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 #define STORVSC_MIN_BUF_NR 64
47 #define STORVSC_RING_BUFFER_SIZE (20*PAGE_SIZE)
48 static int storvsc_ringbuffer_size = STORVSC_RING_BUFFER_SIZE;
50 module_param(storvsc_ringbuffer_size, int, S_IRUGO);
51 MODULE_PARM_DESC(storvsc_ringbuffer_size, "Ring buffer size (bytes)");
53 /* to alert the user that structure sizes may be mismatched even though the */
54 /* protocol versions match. */
57 #define REVISION_STRING(REVISION_) #REVISION_
58 #define FILL_VMSTOR_REVISION(RESULT_LVALUE_) \
60 char *revision_string \
61 = REVISION_STRING($Rev : 6 $) + 6; \
63 while (*revision_string >= '0' \
64 && *revision_string <= '9') { \
65 RESULT_LVALUE_ *= 10; \
66 RESULT_LVALUE_ += *revision_string - '0'; \
71 /* Major/minor macros. Minor version is in LSB, meaning that earlier flat */
72 /* version numbers will be interpreted as "0.x" (i.e., 1 becomes 0.1). */
73 #define VMSTOR_PROTOCOL_MAJOR(VERSION_) (((VERSION_) >> 8) & 0xff)
74 #define VMSTOR_PROTOCOL_MINOR(VERSION_) (((VERSION_)) & 0xff)
75 #define VMSTOR_PROTOCOL_VERSION(MAJOR_, MINOR_) ((((MAJOR_) & 0xff) << 8) | \
77 #define VMSTOR_INVALID_PROTOCOL_VERSION (-1)
79 /* Version history: */
81 /* V1 RC < 2008/1/31 1.0 */
82 /* V1 RC > 2008/1/31 2.0 */
83 #define VMSTOR_PROTOCOL_VERSION_CURRENT VMSTOR_PROTOCOL_VERSION(4, 2)
88 /* This will get replaced with the max transfer length that is possible on */
89 /* the host adapter. */
90 /* The max transfer length will be published when we offer a vmbus channel. */
91 #define MAX_TRANSFER_LENGTH 0x40000
92 #define DEFAULT_PACKET_SIZE (sizeof(struct vmdata_gpa_direct) + \
93 sizeof(struct vstor_packet) + \
94 sizesizeof(u64) * (MAX_TRANSFER_LENGTH / PAGE_SIZE)))
97 /* Packet structure describing virtual storage requests. */
98 enum vstor_packet_operation {
99 VSTOR_OPERATION_COMPLETE_IO = 1,
100 VSTOR_OPERATION_REMOVE_DEVICE = 2,
101 VSTOR_OPERATION_EXECUTE_SRB = 3,
102 VSTOR_OPERATION_RESET_LUN = 4,
103 VSTOR_OPERATION_RESET_ADAPTER = 5,
104 VSTOR_OPERATION_RESET_BUS = 6,
105 VSTOR_OPERATION_BEGIN_INITIALIZATION = 7,
106 VSTOR_OPERATION_END_INITIALIZATION = 8,
107 VSTOR_OPERATION_QUERY_PROTOCOL_VERSION = 9,
108 VSTOR_OPERATION_QUERY_PROPERTIES = 10,
109 VSTOR_OPERATION_ENUMERATE_BUS = 11,
110 VSTOR_OPERATION_MAXIMUM = 11
114 * Platform neutral description of a scsi request -
115 * this remains the same across the write regardless of 32/64 bit
116 * note: it's patterned off the SCSI_PASS_THROUGH structure
118 #define CDB16GENERIC_LENGTH 0x10
120 #ifndef SENSE_BUFFER_SIZE
121 #define SENSE_BUFFER_SIZE 0x12
124 #define MAX_DATA_BUF_LEN_WITH_PADDING 0x14
126 struct vmscsi_request {
127 unsigned short length;
128 unsigned char srb_status;
129 unsigned char scsi_status;
131 unsigned char port_number;
132 unsigned char path_id;
133 unsigned char target_id;
136 unsigned char cdb_length;
137 unsigned char sense_info_length;
138 unsigned char data_in;
139 unsigned char reserved;
141 unsigned int data_transfer_length;
144 unsigned char cdb[CDB16GENERIC_LENGTH];
145 unsigned char sense_data[SENSE_BUFFER_SIZE];
146 unsigned char reserved_array[MAX_DATA_BUF_LEN_WITH_PADDING];
148 } __attribute((packed));
152 * This structure is sent during the intialization phase to get the different
153 * properties of the channel.
155 struct vmstorage_channel_properties {
156 unsigned short protocol_version;
157 unsigned char path_id;
158 unsigned char target_id;
160 /* Note: port number is only really known on the client side */
161 unsigned int port_number;
163 unsigned int max_transfer_bytes;
165 /* This id is unique for each channel and will correspond with */
166 /* vendor specific data in the inquirydata */
167 unsigned long long unique_id;
170 /* This structure is sent during the storage protocol negotiations. */
171 struct vmstorage_protocol_version {
172 /* Major (MSW) and minor (LSW) version numbers. */
173 unsigned short major_minor;
176 * Revision number is auto-incremented whenever this file is changed
177 * (See FILL_VMSTOR_REVISION macro above). Mismatch does not
178 * definitely indicate incompatibility--but it does indicate mismatched
181 unsigned short revision;
184 /* Channel Property Flags */
185 #define STORAGE_CHANNEL_REMOVABLE_FLAG 0x1
186 #define STORAGE_CHANNEL_EMULATED_IDE_FLAG 0x2
188 struct vstor_packet {
189 /* Requested operation type */
190 enum vstor_packet_operation operation;
192 /* Flags - see below for values */
195 /* Status of the request returned from the server side. */
198 /* Data payload area */
201 * Structure used to forward SCSI commands from the
202 * client to the server.
204 struct vmscsi_request vm_srb;
206 /* Structure used to query channel properties. */
207 struct vmstorage_channel_properties storage_channel_properties;
209 /* Used during version negotiations. */
210 struct vmstorage_protocol_version version;
216 * This flag indicates that the server should send back a completion for this
219 #define REQUEST_COMPLETION_FLAG 0x1
221 /* This is the set of flags that the vsc can set in any packets it sends */
222 #define VSC_LEGAL_FLAGS (REQUEST_COMPLETION_FLAG)
227 #define STORVSC_MAX_IO_REQUESTS 128
230 * In Hyper-V, each port/path/target maps to 1 scsi host adapter. In
231 * reality, the path/target is not used (ie always set to 0) so our
232 * scsi host adapter essentially has 1 bus with 1 target that contains
235 #define STORVSC_MAX_LUNS_PER_TARGET 64
236 #define STORVSC_MAX_TARGETS 1
237 #define STORVSC_MAX_CHANNELS 1
238 #define STORVSC_MAX_CMD_LEN 16
240 /* Matches Windows-end */
241 enum storvsc_request_type {
248 struct hv_storvsc_request {
249 struct hv_device *device;
251 /* Synchronize the request/response if needed */
252 struct completion wait_event;
254 unsigned char *sense_buffer;
256 void (*on_io_completion)(struct hv_storvsc_request *request);
257 struct hv_multipage_buffer data_buffer;
259 struct vstor_packet vstor_packet;
263 /* A storvsc device is a device object that contains a vmbus channel */
264 struct storvsc_device {
265 struct hv_device *device;
269 atomic_t num_outstanding_req;
270 struct Scsi_Host *host;
272 wait_queue_head_t waiting_to_drain;
275 * Each unique Port/Path/Target represents 1 channel ie scsi
276 * controller. In reality, the pathid, targetid is always 0
277 * and the port is set by us
279 unsigned int port_number;
280 unsigned char path_id;
281 unsigned char target_id;
283 /* Used for vsc/vsp channel reset process */
284 struct hv_storvsc_request init_request;
285 struct hv_storvsc_request reset_request;
288 struct hv_host_device {
289 struct hv_device *dev;
290 struct kmem_cache *request_pool;
291 mempool_t *request_mempool;
294 unsigned char target;
297 struct storvsc_cmd_request {
298 struct list_head entry;
299 struct scsi_cmnd *cmd;
301 unsigned int bounce_sgl_count;
302 struct scatterlist *bounce_sgl;
304 struct hv_storvsc_request request;
307 struct storvsc_scan_work {
308 struct work_struct work;
309 struct Scsi_Host *host;
313 static void storvsc_bus_scan(struct work_struct *work)
315 struct storvsc_scan_work *wrk;
318 wrk = container_of(work, struct storvsc_scan_work, work);
319 for (id = 0; id < wrk->host->max_id; ++id) {
320 if (wrk->host->reverse_ordering)
321 order_id = wrk->host->max_id - id - 1;
325 scsi_scan_target(&wrk->host->shost_gendev, 0,
326 order_id, SCAN_WILD_CARD, 1);
331 static inline struct storvsc_device *get_out_stor_device(
332 struct hv_device *device)
334 struct storvsc_device *stor_device;
336 stor_device = hv_get_drvdata(device);
338 if (stor_device && stor_device->destroy)
345 static inline void storvsc_wait_to_drain(struct storvsc_device *dev)
347 dev->drain_notify = true;
348 wait_event(dev->waiting_to_drain,
349 atomic_read(&dev->num_outstanding_req) == 0);
350 dev->drain_notify = false;
353 static inline struct storvsc_device *get_in_stor_device(
354 struct hv_device *device)
356 struct storvsc_device *stor_device;
358 stor_device = hv_get_drvdata(device);
364 * If the device is being destroyed; allow incoming
365 * traffic only to cleanup outstanding requests.
368 if (stor_device->destroy &&
369 (atomic_read(&stor_device->num_outstanding_req) == 0))
377 static int storvsc_channel_init(struct hv_device *device)
379 struct storvsc_device *stor_device;
380 struct hv_storvsc_request *request;
381 struct vstor_packet *vstor_packet;
384 stor_device = get_out_stor_device(device);
388 request = &stor_device->init_request;
389 vstor_packet = &request->vstor_packet;
392 * Now, initiate the vsc/vsp initialization protocol on the open
395 memset(request, 0, sizeof(struct hv_storvsc_request));
396 init_completion(&request->wait_event);
397 vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION;
398 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
400 ret = vmbus_sendpacket(device->channel, vstor_packet,
401 sizeof(struct vstor_packet),
402 (unsigned long)request,
404 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
408 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
414 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
415 vstor_packet->status != 0)
419 /* reuse the packet for version range supported */
420 memset(vstor_packet, 0, sizeof(struct vstor_packet));
421 vstor_packet->operation = VSTOR_OPERATION_QUERY_PROTOCOL_VERSION;
422 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
424 vstor_packet->version.major_minor = VMSTOR_PROTOCOL_VERSION_CURRENT;
425 FILL_VMSTOR_REVISION(vstor_packet->version.revision);
427 ret = vmbus_sendpacket(device->channel, vstor_packet,
428 sizeof(struct vstor_packet),
429 (unsigned long)request,
431 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)
446 memset(vstor_packet, 0, sizeof(struct vstor_packet));
447 vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES;
448 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
449 vstor_packet->storage_channel_properties.port_number =
450 stor_device->port_number;
452 ret = vmbus_sendpacket(device->channel, vstor_packet,
453 sizeof(struct vstor_packet),
454 (unsigned long)request,
456 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
461 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
467 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
468 vstor_packet->status != 0)
471 stor_device->path_id = vstor_packet->storage_channel_properties.path_id;
472 stor_device->target_id
473 = vstor_packet->storage_channel_properties.target_id;
475 memset(vstor_packet, 0, sizeof(struct vstor_packet));
476 vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION;
477 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
479 ret = vmbus_sendpacket(device->channel, vstor_packet,
480 sizeof(struct vstor_packet),
481 (unsigned long)request,
483 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
488 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
494 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
495 vstor_packet->status != 0)
503 static void storvsc_on_io_completion(struct hv_device *device,
504 struct vstor_packet *vstor_packet,
505 struct hv_storvsc_request *request)
507 struct storvsc_device *stor_device;
508 struct vstor_packet *stor_pkt;
510 stor_device = hv_get_drvdata(device);
511 stor_pkt = &request->vstor_packet;
514 * The current SCSI handling on the host side does
515 * not correctly handle:
516 * INQUIRY command with page code parameter set to 0x80
517 * MODE_SENSE command with cmd[2] == 0x1c
519 * Setup srb and scsi status so this won't be fatal.
520 * We do this so we can distinguish truly fatal failues
521 * (srb status == 0x4) and off-line the device in that case.
524 if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) ||
525 (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) {
526 vstor_packet->vm_srb.scsi_status = 0;
527 vstor_packet->vm_srb.srb_status = 0x1;
531 /* Copy over the status...etc */
532 stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status;
533 stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status;
534 stor_pkt->vm_srb.sense_info_length =
535 vstor_packet->vm_srb.sense_info_length;
537 if (vstor_packet->vm_srb.scsi_status != 0 ||
538 vstor_packet->vm_srb.srb_status != 1){
539 dev_warn(&device->device,
540 "cmd 0x%x scsi status 0x%x srb status 0x%x\n",
541 stor_pkt->vm_srb.cdb[0],
542 vstor_packet->vm_srb.scsi_status,
543 vstor_packet->vm_srb.srb_status);
546 if ((vstor_packet->vm_srb.scsi_status & 0xFF) == 0x02) {
547 /* CHECK_CONDITION */
548 if (vstor_packet->vm_srb.srb_status & 0x80) {
549 /* autosense data available */
550 dev_warn(&device->device,
551 "stor pkt %p autosense data valid - len %d\n",
553 vstor_packet->vm_srb.sense_info_length);
555 memcpy(request->sense_buffer,
556 vstor_packet->vm_srb.sense_data,
557 vstor_packet->vm_srb.sense_info_length);
562 stor_pkt->vm_srb.data_transfer_length =
563 vstor_packet->vm_srb.data_transfer_length;
565 request->on_io_completion(request);
567 if (atomic_dec_and_test(&stor_device->num_outstanding_req) &&
568 stor_device->drain_notify)
569 wake_up(&stor_device->waiting_to_drain);
574 static void storvsc_on_receive(struct hv_device *device,
575 struct vstor_packet *vstor_packet,
576 struct hv_storvsc_request *request)
578 struct storvsc_scan_work *work;
579 struct storvsc_device *stor_device;
581 switch (vstor_packet->operation) {
582 case VSTOR_OPERATION_COMPLETE_IO:
583 storvsc_on_io_completion(device, vstor_packet, request);
586 case VSTOR_OPERATION_REMOVE_DEVICE:
587 case VSTOR_OPERATION_ENUMERATE_BUS:
588 stor_device = get_in_stor_device(device);
589 work = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC);
593 INIT_WORK(&work->work, storvsc_bus_scan);
594 work->host = stor_device->host;
595 schedule_work(&work->work);
603 static void storvsc_on_channel_callback(void *context)
605 struct hv_device *device = (struct hv_device *)context;
606 struct storvsc_device *stor_device;
609 unsigned char packet[ALIGN(sizeof(struct vstor_packet), 8)];
610 struct hv_storvsc_request *request;
614 stor_device = get_in_stor_device(device);
619 ret = vmbus_recvpacket(device->channel, packet,
620 ALIGN(sizeof(struct vstor_packet), 8),
621 &bytes_recvd, &request_id);
622 if (ret == 0 && bytes_recvd > 0) {
624 request = (struct hv_storvsc_request *)
625 (unsigned long)request_id;
627 if ((request == &stor_device->init_request) ||
628 (request == &stor_device->reset_request)) {
630 memcpy(&request->vstor_packet, packet,
631 sizeof(struct vstor_packet));
632 complete(&request->wait_event);
634 storvsc_on_receive(device,
635 (struct vstor_packet *)packet,
646 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size)
648 struct vmstorage_channel_properties props;
651 memset(&props, 0, sizeof(struct vmstorage_channel_properties));
653 /* Open the channel */
654 ret = vmbus_open(device->channel,
658 sizeof(struct vmstorage_channel_properties),
659 storvsc_on_channel_callback, device);
664 ret = storvsc_channel_init(device);
669 static int storvsc_dev_remove(struct hv_device *device)
671 struct storvsc_device *stor_device;
674 stor_device = hv_get_drvdata(device);
676 spin_lock_irqsave(&device->channel->inbound_lock, flags);
677 stor_device->destroy = true;
678 spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
681 * At this point, all outbound traffic should be disable. We
682 * only allow inbound traffic (responses) to proceed so that
683 * outstanding requests can be completed.
686 storvsc_wait_to_drain(stor_device);
689 * Since we have already drained, we don't need to busy wait
690 * as was done in final_release_stor_device()
691 * Note that we cannot set the ext pointer to NULL until
692 * we have drained - to drain the outgoing packets, we need to
693 * allow incoming packets.
695 spin_lock_irqsave(&device->channel->inbound_lock, flags);
696 hv_set_drvdata(device, NULL);
697 spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
699 /* Close the channel */
700 vmbus_close(device->channel);
706 static int storvsc_do_io(struct hv_device *device,
707 struct hv_storvsc_request *request)
709 struct storvsc_device *stor_device;
710 struct vstor_packet *vstor_packet;
713 vstor_packet = &request->vstor_packet;
714 stor_device = get_out_stor_device(device);
720 request->device = device;
723 vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
725 vstor_packet->vm_srb.length = sizeof(struct vmscsi_request);
728 vstor_packet->vm_srb.sense_info_length = SENSE_BUFFER_SIZE;
731 vstor_packet->vm_srb.data_transfer_length =
732 request->data_buffer.len;
734 vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
736 if (request->data_buffer.len) {
737 ret = vmbus_sendpacket_multipagebuffer(device->channel,
738 &request->data_buffer,
740 sizeof(struct vstor_packet),
741 (unsigned long)request);
743 ret = vmbus_sendpacket(device->channel, vstor_packet,
744 sizeof(struct vstor_packet),
745 (unsigned long)request,
747 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
753 atomic_inc(&stor_device->num_outstanding_req);
758 static void storvsc_get_ide_info(struct hv_device *dev, int *target, int *path)
761 dev->dev_instance.b[5] << 8 | dev->dev_instance.b[4];
764 dev->dev_instance.b[3] << 24 |
765 dev->dev_instance.b[2] << 16 |
766 dev->dev_instance.b[1] << 8 | dev->dev_instance.b[0];
770 static int storvsc_device_alloc(struct scsi_device *sdevice)
773 * This enables luns to be located sparsely. Otherwise, we may not
776 sdevice->sdev_bflags |= BLIST_SPARSELUN | BLIST_LARGELUN;
780 static int storvsc_merge_bvec(struct request_queue *q,
781 struct bvec_merge_data *bmd, struct bio_vec *bvec)
783 /* checking done by caller. */
787 static int storvsc_device_configure(struct scsi_device *sdevice)
789 scsi_adjust_queue_depth(sdevice, MSG_SIMPLE_TAG,
790 STORVSC_MAX_IO_REQUESTS);
792 blk_queue_max_segment_size(sdevice->request_queue, PAGE_SIZE);
794 blk_queue_merge_bvec(sdevice->request_queue, storvsc_merge_bvec);
796 blk_queue_bounce_limit(sdevice->request_queue, BLK_BOUNCE_ANY);
801 static void destroy_bounce_buffer(struct scatterlist *sgl,
802 unsigned int sg_count)
805 struct page *page_buf;
807 for (i = 0; i < sg_count; i++) {
808 page_buf = sg_page((&sgl[i]));
809 if (page_buf != NULL)
810 __free_page(page_buf);
816 static int do_bounce_buffer(struct scatterlist *sgl, unsigned int sg_count)
820 /* No need to check */
824 /* We have at least 2 sg entries */
825 for (i = 0; i < sg_count; i++) {
827 /* make sure 1st one does not have hole */
828 if (sgl[i].offset + sgl[i].length != PAGE_SIZE)
830 } else if (i == sg_count - 1) {
831 /* make sure last one does not have hole */
832 if (sgl[i].offset != 0)
835 /* make sure no hole in the middle */
836 if (sgl[i].length != PAGE_SIZE || sgl[i].offset != 0)
843 static struct scatterlist *create_bounce_buffer(struct scatterlist *sgl,
844 unsigned int sg_count,
849 struct scatterlist *bounce_sgl;
850 struct page *page_buf;
852 num_pages = ALIGN(len, PAGE_SIZE) >> PAGE_SHIFT;
854 bounce_sgl = kcalloc(num_pages, sizeof(struct scatterlist), GFP_ATOMIC);
858 for (i = 0; i < num_pages; i++) {
859 page_buf = alloc_page(GFP_ATOMIC);
862 sg_set_page(&bounce_sgl[i], page_buf, 0, 0);
868 destroy_bounce_buffer(bounce_sgl, num_pages);
873 /* Assume the original sgl has enough room */
874 static unsigned int copy_from_bounce_buffer(struct scatterlist *orig_sgl,
875 struct scatterlist *bounce_sgl,
876 unsigned int orig_sgl_count)
880 unsigned long src, dest;
881 unsigned int srclen, destlen, copylen;
882 unsigned int total_copied = 0;
883 unsigned long bounce_addr = 0;
884 unsigned long dest_addr = 0;
887 local_irq_save(flags);
889 for (i = 0; i < orig_sgl_count; i++) {
890 dest_addr = (unsigned long)kmap_atomic(sg_page((&orig_sgl[i])),
891 KM_IRQ0) + orig_sgl[i].offset;
893 destlen = orig_sgl[i].length;
895 if (bounce_addr == 0)
897 (unsigned long)kmap_atomic(sg_page((&bounce_sgl[j])),
901 src = bounce_addr + bounce_sgl[j].offset;
902 srclen = bounce_sgl[j].length - bounce_sgl[j].offset;
904 copylen = min(srclen, destlen);
905 memcpy((void *)dest, (void *)src, copylen);
907 total_copied += copylen;
908 bounce_sgl[j].offset += copylen;
912 if (bounce_sgl[j].offset == bounce_sgl[j].length) {
914 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
917 /* if we need to use another bounce buffer */
918 if (destlen || i != orig_sgl_count - 1)
920 (unsigned long)kmap_atomic(
921 sg_page((&bounce_sgl[j])), KM_IRQ0);
922 } else if (destlen == 0 && i == orig_sgl_count - 1) {
923 /* unmap the last bounce that is < PAGE_SIZE */
924 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
928 kunmap_atomic((void *)(dest_addr - orig_sgl[i].offset),
932 local_irq_restore(flags);
938 /* Assume the bounce_sgl has enough room ie using the create_bounce_buffer() */
939 static unsigned int copy_to_bounce_buffer(struct scatterlist *orig_sgl,
940 struct scatterlist *bounce_sgl,
941 unsigned int orig_sgl_count)
945 unsigned long src, dest;
946 unsigned int srclen, destlen, copylen;
947 unsigned int total_copied = 0;
948 unsigned long bounce_addr = 0;
949 unsigned long src_addr = 0;
952 local_irq_save(flags);
954 for (i = 0; i < orig_sgl_count; i++) {
955 src_addr = (unsigned long)kmap_atomic(sg_page((&orig_sgl[i])),
956 KM_IRQ0) + orig_sgl[i].offset;
958 srclen = orig_sgl[i].length;
960 if (bounce_addr == 0)
962 (unsigned long)kmap_atomic(sg_page((&bounce_sgl[j])),
966 /* assume bounce offset always == 0 */
967 dest = bounce_addr + bounce_sgl[j].length;
968 destlen = PAGE_SIZE - bounce_sgl[j].length;
970 copylen = min(srclen, destlen);
971 memcpy((void *)dest, (void *)src, copylen);
973 total_copied += copylen;
974 bounce_sgl[j].length += copylen;
978 if (bounce_sgl[j].length == PAGE_SIZE) {
979 /* full..move to next entry */
980 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
983 /* if we need to use another bounce buffer */
984 if (srclen || i != orig_sgl_count - 1)
986 (unsigned long)kmap_atomic(
987 sg_page((&bounce_sgl[j])), KM_IRQ0);
989 } else if (srclen == 0 && i == orig_sgl_count - 1) {
990 /* unmap the last bounce that is < PAGE_SIZE */
991 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
995 kunmap_atomic((void *)(src_addr - orig_sgl[i].offset), KM_IRQ0);
998 local_irq_restore(flags);
1000 return total_copied;
1004 static int storvsc_remove(struct hv_device *dev)
1006 struct storvsc_device *stor_device = hv_get_drvdata(dev);
1007 struct Scsi_Host *host = stor_device->host;
1008 struct hv_host_device *host_dev = shost_priv(host);
1010 scsi_remove_host(host);
1012 scsi_host_put(host);
1014 storvsc_dev_remove(dev);
1015 if (host_dev->request_pool) {
1016 mempool_destroy(host_dev->request_mempool);
1017 kmem_cache_destroy(host_dev->request_pool);
1018 host_dev->request_pool = NULL;
1019 host_dev->request_mempool = NULL;
1025 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev,
1026 sector_t capacity, int *info)
1028 sector_t nsect = capacity;
1029 sector_t cylinders = nsect;
1030 int heads, sectors_pt;
1033 * We are making up these values; let us keep it simple.
1036 sectors_pt = 0x3f; /* Sectors per track */
1037 sector_div(cylinders, heads * sectors_pt);
1038 if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
1042 info[1] = sectors_pt;
1043 info[2] = (int)cylinders;
1048 static int storvsc_host_reset(struct hv_device *device)
1050 struct storvsc_device *stor_device;
1051 struct hv_storvsc_request *request;
1052 struct vstor_packet *vstor_packet;
1056 stor_device = get_out_stor_device(device);
1060 request = &stor_device->reset_request;
1061 vstor_packet = &request->vstor_packet;
1063 init_completion(&request->wait_event);
1065 vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
1066 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
1067 vstor_packet->vm_srb.path_id = stor_device->path_id;
1069 ret = vmbus_sendpacket(device->channel, vstor_packet,
1070 sizeof(struct vstor_packet),
1071 (unsigned long)&stor_device->reset_request,
1073 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1077 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
1079 return TIMEOUT_ERROR;
1083 * At this point, all outstanding requests in the adapter
1084 * should have been flushed out and return to us
1092 * storvsc_host_reset_handler - Reset the scsi HBA
1094 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
1096 struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
1097 struct hv_device *dev = host_dev->dev;
1099 return storvsc_host_reset(dev);
1104 * storvsc_command_completion - Command completion processing
1106 static void storvsc_command_completion(struct hv_storvsc_request *request)
1108 struct storvsc_cmd_request *cmd_request =
1109 (struct storvsc_cmd_request *)request->context;
1110 struct scsi_cmnd *scmnd = cmd_request->cmd;
1111 struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
1112 void (*scsi_done_fn)(struct scsi_cmnd *);
1113 struct scsi_sense_hdr sense_hdr;
1114 struct vmscsi_request *vm_srb;
1116 vm_srb = &request->vstor_packet.vm_srb;
1117 if (cmd_request->bounce_sgl_count) {
1118 if (vm_srb->data_in == READ_TYPE) {
1119 copy_from_bounce_buffer(scsi_sglist(scmnd),
1120 cmd_request->bounce_sgl,
1121 scsi_sg_count(scmnd));
1122 destroy_bounce_buffer(cmd_request->bounce_sgl,
1123 cmd_request->bounce_sgl_count);
1128 * If there is an error; offline the device since all
1129 * error recovery strategies would have already been
1130 * deployed on the host side.
1132 if (vm_srb->srb_status == 0x4)
1133 scmnd->result = DID_TARGET_FAILURE << 16;
1135 scmnd->result = vm_srb->scsi_status;
1137 if (scmnd->result) {
1138 if (scsi_normalize_sense(scmnd->sense_buffer,
1139 SCSI_SENSE_BUFFERSIZE, &sense_hdr))
1140 scsi_print_sense_hdr("storvsc", &sense_hdr);
1143 scsi_set_resid(scmnd,
1144 request->data_buffer.len -
1145 vm_srb->data_transfer_length);
1147 scsi_done_fn = scmnd->scsi_done;
1149 scmnd->host_scribble = NULL;
1150 scmnd->scsi_done = NULL;
1152 scsi_done_fn(scmnd);
1154 mempool_free(cmd_request, host_dev->request_mempool);
1157 static bool storvsc_check_scsi_cmd(struct scsi_cmnd *scmnd)
1159 bool allowed = true;
1160 u8 scsi_op = scmnd->cmnd[0];
1163 /* smartd sends this command, which will offline the device */
1165 scmnd->result = ILLEGAL_REQUEST << 16;
1175 * storvsc_queuecommand - Initiate command processing
1177 static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
1180 struct hv_host_device *host_dev = shost_priv(host);
1181 struct hv_device *dev = host_dev->dev;
1182 struct hv_storvsc_request *request;
1183 struct storvsc_cmd_request *cmd_request;
1184 unsigned int request_size = 0;
1186 struct scatterlist *sgl;
1187 unsigned int sg_count = 0;
1188 struct vmscsi_request *vm_srb;
1190 if (storvsc_check_scsi_cmd(scmnd) == false) {
1191 scmnd->scsi_done(scmnd);
1195 /* If retrying, no need to prep the cmd */
1196 if (scmnd->host_scribble) {
1199 (struct storvsc_cmd_request *)scmnd->host_scribble;
1204 request_size = sizeof(struct storvsc_cmd_request);
1206 cmd_request = mempool_alloc(host_dev->request_mempool,
1209 return SCSI_MLQUEUE_DEVICE_BUSY;
1211 memset(cmd_request, 0, sizeof(struct storvsc_cmd_request));
1213 /* Setup the cmd request */
1214 cmd_request->bounce_sgl_count = 0;
1215 cmd_request->bounce_sgl = NULL;
1216 cmd_request->cmd = scmnd;
1218 scmnd->host_scribble = (unsigned char *)cmd_request;
1220 request = &cmd_request->request;
1221 vm_srb = &request->vstor_packet.vm_srb;
1225 switch (scmnd->sc_data_direction) {
1227 vm_srb->data_in = WRITE_TYPE;
1229 case DMA_FROM_DEVICE:
1230 vm_srb->data_in = READ_TYPE;
1233 vm_srb->data_in = UNKNOWN_TYPE;
1237 request->on_io_completion = storvsc_command_completion;
1238 request->context = cmd_request;/* scmnd; */
1240 vm_srb->port_number = host_dev->port;
1241 vm_srb->path_id = scmnd->device->channel;
1242 vm_srb->target_id = scmnd->device->id;
1243 vm_srb->lun = scmnd->device->lun;
1245 vm_srb->cdb_length = scmnd->cmd_len;
1247 memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
1249 request->sense_buffer = scmnd->sense_buffer;
1252 request->data_buffer.len = scsi_bufflen(scmnd);
1253 if (scsi_sg_count(scmnd)) {
1254 sgl = (struct scatterlist *)scsi_sglist(scmnd);
1255 sg_count = scsi_sg_count(scmnd);
1257 /* check if we need to bounce the sgl */
1258 if (do_bounce_buffer(sgl, scsi_sg_count(scmnd)) != -1) {
1259 cmd_request->bounce_sgl =
1260 create_bounce_buffer(sgl, scsi_sg_count(scmnd),
1261 scsi_bufflen(scmnd));
1262 if (!cmd_request->bounce_sgl) {
1263 scmnd->host_scribble = NULL;
1264 mempool_free(cmd_request,
1265 host_dev->request_mempool);
1267 return SCSI_MLQUEUE_HOST_BUSY;
1270 cmd_request->bounce_sgl_count =
1271 ALIGN(scsi_bufflen(scmnd), PAGE_SIZE) >>
1274 if (vm_srb->data_in == WRITE_TYPE)
1275 copy_to_bounce_buffer(sgl,
1276 cmd_request->bounce_sgl,
1277 scsi_sg_count(scmnd));
1279 sgl = cmd_request->bounce_sgl;
1280 sg_count = cmd_request->bounce_sgl_count;
1283 request->data_buffer.offset = sgl[0].offset;
1285 for (i = 0; i < sg_count; i++)
1286 request->data_buffer.pfn_array[i] =
1287 page_to_pfn(sg_page((&sgl[i])));
1289 } else if (scsi_sglist(scmnd)) {
1290 request->data_buffer.offset =
1291 virt_to_phys(scsi_sglist(scmnd)) & (PAGE_SIZE-1);
1292 request->data_buffer.pfn_array[0] =
1293 virt_to_phys(scsi_sglist(scmnd)) >> PAGE_SHIFT;
1297 /* Invokes the vsc to start an IO */
1298 ret = storvsc_do_io(dev, &cmd_request->request);
1300 if (ret == -EAGAIN) {
1303 if (cmd_request->bounce_sgl_count)
1304 destroy_bounce_buffer(cmd_request->bounce_sgl,
1305 cmd_request->bounce_sgl_count);
1307 mempool_free(cmd_request, host_dev->request_mempool);
1309 scmnd->host_scribble = NULL;
1311 ret = SCSI_MLQUEUE_DEVICE_BUSY;
1318 static struct scsi_host_template scsi_driver = {
1319 .module = THIS_MODULE,
1320 .name = "storvsc_host_t",
1321 .bios_param = storvsc_get_chs,
1322 .queuecommand = storvsc_queuecommand,
1323 .eh_host_reset_handler = storvsc_host_reset_handler,
1324 .slave_alloc = storvsc_device_alloc,
1325 .slave_configure = storvsc_device_configure,
1327 /* 64 max_queue * 1 target */
1328 .can_queue = STORVSC_MAX_IO_REQUESTS*STORVSC_MAX_TARGETS,
1330 /* no use setting to 0 since ll_blk_rw reset it to 1 */
1332 .sg_tablesize = MAX_MULTIPAGE_BUFFER_COUNT,
1334 * ENABLE_CLUSTERING allows mutiple physically contig bio_vecs to merge
1335 * into 1 sg element. If set, we must limit the max_segment_size to
1336 * PAGE_SIZE, otherwise we may get 1 sg element that represents
1339 /* physically contig pfns (ie sg[x].length > PAGE_SIZE). */
1340 .use_clustering = ENABLE_CLUSTERING,
1341 /* Make sure we dont get a sg segment crosses a page boundary */
1342 .dma_boundary = PAGE_SIZE-1,
1350 static const struct hv_vmbus_device_id id_table[] = {
1352 { VMBUS_DEVICE(0xd9, 0x63, 0x61, 0xba, 0xa1, 0x04, 0x29, 0x4d,
1353 0xb6, 0x05, 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f)
1354 .driver_data = SCSI_GUID },
1356 { VMBUS_DEVICE(0x32, 0x26, 0x41, 0x32, 0xcb, 0x86, 0xa2, 0x44,
1357 0x9b, 0x5c, 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5)
1358 .driver_data = IDE_GUID },
1362 MODULE_DEVICE_TABLE(vmbus, id_table);
1366 * storvsc_probe - Add a new device for this driver
1369 static int storvsc_probe(struct hv_device *device,
1370 const struct hv_vmbus_device_id *dev_id)
1373 int number = STORVSC_MIN_BUF_NR;
1374 struct Scsi_Host *host;
1375 struct hv_host_device *host_dev;
1376 bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false);
1379 struct storvsc_device *stor_device;
1381 host = scsi_host_alloc(&scsi_driver,
1382 sizeof(struct hv_host_device));
1386 host_dev = shost_priv(host);
1387 memset(host_dev, 0, sizeof(struct hv_host_device));
1389 host_dev->port = host->host_no;
1390 host_dev->dev = device;
1392 host_dev->request_pool =
1393 kmem_cache_create(dev_name(&device->device),
1394 sizeof(struct storvsc_cmd_request), 0,
1395 SLAB_HWCACHE_ALIGN, NULL);
1397 if (!host_dev->request_pool) {
1398 scsi_host_put(host);
1402 host_dev->request_mempool = mempool_create(number, mempool_alloc_slab,
1404 host_dev->request_pool);
1406 if (!host_dev->request_mempool) {
1411 stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
1417 stor_device->destroy = false;
1418 init_waitqueue_head(&stor_device->waiting_to_drain);
1419 stor_device->device = device;
1420 stor_device->host = host;
1421 hv_set_drvdata(device, stor_device);
1423 stor_device->port_number = host->host_no;
1424 ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size);
1429 storvsc_get_ide_info(device, &target, &path);
1431 host_dev->path = stor_device->path_id;
1432 host_dev->target = stor_device->target_id;
1434 /* max # of devices per target */
1435 host->max_lun = STORVSC_MAX_LUNS_PER_TARGET;
1436 /* max # of targets per channel */
1437 host->max_id = STORVSC_MAX_TARGETS;
1438 /* max # of channels */
1439 host->max_channel = STORVSC_MAX_CHANNELS - 1;
1440 /* max cmd length */
1441 host->max_cmd_len = STORVSC_MAX_CMD_LEN;
1443 /* Register the HBA and start the scsi bus scan */
1444 ret = scsi_add_host(host, &device->device);
1449 scsi_scan_host(host);
1452 ret = scsi_add_device(host, 0, target, 0);
1454 scsi_remove_host(host);
1461 * Once we have connected with the host, we would need to
1462 * to invoke storvsc_dev_remove() to rollback this state and
1463 * this call also frees up the stor_device; hence the jump around
1466 storvsc_dev_remove(device);
1473 mempool_destroy(host_dev->request_mempool);
1476 kmem_cache_destroy(host_dev->request_pool);
1477 scsi_host_put(host);
1481 /* The one and only one */
1483 static struct hv_driver storvsc_drv = {
1484 .name = KBUILD_MODNAME,
1485 .id_table = id_table,
1486 .probe = storvsc_probe,
1487 .remove = storvsc_remove,
1490 static int __init storvsc_drv_init(void)
1492 u32 max_outstanding_req_per_channel;
1495 * Divide the ring buffer data size (which is 1 page less
1496 * than the ring buffer size since that page is reserved for
1497 * the ring buffer indices) by the max request size (which is
1498 * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
1500 max_outstanding_req_per_channel =
1501 ((storvsc_ringbuffer_size - PAGE_SIZE) /
1502 ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
1503 sizeof(struct vstor_packet) + sizeof(u64),
1506 if (max_outstanding_req_per_channel <
1507 STORVSC_MAX_IO_REQUESTS)
1510 return vmbus_driver_register(&storvsc_drv);
1513 static void __exit storvsc_drv_exit(void)
1515 vmbus_driver_unregister(&storvsc_drv);
1518 MODULE_LICENSE("GPL");
1519 MODULE_VERSION(HV_DRV_VERSION);
1520 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
1521 module_init(storvsc_drv_init);
1522 module_exit(storvsc_drv_exit);