1 // SPDX-License-Identifier: GPL-2.0-only
2 /* sunvdc.c: Sun LDOM Virtual Disk Client.
4 * Copyright (C) 2007, 2008 David S. Miller <davem@davemloft.net>
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/genhd.h>
13 #include <linux/cdrom.h>
14 #include <linux/slab.h>
15 #include <linux/spinlock.h>
16 #include <linux/completion.h>
17 #include <linux/delay.h>
18 #include <linux/init.h>
19 #include <linux/list.h>
20 #include <linux/scatterlist.h>
25 #define DRV_MODULE_NAME "sunvdc"
26 #define PFX DRV_MODULE_NAME ": "
27 #define DRV_MODULE_VERSION "1.2"
28 #define DRV_MODULE_RELDATE "November 24, 2014"
30 static char version[] =
31 DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
32 MODULE_AUTHOR("David S. Miller (davem@davemloft.net)");
33 MODULE_DESCRIPTION("Sun LDOM virtual disk client driver");
34 MODULE_LICENSE("GPL");
35 MODULE_VERSION(DRV_MODULE_VERSION);
37 #define VDC_TX_RING_SIZE 512
38 #define VDC_DEFAULT_BLK_SIZE 512
40 #define MAX_XFER_BLKS (128 * 1024)
41 #define MAX_XFER_SIZE (MAX_XFER_BLKS / VDC_DEFAULT_BLK_SIZE)
42 #define MAX_RING_COOKIES ((MAX_XFER_BLKS / PAGE_SIZE) + 2)
44 #define WAITING_FOR_LINK_UP 0x01
45 #define WAITING_FOR_TX_SPACE 0x02
46 #define WAITING_FOR_GEN_CMD 0x04
47 #define WAITING_FOR_ANY -1
49 #define VDC_MAX_RETRIES 10
51 static struct workqueue_struct *sunvdc_wq;
53 struct vdc_req_entry {
58 struct vio_driver_state vio;
62 struct vdc_completion *cmp;
66 struct vdc_req_entry rq_arr[VDC_TX_RING_SIZE];
68 unsigned long ring_cookies;
75 struct delayed_work ldc_reset_timer_work;
76 struct work_struct ldc_reset_work;
78 /* The server fills these in for us in the disk attribute
87 struct blk_mq_tag_set tag_set;
92 static void vdc_ldc_reset(struct vdc_port *port);
93 static void vdc_ldc_reset_work(struct work_struct *work);
94 static void vdc_ldc_reset_timer_work(struct work_struct *work);
96 static inline struct vdc_port *to_vdc_port(struct vio_driver_state *vio)
98 return container_of(vio, struct vdc_port, vio);
101 /* Ordered from largest major to lowest */
102 static struct vio_version vdc_versions[] = {
103 { .major = 1, .minor = 2 },
104 { .major = 1, .minor = 1 },
105 { .major = 1, .minor = 0 },
108 static inline int vdc_version_supported(struct vdc_port *port,
109 u16 major, u16 minor)
111 return port->vio.ver.major == major && port->vio.ver.minor >= minor;
114 #define VDCBLK_NAME "vdisk"
115 static int vdc_major;
116 #define PARTITION_SHIFT 3
118 static inline u32 vdc_tx_dring_avail(struct vio_dring_state *dr)
120 return vio_dring_avail(dr, VDC_TX_RING_SIZE);
123 static int vdc_getgeo(struct block_device *bdev, struct hd_geometry *geo)
125 struct gendisk *disk = bdev->bd_disk;
126 sector_t nsect = get_capacity(disk);
127 sector_t cylinders = nsect;
131 sector_div(cylinders, geo->heads * geo->sectors);
132 geo->cylinders = cylinders;
133 if ((sector_t)(geo->cylinders + 1) * geo->heads * geo->sectors < nsect)
134 geo->cylinders = 0xffff;
139 /* Add ioctl/CDROM_GET_CAPABILITY to support cdrom_id in udev
140 * when vdisk_mtype is VD_MEDIA_TYPE_CD or VD_MEDIA_TYPE_DVD.
141 * Needed to be able to install inside an ldom from an iso image.
143 static int vdc_ioctl(struct block_device *bdev, fmode_t mode,
144 unsigned command, unsigned long argument)
146 struct vdc_port *port = bdev->bd_disk->private_data;
150 case CDROMMULTISESSION:
151 pr_debug(PFX "Multisession CDs not supported\n");
152 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
153 if (put_user(0, (char __user *)(argument + i)))
157 case CDROM_GET_CAPABILITY:
158 if (!vdc_version_supported(port, 1, 1))
160 switch (port->vdisk_mtype) {
161 case VD_MEDIA_TYPE_CD:
162 case VD_MEDIA_TYPE_DVD:
168 pr_debug(PFX "ioctl %08x not supported\n", command);
173 static const struct block_device_operations vdc_fops = {
174 .owner = THIS_MODULE,
175 .getgeo = vdc_getgeo,
177 .compat_ioctl = blkdev_compat_ptr_ioctl,
180 static void vdc_blk_queue_start(struct vdc_port *port)
182 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
184 /* restart blk queue when ring is half emptied. also called after
185 * handshake completes, so check for initial handshake before we've
188 if (port->disk && vdc_tx_dring_avail(dr) * 100 / VDC_TX_RING_SIZE >= 50)
189 blk_mq_start_stopped_hw_queues(port->disk->queue, true);
192 static void vdc_finish(struct vio_driver_state *vio, int err, int waiting_for)
195 (waiting_for == -1 ||
196 vio->cmp->waiting_for == waiting_for)) {
198 complete(&vio->cmp->com);
203 static void vdc_handshake_complete(struct vio_driver_state *vio)
205 struct vdc_port *port = to_vdc_port(vio);
207 cancel_delayed_work(&port->ldc_reset_timer_work);
208 vdc_finish(vio, 0, WAITING_FOR_LINK_UP);
209 vdc_blk_queue_start(port);
212 static int vdc_handle_unknown(struct vdc_port *port, void *arg)
214 struct vio_msg_tag *pkt = arg;
216 printk(KERN_ERR PFX "Received unknown msg [%02x:%02x:%04x:%08x]\n",
217 pkt->type, pkt->stype, pkt->stype_env, pkt->sid);
218 printk(KERN_ERR PFX "Resetting connection.\n");
220 ldc_disconnect(port->vio.lp);
225 static int vdc_send_attr(struct vio_driver_state *vio)
227 struct vdc_port *port = to_vdc_port(vio);
228 struct vio_disk_attr_info pkt;
230 memset(&pkt, 0, sizeof(pkt));
232 pkt.tag.type = VIO_TYPE_CTRL;
233 pkt.tag.stype = VIO_SUBTYPE_INFO;
234 pkt.tag.stype_env = VIO_ATTR_INFO;
235 pkt.tag.sid = vio_send_sid(vio);
237 pkt.xfer_mode = VIO_DRING_MODE;
238 pkt.vdisk_block_size = port->vdisk_block_size;
239 pkt.max_xfer_size = port->max_xfer_size;
241 viodbg(HS, "SEND ATTR xfer_mode[0x%x] blksz[%u] max_xfer[%llu]\n",
242 pkt.xfer_mode, pkt.vdisk_block_size, pkt.max_xfer_size);
244 return vio_ldc_send(&port->vio, &pkt, sizeof(pkt));
247 static int vdc_handle_attr(struct vio_driver_state *vio, void *arg)
249 struct vdc_port *port = to_vdc_port(vio);
250 struct vio_disk_attr_info *pkt = arg;
252 viodbg(HS, "GOT ATTR stype[0x%x] ops[%llx] disk_size[%llu] disk_type[%x] "
253 "mtype[0x%x] xfer_mode[0x%x] blksz[%u] max_xfer[%llu]\n",
254 pkt->tag.stype, pkt->operations,
255 pkt->vdisk_size, pkt->vdisk_type, pkt->vdisk_mtype,
256 pkt->xfer_mode, pkt->vdisk_block_size,
259 if (pkt->tag.stype == VIO_SUBTYPE_ACK) {
260 switch (pkt->vdisk_type) {
261 case VD_DISK_TYPE_DISK:
262 case VD_DISK_TYPE_SLICE:
266 printk(KERN_ERR PFX "%s: Bogus vdisk_type 0x%x\n",
267 vio->name, pkt->vdisk_type);
271 if (pkt->vdisk_block_size > port->vdisk_block_size) {
272 printk(KERN_ERR PFX "%s: BLOCK size increased "
275 port->vdisk_block_size, pkt->vdisk_block_size);
279 port->operations = pkt->operations;
280 port->vdisk_type = pkt->vdisk_type;
281 if (vdc_version_supported(port, 1, 1)) {
282 port->vdisk_size = pkt->vdisk_size;
283 port->vdisk_mtype = pkt->vdisk_mtype;
285 if (pkt->max_xfer_size < port->max_xfer_size)
286 port->max_xfer_size = pkt->max_xfer_size;
287 port->vdisk_block_size = pkt->vdisk_block_size;
289 port->vdisk_phys_blksz = VDC_DEFAULT_BLK_SIZE;
290 if (vdc_version_supported(port, 1, 2))
291 port->vdisk_phys_blksz = pkt->phys_block_size;
295 printk(KERN_ERR PFX "%s: Attribute NACK\n", vio->name);
301 static void vdc_end_special(struct vdc_port *port, struct vio_disk_desc *desc)
303 int err = desc->status;
305 vdc_finish(&port->vio, -err, WAITING_FOR_GEN_CMD);
308 static void vdc_end_one(struct vdc_port *port, struct vio_dring_state *dr,
311 struct vio_disk_desc *desc = vio_dring_entry(dr, index);
312 struct vdc_req_entry *rqe = &port->rq_arr[index];
315 if (unlikely(desc->hdr.state != VIO_DESC_DONE))
318 ldc_unmap(port->vio.lp, desc->cookies, desc->ncookies);
319 desc->hdr.state = VIO_DESC_FREE;
320 dr->cons = vio_dring_next(dr, index);
324 vdc_end_special(port, desc);
330 blk_mq_end_request(req, desc->status ? BLK_STS_IOERR : 0);
332 vdc_blk_queue_start(port);
335 static int vdc_ack(struct vdc_port *port, void *msgbuf)
337 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
338 struct vio_dring_data *pkt = msgbuf;
340 if (unlikely(pkt->dring_ident != dr->ident ||
341 pkt->start_idx != pkt->end_idx ||
342 pkt->start_idx >= VDC_TX_RING_SIZE))
345 vdc_end_one(port, dr, pkt->start_idx);
350 static int vdc_nack(struct vdc_port *port, void *msgbuf)
352 /* XXX Implement me XXX */
356 static void vdc_event(void *arg, int event)
358 struct vdc_port *port = arg;
359 struct vio_driver_state *vio = &port->vio;
363 spin_lock_irqsave(&vio->lock, flags);
365 if (unlikely(event == LDC_EVENT_RESET)) {
366 vio_link_state_change(vio, event);
367 queue_work(sunvdc_wq, &port->ldc_reset_work);
371 if (unlikely(event == LDC_EVENT_UP)) {
372 vio_link_state_change(vio, event);
376 if (unlikely(event != LDC_EVENT_DATA_READY)) {
377 pr_warn(PFX "Unexpected LDC event %d\n", event);
384 struct vio_msg_tag tag;
388 err = ldc_read(vio->lp, &msgbuf, sizeof(msgbuf));
389 if (unlikely(err < 0)) {
390 if (err == -ECONNRESET)
396 viodbg(DATA, "TAG [%02x:%02x:%04x:%08x]\n",
399 msgbuf.tag.stype_env,
401 err = vio_validate_sid(vio, &msgbuf.tag);
405 if (likely(msgbuf.tag.type == VIO_TYPE_DATA)) {
406 if (msgbuf.tag.stype == VIO_SUBTYPE_ACK)
407 err = vdc_ack(port, &msgbuf);
408 else if (msgbuf.tag.stype == VIO_SUBTYPE_NACK)
409 err = vdc_nack(port, &msgbuf);
411 err = vdc_handle_unknown(port, &msgbuf);
412 } else if (msgbuf.tag.type == VIO_TYPE_CTRL) {
413 err = vio_control_pkt_engine(vio, &msgbuf);
415 err = vdc_handle_unknown(port, &msgbuf);
421 vdc_finish(&port->vio, err, WAITING_FOR_ANY);
423 spin_unlock_irqrestore(&vio->lock, flags);
426 static int __vdc_tx_trigger(struct vdc_port *port)
428 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
429 struct vio_dring_data hdr = {
431 .type = VIO_TYPE_DATA,
432 .stype = VIO_SUBTYPE_INFO,
433 .stype_env = VIO_DRING_DATA,
434 .sid = vio_send_sid(&port->vio),
436 .dring_ident = dr->ident,
437 .start_idx = dr->prod,
443 hdr.seq = dr->snd_nxt;
446 err = vio_ldc_send(&port->vio, &hdr, sizeof(hdr));
452 if ((delay <<= 1) > 128)
454 if (retries++ > VDC_MAX_RETRIES)
456 } while (err == -EAGAIN);
458 if (err == -ENOTCONN)
463 static int __send_request(struct request *req)
465 struct vdc_port *port = req->q->disk->private_data;
466 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
467 struct scatterlist sg[MAX_RING_COOKIES];
468 struct vdc_req_entry *rqe;
469 struct vio_disk_desc *desc;
470 unsigned int map_perm;
475 if (WARN_ON(port->ring_cookies > MAX_RING_COOKIES))
478 map_perm = LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_IO;
480 if (rq_data_dir(req) == READ) {
481 map_perm |= LDC_MAP_W;
484 map_perm |= LDC_MAP_R;
488 sg_init_table(sg, port->ring_cookies);
489 nsg = blk_rq_map_sg(req->q, req, sg);
492 for (i = 0; i < nsg; i++)
495 desc = vio_dring_cur(dr);
497 err = ldc_map_sg(port->vio.lp, sg, nsg,
498 desc->cookies, port->ring_cookies,
501 printk(KERN_ERR PFX "ldc_map_sg() failure, err=%d.\n", err);
505 rqe = &port->rq_arr[dr->prod];
508 desc->hdr.ack = VIO_ACK_ENABLE;
509 desc->req_id = port->req_id;
510 desc->operation = op;
511 if (port->vdisk_type == VD_DISK_TYPE_DISK) {
517 desc->offset = (blk_rq_pos(req) << 9) / port->vdisk_block_size;
519 desc->ncookies = err;
521 /* This has to be a non-SMP write barrier because we are writing
522 * to memory which is shared with the peer LDOM.
525 desc->hdr.state = VIO_DESC_READY;
527 err = __vdc_tx_trigger(port);
529 printk(KERN_ERR PFX "vdc_tx_trigger() failure, err=%d\n", err);
532 dr->prod = vio_dring_next(dr, dr->prod);
538 static blk_status_t vdc_queue_rq(struct blk_mq_hw_ctx *hctx,
539 const struct blk_mq_queue_data *bd)
541 struct vdc_port *port = hctx->queue->queuedata;
542 struct vio_dring_state *dr;
545 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
547 blk_mq_start_request(bd->rq);
549 spin_lock_irqsave(&port->vio.lock, flags);
552 * Doing drain, just end the request in error
554 if (unlikely(port->drain)) {
555 spin_unlock_irqrestore(&port->vio.lock, flags);
556 return BLK_STS_IOERR;
559 if (unlikely(vdc_tx_dring_avail(dr) < 1)) {
560 spin_unlock_irqrestore(&port->vio.lock, flags);
561 blk_mq_stop_hw_queue(hctx);
562 return BLK_STS_DEV_RESOURCE;
565 if (__send_request(bd->rq) < 0) {
566 spin_unlock_irqrestore(&port->vio.lock, flags);
567 return BLK_STS_IOERR;
570 spin_unlock_irqrestore(&port->vio.lock, flags);
574 static int generic_request(struct vdc_port *port, u8 op, void *buf, int len)
576 struct vio_dring_state *dr;
577 struct vio_completion comp;
578 struct vio_disk_desc *desc;
579 unsigned int map_perm;
584 if (!(((u64)1 << (u64)op) & port->operations))
599 op_len = sizeof(u32);
600 map_perm = LDC_MAP_W;
604 op_len = sizeof(u32);
605 map_perm = LDC_MAP_R;
609 op_len = sizeof(struct vio_disk_vtoc);
610 map_perm = LDC_MAP_W;
614 op_len = sizeof(struct vio_disk_vtoc);
615 map_perm = LDC_MAP_R;
618 case VD_OP_GET_DISKGEOM:
619 op_len = sizeof(struct vio_disk_geom);
620 map_perm = LDC_MAP_W;
623 case VD_OP_SET_DISKGEOM:
624 op_len = sizeof(struct vio_disk_geom);
625 map_perm = LDC_MAP_R;
630 map_perm = LDC_MAP_RW;
633 case VD_OP_GET_DEVID:
634 op_len = sizeof(struct vio_disk_devid);
635 map_perm = LDC_MAP_W;
643 map_perm |= LDC_MAP_SHADOW | LDC_MAP_DIRECT | LDC_MAP_IO;
645 op_len = (op_len + 7) & ~7;
646 req_buf = kzalloc(op_len, GFP_KERNEL);
653 if (map_perm & LDC_MAP_R)
654 memcpy(req_buf, buf, len);
656 spin_lock_irqsave(&port->vio.lock, flags);
658 dr = &port->vio.drings[VIO_DRIVER_TX_RING];
660 /* XXX If we want to use this code generically we have to
661 * XXX handle TX ring exhaustion etc.
663 desc = vio_dring_cur(dr);
665 err = ldc_map_single(port->vio.lp, req_buf, op_len,
666 desc->cookies, port->ring_cookies,
669 spin_unlock_irqrestore(&port->vio.lock, flags);
674 init_completion(&comp.com);
675 comp.waiting_for = WAITING_FOR_GEN_CMD;
676 port->vio.cmp = ∁
678 desc->hdr.ack = VIO_ACK_ENABLE;
679 desc->req_id = port->req_id;
680 desc->operation = op;
685 desc->ncookies = err;
687 /* This has to be a non-SMP write barrier because we are writing
688 * to memory which is shared with the peer LDOM.
691 desc->hdr.state = VIO_DESC_READY;
693 err = __vdc_tx_trigger(port);
696 dr->prod = vio_dring_next(dr, dr->prod);
697 spin_unlock_irqrestore(&port->vio.lock, flags);
699 wait_for_completion(&comp.com);
702 port->vio.cmp = NULL;
703 spin_unlock_irqrestore(&port->vio.lock, flags);
706 if (map_perm & LDC_MAP_W)
707 memcpy(buf, req_buf, len);
714 static int vdc_alloc_tx_ring(struct vdc_port *port)
716 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
717 unsigned long len, entry_size;
721 entry_size = sizeof(struct vio_disk_desc) +
722 (sizeof(struct ldc_trans_cookie) * port->ring_cookies);
723 len = (VDC_TX_RING_SIZE * entry_size);
725 ncookies = VIO_MAX_RING_COOKIES;
726 dring = ldc_alloc_exp_dring(port->vio.lp, len,
727 dr->cookies, &ncookies,
732 return PTR_ERR(dring);
735 dr->entry_size = entry_size;
736 dr->num_entries = VDC_TX_RING_SIZE;
737 dr->prod = dr->cons = 0;
738 dr->pending = VDC_TX_RING_SIZE;
739 dr->ncookies = ncookies;
744 static void vdc_free_tx_ring(struct vdc_port *port)
746 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
749 ldc_free_exp_dring(port->vio.lp, dr->base,
750 (dr->entry_size * dr->num_entries),
751 dr->cookies, dr->ncookies);
760 static int vdc_port_up(struct vdc_port *port)
762 struct vio_completion comp;
764 init_completion(&comp.com);
766 comp.waiting_for = WAITING_FOR_LINK_UP;
767 port->vio.cmp = ∁
769 vio_port_up(&port->vio);
770 wait_for_completion(&comp.com);
774 static void vdc_port_down(struct vdc_port *port)
776 ldc_disconnect(port->vio.lp);
777 ldc_unbind(port->vio.lp);
778 vdc_free_tx_ring(port);
779 vio_ldc_free(&port->vio);
782 static const struct blk_mq_ops vdc_mq_ops = {
783 .queue_rq = vdc_queue_rq,
786 static int probe_disk(struct vdc_port *port)
788 struct request_queue *q;
792 err = vdc_port_up(port);
796 /* Using version 1.2 means vdisk_phys_blksz should be set unless the
797 * disk is reserved by another system.
799 if (vdc_version_supported(port, 1, 2) && !port->vdisk_phys_blksz)
802 if (vdc_version_supported(port, 1, 1)) {
803 /* vdisk_size should be set during the handshake, if it wasn't
804 * then the underlying disk is reserved by another system
806 if (port->vdisk_size == -1)
809 struct vio_disk_geom geom;
811 err = generic_request(port, VD_OP_GET_DISKGEOM,
812 &geom, sizeof(geom));
814 printk(KERN_ERR PFX "VD_OP_GET_DISKGEOM returns "
818 port->vdisk_size = ((u64)geom.num_cyl *
823 err = blk_mq_alloc_sq_tag_set(&port->tag_set, &vdc_mq_ops,
824 VDC_TX_RING_SIZE, BLK_MQ_F_SHOULD_MERGE);
828 g = blk_mq_alloc_disk(&port->tag_set, port);
830 printk(KERN_ERR PFX "%s: Could not allocate gendisk.\n",
839 /* Each segment in a request is up to an aligned page in size. */
840 blk_queue_segment_boundary(q, PAGE_SIZE - 1);
841 blk_queue_max_segment_size(q, PAGE_SIZE);
843 blk_queue_max_segments(q, port->ring_cookies);
844 blk_queue_max_hw_sectors(q, port->max_xfer_size);
845 g->major = vdc_major;
846 g->first_minor = port->vio.vdev->dev_no << PARTITION_SHIFT;
847 g->minors = 1 << PARTITION_SHIFT;
848 strcpy(g->disk_name, port->disk_name);
852 g->private_data = port;
854 set_capacity(g, port->vdisk_size);
856 if (vdc_version_supported(port, 1, 1)) {
857 switch (port->vdisk_mtype) {
858 case VD_MEDIA_TYPE_CD:
859 pr_info(PFX "Virtual CDROM %s\n", port->disk_name);
860 g->flags |= GENHD_FL_REMOVABLE;
864 case VD_MEDIA_TYPE_DVD:
865 pr_info(PFX "Virtual DVD %s\n", port->disk_name);
866 g->flags |= GENHD_FL_REMOVABLE;
870 case VD_MEDIA_TYPE_FIXED:
871 pr_info(PFX "Virtual Hard disk %s\n", port->disk_name);
876 blk_queue_physical_block_size(q, port->vdisk_phys_blksz);
878 pr_info(PFX "%s: %u sectors (%u MB) protocol %d.%d\n",
880 port->vdisk_size, (port->vdisk_size >> (20 - 9)),
881 port->vio.ver.major, port->vio.ver.minor);
883 err = device_add_disk(&port->vio.vdev->dev, g, NULL);
885 goto out_cleanup_disk;
892 blk_mq_free_tag_set(&port->tag_set);
896 static struct ldc_channel_config vdc_ldc_cfg = {
899 .mode = LDC_MODE_UNRELIABLE,
902 static struct vio_driver_ops vdc_vio_ops = {
903 .send_attr = vdc_send_attr,
904 .handle_attr = vdc_handle_attr,
905 .handshake_complete = vdc_handshake_complete,
908 static void print_version(void)
910 static int version_printed;
912 if (version_printed++ == 0)
913 printk(KERN_INFO "%s", version);
916 struct vdc_check_port_data {
921 static int vdc_device_probed(struct device *dev, void *arg)
923 struct vio_dev *vdev = to_vio_dev(dev);
924 struct vdc_check_port_data *port_data;
926 port_data = (struct vdc_check_port_data *)arg;
928 if ((vdev->dev_no == port_data->dev_no) &&
929 (!(strcmp((char *)&vdev->type, port_data->type))) &&
930 dev_get_drvdata(dev)) {
931 /* This device has already been configured
932 * by vdc_port_probe()
940 /* Determine whether the VIO device is part of an mpgroup
941 * by locating all the virtual-device-port nodes associated
942 * with the parent virtual-device node for the VIO device
943 * and checking whether any of these nodes are vdc-ports
944 * which have already been configured.
946 * Returns true if this device is part of an mpgroup and has
947 * already been probed.
949 static bool vdc_port_mpgroup_check(struct vio_dev *vdev)
951 struct vdc_check_port_data port_data;
954 port_data.dev_no = vdev->dev_no;
955 port_data.type = (char *)&vdev->type;
957 dev = device_find_child(vdev->dev.parent, &port_data,
966 static int vdc_port_probe(struct vio_dev *vdev, const struct vio_device_id *id)
968 struct mdesc_handle *hp;
969 struct vdc_port *port;
971 const u64 *ldc_timeout;
978 if ((vdev->dev_no << PARTITION_SHIFT) & ~(u64)MINORMASK) {
979 printk(KERN_ERR PFX "Port id [%llu] too large.\n",
981 goto err_out_release_mdesc;
984 /* Check if this device is part of an mpgroup */
985 if (vdc_port_mpgroup_check(vdev)) {
987 "VIO: Ignoring extra vdisk port %s",
988 dev_name(&vdev->dev));
989 goto err_out_release_mdesc;
992 port = kzalloc(sizeof(*port), GFP_KERNEL);
995 goto err_out_release_mdesc;
998 if (vdev->dev_no >= 26)
999 snprintf(port->disk_name, sizeof(port->disk_name),
1001 'a' + ((int)vdev->dev_no / 26) - 1,
1002 'a' + ((int)vdev->dev_no % 26));
1004 snprintf(port->disk_name, sizeof(port->disk_name),
1005 VDCBLK_NAME "%c", 'a' + ((int)vdev->dev_no % 26));
1006 port->vdisk_size = -1;
1008 /* Actual wall time may be double due to do_generic_file_read() doing
1009 * a readahead I/O first, and once that fails it will try to read a
1012 ldc_timeout = mdesc_get_property(hp, vdev->mp, "vdc-timeout", NULL);
1013 port->ldc_timeout = ldc_timeout ? *ldc_timeout : 0;
1014 INIT_DELAYED_WORK(&port->ldc_reset_timer_work, vdc_ldc_reset_timer_work);
1015 INIT_WORK(&port->ldc_reset_work, vdc_ldc_reset_work);
1017 err = vio_driver_init(&port->vio, vdev, VDEV_DISK,
1018 vdc_versions, ARRAY_SIZE(vdc_versions),
1019 &vdc_vio_ops, port->disk_name);
1021 goto err_out_free_port;
1023 port->vdisk_block_size = VDC_DEFAULT_BLK_SIZE;
1024 port->max_xfer_size = MAX_XFER_SIZE;
1025 port->ring_cookies = MAX_RING_COOKIES;
1027 err = vio_ldc_alloc(&port->vio, &vdc_ldc_cfg, port);
1029 goto err_out_free_port;
1031 err = vdc_alloc_tx_ring(port);
1033 goto err_out_free_ldc;
1035 err = probe_disk(port);
1037 goto err_out_free_tx_ring;
1039 /* Note that the device driver_data is used to determine
1040 * whether the port has been probed.
1042 dev_set_drvdata(&vdev->dev, port);
1048 err_out_free_tx_ring:
1049 vdc_free_tx_ring(port);
1052 vio_ldc_free(&port->vio);
1057 err_out_release_mdesc:
1062 static void vdc_port_remove(struct vio_dev *vdev)
1064 struct vdc_port *port = dev_get_drvdata(&vdev->dev);
1067 blk_mq_stop_hw_queues(port->disk->queue);
1069 flush_work(&port->ldc_reset_work);
1070 cancel_delayed_work_sync(&port->ldc_reset_timer_work);
1071 del_timer_sync(&port->vio.timer);
1073 del_gendisk(port->disk);
1074 blk_cleanup_disk(port->disk);
1075 blk_mq_free_tag_set(&port->tag_set);
1077 vdc_free_tx_ring(port);
1078 vio_ldc_free(&port->vio);
1080 dev_set_drvdata(&vdev->dev, NULL);
1086 static void vdc_requeue_inflight(struct vdc_port *port)
1088 struct vio_dring_state *dr = &port->vio.drings[VIO_DRIVER_TX_RING];
1091 for (idx = dr->cons; idx != dr->prod; idx = vio_dring_next(dr, idx)) {
1092 struct vio_disk_desc *desc = vio_dring_entry(dr, idx);
1093 struct vdc_req_entry *rqe = &port->rq_arr[idx];
1094 struct request *req;
1096 ldc_unmap(port->vio.lp, desc->cookies, desc->ncookies);
1097 desc->hdr.state = VIO_DESC_FREE;
1098 dr->cons = vio_dring_next(dr, idx);
1102 vdc_end_special(port, desc);
1107 blk_mq_requeue_request(req, false);
1111 static void vdc_queue_drain(struct vdc_port *port)
1113 struct request_queue *q = port->disk->queue;
1116 * Mark the queue as draining, then freeze/quiesce to ensure
1117 * that all existing requests are seen in ->queue_rq() and killed
1120 spin_unlock_irq(&port->vio.lock);
1122 blk_mq_freeze_queue(q);
1123 blk_mq_quiesce_queue(q);
1125 spin_lock_irq(&port->vio.lock);
1127 blk_mq_unquiesce_queue(q);
1128 blk_mq_unfreeze_queue(q);
1131 static void vdc_ldc_reset_timer_work(struct work_struct *work)
1133 struct vdc_port *port;
1134 struct vio_driver_state *vio;
1136 port = container_of(work, struct vdc_port, ldc_reset_timer_work.work);
1139 spin_lock_irq(&vio->lock);
1140 if (!(port->vio.hs_state & VIO_HS_COMPLETE)) {
1141 pr_warn(PFX "%s ldc down %llu seconds, draining queue\n",
1142 port->disk_name, port->ldc_timeout);
1143 vdc_queue_drain(port);
1144 vdc_blk_queue_start(port);
1146 spin_unlock_irq(&vio->lock);
1149 static void vdc_ldc_reset_work(struct work_struct *work)
1151 struct vdc_port *port;
1152 struct vio_driver_state *vio;
1153 unsigned long flags;
1155 port = container_of(work, struct vdc_port, ldc_reset_work);
1158 spin_lock_irqsave(&vio->lock, flags);
1159 vdc_ldc_reset(port);
1160 spin_unlock_irqrestore(&vio->lock, flags);
1163 static void vdc_ldc_reset(struct vdc_port *port)
1167 assert_spin_locked(&port->vio.lock);
1169 pr_warn(PFX "%s ldc link reset\n", port->disk_name);
1170 blk_mq_stop_hw_queues(port->disk->queue);
1171 vdc_requeue_inflight(port);
1172 vdc_port_down(port);
1174 err = vio_ldc_alloc(&port->vio, &vdc_ldc_cfg, port);
1176 pr_err(PFX "%s vio_ldc_alloc:%d\n", port->disk_name, err);
1180 err = vdc_alloc_tx_ring(port);
1182 pr_err(PFX "%s vio_alloc_tx_ring:%d\n", port->disk_name, err);
1186 if (port->ldc_timeout)
1187 mod_delayed_work(system_wq, &port->ldc_reset_timer_work,
1188 round_jiffies(jiffies + HZ * port->ldc_timeout));
1189 mod_timer(&port->vio.timer, round_jiffies(jiffies + HZ));
1193 vio_ldc_free(&port->vio);
1196 static const struct vio_device_id vdc_port_match[] = {
1202 MODULE_DEVICE_TABLE(vio, vdc_port_match);
1204 static struct vio_driver vdc_port_driver = {
1205 .id_table = vdc_port_match,
1206 .probe = vdc_port_probe,
1207 .remove = vdc_port_remove,
1211 static int __init vdc_init(void)
1215 sunvdc_wq = alloc_workqueue("sunvdc", 0, 0);
1219 err = register_blkdev(0, VDCBLK_NAME);
1225 err = vio_register_driver(&vdc_port_driver);
1227 goto out_unregister_blkdev;
1231 out_unregister_blkdev:
1232 unregister_blkdev(vdc_major, VDCBLK_NAME);
1236 destroy_workqueue(sunvdc_wq);
1240 static void __exit vdc_exit(void)
1242 vio_unregister_driver(&vdc_port_driver);
1243 unregister_blkdev(vdc_major, VDCBLK_NAME);
1244 destroy_workqueue(sunvdc_wq);
1247 module_init(vdc_init);
1248 module_exit(vdc_exit);