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 <scsi/scsi.h>
35 #include <scsi/scsi_cmnd.h>
36 #include <scsi/scsi_host.h>
37 #include <scsi/scsi_device.h>
38 #include <scsi/scsi_tcq.h>
39 #include <scsi/scsi_eh.h>
40 #include <scsi/scsi_devinfo.h>
41 #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_storvsc_request *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 struct storvsc_device_info {
264 u32 ring_buffer_size;
265 unsigned int port_number;
266 unsigned char path_id;
267 unsigned char target_id;
271 /* A storvsc device is a device object that contains a vmbus channel */
272 struct storvsc_device {
273 struct hv_device *device;
277 atomic_t num_outstanding_req;
279 wait_queue_head_t waiting_to_drain;
282 * Each unique Port/Path/Target represents 1 channel ie scsi
283 * controller. In reality, the pathid, targetid is always 0
284 * and the port is set by us
286 unsigned int port_number;
287 unsigned char path_id;
288 unsigned char target_id;
290 /* Used for vsc/vsp channel reset process */
291 struct hv_storvsc_request init_request;
292 struct hv_storvsc_request reset_request;
295 struct hv_host_device {
296 struct hv_device *dev;
297 struct kmem_cache *request_pool;
300 unsigned char target;
303 struct storvsc_cmd_request {
304 struct list_head entry;
305 struct scsi_cmnd *cmd;
307 unsigned int bounce_sgl_count;
308 struct scatterlist *bounce_sgl;
310 struct hv_storvsc_request request;
313 static inline struct storvsc_device *get_out_stor_device(
314 struct hv_device *device)
316 struct storvsc_device *stor_device;
318 stor_device = (struct storvsc_device *)device->ext;
320 if (stor_device && stor_device->destroy)
327 static inline void storvsc_wait_to_drain(struct storvsc_device *dev)
329 dev->drain_notify = true;
330 wait_event(dev->waiting_to_drain,
331 atomic_read(&dev->num_outstanding_req) == 0);
332 dev->drain_notify = false;
335 static inline struct storvsc_device *get_in_stor_device(
336 struct hv_device *device)
338 struct storvsc_device *stor_device;
340 stor_device = (struct storvsc_device *)device->ext;
346 * If the device is being destroyed; allow incoming
347 * traffic only to cleanup outstanding requests.
350 if (stor_device->destroy &&
351 (atomic_read(&stor_device->num_outstanding_req) == 0))
359 static int storvsc_channel_init(struct hv_device *device)
361 struct storvsc_device *stor_device;
362 struct hv_storvsc_request *request;
363 struct vstor_packet *vstor_packet;
366 stor_device = get_out_stor_device(device);
370 request = &stor_device->init_request;
371 vstor_packet = &request->vstor_packet;
374 * Now, initiate the vsc/vsp initialization protocol on the open
377 memset(request, 0, sizeof(struct hv_storvsc_request));
378 init_completion(&request->wait_event);
379 vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION;
380 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
382 ret = vmbus_sendpacket(device->channel, vstor_packet,
383 sizeof(struct vstor_packet),
384 (unsigned long)request,
386 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
390 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
396 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
397 vstor_packet->status != 0)
401 /* reuse the packet for version range supported */
402 memset(vstor_packet, 0, sizeof(struct vstor_packet));
403 vstor_packet->operation = VSTOR_OPERATION_QUERY_PROTOCOL_VERSION;
404 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
406 vstor_packet->version.major_minor = VMSTOR_PROTOCOL_VERSION_CURRENT;
407 FILL_VMSTOR_REVISION(vstor_packet->version.revision);
409 ret = vmbus_sendpacket(device->channel, vstor_packet,
410 sizeof(struct vstor_packet),
411 (unsigned long)request,
413 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
417 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
423 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
424 vstor_packet->status != 0)
428 memset(vstor_packet, 0, sizeof(struct vstor_packet));
429 vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES;
430 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
431 vstor_packet->storage_channel_properties.port_number =
432 stor_device->port_number;
434 ret = vmbus_sendpacket(device->channel, vstor_packet,
435 sizeof(struct vstor_packet),
436 (unsigned long)request,
438 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
443 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
449 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
450 vstor_packet->status != 0)
453 stor_device->path_id = vstor_packet->storage_channel_properties.path_id;
454 stor_device->target_id
455 = vstor_packet->storage_channel_properties.target_id;
457 memset(vstor_packet, 0, sizeof(struct vstor_packet));
458 vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION;
459 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
461 ret = vmbus_sendpacket(device->channel, vstor_packet,
462 sizeof(struct vstor_packet),
463 (unsigned long)request,
465 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
470 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
476 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
477 vstor_packet->status != 0)
485 static void storvsc_on_io_completion(struct hv_device *device,
486 struct vstor_packet *vstor_packet,
487 struct hv_storvsc_request *request)
489 struct storvsc_device *stor_device;
490 struct vstor_packet *stor_pkt;
492 stor_device = (struct storvsc_device *)device->ext;
494 stor_pkt = &request->vstor_packet;
497 * The current SCSI handling on the host side does
498 * not correctly handle:
499 * INQUIRY command with page code parameter set to 0x80
500 * MODE_SENSE command with cmd[2] == 0x1c
502 * Setup srb and scsi status so this won't be fatal.
503 * We do this so we can distinguish truly fatal failues
504 * (srb status == 0x4) and off-line the device in that case.
507 if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) ||
508 (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) {
509 vstor_packet->vm_srb.scsi_status = 0;
510 vstor_packet->vm_srb.srb_status = 0x1;
514 /* Copy over the status...etc */
515 stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status;
516 stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status;
517 stor_pkt->vm_srb.sense_info_length =
518 vstor_packet->vm_srb.sense_info_length;
520 if (vstor_packet->vm_srb.scsi_status != 0 ||
521 vstor_packet->vm_srb.srb_status != 1){
523 "cmd 0x%x scsi status 0x%x srb status 0x%x\n",
524 stor_pkt->vm_srb.cdb[0],
525 vstor_packet->vm_srb.scsi_status,
526 vstor_packet->vm_srb.srb_status);
529 if ((vstor_packet->vm_srb.scsi_status & 0xFF) == 0x02) {
530 /* CHECK_CONDITION */
531 if (vstor_packet->vm_srb.srb_status & 0x80) {
532 /* autosense data available */
533 DPRINT_WARN(STORVSC, "storvsc pkt %p autosense data "
534 "valid - len %d\n", request,
535 vstor_packet->vm_srb.sense_info_length);
537 memcpy(request->sense_buffer,
538 vstor_packet->vm_srb.sense_data,
539 vstor_packet->vm_srb.sense_info_length);
544 stor_pkt->vm_srb.data_transfer_length =
545 vstor_packet->vm_srb.data_transfer_length;
547 request->on_io_completion(request);
549 if (atomic_dec_and_test(&stor_device->num_outstanding_req) &&
550 stor_device->drain_notify)
551 wake_up(&stor_device->waiting_to_drain);
556 static void storvsc_on_receive(struct hv_device *device,
557 struct vstor_packet *vstor_packet,
558 struct hv_storvsc_request *request)
560 switch (vstor_packet->operation) {
561 case VSTOR_OPERATION_COMPLETE_IO:
562 storvsc_on_io_completion(device, vstor_packet, request);
564 case VSTOR_OPERATION_REMOVE_DEVICE:
571 static void storvsc_on_channel_callback(void *context)
573 struct hv_device *device = (struct hv_device *)context;
574 struct storvsc_device *stor_device;
577 unsigned char packet[ALIGN(sizeof(struct vstor_packet), 8)];
578 struct hv_storvsc_request *request;
582 stor_device = get_in_stor_device(device);
587 ret = vmbus_recvpacket(device->channel, packet,
588 ALIGN(sizeof(struct vstor_packet), 8),
589 &bytes_recvd, &request_id);
590 if (ret == 0 && bytes_recvd > 0) {
592 request = (struct hv_storvsc_request *)
593 (unsigned long)request_id;
595 if ((request == &stor_device->init_request) ||
596 (request == &stor_device->reset_request)) {
598 memcpy(&request->vstor_packet, packet,
599 sizeof(struct vstor_packet));
600 complete(&request->wait_event);
602 storvsc_on_receive(device,
603 (struct vstor_packet *)packet,
614 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size)
616 struct vmstorage_channel_properties props;
619 memset(&props, 0, sizeof(struct vmstorage_channel_properties));
621 /* Open the channel */
622 ret = vmbus_open(device->channel,
626 sizeof(struct vmstorage_channel_properties),
627 storvsc_on_channel_callback, device);
632 ret = storvsc_channel_init(device);
637 static int storvsc_dev_remove(struct hv_device *device)
639 struct storvsc_device *stor_device;
642 stor_device = (struct storvsc_device *)device->ext;
644 spin_lock_irqsave(&device->channel->inbound_lock, flags);
645 stor_device->destroy = true;
646 spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
649 * At this point, all outbound traffic should be disable. We
650 * only allow inbound traffic (responses) to proceed so that
651 * outstanding requests can be completed.
654 storvsc_wait_to_drain(stor_device);
657 * Since we have already drained, we don't need to busy wait
658 * as was done in final_release_stor_device()
659 * Note that we cannot set the ext pointer to NULL until
660 * we have drained - to drain the outgoing packets, we need to
661 * allow incoming packets.
663 spin_lock_irqsave(&device->channel->inbound_lock, flags);
665 spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
667 /* Close the channel */
668 vmbus_close(device->channel);
674 static int storvsc_do_io(struct hv_device *device,
675 struct hv_storvsc_request *request)
677 struct storvsc_device *stor_device;
678 struct vstor_packet *vstor_packet;
681 vstor_packet = &request->vstor_packet;
682 stor_device = get_out_stor_device(device);
688 request->device = device;
691 vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
693 vstor_packet->vm_srb.length = sizeof(struct vmscsi_request);
696 vstor_packet->vm_srb.sense_info_length = SENSE_BUFFER_SIZE;
699 vstor_packet->vm_srb.data_transfer_length =
700 request->data_buffer.len;
702 vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
704 if (request->data_buffer.len) {
705 ret = vmbus_sendpacket_multipagebuffer(device->channel,
706 &request->data_buffer,
708 sizeof(struct vstor_packet),
709 (unsigned long)request);
711 ret = vmbus_sendpacket(device->channel, vstor_packet,
712 sizeof(struct vstor_packet),
713 (unsigned long)request,
715 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
721 atomic_inc(&stor_device->num_outstanding_req);
726 static void storvsc_get_ide_info(struct hv_device *dev, int *target, int *path)
729 dev->dev_instance.b[5] << 8 | dev->dev_instance.b[4];
732 dev->dev_instance.b[3] << 24 |
733 dev->dev_instance.b[2] << 16 |
734 dev->dev_instance.b[1] << 8 | dev->dev_instance.b[0];
738 static int storvsc_device_alloc(struct scsi_device *sdevice)
741 * This enables luns to be located sparsely. Otherwise, we may not
744 sdevice->sdev_bflags |= BLIST_SPARSELUN | BLIST_LARGELUN;
748 static int storvsc_merge_bvec(struct request_queue *q,
749 struct bvec_merge_data *bmd, struct bio_vec *bvec)
751 /* checking done by caller. */
755 static int storvsc_device_configure(struct scsi_device *sdevice)
757 scsi_adjust_queue_depth(sdevice, MSG_SIMPLE_TAG,
758 STORVSC_MAX_IO_REQUESTS);
760 blk_queue_max_segment_size(sdevice->request_queue, PAGE_SIZE);
762 blk_queue_merge_bvec(sdevice->request_queue, storvsc_merge_bvec);
764 blk_queue_bounce_limit(sdevice->request_queue, BLK_BOUNCE_ANY);
769 static void destroy_bounce_buffer(struct scatterlist *sgl,
770 unsigned int sg_count)
773 struct page *page_buf;
775 for (i = 0; i < sg_count; i++) {
776 page_buf = sg_page((&sgl[i]));
777 if (page_buf != NULL)
778 __free_page(page_buf);
784 static int do_bounce_buffer(struct scatterlist *sgl, unsigned int sg_count)
788 /* No need to check */
792 /* We have at least 2 sg entries */
793 for (i = 0; i < sg_count; i++) {
795 /* make sure 1st one does not have hole */
796 if (sgl[i].offset + sgl[i].length != PAGE_SIZE)
798 } else if (i == sg_count - 1) {
799 /* make sure last one does not have hole */
800 if (sgl[i].offset != 0)
803 /* make sure no hole in the middle */
804 if (sgl[i].length != PAGE_SIZE || sgl[i].offset != 0)
811 static struct scatterlist *create_bounce_buffer(struct scatterlist *sgl,
812 unsigned int sg_count,
817 struct scatterlist *bounce_sgl;
818 struct page *page_buf;
820 num_pages = ALIGN(len, PAGE_SIZE) >> PAGE_SHIFT;
822 bounce_sgl = kcalloc(num_pages, sizeof(struct scatterlist), GFP_ATOMIC);
826 for (i = 0; i < num_pages; i++) {
827 page_buf = alloc_page(GFP_ATOMIC);
830 sg_set_page(&bounce_sgl[i], page_buf, 0, 0);
836 destroy_bounce_buffer(bounce_sgl, num_pages);
841 /* Assume the original sgl has enough room */
842 static unsigned int copy_from_bounce_buffer(struct scatterlist *orig_sgl,
843 struct scatterlist *bounce_sgl,
844 unsigned int orig_sgl_count)
848 unsigned long src, dest;
849 unsigned int srclen, destlen, copylen;
850 unsigned int total_copied = 0;
851 unsigned long bounce_addr = 0;
852 unsigned long dest_addr = 0;
855 local_irq_save(flags);
857 for (i = 0; i < orig_sgl_count; i++) {
858 dest_addr = (unsigned long)kmap_atomic(sg_page((&orig_sgl[i])),
859 KM_IRQ0) + orig_sgl[i].offset;
861 destlen = orig_sgl[i].length;
863 if (bounce_addr == 0)
865 (unsigned long)kmap_atomic(sg_page((&bounce_sgl[j])),
869 src = bounce_addr + bounce_sgl[j].offset;
870 srclen = bounce_sgl[j].length - bounce_sgl[j].offset;
872 copylen = min(srclen, destlen);
873 memcpy((void *)dest, (void *)src, copylen);
875 total_copied += copylen;
876 bounce_sgl[j].offset += copylen;
880 if (bounce_sgl[j].offset == bounce_sgl[j].length) {
882 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
885 /* if we need to use another bounce buffer */
886 if (destlen || i != orig_sgl_count - 1)
888 (unsigned long)kmap_atomic(
889 sg_page((&bounce_sgl[j])), KM_IRQ0);
890 } else if (destlen == 0 && i == orig_sgl_count - 1) {
891 /* unmap the last bounce that is < PAGE_SIZE */
892 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
896 kunmap_atomic((void *)(dest_addr - orig_sgl[i].offset),
900 local_irq_restore(flags);
906 /* Assume the bounce_sgl has enough room ie using the create_bounce_buffer() */
907 static unsigned int copy_to_bounce_buffer(struct scatterlist *orig_sgl,
908 struct scatterlist *bounce_sgl,
909 unsigned int orig_sgl_count)
913 unsigned long src, dest;
914 unsigned int srclen, destlen, copylen;
915 unsigned int total_copied = 0;
916 unsigned long bounce_addr = 0;
917 unsigned long src_addr = 0;
920 local_irq_save(flags);
922 for (i = 0; i < orig_sgl_count; i++) {
923 src_addr = (unsigned long)kmap_atomic(sg_page((&orig_sgl[i])),
924 KM_IRQ0) + orig_sgl[i].offset;
926 srclen = orig_sgl[i].length;
928 if (bounce_addr == 0)
930 (unsigned long)kmap_atomic(sg_page((&bounce_sgl[j])),
934 /* assume bounce offset always == 0 */
935 dest = bounce_addr + bounce_sgl[j].length;
936 destlen = PAGE_SIZE - bounce_sgl[j].length;
938 copylen = min(srclen, destlen);
939 memcpy((void *)dest, (void *)src, copylen);
941 total_copied += copylen;
942 bounce_sgl[j].length += copylen;
946 if (bounce_sgl[j].length == PAGE_SIZE) {
947 /* full..move to next entry */
948 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
951 /* if we need to use another bounce buffer */
952 if (srclen || i != orig_sgl_count - 1)
954 (unsigned long)kmap_atomic(
955 sg_page((&bounce_sgl[j])), KM_IRQ0);
957 } else if (srclen == 0 && i == orig_sgl_count - 1) {
958 /* unmap the last bounce that is < PAGE_SIZE */
959 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
963 kunmap_atomic((void *)(src_addr - orig_sgl[i].offset), KM_IRQ0);
966 local_irq_restore(flags);
972 static int storvsc_remove(struct hv_device *dev)
974 struct Scsi_Host *host = dev_get_drvdata(&dev->device);
975 struct hv_host_device *host_dev =
976 (struct hv_host_device *)host->hostdata;
978 scsi_remove_host(host);
982 storvsc_dev_remove(dev);
983 if (host_dev->request_pool) {
984 kmem_cache_destroy(host_dev->request_pool);
985 host_dev->request_pool = NULL;
991 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev,
992 sector_t capacity, int *info)
994 sector_t nsect = capacity;
995 sector_t cylinders = nsect;
996 int heads, sectors_pt;
999 * We are making up these values; let us keep it simple.
1002 sectors_pt = 0x3f; /* Sectors per track */
1003 sector_div(cylinders, heads * sectors_pt);
1004 if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
1008 info[1] = sectors_pt;
1009 info[2] = (int)cylinders;
1014 static int storvsc_host_reset(struct hv_device *device)
1016 struct storvsc_device *stor_device;
1017 struct hv_storvsc_request *request;
1018 struct vstor_packet *vstor_packet;
1022 stor_device = get_out_stor_device(device);
1026 request = &stor_device->reset_request;
1027 vstor_packet = &request->vstor_packet;
1029 init_completion(&request->wait_event);
1031 vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
1032 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
1033 vstor_packet->vm_srb.path_id = stor_device->path_id;
1035 ret = vmbus_sendpacket(device->channel, vstor_packet,
1036 sizeof(struct vstor_packet),
1037 (unsigned long)&stor_device->reset_request,
1039 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1043 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
1051 * At this point, all outstanding requests in the adapter
1052 * should have been flushed out and return to us
1061 * storvsc_host_reset_handler - Reset the scsi HBA
1063 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
1066 struct hv_host_device *host_dev =
1067 (struct hv_host_device *)scmnd->device->host->hostdata;
1068 struct hv_device *dev = host_dev->dev;
1070 ret = storvsc_host_reset(dev);
1079 * storvsc_command_completion - Command completion processing
1081 static void storvsc_command_completion(struct hv_storvsc_request *request)
1083 struct storvsc_cmd_request *cmd_request =
1084 (struct storvsc_cmd_request *)request->context;
1085 struct scsi_cmnd *scmnd = cmd_request->cmd;
1086 struct hv_host_device *host_dev =
1087 (struct hv_host_device *)scmnd->device->host->hostdata;
1088 void (*scsi_done_fn)(struct scsi_cmnd *);
1089 struct scsi_sense_hdr sense_hdr;
1090 struct vmscsi_request *vm_srb;
1092 vm_srb = &request->vstor_packet.vm_srb;
1093 if (cmd_request->bounce_sgl_count) {
1094 if (vm_srb->data_in == READ_TYPE) {
1095 copy_from_bounce_buffer(scsi_sglist(scmnd),
1096 cmd_request->bounce_sgl,
1097 scsi_sg_count(scmnd));
1098 destroy_bounce_buffer(cmd_request->bounce_sgl,
1099 cmd_request->bounce_sgl_count);
1104 * If there is an error; offline the device since all
1105 * error recovery strategies would have already been
1106 * deployed on the host side.
1108 if (vm_srb->srb_status == 0x4)
1109 scmnd->result = DID_TARGET_FAILURE << 16;
1111 scmnd->result = vm_srb->scsi_status;
1113 if (scmnd->result) {
1114 if (scsi_normalize_sense(scmnd->sense_buffer,
1115 SCSI_SENSE_BUFFERSIZE, &sense_hdr))
1116 scsi_print_sense_hdr("storvsc", &sense_hdr);
1119 scsi_set_resid(scmnd,
1120 request->data_buffer.len -
1121 vm_srb->data_transfer_length);
1123 scsi_done_fn = scmnd->scsi_done;
1125 scmnd->host_scribble = NULL;
1126 scmnd->scsi_done = NULL;
1128 scsi_done_fn(scmnd);
1130 kmem_cache_free(host_dev->request_pool, cmd_request);
1135 * storvsc_queuecommand - Initiate command processing
1137 static int storvsc_queuecommand_lck(struct scsi_cmnd *scmnd,
1138 void (*done)(struct scsi_cmnd *))
1141 struct hv_host_device *host_dev =
1142 (struct hv_host_device *)scmnd->device->host->hostdata;
1143 struct hv_device *dev = host_dev->dev;
1144 struct hv_storvsc_request *request;
1145 struct storvsc_cmd_request *cmd_request;
1146 unsigned int request_size = 0;
1148 struct scatterlist *sgl;
1149 unsigned int sg_count = 0;
1150 struct vmscsi_request *vm_srb;
1153 /* If retrying, no need to prep the cmd */
1154 if (scmnd->host_scribble) {
1157 (struct storvsc_cmd_request *)scmnd->host_scribble;
1162 scmnd->scsi_done = done;
1164 request_size = sizeof(struct storvsc_cmd_request);
1166 cmd_request = kmem_cache_zalloc(host_dev->request_pool,
1169 scmnd->scsi_done = NULL;
1170 return SCSI_MLQUEUE_DEVICE_BUSY;
1173 /* Setup the cmd request */
1174 cmd_request->bounce_sgl_count = 0;
1175 cmd_request->bounce_sgl = NULL;
1176 cmd_request->cmd = scmnd;
1178 scmnd->host_scribble = (unsigned char *)cmd_request;
1180 request = &cmd_request->request;
1181 vm_srb = &request->vstor_packet.vm_srb;
1185 switch (scmnd->sc_data_direction) {
1187 vm_srb->data_in = WRITE_TYPE;
1189 case DMA_FROM_DEVICE:
1190 vm_srb->data_in = READ_TYPE;
1193 vm_srb->data_in = UNKNOWN_TYPE;
1197 request->on_io_completion = storvsc_command_completion;
1198 request->context = cmd_request;/* scmnd; */
1200 vm_srb->port_number = host_dev->port;
1201 vm_srb->path_id = scmnd->device->channel;
1202 vm_srb->target_id = scmnd->device->id;
1203 vm_srb->lun = scmnd->device->lun;
1205 vm_srb->cdb_length = scmnd->cmd_len;
1207 memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
1209 request->sense_buffer = scmnd->sense_buffer;
1212 request->data_buffer.len = scsi_bufflen(scmnd);
1213 if (scsi_sg_count(scmnd)) {
1214 sgl = (struct scatterlist *)scsi_sglist(scmnd);
1215 sg_count = scsi_sg_count(scmnd);
1217 /* check if we need to bounce the sgl */
1218 if (do_bounce_buffer(sgl, scsi_sg_count(scmnd)) != -1) {
1219 cmd_request->bounce_sgl =
1220 create_bounce_buffer(sgl, scsi_sg_count(scmnd),
1221 scsi_bufflen(scmnd));
1222 if (!cmd_request->bounce_sgl) {
1223 scmnd->scsi_done = NULL;
1224 scmnd->host_scribble = NULL;
1225 kmem_cache_free(host_dev->request_pool,
1228 return SCSI_MLQUEUE_HOST_BUSY;
1231 cmd_request->bounce_sgl_count =
1232 ALIGN(scsi_bufflen(scmnd), PAGE_SIZE) >>
1235 if (vm_srb->data_in == WRITE_TYPE)
1236 copy_to_bounce_buffer(sgl,
1237 cmd_request->bounce_sgl,
1238 scsi_sg_count(scmnd));
1240 sgl = cmd_request->bounce_sgl;
1241 sg_count = cmd_request->bounce_sgl_count;
1244 request->data_buffer.offset = sgl[0].offset;
1246 for (i = 0; i < sg_count; i++)
1247 request->data_buffer.pfn_array[i] =
1248 page_to_pfn(sg_page((&sgl[i])));
1250 } else if (scsi_sglist(scmnd)) {
1251 request->data_buffer.offset =
1252 virt_to_phys(scsi_sglist(scmnd)) & (PAGE_SIZE-1);
1253 request->data_buffer.pfn_array[0] =
1254 virt_to_phys(scsi_sglist(scmnd)) >> PAGE_SHIFT;
1258 /* Invokes the vsc to start an IO */
1259 ret = storvsc_do_io(dev, &cmd_request->request);
1261 if (ret == -EAGAIN) {
1264 if (cmd_request->bounce_sgl_count)
1265 destroy_bounce_buffer(cmd_request->bounce_sgl,
1266 cmd_request->bounce_sgl_count);
1268 kmem_cache_free(host_dev->request_pool, cmd_request);
1270 scmnd->scsi_done = NULL;
1271 scmnd->host_scribble = NULL;
1273 ret = SCSI_MLQUEUE_DEVICE_BUSY;
1279 static DEF_SCSI_QCMD(storvsc_queuecommand)
1283 static struct scsi_host_template scsi_driver = {
1284 .module = THIS_MODULE,
1285 .name = "storvsc_host_t",
1286 .bios_param = storvsc_get_chs,
1287 .queuecommand = storvsc_queuecommand,
1288 .eh_host_reset_handler = storvsc_host_reset_handler,
1289 .slave_alloc = storvsc_device_alloc,
1290 .slave_configure = storvsc_device_configure,
1292 /* 64 max_queue * 1 target */
1293 .can_queue = STORVSC_MAX_IO_REQUESTS*STORVSC_MAX_TARGETS,
1295 /* no use setting to 0 since ll_blk_rw reset it to 1 */
1297 .sg_tablesize = MAX_MULTIPAGE_BUFFER_COUNT,
1299 * ENABLE_CLUSTERING allows mutiple physically contig bio_vecs to merge
1300 * into 1 sg element. If set, we must limit the max_segment_size to
1301 * PAGE_SIZE, otherwise we may get 1 sg element that represents
1304 /* physically contig pfns (ie sg[x].length > PAGE_SIZE). */
1305 .use_clustering = ENABLE_CLUSTERING,
1306 /* Make sure we dont get a sg segment crosses a page boundary */
1307 .dma_boundary = PAGE_SIZE-1,
1315 static const struct hv_vmbus_device_id id_table[] = {
1317 { VMBUS_DEVICE(0xd9, 0x63, 0x61, 0xba, 0xa1, 0x04, 0x29, 0x4d,
1318 0xb6, 0x05, 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f)
1319 .driver_data = SCSI_GUID },
1321 { VMBUS_DEVICE(0x32, 0x26, 0x41, 0x32, 0xcb, 0x86, 0xa2, 0x44,
1322 0x9b, 0x5c, 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5)
1323 .driver_data = IDE_GUID },
1327 MODULE_DEVICE_TABLE(vmbus, id_table);
1331 * storvsc_probe - Add a new device for this driver
1334 static int storvsc_probe(struct hv_device *device,
1335 const struct hv_vmbus_device_id *dev_id)
1338 struct Scsi_Host *host;
1339 struct hv_host_device *host_dev;
1340 bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false);
1343 struct storvsc_device *stor_device;
1345 host = scsi_host_alloc(&scsi_driver,
1346 sizeof(struct hv_host_device));
1350 dev_set_drvdata(&device->device, host);
1352 host_dev = (struct hv_host_device *)host->hostdata;
1353 memset(host_dev, 0, sizeof(struct hv_host_device));
1355 host_dev->port = host->host_no;
1356 host_dev->dev = device;
1358 host_dev->request_pool =
1359 kmem_cache_create(dev_name(&device->device),
1360 sizeof(struct storvsc_cmd_request), 0,
1361 SLAB_HWCACHE_ALIGN, NULL);
1363 if (!host_dev->request_pool) {
1364 scsi_host_put(host);
1368 stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
1370 kmem_cache_destroy(host_dev->request_pool);
1371 scsi_host_put(host);
1375 stor_device->destroy = false;
1376 init_waitqueue_head(&stor_device->waiting_to_drain);
1377 stor_device->device = device;
1378 device->ext = stor_device;
1380 stor_device->port_number = host->host_no;
1381 ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size);
1383 kmem_cache_destroy(host_dev->request_pool);
1384 scsi_host_put(host);
1390 storvsc_get_ide_info(device, &target, &path);
1392 host_dev->path = stor_device->path_id;
1393 host_dev->target = stor_device->target_id;
1395 /* max # of devices per target */
1396 host->max_lun = STORVSC_MAX_LUNS_PER_TARGET;
1397 /* max # of targets per channel */
1398 host->max_id = STORVSC_MAX_TARGETS;
1399 /* max # of channels */
1400 host->max_channel = STORVSC_MAX_CHANNELS - 1;
1401 /* max cmd length */
1402 host->max_cmd_len = STORVSC_MAX_CMD_LEN;
1404 /* Register the HBA and start the scsi bus scan */
1405 ret = scsi_add_host(host, &device->device);
1410 scsi_scan_host(host);
1413 ret = scsi_add_device(host, 0, target, 0);
1415 scsi_remove_host(host);
1421 storvsc_dev_remove(device);
1422 kmem_cache_destroy(host_dev->request_pool);
1423 scsi_host_put(host);
1427 /* The one and only one */
1429 static struct hv_driver storvsc_drv = {
1431 .id_table = id_table,
1432 .probe = storvsc_probe,
1433 .remove = storvsc_remove,
1436 static int __init storvsc_drv_init(void)
1438 u32 max_outstanding_req_per_channel;
1441 * Divide the ring buffer data size (which is 1 page less
1442 * than the ring buffer size since that page is reserved for
1443 * the ring buffer indices) by the max request size (which is
1444 * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
1446 max_outstanding_req_per_channel =
1447 ((storvsc_ringbuffer_size - PAGE_SIZE) /
1448 ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
1449 sizeof(struct vstor_packet) + sizeof(u64),
1452 if (max_outstanding_req_per_channel <
1453 STORVSC_MAX_IO_REQUESTS)
1456 return vmbus_driver_register(&storvsc_drv);
1459 static void __exit storvsc_drv_exit(void)
1461 vmbus_driver_unregister(&storvsc_drv);
1464 MODULE_LICENSE("GPL");
1465 MODULE_VERSION(HV_DRV_VERSION);
1466 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
1467 module_init(storvsc_drv_init);
1468 module_exit(storvsc_drv_exit);