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;
369 static unsigned int storvsc_max_hw_queues;
371 module_param(storvsc_ringbuffer_size, int, S_IRUGO);
372 MODULE_PARM_DESC(storvsc_ringbuffer_size, "Ring buffer size (bytes)");
374 module_param(storvsc_max_hw_queues, uint, 0644);
375 MODULE_PARM_DESC(storvsc_max_hw_queues, "Maximum number of hardware queues");
377 module_param(storvsc_vcpus_per_sub_channel, int, S_IRUGO);
378 MODULE_PARM_DESC(storvsc_vcpus_per_sub_channel, "Ratio of VCPUs to subchannels");
380 static int ring_avail_percent_lowater = 10;
381 module_param(ring_avail_percent_lowater, int, S_IRUGO);
382 MODULE_PARM_DESC(ring_avail_percent_lowater,
383 "Select a channel if available ring size > this in percent");
386 * Timeout in seconds for all devices managed by this driver.
388 static int storvsc_timeout = 180;
390 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
391 static struct scsi_transport_template *fc_transport_template;
394 static struct scsi_host_template scsi_driver;
395 static void storvsc_on_channel_callback(void *context);
397 #define STORVSC_MAX_LUNS_PER_TARGET 255
398 #define STORVSC_MAX_TARGETS 2
399 #define STORVSC_MAX_CHANNELS 8
401 #define STORVSC_FC_MAX_LUNS_PER_TARGET 255
402 #define STORVSC_FC_MAX_TARGETS 128
403 #define STORVSC_FC_MAX_CHANNELS 8
405 #define STORVSC_IDE_MAX_LUNS_PER_TARGET 64
406 #define STORVSC_IDE_MAX_TARGETS 1
407 #define STORVSC_IDE_MAX_CHANNELS 1
409 struct storvsc_cmd_request {
410 struct scsi_cmnd *cmd;
412 struct hv_device *device;
414 /* Synchronize the request/response if needed */
415 struct completion wait_event;
417 struct vmbus_channel_packet_multipage_buffer mpb;
418 struct vmbus_packet_mpb_array *payload;
421 struct vstor_packet vstor_packet;
425 /* A storvsc device is a device object that contains a vmbus channel */
426 struct storvsc_device {
427 struct hv_device *device;
431 atomic_t num_outstanding_req;
432 struct Scsi_Host *host;
434 wait_queue_head_t waiting_to_drain;
437 * Each unique Port/Path/Target represents 1 channel ie scsi
438 * controller. In reality, the pathid, targetid is always 0
439 * and the port is set by us
441 unsigned int port_number;
442 unsigned char path_id;
443 unsigned char target_id;
446 * The size of the vmscsi_request has changed in win8. The
447 * additional size is because of new elements added to the
448 * structure. These elements are valid only when we are talking
450 * Track the correction to size we need to apply. This value
451 * will likely change during protocol negotiation but it is
452 * valid to start by assuming pre-Win8.
454 int vmscsi_size_delta;
457 * Max I/O, the device can support.
459 u32 max_transfer_bytes;
461 * Number of sub-channels we will open.
464 struct vmbus_channel **stor_chns;
466 * Mask of CPUs bound to subchannels.
468 struct cpumask alloced_cpus;
470 * Serializes modifications of stor_chns[] from storvsc_do_io()
471 * and storvsc_change_target_cpu().
474 /* Used for vsc/vsp channel reset process */
475 struct storvsc_cmd_request init_request;
476 struct storvsc_cmd_request reset_request;
478 * Currently active port and node names for FC devices.
482 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
483 struct fc_rport *rport;
487 struct hv_host_device {
488 struct hv_device *dev;
491 unsigned char target;
492 struct workqueue_struct *handle_error_wq;
493 struct work_struct host_scan_work;
494 struct Scsi_Host *host;
497 struct storvsc_scan_work {
498 struct work_struct work;
499 struct Scsi_Host *host;
504 static void storvsc_device_scan(struct work_struct *work)
506 struct storvsc_scan_work *wrk;
507 struct scsi_device *sdev;
509 wrk = container_of(work, struct storvsc_scan_work, work);
511 sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
514 scsi_rescan_device(&sdev->sdev_gendev);
515 scsi_device_put(sdev);
521 static void storvsc_host_scan(struct work_struct *work)
523 struct Scsi_Host *host;
524 struct scsi_device *sdev;
525 struct hv_host_device *host_device =
526 container_of(work, struct hv_host_device, host_scan_work);
528 host = host_device->host;
530 * Before scanning the host, first check to see if any of the
531 * currrently known devices have been hot removed. We issue a
532 * "unit ready" command against all currently known devices.
533 * This I/O will result in an error for devices that have been
534 * removed. As part of handling the I/O error, we remove the device.
536 * When a LUN is added or removed, the host sends us a signal to
537 * scan the host. Thus we are forced to discover the LUNs that
538 * may have been removed this way.
540 mutex_lock(&host->scan_mutex);
541 shost_for_each_device(sdev, host)
542 scsi_test_unit_ready(sdev, 1, 1, NULL);
543 mutex_unlock(&host->scan_mutex);
545 * Now scan the host to discover LUNs that may have been added.
547 scsi_scan_host(host);
550 static void storvsc_remove_lun(struct work_struct *work)
552 struct storvsc_scan_work *wrk;
553 struct scsi_device *sdev;
555 wrk = container_of(work, struct storvsc_scan_work, work);
556 if (!scsi_host_get(wrk->host))
559 sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
562 scsi_remove_device(sdev);
563 scsi_device_put(sdev);
565 scsi_host_put(wrk->host);
573 * We can get incoming messages from the host that are not in response to
574 * messages that we have sent out. An example of this would be messages
575 * received by the guest to notify dynamic addition/removal of LUNs. To
576 * deal with potential race conditions where the driver may be in the
577 * midst of being unloaded when we might receive an unsolicited message
578 * from the host, we have implemented a mechanism to gurantee sequential
581 * 1) Once the device is marked as being destroyed, we will fail all
583 * 2) We permit incoming messages when the device is being destroyed,
584 * only to properly account for messages already sent out.
587 static inline struct storvsc_device *get_out_stor_device(
588 struct hv_device *device)
590 struct storvsc_device *stor_device;
592 stor_device = hv_get_drvdata(device);
594 if (stor_device && stor_device->destroy)
601 static inline void storvsc_wait_to_drain(struct storvsc_device *dev)
603 dev->drain_notify = true;
604 wait_event(dev->waiting_to_drain,
605 atomic_read(&dev->num_outstanding_req) == 0);
606 dev->drain_notify = false;
609 static inline struct storvsc_device *get_in_stor_device(
610 struct hv_device *device)
612 struct storvsc_device *stor_device;
614 stor_device = hv_get_drvdata(device);
620 * If the device is being destroyed; allow incoming
621 * traffic only to cleanup outstanding requests.
624 if (stor_device->destroy &&
625 (atomic_read(&stor_device->num_outstanding_req) == 0))
633 static void storvsc_change_target_cpu(struct vmbus_channel *channel, u32 old,
636 struct storvsc_device *stor_device;
637 struct vmbus_channel *cur_chn;
638 bool old_is_alloced = false;
639 struct hv_device *device;
643 device = channel->primary_channel ?
644 channel->primary_channel->device_obj
645 : channel->device_obj;
646 stor_device = get_out_stor_device(device);
650 /* See storvsc_do_io() -> get_og_chn(). */
651 spin_lock_irqsave(&stor_device->lock, flags);
654 * Determines if the storvsc device has other channels assigned to
655 * the "old" CPU to update the alloced_cpus mask and the stor_chns
658 if (device->channel != channel && device->channel->target_cpu == old) {
659 cur_chn = device->channel;
660 old_is_alloced = true;
663 list_for_each_entry(cur_chn, &device->channel->sc_list, sc_list) {
664 if (cur_chn == channel)
666 if (cur_chn->target_cpu == old) {
667 old_is_alloced = true;
674 WRITE_ONCE(stor_device->stor_chns[old], cur_chn);
676 cpumask_clear_cpu(old, &stor_device->alloced_cpus);
678 /* "Flush" the stor_chns array. */
679 for_each_possible_cpu(cpu) {
680 if (stor_device->stor_chns[cpu] && !cpumask_test_cpu(
681 cpu, &stor_device->alloced_cpus))
682 WRITE_ONCE(stor_device->stor_chns[cpu], NULL);
685 WRITE_ONCE(stor_device->stor_chns[new], channel);
686 cpumask_set_cpu(new, &stor_device->alloced_cpus);
688 spin_unlock_irqrestore(&stor_device->lock, flags);
691 static void handle_sc_creation(struct vmbus_channel *new_sc)
693 struct hv_device *device = new_sc->primary_channel->device_obj;
694 struct device *dev = &device->device;
695 struct storvsc_device *stor_device;
696 struct vmstorage_channel_properties props;
699 stor_device = get_out_stor_device(device);
703 memset(&props, 0, sizeof(struct vmstorage_channel_properties));
706 * The size of vmbus_requestor is an upper bound on the number of requests
707 * that can be in-progress at any one time across all channels.
709 new_sc->rqstor_size = scsi_driver.can_queue;
711 ret = vmbus_open(new_sc,
712 storvsc_ringbuffer_size,
713 storvsc_ringbuffer_size,
715 sizeof(struct vmstorage_channel_properties),
716 storvsc_on_channel_callback, new_sc);
718 /* In case vmbus_open() fails, we don't use the sub-channel. */
720 dev_err(dev, "Failed to open sub-channel: err=%d\n", ret);
724 new_sc->change_target_cpu_callback = storvsc_change_target_cpu;
726 /* Add the sub-channel to the array of available channels. */
727 stor_device->stor_chns[new_sc->target_cpu] = new_sc;
728 cpumask_set_cpu(new_sc->target_cpu, &stor_device->alloced_cpus);
731 static void handle_multichannel_storage(struct hv_device *device, int max_chns)
733 struct device *dev = &device->device;
734 struct storvsc_device *stor_device;
736 struct storvsc_cmd_request *request;
737 struct vstor_packet *vstor_packet;
741 * If the number of CPUs is artificially restricted, such as
742 * with maxcpus=1 on the kernel boot line, Hyper-V could offer
743 * sub-channels >= the number of CPUs. These sub-channels
744 * should not be created. The primary channel is already created
745 * and assigned to one CPU, so check against # CPUs - 1.
747 num_sc = min((int)(num_online_cpus() - 1), max_chns);
751 stor_device = get_out_stor_device(device);
755 stor_device->num_sc = num_sc;
756 request = &stor_device->init_request;
757 vstor_packet = &request->vstor_packet;
760 * Establish a handler for dealing with subchannels.
762 vmbus_set_sc_create_callback(device->channel, handle_sc_creation);
765 * Request the host to create sub-channels.
767 memset(request, 0, sizeof(struct storvsc_cmd_request));
768 init_completion(&request->wait_event);
769 vstor_packet->operation = VSTOR_OPERATION_CREATE_SUB_CHANNELS;
770 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
771 vstor_packet->sub_channel_count = num_sc;
773 ret = vmbus_sendpacket(device->channel, vstor_packet,
774 (sizeof(struct vstor_packet) -
775 stor_device->vmscsi_size_delta),
776 (unsigned long)request,
778 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
781 dev_err(dev, "Failed to create sub-channel: err=%d\n", ret);
785 t = wait_for_completion_timeout(&request->wait_event, 10*HZ);
787 dev_err(dev, "Failed to create sub-channel: timed out\n");
791 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
792 vstor_packet->status != 0) {
793 dev_err(dev, "Failed to create sub-channel: op=%d, sts=%d\n",
794 vstor_packet->operation, vstor_packet->status);
799 * We need to do nothing here, because vmbus_process_offer()
800 * invokes channel->sc_creation_callback, which will open and use
801 * the sub-channel(s).
805 static void cache_wwn(struct storvsc_device *stor_device,
806 struct vstor_packet *vstor_packet)
809 * Cache the currently active port and node ww names.
811 if (vstor_packet->wwn_packet.primary_active) {
812 stor_device->node_name =
813 wwn_to_u64(vstor_packet->wwn_packet.primary_node_wwn);
814 stor_device->port_name =
815 wwn_to_u64(vstor_packet->wwn_packet.primary_port_wwn);
817 stor_device->node_name =
818 wwn_to_u64(vstor_packet->wwn_packet.secondary_node_wwn);
819 stor_device->port_name =
820 wwn_to_u64(vstor_packet->wwn_packet.secondary_port_wwn);
825 static int storvsc_execute_vstor_op(struct hv_device *device,
826 struct storvsc_cmd_request *request,
829 struct storvsc_device *stor_device;
830 struct vstor_packet *vstor_packet;
833 stor_device = get_out_stor_device(device);
837 vstor_packet = &request->vstor_packet;
839 init_completion(&request->wait_event);
840 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
842 ret = vmbus_sendpacket(device->channel, vstor_packet,
843 (sizeof(struct vstor_packet) -
844 stor_device->vmscsi_size_delta),
845 (unsigned long)request,
847 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
851 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
858 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
859 vstor_packet->status != 0)
865 static int storvsc_channel_init(struct hv_device *device, bool is_fc)
867 struct storvsc_device *stor_device;
868 struct storvsc_cmd_request *request;
869 struct vstor_packet *vstor_packet;
872 bool process_sub_channels = false;
874 stor_device = get_out_stor_device(device);
878 request = &stor_device->init_request;
879 vstor_packet = &request->vstor_packet;
882 * Now, initiate the vsc/vsp initialization protocol on the open
885 memset(request, 0, sizeof(struct storvsc_cmd_request));
886 vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION;
887 ret = storvsc_execute_vstor_op(device, request, true);
891 * Query host supported protocol version.
894 for (i = 0; i < ARRAY_SIZE(vmstor_protocols); i++) {
895 /* reuse the packet for version range supported */
896 memset(vstor_packet, 0, sizeof(struct vstor_packet));
897 vstor_packet->operation =
898 VSTOR_OPERATION_QUERY_PROTOCOL_VERSION;
900 vstor_packet->version.major_minor =
901 vmstor_protocols[i].protocol_version;
904 * The revision number is only used in Windows; set it to 0.
906 vstor_packet->version.revision = 0;
907 ret = storvsc_execute_vstor_op(device, request, false);
911 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO)
914 if (vstor_packet->status == 0) {
915 vmstor_proto_version =
916 vmstor_protocols[i].protocol_version;
919 vmstor_protocols[i].sense_buffer_size;
921 stor_device->vmscsi_size_delta =
922 vmstor_protocols[i].vmscsi_size_delta;
928 if (vstor_packet->status != 0)
932 memset(vstor_packet, 0, sizeof(struct vstor_packet));
933 vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES;
934 ret = storvsc_execute_vstor_op(device, request, true);
939 * Check to see if multi-channel support is there.
940 * Hosts that implement protocol version of 5.1 and above
941 * support multi-channel.
943 max_chns = vstor_packet->storage_channel_properties.max_channel_cnt;
946 * Allocate state to manage the sub-channels.
947 * We allocate an array based on the numbers of possible CPUs
948 * (Hyper-V does not support cpu online/offline).
949 * This Array will be sparseley populated with unique
950 * channels - primary + sub-channels.
951 * We will however populate all the slots to evenly distribute
954 stor_device->stor_chns = kcalloc(num_possible_cpus(), sizeof(void *),
956 if (stor_device->stor_chns == NULL)
959 device->channel->change_target_cpu_callback = storvsc_change_target_cpu;
961 stor_device->stor_chns[device->channel->target_cpu] = device->channel;
962 cpumask_set_cpu(device->channel->target_cpu,
963 &stor_device->alloced_cpus);
965 if (vmstor_proto_version >= VMSTOR_PROTO_VERSION_WIN8) {
966 if (vstor_packet->storage_channel_properties.flags &
967 STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL)
968 process_sub_channels = true;
970 stor_device->max_transfer_bytes =
971 vstor_packet->storage_channel_properties.max_transfer_bytes;
977 * For FC devices retrieve FC HBA data.
979 memset(vstor_packet, 0, sizeof(struct vstor_packet));
980 vstor_packet->operation = VSTOR_OPERATION_FCHBA_DATA;
981 ret = storvsc_execute_vstor_op(device, request, true);
986 * Cache the currently active port and node ww names.
988 cache_wwn(stor_device, vstor_packet);
992 memset(vstor_packet, 0, sizeof(struct vstor_packet));
993 vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION;
994 ret = storvsc_execute_vstor_op(device, request, true);
998 if (process_sub_channels)
999 handle_multichannel_storage(device, max_chns);
1004 static void storvsc_handle_error(struct vmscsi_request *vm_srb,
1005 struct scsi_cmnd *scmnd,
1006 struct Scsi_Host *host,
1009 struct storvsc_scan_work *wrk;
1010 void (*process_err_fn)(struct work_struct *work);
1011 struct hv_host_device *host_dev = shost_priv(host);
1012 bool do_work = false;
1014 switch (SRB_STATUS(vm_srb->srb_status)) {
1015 case SRB_STATUS_ERROR:
1017 * Let upper layer deal with error when
1018 * sense message is present.
1021 if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID)
1024 * If there is an error; offline the device since all
1025 * error recovery strategies would have already been
1026 * deployed on the host side. However, if the command
1027 * were a pass-through command deal with it appropriately.
1029 switch (scmnd->cmnd[0]) {
1032 set_host_byte(scmnd, DID_PASSTHROUGH);
1035 * On Some Windows hosts TEST_UNIT_READY command can return
1036 * SRB_STATUS_ERROR, let the upper level code deal with it
1037 * based on the sense information.
1039 case TEST_UNIT_READY:
1042 set_host_byte(scmnd, DID_ERROR);
1045 case SRB_STATUS_INVALID_LUN:
1046 set_host_byte(scmnd, DID_NO_CONNECT);
1048 process_err_fn = storvsc_remove_lun;
1050 case SRB_STATUS_ABORTED:
1051 if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID &&
1052 (asc == 0x2a) && (ascq == 0x9)) {
1054 process_err_fn = storvsc_device_scan;
1056 * Retry the I/O that triggered this.
1058 set_host_byte(scmnd, DID_REQUEUE);
1067 * We need to schedule work to process this error; schedule it.
1069 wrk = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC);
1071 set_host_byte(scmnd, DID_TARGET_FAILURE);
1076 wrk->lun = vm_srb->lun;
1077 wrk->tgt_id = vm_srb->target_id;
1078 INIT_WORK(&wrk->work, process_err_fn);
1079 queue_work(host_dev->handle_error_wq, &wrk->work);
1083 static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request,
1084 struct storvsc_device *stor_dev)
1086 struct scsi_cmnd *scmnd = cmd_request->cmd;
1087 struct scsi_sense_hdr sense_hdr;
1088 struct vmscsi_request *vm_srb;
1089 u32 data_transfer_length;
1090 struct Scsi_Host *host;
1091 u32 payload_sz = cmd_request->payload_sz;
1092 void *payload = cmd_request->payload;
1094 host = stor_dev->host;
1096 vm_srb = &cmd_request->vstor_packet.vm_srb;
1097 data_transfer_length = vm_srb->data_transfer_length;
1099 scmnd->result = vm_srb->scsi_status;
1101 if (scmnd->result) {
1102 if (scsi_normalize_sense(scmnd->sense_buffer,
1103 SCSI_SENSE_BUFFERSIZE, &sense_hdr) &&
1104 !(sense_hdr.sense_key == NOT_READY &&
1105 sense_hdr.asc == 0x03A) &&
1106 do_logging(STORVSC_LOGGING_ERROR))
1107 scsi_print_sense_hdr(scmnd->device, "storvsc",
1111 if (vm_srb->srb_status != SRB_STATUS_SUCCESS) {
1112 storvsc_handle_error(vm_srb, scmnd, host, sense_hdr.asc,
1115 * The Windows driver set data_transfer_length on
1116 * SRB_STATUS_DATA_OVERRUN. On other errors, this value
1117 * is untouched. In these cases we set it to 0.
1119 if (vm_srb->srb_status != SRB_STATUS_DATA_OVERRUN)
1120 data_transfer_length = 0;
1123 /* Validate data_transfer_length (from Hyper-V) */
1124 if (data_transfer_length > cmd_request->payload->range.len)
1125 data_transfer_length = cmd_request->payload->range.len;
1127 scsi_set_resid(scmnd,
1128 cmd_request->payload->range.len - data_transfer_length);
1130 scmnd->scsi_done(scmnd);
1133 sizeof(struct vmbus_channel_packet_multipage_buffer))
1137 static void storvsc_on_io_completion(struct storvsc_device *stor_device,
1138 struct vstor_packet *vstor_packet,
1139 struct storvsc_cmd_request *request)
1141 struct vstor_packet *stor_pkt;
1142 struct hv_device *device = stor_device->device;
1144 stor_pkt = &request->vstor_packet;
1147 * The current SCSI handling on the host side does
1148 * not correctly handle:
1149 * INQUIRY command with page code parameter set to 0x80
1150 * MODE_SENSE command with cmd[2] == 0x1c
1152 * Setup srb and scsi status so this won't be fatal.
1153 * We do this so we can distinguish truly fatal failues
1154 * (srb status == 0x4) and off-line the device in that case.
1157 if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) ||
1158 (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) {
1159 vstor_packet->vm_srb.scsi_status = 0;
1160 vstor_packet->vm_srb.srb_status = SRB_STATUS_SUCCESS;
1164 /* Copy over the status...etc */
1165 stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status;
1166 stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status;
1168 /* Validate sense_info_length (from Hyper-V) */
1169 if (vstor_packet->vm_srb.sense_info_length > sense_buffer_size)
1170 vstor_packet->vm_srb.sense_info_length = sense_buffer_size;
1172 stor_pkt->vm_srb.sense_info_length =
1173 vstor_packet->vm_srb.sense_info_length;
1175 if (vstor_packet->vm_srb.scsi_status != 0 ||
1176 vstor_packet->vm_srb.srb_status != SRB_STATUS_SUCCESS)
1177 storvsc_log(device, STORVSC_LOGGING_WARN,
1178 "cmd 0x%x scsi status 0x%x srb status 0x%x\n",
1179 stor_pkt->vm_srb.cdb[0],
1180 vstor_packet->vm_srb.scsi_status,
1181 vstor_packet->vm_srb.srb_status);
1183 if ((vstor_packet->vm_srb.scsi_status & 0xFF) == 0x02) {
1184 /* CHECK_CONDITION */
1185 if (vstor_packet->vm_srb.srb_status &
1186 SRB_STATUS_AUTOSENSE_VALID) {
1187 /* autosense data available */
1189 storvsc_log(device, STORVSC_LOGGING_WARN,
1190 "stor pkt %p autosense data valid - len %d\n",
1191 request, vstor_packet->vm_srb.sense_info_length);
1193 memcpy(request->cmd->sense_buffer,
1194 vstor_packet->vm_srb.sense_data,
1195 vstor_packet->vm_srb.sense_info_length);
1200 stor_pkt->vm_srb.data_transfer_length =
1201 vstor_packet->vm_srb.data_transfer_length;
1203 storvsc_command_completion(request, stor_device);
1205 if (atomic_dec_and_test(&stor_device->num_outstanding_req) &&
1206 stor_device->drain_notify)
1207 wake_up(&stor_device->waiting_to_drain);
1212 static void storvsc_on_receive(struct storvsc_device *stor_device,
1213 struct vstor_packet *vstor_packet,
1214 struct storvsc_cmd_request *request)
1216 struct hv_host_device *host_dev;
1217 switch (vstor_packet->operation) {
1218 case VSTOR_OPERATION_COMPLETE_IO:
1219 storvsc_on_io_completion(stor_device, vstor_packet, request);
1222 case VSTOR_OPERATION_REMOVE_DEVICE:
1223 case VSTOR_OPERATION_ENUMERATE_BUS:
1224 host_dev = shost_priv(stor_device->host);
1226 host_dev->handle_error_wq, &host_dev->host_scan_work);
1229 case VSTOR_OPERATION_FCHBA_DATA:
1230 cache_wwn(stor_device, vstor_packet);
1231 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1232 fc_host_node_name(stor_device->host) = stor_device->node_name;
1233 fc_host_port_name(stor_device->host) = stor_device->port_name;
1241 static void storvsc_on_channel_callback(void *context)
1243 struct vmbus_channel *channel = (struct vmbus_channel *)context;
1244 const struct vmpacket_descriptor *desc;
1245 struct hv_device *device;
1246 struct storvsc_device *stor_device;
1248 if (channel->primary_channel != NULL)
1249 device = channel->primary_channel->device_obj;
1251 device = channel->device_obj;
1253 stor_device = get_in_stor_device(device);
1257 foreach_vmbus_pkt(desc, channel) {
1258 void *packet = hv_pkt_data(desc);
1259 struct storvsc_cmd_request *request;
1262 cmd_rqst = vmbus_request_addr(&channel->requestor,
1264 if (cmd_rqst == VMBUS_RQST_ERROR) {
1265 dev_err(&device->device,
1266 "Incorrect transaction id\n");
1270 request = (struct storvsc_cmd_request *)(unsigned long)cmd_rqst;
1272 if (hv_pkt_datalen(desc) < sizeof(struct vstor_packet) -
1273 stor_device->vmscsi_size_delta) {
1274 dev_err(&device->device, "Invalid packet len\n");
1278 if (request == &stor_device->init_request ||
1279 request == &stor_device->reset_request) {
1280 memcpy(&request->vstor_packet, packet,
1281 (sizeof(struct vstor_packet) - stor_device->vmscsi_size_delta));
1282 complete(&request->wait_event);
1284 storvsc_on_receive(stor_device, packet, request);
1289 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size,
1292 struct vmstorage_channel_properties props;
1295 memset(&props, 0, sizeof(struct vmstorage_channel_properties));
1298 * The size of vmbus_requestor is an upper bound on the number of requests
1299 * that can be in-progress at any one time across all channels.
1301 device->channel->rqstor_size = scsi_driver.can_queue;
1303 ret = vmbus_open(device->channel,
1307 sizeof(struct vmstorage_channel_properties),
1308 storvsc_on_channel_callback, device->channel);
1313 ret = storvsc_channel_init(device, is_fc);
1318 static int storvsc_dev_remove(struct hv_device *device)
1320 struct storvsc_device *stor_device;
1322 stor_device = hv_get_drvdata(device);
1324 stor_device->destroy = true;
1326 /* Make sure flag is set before waiting */
1330 * At this point, all outbound traffic should be disable. We
1331 * only allow inbound traffic (responses) to proceed so that
1332 * outstanding requests can be completed.
1335 storvsc_wait_to_drain(stor_device);
1338 * Since we have already drained, we don't need to busy wait
1339 * as was done in final_release_stor_device()
1340 * Note that we cannot set the ext pointer to NULL until
1341 * we have drained - to drain the outgoing packets, we need to
1342 * allow incoming packets.
1344 hv_set_drvdata(device, NULL);
1346 /* Close the channel */
1347 vmbus_close(device->channel);
1349 kfree(stor_device->stor_chns);
1354 static struct vmbus_channel *get_og_chn(struct storvsc_device *stor_device,
1359 const struct cpumask *node_mask;
1360 int num_channels, tgt_cpu;
1362 if (stor_device->num_sc == 0) {
1363 stor_device->stor_chns[q_num] = stor_device->device->channel;
1364 return stor_device->device->channel;
1368 * Our channel array is sparsley populated and we
1369 * initiated I/O on a processor/hw-q that does not
1370 * currently have a designated channel. Fix this.
1371 * The strategy is simple:
1372 * I. Ensure NUMA locality
1373 * II. Distribute evenly (best effort)
1376 node_mask = cpumask_of_node(cpu_to_node(q_num));
1379 for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1380 if (cpumask_test_cpu(tgt_cpu, node_mask))
1383 if (num_channels == 0) {
1384 stor_device->stor_chns[q_num] = stor_device->device->channel;
1385 return stor_device->device->channel;
1389 while (hash_qnum >= num_channels)
1390 hash_qnum -= num_channels;
1392 for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1393 if (!cpumask_test_cpu(tgt_cpu, node_mask))
1395 if (slot == hash_qnum)
1400 stor_device->stor_chns[q_num] = stor_device->stor_chns[tgt_cpu];
1402 return stor_device->stor_chns[q_num];
1406 static int storvsc_do_io(struct hv_device *device,
1407 struct storvsc_cmd_request *request, u16 q_num)
1409 struct storvsc_device *stor_device;
1410 struct vstor_packet *vstor_packet;
1411 struct vmbus_channel *outgoing_channel, *channel;
1412 unsigned long flags;
1414 const struct cpumask *node_mask;
1417 vstor_packet = &request->vstor_packet;
1418 stor_device = get_out_stor_device(device);
1424 request->device = device;
1426 * Select an appropriate channel to send the request out.
1428 /* See storvsc_change_target_cpu(). */
1429 outgoing_channel = READ_ONCE(stor_device->stor_chns[q_num]);
1430 if (outgoing_channel != NULL) {
1431 if (outgoing_channel->target_cpu == q_num) {
1433 * Ideally, we want to pick a different channel if
1434 * available on the same NUMA node.
1436 node_mask = cpumask_of_node(cpu_to_node(q_num));
1437 for_each_cpu_wrap(tgt_cpu,
1438 &stor_device->alloced_cpus, q_num + 1) {
1439 if (!cpumask_test_cpu(tgt_cpu, node_mask))
1441 if (tgt_cpu == q_num)
1443 channel = READ_ONCE(
1444 stor_device->stor_chns[tgt_cpu]);
1445 if (channel == NULL)
1447 if (hv_get_avail_to_write_percent(
1449 > ring_avail_percent_lowater) {
1450 outgoing_channel = channel;
1456 * All the other channels on the same NUMA node are
1457 * busy. Try to use the channel on the current CPU
1459 if (hv_get_avail_to_write_percent(
1460 &outgoing_channel->outbound)
1461 > ring_avail_percent_lowater)
1465 * If we reach here, all the channels on the current
1466 * NUMA node are busy. Try to find a channel in
1469 for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1470 if (cpumask_test_cpu(tgt_cpu, node_mask))
1472 channel = READ_ONCE(
1473 stor_device->stor_chns[tgt_cpu]);
1474 if (channel == NULL)
1476 if (hv_get_avail_to_write_percent(
1478 > ring_avail_percent_lowater) {
1479 outgoing_channel = channel;
1485 spin_lock_irqsave(&stor_device->lock, flags);
1486 outgoing_channel = stor_device->stor_chns[q_num];
1487 if (outgoing_channel != NULL) {
1488 spin_unlock_irqrestore(&stor_device->lock, flags);
1491 outgoing_channel = get_og_chn(stor_device, q_num);
1492 spin_unlock_irqrestore(&stor_device->lock, flags);
1496 vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
1498 vstor_packet->vm_srb.length = (sizeof(struct vmscsi_request) -
1499 stor_device->vmscsi_size_delta);
1502 vstor_packet->vm_srb.sense_info_length = sense_buffer_size;
1505 vstor_packet->vm_srb.data_transfer_length =
1506 request->payload->range.len;
1508 vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
1510 if (request->payload->range.len) {
1512 ret = vmbus_sendpacket_mpb_desc(outgoing_channel,
1513 request->payload, request->payload_sz,
1515 (sizeof(struct vstor_packet) -
1516 stor_device->vmscsi_size_delta),
1517 (unsigned long)request);
1519 ret = vmbus_sendpacket(outgoing_channel, vstor_packet,
1520 (sizeof(struct vstor_packet) -
1521 stor_device->vmscsi_size_delta),
1522 (unsigned long)request,
1524 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1530 atomic_inc(&stor_device->num_outstanding_req);
1535 static int storvsc_device_alloc(struct scsi_device *sdevice)
1538 * Set blist flag to permit the reading of the VPD pages even when
1539 * the target may claim SPC-2 compliance. MSFT targets currently
1540 * claim SPC-2 compliance while they implement post SPC-2 features.
1541 * With this flag we can correctly handle WRITE_SAME_16 issues.
1543 * Hypervisor reports SCSI_UNKNOWN type for DVD ROM device but
1544 * still supports REPORT LUN.
1546 sdevice->sdev_bflags = BLIST_REPORTLUN2 | BLIST_TRY_VPD_PAGES;
1551 static int storvsc_device_configure(struct scsi_device *sdevice)
1553 blk_queue_rq_timeout(sdevice->request_queue, (storvsc_timeout * HZ));
1555 sdevice->no_write_same = 1;
1558 * If the host is WIN8 or WIN8 R2, claim conformance to SPC-3
1559 * if the device is a MSFT virtual device. If the host is
1560 * WIN10 or newer, allow write_same.
1562 if (!strncmp(sdevice->vendor, "Msft", 4)) {
1563 switch (vmstor_proto_version) {
1564 case VMSTOR_PROTO_VERSION_WIN8:
1565 case VMSTOR_PROTO_VERSION_WIN8_1:
1566 sdevice->scsi_level = SCSI_SPC_3;
1570 if (vmstor_proto_version >= VMSTOR_PROTO_VERSION_WIN10)
1571 sdevice->no_write_same = 0;
1577 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev,
1578 sector_t capacity, int *info)
1580 sector_t nsect = capacity;
1581 sector_t cylinders = nsect;
1582 int heads, sectors_pt;
1585 * We are making up these values; let us keep it simple.
1588 sectors_pt = 0x3f; /* Sectors per track */
1589 sector_div(cylinders, heads * sectors_pt);
1590 if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
1594 info[1] = sectors_pt;
1595 info[2] = (int)cylinders;
1600 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
1602 struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
1603 struct hv_device *device = host_dev->dev;
1605 struct storvsc_device *stor_device;
1606 struct storvsc_cmd_request *request;
1607 struct vstor_packet *vstor_packet;
1610 stor_device = get_out_stor_device(device);
1614 request = &stor_device->reset_request;
1615 vstor_packet = &request->vstor_packet;
1616 memset(vstor_packet, 0, sizeof(struct vstor_packet));
1618 init_completion(&request->wait_event);
1620 vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
1621 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
1622 vstor_packet->vm_srb.path_id = stor_device->path_id;
1624 ret = vmbus_sendpacket(device->channel, vstor_packet,
1625 (sizeof(struct vstor_packet) -
1626 stor_device->vmscsi_size_delta),
1627 (unsigned long)&stor_device->reset_request,
1629 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1633 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
1635 return TIMEOUT_ERROR;
1639 * At this point, all outstanding requests in the adapter
1640 * should have been flushed out and return to us
1641 * There is a potential race here where the host may be in
1642 * the process of responding when we return from here.
1643 * Just wait for all in-transit packets to be accounted for
1644 * before we return from here.
1646 storvsc_wait_to_drain(stor_device);
1652 * The host guarantees to respond to each command, although I/O latencies might
1653 * be unbounded on Azure. Reset the timer unconditionally to give the host a
1654 * chance to perform EH.
1656 static enum blk_eh_timer_return storvsc_eh_timed_out(struct scsi_cmnd *scmnd)
1658 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1659 if (scmnd->device->host->transportt == fc_transport_template)
1660 return fc_eh_timed_out(scmnd);
1662 return BLK_EH_RESET_TIMER;
1665 static bool storvsc_scsi_cmd_ok(struct scsi_cmnd *scmnd)
1667 bool allowed = true;
1668 u8 scsi_op = scmnd->cmnd[0];
1671 /* the host does not handle WRITE_SAME, log accident usage */
1674 * smartd sends this command and the host does not handle
1675 * this. So, don't send it.
1678 scmnd->result = DID_ERROR << 16;
1687 static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
1690 struct hv_host_device *host_dev = shost_priv(host);
1691 struct hv_device *dev = host_dev->dev;
1692 struct storvsc_cmd_request *cmd_request = scsi_cmd_priv(scmnd);
1694 struct scatterlist *sgl;
1695 unsigned int sg_count;
1696 struct vmscsi_request *vm_srb;
1697 struct vmbus_packet_mpb_array *payload;
1701 if (vmstor_proto_version <= VMSTOR_PROTO_VERSION_WIN8) {
1703 * On legacy hosts filter unimplemented commands.
1704 * Future hosts are expected to correctly handle
1705 * unsupported commands. Furthermore, it is
1706 * possible that some of the currently
1707 * unsupported commands maybe supported in
1708 * future versions of the host.
1710 if (!storvsc_scsi_cmd_ok(scmnd)) {
1711 scmnd->scsi_done(scmnd);
1716 /* Setup the cmd request */
1717 cmd_request->cmd = scmnd;
1719 memset(&cmd_request->vstor_packet, 0, sizeof(struct vstor_packet));
1720 vm_srb = &cmd_request->vstor_packet.vm_srb;
1721 vm_srb->win8_extension.time_out_value = 60;
1723 vm_srb->win8_extension.srb_flags |=
1724 SRB_FLAGS_DISABLE_SYNCH_TRANSFER;
1726 if (scmnd->device->tagged_supported) {
1727 vm_srb->win8_extension.srb_flags |=
1728 (SRB_FLAGS_QUEUE_ACTION_ENABLE | SRB_FLAGS_NO_QUEUE_FREEZE);
1729 vm_srb->win8_extension.queue_tag = SP_UNTAGGED;
1730 vm_srb->win8_extension.queue_action = SRB_SIMPLE_TAG_REQUEST;
1734 switch (scmnd->sc_data_direction) {
1736 vm_srb->data_in = WRITE_TYPE;
1737 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_DATA_OUT;
1739 case DMA_FROM_DEVICE:
1740 vm_srb->data_in = READ_TYPE;
1741 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_DATA_IN;
1744 vm_srb->data_in = UNKNOWN_TYPE;
1745 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_NO_DATA_TRANSFER;
1749 * This is DMA_BIDIRECTIONAL or something else we are never
1750 * supposed to see here.
1752 WARN(1, "Unexpected data direction: %d\n",
1753 scmnd->sc_data_direction);
1758 vm_srb->port_number = host_dev->port;
1759 vm_srb->path_id = scmnd->device->channel;
1760 vm_srb->target_id = scmnd->device->id;
1761 vm_srb->lun = scmnd->device->lun;
1763 vm_srb->cdb_length = scmnd->cmd_len;
1765 memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
1767 sgl = (struct scatterlist *)scsi_sglist(scmnd);
1768 sg_count = scsi_sg_count(scmnd);
1770 length = scsi_bufflen(scmnd);
1771 payload = (struct vmbus_packet_mpb_array *)&cmd_request->mpb;
1772 payload_sz = sizeof(cmd_request->mpb);
1775 unsigned int hvpgoff, hvpfns_to_add;
1776 unsigned long offset_in_hvpg = offset_in_hvpage(sgl->offset);
1777 unsigned int hvpg_count = HVPFN_UP(offset_in_hvpg + length);
1780 if (hvpg_count > MAX_PAGE_BUFFER_COUNT) {
1782 payload_sz = (hvpg_count * sizeof(u64) +
1783 sizeof(struct vmbus_packet_mpb_array));
1784 payload = kzalloc(payload_sz, GFP_ATOMIC);
1786 return SCSI_MLQUEUE_DEVICE_BUSY;
1789 payload->range.len = length;
1790 payload->range.offset = offset_in_hvpg;
1793 for (i = 0; sgl != NULL; sgl = sg_next(sgl)) {
1795 * Init values for the current sgl entry. hvpgoff
1796 * and hvpfns_to_add are in units of Hyper-V size
1797 * pages. Handling the PAGE_SIZE != HV_HYP_PAGE_SIZE
1798 * case also handles values of sgl->offset that are
1799 * larger than PAGE_SIZE. Such offsets are handled
1800 * even on other than the first sgl entry, provided
1801 * they are a multiple of PAGE_SIZE.
1803 hvpgoff = HVPFN_DOWN(sgl->offset);
1804 hvpfn = page_to_hvpfn(sg_page(sgl)) + hvpgoff;
1805 hvpfns_to_add = HVPFN_UP(sgl->offset + sgl->length) -
1809 * Fill the next portion of the PFN array with
1810 * sequential Hyper-V PFNs for the continguous physical
1811 * memory described by the sgl entry. The end of the
1812 * last sgl should be reached at the same time that
1813 * the PFN array is filled.
1815 while (hvpfns_to_add--)
1816 payload->range.pfn_array[i++] = hvpfn++;
1820 cmd_request->payload = payload;
1821 cmd_request->payload_sz = payload_sz;
1823 /* Invokes the vsc to start an IO */
1824 ret = storvsc_do_io(dev, cmd_request, get_cpu());
1827 if (ret == -EAGAIN) {
1828 if (payload_sz > sizeof(cmd_request->mpb))
1831 return SCSI_MLQUEUE_DEVICE_BUSY;
1837 static struct scsi_host_template scsi_driver = {
1838 .module = THIS_MODULE,
1839 .name = "storvsc_host_t",
1840 .cmd_size = sizeof(struct storvsc_cmd_request),
1841 .bios_param = storvsc_get_chs,
1842 .queuecommand = storvsc_queuecommand,
1843 .eh_host_reset_handler = storvsc_host_reset_handler,
1844 .proc_name = "storvsc_host",
1845 .eh_timed_out = storvsc_eh_timed_out,
1846 .slave_alloc = storvsc_device_alloc,
1847 .slave_configure = storvsc_device_configure,
1848 .cmd_per_lun = 2048,
1850 /* Ensure there are no gaps in presented sgls */
1851 .virt_boundary_mask = PAGE_SIZE-1,
1853 .track_queue_depth = 1,
1854 .change_queue_depth = storvsc_change_queue_depth,
1863 static const struct hv_vmbus_device_id id_table[] = {
1866 .driver_data = SCSI_GUID
1870 .driver_data = IDE_GUID
1872 /* Fibre Channel GUID */
1875 .driver_data = SFC_GUID
1880 MODULE_DEVICE_TABLE(vmbus, id_table);
1882 static const struct { guid_t guid; } fc_guid = { HV_SYNTHFC_GUID };
1884 static bool hv_dev_is_fc(struct hv_device *hv_dev)
1886 return guid_equal(&fc_guid.guid, &hv_dev->dev_type);
1889 static int storvsc_probe(struct hv_device *device,
1890 const struct hv_vmbus_device_id *dev_id)
1893 int num_cpus = num_online_cpus();
1894 int num_present_cpus = num_present_cpus();
1895 struct Scsi_Host *host;
1896 struct hv_host_device *host_dev;
1897 bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false);
1898 bool is_fc = ((dev_id->driver_data == SFC_GUID) ? true : false);
1900 struct storvsc_device *stor_device;
1901 int max_luns_per_target;
1904 int max_sub_channels = 0;
1907 * Based on the windows host we are running on,
1908 * set state to properly communicate with the host.
1911 if (vmbus_proto_version < VERSION_WIN8) {
1912 max_luns_per_target = STORVSC_IDE_MAX_LUNS_PER_TARGET;
1913 max_targets = STORVSC_IDE_MAX_TARGETS;
1914 max_channels = STORVSC_IDE_MAX_CHANNELS;
1916 max_luns_per_target = STORVSC_MAX_LUNS_PER_TARGET;
1917 max_targets = STORVSC_MAX_TARGETS;
1918 max_channels = STORVSC_MAX_CHANNELS;
1920 * On Windows8 and above, we support sub-channels for storage
1921 * on SCSI and FC controllers.
1922 * The number of sub-channels offerred is based on the number of
1923 * VCPUs in the guest.
1927 (num_cpus - 1) / storvsc_vcpus_per_sub_channel;
1930 scsi_driver.can_queue = max_outstanding_req_per_channel *
1931 (max_sub_channels + 1) *
1932 (100 - ring_avail_percent_lowater) / 100;
1934 host = scsi_host_alloc(&scsi_driver,
1935 sizeof(struct hv_host_device));
1939 host_dev = shost_priv(host);
1940 memset(host_dev, 0, sizeof(struct hv_host_device));
1942 host_dev->port = host->host_no;
1943 host_dev->dev = device;
1944 host_dev->host = host;
1947 stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
1953 stor_device->destroy = false;
1954 init_waitqueue_head(&stor_device->waiting_to_drain);
1955 stor_device->device = device;
1956 stor_device->host = host;
1957 stor_device->vmscsi_size_delta = sizeof(struct vmscsi_win8_extension);
1958 spin_lock_init(&stor_device->lock);
1959 hv_set_drvdata(device, stor_device);
1961 stor_device->port_number = host->host_no;
1962 ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size, is_fc);
1966 host_dev->path = stor_device->path_id;
1967 host_dev->target = stor_device->target_id;
1969 switch (dev_id->driver_data) {
1971 host->max_lun = STORVSC_FC_MAX_LUNS_PER_TARGET;
1972 host->max_id = STORVSC_FC_MAX_TARGETS;
1973 host->max_channel = STORVSC_FC_MAX_CHANNELS - 1;
1974 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1975 host->transportt = fc_transport_template;
1980 host->max_lun = max_luns_per_target;
1981 host->max_id = max_targets;
1982 host->max_channel = max_channels - 1;
1986 host->max_lun = STORVSC_IDE_MAX_LUNS_PER_TARGET;
1987 host->max_id = STORVSC_IDE_MAX_TARGETS;
1988 host->max_channel = STORVSC_IDE_MAX_CHANNELS - 1;
1991 /* max cmd length */
1992 host->max_cmd_len = STORVSC_MAX_CMD_LEN;
1995 * set the table size based on the info we got
1998 host->sg_tablesize = (stor_device->max_transfer_bytes >> PAGE_SHIFT);
2000 * For non-IDE disks, the host supports multiple channels.
2001 * Set the number of HW queues we are supporting.
2004 if (storvsc_max_hw_queues > num_present_cpus) {
2005 storvsc_max_hw_queues = 0;
2006 storvsc_log(device, STORVSC_LOGGING_WARN,
2007 "Resetting invalid storvsc_max_hw_queues value to default.\n");
2009 if (storvsc_max_hw_queues)
2010 host->nr_hw_queues = storvsc_max_hw_queues;
2012 host->nr_hw_queues = num_present_cpus;
2016 * Set the error handler work queue.
2018 host_dev->handle_error_wq =
2019 alloc_ordered_workqueue("storvsc_error_wq_%d",
2022 if (!host_dev->handle_error_wq) {
2026 INIT_WORK(&host_dev->host_scan_work, storvsc_host_scan);
2027 /* Register the HBA and start the scsi bus scan */
2028 ret = scsi_add_host(host, &device->device);
2033 scsi_scan_host(host);
2035 target = (device->dev_instance.b[5] << 8 |
2036 device->dev_instance.b[4]);
2037 ret = scsi_add_device(host, 0, target, 0);
2041 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2042 if (host->transportt == fc_transport_template) {
2043 struct fc_rport_identifiers ids = {
2044 .roles = FC_PORT_ROLE_FCP_DUMMY_INITIATOR,
2047 fc_host_node_name(host) = stor_device->node_name;
2048 fc_host_port_name(host) = stor_device->port_name;
2049 stor_device->rport = fc_remote_port_add(host, 0, &ids);
2050 if (!stor_device->rport) {
2059 scsi_remove_host(host);
2062 destroy_workqueue(host_dev->handle_error_wq);
2066 * Once we have connected with the host, we would need to
2067 * to invoke storvsc_dev_remove() to rollback this state and
2068 * this call also frees up the stor_device; hence the jump around
2071 storvsc_dev_remove(device);
2075 kfree(stor_device->stor_chns);
2079 scsi_host_put(host);
2083 /* Change a scsi target's queue depth */
2084 static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth)
2086 if (queue_depth > scsi_driver.can_queue)
2087 queue_depth = scsi_driver.can_queue;
2089 return scsi_change_queue_depth(sdev, queue_depth);
2092 static int storvsc_remove(struct hv_device *dev)
2094 struct storvsc_device *stor_device = hv_get_drvdata(dev);
2095 struct Scsi_Host *host = stor_device->host;
2096 struct hv_host_device *host_dev = shost_priv(host);
2098 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2099 if (host->transportt == fc_transport_template) {
2100 fc_remote_port_delete(stor_device->rport);
2101 fc_remove_host(host);
2104 destroy_workqueue(host_dev->handle_error_wq);
2105 scsi_remove_host(host);
2106 storvsc_dev_remove(dev);
2107 scsi_host_put(host);
2112 static int storvsc_suspend(struct hv_device *hv_dev)
2114 struct storvsc_device *stor_device = hv_get_drvdata(hv_dev);
2115 struct Scsi_Host *host = stor_device->host;
2116 struct hv_host_device *host_dev = shost_priv(host);
2118 storvsc_wait_to_drain(stor_device);
2120 drain_workqueue(host_dev->handle_error_wq);
2122 vmbus_close(hv_dev->channel);
2124 kfree(stor_device->stor_chns);
2125 stor_device->stor_chns = NULL;
2127 cpumask_clear(&stor_device->alloced_cpus);
2132 static int storvsc_resume(struct hv_device *hv_dev)
2136 ret = storvsc_connect_to_vsp(hv_dev, storvsc_ringbuffer_size,
2137 hv_dev_is_fc(hv_dev));
2141 static struct hv_driver storvsc_drv = {
2142 .name = KBUILD_MODNAME,
2143 .id_table = id_table,
2144 .probe = storvsc_probe,
2145 .remove = storvsc_remove,
2146 .suspend = storvsc_suspend,
2147 .resume = storvsc_resume,
2149 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
2153 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2154 static struct fc_function_template fc_transport_functions = {
2155 .show_host_node_name = 1,
2156 .show_host_port_name = 1,
2160 static int __init storvsc_drv_init(void)
2165 * Divide the ring buffer data size (which is 1 page less
2166 * than the ring buffer size since that page is reserved for
2167 * the ring buffer indices) by the max request size (which is
2168 * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
2170 * The computation underestimates max_outstanding_req_per_channel
2171 * for Win7 and older hosts because it does not take into account
2172 * the vmscsi_size_delta correction to the max request size.
2174 max_outstanding_req_per_channel =
2175 ((storvsc_ringbuffer_size - PAGE_SIZE) /
2176 ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
2177 sizeof(struct vstor_packet) + sizeof(u64),
2180 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2181 fc_transport_template = fc_attach_transport(&fc_transport_functions);
2182 if (!fc_transport_template)
2186 ret = vmbus_driver_register(&storvsc_drv);
2188 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2190 fc_release_transport(fc_transport_template);
2196 static void __exit storvsc_drv_exit(void)
2198 vmbus_driver_unregister(&storvsc_drv);
2199 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2200 fc_release_transport(fc_transport_template);
2204 MODULE_LICENSE("GPL");
2205 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
2206 module_init(storvsc_drv_init);
2207 module_exit(storvsc_drv_exit);