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
to_vdc_port(struct vio_driver_state * vio)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
vdc_version_supported(struct vdc_port * port,u16 major,u16 minor)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
vdc_tx_dring_avail(struct vio_dring_state * dr)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
vdc_getgeo(struct gendisk * disk,struct hd_geometry * geo)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 */
vdc_ioctl(struct block_device * bdev,blk_mode_t mode,unsigned command,unsigned long argument)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
vdc_blk_queue_start(struct vdc_port * port)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
vdc_finish(struct vio_driver_state * vio,int err,int waiting_for)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
vdc_handshake_complete(struct vio_driver_state * vio)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
vdc_handle_unknown(struct vdc_port * port,void * arg)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
vdc_send_attr(struct vio_driver_state * vio)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
vdc_handle_attr(struct vio_driver_state * vio,void * arg)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
vdc_end_special(struct vdc_port * port,struct vio_disk_desc * desc)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
vdc_end_one(struct vdc_port * port,struct vio_dring_state * dr,unsigned int index)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
vdc_ack(struct vdc_port * port,void * msgbuf)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
vdc_nack(struct vdc_port * port,void * msgbuf)348 static int vdc_nack(struct vdc_port *port, void *msgbuf)
349 {
350 /* XXX Implement me XXX */
351 return 0;
352 }
353
vdc_event(void * arg,int event)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
__vdc_tx_trigger(struct vdc_port * port)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
__send_request(struct request * req)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 /*
529 * If the port was reset (-ENOTCONN), the dring and the
530 * LDC channel including all of its mappings are already
531 * torn down and reallocated - there is nothing to undo
532 * and @desc must not be touched.
533 *
534 * For any other failure the descriptor was never handed
535 * to the peer: unmap the cookies and free the descriptor
536 * again, so that a later retry of the request does not
537 * leak LDC map table entries.
538 */
539 if (err != -ENOTCONN) {
540 ldc_unmap(port->vio.lp, desc->cookies,
541 desc->ncookies);
542 desc->hdr.state = VIO_DESC_FREE;
543 rqe->req = NULL;
544 }
545 } else {
546 port->req_id++;
547 dr->prod = vio_dring_next(dr, dr->prod);
548 }
549
550 return err;
551 }
552
vdc_queue_rq(struct blk_mq_hw_ctx * hctx,const struct blk_mq_queue_data * bd)553 static blk_status_t vdc_queue_rq(struct blk_mq_hw_ctx *hctx,
554 const struct blk_mq_queue_data *bd)
555 {
556 struct vdc_port *port = hctx->queue->queuedata;
557 struct vio_dring_state *dr;
558 unsigned long flags;
559 int ret;
560
561 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
562
563 blk_mq_start_request(bd->rq);
564
565 spin_lock_irqsave(&port->vio.lock, flags);
566
567 /*
568 * Doing drain, just end the request in error
569 */
570 if (unlikely(port->drain)) {
571 spin_unlock_irqrestore(&port->vio.lock, flags);
572 return BLK_STS_IOERR;
573 }
574
575 if (unlikely(vdc_tx_dring_avail(dr) < 1)) {
576 spin_unlock_irqrestore(&port->vio.lock, flags);
577 blk_mq_stop_hw_queue(hctx);
578 return BLK_STS_DEV_RESOURCE;
579 }
580
581 ret = __send_request(bd->rq);
582 if (ret == -EAGAIN) {
583 spin_unlock_irqrestore(&port->vio.lock, flags);
584 /* already spun for 10msec, defer 10msec and retry */
585 blk_mq_delay_kick_requeue_list(hctx->queue, 10);
586 return BLK_STS_DEV_RESOURCE;
587 } else if (ret < 0) {
588 spin_unlock_irqrestore(&port->vio.lock, flags);
589 return BLK_STS_IOERR;
590 }
591
592 spin_unlock_irqrestore(&port->vio.lock, flags);
593 return BLK_STS_OK;
594 }
595
generic_request(struct vdc_port * port,u8 op,void * buf,int len)596 static int generic_request(struct vdc_port *port, u8 op, void *buf, int len)
597 {
598 struct vio_dring_state *dr;
599 struct vio_completion comp;
600 struct vio_disk_desc *desc;
601 unsigned int map_perm;
602 unsigned long flags;
603 int op_len, err;
604 void *req_buf;
605
606 if (!(((u64)1 << (u64)op) & port->operations))
607 return -EOPNOTSUPP;
608
609 switch (op) {
610 case VD_OP_BREAD:
611 case VD_OP_BWRITE:
612 default:
613 return -EINVAL;
614
615 case VD_OP_FLUSH:
616 op_len = 0;
617 map_perm = 0;
618 break;
619
620 case VD_OP_GET_WCE:
621 op_len = sizeof(u32);
622 map_perm = LDC_MAP_W;
623 break;
624
625 case VD_OP_SET_WCE:
626 op_len = sizeof(u32);
627 map_perm = LDC_MAP_R;
628 break;
629
630 case VD_OP_GET_VTOC:
631 op_len = sizeof(struct vio_disk_vtoc);
632 map_perm = LDC_MAP_W;
633 break;
634
635 case VD_OP_SET_VTOC:
636 op_len = sizeof(struct vio_disk_vtoc);
637 map_perm = LDC_MAP_R;
638 break;
639
640 case VD_OP_GET_DISKGEOM:
641 op_len = sizeof(struct vio_disk_geom);
642 map_perm = LDC_MAP_W;
643 break;
644
645 case VD_OP_SET_DISKGEOM:
646 op_len = sizeof(struct vio_disk_geom);
647 map_perm = LDC_MAP_R;
648 break;
649
650 case VD_OP_SCSICMD:
651 op_len = 16;
652 map_perm = LDC_MAP_RW;
653 break;
654
655 case VD_OP_GET_DEVID:
656 op_len = sizeof(struct vio_disk_devid);
657 map_perm = LDC_MAP_W;
658 break;
659
660 case VD_OP_GET_EFI:
661 case VD_OP_SET_EFI:
662 return -EOPNOTSUPP;
663 }
664
665 map_perm |= LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_IO;
666
667 op_len = (op_len + 7) & ~7;
668 req_buf = kzalloc(op_len, GFP_KERNEL);
669 if (!req_buf)
670 return -ENOMEM;
671
672 if (len > op_len)
673 len = op_len;
674
675 if (map_perm & LDC_MAP_R)
676 memcpy(req_buf, buf, len);
677
678 spin_lock_irqsave(&port->vio.lock, flags);
679
680 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
681
682 /* XXX If we want to use this code generically we have to
683 * XXX handle TX ring exhaustion etc.
684 */
685 desc = vio_dring_cur(dr);
686
687 err = ldc_map_single(port->vio.lp, req_buf, op_len,
688 desc->cookies, port->ring_cookies,
689 map_perm);
690 if (err < 0) {
691 spin_unlock_irqrestore(&port->vio.lock, flags);
692 kfree(req_buf);
693 return err;
694 }
695
696 init_completion(&comp.com);
697 comp.waiting_for = WAITING_FOR_GEN_CMD;
698 port->vio.cmp = ∁
699
700 desc->hdr.ack = VIO_ACK_ENABLE;
701 desc->req_id = port->req_id;
702 desc->operation = op;
703 desc->slice = 0;
704 desc->status = ~0;
705 desc->offset = 0;
706 desc->size = op_len;
707 desc->ncookies = err;
708
709 /* This has to be a non-SMP write barrier because we are writing
710 * to memory which is shared with the peer LDOM.
711 */
712 wmb();
713 desc->hdr.state = VIO_DESC_READY;
714
715 err = __vdc_tx_trigger(port);
716 if (err >= 0) {
717 port->req_id++;
718 dr->prod = vio_dring_next(dr, dr->prod);
719 spin_unlock_irqrestore(&port->vio.lock, flags);
720
721 wait_for_completion(&comp.com);
722 err = comp.err;
723 } else {
724 port->vio.cmp = NULL;
725 spin_unlock_irqrestore(&port->vio.lock, flags);
726 }
727
728 if (map_perm & LDC_MAP_W)
729 memcpy(buf, req_buf, len);
730
731 kfree(req_buf);
732
733 return err;
734 }
735
vdc_alloc_tx_ring(struct vdc_port * port)736 static int vdc_alloc_tx_ring(struct vdc_port *port)
737 {
738 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
739 unsigned long len, entry_size;
740 int ncookies;
741 void *dring;
742
743 entry_size = sizeof(struct vio_disk_desc) +
744 (sizeof(struct ldc_trans_cookie) * port->ring_cookies);
745 len = (VDC_TX_RING_SIZE * entry_size);
746
747 ncookies = VIO_MAX_RING_COOKIES;
748 dring = ldc_alloc_exp_dring(port->vio.lp, len,
749 dr->cookies, &ncookies,
750 (LDC_MAP_SHADOW |
751 LDC_MAP_DIRECT |
752 LDC_MAP_RW));
753 if (IS_ERR(dring))
754 return PTR_ERR(dring);
755
756 dr->base = dring;
757 dr->entry_size = entry_size;
758 dr->num_entries = VDC_TX_RING_SIZE;
759 dr->prod = dr->cons = 0;
760 dr->pending = VDC_TX_RING_SIZE;
761 dr->ncookies = ncookies;
762
763 return 0;
764 }
765
vdc_free_tx_ring(struct vdc_port * port)766 static void vdc_free_tx_ring(struct vdc_port *port)
767 {
768 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
769
770 if (dr->base) {
771 ldc_free_exp_dring(port->vio.lp, dr->base,
772 (dr->entry_size * dr->num_entries),
773 dr->cookies, dr->ncookies);
774 dr->base = NULL;
775 dr->entry_size = 0;
776 dr->num_entries = 0;
777 dr->pending = 0;
778 dr->ncookies = 0;
779 }
780 }
781
vdc_port_up(struct vdc_port * port)782 static int vdc_port_up(struct vdc_port *port)
783 {
784 struct vio_completion comp;
785
786 init_completion(&comp.com);
787 comp.err = 0;
788 comp.waiting_for = WAITING_FOR_LINK_UP;
789 port->vio.cmp = ∁
790
791 vio_port_up(&port->vio);
792 wait_for_completion(&comp.com);
793 return comp.err;
794 }
795
vdc_port_down(struct vdc_port * port)796 static void vdc_port_down(struct vdc_port *port)
797 {
798 ldc_disconnect(port->vio.lp);
799 ldc_unbind(port->vio.lp);
800 vdc_free_tx_ring(port);
801 vio_ldc_free(&port->vio);
802 }
803
804 static const struct blk_mq_ops vdc_mq_ops = {
805 .queue_rq = vdc_queue_rq,
806 };
807
probe_disk(struct vdc_port * port)808 static int probe_disk(struct vdc_port *port)
809 {
810 struct queue_limits lim = {
811 .physical_block_size = port->vdisk_phys_blksz,
812 .max_hw_sectors = port->max_xfer_size,
813 /* Each segment in a request is up to an aligned page in size. */
814 .seg_boundary_mask = PAGE_SIZE - 1,
815 .max_segment_size = PAGE_SIZE,
816 .max_segments = port->ring_cookies,
817 .features = BLK_FEAT_ROTATIONAL,
818 };
819 struct request_queue *q;
820 struct gendisk *g;
821 int err;
822
823 err = vdc_port_up(port);
824 if (err)
825 return err;
826
827 /* Using version 1.2 means vdisk_phys_blksz should be set unless the
828 * disk is reserved by another system.
829 */
830 if (vdc_version_supported(port, 1, 2) && !port->vdisk_phys_blksz)
831 return -ENODEV;
832
833 if (vdc_version_supported(port, 1, 1)) {
834 /* vdisk_size should be set during the handshake, if it wasn't
835 * then the underlying disk is reserved by another system
836 */
837 if (port->vdisk_size == -1)
838 return -ENODEV;
839 } else {
840 struct vio_disk_geom geom;
841
842 err = generic_request(port, VD_OP_GET_DISKGEOM,
843 &geom, sizeof(geom));
844 if (err < 0) {
845 printk(KERN_ERR PFX "VD_OP_GET_DISKGEOM returns "
846 "error %d\n", err);
847 return err;
848 }
849 port->vdisk_size = ((u64)geom.num_cyl *
850 (u64)geom.num_hd *
851 (u64)geom.num_sec);
852 }
853
854 err = blk_mq_alloc_sq_tag_set(&port->tag_set, &vdc_mq_ops,
855 VDC_TX_RING_SIZE, 0);
856 if (err)
857 return err;
858
859 g = blk_mq_alloc_disk(&port->tag_set, &lim, port);
860 if (IS_ERR(g)) {
861 printk(KERN_ERR PFX "%s: Could not allocate gendisk.\n",
862 port->vio.name);
863 err = PTR_ERR(g);
864 goto out_free_tag;
865 }
866
867 port->disk = g;
868 q = g->queue;
869
870 g->major = vdc_major;
871 g->first_minor = port->vio.vdev->dev_no << PARTITION_SHIFT;
872 g->minors = 1 << PARTITION_SHIFT;
873 strcpy(g->disk_name, port->disk_name);
874
875 g->fops = &vdc_fops;
876 g->queue = q;
877 g->private_data = port;
878
879 set_capacity(g, port->vdisk_size);
880
881 if (vdc_version_supported(port, 1, 1)) {
882 switch (port->vdisk_mtype) {
883 case VD_MEDIA_TYPE_CD:
884 pr_info(PFX "Virtual CDROM %s\n", port->disk_name);
885 g->flags |= GENHD_FL_REMOVABLE;
886 set_disk_ro(g, 1);
887 break;
888
889 case VD_MEDIA_TYPE_DVD:
890 pr_info(PFX "Virtual DVD %s\n", port->disk_name);
891 g->flags |= GENHD_FL_REMOVABLE;
892 set_disk_ro(g, 1);
893 break;
894
895 case VD_MEDIA_TYPE_FIXED:
896 pr_info(PFX "Virtual Hard disk %s\n", port->disk_name);
897 break;
898 }
899 }
900
901 pr_info(PFX "%s: %u sectors (%u MB) protocol %d.%d\n",
902 g->disk_name,
903 port->vdisk_size, (port->vdisk_size >> (20 - 9)),
904 port->vio.ver.major, port->vio.ver.minor);
905
906 err = device_add_disk(&port->vio.vdev->dev, g, NULL);
907 if (err)
908 goto out_cleanup_disk;
909
910 return 0;
911
912 out_cleanup_disk:
913 put_disk(g);
914 out_free_tag:
915 blk_mq_free_tag_set(&port->tag_set);
916 return err;
917 }
918
919 static struct ldc_channel_config vdc_ldc_cfg = {
920 .event = vdc_event,
921 .mtu = 64,
922 .mode = LDC_MODE_UNRELIABLE,
923 };
924
925 static struct vio_driver_ops vdc_vio_ops = {
926 .send_attr = vdc_send_attr,
927 .handle_attr = vdc_handle_attr,
928 .handshake_complete = vdc_handshake_complete,
929 };
930
print_version(void)931 static void print_version(void)
932 {
933 static int version_printed;
934
935 if (version_printed++ == 0)
936 printk(KERN_INFO "%s", version);
937 }
938
939 struct vdc_check_port_data {
940 int dev_no;
941 char *type;
942 };
943
vdc_device_probed(struct device * dev,const void * arg)944 static int vdc_device_probed(struct device *dev, const void *arg)
945 {
946 struct vio_dev *vdev = to_vio_dev(dev);
947 const struct vdc_check_port_data *port_data;
948
949 port_data = (const struct vdc_check_port_data *)arg;
950
951 if ((vdev->dev_no == port_data->dev_no) &&
952 (!(strcmp((char *)&vdev->type, port_data->type))) &&
953 dev_get_drvdata(dev)) {
954 /* This device has already been configured
955 * by vdc_port_probe()
956 */
957 return 1;
958 } else {
959 return 0;
960 }
961 }
962
963 /* Determine whether the VIO device is part of an mpgroup
964 * by locating all the virtual-device-port nodes associated
965 * with the parent virtual-device node for the VIO device
966 * and checking whether any of these nodes are vdc-ports
967 * which have already been configured.
968 *
969 * Returns true if this device is part of an mpgroup and has
970 * already been probed.
971 */
vdc_port_mpgroup_check(struct vio_dev * vdev)972 static bool vdc_port_mpgroup_check(struct vio_dev *vdev)
973 {
974 struct vdc_check_port_data port_data;
975 struct device *dev;
976
977 port_data.dev_no = vdev->dev_no;
978 port_data.type = (char *)&vdev->type;
979
980 dev = device_find_child(vdev->dev.parent, &port_data,
981 vdc_device_probed);
982
983 if (dev) {
984 put_device(dev);
985 return true;
986 }
987
988 return false;
989 }
990
vdc_port_probe(struct vio_dev * vdev,const struct vio_device_id * id)991 static int vdc_port_probe(struct vio_dev *vdev, const struct vio_device_id *id)
992 {
993 struct mdesc_handle *hp;
994 struct vdc_port *port;
995 int err;
996 const u64 *ldc_timeout;
997
998 print_version();
999
1000 hp = mdesc_grab();
1001 if (!hp)
1002 return -ENODEV;
1003
1004 err = -ENODEV;
1005 if ((vdev->dev_no << PARTITION_SHIFT) & ~(u64)MINORMASK) {
1006 printk(KERN_ERR PFX "Port id [%llu] too large.\n",
1007 vdev->dev_no);
1008 goto err_out_release_mdesc;
1009 }
1010
1011 /* Check if this device is part of an mpgroup */
1012 if (vdc_port_mpgroup_check(vdev)) {
1013 printk(KERN_WARNING
1014 "VIO: Ignoring extra vdisk port %s",
1015 dev_name(&vdev->dev));
1016 goto err_out_release_mdesc;
1017 }
1018
1019 port = kzalloc_obj(*port);
1020 if (!port) {
1021 err = -ENOMEM;
1022 goto err_out_release_mdesc;
1023 }
1024
1025 if (vdev->dev_no >= 26)
1026 snprintf(port->disk_name, sizeof(port->disk_name),
1027 VDCBLK_NAME "%c%c",
1028 'a' + ((int)vdev->dev_no / 26) - 1,
1029 'a' + ((int)vdev->dev_no % 26));
1030 else
1031 snprintf(port->disk_name, sizeof(port->disk_name),
1032 VDCBLK_NAME "%c", 'a' + ((int)vdev->dev_no % 26));
1033 port->vdisk_size = -1;
1034
1035 /* Actual wall time may be double due to do_generic_file_read() doing
1036 * a readahead I/O first, and once that fails it will try to read a
1037 * single page.
1038 */
1039 ldc_timeout = mdesc_get_property(hp, vdev->mp, "vdc-timeout", NULL);
1040 port->ldc_timeout = ldc_timeout ? *ldc_timeout : 0;
1041 INIT_DELAYED_WORK(&port->ldc_reset_timer_work, vdc_ldc_reset_timer_work);
1042 INIT_WORK(&port->ldc_reset_work, vdc_ldc_reset_work);
1043
1044 err = vio_driver_init(&port->vio, vdev, VDEV_DISK,
1045 vdc_versions, ARRAY_SIZE(vdc_versions),
1046 &vdc_vio_ops, port->disk_name);
1047 if (err)
1048 goto err_out_free_port;
1049
1050 port->vdisk_block_size = VDC_DEFAULT_BLK_SIZE;
1051 port->max_xfer_size = MAX_XFER_SIZE;
1052 port->ring_cookies = MAX_RING_COOKIES;
1053
1054 err = vio_ldc_alloc(&port->vio, &vdc_ldc_cfg, port);
1055 if (err)
1056 goto err_out_free_port;
1057
1058 err = vdc_alloc_tx_ring(port);
1059 if (err)
1060 goto err_out_free_ldc;
1061
1062 err = probe_disk(port);
1063 if (err)
1064 goto err_out_free_tx_ring;
1065
1066 /* Note that the device driver_data is used to determine
1067 * whether the port has been probed.
1068 */
1069 dev_set_drvdata(&vdev->dev, port);
1070
1071 mdesc_release(hp);
1072
1073 return 0;
1074
1075 err_out_free_tx_ring:
1076 vdc_free_tx_ring(port);
1077
1078 err_out_free_ldc:
1079 vio_ldc_free(&port->vio);
1080
1081 err_out_free_port:
1082 kfree(port);
1083
1084 err_out_release_mdesc:
1085 mdesc_release(hp);
1086 return err;
1087 }
1088
vdc_port_remove(struct vio_dev * vdev)1089 static void vdc_port_remove(struct vio_dev *vdev)
1090 {
1091 struct vdc_port *port = dev_get_drvdata(&vdev->dev);
1092
1093 if (port) {
1094 blk_mq_stop_hw_queues(port->disk->queue);
1095
1096 flush_work(&port->ldc_reset_work);
1097 cancel_delayed_work_sync(&port->ldc_reset_timer_work);
1098 timer_delete_sync(&port->vio.timer);
1099
1100 del_gendisk(port->disk);
1101 put_disk(port->disk);
1102 blk_mq_free_tag_set(&port->tag_set);
1103
1104 vdc_free_tx_ring(port);
1105 vio_ldc_free(&port->vio);
1106
1107 dev_set_drvdata(&vdev->dev, NULL);
1108
1109 kfree(port);
1110 }
1111 }
1112
vdc_requeue_inflight(struct vdc_port * port)1113 static void vdc_requeue_inflight(struct vdc_port *port)
1114 {
1115 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1116 u32 idx;
1117
1118 for (idx = dr->cons; idx != dr->prod; idx = vio_dring_next(dr, idx)) {
1119 struct vio_disk_desc *desc = vio_dring_entry(dr, idx);
1120 struct vdc_req_entry *rqe = &port->rq_arr[idx];
1121 struct request *req;
1122
1123 ldc_unmap(port->vio.lp, desc->cookies, desc->ncookies);
1124 desc->hdr.state = VIO_DESC_FREE;
1125 dr->cons = vio_dring_next(dr, idx);
1126
1127 req = rqe->req;
1128 if (req == NULL) {
1129 vdc_end_special(port, desc);
1130 continue;
1131 }
1132
1133 rqe->req = NULL;
1134 blk_mq_requeue_request(req, false);
1135 }
1136 }
1137
vdc_queue_drain(struct vdc_port * port)1138 static void vdc_queue_drain(struct vdc_port *port)
1139 {
1140 struct request_queue *q = port->disk->queue;
1141 unsigned int memflags;
1142
1143 /*
1144 * Mark the queue as draining, then freeze/quiesce to ensure
1145 * that all existing requests are seen in ->queue_rq() and killed
1146 */
1147 port->drain = 1;
1148 spin_unlock_irq(&port->vio.lock);
1149
1150 memflags = blk_mq_freeze_queue(q);
1151 blk_mq_quiesce_queue(q);
1152
1153 spin_lock_irq(&port->vio.lock);
1154 port->drain = 0;
1155 blk_mq_unquiesce_queue(q);
1156 blk_mq_unfreeze_queue(q, memflags);
1157 }
1158
vdc_ldc_reset_timer_work(struct work_struct * work)1159 static void vdc_ldc_reset_timer_work(struct work_struct *work)
1160 {
1161 struct vdc_port *port;
1162 struct vio_driver_state *vio;
1163
1164 port = container_of(work, struct vdc_port, ldc_reset_timer_work.work);
1165 vio = &port->vio;
1166
1167 spin_lock_irq(&vio->lock);
1168 if (!(port->vio.hs_state & VIO_HS_COMPLETE)) {
1169 pr_warn(PFX "%s ldc down %llu seconds, draining queue\n",
1170 port->disk_name, port->ldc_timeout);
1171 vdc_queue_drain(port);
1172 vdc_blk_queue_start(port);
1173 }
1174 spin_unlock_irq(&vio->lock);
1175 }
1176
vdc_ldc_reset_work(struct work_struct * work)1177 static void vdc_ldc_reset_work(struct work_struct *work)
1178 {
1179 struct vdc_port *port;
1180 struct vio_driver_state *vio;
1181 unsigned long flags;
1182
1183 port = container_of(work, struct vdc_port, ldc_reset_work);
1184 vio = &port->vio;
1185
1186 spin_lock_irqsave(&vio->lock, flags);
1187 vdc_ldc_reset(port);
1188 spin_unlock_irqrestore(&vio->lock, flags);
1189 }
1190
vdc_ldc_reset(struct vdc_port * port)1191 static void vdc_ldc_reset(struct vdc_port *port)
1192 {
1193 int err;
1194
1195 assert_spin_locked(&port->vio.lock);
1196
1197 pr_warn(PFX "%s ldc link reset\n", port->disk_name);
1198 blk_mq_stop_hw_queues(port->disk->queue);
1199 vdc_requeue_inflight(port);
1200 vdc_port_down(port);
1201
1202 err = vio_ldc_alloc(&port->vio, &vdc_ldc_cfg, port);
1203 if (err) {
1204 pr_err(PFX "%s vio_ldc_alloc:%d\n", port->disk_name, err);
1205 return;
1206 }
1207
1208 err = vdc_alloc_tx_ring(port);
1209 if (err) {
1210 pr_err(PFX "%s vio_alloc_tx_ring:%d\n", port->disk_name, err);
1211 goto err_free_ldc;
1212 }
1213
1214 if (port->ldc_timeout)
1215 mod_delayed_work(system_percpu_wq, &port->ldc_reset_timer_work,
1216 round_jiffies(jiffies + HZ * port->ldc_timeout));
1217 mod_timer(&port->vio.timer, round_jiffies(jiffies + HZ));
1218 return;
1219
1220 err_free_ldc:
1221 vio_ldc_free(&port->vio);
1222 }
1223
1224 static const struct vio_device_id vdc_port_match[] = {
1225 {
1226 .type = "vdc-port",
1227 },
1228 {},
1229 };
1230 MODULE_DEVICE_TABLE(vio, vdc_port_match);
1231
1232 static struct vio_driver vdc_port_driver = {
1233 .id_table = vdc_port_match,
1234 .probe = vdc_port_probe,
1235 .remove = vdc_port_remove,
1236 .name = "vdc_port",
1237 };
1238
vdc_init(void)1239 static int __init vdc_init(void)
1240 {
1241 int err;
1242
1243 sunvdc_wq = alloc_workqueue("sunvdc", WQ_PERCPU, 0);
1244 if (!sunvdc_wq)
1245 return -ENOMEM;
1246
1247 err = register_blkdev(0, VDCBLK_NAME);
1248 if (err < 0)
1249 goto out_free_wq;
1250
1251 vdc_major = err;
1252
1253 err = vio_register_driver(&vdc_port_driver);
1254 if (err)
1255 goto out_unregister_blkdev;
1256
1257 return 0;
1258
1259 out_unregister_blkdev:
1260 unregister_blkdev(vdc_major, VDCBLK_NAME);
1261 vdc_major = 0;
1262
1263 out_free_wq:
1264 destroy_workqueue(sunvdc_wq);
1265 return err;
1266 }
1267
vdc_exit(void)1268 static void __exit vdc_exit(void)
1269 {
1270 vio_unregister_driver(&vdc_port_driver);
1271 unregister_blkdev(vdc_major, VDCBLK_NAME);
1272 destroy_workqueue(sunvdc_wq);
1273 }
1274
1275 module_init(vdc_init);
1276 module_exit(vdc_exit);
1277