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 <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>
46 * All wire protocol details (storage protocol between the guest and the host)
47 * are consolidated here.
49 * Begin protocol definitions.
55 * V1 RC < 2008/1/31: 1.0
56 * V1 RC > 2008/1/31: 2.0
60 #define VMSTOR_CURRENT_MAJOR 4
61 #define VMSTOR_CURRENT_MINOR 2
64 /* Packet structure describing virtual storage requests. */
65 enum vstor_packet_operation {
66 VSTOR_OPERATION_COMPLETE_IO = 1,
67 VSTOR_OPERATION_REMOVE_DEVICE = 2,
68 VSTOR_OPERATION_EXECUTE_SRB = 3,
69 VSTOR_OPERATION_RESET_LUN = 4,
70 VSTOR_OPERATION_RESET_ADAPTER = 5,
71 VSTOR_OPERATION_RESET_BUS = 6,
72 VSTOR_OPERATION_BEGIN_INITIALIZATION = 7,
73 VSTOR_OPERATION_END_INITIALIZATION = 8,
74 VSTOR_OPERATION_QUERY_PROTOCOL_VERSION = 9,
75 VSTOR_OPERATION_QUERY_PROPERTIES = 10,
76 VSTOR_OPERATION_ENUMERATE_BUS = 11,
77 VSTOR_OPERATION_MAXIMUM = 11
81 * Platform neutral description of a scsi request -
82 * this remains the same across the write regardless of 32/64 bit
83 * note: it's patterned off the SCSI_PASS_THROUGH structure
85 #define STORVSC_MAX_CMD_LEN 0x10
86 #define STORVSC_SENSE_BUFFER_SIZE 0x12
87 #define STORVSC_MAX_BUF_LEN_WITH_PADDING 0x14
89 struct vmscsi_request {
100 u8 sense_info_length;
104 u32 data_transfer_length;
107 u8 cdb[STORVSC_MAX_CMD_LEN];
108 u8 sense_data[STORVSC_SENSE_BUFFER_SIZE];
109 u8 reserved_array[STORVSC_MAX_BUF_LEN_WITH_PADDING];
111 } __attribute((packed));
115 * This structure is sent during the intialization phase to get the different
116 * properties of the channel.
118 struct vmstorage_channel_properties {
119 u16 protocol_version;
123 /* Note: port number is only really known on the client side */
126 u32 max_transfer_bytes;
129 * This id is unique for each channel and will correspond with
130 * vendor specific data in the inquiry data.
136 /* This structure is sent during the storage protocol negotiations. */
137 struct vmstorage_protocol_version {
138 /* Major (MSW) and minor (LSW) version numbers. */
142 * Revision number is auto-incremented whenever this file is changed
143 * (See FILL_VMSTOR_REVISION macro above). Mismatch does not
144 * definitely indicate incompatibility--but it does indicate mismatched
146 * This is only used on the windows side. Just set it to 0.
151 /* Channel Property Flags */
152 #define STORAGE_CHANNEL_REMOVABLE_FLAG 0x1
153 #define STORAGE_CHANNEL_EMULATED_IDE_FLAG 0x2
155 struct vstor_packet {
156 /* Requested operation type */
157 enum vstor_packet_operation operation;
159 /* Flags - see below for values */
162 /* Status of the request returned from the server side. */
165 /* Data payload area */
168 * Structure used to forward SCSI commands from the
169 * client to the server.
171 struct vmscsi_request vm_srb;
173 /* Structure used to query channel properties. */
174 struct vmstorage_channel_properties storage_channel_properties;
176 /* Used during version negotiations. */
177 struct vmstorage_protocol_version version;
184 * This flag indicates that the server should send back a completion for this
188 #define REQUEST_COMPLETION_FLAG 0x1
190 /* Matches Windows-end */
191 enum storvsc_request_type {
198 * SRB status codes and masks; a subset of the codes used here.
201 #define SRB_STATUS_AUTOSENSE_VALID 0x80
202 #define SRB_STATUS_INVALID_LUN 0x20
203 #define SRB_STATUS_SUCCESS 0x01
204 #define SRB_STATUS_ERROR 0x04
207 * This is the end of Protocol specific defines.
212 * We setup a mempool to allocate request structures for this driver
213 * on a per-lun basis. The following define specifies the number of
214 * elements in the pool.
217 #define STORVSC_MIN_BUF_NR 64
218 static int storvsc_ringbuffer_size = (20 * PAGE_SIZE);
220 module_param(storvsc_ringbuffer_size, int, S_IRUGO);
221 MODULE_PARM_DESC(storvsc_ringbuffer_size, "Ring buffer size (bytes)");
223 #define STORVSC_MAX_IO_REQUESTS 128
226 * In Hyper-V, each port/path/target maps to 1 scsi host adapter. In
227 * reality, the path/target is not used (ie always set to 0) so our
228 * scsi host adapter essentially has 1 bus with 1 target that contains
231 #define STORVSC_MAX_LUNS_PER_TARGET 64
232 #define STORVSC_MAX_TARGETS 1
233 #define STORVSC_MAX_CHANNELS 1
237 struct storvsc_cmd_request {
238 struct list_head entry;
239 struct scsi_cmnd *cmd;
241 unsigned int bounce_sgl_count;
242 struct scatterlist *bounce_sgl;
244 struct hv_device *device;
246 /* Synchronize the request/response if needed */
247 struct completion wait_event;
249 unsigned char *sense_buffer;
250 struct hv_multipage_buffer data_buffer;
251 struct vstor_packet vstor_packet;
255 /* A storvsc device is a device object that contains a vmbus channel */
256 struct storvsc_device {
257 struct hv_device *device;
261 atomic_t num_outstanding_req;
262 struct Scsi_Host *host;
264 wait_queue_head_t waiting_to_drain;
267 * Each unique Port/Path/Target represents 1 channel ie scsi
268 * controller. In reality, the pathid, targetid is always 0
269 * and the port is set by us
271 unsigned int port_number;
272 unsigned char path_id;
273 unsigned char target_id;
275 /* Used for vsc/vsp channel reset process */
276 struct storvsc_cmd_request init_request;
277 struct storvsc_cmd_request reset_request;
280 struct stor_mem_pools {
281 struct kmem_cache *request_pool;
282 mempool_t *request_mempool;
285 struct hv_host_device {
286 struct hv_device *dev;
289 unsigned char target;
292 struct storvsc_scan_work {
293 struct work_struct work;
294 struct Scsi_Host *host;
298 static void storvsc_bus_scan(struct work_struct *work)
300 struct storvsc_scan_work *wrk;
303 wrk = container_of(work, struct storvsc_scan_work, work);
304 for (id = 0; id < wrk->host->max_id; ++id) {
305 if (wrk->host->reverse_ordering)
306 order_id = wrk->host->max_id - id - 1;
310 scsi_scan_target(&wrk->host->shost_gendev, 0,
311 order_id, SCAN_WILD_CARD, 1);
316 static void storvsc_remove_lun(struct work_struct *work)
318 struct storvsc_scan_work *wrk;
319 struct scsi_device *sdev;
321 wrk = container_of(work, struct storvsc_scan_work, work);
322 if (!scsi_host_get(wrk->host))
325 sdev = scsi_device_lookup(wrk->host, 0, 0, wrk->lun);
328 scsi_remove_device(sdev);
329 scsi_device_put(sdev);
331 scsi_host_put(wrk->host);
338 * Major/minor macros. Minor version is in LSB, meaning that earlier flat
339 * version numbers will be interpreted as "0.x" (i.e., 1 becomes 0.1).
342 static inline u16 storvsc_get_version(u8 major, u8 minor)
346 version = ((major << 8) | minor);
351 * We can get incoming messages from the host that are not in response to
352 * messages that we have sent out. An example of this would be messages
353 * received by the guest to notify dynamic addition/removal of LUNs. To
354 * deal with potential race conditions where the driver may be in the
355 * midst of being unloaded when we might receive an unsolicited message
356 * from the host, we have implemented a mechanism to gurantee sequential
359 * 1) Once the device is marked as being destroyed, we will fail all
361 * 2) We permit incoming messages when the device is being destroyed,
362 * only to properly account for messages already sent out.
365 static inline struct storvsc_device *get_out_stor_device(
366 struct hv_device *device)
368 struct storvsc_device *stor_device;
370 stor_device = hv_get_drvdata(device);
372 if (stor_device && stor_device->destroy)
379 static inline void storvsc_wait_to_drain(struct storvsc_device *dev)
381 dev->drain_notify = true;
382 wait_event(dev->waiting_to_drain,
383 atomic_read(&dev->num_outstanding_req) == 0);
384 dev->drain_notify = false;
387 static inline struct storvsc_device *get_in_stor_device(
388 struct hv_device *device)
390 struct storvsc_device *stor_device;
392 stor_device = hv_get_drvdata(device);
398 * If the device is being destroyed; allow incoming
399 * traffic only to cleanup outstanding requests.
402 if (stor_device->destroy &&
403 (atomic_read(&stor_device->num_outstanding_req) == 0))
411 static void destroy_bounce_buffer(struct scatterlist *sgl,
412 unsigned int sg_count)
415 struct page *page_buf;
417 for (i = 0; i < sg_count; i++) {
418 page_buf = sg_page((&sgl[i]));
419 if (page_buf != NULL)
420 __free_page(page_buf);
426 static int do_bounce_buffer(struct scatterlist *sgl, unsigned int sg_count)
430 /* No need to check */
434 /* We have at least 2 sg entries */
435 for (i = 0; i < sg_count; i++) {
437 /* make sure 1st one does not have hole */
438 if (sgl[i].offset + sgl[i].length != PAGE_SIZE)
440 } else if (i == sg_count - 1) {
441 /* make sure last one does not have hole */
442 if (sgl[i].offset != 0)
445 /* make sure no hole in the middle */
446 if (sgl[i].length != PAGE_SIZE || sgl[i].offset != 0)
453 static struct scatterlist *create_bounce_buffer(struct scatterlist *sgl,
454 unsigned int sg_count,
460 struct scatterlist *bounce_sgl;
461 struct page *page_buf;
462 unsigned int buf_len = ((write == WRITE_TYPE) ? 0 : PAGE_SIZE);
464 num_pages = ALIGN(len, PAGE_SIZE) >> PAGE_SHIFT;
466 bounce_sgl = kcalloc(num_pages, sizeof(struct scatterlist), GFP_ATOMIC);
470 for (i = 0; i < num_pages; i++) {
471 page_buf = alloc_page(GFP_ATOMIC);
474 sg_set_page(&bounce_sgl[i], page_buf, buf_len, 0);
480 destroy_bounce_buffer(bounce_sgl, num_pages);
484 /* Disgusting wrapper functions */
485 static inline unsigned long sg_kmap_atomic(struct scatterlist *sgl, int idx)
487 void *addr = kmap_atomic(sg_page(sgl + idx));
488 return (unsigned long)addr;
491 static inline void sg_kunmap_atomic(unsigned long addr)
493 kunmap_atomic((void *)addr);
497 /* Assume the original sgl has enough room */
498 static unsigned int copy_from_bounce_buffer(struct scatterlist *orig_sgl,
499 struct scatterlist *bounce_sgl,
500 unsigned int orig_sgl_count,
501 unsigned int bounce_sgl_count)
505 unsigned long src, dest;
506 unsigned int srclen, destlen, copylen;
507 unsigned int total_copied = 0;
508 unsigned long bounce_addr = 0;
509 unsigned long dest_addr = 0;
512 local_irq_save(flags);
514 for (i = 0; i < orig_sgl_count; i++) {
515 dest_addr = sg_kmap_atomic(orig_sgl,i) + orig_sgl[i].offset;
517 destlen = orig_sgl[i].length;
519 if (bounce_addr == 0)
520 bounce_addr = sg_kmap_atomic(bounce_sgl,j);
523 src = bounce_addr + bounce_sgl[j].offset;
524 srclen = bounce_sgl[j].length - bounce_sgl[j].offset;
526 copylen = min(srclen, destlen);
527 memcpy((void *)dest, (void *)src, copylen);
529 total_copied += copylen;
530 bounce_sgl[j].offset += copylen;
534 if (bounce_sgl[j].offset == bounce_sgl[j].length) {
536 sg_kunmap_atomic(bounce_addr);
540 * It is possible that the number of elements
541 * in the bounce buffer may not be equal to
542 * the number of elements in the original
543 * scatter list. Handle this correctly.
546 if (j == bounce_sgl_count) {
548 * We are done; cleanup and return.
550 sg_kunmap_atomic(dest_addr - orig_sgl[i].offset);
551 local_irq_restore(flags);
555 /* if we need to use another bounce buffer */
556 if (destlen || i != orig_sgl_count - 1)
557 bounce_addr = sg_kmap_atomic(bounce_sgl,j);
558 } else if (destlen == 0 && i == orig_sgl_count - 1) {
559 /* unmap the last bounce that is < PAGE_SIZE */
560 sg_kunmap_atomic(bounce_addr);
564 sg_kunmap_atomic(dest_addr - orig_sgl[i].offset);
567 local_irq_restore(flags);
572 /* Assume the bounce_sgl has enough room ie using the create_bounce_buffer() */
573 static unsigned int copy_to_bounce_buffer(struct scatterlist *orig_sgl,
574 struct scatterlist *bounce_sgl,
575 unsigned int orig_sgl_count)
579 unsigned long src, dest;
580 unsigned int srclen, destlen, copylen;
581 unsigned int total_copied = 0;
582 unsigned long bounce_addr = 0;
583 unsigned long src_addr = 0;
586 local_irq_save(flags);
588 for (i = 0; i < orig_sgl_count; i++) {
589 src_addr = sg_kmap_atomic(orig_sgl,i) + orig_sgl[i].offset;
591 srclen = orig_sgl[i].length;
593 if (bounce_addr == 0)
594 bounce_addr = sg_kmap_atomic(bounce_sgl,j);
597 /* assume bounce offset always == 0 */
598 dest = bounce_addr + bounce_sgl[j].length;
599 destlen = PAGE_SIZE - bounce_sgl[j].length;
601 copylen = min(srclen, destlen);
602 memcpy((void *)dest, (void *)src, copylen);
604 total_copied += copylen;
605 bounce_sgl[j].length += copylen;
609 if (bounce_sgl[j].length == PAGE_SIZE) {
610 /* full..move to next entry */
611 sg_kunmap_atomic(bounce_addr);
614 /* if we need to use another bounce buffer */
615 if (srclen || i != orig_sgl_count - 1)
616 bounce_addr = sg_kmap_atomic(bounce_sgl,j);
618 } else if (srclen == 0 && i == orig_sgl_count - 1) {
619 /* unmap the last bounce that is < PAGE_SIZE */
620 sg_kunmap_atomic(bounce_addr);
624 sg_kunmap_atomic(src_addr - orig_sgl[i].offset);
627 local_irq_restore(flags);
632 static int storvsc_channel_init(struct hv_device *device)
634 struct storvsc_device *stor_device;
635 struct storvsc_cmd_request *request;
636 struct vstor_packet *vstor_packet;
639 stor_device = get_out_stor_device(device);
643 request = &stor_device->init_request;
644 vstor_packet = &request->vstor_packet;
647 * Now, initiate the vsc/vsp initialization protocol on the open
650 memset(request, 0, sizeof(struct storvsc_cmd_request));
651 init_completion(&request->wait_event);
652 vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION;
653 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
655 ret = vmbus_sendpacket(device->channel, vstor_packet,
656 sizeof(struct vstor_packet),
657 (unsigned long)request,
659 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
663 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
669 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
670 vstor_packet->status != 0)
674 /* reuse the packet for version range supported */
675 memset(vstor_packet, 0, sizeof(struct vstor_packet));
676 vstor_packet->operation = VSTOR_OPERATION_QUERY_PROTOCOL_VERSION;
677 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
679 vstor_packet->version.major_minor =
680 storvsc_get_version(VMSTOR_CURRENT_MAJOR, VMSTOR_CURRENT_MINOR);
683 * The revision number is only used in Windows; set it to 0.
685 vstor_packet->version.revision = 0;
687 ret = vmbus_sendpacket(device->channel, vstor_packet,
688 sizeof(struct vstor_packet),
689 (unsigned long)request,
691 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
695 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
701 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
702 vstor_packet->status != 0)
706 memset(vstor_packet, 0, sizeof(struct vstor_packet));
707 vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES;
708 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
709 vstor_packet->storage_channel_properties.port_number =
710 stor_device->port_number;
712 ret = vmbus_sendpacket(device->channel, vstor_packet,
713 sizeof(struct vstor_packet),
714 (unsigned long)request,
716 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
721 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
727 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
728 vstor_packet->status != 0)
731 stor_device->path_id = vstor_packet->storage_channel_properties.path_id;
732 stor_device->target_id
733 = vstor_packet->storage_channel_properties.target_id;
735 memset(vstor_packet, 0, sizeof(struct vstor_packet));
736 vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION;
737 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
739 ret = vmbus_sendpacket(device->channel, vstor_packet,
740 sizeof(struct vstor_packet),
741 (unsigned long)request,
743 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
748 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
754 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
755 vstor_packet->status != 0)
764 static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request)
766 struct scsi_cmnd *scmnd = cmd_request->cmd;
767 struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
768 void (*scsi_done_fn)(struct scsi_cmnd *);
769 struct scsi_sense_hdr sense_hdr;
770 struct vmscsi_request *vm_srb;
771 struct storvsc_scan_work *wrk;
772 struct stor_mem_pools *memp = scmnd->device->hostdata;
774 vm_srb = &cmd_request->vstor_packet.vm_srb;
775 if (cmd_request->bounce_sgl_count) {
776 if (vm_srb->data_in == READ_TYPE)
777 copy_from_bounce_buffer(scsi_sglist(scmnd),
778 cmd_request->bounce_sgl,
779 scsi_sg_count(scmnd),
780 cmd_request->bounce_sgl_count);
781 destroy_bounce_buffer(cmd_request->bounce_sgl,
782 cmd_request->bounce_sgl_count);
786 * If there is an error; offline the device since all
787 * error recovery strategies would have already been
788 * deployed on the host side.
790 if (vm_srb->srb_status == SRB_STATUS_ERROR)
791 scmnd->result = DID_TARGET_FAILURE << 16;
793 scmnd->result = vm_srb->scsi_status;
796 * If the LUN is invalid; remove the device.
798 if (vm_srb->srb_status == SRB_STATUS_INVALID_LUN) {
799 struct storvsc_device *stor_dev;
800 struct hv_device *dev = host_dev->dev;
801 struct Scsi_Host *host;
803 stor_dev = get_in_stor_device(dev);
804 host = stor_dev->host;
806 wrk = kmalloc(sizeof(struct storvsc_scan_work),
809 scmnd->result = DID_TARGET_FAILURE << 16;
812 wrk->lun = vm_srb->lun;
813 INIT_WORK(&wrk->work, storvsc_remove_lun);
814 schedule_work(&wrk->work);
819 if (scsi_normalize_sense(scmnd->sense_buffer,
820 SCSI_SENSE_BUFFERSIZE, &sense_hdr))
821 scsi_print_sense_hdr("storvsc", &sense_hdr);
824 scsi_set_resid(scmnd,
825 cmd_request->data_buffer.len -
826 vm_srb->data_transfer_length);
828 scsi_done_fn = scmnd->scsi_done;
830 scmnd->host_scribble = NULL;
831 scmnd->scsi_done = NULL;
835 mempool_free(cmd_request, memp->request_mempool);
838 static void storvsc_on_io_completion(struct hv_device *device,
839 struct vstor_packet *vstor_packet,
840 struct storvsc_cmd_request *request)
842 struct storvsc_device *stor_device;
843 struct vstor_packet *stor_pkt;
845 stor_device = hv_get_drvdata(device);
846 stor_pkt = &request->vstor_packet;
849 * The current SCSI handling on the host side does
850 * not correctly handle:
851 * INQUIRY command with page code parameter set to 0x80
852 * MODE_SENSE command with cmd[2] == 0x1c
854 * Setup srb and scsi status so this won't be fatal.
855 * We do this so we can distinguish truly fatal failues
856 * (srb status == 0x4) and off-line the device in that case.
859 if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) ||
860 (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) {
861 vstor_packet->vm_srb.scsi_status = 0;
862 vstor_packet->vm_srb.srb_status = SRB_STATUS_SUCCESS;
866 /* Copy over the status...etc */
867 stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status;
868 stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status;
869 stor_pkt->vm_srb.sense_info_length =
870 vstor_packet->vm_srb.sense_info_length;
872 if (vstor_packet->vm_srb.scsi_status != 0 ||
873 vstor_packet->vm_srb.srb_status != SRB_STATUS_SUCCESS){
874 dev_warn(&device->device,
875 "cmd 0x%x scsi status 0x%x srb status 0x%x\n",
876 stor_pkt->vm_srb.cdb[0],
877 vstor_packet->vm_srb.scsi_status,
878 vstor_packet->vm_srb.srb_status);
881 if ((vstor_packet->vm_srb.scsi_status & 0xFF) == 0x02) {
882 /* CHECK_CONDITION */
883 if (vstor_packet->vm_srb.srb_status &
884 SRB_STATUS_AUTOSENSE_VALID) {
885 /* autosense data available */
886 dev_warn(&device->device,
887 "stor pkt %p autosense data valid - len %d\n",
889 vstor_packet->vm_srb.sense_info_length);
891 memcpy(request->sense_buffer,
892 vstor_packet->vm_srb.sense_data,
893 vstor_packet->vm_srb.sense_info_length);
898 stor_pkt->vm_srb.data_transfer_length =
899 vstor_packet->vm_srb.data_transfer_length;
901 storvsc_command_completion(request);
903 if (atomic_dec_and_test(&stor_device->num_outstanding_req) &&
904 stor_device->drain_notify)
905 wake_up(&stor_device->waiting_to_drain);
910 static void storvsc_on_receive(struct hv_device *device,
911 struct vstor_packet *vstor_packet,
912 struct storvsc_cmd_request *request)
914 struct storvsc_scan_work *work;
915 struct storvsc_device *stor_device;
917 switch (vstor_packet->operation) {
918 case VSTOR_OPERATION_COMPLETE_IO:
919 storvsc_on_io_completion(device, vstor_packet, request);
922 case VSTOR_OPERATION_REMOVE_DEVICE:
923 case VSTOR_OPERATION_ENUMERATE_BUS:
924 stor_device = get_in_stor_device(device);
925 work = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC);
929 INIT_WORK(&work->work, storvsc_bus_scan);
930 work->host = stor_device->host;
931 schedule_work(&work->work);
939 static void storvsc_on_channel_callback(void *context)
941 struct hv_device *device = (struct hv_device *)context;
942 struct storvsc_device *stor_device;
945 unsigned char packet[ALIGN(sizeof(struct vstor_packet), 8)];
946 struct storvsc_cmd_request *request;
950 stor_device = get_in_stor_device(device);
955 ret = vmbus_recvpacket(device->channel, packet,
956 ALIGN(sizeof(struct vstor_packet), 8),
957 &bytes_recvd, &request_id);
958 if (ret == 0 && bytes_recvd > 0) {
960 request = (struct storvsc_cmd_request *)
961 (unsigned long)request_id;
963 if ((request == &stor_device->init_request) ||
964 (request == &stor_device->reset_request)) {
966 memcpy(&request->vstor_packet, packet,
967 sizeof(struct vstor_packet));
968 complete(&request->wait_event);
970 storvsc_on_receive(device,
971 (struct vstor_packet *)packet,
982 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size)
984 struct vmstorage_channel_properties props;
987 memset(&props, 0, sizeof(struct vmstorage_channel_properties));
989 ret = vmbus_open(device->channel,
993 sizeof(struct vmstorage_channel_properties),
994 storvsc_on_channel_callback, device);
999 ret = storvsc_channel_init(device);
1004 static int storvsc_dev_remove(struct hv_device *device)
1006 struct storvsc_device *stor_device;
1007 unsigned long flags;
1009 stor_device = hv_get_drvdata(device);
1011 spin_lock_irqsave(&device->channel->inbound_lock, flags);
1012 stor_device->destroy = true;
1013 spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
1016 * At this point, all outbound traffic should be disable. We
1017 * only allow inbound traffic (responses) to proceed so that
1018 * outstanding requests can be completed.
1021 storvsc_wait_to_drain(stor_device);
1024 * Since we have already drained, we don't need to busy wait
1025 * as was done in final_release_stor_device()
1026 * Note that we cannot set the ext pointer to NULL until
1027 * we have drained - to drain the outgoing packets, we need to
1028 * allow incoming packets.
1030 spin_lock_irqsave(&device->channel->inbound_lock, flags);
1031 hv_set_drvdata(device, NULL);
1032 spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
1034 /* Close the channel */
1035 vmbus_close(device->channel);
1041 static int storvsc_do_io(struct hv_device *device,
1042 struct storvsc_cmd_request *request)
1044 struct storvsc_device *stor_device;
1045 struct vstor_packet *vstor_packet;
1048 vstor_packet = &request->vstor_packet;
1049 stor_device = get_out_stor_device(device);
1055 request->device = device;
1058 vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
1060 vstor_packet->vm_srb.length = sizeof(struct vmscsi_request);
1063 vstor_packet->vm_srb.sense_info_length = STORVSC_SENSE_BUFFER_SIZE;
1066 vstor_packet->vm_srb.data_transfer_length =
1067 request->data_buffer.len;
1069 vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
1071 if (request->data_buffer.len) {
1072 ret = vmbus_sendpacket_multipagebuffer(device->channel,
1073 &request->data_buffer,
1075 sizeof(struct vstor_packet),
1076 (unsigned long)request);
1078 ret = vmbus_sendpacket(device->channel, vstor_packet,
1079 sizeof(struct vstor_packet),
1080 (unsigned long)request,
1082 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1088 atomic_inc(&stor_device->num_outstanding_req);
1093 static int storvsc_device_alloc(struct scsi_device *sdevice)
1095 struct stor_mem_pools *memp;
1096 int number = STORVSC_MIN_BUF_NR;
1098 memp = kzalloc(sizeof(struct stor_mem_pools), GFP_KERNEL);
1102 memp->request_pool =
1103 kmem_cache_create(dev_name(&sdevice->sdev_dev),
1104 sizeof(struct storvsc_cmd_request), 0,
1105 SLAB_HWCACHE_ALIGN, NULL);
1107 if (!memp->request_pool)
1110 memp->request_mempool = mempool_create(number, mempool_alloc_slab,
1112 memp->request_pool);
1114 if (!memp->request_mempool)
1117 sdevice->hostdata = memp;
1122 kmem_cache_destroy(memp->request_pool);
1129 static void storvsc_device_destroy(struct scsi_device *sdevice)
1131 struct stor_mem_pools *memp = sdevice->hostdata;
1133 mempool_destroy(memp->request_mempool);
1134 kmem_cache_destroy(memp->request_pool);
1136 sdevice->hostdata = NULL;
1139 static int storvsc_device_configure(struct scsi_device *sdevice)
1141 scsi_adjust_queue_depth(sdevice, MSG_SIMPLE_TAG,
1142 STORVSC_MAX_IO_REQUESTS);
1144 blk_queue_max_segment_size(sdevice->request_queue, PAGE_SIZE);
1146 blk_queue_bounce_limit(sdevice->request_queue, BLK_BOUNCE_ANY);
1151 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev,
1152 sector_t capacity, int *info)
1154 sector_t nsect = capacity;
1155 sector_t cylinders = nsect;
1156 int heads, sectors_pt;
1159 * We are making up these values; let us keep it simple.
1162 sectors_pt = 0x3f; /* Sectors per track */
1163 sector_div(cylinders, heads * sectors_pt);
1164 if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
1168 info[1] = sectors_pt;
1169 info[2] = (int)cylinders;
1174 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
1176 struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
1177 struct hv_device *device = host_dev->dev;
1179 struct storvsc_device *stor_device;
1180 struct storvsc_cmd_request *request;
1181 struct vstor_packet *vstor_packet;
1185 stor_device = get_out_stor_device(device);
1189 request = &stor_device->reset_request;
1190 vstor_packet = &request->vstor_packet;
1192 init_completion(&request->wait_event);
1194 vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
1195 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
1196 vstor_packet->vm_srb.path_id = stor_device->path_id;
1198 ret = vmbus_sendpacket(device->channel, vstor_packet,
1199 sizeof(struct vstor_packet),
1200 (unsigned long)&stor_device->reset_request,
1202 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1206 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
1208 return TIMEOUT_ERROR;
1212 * At this point, all outstanding requests in the adapter
1213 * should have been flushed out and return to us
1219 static bool storvsc_scsi_cmd_ok(struct scsi_cmnd *scmnd)
1221 bool allowed = true;
1222 u8 scsi_op = scmnd->cmnd[0];
1226 * smartd sends this command and the host does not handle
1227 * this. So, don't send it.
1230 scmnd->result = ILLEGAL_REQUEST << 16;
1239 static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
1242 struct hv_host_device *host_dev = shost_priv(host);
1243 struct hv_device *dev = host_dev->dev;
1244 struct storvsc_cmd_request *cmd_request;
1245 unsigned int request_size = 0;
1247 struct scatterlist *sgl;
1248 unsigned int sg_count = 0;
1249 struct vmscsi_request *vm_srb;
1250 struct stor_mem_pools *memp = scmnd->device->hostdata;
1252 if (!storvsc_scsi_cmd_ok(scmnd)) {
1253 scmnd->scsi_done(scmnd);
1257 request_size = sizeof(struct storvsc_cmd_request);
1259 cmd_request = mempool_alloc(memp->request_mempool,
1263 * We might be invoked in an interrupt context; hence
1264 * mempool_alloc() can fail.
1267 return SCSI_MLQUEUE_DEVICE_BUSY;
1269 memset(cmd_request, 0, sizeof(struct storvsc_cmd_request));
1271 /* Setup the cmd request */
1272 cmd_request->cmd = scmnd;
1274 scmnd->host_scribble = (unsigned char *)cmd_request;
1276 vm_srb = &cmd_request->vstor_packet.vm_srb;
1280 switch (scmnd->sc_data_direction) {
1282 vm_srb->data_in = WRITE_TYPE;
1284 case DMA_FROM_DEVICE:
1285 vm_srb->data_in = READ_TYPE;
1288 vm_srb->data_in = UNKNOWN_TYPE;
1293 vm_srb->port_number = host_dev->port;
1294 vm_srb->path_id = scmnd->device->channel;
1295 vm_srb->target_id = scmnd->device->id;
1296 vm_srb->lun = scmnd->device->lun;
1298 vm_srb->cdb_length = scmnd->cmd_len;
1300 memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
1302 cmd_request->sense_buffer = scmnd->sense_buffer;
1305 cmd_request->data_buffer.len = scsi_bufflen(scmnd);
1306 if (scsi_sg_count(scmnd)) {
1307 sgl = (struct scatterlist *)scsi_sglist(scmnd);
1308 sg_count = scsi_sg_count(scmnd);
1310 /* check if we need to bounce the sgl */
1311 if (do_bounce_buffer(sgl, scsi_sg_count(scmnd)) != -1) {
1312 cmd_request->bounce_sgl =
1313 create_bounce_buffer(sgl, scsi_sg_count(scmnd),
1314 scsi_bufflen(scmnd),
1316 if (!cmd_request->bounce_sgl) {
1317 ret = SCSI_MLQUEUE_HOST_BUSY;
1321 cmd_request->bounce_sgl_count =
1322 ALIGN(scsi_bufflen(scmnd), PAGE_SIZE) >>
1325 if (vm_srb->data_in == WRITE_TYPE)
1326 copy_to_bounce_buffer(sgl,
1327 cmd_request->bounce_sgl,
1328 scsi_sg_count(scmnd));
1330 sgl = cmd_request->bounce_sgl;
1331 sg_count = cmd_request->bounce_sgl_count;
1334 cmd_request->data_buffer.offset = sgl[0].offset;
1336 for (i = 0; i < sg_count; i++)
1337 cmd_request->data_buffer.pfn_array[i] =
1338 page_to_pfn(sg_page((&sgl[i])));
1340 } else if (scsi_sglist(scmnd)) {
1341 cmd_request->data_buffer.offset =
1342 virt_to_phys(scsi_sglist(scmnd)) & (PAGE_SIZE-1);
1343 cmd_request->data_buffer.pfn_array[0] =
1344 virt_to_phys(scsi_sglist(scmnd)) >> PAGE_SHIFT;
1347 /* Invokes the vsc to start an IO */
1348 ret = storvsc_do_io(dev, cmd_request);
1350 if (ret == -EAGAIN) {
1353 if (cmd_request->bounce_sgl_count) {
1354 destroy_bounce_buffer(cmd_request->bounce_sgl,
1355 cmd_request->bounce_sgl_count);
1357 ret = SCSI_MLQUEUE_DEVICE_BUSY;
1365 mempool_free(cmd_request, memp->request_mempool);
1366 scmnd->host_scribble = NULL;
1370 static struct scsi_host_template scsi_driver = {
1371 .module = THIS_MODULE,
1372 .name = "storvsc_host_t",
1373 .bios_param = storvsc_get_chs,
1374 .queuecommand = storvsc_queuecommand,
1375 .eh_host_reset_handler = storvsc_host_reset_handler,
1376 .slave_alloc = storvsc_device_alloc,
1377 .slave_destroy = storvsc_device_destroy,
1378 .slave_configure = storvsc_device_configure,
1380 /* 64 max_queue * 1 target */
1381 .can_queue = STORVSC_MAX_IO_REQUESTS*STORVSC_MAX_TARGETS,
1383 /* no use setting to 0 since ll_blk_rw reset it to 1 */
1385 .sg_tablesize = MAX_MULTIPAGE_BUFFER_COUNT,
1386 .use_clustering = DISABLE_CLUSTERING,
1387 /* Make sure we dont get a sg segment crosses a page boundary */
1388 .dma_boundary = PAGE_SIZE-1,
1396 static const struct hv_vmbus_device_id id_table[] = {
1398 { VMBUS_DEVICE(0xd9, 0x63, 0x61, 0xba, 0xa1, 0x04, 0x29, 0x4d,
1399 0xb6, 0x05, 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f)
1400 .driver_data = SCSI_GUID },
1402 { VMBUS_DEVICE(0x32, 0x26, 0x41, 0x32, 0xcb, 0x86, 0xa2, 0x44,
1403 0x9b, 0x5c, 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5)
1404 .driver_data = IDE_GUID },
1408 MODULE_DEVICE_TABLE(vmbus, id_table);
1410 static int storvsc_probe(struct hv_device *device,
1411 const struct hv_vmbus_device_id *dev_id)
1414 struct Scsi_Host *host;
1415 struct hv_host_device *host_dev;
1416 bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false);
1418 struct storvsc_device *stor_device;
1420 host = scsi_host_alloc(&scsi_driver,
1421 sizeof(struct hv_host_device));
1425 host_dev = shost_priv(host);
1426 memset(host_dev, 0, sizeof(struct hv_host_device));
1428 host_dev->port = host->host_no;
1429 host_dev->dev = device;
1432 stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
1438 stor_device->destroy = false;
1439 init_waitqueue_head(&stor_device->waiting_to_drain);
1440 stor_device->device = device;
1441 stor_device->host = host;
1442 hv_set_drvdata(device, stor_device);
1444 stor_device->port_number = host->host_no;
1445 ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size);
1449 host_dev->path = stor_device->path_id;
1450 host_dev->target = stor_device->target_id;
1452 /* max # of devices per target */
1453 host->max_lun = STORVSC_MAX_LUNS_PER_TARGET;
1454 /* max # of targets per channel */
1455 host->max_id = STORVSC_MAX_TARGETS;
1456 /* max # of channels */
1457 host->max_channel = STORVSC_MAX_CHANNELS - 1;
1458 /* max cmd length */
1459 host->max_cmd_len = STORVSC_MAX_CMD_LEN;
1461 /* Register the HBA and start the scsi bus scan */
1462 ret = scsi_add_host(host, &device->device);
1467 scsi_scan_host(host);
1469 target = (device->dev_instance.b[5] << 8 |
1470 device->dev_instance.b[4]);
1471 ret = scsi_add_device(host, 0, target, 0);
1473 scsi_remove_host(host);
1481 * Once we have connected with the host, we would need to
1482 * to invoke storvsc_dev_remove() to rollback this state and
1483 * this call also frees up the stor_device; hence the jump around
1486 storvsc_dev_remove(device);
1493 scsi_host_put(host);
1497 static int storvsc_remove(struct hv_device *dev)
1499 struct storvsc_device *stor_device = hv_get_drvdata(dev);
1500 struct Scsi_Host *host = stor_device->host;
1502 scsi_remove_host(host);
1503 storvsc_dev_remove(dev);
1504 scsi_host_put(host);
1509 static struct hv_driver storvsc_drv = {
1510 .name = KBUILD_MODNAME,
1511 .id_table = id_table,
1512 .probe = storvsc_probe,
1513 .remove = storvsc_remove,
1516 static int __init storvsc_drv_init(void)
1518 u32 max_outstanding_req_per_channel;
1521 * Divide the ring buffer data size (which is 1 page less
1522 * than the ring buffer size since that page is reserved for
1523 * the ring buffer indices) by the max request size (which is
1524 * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
1526 max_outstanding_req_per_channel =
1527 ((storvsc_ringbuffer_size - PAGE_SIZE) /
1528 ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
1529 sizeof(struct vstor_packet) + sizeof(u64),
1532 if (max_outstanding_req_per_channel <
1533 STORVSC_MAX_IO_REQUESTS)
1536 return vmbus_driver_register(&storvsc_drv);
1539 static void __exit storvsc_drv_exit(void)
1541 vmbus_driver_unregister(&storvsc_drv);
1544 MODULE_LICENSE("GPL");
1545 MODULE_VERSION(HV_DRV_VERSION);
1546 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
1547 module_init(storvsc_drv_init);
1548 module_exit(storvsc_drv_exit);