1
/* sunvdc.c: Sun LDOM Virtual Disk Client.
3
* Copyright (C) 2007, 2008 David S. Miller <davem@davemloft.net>
6
#include <linux/module.h>
7
#include <linux/kernel.h>
8
#include <linux/types.h>
9
#include <linux/blkdev.h>
10
#include <linux/hdreg.h>
11
#include <linux/genhd.h>
12
#include <linux/slab.h>
13
#include <linux/spinlock.h>
14
#include <linux/completion.h>
15
#include <linux/delay.h>
16
#include <linux/init.h>
17
#include <linux/list.h>
18
#include <linux/scatterlist.h>
23
#define DRV_MODULE_NAME "sunvdc"
24
#define PFX DRV_MODULE_NAME ": "
25
#define DRV_MODULE_VERSION "1.0"
26
#define DRV_MODULE_RELDATE "June 25, 2007"
28
static char version[] __devinitdata =
29
DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
30
MODULE_AUTHOR("David S. Miller (davem@davemloft.net)");
31
MODULE_DESCRIPTION("Sun LDOM virtual disk client driver");
32
MODULE_LICENSE("GPL");
33
MODULE_VERSION(DRV_MODULE_VERSION);
35
#define VDC_TX_RING_SIZE 256
37
#define WAITING_FOR_LINK_UP 0x01
38
#define WAITING_FOR_TX_SPACE 0x02
39
#define WAITING_FOR_GEN_CMD 0x04
40
#define WAITING_FOR_ANY -1
42
struct vdc_req_entry {
47
struct vio_driver_state vio;
51
struct vdc_completion *cmp;
55
struct vdc_req_entry rq_arr[VDC_TX_RING_SIZE];
57
unsigned long ring_cookies;
62
/* The server fills these in for us in the disk attribute
71
struct vio_disk_geom geom;
72
struct vio_disk_vtoc label;
75
static inline struct vdc_port *to_vdc_port(struct vio_driver_state *vio)
77
return container_of(vio, struct vdc_port, vio);
80
/* Ordered from largest major to lowest */
81
static struct vio_version vdc_versions[] = {
82
{ .major = 1, .minor = 0 },
85
#define VDCBLK_NAME "vdisk"
87
#define PARTITION_SHIFT 3
89
static inline u32 vdc_tx_dring_avail(struct vio_dring_state *dr)
91
return vio_dring_avail(dr, VDC_TX_RING_SIZE);
94
static int vdc_getgeo(struct block_device *bdev, struct hd_geometry *geo)
96
struct gendisk *disk = bdev->bd_disk;
97
struct vdc_port *port = disk->private_data;
99
geo->heads = (u8) port->geom.num_hd;
100
geo->sectors = (u8) port->geom.num_sec;
101
geo->cylinders = port->geom.num_cyl;
106
static const struct block_device_operations vdc_fops = {
107
.owner = THIS_MODULE,
108
.getgeo = vdc_getgeo,
111
static void vdc_finish(struct vio_driver_state *vio, int err, int waiting_for)
114
(waiting_for == -1 ||
115
vio->cmp->waiting_for == waiting_for)) {
117
complete(&vio->cmp->com);
122
static void vdc_handshake_complete(struct vio_driver_state *vio)
124
vdc_finish(vio, 0, WAITING_FOR_LINK_UP);
127
static int vdc_handle_unknown(struct vdc_port *port, void *arg)
129
struct vio_msg_tag *pkt = arg;
131
printk(KERN_ERR PFX "Received unknown msg [%02x:%02x:%04x:%08x]\n",
132
pkt->type, pkt->stype, pkt->stype_env, pkt->sid);
133
printk(KERN_ERR PFX "Resetting connection.\n");
135
ldc_disconnect(port->vio.lp);
140
static int vdc_send_attr(struct vio_driver_state *vio)
142
struct vdc_port *port = to_vdc_port(vio);
143
struct vio_disk_attr_info pkt;
145
memset(&pkt, 0, sizeof(pkt));
147
pkt.tag.type = VIO_TYPE_CTRL;
148
pkt.tag.stype = VIO_SUBTYPE_INFO;
149
pkt.tag.stype_env = VIO_ATTR_INFO;
150
pkt.tag.sid = vio_send_sid(vio);
152
pkt.xfer_mode = VIO_DRING_MODE;
153
pkt.vdisk_block_size = port->vdisk_block_size;
154
pkt.max_xfer_size = port->max_xfer_size;
156
viodbg(HS, "SEND ATTR xfer_mode[0x%x] blksz[%u] max_xfer[%llu]\n",
157
pkt.xfer_mode, pkt.vdisk_block_size, pkt.max_xfer_size);
159
return vio_ldc_send(&port->vio, &pkt, sizeof(pkt));
162
static int vdc_handle_attr(struct vio_driver_state *vio, void *arg)
164
struct vdc_port *port = to_vdc_port(vio);
165
struct vio_disk_attr_info *pkt = arg;
167
viodbg(HS, "GOT ATTR stype[0x%x] ops[%llx] disk_size[%llu] disk_type[%x] "
168
"xfer_mode[0x%x] blksz[%u] max_xfer[%llu]\n",
169
pkt->tag.stype, pkt->operations,
170
pkt->vdisk_size, pkt->vdisk_type,
171
pkt->xfer_mode, pkt->vdisk_block_size,
174
if (pkt->tag.stype == VIO_SUBTYPE_ACK) {
175
switch (pkt->vdisk_type) {
176
case VD_DISK_TYPE_DISK:
177
case VD_DISK_TYPE_SLICE:
181
printk(KERN_ERR PFX "%s: Bogus vdisk_type 0x%x\n",
182
vio->name, pkt->vdisk_type);
186
if (pkt->vdisk_block_size > port->vdisk_block_size) {
187
printk(KERN_ERR PFX "%s: BLOCK size increased "
190
port->vdisk_block_size, pkt->vdisk_block_size);
194
port->operations = pkt->operations;
195
port->vdisk_size = pkt->vdisk_size;
196
port->vdisk_type = pkt->vdisk_type;
197
if (pkt->max_xfer_size < port->max_xfer_size)
198
port->max_xfer_size = pkt->max_xfer_size;
199
port->vdisk_block_size = pkt->vdisk_block_size;
202
printk(KERN_ERR PFX "%s: Attribute NACK\n", vio->name);
208
static void vdc_end_special(struct vdc_port *port, struct vio_disk_desc *desc)
210
int err = desc->status;
212
vdc_finish(&port->vio, -err, WAITING_FOR_GEN_CMD);
215
static void vdc_end_one(struct vdc_port *port, struct vio_dring_state *dr,
218
struct vio_disk_desc *desc = vio_dring_entry(dr, index);
219
struct vdc_req_entry *rqe = &port->rq_arr[index];
222
if (unlikely(desc->hdr.state != VIO_DESC_DONE))
225
ldc_unmap(port->vio.lp, desc->cookies, desc->ncookies);
226
desc->hdr.state = VIO_DESC_FREE;
227
dr->cons = (index + 1) & (VDC_TX_RING_SIZE - 1);
231
vdc_end_special(port, desc);
237
__blk_end_request(req, (desc->status ? -EIO : 0), desc->size);
239
if (blk_queue_stopped(port->disk->queue))
240
blk_start_queue(port->disk->queue);
243
static int vdc_ack(struct vdc_port *port, void *msgbuf)
245
struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
246
struct vio_dring_data *pkt = msgbuf;
248
if (unlikely(pkt->dring_ident != dr->ident ||
249
pkt->start_idx != pkt->end_idx ||
250
pkt->start_idx >= VDC_TX_RING_SIZE))
253
vdc_end_one(port, dr, pkt->start_idx);
258
static int vdc_nack(struct vdc_port *port, void *msgbuf)
260
/* XXX Implement me XXX */
264
static void vdc_event(void *arg, int event)
266
struct vdc_port *port = arg;
267
struct vio_driver_state *vio = &port->vio;
271
spin_lock_irqsave(&vio->lock, flags);
273
if (unlikely(event == LDC_EVENT_RESET ||
274
event == LDC_EVENT_UP)) {
275
vio_link_state_change(vio, event);
276
spin_unlock_irqrestore(&vio->lock, flags);
280
if (unlikely(event != LDC_EVENT_DATA_READY)) {
281
printk(KERN_WARNING PFX "Unexpected LDC event %d\n", event);
282
spin_unlock_irqrestore(&vio->lock, flags);
289
struct vio_msg_tag tag;
293
err = ldc_read(vio->lp, &msgbuf, sizeof(msgbuf));
294
if (unlikely(err < 0)) {
295
if (err == -ECONNRESET)
301
viodbg(DATA, "TAG [%02x:%02x:%04x:%08x]\n",
304
msgbuf.tag.stype_env,
306
err = vio_validate_sid(vio, &msgbuf.tag);
310
if (likely(msgbuf.tag.type == VIO_TYPE_DATA)) {
311
if (msgbuf.tag.stype == VIO_SUBTYPE_ACK)
312
err = vdc_ack(port, &msgbuf);
313
else if (msgbuf.tag.stype == VIO_SUBTYPE_NACK)
314
err = vdc_nack(port, &msgbuf);
316
err = vdc_handle_unknown(port, &msgbuf);
317
} else if (msgbuf.tag.type == VIO_TYPE_CTRL) {
318
err = vio_control_pkt_engine(vio, &msgbuf);
320
err = vdc_handle_unknown(port, &msgbuf);
326
vdc_finish(&port->vio, err, WAITING_FOR_ANY);
327
spin_unlock_irqrestore(&vio->lock, flags);
330
static int __vdc_tx_trigger(struct vdc_port *port)
332
struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
333
struct vio_dring_data hdr = {
335
.type = VIO_TYPE_DATA,
336
.stype = VIO_SUBTYPE_INFO,
337
.stype_env = VIO_DRING_DATA,
338
.sid = vio_send_sid(&port->vio),
340
.dring_ident = dr->ident,
341
.start_idx = dr->prod,
346
hdr.seq = dr->snd_nxt;
349
err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
355
if ((delay <<= 1) > 128)
357
} while (err == -EAGAIN);
362
static int __send_request(struct request *req)
364
struct vdc_port *port = req->rq_disk->private_data;
365
struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
366
struct scatterlist sg[port->ring_cookies];
367
struct vdc_req_entry *rqe;
368
struct vio_disk_desc *desc;
369
unsigned int map_perm;
374
map_perm = LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_IO;
376
if (rq_data_dir(req) == READ) {
377
map_perm |= LDC_MAP_W;
380
map_perm |= LDC_MAP_R;
384
sg_init_table(sg, port->ring_cookies);
385
nsg = blk_rq_map_sg(req->q, req, sg);
388
for (i = 0; i < nsg; i++)
391
if (unlikely(vdc_tx_dring_avail(dr) < 1)) {
392
blk_stop_queue(port->disk->queue);
397
desc = vio_dring_cur(dr);
399
err = ldc_map_sg(port->vio.lp, sg, nsg,
400
desc->cookies, port->ring_cookies,
403
printk(KERN_ERR PFX "ldc_map_sg() failure, err=%d.\n", err);
407
rqe = &port->rq_arr[dr->prod];
410
desc->hdr.ack = VIO_ACK_ENABLE;
411
desc->req_id = port->req_id;
412
desc->operation = op;
413
if (port->vdisk_type == VD_DISK_TYPE_DISK) {
419
desc->offset = (blk_rq_pos(req) << 9) / port->vdisk_block_size;
421
desc->ncookies = err;
423
/* This has to be a non-SMP write barrier because we are writing
424
* to memory which is shared with the peer LDOM.
427
desc->hdr.state = VIO_DESC_READY;
429
err = __vdc_tx_trigger(port);
431
printk(KERN_ERR PFX "vdc_tx_trigger() failure, err=%d\n", err);
434
dr->prod = (dr->prod + 1) & (VDC_TX_RING_SIZE - 1);
441
static void do_vdc_request(struct request_queue *q)
444
struct request *req = blk_fetch_request(q);
449
if (__send_request(req) < 0)
450
__blk_end_request_all(req, -EIO);
454
static int generic_request(struct vdc_port *port, u8 op, void *buf, int len)
456
struct vio_dring_state *dr;
457
struct vio_completion comp;
458
struct vio_disk_desc *desc;
459
unsigned int map_perm;
464
if (!(((u64)1 << ((u64)op - 1)) & port->operations))
479
op_len = sizeof(u32);
480
map_perm = LDC_MAP_W;
484
op_len = sizeof(u32);
485
map_perm = LDC_MAP_R;
489
op_len = sizeof(struct vio_disk_vtoc);
490
map_perm = LDC_MAP_W;
494
op_len = sizeof(struct vio_disk_vtoc);
495
map_perm = LDC_MAP_R;
498
case VD_OP_GET_DISKGEOM:
499
op_len = sizeof(struct vio_disk_geom);
500
map_perm = LDC_MAP_W;
503
case VD_OP_SET_DISKGEOM:
504
op_len = sizeof(struct vio_disk_geom);
505
map_perm = LDC_MAP_R;
510
map_perm = LDC_MAP_RW;
513
case VD_OP_GET_DEVID:
514
op_len = sizeof(struct vio_disk_devid);
515
map_perm = LDC_MAP_W;
524
map_perm |= LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_IO;
526
op_len = (op_len + 7) & ~7;
527
req_buf = kzalloc(op_len, GFP_KERNEL);
534
if (map_perm & LDC_MAP_R)
535
memcpy(req_buf, buf, len);
537
spin_lock_irqsave(&port->vio.lock, flags);
539
dr = &port->vio.drings[VIO_DRIVER_TX_RING];
541
/* XXX If we want to use this code generically we have to
542
* XXX handle TX ring exhaustion etc.
544
desc = vio_dring_cur(dr);
546
err = ldc_map_single(port->vio.lp, req_buf, op_len,
547
desc->cookies, port->ring_cookies,
550
spin_unlock_irqrestore(&port->vio.lock, flags);
555
init_completion(&comp.com);
556
comp.waiting_for = WAITING_FOR_GEN_CMD;
557
port->vio.cmp = ∁
559
desc->hdr.ack = VIO_ACK_ENABLE;
560
desc->req_id = port->req_id;
561
desc->operation = op;
566
desc->ncookies = err;
568
/* This has to be a non-SMP write barrier because we are writing
569
* to memory which is shared with the peer LDOM.
572
desc->hdr.state = VIO_DESC_READY;
574
err = __vdc_tx_trigger(port);
577
dr->prod = (dr->prod + 1) & (VDC_TX_RING_SIZE - 1);
578
spin_unlock_irqrestore(&port->vio.lock, flags);
580
wait_for_completion(&comp.com);
583
port->vio.cmp = NULL;
584
spin_unlock_irqrestore(&port->vio.lock, flags);
587
if (map_perm & LDC_MAP_W)
588
memcpy(buf, req_buf, len);
595
static int __devinit vdc_alloc_tx_ring(struct vdc_port *port)
597
struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
598
unsigned long len, entry_size;
602
entry_size = sizeof(struct vio_disk_desc) +
603
(sizeof(struct ldc_trans_cookie) * port->ring_cookies);
604
len = (VDC_TX_RING_SIZE * entry_size);
606
ncookies = VIO_MAX_RING_COOKIES;
607
dring = ldc_alloc_exp_dring(port->vio.lp, len,
608
dr->cookies, &ncookies,
613
return PTR_ERR(dring);
616
dr->entry_size = entry_size;
617
dr->num_entries = VDC_TX_RING_SIZE;
618
dr->prod = dr->cons = 0;
619
dr->pending = VDC_TX_RING_SIZE;
620
dr->ncookies = ncookies;
625
static void vdc_free_tx_ring(struct vdc_port *port)
627
struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
630
ldc_free_exp_dring(port->vio.lp, dr->base,
631
(dr->entry_size * dr->num_entries),
632
dr->cookies, dr->ncookies);
641
static int probe_disk(struct vdc_port *port)
643
struct vio_completion comp;
644
struct request_queue *q;
648
init_completion(&comp.com);
650
comp.waiting_for = WAITING_FOR_LINK_UP;
651
port->vio.cmp = ∁
653
vio_port_up(&port->vio);
655
wait_for_completion(&comp.com);
659
err = generic_request(port, VD_OP_GET_VTOC,
660
&port->label, sizeof(port->label));
662
printk(KERN_ERR PFX "VD_OP_GET_VTOC returns error %d\n", err);
666
err = generic_request(port, VD_OP_GET_DISKGEOM,
667
&port->geom, sizeof(port->geom));
669
printk(KERN_ERR PFX "VD_OP_GET_DISKGEOM returns "
674
port->vdisk_size = ((u64)port->geom.num_cyl *
675
(u64)port->geom.num_hd *
676
(u64)port->geom.num_sec);
678
q = blk_init_queue(do_vdc_request, &port->vio.lock);
680
printk(KERN_ERR PFX "%s: Could not allocate queue.\n",
684
g = alloc_disk(1 << PARTITION_SHIFT);
686
printk(KERN_ERR PFX "%s: Could not allocate gendisk.\n",
688
blk_cleanup_queue(q);
694
blk_queue_max_segments(q, port->ring_cookies);
695
blk_queue_max_hw_sectors(q, port->max_xfer_size);
696
g->major = vdc_major;
697
g->first_minor = port->vio.vdev->dev_no << PARTITION_SHIFT;
698
strcpy(g->disk_name, port->disk_name);
702
g->private_data = port;
703
g->driverfs_dev = &port->vio.vdev->dev;
705
set_capacity(g, port->vdisk_size);
707
printk(KERN_INFO PFX "%s: %u sectors (%u MB)\n",
709
port->vdisk_size, (port->vdisk_size >> (20 - 9)));
716
static struct ldc_channel_config vdc_ldc_cfg = {
719
.mode = LDC_MODE_UNRELIABLE,
722
static struct vio_driver_ops vdc_vio_ops = {
723
.send_attr = vdc_send_attr,
724
.handle_attr = vdc_handle_attr,
725
.handshake_complete = vdc_handshake_complete,
728
static void __devinit print_version(void)
730
static int version_printed;
732
if (version_printed++ == 0)
733
printk(KERN_INFO "%s", version);
736
static int __devinit vdc_port_probe(struct vio_dev *vdev,
737
const struct vio_device_id *id)
739
struct mdesc_handle *hp;
740
struct vdc_port *port;
748
if ((vdev->dev_no << PARTITION_SHIFT) & ~(u64)MINORMASK) {
749
printk(KERN_ERR PFX "Port id [%llu] too large.\n",
751
goto err_out_release_mdesc;
754
port = kzalloc(sizeof(*port), GFP_KERNEL);
757
printk(KERN_ERR PFX "Cannot allocate vdc_port.\n");
758
goto err_out_release_mdesc;
761
if (vdev->dev_no >= 26)
762
snprintf(port->disk_name, sizeof(port->disk_name),
764
'a' + ((int)vdev->dev_no / 26) - 1,
765
'a' + ((int)vdev->dev_no % 26));
767
snprintf(port->disk_name, sizeof(port->disk_name),
768
VDCBLK_NAME "%c", 'a' + ((int)vdev->dev_no % 26));
770
err = vio_driver_init(&port->vio, vdev, VDEV_DISK,
771
vdc_versions, ARRAY_SIZE(vdc_versions),
772
&vdc_vio_ops, port->disk_name);
774
goto err_out_free_port;
776
port->vdisk_block_size = 512;
777
port->max_xfer_size = ((128 * 1024) / port->vdisk_block_size);
778
port->ring_cookies = ((port->max_xfer_size *
779
port->vdisk_block_size) / PAGE_SIZE) + 2;
781
err = vio_ldc_alloc(&port->vio, &vdc_ldc_cfg, port);
783
goto err_out_free_port;
785
err = vdc_alloc_tx_ring(port);
787
goto err_out_free_ldc;
789
err = probe_disk(port);
791
goto err_out_free_tx_ring;
793
dev_set_drvdata(&vdev->dev, port);
799
err_out_free_tx_ring:
800
vdc_free_tx_ring(port);
803
vio_ldc_free(&port->vio);
808
err_out_release_mdesc:
813
static int vdc_port_remove(struct vio_dev *vdev)
815
struct vdc_port *port = dev_get_drvdata(&vdev->dev);
818
del_timer_sync(&port->vio.timer);
820
vdc_free_tx_ring(port);
821
vio_ldc_free(&port->vio);
823
dev_set_drvdata(&vdev->dev, NULL);
830
static const struct vio_device_id vdc_port_match[] = {
836
MODULE_DEVICE_TABLE(vio, vdc_port_match);
838
static struct vio_driver vdc_port_driver = {
839
.id_table = vdc_port_match,
840
.probe = vdc_port_probe,
841
.remove = vdc_port_remove,
844
.owner = THIS_MODULE,
848
static int __init vdc_init(void)
852
err = register_blkdev(0, VDCBLK_NAME);
858
err = vio_register_driver(&vdc_port_driver);
860
goto out_unregister_blkdev;
864
out_unregister_blkdev:
865
unregister_blkdev(vdc_major, VDCBLK_NAME);
872
static void __exit vdc_exit(void)
874
vio_unregister_driver(&vdc_port_driver);
875
unregister_blkdev(vdc_major, VDCBLK_NAME);
878
module_init(vdc_init);
879
module_exit(vdc_exit);