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){
513 dev_warn(&device->device,
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 dev_warn(&device->device,
525 "stor pkt %p autosense data valid - len %d\n",
527 vstor_packet->vm_srb.sense_info_length);
529 memcpy(request->sense_buffer,
530 vstor_packet->vm_srb.sense_data,
531 vstor_packet->vm_srb.sense_info_length);
536 stor_pkt->vm_srb.data_transfer_length =
537 vstor_packet->vm_srb.data_transfer_length;
539 request->on_io_completion(request);
541 if (atomic_dec_and_test(&stor_device->num_outstanding_req) &&
542 stor_device->drain_notify)
543 wake_up(&stor_device->waiting_to_drain);
548 static void storvsc_on_receive(struct hv_device *device,
549 struct vstor_packet *vstor_packet,
550 struct hv_storvsc_request *request)
552 switch (vstor_packet->operation) {
553 case VSTOR_OPERATION_COMPLETE_IO:
554 storvsc_on_io_completion(device, vstor_packet, request);
556 case VSTOR_OPERATION_REMOVE_DEVICE:
563 static void storvsc_on_channel_callback(void *context)
565 struct hv_device *device = (struct hv_device *)context;
566 struct storvsc_device *stor_device;
569 unsigned char packet[ALIGN(sizeof(struct vstor_packet), 8)];
570 struct hv_storvsc_request *request;
574 stor_device = get_in_stor_device(device);
579 ret = vmbus_recvpacket(device->channel, packet,
580 ALIGN(sizeof(struct vstor_packet), 8),
581 &bytes_recvd, &request_id);
582 if (ret == 0 && bytes_recvd > 0) {
584 request = (struct hv_storvsc_request *)
585 (unsigned long)request_id;
587 if ((request == &stor_device->init_request) ||
588 (request == &stor_device->reset_request)) {
590 memcpy(&request->vstor_packet, packet,
591 sizeof(struct vstor_packet));
592 complete(&request->wait_event);
594 storvsc_on_receive(device,
595 (struct vstor_packet *)packet,
606 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size)
608 struct vmstorage_channel_properties props;
611 memset(&props, 0, sizeof(struct vmstorage_channel_properties));
613 /* Open the channel */
614 ret = vmbus_open(device->channel,
618 sizeof(struct vmstorage_channel_properties),
619 storvsc_on_channel_callback, device);
624 ret = storvsc_channel_init(device);
629 static int storvsc_dev_remove(struct hv_device *device)
631 struct storvsc_device *stor_device;
634 stor_device = hv_get_drvdata(device);
636 spin_lock_irqsave(&device->channel->inbound_lock, flags);
637 stor_device->destroy = true;
638 spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
641 * At this point, all outbound traffic should be disable. We
642 * only allow inbound traffic (responses) to proceed so that
643 * outstanding requests can be completed.
646 storvsc_wait_to_drain(stor_device);
649 * Since we have already drained, we don't need to busy wait
650 * as was done in final_release_stor_device()
651 * Note that we cannot set the ext pointer to NULL until
652 * we have drained - to drain the outgoing packets, we need to
653 * allow incoming packets.
655 spin_lock_irqsave(&device->channel->inbound_lock, flags);
656 hv_set_drvdata(device, NULL);
657 spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
659 /* Close the channel */
660 vmbus_close(device->channel);
666 static int storvsc_do_io(struct hv_device *device,
667 struct hv_storvsc_request *request)
669 struct storvsc_device *stor_device;
670 struct vstor_packet *vstor_packet;
673 vstor_packet = &request->vstor_packet;
674 stor_device = get_out_stor_device(device);
680 request->device = device;
683 vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
685 vstor_packet->vm_srb.length = sizeof(struct vmscsi_request);
688 vstor_packet->vm_srb.sense_info_length = SENSE_BUFFER_SIZE;
691 vstor_packet->vm_srb.data_transfer_length =
692 request->data_buffer.len;
694 vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
696 if (request->data_buffer.len) {
697 ret = vmbus_sendpacket_multipagebuffer(device->channel,
698 &request->data_buffer,
700 sizeof(struct vstor_packet),
701 (unsigned long)request);
703 ret = vmbus_sendpacket(device->channel, vstor_packet,
704 sizeof(struct vstor_packet),
705 (unsigned long)request,
707 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
713 atomic_inc(&stor_device->num_outstanding_req);
718 static void storvsc_get_ide_info(struct hv_device *dev, int *target, int *path)
721 dev->dev_instance.b[5] << 8 | dev->dev_instance.b[4];
724 dev->dev_instance.b[3] << 24 |
725 dev->dev_instance.b[2] << 16 |
726 dev->dev_instance.b[1] << 8 | dev->dev_instance.b[0];
730 static int storvsc_device_alloc(struct scsi_device *sdevice)
733 * This enables luns to be located sparsely. Otherwise, we may not
736 sdevice->sdev_bflags |= BLIST_SPARSELUN | BLIST_LARGELUN;
740 static int storvsc_merge_bvec(struct request_queue *q,
741 struct bvec_merge_data *bmd, struct bio_vec *bvec)
743 /* checking done by caller. */
747 static int storvsc_device_configure(struct scsi_device *sdevice)
749 scsi_adjust_queue_depth(sdevice, MSG_SIMPLE_TAG,
750 STORVSC_MAX_IO_REQUESTS);
752 blk_queue_max_segment_size(sdevice->request_queue, PAGE_SIZE);
754 blk_queue_merge_bvec(sdevice->request_queue, storvsc_merge_bvec);
756 blk_queue_bounce_limit(sdevice->request_queue, BLK_BOUNCE_ANY);
761 static void destroy_bounce_buffer(struct scatterlist *sgl,
762 unsigned int sg_count)
765 struct page *page_buf;
767 for (i = 0; i < sg_count; i++) {
768 page_buf = sg_page((&sgl[i]));
769 if (page_buf != NULL)
770 __free_page(page_buf);
776 static int do_bounce_buffer(struct scatterlist *sgl, unsigned int sg_count)
780 /* No need to check */
784 /* We have at least 2 sg entries */
785 for (i = 0; i < sg_count; i++) {
787 /* make sure 1st one does not have hole */
788 if (sgl[i].offset + sgl[i].length != PAGE_SIZE)
790 } else if (i == sg_count - 1) {
791 /* make sure last one does not have hole */
792 if (sgl[i].offset != 0)
795 /* make sure no hole in the middle */
796 if (sgl[i].length != PAGE_SIZE || sgl[i].offset != 0)
803 static struct scatterlist *create_bounce_buffer(struct scatterlist *sgl,
804 unsigned int sg_count,
809 struct scatterlist *bounce_sgl;
810 struct page *page_buf;
812 num_pages = ALIGN(len, PAGE_SIZE) >> PAGE_SHIFT;
814 bounce_sgl = kcalloc(num_pages, sizeof(struct scatterlist), GFP_ATOMIC);
818 for (i = 0; i < num_pages; i++) {
819 page_buf = alloc_page(GFP_ATOMIC);
822 sg_set_page(&bounce_sgl[i], page_buf, 0, 0);
828 destroy_bounce_buffer(bounce_sgl, num_pages);
833 /* Assume the original sgl has enough room */
834 static unsigned int copy_from_bounce_buffer(struct scatterlist *orig_sgl,
835 struct scatterlist *bounce_sgl,
836 unsigned int orig_sgl_count)
840 unsigned long src, dest;
841 unsigned int srclen, destlen, copylen;
842 unsigned int total_copied = 0;
843 unsigned long bounce_addr = 0;
844 unsigned long dest_addr = 0;
847 local_irq_save(flags);
849 for (i = 0; i < orig_sgl_count; i++) {
850 dest_addr = (unsigned long)kmap_atomic(sg_page((&orig_sgl[i])),
851 KM_IRQ0) + orig_sgl[i].offset;
853 destlen = orig_sgl[i].length;
855 if (bounce_addr == 0)
857 (unsigned long)kmap_atomic(sg_page((&bounce_sgl[j])),
861 src = bounce_addr + bounce_sgl[j].offset;
862 srclen = bounce_sgl[j].length - bounce_sgl[j].offset;
864 copylen = min(srclen, destlen);
865 memcpy((void *)dest, (void *)src, copylen);
867 total_copied += copylen;
868 bounce_sgl[j].offset += copylen;
872 if (bounce_sgl[j].offset == bounce_sgl[j].length) {
874 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
877 /* if we need to use another bounce buffer */
878 if (destlen || i != orig_sgl_count - 1)
880 (unsigned long)kmap_atomic(
881 sg_page((&bounce_sgl[j])), KM_IRQ0);
882 } else if (destlen == 0 && i == orig_sgl_count - 1) {
883 /* unmap the last bounce that is < PAGE_SIZE */
884 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
888 kunmap_atomic((void *)(dest_addr - orig_sgl[i].offset),
892 local_irq_restore(flags);
898 /* Assume the bounce_sgl has enough room ie using the create_bounce_buffer() */
899 static unsigned int copy_to_bounce_buffer(struct scatterlist *orig_sgl,
900 struct scatterlist *bounce_sgl,
901 unsigned int orig_sgl_count)
905 unsigned long src, dest;
906 unsigned int srclen, destlen, copylen;
907 unsigned int total_copied = 0;
908 unsigned long bounce_addr = 0;
909 unsigned long src_addr = 0;
912 local_irq_save(flags);
914 for (i = 0; i < orig_sgl_count; i++) {
915 src_addr = (unsigned long)kmap_atomic(sg_page((&orig_sgl[i])),
916 KM_IRQ0) + orig_sgl[i].offset;
918 srclen = orig_sgl[i].length;
920 if (bounce_addr == 0)
922 (unsigned long)kmap_atomic(sg_page((&bounce_sgl[j])),
926 /* assume bounce offset always == 0 */
927 dest = bounce_addr + bounce_sgl[j].length;
928 destlen = PAGE_SIZE - bounce_sgl[j].length;
930 copylen = min(srclen, destlen);
931 memcpy((void *)dest, (void *)src, copylen);
933 total_copied += copylen;
934 bounce_sgl[j].length += copylen;
938 if (bounce_sgl[j].length == PAGE_SIZE) {
939 /* full..move to next entry */
940 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
943 /* if we need to use another bounce buffer */
944 if (srclen || i != orig_sgl_count - 1)
946 (unsigned long)kmap_atomic(
947 sg_page((&bounce_sgl[j])), KM_IRQ0);
949 } else if (srclen == 0 && i == orig_sgl_count - 1) {
950 /* unmap the last bounce that is < PAGE_SIZE */
951 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
955 kunmap_atomic((void *)(src_addr - orig_sgl[i].offset), KM_IRQ0);
958 local_irq_restore(flags);
964 static int storvsc_remove(struct hv_device *dev)
966 struct storvsc_device *stor_device = hv_get_drvdata(dev);
967 struct Scsi_Host *host = stor_device->host;
968 struct hv_host_device *host_dev =
969 (struct hv_host_device *)host->hostdata;
971 scsi_remove_host(host);
975 storvsc_dev_remove(dev);
976 if (host_dev->request_pool) {
977 kmem_cache_destroy(host_dev->request_pool);
978 host_dev->request_pool = NULL;
984 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev,
985 sector_t capacity, int *info)
987 sector_t nsect = capacity;
988 sector_t cylinders = nsect;
989 int heads, sectors_pt;
992 * We are making up these values; let us keep it simple.
995 sectors_pt = 0x3f; /* Sectors per track */
996 sector_div(cylinders, heads * sectors_pt);
997 if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
1001 info[1] = sectors_pt;
1002 info[2] = (int)cylinders;
1007 static int storvsc_host_reset(struct hv_device *device)
1009 struct storvsc_device *stor_device;
1010 struct hv_storvsc_request *request;
1011 struct vstor_packet *vstor_packet;
1015 stor_device = get_out_stor_device(device);
1019 request = &stor_device->reset_request;
1020 vstor_packet = &request->vstor_packet;
1022 init_completion(&request->wait_event);
1024 vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
1025 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
1026 vstor_packet->vm_srb.path_id = stor_device->path_id;
1028 ret = vmbus_sendpacket(device->channel, vstor_packet,
1029 sizeof(struct vstor_packet),
1030 (unsigned long)&stor_device->reset_request,
1032 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1036 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
1044 * At this point, all outstanding requests in the adapter
1045 * should have been flushed out and return to us
1054 * storvsc_host_reset_handler - Reset the scsi HBA
1056 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
1059 struct hv_host_device *host_dev =
1060 (struct hv_host_device *)scmnd->device->host->hostdata;
1061 struct hv_device *dev = host_dev->dev;
1063 ret = storvsc_host_reset(dev);
1072 * storvsc_command_completion - Command completion processing
1074 static void storvsc_command_completion(struct hv_storvsc_request *request)
1076 struct storvsc_cmd_request *cmd_request =
1077 (struct storvsc_cmd_request *)request->context;
1078 struct scsi_cmnd *scmnd = cmd_request->cmd;
1079 struct hv_host_device *host_dev =
1080 (struct hv_host_device *)scmnd->device->host->hostdata;
1081 void (*scsi_done_fn)(struct scsi_cmnd *);
1082 struct scsi_sense_hdr sense_hdr;
1083 struct vmscsi_request *vm_srb;
1085 vm_srb = &request->vstor_packet.vm_srb;
1086 if (cmd_request->bounce_sgl_count) {
1087 if (vm_srb->data_in == READ_TYPE) {
1088 copy_from_bounce_buffer(scsi_sglist(scmnd),
1089 cmd_request->bounce_sgl,
1090 scsi_sg_count(scmnd));
1091 destroy_bounce_buffer(cmd_request->bounce_sgl,
1092 cmd_request->bounce_sgl_count);
1097 * If there is an error; offline the device since all
1098 * error recovery strategies would have already been
1099 * deployed on the host side.
1101 if (vm_srb->srb_status == 0x4)
1102 scmnd->result = DID_TARGET_FAILURE << 16;
1104 scmnd->result = vm_srb->scsi_status;
1106 if (scmnd->result) {
1107 if (scsi_normalize_sense(scmnd->sense_buffer,
1108 SCSI_SENSE_BUFFERSIZE, &sense_hdr))
1109 scsi_print_sense_hdr("storvsc", &sense_hdr);
1112 scsi_set_resid(scmnd,
1113 request->data_buffer.len -
1114 vm_srb->data_transfer_length);
1116 scsi_done_fn = scmnd->scsi_done;
1118 scmnd->host_scribble = NULL;
1119 scmnd->scsi_done = NULL;
1121 scsi_done_fn(scmnd);
1123 kmem_cache_free(host_dev->request_pool, cmd_request);
1126 static bool storvsc_check_scsi_cmd(struct scsi_cmnd *scmnd)
1128 bool allowed = true;
1129 u8 scsi_op = scmnd->cmnd[0];
1132 /* smartd sends this command, which will offline the device */
1134 scmnd->result = DID_ERROR << 16;
1144 * storvsc_queuecommand - Initiate command processing
1146 static int storvsc_queuecommand_lck(struct scsi_cmnd *scmnd,
1147 void (*done)(struct scsi_cmnd *))
1150 struct hv_host_device *host_dev =
1151 (struct hv_host_device *)scmnd->device->host->hostdata;
1152 struct hv_device *dev = host_dev->dev;
1153 struct hv_storvsc_request *request;
1154 struct storvsc_cmd_request *cmd_request;
1155 unsigned int request_size = 0;
1157 struct scatterlist *sgl;
1158 unsigned int sg_count = 0;
1159 struct vmscsi_request *vm_srb;
1161 if (storvsc_check_scsi_cmd(scmnd) == false) {
1166 /* If retrying, no need to prep the cmd */
1167 if (scmnd->host_scribble) {
1170 (struct storvsc_cmd_request *)scmnd->host_scribble;
1175 scmnd->scsi_done = done;
1177 request_size = sizeof(struct storvsc_cmd_request);
1179 cmd_request = kmem_cache_zalloc(host_dev->request_pool,
1182 scmnd->scsi_done = NULL;
1183 return SCSI_MLQUEUE_DEVICE_BUSY;
1186 /* Setup the cmd request */
1187 cmd_request->bounce_sgl_count = 0;
1188 cmd_request->bounce_sgl = NULL;
1189 cmd_request->cmd = scmnd;
1191 scmnd->host_scribble = (unsigned char *)cmd_request;
1193 request = &cmd_request->request;
1194 vm_srb = &request->vstor_packet.vm_srb;
1198 switch (scmnd->sc_data_direction) {
1200 vm_srb->data_in = WRITE_TYPE;
1202 case DMA_FROM_DEVICE:
1203 vm_srb->data_in = READ_TYPE;
1206 vm_srb->data_in = UNKNOWN_TYPE;
1210 request->on_io_completion = storvsc_command_completion;
1211 request->context = cmd_request;/* scmnd; */
1213 vm_srb->port_number = host_dev->port;
1214 vm_srb->path_id = scmnd->device->channel;
1215 vm_srb->target_id = scmnd->device->id;
1216 vm_srb->lun = scmnd->device->lun;
1218 vm_srb->cdb_length = scmnd->cmd_len;
1220 memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
1222 request->sense_buffer = scmnd->sense_buffer;
1225 request->data_buffer.len = scsi_bufflen(scmnd);
1226 if (scsi_sg_count(scmnd)) {
1227 sgl = (struct scatterlist *)scsi_sglist(scmnd);
1228 sg_count = scsi_sg_count(scmnd);
1230 /* check if we need to bounce the sgl */
1231 if (do_bounce_buffer(sgl, scsi_sg_count(scmnd)) != -1) {
1232 cmd_request->bounce_sgl =
1233 create_bounce_buffer(sgl, scsi_sg_count(scmnd),
1234 scsi_bufflen(scmnd));
1235 if (!cmd_request->bounce_sgl) {
1236 scmnd->scsi_done = NULL;
1237 scmnd->host_scribble = NULL;
1238 kmem_cache_free(host_dev->request_pool,
1241 return SCSI_MLQUEUE_HOST_BUSY;
1244 cmd_request->bounce_sgl_count =
1245 ALIGN(scsi_bufflen(scmnd), PAGE_SIZE) >>
1248 if (vm_srb->data_in == WRITE_TYPE)
1249 copy_to_bounce_buffer(sgl,
1250 cmd_request->bounce_sgl,
1251 scsi_sg_count(scmnd));
1253 sgl = cmd_request->bounce_sgl;
1254 sg_count = cmd_request->bounce_sgl_count;
1257 request->data_buffer.offset = sgl[0].offset;
1259 for (i = 0; i < sg_count; i++)
1260 request->data_buffer.pfn_array[i] =
1261 page_to_pfn(sg_page((&sgl[i])));
1263 } else if (scsi_sglist(scmnd)) {
1264 request->data_buffer.offset =
1265 virt_to_phys(scsi_sglist(scmnd)) & (PAGE_SIZE-1);
1266 request->data_buffer.pfn_array[0] =
1267 virt_to_phys(scsi_sglist(scmnd)) >> PAGE_SHIFT;
1271 /* Invokes the vsc to start an IO */
1272 ret = storvsc_do_io(dev, &cmd_request->request);
1274 if (ret == -EAGAIN) {
1277 if (cmd_request->bounce_sgl_count)
1278 destroy_bounce_buffer(cmd_request->bounce_sgl,
1279 cmd_request->bounce_sgl_count);
1281 kmem_cache_free(host_dev->request_pool, cmd_request);
1283 scmnd->scsi_done = NULL;
1284 scmnd->host_scribble = NULL;
1286 ret = SCSI_MLQUEUE_DEVICE_BUSY;
1292 static DEF_SCSI_QCMD(storvsc_queuecommand)
1296 static struct scsi_host_template scsi_driver = {
1297 .module = THIS_MODULE,
1298 .name = "storvsc_host_t",
1299 .bios_param = storvsc_get_chs,
1300 .queuecommand = storvsc_queuecommand,
1301 .eh_host_reset_handler = storvsc_host_reset_handler,
1302 .slave_alloc = storvsc_device_alloc,
1303 .slave_configure = storvsc_device_configure,
1305 /* 64 max_queue * 1 target */
1306 .can_queue = STORVSC_MAX_IO_REQUESTS*STORVSC_MAX_TARGETS,
1308 /* no use setting to 0 since ll_blk_rw reset it to 1 */
1310 .sg_tablesize = MAX_MULTIPAGE_BUFFER_COUNT,
1312 * ENABLE_CLUSTERING allows mutiple physically contig bio_vecs to merge
1313 * into 1 sg element. If set, we must limit the max_segment_size to
1314 * PAGE_SIZE, otherwise we may get 1 sg element that represents
1317 /* physically contig pfns (ie sg[x].length > PAGE_SIZE). */
1318 .use_clustering = ENABLE_CLUSTERING,
1319 /* Make sure we dont get a sg segment crosses a page boundary */
1320 .dma_boundary = PAGE_SIZE-1,
1328 static const struct hv_vmbus_device_id id_table[] = {
1330 { VMBUS_DEVICE(0xd9, 0x63, 0x61, 0xba, 0xa1, 0x04, 0x29, 0x4d,
1331 0xb6, 0x05, 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f)
1332 .driver_data = SCSI_GUID },
1334 { VMBUS_DEVICE(0x32, 0x26, 0x41, 0x32, 0xcb, 0x86, 0xa2, 0x44,
1335 0x9b, 0x5c, 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5)
1336 .driver_data = IDE_GUID },
1340 MODULE_DEVICE_TABLE(vmbus, id_table);
1344 * storvsc_probe - Add a new device for this driver
1347 static int storvsc_probe(struct hv_device *device,
1348 const struct hv_vmbus_device_id *dev_id)
1351 struct Scsi_Host *host;
1352 struct hv_host_device *host_dev;
1353 bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false);
1356 struct storvsc_device *stor_device;
1358 host = scsi_host_alloc(&scsi_driver,
1359 sizeof(struct hv_host_device));
1363 host_dev = (struct hv_host_device *)host->hostdata;
1364 memset(host_dev, 0, sizeof(struct hv_host_device));
1366 host_dev->port = host->host_no;
1367 host_dev->dev = device;
1369 host_dev->request_pool =
1370 kmem_cache_create(dev_name(&device->device),
1371 sizeof(struct storvsc_cmd_request), 0,
1372 SLAB_HWCACHE_ALIGN, NULL);
1374 if (!host_dev->request_pool) {
1375 scsi_host_put(host);
1379 stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
1381 kmem_cache_destroy(host_dev->request_pool);
1382 scsi_host_put(host);
1386 stor_device->destroy = false;
1387 init_waitqueue_head(&stor_device->waiting_to_drain);
1388 stor_device->device = device;
1389 stor_device->host = host;
1390 hv_set_drvdata(device, stor_device);
1392 stor_device->port_number = host->host_no;
1393 ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size);
1395 kmem_cache_destroy(host_dev->request_pool);
1396 scsi_host_put(host);
1402 storvsc_get_ide_info(device, &target, &path);
1404 host_dev->path = stor_device->path_id;
1405 host_dev->target = stor_device->target_id;
1407 /* max # of devices per target */
1408 host->max_lun = STORVSC_MAX_LUNS_PER_TARGET;
1409 /* max # of targets per channel */
1410 host->max_id = STORVSC_MAX_TARGETS;
1411 /* max # of channels */
1412 host->max_channel = STORVSC_MAX_CHANNELS - 1;
1413 /* max cmd length */
1414 host->max_cmd_len = STORVSC_MAX_CMD_LEN;
1416 /* Register the HBA and start the scsi bus scan */
1417 ret = scsi_add_host(host, &device->device);
1422 scsi_scan_host(host);
1425 ret = scsi_add_device(host, 0, target, 0);
1427 scsi_remove_host(host);
1433 storvsc_dev_remove(device);
1434 kmem_cache_destroy(host_dev->request_pool);
1435 scsi_host_put(host);
1439 /* The one and only one */
1441 static struct hv_driver storvsc_drv = {
1443 .id_table = id_table,
1444 .probe = storvsc_probe,
1445 .remove = storvsc_remove,
1448 static int __init storvsc_drv_init(void)
1450 u32 max_outstanding_req_per_channel;
1453 * Divide the ring buffer data size (which is 1 page less
1454 * than the ring buffer size since that page is reserved for
1455 * the ring buffer indices) by the max request size (which is
1456 * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
1458 max_outstanding_req_per_channel =
1459 ((storvsc_ringbuffer_size - PAGE_SIZE) /
1460 ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
1461 sizeof(struct vstor_packet) + sizeof(u64),
1464 if (max_outstanding_req_per_channel <
1465 STORVSC_MAX_IO_REQUESTS)
1468 return vmbus_driver_register(&storvsc_drv);
1471 static void __exit storvsc_drv_exit(void)
1473 vmbus_driver_unregister(&storvsc_drv);
1476 MODULE_LICENSE("GPL");
1477 MODULE_VERSION(HV_DRV_VERSION);
1478 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
1479 module_init(storvsc_drv_init);
1480 module_exit(storvsc_drv_exit);