1/* -*- linux-c -*-
2 * viodasd.c
3 *  Authors: Dave Boutcher <boutcher@us.ibm.com>
4 *           Ryan Arnold <ryanarn@us.ibm.com>
5 *           Colin Devilbiss <devilbis@us.ibm.com>
6 *           Stephen Rothwell <sfr@au1.ibm.com>
7 *
8 * (C) Copyright 2000-2004 IBM Corporation
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License as
12 * published by the Free Software Foundation; either version 2 of the
13 * License, or (at your option) any later version.
14 *
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
18 * GNU General Public License for more details.
19 *
20 * You should have received a copy of the GNU General Public License
21 * along with this program; if not, write to the Free Software
22 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 *
24 * This routine provides access to disk space (termed "DASD" in historical
25 * IBM terms) owned and managed by an OS/400 partition running on the
26 * same box as this Linux partition.
27 *
28 * All disk operations are performed by sending messages back and forth to
29 * the OS/400 partition.
30 */
31#include <linux/major.h>
32#include <linux/fs.h>
33#include <linux/module.h>
34#include <linux/kernel.h>
35#include <linux/blkdev.h>
36#include <linux/genhd.h>
37#include <linux/hdreg.h>
38#include <linux/errno.h>
39#include <linux/init.h>
40#include <linux/string.h>
41#include <linux/dma-mapping.h>
42#include <linux/completion.h>
43#include <linux/device.h>
44#include <linux/kernel.h>
45
46#include <asm/uaccess.h>
47#include <asm/vio.h>
48#include <asm/iseries/hv_types.h>
49#include <asm/iseries/hv_lp_event.h>
50#include <asm/iseries/hv_lp_config.h>
51#include <asm/iseries/vio.h>
52#include <asm/firmware.h>
53
54MODULE_DESCRIPTION("iSeries Virtual DASD");
55MODULE_AUTHOR("Dave Boutcher");
56MODULE_LICENSE("GPL");
57
58/*
59 * We only support 7 partitions per physical disk....so with minor
60 * numbers 0-255 we get a maximum of 32 disks.
61 */
62#define VIOD_GENHD_NAME		"iseries/vd"
63
64#define VIOD_VERS		"1.64"
65
66#define VIOD_KERN_WARNING	KERN_WARNING "viod: "
67#define VIOD_KERN_INFO		KERN_INFO "viod: "
68
69enum {
70	PARTITION_SHIFT = 3,
71	MAX_DISKNO = HVMAXARCHITECTEDVIRTUALDISKS,
72	MAX_DISK_NAME = sizeof(((struct gendisk *)0)->disk_name)
73};
74
75static DEFINE_SPINLOCK(viodasd_spinlock);
76
77#define VIOMAXREQ		16
78#define VIOMAXBLOCKDMA		12
79
80#define DEVICE_NO(cell)	((struct viodasd_device *)(cell) - &viodasd_devices[0])
81
82struct open_data {
83	u64	disk_size;
84	u16	max_disk;
85	u16	cylinders;
86	u16	tracks;
87	u16	sectors;
88	u16	bytes_per_sector;
89};
90
91struct rw_data {
92	u64	offset;
93	struct {
94		u32	token;
95		u32	reserved;
96		u64	len;
97	} dma_info[VIOMAXBLOCKDMA];
98};
99
100struct vioblocklpevent {
101	struct HvLpEvent	event;
102	u32			reserved;
103	u16			version;
104	u16			sub_result;
105	u16			disk;
106	u16			flags;
107	union {
108		struct open_data	open_data;
109		struct rw_data		rw_data;
110		u64			changed;
111	} u;
112};
113
114#define vioblockflags_ro   0x0001
115
116enum vioblocksubtype {
117	vioblockopen = 0x0001,
118	vioblockclose = 0x0002,
119	vioblockread = 0x0003,
120	vioblockwrite = 0x0004,
121	vioblockflush = 0x0005,
122	vioblockcheck = 0x0007
123};
124
125struct viodasd_waitevent {
126	struct completion	com;
127	int			rc;
128	u16			sub_result;
129	int			max_disk;	/* open */
130};
131
132static const struct vio_error_entry viodasd_err_table[] = {
133	{ 0x0201, EINVAL, "Invalid Range" },
134	{ 0x0202, EINVAL, "Invalid Token" },
135	{ 0x0203, EIO, "DMA Error" },
136	{ 0x0204, EIO, "Use Error" },
137	{ 0x0205, EIO, "Release Error" },
138	{ 0x0206, EINVAL, "Invalid Disk" },
139	{ 0x0207, EBUSY, "Cant Lock" },
140	{ 0x0208, EIO, "Already Locked" },
141	{ 0x0209, EIO, "Already Unlocked" },
142	{ 0x020A, EIO, "Invalid Arg" },
143	{ 0x020B, EIO, "Bad IFS File" },
144	{ 0x020C, EROFS, "Read Only Device" },
145	{ 0x02FF, EIO, "Internal Error" },
146	{ 0x0000, 0, NULL },
147};
148
149/*
150 * Figure out the biggest I/O request (in sectors) we can accept
151 */
152#define VIODASD_MAXSECTORS (4096 / 512 * VIOMAXBLOCKDMA)
153
154/*
155 * Number of disk I/O requests we've sent to OS/400
156 */
157static int num_req_outstanding;
158
159/*
160 * This is our internal structure for keeping track of disk devices
161 */
162struct viodasd_device {
163	u16		cylinders;
164	u16		tracks;
165	u16		sectors;
166	u16		bytes_per_sector;
167	u64		size;
168	int		read_only;
169	spinlock_t	q_lock;
170	struct gendisk	*disk;
171	struct device	*dev;
172} viodasd_devices[MAX_DISKNO];
173
174/*
175 * External open entry point.
176 */
177static int viodasd_open(struct inode *ino, struct file *fil)
178{
179	struct viodasd_device *d = ino->i_bdev->bd_disk->private_data;
180	HvLpEvent_Rc hvrc;
181	struct viodasd_waitevent we;
182	u16 flags = 0;
183
184	if (d->read_only) {
185		if ((fil != NULL) && (fil->f_mode & FMODE_WRITE))
186			return -EROFS;
187		flags = vioblockflags_ro;
188	}
189
190	init_completion(&we.com);
191
192	/* Send the open event to OS/400 */
193	hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
194			HvLpEvent_Type_VirtualIo,
195			viomajorsubtype_blockio | vioblockopen,
196			HvLpEvent_AckInd_DoAck, HvLpEvent_AckType_ImmediateAck,
197			viopath_sourceinst(viopath_hostLp),
198			viopath_targetinst(viopath_hostLp),
199			(u64)(unsigned long)&we, VIOVERSION << 16,
200			((u64)DEVICE_NO(d) << 48) | ((u64)flags << 32),
201			0, 0, 0);
202	if (hvrc != 0) {
203		printk(VIOD_KERN_WARNING "HV open failed %d\n", (int)hvrc);
204		return -EIO;
205	}
206
207	wait_for_completion(&we.com);
208
209	/* Check the return code */
210	if (we.rc != 0) {
211		const struct vio_error_entry *err =
212			vio_lookup_rc(viodasd_err_table, we.sub_result);
213
214		printk(VIOD_KERN_WARNING
215				"bad rc opening disk: %d:0x%04x (%s)\n",
216				(int)we.rc, we.sub_result, err->msg);
217		return -EIO;
218	}
219
220	return 0;
221}
222
223/*
224 * External release entry point.
225 */
226static int viodasd_release(struct inode *ino, struct file *fil)
227{
228	struct viodasd_device *d = ino->i_bdev->bd_disk->private_data;
229	HvLpEvent_Rc hvrc;
230
231	/* Send the event to OS/400.  We DON'T expect a response */
232	hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
233			HvLpEvent_Type_VirtualIo,
234			viomajorsubtype_blockio | vioblockclose,
235			HvLpEvent_AckInd_NoAck, HvLpEvent_AckType_ImmediateAck,
236			viopath_sourceinst(viopath_hostLp),
237			viopath_targetinst(viopath_hostLp),
238			0, VIOVERSION << 16,
239			((u64)DEVICE_NO(d) << 48) /* | ((u64)flags << 32) */,
240			0, 0, 0);
241	if (hvrc != 0)
242		printk(VIOD_KERN_WARNING "HV close call failed %d\n",
243				(int)hvrc);
244	return 0;
245}
246
247
248/* External ioctl entry point.
249 */
250static int viodasd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
251{
252	struct gendisk *disk = bdev->bd_disk;
253	struct viodasd_device *d = disk->private_data;
254
255	geo->sectors = d->sectors ? d->sectors : 0;
256	geo->heads = d->tracks ? d->tracks  : 64;
257	geo->cylinders = d->cylinders ? d->cylinders :
258		get_capacity(disk) / (geo->cylinders * geo->heads);
259
260	return 0;
261}
262
263/*
264 * Our file operations table
265 */
266static struct block_device_operations viodasd_fops = {
267	.owner = THIS_MODULE,
268	.open = viodasd_open,
269	.release = viodasd_release,
270	.getgeo = viodasd_getgeo,
271};
272
273/*
274 * End a request
275 */
276static void viodasd_end_request(struct request *req, int uptodate,
277		int num_sectors)
278{
279	if (end_that_request_first(req, uptodate, num_sectors))
280		return;
281	add_disk_randomness(req->rq_disk);
282	end_that_request_last(req, uptodate);
283}
284
285/*
286 * Send an actual I/O request to OS/400
287 */
288static int send_request(struct request *req)
289{
290	u64 start;
291	int direction;
292	int nsg;
293	u16 viocmd;
294	HvLpEvent_Rc hvrc;
295	struct vioblocklpevent *bevent;
296	struct HvLpEvent *hev;
297	struct scatterlist sg[VIOMAXBLOCKDMA];
298	int sgindex;
299	int statindex;
300	struct viodasd_device *d;
301	unsigned long flags;
302
303	start = (u64)req->sector << 9;
304
305	if (rq_data_dir(req) == READ) {
306		direction = DMA_FROM_DEVICE;
307		viocmd = viomajorsubtype_blockio | vioblockread;
308		statindex = 0;
309	} else {
310		direction = DMA_TO_DEVICE;
311		viocmd = viomajorsubtype_blockio | vioblockwrite;
312		statindex = 1;
313	}
314
315        d = req->rq_disk->private_data;
316
317	/* Now build the scatter-gather list */
318	nsg = blk_rq_map_sg(req->q, req, sg);
319	nsg = dma_map_sg(d->dev, sg, nsg, direction);
320
321	spin_lock_irqsave(&viodasd_spinlock, flags);
322	num_req_outstanding++;
323
324	/* This optimization handles a single DMA block */
325	if (nsg == 1)
326		hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
327				HvLpEvent_Type_VirtualIo, viocmd,
328				HvLpEvent_AckInd_DoAck,
329				HvLpEvent_AckType_ImmediateAck,
330				viopath_sourceinst(viopath_hostLp),
331				viopath_targetinst(viopath_hostLp),
332				(u64)(unsigned long)req, VIOVERSION << 16,
333				((u64)DEVICE_NO(d) << 48), start,
334				((u64)sg_dma_address(&sg[0])) << 32,
335				sg_dma_len(&sg[0]));
336	else {
337		bevent = (struct vioblocklpevent *)
338			vio_get_event_buffer(viomajorsubtype_blockio);
339		if (bevent == NULL) {
340			printk(VIOD_KERN_WARNING
341			       "error allocating disk event buffer\n");
342			goto error_ret;
343		}
344
345		/*
346		 * Now build up the actual request.  Note that we store
347		 * the pointer to the request in the correlation
348		 * token so we can match the response up later
349		 */
350		memset(bevent, 0, sizeof(struct vioblocklpevent));
351		hev = &bevent->event;
352		hev->flags = HV_LP_EVENT_VALID | HV_LP_EVENT_DO_ACK |
353			HV_LP_EVENT_INT;
354		hev->xType = HvLpEvent_Type_VirtualIo;
355		hev->xSubtype = viocmd;
356		hev->xSourceLp = HvLpConfig_getLpIndex();
357		hev->xTargetLp = viopath_hostLp;
358		hev->xSizeMinus1 =
359			offsetof(struct vioblocklpevent, u.rw_data.dma_info) +
360			(sizeof(bevent->u.rw_data.dma_info[0]) * nsg) - 1;
361		hev->xSourceInstanceId = viopath_sourceinst(viopath_hostLp);
362		hev->xTargetInstanceId = viopath_targetinst(viopath_hostLp);
363		hev->xCorrelationToken = (u64)req;
364		bevent->version = VIOVERSION;
365		bevent->disk = DEVICE_NO(d);
366		bevent->u.rw_data.offset = start;
367
368		/*
369		 * Copy just the dma information from the sg list
370		 * into the request
371		 */
372		for (sgindex = 0; sgindex < nsg; sgindex++) {
373			bevent->u.rw_data.dma_info[sgindex].token =
374				sg_dma_address(&sg[sgindex]);
375			bevent->u.rw_data.dma_info[sgindex].len =
376				sg_dma_len(&sg[sgindex]);
377		}
378
379		/* Send the request */
380		hvrc = HvCallEvent_signalLpEvent(&bevent->event);
381		vio_free_event_buffer(viomajorsubtype_blockio, bevent);
382	}
383
384	if (hvrc != HvLpEvent_Rc_Good) {
385		printk(VIOD_KERN_WARNING
386		       "error sending disk event to OS/400 (rc %d)\n",
387		       (int)hvrc);
388		goto error_ret;
389	}
390	spin_unlock_irqrestore(&viodasd_spinlock, flags);
391	return 0;
392
393error_ret:
394	num_req_outstanding--;
395	spin_unlock_irqrestore(&viodasd_spinlock, flags);
396	dma_unmap_sg(d->dev, sg, nsg, direction);
397	return -1;
398}
399
400/*
401 * This is the external request processing routine
402 */
403static void do_viodasd_request(request_queue_t *q)
404{
405	struct request *req;
406
407	/*
408	 * If we already have the maximum number of requests
409	 * outstanding to OS/400 just bail out. We'll come
410	 * back later.
411	 */
412	while (num_req_outstanding < VIOMAXREQ) {
413		req = elv_next_request(q);
414		if (req == NULL)
415			return;
416		/* dequeue the current request from the queue */
417		blkdev_dequeue_request(req);
418		/* check that request contains a valid command */
419		if (!blk_fs_request(req)) {
420			viodasd_end_request(req, 0, req->hard_nr_sectors);
421			continue;
422		}
423		/* Try sending the request */
424		if (send_request(req) != 0)
425			viodasd_end_request(req, 0, req->hard_nr_sectors);
426	}
427}
428
429/*
430 * Probe a single disk and fill in the viodasd_device structure
431 * for it.
432 */
433static void probe_disk(struct viodasd_device *d)
434{
435	HvLpEvent_Rc hvrc;
436	struct viodasd_waitevent we;
437	int dev_no = DEVICE_NO(d);
438	struct gendisk *g;
439	struct request_queue *q;
440	u16 flags = 0;
441
442retry:
443	init_completion(&we.com);
444
445	/* Send the open event to OS/400 */
446	hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
447			HvLpEvent_Type_VirtualIo,
448			viomajorsubtype_blockio | vioblockopen,
449			HvLpEvent_AckInd_DoAck, HvLpEvent_AckType_ImmediateAck,
450			viopath_sourceinst(viopath_hostLp),
451			viopath_targetinst(viopath_hostLp),
452			(u64)(unsigned long)&we, VIOVERSION << 16,
453			((u64)dev_no << 48) | ((u64)flags<< 32),
454			0, 0, 0);
455	if (hvrc != 0) {
456		printk(VIOD_KERN_WARNING "bad rc on HV open %d\n", (int)hvrc);
457		return;
458	}
459
460	wait_for_completion(&we.com);
461
462	if (we.rc != 0) {
463		if (flags != 0)
464			return;
465		/* try again with read only flag set */
466		flags = vioblockflags_ro;
467		goto retry;
468	}
469	if (we.max_disk > (MAX_DISKNO - 1)) {
470		static int warned;
471
472		if (warned == 0) {
473			warned++;
474			printk(VIOD_KERN_INFO
475				"Only examining the first %d "
476				"of %d disks connected\n",
477				MAX_DISKNO, we.max_disk + 1);
478		}
479	}
480
481	/* Send the close event to OS/400.  We DON'T expect a response */
482	hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
483			HvLpEvent_Type_VirtualIo,
484			viomajorsubtype_blockio | vioblockclose,
485			HvLpEvent_AckInd_NoAck, HvLpEvent_AckType_ImmediateAck,
486			viopath_sourceinst(viopath_hostLp),
487			viopath_targetinst(viopath_hostLp),
488			0, VIOVERSION << 16,
489			((u64)dev_no << 48) | ((u64)flags << 32),
490			0, 0, 0);
491	if (hvrc != 0) {
492		printk(VIOD_KERN_WARNING
493		       "bad rc sending event to OS/400 %d\n", (int)hvrc);
494		return;
495	}
496	/* create the request queue for the disk */
497	spin_lock_init(&d->q_lock);
498	q = blk_init_queue(do_viodasd_request, &d->q_lock);
499	if (q == NULL) {
500		printk(VIOD_KERN_WARNING "cannot allocate queue for disk %d\n",
501				dev_no);
502		return;
503	}
504	g = alloc_disk(1 << PARTITION_SHIFT);
505	if (g == NULL) {
506		printk(VIOD_KERN_WARNING
507				"cannot allocate disk structure for disk %d\n",
508				dev_no);
509		blk_cleanup_queue(q);
510		return;
511	}
512
513	d->disk = g;
514	blk_queue_max_hw_segments(q, VIOMAXBLOCKDMA);
515	blk_queue_max_phys_segments(q, VIOMAXBLOCKDMA);
516	blk_queue_max_sectors(q, VIODASD_MAXSECTORS);
517	g->major = VIODASD_MAJOR;
518	g->first_minor = dev_no << PARTITION_SHIFT;
519	if (dev_no >= 26)
520		snprintf(g->disk_name, sizeof(g->disk_name),
521				VIOD_GENHD_NAME "%c%c",
522				'a' + (dev_no / 26) - 1, 'a' + (dev_no % 26));
523	else
524		snprintf(g->disk_name, sizeof(g->disk_name),
525				VIOD_GENHD_NAME "%c", 'a' + (dev_no % 26));
526	g->fops = &viodasd_fops;
527	g->queue = q;
528	g->private_data = d;
529	g->driverfs_dev = d->dev;
530	set_capacity(g, d->size >> 9);
531
532	printk(VIOD_KERN_INFO "disk %d: %lu sectors (%lu MB) "
533			"CHS=%d/%d/%d sector size %d%s\n",
534			dev_no, (unsigned long)(d->size >> 9),
535			(unsigned long)(d->size >> 20),
536			(int)d->cylinders, (int)d->tracks,
537			(int)d->sectors, (int)d->bytes_per_sector,
538			d->read_only ? " (RO)" : "");
539
540	/* register us in the global list */
541	add_disk(g);
542}
543
544/* returns the total number of scatterlist elements converted */
545static int block_event_to_scatterlist(const struct vioblocklpevent *bevent,
546		struct scatterlist *sg, int *total_len)
547{
548	int i, numsg;
549	const struct rw_data *rw_data = &bevent->u.rw_data;
550	static const int offset =
551		offsetof(struct vioblocklpevent, u.rw_data.dma_info);
552	static const int element_size = sizeof(rw_data->dma_info[0]);
553
554	numsg = ((bevent->event.xSizeMinus1 + 1) - offset) / element_size;
555	if (numsg > VIOMAXBLOCKDMA)
556		numsg = VIOMAXBLOCKDMA;
557
558	*total_len = 0;
559	memset(sg, 0, sizeof(sg[0]) * VIOMAXBLOCKDMA);
560
561	for (i = 0; (i < numsg) && (rw_data->dma_info[i].len > 0); ++i) {
562		sg_dma_address(&sg[i]) = rw_data->dma_info[i].token;
563		sg_dma_len(&sg[i]) = rw_data->dma_info[i].len;
564		*total_len += rw_data->dma_info[i].len;
565	}
566	return i;
567}
568
569/*
570 * Restart all queues, starting with the one _after_ the disk given,
571 * thus reducing the chance of starvation of higher numbered disks.
572 */
573static void viodasd_restart_all_queues_starting_from(int first_index)
574{
575	int i;
576
577	for (i = first_index + 1; i < MAX_DISKNO; ++i)
578		if (viodasd_devices[i].disk)
579			blk_run_queue(viodasd_devices[i].disk->queue);
580	for (i = 0; i <= first_index; ++i)
581		if (viodasd_devices[i].disk)
582			blk_run_queue(viodasd_devices[i].disk->queue);
583}
584
585/*
586 * For read and write requests, decrement the number of outstanding requests,
587 * Free the DMA buffers we allocated.
588 */
589static int viodasd_handle_read_write(struct vioblocklpevent *bevent)
590{
591	int num_sg, num_sect, pci_direction, total_len;
592	struct request *req;
593	struct scatterlist sg[VIOMAXBLOCKDMA];
594	struct HvLpEvent *event = &bevent->event;
595	unsigned long irq_flags;
596	struct viodasd_device *d;
597	int error;
598	spinlock_t *qlock;
599
600	num_sg = block_event_to_scatterlist(bevent, sg, &total_len);
601	num_sect = total_len >> 9;
602	if (event->xSubtype == (viomajorsubtype_blockio | vioblockread))
603		pci_direction = DMA_FROM_DEVICE;
604	else
605		pci_direction = DMA_TO_DEVICE;
606	req = (struct request *)bevent->event.xCorrelationToken;
607	d = req->rq_disk->private_data;
608
609	dma_unmap_sg(d->dev, sg, num_sg, pci_direction);
610
611	/*
612	 * Since this is running in interrupt mode, we need to make sure
613	 * we're not stepping on any global I/O operations
614	 */
615	spin_lock_irqsave(&viodasd_spinlock, irq_flags);
616	num_req_outstanding--;
617	spin_unlock_irqrestore(&viodasd_spinlock, irq_flags);
618
619	error = event->xRc != HvLpEvent_Rc_Good;
620	if (error) {
621		const struct vio_error_entry *err;
622		err = vio_lookup_rc(viodasd_err_table, bevent->sub_result);
623		printk(VIOD_KERN_WARNING "read/write error %d:0x%04x (%s)\n",
624				event->xRc, bevent->sub_result, err->msg);
625		num_sect = req->hard_nr_sectors;
626	}
627	qlock = req->q->queue_lock;
628	spin_lock_irqsave(qlock, irq_flags);
629	viodasd_end_request(req, !error, num_sect);
630	spin_unlock_irqrestore(qlock, irq_flags);
631
632	/* Finally, try to get more requests off of this device's queue */
633	viodasd_restart_all_queues_starting_from(DEVICE_NO(d));
634
635	return 0;
636}
637
638/* This routine handles incoming block LP events */
639static void handle_block_event(struct HvLpEvent *event)
640{
641	struct vioblocklpevent *bevent = (struct vioblocklpevent *)event;
642	struct viodasd_waitevent *pwe;
643
644	if (event == NULL)
645		/* Notification that a partition went away! */
646		return;
647	/* First, we should NEVER get an int here...only acks */
648	if (hvlpevent_is_int(event)) {
649		printk(VIOD_KERN_WARNING
650		       "Yikes! got an int in viodasd event handler!\n");
651		if (hvlpevent_need_ack(event)) {
652			event->xRc = HvLpEvent_Rc_InvalidSubtype;
653			HvCallEvent_ackLpEvent(event);
654		}
655	}
656
657	switch (event->xSubtype & VIOMINOR_SUBTYPE_MASK) {
658	case vioblockopen:
659		/*
660		 * Handle a response to an open request.  We get all the
661		 * disk information in the response, so update it.  The
662		 * correlation token contains a pointer to a waitevent
663		 * structure that has a completion in it.  update the
664		 * return code in the waitevent structure and post the
665		 * completion to wake up the guy who sent the request
666		 */
667		pwe = (struct viodasd_waitevent *)event->xCorrelationToken;
668		pwe->rc = event->xRc;
669		pwe->sub_result = bevent->sub_result;
670		if (event->xRc == HvLpEvent_Rc_Good) {
671			const struct open_data *data = &bevent->u.open_data;
672			struct viodasd_device *device =
673				&viodasd_devices[bevent->disk];
674			device->read_only =
675				bevent->flags & vioblockflags_ro;
676			device->size = data->disk_size;
677			device->cylinders = data->cylinders;
678			device->tracks = data->tracks;
679			device->sectors = data->sectors;
680			device->bytes_per_sector = data->bytes_per_sector;
681			pwe->max_disk = data->max_disk;
682		}
683		complete(&pwe->com);
684		break;
685	case vioblockclose:
686		break;
687	case vioblockread:
688	case vioblockwrite:
689		viodasd_handle_read_write(bevent);
690		break;
691
692	default:
693		printk(VIOD_KERN_WARNING "invalid subtype!");
694		if (hvlpevent_need_ack(event)) {
695			event->xRc = HvLpEvent_Rc_InvalidSubtype;
696			HvCallEvent_ackLpEvent(event);
697		}
698	}
699}
700
701/*
702 * Get the driver to reprobe for more disks.
703 */
704static ssize_t probe_disks(struct device_driver *drv, const char *buf,
705		size_t count)
706{
707	struct viodasd_device *d;
708
709	for (d = viodasd_devices; d < &viodasd_devices[MAX_DISKNO]; d++) {
710		if (d->disk == NULL)
711			probe_disk(d);
712	}
713	return count;
714}
715static DRIVER_ATTR(probe, S_IWUSR, NULL, probe_disks);
716
717static int viodasd_probe(struct vio_dev *vdev, const struct vio_device_id *id)
718{
719	struct viodasd_device *d = &viodasd_devices[vdev->unit_address];
720
721	d->dev = &vdev->dev;
722	probe_disk(d);
723	if (d->disk == NULL)
724		return -ENODEV;
725	return 0;
726}
727
728static int viodasd_remove(struct vio_dev *vdev)
729{
730	struct viodasd_device *d;
731
732	d = &viodasd_devices[vdev->unit_address];
733	if (d->disk) {
734		del_gendisk(d->disk);
735		blk_cleanup_queue(d->disk->queue);
736		put_disk(d->disk);
737		d->disk = NULL;
738	}
739	d->dev = NULL;
740	return 0;
741}
742
743/**
744 * viodasd_device_table: Used by vio.c to match devices that we
745 * support.
746 */
747static struct vio_device_id viodasd_device_table[] __devinitdata = {
748	{ "block", "IBM,iSeries-viodasd" },
749	{ "", "" }
750};
751MODULE_DEVICE_TABLE(vio, viodasd_device_table);
752
753static struct vio_driver viodasd_driver = {
754	.id_table = viodasd_device_table,
755	.probe = viodasd_probe,
756	.remove = viodasd_remove,
757	.driver = {
758		.name = "viodasd",
759		.owner = THIS_MODULE,
760	}
761};
762
763static int need_delete_probe;
764
765/*
766 * Initialize the whole device driver.  Handle module and non-module
767 * versions
768 */
769static int __init viodasd_init(void)
770{
771	int rc;
772
773	if (!firmware_has_feature(FW_FEATURE_ISERIES)) {
774		rc = -ENODEV;
775		goto early_fail;
776	}
777
778	/* Try to open to our host lp */
779	if (viopath_hostLp == HvLpIndexInvalid)
780		vio_set_hostlp();
781
782	if (viopath_hostLp == HvLpIndexInvalid) {
783		printk(VIOD_KERN_WARNING "invalid hosting partition\n");
784		rc = -EIO;
785		goto early_fail;
786	}
787
788	printk(VIOD_KERN_INFO "vers " VIOD_VERS ", hosting partition %d\n",
789			viopath_hostLp);
790
791        /* register the block device */
792	rc =  register_blkdev(VIODASD_MAJOR, VIOD_GENHD_NAME);
793	if (rc) {
794		printk(VIOD_KERN_WARNING
795				"Unable to get major number %d for %s\n",
796				VIODASD_MAJOR, VIOD_GENHD_NAME);
797		goto early_fail;
798	}
799	/* Actually open the path to the hosting partition */
800	rc = viopath_open(viopath_hostLp, viomajorsubtype_blockio,
801				VIOMAXREQ + 2);
802	if (rc) {
803		printk(VIOD_KERN_WARNING
804		       "error opening path to host partition %d\n",
805		       viopath_hostLp);
806		goto unregister_blk;
807	}
808
809	/* Initialize our request handler */
810	vio_setHandler(viomajorsubtype_blockio, handle_block_event);
811
812	rc = vio_register_driver(&viodasd_driver);
813	if (rc) {
814		printk(VIOD_KERN_WARNING "vio_register_driver failed\n");
815		goto unset_handler;
816	}
817
818	/*
819	 * If this call fails, it just means that we cannot dynamically
820	 * add virtual disks, but the driver will still work fine for
821	 * all existing disk, so ignore the failure.
822	 */
823	if (!driver_create_file(&viodasd_driver.driver, &driver_attr_probe))
824		need_delete_probe = 1;
825
826	return 0;
827
828unset_handler:
829	vio_clearHandler(viomajorsubtype_blockio);
830	viopath_close(viopath_hostLp, viomajorsubtype_blockio, VIOMAXREQ + 2);
831unregister_blk:
832	unregister_blkdev(VIODASD_MAJOR, VIOD_GENHD_NAME);
833early_fail:
834	return rc;
835}
836module_init(viodasd_init);
837
838void __exit viodasd_exit(void)
839{
840	if (need_delete_probe)
841		driver_remove_file(&viodasd_driver.driver, &driver_attr_probe);
842	vio_unregister_driver(&viodasd_driver);
843	vio_clearHandler(viomajorsubtype_blockio);
844	viopath_close(viopath_hostLp, viomajorsubtype_blockio, VIOMAXREQ + 2);
845	unregister_blkdev(VIODASD_MAJOR, VIOD_GENHD_NAME);
846}
847module_exit(viodasd_exit);
848