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 stor_mem_pools {
289 struct kmem_cache *request_pool;
290 mempool_t *request_mempool;
293 struct hv_host_device {
294 struct hv_device *dev;
297 unsigned char target;
300 struct storvsc_cmd_request {
301 struct list_head entry;
302 struct scsi_cmnd *cmd;
304 unsigned int bounce_sgl_count;
305 struct scatterlist *bounce_sgl;
307 struct hv_storvsc_request request;
310 struct storvsc_scan_work {
311 struct work_struct work;
312 struct Scsi_Host *host;
316 static void storvsc_bus_scan(struct work_struct *work)
318 struct storvsc_scan_work *wrk;
321 wrk = container_of(work, struct storvsc_scan_work, work);
322 for (id = 0; id < wrk->host->max_id; ++id) {
323 if (wrk->host->reverse_ordering)
324 order_id = wrk->host->max_id - id - 1;
328 scsi_scan_target(&wrk->host->shost_gendev, 0,
329 order_id, SCAN_WILD_CARD, 1);
334 static void storvsc_remove_lun(struct work_struct *work)
336 struct storvsc_scan_work *wrk;
337 struct scsi_device *sdev;
339 wrk = container_of(work, struct storvsc_scan_work, work);
340 if (!scsi_host_get(wrk->host))
343 sdev = scsi_device_lookup(wrk->host, 0, 0, wrk->lun);
346 scsi_remove_device(sdev);
347 scsi_device_put(sdev);
349 scsi_host_put(wrk->host);
355 static inline struct storvsc_device *get_out_stor_device(
356 struct hv_device *device)
358 struct storvsc_device *stor_device;
360 stor_device = hv_get_drvdata(device);
362 if (stor_device && stor_device->destroy)
369 static inline void storvsc_wait_to_drain(struct storvsc_device *dev)
371 dev->drain_notify = true;
372 wait_event(dev->waiting_to_drain,
373 atomic_read(&dev->num_outstanding_req) == 0);
374 dev->drain_notify = false;
377 static inline struct storvsc_device *get_in_stor_device(
378 struct hv_device *device)
380 struct storvsc_device *stor_device;
382 stor_device = hv_get_drvdata(device);
388 * If the device is being destroyed; allow incoming
389 * traffic only to cleanup outstanding requests.
392 if (stor_device->destroy &&
393 (atomic_read(&stor_device->num_outstanding_req) == 0))
401 static int storvsc_channel_init(struct hv_device *device)
403 struct storvsc_device *stor_device;
404 struct hv_storvsc_request *request;
405 struct vstor_packet *vstor_packet;
408 stor_device = get_out_stor_device(device);
412 request = &stor_device->init_request;
413 vstor_packet = &request->vstor_packet;
416 * Now, initiate the vsc/vsp initialization protocol on the open
419 memset(request, 0, sizeof(struct hv_storvsc_request));
420 init_completion(&request->wait_event);
421 vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION;
422 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
424 ret = vmbus_sendpacket(device->channel, vstor_packet,
425 sizeof(struct vstor_packet),
426 (unsigned long)request,
428 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
432 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
438 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
439 vstor_packet->status != 0)
443 /* reuse the packet for version range supported */
444 memset(vstor_packet, 0, sizeof(struct vstor_packet));
445 vstor_packet->operation = VSTOR_OPERATION_QUERY_PROTOCOL_VERSION;
446 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
448 vstor_packet->version.major_minor = VMSTOR_PROTOCOL_VERSION_CURRENT;
449 FILL_VMSTOR_REVISION(vstor_packet->version.revision);
451 ret = vmbus_sendpacket(device->channel, vstor_packet,
452 sizeof(struct vstor_packet),
453 (unsigned long)request,
455 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
459 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
465 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
466 vstor_packet->status != 0)
470 memset(vstor_packet, 0, sizeof(struct vstor_packet));
471 vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES;
472 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
473 vstor_packet->storage_channel_properties.port_number =
474 stor_device->port_number;
476 ret = vmbus_sendpacket(device->channel, vstor_packet,
477 sizeof(struct vstor_packet),
478 (unsigned long)request,
480 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
485 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
491 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
492 vstor_packet->status != 0)
495 stor_device->path_id = vstor_packet->storage_channel_properties.path_id;
496 stor_device->target_id
497 = vstor_packet->storage_channel_properties.target_id;
499 memset(vstor_packet, 0, sizeof(struct vstor_packet));
500 vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION;
501 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
503 ret = vmbus_sendpacket(device->channel, vstor_packet,
504 sizeof(struct vstor_packet),
505 (unsigned long)request,
507 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
512 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
518 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
519 vstor_packet->status != 0)
527 static void storvsc_on_io_completion(struct hv_device *device,
528 struct vstor_packet *vstor_packet,
529 struct hv_storvsc_request *request)
531 struct storvsc_device *stor_device;
532 struct vstor_packet *stor_pkt;
534 stor_device = hv_get_drvdata(device);
535 stor_pkt = &request->vstor_packet;
538 * The current SCSI handling on the host side does
539 * not correctly handle:
540 * INQUIRY command with page code parameter set to 0x80
541 * MODE_SENSE command with cmd[2] == 0x1c
543 * Setup srb and scsi status so this won't be fatal.
544 * We do this so we can distinguish truly fatal failues
545 * (srb status == 0x4) and off-line the device in that case.
548 if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) ||
549 (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) {
550 vstor_packet->vm_srb.scsi_status = 0;
551 vstor_packet->vm_srb.srb_status = 0x1;
555 /* Copy over the status...etc */
556 stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status;
557 stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status;
558 stor_pkt->vm_srb.sense_info_length =
559 vstor_packet->vm_srb.sense_info_length;
561 if (vstor_packet->vm_srb.scsi_status != 0 ||
562 vstor_packet->vm_srb.srb_status != 1){
563 dev_warn(&device->device,
564 "cmd 0x%x scsi status 0x%x srb status 0x%x\n",
565 stor_pkt->vm_srb.cdb[0],
566 vstor_packet->vm_srb.scsi_status,
567 vstor_packet->vm_srb.srb_status);
570 if ((vstor_packet->vm_srb.scsi_status & 0xFF) == 0x02) {
571 /* CHECK_CONDITION */
572 if (vstor_packet->vm_srb.srb_status & 0x80) {
573 /* autosense data available */
574 dev_warn(&device->device,
575 "stor pkt %p autosense data valid - len %d\n",
577 vstor_packet->vm_srb.sense_info_length);
579 memcpy(request->sense_buffer,
580 vstor_packet->vm_srb.sense_data,
581 vstor_packet->vm_srb.sense_info_length);
586 stor_pkt->vm_srb.data_transfer_length =
587 vstor_packet->vm_srb.data_transfer_length;
589 request->on_io_completion(request);
591 if (atomic_dec_and_test(&stor_device->num_outstanding_req) &&
592 stor_device->drain_notify)
593 wake_up(&stor_device->waiting_to_drain);
598 static void storvsc_on_receive(struct hv_device *device,
599 struct vstor_packet *vstor_packet,
600 struct hv_storvsc_request *request)
602 struct storvsc_scan_work *work;
603 struct storvsc_device *stor_device;
605 switch (vstor_packet->operation) {
606 case VSTOR_OPERATION_COMPLETE_IO:
607 storvsc_on_io_completion(device, vstor_packet, request);
610 case VSTOR_OPERATION_REMOVE_DEVICE:
611 case VSTOR_OPERATION_ENUMERATE_BUS:
612 stor_device = get_in_stor_device(device);
613 work = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC);
617 INIT_WORK(&work->work, storvsc_bus_scan);
618 work->host = stor_device->host;
619 schedule_work(&work->work);
627 static void storvsc_on_channel_callback(void *context)
629 struct hv_device *device = (struct hv_device *)context;
630 struct storvsc_device *stor_device;
633 unsigned char packet[ALIGN(sizeof(struct vstor_packet), 8)];
634 struct hv_storvsc_request *request;
638 stor_device = get_in_stor_device(device);
643 ret = vmbus_recvpacket(device->channel, packet,
644 ALIGN(sizeof(struct vstor_packet), 8),
645 &bytes_recvd, &request_id);
646 if (ret == 0 && bytes_recvd > 0) {
648 request = (struct hv_storvsc_request *)
649 (unsigned long)request_id;
651 if ((request == &stor_device->init_request) ||
652 (request == &stor_device->reset_request)) {
654 memcpy(&request->vstor_packet, packet,
655 sizeof(struct vstor_packet));
656 complete(&request->wait_event);
658 storvsc_on_receive(device,
659 (struct vstor_packet *)packet,
670 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size)
672 struct vmstorage_channel_properties props;
675 memset(&props, 0, sizeof(struct vmstorage_channel_properties));
677 /* Open the channel */
678 ret = vmbus_open(device->channel,
682 sizeof(struct vmstorage_channel_properties),
683 storvsc_on_channel_callback, device);
688 ret = storvsc_channel_init(device);
693 static int storvsc_dev_remove(struct hv_device *device)
695 struct storvsc_device *stor_device;
698 stor_device = hv_get_drvdata(device);
700 spin_lock_irqsave(&device->channel->inbound_lock, flags);
701 stor_device->destroy = true;
702 spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
705 * At this point, all outbound traffic should be disable. We
706 * only allow inbound traffic (responses) to proceed so that
707 * outstanding requests can be completed.
710 storvsc_wait_to_drain(stor_device);
713 * Since we have already drained, we don't need to busy wait
714 * as was done in final_release_stor_device()
715 * Note that we cannot set the ext pointer to NULL until
716 * we have drained - to drain the outgoing packets, we need to
717 * allow incoming packets.
719 spin_lock_irqsave(&device->channel->inbound_lock, flags);
720 hv_set_drvdata(device, NULL);
721 spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
723 /* Close the channel */
724 vmbus_close(device->channel);
730 static int storvsc_do_io(struct hv_device *device,
731 struct hv_storvsc_request *request)
733 struct storvsc_device *stor_device;
734 struct vstor_packet *vstor_packet;
737 vstor_packet = &request->vstor_packet;
738 stor_device = get_out_stor_device(device);
744 request->device = device;
747 vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
749 vstor_packet->vm_srb.length = sizeof(struct vmscsi_request);
752 vstor_packet->vm_srb.sense_info_length = SENSE_BUFFER_SIZE;
755 vstor_packet->vm_srb.data_transfer_length =
756 request->data_buffer.len;
758 vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
760 if (request->data_buffer.len) {
761 ret = vmbus_sendpacket_multipagebuffer(device->channel,
762 &request->data_buffer,
764 sizeof(struct vstor_packet),
765 (unsigned long)request);
767 ret = vmbus_sendpacket(device->channel, vstor_packet,
768 sizeof(struct vstor_packet),
769 (unsigned long)request,
771 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
777 atomic_inc(&stor_device->num_outstanding_req);
782 static void storvsc_get_ide_info(struct hv_device *dev, int *target, int *path)
785 dev->dev_instance.b[5] << 8 | dev->dev_instance.b[4];
788 dev->dev_instance.b[3] << 24 |
789 dev->dev_instance.b[2] << 16 |
790 dev->dev_instance.b[1] << 8 | dev->dev_instance.b[0];
794 static int storvsc_device_alloc(struct scsi_device *sdevice)
796 struct stor_mem_pools *memp;
797 int number = STORVSC_MIN_BUF_NR;
799 memp = kzalloc(sizeof(struct stor_mem_pools), GFP_KERNEL);
804 kmem_cache_create(dev_name(&sdevice->sdev_dev),
805 sizeof(struct storvsc_cmd_request), 0,
806 SLAB_HWCACHE_ALIGN, NULL);
808 if (!memp->request_pool)
811 memp->request_mempool = mempool_create(number, mempool_alloc_slab,
815 if (!memp->request_mempool)
818 sdevice->hostdata = memp;
823 kmem_cache_destroy(memp->request_pool);
830 static void storvsc_device_destroy(struct scsi_device *sdevice)
832 struct stor_mem_pools *memp = sdevice->hostdata;
834 mempool_destroy(memp->request_mempool);
835 kmem_cache_destroy(memp->request_pool);
837 sdevice->hostdata = NULL;
840 static int storvsc_device_configure(struct scsi_device *sdevice)
842 scsi_adjust_queue_depth(sdevice, MSG_SIMPLE_TAG,
843 STORVSC_MAX_IO_REQUESTS);
845 blk_queue_max_segment_size(sdevice->request_queue, PAGE_SIZE);
847 blk_queue_bounce_limit(sdevice->request_queue, BLK_BOUNCE_ANY);
852 static void destroy_bounce_buffer(struct scatterlist *sgl,
853 unsigned int sg_count)
856 struct page *page_buf;
858 for (i = 0; i < sg_count; i++) {
859 page_buf = sg_page((&sgl[i]));
860 if (page_buf != NULL)
861 __free_page(page_buf);
867 static int do_bounce_buffer(struct scatterlist *sgl, unsigned int sg_count)
871 /* No need to check */
875 /* We have at least 2 sg entries */
876 for (i = 0; i < sg_count; i++) {
878 /* make sure 1st one does not have hole */
879 if (sgl[i].offset + sgl[i].length != PAGE_SIZE)
881 } else if (i == sg_count - 1) {
882 /* make sure last one does not have hole */
883 if (sgl[i].offset != 0)
886 /* make sure no hole in the middle */
887 if (sgl[i].length != PAGE_SIZE || sgl[i].offset != 0)
894 static struct scatterlist *create_bounce_buffer(struct scatterlist *sgl,
895 unsigned int sg_count,
901 struct scatterlist *bounce_sgl;
902 struct page *page_buf;
903 unsigned int buf_len = ((write == WRITE_TYPE) ? 0 : PAGE_SIZE);
905 num_pages = ALIGN(len, PAGE_SIZE) >> PAGE_SHIFT;
907 bounce_sgl = kcalloc(num_pages, sizeof(struct scatterlist), GFP_ATOMIC);
911 for (i = 0; i < num_pages; i++) {
912 page_buf = alloc_page(GFP_ATOMIC);
915 sg_set_page(&bounce_sgl[i], page_buf, buf_len, 0);
921 destroy_bounce_buffer(bounce_sgl, num_pages);
926 /* Assume the original sgl has enough room */
927 static unsigned int copy_from_bounce_buffer(struct scatterlist *orig_sgl,
928 struct scatterlist *bounce_sgl,
929 unsigned int orig_sgl_count,
930 unsigned int bounce_sgl_count)
934 unsigned long src, dest;
935 unsigned int srclen, destlen, copylen;
936 unsigned int total_copied = 0;
937 unsigned long bounce_addr = 0;
938 unsigned long dest_addr = 0;
941 local_irq_save(flags);
943 for (i = 0; i < orig_sgl_count; i++) {
944 dest_addr = (unsigned long)kmap_atomic(sg_page((&orig_sgl[i])),
945 KM_IRQ0) + orig_sgl[i].offset;
947 destlen = orig_sgl[i].length;
949 if (bounce_addr == 0)
951 (unsigned long)kmap_atomic(sg_page((&bounce_sgl[j])),
955 src = bounce_addr + bounce_sgl[j].offset;
956 srclen = bounce_sgl[j].length - bounce_sgl[j].offset;
958 copylen = min(srclen, destlen);
959 memcpy((void *)dest, (void *)src, copylen);
961 total_copied += copylen;
962 bounce_sgl[j].offset += copylen;
966 if (bounce_sgl[j].offset == bounce_sgl[j].length) {
968 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
972 * It is possible that the number of elements
973 * in the bounce buffer may not be equal to
974 * the number of elements in the original
975 * scatter list. Handle this correctly.
978 if (j == bounce_sgl_count) {
980 * We are done; cleanup and return.
982 kunmap_atomic((void *)(dest_addr -
985 local_irq_restore(flags);
989 /* if we need to use another bounce buffer */
990 if (destlen || i != orig_sgl_count - 1)
992 (unsigned long)kmap_atomic(
993 sg_page((&bounce_sgl[j])), KM_IRQ0);
994 } else if (destlen == 0 && i == orig_sgl_count - 1) {
995 /* unmap the last bounce that is < PAGE_SIZE */
996 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
1000 kunmap_atomic((void *)(dest_addr - orig_sgl[i].offset),
1004 local_irq_restore(flags);
1006 return total_copied;
1010 /* Assume the bounce_sgl has enough room ie using the create_bounce_buffer() */
1011 static unsigned int copy_to_bounce_buffer(struct scatterlist *orig_sgl,
1012 struct scatterlist *bounce_sgl,
1013 unsigned int orig_sgl_count)
1017 unsigned long src, dest;
1018 unsigned int srclen, destlen, copylen;
1019 unsigned int total_copied = 0;
1020 unsigned long bounce_addr = 0;
1021 unsigned long src_addr = 0;
1022 unsigned long flags;
1024 local_irq_save(flags);
1026 for (i = 0; i < orig_sgl_count; i++) {
1027 src_addr = (unsigned long)kmap_atomic(sg_page((&orig_sgl[i])),
1028 KM_IRQ0) + orig_sgl[i].offset;
1030 srclen = orig_sgl[i].length;
1032 if (bounce_addr == 0)
1034 (unsigned long)kmap_atomic(sg_page((&bounce_sgl[j])),
1038 /* assume bounce offset always == 0 */
1039 dest = bounce_addr + bounce_sgl[j].length;
1040 destlen = PAGE_SIZE - bounce_sgl[j].length;
1042 copylen = min(srclen, destlen);
1043 memcpy((void *)dest, (void *)src, copylen);
1045 total_copied += copylen;
1046 bounce_sgl[j].length += copylen;
1050 if (bounce_sgl[j].length == PAGE_SIZE) {
1051 /* full..move to next entry */
1052 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
1055 /* if we need to use another bounce buffer */
1056 if (srclen || i != orig_sgl_count - 1)
1058 (unsigned long)kmap_atomic(
1059 sg_page((&bounce_sgl[j])), KM_IRQ0);
1061 } else if (srclen == 0 && i == orig_sgl_count - 1) {
1062 /* unmap the last bounce that is < PAGE_SIZE */
1063 kunmap_atomic((void *)bounce_addr, KM_IRQ0);
1067 kunmap_atomic((void *)(src_addr - orig_sgl[i].offset), KM_IRQ0);
1070 local_irq_restore(flags);
1072 return total_copied;
1076 static int storvsc_remove(struct hv_device *dev)
1078 struct storvsc_device *stor_device = hv_get_drvdata(dev);
1079 struct Scsi_Host *host = stor_device->host;
1081 scsi_remove_host(host);
1083 scsi_host_put(host);
1085 storvsc_dev_remove(dev);
1091 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev,
1092 sector_t capacity, int *info)
1094 sector_t nsect = capacity;
1095 sector_t cylinders = nsect;
1096 int heads, sectors_pt;
1099 * We are making up these values; let us keep it simple.
1102 sectors_pt = 0x3f; /* Sectors per track */
1103 sector_div(cylinders, heads * sectors_pt);
1104 if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
1108 info[1] = sectors_pt;
1109 info[2] = (int)cylinders;
1114 static int storvsc_host_reset(struct hv_device *device)
1116 struct storvsc_device *stor_device;
1117 struct hv_storvsc_request *request;
1118 struct vstor_packet *vstor_packet;
1122 stor_device = get_out_stor_device(device);
1126 request = &stor_device->reset_request;
1127 vstor_packet = &request->vstor_packet;
1129 init_completion(&request->wait_event);
1131 vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
1132 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
1133 vstor_packet->vm_srb.path_id = stor_device->path_id;
1135 ret = vmbus_sendpacket(device->channel, vstor_packet,
1136 sizeof(struct vstor_packet),
1137 (unsigned long)&stor_device->reset_request,
1139 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1143 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
1145 return TIMEOUT_ERROR;
1149 * At this point, all outstanding requests in the adapter
1150 * should have been flushed out and return to us
1158 * storvsc_host_reset_handler - Reset the scsi HBA
1160 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
1162 struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
1163 struct hv_device *dev = host_dev->dev;
1165 return storvsc_host_reset(dev);
1170 * storvsc_command_completion - Command completion processing
1172 static void storvsc_command_completion(struct hv_storvsc_request *request)
1174 struct storvsc_cmd_request *cmd_request =
1175 (struct storvsc_cmd_request *)request->context;
1176 struct scsi_cmnd *scmnd = cmd_request->cmd;
1177 struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
1178 void (*scsi_done_fn)(struct scsi_cmnd *);
1179 struct scsi_sense_hdr sense_hdr;
1180 struct vmscsi_request *vm_srb;
1181 struct storvsc_scan_work *wrk;
1182 struct stor_mem_pools *memp = scmnd->device->hostdata;
1184 vm_srb = &request->vstor_packet.vm_srb;
1185 if (cmd_request->bounce_sgl_count) {
1186 if (vm_srb->data_in == READ_TYPE)
1187 copy_from_bounce_buffer(scsi_sglist(scmnd),
1188 cmd_request->bounce_sgl,
1189 scsi_sg_count(scmnd),
1190 cmd_request->bounce_sgl_count);
1191 destroy_bounce_buffer(cmd_request->bounce_sgl,
1192 cmd_request->bounce_sgl_count);
1196 * If there is an error; offline the device since all
1197 * error recovery strategies would have already been
1198 * deployed on the host side.
1200 if (vm_srb->srb_status == 0x4)
1201 scmnd->result = DID_TARGET_FAILURE << 16;
1203 scmnd->result = vm_srb->scsi_status;
1206 * If the LUN is invalid; remove the device.
1208 if (vm_srb->srb_status == 0x20) {
1209 struct storvsc_device *stor_dev;
1210 struct hv_device *dev = host_dev->dev;
1211 struct Scsi_Host *host;
1213 stor_dev = get_in_stor_device(dev);
1214 host = stor_dev->host;
1216 wrk = kmalloc(sizeof(struct storvsc_scan_work),
1219 scmnd->result = DID_TARGET_FAILURE << 16;
1222 wrk->lun = vm_srb->lun;
1223 INIT_WORK(&wrk->work, storvsc_remove_lun);
1224 schedule_work(&wrk->work);
1228 if (scmnd->result) {
1229 if (scsi_normalize_sense(scmnd->sense_buffer,
1230 SCSI_SENSE_BUFFERSIZE, &sense_hdr))
1231 scsi_print_sense_hdr("storvsc", &sense_hdr);
1234 scsi_set_resid(scmnd,
1235 request->data_buffer.len -
1236 vm_srb->data_transfer_length);
1238 scsi_done_fn = scmnd->scsi_done;
1240 scmnd->host_scribble = NULL;
1241 scmnd->scsi_done = NULL;
1243 scsi_done_fn(scmnd);
1245 mempool_free(cmd_request, memp->request_mempool);
1248 static bool storvsc_check_scsi_cmd(struct scsi_cmnd *scmnd)
1250 bool allowed = true;
1251 u8 scsi_op = scmnd->cmnd[0];
1254 /* smartd sends this command, which will offline the device */
1256 scmnd->result = ILLEGAL_REQUEST << 16;
1266 * storvsc_queuecommand - Initiate command processing
1268 static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
1271 struct hv_host_device *host_dev = shost_priv(host);
1272 struct hv_device *dev = host_dev->dev;
1273 struct hv_storvsc_request *request;
1274 struct storvsc_cmd_request *cmd_request;
1275 unsigned int request_size = 0;
1277 struct scatterlist *sgl;
1278 unsigned int sg_count = 0;
1279 struct vmscsi_request *vm_srb;
1280 struct stor_mem_pools *memp = scmnd->device->hostdata;
1282 if (storvsc_check_scsi_cmd(scmnd) == false) {
1283 scmnd->scsi_done(scmnd);
1287 /* If retrying, no need to prep the cmd */
1288 if (scmnd->host_scribble) {
1291 (struct storvsc_cmd_request *)scmnd->host_scribble;
1296 request_size = sizeof(struct storvsc_cmd_request);
1298 cmd_request = mempool_alloc(memp->request_mempool,
1301 return SCSI_MLQUEUE_DEVICE_BUSY;
1303 memset(cmd_request, 0, sizeof(struct storvsc_cmd_request));
1305 /* Setup the cmd request */
1306 cmd_request->bounce_sgl_count = 0;
1307 cmd_request->bounce_sgl = NULL;
1308 cmd_request->cmd = scmnd;
1310 scmnd->host_scribble = (unsigned char *)cmd_request;
1312 request = &cmd_request->request;
1313 vm_srb = &request->vstor_packet.vm_srb;
1317 switch (scmnd->sc_data_direction) {
1319 vm_srb->data_in = WRITE_TYPE;
1321 case DMA_FROM_DEVICE:
1322 vm_srb->data_in = READ_TYPE;
1325 vm_srb->data_in = UNKNOWN_TYPE;
1329 request->on_io_completion = storvsc_command_completion;
1330 request->context = cmd_request;/* scmnd; */
1332 vm_srb->port_number = host_dev->port;
1333 vm_srb->path_id = scmnd->device->channel;
1334 vm_srb->target_id = scmnd->device->id;
1335 vm_srb->lun = scmnd->device->lun;
1337 vm_srb->cdb_length = scmnd->cmd_len;
1339 memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
1341 request->sense_buffer = scmnd->sense_buffer;
1344 request->data_buffer.len = scsi_bufflen(scmnd);
1345 if (scsi_sg_count(scmnd)) {
1346 sgl = (struct scatterlist *)scsi_sglist(scmnd);
1347 sg_count = scsi_sg_count(scmnd);
1349 /* check if we need to bounce the sgl */
1350 if (do_bounce_buffer(sgl, scsi_sg_count(scmnd)) != -1) {
1351 cmd_request->bounce_sgl =
1352 create_bounce_buffer(sgl, scsi_sg_count(scmnd),
1353 scsi_bufflen(scmnd),
1355 if (!cmd_request->bounce_sgl) {
1356 scmnd->host_scribble = NULL;
1357 mempool_free(cmd_request,
1358 memp->request_mempool);
1360 return SCSI_MLQUEUE_HOST_BUSY;
1363 cmd_request->bounce_sgl_count =
1364 ALIGN(scsi_bufflen(scmnd), PAGE_SIZE) >>
1367 if (vm_srb->data_in == WRITE_TYPE)
1368 copy_to_bounce_buffer(sgl,
1369 cmd_request->bounce_sgl,
1370 scsi_sg_count(scmnd));
1372 sgl = cmd_request->bounce_sgl;
1373 sg_count = cmd_request->bounce_sgl_count;
1376 request->data_buffer.offset = sgl[0].offset;
1378 for (i = 0; i < sg_count; i++)
1379 request->data_buffer.pfn_array[i] =
1380 page_to_pfn(sg_page((&sgl[i])));
1382 } else if (scsi_sglist(scmnd)) {
1383 request->data_buffer.offset =
1384 virt_to_phys(scsi_sglist(scmnd)) & (PAGE_SIZE-1);
1385 request->data_buffer.pfn_array[0] =
1386 virt_to_phys(scsi_sglist(scmnd)) >> PAGE_SHIFT;
1390 /* Invokes the vsc to start an IO */
1391 ret = storvsc_do_io(dev, &cmd_request->request);
1393 if (ret == -EAGAIN) {
1396 if (cmd_request->bounce_sgl_count)
1397 destroy_bounce_buffer(cmd_request->bounce_sgl,
1398 cmd_request->bounce_sgl_count);
1400 mempool_free(cmd_request, memp->request_mempool);
1402 scmnd->host_scribble = NULL;
1404 ret = SCSI_MLQUEUE_DEVICE_BUSY;
1411 static struct scsi_host_template scsi_driver = {
1412 .module = THIS_MODULE,
1413 .name = "storvsc_host_t",
1414 .bios_param = storvsc_get_chs,
1415 .queuecommand = storvsc_queuecommand,
1416 .eh_host_reset_handler = storvsc_host_reset_handler,
1417 .slave_alloc = storvsc_device_alloc,
1418 .slave_destroy = storvsc_device_destroy,
1419 .slave_configure = storvsc_device_configure,
1421 /* 64 max_queue * 1 target */
1422 .can_queue = STORVSC_MAX_IO_REQUESTS*STORVSC_MAX_TARGETS,
1424 /* no use setting to 0 since ll_blk_rw reset it to 1 */
1426 .sg_tablesize = MAX_MULTIPAGE_BUFFER_COUNT,
1427 .use_clustering = DISABLE_CLUSTERING,
1428 /* Make sure we dont get a sg segment crosses a page boundary */
1429 .dma_boundary = PAGE_SIZE-1,
1437 static const struct hv_vmbus_device_id id_table[] = {
1439 { VMBUS_DEVICE(0xd9, 0x63, 0x61, 0xba, 0xa1, 0x04, 0x29, 0x4d,
1440 0xb6, 0x05, 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f)
1441 .driver_data = SCSI_GUID },
1443 { VMBUS_DEVICE(0x32, 0x26, 0x41, 0x32, 0xcb, 0x86, 0xa2, 0x44,
1444 0x9b, 0x5c, 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5)
1445 .driver_data = IDE_GUID },
1449 MODULE_DEVICE_TABLE(vmbus, id_table);
1453 * storvsc_probe - Add a new device for this driver
1456 static int storvsc_probe(struct hv_device *device,
1457 const struct hv_vmbus_device_id *dev_id)
1460 struct Scsi_Host *host;
1461 struct hv_host_device *host_dev;
1462 bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false);
1465 struct storvsc_device *stor_device;
1467 host = scsi_host_alloc(&scsi_driver,
1468 sizeof(struct hv_host_device));
1472 host_dev = shost_priv(host);
1473 memset(host_dev, 0, sizeof(struct hv_host_device));
1475 host_dev->port = host->host_no;
1476 host_dev->dev = device;
1479 stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
1485 stor_device->destroy = false;
1486 init_waitqueue_head(&stor_device->waiting_to_drain);
1487 stor_device->device = device;
1488 stor_device->host = host;
1489 hv_set_drvdata(device, stor_device);
1491 stor_device->port_number = host->host_no;
1492 ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size);
1497 storvsc_get_ide_info(device, &target, &path);
1499 host_dev->path = stor_device->path_id;
1500 host_dev->target = stor_device->target_id;
1502 /* max # of devices per target */
1503 host->max_lun = STORVSC_MAX_LUNS_PER_TARGET;
1504 /* max # of targets per channel */
1505 host->max_id = STORVSC_MAX_TARGETS;
1506 /* max # of channels */
1507 host->max_channel = STORVSC_MAX_CHANNELS - 1;
1508 /* max cmd length */
1509 host->max_cmd_len = STORVSC_MAX_CMD_LEN;
1511 /* Register the HBA and start the scsi bus scan */
1512 ret = scsi_add_host(host, &device->device);
1517 scsi_scan_host(host);
1520 ret = scsi_add_device(host, 0, target, 0);
1522 scsi_remove_host(host);
1529 * Once we have connected with the host, we would need to
1530 * to invoke storvsc_dev_remove() to rollback this state and
1531 * this call also frees up the stor_device; hence the jump around
1534 storvsc_dev_remove(device);
1541 scsi_host_put(host);
1545 /* The one and only one */
1547 static struct hv_driver storvsc_drv = {
1548 .name = KBUILD_MODNAME,
1549 .id_table = id_table,
1550 .probe = storvsc_probe,
1551 .remove = storvsc_remove,
1554 static int __init storvsc_drv_init(void)
1556 u32 max_outstanding_req_per_channel;
1559 * Divide the ring buffer data size (which is 1 page less
1560 * than the ring buffer size since that page is reserved for
1561 * the ring buffer indices) by the max request size (which is
1562 * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
1564 max_outstanding_req_per_channel =
1565 ((storvsc_ringbuffer_size - PAGE_SIZE) /
1566 ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
1567 sizeof(struct vstor_packet) + sizeof(u64),
1570 if (max_outstanding_req_per_channel <
1571 STORVSC_MAX_IO_REQUESTS)
1574 return vmbus_driver_register(&storvsc_drv);
1577 static void __exit storvsc_drv_exit(void)
1579 vmbus_driver_unregister(&storvsc_drv);
1582 MODULE_LICENSE("GPL");
1583 MODULE_VERSION(HV_DRV_VERSION);
1584 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
1585 module_init(storvsc_drv_init);
1586 module_exit(storvsc_drv_exit);