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/blk-mq.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 "1.2"
27 #define DRV_MODULE_RELDATE "November 24, 2014"
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
37 #define VDC_DEFAULT_BLK_SIZE 512
39 #define MAX_XFER_BLKS (128 * 1024)
40 #define MAX_XFER_SIZE (MAX_XFER_BLKS / VDC_DEFAULT_BLK_SIZE)
41 #define MAX_RING_COOKIES ((MAX_XFER_BLKS / PAGE_SIZE) + 2)
43 #define WAITING_FOR_LINK_UP 0x01
44 #define WAITING_FOR_TX_SPACE 0x02
45 #define WAITING_FOR_GEN_CMD 0x04
46 #define WAITING_FOR_ANY -1
48 #define VDC_MAX_RETRIES 10
50 static struct workqueue_struct
*sunvdc_wq
;
52 struct vdc_req_entry
{
57 struct vio_driver_state vio
;
61 struct vdc_completion
*cmp
;
65 struct vdc_req_entry rq_arr
[VDC_TX_RING_SIZE
];
67 unsigned long ring_cookies
;
74 struct delayed_work ldc_reset_timer_work
;
75 struct work_struct ldc_reset_work
;
77 /* The server fills these in for us in the disk attribute
86 struct blk_mq_tag_set tag_set
;
91 static void vdc_ldc_reset(struct vdc_port
*port
);
92 static void vdc_ldc_reset_work(struct work_struct
*work
);
93 static void vdc_ldc_reset_timer_work(struct work_struct
*work
);
95 static inline struct vdc_port
*to_vdc_port(struct vio_driver_state
*vio
)
97 return container_of(vio
, struct vdc_port
, vio
);
100 /* Ordered from largest major to lowest */
101 static struct vio_version vdc_versions
[] = {
102 { .major
= 1, .minor
= 2 },
103 { .major
= 1, .minor
= 1 },
104 { .major
= 1, .minor
= 0 },
107 static inline int vdc_version_supported(struct vdc_port
*port
,
108 u16 major
, u16 minor
)
110 return port
->vio
.ver
.major
== major
&& port
->vio
.ver
.minor
>= minor
;
113 #define VDCBLK_NAME "vdisk"
114 static int vdc_major
;
115 #define PARTITION_SHIFT 3
117 static inline u32
vdc_tx_dring_avail(struct vio_dring_state
*dr
)
119 return vio_dring_avail(dr
, VDC_TX_RING_SIZE
);
122 static int vdc_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
124 struct gendisk
*disk
= bdev
->bd_disk
;
125 sector_t nsect
= get_capacity(disk
);
126 sector_t cylinders
= nsect
;
130 sector_div(cylinders
, geo
->heads
* geo
->sectors
);
131 geo
->cylinders
= cylinders
;
132 if ((sector_t
)(geo
->cylinders
+ 1) * geo
->heads
* geo
->sectors
< nsect
)
133 geo
->cylinders
= 0xffff;
138 /* Add ioctl/CDROM_GET_CAPABILITY to support cdrom_id in udev
139 * when vdisk_mtype is VD_MEDIA_TYPE_CD or VD_MEDIA_TYPE_DVD.
140 * Needed to be able to install inside an ldom from an iso image.
142 static int vdc_ioctl(struct block_device
*bdev
, fmode_t mode
,
143 unsigned command
, unsigned long argument
)
146 struct gendisk
*disk
;
149 case CDROMMULTISESSION
:
150 pr_debug(PFX
"Multisession CDs not supported\n");
151 for (i
= 0; i
< sizeof(struct cdrom_multisession
); i
++)
152 if (put_user(0, (char __user
*)(argument
+ i
)))
156 case CDROM_GET_CAPABILITY
:
157 disk
= bdev
->bd_disk
;
159 if (bdev
->bd_disk
&& (disk
->flags
& GENHD_FL_CD
))
164 pr_debug(PFX
"ioctl %08x not supported\n", command
);
169 static const struct block_device_operations vdc_fops
= {
170 .owner
= THIS_MODULE
,
171 .getgeo
= vdc_getgeo
,
175 static void vdc_blk_queue_start(struct vdc_port
*port
)
177 struct vio_dring_state
*dr
= &port
->vio
.drings
[VIO_DRIVER_TX_RING
];
179 /* restart blk queue when ring is half emptied. also called after
180 * handshake completes, so check for initial handshake before we've
183 if (port
->disk
&& vdc_tx_dring_avail(dr
) * 100 / VDC_TX_RING_SIZE
>= 50)
184 blk_mq_start_stopped_hw_queues(port
->disk
->queue
, true);
187 static void vdc_finish(struct vio_driver_state
*vio
, int err
, int waiting_for
)
190 (waiting_for
== -1 ||
191 vio
->cmp
->waiting_for
== waiting_for
)) {
193 complete(&vio
->cmp
->com
);
198 static void vdc_handshake_complete(struct vio_driver_state
*vio
)
200 struct vdc_port
*port
= to_vdc_port(vio
);
202 cancel_delayed_work(&port
->ldc_reset_timer_work
);
203 vdc_finish(vio
, 0, WAITING_FOR_LINK_UP
);
204 vdc_blk_queue_start(port
);
207 static int vdc_handle_unknown(struct vdc_port
*port
, void *arg
)
209 struct vio_msg_tag
*pkt
= arg
;
211 printk(KERN_ERR PFX
"Received unknown msg [%02x:%02x:%04x:%08x]\n",
212 pkt
->type
, pkt
->stype
, pkt
->stype_env
, pkt
->sid
);
213 printk(KERN_ERR PFX
"Resetting connection.\n");
215 ldc_disconnect(port
->vio
.lp
);
220 static int vdc_send_attr(struct vio_driver_state
*vio
)
222 struct vdc_port
*port
= to_vdc_port(vio
);
223 struct vio_disk_attr_info pkt
;
225 memset(&pkt
, 0, sizeof(pkt
));
227 pkt
.tag
.type
= VIO_TYPE_CTRL
;
228 pkt
.tag
.stype
= VIO_SUBTYPE_INFO
;
229 pkt
.tag
.stype_env
= VIO_ATTR_INFO
;
230 pkt
.tag
.sid
= vio_send_sid(vio
);
232 pkt
.xfer_mode
= VIO_DRING_MODE
;
233 pkt
.vdisk_block_size
= port
->vdisk_block_size
;
234 pkt
.max_xfer_size
= port
->max_xfer_size
;
236 viodbg(HS
, "SEND ATTR xfer_mode[0x%x] blksz[%u] max_xfer[%llu]\n",
237 pkt
.xfer_mode
, pkt
.vdisk_block_size
, pkt
.max_xfer_size
);
239 return vio_ldc_send(&port
->vio
, &pkt
, sizeof(pkt
));
242 static int vdc_handle_attr(struct vio_driver_state
*vio
, void *arg
)
244 struct vdc_port
*port
= to_vdc_port(vio
);
245 struct vio_disk_attr_info
*pkt
= arg
;
247 viodbg(HS
, "GOT ATTR stype[0x%x] ops[%llx] disk_size[%llu] disk_type[%x] "
248 "mtype[0x%x] xfer_mode[0x%x] blksz[%u] max_xfer[%llu]\n",
249 pkt
->tag
.stype
, pkt
->operations
,
250 pkt
->vdisk_size
, pkt
->vdisk_type
, pkt
->vdisk_mtype
,
251 pkt
->xfer_mode
, pkt
->vdisk_block_size
,
254 if (pkt
->tag
.stype
== VIO_SUBTYPE_ACK
) {
255 switch (pkt
->vdisk_type
) {
256 case VD_DISK_TYPE_DISK
:
257 case VD_DISK_TYPE_SLICE
:
261 printk(KERN_ERR PFX
"%s: Bogus vdisk_type 0x%x\n",
262 vio
->name
, pkt
->vdisk_type
);
266 if (pkt
->vdisk_block_size
> port
->vdisk_block_size
) {
267 printk(KERN_ERR PFX
"%s: BLOCK size increased "
270 port
->vdisk_block_size
, pkt
->vdisk_block_size
);
274 port
->operations
= pkt
->operations
;
275 port
->vdisk_type
= pkt
->vdisk_type
;
276 if (vdc_version_supported(port
, 1, 1)) {
277 port
->vdisk_size
= pkt
->vdisk_size
;
278 port
->vdisk_mtype
= pkt
->vdisk_mtype
;
280 if (pkt
->max_xfer_size
< port
->max_xfer_size
)
281 port
->max_xfer_size
= pkt
->max_xfer_size
;
282 port
->vdisk_block_size
= pkt
->vdisk_block_size
;
284 port
->vdisk_phys_blksz
= VDC_DEFAULT_BLK_SIZE
;
285 if (vdc_version_supported(port
, 1, 2))
286 port
->vdisk_phys_blksz
= pkt
->phys_block_size
;
290 printk(KERN_ERR PFX
"%s: Attribute NACK\n", vio
->name
);
296 static void vdc_end_special(struct vdc_port
*port
, struct vio_disk_desc
*desc
)
298 int err
= desc
->status
;
300 vdc_finish(&port
->vio
, -err
, WAITING_FOR_GEN_CMD
);
303 static void vdc_end_one(struct vdc_port
*port
, struct vio_dring_state
*dr
,
306 struct vio_disk_desc
*desc
= vio_dring_entry(dr
, index
);
307 struct vdc_req_entry
*rqe
= &port
->rq_arr
[index
];
310 if (unlikely(desc
->hdr
.state
!= VIO_DESC_DONE
))
313 ldc_unmap(port
->vio
.lp
, desc
->cookies
, desc
->ncookies
);
314 desc
->hdr
.state
= VIO_DESC_FREE
;
315 dr
->cons
= vio_dring_next(dr
, index
);
319 vdc_end_special(port
, desc
);
325 blk_mq_end_request(req
, desc
->status
? BLK_STS_IOERR
: 0);
327 vdc_blk_queue_start(port
);
330 static int vdc_ack(struct vdc_port
*port
, void *msgbuf
)
332 struct vio_dring_state
*dr
= &port
->vio
.drings
[VIO_DRIVER_TX_RING
];
333 struct vio_dring_data
*pkt
= msgbuf
;
335 if (unlikely(pkt
->dring_ident
!= dr
->ident
||
336 pkt
->start_idx
!= pkt
->end_idx
||
337 pkt
->start_idx
>= VDC_TX_RING_SIZE
))
340 vdc_end_one(port
, dr
, pkt
->start_idx
);
345 static int vdc_nack(struct vdc_port
*port
, void *msgbuf
)
347 /* XXX Implement me XXX */
351 static void vdc_event(void *arg
, int event
)
353 struct vdc_port
*port
= arg
;
354 struct vio_driver_state
*vio
= &port
->vio
;
358 spin_lock_irqsave(&vio
->lock
, flags
);
360 if (unlikely(event
== LDC_EVENT_RESET
)) {
361 vio_link_state_change(vio
, event
);
362 queue_work(sunvdc_wq
, &port
->ldc_reset_work
);
366 if (unlikely(event
== LDC_EVENT_UP
)) {
367 vio_link_state_change(vio
, event
);
371 if (unlikely(event
!= LDC_EVENT_DATA_READY
)) {
372 pr_warn(PFX
"Unexpected LDC event %d\n", event
);
379 struct vio_msg_tag tag
;
383 err
= ldc_read(vio
->lp
, &msgbuf
, sizeof(msgbuf
));
384 if (unlikely(err
< 0)) {
385 if (err
== -ECONNRESET
)
391 viodbg(DATA
, "TAG [%02x:%02x:%04x:%08x]\n",
394 msgbuf
.tag
.stype_env
,
396 err
= vio_validate_sid(vio
, &msgbuf
.tag
);
400 if (likely(msgbuf
.tag
.type
== VIO_TYPE_DATA
)) {
401 if (msgbuf
.tag
.stype
== VIO_SUBTYPE_ACK
)
402 err
= vdc_ack(port
, &msgbuf
);
403 else if (msgbuf
.tag
.stype
== VIO_SUBTYPE_NACK
)
404 err
= vdc_nack(port
, &msgbuf
);
406 err
= vdc_handle_unknown(port
, &msgbuf
);
407 } else if (msgbuf
.tag
.type
== VIO_TYPE_CTRL
) {
408 err
= vio_control_pkt_engine(vio
, &msgbuf
);
410 err
= vdc_handle_unknown(port
, &msgbuf
);
416 vdc_finish(&port
->vio
, err
, WAITING_FOR_ANY
);
418 spin_unlock_irqrestore(&vio
->lock
, flags
);
421 static int __vdc_tx_trigger(struct vdc_port
*port
)
423 struct vio_dring_state
*dr
= &port
->vio
.drings
[VIO_DRIVER_TX_RING
];
424 struct vio_dring_data hdr
= {
426 .type
= VIO_TYPE_DATA
,
427 .stype
= VIO_SUBTYPE_INFO
,
428 .stype_env
= VIO_DRING_DATA
,
429 .sid
= vio_send_sid(&port
->vio
),
431 .dring_ident
= dr
->ident
,
432 .start_idx
= dr
->prod
,
438 hdr
.seq
= dr
->snd_nxt
;
441 err
= vio_ldc_send(&port
->vio
, &hdr
, sizeof(hdr
));
447 if ((delay
<<= 1) > 128)
449 if (retries
++ > VDC_MAX_RETRIES
)
451 } while (err
== -EAGAIN
);
453 if (err
== -ENOTCONN
)
458 static int __send_request(struct request
*req
)
460 struct vdc_port
*port
= req
->rq_disk
->private_data
;
461 struct vio_dring_state
*dr
= &port
->vio
.drings
[VIO_DRIVER_TX_RING
];
462 struct scatterlist sg
[MAX_RING_COOKIES
];
463 struct vdc_req_entry
*rqe
;
464 struct vio_disk_desc
*desc
;
465 unsigned int map_perm
;
470 if (WARN_ON(port
->ring_cookies
> MAX_RING_COOKIES
))
473 map_perm
= LDC_MAP_SHADOW
| LDC_MAP_DIRECT
| LDC_MAP_IO
;
475 if (rq_data_dir(req
) == READ
) {
476 map_perm
|= LDC_MAP_W
;
479 map_perm
|= LDC_MAP_R
;
483 sg_init_table(sg
, port
->ring_cookies
);
484 nsg
= blk_rq_map_sg(req
->q
, req
, sg
);
487 for (i
= 0; i
< nsg
; i
++)
490 desc
= vio_dring_cur(dr
);
492 err
= ldc_map_sg(port
->vio
.lp
, sg
, nsg
,
493 desc
->cookies
, port
->ring_cookies
,
496 printk(KERN_ERR PFX
"ldc_map_sg() failure, err=%d.\n", err
);
500 rqe
= &port
->rq_arr
[dr
->prod
];
503 desc
->hdr
.ack
= VIO_ACK_ENABLE
;
504 desc
->req_id
= port
->req_id
;
505 desc
->operation
= op
;
506 if (port
->vdisk_type
== VD_DISK_TYPE_DISK
) {
512 desc
->offset
= (blk_rq_pos(req
) << 9) / port
->vdisk_block_size
;
514 desc
->ncookies
= err
;
516 /* This has to be a non-SMP write barrier because we are writing
517 * to memory which is shared with the peer LDOM.
520 desc
->hdr
.state
= VIO_DESC_READY
;
522 err
= __vdc_tx_trigger(port
);
524 printk(KERN_ERR PFX
"vdc_tx_trigger() failure, err=%d\n", err
);
527 dr
->prod
= vio_dring_next(dr
, dr
->prod
);
533 static blk_status_t
vdc_queue_rq(struct blk_mq_hw_ctx
*hctx
,
534 const struct blk_mq_queue_data
*bd
)
536 struct vdc_port
*port
= hctx
->queue
->queuedata
;
537 struct vio_dring_state
*dr
;
540 dr
= &port
->vio
.drings
[VIO_DRIVER_TX_RING
];
542 blk_mq_start_request(bd
->rq
);
544 spin_lock_irqsave(&port
->vio
.lock
, flags
);
547 * Doing drain, just end the request in error
549 if (unlikely(port
->drain
)) {
550 spin_unlock_irqrestore(&port
->vio
.lock
, flags
);
551 return BLK_STS_IOERR
;
554 if (unlikely(vdc_tx_dring_avail(dr
) < 1)) {
555 spin_unlock_irqrestore(&port
->vio
.lock
, flags
);
556 blk_mq_stop_hw_queue(hctx
);
557 return BLK_STS_DEV_RESOURCE
;
560 if (__send_request(bd
->rq
) < 0) {
561 spin_unlock_irqrestore(&port
->vio
.lock
, flags
);
562 return BLK_STS_IOERR
;
565 spin_unlock_irqrestore(&port
->vio
.lock
, flags
);
569 static int generic_request(struct vdc_port
*port
, u8 op
, void *buf
, int len
)
571 struct vio_dring_state
*dr
;
572 struct vio_completion comp
;
573 struct vio_disk_desc
*desc
;
574 unsigned int map_perm
;
579 if (!(((u64
)1 << (u64
)op
) & port
->operations
))
594 op_len
= sizeof(u32
);
595 map_perm
= LDC_MAP_W
;
599 op_len
= sizeof(u32
);
600 map_perm
= LDC_MAP_R
;
604 op_len
= sizeof(struct vio_disk_vtoc
);
605 map_perm
= LDC_MAP_W
;
609 op_len
= sizeof(struct vio_disk_vtoc
);
610 map_perm
= LDC_MAP_R
;
613 case VD_OP_GET_DISKGEOM
:
614 op_len
= sizeof(struct vio_disk_geom
);
615 map_perm
= LDC_MAP_W
;
618 case VD_OP_SET_DISKGEOM
:
619 op_len
= sizeof(struct vio_disk_geom
);
620 map_perm
= LDC_MAP_R
;
625 map_perm
= LDC_MAP_RW
;
628 case VD_OP_GET_DEVID
:
629 op_len
= sizeof(struct vio_disk_devid
);
630 map_perm
= LDC_MAP_W
;
638 map_perm
|= LDC_MAP_SHADOW
| LDC_MAP_DIRECT
| LDC_MAP_IO
;
640 op_len
= (op_len
+ 7) & ~7;
641 req_buf
= kzalloc(op_len
, GFP_KERNEL
);
648 if (map_perm
& LDC_MAP_R
)
649 memcpy(req_buf
, buf
, len
);
651 spin_lock_irqsave(&port
->vio
.lock
, flags
);
653 dr
= &port
->vio
.drings
[VIO_DRIVER_TX_RING
];
655 /* XXX If we want to use this code generically we have to
656 * XXX handle TX ring exhaustion etc.
658 desc
= vio_dring_cur(dr
);
660 err
= ldc_map_single(port
->vio
.lp
, req_buf
, op_len
,
661 desc
->cookies
, port
->ring_cookies
,
664 spin_unlock_irqrestore(&port
->vio
.lock
, flags
);
669 init_completion(&comp
.com
);
670 comp
.waiting_for
= WAITING_FOR_GEN_CMD
;
671 port
->vio
.cmp
= &comp
;
673 desc
->hdr
.ack
= VIO_ACK_ENABLE
;
674 desc
->req_id
= port
->req_id
;
675 desc
->operation
= op
;
680 desc
->ncookies
= err
;
682 /* This has to be a non-SMP write barrier because we are writing
683 * to memory which is shared with the peer LDOM.
686 desc
->hdr
.state
= VIO_DESC_READY
;
688 err
= __vdc_tx_trigger(port
);
691 dr
->prod
= vio_dring_next(dr
, dr
->prod
);
692 spin_unlock_irqrestore(&port
->vio
.lock
, flags
);
694 wait_for_completion(&comp
.com
);
697 port
->vio
.cmp
= NULL
;
698 spin_unlock_irqrestore(&port
->vio
.lock
, flags
);
701 if (map_perm
& LDC_MAP_W
)
702 memcpy(buf
, req_buf
, len
);
709 static int vdc_alloc_tx_ring(struct vdc_port
*port
)
711 struct vio_dring_state
*dr
= &port
->vio
.drings
[VIO_DRIVER_TX_RING
];
712 unsigned long len
, entry_size
;
716 entry_size
= sizeof(struct vio_disk_desc
) +
717 (sizeof(struct ldc_trans_cookie
) * port
->ring_cookies
);
718 len
= (VDC_TX_RING_SIZE
* entry_size
);
720 ncookies
= VIO_MAX_RING_COOKIES
;
721 dring
= ldc_alloc_exp_dring(port
->vio
.lp
, len
,
722 dr
->cookies
, &ncookies
,
727 return PTR_ERR(dring
);
730 dr
->entry_size
= entry_size
;
731 dr
->num_entries
= VDC_TX_RING_SIZE
;
732 dr
->prod
= dr
->cons
= 0;
733 dr
->pending
= VDC_TX_RING_SIZE
;
734 dr
->ncookies
= ncookies
;
739 static void vdc_free_tx_ring(struct vdc_port
*port
)
741 struct vio_dring_state
*dr
= &port
->vio
.drings
[VIO_DRIVER_TX_RING
];
744 ldc_free_exp_dring(port
->vio
.lp
, dr
->base
,
745 (dr
->entry_size
* dr
->num_entries
),
746 dr
->cookies
, dr
->ncookies
);
755 static int vdc_port_up(struct vdc_port
*port
)
757 struct vio_completion comp
;
759 init_completion(&comp
.com
);
761 comp
.waiting_for
= WAITING_FOR_LINK_UP
;
762 port
->vio
.cmp
= &comp
;
764 vio_port_up(&port
->vio
);
765 wait_for_completion(&comp
.com
);
769 static void vdc_port_down(struct vdc_port
*port
)
771 ldc_disconnect(port
->vio
.lp
);
772 ldc_unbind(port
->vio
.lp
);
773 vdc_free_tx_ring(port
);
774 vio_ldc_free(&port
->vio
);
777 static const struct blk_mq_ops vdc_mq_ops
= {
778 .queue_rq
= vdc_queue_rq
,
781 static void cleanup_queue(struct request_queue
*q
)
783 struct vdc_port
*port
= q
->queuedata
;
785 blk_cleanup_queue(q
);
786 blk_mq_free_tag_set(&port
->tag_set
);
789 static struct request_queue
*init_queue(struct vdc_port
*port
)
791 struct request_queue
*q
;
793 q
= blk_mq_init_sq_queue(&port
->tag_set
, &vdc_mq_ops
, VDC_TX_RING_SIZE
,
794 BLK_MQ_F_SHOULD_MERGE
);
802 static int probe_disk(struct vdc_port
*port
)
804 struct request_queue
*q
;
808 err
= vdc_port_up(port
);
812 /* Using version 1.2 means vdisk_phys_blksz should be set unless the
813 * disk is reserved by another system.
815 if (vdc_version_supported(port
, 1, 2) && !port
->vdisk_phys_blksz
)
818 if (vdc_version_supported(port
, 1, 1)) {
819 /* vdisk_size should be set during the handshake, if it wasn't
820 * then the underlying disk is reserved by another system
822 if (port
->vdisk_size
== -1)
825 struct vio_disk_geom geom
;
827 err
= generic_request(port
, VD_OP_GET_DISKGEOM
,
828 &geom
, sizeof(geom
));
830 printk(KERN_ERR PFX
"VD_OP_GET_DISKGEOM returns "
834 port
->vdisk_size
= ((u64
)geom
.num_cyl
*
839 q
= init_queue(port
);
841 printk(KERN_ERR PFX
"%s: Could not allocate queue.\n",
845 g
= alloc_disk(1 << PARTITION_SHIFT
);
847 printk(KERN_ERR PFX
"%s: Could not allocate gendisk.\n",
855 /* Each segment in a request is up to an aligned page in size. */
856 blk_queue_segment_boundary(q
, PAGE_SIZE
- 1);
857 blk_queue_max_segment_size(q
, PAGE_SIZE
);
859 blk_queue_max_segments(q
, port
->ring_cookies
);
860 blk_queue_max_hw_sectors(q
, port
->max_xfer_size
);
861 g
->major
= vdc_major
;
862 g
->first_minor
= port
->vio
.vdev
->dev_no
<< PARTITION_SHIFT
;
863 strcpy(g
->disk_name
, port
->disk_name
);
867 g
->private_data
= port
;
869 set_capacity(g
, port
->vdisk_size
);
871 if (vdc_version_supported(port
, 1, 1)) {
872 switch (port
->vdisk_mtype
) {
873 case VD_MEDIA_TYPE_CD
:
874 pr_info(PFX
"Virtual CDROM %s\n", port
->disk_name
);
875 g
->flags
|= GENHD_FL_CD
;
876 g
->flags
|= GENHD_FL_REMOVABLE
;
880 case VD_MEDIA_TYPE_DVD
:
881 pr_info(PFX
"Virtual DVD %s\n", port
->disk_name
);
882 g
->flags
|= GENHD_FL_CD
;
883 g
->flags
|= GENHD_FL_REMOVABLE
;
887 case VD_MEDIA_TYPE_FIXED
:
888 pr_info(PFX
"Virtual Hard disk %s\n", port
->disk_name
);
893 blk_queue_physical_block_size(q
, port
->vdisk_phys_blksz
);
895 pr_info(PFX
"%s: %u sectors (%u MB) protocol %d.%d\n",
897 port
->vdisk_size
, (port
->vdisk_size
>> (20 - 9)),
898 port
->vio
.ver
.major
, port
->vio
.ver
.minor
);
900 device_add_disk(&port
->vio
.vdev
->dev
, g
, NULL
);
905 static struct ldc_channel_config vdc_ldc_cfg
= {
908 .mode
= LDC_MODE_UNRELIABLE
,
911 static struct vio_driver_ops vdc_vio_ops
= {
912 .send_attr
= vdc_send_attr
,
913 .handle_attr
= vdc_handle_attr
,
914 .handshake_complete
= vdc_handshake_complete
,
917 static void print_version(void)
919 static int version_printed
;
921 if (version_printed
++ == 0)
922 printk(KERN_INFO
"%s", version
);
925 struct vdc_check_port_data
{
930 static int vdc_device_probed(struct device
*dev
, void *arg
)
932 struct vio_dev
*vdev
= to_vio_dev(dev
);
933 struct vdc_check_port_data
*port_data
;
935 port_data
= (struct vdc_check_port_data
*)arg
;
937 if ((vdev
->dev_no
== port_data
->dev_no
) &&
938 (!(strcmp((char *)&vdev
->type
, port_data
->type
))) &&
939 dev_get_drvdata(dev
)) {
940 /* This device has already been configured
941 * by vdc_port_probe()
949 /* Determine whether the VIO device is part of an mpgroup
950 * by locating all the virtual-device-port nodes associated
951 * with the parent virtual-device node for the VIO device
952 * and checking whether any of these nodes are vdc-ports
953 * which have already been configured.
955 * Returns true if this device is part of an mpgroup and has
956 * already been probed.
958 static bool vdc_port_mpgroup_check(struct vio_dev
*vdev
)
960 struct vdc_check_port_data port_data
;
963 port_data
.dev_no
= vdev
->dev_no
;
964 port_data
.type
= (char *)&vdev
->type
;
966 dev
= device_find_child(vdev
->dev
.parent
, &port_data
,
975 static int vdc_port_probe(struct vio_dev
*vdev
, const struct vio_device_id
*id
)
977 struct mdesc_handle
*hp
;
978 struct vdc_port
*port
;
980 const u64
*ldc_timeout
;
987 if ((vdev
->dev_no
<< PARTITION_SHIFT
) & ~(u64
)MINORMASK
) {
988 printk(KERN_ERR PFX
"Port id [%llu] too large.\n",
990 goto err_out_release_mdesc
;
993 /* Check if this device is part of an mpgroup */
994 if (vdc_port_mpgroup_check(vdev
)) {
996 "VIO: Ignoring extra vdisk port %s",
997 dev_name(&vdev
->dev
));
998 goto err_out_release_mdesc
;
1001 port
= kzalloc(sizeof(*port
), GFP_KERNEL
);
1004 printk(KERN_ERR PFX
"Cannot allocate vdc_port.\n");
1005 goto err_out_release_mdesc
;
1008 if (vdev
->dev_no
>= 26)
1009 snprintf(port
->disk_name
, sizeof(port
->disk_name
),
1011 'a' + ((int)vdev
->dev_no
/ 26) - 1,
1012 'a' + ((int)vdev
->dev_no
% 26));
1014 snprintf(port
->disk_name
, sizeof(port
->disk_name
),
1015 VDCBLK_NAME
"%c", 'a' + ((int)vdev
->dev_no
% 26));
1016 port
->vdisk_size
= -1;
1018 /* Actual wall time may be double due to do_generic_file_read() doing
1019 * a readahead I/O first, and once that fails it will try to read a
1022 ldc_timeout
= mdesc_get_property(hp
, vdev
->mp
, "vdc-timeout", NULL
);
1023 port
->ldc_timeout
= ldc_timeout
? *ldc_timeout
: 0;
1024 INIT_DELAYED_WORK(&port
->ldc_reset_timer_work
, vdc_ldc_reset_timer_work
);
1025 INIT_WORK(&port
->ldc_reset_work
, vdc_ldc_reset_work
);
1027 err
= vio_driver_init(&port
->vio
, vdev
, VDEV_DISK
,
1028 vdc_versions
, ARRAY_SIZE(vdc_versions
),
1029 &vdc_vio_ops
, port
->disk_name
);
1031 goto err_out_free_port
;
1033 port
->vdisk_block_size
= VDC_DEFAULT_BLK_SIZE
;
1034 port
->max_xfer_size
= MAX_XFER_SIZE
;
1035 port
->ring_cookies
= MAX_RING_COOKIES
;
1037 err
= vio_ldc_alloc(&port
->vio
, &vdc_ldc_cfg
, port
);
1039 goto err_out_free_port
;
1041 err
= vdc_alloc_tx_ring(port
);
1043 goto err_out_free_ldc
;
1045 err
= probe_disk(port
);
1047 goto err_out_free_tx_ring
;
1049 /* Note that the device driver_data is used to determine
1050 * whether the port has been probed.
1052 dev_set_drvdata(&vdev
->dev
, port
);
1058 err_out_free_tx_ring
:
1059 vdc_free_tx_ring(port
);
1062 vio_ldc_free(&port
->vio
);
1067 err_out_release_mdesc
:
1072 static int vdc_port_remove(struct vio_dev
*vdev
)
1074 struct vdc_port
*port
= dev_get_drvdata(&vdev
->dev
);
1077 blk_mq_stop_hw_queues(port
->disk
->queue
);
1079 flush_work(&port
->ldc_reset_work
);
1080 cancel_delayed_work_sync(&port
->ldc_reset_timer_work
);
1081 del_timer_sync(&port
->vio
.timer
);
1083 del_gendisk(port
->disk
);
1084 cleanup_queue(port
->disk
->queue
);
1085 put_disk(port
->disk
);
1088 vdc_free_tx_ring(port
);
1089 vio_ldc_free(&port
->vio
);
1091 dev_set_drvdata(&vdev
->dev
, NULL
);
1098 static void vdc_requeue_inflight(struct vdc_port
*port
)
1100 struct vio_dring_state
*dr
= &port
->vio
.drings
[VIO_DRIVER_TX_RING
];
1103 for (idx
= dr
->cons
; idx
!= dr
->prod
; idx
= vio_dring_next(dr
, idx
)) {
1104 struct vio_disk_desc
*desc
= vio_dring_entry(dr
, idx
);
1105 struct vdc_req_entry
*rqe
= &port
->rq_arr
[idx
];
1106 struct request
*req
;
1108 ldc_unmap(port
->vio
.lp
, desc
->cookies
, desc
->ncookies
);
1109 desc
->hdr
.state
= VIO_DESC_FREE
;
1110 dr
->cons
= vio_dring_next(dr
, idx
);
1114 vdc_end_special(port
, desc
);
1119 blk_mq_requeue_request(req
, false);
1123 static void vdc_queue_drain(struct vdc_port
*port
)
1125 struct request_queue
*q
= port
->disk
->queue
;
1128 * Mark the queue as draining, then freeze/quiesce to ensure
1129 * that all existing requests are seen in ->queue_rq() and killed
1132 spin_unlock_irq(&port
->vio
.lock
);
1134 blk_mq_freeze_queue(q
);
1135 blk_mq_quiesce_queue(q
);
1137 spin_lock_irq(&port
->vio
.lock
);
1139 blk_mq_unquiesce_queue(q
);
1140 blk_mq_unfreeze_queue(q
);
1143 static void vdc_ldc_reset_timer_work(struct work_struct
*work
)
1145 struct vdc_port
*port
;
1146 struct vio_driver_state
*vio
;
1148 port
= container_of(work
, struct vdc_port
, ldc_reset_timer_work
.work
);
1151 spin_lock_irq(&vio
->lock
);
1152 if (!(port
->vio
.hs_state
& VIO_HS_COMPLETE
)) {
1153 pr_warn(PFX
"%s ldc down %llu seconds, draining queue\n",
1154 port
->disk_name
, port
->ldc_timeout
);
1155 vdc_queue_drain(port
);
1156 vdc_blk_queue_start(port
);
1158 spin_unlock_irq(&vio
->lock
);
1161 static void vdc_ldc_reset_work(struct work_struct
*work
)
1163 struct vdc_port
*port
;
1164 struct vio_driver_state
*vio
;
1165 unsigned long flags
;
1167 port
= container_of(work
, struct vdc_port
, ldc_reset_work
);
1170 spin_lock_irqsave(&vio
->lock
, flags
);
1171 vdc_ldc_reset(port
);
1172 spin_unlock_irqrestore(&vio
->lock
, flags
);
1175 static void vdc_ldc_reset(struct vdc_port
*port
)
1179 assert_spin_locked(&port
->vio
.lock
);
1181 pr_warn(PFX
"%s ldc link reset\n", port
->disk_name
);
1182 blk_mq_stop_hw_queues(port
->disk
->queue
);
1183 vdc_requeue_inflight(port
);
1184 vdc_port_down(port
);
1186 err
= vio_ldc_alloc(&port
->vio
, &vdc_ldc_cfg
, port
);
1188 pr_err(PFX
"%s vio_ldc_alloc:%d\n", port
->disk_name
, err
);
1192 err
= vdc_alloc_tx_ring(port
);
1194 pr_err(PFX
"%s vio_alloc_tx_ring:%d\n", port
->disk_name
, err
);
1198 if (port
->ldc_timeout
)
1199 mod_delayed_work(system_wq
, &port
->ldc_reset_timer_work
,
1200 round_jiffies(jiffies
+ HZ
* port
->ldc_timeout
));
1201 mod_timer(&port
->vio
.timer
, round_jiffies(jiffies
+ HZ
));
1205 vio_ldc_free(&port
->vio
);
1208 static const struct vio_device_id vdc_port_match
[] = {
1214 MODULE_DEVICE_TABLE(vio
, vdc_port_match
);
1216 static struct vio_driver vdc_port_driver
= {
1217 .id_table
= vdc_port_match
,
1218 .probe
= vdc_port_probe
,
1219 .remove
= vdc_port_remove
,
1223 static int __init
vdc_init(void)
1227 sunvdc_wq
= alloc_workqueue("sunvdc", 0, 0);
1231 err
= register_blkdev(0, VDCBLK_NAME
);
1237 err
= vio_register_driver(&vdc_port_driver
);
1239 goto out_unregister_blkdev
;
1243 out_unregister_blkdev
:
1244 unregister_blkdev(vdc_major
, VDCBLK_NAME
);
1248 destroy_workqueue(sunvdc_wq
);
1252 static void __exit
vdc_exit(void)
1254 vio_unregister_driver(&vdc_port_driver
);
1255 unregister_blkdev(vdc_major
, VDCBLK_NAME
);
1256 destroy_workqueue(sunvdc_wq
);
1259 module_init(vdc_init
);
1260 module_exit(vdc_exit
);