7c82d148f32e5d3b9c607b16faba0bdee61ef8ed
[cascardo/linux.git] / drivers / staging / hv / storvsc_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  *   K. Y. Srinivasan <kys@microsoft.com>
21  */
22
23 #include <linux/kernel.h>
24 #include <linux/wait.h>
25 #include <linux/sched.h>
26 #include <linux/completion.h>
27 #include <linux/string.h>
28 #include <linux/mm.h>
29 #include <linux/delay.h>
30 #include <linux/init.h>
31 #include <linux/slab.h>
32 #include <linux/module.h>
33 #include <linux/device.h>
34 #include <linux/hyperv.h>
35 #include <linux/mempool.h>
36 #include <scsi/scsi.h>
37 #include <scsi/scsi_cmnd.h>
38 #include <scsi/scsi_host.h>
39 #include <scsi/scsi_device.h>
40 #include <scsi/scsi_tcq.h>
41 #include <scsi/scsi_eh.h>
42 #include <scsi/scsi_devinfo.h>
43 #include <scsi/scsi_dbg.h>
44
45
46 #define STORVSC_MIN_BUF_NR                              64
47 #define STORVSC_RING_BUFFER_SIZE                        (20*PAGE_SIZE)
48 static int storvsc_ringbuffer_size = STORVSC_RING_BUFFER_SIZE;
49
50 module_param(storvsc_ringbuffer_size, int, S_IRUGO);
51 MODULE_PARM_DESC(storvsc_ringbuffer_size, "Ring buffer size (bytes)");
52
53 /* to alert the user that structure sizes may be mismatched even though the */
54 /* protocol versions match. */
55
56
57 #define REVISION_STRING(REVISION_) #REVISION_
58 #define FILL_VMSTOR_REVISION(RESULT_LVALUE_)                            \
59         do {                                                            \
60                 char *revision_string                                   \
61                         = REVISION_STRING($Rev : 6 $) + 6;              \
62                 RESULT_LVALUE_ = 0;                                     \
63                 while (*revision_string >= '0'                          \
64                         && *revision_string <= '9') {                   \
65                         RESULT_LVALUE_ *= 10;                           \
66                         RESULT_LVALUE_ += *revision_string - '0';       \
67                         revision_string++;                              \
68                 }                                                       \
69         } while (0)
70
71 /* Major/minor macros.  Minor version is in LSB, meaning that earlier flat */
72 /* version numbers will be interpreted as "0.x" (i.e., 1 becomes 0.1). */
73 #define VMSTOR_PROTOCOL_MAJOR(VERSION_)         (((VERSION_) >> 8) & 0xff)
74 #define VMSTOR_PROTOCOL_MINOR(VERSION_)         (((VERSION_))      & 0xff)
75 #define VMSTOR_PROTOCOL_VERSION(MAJOR_, MINOR_) ((((MAJOR_) & 0xff) << 8) | \
76                                                  (((MINOR_) & 0xff)))
77 #define VMSTOR_INVALID_PROTOCOL_VERSION         (-1)
78
79 /* Version history: */
80 /* V1 Beta                    0.1 */
81 /* V1 RC < 2008/1/31          1.0 */
82 /* V1 RC > 2008/1/31          2.0 */
83 #define VMSTOR_PROTOCOL_VERSION_CURRENT VMSTOR_PROTOCOL_VERSION(4, 2)
84
85
86
87
88 /*  This will get replaced with the max transfer length that is possible on */
89 /*  the host adapter. */
90 /*  The max transfer length will be published when we offer a vmbus channel. */
91 #define MAX_TRANSFER_LENGTH     0x40000
92 #define DEFAULT_PACKET_SIZE (sizeof(struct vmdata_gpa_direct) + \
93                         sizeof(struct vstor_packet) +           \
94                         sizesizeof(u64) * (MAX_TRANSFER_LENGTH / PAGE_SIZE)))
95
96
97 /*  Packet structure describing virtual storage requests. */
98 enum vstor_packet_operation {
99         VSTOR_OPERATION_COMPLETE_IO             = 1,
100         VSTOR_OPERATION_REMOVE_DEVICE           = 2,
101         VSTOR_OPERATION_EXECUTE_SRB             = 3,
102         VSTOR_OPERATION_RESET_LUN               = 4,
103         VSTOR_OPERATION_RESET_ADAPTER           = 5,
104         VSTOR_OPERATION_RESET_BUS               = 6,
105         VSTOR_OPERATION_BEGIN_INITIALIZATION    = 7,
106         VSTOR_OPERATION_END_INITIALIZATION      = 8,
107         VSTOR_OPERATION_QUERY_PROTOCOL_VERSION  = 9,
108         VSTOR_OPERATION_QUERY_PROPERTIES        = 10,
109         VSTOR_OPERATION_ENUMERATE_BUS           = 11,
110         VSTOR_OPERATION_MAXIMUM                 = 11
111 };
112
113 /*
114  * Platform neutral description of a scsi request -
115  * this remains the same across the write regardless of 32/64 bit
116  * note: it's patterned off the SCSI_PASS_THROUGH structure
117  */
118 #define CDB16GENERIC_LENGTH                     0x10
119
120 #ifndef SENSE_BUFFER_SIZE
121 #define SENSE_BUFFER_SIZE                       0x12
122 #endif
123
124 #define MAX_DATA_BUF_LEN_WITH_PADDING           0x14
125
126 struct vmscsi_request {
127         unsigned short length;
128         unsigned char srb_status;
129         unsigned char scsi_status;
130
131         unsigned char port_number;
132         unsigned char path_id;
133         unsigned char target_id;
134         unsigned char lun;
135
136         unsigned char cdb_length;
137         unsigned char sense_info_length;
138         unsigned char data_in;
139         unsigned char reserved;
140
141         unsigned int data_transfer_length;
142
143         union {
144                 unsigned char cdb[CDB16GENERIC_LENGTH];
145                 unsigned char sense_data[SENSE_BUFFER_SIZE];
146                 unsigned char reserved_array[MAX_DATA_BUF_LEN_WITH_PADDING];
147         };
148 } __attribute((packed));
149
150
151 /*
152  * This structure is sent during the intialization phase to get the different
153  * properties of the channel.
154  */
155 struct vmstorage_channel_properties {
156         unsigned short protocol_version;
157         unsigned char path_id;
158         unsigned char target_id;
159
160         /* Note: port number is only really known on the client side */
161         unsigned int port_number;
162         unsigned int flags;
163         unsigned int max_transfer_bytes;
164
165         /*  This id is unique for each channel and will correspond with */
166         /*  vendor specific data in the inquirydata */
167         unsigned long long unique_id;
168 } __packed;
169
170 /*  This structure is sent during the storage protocol negotiations. */
171 struct vmstorage_protocol_version {
172         /* Major (MSW) and minor (LSW) version numbers. */
173         unsigned short major_minor;
174
175         /*
176          * Revision number is auto-incremented whenever this file is changed
177          * (See FILL_VMSTOR_REVISION macro above).  Mismatch does not
178          * definitely indicate incompatibility--but it does indicate mismatched
179          * builds.
180          */
181         unsigned short revision;
182 } __packed;
183
184 /* Channel Property Flags */
185 #define STORAGE_CHANNEL_REMOVABLE_FLAG          0x1
186 #define STORAGE_CHANNEL_EMULATED_IDE_FLAG       0x2
187
188 struct vstor_packet {
189         /* Requested operation type */
190         enum vstor_packet_operation operation;
191
192         /*  Flags - see below for values */
193         unsigned int flags;
194
195         /* Status of the request returned from the server side. */
196         unsigned int status;
197
198         /* Data payload area */
199         union {
200                 /*
201                  * Structure used to forward SCSI commands from the
202                  * client to the server.
203                  */
204                 struct vmscsi_request vm_srb;
205
206                 /* Structure used to query channel properties. */
207                 struct vmstorage_channel_properties storage_channel_properties;
208
209                 /* Used during version negotiations. */
210                 struct vmstorage_protocol_version version;
211         };
212 } __packed;
213
214 /* Packet flags */
215 /*
216  * This flag indicates that the server should send back a completion for this
217  * packet.
218  */
219 #define REQUEST_COMPLETION_FLAG 0x1
220
221 /*  This is the set of flags that the vsc can set in any packets it sends */
222 #define VSC_LEGAL_FLAGS         (REQUEST_COMPLETION_FLAG)
223
224
225 /* Defines */
226
227 #define STORVSC_MAX_IO_REQUESTS                         128
228
229 /*
230  * In Hyper-V, each port/path/target maps to 1 scsi host adapter.  In
231  * reality, the path/target is not used (ie always set to 0) so our
232  * scsi host adapter essentially has 1 bus with 1 target that contains
233  * up to 256 luns.
234  */
235 #define STORVSC_MAX_LUNS_PER_TARGET                     64
236 #define STORVSC_MAX_TARGETS                             1
237 #define STORVSC_MAX_CHANNELS                            1
238 #define STORVSC_MAX_CMD_LEN                             16
239
240 /* Matches Windows-end */
241 enum storvsc_request_type {
242         WRITE_TYPE,
243         READ_TYPE,
244         UNKNOWN_TYPE,
245 };
246
247
248 struct hv_storvsc_request {
249         struct hv_device *device;
250
251         /* Synchronize the request/response if needed */
252         struct completion wait_event;
253
254         unsigned char *sense_buffer;
255         void *context;
256         void (*on_io_completion)(struct hv_storvsc_request *request);
257         struct hv_multipage_buffer data_buffer;
258
259         struct vstor_packet vstor_packet;
260 };
261
262
263 /* A storvsc device is a device object that contains a vmbus channel */
264 struct storvsc_device {
265         struct hv_device *device;
266
267         bool     destroy;
268         bool     drain_notify;
269         atomic_t num_outstanding_req;
270         struct Scsi_Host *host;
271
272         wait_queue_head_t waiting_to_drain;
273
274         /*
275          * Each unique Port/Path/Target represents 1 channel ie scsi
276          * controller. In reality, the pathid, targetid is always 0
277          * and the port is set by us
278          */
279         unsigned int port_number;
280         unsigned char path_id;
281         unsigned char target_id;
282
283         /* Used for vsc/vsp channel reset process */
284         struct hv_storvsc_request init_request;
285         struct hv_storvsc_request reset_request;
286 };
287
288 struct hv_host_device {
289         struct hv_device *dev;
290         struct kmem_cache *request_pool;
291         mempool_t *request_mempool;
292         unsigned int port;
293         unsigned char path;
294         unsigned char target;
295 };
296
297 struct storvsc_cmd_request {
298         struct list_head entry;
299         struct scsi_cmnd *cmd;
300
301         unsigned int bounce_sgl_count;
302         struct scatterlist *bounce_sgl;
303
304         struct hv_storvsc_request request;
305 };
306
307 struct storvsc_scan_work {
308         struct work_struct work;
309         struct Scsi_Host *host;
310         uint lun;
311 };
312
313 static void storvsc_bus_scan(struct work_struct *work)
314 {
315         struct storvsc_scan_work *wrk;
316         int id, order_id;
317
318         wrk = container_of(work, struct storvsc_scan_work, work);
319         for (id = 0; id < wrk->host->max_id; ++id) {
320                 if (wrk->host->reverse_ordering)
321                         order_id = wrk->host->max_id - id - 1;
322                 else
323                         order_id = id;
324
325                 scsi_scan_target(&wrk->host->shost_gendev, 0,
326                                 order_id, SCAN_WILD_CARD, 1);
327         }
328         kfree(wrk);
329 }
330
331 static inline struct storvsc_device *get_out_stor_device(
332                                         struct hv_device *device)
333 {
334         struct storvsc_device *stor_device;
335
336         stor_device = hv_get_drvdata(device);
337
338         if (stor_device && stor_device->destroy)
339                 stor_device = NULL;
340
341         return stor_device;
342 }
343
344
345 static inline void storvsc_wait_to_drain(struct storvsc_device *dev)
346 {
347         dev->drain_notify = true;
348         wait_event(dev->waiting_to_drain,
349                    atomic_read(&dev->num_outstanding_req) == 0);
350         dev->drain_notify = false;
351 }
352
353 static inline struct storvsc_device *get_in_stor_device(
354                                         struct hv_device *device)
355 {
356         struct storvsc_device *stor_device;
357
358         stor_device = hv_get_drvdata(device);
359
360         if (!stor_device)
361                 goto get_in_err;
362
363         /*
364          * If the device is being destroyed; allow incoming
365          * traffic only to cleanup outstanding requests.
366          */
367
368         if (stor_device->destroy  &&
369                 (atomic_read(&stor_device->num_outstanding_req) == 0))
370                 stor_device = NULL;
371
372 get_in_err:
373         return stor_device;
374
375 }
376
377 static int storvsc_channel_init(struct hv_device *device)
378 {
379         struct storvsc_device *stor_device;
380         struct hv_storvsc_request *request;
381         struct vstor_packet *vstor_packet;
382         int ret, t;
383
384         stor_device = get_out_stor_device(device);
385         if (!stor_device)
386                 return -ENODEV;
387
388         request = &stor_device->init_request;
389         vstor_packet = &request->vstor_packet;
390
391         /*
392          * Now, initiate the vsc/vsp initialization protocol on the open
393          * channel
394          */
395         memset(request, 0, sizeof(struct hv_storvsc_request));
396         init_completion(&request->wait_event);
397         vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION;
398         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
399
400         ret = vmbus_sendpacket(device->channel, vstor_packet,
401                                sizeof(struct vstor_packet),
402                                (unsigned long)request,
403                                VM_PKT_DATA_INBAND,
404                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
405         if (ret != 0)
406                 goto cleanup;
407
408         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
409         if (t == 0) {
410                 ret = -ETIMEDOUT;
411                 goto cleanup;
412         }
413
414         if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
415             vstor_packet->status != 0)
416                 goto cleanup;
417
418
419         /* reuse the packet for version range supported */
420         memset(vstor_packet, 0, sizeof(struct vstor_packet));
421         vstor_packet->operation = VSTOR_OPERATION_QUERY_PROTOCOL_VERSION;
422         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
423
424         vstor_packet->version.major_minor = VMSTOR_PROTOCOL_VERSION_CURRENT;
425         FILL_VMSTOR_REVISION(vstor_packet->version.revision);
426
427         ret = vmbus_sendpacket(device->channel, vstor_packet,
428                                sizeof(struct vstor_packet),
429                                (unsigned long)request,
430                                VM_PKT_DATA_INBAND,
431                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
432         if (ret != 0)
433                 goto cleanup;
434
435         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
436         if (t == 0) {
437                 ret = -ETIMEDOUT;
438                 goto cleanup;
439         }
440
441         if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
442             vstor_packet->status != 0)
443                 goto cleanup;
444
445
446         memset(vstor_packet, 0, sizeof(struct vstor_packet));
447         vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES;
448         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
449         vstor_packet->storage_channel_properties.port_number =
450                                         stor_device->port_number;
451
452         ret = vmbus_sendpacket(device->channel, vstor_packet,
453                                sizeof(struct vstor_packet),
454                                (unsigned long)request,
455                                VM_PKT_DATA_INBAND,
456                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
457
458         if (ret != 0)
459                 goto cleanup;
460
461         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
462         if (t == 0) {
463                 ret = -ETIMEDOUT;
464                 goto cleanup;
465         }
466
467         if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
468             vstor_packet->status != 0)
469                 goto cleanup;
470
471         stor_device->path_id = vstor_packet->storage_channel_properties.path_id;
472         stor_device->target_id
473                 = vstor_packet->storage_channel_properties.target_id;
474
475         memset(vstor_packet, 0, sizeof(struct vstor_packet));
476         vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION;
477         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
478
479         ret = vmbus_sendpacket(device->channel, vstor_packet,
480                                sizeof(struct vstor_packet),
481                                (unsigned long)request,
482                                VM_PKT_DATA_INBAND,
483                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
484
485         if (ret != 0)
486                 goto cleanup;
487
488         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
489         if (t == 0) {
490                 ret = -ETIMEDOUT;
491                 goto cleanup;
492         }
493
494         if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
495             vstor_packet->status != 0)
496                 goto cleanup;
497
498
499 cleanup:
500         return ret;
501 }
502
503 static void storvsc_on_io_completion(struct hv_device *device,
504                                   struct vstor_packet *vstor_packet,
505                                   struct hv_storvsc_request *request)
506 {
507         struct storvsc_device *stor_device;
508         struct vstor_packet *stor_pkt;
509
510         stor_device = hv_get_drvdata(device);
511         stor_pkt = &request->vstor_packet;
512
513         /*
514          * The current SCSI handling on the host side does
515          * not correctly handle:
516          * INQUIRY command with page code parameter set to 0x80
517          * MODE_SENSE command with cmd[2] == 0x1c
518          *
519          * Setup srb and scsi status so this won't be fatal.
520          * We do this so we can distinguish truly fatal failues
521          * (srb status == 0x4) and off-line the device in that case.
522          */
523
524         if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) ||
525                 (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) {
526                 vstor_packet->vm_srb.scsi_status = 0;
527                 vstor_packet->vm_srb.srb_status = 0x1;
528         }
529
530
531         /* Copy over the status...etc */
532         stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status;
533         stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status;
534         stor_pkt->vm_srb.sense_info_length =
535         vstor_packet->vm_srb.sense_info_length;
536
537         if (vstor_packet->vm_srb.scsi_status != 0 ||
538                 vstor_packet->vm_srb.srb_status != 1){
539                 dev_warn(&device->device,
540                          "cmd 0x%x scsi status 0x%x srb status 0x%x\n",
541                          stor_pkt->vm_srb.cdb[0],
542                          vstor_packet->vm_srb.scsi_status,
543                          vstor_packet->vm_srb.srb_status);
544         }
545
546         if ((vstor_packet->vm_srb.scsi_status & 0xFF) == 0x02) {
547                 /* CHECK_CONDITION */
548                 if (vstor_packet->vm_srb.srb_status & 0x80) {
549                         /* autosense data available */
550                         dev_warn(&device->device,
551                                  "stor pkt %p autosense data valid - len %d\n",
552                                  request,
553                                  vstor_packet->vm_srb.sense_info_length);
554
555                         memcpy(request->sense_buffer,
556                                vstor_packet->vm_srb.sense_data,
557                                vstor_packet->vm_srb.sense_info_length);
558
559                 }
560         }
561
562         stor_pkt->vm_srb.data_transfer_length =
563         vstor_packet->vm_srb.data_transfer_length;
564
565         request->on_io_completion(request);
566
567         if (atomic_dec_and_test(&stor_device->num_outstanding_req) &&
568                 stor_device->drain_notify)
569                 wake_up(&stor_device->waiting_to_drain);
570
571
572 }
573
574 static void storvsc_on_receive(struct hv_device *device,
575                              struct vstor_packet *vstor_packet,
576                              struct hv_storvsc_request *request)
577 {
578         struct storvsc_scan_work *work;
579         struct storvsc_device *stor_device;
580
581         switch (vstor_packet->operation) {
582         case VSTOR_OPERATION_COMPLETE_IO:
583                 storvsc_on_io_completion(device, vstor_packet, request);
584                 break;
585
586         case VSTOR_OPERATION_REMOVE_DEVICE:
587         case VSTOR_OPERATION_ENUMERATE_BUS:
588                 stor_device = get_in_stor_device(device);
589                 work = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC);
590                 if (!work)
591                         return;
592
593                 INIT_WORK(&work->work, storvsc_bus_scan);
594                 work->host = stor_device->host;
595                 schedule_work(&work->work);
596                 break;
597
598         default:
599                 break;
600         }
601 }
602
603 static void storvsc_on_channel_callback(void *context)
604 {
605         struct hv_device *device = (struct hv_device *)context;
606         struct storvsc_device *stor_device;
607         u32 bytes_recvd;
608         u64 request_id;
609         unsigned char packet[ALIGN(sizeof(struct vstor_packet), 8)];
610         struct hv_storvsc_request *request;
611         int ret;
612
613
614         stor_device = get_in_stor_device(device);
615         if (!stor_device)
616                 return;
617
618         do {
619                 ret = vmbus_recvpacket(device->channel, packet,
620                                        ALIGN(sizeof(struct vstor_packet), 8),
621                                        &bytes_recvd, &request_id);
622                 if (ret == 0 && bytes_recvd > 0) {
623
624                         request = (struct hv_storvsc_request *)
625                                         (unsigned long)request_id;
626
627                         if ((request == &stor_device->init_request) ||
628                             (request == &stor_device->reset_request)) {
629
630                                 memcpy(&request->vstor_packet, packet,
631                                        sizeof(struct vstor_packet));
632                                 complete(&request->wait_event);
633                         } else {
634                                 storvsc_on_receive(device,
635                                                 (struct vstor_packet *)packet,
636                                                 request);
637                         }
638                 } else {
639                         break;
640                 }
641         } while (1);
642
643         return;
644 }
645
646 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size)
647 {
648         struct vmstorage_channel_properties props;
649         int ret;
650
651         memset(&props, 0, sizeof(struct vmstorage_channel_properties));
652
653         /* Open the channel */
654         ret = vmbus_open(device->channel,
655                          ring_size,
656                          ring_size,
657                          (void *)&props,
658                          sizeof(struct vmstorage_channel_properties),
659                          storvsc_on_channel_callback, device);
660
661         if (ret != 0)
662                 return ret;
663
664         ret = storvsc_channel_init(device);
665
666         return ret;
667 }
668
669 static int storvsc_dev_remove(struct hv_device *device)
670 {
671         struct storvsc_device *stor_device;
672         unsigned long flags;
673
674         stor_device = hv_get_drvdata(device);
675
676         spin_lock_irqsave(&device->channel->inbound_lock, flags);
677         stor_device->destroy = true;
678         spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
679
680         /*
681          * At this point, all outbound traffic should be disable. We
682          * only allow inbound traffic (responses) to proceed so that
683          * outstanding requests can be completed.
684          */
685
686         storvsc_wait_to_drain(stor_device);
687
688         /*
689          * Since we have already drained, we don't need to busy wait
690          * as was done in final_release_stor_device()
691          * Note that we cannot set the ext pointer to NULL until
692          * we have drained - to drain the outgoing packets, we need to
693          * allow incoming packets.
694          */
695         spin_lock_irqsave(&device->channel->inbound_lock, flags);
696         hv_set_drvdata(device, NULL);
697         spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
698
699         /* Close the channel */
700         vmbus_close(device->channel);
701
702         kfree(stor_device);
703         return 0;
704 }
705
706 static int storvsc_do_io(struct hv_device *device,
707                               struct hv_storvsc_request *request)
708 {
709         struct storvsc_device *stor_device;
710         struct vstor_packet *vstor_packet;
711         int ret = 0;
712
713         vstor_packet = &request->vstor_packet;
714         stor_device = get_out_stor_device(device);
715
716         if (!stor_device)
717                 return -ENODEV;
718
719
720         request->device  = device;
721
722
723         vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
724
725         vstor_packet->vm_srb.length = sizeof(struct vmscsi_request);
726
727
728         vstor_packet->vm_srb.sense_info_length = SENSE_BUFFER_SIZE;
729
730
731         vstor_packet->vm_srb.data_transfer_length =
732         request->data_buffer.len;
733
734         vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
735
736         if (request->data_buffer.len) {
737                 ret = vmbus_sendpacket_multipagebuffer(device->channel,
738                                 &request->data_buffer,
739                                 vstor_packet,
740                                 sizeof(struct vstor_packet),
741                                 (unsigned long)request);
742         } else {
743                 ret = vmbus_sendpacket(device->channel, vstor_packet,
744                                sizeof(struct vstor_packet),
745                                (unsigned long)request,
746                                VM_PKT_DATA_INBAND,
747                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
748         }
749
750         if (ret != 0)
751                 return ret;
752
753         atomic_inc(&stor_device->num_outstanding_req);
754
755         return ret;
756 }
757
758 static void storvsc_get_ide_info(struct hv_device *dev, int *target, int *path)
759 {
760         *target =
761                 dev->dev_instance.b[5] << 8 | dev->dev_instance.b[4];
762
763         *path =
764                 dev->dev_instance.b[3] << 24 |
765                 dev->dev_instance.b[2] << 16 |
766                 dev->dev_instance.b[1] << 8  | dev->dev_instance.b[0];
767 }
768
769
770 static int storvsc_device_alloc(struct scsi_device *sdevice)
771 {
772         /*
773          * This enables luns to be located sparsely. Otherwise, we may not
774          * discovered them.
775          */
776         sdevice->sdev_bflags |= BLIST_SPARSELUN | BLIST_LARGELUN;
777         return 0;
778 }
779
780 static int storvsc_merge_bvec(struct request_queue *q,
781                               struct bvec_merge_data *bmd, struct bio_vec *bvec)
782 {
783         /* checking done by caller. */
784         return bvec->bv_len;
785 }
786
787 static int storvsc_device_configure(struct scsi_device *sdevice)
788 {
789         scsi_adjust_queue_depth(sdevice, MSG_SIMPLE_TAG,
790                                 STORVSC_MAX_IO_REQUESTS);
791
792         blk_queue_max_segment_size(sdevice->request_queue, PAGE_SIZE);
793
794         blk_queue_merge_bvec(sdevice->request_queue, storvsc_merge_bvec);
795
796         blk_queue_bounce_limit(sdevice->request_queue, BLK_BOUNCE_ANY);
797
798         return 0;
799 }
800
801 static void destroy_bounce_buffer(struct scatterlist *sgl,
802                                   unsigned int sg_count)
803 {
804         int i;
805         struct page *page_buf;
806
807         for (i = 0; i < sg_count; i++) {
808                 page_buf = sg_page((&sgl[i]));
809                 if (page_buf != NULL)
810                         __free_page(page_buf);
811         }
812
813         kfree(sgl);
814 }
815
816 static int do_bounce_buffer(struct scatterlist *sgl, unsigned int sg_count)
817 {
818         int i;
819
820         /* No need to check */
821         if (sg_count < 2)
822                 return -1;
823
824         /* We have at least 2 sg entries */
825         for (i = 0; i < sg_count; i++) {
826                 if (i == 0) {
827                         /* make sure 1st one does not have hole */
828                         if (sgl[i].offset + sgl[i].length != PAGE_SIZE)
829                                 return i;
830                 } else if (i == sg_count - 1) {
831                         /* make sure last one does not have hole */
832                         if (sgl[i].offset != 0)
833                                 return i;
834                 } else {
835                         /* make sure no hole in the middle */
836                         if (sgl[i].length != PAGE_SIZE || sgl[i].offset != 0)
837                                 return i;
838                 }
839         }
840         return -1;
841 }
842
843 static struct scatterlist *create_bounce_buffer(struct scatterlist *sgl,
844                                                 unsigned int sg_count,
845                                                 unsigned int len)
846 {
847         int i;
848         int num_pages;
849         struct scatterlist *bounce_sgl;
850         struct page *page_buf;
851
852         num_pages = ALIGN(len, PAGE_SIZE) >> PAGE_SHIFT;
853
854         bounce_sgl = kcalloc(num_pages, sizeof(struct scatterlist), GFP_ATOMIC);
855         if (!bounce_sgl)
856                 return NULL;
857
858         for (i = 0; i < num_pages; i++) {
859                 page_buf = alloc_page(GFP_ATOMIC);
860                 if (!page_buf)
861                         goto cleanup;
862                 sg_set_page(&bounce_sgl[i], page_buf, 0, 0);
863         }
864
865         return bounce_sgl;
866
867 cleanup:
868         destroy_bounce_buffer(bounce_sgl, num_pages);
869         return NULL;
870 }
871
872
873 /* Assume the original sgl has enough room */
874 static unsigned int copy_from_bounce_buffer(struct scatterlist *orig_sgl,
875                                             struct scatterlist *bounce_sgl,
876                                             unsigned int orig_sgl_count)
877 {
878         int i;
879         int j = 0;
880         unsigned long src, dest;
881         unsigned int srclen, destlen, copylen;
882         unsigned int total_copied = 0;
883         unsigned long bounce_addr = 0;
884         unsigned long dest_addr = 0;
885         unsigned long flags;
886
887         local_irq_save(flags);
888
889         for (i = 0; i < orig_sgl_count; i++) {
890                 dest_addr = (unsigned long)kmap_atomic(sg_page((&orig_sgl[i])),
891                                         KM_IRQ0) + orig_sgl[i].offset;
892                 dest = dest_addr;
893                 destlen = orig_sgl[i].length;
894
895                 if (bounce_addr == 0)
896                         bounce_addr =
897                         (unsigned long)kmap_atomic(sg_page((&bounce_sgl[j])),
898                                                         KM_IRQ0);
899
900                 while (destlen) {
901                         src = bounce_addr + bounce_sgl[j].offset;
902                         srclen = bounce_sgl[j].length - bounce_sgl[j].offset;
903
904                         copylen = min(srclen, destlen);
905                         memcpy((void *)dest, (void *)src, copylen);
906
907                         total_copied += copylen;
908                         bounce_sgl[j].offset += copylen;
909                         destlen -= copylen;
910                         dest += copylen;
911
912                         if (bounce_sgl[j].offset == bounce_sgl[j].length) {
913                                 /* full */
914                                 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
915                                 j++;
916
917                                 /* if we need to use another bounce buffer */
918                                 if (destlen || i != orig_sgl_count - 1)
919                                         bounce_addr =
920                                         (unsigned long)kmap_atomic(
921                                         sg_page((&bounce_sgl[j])), KM_IRQ0);
922                         } else if (destlen == 0 && i == orig_sgl_count - 1) {
923                                 /* unmap the last bounce that is < PAGE_SIZE */
924                                 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
925                         }
926                 }
927
928                 kunmap_atomic((void *)(dest_addr - orig_sgl[i].offset),
929                               KM_IRQ0);
930         }
931
932         local_irq_restore(flags);
933
934         return total_copied;
935 }
936
937
938 /* Assume the bounce_sgl has enough room ie using the create_bounce_buffer() */
939 static unsigned int copy_to_bounce_buffer(struct scatterlist *orig_sgl,
940                                           struct scatterlist *bounce_sgl,
941                                           unsigned int orig_sgl_count)
942 {
943         int i;
944         int j = 0;
945         unsigned long src, dest;
946         unsigned int srclen, destlen, copylen;
947         unsigned int total_copied = 0;
948         unsigned long bounce_addr = 0;
949         unsigned long src_addr = 0;
950         unsigned long flags;
951
952         local_irq_save(flags);
953
954         for (i = 0; i < orig_sgl_count; i++) {
955                 src_addr = (unsigned long)kmap_atomic(sg_page((&orig_sgl[i])),
956                                 KM_IRQ0) + orig_sgl[i].offset;
957                 src = src_addr;
958                 srclen = orig_sgl[i].length;
959
960                 if (bounce_addr == 0)
961                         bounce_addr =
962                         (unsigned long)kmap_atomic(sg_page((&bounce_sgl[j])),
963                                                 KM_IRQ0);
964
965                 while (srclen) {
966                         /* assume bounce offset always == 0 */
967                         dest = bounce_addr + bounce_sgl[j].length;
968                         destlen = PAGE_SIZE - bounce_sgl[j].length;
969
970                         copylen = min(srclen, destlen);
971                         memcpy((void *)dest, (void *)src, copylen);
972
973                         total_copied += copylen;
974                         bounce_sgl[j].length += copylen;
975                         srclen -= copylen;
976                         src += copylen;
977
978                         if (bounce_sgl[j].length == PAGE_SIZE) {
979                                 /* full..move to next entry */
980                                 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
981                                 j++;
982
983                                 /* if we need to use another bounce buffer */
984                                 if (srclen || i != orig_sgl_count - 1)
985                                         bounce_addr =
986                                         (unsigned long)kmap_atomic(
987                                         sg_page((&bounce_sgl[j])), KM_IRQ0);
988
989                         } else if (srclen == 0 && i == orig_sgl_count - 1) {
990                                 /* unmap the last bounce that is < PAGE_SIZE */
991                                 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
992                         }
993                 }
994
995                 kunmap_atomic((void *)(src_addr - orig_sgl[i].offset), KM_IRQ0);
996         }
997
998         local_irq_restore(flags);
999
1000         return total_copied;
1001 }
1002
1003
1004 static int storvsc_remove(struct hv_device *dev)
1005 {
1006         struct storvsc_device *stor_device = hv_get_drvdata(dev);
1007         struct Scsi_Host *host = stor_device->host;
1008         struct hv_host_device *host_dev = shost_priv(host);
1009
1010         scsi_remove_host(host);
1011
1012         scsi_host_put(host);
1013
1014         storvsc_dev_remove(dev);
1015         if (host_dev->request_pool) {
1016                 mempool_destroy(host_dev->request_mempool);
1017                 kmem_cache_destroy(host_dev->request_pool);
1018                 host_dev->request_pool = NULL;
1019                 host_dev->request_mempool = NULL;
1020         }
1021         return 0;
1022 }
1023
1024
1025 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev,
1026                            sector_t capacity, int *info)
1027 {
1028         sector_t nsect = capacity;
1029         sector_t cylinders = nsect;
1030         int heads, sectors_pt;
1031
1032         /*
1033          * We are making up these values; let us keep it simple.
1034          */
1035         heads = 0xff;
1036         sectors_pt = 0x3f;      /* Sectors per track */
1037         sector_div(cylinders, heads * sectors_pt);
1038         if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
1039                 cylinders = 0xffff;
1040
1041         info[0] = heads;
1042         info[1] = sectors_pt;
1043         info[2] = (int)cylinders;
1044
1045         return 0;
1046 }
1047
1048 static int storvsc_host_reset(struct hv_device *device)
1049 {
1050         struct storvsc_device *stor_device;
1051         struct hv_storvsc_request *request;
1052         struct vstor_packet *vstor_packet;
1053         int ret, t;
1054
1055
1056         stor_device = get_out_stor_device(device);
1057         if (!stor_device)
1058                 return FAILED;
1059
1060         request = &stor_device->reset_request;
1061         vstor_packet = &request->vstor_packet;
1062
1063         init_completion(&request->wait_event);
1064
1065         vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
1066         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
1067         vstor_packet->vm_srb.path_id = stor_device->path_id;
1068
1069         ret = vmbus_sendpacket(device->channel, vstor_packet,
1070                                sizeof(struct vstor_packet),
1071                                (unsigned long)&stor_device->reset_request,
1072                                VM_PKT_DATA_INBAND,
1073                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1074         if (ret != 0)
1075                 return FAILED;
1076
1077         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
1078         if (t == 0)
1079                 return TIMEOUT_ERROR;
1080
1081
1082         /*
1083          * At this point, all outstanding requests in the adapter
1084          * should have been flushed out and return to us
1085          */
1086
1087         return SUCCESS;
1088 }
1089
1090
1091 /*
1092  * storvsc_host_reset_handler - Reset the scsi HBA
1093  */
1094 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
1095 {
1096         struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
1097         struct hv_device *dev = host_dev->dev;
1098
1099         return storvsc_host_reset(dev);
1100 }
1101
1102
1103 /*
1104  * storvsc_command_completion - Command completion processing
1105  */
1106 static void storvsc_command_completion(struct hv_storvsc_request *request)
1107 {
1108         struct storvsc_cmd_request *cmd_request =
1109                 (struct storvsc_cmd_request *)request->context;
1110         struct scsi_cmnd *scmnd = cmd_request->cmd;
1111         struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
1112         void (*scsi_done_fn)(struct scsi_cmnd *);
1113         struct scsi_sense_hdr sense_hdr;
1114         struct vmscsi_request *vm_srb;
1115
1116         vm_srb = &request->vstor_packet.vm_srb;
1117         if (cmd_request->bounce_sgl_count) {
1118                 if (vm_srb->data_in == READ_TYPE) {
1119                         copy_from_bounce_buffer(scsi_sglist(scmnd),
1120                                         cmd_request->bounce_sgl,
1121                                         scsi_sg_count(scmnd));
1122                         destroy_bounce_buffer(cmd_request->bounce_sgl,
1123                                         cmd_request->bounce_sgl_count);
1124                 }
1125         }
1126
1127         /*
1128          * If there is an error; offline the device since all
1129          * error recovery strategies would have already been
1130          * deployed on the host side.
1131          */
1132         if (vm_srb->srb_status == 0x4)
1133                 scmnd->result = DID_TARGET_FAILURE << 16;
1134         else
1135                 scmnd->result = vm_srb->scsi_status;
1136
1137         if (scmnd->result) {
1138                 if (scsi_normalize_sense(scmnd->sense_buffer,
1139                                 SCSI_SENSE_BUFFERSIZE, &sense_hdr))
1140                         scsi_print_sense_hdr("storvsc", &sense_hdr);
1141         }
1142
1143         scsi_set_resid(scmnd,
1144                 request->data_buffer.len -
1145                 vm_srb->data_transfer_length);
1146
1147         scsi_done_fn = scmnd->scsi_done;
1148
1149         scmnd->host_scribble = NULL;
1150         scmnd->scsi_done = NULL;
1151
1152         scsi_done_fn(scmnd);
1153
1154         mempool_free(cmd_request, host_dev->request_mempool);
1155 }
1156
1157 static bool storvsc_check_scsi_cmd(struct scsi_cmnd *scmnd)
1158 {
1159         bool allowed = true;
1160         u8 scsi_op = scmnd->cmnd[0];
1161
1162         switch (scsi_op) {
1163         /* smartd sends this command, which will offline the device */
1164         case SET_WINDOW:
1165                 scmnd->result = ILLEGAL_REQUEST << 16;
1166                 allowed = false;
1167                 break;
1168         default:
1169                 break;
1170         }
1171         return allowed;
1172 }
1173
1174 /*
1175  * storvsc_queuecommand - Initiate command processing
1176  */
1177 static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
1178 {
1179         int ret;
1180         struct hv_host_device *host_dev = shost_priv(host);
1181         struct hv_device *dev = host_dev->dev;
1182         struct hv_storvsc_request *request;
1183         struct storvsc_cmd_request *cmd_request;
1184         unsigned int request_size = 0;
1185         int i;
1186         struct scatterlist *sgl;
1187         unsigned int sg_count = 0;
1188         struct vmscsi_request *vm_srb;
1189
1190         if (storvsc_check_scsi_cmd(scmnd) == false) {
1191                 scmnd->scsi_done(scmnd);
1192                 return 0;
1193         }
1194
1195         /* If retrying, no need to prep the cmd */
1196         if (scmnd->host_scribble) {
1197
1198                 cmd_request =
1199                         (struct storvsc_cmd_request *)scmnd->host_scribble;
1200
1201                 goto retry_request;
1202         }
1203
1204         request_size = sizeof(struct storvsc_cmd_request);
1205
1206         cmd_request = mempool_alloc(host_dev->request_mempool,
1207                                        GFP_ATOMIC);
1208         if (!cmd_request)
1209                 return SCSI_MLQUEUE_DEVICE_BUSY;
1210
1211         memset(cmd_request, 0, sizeof(struct storvsc_cmd_request));
1212
1213         /* Setup the cmd request */
1214         cmd_request->bounce_sgl_count = 0;
1215         cmd_request->bounce_sgl = NULL;
1216         cmd_request->cmd = scmnd;
1217
1218         scmnd->host_scribble = (unsigned char *)cmd_request;
1219
1220         request = &cmd_request->request;
1221         vm_srb = &request->vstor_packet.vm_srb;
1222
1223
1224         /* Build the SRB */
1225         switch (scmnd->sc_data_direction) {
1226         case DMA_TO_DEVICE:
1227                 vm_srb->data_in = WRITE_TYPE;
1228                 break;
1229         case DMA_FROM_DEVICE:
1230                 vm_srb->data_in = READ_TYPE;
1231                 break;
1232         default:
1233                 vm_srb->data_in = UNKNOWN_TYPE;
1234                 break;
1235         }
1236
1237         request->on_io_completion = storvsc_command_completion;
1238         request->context = cmd_request;/* scmnd; */
1239
1240         vm_srb->port_number = host_dev->port;
1241         vm_srb->path_id = scmnd->device->channel;
1242         vm_srb->target_id = scmnd->device->id;
1243         vm_srb->lun = scmnd->device->lun;
1244
1245         vm_srb->cdb_length = scmnd->cmd_len;
1246
1247         memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
1248
1249         request->sense_buffer = scmnd->sense_buffer;
1250
1251
1252         request->data_buffer.len = scsi_bufflen(scmnd);
1253         if (scsi_sg_count(scmnd)) {
1254                 sgl = (struct scatterlist *)scsi_sglist(scmnd);
1255                 sg_count = scsi_sg_count(scmnd);
1256
1257                 /* check if we need to bounce the sgl */
1258                 if (do_bounce_buffer(sgl, scsi_sg_count(scmnd)) != -1) {
1259                         cmd_request->bounce_sgl =
1260                                 create_bounce_buffer(sgl, scsi_sg_count(scmnd),
1261                                                      scsi_bufflen(scmnd));
1262                         if (!cmd_request->bounce_sgl) {
1263                                 scmnd->host_scribble = NULL;
1264                                 mempool_free(cmd_request,
1265                                                 host_dev->request_mempool);
1266
1267                                 return SCSI_MLQUEUE_HOST_BUSY;
1268                         }
1269
1270                         cmd_request->bounce_sgl_count =
1271                                 ALIGN(scsi_bufflen(scmnd), PAGE_SIZE) >>
1272                                         PAGE_SHIFT;
1273
1274                         if (vm_srb->data_in == WRITE_TYPE)
1275                                 copy_to_bounce_buffer(sgl,
1276                                         cmd_request->bounce_sgl,
1277                                         scsi_sg_count(scmnd));
1278
1279                         sgl = cmd_request->bounce_sgl;
1280                         sg_count = cmd_request->bounce_sgl_count;
1281                 }
1282
1283                 request->data_buffer.offset = sgl[0].offset;
1284
1285                 for (i = 0; i < sg_count; i++)
1286                         request->data_buffer.pfn_array[i] =
1287                                 page_to_pfn(sg_page((&sgl[i])));
1288
1289         } else if (scsi_sglist(scmnd)) {
1290                 request->data_buffer.offset =
1291                         virt_to_phys(scsi_sglist(scmnd)) & (PAGE_SIZE-1);
1292                 request->data_buffer.pfn_array[0] =
1293                         virt_to_phys(scsi_sglist(scmnd)) >> PAGE_SHIFT;
1294         }
1295
1296 retry_request:
1297         /* Invokes the vsc to start an IO */
1298         ret = storvsc_do_io(dev, &cmd_request->request);
1299
1300         if (ret == -EAGAIN) {
1301                 /* no more space */
1302
1303                 if (cmd_request->bounce_sgl_count)
1304                         destroy_bounce_buffer(cmd_request->bounce_sgl,
1305                                         cmd_request->bounce_sgl_count);
1306
1307                 mempool_free(cmd_request, host_dev->request_mempool);
1308
1309                 scmnd->host_scribble = NULL;
1310
1311                 ret = SCSI_MLQUEUE_DEVICE_BUSY;
1312         }
1313
1314         return ret;
1315 }
1316
1317 /* Scsi driver */
1318 static struct scsi_host_template scsi_driver = {
1319         .module =               THIS_MODULE,
1320         .name =                 "storvsc_host_t",
1321         .bios_param =           storvsc_get_chs,
1322         .queuecommand =         storvsc_queuecommand,
1323         .eh_host_reset_handler =        storvsc_host_reset_handler,
1324         .slave_alloc =          storvsc_device_alloc,
1325         .slave_configure =      storvsc_device_configure,
1326         .cmd_per_lun =          1,
1327         /* 64 max_queue * 1 target */
1328         .can_queue =            STORVSC_MAX_IO_REQUESTS*STORVSC_MAX_TARGETS,
1329         .this_id =              -1,
1330         /* no use setting to 0 since ll_blk_rw reset it to 1 */
1331         /* currently 32 */
1332         .sg_tablesize =         MAX_MULTIPAGE_BUFFER_COUNT,
1333         /*
1334          * ENABLE_CLUSTERING allows mutiple physically contig bio_vecs to merge
1335          * into 1 sg element. If set, we must limit the max_segment_size to
1336          * PAGE_SIZE, otherwise we may get 1 sg element that represents
1337          * multiple
1338          */
1339         /* physically contig pfns (ie sg[x].length > PAGE_SIZE). */
1340         .use_clustering =       ENABLE_CLUSTERING,
1341         /* Make sure we dont get a sg segment crosses a page boundary */
1342         .dma_boundary =         PAGE_SIZE-1,
1343 };
1344
1345 enum {
1346         SCSI_GUID,
1347         IDE_GUID,
1348 };
1349
1350 static const struct hv_vmbus_device_id id_table[] = {
1351         /* SCSI guid */
1352         { VMBUS_DEVICE(0xd9, 0x63, 0x61, 0xba, 0xa1, 0x04, 0x29, 0x4d,
1353                        0xb6, 0x05, 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f)
1354           .driver_data = SCSI_GUID },
1355         /* IDE guid */
1356         { VMBUS_DEVICE(0x32, 0x26, 0x41, 0x32, 0xcb, 0x86, 0xa2, 0x44,
1357                        0x9b, 0x5c, 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5)
1358           .driver_data = IDE_GUID },
1359         { },
1360 };
1361
1362 MODULE_DEVICE_TABLE(vmbus, id_table);
1363
1364
1365 /*
1366  * storvsc_probe - Add a new device for this driver
1367  */
1368
1369 static int storvsc_probe(struct hv_device *device,
1370                         const struct hv_vmbus_device_id *dev_id)
1371 {
1372         int ret;
1373         int number = STORVSC_MIN_BUF_NR;
1374         struct Scsi_Host *host;
1375         struct hv_host_device *host_dev;
1376         bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false);
1377         int path = 0;
1378         int target = 0;
1379         struct storvsc_device *stor_device;
1380
1381         host = scsi_host_alloc(&scsi_driver,
1382                                sizeof(struct hv_host_device));
1383         if (!host)
1384                 return -ENOMEM;
1385
1386         host_dev = shost_priv(host);
1387         memset(host_dev, 0, sizeof(struct hv_host_device));
1388
1389         host_dev->port = host->host_no;
1390         host_dev->dev = device;
1391
1392         host_dev->request_pool =
1393                                 kmem_cache_create(dev_name(&device->device),
1394                                         sizeof(struct storvsc_cmd_request), 0,
1395                                         SLAB_HWCACHE_ALIGN, NULL);
1396
1397         if (!host_dev->request_pool) {
1398                 scsi_host_put(host);
1399                 return -ENOMEM;
1400         }
1401
1402         host_dev->request_mempool = mempool_create(number, mempool_alloc_slab,
1403                                                 mempool_free_slab,
1404                                                 host_dev->request_pool);
1405
1406         if (!host_dev->request_mempool) {
1407                 ret = -ENOMEM;
1408                 goto err_out0;
1409         }
1410
1411         stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
1412         if (!stor_device) {
1413                 ret = -ENOMEM;
1414                 goto err_out1;
1415         }
1416
1417         stor_device->destroy = false;
1418         init_waitqueue_head(&stor_device->waiting_to_drain);
1419         stor_device->device = device;
1420         stor_device->host = host;
1421         hv_set_drvdata(device, stor_device);
1422
1423         stor_device->port_number = host->host_no;
1424         ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size);
1425         if (ret)
1426                 goto err_out2;
1427
1428         if (dev_is_ide)
1429                 storvsc_get_ide_info(device, &target, &path);
1430
1431         host_dev->path = stor_device->path_id;
1432         host_dev->target = stor_device->target_id;
1433
1434         /* max # of devices per target */
1435         host->max_lun = STORVSC_MAX_LUNS_PER_TARGET;
1436         /* max # of targets per channel */
1437         host->max_id = STORVSC_MAX_TARGETS;
1438         /* max # of channels */
1439         host->max_channel = STORVSC_MAX_CHANNELS - 1;
1440         /* max cmd length */
1441         host->max_cmd_len = STORVSC_MAX_CMD_LEN;
1442
1443         /* Register the HBA and start the scsi bus scan */
1444         ret = scsi_add_host(host, &device->device);
1445         if (ret != 0)
1446                 goto err_out3;
1447
1448         if (!dev_is_ide) {
1449                 scsi_scan_host(host);
1450                 return 0;
1451         }
1452         ret = scsi_add_device(host, 0, target, 0);
1453         if (ret) {
1454                 scsi_remove_host(host);
1455                 goto err_out3;
1456         }
1457         return 0;
1458
1459 err_out3:
1460         /*
1461          * Once we have connected with the host, we would need to
1462          * to invoke storvsc_dev_remove() to rollback this state and
1463          * this call also frees up the stor_device; hence the jump around
1464          * err_out2 label.
1465          */
1466         storvsc_dev_remove(device);
1467         goto err_out1;
1468
1469 err_out2:
1470         kfree(stor_device);
1471
1472 err_out1:
1473         mempool_destroy(host_dev->request_mempool);
1474
1475 err_out0:
1476         kmem_cache_destroy(host_dev->request_pool);
1477         scsi_host_put(host);
1478         return ret;
1479 }
1480
1481 /* The one and only one */
1482
1483 static struct hv_driver storvsc_drv = {
1484         .name = KBUILD_MODNAME,
1485         .id_table = id_table,
1486         .probe = storvsc_probe,
1487         .remove = storvsc_remove,
1488 };
1489
1490 static int __init storvsc_drv_init(void)
1491 {
1492         u32 max_outstanding_req_per_channel;
1493
1494         /*
1495          * Divide the ring buffer data size (which is 1 page less
1496          * than the ring buffer size since that page is reserved for
1497          * the ring buffer indices) by the max request size (which is
1498          * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
1499          */
1500         max_outstanding_req_per_channel =
1501                 ((storvsc_ringbuffer_size - PAGE_SIZE) /
1502                 ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
1503                 sizeof(struct vstor_packet) + sizeof(u64),
1504                 sizeof(u64)));
1505
1506         if (max_outstanding_req_per_channel <
1507             STORVSC_MAX_IO_REQUESTS)
1508                 return -EINVAL;
1509
1510         return vmbus_driver_register(&storvsc_drv);
1511 }
1512
1513 static void __exit storvsc_drv_exit(void)
1514 {
1515         vmbus_driver_unregister(&storvsc_drv);
1516 }
1517
1518 MODULE_LICENSE("GPL");
1519 MODULE_VERSION(HV_DRV_VERSION);
1520 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
1521 module_init(storvsc_drv_init);
1522 module_exit(storvsc_drv_exit);