xref: /linux/drivers/block/sunvdc.c (revision b615879dbfea6cf1236acbc3f2fb25ae84e07071)
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 gendisk *disk, struct hd_geometry *geo)
123 {
124 	sector_t nsect = get_capacity(disk);
125 	sector_t cylinders = nsect;
126 
127 	geo->heads = 0xff;
128 	geo->sectors = 0x3f;
129 	sector_div(cylinders, geo->heads * geo->sectors);
130 	geo->cylinders = cylinders;
131 	if ((sector_t)(geo->cylinders + 1) * geo->heads * geo->sectors < nsect)
132 		geo->cylinders = 0xffff;
133 
134 	return 0;
135 }
136 
137 /* Add ioctl/CDROM_GET_CAPABILITY to support cdrom_id in udev
138  * when vdisk_mtype is VD_MEDIA_TYPE_CD or VD_MEDIA_TYPE_DVD.
139  * Needed to be able to install inside an ldom from an iso image.
140  */
141 static int vdc_ioctl(struct block_device *bdev, blk_mode_t mode,
142 		     unsigned command, unsigned long argument)
143 {
144 	struct vdc_port *port = bdev->bd_disk->private_data;
145 	int i;
146 
147 	switch (command) {
148 	case CDROMMULTISESSION:
149 		pr_debug(PFX "Multisession CDs not supported\n");
150 		for (i = 0; i < sizeof(struct cdrom_multisession); i++)
151 			if (put_user(0, (char __user *)(argument + i)))
152 				return -EFAULT;
153 		return 0;
154 
155 	case CDROM_GET_CAPABILITY:
156 		if (!vdc_version_supported(port, 1, 1))
157 			return -EINVAL;
158 		switch (port->vdisk_mtype) {
159 		case VD_MEDIA_TYPE_CD:
160 		case VD_MEDIA_TYPE_DVD:
161 			return 0;
162 		default:
163 			return -EINVAL;
164 		}
165 	default:
166 		pr_debug(PFX "ioctl %08x not supported\n", command);
167 		return -EINVAL;
168 	}
169 }
170 
171 static const struct block_device_operations vdc_fops = {
172 	.owner		= THIS_MODULE,
173 	.getgeo		= vdc_getgeo,
174 	.ioctl		= vdc_ioctl,
175 	.compat_ioctl	= blkdev_compat_ptr_ioctl,
176 };
177 
178 static void vdc_blk_queue_start(struct vdc_port *port)
179 {
180 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
181 
182 	/* restart blk queue when ring is half emptied. also called after
183 	 * handshake completes, so check for initial handshake before we've
184 	 * allocated a disk.
185 	 */
186 	if (port->disk && vdc_tx_dring_avail(dr) * 100 / VDC_TX_RING_SIZE >= 50)
187 		blk_mq_start_stopped_hw_queues(port->disk->queue, true);
188 }
189 
190 static void vdc_finish(struct vio_driver_state *vio, int err, int waiting_for)
191 {
192 	if (vio->cmp &&
193 	    (waiting_for == -1 ||
194 	     vio->cmp->waiting_for == waiting_for)) {
195 		vio->cmp->err = err;
196 		complete(&vio->cmp->com);
197 		vio->cmp = NULL;
198 	}
199 }
200 
201 static void vdc_handshake_complete(struct vio_driver_state *vio)
202 {
203 	struct vdc_port *port = to_vdc_port(vio);
204 
205 	cancel_delayed_work(&port->ldc_reset_timer_work);
206 	vdc_finish(vio, 0, WAITING_FOR_LINK_UP);
207 	vdc_blk_queue_start(port);
208 }
209 
210 static int vdc_handle_unknown(struct vdc_port *port, void *arg)
211 {
212 	struct vio_msg_tag *pkt = arg;
213 
214 	printk(KERN_ERR PFX "Received unknown msg [%02x:%02x:%04x:%08x]\n",
215 	       pkt->type, pkt->stype, pkt->stype_env, pkt->sid);
216 	printk(KERN_ERR PFX "Resetting connection.\n");
217 
218 	ldc_disconnect(port->vio.lp);
219 
220 	return -ECONNRESET;
221 }
222 
223 static int vdc_send_attr(struct vio_driver_state *vio)
224 {
225 	struct vdc_port *port = to_vdc_port(vio);
226 	struct vio_disk_attr_info pkt;
227 
228 	memset(&pkt, 0, sizeof(pkt));
229 
230 	pkt.tag.type = VIO_TYPE_CTRL;
231 	pkt.tag.stype = VIO_SUBTYPE_INFO;
232 	pkt.tag.stype_env = VIO_ATTR_INFO;
233 	pkt.tag.sid = vio_send_sid(vio);
234 
235 	pkt.xfer_mode = VIO_DRING_MODE;
236 	pkt.vdisk_block_size = port->vdisk_block_size;
237 	pkt.max_xfer_size = port->max_xfer_size;
238 
239 	viodbg(HS, "SEND ATTR xfer_mode[0x%x] blksz[%u] max_xfer[%llu]\n",
240 	       pkt.xfer_mode, pkt.vdisk_block_size, pkt.max_xfer_size);
241 
242 	return vio_ldc_send(&port->vio, &pkt, sizeof(pkt));
243 }
244 
245 static int vdc_handle_attr(struct vio_driver_state *vio, void *arg)
246 {
247 	struct vdc_port *port = to_vdc_port(vio);
248 	struct vio_disk_attr_info *pkt = arg;
249 
250 	viodbg(HS, "GOT ATTR stype[0x%x] ops[%llx] disk_size[%llu] disk_type[%x] "
251 	       "mtype[0x%x] xfer_mode[0x%x] blksz[%u] max_xfer[%llu]\n",
252 	       pkt->tag.stype, pkt->operations,
253 	       pkt->vdisk_size, pkt->vdisk_type, pkt->vdisk_mtype,
254 	       pkt->xfer_mode, pkt->vdisk_block_size,
255 	       pkt->max_xfer_size);
256 
257 	if (pkt->tag.stype == VIO_SUBTYPE_ACK) {
258 		switch (pkt->vdisk_type) {
259 		case VD_DISK_TYPE_DISK:
260 		case VD_DISK_TYPE_SLICE:
261 			break;
262 
263 		default:
264 			printk(KERN_ERR PFX "%s: Bogus vdisk_type 0x%x\n",
265 			       vio->name, pkt->vdisk_type);
266 			return -ECONNRESET;
267 		}
268 
269 		if (pkt->vdisk_block_size > port->vdisk_block_size) {
270 			printk(KERN_ERR PFX "%s: BLOCK size increased "
271 			       "%u --> %u\n",
272 			       vio->name,
273 			       port->vdisk_block_size, pkt->vdisk_block_size);
274 			return -ECONNRESET;
275 		}
276 
277 		port->operations = pkt->operations;
278 		port->vdisk_type = pkt->vdisk_type;
279 		if (vdc_version_supported(port, 1, 1)) {
280 			port->vdisk_size = pkt->vdisk_size;
281 			port->vdisk_mtype = pkt->vdisk_mtype;
282 		}
283 		if (pkt->max_xfer_size < port->max_xfer_size)
284 			port->max_xfer_size = pkt->max_xfer_size;
285 		port->vdisk_block_size = pkt->vdisk_block_size;
286 
287 		port->vdisk_phys_blksz = VDC_DEFAULT_BLK_SIZE;
288 		if (vdc_version_supported(port, 1, 2))
289 			port->vdisk_phys_blksz = pkt->phys_block_size;
290 
291 		return 0;
292 	} else {
293 		printk(KERN_ERR PFX "%s: Attribute NACK\n", vio->name);
294 
295 		return -ECONNRESET;
296 	}
297 }
298 
299 static void vdc_end_special(struct vdc_port *port, struct vio_disk_desc *desc)
300 {
301 	int err = desc->status;
302 
303 	vdc_finish(&port->vio, -err, WAITING_FOR_GEN_CMD);
304 }
305 
306 static void vdc_end_one(struct vdc_port *port, struct vio_dring_state *dr,
307 			unsigned int index)
308 {
309 	struct vio_disk_desc *desc = vio_dring_entry(dr, index);
310 	struct vdc_req_entry *rqe = &port->rq_arr[index];
311 	struct request *req;
312 
313 	if (unlikely(desc->hdr.state != VIO_DESC_DONE))
314 		return;
315 
316 	ldc_unmap(port->vio.lp, desc->cookies, desc->ncookies);
317 	desc->hdr.state = VIO_DESC_FREE;
318 	dr->cons = vio_dring_next(dr, index);
319 
320 	req = rqe->req;
321 	if (req == NULL) {
322 		vdc_end_special(port, desc);
323 		return;
324 	}
325 
326 	rqe->req = NULL;
327 
328 	blk_mq_end_request(req, desc->status ? BLK_STS_IOERR : 0);
329 
330 	vdc_blk_queue_start(port);
331 }
332 
333 static int vdc_ack(struct vdc_port *port, void *msgbuf)
334 {
335 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
336 	struct vio_dring_data *pkt = msgbuf;
337 
338 	if (unlikely(pkt->dring_ident != dr->ident ||
339 		     pkt->start_idx != pkt->end_idx ||
340 		     pkt->start_idx >= VDC_TX_RING_SIZE))
341 		return 0;
342 
343 	vdc_end_one(port, dr, pkt->start_idx);
344 
345 	return 0;
346 }
347 
348 static int vdc_nack(struct vdc_port *port, void *msgbuf)
349 {
350 	/* XXX Implement me XXX */
351 	return 0;
352 }
353 
354 static void vdc_event(void *arg, int event)
355 {
356 	struct vdc_port *port = arg;
357 	struct vio_driver_state *vio = &port->vio;
358 	unsigned long flags;
359 	int err;
360 
361 	spin_lock_irqsave(&vio->lock, flags);
362 
363 	if (unlikely(event == LDC_EVENT_RESET)) {
364 		vio_link_state_change(vio, event);
365 		queue_work(sunvdc_wq, &port->ldc_reset_work);
366 		goto out;
367 	}
368 
369 	if (unlikely(event == LDC_EVENT_UP)) {
370 		vio_link_state_change(vio, event);
371 		goto out;
372 	}
373 
374 	if (unlikely(event != LDC_EVENT_DATA_READY)) {
375 		pr_warn(PFX "Unexpected LDC event %d\n", event);
376 		goto out;
377 	}
378 
379 	err = 0;
380 	while (1) {
381 		union {
382 			struct vio_msg_tag tag;
383 			u64 raw[8];
384 		} msgbuf;
385 
386 		err = ldc_read(vio->lp, &msgbuf, sizeof(msgbuf));
387 		if (unlikely(err < 0)) {
388 			if (err == -ECONNRESET)
389 				vio_conn_reset(vio);
390 			break;
391 		}
392 		if (err == 0)
393 			break;
394 		viodbg(DATA, "TAG [%02x:%02x:%04x:%08x]\n",
395 		       msgbuf.tag.type,
396 		       msgbuf.tag.stype,
397 		       msgbuf.tag.stype_env,
398 		       msgbuf.tag.sid);
399 		err = vio_validate_sid(vio, &msgbuf.tag);
400 		if (err < 0)
401 			break;
402 
403 		if (likely(msgbuf.tag.type == VIO_TYPE_DATA)) {
404 			if (msgbuf.tag.stype == VIO_SUBTYPE_ACK)
405 				err = vdc_ack(port, &msgbuf);
406 			else if (msgbuf.tag.stype == VIO_SUBTYPE_NACK)
407 				err = vdc_nack(port, &msgbuf);
408 			else
409 				err = vdc_handle_unknown(port, &msgbuf);
410 		} else if (msgbuf.tag.type == VIO_TYPE_CTRL) {
411 			err = vio_control_pkt_engine(vio, &msgbuf);
412 		} else {
413 			err = vdc_handle_unknown(port, &msgbuf);
414 		}
415 		if (err < 0)
416 			break;
417 	}
418 	if (err < 0)
419 		vdc_finish(&port->vio, err, WAITING_FOR_ANY);
420 out:
421 	spin_unlock_irqrestore(&vio->lock, flags);
422 }
423 
424 static int __vdc_tx_trigger(struct vdc_port *port)
425 {
426 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
427 	struct vio_dring_data hdr = {
428 		.tag = {
429 			.type		= VIO_TYPE_DATA,
430 			.stype		= VIO_SUBTYPE_INFO,
431 			.stype_env	= VIO_DRING_DATA,
432 			.sid		= vio_send_sid(&port->vio),
433 		},
434 		.dring_ident		= dr->ident,
435 		.start_idx		= dr->prod,
436 		.end_idx		= dr->prod,
437 	};
438 	int err, delay;
439 	int retries = 0;
440 
441 	hdr.seq = dr->snd_nxt;
442 	delay = 1;
443 	do {
444 		err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
445 		if (err > 0) {
446 			dr->snd_nxt++;
447 			break;
448 		}
449 		udelay(delay);
450 		if ((delay <<= 1) > 128)
451 			delay = 128;
452 		if (retries++ > VDC_MAX_RETRIES)
453 			break;
454 	} while (err == -EAGAIN);
455 
456 	if (err == -ENOTCONN)
457 		vdc_ldc_reset(port);
458 	return err;
459 }
460 
461 static int __send_request(struct request *req)
462 {
463 	struct vdc_port *port = req->q->disk->private_data;
464 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
465 	struct scatterlist sg[MAX_RING_COOKIES];
466 	struct vdc_req_entry *rqe;
467 	struct vio_disk_desc *desc;
468 	unsigned int map_perm;
469 	int nsg, err, i;
470 	u64 len;
471 	u8 op;
472 
473 	if (WARN_ON(port->ring_cookies > MAX_RING_COOKIES))
474 		return -EINVAL;
475 
476 	map_perm = LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_IO;
477 
478 	if (rq_data_dir(req) == READ) {
479 		map_perm |= LDC_MAP_W;
480 		op = VD_OP_BREAD;
481 	} else {
482 		map_perm |= LDC_MAP_R;
483 		op = VD_OP_BWRITE;
484 	}
485 
486 	sg_init_table(sg, port->ring_cookies);
487 	nsg = blk_rq_map_sg(req, sg);
488 
489 	len = 0;
490 	for (i = 0; i < nsg; i++)
491 		len += sg[i].length;
492 
493 	desc = vio_dring_cur(dr);
494 
495 	err = ldc_map_sg(port->vio.lp, sg, nsg,
496 			 desc->cookies, port->ring_cookies,
497 			 map_perm);
498 	if (err < 0) {
499 		printk(KERN_ERR PFX "ldc_map_sg() failure, err=%d.\n", err);
500 		return err;
501 	}
502 
503 	rqe = &port->rq_arr[dr->prod];
504 	rqe->req = req;
505 
506 	desc->hdr.ack = VIO_ACK_ENABLE;
507 	desc->req_id = port->req_id;
508 	desc->operation = op;
509 	if (port->vdisk_type == VD_DISK_TYPE_DISK) {
510 		desc->slice = 0xff;
511 	} else {
512 		desc->slice = 0;
513 	}
514 	desc->status = ~0;
515 	desc->offset = (blk_rq_pos(req) << 9) / port->vdisk_block_size;
516 	desc->size = len;
517 	desc->ncookies = err;
518 
519 	/* This has to be a non-SMP write barrier because we are writing
520 	 * to memory which is shared with the peer LDOM.
521 	 */
522 	wmb();
523 	desc->hdr.state = VIO_DESC_READY;
524 
525 	err = __vdc_tx_trigger(port);
526 	if (err < 0) {
527 		printk(KERN_ERR PFX "vdc_tx_trigger() failure, err=%d\n", err);
528 	} else {
529 		port->req_id++;
530 		dr->prod = vio_dring_next(dr, dr->prod);
531 	}
532 
533 	return err;
534 }
535 
536 static blk_status_t vdc_queue_rq(struct blk_mq_hw_ctx *hctx,
537 				 const struct blk_mq_queue_data *bd)
538 {
539 	struct vdc_port *port = hctx->queue->queuedata;
540 	struct vio_dring_state *dr;
541 	unsigned long flags;
542 
543 	dr = &port->vio.drings[VIO_DRIVER_TX_RING];
544 
545 	blk_mq_start_request(bd->rq);
546 
547 	spin_lock_irqsave(&port->vio.lock, flags);
548 
549 	/*
550 	 * Doing drain, just end the request in error
551 	 */
552 	if (unlikely(port->drain)) {
553 		spin_unlock_irqrestore(&port->vio.lock, flags);
554 		return BLK_STS_IOERR;
555 	}
556 
557 	if (unlikely(vdc_tx_dring_avail(dr) < 1)) {
558 		spin_unlock_irqrestore(&port->vio.lock, flags);
559 		blk_mq_stop_hw_queue(hctx);
560 		return BLK_STS_DEV_RESOURCE;
561 	}
562 
563 	if (__send_request(bd->rq) < 0) {
564 		spin_unlock_irqrestore(&port->vio.lock, flags);
565 		return BLK_STS_IOERR;
566 	}
567 
568 	spin_unlock_irqrestore(&port->vio.lock, flags);
569 	return BLK_STS_OK;
570 }
571 
572 static int generic_request(struct vdc_port *port, u8 op, void *buf, int len)
573 {
574 	struct vio_dring_state *dr;
575 	struct vio_completion comp;
576 	struct vio_disk_desc *desc;
577 	unsigned int map_perm;
578 	unsigned long flags;
579 	int op_len, err;
580 	void *req_buf;
581 
582 	if (!(((u64)1 << (u64)op) & port->operations))
583 		return -EOPNOTSUPP;
584 
585 	switch (op) {
586 	case VD_OP_BREAD:
587 	case VD_OP_BWRITE:
588 	default:
589 		return -EINVAL;
590 
591 	case VD_OP_FLUSH:
592 		op_len = 0;
593 		map_perm = 0;
594 		break;
595 
596 	case VD_OP_GET_WCE:
597 		op_len = sizeof(u32);
598 		map_perm = LDC_MAP_W;
599 		break;
600 
601 	case VD_OP_SET_WCE:
602 		op_len = sizeof(u32);
603 		map_perm = LDC_MAP_R;
604 		break;
605 
606 	case VD_OP_GET_VTOC:
607 		op_len = sizeof(struct vio_disk_vtoc);
608 		map_perm = LDC_MAP_W;
609 		break;
610 
611 	case VD_OP_SET_VTOC:
612 		op_len = sizeof(struct vio_disk_vtoc);
613 		map_perm = LDC_MAP_R;
614 		break;
615 
616 	case VD_OP_GET_DISKGEOM:
617 		op_len = sizeof(struct vio_disk_geom);
618 		map_perm = LDC_MAP_W;
619 		break;
620 
621 	case VD_OP_SET_DISKGEOM:
622 		op_len = sizeof(struct vio_disk_geom);
623 		map_perm = LDC_MAP_R;
624 		break;
625 
626 	case VD_OP_SCSICMD:
627 		op_len = 16;
628 		map_perm = LDC_MAP_RW;
629 		break;
630 
631 	case VD_OP_GET_DEVID:
632 		op_len = sizeof(struct vio_disk_devid);
633 		map_perm = LDC_MAP_W;
634 		break;
635 
636 	case VD_OP_GET_EFI:
637 	case VD_OP_SET_EFI:
638 		return -EOPNOTSUPP;
639 	}
640 
641 	map_perm |= LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_IO;
642 
643 	op_len = (op_len + 7) & ~7;
644 	req_buf = kzalloc(op_len, GFP_KERNEL);
645 	if (!req_buf)
646 		return -ENOMEM;
647 
648 	if (len > op_len)
649 		len = op_len;
650 
651 	if (map_perm & LDC_MAP_R)
652 		memcpy(req_buf, buf, len);
653 
654 	spin_lock_irqsave(&port->vio.lock, flags);
655 
656 	dr = &port->vio.drings[VIO_DRIVER_TX_RING];
657 
658 	/* XXX If we want to use this code generically we have to
659 	 * XXX handle TX ring exhaustion etc.
660 	 */
661 	desc = vio_dring_cur(dr);
662 
663 	err = ldc_map_single(port->vio.lp, req_buf, op_len,
664 			     desc->cookies, port->ring_cookies,
665 			     map_perm);
666 	if (err < 0) {
667 		spin_unlock_irqrestore(&port->vio.lock, flags);
668 		kfree(req_buf);
669 		return err;
670 	}
671 
672 	init_completion(&comp.com);
673 	comp.waiting_for = WAITING_FOR_GEN_CMD;
674 	port->vio.cmp = &comp;
675 
676 	desc->hdr.ack = VIO_ACK_ENABLE;
677 	desc->req_id = port->req_id;
678 	desc->operation = op;
679 	desc->slice = 0;
680 	desc->status = ~0;
681 	desc->offset = 0;
682 	desc->size = op_len;
683 	desc->ncookies = err;
684 
685 	/* This has to be a non-SMP write barrier because we are writing
686 	 * to memory which is shared with the peer LDOM.
687 	 */
688 	wmb();
689 	desc->hdr.state = VIO_DESC_READY;
690 
691 	err = __vdc_tx_trigger(port);
692 	if (err >= 0) {
693 		port->req_id++;
694 		dr->prod = vio_dring_next(dr, dr->prod);
695 		spin_unlock_irqrestore(&port->vio.lock, flags);
696 
697 		wait_for_completion(&comp.com);
698 		err = comp.err;
699 	} else {
700 		port->vio.cmp = NULL;
701 		spin_unlock_irqrestore(&port->vio.lock, flags);
702 	}
703 
704 	if (map_perm & LDC_MAP_W)
705 		memcpy(buf, req_buf, len);
706 
707 	kfree(req_buf);
708 
709 	return err;
710 }
711 
712 static int vdc_alloc_tx_ring(struct vdc_port *port)
713 {
714 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
715 	unsigned long len, entry_size;
716 	int ncookies;
717 	void *dring;
718 
719 	entry_size = sizeof(struct vio_disk_desc) +
720 		(sizeof(struct ldc_trans_cookie) * port->ring_cookies);
721 	len = (VDC_TX_RING_SIZE * entry_size);
722 
723 	ncookies = VIO_MAX_RING_COOKIES;
724 	dring = ldc_alloc_exp_dring(port->vio.lp, len,
725 				    dr->cookies, &ncookies,
726 				    (LDC_MAP_SHADOW |
727 				     LDC_MAP_DIRECT |
728 				     LDC_MAP_RW));
729 	if (IS_ERR(dring))
730 		return PTR_ERR(dring);
731 
732 	dr->base = dring;
733 	dr->entry_size = entry_size;
734 	dr->num_entries = VDC_TX_RING_SIZE;
735 	dr->prod = dr->cons = 0;
736 	dr->pending = VDC_TX_RING_SIZE;
737 	dr->ncookies = ncookies;
738 
739 	return 0;
740 }
741 
742 static void vdc_free_tx_ring(struct vdc_port *port)
743 {
744 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
745 
746 	if (dr->base) {
747 		ldc_free_exp_dring(port->vio.lp, dr->base,
748 				   (dr->entry_size * dr->num_entries),
749 				   dr->cookies, dr->ncookies);
750 		dr->base = NULL;
751 		dr->entry_size = 0;
752 		dr->num_entries = 0;
753 		dr->pending = 0;
754 		dr->ncookies = 0;
755 	}
756 }
757 
758 static int vdc_port_up(struct vdc_port *port)
759 {
760 	struct vio_completion comp;
761 
762 	init_completion(&comp.com);
763 	comp.err = 0;
764 	comp.waiting_for = WAITING_FOR_LINK_UP;
765 	port->vio.cmp = &comp;
766 
767 	vio_port_up(&port->vio);
768 	wait_for_completion(&comp.com);
769 	return comp.err;
770 }
771 
772 static void vdc_port_down(struct vdc_port *port)
773 {
774 	ldc_disconnect(port->vio.lp);
775 	ldc_unbind(port->vio.lp);
776 	vdc_free_tx_ring(port);
777 	vio_ldc_free(&port->vio);
778 }
779 
780 static const struct blk_mq_ops vdc_mq_ops = {
781 	.queue_rq	= vdc_queue_rq,
782 };
783 
784 static int probe_disk(struct vdc_port *port)
785 {
786 	struct queue_limits lim = {
787 		.physical_block_size		= port->vdisk_phys_blksz,
788 		.max_hw_sectors			= port->max_xfer_size,
789 		/* Each segment in a request is up to an aligned page in size. */
790 		.seg_boundary_mask		= PAGE_SIZE - 1,
791 		.max_segment_size		= PAGE_SIZE,
792 		.max_segments			= port->ring_cookies,
793 		.features			= BLK_FEAT_ROTATIONAL,
794 	};
795 	struct request_queue *q;
796 	struct gendisk *g;
797 	int err;
798 
799 	err = vdc_port_up(port);
800 	if (err)
801 		return err;
802 
803 	/* Using version 1.2 means vdisk_phys_blksz should be set unless the
804 	 * disk is reserved by another system.
805 	 */
806 	if (vdc_version_supported(port, 1, 2) && !port->vdisk_phys_blksz)
807 		return -ENODEV;
808 
809 	if (vdc_version_supported(port, 1, 1)) {
810 		/* vdisk_size should be set during the handshake, if it wasn't
811 		 * then the underlying disk is reserved by another system
812 		 */
813 		if (port->vdisk_size == -1)
814 			return -ENODEV;
815 	} else {
816 		struct vio_disk_geom geom;
817 
818 		err = generic_request(port, VD_OP_GET_DISKGEOM,
819 				      &geom, sizeof(geom));
820 		if (err < 0) {
821 			printk(KERN_ERR PFX "VD_OP_GET_DISKGEOM returns "
822 			       "error %d\n", err);
823 			return err;
824 		}
825 		port->vdisk_size = ((u64)geom.num_cyl *
826 				    (u64)geom.num_hd *
827 				    (u64)geom.num_sec);
828 	}
829 
830 	err = blk_mq_alloc_sq_tag_set(&port->tag_set, &vdc_mq_ops,
831 			VDC_TX_RING_SIZE, 0);
832 	if (err)
833 		return err;
834 
835 	g = blk_mq_alloc_disk(&port->tag_set, &lim, port);
836 	if (IS_ERR(g)) {
837 		printk(KERN_ERR PFX "%s: Could not allocate gendisk.\n",
838 		       port->vio.name);
839 		err = PTR_ERR(g);
840 		goto out_free_tag;
841 	}
842 
843 	port->disk = g;
844 	q = g->queue;
845 
846 	g->major = vdc_major;
847 	g->first_minor = port->vio.vdev->dev_no << PARTITION_SHIFT;
848 	g->minors = 1 << PARTITION_SHIFT;
849 	strcpy(g->disk_name, port->disk_name);
850 
851 	g->fops = &vdc_fops;
852 	g->queue = q;
853 	g->private_data = port;
854 
855 	set_capacity(g, port->vdisk_size);
856 
857 	if (vdc_version_supported(port, 1, 1)) {
858 		switch (port->vdisk_mtype) {
859 		case VD_MEDIA_TYPE_CD:
860 			pr_info(PFX "Virtual CDROM %s\n", port->disk_name);
861 			g->flags |= GENHD_FL_REMOVABLE;
862 			set_disk_ro(g, 1);
863 			break;
864 
865 		case VD_MEDIA_TYPE_DVD:
866 			pr_info(PFX "Virtual DVD %s\n", port->disk_name);
867 			g->flags |= GENHD_FL_REMOVABLE;
868 			set_disk_ro(g, 1);
869 			break;
870 
871 		case VD_MEDIA_TYPE_FIXED:
872 			pr_info(PFX "Virtual Hard disk %s\n", port->disk_name);
873 			break;
874 		}
875 	}
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 	put_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, const void *arg)
921 {
922 	struct vio_dev *vdev = to_vio_dev(dev);
923 	const struct vdc_check_port_data *port_data;
924 
925 	port_data = (const 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 		put_device(dev);
961 		return true;
962 	}
963 
964 	return false;
965 }
966 
967 static int vdc_port_probe(struct vio_dev *vdev, const struct vio_device_id *id)
968 {
969 	struct mdesc_handle *hp;
970 	struct vdc_port *port;
971 	int err;
972 	const u64 *ldc_timeout;
973 
974 	print_version();
975 
976 	hp = mdesc_grab();
977 	if (!hp)
978 		return -ENODEV;
979 
980 	err = -ENODEV;
981 	if ((vdev->dev_no << PARTITION_SHIFT) & ~(u64)MINORMASK) {
982 		printk(KERN_ERR PFX "Port id [%llu] too large.\n",
983 		       vdev->dev_no);
984 		goto err_out_release_mdesc;
985 	}
986 
987 	/* Check if this device is part of an mpgroup */
988 	if (vdc_port_mpgroup_check(vdev)) {
989 		printk(KERN_WARNING
990 			"VIO: Ignoring extra vdisk port %s",
991 			dev_name(&vdev->dev));
992 		goto err_out_release_mdesc;
993 	}
994 
995 	port = kzalloc(sizeof(*port), GFP_KERNEL);
996 	if (!port) {
997 		err = -ENOMEM;
998 		goto err_out_release_mdesc;
999 	}
1000 
1001 	if (vdev->dev_no >= 26)
1002 		snprintf(port->disk_name, sizeof(port->disk_name),
1003 			 VDCBLK_NAME "%c%c",
1004 			 'a' + ((int)vdev->dev_no / 26) - 1,
1005 			 'a' + ((int)vdev->dev_no % 26));
1006 	else
1007 		snprintf(port->disk_name, sizeof(port->disk_name),
1008 			 VDCBLK_NAME "%c", 'a' + ((int)vdev->dev_no % 26));
1009 	port->vdisk_size = -1;
1010 
1011 	/* Actual wall time may be double due to do_generic_file_read() doing
1012 	 * a readahead I/O first, and once that fails it will try to read a
1013 	 * single page.
1014 	 */
1015 	ldc_timeout = mdesc_get_property(hp, vdev->mp, "vdc-timeout", NULL);
1016 	port->ldc_timeout = ldc_timeout ? *ldc_timeout : 0;
1017 	INIT_DELAYED_WORK(&port->ldc_reset_timer_work, vdc_ldc_reset_timer_work);
1018 	INIT_WORK(&port->ldc_reset_work, vdc_ldc_reset_work);
1019 
1020 	err = vio_driver_init(&port->vio, vdev, VDEV_DISK,
1021 			      vdc_versions, ARRAY_SIZE(vdc_versions),
1022 			      &vdc_vio_ops, port->disk_name);
1023 	if (err)
1024 		goto err_out_free_port;
1025 
1026 	port->vdisk_block_size = VDC_DEFAULT_BLK_SIZE;
1027 	port->max_xfer_size = MAX_XFER_SIZE;
1028 	port->ring_cookies = MAX_RING_COOKIES;
1029 
1030 	err = vio_ldc_alloc(&port->vio, &vdc_ldc_cfg, port);
1031 	if (err)
1032 		goto err_out_free_port;
1033 
1034 	err = vdc_alloc_tx_ring(port);
1035 	if (err)
1036 		goto err_out_free_ldc;
1037 
1038 	err = probe_disk(port);
1039 	if (err)
1040 		goto err_out_free_tx_ring;
1041 
1042 	/* Note that the device driver_data is used to determine
1043 	 * whether the port has been probed.
1044 	 */
1045 	dev_set_drvdata(&vdev->dev, port);
1046 
1047 	mdesc_release(hp);
1048 
1049 	return 0;
1050 
1051 err_out_free_tx_ring:
1052 	vdc_free_tx_ring(port);
1053 
1054 err_out_free_ldc:
1055 	vio_ldc_free(&port->vio);
1056 
1057 err_out_free_port:
1058 	kfree(port);
1059 
1060 err_out_release_mdesc:
1061 	mdesc_release(hp);
1062 	return err;
1063 }
1064 
1065 static void vdc_port_remove(struct vio_dev *vdev)
1066 {
1067 	struct vdc_port *port = dev_get_drvdata(&vdev->dev);
1068 
1069 	if (port) {
1070 		blk_mq_stop_hw_queues(port->disk->queue);
1071 
1072 		flush_work(&port->ldc_reset_work);
1073 		cancel_delayed_work_sync(&port->ldc_reset_timer_work);
1074 		timer_delete_sync(&port->vio.timer);
1075 
1076 		del_gendisk(port->disk);
1077 		put_disk(port->disk);
1078 		blk_mq_free_tag_set(&port->tag_set);
1079 
1080 		vdc_free_tx_ring(port);
1081 		vio_ldc_free(&port->vio);
1082 
1083 		dev_set_drvdata(&vdev->dev, NULL);
1084 
1085 		kfree(port);
1086 	}
1087 }
1088 
1089 static void vdc_requeue_inflight(struct vdc_port *port)
1090 {
1091 	struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1092 	u32 idx;
1093 
1094 	for (idx = dr->cons; idx != dr->prod; idx = vio_dring_next(dr, idx)) {
1095 		struct vio_disk_desc *desc = vio_dring_entry(dr, idx);
1096 		struct vdc_req_entry *rqe = &port->rq_arr[idx];
1097 		struct request *req;
1098 
1099 		ldc_unmap(port->vio.lp, desc->cookies, desc->ncookies);
1100 		desc->hdr.state = VIO_DESC_FREE;
1101 		dr->cons = vio_dring_next(dr, idx);
1102 
1103 		req = rqe->req;
1104 		if (req == NULL) {
1105 			vdc_end_special(port, desc);
1106 			continue;
1107 		}
1108 
1109 		rqe->req = NULL;
1110 		blk_mq_requeue_request(req, false);
1111 	}
1112 }
1113 
1114 static void vdc_queue_drain(struct vdc_port *port)
1115 {
1116 	struct request_queue *q = port->disk->queue;
1117 	unsigned int memflags;
1118 
1119 	/*
1120 	 * Mark the queue as draining, then freeze/quiesce to ensure
1121 	 * that all existing requests are seen in ->queue_rq() and killed
1122 	 */
1123 	port->drain = 1;
1124 	spin_unlock_irq(&port->vio.lock);
1125 
1126 	memflags = blk_mq_freeze_queue(q);
1127 	blk_mq_quiesce_queue(q);
1128 
1129 	spin_lock_irq(&port->vio.lock);
1130 	port->drain = 0;
1131 	blk_mq_unquiesce_queue(q);
1132 	blk_mq_unfreeze_queue(q, memflags);
1133 }
1134 
1135 static void vdc_ldc_reset_timer_work(struct work_struct *work)
1136 {
1137 	struct vdc_port *port;
1138 	struct vio_driver_state *vio;
1139 
1140 	port = container_of(work, struct vdc_port, ldc_reset_timer_work.work);
1141 	vio = &port->vio;
1142 
1143 	spin_lock_irq(&vio->lock);
1144 	if (!(port->vio.hs_state & VIO_HS_COMPLETE)) {
1145 		pr_warn(PFX "%s ldc down %llu seconds, draining queue\n",
1146 			port->disk_name, port->ldc_timeout);
1147 		vdc_queue_drain(port);
1148 		vdc_blk_queue_start(port);
1149 	}
1150 	spin_unlock_irq(&vio->lock);
1151 }
1152 
1153 static void vdc_ldc_reset_work(struct work_struct *work)
1154 {
1155 	struct vdc_port *port;
1156 	struct vio_driver_state *vio;
1157 	unsigned long flags;
1158 
1159 	port = container_of(work, struct vdc_port, ldc_reset_work);
1160 	vio = &port->vio;
1161 
1162 	spin_lock_irqsave(&vio->lock, flags);
1163 	vdc_ldc_reset(port);
1164 	spin_unlock_irqrestore(&vio->lock, flags);
1165 }
1166 
1167 static void vdc_ldc_reset(struct vdc_port *port)
1168 {
1169 	int err;
1170 
1171 	assert_spin_locked(&port->vio.lock);
1172 
1173 	pr_warn(PFX "%s ldc link reset\n", port->disk_name);
1174 	blk_mq_stop_hw_queues(port->disk->queue);
1175 	vdc_requeue_inflight(port);
1176 	vdc_port_down(port);
1177 
1178 	err = vio_ldc_alloc(&port->vio, &vdc_ldc_cfg, port);
1179 	if (err) {
1180 		pr_err(PFX "%s vio_ldc_alloc:%d\n", port->disk_name, err);
1181 		return;
1182 	}
1183 
1184 	err = vdc_alloc_tx_ring(port);
1185 	if (err) {
1186 		pr_err(PFX "%s vio_alloc_tx_ring:%d\n", port->disk_name, err);
1187 		goto err_free_ldc;
1188 	}
1189 
1190 	if (port->ldc_timeout)
1191 		mod_delayed_work(system_percpu_wq, &port->ldc_reset_timer_work,
1192 			  round_jiffies(jiffies + HZ * port->ldc_timeout));
1193 	mod_timer(&port->vio.timer, round_jiffies(jiffies + HZ));
1194 	return;
1195 
1196 err_free_ldc:
1197 	vio_ldc_free(&port->vio);
1198 }
1199 
1200 static const struct vio_device_id vdc_port_match[] = {
1201 	{
1202 		.type = "vdc-port",
1203 	},
1204 	{},
1205 };
1206 MODULE_DEVICE_TABLE(vio, vdc_port_match);
1207 
1208 static struct vio_driver vdc_port_driver = {
1209 	.id_table	= vdc_port_match,
1210 	.probe		= vdc_port_probe,
1211 	.remove		= vdc_port_remove,
1212 	.name		= "vdc_port",
1213 };
1214 
1215 static int __init vdc_init(void)
1216 {
1217 	int err;
1218 
1219 	sunvdc_wq = alloc_workqueue("sunvdc", WQ_PERCPU, 0);
1220 	if (!sunvdc_wq)
1221 		return -ENOMEM;
1222 
1223 	err = register_blkdev(0, VDCBLK_NAME);
1224 	if (err < 0)
1225 		goto out_free_wq;
1226 
1227 	vdc_major = err;
1228 
1229 	err = vio_register_driver(&vdc_port_driver);
1230 	if (err)
1231 		goto out_unregister_blkdev;
1232 
1233 	return 0;
1234 
1235 out_unregister_blkdev:
1236 	unregister_blkdev(vdc_major, VDCBLK_NAME);
1237 	vdc_major = 0;
1238 
1239 out_free_wq:
1240 	destroy_workqueue(sunvdc_wq);
1241 	return err;
1242 }
1243 
1244 static void __exit vdc_exit(void)
1245 {
1246 	vio_unregister_driver(&vdc_port_driver);
1247 	unregister_blkdev(vdc_major, VDCBLK_NAME);
1248 	destroy_workqueue(sunvdc_wq);
1249 }
1250 
1251 module_init(vdc_init);
1252 module_exit(vdc_exit);
1253