Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/kaber/nf-next-2.6
[pandora-kernel.git] / drivers / staging / hv / blkvsc_drv.c
1 /*
2  * Copyright (c) 2009, Microsoft Corporation.
3  *
4  * This program is free software; you can redistribute it and/or modify it
5  * under the terms and conditions of the GNU General Public License,
6  * version 2, as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope it will be useful, but WITHOUT
9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
11  * more details.
12  *
13  * You should have received a copy of the GNU General Public License along with
14  * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
15  * Place - Suite 330, Boston, MA 02111-1307 USA.
16  *
17  * Authors:
18  *   Haiyang Zhang <haiyangz@microsoft.com>
19  *   Hank Janssen  <hjanssen@microsoft.com>
20  */
21 #include <linux/init.h>
22 #include <linux/module.h>
23 #include <linux/device.h>
24 #include <linux/blkdev.h>
25 #include <linux/major.h>
26 #include <linux/delay.h>
27 #include <linux/hdreg.h>
28 #include <linux/mutex.h>
29 #include <linux/slab.h>
30 #include <scsi/scsi.h>
31 #include <scsi/scsi_cmnd.h>
32 #include <scsi/scsi_eh.h>
33 #include <scsi/scsi_dbg.h>
34 #include "osd.h"
35 #include "logging.h"
36 #include "version_info.h"
37 #include "vmbus.h"
38 #include "storvsc_api.h"
39
40
41 #define BLKVSC_MINORS   64
42
43 enum blkvsc_device_type {
44         UNKNOWN_DEV_TYPE,
45         HARDDISK_TYPE,
46         DVD_TYPE,
47 };
48
49 /*
50  * This request ties the struct request and struct
51  * blkvsc_request/hv_storvsc_request together A struct request may be
52  * represented by 1 or more struct blkvsc_request
53  */
54 struct blkvsc_request_group {
55         int outstanding;
56         int status;
57         struct list_head blkvsc_req_list;       /* list of blkvsc_requests */
58 };
59
60 struct blkvsc_request {
61         /* blkvsc_request_group.blkvsc_req_list */
62         struct list_head req_entry;
63
64         /* block_device_context.pending_list */
65         struct list_head pend_entry;
66
67         /* This may be null if we generate a request internally */
68         struct request *req;
69
70         struct block_device_context *dev;
71
72         /* The group this request is part of. Maybe null */
73         struct blkvsc_request_group *group;
74
75         wait_queue_head_t wevent;
76         int cond;
77
78         int write;
79         sector_t sector_start;
80         unsigned long sector_count;
81
82         unsigned char sense_buffer[SCSI_SENSE_BUFFERSIZE];
83         unsigned char cmd_len;
84         unsigned char cmnd[MAX_COMMAND_SIZE];
85
86         struct hv_storvsc_request request;
87         /*
88          * !!!DO NOT ADD ANYTHING BELOW HERE!!! Otherwise, memory can overlap,
89          * because - The extension buffer falls right here and is pointed to by
90          * request.Extension;
91          * Which sounds like a horrible idea, who designed this?
92          */
93 };
94
95 /* Per device structure */
96 struct block_device_context {
97         /* point back to our device context */
98         struct vm_device *device_ctx;
99         struct kmem_cache *request_pool;
100         spinlock_t lock;
101         struct gendisk *gd;
102         enum blkvsc_device_type device_type;
103         struct list_head pending_list;
104
105         unsigned char device_id[64];
106         unsigned int device_id_len;
107         int num_outstanding_reqs;
108         int shutting_down;
109         int media_not_present;
110         unsigned int sector_size;
111         sector_t capacity;
112         unsigned int port;
113         unsigned char path;
114         unsigned char target;
115         int users;
116 };
117
118 /* Per driver */
119 struct blkvsc_driver_context {
120         /* !! These must be the first 2 fields !! */
121         /* FIXME this is a bug! */
122         struct driver_context drv_ctx;
123         struct storvsc_driver_object drv_obj;
124 };
125
126 /* Static decl */
127 static DEFINE_MUTEX(blkvsc_mutex);
128 static int blkvsc_probe(struct device *dev);
129 static int blkvsc_remove(struct device *device);
130 static void blkvsc_shutdown(struct device *device);
131
132 static int blkvsc_open(struct block_device *bdev,  fmode_t mode);
133 static int blkvsc_release(struct gendisk *disk, fmode_t mode);
134 static int blkvsc_media_changed(struct gendisk *gd);
135 static int blkvsc_revalidate_disk(struct gendisk *gd);
136 static int blkvsc_getgeo(struct block_device *bd, struct hd_geometry *hg);
137 static int blkvsc_ioctl(struct block_device *bd, fmode_t mode,
138                         unsigned cmd, unsigned long argument);
139 static void blkvsc_request(struct request_queue *queue);
140 static void blkvsc_request_completion(struct hv_storvsc_request *request);
141 static int blkvsc_do_request(struct block_device_context *blkdev,
142                              struct request *req);
143 static int blkvsc_submit_request(struct blkvsc_request *blkvsc_req,
144                 void (*request_completion)(struct hv_storvsc_request *));
145 static void blkvsc_init_rw(struct blkvsc_request *blkvsc_req);
146 static void blkvsc_cmd_completion(struct hv_storvsc_request *request);
147 static int blkvsc_do_inquiry(struct block_device_context *blkdev);
148 static int blkvsc_do_read_capacity(struct block_device_context *blkdev);
149 static int blkvsc_do_read_capacity16(struct block_device_context *blkdev);
150 static int blkvsc_do_flush(struct block_device_context *blkdev);
151 static int blkvsc_cancel_pending_reqs(struct block_device_context *blkdev);
152 static int blkvsc_do_pending_reqs(struct block_device_context *blkdev);
153
154 static int blkvsc_ringbuffer_size = BLKVSC_RING_BUFFER_SIZE;
155 module_param(blkvsc_ringbuffer_size, int, S_IRUGO);
156 MODULE_PARM_DESC(ring_size, "Ring buffer size (in bytes)");
157
158 /* The one and only one */
159 static struct blkvsc_driver_context g_blkvsc_drv;
160
161 static const struct block_device_operations block_ops = {
162         .owner = THIS_MODULE,
163         .open = blkvsc_open,
164         .release = blkvsc_release,
165         .media_changed = blkvsc_media_changed,
166         .revalidate_disk = blkvsc_revalidate_disk,
167         .getgeo = blkvsc_getgeo,
168         .ioctl  = blkvsc_ioctl,
169 };
170
171 /*
172  * blkvsc_drv_init -  BlkVsc driver initialization.
173  */
174 static int blkvsc_drv_init(int (*drv_init)(struct hv_driver *drv))
175 {
176         struct storvsc_driver_object *storvsc_drv_obj = &g_blkvsc_drv.drv_obj;
177         struct driver_context *drv_ctx = &g_blkvsc_drv.drv_ctx;
178         int ret;
179
180         storvsc_drv_obj->ring_buffer_size = blkvsc_ringbuffer_size;
181
182         /* Callback to client driver to complete the initialization */
183         drv_init(&storvsc_drv_obj->base);
184
185         drv_ctx->driver.name = storvsc_drv_obj->base.name;
186         memcpy(&drv_ctx->class_id, &storvsc_drv_obj->base.deviceType,
187                sizeof(struct hv_guid));
188
189         drv_ctx->probe = blkvsc_probe;
190         drv_ctx->remove = blkvsc_remove;
191         drv_ctx->shutdown = blkvsc_shutdown;
192
193         /* The driver belongs to vmbus */
194         ret = vmbus_child_driver_register(drv_ctx);
195
196         return ret;
197 }
198
199 static int blkvsc_drv_exit_cb(struct device *dev, void *data)
200 {
201         struct device **curr = (struct device **)data;
202         *curr = dev;
203         return 1; /* stop iterating */
204 }
205
206 static void blkvsc_drv_exit(void)
207 {
208         struct storvsc_driver_object *storvsc_drv_obj = &g_blkvsc_drv.drv_obj;
209         struct driver_context *drv_ctx = &g_blkvsc_drv.drv_ctx;
210         struct device *current_dev;
211         int ret;
212
213         while (1) {
214                 current_dev = NULL;
215
216                 /* Get the device */
217                 ret = driver_for_each_device(&drv_ctx->driver, NULL,
218                                              (void *) &current_dev,
219                                              blkvsc_drv_exit_cb);
220
221                 if (ret)
222                         DPRINT_WARN(BLKVSC_DRV,
223                                     "driver_for_each_device returned %d", ret);
224
225
226                 if (current_dev == NULL)
227                         break;
228
229                 /* Initiate removal from the top-down */
230                 device_unregister(current_dev);
231         }
232
233         if (storvsc_drv_obj->base.OnCleanup)
234                 storvsc_drv_obj->base.OnCleanup(&storvsc_drv_obj->base);
235
236         vmbus_child_driver_unregister(drv_ctx);
237
238         return;
239 }
240
241 /*
242  * blkvsc_probe - Add a new device for this driver
243  */
244 static int blkvsc_probe(struct device *device)
245 {
246         struct driver_context *driver_ctx =
247                                 driver_to_driver_context(device->driver);
248         struct blkvsc_driver_context *blkvsc_drv_ctx =
249                                 (struct blkvsc_driver_context *)driver_ctx;
250         struct storvsc_driver_object *storvsc_drv_obj =
251                                 &blkvsc_drv_ctx->drv_obj;
252         struct vm_device *device_ctx = device_to_vm_device(device);
253         struct hv_device *device_obj = &device_ctx->device_obj;
254
255         struct block_device_context *blkdev = NULL;
256         struct storvsc_device_info device_info;
257         int major = 0;
258         int devnum = 0;
259         int ret = 0;
260         static int ide0_registered;
261         static int ide1_registered;
262
263         DPRINT_DBG(BLKVSC_DRV, "blkvsc_probe - enter");
264
265         if (!storvsc_drv_obj->base.OnDeviceAdd) {
266                 DPRINT_ERR(BLKVSC_DRV, "OnDeviceAdd() not set");
267                 ret = -1;
268                 goto Cleanup;
269         }
270
271         blkdev = kzalloc(sizeof(struct block_device_context), GFP_KERNEL);
272         if (!blkdev) {
273                 ret = -ENOMEM;
274                 goto Cleanup;
275         }
276
277         INIT_LIST_HEAD(&blkdev->pending_list);
278
279         /* Initialize what we can here */
280         spin_lock_init(&blkdev->lock);
281
282         /* ASSERT(sizeof(struct blkvsc_request_group) <= */
283         /*      sizeof(struct blkvsc_request)); */
284
285         blkdev->request_pool = kmem_cache_create(dev_name(&device_ctx->device),
286                                         sizeof(struct blkvsc_request) +
287                                         storvsc_drv_obj->request_ext_size, 0,
288                                         SLAB_HWCACHE_ALIGN, NULL);
289         if (!blkdev->request_pool) {
290                 ret = -ENOMEM;
291                 goto Cleanup;
292         }
293
294
295         /* Call to the vsc driver to add the device */
296         ret = storvsc_drv_obj->base.OnDeviceAdd(device_obj, &device_info);
297         if (ret != 0) {
298                 DPRINT_ERR(BLKVSC_DRV, "unable to add blkvsc device");
299                 goto Cleanup;
300         }
301
302         blkdev->device_ctx = device_ctx;
303         /* this identified the device 0 or 1 */
304         blkdev->target = device_info.target_id;
305         /* this identified the ide ctrl 0 or 1 */
306         blkdev->path = device_info.path_id;
307
308         dev_set_drvdata(device, blkdev);
309
310         /* Calculate the major and device num */
311         if (blkdev->path == 0) {
312                 major = IDE0_MAJOR;
313                 devnum = blkdev->path + blkdev->target;         /* 0 or 1 */
314
315                 if (!ide0_registered) {
316                         ret = register_blkdev(major, "ide");
317                         if (ret != 0) {
318                                 DPRINT_ERR(BLKVSC_DRV,
319                                            "register_blkdev() failed! ret %d",
320                                            ret);
321                                 goto Remove;
322                         }
323
324                         ide0_registered = 1;
325                 }
326         } else if (blkdev->path == 1) {
327                 major = IDE1_MAJOR;
328                 devnum = blkdev->path + blkdev->target + 1; /* 2 or 3 */
329
330                 if (!ide1_registered) {
331                         ret = register_blkdev(major, "ide");
332                         if (ret != 0) {
333                                 DPRINT_ERR(BLKVSC_DRV,
334                                            "register_blkdev() failed! ret %d",
335                                            ret);
336                                 goto Remove;
337                         }
338
339                         ide1_registered = 1;
340                 }
341         } else {
342                 DPRINT_ERR(BLKVSC_DRV, "invalid pathid");
343                 ret = -1;
344                 goto Cleanup;
345         }
346
347         DPRINT_INFO(BLKVSC_DRV, "blkvsc registered for major %d!!", major);
348
349         blkdev->gd = alloc_disk(BLKVSC_MINORS);
350         if (!blkdev->gd) {
351                 DPRINT_ERR(BLKVSC_DRV, "register_blkdev() failed! ret %d", ret);
352                 ret = -1;
353                 goto Cleanup;
354         }
355
356         blkdev->gd->queue = blk_init_queue(blkvsc_request, &blkdev->lock);
357
358         blk_queue_max_segment_size(blkdev->gd->queue, PAGE_SIZE);
359         blk_queue_max_segments(blkdev->gd->queue, MAX_MULTIPAGE_BUFFER_COUNT);
360         blk_queue_segment_boundary(blkdev->gd->queue, PAGE_SIZE-1);
361         blk_queue_bounce_limit(blkdev->gd->queue, BLK_BOUNCE_ANY);
362         blk_queue_dma_alignment(blkdev->gd->queue, 511);
363
364         blkdev->gd->major = major;
365         if (devnum == 1 || devnum == 3)
366                 blkdev->gd->first_minor = BLKVSC_MINORS;
367         else
368                 blkdev->gd->first_minor = 0;
369         blkdev->gd->fops = &block_ops;
370         blkdev->gd->private_data = blkdev;
371         blkdev->gd->driverfs_dev = &(blkdev->device_ctx->device);
372         sprintf(blkdev->gd->disk_name, "hd%c", 'a' + devnum);
373
374         blkvsc_do_inquiry(blkdev);
375         if (blkdev->device_type == DVD_TYPE) {
376                 set_disk_ro(blkdev->gd, 1);
377                 blkdev->gd->flags |= GENHD_FL_REMOVABLE;
378                 blkvsc_do_read_capacity(blkdev);
379         } else {
380                 blkvsc_do_read_capacity16(blkdev);
381         }
382
383         set_capacity(blkdev->gd, blkdev->capacity * (blkdev->sector_size/512));
384         blk_queue_logical_block_size(blkdev->gd->queue, blkdev->sector_size);
385         /* go! */
386         add_disk(blkdev->gd);
387
388         DPRINT_INFO(BLKVSC_DRV, "%s added!! capacity %lu sector_size %d",
389                     blkdev->gd->disk_name, (unsigned long)blkdev->capacity,
390                     blkdev->sector_size);
391
392         return ret;
393
394 Remove:
395         storvsc_drv_obj->base.OnDeviceRemove(device_obj);
396
397 Cleanup:
398         if (blkdev) {
399                 if (blkdev->request_pool) {
400                         kmem_cache_destroy(blkdev->request_pool);
401                         blkdev->request_pool = NULL;
402                 }
403                 kfree(blkdev);
404                 blkdev = NULL;
405         }
406
407         return ret;
408 }
409
410 static void blkvsc_shutdown(struct device *device)
411 {
412         struct block_device_context *blkdev = dev_get_drvdata(device);
413         unsigned long flags;
414
415         if (!blkdev)
416                 return;
417
418         DPRINT_DBG(BLKVSC_DRV, "blkvsc_shutdown - users %d disk %s\n",
419                    blkdev->users, blkdev->gd->disk_name);
420
421         spin_lock_irqsave(&blkdev->lock, flags);
422
423         blkdev->shutting_down = 1;
424
425         blk_stop_queue(blkdev->gd->queue);
426
427         spin_unlock_irqrestore(&blkdev->lock, flags);
428
429         while (blkdev->num_outstanding_reqs) {
430                 DPRINT_INFO(STORVSC, "waiting for %d requests to complete...",
431                             blkdev->num_outstanding_reqs);
432                 udelay(100);
433         }
434
435         blkvsc_do_flush(blkdev);
436
437         spin_lock_irqsave(&blkdev->lock, flags);
438
439         blkvsc_cancel_pending_reqs(blkdev);
440
441         spin_unlock_irqrestore(&blkdev->lock, flags);
442 }
443
444 static int blkvsc_do_flush(struct block_device_context *blkdev)
445 {
446         struct blkvsc_request *blkvsc_req;
447
448         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_flush()\n");
449
450         if (blkdev->device_type != HARDDISK_TYPE)
451                 return 0;
452
453         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
454         if (!blkvsc_req)
455                 return -ENOMEM;
456
457         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
458         init_waitqueue_head(&blkvsc_req->wevent);
459         blkvsc_req->dev = blkdev;
460         blkvsc_req->req = NULL;
461         blkvsc_req->write = 0;
462
463         blkvsc_req->request.data_buffer.PfnArray[0] = 0;
464         blkvsc_req->request.data_buffer.Offset = 0;
465         blkvsc_req->request.data_buffer.Length = 0;
466
467         blkvsc_req->cmnd[0] = SYNCHRONIZE_CACHE;
468         blkvsc_req->cmd_len = 10;
469
470         /*
471          * Set this here since the completion routine may be invoked and
472          * completed before we return
473          */
474         blkvsc_req->cond = 0;
475         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
476
477         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
478
479         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
480
481         return 0;
482 }
483
484 /* Do a scsi INQUIRY cmd here to get the device type (ie disk or dvd) */
485 static int blkvsc_do_inquiry(struct block_device_context *blkdev)
486 {
487         struct blkvsc_request *blkvsc_req;
488         struct page *page_buf;
489         unsigned char *buf;
490         unsigned char device_type;
491
492         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_inquiry()\n");
493
494         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
495         if (!blkvsc_req)
496                 return -ENOMEM;
497
498         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
499         page_buf = alloc_page(GFP_KERNEL);
500         if (!page_buf) {
501                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
502                 return -ENOMEM;
503         }
504
505         init_waitqueue_head(&blkvsc_req->wevent);
506         blkvsc_req->dev = blkdev;
507         blkvsc_req->req = NULL;
508         blkvsc_req->write = 0;
509
510         blkvsc_req->request.data_buffer.PfnArray[0] = page_to_pfn(page_buf);
511         blkvsc_req->request.data_buffer.Offset = 0;
512         blkvsc_req->request.data_buffer.Length = 64;
513
514         blkvsc_req->cmnd[0] = INQUIRY;
515         blkvsc_req->cmnd[1] = 0x1;              /* Get product data */
516         blkvsc_req->cmnd[2] = 0x83;             /* mode page 83 */
517         blkvsc_req->cmnd[4] = 64;
518         blkvsc_req->cmd_len = 6;
519
520         /*
521          * Set this here since the completion routine may be invoked and
522          * completed before we return
523          */
524         blkvsc_req->cond = 0;
525
526         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
527
528         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete - cond %d\n",
529                    blkvsc_req, blkvsc_req->cond);
530
531         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
532
533         buf = kmap(page_buf);
534
535         /* print_hex_dump_bytes("", DUMP_PREFIX_NONE, buf, 64); */
536         /* be to le */
537         device_type = buf[0] & 0x1F;
538
539         if (device_type == 0x0) {
540                 blkdev->device_type = HARDDISK_TYPE;
541         } else if (device_type == 0x5) {
542                 blkdev->device_type = DVD_TYPE;
543         } else {
544                 /* TODO: this is currently unsupported device type */
545                 blkdev->device_type = UNKNOWN_DEV_TYPE;
546         }
547
548         DPRINT_DBG(BLKVSC_DRV, "device type %d\n", device_type);
549
550         blkdev->device_id_len = buf[7];
551         if (blkdev->device_id_len > 64)
552                 blkdev->device_id_len = 64;
553
554         memcpy(blkdev->device_id, &buf[8], blkdev->device_id_len);
555         /* printk_hex_dump_bytes("", DUMP_PREFIX_NONE, blkdev->device_id,
556          * blkdev->device_id_len); */
557
558         kunmap(page_buf);
559
560         __free_page(page_buf);
561
562         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
563
564         return 0;
565 }
566
567 /* Do a scsi READ_CAPACITY cmd here to get the size of the disk */
568 static int blkvsc_do_read_capacity(struct block_device_context *blkdev)
569 {
570         struct blkvsc_request *blkvsc_req;
571         struct page *page_buf;
572         unsigned char *buf;
573         struct scsi_sense_hdr sense_hdr;
574
575         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_read_capacity()\n");
576
577         blkdev->sector_size = 0;
578         blkdev->capacity = 0;
579         blkdev->media_not_present = 0; /* assume a disk is present */
580
581         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
582         if (!blkvsc_req)
583                 return -ENOMEM;
584
585         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
586         page_buf = alloc_page(GFP_KERNEL);
587         if (!page_buf) {
588                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
589                 return -ENOMEM;
590         }
591
592         init_waitqueue_head(&blkvsc_req->wevent);
593         blkvsc_req->dev = blkdev;
594         blkvsc_req->req = NULL;
595         blkvsc_req->write = 0;
596
597         blkvsc_req->request.data_buffer.PfnArray[0] = page_to_pfn(page_buf);
598         blkvsc_req->request.data_buffer.Offset = 0;
599         blkvsc_req->request.data_buffer.Length = 8;
600
601         blkvsc_req->cmnd[0] = READ_CAPACITY;
602         blkvsc_req->cmd_len = 16;
603
604         /*
605          * Set this here since the completion routine may be invoked
606          * and completed before we return
607          */
608         blkvsc_req->cond = 0;
609
610         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
611
612         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete - cond %d\n",
613                    blkvsc_req, blkvsc_req->cond);
614
615         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
616
617         /* check error */
618         if (blkvsc_req->request.status) {
619                 scsi_normalize_sense(blkvsc_req->sense_buffer,
620                                      SCSI_SENSE_BUFFERSIZE, &sense_hdr);
621
622                 if (sense_hdr.asc == 0x3A) {
623                         /* Medium not present */
624                         blkdev->media_not_present = 1;
625                 }
626                 return 0;
627         }
628         buf = kmap(page_buf);
629
630         /* be to le */
631         blkdev->capacity = ((buf[0] << 24) | (buf[1] << 16) |
632                             (buf[2] << 8) | buf[3]) + 1;
633         blkdev->sector_size = (buf[4] << 24) | (buf[5] << 16) |
634                               (buf[6] << 8) | buf[7];
635
636         kunmap(page_buf);
637
638         __free_page(page_buf);
639
640         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
641
642         return 0;
643 }
644
645 static int blkvsc_do_read_capacity16(struct block_device_context *blkdev)
646 {
647         struct blkvsc_request *blkvsc_req;
648         struct page *page_buf;
649         unsigned char *buf;
650         struct scsi_sense_hdr sense_hdr;
651
652         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_read_capacity16()\n");
653
654         blkdev->sector_size = 0;
655         blkdev->capacity = 0;
656         blkdev->media_not_present = 0; /* assume a disk is present */
657
658         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
659         if (!blkvsc_req)
660                 return -ENOMEM;
661
662         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
663         page_buf = alloc_page(GFP_KERNEL);
664         if (!page_buf) {
665                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
666                 return -ENOMEM;
667         }
668
669         init_waitqueue_head(&blkvsc_req->wevent);
670         blkvsc_req->dev = blkdev;
671         blkvsc_req->req = NULL;
672         blkvsc_req->write = 0;
673
674         blkvsc_req->request.data_buffer.PfnArray[0] = page_to_pfn(page_buf);
675         blkvsc_req->request.data_buffer.Offset = 0;
676         blkvsc_req->request.data_buffer.Length = 12;
677
678         blkvsc_req->cmnd[0] = 0x9E; /* READ_CAPACITY16; */
679         blkvsc_req->cmd_len = 16;
680
681         /*
682          * Set this here since the completion routine may be invoked
683          * and completed before we return
684          */
685         blkvsc_req->cond = 0;
686
687         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
688
689         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete - cond %d\n",
690                    blkvsc_req, blkvsc_req->cond);
691
692         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
693
694         /* check error */
695         if (blkvsc_req->request.status) {
696                 scsi_normalize_sense(blkvsc_req->sense_buffer,
697                                      SCSI_SENSE_BUFFERSIZE, &sense_hdr);
698                 if (sense_hdr.asc == 0x3A) {
699                         /* Medium not present */
700                         blkdev->media_not_present = 1;
701                 }
702                 return 0;
703         }
704         buf = kmap(page_buf);
705
706         /* be to le */
707         blkdev->capacity = be64_to_cpu(*(unsigned long long *) &buf[0]) + 1;
708         blkdev->sector_size = be32_to_cpu(*(unsigned int *)&buf[8]);
709
710 #if 0
711         blkdev->capacity = ((buf[0] << 24) | (buf[1] << 16) |
712                             (buf[2] << 8) | buf[3]) + 1;
713         blkdev->sector_size = (buf[4] << 24) | (buf[5] << 16) |
714                               (buf[6] << 8) | buf[7];
715 #endif
716
717         kunmap(page_buf);
718
719         __free_page(page_buf);
720
721         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
722
723         return 0;
724 }
725
726 /*
727  * blkvsc_remove() - Callback when our device is removed
728  */
729 static int blkvsc_remove(struct device *device)
730 {
731         struct driver_context *driver_ctx =
732                                 driver_to_driver_context(device->driver);
733         struct blkvsc_driver_context *blkvsc_drv_ctx =
734                                 (struct blkvsc_driver_context *)driver_ctx;
735         struct storvsc_driver_object *storvsc_drv_obj =
736                                 &blkvsc_drv_ctx->drv_obj;
737         struct vm_device *device_ctx = device_to_vm_device(device);
738         struct hv_device *device_obj = &device_ctx->device_obj;
739         struct block_device_context *blkdev = dev_get_drvdata(device);
740         unsigned long flags;
741         int ret;
742
743         DPRINT_DBG(BLKVSC_DRV, "blkvsc_remove()\n");
744
745         if (!storvsc_drv_obj->base.OnDeviceRemove)
746                 return -1;
747
748         /*
749          * Call to the vsc driver to let it know that the device is being
750          * removed
751          */
752         ret = storvsc_drv_obj->base.OnDeviceRemove(device_obj);
753         if (ret != 0) {
754                 /* TODO: */
755                 DPRINT_ERR(BLKVSC_DRV,
756                            "unable to remove blkvsc device (ret %d)", ret);
757         }
758
759         /* Get to a known state */
760         spin_lock_irqsave(&blkdev->lock, flags);
761
762         blkdev->shutting_down = 1;
763
764         blk_stop_queue(blkdev->gd->queue);
765
766         spin_unlock_irqrestore(&blkdev->lock, flags);
767
768         while (blkdev->num_outstanding_reqs) {
769                 DPRINT_INFO(STORVSC, "waiting for %d requests to complete...",
770                             blkdev->num_outstanding_reqs);
771                 udelay(100);
772         }
773
774         blkvsc_do_flush(blkdev);
775
776         spin_lock_irqsave(&blkdev->lock, flags);
777
778         blkvsc_cancel_pending_reqs(blkdev);
779
780         spin_unlock_irqrestore(&blkdev->lock, flags);
781
782         blk_cleanup_queue(blkdev->gd->queue);
783
784         del_gendisk(blkdev->gd);
785
786         kmem_cache_destroy(blkdev->request_pool);
787
788         kfree(blkdev);
789
790         return ret;
791 }
792
793 static void blkvsc_init_rw(struct blkvsc_request *blkvsc_req)
794 {
795         /* ASSERT(blkvsc_req->req); */
796         /* ASSERT(blkvsc_req->sector_count <= (MAX_MULTIPAGE_BUFFER_COUNT*8)); */
797
798         blkvsc_req->cmd_len = 16;
799
800         if (blkvsc_req->sector_start > 0xffffffff) {
801                 if (rq_data_dir(blkvsc_req->req)) {
802                         blkvsc_req->write = 1;
803                         blkvsc_req->cmnd[0] = WRITE_16;
804                 } else {
805                         blkvsc_req->write = 0;
806                         blkvsc_req->cmnd[0] = READ_16;
807                 }
808
809                 blkvsc_req->cmnd[1] |=
810                         (blkvsc_req->req->cmd_flags & REQ_FUA) ? 0x8 : 0;
811
812                 *(unsigned long long *)&blkvsc_req->cmnd[2] =
813                                 cpu_to_be64(blkvsc_req->sector_start);
814                 *(unsigned int *)&blkvsc_req->cmnd[10] =
815                                 cpu_to_be32(blkvsc_req->sector_count);
816         } else if ((blkvsc_req->sector_count > 0xff) ||
817                    (blkvsc_req->sector_start > 0x1fffff)) {
818                 if (rq_data_dir(blkvsc_req->req)) {
819                         blkvsc_req->write = 1;
820                         blkvsc_req->cmnd[0] = WRITE_10;
821                 } else {
822                         blkvsc_req->write = 0;
823                         blkvsc_req->cmnd[0] = READ_10;
824                 }
825
826                 blkvsc_req->cmnd[1] |=
827                         (blkvsc_req->req->cmd_flags & REQ_FUA) ? 0x8 : 0;
828
829                 *(unsigned int *)&blkvsc_req->cmnd[2] =
830                                 cpu_to_be32(blkvsc_req->sector_start);
831                 *(unsigned short *)&blkvsc_req->cmnd[7] =
832                                 cpu_to_be16(blkvsc_req->sector_count);
833         } else {
834                 if (rq_data_dir(blkvsc_req->req)) {
835                         blkvsc_req->write = 1;
836                         blkvsc_req->cmnd[0] = WRITE_6;
837                 } else {
838                         blkvsc_req->write = 0;
839                         blkvsc_req->cmnd[0] = READ_6;
840                 }
841
842                 *(unsigned int *)&blkvsc_req->cmnd[1] =
843                                 cpu_to_be32(blkvsc_req->sector_start) >> 8;
844                 blkvsc_req->cmnd[1] &= 0x1f;
845                 blkvsc_req->cmnd[4] = (unsigned char)blkvsc_req->sector_count;
846         }
847 }
848
849 static int blkvsc_submit_request(struct blkvsc_request *blkvsc_req,
850                         void (*request_completion)(struct hv_storvsc_request *))
851 {
852         struct block_device_context *blkdev = blkvsc_req->dev;
853         struct vm_device *device_ctx = blkdev->device_ctx;
854         struct driver_context *driver_ctx =
855                         driver_to_driver_context(device_ctx->device.driver);
856         struct blkvsc_driver_context *blkvsc_drv_ctx =
857                         (struct blkvsc_driver_context *)driver_ctx;
858         struct storvsc_driver_object *storvsc_drv_obj =
859                         &blkvsc_drv_ctx->drv_obj;
860         struct hv_storvsc_request *storvsc_req;
861         int ret;
862
863         DPRINT_DBG(BLKVSC_DRV, "blkvsc_submit_request() - "
864                    "req %p type %s start_sector %lu count %ld offset %d "
865                    "len %d\n", blkvsc_req,
866                    (blkvsc_req->write) ? "WRITE" : "READ",
867                    (unsigned long) blkvsc_req->sector_start,
868                    blkvsc_req->sector_count,
869                    blkvsc_req->request.data_buffer.Offset,
870                    blkvsc_req->request.data_buffer.Length);
871 #if 0
872         for (i = 0; i < (blkvsc_req->request.data_buffer.Length >> 12); i++) {
873                 DPRINT_DBG(BLKVSC_DRV, "blkvsc_submit_request() - "
874                            "req %p pfn[%d] %llx\n",
875                            blkvsc_req, i,
876                            blkvsc_req->request.data_buffer.PfnArray[i]);
877         }
878 #endif
879
880         storvsc_req = &blkvsc_req->request;
881         storvsc_req->extension = (void *)((unsigned long)blkvsc_req +
882                                           sizeof(struct blkvsc_request));
883
884         storvsc_req->type = blkvsc_req->write ? WRITE_TYPE : READ_TYPE;
885
886         storvsc_req->on_io_completion = request_completion;
887         storvsc_req->context = blkvsc_req;
888
889         storvsc_req->host = blkdev->port;
890         storvsc_req->bus = blkdev->path;
891         storvsc_req->target_id = blkdev->target;
892         storvsc_req->lun_id = 0;         /* this is not really used at all */
893
894         storvsc_req->cdb_len = blkvsc_req->cmd_len;
895         storvsc_req->cdb = blkvsc_req->cmnd;
896
897         storvsc_req->sense_buffer = blkvsc_req->sense_buffer;
898         storvsc_req->sense_buffer_size = SCSI_SENSE_BUFFERSIZE;
899
900         ret = storvsc_drv_obj->on_io_request(&blkdev->device_ctx->device_obj,
901                                            &blkvsc_req->request);
902         if (ret == 0)
903                 blkdev->num_outstanding_reqs++;
904
905         return ret;
906 }
907
908 /*
909  * We break the request into 1 or more blkvsc_requests and submit
910  * them.  If we cant submit them all, we put them on the
911  * pending_list. The blkvsc_request() will work on the pending_list.
912  */
913 static int blkvsc_do_request(struct block_device_context *blkdev,
914                              struct request *req)
915 {
916         struct bio *bio = NULL;
917         struct bio_vec *bvec = NULL;
918         struct bio_vec *prev_bvec = NULL;
919         struct blkvsc_request *blkvsc_req = NULL;
920         struct blkvsc_request *tmp;
921         int databuf_idx = 0;
922         int seg_idx = 0;
923         sector_t start_sector;
924         unsigned long num_sectors = 0;
925         int ret = 0;
926         int pending = 0;
927         struct blkvsc_request_group *group = NULL;
928
929         DPRINT_DBG(BLKVSC_DRV, "blkdev %p req %p sect %lu\n", blkdev, req,
930                   (unsigned long)blk_rq_pos(req));
931
932         /* Create a group to tie req to list of blkvsc_reqs */
933         group = kmem_cache_alloc(blkdev->request_pool, GFP_ATOMIC);
934         if (!group)
935                 return -ENOMEM;
936
937         INIT_LIST_HEAD(&group->blkvsc_req_list);
938         group->outstanding = group->status = 0;
939
940         start_sector = blk_rq_pos(req);
941
942         /* foreach bio in the request */
943         if (req->bio) {
944                 for (bio = req->bio; bio; bio = bio->bi_next) {
945                         /*
946                          * Map this bio into an existing or new storvsc request
947                          */
948                         bio_for_each_segment(bvec, bio, seg_idx) {
949                                 DPRINT_DBG(BLKVSC_DRV, "bio_for_each_segment() "
950                                            "- req %p bio %p bvec %p seg_idx %d "
951                                            "databuf_idx %d\n", req, bio, bvec,
952                                            seg_idx, databuf_idx);
953
954                                 /* Get a new storvsc request */
955                                 /* 1st-time */
956                                 if ((!blkvsc_req) ||
957                                     (databuf_idx >= MAX_MULTIPAGE_BUFFER_COUNT)
958                                     /* hole at the begin of page */
959                                     || (bvec->bv_offset != 0) ||
960                                     /* hold at the end of page */
961                                     (prev_bvec &&
962                                      (prev_bvec->bv_len != PAGE_SIZE))) {
963                                         /* submit the prev one */
964                                         if (blkvsc_req) {
965                                                 blkvsc_req->sector_start = start_sector;
966                                                 sector_div(blkvsc_req->sector_start, (blkdev->sector_size >> 9));
967
968                                                 blkvsc_req->sector_count = num_sectors / (blkdev->sector_size >> 9);
969                                                 blkvsc_init_rw(blkvsc_req);
970                                         }
971
972                                         /*
973                                          * Create new blkvsc_req to represent
974                                          * the current bvec
975                                          */
976                                         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_ATOMIC);
977                                         if (!blkvsc_req) {
978                                                 /* free up everything */
979                                                 list_for_each_entry_safe(
980                                                         blkvsc_req, tmp,
981                                                         &group->blkvsc_req_list,
982                                                         req_entry) {
983                                                         list_del(&blkvsc_req->req_entry);
984                                                         kmem_cache_free(blkdev->request_pool, blkvsc_req);
985                                                 }
986
987                                                 kmem_cache_free(blkdev->request_pool, group);
988                                                 return -ENOMEM;
989                                         }
990
991                                         memset(blkvsc_req, 0,
992                                                sizeof(struct blkvsc_request));
993
994                                         blkvsc_req->dev = blkdev;
995                                         blkvsc_req->req = req;
996                                         blkvsc_req->request.data_buffer.Offset
997                                                 = bvec->bv_offset;
998                                         blkvsc_req->request.data_buffer.Length
999                                                 = 0;
1000
1001                                         /* Add to the group */
1002                                         blkvsc_req->group = group;
1003                                         blkvsc_req->group->outstanding++;
1004                                         list_add_tail(&blkvsc_req->req_entry,
1005                                                 &blkvsc_req->group->blkvsc_req_list);
1006
1007                                         start_sector += num_sectors;
1008                                         num_sectors = 0;
1009                                         databuf_idx = 0;
1010                                 }
1011
1012                                 /* Add the curr bvec/segment to the curr blkvsc_req */
1013                                 blkvsc_req->request.data_buffer.
1014                                         PfnArray[databuf_idx]
1015                                                 = page_to_pfn(bvec->bv_page);
1016                                 blkvsc_req->request.data_buffer.Length
1017                                         += bvec->bv_len;
1018
1019                                 prev_bvec = bvec;
1020
1021                                 databuf_idx++;
1022                                 num_sectors += bvec->bv_len >> 9;
1023
1024                         } /* bio_for_each_segment */
1025
1026                 } /* rq_for_each_bio */
1027         }
1028
1029         /* Handle the last one */
1030         if (blkvsc_req) {
1031                 DPRINT_DBG(BLKVSC_DRV, "blkdev %p req %p group %p count %d\n",
1032                            blkdev, req, blkvsc_req->group,
1033                            blkvsc_req->group->outstanding);
1034
1035                 blkvsc_req->sector_start = start_sector;
1036                 sector_div(blkvsc_req->sector_start,
1037                            (blkdev->sector_size >> 9));
1038
1039                 blkvsc_req->sector_count = num_sectors /
1040                                            (blkdev->sector_size >> 9);
1041
1042                 blkvsc_init_rw(blkvsc_req);
1043         }
1044
1045         list_for_each_entry(blkvsc_req, &group->blkvsc_req_list, req_entry) {
1046                 if (pending) {
1047                         DPRINT_DBG(BLKVSC_DRV, "adding blkvsc_req to "
1048                                    "pending_list - blkvsc_req %p start_sect %lu"
1049                                    " sect_count %ld (%lu %ld)\n", blkvsc_req,
1050                                    (unsigned long)blkvsc_req->sector_start,
1051                                    blkvsc_req->sector_count,
1052                                    (unsigned long)start_sector,
1053                                    (unsigned long)num_sectors);
1054
1055                         list_add_tail(&blkvsc_req->pend_entry,
1056                                       &blkdev->pending_list);
1057                 } else {
1058                         ret = blkvsc_submit_request(blkvsc_req,
1059                                                     blkvsc_request_completion);
1060                         if (ret == -1) {
1061                                 pending = 1;
1062                                 list_add_tail(&blkvsc_req->pend_entry,
1063                                               &blkdev->pending_list);
1064                         }
1065
1066                         DPRINT_DBG(BLKVSC_DRV, "submitted blkvsc_req %p "
1067                                    "start_sect %lu sect_count %ld (%lu %ld) "
1068                                    "ret %d\n", blkvsc_req,
1069                                    (unsigned long)blkvsc_req->sector_start,
1070                                    blkvsc_req->sector_count,
1071                                    (unsigned long)start_sector,
1072                                    num_sectors, ret);
1073                 }
1074         }
1075
1076         return pending;
1077 }
1078
1079 static void blkvsc_cmd_completion(struct hv_storvsc_request *request)
1080 {
1081         struct blkvsc_request *blkvsc_req =
1082                         (struct blkvsc_request *)request->context;
1083         struct block_device_context *blkdev =
1084                         (struct block_device_context *)blkvsc_req->dev;
1085         struct scsi_sense_hdr sense_hdr;
1086
1087         DPRINT_DBG(BLKVSC_DRV, "blkvsc_cmd_completion() - req %p\n",
1088                    blkvsc_req);
1089
1090         blkdev->num_outstanding_reqs--;
1091
1092         if (blkvsc_req->request.status)
1093                 if (scsi_normalize_sense(blkvsc_req->sense_buffer,
1094                                          SCSI_SENSE_BUFFERSIZE, &sense_hdr))
1095                         scsi_print_sense_hdr("blkvsc", &sense_hdr);
1096
1097         blkvsc_req->cond = 1;
1098         wake_up_interruptible(&blkvsc_req->wevent);
1099 }
1100
1101 static void blkvsc_request_completion(struct hv_storvsc_request *request)
1102 {
1103         struct blkvsc_request *blkvsc_req =
1104                         (struct blkvsc_request *)request->context;
1105         struct block_device_context *blkdev =
1106                         (struct block_device_context *)blkvsc_req->dev;
1107         unsigned long flags;
1108         struct blkvsc_request *comp_req, *tmp;
1109
1110         /* ASSERT(blkvsc_req->group); */
1111
1112         DPRINT_DBG(BLKVSC_DRV, "blkdev %p blkvsc_req %p group %p type %s "
1113                    "sect_start %lu sect_count %ld len %d group outstd %d "
1114                    "total outstd %d\n",
1115                    blkdev, blkvsc_req, blkvsc_req->group,
1116                    (blkvsc_req->write) ? "WRITE" : "READ",
1117                    (unsigned long)blkvsc_req->sector_start,
1118                    blkvsc_req->sector_count,
1119                    blkvsc_req->request.data_buffer.Length,
1120                    blkvsc_req->group->outstanding,
1121                    blkdev->num_outstanding_reqs);
1122
1123         spin_lock_irqsave(&blkdev->lock, flags);
1124
1125         blkdev->num_outstanding_reqs--;
1126         blkvsc_req->group->outstanding--;
1127
1128         /*
1129          * Only start processing when all the blkvsc_reqs are
1130          * completed. This guarantees no out-of-order blkvsc_req
1131          * completion when calling end_that_request_first()
1132          */
1133         if (blkvsc_req->group->outstanding == 0) {
1134                 list_for_each_entry_safe(comp_req, tmp,
1135                                          &blkvsc_req->group->blkvsc_req_list,
1136                                          req_entry) {
1137                         DPRINT_DBG(BLKVSC_DRV, "completing blkvsc_req %p "
1138                                    "sect_start %lu sect_count %ld\n",
1139                                    comp_req,
1140                                    (unsigned long)comp_req->sector_start,
1141                                    comp_req->sector_count);
1142
1143                         list_del(&comp_req->req_entry);
1144
1145                         if (!__blk_end_request(comp_req->req,
1146                                 (!comp_req->request.status ? 0 : -EIO),
1147                                 comp_req->sector_count * blkdev->sector_size)) {
1148                                 /*
1149                                  * All the sectors have been xferred ie the
1150                                  * request is done
1151                                  */
1152                                 DPRINT_DBG(BLKVSC_DRV, "req %p COMPLETED\n",
1153                                            comp_req->req);
1154                                 kmem_cache_free(blkdev->request_pool,
1155                                                 comp_req->group);
1156                         }
1157
1158                         kmem_cache_free(blkdev->request_pool, comp_req);
1159                 }
1160
1161                 if (!blkdev->shutting_down) {
1162                         blkvsc_do_pending_reqs(blkdev);
1163                         blk_start_queue(blkdev->gd->queue);
1164                         blkvsc_request(blkdev->gd->queue);
1165                 }
1166         }
1167
1168         spin_unlock_irqrestore(&blkdev->lock, flags);
1169 }
1170
1171 static int blkvsc_cancel_pending_reqs(struct block_device_context *blkdev)
1172 {
1173         struct blkvsc_request *pend_req, *tmp;
1174         struct blkvsc_request *comp_req, *tmp2;
1175
1176         int ret = 0;
1177
1178         DPRINT_DBG(BLKVSC_DRV, "blkvsc_cancel_pending_reqs()");
1179
1180         /* Flush the pending list first */
1181         list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
1182                                  pend_entry) {
1183                 /*
1184                  * The pend_req could be part of a partially completed
1185                  * request. If so, complete those req first until we
1186                  * hit the pend_req
1187                  */
1188                 list_for_each_entry_safe(comp_req, tmp2,
1189                                          &pend_req->group->blkvsc_req_list,
1190                                          req_entry) {
1191                         DPRINT_DBG(BLKVSC_DRV, "completing blkvsc_req %p "
1192                                    "sect_start %lu sect_count %ld\n",
1193                                    comp_req,
1194                                    (unsigned long) comp_req->sector_start,
1195                                    comp_req->sector_count);
1196
1197                         if (comp_req == pend_req)
1198                                 break;
1199
1200                         list_del(&comp_req->req_entry);
1201
1202                         if (comp_req->req) {
1203                                 ret = __blk_end_request(comp_req->req,
1204                                         (!comp_req->request.status ? 0 : -EIO),
1205                                         comp_req->sector_count *
1206                                         blkdev->sector_size);
1207
1208                                 /* FIXME: shouldn't this do more than return? */
1209                                 if (ret)
1210                                         goto out;
1211                         }
1212
1213                         kmem_cache_free(blkdev->request_pool, comp_req);
1214                 }
1215
1216                 DPRINT_DBG(BLKVSC_DRV, "cancelling pending request - %p\n",
1217                            pend_req);
1218
1219                 list_del(&pend_req->pend_entry);
1220
1221                 list_del(&pend_req->req_entry);
1222
1223                 if (comp_req->req) {
1224                         if (!__blk_end_request(pend_req->req, -EIO,
1225                                                pend_req->sector_count *
1226                                                blkdev->sector_size)) {
1227                                 /*
1228                                  * All the sectors have been xferred ie the
1229                                  * request is done
1230                                  */
1231                                 DPRINT_DBG(BLKVSC_DRV,
1232                                            "blkvsc_cancel_pending_reqs() - "
1233                                            "req %p COMPLETED\n", pend_req->req);
1234                                 kmem_cache_free(blkdev->request_pool,
1235                                                 pend_req->group);
1236                         }
1237                 }
1238
1239                 kmem_cache_free(blkdev->request_pool, pend_req);
1240         }
1241
1242 out:
1243         return ret;
1244 }
1245
1246 static int blkvsc_do_pending_reqs(struct block_device_context *blkdev)
1247 {
1248         struct blkvsc_request *pend_req, *tmp;
1249         int ret = 0;
1250
1251         /* Flush the pending list first */
1252         list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
1253                                  pend_entry) {
1254                 DPRINT_DBG(BLKVSC_DRV, "working off pending_list - %p\n",
1255                            pend_req);
1256
1257                 ret = blkvsc_submit_request(pend_req,
1258                                             blkvsc_request_completion);
1259                 if (ret != 0)
1260                         break;
1261                 else
1262                         list_del(&pend_req->pend_entry);
1263         }
1264
1265         return ret;
1266 }
1267
1268 static void blkvsc_request(struct request_queue *queue)
1269 {
1270         struct block_device_context *blkdev = NULL;
1271         struct request *req;
1272         int ret = 0;
1273
1274         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1275         while ((req = blk_peek_request(queue)) != NULL) {
1276                 DPRINT_DBG(BLKVSC_DRV, "- req %p\n", req);
1277
1278                 blkdev = req->rq_disk->private_data;
1279                 if (blkdev->shutting_down || req->cmd_type != REQ_TYPE_FS ||
1280                     blkdev->media_not_present) {
1281                         __blk_end_request_cur(req, 0);
1282                         continue;
1283                 }
1284
1285                 ret = blkvsc_do_pending_reqs(blkdev);
1286
1287                 if (ret != 0) {
1288                         DPRINT_DBG(BLKVSC_DRV,
1289                                    "- stop queue - pending_list not empty\n");
1290                         blk_stop_queue(queue);
1291                         break;
1292                 }
1293
1294                 blk_start_request(req);
1295
1296                 ret = blkvsc_do_request(blkdev, req);
1297                 if (ret > 0) {
1298                         DPRINT_DBG(BLKVSC_DRV, "- stop queue - no room\n");
1299                         blk_stop_queue(queue);
1300                         break;
1301                 } else if (ret < 0) {
1302                         DPRINT_DBG(BLKVSC_DRV, "- stop queue - no mem\n");
1303                         blk_requeue_request(queue, req);
1304                         blk_stop_queue(queue);
1305                         break;
1306                 }
1307         }
1308 }
1309
1310 static int blkvsc_open(struct block_device *bdev, fmode_t mode)
1311 {
1312         struct block_device_context *blkdev = bdev->bd_disk->private_data;
1313
1314         DPRINT_DBG(BLKVSC_DRV, "- users %d disk %s\n", blkdev->users,
1315                    blkdev->gd->disk_name);
1316
1317         mutex_lock(&blkvsc_mutex);
1318         spin_lock(&blkdev->lock);
1319
1320         if (!blkdev->users && blkdev->device_type == DVD_TYPE) {
1321                 spin_unlock(&blkdev->lock);
1322                 check_disk_change(bdev);
1323                 spin_lock(&blkdev->lock);
1324         }
1325
1326         blkdev->users++;
1327
1328         spin_unlock(&blkdev->lock);
1329         mutex_unlock(&blkvsc_mutex);
1330         return 0;
1331 }
1332
1333 static int blkvsc_release(struct gendisk *disk, fmode_t mode)
1334 {
1335         struct block_device_context *blkdev = disk->private_data;
1336
1337         DPRINT_DBG(BLKVSC_DRV, "- users %d disk %s\n", blkdev->users,
1338                    blkdev->gd->disk_name);
1339
1340         mutex_lock(&blkvsc_mutex);
1341         spin_lock(&blkdev->lock);
1342         if (blkdev->users == 1) {
1343                 spin_unlock(&blkdev->lock);
1344                 blkvsc_do_flush(blkdev);
1345                 spin_lock(&blkdev->lock);
1346         }
1347
1348         blkdev->users--;
1349
1350         spin_unlock(&blkdev->lock);
1351         mutex_unlock(&blkvsc_mutex);
1352         return 0;
1353 }
1354
1355 static int blkvsc_media_changed(struct gendisk *gd)
1356 {
1357         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1358         return 1;
1359 }
1360
1361 static int blkvsc_revalidate_disk(struct gendisk *gd)
1362 {
1363         struct block_device_context *blkdev = gd->private_data;
1364
1365         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1366
1367         if (blkdev->device_type == DVD_TYPE) {
1368                 blkvsc_do_read_capacity(blkdev);
1369                 set_capacity(blkdev->gd, blkdev->capacity *
1370                             (blkdev->sector_size/512));
1371                 blk_queue_logical_block_size(gd->queue, blkdev->sector_size);
1372         }
1373         return 0;
1374 }
1375
1376 static int blkvsc_getgeo(struct block_device *bd, struct hd_geometry *hg)
1377 {
1378         sector_t total_sectors = get_capacity(bd->bd_disk);
1379         sector_t cylinder_times_heads = 0;
1380         sector_t temp = 0;
1381
1382         int sectors_per_track = 0;
1383         int heads = 0;
1384         int cylinders = 0;
1385         int rem = 0;
1386
1387         if (total_sectors > (65535 * 16 * 255))
1388                 total_sectors = (65535 * 16 * 255);
1389
1390         if (total_sectors >= (65535 * 16 * 63)) {
1391                 sectors_per_track = 255;
1392                 heads = 16;
1393
1394                 cylinder_times_heads = total_sectors;
1395                 /* sector_div stores the quotient in cylinder_times_heads */
1396                 rem = sector_div(cylinder_times_heads, sectors_per_track);
1397         } else {
1398                 sectors_per_track = 17;
1399
1400                 cylinder_times_heads = total_sectors;
1401                 /* sector_div stores the quotient in cylinder_times_heads */
1402                 rem = sector_div(cylinder_times_heads, sectors_per_track);
1403
1404                 temp = cylinder_times_heads + 1023;
1405                 /* sector_div stores the quotient in temp */
1406                 rem = sector_div(temp, 1024);
1407
1408                 heads = temp;
1409
1410                 if (heads < 4)
1411                         heads = 4;
1412
1413
1414                 if (cylinder_times_heads >= (heads * 1024) || (heads > 16)) {
1415                         sectors_per_track = 31;
1416                         heads = 16;
1417
1418                         cylinder_times_heads = total_sectors;
1419                         /*
1420                          * sector_div stores the quotient in
1421                          * cylinder_times_heads
1422                          */
1423                         rem = sector_div(cylinder_times_heads,
1424                                          sectors_per_track);
1425                 }
1426
1427                 if (cylinder_times_heads >= (heads * 1024)) {
1428                         sectors_per_track = 63;
1429                         heads = 16;
1430
1431                         cylinder_times_heads = total_sectors;
1432                         /*
1433                          * sector_div stores the quotient in
1434                          * cylinder_times_heads
1435                          */
1436                         rem = sector_div(cylinder_times_heads,
1437                                          sectors_per_track);
1438                 }
1439         }
1440
1441         temp = cylinder_times_heads;
1442         /* sector_div stores the quotient in temp */
1443         rem = sector_div(temp, heads);
1444         cylinders = temp;
1445
1446         hg->heads = heads;
1447         hg->sectors = sectors_per_track;
1448         hg->cylinders = cylinders;
1449
1450         DPRINT_INFO(BLKVSC_DRV, "CHS (%d, %d, %d)", cylinders, heads,
1451                     sectors_per_track);
1452
1453     return 0;
1454 }
1455
1456 static int blkvsc_ioctl(struct block_device *bd, fmode_t mode,
1457                         unsigned cmd, unsigned long argument)
1458 {
1459 /*      struct block_device_context *blkdev = bd->bd_disk->private_data; */
1460         int ret;
1461
1462         switch (cmd) {
1463         /*
1464          * TODO: I think there is certain format for HDIO_GET_IDENTITY rather
1465          * than just a GUID. Commented it out for now.
1466          */
1467 #if 0
1468         case HDIO_GET_IDENTITY:
1469                 DPRINT_INFO(BLKVSC_DRV, "HDIO_GET_IDENTITY\n");
1470                 if (copy_to_user((void __user *)arg, blkdev->device_id,
1471                                  blkdev->device_id_len))
1472                         ret = -EFAULT;
1473                 break;
1474 #endif
1475         default:
1476                 ret = -EINVAL;
1477                 break;
1478         }
1479
1480         return ret;
1481 }
1482
1483 static int __init blkvsc_init(void)
1484 {
1485         int ret;
1486
1487         BUILD_BUG_ON(sizeof(sector_t) != 8);
1488
1489         DPRINT_INFO(BLKVSC_DRV, "Blkvsc initializing....");
1490
1491         ret = blkvsc_drv_init(blk_vsc_initialize);
1492
1493         return ret;
1494 }
1495
1496 static void __exit blkvsc_exit(void)
1497 {
1498         blkvsc_drv_exit();
1499 }
1500
1501 MODULE_LICENSE("GPL");
1502 MODULE_VERSION(HV_DRV_VERSION);
1503 MODULE_DESCRIPTION("Microsoft Hyper-V virtual block driver");
1504 module_init(blkvsc_init);
1505 module_exit(blkvsc_exit);