1 /* sunvdc.c: Sun LDOM Virtual Disk Client.
3 * Copyright (C) 2007 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>
22 #define DRV_MODULE_NAME "sunvdc"
23 #define PFX DRV_MODULE_NAME ": "
24 #define DRV_MODULE_VERSION "1.0"
25 #define DRV_MODULE_RELDATE "June 25, 2007"
27 static char version[] __devinitdata =
28 DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
29 MODULE_AUTHOR("David S. Miller (davem@davemloft.net)");
30 MODULE_DESCRIPTION("Sun LDOM virtual disk client driver");
31 MODULE_LICENSE("GPL");
32 MODULE_VERSION(DRV_MODULE_VERSION);
34 #define VDC_TX_RING_SIZE 256
36 #define WAITING_FOR_LINK_UP 0x01
37 #define WAITING_FOR_TX_SPACE 0x02
38 #define WAITING_FOR_GEN_CMD 0x04
39 #define WAITING_FOR_ANY -1
41 struct vdc_req_entry {
46 struct vio_driver_state vio;
50 struct vdc_completion *cmp;
54 struct vdc_req_entry rq_arr[VDC_TX_RING_SIZE];
56 unsigned long ring_cookies;
61 /* The server fills these in for us in the disk attribute
70 struct vio_disk_geom geom;
71 struct vio_disk_vtoc label;
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 = 0 },
84 #define VDCBLK_NAME "vdisk"
86 #define PARTITION_SHIFT 3
88 static inline u32 vdc_tx_dring_avail(struct vio_dring_state *dr)
90 return vio_dring_avail(dr, VDC_TX_RING_SIZE);
93 static int vdc_getgeo(struct block_device *bdev, struct hd_geometry *geo)
95 struct gendisk *disk = bdev->bd_disk;
96 struct vdc_port *port = disk->private_data;
98 geo->heads = (u8) port->geom.num_hd;
99 geo->sectors = (u8) port->geom.num_sec;
100 geo->cylinders = port->geom.num_cyl;
105 static struct block_device_operations vdc_fops = {
106 .owner = THIS_MODULE,
107 .getgeo = vdc_getgeo,
110 static void vdc_finish(struct vio_driver_state *vio, int err, int waiting_for)
113 (waiting_for == -1 ||
114 vio->cmp->waiting_for == waiting_for)) {
116 complete(&vio->cmp->com);
121 static void vdc_handshake_complete(struct vio_driver_state *vio)
123 vdc_finish(vio, 0, WAITING_FOR_LINK_UP);
126 static int vdc_handle_unknown(struct vdc_port *port, void *arg)
128 struct vio_msg_tag *pkt = arg;
130 printk(KERN_ERR PFX "Received unknown msg [%02x:%02x:%04x:%08x]\n",
131 pkt->type, pkt->stype, pkt->stype_env, pkt->sid);
132 printk(KERN_ERR PFX "Resetting connection.\n");
134 ldc_disconnect(port->vio.lp);
139 static int vdc_send_attr(struct vio_driver_state *vio)
141 struct vdc_port *port = to_vdc_port(vio);
142 struct vio_disk_attr_info pkt;
144 memset(&pkt, 0, sizeof(pkt));
146 pkt.tag.type = VIO_TYPE_CTRL;
147 pkt.tag.stype = VIO_SUBTYPE_INFO;
148 pkt.tag.stype_env = VIO_ATTR_INFO;
149 pkt.tag.sid = vio_send_sid(vio);
151 pkt.xfer_mode = VIO_DRING_MODE;
152 pkt.vdisk_block_size = port->vdisk_block_size;
153 pkt.max_xfer_size = port->max_xfer_size;
155 viodbg(HS, "SEND ATTR xfer_mode[0x%x] blksz[%u] max_xfer[%lu]\n",
156 pkt.xfer_mode, pkt.vdisk_block_size, pkt.max_xfer_size);
158 return vio_ldc_send(&port->vio, &pkt, sizeof(pkt));
161 static int vdc_handle_attr(struct vio_driver_state *vio, void *arg)
163 struct vdc_port *port = to_vdc_port(vio);
164 struct vio_disk_attr_info *pkt = arg;
166 viodbg(HS, "GOT ATTR stype[0x%x] ops[%lx] disk_size[%lu] disk_type[%x] "
167 "xfer_mode[0x%x] blksz[%u] max_xfer[%lu]\n",
168 pkt->tag.stype, pkt->operations,
169 pkt->vdisk_size, pkt->vdisk_type,
170 pkt->xfer_mode, pkt->vdisk_block_size,
173 if (pkt->tag.stype == VIO_SUBTYPE_ACK) {
174 switch (pkt->vdisk_type) {
175 case VD_DISK_TYPE_DISK:
176 case VD_DISK_TYPE_SLICE:
180 printk(KERN_ERR PFX "%s: Bogus vdisk_type 0x%x\n",
181 vio->name, pkt->vdisk_type);
185 if (pkt->vdisk_block_size > port->vdisk_block_size) {
186 printk(KERN_ERR PFX "%s: BLOCK size increased "
189 port->vdisk_block_size, pkt->vdisk_block_size);
193 port->operations = pkt->operations;
194 port->vdisk_size = pkt->vdisk_size;
195 port->vdisk_type = pkt->vdisk_type;
196 if (pkt->max_xfer_size < port->max_xfer_size)
197 port->max_xfer_size = pkt->max_xfer_size;
198 port->vdisk_block_size = pkt->vdisk_block_size;
201 printk(KERN_ERR PFX "%s: Attribute NACK\n", vio->name);
207 static void vdc_end_special(struct vdc_port *port, struct vio_disk_desc *desc)
209 int err = desc->status;
211 vdc_finish(&port->vio, -err, WAITING_FOR_GEN_CMD);
214 static void vdc_end_request(struct request *req, int uptodate, int num_sectors)
216 if (end_that_request_first(req, uptodate, num_sectors))
218 add_disk_randomness(req->rq_disk);
219 end_that_request_last(req, uptodate);
222 static void vdc_end_one(struct vdc_port *port, struct vio_dring_state *dr,
225 struct vio_disk_desc *desc = vio_dring_entry(dr, index);
226 struct vdc_req_entry *rqe = &port->rq_arr[index];
229 if (unlikely(desc->hdr.state != VIO_DESC_DONE))
232 ldc_unmap(port->vio.lp, desc->cookies, desc->ncookies);
233 desc->hdr.state = VIO_DESC_FREE;
234 dr->cons = (index + 1) & (VDC_TX_RING_SIZE - 1);
238 vdc_end_special(port, desc);
244 vdc_end_request(req, !desc->status, desc->size >> 9);
246 if (blk_queue_stopped(port->disk->queue))
247 blk_start_queue(port->disk->queue);
250 static int vdc_ack(struct vdc_port *port, void *msgbuf)
252 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
253 struct vio_dring_data *pkt = msgbuf;
255 if (unlikely(pkt->dring_ident != dr->ident ||
256 pkt->start_idx != pkt->end_idx ||
257 pkt->start_idx >= VDC_TX_RING_SIZE))
260 vdc_end_one(port, dr, pkt->start_idx);
265 static int vdc_nack(struct vdc_port *port, void *msgbuf)
267 /* XXX Implement me XXX */
271 static void vdc_event(void *arg, int event)
273 struct vdc_port *port = arg;
274 struct vio_driver_state *vio = &port->vio;
278 spin_lock_irqsave(&vio->lock, flags);
280 if (unlikely(event == LDC_EVENT_RESET ||
281 event == LDC_EVENT_UP)) {
282 vio_link_state_change(vio, event);
283 spin_unlock_irqrestore(&vio->lock, flags);
287 if (unlikely(event != LDC_EVENT_DATA_READY)) {
288 printk(KERN_WARNING PFX "Unexpected LDC event %d\n", event);
289 spin_unlock_irqrestore(&vio->lock, flags);
296 struct vio_msg_tag tag;
300 err = ldc_read(vio->lp, &msgbuf, sizeof(msgbuf));
301 if (unlikely(err < 0)) {
302 if (err == -ECONNRESET)
308 viodbg(DATA, "TAG [%02x:%02x:%04x:%08x]\n",
311 msgbuf.tag.stype_env,
313 err = vio_validate_sid(vio, &msgbuf.tag);
317 if (likely(msgbuf.tag.type == VIO_TYPE_DATA)) {
318 if (msgbuf.tag.stype == VIO_SUBTYPE_ACK)
319 err = vdc_ack(port, &msgbuf);
320 else if (msgbuf.tag.stype == VIO_SUBTYPE_NACK)
321 err = vdc_nack(port, &msgbuf);
323 err = vdc_handle_unknown(port, &msgbuf);
324 } else if (msgbuf.tag.type == VIO_TYPE_CTRL) {
325 err = vio_control_pkt_engine(vio, &msgbuf);
327 err = vdc_handle_unknown(port, &msgbuf);
333 vdc_finish(&port->vio, err, WAITING_FOR_ANY);
334 spin_unlock_irqrestore(&vio->lock, flags);
337 static int __vdc_tx_trigger(struct vdc_port *port)
339 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
340 struct vio_dring_data hdr = {
342 .type = VIO_TYPE_DATA,
343 .stype = VIO_SUBTYPE_INFO,
344 .stype_env = VIO_DRING_DATA,
345 .sid = vio_send_sid(&port->vio),
347 .dring_ident = dr->ident,
348 .start_idx = dr->prod,
353 hdr.seq = dr->snd_nxt;
356 err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
362 if ((delay <<= 1) > 128)
364 } while (err == -EAGAIN);
369 static int __send_request(struct request *req)
371 struct vdc_port *port = req->rq_disk->private_data;
372 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
373 struct scatterlist sg[port->ring_cookies];
374 struct vdc_req_entry *rqe;
375 struct vio_disk_desc *desc;
376 unsigned int map_perm;
381 map_perm = LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_IO;
383 if (rq_data_dir(req) == READ) {
384 map_perm |= LDC_MAP_W;
387 map_perm |= LDC_MAP_R;
391 nsg = blk_rq_map_sg(req->q, req, sg);
394 for (i = 0; i < nsg; i++)
397 if (unlikely(vdc_tx_dring_avail(dr) < 1)) {
398 blk_stop_queue(port->disk->queue);
403 desc = vio_dring_cur(dr);
405 err = ldc_map_sg(port->vio.lp, sg, nsg,
406 desc->cookies, port->ring_cookies,
409 printk(KERN_ERR PFX "ldc_map_sg() failure, err=%d.\n", err);
413 rqe = &port->rq_arr[dr->prod];
416 desc->hdr.ack = VIO_ACK_ENABLE;
417 desc->req_id = port->req_id;
418 desc->operation = op;
419 if (port->vdisk_type == VD_DISK_TYPE_DISK) {
425 desc->offset = (req->sector << 9) / port->vdisk_block_size;
427 desc->ncookies = err;
429 /* This has to be a non-SMP write barrier because we are writing
430 * to memory which is shared with the peer LDOM.
433 desc->hdr.state = VIO_DESC_READY;
435 err = __vdc_tx_trigger(port);
437 printk(KERN_ERR PFX "vdc_tx_trigger() failure, err=%d\n", err);
440 dr->prod = (dr->prod + 1) & (VDC_TX_RING_SIZE - 1);
447 static void do_vdc_request(request_queue_t *q)
450 struct request *req = elv_next_request(q);
455 blkdev_dequeue_request(req);
456 if (__send_request(req) < 0)
457 vdc_end_request(req, 0, req->hard_nr_sectors);
461 static int generic_request(struct vdc_port *port, u8 op, void *buf, int len)
463 struct vio_dring_state *dr;
464 struct vio_completion comp;
465 struct vio_disk_desc *desc;
466 unsigned int map_perm;
471 if (!(((u64)1 << ((u64)op - 1)) & port->operations))
486 op_len = sizeof(u32);
487 map_perm = LDC_MAP_W;
491 op_len = sizeof(u32);
492 map_perm = LDC_MAP_R;
496 op_len = sizeof(struct vio_disk_vtoc);
497 map_perm = LDC_MAP_W;
501 op_len = sizeof(struct vio_disk_vtoc);
502 map_perm = LDC_MAP_R;
505 case VD_OP_GET_DISKGEOM:
506 op_len = sizeof(struct vio_disk_geom);
507 map_perm = LDC_MAP_W;
510 case VD_OP_SET_DISKGEOM:
511 op_len = sizeof(struct vio_disk_geom);
512 map_perm = LDC_MAP_R;
517 map_perm = LDC_MAP_RW;
520 case VD_OP_GET_DEVID:
521 op_len = sizeof(struct vio_disk_devid);
522 map_perm = LDC_MAP_W;
531 map_perm |= LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_IO;
533 op_len = (op_len + 7) & ~7;
534 req_buf = kzalloc(op_len, GFP_KERNEL);
541 if (map_perm & LDC_MAP_R)
542 memcpy(req_buf, buf, len);
544 spin_lock_irqsave(&port->vio.lock, flags);
546 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
548 /* XXX If we want to use this code generically we have to
549 * XXX handle TX ring exhaustion etc.
551 desc = vio_dring_cur(dr);
553 err = ldc_map_single(port->vio.lp, req_buf, op_len,
554 desc->cookies, port->ring_cookies,
557 spin_unlock_irqrestore(&port->vio.lock, flags);
562 init_completion(&comp.com);
563 comp.waiting_for = WAITING_FOR_GEN_CMD;
564 port->vio.cmp = ∁
566 desc->hdr.ack = VIO_ACK_ENABLE;
567 desc->req_id = port->req_id;
568 desc->operation = op;
573 desc->ncookies = err;
575 /* This has to be a non-SMP write barrier because we are writing
576 * to memory which is shared with the peer LDOM.
579 desc->hdr.state = VIO_DESC_READY;
581 err = __vdc_tx_trigger(port);
584 dr->prod = (dr->prod + 1) & (VDC_TX_RING_SIZE - 1);
585 spin_unlock_irqrestore(&port->vio.lock, flags);
587 wait_for_completion(&comp.com);
590 port->vio.cmp = NULL;
591 spin_unlock_irqrestore(&port->vio.lock, flags);
594 if (map_perm & LDC_MAP_W)
595 memcpy(buf, req_buf, len);
602 static int __devinit vdc_alloc_tx_ring(struct vdc_port *port)
604 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
605 unsigned long len, entry_size;
609 entry_size = sizeof(struct vio_disk_desc) +
610 (sizeof(struct ldc_trans_cookie) * port->ring_cookies);
611 len = (VDC_TX_RING_SIZE * entry_size);
613 ncookies = VIO_MAX_RING_COOKIES;
614 dring = ldc_alloc_exp_dring(port->vio.lp, len,
615 dr->cookies, &ncookies,
620 return PTR_ERR(dring);
623 dr->entry_size = entry_size;
624 dr->num_entries = VDC_TX_RING_SIZE;
625 dr->prod = dr->cons = 0;
626 dr->pending = VDC_TX_RING_SIZE;
627 dr->ncookies = ncookies;
632 static void vdc_free_tx_ring(struct vdc_port *port)
634 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
637 ldc_free_exp_dring(port->vio.lp, dr->base,
638 (dr->entry_size * dr->num_entries),
639 dr->cookies, dr->ncookies);
648 static int probe_disk(struct vdc_port *port)
650 struct vio_completion comp;
651 struct request_queue *q;
655 init_completion(&comp.com);
657 comp.waiting_for = WAITING_FOR_LINK_UP;
658 port->vio.cmp = ∁
660 vio_port_up(&port->vio);
662 wait_for_completion(&comp.com);
666 err = generic_request(port, VD_OP_GET_VTOC,
667 &port->label, sizeof(port->label));
669 printk(KERN_ERR PFX "VD_OP_GET_VTOC returns error %d\n", err);
673 err = generic_request(port, VD_OP_GET_DISKGEOM,
674 &port->geom, sizeof(port->geom));
676 printk(KERN_ERR PFX "VD_OP_GET_DISKGEOM returns "
681 port->vdisk_size = ((u64)port->geom.num_cyl *
682 (u64)port->geom.num_hd *
683 (u64)port->geom.num_sec);
685 q = blk_init_queue(do_vdc_request, &port->vio.lock);
687 printk(KERN_ERR PFX "%s: Could not allocate queue.\n",
691 g = alloc_disk(1 << PARTITION_SHIFT);
693 printk(KERN_ERR PFX "%s: Could not allocate gendisk.\n",
695 blk_cleanup_queue(q);
701 blk_queue_max_hw_segments(q, port->ring_cookies);
702 blk_queue_max_phys_segments(q, port->ring_cookies);
703 blk_queue_max_sectors(q, port->max_xfer_size);
704 g->major = vdc_major;
705 g->first_minor = port->vio.vdev->dev_no << PARTITION_SHIFT;
706 strcpy(g->disk_name, port->disk_name);
710 g->private_data = port;
711 g->driverfs_dev = &port->vio.vdev->dev;
713 set_capacity(g, port->vdisk_size);
715 printk(KERN_INFO PFX "%s: %u sectors (%u MB)\n",
717 port->vdisk_size, (port->vdisk_size >> (20 - 9)));
724 static struct ldc_channel_config vdc_ldc_cfg = {
727 .mode = LDC_MODE_UNRELIABLE,
730 static struct vio_driver_ops vdc_vio_ops = {
731 .send_attr = vdc_send_attr,
732 .handle_attr = vdc_handle_attr,
733 .handshake_complete = vdc_handshake_complete,
736 static void print_version(void)
738 static int version_printed;
740 if (version_printed++ == 0)
741 printk(KERN_INFO "%s", version);
744 static int __devinit vdc_port_probe(struct vio_dev *vdev,
745 const struct vio_device_id *id)
747 struct mdesc_handle *hp;
748 struct vdc_port *port;
756 if ((vdev->dev_no << PARTITION_SHIFT) & ~(u64)MINORMASK) {
757 printk(KERN_ERR PFX "Port id [%lu] too large.\n",
759 goto err_out_release_mdesc;
762 port = kzalloc(sizeof(*port), GFP_KERNEL);
765 printk(KERN_ERR PFX "Cannot allocate vdc_port.\n");
766 goto err_out_release_mdesc;
769 if (vdev->dev_no >= 26)
770 snprintf(port->disk_name, sizeof(port->disk_name),
772 'a' + ((int)vdev->dev_no / 26) - 1,
773 'a' + ((int)vdev->dev_no % 26));
775 snprintf(port->disk_name, sizeof(port->disk_name),
776 VDCBLK_NAME "%c", 'a' + ((int)vdev->dev_no % 26));
778 err = vio_driver_init(&port->vio, vdev, VDEV_DISK,
779 vdc_versions, ARRAY_SIZE(vdc_versions),
780 &vdc_vio_ops, port->disk_name);
782 goto err_out_free_port;
784 port->vdisk_block_size = 512;
785 port->max_xfer_size = ((128 * 1024) / port->vdisk_block_size);
786 port->ring_cookies = ((port->max_xfer_size *
787 port->vdisk_block_size) / PAGE_SIZE) + 2;
789 err = vio_ldc_alloc(&port->vio, &vdc_ldc_cfg, port);
791 goto err_out_free_port;
793 err = vdc_alloc_tx_ring(port);
795 goto err_out_free_ldc;
797 err = probe_disk(port);
799 goto err_out_free_tx_ring;
801 dev_set_drvdata(&vdev->dev, port);
807 err_out_free_tx_ring:
808 vdc_free_tx_ring(port);
811 vio_ldc_free(&port->vio);
816 err_out_release_mdesc:
821 static int vdc_port_remove(struct vio_dev *vdev)
823 struct vdc_port *port = dev_get_drvdata(&vdev->dev);
826 del_timer_sync(&port->vio.timer);
828 vdc_free_tx_ring(port);
829 vio_ldc_free(&port->vio);
831 dev_set_drvdata(&vdev->dev, NULL);
838 static struct vio_device_id vdc_port_match[] = {
844 MODULE_DEVICE_TABLE(vio, vdc_port_match);
846 static struct vio_driver vdc_port_driver = {
847 .id_table = vdc_port_match,
848 .probe = vdc_port_probe,
849 .remove = vdc_port_remove,
852 .owner = THIS_MODULE,
856 static int __init vdc_init(void)
860 err = register_blkdev(0, VDCBLK_NAME);
866 err = vio_register_driver(&vdc_port_driver);
868 goto out_unregister_blkdev;
872 out_unregister_blkdev:
873 unregister_blkdev(vdc_major, VDCBLK_NAME);
880 static void __exit vdc_exit(void)
882 vio_unregister_driver(&vdc_port_driver);
883 unregister_blkdev(vdc_major, VDCBLK_NAME);
886 module_init(vdc_init);
887 module_exit(vdc_exit);