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
74 static inline struct vdc_port *to_vdc_port(struct vio_driver_state *vio)
76 return container_of(vio, struct vdc_port, vio);
79 /* Ordered from largest major to lowest */
80 static struct vio_version vdc_versions[] = {
81 { .major = 1, .minor = 1 },
82 { .major = 1, .minor = 0 },
85 static inline int vdc_version_supported(struct vdc_port *port,
88 return port->vio.ver.major == major && port->vio.ver.minor >= minor;
91 #define VDCBLK_NAME "vdisk"
93 #define PARTITION_SHIFT 3
95 static inline u32 vdc_tx_dring_avail(struct vio_dring_state *dr)
97 return vio_dring_avail(dr, VDC_TX_RING_SIZE);
100 static int vdc_getgeo(struct block_device *bdev, struct hd_geometry *geo)
102 struct gendisk *disk = bdev->bd_disk;
103 sector_t nsect = get_capacity(disk);
104 sector_t cylinders = nsect;
108 sector_div(cylinders, geo->heads * geo->sectors);
109 geo->cylinders = cylinders;
110 if ((sector_t)(geo->cylinders + 1) * geo->heads * geo->sectors < nsect)
111 geo->cylinders = 0xffff;
116 /* Add ioctl/CDROM_GET_CAPABILITY to support cdrom_id in udev
117 * when vdisk_mtype is VD_MEDIA_TYPE_CD or VD_MEDIA_TYPE_DVD.
118 * Needed to be able to install inside an ldom from an iso image.
120 static int vdc_ioctl(struct block_device *bdev, fmode_t mode,
121 unsigned command, unsigned long argument)
124 struct gendisk *disk;
127 case CDROMMULTISESSION:
128 pr_debug(PFX "Multisession CDs not supported\n");
129 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
130 if (put_user(0, (char __user *)(argument + i)))
134 case CDROM_GET_CAPABILITY:
135 disk = bdev->bd_disk;
137 if (bdev->bd_disk && (disk->flags & GENHD_FL_CD))
142 pr_debug(PFX "ioctl %08x not supported\n", command);
147 static const struct block_device_operations vdc_fops = {
148 .owner = THIS_MODULE,
149 .getgeo = vdc_getgeo,
153 static void vdc_finish(struct vio_driver_state *vio, int err, int waiting_for)
156 (waiting_for == -1 ||
157 vio->cmp->waiting_for == waiting_for)) {
159 complete(&vio->cmp->com);
164 static void vdc_handshake_complete(struct vio_driver_state *vio)
166 vdc_finish(vio, 0, WAITING_FOR_LINK_UP);
169 static int vdc_handle_unknown(struct vdc_port *port, void *arg)
171 struct vio_msg_tag *pkt = arg;
173 printk(KERN_ERR PFX "Received unknown msg [%02x:%02x:%04x:%08x]\n",
174 pkt->type, pkt->stype, pkt->stype_env, pkt->sid);
175 printk(KERN_ERR PFX "Resetting connection.\n");
177 ldc_disconnect(port->vio.lp);
182 static int vdc_send_attr(struct vio_driver_state *vio)
184 struct vdc_port *port = to_vdc_port(vio);
185 struct vio_disk_attr_info pkt;
187 memset(&pkt, 0, sizeof(pkt));
189 pkt.tag.type = VIO_TYPE_CTRL;
190 pkt.tag.stype = VIO_SUBTYPE_INFO;
191 pkt.tag.stype_env = VIO_ATTR_INFO;
192 pkt.tag.sid = vio_send_sid(vio);
194 pkt.xfer_mode = VIO_DRING_MODE;
195 pkt.vdisk_block_size = port->vdisk_block_size;
196 pkt.max_xfer_size = port->max_xfer_size;
198 viodbg(HS, "SEND ATTR xfer_mode[0x%x] blksz[%u] max_xfer[%llu]\n",
199 pkt.xfer_mode, pkt.vdisk_block_size, pkt.max_xfer_size);
201 return vio_ldc_send(&port->vio, &pkt, sizeof(pkt));
204 static int vdc_handle_attr(struct vio_driver_state *vio, void *arg)
206 struct vdc_port *port = to_vdc_port(vio);
207 struct vio_disk_attr_info *pkt = arg;
209 viodbg(HS, "GOT ATTR stype[0x%x] ops[%llx] disk_size[%llu] disk_type[%x] "
210 "mtype[0x%x] xfer_mode[0x%x] blksz[%u] max_xfer[%llu]\n",
211 pkt->tag.stype, pkt->operations,
212 pkt->vdisk_size, pkt->vdisk_type, pkt->vdisk_mtype,
213 pkt->xfer_mode, pkt->vdisk_block_size,
216 if (pkt->tag.stype == VIO_SUBTYPE_ACK) {
217 switch (pkt->vdisk_type) {
218 case VD_DISK_TYPE_DISK:
219 case VD_DISK_TYPE_SLICE:
223 printk(KERN_ERR PFX "%s: Bogus vdisk_type 0x%x\n",
224 vio->name, pkt->vdisk_type);
228 if (pkt->vdisk_block_size > port->vdisk_block_size) {
229 printk(KERN_ERR PFX "%s: BLOCK size increased "
232 port->vdisk_block_size, pkt->vdisk_block_size);
236 port->operations = pkt->operations;
237 port->vdisk_type = pkt->vdisk_type;
238 if (vdc_version_supported(port, 1, 1)) {
239 port->vdisk_size = pkt->vdisk_size;
240 port->vdisk_mtype = pkt->vdisk_mtype;
242 if (pkt->max_xfer_size < port->max_xfer_size)
243 port->max_xfer_size = pkt->max_xfer_size;
244 port->vdisk_block_size = pkt->vdisk_block_size;
247 printk(KERN_ERR PFX "%s: Attribute NACK\n", vio->name);
253 static void vdc_end_special(struct vdc_port *port, struct vio_disk_desc *desc)
255 int err = desc->status;
257 vdc_finish(&port->vio, -err, WAITING_FOR_GEN_CMD);
260 static void vdc_end_one(struct vdc_port *port, struct vio_dring_state *dr,
263 struct vio_disk_desc *desc = vio_dring_entry(dr, index);
264 struct vdc_req_entry *rqe = &port->rq_arr[index];
267 if (unlikely(desc->hdr.state != VIO_DESC_DONE))
270 ldc_unmap(port->vio.lp, desc->cookies, desc->ncookies);
271 desc->hdr.state = VIO_DESC_FREE;
272 dr->cons = (index + 1) & (VDC_TX_RING_SIZE - 1);
276 vdc_end_special(port, desc);
282 __blk_end_request(req, (desc->status ? -EIO : 0), desc->size);
284 /* restart blk queue when ring is half emptied */
285 if (blk_queue_stopped(port->disk->queue) &&
286 vdc_tx_dring_avail(dr) * 100 / VDC_TX_RING_SIZE >= 50)
287 blk_start_queue(port->disk->queue);
290 static int vdc_ack(struct vdc_port *port, void *msgbuf)
292 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
293 struct vio_dring_data *pkt = msgbuf;
295 if (unlikely(pkt->dring_ident != dr->ident ||
296 pkt->start_idx != pkt->end_idx ||
297 pkt->start_idx >= VDC_TX_RING_SIZE))
300 vdc_end_one(port, dr, pkt->start_idx);
305 static int vdc_nack(struct vdc_port *port, void *msgbuf)
307 /* XXX Implement me XXX */
311 static void vdc_event(void *arg, int event)
313 struct vdc_port *port = arg;
314 struct vio_driver_state *vio = &port->vio;
318 spin_lock_irqsave(&vio->lock, flags);
320 if (unlikely(event == LDC_EVENT_RESET ||
321 event == LDC_EVENT_UP)) {
322 vio_link_state_change(vio, event);
323 spin_unlock_irqrestore(&vio->lock, flags);
327 if (unlikely(event != LDC_EVENT_DATA_READY)) {
328 printk(KERN_WARNING PFX "Unexpected LDC event %d\n", event);
329 spin_unlock_irqrestore(&vio->lock, flags);
336 struct vio_msg_tag tag;
340 err = ldc_read(vio->lp, &msgbuf, sizeof(msgbuf));
341 if (unlikely(err < 0)) {
342 if (err == -ECONNRESET)
348 viodbg(DATA, "TAG [%02x:%02x:%04x:%08x]\n",
351 msgbuf.tag.stype_env,
353 err = vio_validate_sid(vio, &msgbuf.tag);
357 if (likely(msgbuf.tag.type == VIO_TYPE_DATA)) {
358 if (msgbuf.tag.stype == VIO_SUBTYPE_ACK)
359 err = vdc_ack(port, &msgbuf);
360 else if (msgbuf.tag.stype == VIO_SUBTYPE_NACK)
361 err = vdc_nack(port, &msgbuf);
363 err = vdc_handle_unknown(port, &msgbuf);
364 } else if (msgbuf.tag.type == VIO_TYPE_CTRL) {
365 err = vio_control_pkt_engine(vio, &msgbuf);
367 err = vdc_handle_unknown(port, &msgbuf);
373 vdc_finish(&port->vio, err, WAITING_FOR_ANY);
374 spin_unlock_irqrestore(&vio->lock, flags);
377 static int __vdc_tx_trigger(struct vdc_port *port)
379 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
380 struct vio_dring_data hdr = {
382 .type = VIO_TYPE_DATA,
383 .stype = VIO_SUBTYPE_INFO,
384 .stype_env = VIO_DRING_DATA,
385 .sid = vio_send_sid(&port->vio),
387 .dring_ident = dr->ident,
388 .start_idx = dr->prod,
393 hdr.seq = dr->snd_nxt;
396 err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
402 if ((delay <<= 1) > 128)
404 } while (err == -EAGAIN);
409 static int __send_request(struct request *req)
411 struct vdc_port *port = req->rq_disk->private_data;
412 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
413 struct scatterlist sg[port->ring_cookies];
414 struct vdc_req_entry *rqe;
415 struct vio_disk_desc *desc;
416 unsigned int map_perm;
421 map_perm = LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_IO;
423 if (rq_data_dir(req) == READ) {
424 map_perm |= LDC_MAP_W;
427 map_perm |= LDC_MAP_R;
431 sg_init_table(sg, port->ring_cookies);
432 nsg = blk_rq_map_sg(req->q, req, sg);
435 for (i = 0; i < nsg; i++)
438 desc = vio_dring_cur(dr);
440 err = ldc_map_sg(port->vio.lp, sg, nsg,
441 desc->cookies, port->ring_cookies,
444 printk(KERN_ERR PFX "ldc_map_sg() failure, err=%d.\n", err);
448 rqe = &port->rq_arr[dr->prod];
451 desc->hdr.ack = VIO_ACK_ENABLE;
452 desc->req_id = port->req_id;
453 desc->operation = op;
454 if (port->vdisk_type == VD_DISK_TYPE_DISK) {
460 desc->offset = (blk_rq_pos(req) << 9) / port->vdisk_block_size;
462 desc->ncookies = err;
464 /* This has to be a non-SMP write barrier because we are writing
465 * to memory which is shared with the peer LDOM.
468 desc->hdr.state = VIO_DESC_READY;
470 err = __vdc_tx_trigger(port);
472 printk(KERN_ERR PFX "vdc_tx_trigger() failure, err=%d\n", err);
475 dr->prod = (dr->prod + 1) & (VDC_TX_RING_SIZE - 1);
481 static void do_vdc_request(struct request_queue *rq)
485 while ((req = blk_peek_request(rq)) != NULL) {
486 struct vdc_port *port;
487 struct vio_dring_state *dr;
489 port = req->rq_disk->private_data;
490 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
491 if (unlikely(vdc_tx_dring_avail(dr) < 1))
494 blk_start_request(req);
496 if (__send_request(req) < 0) {
497 blk_requeue_request(rq, req);
499 /* Avoid pointless unplugs. */
506 static int generic_request(struct vdc_port *port, u8 op, void *buf, int len)
508 struct vio_dring_state *dr;
509 struct vio_completion comp;
510 struct vio_disk_desc *desc;
511 unsigned int map_perm;
516 if (!(((u64)1 << (u64)op) & port->operations))
531 op_len = sizeof(u32);
532 map_perm = LDC_MAP_W;
536 op_len = sizeof(u32);
537 map_perm = LDC_MAP_R;
541 op_len = sizeof(struct vio_disk_vtoc);
542 map_perm = LDC_MAP_W;
546 op_len = sizeof(struct vio_disk_vtoc);
547 map_perm = LDC_MAP_R;
550 case VD_OP_GET_DISKGEOM:
551 op_len = sizeof(struct vio_disk_geom);
552 map_perm = LDC_MAP_W;
555 case VD_OP_SET_DISKGEOM:
556 op_len = sizeof(struct vio_disk_geom);
557 map_perm = LDC_MAP_R;
562 map_perm = LDC_MAP_RW;
565 case VD_OP_GET_DEVID:
566 op_len = sizeof(struct vio_disk_devid);
567 map_perm = LDC_MAP_W;
576 map_perm |= LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_IO;
578 op_len = (op_len + 7) & ~7;
579 req_buf = kzalloc(op_len, GFP_KERNEL);
586 if (map_perm & LDC_MAP_R)
587 memcpy(req_buf, buf, len);
589 spin_lock_irqsave(&port->vio.lock, flags);
591 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
593 /* XXX If we want to use this code generically we have to
594 * XXX handle TX ring exhaustion etc.
596 desc = vio_dring_cur(dr);
598 err = ldc_map_single(port->vio.lp, req_buf, op_len,
599 desc->cookies, port->ring_cookies,
602 spin_unlock_irqrestore(&port->vio.lock, flags);
607 init_completion(&comp.com);
608 comp.waiting_for = WAITING_FOR_GEN_CMD;
609 port->vio.cmp = ∁
611 desc->hdr.ack = VIO_ACK_ENABLE;
612 desc->req_id = port->req_id;
613 desc->operation = op;
618 desc->ncookies = err;
620 /* This has to be a non-SMP write barrier because we are writing
621 * to memory which is shared with the peer LDOM.
624 desc->hdr.state = VIO_DESC_READY;
626 err = __vdc_tx_trigger(port);
629 dr->prod = (dr->prod + 1) & (VDC_TX_RING_SIZE - 1);
630 spin_unlock_irqrestore(&port->vio.lock, flags);
632 wait_for_completion(&comp.com);
635 port->vio.cmp = NULL;
636 spin_unlock_irqrestore(&port->vio.lock, flags);
639 if (map_perm & LDC_MAP_W)
640 memcpy(buf, req_buf, len);
647 static int vdc_alloc_tx_ring(struct vdc_port *port)
649 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
650 unsigned long len, entry_size;
654 entry_size = sizeof(struct vio_disk_desc) +
655 (sizeof(struct ldc_trans_cookie) * port->ring_cookies);
656 len = (VDC_TX_RING_SIZE * entry_size);
658 ncookies = VIO_MAX_RING_COOKIES;
659 dring = ldc_alloc_exp_dring(port->vio.lp, len,
660 dr->cookies, &ncookies,
665 return PTR_ERR(dring);
668 dr->entry_size = entry_size;
669 dr->num_entries = VDC_TX_RING_SIZE;
670 dr->prod = dr->cons = 0;
671 dr->pending = VDC_TX_RING_SIZE;
672 dr->ncookies = ncookies;
677 static void vdc_free_tx_ring(struct vdc_port *port)
679 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
682 ldc_free_exp_dring(port->vio.lp, dr->base,
683 (dr->entry_size * dr->num_entries),
684 dr->cookies, dr->ncookies);
693 static int probe_disk(struct vdc_port *port)
695 struct vio_completion comp;
696 struct request_queue *q;
700 init_completion(&comp.com);
702 comp.waiting_for = WAITING_FOR_LINK_UP;
703 port->vio.cmp = ∁
705 vio_port_up(&port->vio);
707 wait_for_completion(&comp.com);
711 if (vdc_version_supported(port, 1, 1)) {
712 /* vdisk_size should be set during the handshake, if it wasn't
713 * then the underlying disk is reserved by another system
715 if (port->vdisk_size == -1)
718 struct vio_disk_geom geom;
720 err = generic_request(port, VD_OP_GET_DISKGEOM,
721 &geom, sizeof(geom));
723 printk(KERN_ERR PFX "VD_OP_GET_DISKGEOM returns "
727 port->vdisk_size = ((u64)geom.num_cyl *
732 q = blk_init_queue(do_vdc_request, &port->vio.lock);
734 printk(KERN_ERR PFX "%s: Could not allocate queue.\n",
738 g = alloc_disk(1 << PARTITION_SHIFT);
740 printk(KERN_ERR PFX "%s: Could not allocate gendisk.\n",
742 blk_cleanup_queue(q);
748 /* Each segment in a request is up to an aligned page in size. */
749 blk_queue_segment_boundary(q, PAGE_SIZE - 1);
750 blk_queue_max_segment_size(q, PAGE_SIZE);
752 blk_queue_max_segments(q, port->ring_cookies);
753 blk_queue_max_hw_sectors(q, port->max_xfer_size);
754 g->major = vdc_major;
755 g->first_minor = port->vio.vdev->dev_no << PARTITION_SHIFT;
756 strcpy(g->disk_name, port->disk_name);
760 g->private_data = port;
761 g->driverfs_dev = &port->vio.vdev->dev;
763 set_capacity(g, port->vdisk_size);
765 if (vdc_version_supported(port, 1, 1)) {
766 switch (port->vdisk_mtype) {
767 case VD_MEDIA_TYPE_CD:
768 pr_info(PFX "Virtual CDROM %s\n", port->disk_name);
769 g->flags |= GENHD_FL_CD;
770 g->flags |= GENHD_FL_REMOVABLE;
774 case VD_MEDIA_TYPE_DVD:
775 pr_info(PFX "Virtual DVD %s\n", port->disk_name);
776 g->flags |= GENHD_FL_CD;
777 g->flags |= GENHD_FL_REMOVABLE;
781 case VD_MEDIA_TYPE_FIXED:
782 pr_info(PFX "Virtual Hard disk %s\n", port->disk_name);
787 pr_info(PFX "%s: %u sectors (%u MB) protocol %d.%d\n",
789 port->vdisk_size, (port->vdisk_size >> (20 - 9)),
790 port->vio.ver.major, port->vio.ver.minor);
797 static struct ldc_channel_config vdc_ldc_cfg = {
800 .mode = LDC_MODE_UNRELIABLE,
803 static struct vio_driver_ops vdc_vio_ops = {
804 .send_attr = vdc_send_attr,
805 .handle_attr = vdc_handle_attr,
806 .handshake_complete = vdc_handshake_complete,
809 static void print_version(void)
811 static int version_printed;
813 if (version_printed++ == 0)
814 printk(KERN_INFO "%s", version);
817 static int vdc_port_probe(struct vio_dev *vdev, const struct vio_device_id *id)
819 struct mdesc_handle *hp;
820 struct vdc_port *port;
828 if ((vdev->dev_no << PARTITION_SHIFT) & ~(u64)MINORMASK) {
829 printk(KERN_ERR PFX "Port id [%llu] too large.\n",
831 goto err_out_release_mdesc;
834 port = kzalloc(sizeof(*port), GFP_KERNEL);
837 printk(KERN_ERR PFX "Cannot allocate vdc_port.\n");
838 goto err_out_release_mdesc;
841 if (vdev->dev_no >= 26)
842 snprintf(port->disk_name, sizeof(port->disk_name),
844 'a' + ((int)vdev->dev_no / 26) - 1,
845 'a' + ((int)vdev->dev_no % 26));
847 snprintf(port->disk_name, sizeof(port->disk_name),
848 VDCBLK_NAME "%c", 'a' + ((int)vdev->dev_no % 26));
849 port->vdisk_size = -1;
851 err = vio_driver_init(&port->vio, vdev, VDEV_DISK,
852 vdc_versions, ARRAY_SIZE(vdc_versions),
853 &vdc_vio_ops, port->disk_name);
855 goto err_out_free_port;
857 port->vdisk_block_size = 512;
858 port->max_xfer_size = ((128 * 1024) / port->vdisk_block_size);
859 port->ring_cookies = ((port->max_xfer_size *
860 port->vdisk_block_size) / PAGE_SIZE) + 2;
862 err = vio_ldc_alloc(&port->vio, &vdc_ldc_cfg, port);
864 goto err_out_free_port;
866 err = vdc_alloc_tx_ring(port);
868 goto err_out_free_ldc;
870 err = probe_disk(port);
872 goto err_out_free_tx_ring;
874 dev_set_drvdata(&vdev->dev, port);
880 err_out_free_tx_ring:
881 vdc_free_tx_ring(port);
884 vio_ldc_free(&port->vio);
889 err_out_release_mdesc:
894 static int vdc_port_remove(struct vio_dev *vdev)
896 struct vdc_port *port = dev_get_drvdata(&vdev->dev);
899 del_timer_sync(&port->vio.timer);
901 vdc_free_tx_ring(port);
902 vio_ldc_free(&port->vio);
904 dev_set_drvdata(&vdev->dev, NULL);
911 static const struct vio_device_id vdc_port_match[] = {
917 MODULE_DEVICE_TABLE(vio, vdc_port_match);
919 static struct vio_driver vdc_port_driver = {
920 .id_table = vdc_port_match,
921 .probe = vdc_port_probe,
922 .remove = vdc_port_remove,
926 static int __init vdc_init(void)
930 err = register_blkdev(0, VDCBLK_NAME);
936 err = vio_register_driver(&vdc_port_driver);
938 goto out_unregister_blkdev;
942 out_unregister_blkdev:
943 unregister_blkdev(vdc_major, VDCBLK_NAME);
950 static void __exit vdc_exit(void)
952 vio_unregister_driver(&vdc_port_driver);
953 unregister_blkdev(vdc_major, VDCBLK_NAME);
956 module_init(vdc_init);
957 module_exit(vdc_exit);