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 <scsi/scsi.h>
36 #include <scsi/scsi_cmnd.h>
37 #include <scsi/scsi_host.h>
38 #include <scsi/scsi_device.h>
39 #include <scsi/scsi_tcq.h>
40 #include <scsi/scsi_eh.h>
41 #include <scsi/scsi_devinfo.h>
42 #include <scsi/scsi_dbg.h>
45 #define STORVSC_RING_BUFFER_SIZE (20*PAGE_SIZE)
46 static int storvsc_ringbuffer_size = STORVSC_RING_BUFFER_SIZE;
48 module_param(storvsc_ringbuffer_size, int, S_IRUGO);
49 MODULE_PARM_DESC(storvsc_ringbuffer_size, "Ring buffer size (bytes)");
51 /* to alert the user that structure sizes may be mismatched even though the */
52 /* protocol versions match. */
55 #define REVISION_STRING(REVISION_) #REVISION_
56 #define FILL_VMSTOR_REVISION(RESULT_LVALUE_) \
58 char *revision_string \
59 = REVISION_STRING($Rev : 6 $) + 6; \
61 while (*revision_string >= '0' \
62 && *revision_string <= '9') { \
63 RESULT_LVALUE_ *= 10; \
64 RESULT_LVALUE_ += *revision_string - '0'; \
69 /* Major/minor macros. Minor version is in LSB, meaning that earlier flat */
70 /* version numbers will be interpreted as "0.x" (i.e., 1 becomes 0.1). */
71 #define VMSTOR_PROTOCOL_MAJOR(VERSION_) (((VERSION_) >> 8) & 0xff)
72 #define VMSTOR_PROTOCOL_MINOR(VERSION_) (((VERSION_)) & 0xff)
73 #define VMSTOR_PROTOCOL_VERSION(MAJOR_, MINOR_) ((((MAJOR_) & 0xff) << 8) | \
75 #define VMSTOR_INVALID_PROTOCOL_VERSION (-1)
77 /* Version history: */
79 /* V1 RC < 2008/1/31 1.0 */
80 /* V1 RC > 2008/1/31 2.0 */
81 #define VMSTOR_PROTOCOL_VERSION_CURRENT VMSTOR_PROTOCOL_VERSION(2, 0)
86 /* This will get replaced with the max transfer length that is possible on */
87 /* the host adapter. */
88 /* The max transfer length will be published when we offer a vmbus channel. */
89 #define MAX_TRANSFER_LENGTH 0x40000
90 #define DEFAULT_PACKET_SIZE (sizeof(struct vmdata_gpa_direct) + \
91 sizeof(struct vstor_packet) + \
92 sizesizeof(u64) * (MAX_TRANSFER_LENGTH / PAGE_SIZE)))
95 /* Packet structure describing virtual storage requests. */
96 enum vstor_packet_operation {
97 VSTOR_OPERATION_COMPLETE_IO = 1,
98 VSTOR_OPERATION_REMOVE_DEVICE = 2,
99 VSTOR_OPERATION_EXECUTE_SRB = 3,
100 VSTOR_OPERATION_RESET_LUN = 4,
101 VSTOR_OPERATION_RESET_ADAPTER = 5,
102 VSTOR_OPERATION_RESET_BUS = 6,
103 VSTOR_OPERATION_BEGIN_INITIALIZATION = 7,
104 VSTOR_OPERATION_END_INITIALIZATION = 8,
105 VSTOR_OPERATION_QUERY_PROTOCOL_VERSION = 9,
106 VSTOR_OPERATION_QUERY_PROPERTIES = 10,
107 VSTOR_OPERATION_MAXIMUM = 10
111 * Platform neutral description of a scsi request -
112 * this remains the same across the write regardless of 32/64 bit
113 * note: it's patterned off the SCSI_PASS_THROUGH structure
115 #define CDB16GENERIC_LENGTH 0x10
117 #ifndef SENSE_BUFFER_SIZE
118 #define SENSE_BUFFER_SIZE 0x12
121 #define MAX_DATA_BUF_LEN_WITH_PADDING 0x14
123 struct vmscsi_request {
124 unsigned short length;
125 unsigned char srb_status;
126 unsigned char scsi_status;
128 unsigned char port_number;
129 unsigned char path_id;
130 unsigned char target_id;
133 unsigned char cdb_length;
134 unsigned char sense_info_length;
135 unsigned char data_in;
136 unsigned char reserved;
138 unsigned int data_transfer_length;
141 unsigned char cdb[CDB16GENERIC_LENGTH];
142 unsigned char sense_data[SENSE_BUFFER_SIZE];
143 unsigned char reserved_array[MAX_DATA_BUF_LEN_WITH_PADDING];
145 } __attribute((packed));
149 * This structure is sent during the intialization phase to get the different
150 * properties of the channel.
152 struct vmstorage_channel_properties {
153 unsigned short protocol_version;
154 unsigned char path_id;
155 unsigned char target_id;
157 /* Note: port number is only really known on the client side */
158 unsigned int port_number;
160 unsigned int max_transfer_bytes;
162 /* This id is unique for each channel and will correspond with */
163 /* vendor specific data in the inquirydata */
164 unsigned long long unique_id;
167 /* This structure is sent during the storage protocol negotiations. */
168 struct vmstorage_protocol_version {
169 /* Major (MSW) and minor (LSW) version numbers. */
170 unsigned short major_minor;
173 * Revision number is auto-incremented whenever this file is changed
174 * (See FILL_VMSTOR_REVISION macro above). Mismatch does not
175 * definitely indicate incompatibility--but it does indicate mismatched
178 unsigned short revision;
181 /* Channel Property Flags */
182 #define STORAGE_CHANNEL_REMOVABLE_FLAG 0x1
183 #define STORAGE_CHANNEL_EMULATED_IDE_FLAG 0x2
185 struct vstor_packet {
186 /* Requested operation type */
187 enum vstor_packet_operation operation;
189 /* Flags - see below for values */
192 /* Status of the request returned from the server side. */
195 /* Data payload area */
198 * Structure used to forward SCSI commands from the
199 * client to the server.
201 struct vmscsi_request vm_srb;
203 /* Structure used to query channel properties. */
204 struct vmstorage_channel_properties storage_channel_properties;
206 /* Used during version negotiations. */
207 struct vmstorage_protocol_version version;
213 * This flag indicates that the server should send back a completion for this
216 #define REQUEST_COMPLETION_FLAG 0x1
218 /* This is the set of flags that the vsc can set in any packets it sends */
219 #define VSC_LEGAL_FLAGS (REQUEST_COMPLETION_FLAG)
224 #define STORVSC_MAX_IO_REQUESTS 128
227 * In Hyper-V, each port/path/target maps to 1 scsi host adapter. In
228 * reality, the path/target is not used (ie always set to 0) so our
229 * scsi host adapter essentially has 1 bus with 1 target that contains
232 #define STORVSC_MAX_LUNS_PER_TARGET 64
233 #define STORVSC_MAX_TARGETS 1
234 #define STORVSC_MAX_CHANNELS 1
235 #define STORVSC_MAX_CMD_LEN 16
237 struct hv_storvsc_request;
239 /* Matches Windows-end */
240 enum storvsc_request_type {
247 struct hv_storvsc_request {
248 struct hv_device *device;
250 /* Synchronize the request/response if needed */
251 struct completion wait_event;
253 unsigned char *sense_buffer;
255 void (*on_io_completion)(struct hv_storvsc_request *request);
256 struct hv_multipage_buffer data_buffer;
258 struct vstor_packet vstor_packet;
262 /* A storvsc device is a device object that contains a vmbus channel */
263 struct storvsc_device {
264 struct hv_device *device;
268 atomic_t num_outstanding_req;
269 struct Scsi_Host *host;
271 wait_queue_head_t waiting_to_drain;
274 * Each unique Port/Path/Target represents 1 channel ie scsi
275 * controller. In reality, the pathid, targetid is always 0
276 * and the port is set by us
278 unsigned int port_number;
279 unsigned char path_id;
280 unsigned char target_id;
282 /* Used for vsc/vsp channel reset process */
283 struct hv_storvsc_request init_request;
284 struct hv_storvsc_request reset_request;
287 struct hv_host_device {
288 struct hv_device *dev;
289 struct kmem_cache *request_pool;
292 unsigned char target;
295 struct storvsc_cmd_request {
296 struct list_head entry;
297 struct scsi_cmnd *cmd;
299 unsigned int bounce_sgl_count;
300 struct scatterlist *bounce_sgl;
302 struct hv_storvsc_request request;
305 static inline struct storvsc_device *get_out_stor_device(
306 struct hv_device *device)
308 struct storvsc_device *stor_device;
310 stor_device = hv_get_drvdata(device);
312 if (stor_device && stor_device->destroy)
319 static inline void storvsc_wait_to_drain(struct storvsc_device *dev)
321 dev->drain_notify = true;
322 wait_event(dev->waiting_to_drain,
323 atomic_read(&dev->num_outstanding_req) == 0);
324 dev->drain_notify = false;
327 static inline struct storvsc_device *get_in_stor_device(
328 struct hv_device *device)
330 struct storvsc_device *stor_device;
332 stor_device = hv_get_drvdata(device);
338 * If the device is being destroyed; allow incoming
339 * traffic only to cleanup outstanding requests.
342 if (stor_device->destroy &&
343 (atomic_read(&stor_device->num_outstanding_req) == 0))
351 static int storvsc_channel_init(struct hv_device *device)
353 struct storvsc_device *stor_device;
354 struct hv_storvsc_request *request;
355 struct vstor_packet *vstor_packet;
358 stor_device = get_out_stor_device(device);
362 request = &stor_device->init_request;
363 vstor_packet = &request->vstor_packet;
366 * Now, initiate the vsc/vsp initialization protocol on the open
369 memset(request, 0, sizeof(struct hv_storvsc_request));
370 init_completion(&request->wait_event);
371 vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION;
372 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
374 ret = vmbus_sendpacket(device->channel, vstor_packet,
375 sizeof(struct vstor_packet),
376 (unsigned long)request,
378 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
382 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
388 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
389 vstor_packet->status != 0)
393 /* reuse the packet for version range supported */
394 memset(vstor_packet, 0, sizeof(struct vstor_packet));
395 vstor_packet->operation = VSTOR_OPERATION_QUERY_PROTOCOL_VERSION;
396 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
398 vstor_packet->version.major_minor = VMSTOR_PROTOCOL_VERSION_CURRENT;
399 FILL_VMSTOR_REVISION(vstor_packet->version.revision);
401 ret = vmbus_sendpacket(device->channel, vstor_packet,
402 sizeof(struct vstor_packet),
403 (unsigned long)request,
405 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
409 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
415 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
416 vstor_packet->status != 0)
420 memset(vstor_packet, 0, sizeof(struct vstor_packet));
421 vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES;
422 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
423 vstor_packet->storage_channel_properties.port_number =
424 stor_device->port_number;
426 ret = vmbus_sendpacket(device->channel, vstor_packet,
427 sizeof(struct vstor_packet),
428 (unsigned long)request,
430 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
435 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
441 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
442 vstor_packet->status != 0)
445 stor_device->path_id = vstor_packet->storage_channel_properties.path_id;
446 stor_device->target_id
447 = vstor_packet->storage_channel_properties.target_id;
449 memset(vstor_packet, 0, sizeof(struct vstor_packet));
450 vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION;
451 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
453 ret = vmbus_sendpacket(device->channel, vstor_packet,
454 sizeof(struct vstor_packet),
455 (unsigned long)request,
457 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
462 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
468 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
469 vstor_packet->status != 0)
477 static void storvsc_on_io_completion(struct hv_device *device,
478 struct vstor_packet *vstor_packet,
479 struct hv_storvsc_request *request)
481 struct storvsc_device *stor_device;
482 struct vstor_packet *stor_pkt;
484 stor_device = hv_get_drvdata(device);
485 stor_pkt = &request->vstor_packet;
488 * The current SCSI handling on the host side does
489 * not correctly handle:
490 * INQUIRY command with page code parameter set to 0x80
491 * MODE_SENSE command with cmd[2] == 0x1c
493 * Setup srb and scsi status so this won't be fatal.
494 * We do this so we can distinguish truly fatal failues
495 * (srb status == 0x4) and off-line the device in that case.
498 if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) ||
499 (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) {
500 vstor_packet->vm_srb.scsi_status = 0;
501 vstor_packet->vm_srb.srb_status = 0x1;
505 /* Copy over the status...etc */
506 stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status;
507 stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status;
508 stor_pkt->vm_srb.sense_info_length =
509 vstor_packet->vm_srb.sense_info_length;
511 if (vstor_packet->vm_srb.scsi_status != 0 ||
512 vstor_packet->vm_srb.srb_status != 1){
514 "cmd 0x%x scsi status 0x%x srb status 0x%x\n",
515 stor_pkt->vm_srb.cdb[0],
516 vstor_packet->vm_srb.scsi_status,
517 vstor_packet->vm_srb.srb_status);
520 if ((vstor_packet->vm_srb.scsi_status & 0xFF) == 0x02) {
521 /* CHECK_CONDITION */
522 if (vstor_packet->vm_srb.srb_status & 0x80) {
523 /* autosense data available */
524 DPRINT_WARN(STORVSC, "storvsc pkt %p autosense data "
525 "valid - len %d\n", request,
526 vstor_packet->vm_srb.sense_info_length);
528 memcpy(request->sense_buffer,
529 vstor_packet->vm_srb.sense_data,
530 vstor_packet->vm_srb.sense_info_length);
535 stor_pkt->vm_srb.data_transfer_length =
536 vstor_packet->vm_srb.data_transfer_length;
538 request->on_io_completion(request);
540 if (atomic_dec_and_test(&stor_device->num_outstanding_req) &&
541 stor_device->drain_notify)
542 wake_up(&stor_device->waiting_to_drain);
547 static void storvsc_on_receive(struct hv_device *device,
548 struct vstor_packet *vstor_packet,
549 struct hv_storvsc_request *request)
551 switch (vstor_packet->operation) {
552 case VSTOR_OPERATION_COMPLETE_IO:
553 storvsc_on_io_completion(device, vstor_packet, request);
555 case VSTOR_OPERATION_REMOVE_DEVICE:
562 static void storvsc_on_channel_callback(void *context)
564 struct hv_device *device = (struct hv_device *)context;
565 struct storvsc_device *stor_device;
568 unsigned char packet[ALIGN(sizeof(struct vstor_packet), 8)];
569 struct hv_storvsc_request *request;
573 stor_device = get_in_stor_device(device);
578 ret = vmbus_recvpacket(device->channel, packet,
579 ALIGN(sizeof(struct vstor_packet), 8),
580 &bytes_recvd, &request_id);
581 if (ret == 0 && bytes_recvd > 0) {
583 request = (struct hv_storvsc_request *)
584 (unsigned long)request_id;
586 if ((request == &stor_device->init_request) ||
587 (request == &stor_device->reset_request)) {
589 memcpy(&request->vstor_packet, packet,
590 sizeof(struct vstor_packet));
591 complete(&request->wait_event);
593 storvsc_on_receive(device,
594 (struct vstor_packet *)packet,
605 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size)
607 struct vmstorage_channel_properties props;
610 memset(&props, 0, sizeof(struct vmstorage_channel_properties));
612 /* Open the channel */
613 ret = vmbus_open(device->channel,
617 sizeof(struct vmstorage_channel_properties),
618 storvsc_on_channel_callback, device);
623 ret = storvsc_channel_init(device);
628 static int storvsc_dev_remove(struct hv_device *device)
630 struct storvsc_device *stor_device;
633 stor_device = hv_get_drvdata(device);
635 spin_lock_irqsave(&device->channel->inbound_lock, flags);
636 stor_device->destroy = true;
637 spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
640 * At this point, all outbound traffic should be disable. We
641 * only allow inbound traffic (responses) to proceed so that
642 * outstanding requests can be completed.
645 storvsc_wait_to_drain(stor_device);
648 * Since we have already drained, we don't need to busy wait
649 * as was done in final_release_stor_device()
650 * Note that we cannot set the ext pointer to NULL until
651 * we have drained - to drain the outgoing packets, we need to
652 * allow incoming packets.
654 spin_lock_irqsave(&device->channel->inbound_lock, flags);
655 hv_set_drvdata(device, NULL);
656 spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
658 /* Close the channel */
659 vmbus_close(device->channel);
665 static int storvsc_do_io(struct hv_device *device,
666 struct hv_storvsc_request *request)
668 struct storvsc_device *stor_device;
669 struct vstor_packet *vstor_packet;
672 vstor_packet = &request->vstor_packet;
673 stor_device = get_out_stor_device(device);
679 request->device = device;
682 vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
684 vstor_packet->vm_srb.length = sizeof(struct vmscsi_request);
687 vstor_packet->vm_srb.sense_info_length = SENSE_BUFFER_SIZE;
690 vstor_packet->vm_srb.data_transfer_length =
691 request->data_buffer.len;
693 vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
695 if (request->data_buffer.len) {
696 ret = vmbus_sendpacket_multipagebuffer(device->channel,
697 &request->data_buffer,
699 sizeof(struct vstor_packet),
700 (unsigned long)request);
702 ret = vmbus_sendpacket(device->channel, vstor_packet,
703 sizeof(struct vstor_packet),
704 (unsigned long)request,
706 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
712 atomic_inc(&stor_device->num_outstanding_req);
717 static void storvsc_get_ide_info(struct hv_device *dev, int *target, int *path)
720 dev->dev_instance.b[5] << 8 | dev->dev_instance.b[4];
723 dev->dev_instance.b[3] << 24 |
724 dev->dev_instance.b[2] << 16 |
725 dev->dev_instance.b[1] << 8 | dev->dev_instance.b[0];
729 static int storvsc_device_alloc(struct scsi_device *sdevice)
732 * This enables luns to be located sparsely. Otherwise, we may not
735 sdevice->sdev_bflags |= BLIST_SPARSELUN | BLIST_LARGELUN;
739 static int storvsc_merge_bvec(struct request_queue *q,
740 struct bvec_merge_data *bmd, struct bio_vec *bvec)
742 /* checking done by caller. */
746 static int storvsc_device_configure(struct scsi_device *sdevice)
748 scsi_adjust_queue_depth(sdevice, MSG_SIMPLE_TAG,
749 STORVSC_MAX_IO_REQUESTS);
751 blk_queue_max_segment_size(sdevice->request_queue, PAGE_SIZE);
753 blk_queue_merge_bvec(sdevice->request_queue, storvsc_merge_bvec);
755 blk_queue_bounce_limit(sdevice->request_queue, BLK_BOUNCE_ANY);
760 static void destroy_bounce_buffer(struct scatterlist *sgl,
761 unsigned int sg_count)
764 struct page *page_buf;
766 for (i = 0; i < sg_count; i++) {
767 page_buf = sg_page((&sgl[i]));
768 if (page_buf != NULL)
769 __free_page(page_buf);
775 static int do_bounce_buffer(struct scatterlist *sgl, unsigned int sg_count)
779 /* No need to check */
783 /* We have at least 2 sg entries */
784 for (i = 0; i < sg_count; i++) {
786 /* make sure 1st one does not have hole */
787 if (sgl[i].offset + sgl[i].length != PAGE_SIZE)
789 } else if (i == sg_count - 1) {
790 /* make sure last one does not have hole */
791 if (sgl[i].offset != 0)
794 /* make sure no hole in the middle */
795 if (sgl[i].length != PAGE_SIZE || sgl[i].offset != 0)
802 static struct scatterlist *create_bounce_buffer(struct scatterlist *sgl,
803 unsigned int sg_count,
808 struct scatterlist *bounce_sgl;
809 struct page *page_buf;
811 num_pages = ALIGN(len, PAGE_SIZE) >> PAGE_SHIFT;
813 bounce_sgl = kcalloc(num_pages, sizeof(struct scatterlist), GFP_ATOMIC);
817 for (i = 0; i < num_pages; i++) {
818 page_buf = alloc_page(GFP_ATOMIC);
821 sg_set_page(&bounce_sgl[i], page_buf, 0, 0);
827 destroy_bounce_buffer(bounce_sgl, num_pages);
832 /* Assume the original sgl has enough room */
833 static unsigned int copy_from_bounce_buffer(struct scatterlist *orig_sgl,
834 struct scatterlist *bounce_sgl,
835 unsigned int orig_sgl_count)
839 unsigned long src, dest;
840 unsigned int srclen, destlen, copylen;
841 unsigned int total_copied = 0;
842 unsigned long bounce_addr = 0;
843 unsigned long dest_addr = 0;
846 local_irq_save(flags);
848 for (i = 0; i < orig_sgl_count; i++) {
849 dest_addr = (unsigned long)kmap_atomic(sg_page((&orig_sgl[i])),
850 KM_IRQ0) + orig_sgl[i].offset;
852 destlen = orig_sgl[i].length;
854 if (bounce_addr == 0)
856 (unsigned long)kmap_atomic(sg_page((&bounce_sgl[j])),
860 src = bounce_addr + bounce_sgl[j].offset;
861 srclen = bounce_sgl[j].length - bounce_sgl[j].offset;
863 copylen = min(srclen, destlen);
864 memcpy((void *)dest, (void *)src, copylen);
866 total_copied += copylen;
867 bounce_sgl[j].offset += copylen;
871 if (bounce_sgl[j].offset == bounce_sgl[j].length) {
873 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
876 /* if we need to use another bounce buffer */
877 if (destlen || i != orig_sgl_count - 1)
879 (unsigned long)kmap_atomic(
880 sg_page((&bounce_sgl[j])), KM_IRQ0);
881 } else if (destlen == 0 && i == orig_sgl_count - 1) {
882 /* unmap the last bounce that is < PAGE_SIZE */
883 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
887 kunmap_atomic((void *)(dest_addr - orig_sgl[i].offset),
891 local_irq_restore(flags);
897 /* Assume the bounce_sgl has enough room ie using the create_bounce_buffer() */
898 static unsigned int copy_to_bounce_buffer(struct scatterlist *orig_sgl,
899 struct scatterlist *bounce_sgl,
900 unsigned int orig_sgl_count)
904 unsigned long src, dest;
905 unsigned int srclen, destlen, copylen;
906 unsigned int total_copied = 0;
907 unsigned long bounce_addr = 0;
908 unsigned long src_addr = 0;
911 local_irq_save(flags);
913 for (i = 0; i < orig_sgl_count; i++) {
914 src_addr = (unsigned long)kmap_atomic(sg_page((&orig_sgl[i])),
915 KM_IRQ0) + orig_sgl[i].offset;
917 srclen = orig_sgl[i].length;
919 if (bounce_addr == 0)
921 (unsigned long)kmap_atomic(sg_page((&bounce_sgl[j])),
925 /* assume bounce offset always == 0 */
926 dest = bounce_addr + bounce_sgl[j].length;
927 destlen = PAGE_SIZE - bounce_sgl[j].length;
929 copylen = min(srclen, destlen);
930 memcpy((void *)dest, (void *)src, copylen);
932 total_copied += copylen;
933 bounce_sgl[j].length += copylen;
937 if (bounce_sgl[j].length == PAGE_SIZE) {
938 /* full..move to next entry */
939 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
942 /* if we need to use another bounce buffer */
943 if (srclen || i != orig_sgl_count - 1)
945 (unsigned long)kmap_atomic(
946 sg_page((&bounce_sgl[j])), KM_IRQ0);
948 } else if (srclen == 0 && i == orig_sgl_count - 1) {
949 /* unmap the last bounce that is < PAGE_SIZE */
950 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
954 kunmap_atomic((void *)(src_addr - orig_sgl[i].offset), KM_IRQ0);
957 local_irq_restore(flags);
963 static int storvsc_remove(struct hv_device *dev)
965 struct storvsc_device *stor_device = hv_get_drvdata(dev);
966 struct Scsi_Host *host = stor_device->host;
967 struct hv_host_device *host_dev =
968 (struct hv_host_device *)host->hostdata;
970 scsi_remove_host(host);
974 storvsc_dev_remove(dev);
975 if (host_dev->request_pool) {
976 kmem_cache_destroy(host_dev->request_pool);
977 host_dev->request_pool = NULL;
983 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev,
984 sector_t capacity, int *info)
986 sector_t nsect = capacity;
987 sector_t cylinders = nsect;
988 int heads, sectors_pt;
991 * We are making up these values; let us keep it simple.
994 sectors_pt = 0x3f; /* Sectors per track */
995 sector_div(cylinders, heads * sectors_pt);
996 if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
1000 info[1] = sectors_pt;
1001 info[2] = (int)cylinders;
1006 static int storvsc_host_reset(struct hv_device *device)
1008 struct storvsc_device *stor_device;
1009 struct hv_storvsc_request *request;
1010 struct vstor_packet *vstor_packet;
1014 stor_device = get_out_stor_device(device);
1018 request = &stor_device->reset_request;
1019 vstor_packet = &request->vstor_packet;
1021 init_completion(&request->wait_event);
1023 vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
1024 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
1025 vstor_packet->vm_srb.path_id = stor_device->path_id;
1027 ret = vmbus_sendpacket(device->channel, vstor_packet,
1028 sizeof(struct vstor_packet),
1029 (unsigned long)&stor_device->reset_request,
1031 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1035 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
1043 * At this point, all outstanding requests in the adapter
1044 * should have been flushed out and return to us
1053 * storvsc_host_reset_handler - Reset the scsi HBA
1055 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
1058 struct hv_host_device *host_dev =
1059 (struct hv_host_device *)scmnd->device->host->hostdata;
1060 struct hv_device *dev = host_dev->dev;
1062 ret = storvsc_host_reset(dev);
1071 * storvsc_command_completion - Command completion processing
1073 static void storvsc_command_completion(struct hv_storvsc_request *request)
1075 struct storvsc_cmd_request *cmd_request =
1076 (struct storvsc_cmd_request *)request->context;
1077 struct scsi_cmnd *scmnd = cmd_request->cmd;
1078 struct hv_host_device *host_dev =
1079 (struct hv_host_device *)scmnd->device->host->hostdata;
1080 void (*scsi_done_fn)(struct scsi_cmnd *);
1081 struct scsi_sense_hdr sense_hdr;
1082 struct vmscsi_request *vm_srb;
1084 vm_srb = &request->vstor_packet.vm_srb;
1085 if (cmd_request->bounce_sgl_count) {
1086 if (vm_srb->data_in == READ_TYPE) {
1087 copy_from_bounce_buffer(scsi_sglist(scmnd),
1088 cmd_request->bounce_sgl,
1089 scsi_sg_count(scmnd));
1090 destroy_bounce_buffer(cmd_request->bounce_sgl,
1091 cmd_request->bounce_sgl_count);
1096 * If there is an error; offline the device since all
1097 * error recovery strategies would have already been
1098 * deployed on the host side.
1100 if (vm_srb->srb_status == 0x4)
1101 scmnd->result = DID_TARGET_FAILURE << 16;
1103 scmnd->result = vm_srb->scsi_status;
1105 if (scmnd->result) {
1106 if (scsi_normalize_sense(scmnd->sense_buffer,
1107 SCSI_SENSE_BUFFERSIZE, &sense_hdr))
1108 scsi_print_sense_hdr("storvsc", &sense_hdr);
1111 scsi_set_resid(scmnd,
1112 request->data_buffer.len -
1113 vm_srb->data_transfer_length);
1115 scsi_done_fn = scmnd->scsi_done;
1117 scmnd->host_scribble = NULL;
1118 scmnd->scsi_done = NULL;
1120 scsi_done_fn(scmnd);
1122 kmem_cache_free(host_dev->request_pool, cmd_request);
1125 static bool storvsc_check_scsi_cmd(struct scsi_cmnd *scmnd)
1127 bool allowed = true;
1128 u8 scsi_op = scmnd->cmnd[0];
1131 /* smartd sends this command, which will offline the device */
1133 scmnd->result = DID_ERROR << 16;
1143 * storvsc_queuecommand - Initiate command processing
1145 static int storvsc_queuecommand_lck(struct scsi_cmnd *scmnd,
1146 void (*done)(struct scsi_cmnd *))
1149 struct hv_host_device *host_dev =
1150 (struct hv_host_device *)scmnd->device->host->hostdata;
1151 struct hv_device *dev = host_dev->dev;
1152 struct hv_storvsc_request *request;
1153 struct storvsc_cmd_request *cmd_request;
1154 unsigned int request_size = 0;
1156 struct scatterlist *sgl;
1157 unsigned int sg_count = 0;
1158 struct vmscsi_request *vm_srb;
1160 if (storvsc_check_scsi_cmd(scmnd) == false) {
1165 /* If retrying, no need to prep the cmd */
1166 if (scmnd->host_scribble) {
1169 (struct storvsc_cmd_request *)scmnd->host_scribble;
1174 scmnd->scsi_done = done;
1176 request_size = sizeof(struct storvsc_cmd_request);
1178 cmd_request = kmem_cache_zalloc(host_dev->request_pool,
1181 scmnd->scsi_done = NULL;
1182 return SCSI_MLQUEUE_DEVICE_BUSY;
1185 /* Setup the cmd request */
1186 cmd_request->bounce_sgl_count = 0;
1187 cmd_request->bounce_sgl = NULL;
1188 cmd_request->cmd = scmnd;
1190 scmnd->host_scribble = (unsigned char *)cmd_request;
1192 request = &cmd_request->request;
1193 vm_srb = &request->vstor_packet.vm_srb;
1197 switch (scmnd->sc_data_direction) {
1199 vm_srb->data_in = WRITE_TYPE;
1201 case DMA_FROM_DEVICE:
1202 vm_srb->data_in = READ_TYPE;
1205 vm_srb->data_in = UNKNOWN_TYPE;
1209 request->on_io_completion = storvsc_command_completion;
1210 request->context = cmd_request;/* scmnd; */
1212 vm_srb->port_number = host_dev->port;
1213 vm_srb->path_id = scmnd->device->channel;
1214 vm_srb->target_id = scmnd->device->id;
1215 vm_srb->lun = scmnd->device->lun;
1217 vm_srb->cdb_length = scmnd->cmd_len;
1219 memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
1221 request->sense_buffer = scmnd->sense_buffer;
1224 request->data_buffer.len = scsi_bufflen(scmnd);
1225 if (scsi_sg_count(scmnd)) {
1226 sgl = (struct scatterlist *)scsi_sglist(scmnd);
1227 sg_count = scsi_sg_count(scmnd);
1229 /* check if we need to bounce the sgl */
1230 if (do_bounce_buffer(sgl, scsi_sg_count(scmnd)) != -1) {
1231 cmd_request->bounce_sgl =
1232 create_bounce_buffer(sgl, scsi_sg_count(scmnd),
1233 scsi_bufflen(scmnd));
1234 if (!cmd_request->bounce_sgl) {
1235 scmnd->scsi_done = NULL;
1236 scmnd->host_scribble = NULL;
1237 kmem_cache_free(host_dev->request_pool,
1240 return SCSI_MLQUEUE_HOST_BUSY;
1243 cmd_request->bounce_sgl_count =
1244 ALIGN(scsi_bufflen(scmnd), PAGE_SIZE) >>
1247 if (vm_srb->data_in == WRITE_TYPE)
1248 copy_to_bounce_buffer(sgl,
1249 cmd_request->bounce_sgl,
1250 scsi_sg_count(scmnd));
1252 sgl = cmd_request->bounce_sgl;
1253 sg_count = cmd_request->bounce_sgl_count;
1256 request->data_buffer.offset = sgl[0].offset;
1258 for (i = 0; i < sg_count; i++)
1259 request->data_buffer.pfn_array[i] =
1260 page_to_pfn(sg_page((&sgl[i])));
1262 } else if (scsi_sglist(scmnd)) {
1263 request->data_buffer.offset =
1264 virt_to_phys(scsi_sglist(scmnd)) & (PAGE_SIZE-1);
1265 request->data_buffer.pfn_array[0] =
1266 virt_to_phys(scsi_sglist(scmnd)) >> PAGE_SHIFT;
1270 /* Invokes the vsc to start an IO */
1271 ret = storvsc_do_io(dev, &cmd_request->request);
1273 if (ret == -EAGAIN) {
1276 if (cmd_request->bounce_sgl_count)
1277 destroy_bounce_buffer(cmd_request->bounce_sgl,
1278 cmd_request->bounce_sgl_count);
1280 kmem_cache_free(host_dev->request_pool, cmd_request);
1282 scmnd->scsi_done = NULL;
1283 scmnd->host_scribble = NULL;
1285 ret = SCSI_MLQUEUE_DEVICE_BUSY;
1291 static DEF_SCSI_QCMD(storvsc_queuecommand)
1295 static struct scsi_host_template scsi_driver = {
1296 .module = THIS_MODULE,
1297 .name = "storvsc_host_t",
1298 .bios_param = storvsc_get_chs,
1299 .queuecommand = storvsc_queuecommand,
1300 .eh_host_reset_handler = storvsc_host_reset_handler,
1301 .slave_alloc = storvsc_device_alloc,
1302 .slave_configure = storvsc_device_configure,
1304 /* 64 max_queue * 1 target */
1305 .can_queue = STORVSC_MAX_IO_REQUESTS*STORVSC_MAX_TARGETS,
1307 /* no use setting to 0 since ll_blk_rw reset it to 1 */
1309 .sg_tablesize = MAX_MULTIPAGE_BUFFER_COUNT,
1311 * ENABLE_CLUSTERING allows mutiple physically contig bio_vecs to merge
1312 * into 1 sg element. If set, we must limit the max_segment_size to
1313 * PAGE_SIZE, otherwise we may get 1 sg element that represents
1316 /* physically contig pfns (ie sg[x].length > PAGE_SIZE). */
1317 .use_clustering = ENABLE_CLUSTERING,
1318 /* Make sure we dont get a sg segment crosses a page boundary */
1319 .dma_boundary = PAGE_SIZE-1,
1327 static const struct hv_vmbus_device_id id_table[] = {
1329 { VMBUS_DEVICE(0xd9, 0x63, 0x61, 0xba, 0xa1, 0x04, 0x29, 0x4d,
1330 0xb6, 0x05, 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f)
1331 .driver_data = SCSI_GUID },
1333 { VMBUS_DEVICE(0x32, 0x26, 0x41, 0x32, 0xcb, 0x86, 0xa2, 0x44,
1334 0x9b, 0x5c, 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5)
1335 .driver_data = IDE_GUID },
1339 MODULE_DEVICE_TABLE(vmbus, id_table);
1343 * storvsc_probe - Add a new device for this driver
1346 static int storvsc_probe(struct hv_device *device,
1347 const struct hv_vmbus_device_id *dev_id)
1350 struct Scsi_Host *host;
1351 struct hv_host_device *host_dev;
1352 bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false);
1355 struct storvsc_device *stor_device;
1357 host = scsi_host_alloc(&scsi_driver,
1358 sizeof(struct hv_host_device));
1362 host_dev = (struct hv_host_device *)host->hostdata;
1363 memset(host_dev, 0, sizeof(struct hv_host_device));
1365 host_dev->port = host->host_no;
1366 host_dev->dev = device;
1368 host_dev->request_pool =
1369 kmem_cache_create(dev_name(&device->device),
1370 sizeof(struct storvsc_cmd_request), 0,
1371 SLAB_HWCACHE_ALIGN, NULL);
1373 if (!host_dev->request_pool) {
1374 scsi_host_put(host);
1378 stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
1380 kmem_cache_destroy(host_dev->request_pool);
1381 scsi_host_put(host);
1385 stor_device->destroy = false;
1386 init_waitqueue_head(&stor_device->waiting_to_drain);
1387 stor_device->device = device;
1388 stor_device->host = host;
1389 hv_set_drvdata(device, stor_device);
1391 stor_device->port_number = host->host_no;
1392 ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size);
1394 kmem_cache_destroy(host_dev->request_pool);
1395 scsi_host_put(host);
1401 storvsc_get_ide_info(device, &target, &path);
1403 host_dev->path = stor_device->path_id;
1404 host_dev->target = stor_device->target_id;
1406 /* max # of devices per target */
1407 host->max_lun = STORVSC_MAX_LUNS_PER_TARGET;
1408 /* max # of targets per channel */
1409 host->max_id = STORVSC_MAX_TARGETS;
1410 /* max # of channels */
1411 host->max_channel = STORVSC_MAX_CHANNELS - 1;
1412 /* max cmd length */
1413 host->max_cmd_len = STORVSC_MAX_CMD_LEN;
1415 /* Register the HBA and start the scsi bus scan */
1416 ret = scsi_add_host(host, &device->device);
1421 scsi_scan_host(host);
1424 ret = scsi_add_device(host, 0, target, 0);
1426 scsi_remove_host(host);
1432 storvsc_dev_remove(device);
1433 kmem_cache_destroy(host_dev->request_pool);
1434 scsi_host_put(host);
1438 /* The one and only one */
1440 static struct hv_driver storvsc_drv = {
1442 .id_table = id_table,
1443 .probe = storvsc_probe,
1444 .remove = storvsc_remove,
1447 static int __init storvsc_drv_init(void)
1449 u32 max_outstanding_req_per_channel;
1452 * Divide the ring buffer data size (which is 1 page less
1453 * than the ring buffer size since that page is reserved for
1454 * the ring buffer indices) by the max request size (which is
1455 * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
1457 max_outstanding_req_per_channel =
1458 ((storvsc_ringbuffer_size - PAGE_SIZE) /
1459 ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
1460 sizeof(struct vstor_packet) + sizeof(u64),
1463 if (max_outstanding_req_per_channel <
1464 STORVSC_MAX_IO_REQUESTS)
1467 return vmbus_driver_register(&storvsc_drv);
1470 static void __exit storvsc_drv_exit(void)
1472 vmbus_driver_unregister(&storvsc_drv);
1475 MODULE_LICENSE("GPL");
1476 MODULE_VERSION(HV_DRV_VERSION);
1477 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
1478 module_init(storvsc_drv_init);
1479 module_exit(storvsc_drv_exit);