1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (c) 2009, Microsoft Corporation.
6 * Haiyang Zhang <haiyangz@microsoft.com>
7 * Hank Janssen <hjanssen@microsoft.com>
8 * K. Y. Srinivasan <kys@microsoft.com>
11 #include <linux/kernel.h>
12 #include <linux/wait.h>
13 #include <linux/sched.h>
14 #include <linux/completion.h>
15 #include <linux/string.h>
17 #include <linux/delay.h>
18 #include <linux/init.h>
19 #include <linux/slab.h>
20 #include <linux/module.h>
21 #include <linux/device.h>
22 #include <linux/hyperv.h>
23 #include <linux/blkdev.h>
24 #include <scsi/scsi.h>
25 #include <scsi/scsi_cmnd.h>
26 #include <scsi/scsi_host.h>
27 #include <scsi/scsi_device.h>
28 #include <scsi/scsi_tcq.h>
29 #include <scsi/scsi_eh.h>
30 #include <scsi/scsi_devinfo.h>
31 #include <scsi/scsi_dbg.h>
32 #include <scsi/scsi_transport_fc.h>
33 #include <scsi/scsi_transport.h>
36 * All wire protocol details (storage protocol between the guest and the host)
37 * are consolidated here.
39 * Begin protocol definitions.
45 * V1 RC < 2008/1/31: 1.0
46 * V1 RC > 2008/1/31: 2.0
53 #define VMSTOR_PROTO_VERSION(MAJOR_, MINOR_) ((((MAJOR_) & 0xff) << 8) | \
56 #define VMSTOR_PROTO_VERSION_WIN6 VMSTOR_PROTO_VERSION(2, 0)
57 #define VMSTOR_PROTO_VERSION_WIN7 VMSTOR_PROTO_VERSION(4, 2)
58 #define VMSTOR_PROTO_VERSION_WIN8 VMSTOR_PROTO_VERSION(5, 1)
59 #define VMSTOR_PROTO_VERSION_WIN8_1 VMSTOR_PROTO_VERSION(6, 0)
60 #define VMSTOR_PROTO_VERSION_WIN10 VMSTOR_PROTO_VERSION(6, 2)
62 /* Packet structure describing virtual storage requests. */
63 enum vstor_packet_operation {
64 VSTOR_OPERATION_COMPLETE_IO = 1,
65 VSTOR_OPERATION_REMOVE_DEVICE = 2,
66 VSTOR_OPERATION_EXECUTE_SRB = 3,
67 VSTOR_OPERATION_RESET_LUN = 4,
68 VSTOR_OPERATION_RESET_ADAPTER = 5,
69 VSTOR_OPERATION_RESET_BUS = 6,
70 VSTOR_OPERATION_BEGIN_INITIALIZATION = 7,
71 VSTOR_OPERATION_END_INITIALIZATION = 8,
72 VSTOR_OPERATION_QUERY_PROTOCOL_VERSION = 9,
73 VSTOR_OPERATION_QUERY_PROPERTIES = 10,
74 VSTOR_OPERATION_ENUMERATE_BUS = 11,
75 VSTOR_OPERATION_FCHBA_DATA = 12,
76 VSTOR_OPERATION_CREATE_SUB_CHANNELS = 13,
77 VSTOR_OPERATION_MAXIMUM = 13
81 * WWN packet for Fibre Channel HBA
84 struct hv_fc_wwn_packet {
87 u8 primary_port_wwn[8];
88 u8 primary_node_wwn[8];
89 u8 secondary_port_wwn[8];
90 u8 secondary_node_wwn[8];
99 #define SRB_FLAGS_QUEUE_ACTION_ENABLE 0x00000002
100 #define SRB_FLAGS_DISABLE_DISCONNECT 0x00000004
101 #define SRB_FLAGS_DISABLE_SYNCH_TRANSFER 0x00000008
102 #define SRB_FLAGS_BYPASS_FROZEN_QUEUE 0x00000010
103 #define SRB_FLAGS_DISABLE_AUTOSENSE 0x00000020
104 #define SRB_FLAGS_DATA_IN 0x00000040
105 #define SRB_FLAGS_DATA_OUT 0x00000080
106 #define SRB_FLAGS_NO_DATA_TRANSFER 0x00000000
107 #define SRB_FLAGS_UNSPECIFIED_DIRECTION (SRB_FLAGS_DATA_IN | SRB_FLAGS_DATA_OUT)
108 #define SRB_FLAGS_NO_QUEUE_FREEZE 0x00000100
109 #define SRB_FLAGS_ADAPTER_CACHE_ENABLE 0x00000200
110 #define SRB_FLAGS_FREE_SENSE_BUFFER 0x00000400
113 * This flag indicates the request is part of the workflow for processing a D3.
115 #define SRB_FLAGS_D3_PROCESSING 0x00000800
116 #define SRB_FLAGS_IS_ACTIVE 0x00010000
117 #define SRB_FLAGS_ALLOCATED_FROM_ZONE 0x00020000
118 #define SRB_FLAGS_SGLIST_FROM_POOL 0x00040000
119 #define SRB_FLAGS_BYPASS_LOCKED_QUEUE 0x00080000
120 #define SRB_FLAGS_NO_KEEP_AWAKE 0x00100000
121 #define SRB_FLAGS_PORT_DRIVER_ALLOCSENSE 0x00200000
122 #define SRB_FLAGS_PORT_DRIVER_SENSEHASPORT 0x00400000
123 #define SRB_FLAGS_DONT_START_NEXT_PACKET 0x00800000
124 #define SRB_FLAGS_PORT_DRIVER_RESERVED 0x0F000000
125 #define SRB_FLAGS_CLASS_DRIVER_RESERVED 0xF0000000
127 #define SP_UNTAGGED ((unsigned char) ~0)
128 #define SRB_SIMPLE_TAG_REQUEST 0x20
131 * Platform neutral description of a scsi request -
132 * this remains the same across the write regardless of 32/64 bit
133 * note: it's patterned off the SCSI_PASS_THROUGH structure
135 #define STORVSC_MAX_CMD_LEN 0x10
137 #define POST_WIN7_STORVSC_SENSE_BUFFER_SIZE 0x14
138 #define PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE 0x12
140 #define STORVSC_SENSE_BUFFER_SIZE 0x14
141 #define STORVSC_MAX_BUF_LEN_WITH_PADDING 0x14
144 * Sense buffer size changed in win8; have a run-time
145 * variable to track the size we should use. This value will
146 * likely change during protocol negotiation but it is valid
147 * to start by assuming pre-Win8.
149 static int sense_buffer_size = PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE;
152 * The storage protocol version is determined during the
153 * initial exchange with the host. It will indicate which
154 * storage functionality is available in the host.
156 static int vmstor_proto_version;
158 #define STORVSC_LOGGING_NONE 0
159 #define STORVSC_LOGGING_ERROR 1
160 #define STORVSC_LOGGING_WARN 2
162 static int logging_level = STORVSC_LOGGING_ERROR;
163 module_param(logging_level, int, S_IRUGO|S_IWUSR);
164 MODULE_PARM_DESC(logging_level,
165 "Logging level, 0 - None, 1 - Error (default), 2 - Warning.");
167 static inline bool do_logging(int level)
169 return logging_level >= level;
172 #define storvsc_log(dev, level, fmt, ...) \
174 if (do_logging(level)) \
175 dev_warn(&(dev)->device, fmt, ##__VA_ARGS__); \
178 struct vmscsi_win8_extension {
180 * The following were added in Windows 8
190 struct vmscsi_request {
201 u8 sense_info_length;
205 u32 data_transfer_length;
208 u8 cdb[STORVSC_MAX_CMD_LEN];
209 u8 sense_data[STORVSC_SENSE_BUFFER_SIZE];
210 u8 reserved_array[STORVSC_MAX_BUF_LEN_WITH_PADDING];
213 * The following was added in win8.
215 struct vmscsi_win8_extension win8_extension;
217 } __attribute((packed));
220 * The list of storage protocols in order of preference.
222 struct vmstor_protocol {
223 int protocol_version;
224 int sense_buffer_size;
225 int vmscsi_size_delta;
229 static const struct vmstor_protocol vmstor_protocols[] = {
231 VMSTOR_PROTO_VERSION_WIN10,
232 POST_WIN7_STORVSC_SENSE_BUFFER_SIZE,
236 VMSTOR_PROTO_VERSION_WIN8_1,
237 POST_WIN7_STORVSC_SENSE_BUFFER_SIZE,
241 VMSTOR_PROTO_VERSION_WIN8,
242 POST_WIN7_STORVSC_SENSE_BUFFER_SIZE,
246 VMSTOR_PROTO_VERSION_WIN7,
247 PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE,
248 sizeof(struct vmscsi_win8_extension),
251 VMSTOR_PROTO_VERSION_WIN6,
252 PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE,
253 sizeof(struct vmscsi_win8_extension),
259 * This structure is sent during the initialization phase to get the different
260 * properties of the channel.
263 #define STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL 0x1
265 struct vmstorage_channel_properties {
271 u32 max_transfer_bytes;
276 /* This structure is sent during the storage protocol negotiations. */
277 struct vmstorage_protocol_version {
278 /* Major (MSW) and minor (LSW) version numbers. */
282 * Revision number is auto-incremented whenever this file is changed
283 * (See FILL_VMSTOR_REVISION macro above). Mismatch does not
284 * definitely indicate incompatibility--but it does indicate mismatched
286 * This is only used on the windows side. Just set it to 0.
291 /* Channel Property Flags */
292 #define STORAGE_CHANNEL_REMOVABLE_FLAG 0x1
293 #define STORAGE_CHANNEL_EMULATED_IDE_FLAG 0x2
295 struct vstor_packet {
296 /* Requested operation type */
297 enum vstor_packet_operation operation;
299 /* Flags - see below for values */
302 /* Status of the request returned from the server side. */
305 /* Data payload area */
308 * Structure used to forward SCSI commands from the
309 * client to the server.
311 struct vmscsi_request vm_srb;
313 /* Structure used to query channel properties. */
314 struct vmstorage_channel_properties storage_channel_properties;
316 /* Used during version negotiations. */
317 struct vmstorage_protocol_version version;
319 /* Fibre channel address packet */
320 struct hv_fc_wwn_packet wwn_packet;
322 /* Number of sub-channels to create */
323 u16 sub_channel_count;
325 /* This will be the maximum of the union members */
333 * This flag indicates that the server should send back a completion for this
337 #define REQUEST_COMPLETION_FLAG 0x1
339 /* Matches Windows-end */
340 enum storvsc_request_type {
347 * SRB status codes and masks; a subset of the codes used here.
350 #define SRB_STATUS_AUTOSENSE_VALID 0x80
351 #define SRB_STATUS_QUEUE_FROZEN 0x40
352 #define SRB_STATUS_INVALID_LUN 0x20
353 #define SRB_STATUS_SUCCESS 0x01
354 #define SRB_STATUS_ABORTED 0x02
355 #define SRB_STATUS_ERROR 0x04
356 #define SRB_STATUS_DATA_OVERRUN 0x12
358 #define SRB_STATUS(status) \
359 (status & ~(SRB_STATUS_AUTOSENSE_VALID | SRB_STATUS_QUEUE_FROZEN))
361 * This is the end of Protocol specific defines.
364 static int storvsc_ringbuffer_size = (128 * 1024);
365 static u32 max_outstanding_req_per_channel;
366 static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth);
368 static int storvsc_vcpus_per_sub_channel = 4;
370 module_param(storvsc_ringbuffer_size, int, S_IRUGO);
371 MODULE_PARM_DESC(storvsc_ringbuffer_size, "Ring buffer size (bytes)");
373 module_param(storvsc_vcpus_per_sub_channel, int, S_IRUGO);
374 MODULE_PARM_DESC(storvsc_vcpus_per_sub_channel, "Ratio of VCPUs to subchannels");
376 static int ring_avail_percent_lowater = 10;
377 module_param(ring_avail_percent_lowater, int, S_IRUGO);
378 MODULE_PARM_DESC(ring_avail_percent_lowater,
379 "Select a channel if available ring size > this in percent");
382 * Timeout in seconds for all devices managed by this driver.
384 static int storvsc_timeout = 180;
386 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
387 static struct scsi_transport_template *fc_transport_template;
390 static struct scsi_host_template scsi_driver;
391 static void storvsc_on_channel_callback(void *context);
393 #define STORVSC_MAX_LUNS_PER_TARGET 255
394 #define STORVSC_MAX_TARGETS 2
395 #define STORVSC_MAX_CHANNELS 8
397 #define STORVSC_FC_MAX_LUNS_PER_TARGET 255
398 #define STORVSC_FC_MAX_TARGETS 128
399 #define STORVSC_FC_MAX_CHANNELS 8
401 #define STORVSC_IDE_MAX_LUNS_PER_TARGET 64
402 #define STORVSC_IDE_MAX_TARGETS 1
403 #define STORVSC_IDE_MAX_CHANNELS 1
405 struct storvsc_cmd_request {
406 struct scsi_cmnd *cmd;
408 struct hv_device *device;
410 /* Synchronize the request/response if needed */
411 struct completion wait_event;
413 struct vmbus_channel_packet_multipage_buffer mpb;
414 struct vmbus_packet_mpb_array *payload;
417 struct vstor_packet vstor_packet;
421 /* A storvsc device is a device object that contains a vmbus channel */
422 struct storvsc_device {
423 struct hv_device *device;
427 atomic_t num_outstanding_req;
428 struct Scsi_Host *host;
430 wait_queue_head_t waiting_to_drain;
433 * Each unique Port/Path/Target represents 1 channel ie scsi
434 * controller. In reality, the pathid, targetid is always 0
435 * and the port is set by us
437 unsigned int port_number;
438 unsigned char path_id;
439 unsigned char target_id;
442 * The size of the vmscsi_request has changed in win8. The
443 * additional size is because of new elements added to the
444 * structure. These elements are valid only when we are talking
446 * Track the correction to size we need to apply. This value
447 * will likely change during protocol negotiation but it is
448 * valid to start by assuming pre-Win8.
450 int vmscsi_size_delta;
453 * Max I/O, the device can support.
455 u32 max_transfer_bytes;
457 * Number of sub-channels we will open.
460 struct vmbus_channel **stor_chns;
462 * Mask of CPUs bound to subchannels.
464 struct cpumask alloced_cpus;
466 * Serializes modifications of stor_chns[] from storvsc_do_io()
467 * and storvsc_change_target_cpu().
470 /* Used for vsc/vsp channel reset process */
471 struct storvsc_cmd_request init_request;
472 struct storvsc_cmd_request reset_request;
474 * Currently active port and node names for FC devices.
478 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
479 struct fc_rport *rport;
483 struct hv_host_device {
484 struct hv_device *dev;
487 unsigned char target;
488 struct workqueue_struct *handle_error_wq;
489 struct work_struct host_scan_work;
490 struct Scsi_Host *host;
493 struct storvsc_scan_work {
494 struct work_struct work;
495 struct Scsi_Host *host;
500 static void storvsc_device_scan(struct work_struct *work)
502 struct storvsc_scan_work *wrk;
503 struct scsi_device *sdev;
505 wrk = container_of(work, struct storvsc_scan_work, work);
507 sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
510 scsi_rescan_device(&sdev->sdev_gendev);
511 scsi_device_put(sdev);
517 static void storvsc_host_scan(struct work_struct *work)
519 struct Scsi_Host *host;
520 struct scsi_device *sdev;
521 struct hv_host_device *host_device =
522 container_of(work, struct hv_host_device, host_scan_work);
524 host = host_device->host;
526 * Before scanning the host, first check to see if any of the
527 * currrently known devices have been hot removed. We issue a
528 * "unit ready" command against all currently known devices.
529 * This I/O will result in an error for devices that have been
530 * removed. As part of handling the I/O error, we remove the device.
532 * When a LUN is added or removed, the host sends us a signal to
533 * scan the host. Thus we are forced to discover the LUNs that
534 * may have been removed this way.
536 mutex_lock(&host->scan_mutex);
537 shost_for_each_device(sdev, host)
538 scsi_test_unit_ready(sdev, 1, 1, NULL);
539 mutex_unlock(&host->scan_mutex);
541 * Now scan the host to discover LUNs that may have been added.
543 scsi_scan_host(host);
546 static void storvsc_remove_lun(struct work_struct *work)
548 struct storvsc_scan_work *wrk;
549 struct scsi_device *sdev;
551 wrk = container_of(work, struct storvsc_scan_work, work);
552 if (!scsi_host_get(wrk->host))
555 sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
558 scsi_remove_device(sdev);
559 scsi_device_put(sdev);
561 scsi_host_put(wrk->host);
569 * We can get incoming messages from the host that are not in response to
570 * messages that we have sent out. An example of this would be messages
571 * received by the guest to notify dynamic addition/removal of LUNs. To
572 * deal with potential race conditions where the driver may be in the
573 * midst of being unloaded when we might receive an unsolicited message
574 * from the host, we have implemented a mechanism to gurantee sequential
577 * 1) Once the device is marked as being destroyed, we will fail all
579 * 2) We permit incoming messages when the device is being destroyed,
580 * only to properly account for messages already sent out.
583 static inline struct storvsc_device *get_out_stor_device(
584 struct hv_device *device)
586 struct storvsc_device *stor_device;
588 stor_device = hv_get_drvdata(device);
590 if (stor_device && stor_device->destroy)
597 static inline void storvsc_wait_to_drain(struct storvsc_device *dev)
599 dev->drain_notify = true;
600 wait_event(dev->waiting_to_drain,
601 atomic_read(&dev->num_outstanding_req) == 0);
602 dev->drain_notify = false;
605 static inline struct storvsc_device *get_in_stor_device(
606 struct hv_device *device)
608 struct storvsc_device *stor_device;
610 stor_device = hv_get_drvdata(device);
616 * If the device is being destroyed; allow incoming
617 * traffic only to cleanup outstanding requests.
620 if (stor_device->destroy &&
621 (atomic_read(&stor_device->num_outstanding_req) == 0))
629 static void storvsc_change_target_cpu(struct vmbus_channel *channel, u32 old,
632 struct storvsc_device *stor_device;
633 struct vmbus_channel *cur_chn;
634 bool old_is_alloced = false;
635 struct hv_device *device;
639 device = channel->primary_channel ?
640 channel->primary_channel->device_obj
641 : channel->device_obj;
642 stor_device = get_out_stor_device(device);
646 /* See storvsc_do_io() -> get_og_chn(). */
647 spin_lock_irqsave(&stor_device->lock, flags);
650 * Determines if the storvsc device has other channels assigned to
651 * the "old" CPU to update the alloced_cpus mask and the stor_chns
654 if (device->channel != channel && device->channel->target_cpu == old) {
655 cur_chn = device->channel;
656 old_is_alloced = true;
659 list_for_each_entry(cur_chn, &device->channel->sc_list, sc_list) {
660 if (cur_chn == channel)
662 if (cur_chn->target_cpu == old) {
663 old_is_alloced = true;
670 WRITE_ONCE(stor_device->stor_chns[old], cur_chn);
672 cpumask_clear_cpu(old, &stor_device->alloced_cpus);
674 /* "Flush" the stor_chns array. */
675 for_each_possible_cpu(cpu) {
676 if (stor_device->stor_chns[cpu] && !cpumask_test_cpu(
677 cpu, &stor_device->alloced_cpus))
678 WRITE_ONCE(stor_device->stor_chns[cpu], NULL);
681 WRITE_ONCE(stor_device->stor_chns[new], channel);
682 cpumask_set_cpu(new, &stor_device->alloced_cpus);
684 spin_unlock_irqrestore(&stor_device->lock, flags);
687 static void handle_sc_creation(struct vmbus_channel *new_sc)
689 struct hv_device *device = new_sc->primary_channel->device_obj;
690 struct device *dev = &device->device;
691 struct storvsc_device *stor_device;
692 struct vmstorage_channel_properties props;
695 stor_device = get_out_stor_device(device);
699 memset(&props, 0, sizeof(struct vmstorage_channel_properties));
702 * The size of vmbus_requestor is an upper bound on the number of requests
703 * that can be in-progress at any one time across all channels.
705 new_sc->rqstor_size = scsi_driver.can_queue;
707 ret = vmbus_open(new_sc,
708 storvsc_ringbuffer_size,
709 storvsc_ringbuffer_size,
711 sizeof(struct vmstorage_channel_properties),
712 storvsc_on_channel_callback, new_sc);
714 /* In case vmbus_open() fails, we don't use the sub-channel. */
716 dev_err(dev, "Failed to open sub-channel: err=%d\n", ret);
720 new_sc->change_target_cpu_callback = storvsc_change_target_cpu;
722 /* Add the sub-channel to the array of available channels. */
723 stor_device->stor_chns[new_sc->target_cpu] = new_sc;
724 cpumask_set_cpu(new_sc->target_cpu, &stor_device->alloced_cpus);
727 static void handle_multichannel_storage(struct hv_device *device, int max_chns)
729 struct device *dev = &device->device;
730 struct storvsc_device *stor_device;
732 struct storvsc_cmd_request *request;
733 struct vstor_packet *vstor_packet;
737 * If the number of CPUs is artificially restricted, such as
738 * with maxcpus=1 on the kernel boot line, Hyper-V could offer
739 * sub-channels >= the number of CPUs. These sub-channels
740 * should not be created. The primary channel is already created
741 * and assigned to one CPU, so check against # CPUs - 1.
743 num_sc = min((int)(num_online_cpus() - 1), max_chns);
747 stor_device = get_out_stor_device(device);
751 stor_device->num_sc = num_sc;
752 request = &stor_device->init_request;
753 vstor_packet = &request->vstor_packet;
756 * Establish a handler for dealing with subchannels.
758 vmbus_set_sc_create_callback(device->channel, handle_sc_creation);
761 * Request the host to create sub-channels.
763 memset(request, 0, sizeof(struct storvsc_cmd_request));
764 init_completion(&request->wait_event);
765 vstor_packet->operation = VSTOR_OPERATION_CREATE_SUB_CHANNELS;
766 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
767 vstor_packet->sub_channel_count = num_sc;
769 ret = vmbus_sendpacket(device->channel, vstor_packet,
770 (sizeof(struct vstor_packet) -
771 stor_device->vmscsi_size_delta),
772 (unsigned long)request,
774 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
777 dev_err(dev, "Failed to create sub-channel: err=%d\n", ret);
781 t = wait_for_completion_timeout(&request->wait_event, 10*HZ);
783 dev_err(dev, "Failed to create sub-channel: timed out\n");
787 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
788 vstor_packet->status != 0) {
789 dev_err(dev, "Failed to create sub-channel: op=%d, sts=%d\n",
790 vstor_packet->operation, vstor_packet->status);
795 * We need to do nothing here, because vmbus_process_offer()
796 * invokes channel->sc_creation_callback, which will open and use
797 * the sub-channel(s).
801 static void cache_wwn(struct storvsc_device *stor_device,
802 struct vstor_packet *vstor_packet)
805 * Cache the currently active port and node ww names.
807 if (vstor_packet->wwn_packet.primary_active) {
808 stor_device->node_name =
809 wwn_to_u64(vstor_packet->wwn_packet.primary_node_wwn);
810 stor_device->port_name =
811 wwn_to_u64(vstor_packet->wwn_packet.primary_port_wwn);
813 stor_device->node_name =
814 wwn_to_u64(vstor_packet->wwn_packet.secondary_node_wwn);
815 stor_device->port_name =
816 wwn_to_u64(vstor_packet->wwn_packet.secondary_port_wwn);
821 static int storvsc_execute_vstor_op(struct hv_device *device,
822 struct storvsc_cmd_request *request,
825 struct storvsc_device *stor_device;
826 struct vstor_packet *vstor_packet;
829 stor_device = get_out_stor_device(device);
833 vstor_packet = &request->vstor_packet;
835 init_completion(&request->wait_event);
836 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
838 ret = vmbus_sendpacket(device->channel, vstor_packet,
839 (sizeof(struct vstor_packet) -
840 stor_device->vmscsi_size_delta),
841 (unsigned long)request,
843 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
847 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
854 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
855 vstor_packet->status != 0)
861 static int storvsc_channel_init(struct hv_device *device, bool is_fc)
863 struct storvsc_device *stor_device;
864 struct storvsc_cmd_request *request;
865 struct vstor_packet *vstor_packet;
868 bool process_sub_channels = false;
870 stor_device = get_out_stor_device(device);
874 request = &stor_device->init_request;
875 vstor_packet = &request->vstor_packet;
878 * Now, initiate the vsc/vsp initialization protocol on the open
881 memset(request, 0, sizeof(struct storvsc_cmd_request));
882 vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION;
883 ret = storvsc_execute_vstor_op(device, request, true);
887 * Query host supported protocol version.
890 for (i = 0; i < ARRAY_SIZE(vmstor_protocols); i++) {
891 /* reuse the packet for version range supported */
892 memset(vstor_packet, 0, sizeof(struct vstor_packet));
893 vstor_packet->operation =
894 VSTOR_OPERATION_QUERY_PROTOCOL_VERSION;
896 vstor_packet->version.major_minor =
897 vmstor_protocols[i].protocol_version;
900 * The revision number is only used in Windows; set it to 0.
902 vstor_packet->version.revision = 0;
903 ret = storvsc_execute_vstor_op(device, request, false);
907 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO)
910 if (vstor_packet->status == 0) {
911 vmstor_proto_version =
912 vmstor_protocols[i].protocol_version;
915 vmstor_protocols[i].sense_buffer_size;
917 stor_device->vmscsi_size_delta =
918 vmstor_protocols[i].vmscsi_size_delta;
924 if (vstor_packet->status != 0)
928 memset(vstor_packet, 0, sizeof(struct vstor_packet));
929 vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES;
930 ret = storvsc_execute_vstor_op(device, request, true);
935 * Check to see if multi-channel support is there.
936 * Hosts that implement protocol version of 5.1 and above
937 * support multi-channel.
939 max_chns = vstor_packet->storage_channel_properties.max_channel_cnt;
942 * Allocate state to manage the sub-channels.
943 * We allocate an array based on the numbers of possible CPUs
944 * (Hyper-V does not support cpu online/offline).
945 * This Array will be sparseley populated with unique
946 * channels - primary + sub-channels.
947 * We will however populate all the slots to evenly distribute
950 stor_device->stor_chns = kcalloc(num_possible_cpus(), sizeof(void *),
952 if (stor_device->stor_chns == NULL)
955 device->channel->change_target_cpu_callback = storvsc_change_target_cpu;
957 stor_device->stor_chns[device->channel->target_cpu] = device->channel;
958 cpumask_set_cpu(device->channel->target_cpu,
959 &stor_device->alloced_cpus);
961 if (vmstor_proto_version >= VMSTOR_PROTO_VERSION_WIN8) {
962 if (vstor_packet->storage_channel_properties.flags &
963 STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL)
964 process_sub_channels = true;
966 stor_device->max_transfer_bytes =
967 vstor_packet->storage_channel_properties.max_transfer_bytes;
973 * For FC devices retrieve FC HBA data.
975 memset(vstor_packet, 0, sizeof(struct vstor_packet));
976 vstor_packet->operation = VSTOR_OPERATION_FCHBA_DATA;
977 ret = storvsc_execute_vstor_op(device, request, true);
982 * Cache the currently active port and node ww names.
984 cache_wwn(stor_device, vstor_packet);
988 memset(vstor_packet, 0, sizeof(struct vstor_packet));
989 vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION;
990 ret = storvsc_execute_vstor_op(device, request, true);
994 if (process_sub_channels)
995 handle_multichannel_storage(device, max_chns);
1000 static void storvsc_handle_error(struct vmscsi_request *vm_srb,
1001 struct scsi_cmnd *scmnd,
1002 struct Scsi_Host *host,
1005 struct storvsc_scan_work *wrk;
1006 void (*process_err_fn)(struct work_struct *work);
1007 struct hv_host_device *host_dev = shost_priv(host);
1008 bool do_work = false;
1010 switch (SRB_STATUS(vm_srb->srb_status)) {
1011 case SRB_STATUS_ERROR:
1013 * Let upper layer deal with error when
1014 * sense message is present.
1017 if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID)
1020 * If there is an error; offline the device since all
1021 * error recovery strategies would have already been
1022 * deployed on the host side. However, if the command
1023 * were a pass-through command deal with it appropriately.
1025 switch (scmnd->cmnd[0]) {
1028 set_host_byte(scmnd, DID_PASSTHROUGH);
1031 * On Some Windows hosts TEST_UNIT_READY command can return
1032 * SRB_STATUS_ERROR, let the upper level code deal with it
1033 * based on the sense information.
1035 case TEST_UNIT_READY:
1038 set_host_byte(scmnd, DID_ERROR);
1041 case SRB_STATUS_INVALID_LUN:
1042 set_host_byte(scmnd, DID_NO_CONNECT);
1044 process_err_fn = storvsc_remove_lun;
1046 case SRB_STATUS_ABORTED:
1047 if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID &&
1048 (asc == 0x2a) && (ascq == 0x9)) {
1050 process_err_fn = storvsc_device_scan;
1052 * Retry the I/O that triggered this.
1054 set_host_byte(scmnd, DID_REQUEUE);
1063 * We need to schedule work to process this error; schedule it.
1065 wrk = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC);
1067 set_host_byte(scmnd, DID_TARGET_FAILURE);
1072 wrk->lun = vm_srb->lun;
1073 wrk->tgt_id = vm_srb->target_id;
1074 INIT_WORK(&wrk->work, process_err_fn);
1075 queue_work(host_dev->handle_error_wq, &wrk->work);
1079 static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request,
1080 struct storvsc_device *stor_dev)
1082 struct scsi_cmnd *scmnd = cmd_request->cmd;
1083 struct scsi_sense_hdr sense_hdr;
1084 struct vmscsi_request *vm_srb;
1085 u32 data_transfer_length;
1086 struct Scsi_Host *host;
1087 u32 payload_sz = cmd_request->payload_sz;
1088 void *payload = cmd_request->payload;
1090 host = stor_dev->host;
1092 vm_srb = &cmd_request->vstor_packet.vm_srb;
1093 data_transfer_length = vm_srb->data_transfer_length;
1095 scmnd->result = vm_srb->scsi_status;
1097 if (scmnd->result) {
1098 if (scsi_normalize_sense(scmnd->sense_buffer,
1099 SCSI_SENSE_BUFFERSIZE, &sense_hdr) &&
1100 !(sense_hdr.sense_key == NOT_READY &&
1101 sense_hdr.asc == 0x03A) &&
1102 do_logging(STORVSC_LOGGING_ERROR))
1103 scsi_print_sense_hdr(scmnd->device, "storvsc",
1107 if (vm_srb->srb_status != SRB_STATUS_SUCCESS) {
1108 storvsc_handle_error(vm_srb, scmnd, host, sense_hdr.asc,
1111 * The Windows driver set data_transfer_length on
1112 * SRB_STATUS_DATA_OVERRUN. On other errors, this value
1113 * is untouched. In these cases we set it to 0.
1115 if (vm_srb->srb_status != SRB_STATUS_DATA_OVERRUN)
1116 data_transfer_length = 0;
1119 /* Validate data_transfer_length (from Hyper-V) */
1120 if (data_transfer_length > cmd_request->payload->range.len)
1121 data_transfer_length = cmd_request->payload->range.len;
1123 scsi_set_resid(scmnd,
1124 cmd_request->payload->range.len - data_transfer_length);
1126 scmnd->scsi_done(scmnd);
1129 sizeof(struct vmbus_channel_packet_multipage_buffer))
1133 static void storvsc_on_io_completion(struct storvsc_device *stor_device,
1134 struct vstor_packet *vstor_packet,
1135 struct storvsc_cmd_request *request)
1137 struct vstor_packet *stor_pkt;
1138 struct hv_device *device = stor_device->device;
1140 stor_pkt = &request->vstor_packet;
1143 * The current SCSI handling on the host side does
1144 * not correctly handle:
1145 * INQUIRY command with page code parameter set to 0x80
1146 * MODE_SENSE command with cmd[2] == 0x1c
1148 * Setup srb and scsi status so this won't be fatal.
1149 * We do this so we can distinguish truly fatal failues
1150 * (srb status == 0x4) and off-line the device in that case.
1153 if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) ||
1154 (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) {
1155 vstor_packet->vm_srb.scsi_status = 0;
1156 vstor_packet->vm_srb.srb_status = SRB_STATUS_SUCCESS;
1160 /* Copy over the status...etc */
1161 stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status;
1162 stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status;
1164 /* Validate sense_info_length (from Hyper-V) */
1165 if (vstor_packet->vm_srb.sense_info_length > sense_buffer_size)
1166 vstor_packet->vm_srb.sense_info_length = sense_buffer_size;
1168 stor_pkt->vm_srb.sense_info_length =
1169 vstor_packet->vm_srb.sense_info_length;
1171 if (vstor_packet->vm_srb.scsi_status != 0 ||
1172 vstor_packet->vm_srb.srb_status != SRB_STATUS_SUCCESS)
1173 storvsc_log(device, STORVSC_LOGGING_WARN,
1174 "cmd 0x%x scsi status 0x%x srb status 0x%x\n",
1175 stor_pkt->vm_srb.cdb[0],
1176 vstor_packet->vm_srb.scsi_status,
1177 vstor_packet->vm_srb.srb_status);
1179 if ((vstor_packet->vm_srb.scsi_status & 0xFF) == 0x02) {
1180 /* CHECK_CONDITION */
1181 if (vstor_packet->vm_srb.srb_status &
1182 SRB_STATUS_AUTOSENSE_VALID) {
1183 /* autosense data available */
1185 storvsc_log(device, STORVSC_LOGGING_WARN,
1186 "stor pkt %p autosense data valid - len %d\n",
1187 request, vstor_packet->vm_srb.sense_info_length);
1189 memcpy(request->cmd->sense_buffer,
1190 vstor_packet->vm_srb.sense_data,
1191 vstor_packet->vm_srb.sense_info_length);
1196 stor_pkt->vm_srb.data_transfer_length =
1197 vstor_packet->vm_srb.data_transfer_length;
1199 storvsc_command_completion(request, stor_device);
1201 if (atomic_dec_and_test(&stor_device->num_outstanding_req) &&
1202 stor_device->drain_notify)
1203 wake_up(&stor_device->waiting_to_drain);
1208 static void storvsc_on_receive(struct storvsc_device *stor_device,
1209 struct vstor_packet *vstor_packet,
1210 struct storvsc_cmd_request *request)
1212 struct hv_host_device *host_dev;
1213 switch (vstor_packet->operation) {
1214 case VSTOR_OPERATION_COMPLETE_IO:
1215 storvsc_on_io_completion(stor_device, vstor_packet, request);
1218 case VSTOR_OPERATION_REMOVE_DEVICE:
1219 case VSTOR_OPERATION_ENUMERATE_BUS:
1220 host_dev = shost_priv(stor_device->host);
1222 host_dev->handle_error_wq, &host_dev->host_scan_work);
1225 case VSTOR_OPERATION_FCHBA_DATA:
1226 cache_wwn(stor_device, vstor_packet);
1227 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1228 fc_host_node_name(stor_device->host) = stor_device->node_name;
1229 fc_host_port_name(stor_device->host) = stor_device->port_name;
1237 static void storvsc_on_channel_callback(void *context)
1239 struct vmbus_channel *channel = (struct vmbus_channel *)context;
1240 const struct vmpacket_descriptor *desc;
1241 struct hv_device *device;
1242 struct storvsc_device *stor_device;
1244 if (channel->primary_channel != NULL)
1245 device = channel->primary_channel->device_obj;
1247 device = channel->device_obj;
1249 stor_device = get_in_stor_device(device);
1253 foreach_vmbus_pkt(desc, channel) {
1254 void *packet = hv_pkt_data(desc);
1255 struct storvsc_cmd_request *request;
1258 cmd_rqst = vmbus_request_addr(&channel->requestor,
1260 if (cmd_rqst == VMBUS_RQST_ERROR) {
1261 dev_err(&device->device,
1262 "Incorrect transaction id\n");
1266 request = (struct storvsc_cmd_request *)(unsigned long)cmd_rqst;
1268 if (hv_pkt_datalen(desc) < sizeof(struct vstor_packet) -
1269 stor_device->vmscsi_size_delta) {
1270 dev_err(&device->device, "Invalid packet len\n");
1274 if (request == &stor_device->init_request ||
1275 request == &stor_device->reset_request) {
1276 memcpy(&request->vstor_packet, packet,
1277 (sizeof(struct vstor_packet) - stor_device->vmscsi_size_delta));
1278 complete(&request->wait_event);
1280 storvsc_on_receive(stor_device, packet, request);
1285 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size,
1288 struct vmstorage_channel_properties props;
1291 memset(&props, 0, sizeof(struct vmstorage_channel_properties));
1294 * The size of vmbus_requestor is an upper bound on the number of requests
1295 * that can be in-progress at any one time across all channels.
1297 device->channel->rqstor_size = scsi_driver.can_queue;
1299 ret = vmbus_open(device->channel,
1303 sizeof(struct vmstorage_channel_properties),
1304 storvsc_on_channel_callback, device->channel);
1309 ret = storvsc_channel_init(device, is_fc);
1314 static int storvsc_dev_remove(struct hv_device *device)
1316 struct storvsc_device *stor_device;
1318 stor_device = hv_get_drvdata(device);
1320 stor_device->destroy = true;
1322 /* Make sure flag is set before waiting */
1326 * At this point, all outbound traffic should be disable. We
1327 * only allow inbound traffic (responses) to proceed so that
1328 * outstanding requests can be completed.
1331 storvsc_wait_to_drain(stor_device);
1334 * Since we have already drained, we don't need to busy wait
1335 * as was done in final_release_stor_device()
1336 * Note that we cannot set the ext pointer to NULL until
1337 * we have drained - to drain the outgoing packets, we need to
1338 * allow incoming packets.
1340 hv_set_drvdata(device, NULL);
1342 /* Close the channel */
1343 vmbus_close(device->channel);
1345 kfree(stor_device->stor_chns);
1350 static struct vmbus_channel *get_og_chn(struct storvsc_device *stor_device,
1355 const struct cpumask *node_mask;
1356 int num_channels, tgt_cpu;
1358 if (stor_device->num_sc == 0) {
1359 stor_device->stor_chns[q_num] = stor_device->device->channel;
1360 return stor_device->device->channel;
1364 * Our channel array is sparsley populated and we
1365 * initiated I/O on a processor/hw-q that does not
1366 * currently have a designated channel. Fix this.
1367 * The strategy is simple:
1368 * I. Ensure NUMA locality
1369 * II. Distribute evenly (best effort)
1372 node_mask = cpumask_of_node(cpu_to_node(q_num));
1375 for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1376 if (cpumask_test_cpu(tgt_cpu, node_mask))
1379 if (num_channels == 0) {
1380 stor_device->stor_chns[q_num] = stor_device->device->channel;
1381 return stor_device->device->channel;
1385 while (hash_qnum >= num_channels)
1386 hash_qnum -= num_channels;
1388 for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1389 if (!cpumask_test_cpu(tgt_cpu, node_mask))
1391 if (slot == hash_qnum)
1396 stor_device->stor_chns[q_num] = stor_device->stor_chns[tgt_cpu];
1398 return stor_device->stor_chns[q_num];
1402 static int storvsc_do_io(struct hv_device *device,
1403 struct storvsc_cmd_request *request, u16 q_num)
1405 struct storvsc_device *stor_device;
1406 struct vstor_packet *vstor_packet;
1407 struct vmbus_channel *outgoing_channel, *channel;
1408 unsigned long flags;
1410 const struct cpumask *node_mask;
1413 vstor_packet = &request->vstor_packet;
1414 stor_device = get_out_stor_device(device);
1420 request->device = device;
1422 * Select an appropriate channel to send the request out.
1424 /* See storvsc_change_target_cpu(). */
1425 outgoing_channel = READ_ONCE(stor_device->stor_chns[q_num]);
1426 if (outgoing_channel != NULL) {
1427 if (outgoing_channel->target_cpu == q_num) {
1429 * Ideally, we want to pick a different channel if
1430 * available on the same NUMA node.
1432 node_mask = cpumask_of_node(cpu_to_node(q_num));
1433 for_each_cpu_wrap(tgt_cpu,
1434 &stor_device->alloced_cpus, q_num + 1) {
1435 if (!cpumask_test_cpu(tgt_cpu, node_mask))
1437 if (tgt_cpu == q_num)
1439 channel = READ_ONCE(
1440 stor_device->stor_chns[tgt_cpu]);
1441 if (channel == NULL)
1443 if (hv_get_avail_to_write_percent(
1445 > ring_avail_percent_lowater) {
1446 outgoing_channel = channel;
1452 * All the other channels on the same NUMA node are
1453 * busy. Try to use the channel on the current CPU
1455 if (hv_get_avail_to_write_percent(
1456 &outgoing_channel->outbound)
1457 > ring_avail_percent_lowater)
1461 * If we reach here, all the channels on the current
1462 * NUMA node are busy. Try to find a channel in
1465 for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1466 if (cpumask_test_cpu(tgt_cpu, node_mask))
1468 channel = READ_ONCE(
1469 stor_device->stor_chns[tgt_cpu]);
1470 if (channel == NULL)
1472 if (hv_get_avail_to_write_percent(
1474 > ring_avail_percent_lowater) {
1475 outgoing_channel = channel;
1481 spin_lock_irqsave(&stor_device->lock, flags);
1482 outgoing_channel = stor_device->stor_chns[q_num];
1483 if (outgoing_channel != NULL) {
1484 spin_unlock_irqrestore(&stor_device->lock, flags);
1487 outgoing_channel = get_og_chn(stor_device, q_num);
1488 spin_unlock_irqrestore(&stor_device->lock, flags);
1492 vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
1494 vstor_packet->vm_srb.length = (sizeof(struct vmscsi_request) -
1495 stor_device->vmscsi_size_delta);
1498 vstor_packet->vm_srb.sense_info_length = sense_buffer_size;
1501 vstor_packet->vm_srb.data_transfer_length =
1502 request->payload->range.len;
1504 vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
1506 if (request->payload->range.len) {
1508 ret = vmbus_sendpacket_mpb_desc(outgoing_channel,
1509 request->payload, request->payload_sz,
1511 (sizeof(struct vstor_packet) -
1512 stor_device->vmscsi_size_delta),
1513 (unsigned long)request);
1515 ret = vmbus_sendpacket(outgoing_channel, vstor_packet,
1516 (sizeof(struct vstor_packet) -
1517 stor_device->vmscsi_size_delta),
1518 (unsigned long)request,
1520 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1526 atomic_inc(&stor_device->num_outstanding_req);
1531 static int storvsc_device_alloc(struct scsi_device *sdevice)
1534 * Set blist flag to permit the reading of the VPD pages even when
1535 * the target may claim SPC-2 compliance. MSFT targets currently
1536 * claim SPC-2 compliance while they implement post SPC-2 features.
1537 * With this flag we can correctly handle WRITE_SAME_16 issues.
1539 * Hypervisor reports SCSI_UNKNOWN type for DVD ROM device but
1540 * still supports REPORT LUN.
1542 sdevice->sdev_bflags = BLIST_REPORTLUN2 | BLIST_TRY_VPD_PAGES;
1547 static int storvsc_device_configure(struct scsi_device *sdevice)
1549 blk_queue_rq_timeout(sdevice->request_queue, (storvsc_timeout * HZ));
1551 sdevice->no_write_same = 1;
1554 * If the host is WIN8 or WIN8 R2, claim conformance to SPC-3
1555 * if the device is a MSFT virtual device. If the host is
1556 * WIN10 or newer, allow write_same.
1558 if (!strncmp(sdevice->vendor, "Msft", 4)) {
1559 switch (vmstor_proto_version) {
1560 case VMSTOR_PROTO_VERSION_WIN8:
1561 case VMSTOR_PROTO_VERSION_WIN8_1:
1562 sdevice->scsi_level = SCSI_SPC_3;
1566 if (vmstor_proto_version >= VMSTOR_PROTO_VERSION_WIN10)
1567 sdevice->no_write_same = 0;
1573 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev,
1574 sector_t capacity, int *info)
1576 sector_t nsect = capacity;
1577 sector_t cylinders = nsect;
1578 int heads, sectors_pt;
1581 * We are making up these values; let us keep it simple.
1584 sectors_pt = 0x3f; /* Sectors per track */
1585 sector_div(cylinders, heads * sectors_pt);
1586 if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
1590 info[1] = sectors_pt;
1591 info[2] = (int)cylinders;
1596 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
1598 struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
1599 struct hv_device *device = host_dev->dev;
1601 struct storvsc_device *stor_device;
1602 struct storvsc_cmd_request *request;
1603 struct vstor_packet *vstor_packet;
1606 stor_device = get_out_stor_device(device);
1610 request = &stor_device->reset_request;
1611 vstor_packet = &request->vstor_packet;
1612 memset(vstor_packet, 0, sizeof(struct vstor_packet));
1614 init_completion(&request->wait_event);
1616 vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
1617 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
1618 vstor_packet->vm_srb.path_id = stor_device->path_id;
1620 ret = vmbus_sendpacket(device->channel, vstor_packet,
1621 (sizeof(struct vstor_packet) -
1622 stor_device->vmscsi_size_delta),
1623 (unsigned long)&stor_device->reset_request,
1625 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1629 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
1631 return TIMEOUT_ERROR;
1635 * At this point, all outstanding requests in the adapter
1636 * should have been flushed out and return to us
1637 * There is a potential race here where the host may be in
1638 * the process of responding when we return from here.
1639 * Just wait for all in-transit packets to be accounted for
1640 * before we return from here.
1642 storvsc_wait_to_drain(stor_device);
1648 * The host guarantees to respond to each command, although I/O latencies might
1649 * be unbounded on Azure. Reset the timer unconditionally to give the host a
1650 * chance to perform EH.
1652 static enum blk_eh_timer_return storvsc_eh_timed_out(struct scsi_cmnd *scmnd)
1654 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1655 if (scmnd->device->host->transportt == fc_transport_template)
1656 return fc_eh_timed_out(scmnd);
1658 return BLK_EH_RESET_TIMER;
1661 static bool storvsc_scsi_cmd_ok(struct scsi_cmnd *scmnd)
1663 bool allowed = true;
1664 u8 scsi_op = scmnd->cmnd[0];
1667 /* the host does not handle WRITE_SAME, log accident usage */
1670 * smartd sends this command and the host does not handle
1671 * this. So, don't send it.
1674 scmnd->result = DID_ERROR << 16;
1683 static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
1686 struct hv_host_device *host_dev = shost_priv(host);
1687 struct hv_device *dev = host_dev->dev;
1688 struct storvsc_cmd_request *cmd_request = scsi_cmd_priv(scmnd);
1690 struct scatterlist *sgl;
1691 unsigned int sg_count = 0;
1692 struct vmscsi_request *vm_srb;
1693 struct scatterlist *cur_sgl;
1694 struct vmbus_packet_mpb_array *payload;
1698 if (vmstor_proto_version <= VMSTOR_PROTO_VERSION_WIN8) {
1700 * On legacy hosts filter unimplemented commands.
1701 * Future hosts are expected to correctly handle
1702 * unsupported commands. Furthermore, it is
1703 * possible that some of the currently
1704 * unsupported commands maybe supported in
1705 * future versions of the host.
1707 if (!storvsc_scsi_cmd_ok(scmnd)) {
1708 scmnd->scsi_done(scmnd);
1713 /* Setup the cmd request */
1714 cmd_request->cmd = scmnd;
1716 memset(&cmd_request->vstor_packet, 0, sizeof(struct vstor_packet));
1717 vm_srb = &cmd_request->vstor_packet.vm_srb;
1718 vm_srb->win8_extension.time_out_value = 60;
1720 vm_srb->win8_extension.srb_flags |=
1721 SRB_FLAGS_DISABLE_SYNCH_TRANSFER;
1723 if (scmnd->device->tagged_supported) {
1724 vm_srb->win8_extension.srb_flags |=
1725 (SRB_FLAGS_QUEUE_ACTION_ENABLE | SRB_FLAGS_NO_QUEUE_FREEZE);
1726 vm_srb->win8_extension.queue_tag = SP_UNTAGGED;
1727 vm_srb->win8_extension.queue_action = SRB_SIMPLE_TAG_REQUEST;
1731 switch (scmnd->sc_data_direction) {
1733 vm_srb->data_in = WRITE_TYPE;
1734 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_DATA_OUT;
1736 case DMA_FROM_DEVICE:
1737 vm_srb->data_in = READ_TYPE;
1738 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_DATA_IN;
1741 vm_srb->data_in = UNKNOWN_TYPE;
1742 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_NO_DATA_TRANSFER;
1746 * This is DMA_BIDIRECTIONAL or something else we are never
1747 * supposed to see here.
1749 WARN(1, "Unexpected data direction: %d\n",
1750 scmnd->sc_data_direction);
1755 vm_srb->port_number = host_dev->port;
1756 vm_srb->path_id = scmnd->device->channel;
1757 vm_srb->target_id = scmnd->device->id;
1758 vm_srb->lun = scmnd->device->lun;
1760 vm_srb->cdb_length = scmnd->cmd_len;
1762 memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
1764 sgl = (struct scatterlist *)scsi_sglist(scmnd);
1765 sg_count = scsi_sg_count(scmnd);
1767 length = scsi_bufflen(scmnd);
1768 payload = (struct vmbus_packet_mpb_array *)&cmd_request->mpb;
1769 payload_sz = sizeof(cmd_request->mpb);
1772 unsigned int hvpgoff = 0;
1773 unsigned long offset_in_hvpg = sgl->offset & ~HV_HYP_PAGE_MASK;
1774 unsigned int hvpg_count = HVPFN_UP(offset_in_hvpg + length);
1777 if (hvpg_count > MAX_PAGE_BUFFER_COUNT) {
1779 payload_sz = (hvpg_count * sizeof(u64) +
1780 sizeof(struct vmbus_packet_mpb_array));
1781 payload = kzalloc(payload_sz, GFP_ATOMIC);
1783 return SCSI_MLQUEUE_DEVICE_BUSY;
1787 * sgl is a list of PAGEs, and payload->range.pfn_array
1788 * expects the page number in the unit of HV_HYP_PAGE_SIZE (the
1789 * page size that Hyper-V uses, so here we need to divide PAGEs
1790 * into HV_HYP_PAGE in case that PAGE_SIZE > HV_HYP_PAGE_SIZE.
1791 * Besides, payload->range.offset should be the offset in one
1794 payload->range.len = length;
1795 payload->range.offset = offset_in_hvpg;
1796 hvpgoff = sgl->offset >> HV_HYP_PAGE_SHIFT;
1799 for (i = 0; i < hvpg_count; i++) {
1801 * 'i' is the index of hv pages in the payload and
1802 * 'hvpgoff' is the offset (in hv pages) of the first
1803 * hv page in the the first page. The relationship
1804 * between the sum of 'i' and 'hvpgoff' and the offset
1805 * (in hv pages) in a payload page ('hvpgoff_in_page')
1808 * |------------------ PAGE -------------------|
1809 * | NR_HV_HYP_PAGES_IN_PAGE hvpgs in total |
1810 * |hvpg|hvpg| ... |hvpg|... |hvpg|
1812 * +-hvpgoff-+ +-hvpgoff_in_page-+
1814 * +--------------------- i ---------------------------+
1816 unsigned int hvpgoff_in_page =
1817 (i + hvpgoff) % NR_HV_HYP_PAGES_IN_PAGE;
1820 * Two cases that we need to fetch a page:
1821 * 1) i == 0, the first step or
1822 * 2) hvpgoff_in_page == 0, when we reach the boundary
1825 if (hvpgoff_in_page == 0 || i == 0) {
1826 hvpfn = page_to_hvpfn(sg_page(cur_sgl));
1827 cur_sgl = sg_next(cur_sgl);
1830 payload->range.pfn_array[i] = hvpfn + hvpgoff_in_page;
1834 cmd_request->payload = payload;
1835 cmd_request->payload_sz = payload_sz;
1837 /* Invokes the vsc to start an IO */
1838 ret = storvsc_do_io(dev, cmd_request, get_cpu());
1841 if (ret == -EAGAIN) {
1842 if (payload_sz > sizeof(cmd_request->mpb))
1845 return SCSI_MLQUEUE_DEVICE_BUSY;
1851 static struct scsi_host_template scsi_driver = {
1852 .module = THIS_MODULE,
1853 .name = "storvsc_host_t",
1854 .cmd_size = sizeof(struct storvsc_cmd_request),
1855 .bios_param = storvsc_get_chs,
1856 .queuecommand = storvsc_queuecommand,
1857 .eh_host_reset_handler = storvsc_host_reset_handler,
1858 .proc_name = "storvsc_host",
1859 .eh_timed_out = storvsc_eh_timed_out,
1860 .slave_alloc = storvsc_device_alloc,
1861 .slave_configure = storvsc_device_configure,
1862 .cmd_per_lun = 2048,
1864 /* Make sure we dont get a sg segment crosses a page boundary */
1865 .dma_boundary = PAGE_SIZE-1,
1866 /* Ensure there are no gaps in presented sgls */
1867 .virt_boundary_mask = PAGE_SIZE-1,
1869 .track_queue_depth = 1,
1870 .change_queue_depth = storvsc_change_queue_depth,
1879 static const struct hv_vmbus_device_id id_table[] = {
1882 .driver_data = SCSI_GUID
1886 .driver_data = IDE_GUID
1888 /* Fibre Channel GUID */
1891 .driver_data = SFC_GUID
1896 MODULE_DEVICE_TABLE(vmbus, id_table);
1898 static const struct { guid_t guid; } fc_guid = { HV_SYNTHFC_GUID };
1900 static bool hv_dev_is_fc(struct hv_device *hv_dev)
1902 return guid_equal(&fc_guid.guid, &hv_dev->dev_type);
1905 static int storvsc_probe(struct hv_device *device,
1906 const struct hv_vmbus_device_id *dev_id)
1909 int num_cpus = num_online_cpus();
1910 struct Scsi_Host *host;
1911 struct hv_host_device *host_dev;
1912 bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false);
1913 bool is_fc = ((dev_id->driver_data == SFC_GUID) ? true : false);
1915 struct storvsc_device *stor_device;
1916 int max_luns_per_target;
1919 int max_sub_channels = 0;
1922 * Based on the windows host we are running on,
1923 * set state to properly communicate with the host.
1926 if (vmbus_proto_version < VERSION_WIN8) {
1927 max_luns_per_target = STORVSC_IDE_MAX_LUNS_PER_TARGET;
1928 max_targets = STORVSC_IDE_MAX_TARGETS;
1929 max_channels = STORVSC_IDE_MAX_CHANNELS;
1931 max_luns_per_target = STORVSC_MAX_LUNS_PER_TARGET;
1932 max_targets = STORVSC_MAX_TARGETS;
1933 max_channels = STORVSC_MAX_CHANNELS;
1935 * On Windows8 and above, we support sub-channels for storage
1936 * on SCSI and FC controllers.
1937 * The number of sub-channels offerred is based on the number of
1938 * VCPUs in the guest.
1942 (num_cpus - 1) / storvsc_vcpus_per_sub_channel;
1945 scsi_driver.can_queue = max_outstanding_req_per_channel *
1946 (max_sub_channels + 1) *
1947 (100 - ring_avail_percent_lowater) / 100;
1949 host = scsi_host_alloc(&scsi_driver,
1950 sizeof(struct hv_host_device));
1954 host_dev = shost_priv(host);
1955 memset(host_dev, 0, sizeof(struct hv_host_device));
1957 host_dev->port = host->host_no;
1958 host_dev->dev = device;
1959 host_dev->host = host;
1962 stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
1968 stor_device->destroy = false;
1969 init_waitqueue_head(&stor_device->waiting_to_drain);
1970 stor_device->device = device;
1971 stor_device->host = host;
1972 stor_device->vmscsi_size_delta = sizeof(struct vmscsi_win8_extension);
1973 spin_lock_init(&stor_device->lock);
1974 hv_set_drvdata(device, stor_device);
1976 stor_device->port_number = host->host_no;
1977 ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size, is_fc);
1981 host_dev->path = stor_device->path_id;
1982 host_dev->target = stor_device->target_id;
1984 switch (dev_id->driver_data) {
1986 host->max_lun = STORVSC_FC_MAX_LUNS_PER_TARGET;
1987 host->max_id = STORVSC_FC_MAX_TARGETS;
1988 host->max_channel = STORVSC_FC_MAX_CHANNELS - 1;
1989 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1990 host->transportt = fc_transport_template;
1995 host->max_lun = max_luns_per_target;
1996 host->max_id = max_targets;
1997 host->max_channel = max_channels - 1;
2001 host->max_lun = STORVSC_IDE_MAX_LUNS_PER_TARGET;
2002 host->max_id = STORVSC_IDE_MAX_TARGETS;
2003 host->max_channel = STORVSC_IDE_MAX_CHANNELS - 1;
2006 /* max cmd length */
2007 host->max_cmd_len = STORVSC_MAX_CMD_LEN;
2010 * set the table size based on the info we got
2013 host->sg_tablesize = (stor_device->max_transfer_bytes >> PAGE_SHIFT);
2015 * For non-IDE disks, the host supports multiple channels.
2016 * Set the number of HW queues we are supporting.
2019 host->nr_hw_queues = num_present_cpus();
2022 * Set the error handler work queue.
2024 host_dev->handle_error_wq =
2025 alloc_ordered_workqueue("storvsc_error_wq_%d",
2028 if (!host_dev->handle_error_wq) {
2032 INIT_WORK(&host_dev->host_scan_work, storvsc_host_scan);
2033 /* Register the HBA and start the scsi bus scan */
2034 ret = scsi_add_host(host, &device->device);
2039 scsi_scan_host(host);
2041 target = (device->dev_instance.b[5] << 8 |
2042 device->dev_instance.b[4]);
2043 ret = scsi_add_device(host, 0, target, 0);
2047 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2048 if (host->transportt == fc_transport_template) {
2049 struct fc_rport_identifiers ids = {
2050 .roles = FC_PORT_ROLE_FCP_DUMMY_INITIATOR,
2053 fc_host_node_name(host) = stor_device->node_name;
2054 fc_host_port_name(host) = stor_device->port_name;
2055 stor_device->rport = fc_remote_port_add(host, 0, &ids);
2056 if (!stor_device->rport) {
2065 scsi_remove_host(host);
2068 destroy_workqueue(host_dev->handle_error_wq);
2072 * Once we have connected with the host, we would need to
2073 * to invoke storvsc_dev_remove() to rollback this state and
2074 * this call also frees up the stor_device; hence the jump around
2077 storvsc_dev_remove(device);
2081 kfree(stor_device->stor_chns);
2085 scsi_host_put(host);
2089 /* Change a scsi target's queue depth */
2090 static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth)
2092 if (queue_depth > scsi_driver.can_queue)
2093 queue_depth = scsi_driver.can_queue;
2095 return scsi_change_queue_depth(sdev, queue_depth);
2098 static int storvsc_remove(struct hv_device *dev)
2100 struct storvsc_device *stor_device = hv_get_drvdata(dev);
2101 struct Scsi_Host *host = stor_device->host;
2102 struct hv_host_device *host_dev = shost_priv(host);
2104 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2105 if (host->transportt == fc_transport_template) {
2106 fc_remote_port_delete(stor_device->rport);
2107 fc_remove_host(host);
2110 destroy_workqueue(host_dev->handle_error_wq);
2111 scsi_remove_host(host);
2112 storvsc_dev_remove(dev);
2113 scsi_host_put(host);
2118 static int storvsc_suspend(struct hv_device *hv_dev)
2120 struct storvsc_device *stor_device = hv_get_drvdata(hv_dev);
2121 struct Scsi_Host *host = stor_device->host;
2122 struct hv_host_device *host_dev = shost_priv(host);
2124 storvsc_wait_to_drain(stor_device);
2126 drain_workqueue(host_dev->handle_error_wq);
2128 vmbus_close(hv_dev->channel);
2130 kfree(stor_device->stor_chns);
2131 stor_device->stor_chns = NULL;
2133 cpumask_clear(&stor_device->alloced_cpus);
2138 static int storvsc_resume(struct hv_device *hv_dev)
2142 ret = storvsc_connect_to_vsp(hv_dev, storvsc_ringbuffer_size,
2143 hv_dev_is_fc(hv_dev));
2147 static struct hv_driver storvsc_drv = {
2148 .name = KBUILD_MODNAME,
2149 .id_table = id_table,
2150 .probe = storvsc_probe,
2151 .remove = storvsc_remove,
2152 .suspend = storvsc_suspend,
2153 .resume = storvsc_resume,
2155 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
2159 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2160 static struct fc_function_template fc_transport_functions = {
2161 .show_host_node_name = 1,
2162 .show_host_port_name = 1,
2166 static int __init storvsc_drv_init(void)
2171 * Divide the ring buffer data size (which is 1 page less
2172 * than the ring buffer size since that page is reserved for
2173 * the ring buffer indices) by the max request size (which is
2174 * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
2176 * The computation underestimates max_outstanding_req_per_channel
2177 * for Win7 and older hosts because it does not take into account
2178 * the vmscsi_size_delta correction to the max request size.
2180 max_outstanding_req_per_channel =
2181 ((storvsc_ringbuffer_size - PAGE_SIZE) /
2182 ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
2183 sizeof(struct vstor_packet) + sizeof(u64),
2186 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2187 fc_transport_template = fc_attach_transport(&fc_transport_functions);
2188 if (!fc_transport_template)
2192 ret = vmbus_driver_register(&storvsc_drv);
2194 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2196 fc_release_transport(fc_transport_template);
2202 static void __exit storvsc_drv_exit(void)
2204 vmbus_driver_unregister(&storvsc_drv);
2205 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2206 fc_release_transport(fc_transport_template);
2210 MODULE_LICENSE("GPL");
2211 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
2212 module_init(storvsc_drv_init);
2213 module_exit(storvsc_drv_exit);