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/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>
24 #define DRV_MODULE_NAME "sunvdc"
25 #define PFX DRV_MODULE_NAME ": "
26 #define DRV_MODULE_VERSION "1.1"
27 #define DRV_MODULE_RELDATE "February 13, 2013"
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);
36 #define VDC_TX_RING_SIZE 512
38 #define WAITING_FOR_LINK_UP 0x01
39 #define WAITING_FOR_TX_SPACE 0x02
40 #define WAITING_FOR_GEN_CMD 0x04
41 #define WAITING_FOR_ANY -1
43 struct vdc_req_entry {
48 struct vio_driver_state vio;
52 struct vdc_completion *cmp;
56 struct vdc_req_entry rq_arr[VDC_TX_RING_SIZE];
58 unsigned long ring_cookies;
63 /* The server fills these in for us in the disk attribute
73 struct vio_disk_vtoc label;
76 static inline struct vdc_port *to_vdc_port(struct vio_driver_state *vio)
78 return container_of(vio, struct vdc_port, vio);
81 /* Ordered from largest major to lowest */
82 static struct vio_version vdc_versions[] = {
83 { .major = 1, .minor = 1 },
84 { .major = 1, .minor = 0 },
87 static inline int vdc_version_supported(struct vdc_port *port,
90 return port->vio.ver.major == major && port->vio.ver.minor >= minor;
93 #define VDCBLK_NAME "vdisk"
95 #define PARTITION_SHIFT 3
97 static inline u32 vdc_tx_dring_avail(struct vio_dring_state *dr)
99 return vio_dring_avail(dr, VDC_TX_RING_SIZE);
102 static int vdc_getgeo(struct block_device *bdev, struct hd_geometry *geo)
104 struct gendisk *disk = bdev->bd_disk;
105 sector_t nsect = get_capacity(disk);
106 sector_t cylinders = nsect;
110 sector_div(cylinders, geo->heads * geo->sectors);
111 geo->cylinders = cylinders;
112 if ((sector_t)(geo->cylinders + 1) * geo->heads * geo->sectors < nsect)
113 geo->cylinders = 0xffff;
118 /* Add ioctl/CDROM_GET_CAPABILITY to support cdrom_id in udev
119 * when vdisk_mtype is VD_MEDIA_TYPE_CD or VD_MEDIA_TYPE_DVD.
120 * Needed to be able to install inside an ldom from an iso image.
122 static int vdc_ioctl(struct block_device *bdev, fmode_t mode,
123 unsigned command, unsigned long argument)
126 struct gendisk *disk;
129 case CDROMMULTISESSION:
130 pr_debug(PFX "Multisession CDs not supported\n");
131 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
132 if (put_user(0, (char __user *)(argument + i)))
136 case CDROM_GET_CAPABILITY:
137 disk = bdev->bd_disk;
139 if (bdev->bd_disk && (disk->flags & GENHD_FL_CD))
144 pr_debug(PFX "ioctl %08x not supported\n", command);
149 static const struct block_device_operations vdc_fops = {
150 .owner = THIS_MODULE,
151 .getgeo = vdc_getgeo,
155 static void vdc_finish(struct vio_driver_state *vio, int err, int waiting_for)
158 (waiting_for == -1 ||
159 vio->cmp->waiting_for == waiting_for)) {
161 complete(&vio->cmp->com);
166 static void vdc_handshake_complete(struct vio_driver_state *vio)
168 vdc_finish(vio, 0, WAITING_FOR_LINK_UP);
171 static int vdc_handle_unknown(struct vdc_port *port, void *arg)
173 struct vio_msg_tag *pkt = arg;
175 printk(KERN_ERR PFX "Received unknown msg [%02x:%02x:%04x:%08x]\n",
176 pkt->type, pkt->stype, pkt->stype_env, pkt->sid);
177 printk(KERN_ERR PFX "Resetting connection.\n");
179 ldc_disconnect(port->vio.lp);
184 static int vdc_send_attr(struct vio_driver_state *vio)
186 struct vdc_port *port = to_vdc_port(vio);
187 struct vio_disk_attr_info pkt;
189 memset(&pkt, 0, sizeof(pkt));
191 pkt.tag.type = VIO_TYPE_CTRL;
192 pkt.tag.stype = VIO_SUBTYPE_INFO;
193 pkt.tag.stype_env = VIO_ATTR_INFO;
194 pkt.tag.sid = vio_send_sid(vio);
196 pkt.xfer_mode = VIO_DRING_MODE;
197 pkt.vdisk_block_size = port->vdisk_block_size;
198 pkt.max_xfer_size = port->max_xfer_size;
200 viodbg(HS, "SEND ATTR xfer_mode[0x%x] blksz[%u] max_xfer[%llu]\n",
201 pkt.xfer_mode, pkt.vdisk_block_size, pkt.max_xfer_size);
203 return vio_ldc_send(&port->vio, &pkt, sizeof(pkt));
206 static int vdc_handle_attr(struct vio_driver_state *vio, void *arg)
208 struct vdc_port *port = to_vdc_port(vio);
209 struct vio_disk_attr_info *pkt = arg;
211 viodbg(HS, "GOT ATTR stype[0x%x] ops[%llx] disk_size[%llu] disk_type[%x] "
212 "mtype[0x%x] xfer_mode[0x%x] blksz[%u] max_xfer[%llu]\n",
213 pkt->tag.stype, pkt->operations,
214 pkt->vdisk_size, pkt->vdisk_type, pkt->vdisk_mtype,
215 pkt->xfer_mode, pkt->vdisk_block_size,
218 if (pkt->tag.stype == VIO_SUBTYPE_ACK) {
219 switch (pkt->vdisk_type) {
220 case VD_DISK_TYPE_DISK:
221 case VD_DISK_TYPE_SLICE:
225 printk(KERN_ERR PFX "%s: Bogus vdisk_type 0x%x\n",
226 vio->name, pkt->vdisk_type);
230 if (pkt->vdisk_block_size > port->vdisk_block_size) {
231 printk(KERN_ERR PFX "%s: BLOCK size increased "
234 port->vdisk_block_size, pkt->vdisk_block_size);
238 port->operations = pkt->operations;
239 port->vdisk_type = pkt->vdisk_type;
240 if (vdc_version_supported(port, 1, 1)) {
241 port->vdisk_size = pkt->vdisk_size;
242 port->vdisk_mtype = pkt->vdisk_mtype;
244 if (pkt->max_xfer_size < port->max_xfer_size)
245 port->max_xfer_size = pkt->max_xfer_size;
246 port->vdisk_block_size = pkt->vdisk_block_size;
249 printk(KERN_ERR PFX "%s: Attribute NACK\n", vio->name);
255 static void vdc_end_special(struct vdc_port *port, struct vio_disk_desc *desc)
257 int err = desc->status;
259 vdc_finish(&port->vio, -err, WAITING_FOR_GEN_CMD);
262 static void vdc_end_one(struct vdc_port *port, struct vio_dring_state *dr,
265 struct vio_disk_desc *desc = vio_dring_entry(dr, index);
266 struct vdc_req_entry *rqe = &port->rq_arr[index];
269 if (unlikely(desc->hdr.state != VIO_DESC_DONE))
272 ldc_unmap(port->vio.lp, desc->cookies, desc->ncookies);
273 desc->hdr.state = VIO_DESC_FREE;
274 dr->cons = (index + 1) & (VDC_TX_RING_SIZE - 1);
278 vdc_end_special(port, desc);
284 __blk_end_request(req, (desc->status ? -EIO : 0), desc->size);
286 /* restart blk queue when ring is half emptied */
287 if (blk_queue_stopped(port->disk->queue) &&
288 vdc_tx_dring_avail(dr) * 100 / VDC_TX_RING_SIZE >= 50)
289 blk_start_queue(port->disk->queue);
292 static int vdc_ack(struct vdc_port *port, void *msgbuf)
294 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
295 struct vio_dring_data *pkt = msgbuf;
297 if (unlikely(pkt->dring_ident != dr->ident ||
298 pkt->start_idx != pkt->end_idx ||
299 pkt->start_idx >= VDC_TX_RING_SIZE))
302 vdc_end_one(port, dr, pkt->start_idx);
307 static int vdc_nack(struct vdc_port *port, void *msgbuf)
309 /* XXX Implement me XXX */
313 static void vdc_event(void *arg, int event)
315 struct vdc_port *port = arg;
316 struct vio_driver_state *vio = &port->vio;
320 spin_lock_irqsave(&vio->lock, flags);
322 if (unlikely(event == LDC_EVENT_RESET ||
323 event == LDC_EVENT_UP)) {
324 vio_link_state_change(vio, event);
325 spin_unlock_irqrestore(&vio->lock, flags);
329 if (unlikely(event != LDC_EVENT_DATA_READY)) {
330 printk(KERN_WARNING PFX "Unexpected LDC event %d\n", event);
331 spin_unlock_irqrestore(&vio->lock, flags);
338 struct vio_msg_tag tag;
342 err = ldc_read(vio->lp, &msgbuf, sizeof(msgbuf));
343 if (unlikely(err < 0)) {
344 if (err == -ECONNRESET)
350 viodbg(DATA, "TAG [%02x:%02x:%04x:%08x]\n",
353 msgbuf.tag.stype_env,
355 err = vio_validate_sid(vio, &msgbuf.tag);
359 if (likely(msgbuf.tag.type == VIO_TYPE_DATA)) {
360 if (msgbuf.tag.stype == VIO_SUBTYPE_ACK)
361 err = vdc_ack(port, &msgbuf);
362 else if (msgbuf.tag.stype == VIO_SUBTYPE_NACK)
363 err = vdc_nack(port, &msgbuf);
365 err = vdc_handle_unknown(port, &msgbuf);
366 } else if (msgbuf.tag.type == VIO_TYPE_CTRL) {
367 err = vio_control_pkt_engine(vio, &msgbuf);
369 err = vdc_handle_unknown(port, &msgbuf);
375 vdc_finish(&port->vio, err, WAITING_FOR_ANY);
376 spin_unlock_irqrestore(&vio->lock, flags);
379 static int __vdc_tx_trigger(struct vdc_port *port)
381 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
382 struct vio_dring_data hdr = {
384 .type = VIO_TYPE_DATA,
385 .stype = VIO_SUBTYPE_INFO,
386 .stype_env = VIO_DRING_DATA,
387 .sid = vio_send_sid(&port->vio),
389 .dring_ident = dr->ident,
390 .start_idx = dr->prod,
395 hdr.seq = dr->snd_nxt;
398 err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
404 if ((delay <<= 1) > 128)
406 } while (err == -EAGAIN);
411 static int __send_request(struct request *req)
413 struct vdc_port *port = req->rq_disk->private_data;
414 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
415 struct scatterlist sg[port->ring_cookies];
416 struct vdc_req_entry *rqe;
417 struct vio_disk_desc *desc;
418 unsigned int map_perm;
423 map_perm = LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_IO;
425 if (rq_data_dir(req) == READ) {
426 map_perm |= LDC_MAP_W;
429 map_perm |= LDC_MAP_R;
433 sg_init_table(sg, port->ring_cookies);
434 nsg = blk_rq_map_sg(req->q, req, sg);
437 for (i = 0; i < nsg; i++)
440 desc = vio_dring_cur(dr);
442 err = ldc_map_sg(port->vio.lp, sg, nsg,
443 desc->cookies, port->ring_cookies,
446 printk(KERN_ERR PFX "ldc_map_sg() failure, err=%d.\n", err);
450 rqe = &port->rq_arr[dr->prod];
453 desc->hdr.ack = VIO_ACK_ENABLE;
454 desc->req_id = port->req_id;
455 desc->operation = op;
456 if (port->vdisk_type == VD_DISK_TYPE_DISK) {
462 desc->offset = (blk_rq_pos(req) << 9) / port->vdisk_block_size;
464 desc->ncookies = err;
466 /* This has to be a non-SMP write barrier because we are writing
467 * to memory which is shared with the peer LDOM.
470 desc->hdr.state = VIO_DESC_READY;
472 err = __vdc_tx_trigger(port);
474 printk(KERN_ERR PFX "vdc_tx_trigger() failure, err=%d\n", err);
477 dr->prod = (dr->prod + 1) & (VDC_TX_RING_SIZE - 1);
483 static void do_vdc_request(struct request_queue *rq)
487 while ((req = blk_peek_request(rq)) != NULL) {
488 struct vdc_port *port;
489 struct vio_dring_state *dr;
491 port = req->rq_disk->private_data;
492 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
493 if (unlikely(vdc_tx_dring_avail(dr) < 1))
496 blk_start_request(req);
498 if (__send_request(req) < 0) {
499 blk_requeue_request(rq, req);
501 /* Avoid pointless unplugs. */
508 static int generic_request(struct vdc_port *port, u8 op, void *buf, int len)
510 struct vio_dring_state *dr;
511 struct vio_completion comp;
512 struct vio_disk_desc *desc;
513 unsigned int map_perm;
518 if (!(((u64)1 << (u64)op) & port->operations))
533 op_len = sizeof(u32);
534 map_perm = LDC_MAP_W;
538 op_len = sizeof(u32);
539 map_perm = LDC_MAP_R;
543 op_len = sizeof(struct vio_disk_vtoc);
544 map_perm = LDC_MAP_W;
548 op_len = sizeof(struct vio_disk_vtoc);
549 map_perm = LDC_MAP_R;
552 case VD_OP_GET_DISKGEOM:
553 op_len = sizeof(struct vio_disk_geom);
554 map_perm = LDC_MAP_W;
557 case VD_OP_SET_DISKGEOM:
558 op_len = sizeof(struct vio_disk_geom);
559 map_perm = LDC_MAP_R;
564 map_perm = LDC_MAP_RW;
567 case VD_OP_GET_DEVID:
568 op_len = sizeof(struct vio_disk_devid);
569 map_perm = LDC_MAP_W;
578 map_perm |= LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_IO;
580 op_len = (op_len + 7) & ~7;
581 req_buf = kzalloc(op_len, GFP_KERNEL);
588 if (map_perm & LDC_MAP_R)
589 memcpy(req_buf, buf, len);
591 spin_lock_irqsave(&port->vio.lock, flags);
593 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
595 /* XXX If we want to use this code generically we have to
596 * XXX handle TX ring exhaustion etc.
598 desc = vio_dring_cur(dr);
600 err = ldc_map_single(port->vio.lp, req_buf, op_len,
601 desc->cookies, port->ring_cookies,
604 spin_unlock_irqrestore(&port->vio.lock, flags);
609 init_completion(&comp.com);
610 comp.waiting_for = WAITING_FOR_GEN_CMD;
611 port->vio.cmp = ∁
613 desc->hdr.ack = VIO_ACK_ENABLE;
614 desc->req_id = port->req_id;
615 desc->operation = op;
620 desc->ncookies = err;
622 /* This has to be a non-SMP write barrier because we are writing
623 * to memory which is shared with the peer LDOM.
626 desc->hdr.state = VIO_DESC_READY;
628 err = __vdc_tx_trigger(port);
631 dr->prod = (dr->prod + 1) & (VDC_TX_RING_SIZE - 1);
632 spin_unlock_irqrestore(&port->vio.lock, flags);
634 wait_for_completion(&comp.com);
637 port->vio.cmp = NULL;
638 spin_unlock_irqrestore(&port->vio.lock, flags);
641 if (map_perm & LDC_MAP_W)
642 memcpy(buf, req_buf, len);
649 static int vdc_alloc_tx_ring(struct vdc_port *port)
651 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
652 unsigned long len, entry_size;
656 entry_size = sizeof(struct vio_disk_desc) +
657 (sizeof(struct ldc_trans_cookie) * port->ring_cookies);
658 len = (VDC_TX_RING_SIZE * entry_size);
660 ncookies = VIO_MAX_RING_COOKIES;
661 dring = ldc_alloc_exp_dring(port->vio.lp, len,
662 dr->cookies, &ncookies,
667 return PTR_ERR(dring);
670 dr->entry_size = entry_size;
671 dr->num_entries = VDC_TX_RING_SIZE;
672 dr->prod = dr->cons = 0;
673 dr->pending = VDC_TX_RING_SIZE;
674 dr->ncookies = ncookies;
679 static void vdc_free_tx_ring(struct vdc_port *port)
681 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
684 ldc_free_exp_dring(port->vio.lp, dr->base,
685 (dr->entry_size * dr->num_entries),
686 dr->cookies, dr->ncookies);
695 static int probe_disk(struct vdc_port *port)
697 struct vio_completion comp;
698 struct request_queue *q;
702 init_completion(&comp.com);
704 comp.waiting_for = WAITING_FOR_LINK_UP;
705 port->vio.cmp = ∁
707 vio_port_up(&port->vio);
709 wait_for_completion(&comp.com);
713 err = generic_request(port, VD_OP_GET_VTOC,
714 &port->label, sizeof(port->label));
716 printk(KERN_ERR PFX "VD_OP_GET_VTOC returns error %d\n", err);
720 if (vdc_version_supported(port, 1, 1)) {
721 /* vdisk_size should be set during the handshake, if it wasn't
722 * then the underlying disk is reserved by another system
724 if (port->vdisk_size == -1)
727 struct vio_disk_geom geom;
729 err = generic_request(port, VD_OP_GET_DISKGEOM,
730 &geom, sizeof(geom));
732 printk(KERN_ERR PFX "VD_OP_GET_DISKGEOM returns "
736 port->vdisk_size = ((u64)geom.num_cyl *
741 q = blk_init_queue(do_vdc_request, &port->vio.lock);
743 printk(KERN_ERR PFX "%s: Could not allocate queue.\n",
747 g = alloc_disk(1 << PARTITION_SHIFT);
749 printk(KERN_ERR PFX "%s: Could not allocate gendisk.\n",
751 blk_cleanup_queue(q);
757 /* Each segment in a request is up to an aligned page in size. */
758 blk_queue_segment_boundary(q, PAGE_SIZE - 1);
759 blk_queue_max_segment_size(q, PAGE_SIZE);
761 blk_queue_max_segments(q, port->ring_cookies);
762 blk_queue_max_hw_sectors(q, port->max_xfer_size);
763 g->major = vdc_major;
764 g->first_minor = port->vio.vdev->dev_no << PARTITION_SHIFT;
765 strcpy(g->disk_name, port->disk_name);
769 g->private_data = port;
770 g->driverfs_dev = &port->vio.vdev->dev;
772 set_capacity(g, port->vdisk_size);
774 if (vdc_version_supported(port, 1, 1)) {
775 switch (port->vdisk_mtype) {
776 case VD_MEDIA_TYPE_CD:
777 pr_info(PFX "Virtual CDROM %s\n", port->disk_name);
778 g->flags |= GENHD_FL_CD;
779 g->flags |= GENHD_FL_REMOVABLE;
783 case VD_MEDIA_TYPE_DVD:
784 pr_info(PFX "Virtual DVD %s\n", port->disk_name);
785 g->flags |= GENHD_FL_CD;
786 g->flags |= GENHD_FL_REMOVABLE;
790 case VD_MEDIA_TYPE_FIXED:
791 pr_info(PFX "Virtual Hard disk %s\n", port->disk_name);
796 pr_info(PFX "%s: %u sectors (%u MB) protocol %d.%d\n",
798 port->vdisk_size, (port->vdisk_size >> (20 - 9)),
799 port->vio.ver.major, port->vio.ver.minor);
806 static struct ldc_channel_config vdc_ldc_cfg = {
809 .mode = LDC_MODE_UNRELIABLE,
812 static struct vio_driver_ops vdc_vio_ops = {
813 .send_attr = vdc_send_attr,
814 .handle_attr = vdc_handle_attr,
815 .handshake_complete = vdc_handshake_complete,
818 static void print_version(void)
820 static int version_printed;
822 if (version_printed++ == 0)
823 printk(KERN_INFO "%s", version);
826 static int vdc_port_probe(struct vio_dev *vdev, const struct vio_device_id *id)
828 struct mdesc_handle *hp;
829 struct vdc_port *port;
837 if ((vdev->dev_no << PARTITION_SHIFT) & ~(u64)MINORMASK) {
838 printk(KERN_ERR PFX "Port id [%llu] too large.\n",
840 goto err_out_release_mdesc;
843 port = kzalloc(sizeof(*port), GFP_KERNEL);
846 printk(KERN_ERR PFX "Cannot allocate vdc_port.\n");
847 goto err_out_release_mdesc;
850 if (vdev->dev_no >= 26)
851 snprintf(port->disk_name, sizeof(port->disk_name),
853 'a' + ((int)vdev->dev_no / 26) - 1,
854 'a' + ((int)vdev->dev_no % 26));
856 snprintf(port->disk_name, sizeof(port->disk_name),
857 VDCBLK_NAME "%c", 'a' + ((int)vdev->dev_no % 26));
858 port->vdisk_size = -1;
860 err = vio_driver_init(&port->vio, vdev, VDEV_DISK,
861 vdc_versions, ARRAY_SIZE(vdc_versions),
862 &vdc_vio_ops, port->disk_name);
864 goto err_out_free_port;
866 port->vdisk_block_size = 512;
867 port->max_xfer_size = ((128 * 1024) / port->vdisk_block_size);
868 port->ring_cookies = ((port->max_xfer_size *
869 port->vdisk_block_size) / PAGE_SIZE) + 2;
871 err = vio_ldc_alloc(&port->vio, &vdc_ldc_cfg, port);
873 goto err_out_free_port;
875 err = vdc_alloc_tx_ring(port);
877 goto err_out_free_ldc;
879 err = probe_disk(port);
881 goto err_out_free_tx_ring;
883 dev_set_drvdata(&vdev->dev, port);
889 err_out_free_tx_ring:
890 vdc_free_tx_ring(port);
893 vio_ldc_free(&port->vio);
898 err_out_release_mdesc:
903 static int vdc_port_remove(struct vio_dev *vdev)
905 struct vdc_port *port = dev_get_drvdata(&vdev->dev);
908 del_timer_sync(&port->vio.timer);
910 vdc_free_tx_ring(port);
911 vio_ldc_free(&port->vio);
913 dev_set_drvdata(&vdev->dev, NULL);
920 static const struct vio_device_id vdc_port_match[] = {
926 MODULE_DEVICE_TABLE(vio, vdc_port_match);
928 static struct vio_driver vdc_port_driver = {
929 .id_table = vdc_port_match,
930 .probe = vdc_port_probe,
931 .remove = vdc_port_remove,
935 static int __init vdc_init(void)
939 err = register_blkdev(0, VDCBLK_NAME);
945 err = vio_register_driver(&vdc_port_driver);
947 goto out_unregister_blkdev;
951 out_unregister_blkdev:
952 unregister_blkdev(vdc_major, VDCBLK_NAME);
959 static void __exit vdc_exit(void)
961 vio_unregister_driver(&vdc_port_driver);
962 unregister_blkdev(vdc_major, VDCBLK_NAME);
965 module_init(vdc_init);
966 module_exit(vdc_exit);