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 /* Assume the original sgl has enough room */
485 static unsigned int copy_from_bounce_buffer(struct scatterlist *orig_sgl,
486 struct scatterlist *bounce_sgl,
487 unsigned int orig_sgl_count,
488 unsigned int bounce_sgl_count)
492 unsigned long src, dest;
493 unsigned int srclen, destlen, copylen;
494 unsigned int total_copied = 0;
495 unsigned long bounce_addr = 0;
496 unsigned long dest_addr = 0;
499 local_irq_save(flags);
501 for (i = 0; i < orig_sgl_count; i++) {
502 dest_addr = (unsigned long)kmap_atomic(sg_page((&orig_sgl[i])),
503 KM_IRQ0) + orig_sgl[i].offset;
505 destlen = orig_sgl[i].length;
507 if (bounce_addr == 0)
509 (unsigned long)kmap_atomic(sg_page((&bounce_sgl[j])),
513 src = bounce_addr + bounce_sgl[j].offset;
514 srclen = bounce_sgl[j].length - bounce_sgl[j].offset;
516 copylen = min(srclen, destlen);
517 memcpy((void *)dest, (void *)src, copylen);
519 total_copied += copylen;
520 bounce_sgl[j].offset += copylen;
524 if (bounce_sgl[j].offset == bounce_sgl[j].length) {
526 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
530 * It is possible that the number of elements
531 * in the bounce buffer may not be equal to
532 * the number of elements in the original
533 * scatter list. Handle this correctly.
536 if (j == bounce_sgl_count) {
538 * We are done; cleanup and return.
540 kunmap_atomic((void *)(dest_addr -
543 local_irq_restore(flags);
547 /* if we need to use another bounce buffer */
548 if (destlen || i != orig_sgl_count - 1)
550 (unsigned long)kmap_atomic(
551 sg_page((&bounce_sgl[j])), KM_IRQ0);
552 } else if (destlen == 0 && i == orig_sgl_count - 1) {
553 /* unmap the last bounce that is < PAGE_SIZE */
554 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
558 kunmap_atomic((void *)(dest_addr - orig_sgl[i].offset),
562 local_irq_restore(flags);
567 /* Assume the bounce_sgl has enough room ie using the create_bounce_buffer() */
568 static unsigned int copy_to_bounce_buffer(struct scatterlist *orig_sgl,
569 struct scatterlist *bounce_sgl,
570 unsigned int orig_sgl_count)
574 unsigned long src, dest;
575 unsigned int srclen, destlen, copylen;
576 unsigned int total_copied = 0;
577 unsigned long bounce_addr = 0;
578 unsigned long src_addr = 0;
581 local_irq_save(flags);
583 for (i = 0; i < orig_sgl_count; i++) {
584 src_addr = (unsigned long)kmap_atomic(sg_page((&orig_sgl[i])),
585 KM_IRQ0) + orig_sgl[i].offset;
587 srclen = orig_sgl[i].length;
589 if (bounce_addr == 0)
591 (unsigned long)kmap_atomic(sg_page((&bounce_sgl[j])),
595 /* assume bounce offset always == 0 */
596 dest = bounce_addr + bounce_sgl[j].length;
597 destlen = PAGE_SIZE - bounce_sgl[j].length;
599 copylen = min(srclen, destlen);
600 memcpy((void *)dest, (void *)src, copylen);
602 total_copied += copylen;
603 bounce_sgl[j].length += copylen;
607 if (bounce_sgl[j].length == PAGE_SIZE) {
608 /* full..move to next entry */
609 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
612 /* if we need to use another bounce buffer */
613 if (srclen || i != orig_sgl_count - 1)
615 (unsigned long)kmap_atomic(
616 sg_page((&bounce_sgl[j])), KM_IRQ0);
618 } else if (srclen == 0 && i == orig_sgl_count - 1) {
619 /* unmap the last bounce that is < PAGE_SIZE */
620 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
624 kunmap_atomic((void *)(src_addr - orig_sgl[i].offset), KM_IRQ0);
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. However, if the command
789 * were a pass-through command deal with it appropriately.
791 switch (vm_srb->srb_status) {
792 case SRB_STATUS_ERROR:
793 switch (scmnd->cmnd[0]) {
796 scmnd->result = DID_PASSTHROUGH << 16;
799 scmnd->result = DID_TARGET_FAILURE << 16;
803 scmnd->result = vm_srb->scsi_status;
808 * If the LUN is invalid; remove the device.
810 if (vm_srb->srb_status == SRB_STATUS_INVALID_LUN) {
811 struct storvsc_device *stor_dev;
812 struct hv_device *dev = host_dev->dev;
813 struct Scsi_Host *host;
815 stor_dev = get_in_stor_device(dev);
816 host = stor_dev->host;
818 wrk = kmalloc(sizeof(struct storvsc_scan_work),
821 scmnd->result = DID_TARGET_FAILURE << 16;
824 wrk->lun = vm_srb->lun;
825 INIT_WORK(&wrk->work, storvsc_remove_lun);
826 schedule_work(&wrk->work);
831 if (scsi_normalize_sense(scmnd->sense_buffer,
832 SCSI_SENSE_BUFFERSIZE, &sense_hdr))
833 scsi_print_sense_hdr("storvsc", &sense_hdr);
836 scsi_set_resid(scmnd,
837 cmd_request->data_buffer.len -
838 vm_srb->data_transfer_length);
840 scsi_done_fn = scmnd->scsi_done;
842 scmnd->host_scribble = NULL;
843 scmnd->scsi_done = NULL;
847 mempool_free(cmd_request, memp->request_mempool);
850 static void storvsc_on_io_completion(struct hv_device *device,
851 struct vstor_packet *vstor_packet,
852 struct storvsc_cmd_request *request)
854 struct storvsc_device *stor_device;
855 struct vstor_packet *stor_pkt;
857 stor_device = hv_get_drvdata(device);
858 stor_pkt = &request->vstor_packet;
861 * The current SCSI handling on the host side does
862 * not correctly handle:
863 * INQUIRY command with page code parameter set to 0x80
864 * MODE_SENSE command with cmd[2] == 0x1c
866 * Setup srb and scsi status so this won't be fatal.
867 * We do this so we can distinguish truly fatal failues
868 * (srb status == 0x4) and off-line the device in that case.
871 if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) ||
872 (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) {
873 vstor_packet->vm_srb.scsi_status = 0;
874 vstor_packet->vm_srb.srb_status = SRB_STATUS_SUCCESS;
878 /* Copy over the status...etc */
879 stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status;
880 stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status;
881 stor_pkt->vm_srb.sense_info_length =
882 vstor_packet->vm_srb.sense_info_length;
884 if (vstor_packet->vm_srb.scsi_status != 0 ||
885 vstor_packet->vm_srb.srb_status != SRB_STATUS_SUCCESS){
886 dev_warn(&device->device,
887 "cmd 0x%x scsi status 0x%x srb status 0x%x\n",
888 stor_pkt->vm_srb.cdb[0],
889 vstor_packet->vm_srb.scsi_status,
890 vstor_packet->vm_srb.srb_status);
893 if ((vstor_packet->vm_srb.scsi_status & 0xFF) == 0x02) {
894 /* CHECK_CONDITION */
895 if (vstor_packet->vm_srb.srb_status &
896 SRB_STATUS_AUTOSENSE_VALID) {
897 /* autosense data available */
898 dev_warn(&device->device,
899 "stor pkt %p autosense data valid - len %d\n",
901 vstor_packet->vm_srb.sense_info_length);
903 memcpy(request->sense_buffer,
904 vstor_packet->vm_srb.sense_data,
905 vstor_packet->vm_srb.sense_info_length);
910 stor_pkt->vm_srb.data_transfer_length =
911 vstor_packet->vm_srb.data_transfer_length;
913 storvsc_command_completion(request);
915 if (atomic_dec_and_test(&stor_device->num_outstanding_req) &&
916 stor_device->drain_notify)
917 wake_up(&stor_device->waiting_to_drain);
922 static void storvsc_on_receive(struct hv_device *device,
923 struct vstor_packet *vstor_packet,
924 struct storvsc_cmd_request *request)
926 struct storvsc_scan_work *work;
927 struct storvsc_device *stor_device;
929 switch (vstor_packet->operation) {
930 case VSTOR_OPERATION_COMPLETE_IO:
931 storvsc_on_io_completion(device, vstor_packet, request);
934 case VSTOR_OPERATION_REMOVE_DEVICE:
935 case VSTOR_OPERATION_ENUMERATE_BUS:
936 stor_device = get_in_stor_device(device);
937 work = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC);
941 INIT_WORK(&work->work, storvsc_bus_scan);
942 work->host = stor_device->host;
943 schedule_work(&work->work);
951 static void storvsc_on_channel_callback(void *context)
953 struct hv_device *device = (struct hv_device *)context;
954 struct storvsc_device *stor_device;
957 unsigned char packet[ALIGN(sizeof(struct vstor_packet), 8)];
958 struct storvsc_cmd_request *request;
962 stor_device = get_in_stor_device(device);
967 ret = vmbus_recvpacket(device->channel, packet,
968 ALIGN(sizeof(struct vstor_packet), 8),
969 &bytes_recvd, &request_id);
970 if (ret == 0 && bytes_recvd > 0) {
972 request = (struct storvsc_cmd_request *)
973 (unsigned long)request_id;
975 if ((request == &stor_device->init_request) ||
976 (request == &stor_device->reset_request)) {
978 memcpy(&request->vstor_packet, packet,
979 sizeof(struct vstor_packet));
980 complete(&request->wait_event);
982 storvsc_on_receive(device,
983 (struct vstor_packet *)packet,
994 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size)
996 struct vmstorage_channel_properties props;
999 memset(&props, 0, sizeof(struct vmstorage_channel_properties));
1001 ret = vmbus_open(device->channel,
1005 sizeof(struct vmstorage_channel_properties),
1006 storvsc_on_channel_callback, device);
1011 ret = storvsc_channel_init(device);
1016 static int storvsc_dev_remove(struct hv_device *device)
1018 struct storvsc_device *stor_device;
1019 unsigned long flags;
1021 stor_device = hv_get_drvdata(device);
1023 spin_lock_irqsave(&device->channel->inbound_lock, flags);
1024 stor_device->destroy = true;
1025 spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
1028 * At this point, all outbound traffic should be disable. We
1029 * only allow inbound traffic (responses) to proceed so that
1030 * outstanding requests can be completed.
1033 storvsc_wait_to_drain(stor_device);
1036 * Since we have already drained, we don't need to busy wait
1037 * as was done in final_release_stor_device()
1038 * Note that we cannot set the ext pointer to NULL until
1039 * we have drained - to drain the outgoing packets, we need to
1040 * allow incoming packets.
1042 spin_lock_irqsave(&device->channel->inbound_lock, flags);
1043 hv_set_drvdata(device, NULL);
1044 spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
1046 /* Close the channel */
1047 vmbus_close(device->channel);
1053 static int storvsc_do_io(struct hv_device *device,
1054 struct storvsc_cmd_request *request)
1056 struct storvsc_device *stor_device;
1057 struct vstor_packet *vstor_packet;
1060 vstor_packet = &request->vstor_packet;
1061 stor_device = get_out_stor_device(device);
1067 request->device = device;
1070 vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
1072 vstor_packet->vm_srb.length = sizeof(struct vmscsi_request);
1075 vstor_packet->vm_srb.sense_info_length = STORVSC_SENSE_BUFFER_SIZE;
1078 vstor_packet->vm_srb.data_transfer_length =
1079 request->data_buffer.len;
1081 vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
1083 if (request->data_buffer.len) {
1084 ret = vmbus_sendpacket_multipagebuffer(device->channel,
1085 &request->data_buffer,
1087 sizeof(struct vstor_packet),
1088 (unsigned long)request);
1090 ret = vmbus_sendpacket(device->channel, vstor_packet,
1091 sizeof(struct vstor_packet),
1092 (unsigned long)request,
1094 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1100 atomic_inc(&stor_device->num_outstanding_req);
1105 static int storvsc_device_alloc(struct scsi_device *sdevice)
1107 struct stor_mem_pools *memp;
1108 int number = STORVSC_MIN_BUF_NR;
1110 memp = kzalloc(sizeof(struct stor_mem_pools), GFP_KERNEL);
1114 memp->request_pool =
1115 kmem_cache_create(dev_name(&sdevice->sdev_dev),
1116 sizeof(struct storvsc_cmd_request), 0,
1117 SLAB_HWCACHE_ALIGN, NULL);
1119 if (!memp->request_pool)
1122 memp->request_mempool = mempool_create(number, mempool_alloc_slab,
1124 memp->request_pool);
1126 if (!memp->request_mempool)
1129 sdevice->hostdata = memp;
1134 kmem_cache_destroy(memp->request_pool);
1141 static void storvsc_device_destroy(struct scsi_device *sdevice)
1143 struct stor_mem_pools *memp = sdevice->hostdata;
1145 mempool_destroy(memp->request_mempool);
1146 kmem_cache_destroy(memp->request_pool);
1148 sdevice->hostdata = NULL;
1151 static int storvsc_device_configure(struct scsi_device *sdevice)
1153 scsi_adjust_queue_depth(sdevice, MSG_SIMPLE_TAG,
1154 STORVSC_MAX_IO_REQUESTS);
1156 blk_queue_max_segment_size(sdevice->request_queue, PAGE_SIZE);
1158 blk_queue_bounce_limit(sdevice->request_queue, BLK_BOUNCE_ANY);
1163 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev,
1164 sector_t capacity, int *info)
1166 sector_t nsect = capacity;
1167 sector_t cylinders = nsect;
1168 int heads, sectors_pt;
1171 * We are making up these values; let us keep it simple.
1174 sectors_pt = 0x3f; /* Sectors per track */
1175 sector_div(cylinders, heads * sectors_pt);
1176 if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
1180 info[1] = sectors_pt;
1181 info[2] = (int)cylinders;
1186 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
1188 struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
1189 struct hv_device *device = host_dev->dev;
1191 struct storvsc_device *stor_device;
1192 struct storvsc_cmd_request *request;
1193 struct vstor_packet *vstor_packet;
1197 stor_device = get_out_stor_device(device);
1201 request = &stor_device->reset_request;
1202 vstor_packet = &request->vstor_packet;
1204 init_completion(&request->wait_event);
1206 vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
1207 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
1208 vstor_packet->vm_srb.path_id = stor_device->path_id;
1210 ret = vmbus_sendpacket(device->channel, vstor_packet,
1211 sizeof(struct vstor_packet),
1212 (unsigned long)&stor_device->reset_request,
1214 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1218 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
1220 return TIMEOUT_ERROR;
1224 * At this point, all outstanding requests in the adapter
1225 * should have been flushed out and return to us
1226 * There is a potential race here where the host may be in
1227 * the process of responding when we return from here.
1228 * Just wait for all in-transit packets to be accounted for
1229 * before we return from here.
1231 storvsc_wait_to_drain(stor_device);
1236 static bool storvsc_scsi_cmd_ok(struct scsi_cmnd *scmnd)
1238 bool allowed = true;
1239 u8 scsi_op = scmnd->cmnd[0];
1243 * smartd sends this command and the host does not handle
1244 * this. So, don't send it.
1247 scmnd->result = ILLEGAL_REQUEST << 16;
1256 static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
1259 struct hv_host_device *host_dev = shost_priv(host);
1260 struct hv_device *dev = host_dev->dev;
1261 struct storvsc_cmd_request *cmd_request;
1262 unsigned int request_size = 0;
1264 struct scatterlist *sgl;
1265 unsigned int sg_count = 0;
1266 struct vmscsi_request *vm_srb;
1267 struct stor_mem_pools *memp = scmnd->device->hostdata;
1269 if (!storvsc_scsi_cmd_ok(scmnd)) {
1270 scmnd->scsi_done(scmnd);
1274 request_size = sizeof(struct storvsc_cmd_request);
1276 cmd_request = mempool_alloc(memp->request_mempool,
1280 * We might be invoked in an interrupt context; hence
1281 * mempool_alloc() can fail.
1284 return SCSI_MLQUEUE_DEVICE_BUSY;
1286 memset(cmd_request, 0, sizeof(struct storvsc_cmd_request));
1288 /* Setup the cmd request */
1289 cmd_request->cmd = scmnd;
1291 scmnd->host_scribble = (unsigned char *)cmd_request;
1293 vm_srb = &cmd_request->vstor_packet.vm_srb;
1297 switch (scmnd->sc_data_direction) {
1299 vm_srb->data_in = WRITE_TYPE;
1301 case DMA_FROM_DEVICE:
1302 vm_srb->data_in = READ_TYPE;
1305 vm_srb->data_in = UNKNOWN_TYPE;
1310 vm_srb->port_number = host_dev->port;
1311 vm_srb->path_id = scmnd->device->channel;
1312 vm_srb->target_id = scmnd->device->id;
1313 vm_srb->lun = scmnd->device->lun;
1315 vm_srb->cdb_length = scmnd->cmd_len;
1317 memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
1319 cmd_request->sense_buffer = scmnd->sense_buffer;
1322 cmd_request->data_buffer.len = scsi_bufflen(scmnd);
1323 if (scsi_sg_count(scmnd)) {
1324 sgl = (struct scatterlist *)scsi_sglist(scmnd);
1325 sg_count = scsi_sg_count(scmnd);
1327 /* check if we need to bounce the sgl */
1328 if (do_bounce_buffer(sgl, scsi_sg_count(scmnd)) != -1) {
1329 cmd_request->bounce_sgl =
1330 create_bounce_buffer(sgl, scsi_sg_count(scmnd),
1331 scsi_bufflen(scmnd),
1333 if (!cmd_request->bounce_sgl) {
1334 ret = SCSI_MLQUEUE_HOST_BUSY;
1338 cmd_request->bounce_sgl_count =
1339 ALIGN(scsi_bufflen(scmnd), PAGE_SIZE) >>
1342 if (vm_srb->data_in == WRITE_TYPE)
1343 copy_to_bounce_buffer(sgl,
1344 cmd_request->bounce_sgl,
1345 scsi_sg_count(scmnd));
1347 sgl = cmd_request->bounce_sgl;
1348 sg_count = cmd_request->bounce_sgl_count;
1351 cmd_request->data_buffer.offset = sgl[0].offset;
1353 for (i = 0; i < sg_count; i++)
1354 cmd_request->data_buffer.pfn_array[i] =
1355 page_to_pfn(sg_page((&sgl[i])));
1357 } else if (scsi_sglist(scmnd)) {
1358 cmd_request->data_buffer.offset =
1359 virt_to_phys(scsi_sglist(scmnd)) & (PAGE_SIZE-1);
1360 cmd_request->data_buffer.pfn_array[0] =
1361 virt_to_phys(scsi_sglist(scmnd)) >> PAGE_SHIFT;
1364 /* Invokes the vsc to start an IO */
1365 ret = storvsc_do_io(dev, cmd_request);
1367 if (ret == -EAGAIN) {
1370 if (cmd_request->bounce_sgl_count) {
1371 destroy_bounce_buffer(cmd_request->bounce_sgl,
1372 cmd_request->bounce_sgl_count);
1374 ret = SCSI_MLQUEUE_DEVICE_BUSY;
1382 mempool_free(cmd_request, memp->request_mempool);
1383 scmnd->host_scribble = NULL;
1387 static struct scsi_host_template scsi_driver = {
1388 .module = THIS_MODULE,
1389 .name = "storvsc_host_t",
1390 .bios_param = storvsc_get_chs,
1391 .queuecommand = storvsc_queuecommand,
1392 .eh_host_reset_handler = storvsc_host_reset_handler,
1393 .slave_alloc = storvsc_device_alloc,
1394 .slave_destroy = storvsc_device_destroy,
1395 .slave_configure = storvsc_device_configure,
1397 /* 64 max_queue * 1 target */
1398 .can_queue = STORVSC_MAX_IO_REQUESTS*STORVSC_MAX_TARGETS,
1400 /* no use setting to 0 since ll_blk_rw reset it to 1 */
1402 .sg_tablesize = MAX_MULTIPAGE_BUFFER_COUNT,
1403 .use_clustering = DISABLE_CLUSTERING,
1404 /* Make sure we dont get a sg segment crosses a page boundary */
1405 .dma_boundary = PAGE_SIZE-1,
1413 static const struct hv_vmbus_device_id id_table[] = {
1415 { VMBUS_DEVICE(0xd9, 0x63, 0x61, 0xba, 0xa1, 0x04, 0x29, 0x4d,
1416 0xb6, 0x05, 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f)
1417 .driver_data = SCSI_GUID },
1419 { VMBUS_DEVICE(0x32, 0x26, 0x41, 0x32, 0xcb, 0x86, 0xa2, 0x44,
1420 0x9b, 0x5c, 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5)
1421 .driver_data = IDE_GUID },
1425 MODULE_DEVICE_TABLE(vmbus, id_table);
1427 static int storvsc_probe(struct hv_device *device,
1428 const struct hv_vmbus_device_id *dev_id)
1431 struct Scsi_Host *host;
1432 struct hv_host_device *host_dev;
1433 bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false);
1435 struct storvsc_device *stor_device;
1437 host = scsi_host_alloc(&scsi_driver,
1438 sizeof(struct hv_host_device));
1442 host_dev = shost_priv(host);
1443 memset(host_dev, 0, sizeof(struct hv_host_device));
1445 host_dev->port = host->host_no;
1446 host_dev->dev = device;
1449 stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
1455 stor_device->destroy = false;
1456 init_waitqueue_head(&stor_device->waiting_to_drain);
1457 stor_device->device = device;
1458 stor_device->host = host;
1459 hv_set_drvdata(device, stor_device);
1461 stor_device->port_number = host->host_no;
1462 ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size);
1466 host_dev->path = stor_device->path_id;
1467 host_dev->target = stor_device->target_id;
1469 /* max # of devices per target */
1470 host->max_lun = STORVSC_MAX_LUNS_PER_TARGET;
1471 /* max # of targets per channel */
1472 host->max_id = STORVSC_MAX_TARGETS;
1473 /* max # of channels */
1474 host->max_channel = STORVSC_MAX_CHANNELS - 1;
1475 /* max cmd length */
1476 host->max_cmd_len = STORVSC_MAX_CMD_LEN;
1478 /* Register the HBA and start the scsi bus scan */
1479 ret = scsi_add_host(host, &device->device);
1484 scsi_scan_host(host);
1486 target = (device->dev_instance.b[5] << 8 |
1487 device->dev_instance.b[4]);
1488 ret = scsi_add_device(host, 0, target, 0);
1490 scsi_remove_host(host);
1498 * Once we have connected with the host, we would need to
1499 * to invoke storvsc_dev_remove() to rollback this state and
1500 * this call also frees up the stor_device; hence the jump around
1503 storvsc_dev_remove(device);
1510 scsi_host_put(host);
1514 static int storvsc_remove(struct hv_device *dev)
1516 struct storvsc_device *stor_device = hv_get_drvdata(dev);
1517 struct Scsi_Host *host = stor_device->host;
1519 scsi_remove_host(host);
1520 storvsc_dev_remove(dev);
1521 scsi_host_put(host);
1526 static struct hv_driver storvsc_drv = {
1527 .name = KBUILD_MODNAME,
1528 .id_table = id_table,
1529 .probe = storvsc_probe,
1530 .remove = storvsc_remove,
1533 static int __init storvsc_drv_init(void)
1535 u32 max_outstanding_req_per_channel;
1538 * Divide the ring buffer data size (which is 1 page less
1539 * than the ring buffer size since that page is reserved for
1540 * the ring buffer indices) by the max request size (which is
1541 * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
1543 max_outstanding_req_per_channel =
1544 ((storvsc_ringbuffer_size - PAGE_SIZE) /
1545 ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
1546 sizeof(struct vstor_packet) + sizeof(u64),
1549 if (max_outstanding_req_per_channel <
1550 STORVSC_MAX_IO_REQUESTS)
1553 return vmbus_driver_register(&storvsc_drv);
1556 static void __exit storvsc_drv_exit(void)
1558 vmbus_driver_unregister(&storvsc_drv);
1561 MODULE_LICENSE("GPL");
1562 MODULE_VERSION(HV_DRV_VERSION);
1563 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
1564 module_init(storvsc_drv_init);
1565 module_exit(storvsc_drv_exit);