stop_machine: wean Xen off stop_machine_run
[linux-2.6] / drivers / block / viodasd.c
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
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/scatterlist.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
54 MODULE_DESCRIPTION("iSeries Virtual DASD");
55 MODULE_AUTHOR("Dave Boutcher");
56 MODULE_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
69 enum {
70         PARTITION_SHIFT = 3,
71         MAX_DISKNO = HVMAXARCHITECTEDVIRTUALDISKS,
72         MAX_DISK_NAME = FIELD_SIZEOF(struct gendisk, disk_name)
73 };
74
75 static DEFINE_SPINLOCK(viodasd_spinlock);
76
77 #define VIOMAXREQ               16
78
79 #define DEVICE_NO(cell) ((struct viodasd_device *)(cell) - &viodasd_devices[0])
80
81 struct viodasd_waitevent {
82         struct completion       com;
83         int                     rc;
84         u16                     sub_result;
85         int                     max_disk;       /* open */
86 };
87
88 static const struct vio_error_entry viodasd_err_table[] = {
89         { 0x0201, EINVAL, "Invalid Range" },
90         { 0x0202, EINVAL, "Invalid Token" },
91         { 0x0203, EIO, "DMA Error" },
92         { 0x0204, EIO, "Use Error" },
93         { 0x0205, EIO, "Release Error" },
94         { 0x0206, EINVAL, "Invalid Disk" },
95         { 0x0207, EBUSY, "Cant Lock" },
96         { 0x0208, EIO, "Already Locked" },
97         { 0x0209, EIO, "Already Unlocked" },
98         { 0x020A, EIO, "Invalid Arg" },
99         { 0x020B, EIO, "Bad IFS File" },
100         { 0x020C, EROFS, "Read Only Device" },
101         { 0x02FF, EIO, "Internal Error" },
102         { 0x0000, 0, NULL },
103 };
104
105 /*
106  * Figure out the biggest I/O request (in sectors) we can accept
107  */
108 #define VIODASD_MAXSECTORS (4096 / 512 * VIOMAXBLOCKDMA)
109
110 /*
111  * Number of disk I/O requests we've sent to OS/400
112  */
113 static int num_req_outstanding;
114
115 /*
116  * This is our internal structure for keeping track of disk devices
117  */
118 struct viodasd_device {
119         u16             cylinders;
120         u16             tracks;
121         u16             sectors;
122         u16             bytes_per_sector;
123         u64             size;
124         int             read_only;
125         spinlock_t      q_lock;
126         struct gendisk  *disk;
127         struct device   *dev;
128 } viodasd_devices[MAX_DISKNO];
129
130 /*
131  * External open entry point.
132  */
133 static int viodasd_open(struct inode *ino, struct file *fil)
134 {
135         struct viodasd_device *d = ino->i_bdev->bd_disk->private_data;
136         HvLpEvent_Rc hvrc;
137         struct viodasd_waitevent we;
138         u16 flags = 0;
139
140         if (d->read_only) {
141                 if ((fil != NULL) && (fil->f_mode & FMODE_WRITE))
142                         return -EROFS;
143                 flags = vioblockflags_ro;
144         }
145
146         init_completion(&we.com);
147
148         /* Send the open event to OS/400 */
149         hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
150                         HvLpEvent_Type_VirtualIo,
151                         viomajorsubtype_blockio | vioblockopen,
152                         HvLpEvent_AckInd_DoAck, HvLpEvent_AckType_ImmediateAck,
153                         viopath_sourceinst(viopath_hostLp),
154                         viopath_targetinst(viopath_hostLp),
155                         (u64)(unsigned long)&we, VIOVERSION << 16,
156                         ((u64)DEVICE_NO(d) << 48) | ((u64)flags << 32),
157                         0, 0, 0);
158         if (hvrc != 0) {
159                 printk(VIOD_KERN_WARNING "HV open failed %d\n", (int)hvrc);
160                 return -EIO;
161         }
162
163         wait_for_completion(&we.com);
164
165         /* Check the return code */
166         if (we.rc != 0) {
167                 const struct vio_error_entry *err =
168                         vio_lookup_rc(viodasd_err_table, we.sub_result);
169
170                 printk(VIOD_KERN_WARNING
171                                 "bad rc opening disk: %d:0x%04x (%s)\n",
172                                 (int)we.rc, we.sub_result, err->msg);
173                 return -EIO;
174         }
175
176         return 0;
177 }
178
179 /*
180  * External release entry point.
181  */
182 static int viodasd_release(struct inode *ino, struct file *fil)
183 {
184         struct viodasd_device *d = ino->i_bdev->bd_disk->private_data;
185         HvLpEvent_Rc hvrc;
186
187         /* Send the event to OS/400.  We DON'T expect a response */
188         hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
189                         HvLpEvent_Type_VirtualIo,
190                         viomajorsubtype_blockio | vioblockclose,
191                         HvLpEvent_AckInd_NoAck, HvLpEvent_AckType_ImmediateAck,
192                         viopath_sourceinst(viopath_hostLp),
193                         viopath_targetinst(viopath_hostLp),
194                         0, VIOVERSION << 16,
195                         ((u64)DEVICE_NO(d) << 48) /* | ((u64)flags << 32) */,
196                         0, 0, 0);
197         if (hvrc != 0)
198                 printk(VIOD_KERN_WARNING "HV close call failed %d\n",
199                                 (int)hvrc);
200         return 0;
201 }
202
203
204 /* External ioctl entry point.
205  */
206 static int viodasd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
207 {
208         struct gendisk *disk = bdev->bd_disk;
209         struct viodasd_device *d = disk->private_data;
210
211         geo->sectors = d->sectors ? d->sectors : 32;
212         geo->heads = d->tracks ? d->tracks  : 64;
213         geo->cylinders = d->cylinders ? d->cylinders :
214                 get_capacity(disk) / (geo->sectors * geo->heads);
215
216         return 0;
217 }
218
219 /*
220  * Our file operations table
221  */
222 static struct block_device_operations viodasd_fops = {
223         .owner = THIS_MODULE,
224         .open = viodasd_open,
225         .release = viodasd_release,
226         .getgeo = viodasd_getgeo,
227 };
228
229 /*
230  * End a request
231  */
232 static void viodasd_end_request(struct request *req, int error,
233                 int num_sectors)
234 {
235         __blk_end_request(req, error, num_sectors << 9);
236 }
237
238 /*
239  * Send an actual I/O request to OS/400
240  */
241 static int send_request(struct request *req)
242 {
243         u64 start;
244         int direction;
245         int nsg;
246         u16 viocmd;
247         HvLpEvent_Rc hvrc;
248         struct vioblocklpevent *bevent;
249         struct HvLpEvent *hev;
250         struct scatterlist sg[VIOMAXBLOCKDMA];
251         int sgindex;
252         int statindex;
253         struct viodasd_device *d;
254         unsigned long flags;
255
256         start = (u64)req->sector << 9;
257
258         if (rq_data_dir(req) == READ) {
259                 direction = DMA_FROM_DEVICE;
260                 viocmd = viomajorsubtype_blockio | vioblockread;
261                 statindex = 0;
262         } else {
263                 direction = DMA_TO_DEVICE;
264                 viocmd = viomajorsubtype_blockio | vioblockwrite;
265                 statindex = 1;
266         }
267
268         d = req->rq_disk->private_data;
269
270         /* Now build the scatter-gather list */
271         sg_init_table(sg, VIOMAXBLOCKDMA);
272         nsg = blk_rq_map_sg(req->q, req, sg);
273         nsg = dma_map_sg(d->dev, sg, nsg, direction);
274
275         spin_lock_irqsave(&viodasd_spinlock, flags);
276         num_req_outstanding++;
277
278         /* This optimization handles a single DMA block */
279         if (nsg == 1)
280                 hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
281                                 HvLpEvent_Type_VirtualIo, viocmd,
282                                 HvLpEvent_AckInd_DoAck,
283                                 HvLpEvent_AckType_ImmediateAck,
284                                 viopath_sourceinst(viopath_hostLp),
285                                 viopath_targetinst(viopath_hostLp),
286                                 (u64)(unsigned long)req, VIOVERSION << 16,
287                                 ((u64)DEVICE_NO(d) << 48), start,
288                                 ((u64)sg_dma_address(&sg[0])) << 32,
289                                 sg_dma_len(&sg[0]));
290         else {
291                 bevent = (struct vioblocklpevent *)
292                         vio_get_event_buffer(viomajorsubtype_blockio);
293                 if (bevent == NULL) {
294                         printk(VIOD_KERN_WARNING
295                                "error allocating disk event buffer\n");
296                         goto error_ret;
297                 }
298
299                 /*
300                  * Now build up the actual request.  Note that we store
301                  * the pointer to the request in the correlation
302                  * token so we can match the response up later
303                  */
304                 memset(bevent, 0, sizeof(struct vioblocklpevent));
305                 hev = &bevent->event;
306                 hev->flags = HV_LP_EVENT_VALID | HV_LP_EVENT_DO_ACK |
307                         HV_LP_EVENT_INT;
308                 hev->xType = HvLpEvent_Type_VirtualIo;
309                 hev->xSubtype = viocmd;
310                 hev->xSourceLp = HvLpConfig_getLpIndex();
311                 hev->xTargetLp = viopath_hostLp;
312                 hev->xSizeMinus1 =
313                         offsetof(struct vioblocklpevent, u.rw_data.dma_info) +
314                         (sizeof(bevent->u.rw_data.dma_info[0]) * nsg) - 1;
315                 hev->xSourceInstanceId = viopath_sourceinst(viopath_hostLp);
316                 hev->xTargetInstanceId = viopath_targetinst(viopath_hostLp);
317                 hev->xCorrelationToken = (u64)req;
318                 bevent->version = VIOVERSION;
319                 bevent->disk = DEVICE_NO(d);
320                 bevent->u.rw_data.offset = start;
321
322                 /*
323                  * Copy just the dma information from the sg list
324                  * into the request
325                  */
326                 for (sgindex = 0; sgindex < nsg; sgindex++) {
327                         bevent->u.rw_data.dma_info[sgindex].token =
328                                 sg_dma_address(&sg[sgindex]);
329                         bevent->u.rw_data.dma_info[sgindex].len =
330                                 sg_dma_len(&sg[sgindex]);
331                 }
332
333                 /* Send the request */
334                 hvrc = HvCallEvent_signalLpEvent(&bevent->event);
335                 vio_free_event_buffer(viomajorsubtype_blockio, bevent);
336         }
337
338         if (hvrc != HvLpEvent_Rc_Good) {
339                 printk(VIOD_KERN_WARNING
340                        "error sending disk event to OS/400 (rc %d)\n",
341                        (int)hvrc);
342                 goto error_ret;
343         }
344         spin_unlock_irqrestore(&viodasd_spinlock, flags);
345         return 0;
346
347 error_ret:
348         num_req_outstanding--;
349         spin_unlock_irqrestore(&viodasd_spinlock, flags);
350         dma_unmap_sg(d->dev, sg, nsg, direction);
351         return -1;
352 }
353
354 /*
355  * This is the external request processing routine
356  */
357 static void do_viodasd_request(struct request_queue *q)
358 {
359         struct request *req;
360
361         /*
362          * If we already have the maximum number of requests
363          * outstanding to OS/400 just bail out. We'll come
364          * back later.
365          */
366         while (num_req_outstanding < VIOMAXREQ) {
367                 req = elv_next_request(q);
368                 if (req == NULL)
369                         return;
370                 /* dequeue the current request from the queue */
371                 blkdev_dequeue_request(req);
372                 /* check that request contains a valid command */
373                 if (!blk_fs_request(req)) {
374                         viodasd_end_request(req, -EIO, req->hard_nr_sectors);
375                         continue;
376                 }
377                 /* Try sending the request */
378                 if (send_request(req) != 0)
379                         viodasd_end_request(req, -EIO, req->hard_nr_sectors);
380         }
381 }
382
383 /*
384  * Probe a single disk and fill in the viodasd_device structure
385  * for it.
386  */
387 static int probe_disk(struct viodasd_device *d)
388 {
389         HvLpEvent_Rc hvrc;
390         struct viodasd_waitevent we;
391         int dev_no = DEVICE_NO(d);
392         struct gendisk *g;
393         struct request_queue *q;
394         u16 flags = 0;
395
396 retry:
397         init_completion(&we.com);
398
399         /* Send the open event to OS/400 */
400         hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
401                         HvLpEvent_Type_VirtualIo,
402                         viomajorsubtype_blockio | vioblockopen,
403                         HvLpEvent_AckInd_DoAck, HvLpEvent_AckType_ImmediateAck,
404                         viopath_sourceinst(viopath_hostLp),
405                         viopath_targetinst(viopath_hostLp),
406                         (u64)(unsigned long)&we, VIOVERSION << 16,
407                         ((u64)dev_no << 48) | ((u64)flags<< 32),
408                         0, 0, 0);
409         if (hvrc != 0) {
410                 printk(VIOD_KERN_WARNING "bad rc on HV open %d\n", (int)hvrc);
411                 return 0;
412         }
413
414         wait_for_completion(&we.com);
415
416         if (we.rc != 0) {
417                 if (flags != 0)
418                         return 0;
419                 /* try again with read only flag set */
420                 flags = vioblockflags_ro;
421                 goto retry;
422         }
423         if (we.max_disk > (MAX_DISKNO - 1)) {
424                 static int warned;
425
426                 if (warned == 0) {
427                         warned++;
428                         printk(VIOD_KERN_INFO
429                                 "Only examining the first %d "
430                                 "of %d disks connected\n",
431                                 MAX_DISKNO, we.max_disk + 1);
432                 }
433         }
434
435         /* Send the close event to OS/400.  We DON'T expect a response */
436         hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
437                         HvLpEvent_Type_VirtualIo,
438                         viomajorsubtype_blockio | vioblockclose,
439                         HvLpEvent_AckInd_NoAck, HvLpEvent_AckType_ImmediateAck,
440                         viopath_sourceinst(viopath_hostLp),
441                         viopath_targetinst(viopath_hostLp),
442                         0, VIOVERSION << 16,
443                         ((u64)dev_no << 48) | ((u64)flags << 32),
444                         0, 0, 0);
445         if (hvrc != 0) {
446                 printk(VIOD_KERN_WARNING
447                        "bad rc sending event to OS/400 %d\n", (int)hvrc);
448                 return 0;
449         }
450
451         if (d->dev == NULL) {
452                 /* this is when we reprobe for new disks */
453                 if (vio_create_viodasd(dev_no) == NULL) {
454                         printk(VIOD_KERN_WARNING
455                                 "cannot allocate virtual device for disk %d\n",
456                                 dev_no);
457                         return 0;
458                 }
459                 /*
460                  * The vio_create_viodasd will have recursed into this
461                  * routine with d->dev set to the new vio device and
462                  * will finish the setup of the disk below.
463                  */
464                 return 1;
465         }
466
467         /* create the request queue for the disk */
468         spin_lock_init(&d->q_lock);
469         q = blk_init_queue(do_viodasd_request, &d->q_lock);
470         if (q == NULL) {
471                 printk(VIOD_KERN_WARNING "cannot allocate queue for disk %d\n",
472                                 dev_no);
473                 return 0;
474         }
475         g = alloc_disk(1 << PARTITION_SHIFT);
476         if (g == NULL) {
477                 printk(VIOD_KERN_WARNING
478                                 "cannot allocate disk structure for disk %d\n",
479                                 dev_no);
480                 blk_cleanup_queue(q);
481                 return 0;
482         }
483
484         d->disk = g;
485         blk_queue_max_hw_segments(q, VIOMAXBLOCKDMA);
486         blk_queue_max_phys_segments(q, VIOMAXBLOCKDMA);
487         blk_queue_max_sectors(q, VIODASD_MAXSECTORS);
488         g->major = VIODASD_MAJOR;
489         g->first_minor = dev_no << PARTITION_SHIFT;
490         if (dev_no >= 26)
491                 snprintf(g->disk_name, sizeof(g->disk_name),
492                                 VIOD_GENHD_NAME "%c%c",
493                                 'a' + (dev_no / 26) - 1, 'a' + (dev_no % 26));
494         else
495                 snprintf(g->disk_name, sizeof(g->disk_name),
496                                 VIOD_GENHD_NAME "%c", 'a' + (dev_no % 26));
497         g->fops = &viodasd_fops;
498         g->queue = q;
499         g->private_data = d;
500         g->driverfs_dev = d->dev;
501         set_capacity(g, d->size >> 9);
502
503         printk(VIOD_KERN_INFO "disk %d: %lu sectors (%lu MB) "
504                         "CHS=%d/%d/%d sector size %d%s\n",
505                         dev_no, (unsigned long)(d->size >> 9),
506                         (unsigned long)(d->size >> 20),
507                         (int)d->cylinders, (int)d->tracks,
508                         (int)d->sectors, (int)d->bytes_per_sector,
509                         d->read_only ? " (RO)" : "");
510
511         /* register us in the global list */
512         add_disk(g);
513         return 1;
514 }
515
516 /* returns the total number of scatterlist elements converted */
517 static int block_event_to_scatterlist(const struct vioblocklpevent *bevent,
518                 struct scatterlist *sg, int *total_len)
519 {
520         int i, numsg;
521         const struct rw_data *rw_data = &bevent->u.rw_data;
522         static const int offset =
523                 offsetof(struct vioblocklpevent, u.rw_data.dma_info);
524         static const int element_size = sizeof(rw_data->dma_info[0]);
525
526         numsg = ((bevent->event.xSizeMinus1 + 1) - offset) / element_size;
527         if (numsg > VIOMAXBLOCKDMA)
528                 numsg = VIOMAXBLOCKDMA;
529
530         *total_len = 0;
531         sg_init_table(sg, VIOMAXBLOCKDMA);
532         for (i = 0; (i < numsg) && (rw_data->dma_info[i].len > 0); ++i) {
533                 sg_dma_address(&sg[i]) = rw_data->dma_info[i].token;
534                 sg_dma_len(&sg[i]) = rw_data->dma_info[i].len;
535                 *total_len += rw_data->dma_info[i].len;
536         }
537         return i;
538 }
539
540 /*
541  * Restart all queues, starting with the one _after_ the disk given,
542  * thus reducing the chance of starvation of higher numbered disks.
543  */
544 static void viodasd_restart_all_queues_starting_from(int first_index)
545 {
546         int i;
547
548         for (i = first_index + 1; i < MAX_DISKNO; ++i)
549                 if (viodasd_devices[i].disk)
550                         blk_run_queue(viodasd_devices[i].disk->queue);
551         for (i = 0; i <= first_index; ++i)
552                 if (viodasd_devices[i].disk)
553                         blk_run_queue(viodasd_devices[i].disk->queue);
554 }
555
556 /*
557  * For read and write requests, decrement the number of outstanding requests,
558  * Free the DMA buffers we allocated.
559  */
560 static int viodasd_handle_read_write(struct vioblocklpevent *bevent)
561 {
562         int num_sg, num_sect, pci_direction, total_len;
563         struct request *req;
564         struct scatterlist sg[VIOMAXBLOCKDMA];
565         struct HvLpEvent *event = &bevent->event;
566         unsigned long irq_flags;
567         struct viodasd_device *d;
568         int error;
569         spinlock_t *qlock;
570
571         num_sg = block_event_to_scatterlist(bevent, sg, &total_len);
572         num_sect = total_len >> 9;
573         if (event->xSubtype == (viomajorsubtype_blockio | vioblockread))
574                 pci_direction = DMA_FROM_DEVICE;
575         else
576                 pci_direction = DMA_TO_DEVICE;
577         req = (struct request *)bevent->event.xCorrelationToken;
578         d = req->rq_disk->private_data;
579
580         dma_unmap_sg(d->dev, sg, num_sg, pci_direction);
581
582         /*
583          * Since this is running in interrupt mode, we need to make sure
584          * we're not stepping on any global I/O operations
585          */
586         spin_lock_irqsave(&viodasd_spinlock, irq_flags);
587         num_req_outstanding--;
588         spin_unlock_irqrestore(&viodasd_spinlock, irq_flags);
589
590         error = (event->xRc == HvLpEvent_Rc_Good) ? 0 : -EIO;
591         if (error) {
592                 const struct vio_error_entry *err;
593                 err = vio_lookup_rc(viodasd_err_table, bevent->sub_result);
594                 printk(VIOD_KERN_WARNING "read/write error %d:0x%04x (%s)\n",
595                                 event->xRc, bevent->sub_result, err->msg);
596                 num_sect = req->hard_nr_sectors;
597         }
598         qlock = req->q->queue_lock;
599         spin_lock_irqsave(qlock, irq_flags);
600         viodasd_end_request(req, error, num_sect);
601         spin_unlock_irqrestore(qlock, irq_flags);
602
603         /* Finally, try to get more requests off of this device's queue */
604         viodasd_restart_all_queues_starting_from(DEVICE_NO(d));
605
606         return 0;
607 }
608
609 /* This routine handles incoming block LP events */
610 static void handle_block_event(struct HvLpEvent *event)
611 {
612         struct vioblocklpevent *bevent = (struct vioblocklpevent *)event;
613         struct viodasd_waitevent *pwe;
614
615         if (event == NULL)
616                 /* Notification that a partition went away! */
617                 return;
618         /* First, we should NEVER get an int here...only acks */
619         if (hvlpevent_is_int(event)) {
620                 printk(VIOD_KERN_WARNING
621                        "Yikes! got an int in viodasd event handler!\n");
622                 if (hvlpevent_need_ack(event)) {
623                         event->xRc = HvLpEvent_Rc_InvalidSubtype;
624                         HvCallEvent_ackLpEvent(event);
625                 }
626         }
627
628         switch (event->xSubtype & VIOMINOR_SUBTYPE_MASK) {
629         case vioblockopen:
630                 /*
631                  * Handle a response to an open request.  We get all the
632                  * disk information in the response, so update it.  The
633                  * correlation token contains a pointer to a waitevent
634                  * structure that has a completion in it.  update the
635                  * return code in the waitevent structure and post the
636                  * completion to wake up the guy who sent the request
637                  */
638                 pwe = (struct viodasd_waitevent *)event->xCorrelationToken;
639                 pwe->rc = event->xRc;
640                 pwe->sub_result = bevent->sub_result;
641                 if (event->xRc == HvLpEvent_Rc_Good) {
642                         const struct open_data *data = &bevent->u.open_data;
643                         struct viodasd_device *device =
644                                 &viodasd_devices[bevent->disk];
645                         device->read_only =
646                                 bevent->flags & vioblockflags_ro;
647                         device->size = data->disk_size;
648                         device->cylinders = data->cylinders;
649                         device->tracks = data->tracks;
650                         device->sectors = data->sectors;
651                         device->bytes_per_sector = data->bytes_per_sector;
652                         pwe->max_disk = data->max_disk;
653                 }
654                 complete(&pwe->com);
655                 break;
656         case vioblockclose:
657                 break;
658         case vioblockread:
659         case vioblockwrite:
660                 viodasd_handle_read_write(bevent);
661                 break;
662
663         default:
664                 printk(VIOD_KERN_WARNING "invalid subtype!");
665                 if (hvlpevent_need_ack(event)) {
666                         event->xRc = HvLpEvent_Rc_InvalidSubtype;
667                         HvCallEvent_ackLpEvent(event);
668                 }
669         }
670 }
671
672 /*
673  * Get the driver to reprobe for more disks.
674  */
675 static ssize_t probe_disks(struct device_driver *drv, const char *buf,
676                 size_t count)
677 {
678         struct viodasd_device *d;
679
680         for (d = viodasd_devices; d < &viodasd_devices[MAX_DISKNO]; d++) {
681                 if (d->disk == NULL)
682                         probe_disk(d);
683         }
684         return count;
685 }
686 static DRIVER_ATTR(probe, S_IWUSR, NULL, probe_disks);
687
688 static int viodasd_probe(struct vio_dev *vdev, const struct vio_device_id *id)
689 {
690         struct viodasd_device *d = &viodasd_devices[vdev->unit_address];
691
692         d->dev = &vdev->dev;
693         if (!probe_disk(d))
694                 return -ENODEV;
695         return 0;
696 }
697
698 static int viodasd_remove(struct vio_dev *vdev)
699 {
700         struct viodasd_device *d;
701
702         d = &viodasd_devices[vdev->unit_address];
703         if (d->disk) {
704                 del_gendisk(d->disk);
705                 blk_cleanup_queue(d->disk->queue);
706                 put_disk(d->disk);
707                 d->disk = NULL;
708         }
709         d->dev = NULL;
710         return 0;
711 }
712
713 /**
714  * viodasd_device_table: Used by vio.c to match devices that we
715  * support.
716  */
717 static struct vio_device_id viodasd_device_table[] __devinitdata = {
718         { "block", "IBM,iSeries-viodasd" },
719         { "", "" }
720 };
721 MODULE_DEVICE_TABLE(vio, viodasd_device_table);
722
723 static struct vio_driver viodasd_driver = {
724         .id_table = viodasd_device_table,
725         .probe = viodasd_probe,
726         .remove = viodasd_remove,
727         .driver = {
728                 .name = "viodasd",
729                 .owner = THIS_MODULE,
730         }
731 };
732
733 static int need_delete_probe;
734
735 /*
736  * Initialize the whole device driver.  Handle module and non-module
737  * versions
738  */
739 static int __init viodasd_init(void)
740 {
741         int rc;
742
743         if (!firmware_has_feature(FW_FEATURE_ISERIES)) {
744                 rc = -ENODEV;
745                 goto early_fail;
746         }
747
748         /* Try to open to our host lp */
749         if (viopath_hostLp == HvLpIndexInvalid)
750                 vio_set_hostlp();
751
752         if (viopath_hostLp == HvLpIndexInvalid) {
753                 printk(VIOD_KERN_WARNING "invalid hosting partition\n");
754                 rc = -EIO;
755                 goto early_fail;
756         }
757
758         printk(VIOD_KERN_INFO "vers " VIOD_VERS ", hosting partition %d\n",
759                         viopath_hostLp);
760
761         /* register the block device */
762         rc =  register_blkdev(VIODASD_MAJOR, VIOD_GENHD_NAME);
763         if (rc) {
764                 printk(VIOD_KERN_WARNING
765                                 "Unable to get major number %d for %s\n",
766                                 VIODASD_MAJOR, VIOD_GENHD_NAME);
767                 goto early_fail;
768         }
769         /* Actually open the path to the hosting partition */
770         rc = viopath_open(viopath_hostLp, viomajorsubtype_blockio,
771                                 VIOMAXREQ + 2);
772         if (rc) {
773                 printk(VIOD_KERN_WARNING
774                        "error opening path to host partition %d\n",
775                        viopath_hostLp);
776                 goto unregister_blk;
777         }
778
779         /* Initialize our request handler */
780         vio_setHandler(viomajorsubtype_blockio, handle_block_event);
781
782         rc = vio_register_driver(&viodasd_driver);
783         if (rc) {
784                 printk(VIOD_KERN_WARNING "vio_register_driver failed\n");
785                 goto unset_handler;
786         }
787
788         /*
789          * If this call fails, it just means that we cannot dynamically
790          * add virtual disks, but the driver will still work fine for
791          * all existing disk, so ignore the failure.
792          */
793         if (!driver_create_file(&viodasd_driver.driver, &driver_attr_probe))
794                 need_delete_probe = 1;
795
796         return 0;
797
798 unset_handler:
799         vio_clearHandler(viomajorsubtype_blockio);
800         viopath_close(viopath_hostLp, viomajorsubtype_blockio, VIOMAXREQ + 2);
801 unregister_blk:
802         unregister_blkdev(VIODASD_MAJOR, VIOD_GENHD_NAME);
803 early_fail:
804         return rc;
805 }
806 module_init(viodasd_init);
807
808 void __exit viodasd_exit(void)
809 {
810         if (need_delete_probe)
811                 driver_remove_file(&viodasd_driver.driver, &driver_attr_probe);
812         vio_unregister_driver(&viodasd_driver);
813         vio_clearHandler(viomajorsubtype_blockio);
814         viopath_close(viopath_hostLp, viomajorsubtype_blockio, VIOMAXREQ + 2);
815         unregister_blkdev(VIODASD_MAJOR, VIOD_GENHD_NAME);
816 }
817 module_exit(viodasd_exit);