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
61 #define VMSTOR_CURRENT_MAJOR 4
62 #define VMSTOR_CURRENT_MINOR 2
65 /* Packet structure describing virtual storage requests. */
66 enum vstor_packet_operation
{
67 VSTOR_OPERATION_COMPLETE_IO
= 1,
68 VSTOR_OPERATION_REMOVE_DEVICE
= 2,
69 VSTOR_OPERATION_EXECUTE_SRB
= 3,
70 VSTOR_OPERATION_RESET_LUN
= 4,
71 VSTOR_OPERATION_RESET_ADAPTER
= 5,
72 VSTOR_OPERATION_RESET_BUS
= 6,
73 VSTOR_OPERATION_BEGIN_INITIALIZATION
= 7,
74 VSTOR_OPERATION_END_INITIALIZATION
= 8,
75 VSTOR_OPERATION_QUERY_PROTOCOL_VERSION
= 9,
76 VSTOR_OPERATION_QUERY_PROPERTIES
= 10,
77 VSTOR_OPERATION_ENUMERATE_BUS
= 11,
78 VSTOR_OPERATION_MAXIMUM
= 11
82 * Platform neutral description of a scsi request -
83 * this remains the same across the write regardless of 32/64 bit
84 * note: it's patterned off the SCSI_PASS_THROUGH structure
86 #define STORVSC_MAX_CMD_LEN 0x10
87 #define STORVSC_SENSE_BUFFER_SIZE 0x12
88 #define STORVSC_MAX_BUF_LEN_WITH_PADDING 0x14
90 struct vmscsi_request
{
101 u8 sense_info_length
;
105 u32 data_transfer_length
;
108 u8 cdb
[STORVSC_MAX_CMD_LEN
];
109 u8 sense_data
[STORVSC_SENSE_BUFFER_SIZE
];
110 u8 reserved_array
[STORVSC_MAX_BUF_LEN_WITH_PADDING
];
112 } __attribute((packed
));
116 * This structure is sent during the intialization phase to get the different
117 * properties of the channel.
119 struct vmstorage_channel_properties
{
120 u16 protocol_version
;
124 /* Note: port number is only really known on the client side */
127 u32 max_transfer_bytes
;
130 * This id is unique for each channel and will correspond with
131 * vendor specific data in the inquiry data.
137 /* This structure is sent during the storage protocol negotiations. */
138 struct vmstorage_protocol_version
{
139 /* Major (MSW) and minor (LSW) version numbers. */
143 * Revision number is auto-incremented whenever this file is changed
144 * (See FILL_VMSTOR_REVISION macro above). Mismatch does not
145 * definitely indicate incompatibility--but it does indicate mismatched
147 * This is only used on the windows side. Just set it to 0.
152 /* Channel Property Flags */
153 #define STORAGE_CHANNEL_REMOVABLE_FLAG 0x1
154 #define STORAGE_CHANNEL_EMULATED_IDE_FLAG 0x2
156 struct vstor_packet
{
157 /* Requested operation type */
158 enum vstor_packet_operation operation
;
160 /* Flags - see below for values */
163 /* Status of the request returned from the server side. */
166 /* Data payload area */
169 * Structure used to forward SCSI commands from the
170 * client to the server.
172 struct vmscsi_request vm_srb
;
174 /* Structure used to query channel properties. */
175 struct vmstorage_channel_properties storage_channel_properties
;
177 /* Used during version negotiations. */
178 struct vmstorage_protocol_version version
;
185 * This flag indicates that the server should send back a completion for this
189 #define REQUEST_COMPLETION_FLAG 0x1
191 /* Matches Windows-end */
192 enum storvsc_request_type
{
199 * SRB status codes and masks; a subset of the codes used here.
202 #define SRB_STATUS_AUTOSENSE_VALID 0x80
203 #define SRB_STATUS_INVALID_LUN 0x20
204 #define SRB_STATUS_SUCCESS 0x01
205 #define SRB_STATUS_ABORTED 0x02
206 #define SRB_STATUS_ERROR 0x04
209 * This is the end of Protocol specific defines.
214 * We setup a mempool to allocate request structures for this driver
215 * on a per-lun basis. The following define specifies the number of
216 * elements in the pool.
219 #define STORVSC_MIN_BUF_NR 64
220 static int storvsc_ringbuffer_size
= (20 * PAGE_SIZE
);
222 module_param(storvsc_ringbuffer_size
, int, S_IRUGO
);
223 MODULE_PARM_DESC(storvsc_ringbuffer_size
, "Ring buffer size (bytes)");
225 #define STORVSC_MAX_IO_REQUESTS 128
228 * In Hyper-V, each port/path/target maps to 1 scsi host adapter. In
229 * reality, the path/target is not used (ie always set to 0) so our
230 * scsi host adapter essentially has 1 bus with 1 target that contains
233 #define STORVSC_MAX_LUNS_PER_TARGET 64
234 #define STORVSC_MAX_TARGETS 1
235 #define STORVSC_MAX_CHANNELS 1
239 struct storvsc_cmd_request
{
240 struct list_head entry
;
241 struct scsi_cmnd
*cmd
;
243 unsigned int bounce_sgl_count
;
244 struct scatterlist
*bounce_sgl
;
246 struct hv_device
*device
;
248 /* Synchronize the request/response if needed */
249 struct completion wait_event
;
251 unsigned char *sense_buffer
;
252 struct hv_multipage_buffer data_buffer
;
253 struct vstor_packet vstor_packet
;
257 /* A storvsc device is a device object that contains a vmbus channel */
258 struct storvsc_device
{
259 struct hv_device
*device
;
263 atomic_t num_outstanding_req
;
264 struct Scsi_Host
*host
;
266 wait_queue_head_t waiting_to_drain
;
269 * Each unique Port/Path/Target represents 1 channel ie scsi
270 * controller. In reality, the pathid, targetid is always 0
271 * and the port is set by us
273 unsigned int port_number
;
274 unsigned char path_id
;
275 unsigned char target_id
;
277 /* Used for vsc/vsp channel reset process */
278 struct storvsc_cmd_request init_request
;
279 struct storvsc_cmd_request reset_request
;
282 struct stor_mem_pools
{
283 struct kmem_cache
*request_pool
;
284 mempool_t
*request_mempool
;
287 struct hv_host_device
{
288 struct hv_device
*dev
;
291 unsigned char target
;
294 struct storvsc_scan_work
{
295 struct work_struct work
;
296 struct Scsi_Host
*host
;
300 static void storvsc_device_scan(struct work_struct
*work
)
302 struct storvsc_scan_work
*wrk
;
304 struct scsi_device
*sdev
;
306 wrk
= container_of(work
, struct storvsc_scan_work
, work
);
309 sdev
= scsi_device_lookup(wrk
->host
, 0, 0, lun
);
312 scsi_rescan_device(&sdev
->sdev_gendev
);
313 scsi_device_put(sdev
);
319 static void storvsc_bus_scan(struct work_struct
*work
)
321 struct storvsc_scan_work
*wrk
;
324 wrk
= container_of(work
, struct storvsc_scan_work
, work
);
325 for (id
= 0; id
< wrk
->host
->max_id
; ++id
) {
326 if (wrk
->host
->reverse_ordering
)
327 order_id
= wrk
->host
->max_id
- id
- 1;
331 scsi_scan_target(&wrk
->host
->shost_gendev
, 0,
332 order_id
, SCAN_WILD_CARD
, 1);
337 static void storvsc_remove_lun(struct work_struct
*work
)
339 struct storvsc_scan_work
*wrk
;
340 struct scsi_device
*sdev
;
342 wrk
= container_of(work
, struct storvsc_scan_work
, work
);
343 if (!scsi_host_get(wrk
->host
))
346 sdev
= scsi_device_lookup(wrk
->host
, 0, 0, wrk
->lun
);
349 scsi_remove_device(sdev
);
350 scsi_device_put(sdev
);
352 scsi_host_put(wrk
->host
);
359 * Major/minor macros. Minor version is in LSB, meaning that earlier flat
360 * version numbers will be interpreted as "0.x" (i.e., 1 becomes 0.1).
363 static inline u16
storvsc_get_version(u8 major
, u8 minor
)
367 version
= ((major
<< 8) | minor
);
372 * We can get incoming messages from the host that are not in response to
373 * messages that we have sent out. An example of this would be messages
374 * received by the guest to notify dynamic addition/removal of LUNs. To
375 * deal with potential race conditions where the driver may be in the
376 * midst of being unloaded when we might receive an unsolicited message
377 * from the host, we have implemented a mechanism to gurantee sequential
380 * 1) Once the device is marked as being destroyed, we will fail all
382 * 2) We permit incoming messages when the device is being destroyed,
383 * only to properly account for messages already sent out.
386 static inline struct storvsc_device
*get_out_stor_device(
387 struct hv_device
*device
)
389 struct storvsc_device
*stor_device
;
391 stor_device
= hv_get_drvdata(device
);
393 if (stor_device
&& stor_device
->destroy
)
400 static inline void storvsc_wait_to_drain(struct storvsc_device
*dev
)
402 dev
->drain_notify
= true;
403 wait_event(dev
->waiting_to_drain
,
404 atomic_read(&dev
->num_outstanding_req
) == 0);
405 dev
->drain_notify
= false;
408 static inline struct storvsc_device
*get_in_stor_device(
409 struct hv_device
*device
)
411 struct storvsc_device
*stor_device
;
413 stor_device
= hv_get_drvdata(device
);
419 * If the device is being destroyed; allow incoming
420 * traffic only to cleanup outstanding requests.
423 if (stor_device
->destroy
&&
424 (atomic_read(&stor_device
->num_outstanding_req
) == 0))
432 static void destroy_bounce_buffer(struct scatterlist
*sgl
,
433 unsigned int sg_count
)
436 struct page
*page_buf
;
438 for (i
= 0; i
< sg_count
; i
++) {
439 page_buf
= sg_page((&sgl
[i
]));
440 if (page_buf
!= NULL
)
441 __free_page(page_buf
);
447 static int do_bounce_buffer(struct scatterlist
*sgl
, unsigned int sg_count
)
451 /* No need to check */
455 /* We have at least 2 sg entries */
456 for (i
= 0; i
< sg_count
; i
++) {
458 /* make sure 1st one does not have hole */
459 if (sgl
[i
].offset
+ sgl
[i
].length
!= PAGE_SIZE
)
461 } else if (i
== sg_count
- 1) {
462 /* make sure last one does not have hole */
463 if (sgl
[i
].offset
!= 0)
466 /* make sure no hole in the middle */
467 if (sgl
[i
].length
!= PAGE_SIZE
|| sgl
[i
].offset
!= 0)
474 static struct scatterlist
*create_bounce_buffer(struct scatterlist
*sgl
,
475 unsigned int sg_count
,
481 struct scatterlist
*bounce_sgl
;
482 struct page
*page_buf
;
483 unsigned int buf_len
= ((write
== WRITE_TYPE
) ? 0 : PAGE_SIZE
);
485 num_pages
= ALIGN(len
, PAGE_SIZE
) >> PAGE_SHIFT
;
487 bounce_sgl
= kcalloc(num_pages
, sizeof(struct scatterlist
), GFP_ATOMIC
);
491 sg_init_table(bounce_sgl
, num_pages
);
492 for (i
= 0; i
< num_pages
; i
++) {
493 page_buf
= alloc_page(GFP_ATOMIC
);
496 sg_set_page(&bounce_sgl
[i
], page_buf
, buf_len
, 0);
502 destroy_bounce_buffer(bounce_sgl
, num_pages
);
506 /* Disgusting wrapper functions */
507 static inline unsigned long sg_kmap_atomic(struct scatterlist
*sgl
, int idx
)
509 void *addr
= kmap_atomic(sg_page(sgl
+ idx
));
510 return (unsigned long)addr
;
513 static inline void sg_kunmap_atomic(unsigned long addr
)
515 kunmap_atomic((void *)addr
);
519 /* Assume the original sgl has enough room */
520 static unsigned int copy_from_bounce_buffer(struct scatterlist
*orig_sgl
,
521 struct scatterlist
*bounce_sgl
,
522 unsigned int orig_sgl_count
,
523 unsigned int bounce_sgl_count
)
527 unsigned long src
, dest
;
528 unsigned int srclen
, destlen
, copylen
;
529 unsigned int total_copied
= 0;
530 unsigned long bounce_addr
= 0;
531 unsigned long dest_addr
= 0;
534 local_irq_save(flags
);
536 for (i
= 0; i
< orig_sgl_count
; i
++) {
537 dest_addr
= sg_kmap_atomic(orig_sgl
,i
) + orig_sgl
[i
].offset
;
539 destlen
= orig_sgl
[i
].length
;
541 if (bounce_addr
== 0)
542 bounce_addr
= sg_kmap_atomic(bounce_sgl
,j
);
545 src
= bounce_addr
+ bounce_sgl
[j
].offset
;
546 srclen
= bounce_sgl
[j
].length
- bounce_sgl
[j
].offset
;
548 copylen
= min(srclen
, destlen
);
549 memcpy((void *)dest
, (void *)src
, copylen
);
551 total_copied
+= copylen
;
552 bounce_sgl
[j
].offset
+= copylen
;
556 if (bounce_sgl
[j
].offset
== bounce_sgl
[j
].length
) {
558 sg_kunmap_atomic(bounce_addr
);
562 * It is possible that the number of elements
563 * in the bounce buffer may not be equal to
564 * the number of elements in the original
565 * scatter list. Handle this correctly.
568 if (j
== bounce_sgl_count
) {
570 * We are done; cleanup and return.
572 sg_kunmap_atomic(dest_addr
- orig_sgl
[i
].offset
);
573 local_irq_restore(flags
);
577 /* if we need to use another bounce buffer */
578 if (destlen
|| i
!= orig_sgl_count
- 1)
579 bounce_addr
= sg_kmap_atomic(bounce_sgl
,j
);
580 } else if (destlen
== 0 && i
== orig_sgl_count
- 1) {
581 /* unmap the last bounce that is < PAGE_SIZE */
582 sg_kunmap_atomic(bounce_addr
);
586 sg_kunmap_atomic(dest_addr
- orig_sgl
[i
].offset
);
589 local_irq_restore(flags
);
594 /* Assume the bounce_sgl has enough room ie using the create_bounce_buffer() */
595 static unsigned int copy_to_bounce_buffer(struct scatterlist
*orig_sgl
,
596 struct scatterlist
*bounce_sgl
,
597 unsigned int orig_sgl_count
)
601 unsigned long src
, dest
;
602 unsigned int srclen
, destlen
, copylen
;
603 unsigned int total_copied
= 0;
604 unsigned long bounce_addr
= 0;
605 unsigned long src_addr
= 0;
608 local_irq_save(flags
);
610 for (i
= 0; i
< orig_sgl_count
; i
++) {
611 src_addr
= sg_kmap_atomic(orig_sgl
,i
) + orig_sgl
[i
].offset
;
613 srclen
= orig_sgl
[i
].length
;
615 if (bounce_addr
== 0)
616 bounce_addr
= sg_kmap_atomic(bounce_sgl
,j
);
619 /* assume bounce offset always == 0 */
620 dest
= bounce_addr
+ bounce_sgl
[j
].length
;
621 destlen
= PAGE_SIZE
- bounce_sgl
[j
].length
;
623 copylen
= min(srclen
, destlen
);
624 memcpy((void *)dest
, (void *)src
, copylen
);
626 total_copied
+= copylen
;
627 bounce_sgl
[j
].length
+= copylen
;
631 if (bounce_sgl
[j
].length
== PAGE_SIZE
) {
632 /* full..move to next entry */
633 sg_kunmap_atomic(bounce_addr
);
638 /* if we need to use another bounce buffer */
639 if (srclen
&& bounce_addr
== 0)
640 bounce_addr
= sg_kmap_atomic(bounce_sgl
, j
);
644 sg_kunmap_atomic(src_addr
- orig_sgl
[i
].offset
);
648 sg_kunmap_atomic(bounce_addr
);
650 local_irq_restore(flags
);
655 static int storvsc_channel_init(struct hv_device
*device
)
657 struct storvsc_device
*stor_device
;
658 struct storvsc_cmd_request
*request
;
659 struct vstor_packet
*vstor_packet
;
662 stor_device
= get_out_stor_device(device
);
666 request
= &stor_device
->init_request
;
667 vstor_packet
= &request
->vstor_packet
;
670 * Now, initiate the vsc/vsp initialization protocol on the open
673 memset(request
, 0, sizeof(struct storvsc_cmd_request
));
674 init_completion(&request
->wait_event
);
675 vstor_packet
->operation
= VSTOR_OPERATION_BEGIN_INITIALIZATION
;
676 vstor_packet
->flags
= REQUEST_COMPLETION_FLAG
;
678 ret
= vmbus_sendpacket(device
->channel
, vstor_packet
,
679 sizeof(struct vstor_packet
),
680 (unsigned long)request
,
682 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
);
686 t
= wait_for_completion_timeout(&request
->wait_event
, 5*HZ
);
692 if (vstor_packet
->operation
!= VSTOR_OPERATION_COMPLETE_IO
||
693 vstor_packet
->status
!= 0)
697 /* reuse the packet for version range supported */
698 memset(vstor_packet
, 0, sizeof(struct vstor_packet
));
699 vstor_packet
->operation
= VSTOR_OPERATION_QUERY_PROTOCOL_VERSION
;
700 vstor_packet
->flags
= REQUEST_COMPLETION_FLAG
;
702 vstor_packet
->version
.major_minor
=
703 storvsc_get_version(VMSTOR_CURRENT_MAJOR
, VMSTOR_CURRENT_MINOR
);
706 * The revision number is only used in Windows; set it to 0.
708 vstor_packet
->version
.revision
= 0;
710 ret
= vmbus_sendpacket(device
->channel
, vstor_packet
,
711 sizeof(struct vstor_packet
),
712 (unsigned long)request
,
714 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
);
718 t
= wait_for_completion_timeout(&request
->wait_event
, 5*HZ
);
724 if (vstor_packet
->operation
!= VSTOR_OPERATION_COMPLETE_IO
||
725 vstor_packet
->status
!= 0)
729 memset(vstor_packet
, 0, sizeof(struct vstor_packet
));
730 vstor_packet
->operation
= VSTOR_OPERATION_QUERY_PROPERTIES
;
731 vstor_packet
->flags
= REQUEST_COMPLETION_FLAG
;
732 vstor_packet
->storage_channel_properties
.port_number
=
733 stor_device
->port_number
;
735 ret
= vmbus_sendpacket(device
->channel
, vstor_packet
,
736 sizeof(struct vstor_packet
),
737 (unsigned long)request
,
739 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
);
744 t
= wait_for_completion_timeout(&request
->wait_event
, 5*HZ
);
750 if (vstor_packet
->operation
!= VSTOR_OPERATION_COMPLETE_IO
||
751 vstor_packet
->status
!= 0)
754 stor_device
->path_id
= vstor_packet
->storage_channel_properties
.path_id
;
755 stor_device
->target_id
756 = vstor_packet
->storage_channel_properties
.target_id
;
758 memset(vstor_packet
, 0, sizeof(struct vstor_packet
));
759 vstor_packet
->operation
= VSTOR_OPERATION_END_INITIALIZATION
;
760 vstor_packet
->flags
= REQUEST_COMPLETION_FLAG
;
762 ret
= vmbus_sendpacket(device
->channel
, vstor_packet
,
763 sizeof(struct vstor_packet
),
764 (unsigned long)request
,
766 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
);
771 t
= wait_for_completion_timeout(&request
->wait_event
, 5*HZ
);
777 if (vstor_packet
->operation
!= VSTOR_OPERATION_COMPLETE_IO
||
778 vstor_packet
->status
!= 0)
786 static void storvsc_handle_error(struct vmscsi_request
*vm_srb
,
787 struct scsi_cmnd
*scmnd
,
788 struct Scsi_Host
*host
,
791 struct storvsc_scan_work
*wrk
;
792 void (*process_err_fn
)(struct work_struct
*work
);
793 bool do_work
= false;
795 switch (vm_srb
->srb_status
) {
796 case SRB_STATUS_ERROR
:
798 * If there is an error; offline the device since all
799 * error recovery strategies would have already been
800 * deployed on the host side. However, if the command
801 * were a pass-through command deal with it appropriately.
803 switch (scmnd
->cmnd
[0]) {
806 set_host_byte(scmnd
, DID_PASSTHROUGH
);
809 * On Some Windows hosts TEST_UNIT_READY command can return
810 * SRB_STATUS_ERROR, let the upper level code deal with it
811 * based on the sense information.
813 case TEST_UNIT_READY
:
816 set_host_byte(scmnd
, DID_TARGET_FAILURE
);
819 case SRB_STATUS_INVALID_LUN
:
821 process_err_fn
= storvsc_remove_lun
;
823 case (SRB_STATUS_ABORTED
| SRB_STATUS_AUTOSENSE_VALID
):
824 if ((asc
== 0x2a) && (ascq
== 0x9)) {
826 process_err_fn
= storvsc_device_scan
;
828 * Retry the I/O that trigerred this.
830 set_host_byte(scmnd
, DID_REQUEUE
);
839 * We need to schedule work to process this error; schedule it.
841 wrk
= kmalloc(sizeof(struct storvsc_scan_work
), GFP_ATOMIC
);
843 set_host_byte(scmnd
, DID_TARGET_FAILURE
);
848 wrk
->lun
= vm_srb
->lun
;
849 INIT_WORK(&wrk
->work
, process_err_fn
);
850 schedule_work(&wrk
->work
);
854 static void storvsc_command_completion(struct storvsc_cmd_request
*cmd_request
)
856 struct scsi_cmnd
*scmnd
= cmd_request
->cmd
;
857 struct hv_host_device
*host_dev
= shost_priv(scmnd
->device
->host
);
858 void (*scsi_done_fn
)(struct scsi_cmnd
*);
859 struct scsi_sense_hdr sense_hdr
;
860 struct vmscsi_request
*vm_srb
;
861 struct stor_mem_pools
*memp
= scmnd
->device
->hostdata
;
862 struct Scsi_Host
*host
;
863 struct storvsc_device
*stor_dev
;
864 struct hv_device
*dev
= host_dev
->dev
;
866 stor_dev
= get_in_stor_device(dev
);
867 host
= stor_dev
->host
;
869 vm_srb
= &cmd_request
->vstor_packet
.vm_srb
;
870 if (cmd_request
->bounce_sgl_count
) {
871 if (vm_srb
->data_in
== READ_TYPE
)
872 copy_from_bounce_buffer(scsi_sglist(scmnd
),
873 cmd_request
->bounce_sgl
,
874 scsi_sg_count(scmnd
),
875 cmd_request
->bounce_sgl_count
);
876 destroy_bounce_buffer(cmd_request
->bounce_sgl
,
877 cmd_request
->bounce_sgl_count
);
880 scmnd
->result
= vm_srb
->scsi_status
;
883 if (scsi_normalize_sense(scmnd
->sense_buffer
,
884 SCSI_SENSE_BUFFERSIZE
, &sense_hdr
))
885 scsi_print_sense_hdr("storvsc", &sense_hdr
);
888 if (vm_srb
->srb_status
!= SRB_STATUS_SUCCESS
)
889 storvsc_handle_error(vm_srb
, scmnd
, host
, sense_hdr
.asc
,
892 scsi_set_resid(scmnd
,
893 cmd_request
->data_buffer
.len
-
894 vm_srb
->data_transfer_length
);
896 scsi_done_fn
= scmnd
->scsi_done
;
898 scmnd
->host_scribble
= NULL
;
899 scmnd
->scsi_done
= NULL
;
903 mempool_free(cmd_request
, memp
->request_mempool
);
906 static void storvsc_on_io_completion(struct hv_device
*device
,
907 struct vstor_packet
*vstor_packet
,
908 struct storvsc_cmd_request
*request
)
910 struct storvsc_device
*stor_device
;
911 struct vstor_packet
*stor_pkt
;
913 stor_device
= hv_get_drvdata(device
);
914 stor_pkt
= &request
->vstor_packet
;
917 * The current SCSI handling on the host side does
918 * not correctly handle:
919 * INQUIRY command with page code parameter set to 0x80
920 * MODE_SENSE command with cmd[2] == 0x1c
922 * Setup srb and scsi status so this won't be fatal.
923 * We do this so we can distinguish truly fatal failues
924 * (srb status == 0x4) and off-line the device in that case.
927 if ((stor_pkt
->vm_srb
.cdb
[0] == INQUIRY
) ||
928 (stor_pkt
->vm_srb
.cdb
[0] == MODE_SENSE
)) {
929 vstor_packet
->vm_srb
.scsi_status
= 0;
930 vstor_packet
->vm_srb
.srb_status
= SRB_STATUS_SUCCESS
;
934 /* Copy over the status...etc */
935 stor_pkt
->vm_srb
.scsi_status
= vstor_packet
->vm_srb
.scsi_status
;
936 stor_pkt
->vm_srb
.srb_status
= vstor_packet
->vm_srb
.srb_status
;
937 stor_pkt
->vm_srb
.sense_info_length
=
938 vstor_packet
->vm_srb
.sense_info_length
;
940 if (vstor_packet
->vm_srb
.scsi_status
!= 0 ||
941 vstor_packet
->vm_srb
.srb_status
!= SRB_STATUS_SUCCESS
){
942 dev_warn(&device
->device
,
943 "cmd 0x%x scsi status 0x%x srb status 0x%x\n",
944 stor_pkt
->vm_srb
.cdb
[0],
945 vstor_packet
->vm_srb
.scsi_status
,
946 vstor_packet
->vm_srb
.srb_status
);
949 if ((vstor_packet
->vm_srb
.scsi_status
& 0xFF) == 0x02) {
950 /* CHECK_CONDITION */
951 if (vstor_packet
->vm_srb
.srb_status
&
952 SRB_STATUS_AUTOSENSE_VALID
) {
953 /* autosense data available */
954 dev_warn(&device
->device
,
955 "stor pkt %p autosense data valid - len %d\n",
957 vstor_packet
->vm_srb
.sense_info_length
);
959 memcpy(request
->sense_buffer
,
960 vstor_packet
->vm_srb
.sense_data
,
961 vstor_packet
->vm_srb
.sense_info_length
);
966 stor_pkt
->vm_srb
.data_transfer_length
=
967 vstor_packet
->vm_srb
.data_transfer_length
;
969 storvsc_command_completion(request
);
971 if (atomic_dec_and_test(&stor_device
->num_outstanding_req
) &&
972 stor_device
->drain_notify
)
973 wake_up(&stor_device
->waiting_to_drain
);
978 static void storvsc_on_receive(struct hv_device
*device
,
979 struct vstor_packet
*vstor_packet
,
980 struct storvsc_cmd_request
*request
)
982 struct storvsc_scan_work
*work
;
983 struct storvsc_device
*stor_device
;
985 switch (vstor_packet
->operation
) {
986 case VSTOR_OPERATION_COMPLETE_IO
:
987 storvsc_on_io_completion(device
, vstor_packet
, request
);
990 case VSTOR_OPERATION_REMOVE_DEVICE
:
991 case VSTOR_OPERATION_ENUMERATE_BUS
:
992 stor_device
= get_in_stor_device(device
);
993 work
= kmalloc(sizeof(struct storvsc_scan_work
), GFP_ATOMIC
);
997 INIT_WORK(&work
->work
, storvsc_bus_scan
);
998 work
->host
= stor_device
->host
;
999 schedule_work(&work
->work
);
1007 static void storvsc_on_channel_callback(void *context
)
1009 struct hv_device
*device
= (struct hv_device
*)context
;
1010 struct storvsc_device
*stor_device
;
1013 unsigned char packet
[ALIGN(sizeof(struct vstor_packet
), 8)];
1014 struct storvsc_cmd_request
*request
;
1018 stor_device
= get_in_stor_device(device
);
1023 ret
= vmbus_recvpacket(device
->channel
, packet
,
1024 ALIGN(sizeof(struct vstor_packet
), 8),
1025 &bytes_recvd
, &request_id
);
1026 if (ret
== 0 && bytes_recvd
> 0) {
1028 request
= (struct storvsc_cmd_request
*)
1029 (unsigned long)request_id
;
1031 if ((request
== &stor_device
->init_request
) ||
1032 (request
== &stor_device
->reset_request
)) {
1034 memcpy(&request
->vstor_packet
, packet
,
1035 sizeof(struct vstor_packet
));
1036 complete(&request
->wait_event
);
1038 storvsc_on_receive(device
,
1039 (struct vstor_packet
*)packet
,
1050 static int storvsc_connect_to_vsp(struct hv_device
*device
, u32 ring_size
)
1052 struct vmstorage_channel_properties props
;
1055 memset(&props
, 0, sizeof(struct vmstorage_channel_properties
));
1057 ret
= vmbus_open(device
->channel
,
1061 sizeof(struct vmstorage_channel_properties
),
1062 storvsc_on_channel_callback
, device
);
1067 ret
= storvsc_channel_init(device
);
1072 static int storvsc_dev_remove(struct hv_device
*device
)
1074 struct storvsc_device
*stor_device
;
1075 unsigned long flags
;
1077 stor_device
= hv_get_drvdata(device
);
1079 spin_lock_irqsave(&device
->channel
->inbound_lock
, flags
);
1080 stor_device
->destroy
= true;
1081 spin_unlock_irqrestore(&device
->channel
->inbound_lock
, flags
);
1084 * At this point, all outbound traffic should be disable. We
1085 * only allow inbound traffic (responses) to proceed so that
1086 * outstanding requests can be completed.
1089 storvsc_wait_to_drain(stor_device
);
1092 * Since we have already drained, we don't need to busy wait
1093 * as was done in final_release_stor_device()
1094 * Note that we cannot set the ext pointer to NULL until
1095 * we have drained - to drain the outgoing packets, we need to
1096 * allow incoming packets.
1098 spin_lock_irqsave(&device
->channel
->inbound_lock
, flags
);
1099 hv_set_drvdata(device
, NULL
);
1100 spin_unlock_irqrestore(&device
->channel
->inbound_lock
, flags
);
1102 /* Close the channel */
1103 vmbus_close(device
->channel
);
1109 static int storvsc_do_io(struct hv_device
*device
,
1110 struct storvsc_cmd_request
*request
)
1112 struct storvsc_device
*stor_device
;
1113 struct vstor_packet
*vstor_packet
;
1116 vstor_packet
= &request
->vstor_packet
;
1117 stor_device
= get_out_stor_device(device
);
1123 request
->device
= device
;
1126 vstor_packet
->flags
|= REQUEST_COMPLETION_FLAG
;
1128 vstor_packet
->vm_srb
.length
= sizeof(struct vmscsi_request
);
1131 vstor_packet
->vm_srb
.sense_info_length
= STORVSC_SENSE_BUFFER_SIZE
;
1134 vstor_packet
->vm_srb
.data_transfer_length
=
1135 request
->data_buffer
.len
;
1137 vstor_packet
->operation
= VSTOR_OPERATION_EXECUTE_SRB
;
1139 if (request
->data_buffer
.len
) {
1140 ret
= vmbus_sendpacket_multipagebuffer(device
->channel
,
1141 &request
->data_buffer
,
1143 sizeof(struct vstor_packet
),
1144 (unsigned long)request
);
1146 ret
= vmbus_sendpacket(device
->channel
, vstor_packet
,
1147 sizeof(struct vstor_packet
),
1148 (unsigned long)request
,
1150 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
);
1156 atomic_inc(&stor_device
->num_outstanding_req
);
1161 static int storvsc_device_alloc(struct scsi_device
*sdevice
)
1163 struct stor_mem_pools
*memp
;
1164 int number
= STORVSC_MIN_BUF_NR
;
1166 memp
= kzalloc(sizeof(struct stor_mem_pools
), GFP_KERNEL
);
1170 memp
->request_pool
=
1171 kmem_cache_create(dev_name(&sdevice
->sdev_dev
),
1172 sizeof(struct storvsc_cmd_request
), 0,
1173 SLAB_HWCACHE_ALIGN
, NULL
);
1175 if (!memp
->request_pool
)
1178 memp
->request_mempool
= mempool_create(number
, mempool_alloc_slab
,
1180 memp
->request_pool
);
1182 if (!memp
->request_mempool
)
1185 sdevice
->hostdata
= memp
;
1190 kmem_cache_destroy(memp
->request_pool
);
1197 static void storvsc_device_destroy(struct scsi_device
*sdevice
)
1199 struct stor_mem_pools
*memp
= sdevice
->hostdata
;
1204 mempool_destroy(memp
->request_mempool
);
1205 kmem_cache_destroy(memp
->request_pool
);
1207 sdevice
->hostdata
= NULL
;
1210 static int storvsc_device_configure(struct scsi_device
*sdevice
)
1212 scsi_adjust_queue_depth(sdevice
, MSG_SIMPLE_TAG
,
1213 STORVSC_MAX_IO_REQUESTS
);
1215 blk_queue_max_segment_size(sdevice
->request_queue
, PAGE_SIZE
);
1217 blk_queue_bounce_limit(sdevice
->request_queue
, BLK_BOUNCE_ANY
);
1219 sdevice
->no_write_same
= 1;
1224 static int storvsc_get_chs(struct scsi_device
*sdev
, struct block_device
* bdev
,
1225 sector_t capacity
, int *info
)
1227 sector_t nsect
= capacity
;
1228 sector_t cylinders
= nsect
;
1229 int heads
, sectors_pt
;
1232 * We are making up these values; let us keep it simple.
1235 sectors_pt
= 0x3f; /* Sectors per track */
1236 sector_div(cylinders
, heads
* sectors_pt
);
1237 if ((sector_t
)(cylinders
+ 1) * heads
* sectors_pt
< nsect
)
1241 info
[1] = sectors_pt
;
1242 info
[2] = (int)cylinders
;
1247 static int storvsc_host_reset_handler(struct scsi_cmnd
*scmnd
)
1249 struct hv_host_device
*host_dev
= shost_priv(scmnd
->device
->host
);
1250 struct hv_device
*device
= host_dev
->dev
;
1252 struct storvsc_device
*stor_device
;
1253 struct storvsc_cmd_request
*request
;
1254 struct vstor_packet
*vstor_packet
;
1258 stor_device
= get_out_stor_device(device
);
1262 request
= &stor_device
->reset_request
;
1263 vstor_packet
= &request
->vstor_packet
;
1265 init_completion(&request
->wait_event
);
1267 vstor_packet
->operation
= VSTOR_OPERATION_RESET_BUS
;
1268 vstor_packet
->flags
= REQUEST_COMPLETION_FLAG
;
1269 vstor_packet
->vm_srb
.path_id
= stor_device
->path_id
;
1271 ret
= vmbus_sendpacket(device
->channel
, vstor_packet
,
1272 sizeof(struct vstor_packet
),
1273 (unsigned long)&stor_device
->reset_request
,
1275 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
);
1279 t
= wait_for_completion_timeout(&request
->wait_event
, 5*HZ
);
1281 return TIMEOUT_ERROR
;
1285 * At this point, all outstanding requests in the adapter
1286 * should have been flushed out and return to us
1287 * There is a potential race here where the host may be in
1288 * the process of responding when we return from here.
1289 * Just wait for all in-transit packets to be accounted for
1290 * before we return from here.
1292 storvsc_wait_to_drain(stor_device
);
1298 * The host guarantees to respond to each command, although I/O latencies might
1299 * be unbounded on Azure. Reset the timer unconditionally to give the host a
1300 * chance to perform EH.
1302 static enum blk_eh_timer_return
storvsc_eh_timed_out(struct scsi_cmnd
*scmnd
)
1304 return BLK_EH_RESET_TIMER
;
1307 static bool storvsc_scsi_cmd_ok(struct scsi_cmnd
*scmnd
)
1309 bool allowed
= true;
1310 u8 scsi_op
= scmnd
->cmnd
[0];
1313 /* the host does not handle WRITE_SAME, log accident usage */
1316 * smartd sends this command and the host does not handle
1317 * this. So, don't send it.
1320 scmnd
->result
= ILLEGAL_REQUEST
<< 16;
1329 static int storvsc_queuecommand(struct Scsi_Host
*host
, struct scsi_cmnd
*scmnd
)
1332 struct hv_host_device
*host_dev
= shost_priv(host
);
1333 struct hv_device
*dev
= host_dev
->dev
;
1334 struct storvsc_cmd_request
*cmd_request
;
1335 unsigned int request_size
= 0;
1337 struct scatterlist
*sgl
;
1338 unsigned int sg_count
= 0;
1339 struct vmscsi_request
*vm_srb
;
1340 struct stor_mem_pools
*memp
= scmnd
->device
->hostdata
;
1342 if (!storvsc_scsi_cmd_ok(scmnd
)) {
1343 scmnd
->scsi_done(scmnd
);
1347 request_size
= sizeof(struct storvsc_cmd_request
);
1349 cmd_request
= mempool_alloc(memp
->request_mempool
,
1353 * We might be invoked in an interrupt context; hence
1354 * mempool_alloc() can fail.
1357 return SCSI_MLQUEUE_DEVICE_BUSY
;
1359 memset(cmd_request
, 0, sizeof(struct storvsc_cmd_request
));
1361 /* Setup the cmd request */
1362 cmd_request
->cmd
= scmnd
;
1364 scmnd
->host_scribble
= (unsigned char *)cmd_request
;
1366 vm_srb
= &cmd_request
->vstor_packet
.vm_srb
;
1370 switch (scmnd
->sc_data_direction
) {
1372 vm_srb
->data_in
= WRITE_TYPE
;
1374 case DMA_FROM_DEVICE
:
1375 vm_srb
->data_in
= READ_TYPE
;
1378 vm_srb
->data_in
= UNKNOWN_TYPE
;
1383 vm_srb
->port_number
= host_dev
->port
;
1384 vm_srb
->path_id
= scmnd
->device
->channel
;
1385 vm_srb
->target_id
= scmnd
->device
->id
;
1386 vm_srb
->lun
= scmnd
->device
->lun
;
1388 vm_srb
->cdb_length
= scmnd
->cmd_len
;
1390 memcpy(vm_srb
->cdb
, scmnd
->cmnd
, vm_srb
->cdb_length
);
1392 cmd_request
->sense_buffer
= scmnd
->sense_buffer
;
1395 cmd_request
->data_buffer
.len
= scsi_bufflen(scmnd
);
1396 if (scsi_sg_count(scmnd
)) {
1397 sgl
= (struct scatterlist
*)scsi_sglist(scmnd
);
1398 sg_count
= scsi_sg_count(scmnd
);
1400 /* check if we need to bounce the sgl */
1401 if (do_bounce_buffer(sgl
, scsi_sg_count(scmnd
)) != -1) {
1402 cmd_request
->bounce_sgl
=
1403 create_bounce_buffer(sgl
, scsi_sg_count(scmnd
),
1404 scsi_bufflen(scmnd
),
1406 if (!cmd_request
->bounce_sgl
) {
1407 ret
= SCSI_MLQUEUE_HOST_BUSY
;
1411 cmd_request
->bounce_sgl_count
=
1412 ALIGN(scsi_bufflen(scmnd
), PAGE_SIZE
) >>
1415 if (vm_srb
->data_in
== WRITE_TYPE
)
1416 copy_to_bounce_buffer(sgl
,
1417 cmd_request
->bounce_sgl
,
1418 scsi_sg_count(scmnd
));
1420 sgl
= cmd_request
->bounce_sgl
;
1421 sg_count
= cmd_request
->bounce_sgl_count
;
1424 cmd_request
->data_buffer
.offset
= sgl
[0].offset
;
1426 for (i
= 0; i
< sg_count
; i
++)
1427 cmd_request
->data_buffer
.pfn_array
[i
] =
1428 page_to_pfn(sg_page((&sgl
[i
])));
1430 } else if (scsi_sglist(scmnd
)) {
1431 cmd_request
->data_buffer
.offset
=
1432 virt_to_phys(scsi_sglist(scmnd
)) & (PAGE_SIZE
-1);
1433 cmd_request
->data_buffer
.pfn_array
[0] =
1434 virt_to_phys(scsi_sglist(scmnd
)) >> PAGE_SHIFT
;
1437 /* Invokes the vsc to start an IO */
1438 ret
= storvsc_do_io(dev
, cmd_request
);
1440 if (ret
== -EAGAIN
) {
1443 if (cmd_request
->bounce_sgl_count
)
1444 destroy_bounce_buffer(cmd_request
->bounce_sgl
,
1445 cmd_request
->bounce_sgl_count
);
1447 ret
= SCSI_MLQUEUE_DEVICE_BUSY
;
1454 mempool_free(cmd_request
, memp
->request_mempool
);
1455 scmnd
->host_scribble
= NULL
;
1459 static struct scsi_host_template scsi_driver
= {
1460 .module
= THIS_MODULE
,
1461 .name
= "storvsc_host_t",
1462 .bios_param
= storvsc_get_chs
,
1463 .queuecommand
= storvsc_queuecommand
,
1464 .eh_host_reset_handler
= storvsc_host_reset_handler
,
1465 .eh_timed_out
= storvsc_eh_timed_out
,
1466 .slave_alloc
= storvsc_device_alloc
,
1467 .slave_destroy
= storvsc_device_destroy
,
1468 .slave_configure
= storvsc_device_configure
,
1470 /* 64 max_queue * 1 target */
1471 .can_queue
= STORVSC_MAX_IO_REQUESTS
*STORVSC_MAX_TARGETS
,
1473 /* no use setting to 0 since ll_blk_rw reset it to 1 */
1475 .sg_tablesize
= MAX_MULTIPAGE_BUFFER_COUNT
,
1476 .use_clustering
= DISABLE_CLUSTERING
,
1477 /* Make sure we dont get a sg segment crosses a page boundary */
1478 .dma_boundary
= PAGE_SIZE
-1,
1487 static const struct hv_vmbus_device_id id_table
[] = {
1490 .driver_data
= SCSI_GUID
1494 .driver_data
= IDE_GUID
1499 MODULE_DEVICE_TABLE(vmbus
, id_table
);
1501 static int storvsc_probe(struct hv_device
*device
,
1502 const struct hv_vmbus_device_id
*dev_id
)
1505 struct Scsi_Host
*host
;
1506 struct hv_host_device
*host_dev
;
1507 bool dev_is_ide
= ((dev_id
->driver_data
== IDE_GUID
) ? true : false);
1509 struct storvsc_device
*stor_device
;
1511 host
= scsi_host_alloc(&scsi_driver
,
1512 sizeof(struct hv_host_device
));
1516 host_dev
= shost_priv(host
);
1517 memset(host_dev
, 0, sizeof(struct hv_host_device
));
1519 host_dev
->port
= host
->host_no
;
1520 host_dev
->dev
= device
;
1523 stor_device
= kzalloc(sizeof(struct storvsc_device
), GFP_KERNEL
);
1529 stor_device
->destroy
= false;
1530 init_waitqueue_head(&stor_device
->waiting_to_drain
);
1531 stor_device
->device
= device
;
1532 stor_device
->host
= host
;
1533 hv_set_drvdata(device
, stor_device
);
1535 stor_device
->port_number
= host
->host_no
;
1536 ret
= storvsc_connect_to_vsp(device
, storvsc_ringbuffer_size
);
1540 host_dev
->path
= stor_device
->path_id
;
1541 host_dev
->target
= stor_device
->target_id
;
1543 /* max # of devices per target */
1544 host
->max_lun
= STORVSC_MAX_LUNS_PER_TARGET
;
1545 /* max # of targets per channel */
1546 host
->max_id
= STORVSC_MAX_TARGETS
;
1547 /* max # of channels */
1548 host
->max_channel
= STORVSC_MAX_CHANNELS
- 1;
1549 /* max cmd length */
1550 host
->max_cmd_len
= STORVSC_MAX_CMD_LEN
;
1552 /* Register the HBA and start the scsi bus scan */
1553 ret
= scsi_add_host(host
, &device
->device
);
1558 scsi_scan_host(host
);
1560 target
= (device
->dev_instance
.b
[5] << 8 |
1561 device
->dev_instance
.b
[4]);
1562 ret
= scsi_add_device(host
, 0, target
, 0);
1564 scsi_remove_host(host
);
1572 * Once we have connected with the host, we would need to
1573 * to invoke storvsc_dev_remove() to rollback this state and
1574 * this call also frees up the stor_device; hence the jump around
1577 storvsc_dev_remove(device
);
1584 scsi_host_put(host
);
1588 static int storvsc_remove(struct hv_device
*dev
)
1590 struct storvsc_device
*stor_device
= hv_get_drvdata(dev
);
1591 struct Scsi_Host
*host
= stor_device
->host
;
1593 scsi_remove_host(host
);
1594 storvsc_dev_remove(dev
);
1595 scsi_host_put(host
);
1600 static struct hv_driver storvsc_drv
= {
1601 .name
= KBUILD_MODNAME
,
1602 .id_table
= id_table
,
1603 .probe
= storvsc_probe
,
1604 .remove
= storvsc_remove
,
1607 static int __init
storvsc_drv_init(void)
1609 u32 max_outstanding_req_per_channel
;
1612 * Divide the ring buffer data size (which is 1 page less
1613 * than the ring buffer size since that page is reserved for
1614 * the ring buffer indices) by the max request size (which is
1615 * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
1617 max_outstanding_req_per_channel
=
1618 ((storvsc_ringbuffer_size
- PAGE_SIZE
) /
1619 ALIGN(MAX_MULTIPAGE_BUFFER_PACKET
+
1620 sizeof(struct vstor_packet
) + sizeof(u64
),
1623 if (max_outstanding_req_per_channel
<
1624 STORVSC_MAX_IO_REQUESTS
)
1627 return vmbus_driver_register(&storvsc_drv
);
1630 static void __exit
storvsc_drv_exit(void)
1632 vmbus_driver_unregister(&storvsc_drv
);
1635 MODULE_LICENSE("GPL");
1636 MODULE_VERSION(HV_DRV_VERSION
);
1637 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
1638 module_init(storvsc_drv_init
);
1639 module_exit(storvsc_drv_exit
);