2 * Copyright (c) 2009, Microsoft Corporation.
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.
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
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.
18 * Haiyang Zhang <haiyangz@microsoft.com>
19 * Hank Janssen <hjanssen@microsoft.com>
20 * K. Y. Srinivasan <kys@microsoft.com>
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>
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 <linux/blkdev.h>
37 #include <scsi/scsi.h>
38 #include <scsi/scsi_cmnd.h>
39 #include <scsi/scsi_host.h>
40 #include <scsi/scsi_device.h>
41 #include <scsi/scsi_tcq.h>
42 #include <scsi/scsi_eh.h>
43 #include <scsi/scsi_devinfo.h>
44 #include <scsi/scsi_dbg.h>
47 * All wire protocol details (storage protocol between the guest and the host)
48 * are consolidated here.
50 * Begin protocol definitions.
56 * V1 RC < 2008/1/31: 1.0
57 * V1 RC > 2008/1/31: 2.0
63 #define VMSTOR_WIN7_MAJOR 4
64 #define VMSTOR_WIN7_MINOR 2
66 #define VMSTOR_WIN8_MAJOR 5
67 #define VMSTOR_WIN8_MINOR 1
70 /* Packet structure describing virtual storage requests. */
71 enum vstor_packet_operation {
72 VSTOR_OPERATION_COMPLETE_IO = 1,
73 VSTOR_OPERATION_REMOVE_DEVICE = 2,
74 VSTOR_OPERATION_EXECUTE_SRB = 3,
75 VSTOR_OPERATION_RESET_LUN = 4,
76 VSTOR_OPERATION_RESET_ADAPTER = 5,
77 VSTOR_OPERATION_RESET_BUS = 6,
78 VSTOR_OPERATION_BEGIN_INITIALIZATION = 7,
79 VSTOR_OPERATION_END_INITIALIZATION = 8,
80 VSTOR_OPERATION_QUERY_PROTOCOL_VERSION = 9,
81 VSTOR_OPERATION_QUERY_PROPERTIES = 10,
82 VSTOR_OPERATION_ENUMERATE_BUS = 11,
83 VSTOR_OPERATION_FCHBA_DATA = 12,
84 VSTOR_OPERATION_CREATE_SUB_CHANNELS = 13,
85 VSTOR_OPERATION_MAXIMUM = 13
89 * WWN packet for Fibre Channel HBA
92 struct hv_fc_wwn_packet {
96 u8 primary_port_wwn[8];
97 u8 primary_node_wwn[8];
98 u8 secondary_port_wwn[8];
99 u8 secondary_node_wwn[8];
108 #define SRB_FLAGS_QUEUE_ACTION_ENABLE 0x00000002
109 #define SRB_FLAGS_DISABLE_DISCONNECT 0x00000004
110 #define SRB_FLAGS_DISABLE_SYNCH_TRANSFER 0x00000008
111 #define SRB_FLAGS_BYPASS_FROZEN_QUEUE 0x00000010
112 #define SRB_FLAGS_DISABLE_AUTOSENSE 0x00000020
113 #define SRB_FLAGS_DATA_IN 0x00000040
114 #define SRB_FLAGS_DATA_OUT 0x00000080
115 #define SRB_FLAGS_NO_DATA_TRANSFER 0x00000000
116 #define SRB_FLAGS_UNSPECIFIED_DIRECTION (SRB_FLAGS_DATA_IN | SRB_FLAGS_DATA_OUT)
117 #define SRB_FLAGS_NO_QUEUE_FREEZE 0x00000100
118 #define SRB_FLAGS_ADAPTER_CACHE_ENABLE 0x00000200
119 #define SRB_FLAGS_FREE_SENSE_BUFFER 0x00000400
122 * This flag indicates the request is part of the workflow for processing a D3.
124 #define SRB_FLAGS_D3_PROCESSING 0x00000800
125 #define SRB_FLAGS_IS_ACTIVE 0x00010000
126 #define SRB_FLAGS_ALLOCATED_FROM_ZONE 0x00020000
127 #define SRB_FLAGS_SGLIST_FROM_POOL 0x00040000
128 #define SRB_FLAGS_BYPASS_LOCKED_QUEUE 0x00080000
129 #define SRB_FLAGS_NO_KEEP_AWAKE 0x00100000
130 #define SRB_FLAGS_PORT_DRIVER_ALLOCSENSE 0x00200000
131 #define SRB_FLAGS_PORT_DRIVER_SENSEHASPORT 0x00400000
132 #define SRB_FLAGS_DONT_START_NEXT_PACKET 0x00800000
133 #define SRB_FLAGS_PORT_DRIVER_RESERVED 0x0F000000
134 #define SRB_FLAGS_CLASS_DRIVER_RESERVED 0xF0000000
138 * Platform neutral description of a scsi request -
139 * this remains the same across the write regardless of 32/64 bit
140 * note: it's patterned off the SCSI_PASS_THROUGH structure
142 #define STORVSC_MAX_CMD_LEN 0x10
144 #define POST_WIN7_STORVSC_SENSE_BUFFER_SIZE 0x14
145 #define PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE 0x12
147 #define STORVSC_SENSE_BUFFER_SIZE 0x14
148 #define STORVSC_MAX_BUF_LEN_WITH_PADDING 0x14
151 * Sense buffer size changed in win8; have a run-time
152 * variable to track the size we should use.
154 static int sense_buffer_size;
157 * The size of the vmscsi_request has changed in win8. The
158 * additional size is because of new elements added to the
159 * structure. These elements are valid only when we are talking
161 * Track the correction to size we need to apply.
164 static int vmscsi_size_delta;
165 static int vmstor_current_major;
166 static int vmstor_current_minor;
168 struct vmscsi_win8_extension {
170 * The following were added in Windows 8
180 struct vmscsi_request {
191 u8 sense_info_length;
195 u32 data_transfer_length;
198 u8 cdb[STORVSC_MAX_CMD_LEN];
199 u8 sense_data[STORVSC_SENSE_BUFFER_SIZE];
200 u8 reserved_array[STORVSC_MAX_BUF_LEN_WITH_PADDING];
203 * The following was added in win8.
205 struct vmscsi_win8_extension win8_extension;
207 } __attribute((packed));
211 * This structure is sent during the intialization phase to get the different
212 * properties of the channel.
215 #define STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL 0x1
217 struct vmstorage_channel_properties {
223 u32 max_transfer_bytes;
228 /* This structure is sent during the storage protocol negotiations. */
229 struct vmstorage_protocol_version {
230 /* Major (MSW) and minor (LSW) version numbers. */
234 * Revision number is auto-incremented whenever this file is changed
235 * (See FILL_VMSTOR_REVISION macro above). Mismatch does not
236 * definitely indicate incompatibility--but it does indicate mismatched
238 * This is only used on the windows side. Just set it to 0.
243 /* Channel Property Flags */
244 #define STORAGE_CHANNEL_REMOVABLE_FLAG 0x1
245 #define STORAGE_CHANNEL_EMULATED_IDE_FLAG 0x2
247 struct vstor_packet {
248 /* Requested operation type */
249 enum vstor_packet_operation operation;
251 /* Flags - see below for values */
254 /* Status of the request returned from the server side. */
257 /* Data payload area */
260 * Structure used to forward SCSI commands from the
261 * client to the server.
263 struct vmscsi_request vm_srb;
265 /* Structure used to query channel properties. */
266 struct vmstorage_channel_properties storage_channel_properties;
268 /* Used during version negotiations. */
269 struct vmstorage_protocol_version version;
271 /* Fibre channel address packet */
272 struct hv_fc_wwn_packet wwn_packet;
274 /* Number of sub-channels to create */
275 u16 sub_channel_count;
277 /* This will be the maximum of the union members */
285 * This flag indicates that the server should send back a completion for this
289 #define REQUEST_COMPLETION_FLAG 0x1
291 /* Matches Windows-end */
292 enum storvsc_request_type {
299 * SRB status codes and masks; a subset of the codes used here.
302 #define SRB_STATUS_AUTOSENSE_VALID 0x80
303 #define SRB_STATUS_INVALID_LUN 0x20
304 #define SRB_STATUS_SUCCESS 0x01
305 #define SRB_STATUS_ABORTED 0x02
306 #define SRB_STATUS_ERROR 0x04
309 * This is the end of Protocol specific defines.
314 * We setup a mempool to allocate request structures for this driver
315 * on a per-lun basis. The following define specifies the number of
316 * elements in the pool.
319 #define STORVSC_MIN_BUF_NR 64
320 static int storvsc_ringbuffer_size = (20 * PAGE_SIZE);
322 module_param(storvsc_ringbuffer_size, int, S_IRUGO);
323 MODULE_PARM_DESC(storvsc_ringbuffer_size, "Ring buffer size (bytes)");
326 * Timeout in seconds for all devices managed by this driver.
328 static int storvsc_timeout = 180;
330 #define STORVSC_MAX_IO_REQUESTS 200
332 static void storvsc_on_channel_callback(void *context);
334 #define STORVSC_MAX_LUNS_PER_TARGET 255
335 #define STORVSC_MAX_TARGETS 2
336 #define STORVSC_MAX_CHANNELS 8
338 #define STORVSC_FC_MAX_LUNS_PER_TARGET 255
339 #define STORVSC_FC_MAX_TARGETS 128
340 #define STORVSC_FC_MAX_CHANNELS 8
342 #define STORVSC_IDE_MAX_LUNS_PER_TARGET 64
343 #define STORVSC_IDE_MAX_TARGETS 1
344 #define STORVSC_IDE_MAX_CHANNELS 1
346 struct storvsc_cmd_request {
347 struct list_head entry;
348 struct scsi_cmnd *cmd;
350 unsigned int bounce_sgl_count;
351 struct scatterlist *bounce_sgl;
353 struct hv_device *device;
355 /* Synchronize the request/response if needed */
356 struct completion wait_event;
358 unsigned char *sense_buffer;
359 struct hv_multipage_buffer data_buffer;
360 struct vstor_packet vstor_packet;
364 /* A storvsc device is a device object that contains a vmbus channel */
365 struct storvsc_device {
366 struct hv_device *device;
370 bool open_sub_channel;
371 atomic_t num_outstanding_req;
372 struct Scsi_Host *host;
374 wait_queue_head_t waiting_to_drain;
377 * Each unique Port/Path/Target represents 1 channel ie scsi
378 * controller. In reality, the pathid, targetid is always 0
379 * and the port is set by us
381 unsigned int port_number;
382 unsigned char path_id;
383 unsigned char target_id;
385 /* Used for vsc/vsp channel reset process */
386 struct storvsc_cmd_request init_request;
387 struct storvsc_cmd_request reset_request;
390 struct stor_mem_pools {
391 struct kmem_cache *request_pool;
392 mempool_t *request_mempool;
395 struct hv_host_device {
396 struct hv_device *dev;
399 unsigned char target;
402 struct storvsc_scan_work {
403 struct work_struct work;
404 struct Scsi_Host *host;
408 static void storvsc_device_scan(struct work_struct *work)
410 struct storvsc_scan_work *wrk;
412 struct scsi_device *sdev;
414 wrk = container_of(work, struct storvsc_scan_work, work);
417 sdev = scsi_device_lookup(wrk->host, 0, 0, lun);
420 scsi_rescan_device(&sdev->sdev_gendev);
421 scsi_device_put(sdev);
427 static void storvsc_bus_scan(struct work_struct *work)
429 struct storvsc_scan_work *wrk;
432 wrk = container_of(work, struct storvsc_scan_work, work);
433 for (id = 0; id < wrk->host->max_id; ++id) {
434 if (wrk->host->reverse_ordering)
435 order_id = wrk->host->max_id - id - 1;
439 scsi_scan_target(&wrk->host->shost_gendev, 0,
440 order_id, SCAN_WILD_CARD, 1);
445 static void storvsc_remove_lun(struct work_struct *work)
447 struct storvsc_scan_work *wrk;
448 struct scsi_device *sdev;
450 wrk = container_of(work, struct storvsc_scan_work, work);
451 if (!scsi_host_get(wrk->host))
454 sdev = scsi_device_lookup(wrk->host, 0, 0, wrk->lun);
457 scsi_remove_device(sdev);
458 scsi_device_put(sdev);
460 scsi_host_put(wrk->host);
467 * Major/minor macros. Minor version is in LSB, meaning that earlier flat
468 * version numbers will be interpreted as "0.x" (i.e., 1 becomes 0.1).
471 static inline u16 storvsc_get_version(u8 major, u8 minor)
475 version = ((major << 8) | minor);
480 * We can get incoming messages from the host that are not in response to
481 * messages that we have sent out. An example of this would be messages
482 * received by the guest to notify dynamic addition/removal of LUNs. To
483 * deal with potential race conditions where the driver may be in the
484 * midst of being unloaded when we might receive an unsolicited message
485 * from the host, we have implemented a mechanism to gurantee sequential
488 * 1) Once the device is marked as being destroyed, we will fail all
490 * 2) We permit incoming messages when the device is being destroyed,
491 * only to properly account for messages already sent out.
494 static inline struct storvsc_device *get_out_stor_device(
495 struct hv_device *device)
497 struct storvsc_device *stor_device;
499 stor_device = hv_get_drvdata(device);
501 if (stor_device && stor_device->destroy)
508 static inline void storvsc_wait_to_drain(struct storvsc_device *dev)
510 dev->drain_notify = true;
511 wait_event(dev->waiting_to_drain,
512 atomic_read(&dev->num_outstanding_req) == 0);
513 dev->drain_notify = false;
516 static inline struct storvsc_device *get_in_stor_device(
517 struct hv_device *device)
519 struct storvsc_device *stor_device;
521 stor_device = hv_get_drvdata(device);
527 * If the device is being destroyed; allow incoming
528 * traffic only to cleanup outstanding requests.
531 if (stor_device->destroy &&
532 (atomic_read(&stor_device->num_outstanding_req) == 0))
540 static void destroy_bounce_buffer(struct scatterlist *sgl,
541 unsigned int sg_count)
544 struct page *page_buf;
546 for (i = 0; i < sg_count; i++) {
547 page_buf = sg_page((&sgl[i]));
548 if (page_buf != NULL)
549 __free_page(page_buf);
555 static int do_bounce_buffer(struct scatterlist *sgl, unsigned int sg_count)
559 /* No need to check */
563 /* We have at least 2 sg entries */
564 for (i = 0; i < sg_count; i++) {
566 /* make sure 1st one does not have hole */
567 if (sgl[i].offset + sgl[i].length != PAGE_SIZE)
569 } else if (i == sg_count - 1) {
570 /* make sure last one does not have hole */
571 if (sgl[i].offset != 0)
574 /* make sure no hole in the middle */
575 if (sgl[i].length != PAGE_SIZE || sgl[i].offset != 0)
582 static struct scatterlist *create_bounce_buffer(struct scatterlist *sgl,
583 unsigned int sg_count,
589 struct scatterlist *bounce_sgl;
590 struct page *page_buf;
591 unsigned int buf_len = ((write == WRITE_TYPE) ? 0 : PAGE_SIZE);
593 num_pages = ALIGN(len, PAGE_SIZE) >> PAGE_SHIFT;
595 bounce_sgl = kcalloc(num_pages, sizeof(struct scatterlist), GFP_ATOMIC);
599 sg_init_table(bounce_sgl, num_pages);
600 for (i = 0; i < num_pages; i++) {
601 page_buf = alloc_page(GFP_ATOMIC);
604 sg_set_page(&bounce_sgl[i], page_buf, buf_len, 0);
610 destroy_bounce_buffer(bounce_sgl, num_pages);
614 /* Disgusting wrapper functions */
615 static inline unsigned long sg_kmap_atomic(struct scatterlist *sgl, int idx)
617 void *addr = kmap_atomic(sg_page(sgl + idx));
618 return (unsigned long)addr;
621 static inline void sg_kunmap_atomic(unsigned long addr)
623 kunmap_atomic((void *)addr);
627 /* Assume the original sgl has enough room */
628 static unsigned int copy_from_bounce_buffer(struct scatterlist *orig_sgl,
629 struct scatterlist *bounce_sgl,
630 unsigned int orig_sgl_count,
631 unsigned int bounce_sgl_count)
635 unsigned long src, dest;
636 unsigned int srclen, destlen, copylen;
637 unsigned int total_copied = 0;
638 unsigned long bounce_addr = 0;
639 unsigned long dest_addr = 0;
642 local_irq_save(flags);
644 for (i = 0; i < orig_sgl_count; i++) {
645 dest_addr = sg_kmap_atomic(orig_sgl,i) + orig_sgl[i].offset;
647 destlen = orig_sgl[i].length;
649 if (bounce_addr == 0)
650 bounce_addr = sg_kmap_atomic(bounce_sgl,j);
653 src = bounce_addr + bounce_sgl[j].offset;
654 srclen = bounce_sgl[j].length - bounce_sgl[j].offset;
656 copylen = min(srclen, destlen);
657 memcpy((void *)dest, (void *)src, copylen);
659 total_copied += copylen;
660 bounce_sgl[j].offset += copylen;
664 if (bounce_sgl[j].offset == bounce_sgl[j].length) {
666 sg_kunmap_atomic(bounce_addr);
670 * It is possible that the number of elements
671 * in the bounce buffer may not be equal to
672 * the number of elements in the original
673 * scatter list. Handle this correctly.
676 if (j == bounce_sgl_count) {
678 * We are done; cleanup and return.
680 sg_kunmap_atomic(dest_addr - orig_sgl[i].offset);
681 local_irq_restore(flags);
685 /* if we need to use another bounce buffer */
686 if (destlen || i != orig_sgl_count - 1)
687 bounce_addr = sg_kmap_atomic(bounce_sgl,j);
688 } else if (destlen == 0 && i == orig_sgl_count - 1) {
689 /* unmap the last bounce that is < PAGE_SIZE */
690 sg_kunmap_atomic(bounce_addr);
694 sg_kunmap_atomic(dest_addr - orig_sgl[i].offset);
697 local_irq_restore(flags);
702 /* Assume the bounce_sgl has enough room ie using the create_bounce_buffer() */
703 static unsigned int copy_to_bounce_buffer(struct scatterlist *orig_sgl,
704 struct scatterlist *bounce_sgl,
705 unsigned int orig_sgl_count)
709 unsigned long src, dest;
710 unsigned int srclen, destlen, copylen;
711 unsigned int total_copied = 0;
712 unsigned long bounce_addr = 0;
713 unsigned long src_addr = 0;
716 local_irq_save(flags);
718 for (i = 0; i < orig_sgl_count; i++) {
719 src_addr = sg_kmap_atomic(orig_sgl,i) + orig_sgl[i].offset;
721 srclen = orig_sgl[i].length;
723 if (bounce_addr == 0)
724 bounce_addr = sg_kmap_atomic(bounce_sgl,j);
727 /* assume bounce offset always == 0 */
728 dest = bounce_addr + bounce_sgl[j].length;
729 destlen = PAGE_SIZE - bounce_sgl[j].length;
731 copylen = min(srclen, destlen);
732 memcpy((void *)dest, (void *)src, copylen);
734 total_copied += copylen;
735 bounce_sgl[j].length += copylen;
739 if (bounce_sgl[j].length == PAGE_SIZE) {
740 /* full..move to next entry */
741 sg_kunmap_atomic(bounce_addr);
744 /* if we need to use another bounce buffer */
745 if (srclen || i != orig_sgl_count - 1)
746 bounce_addr = sg_kmap_atomic(bounce_sgl,j);
748 } else if (srclen == 0 && i == orig_sgl_count - 1) {
749 /* unmap the last bounce that is < PAGE_SIZE */
750 sg_kunmap_atomic(bounce_addr);
754 sg_kunmap_atomic(src_addr - orig_sgl[i].offset);
757 local_irq_restore(flags);
762 static void handle_sc_creation(struct vmbus_channel *new_sc)
764 struct hv_device *device = new_sc->primary_channel->device_obj;
765 struct storvsc_device *stor_device;
766 struct vmstorage_channel_properties props;
768 stor_device = get_out_stor_device(device);
772 if (stor_device->open_sub_channel == false)
775 memset(&props, 0, sizeof(struct vmstorage_channel_properties));
778 storvsc_ringbuffer_size,
779 storvsc_ringbuffer_size,
781 sizeof(struct vmstorage_channel_properties),
782 storvsc_on_channel_callback, new_sc);
785 static void handle_multichannel_storage(struct hv_device *device, int max_chns)
787 struct storvsc_device *stor_device;
788 int num_cpus = num_online_cpus();
790 struct storvsc_cmd_request *request;
791 struct vstor_packet *vstor_packet;
794 num_sc = ((max_chns > num_cpus) ? num_cpus : max_chns);
795 stor_device = get_out_stor_device(device);
799 request = &stor_device->init_request;
800 vstor_packet = &request->vstor_packet;
802 stor_device->open_sub_channel = true;
804 * Establish a handler for dealing with subchannels.
806 vmbus_set_sc_create_callback(device->channel, handle_sc_creation);
809 * Check to see if sub-channels have already been created. This
810 * can happen when this driver is re-loaded after unloading.
813 if (vmbus_are_subchannels_present(device->channel))
816 stor_device->open_sub_channel = false;
818 * Request the host to create sub-channels.
820 memset(request, 0, sizeof(struct storvsc_cmd_request));
821 init_completion(&request->wait_event);
822 vstor_packet->operation = VSTOR_OPERATION_CREATE_SUB_CHANNELS;
823 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
824 vstor_packet->sub_channel_count = num_sc;
826 ret = vmbus_sendpacket(device->channel, vstor_packet,
827 (sizeof(struct vstor_packet) -
829 (unsigned long)request,
831 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
836 t = wait_for_completion_timeout(&request->wait_event, 10*HZ);
840 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
841 vstor_packet->status != 0)
845 * Now that we created the sub-channels, invoke the check; this
846 * may trigger the callback.
848 stor_device->open_sub_channel = true;
849 vmbus_are_subchannels_present(device->channel);
852 static int storvsc_channel_init(struct hv_device *device)
854 struct storvsc_device *stor_device;
855 struct storvsc_cmd_request *request;
856 struct vstor_packet *vstor_packet;
859 bool process_sub_channels = false;
861 stor_device = get_out_stor_device(device);
865 request = &stor_device->init_request;
866 vstor_packet = &request->vstor_packet;
869 * Now, initiate the vsc/vsp initialization protocol on the open
872 memset(request, 0, sizeof(struct storvsc_cmd_request));
873 init_completion(&request->wait_event);
874 vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION;
875 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
877 ret = vmbus_sendpacket(device->channel, vstor_packet,
878 (sizeof(struct vstor_packet) -
880 (unsigned long)request,
882 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
886 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
892 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
893 vstor_packet->status != 0)
897 /* reuse the packet for version range supported */
898 memset(vstor_packet, 0, sizeof(struct vstor_packet));
899 vstor_packet->operation = VSTOR_OPERATION_QUERY_PROTOCOL_VERSION;
900 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
902 vstor_packet->version.major_minor =
903 storvsc_get_version(vmstor_current_major, vmstor_current_minor);
906 * The revision number is only used in Windows; set it to 0.
908 vstor_packet->version.revision = 0;
910 ret = vmbus_sendpacket(device->channel, vstor_packet,
911 (sizeof(struct vstor_packet) -
913 (unsigned long)request,
915 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
919 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
925 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
926 vstor_packet->status != 0)
930 memset(vstor_packet, 0, sizeof(struct vstor_packet));
931 vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES;
932 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
934 ret = vmbus_sendpacket(device->channel, vstor_packet,
935 (sizeof(struct vstor_packet) -
937 (unsigned long)request,
939 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
944 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
950 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
951 vstor_packet->status != 0)
955 * Check to see if multi-channel support is there.
956 * Hosts that implement protocol version of 5.1 and above
957 * support multi-channel.
959 max_chns = vstor_packet->storage_channel_properties.max_channel_cnt;
960 if ((vmbus_proto_version != VERSION_WIN7) &&
961 (vmbus_proto_version != VERSION_WS2008)) {
962 if (vstor_packet->storage_channel_properties.flags &
963 STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL)
964 process_sub_channels = true;
967 memset(vstor_packet, 0, sizeof(struct vstor_packet));
968 vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION;
969 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
971 ret = vmbus_sendpacket(device->channel, vstor_packet,
972 (sizeof(struct vstor_packet) -
974 (unsigned long)request,
976 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
981 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
987 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
988 vstor_packet->status != 0)
991 if (process_sub_channels)
992 handle_multichannel_storage(device, max_chns);
999 static void storvsc_handle_error(struct vmscsi_request *vm_srb,
1000 struct scsi_cmnd *scmnd,
1001 struct Scsi_Host *host,
1004 struct storvsc_scan_work *wrk;
1005 void (*process_err_fn)(struct work_struct *work);
1006 bool do_work = false;
1008 switch (vm_srb->srb_status) {
1009 case SRB_STATUS_ERROR:
1011 * If there is an error; offline the device since all
1012 * error recovery strategies would have already been
1013 * deployed on the host side. However, if the command
1014 * were a pass-through command deal with it appropriately.
1016 switch (scmnd->cmnd[0]) {
1019 set_host_byte(scmnd, DID_PASSTHROUGH);
1022 set_host_byte(scmnd, DID_TARGET_FAILURE);
1025 case SRB_STATUS_INVALID_LUN:
1027 process_err_fn = storvsc_remove_lun;
1029 case (SRB_STATUS_ABORTED | SRB_STATUS_AUTOSENSE_VALID):
1030 if ((asc == 0x2a) && (ascq == 0x9)) {
1032 process_err_fn = storvsc_device_scan;
1034 * Retry the I/O that trigerred this.
1036 set_host_byte(scmnd, DID_REQUEUE);
1045 * We need to schedule work to process this error; schedule it.
1047 wrk = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC);
1049 set_host_byte(scmnd, DID_TARGET_FAILURE);
1054 wrk->lun = vm_srb->lun;
1055 INIT_WORK(&wrk->work, process_err_fn);
1056 schedule_work(&wrk->work);
1060 static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request)
1062 struct scsi_cmnd *scmnd = cmd_request->cmd;
1063 struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
1064 void (*scsi_done_fn)(struct scsi_cmnd *);
1065 struct scsi_sense_hdr sense_hdr;
1066 struct vmscsi_request *vm_srb;
1067 struct stor_mem_pools *memp = scmnd->device->hostdata;
1068 struct Scsi_Host *host;
1069 struct storvsc_device *stor_dev;
1070 struct hv_device *dev = host_dev->dev;
1072 stor_dev = get_in_stor_device(dev);
1073 host = stor_dev->host;
1075 vm_srb = &cmd_request->vstor_packet.vm_srb;
1076 if (cmd_request->bounce_sgl_count) {
1077 if (vm_srb->data_in == READ_TYPE)
1078 copy_from_bounce_buffer(scsi_sglist(scmnd),
1079 cmd_request->bounce_sgl,
1080 scsi_sg_count(scmnd),
1081 cmd_request->bounce_sgl_count);
1082 destroy_bounce_buffer(cmd_request->bounce_sgl,
1083 cmd_request->bounce_sgl_count);
1086 scmnd->result = vm_srb->scsi_status;
1088 if (scmnd->result) {
1089 if (scsi_normalize_sense(scmnd->sense_buffer,
1090 SCSI_SENSE_BUFFERSIZE, &sense_hdr))
1091 scsi_print_sense_hdr("storvsc", &sense_hdr);
1094 if (vm_srb->srb_status != SRB_STATUS_SUCCESS)
1095 storvsc_handle_error(vm_srb, scmnd, host, sense_hdr.asc,
1098 scsi_set_resid(scmnd,
1099 cmd_request->data_buffer.len -
1100 vm_srb->data_transfer_length);
1102 scsi_done_fn = scmnd->scsi_done;
1104 scmnd->host_scribble = NULL;
1105 scmnd->scsi_done = NULL;
1107 scsi_done_fn(scmnd);
1109 mempool_free(cmd_request, memp->request_mempool);
1112 static void storvsc_on_io_completion(struct hv_device *device,
1113 struct vstor_packet *vstor_packet,
1114 struct storvsc_cmd_request *request)
1116 struct storvsc_device *stor_device;
1117 struct vstor_packet *stor_pkt;
1119 stor_device = hv_get_drvdata(device);
1120 stor_pkt = &request->vstor_packet;
1123 * The current SCSI handling on the host side does
1124 * not correctly handle:
1125 * INQUIRY command with page code parameter set to 0x80
1126 * MODE_SENSE command with cmd[2] == 0x1c
1128 * Setup srb and scsi status so this won't be fatal.
1129 * We do this so we can distinguish truly fatal failues
1130 * (srb status == 0x4) and off-line the device in that case.
1133 if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) ||
1134 (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) {
1135 vstor_packet->vm_srb.scsi_status = 0;
1136 vstor_packet->vm_srb.srb_status = SRB_STATUS_SUCCESS;
1140 /* Copy over the status...etc */
1141 stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status;
1142 stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status;
1143 stor_pkt->vm_srb.sense_info_length =
1144 vstor_packet->vm_srb.sense_info_length;
1146 if (vstor_packet->vm_srb.scsi_status != 0 ||
1147 vstor_packet->vm_srb.srb_status != SRB_STATUS_SUCCESS){
1148 dev_warn(&device->device,
1149 "cmd 0x%x scsi status 0x%x srb status 0x%x\n",
1150 stor_pkt->vm_srb.cdb[0],
1151 vstor_packet->vm_srb.scsi_status,
1152 vstor_packet->vm_srb.srb_status);
1155 if ((vstor_packet->vm_srb.scsi_status & 0xFF) == 0x02) {
1156 /* CHECK_CONDITION */
1157 if (vstor_packet->vm_srb.srb_status &
1158 SRB_STATUS_AUTOSENSE_VALID) {
1159 /* autosense data available */
1160 dev_warn(&device->device,
1161 "stor pkt %p autosense data valid - len %d\n",
1163 vstor_packet->vm_srb.sense_info_length);
1165 memcpy(request->sense_buffer,
1166 vstor_packet->vm_srb.sense_data,
1167 vstor_packet->vm_srb.sense_info_length);
1172 stor_pkt->vm_srb.data_transfer_length =
1173 vstor_packet->vm_srb.data_transfer_length;
1175 storvsc_command_completion(request);
1177 if (atomic_dec_and_test(&stor_device->num_outstanding_req) &&
1178 stor_device->drain_notify)
1179 wake_up(&stor_device->waiting_to_drain);
1184 static void storvsc_on_receive(struct hv_device *device,
1185 struct vstor_packet *vstor_packet,
1186 struct storvsc_cmd_request *request)
1188 struct storvsc_scan_work *work;
1189 struct storvsc_device *stor_device;
1191 switch (vstor_packet->operation) {
1192 case VSTOR_OPERATION_COMPLETE_IO:
1193 storvsc_on_io_completion(device, vstor_packet, request);
1196 case VSTOR_OPERATION_REMOVE_DEVICE:
1197 case VSTOR_OPERATION_ENUMERATE_BUS:
1198 stor_device = get_in_stor_device(device);
1199 work = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC);
1203 INIT_WORK(&work->work, storvsc_bus_scan);
1204 work->host = stor_device->host;
1205 schedule_work(&work->work);
1213 static void storvsc_on_channel_callback(void *context)
1215 struct vmbus_channel *channel = (struct vmbus_channel *)context;
1216 struct hv_device *device;
1217 struct storvsc_device *stor_device;
1220 unsigned char packet[ALIGN(sizeof(struct vstor_packet), 8)];
1221 struct storvsc_cmd_request *request;
1224 if (channel->primary_channel != NULL)
1225 device = channel->primary_channel->device_obj;
1227 device = channel->device_obj;
1229 stor_device = get_in_stor_device(device);
1234 ret = vmbus_recvpacket(channel, packet,
1235 ALIGN((sizeof(struct vstor_packet) -
1236 vmscsi_size_delta), 8),
1237 &bytes_recvd, &request_id);
1238 if (ret == 0 && bytes_recvd > 0) {
1240 request = (struct storvsc_cmd_request *)
1241 (unsigned long)request_id;
1243 if ((request == &stor_device->init_request) ||
1244 (request == &stor_device->reset_request)) {
1246 memcpy(&request->vstor_packet, packet,
1247 (sizeof(struct vstor_packet) -
1248 vmscsi_size_delta));
1249 complete(&request->wait_event);
1251 storvsc_on_receive(device,
1252 (struct vstor_packet *)packet,
1263 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size)
1265 struct vmstorage_channel_properties props;
1268 memset(&props, 0, sizeof(struct vmstorage_channel_properties));
1270 ret = vmbus_open(device->channel,
1274 sizeof(struct vmstorage_channel_properties),
1275 storvsc_on_channel_callback, device->channel);
1280 ret = storvsc_channel_init(device);
1285 static int storvsc_dev_remove(struct hv_device *device)
1287 struct storvsc_device *stor_device;
1288 unsigned long flags;
1290 stor_device = hv_get_drvdata(device);
1292 spin_lock_irqsave(&device->channel->inbound_lock, flags);
1293 stor_device->destroy = true;
1294 spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
1297 * At this point, all outbound traffic should be disable. We
1298 * only allow inbound traffic (responses) to proceed so that
1299 * outstanding requests can be completed.
1302 storvsc_wait_to_drain(stor_device);
1305 * Since we have already drained, we don't need to busy wait
1306 * as was done in final_release_stor_device()
1307 * Note that we cannot set the ext pointer to NULL until
1308 * we have drained - to drain the outgoing packets, we need to
1309 * allow incoming packets.
1311 spin_lock_irqsave(&device->channel->inbound_lock, flags);
1312 hv_set_drvdata(device, NULL);
1313 spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
1315 /* Close the channel */
1316 vmbus_close(device->channel);
1322 static int storvsc_do_io(struct hv_device *device,
1323 struct storvsc_cmd_request *request)
1325 struct storvsc_device *stor_device;
1326 struct vstor_packet *vstor_packet;
1327 struct vmbus_channel *outgoing_channel;
1330 vstor_packet = &request->vstor_packet;
1331 stor_device = get_out_stor_device(device);
1337 request->device = device;
1339 * Select an an appropriate channel to send the request out.
1342 outgoing_channel = vmbus_get_outgoing_channel(device->channel);
1345 vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
1347 vstor_packet->vm_srb.length = (sizeof(struct vmscsi_request) -
1351 vstor_packet->vm_srb.sense_info_length = sense_buffer_size;
1354 vstor_packet->vm_srb.data_transfer_length =
1355 request->data_buffer.len;
1357 vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
1359 if (request->data_buffer.len) {
1360 ret = vmbus_sendpacket_multipagebuffer(outgoing_channel,
1361 &request->data_buffer,
1363 (sizeof(struct vstor_packet) -
1365 (unsigned long)request);
1367 ret = vmbus_sendpacket(device->channel, vstor_packet,
1368 (sizeof(struct vstor_packet) -
1370 (unsigned long)request,
1372 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1378 atomic_inc(&stor_device->num_outstanding_req);
1383 static int storvsc_device_alloc(struct scsi_device *sdevice)
1385 struct stor_mem_pools *memp;
1386 int number = STORVSC_MIN_BUF_NR;
1388 memp = kzalloc(sizeof(struct stor_mem_pools), GFP_KERNEL);
1392 memp->request_pool =
1393 kmem_cache_create(dev_name(&sdevice->sdev_dev),
1394 sizeof(struct storvsc_cmd_request), 0,
1395 SLAB_HWCACHE_ALIGN, NULL);
1397 if (!memp->request_pool)
1400 memp->request_mempool = mempool_create(number, mempool_alloc_slab,
1402 memp->request_pool);
1404 if (!memp->request_mempool)
1407 sdevice->hostdata = memp;
1412 kmem_cache_destroy(memp->request_pool);
1419 static void storvsc_device_destroy(struct scsi_device *sdevice)
1421 struct stor_mem_pools *memp = sdevice->hostdata;
1426 mempool_destroy(memp->request_mempool);
1427 kmem_cache_destroy(memp->request_pool);
1429 sdevice->hostdata = NULL;
1432 static int storvsc_device_configure(struct scsi_device *sdevice)
1434 scsi_adjust_queue_depth(sdevice, MSG_SIMPLE_TAG,
1435 STORVSC_MAX_IO_REQUESTS);
1437 blk_queue_max_segment_size(sdevice->request_queue, PAGE_SIZE);
1439 blk_queue_bounce_limit(sdevice->request_queue, BLK_BOUNCE_ANY);
1441 blk_queue_rq_timeout(sdevice->request_queue, (storvsc_timeout * HZ));
1443 sdevice->no_write_same = 1;
1448 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev,
1449 sector_t capacity, int *info)
1451 sector_t nsect = capacity;
1452 sector_t cylinders = nsect;
1453 int heads, sectors_pt;
1456 * We are making up these values; let us keep it simple.
1459 sectors_pt = 0x3f; /* Sectors per track */
1460 sector_div(cylinders, heads * sectors_pt);
1461 if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
1465 info[1] = sectors_pt;
1466 info[2] = (int)cylinders;
1471 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
1473 struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
1474 struct hv_device *device = host_dev->dev;
1476 struct storvsc_device *stor_device;
1477 struct storvsc_cmd_request *request;
1478 struct vstor_packet *vstor_packet;
1482 stor_device = get_out_stor_device(device);
1486 request = &stor_device->reset_request;
1487 vstor_packet = &request->vstor_packet;
1489 init_completion(&request->wait_event);
1491 vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
1492 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
1493 vstor_packet->vm_srb.path_id = stor_device->path_id;
1495 ret = vmbus_sendpacket(device->channel, vstor_packet,
1496 (sizeof(struct vstor_packet) -
1498 (unsigned long)&stor_device->reset_request,
1500 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1504 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
1506 return TIMEOUT_ERROR;
1510 * At this point, all outstanding requests in the adapter
1511 * should have been flushed out and return to us
1512 * There is a potential race here where the host may be in
1513 * the process of responding when we return from here.
1514 * Just wait for all in-transit packets to be accounted for
1515 * before we return from here.
1517 storvsc_wait_to_drain(stor_device);
1523 * The host guarantees to respond to each command, although I/O latencies might
1524 * be unbounded on Azure. Reset the timer unconditionally to give the host a
1525 * chance to perform EH.
1527 static enum blk_eh_timer_return storvsc_eh_timed_out(struct scsi_cmnd *scmnd)
1529 return BLK_EH_RESET_TIMER;
1532 static bool storvsc_scsi_cmd_ok(struct scsi_cmnd *scmnd)
1534 bool allowed = true;
1535 u8 scsi_op = scmnd->cmnd[0];
1538 /* the host does not handle WRITE_SAME, log accident usage */
1541 * smartd sends this command and the host does not handle
1542 * this. So, don't send it.
1545 scmnd->result = ILLEGAL_REQUEST << 16;
1554 static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
1557 struct hv_host_device *host_dev = shost_priv(host);
1558 struct hv_device *dev = host_dev->dev;
1559 struct storvsc_cmd_request *cmd_request;
1560 unsigned int request_size = 0;
1562 struct scatterlist *sgl;
1563 unsigned int sg_count = 0;
1564 struct vmscsi_request *vm_srb;
1565 struct stor_mem_pools *memp = scmnd->device->hostdata;
1567 if (vmstor_current_major <= VMSTOR_WIN8_MAJOR) {
1569 * On legacy hosts filter unimplemented commands.
1570 * Future hosts are expected to correctly handle
1571 * unsupported commands. Furthermore, it is
1572 * possible that some of the currently
1573 * unsupported commands maybe supported in
1574 * future versions of the host.
1576 if (!storvsc_scsi_cmd_ok(scmnd)) {
1577 scmnd->scsi_done(scmnd);
1582 request_size = sizeof(struct storvsc_cmd_request);
1584 cmd_request = mempool_alloc(memp->request_mempool,
1588 * We might be invoked in an interrupt context; hence
1589 * mempool_alloc() can fail.
1592 return SCSI_MLQUEUE_DEVICE_BUSY;
1594 memset(cmd_request, 0, sizeof(struct storvsc_cmd_request));
1596 /* Setup the cmd request */
1597 cmd_request->cmd = scmnd;
1599 scmnd->host_scribble = (unsigned char *)cmd_request;
1601 vm_srb = &cmd_request->vstor_packet.vm_srb;
1602 vm_srb->win8_extension.time_out_value = 60;
1604 vm_srb->win8_extension.srb_flags |=
1605 (SRB_FLAGS_QUEUE_ACTION_ENABLE |
1606 SRB_FLAGS_DISABLE_SYNCH_TRANSFER);
1609 switch (scmnd->sc_data_direction) {
1611 vm_srb->data_in = WRITE_TYPE;
1612 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_DATA_OUT;
1614 case DMA_FROM_DEVICE:
1615 vm_srb->data_in = READ_TYPE;
1616 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_DATA_IN;
1619 vm_srb->data_in = UNKNOWN_TYPE;
1620 vm_srb->win8_extension.srb_flags |= (SRB_FLAGS_DATA_IN |
1621 SRB_FLAGS_DATA_OUT);
1626 vm_srb->port_number = host_dev->port;
1627 vm_srb->path_id = scmnd->device->channel;
1628 vm_srb->target_id = scmnd->device->id;
1629 vm_srb->lun = scmnd->device->lun;
1631 vm_srb->cdb_length = scmnd->cmd_len;
1633 memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
1635 cmd_request->sense_buffer = scmnd->sense_buffer;
1638 cmd_request->data_buffer.len = scsi_bufflen(scmnd);
1639 if (scsi_sg_count(scmnd)) {
1640 sgl = (struct scatterlist *)scsi_sglist(scmnd);
1641 sg_count = scsi_sg_count(scmnd);
1643 /* check if we need to bounce the sgl */
1644 if (do_bounce_buffer(sgl, scsi_sg_count(scmnd)) != -1) {
1645 cmd_request->bounce_sgl =
1646 create_bounce_buffer(sgl, scsi_sg_count(scmnd),
1647 scsi_bufflen(scmnd),
1649 if (!cmd_request->bounce_sgl) {
1650 ret = SCSI_MLQUEUE_HOST_BUSY;
1654 cmd_request->bounce_sgl_count =
1655 ALIGN(scsi_bufflen(scmnd), PAGE_SIZE) >>
1658 if (vm_srb->data_in == WRITE_TYPE)
1659 copy_to_bounce_buffer(sgl,
1660 cmd_request->bounce_sgl,
1661 scsi_sg_count(scmnd));
1663 sgl = cmd_request->bounce_sgl;
1664 sg_count = cmd_request->bounce_sgl_count;
1667 cmd_request->data_buffer.offset = sgl[0].offset;
1669 for (i = 0; i < sg_count; i++)
1670 cmd_request->data_buffer.pfn_array[i] =
1671 page_to_pfn(sg_page((&sgl[i])));
1673 } else if (scsi_sglist(scmnd)) {
1674 cmd_request->data_buffer.offset =
1675 virt_to_phys(scsi_sglist(scmnd)) & (PAGE_SIZE-1);
1676 cmd_request->data_buffer.pfn_array[0] =
1677 virt_to_phys(scsi_sglist(scmnd)) >> PAGE_SHIFT;
1680 /* Invokes the vsc to start an IO */
1681 ret = storvsc_do_io(dev, cmd_request);
1683 if (ret == -EAGAIN) {
1686 if (cmd_request->bounce_sgl_count) {
1687 destroy_bounce_buffer(cmd_request->bounce_sgl,
1688 cmd_request->bounce_sgl_count);
1690 ret = SCSI_MLQUEUE_DEVICE_BUSY;
1698 mempool_free(cmd_request, memp->request_mempool);
1699 scmnd->host_scribble = NULL;
1703 static struct scsi_host_template scsi_driver = {
1704 .module = THIS_MODULE,
1705 .name = "storvsc_host_t",
1706 .bios_param = storvsc_get_chs,
1707 .queuecommand = storvsc_queuecommand,
1708 .eh_host_reset_handler = storvsc_host_reset_handler,
1709 .eh_timed_out = storvsc_eh_timed_out,
1710 .slave_alloc = storvsc_device_alloc,
1711 .slave_destroy = storvsc_device_destroy,
1712 .slave_configure = storvsc_device_configure,
1714 .can_queue = STORVSC_MAX_IO_REQUESTS*STORVSC_MAX_TARGETS,
1716 /* no use setting to 0 since ll_blk_rw reset it to 1 */
1718 .sg_tablesize = MAX_MULTIPAGE_BUFFER_COUNT,
1719 .use_clustering = DISABLE_CLUSTERING,
1720 /* Make sure we dont get a sg segment crosses a page boundary */
1721 .dma_boundary = PAGE_SIZE-1,
1731 static const struct hv_vmbus_device_id id_table[] = {
1734 .driver_data = SCSI_GUID
1738 .driver_data = IDE_GUID
1740 /* Fibre Channel GUID */
1743 .driver_data = SFC_GUID
1748 MODULE_DEVICE_TABLE(vmbus, id_table);
1750 static int storvsc_probe(struct hv_device *device,
1751 const struct hv_vmbus_device_id *dev_id)
1754 struct Scsi_Host *host;
1755 struct hv_host_device *host_dev;
1756 bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false);
1758 struct storvsc_device *stor_device;
1761 * Based on the windows host we are running on,
1762 * set state to properly communicate with the host.
1765 switch (vmbus_proto_version) {
1766 case VERSION_WS2008:
1768 sense_buffer_size = PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE;
1769 vmscsi_size_delta = sizeof(struct vmscsi_win8_extension);
1770 vmstor_current_major = VMSTOR_WIN7_MAJOR;
1771 vmstor_current_minor = VMSTOR_WIN7_MINOR;
1774 sense_buffer_size = POST_WIN7_STORVSC_SENSE_BUFFER_SIZE;
1775 vmscsi_size_delta = 0;
1776 vmstor_current_major = VMSTOR_WIN8_MAJOR;
1777 vmstor_current_minor = VMSTOR_WIN8_MINOR;
1781 if (dev_id->driver_data == SFC_GUID)
1782 scsi_driver.can_queue = (STORVSC_MAX_IO_REQUESTS *
1783 STORVSC_FC_MAX_TARGETS);
1784 host = scsi_host_alloc(&scsi_driver,
1785 sizeof(struct hv_host_device));
1789 host_dev = shost_priv(host);
1790 memset(host_dev, 0, sizeof(struct hv_host_device));
1792 host_dev->port = host->host_no;
1793 host_dev->dev = device;
1796 stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
1802 stor_device->destroy = false;
1803 stor_device->open_sub_channel = false;
1804 init_waitqueue_head(&stor_device->waiting_to_drain);
1805 stor_device->device = device;
1806 stor_device->host = host;
1807 hv_set_drvdata(device, stor_device);
1809 stor_device->port_number = host->host_no;
1810 ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size);
1814 host_dev->path = stor_device->path_id;
1815 host_dev->target = stor_device->target_id;
1817 switch (dev_id->driver_data) {
1819 host->max_lun = STORVSC_FC_MAX_LUNS_PER_TARGET;
1820 host->max_id = STORVSC_FC_MAX_TARGETS;
1821 host->max_channel = STORVSC_FC_MAX_CHANNELS - 1;
1825 host->max_lun = STORVSC_MAX_LUNS_PER_TARGET;
1826 host->max_id = STORVSC_MAX_TARGETS;
1827 host->max_channel = STORVSC_MAX_CHANNELS - 1;
1831 host->max_lun = STORVSC_IDE_MAX_LUNS_PER_TARGET;
1832 host->max_id = STORVSC_IDE_MAX_TARGETS;
1833 host->max_channel = STORVSC_IDE_MAX_CHANNELS - 1;
1836 /* max cmd length */
1837 host->max_cmd_len = STORVSC_MAX_CMD_LEN;
1839 /* Register the HBA and start the scsi bus scan */
1840 ret = scsi_add_host(host, &device->device);
1845 scsi_scan_host(host);
1847 target = (device->dev_instance.b[5] << 8 |
1848 device->dev_instance.b[4]);
1849 ret = scsi_add_device(host, 0, target, 0);
1851 scsi_remove_host(host);
1859 * Once we have connected with the host, we would need to
1860 * to invoke storvsc_dev_remove() to rollback this state and
1861 * this call also frees up the stor_device; hence the jump around
1864 storvsc_dev_remove(device);
1871 scsi_host_put(host);
1875 static int storvsc_remove(struct hv_device *dev)
1877 struct storvsc_device *stor_device = hv_get_drvdata(dev);
1878 struct Scsi_Host *host = stor_device->host;
1880 scsi_remove_host(host);
1881 storvsc_dev_remove(dev);
1882 scsi_host_put(host);
1887 static struct hv_driver storvsc_drv = {
1888 .name = KBUILD_MODNAME,
1889 .id_table = id_table,
1890 .probe = storvsc_probe,
1891 .remove = storvsc_remove,
1894 static int __init storvsc_drv_init(void)
1896 u32 max_outstanding_req_per_channel;
1899 * Divide the ring buffer data size (which is 1 page less
1900 * than the ring buffer size since that page is reserved for
1901 * the ring buffer indices) by the max request size (which is
1902 * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
1904 max_outstanding_req_per_channel =
1905 ((storvsc_ringbuffer_size - PAGE_SIZE) /
1906 ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
1907 sizeof(struct vstor_packet) + sizeof(u64) -
1911 if (max_outstanding_req_per_channel <
1912 STORVSC_MAX_IO_REQUESTS)
1915 return vmbus_driver_register(&storvsc_drv);
1918 static void __exit storvsc_drv_exit(void)
1920 vmbus_driver_unregister(&storvsc_drv);
1923 MODULE_LICENSE("GPL");
1924 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
1925 module_init(storvsc_drv_init);
1926 module_exit(storvsc_drv_exit);