP_R;
605 		break;
606 
607 	case VD_OP_GET_VTOC:
608 		op_len = sizeof(struct vio_disk_vtoc);
609 		map_perm = LDC_MAP_W;
610 		break;
611 
612 	case VD_OP_SET_VTOC:
613 		op_len = sizeof(struct vio_disk_vtoc);
614 		map_perm = LDC_MAP_R;
615 		break;
616 
617 	case VD_OP_GET_DISKGEOM:
618 		op_len = sizeof(struct vio_disk_geom);
619 		map_perm = LDC_MAP_W;
620 		break;
621 
622 	case VD_OP_SET_DISKGEOM:
623 		op_len = sizeof(struct vio_disk_geom);
624 		map_perm = LDC_MAP_R;
625 		break;
626 
627 	case VD_OP_SCSICMD:
628 		op_len = 16;
629 		map_perm = LDC_MAP_RW;
630 		break;
631 
632 	case VD_OP_GET_DEVID:
633 		op_len = sizeof(struct vio_disk_devid);
634 		map_perm = LDC_MAP_W;
635 		break;
636 
637 	case VD_OP_GET_EFI:
638 	case VD_OP_SET_EFI:
639 		return -EOPNOTSUPP;
640 	}
641 
642 	map_perm |= LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_IO;
643 
644 	op_len = (op_len + 7) & ~7;
645 	req_buf = kzalloc(op_len, GFP_KERNEL);
646 	if (!req_buf)
647 		return -ENOMEM;
648 
649 	if (len > op_len)
650 		len = op_len;
651 
652 	if (map_perm & LDC_MAP_R)
653 		memcpy(req_buf, buf, len);
654 
655 	spin_lock_irqsave(&port->vio.lock, flags);
656 
657 	dr = &port->vio.drings[VIO_DRIVER_TX_RING];
658 
659 	/* XXX If we want to use this code generically we have to
660 	 * XXX handle TX ring exhaustion etc.
661 	 */
662 	desc = vio_dring_cur(dr);
663 
664 	err = ldc_map_single(port->vio.lp, req_buf, op_len,
665 			     desc->cookies, port->ring_cookies,
666 			     map_perm);
667 	if (err < 0) {
668 		spin_unlock_irqrestore(&port->vio.lock, flags);
669 		kfree(req_buf);
670 		return err;
671 	}
672 
673 	init_completion(&comp.com);
674 	comp.waiting_for = WAITING_FOR_GEN_CMD;
675 	port->vio.cmp = &comp;
676 
677 	desc->hdr.ack = VIO_ACK_ENABLE;
678 	desc->req_id = port->req_id;
679 	desc->operation = op;
680 	desc->slice = 0;
681 	desc->status = ~0;
682 	desc->offset = 0;
683 	desc->size = op_len;
684 	desc->ncookies = err;
685 
686 	/* This has to be a non-SMP write barrier because we are writing
687 	 * to memory which is shared with the peer LDOM.
688 	 */
689 	wmb();
690 	desc->hdr.state = VIO_DESC_READY;
691 
692 	err = __vdc_tx_trigger(port);
693 	if (err >= 0) {
694 		port->req_id++;
695 		dr->prod = vio_dring_next(dr, dr->prod);
696 		spin_unlock_irqrestore(&port->vio.lock, flags);
697 
698 		wait_for_completion(&comp.com);
699 		err = comp.err;
700 	} else {
701 		port->vio.cmp = NULL;
702 		spin_unlock_irqrestore(&port->vio.lock, flags);
703 	}
704 
705 	if (map_perm & LDC_MAP_W)
706 		memcpy(buf, req_buf, len);
707 
708 	kfree(req_buf);
709 
710 	return err;
711 }
712 
713 static int vdc_alloc_tx_ring(struct vdc_port *port)
714 {
715 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
716 	unsigned long len, entry_size;
717 	int ncookies;
718 	void *dring;
719 
720 	entry_size = sizeof(struct vio_disk_desc) +
721 		(sizeof(struct ldc_trans_cookie) * port->ring_cookies);
722 	len = (VDC_TX_RING_SIZE * entry_size);
723 
724 	ncookies = VIO_MAX_RING_COOKIES;
725 	dring = ldc_alloc_exp_dring(port->vio.lp, len,
726 				    dr->cookies, &ncookies,
727 				    (LDC_MAP_SHADOW |
728 				     LDC_MAP_DIRECT |
729 				     LDC_MAP_RW));
730 	if (IS_ERR(dring))
731 		return PTR_ERR(dring);
732 
733 	dr->base = dring;
734 	dr->entry_size = entry_size;
735 	dr->num_entries = VDC_TX_RING_SIZE;
736 	dr->prod = dr->cons = 0;
737 	dr->pending = VDC_TX_RING_SIZE;
738 	dr->ncookies = ncookies;
739 
740 	return 0;
741 }
742 
743 static void vdc_free_tx_ring(struct vdc_port *port)
744 {
745 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
746 
747 	if (dr->base) {
748 		ldc_free_exp_dring(port->vio.lp, dr->base,
749 				   (dr->entry_size * dr->num_entries),
750 				   dr->cookies, dr->ncookies);
751 		dr->base = NULL;
752 		dr->entry_size = 0;
753 		dr->num_entries = 0;
754 		dr->pending = 0;
755 		dr->ncookies = 0;
756 	}
757 }
758 
759 static int vdc_port_up(struct vdc_port *port)
760 {
761 	struct vio_completion comp;
762 
763 	init_completion(&comp.com);
764 	comp.err = 0;
765 	comp.waiting_for = WAITING_FOR_LINK_UP;
766 	port->vio.cmp = &comp;
767 
768 	vio_port_up(&port->vio);
769 	wait_for_completion(&comp.com);
770 	return comp.err;
771 }
772 
773 static void vdc_port_down(struct vdc_port *port)
774 {
775 	ldc_disconnect(port->vio.lp);
776 	ldc_unbind(port->vio.lp);
777 	vdc_free_tx_ring(port);
778 	vio_ldc_free(&port->vio);
779 }
780 
781 static const struct blk_mq_ops vdc_mq_ops = {
782 	.queue_rq	= vdc_queue_rq,
783 };
784 
785 static int probe_disk(struct vdc_port *port)
786 {
787 	struct request_queue *q;
788 	struct gendisk *g;
789 	int err;
790 
791 	err = vdc_port_up(port);
792 	if (err)
793 		return err;
794 
795 	/* Using version 1.2 means vdisk_phys_blksz should be set unless the
796 	 * disk is reserved by another system.
797 	 */
798 	if (vdc_version_supported(port, 1, 2) && !port->vdisk_phys_blksz)
799 		return -ENODEV;
800 
801 	if (vdc_version_supported(port, 1, 1)) {
802 		/* vdisk_size should be set during the handshake, if it wasn't
803 		 * then the underlying disk is reserved by another system
804 		 */
805 		if (port->vdisk_size == -1)
806 			return -ENODEV;
807 	} else {
808 		struct vio_disk_geom geom;
809 
810 		err = generic_request(port, VD_OP_GET_DISKGEOM,
811 				      &geom, sizeof(geom));
812 		if (err < 0) {
813 			printk(KERN_ERR PFX "VD_OP_GET_DISKGEOM returns "
814 			       "error %d\n", err);
815 			return err;
816 		}
817 		port->vdisk_size = ((u64)geom.num_cyl *
818 				    (u64)geom.num_hd *
819 				    (u64)geom.num_sec);
820 	}
821 
822 	err = blk_mq_alloc_sq_tag_set(&port->tag_set, &vdc_mq_ops,
823 			VDC_TX_RING_SIZE, BLK_MQ_F_SHOULD_MERGE);
824 	if (err)
825 		return err;
826 
827 	g = blk_mq_alloc_disk(&port->tag_set, port);
828 	if (IS_ERR(g)) {
829 		printk(KERN_ERR PFX "%s: Could not allocate gendisk.\n",
830 		       port->vio.name);
831 		err = PTR_ERR(g);
832 		goto out_free_tag;
833 	}
834 
835 	port->disk = g;
836 	q = g->queue;
837 
838 	/* Each segment in a request is up to an aligned page in size. */
839 	blk_queue_segment_boundary(q, PAGE_SIZE - 1);
840 	blk_queue_max_segment_size(q, PAGE_SIZE);
841 
842 	blk_queue_max_segments(q, port->ring_cookies);
843 	blk_queue_max_hw_sectors(q, port->max_xfer_size);
844 	g->major = vdc_major;
845 	g->first_minor = port->vio.vdev->dev_no << PARTITION_SHIFT;
846 	g->minors = 1 << PARTITION_SHIFT;
847 	strcpy(g->disk_name, port->disk_name);
848 
849 	g->fops = &vdc_fops;
850 	g->queue = q;
851 	g->private_data = port;
852 
853 	set_capacity(g, port->vdisk_size);
854 
855 	if (vdc_version_supported(port, 1, 1)) {
856 		switch (port->vdisk_mtype) {
857 		case VD_MEDIA_TYPE_CD:
858 			pr_info(PFX "Virtual CDROM %s\n", port->disk_name);
859 			g->flags |= GENHD_FL_REMOVABLE;
860 			set_disk_ro(g, 1);
861 			break;
862 
863 		case VD_MEDIA_TYPE_DVD:
864 			pr_info(PFX "Virtual DVD %s\n", port->disk_name);
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 	err = device_add_disk(&port->vio.vdev->dev, g, NULL);
883 	if (err)
884 		goto out_cleanup_disk;
885 
886 	return 0;
887 
888 out_cleanup_disk:
889 	blk_cleanup_disk(g);
890 out_free_tag:
891 	blk_mq_free_tag_set(&port->tag_set);
892 	return err;
893 }
894 
895 static struct ldc_channel_config vdc_ldc_cfg = {
896 	.event		= vdc_event,
897 	.mtu		= 64,
898 	.mode		= LDC_MODE_UNRELIABLE,
899 };
900 
901 static struct vio_driver_ops vdc_vio_ops = {
902 	.send_attr		= vdc_send_attr,
903 	.handle_attr		= vdc_handle_attr,
904 	.handshake_complete	= vdc_handshake_complete,
905 };
906 
907 static void print_version(void)
908 {
909 	static int version_printed;
910 
911 	if (version_printed++ == 0)
912 		printk(KERN_INFO "%s", version);
913 }
914 
915 struct vdc_check_port_data {
916 	int	dev_no;
917 	char	*type;
918 };
919 
920 static int vdc_device_probed(struct device *dev, void *arg)
921 {
922 	struct vio_dev *vdev = to_vio_dev(dev);
923 	struct vdc_check_port_data *port_data;
924 
925 	port_data = (struct vdc_check_port_data *)arg;
926 
927 	if ((vdev->dev_no == port_data->dev_no) &&
928 	    (!(strcmp((char *)&vdev->type, port_data->type))) &&
929 		dev_get_drvdata(dev)) {
930 		/* This device has already been configured
931 		 * by vdc_port_probe()
932 		 */
933 		return 1;
934 	} else {
935 		return 0;
936 	}
937 }
938 
939 /* Determine whether the VIO device is part of an mpgroup
940  * by locating all the virtual-device-port nodes associated
941  * with the parent virtual-device node for the VIO device
942  * and checking whether any of these nodes are vdc-ports
943  * which have already been configured.
944  *
945  * Returns true if this device is part of an mpgroup and has
946  * already been probed.
947  */
948 static bool vdc_port_mpgroup_check(struct vio_dev *vdev)
949 {
950 	struct vdc_check_port_data port_data;
951 	struct device *dev;
952 
953 	port_data.dev_no = vdev->dev_no;
954 	port_data.type = (char *)&vdev->type;
955 
956 	dev = device_find_child(vdev->dev.parent, &port_data,
957 				vdc_device_probed);
958 
959 	if (dev)
960 		return true;
961 
962 	return false;
963 }
964 
965 static int vdc_port_probe(struct vio_dev *vdev, const struct vio_device_id *id)
966 {
967 	struct mdesc_handle *hp;
968 	struct vdc_port *port;
969 	int err;
970 	const u64 *ldc_timeout;
971 
972 	print_version();
973 
974 	hp = mdesc_grab();
975 
976 	err = -ENODEV;
977 	if ((vdev->dev_no << PARTITION_SHIFT) & ~(u64)MINORMASK) {
978 		printk(KERN_ERR PFX "Port id [%llu] too large.\n",
979 		       vdev->dev_no);
980 		goto err_out_release_mdesc;
981 	}
982 
983 	/* Check if this device is part of an mpgroup */
984 	if (vdc_port_mpgroup_check(vdev)) {
985 		printk(KERN_WARNING
986 			"VIO: Ignoring extra vdisk port %s",
987 			dev_name(&vdev->dev));
988 		goto err_out_release_mdesc;
989 	}
990 
991 	port = kzalloc(sizeof(*port), GFP_KERNEL);
992 	if (!port) {
993 		err = -ENOMEM;
994 		goto err_out_release_mdesc;
995 	}
996 
997 	if (vdev->dev_no >= 26)
998 		snprintf(port->disk_name, sizeof(port->disk_name),
999 			 VDCBLK_NAME "%c%c",
1000 			 'a' + ((int)vdev->dev_no / 26) - 1,
1001 			 'a' + ((int)vdev->dev_no % 26));
1002 	else
1003 		snprintf(port->disk_name, sizeof(port->disk_name),
1004 			 VDCBLK_NAME "%c", 'a' + ((int)vdev->dev_no % 26));
1005 	port->vdisk_size = -1;
1006 
1007 	/* Actual wall time may be double due to do_generic_file_read() doing
1008 	 * a readahead I/O first, and once that fails it will try to read a
1009 	 * single page.
1010 	 */
1011 	ldc_timeout = mdesc_get_property(hp, vdev->mp, "vdc-timeout", NULL);
1012 	port->ldc_timeout = ldc_timeout ? *ldc_timeout : 0;
1013 	INIT_DELAYED_WORK(&port->ldc_reset_timer_work, vdc_ldc_reset_timer_work);
1014 	INIT_WORK(&port->ldc_reset_work, vdc_ldc_reset_work);
1015 
1016 	err = vio_driver_init(&port->vio, vdev, VDEV_DISK,
1017 			      vdc_versions, ARRAY_SIZE(vdc_versions),
1018 			      &vdc_vio_ops, port->disk_name);
1019 	if (err)
1020 		goto err_out_free_port;
1021 
1022 	port->vdisk_block_size = VDC_DEFAULT_BLK_SIZE;
1023 	port->max_xfer_size = MAX_XFER_SIZE;
1024 	port->ring_cookies = MAX_RING_COOKIES;
1025 
1026 	err = vio_ldc_alloc(&port->vio, &vdc_ldc_cfg, port);
1027 	if (err)
1028 		goto err_out_free_port;
1029 
1030 	err = vdc_alloc_tx_ring(port);
1031 	if (err)
1032 		goto err_out_free_ldc;
1033 
1034 	err = probe_disk(port);
1035 	if (err)
1036 		goto err_out_free_tx_ring;
1037 
1038 	/* Note that the device driver_data is used to determine
1039 	 * whether the port has been probed.
1040 	 */
1041 	dev_set_drvdata(&vdev->dev, port);
1042 
1043 	mdesc_release(hp);
1044 
1045 	return 0;
1046 
1047 err_out_free_tx_ring:
1048 	vdc_free_tx_ring(port);
1049 
1050 err_out_free_ldc:
1051 	vio_ldc_free(&port->vio);
1052 
1053 err_out_free_port:
1054 	kfree(port);
1055 
1056 err_out_release_mdesc:
1057 	mdesc_release(hp);
1058 	return err;
1059 }
1060 
1061 static void vdc_port_remove(struct vio_dev *vdev)
1062 {
1063 	struct vdc_port *port = dev_get_drvdata(&vdev->dev);
1064 
1065 	if (port) {
1066 		blk_mq_stop_hw_queues(port->disk->queue);
1067 
1068 		flush_work(&port->ldc_reset_work);
1069 		cancel_delayed_work_sync(&port->ldc_reset_timer_work);
1070 		del_timer_sync(&port->vio.timer);
1071 
1072 		del_gendisk(port->disk);
1073 		blk_cleanup_disk(port->disk);
1074 		blk_mq_free_tag_set(&port->tag_set);
1075 
1076 		vdc_free_tx_ring(port);
1077 		vio_ldc_free(&port->vio);
1078 
1079 		dev_set_drvdata(&vdev->dev, NULL);
1080 
1081 		kfree(port);
1082 	}
1083 }
1084 
1085 static void vdc_requeue_inflight(struct vdc_port *port)
1086 {
1087 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1088 	u32 idx;
1089 
1090 	for (idx = dr->cons; idx != dr->prod; idx = vio_dring_next(dr, idx)) {
1091 		struct vio_disk_desc *desc = vio_dring_entry(dr, idx);
1092 		struct vdc_req_entry *rqe = &port->rq_arr[idx];
1093 		struct request *req;
1094 
1095 		ldc_unmap(port->vio.lp, desc->cookies, desc->ncookies);
1096 		desc->hdr.state = VIO_DESC_FREE;
1097 		dr->cons = vio_dring_next(dr, idx);
1098 
1099 		req = rqe->req;
1100 		if (req == NULL) {
1101 			vdc_end_special(port, desc);
1102 			continue;
1103 		}
1104 
1105 		rqe->req = NULL;
1106 		blk_mq_requeue_request(req, false);
1107 	}
1108 }
1109 
1110 static void vdc_queue_drain(struct vdc_port *port)
1111 {
1112 	struct request_queue *q = port->disk->queue;
1113 
1114 	/*
1115 	 * Mark the queue as draining, then freeze/quiesce to ensure
1116 	 * that all existing requests are seen in ->queue_rq() and killed
1117 	 */
1118 	port->drain = 1;
1119 	spin_unlock_irq(&port->vio.lock);
1120 
1121 	blk_mq_freeze_queue(q);
1122 	blk_mq_quiesce_queue(q);
1123 
1124 	spin_lock_irq(&port->vio.lock);
1125 	port->drain = 0;
1126 	blk_mq_unquiesce_queue(q);
1127 	blk_mq_unfreeze_queue(q);
1128 }
1129 
1130 static void vdc_ldc_reset_timer_work(struct work_struct *work)
1131 {
1132 	struct vdc_port *port;
1133 	struct vio_driver_state *vio;
1134 
1135 	port = container_of(work, struct vdc_port, ldc_reset_timer_work.work);
1136 	vio = &port->vio;
1137 
1138 	spin_lock_irq(&vio->lock);
1139 	if (!(port->vio.hs_state & VIO_HS_COMPLETE)) {
1140 		pr_warn(PFX "%s ldc down %llu seconds, draining queue\n",
1141 			port->disk_name, port->ldc_timeout);
1142 		vdc_queue_drain(port);
1143 		vdc_blk_queue_start(port);
1144 	}
1145 	spin_unlock_irq(&vio->lock);
1146 }
1147 
1148 static void vdc_ldc_reset_work(struct work_struct *work)
1149 {
1150 	struct vdc_port *port;
1151 	struct vio_driver_state *vio;
1152 	unsigned long flags;
1153 
1154 	port = container_of(work, struct vdc_port, ldc_reset_work);
1155 	vio = &port->vio;
1156 
1157 	spin_lock_irqsave(&vio->lock, flags);
1158 	vdc_ldc_reset(port);
1159 	spin_unlock_irqrestore(&vio->lock, flags);
1160 }
1161 
1162 static void vdc_ldc_reset(struct vdc_port *port)
1163 {
1164 	int err;
1165 
1166 	assert_spin_locked(&port->vio.lock);
1167 
1168 	pr_warn(PFX "%s ldc link reset\n", port->disk_name);
1169 	blk_mq_stop_hw_queues(port->disk->queue);
1170 	vdc_requeue_inflight(port);
1171 	vdc_port_down(port);
1172 
1173 	err = vio_ldc_alloc(&port->vio, &vdc_ldc_cfg, port);
1174 	if (err) {
1175 		pr_err(PFX "%s vio_ldc_alloc:%d\n", port->disk_name, err);
1176 		return;
1177 	}
1178 
1179 	err = vdc_alloc_tx_ring(port);
1180 	if (err) {
1181 		pr_err(PFX "%s vio_alloc_tx_ring:%d\n", port->disk_name, err);
1182 		goto err_free_ldc;
1183 	}
1184 
1185 	if (port->ldc_timeout)
1186 		mod_delayed_work(system_wq, &port->ldc_reset_timer_work,
1187 			  round_jiffies(jiffies + HZ * port->ldc_timeout));
1188 	mod_timer(&port->vio.timer, round_jiffies(jiffies + HZ));
1189 	return;
1190 
1191 err_free_ldc:
1192 	vio_ldc_free(&port->vio);
1193 }
1194 
1195 static const struct vio_device_id vdc_port_match[] = {
1196 	{
1197 		.type = "vdc-port",
1198 	},
1199 	{},
1200 };
1201 MODULE_DEVICE_TABLE(vio, vdc_port_match);
1202 
1203 static struct vio_driver vdc_port_driver = {
1204 	.id_table	= vdc_port_match,
1205 	.probe		= vdc_port_probe,
1206 	.remove		= vdc_port_remove,
1207 	.name		= "vdc_port",
1208 };
1209 
1210 static int __init vdc_init(void)
1211 {
1212 	int err;
1213 
1214 	sunvdc_wq = alloc_workqueue("sunvdc", 0, 0);
1215 	if (!sunvdc_wq)
1216 		return -ENOMEM;
1217 
1218 	err = register_blkdev(0, VDCBLK_NAME);
1219 	if (err < 0)
1220 		goto out_free_wq;
1221 
1222 	vdc_major = err;
1223 
1224 	err = vio_register_driver(&vdc_port_driver);
1225 	if (err)
1226 		goto out_unregister_blkdev;
1227 
1228 	return 0;
1229 
1230 out_unregister_blkdev:
1231 	unregister_blkdev(vdc_major, VDCBLK_NAME);
1232 	vdc_major = 0;
1233 
1234 out_free_wq:
1235 	destroy_workqueue(sunvdc_wq);
1236 	return err;
1237 }
1238 
1239 static void __exit vdc_exit(void)
1240 {
1241 	vio_unregister_driver(&vdc_port_driver);
1242 	unregister_blkdev(vdc_major, VDCBLK_NAME);
1243 	destroy_workqueue(sunvdc_wq);
1244 }
1245 
1246 module_init(vdc_init);
1247 module_exit(vdc_exit);
1248