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. In the 8-bit field, the two high order bits
348 * are flags, while the remaining 6 bits are an integer status code. The
349 * definitions here include only the subset of the integer status codes that
350 * are tested for in this driver.
352 #define SRB_STATUS_AUTOSENSE_VALID 0x80
353 #define SRB_STATUS_QUEUE_FROZEN 0x40
355 /* SRB status integer codes */
356 #define SRB_STATUS_SUCCESS 0x01
357 #define SRB_STATUS_ABORTED 0x02
358 #define SRB_STATUS_ERROR 0x04
359 #define SRB_STATUS_INVALID_REQUEST 0x06
360 #define SRB_STATUS_DATA_OVERRUN 0x12
361 #define SRB_STATUS_INVALID_LUN 0x20
363 #define SRB_STATUS(status) \
364 (status & ~(SRB_STATUS_AUTOSENSE_VALID | SRB_STATUS_QUEUE_FROZEN))
366 * This is the end of Protocol specific defines.
369 static int storvsc_ringbuffer_size = (128 * 1024);
370 static u32 max_outstanding_req_per_channel;
371 static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth);
373 static int storvsc_vcpus_per_sub_channel = 4;
374 static unsigned int storvsc_max_hw_queues;
376 module_param(storvsc_ringbuffer_size, int, S_IRUGO);
377 MODULE_PARM_DESC(storvsc_ringbuffer_size, "Ring buffer size (bytes)");
379 module_param(storvsc_max_hw_queues, uint, 0644);
380 MODULE_PARM_DESC(storvsc_max_hw_queues, "Maximum number of hardware queues");
382 module_param(storvsc_vcpus_per_sub_channel, int, S_IRUGO);
383 MODULE_PARM_DESC(storvsc_vcpus_per_sub_channel, "Ratio of VCPUs to subchannels");
385 static int ring_avail_percent_lowater = 10;
386 module_param(ring_avail_percent_lowater, int, S_IRUGO);
387 MODULE_PARM_DESC(ring_avail_percent_lowater,
388 "Select a channel if available ring size > this in percent");
391 * Timeout in seconds for all devices managed by this driver.
393 static int storvsc_timeout = 180;
395 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
396 static struct scsi_transport_template *fc_transport_template;
399 static struct scsi_host_template scsi_driver;
400 static void storvsc_on_channel_callback(void *context);
402 #define STORVSC_MAX_LUNS_PER_TARGET 255
403 #define STORVSC_MAX_TARGETS 2
404 #define STORVSC_MAX_CHANNELS 8
406 #define STORVSC_FC_MAX_LUNS_PER_TARGET 255
407 #define STORVSC_FC_MAX_TARGETS 128
408 #define STORVSC_FC_MAX_CHANNELS 8
410 #define STORVSC_IDE_MAX_LUNS_PER_TARGET 64
411 #define STORVSC_IDE_MAX_TARGETS 1
412 #define STORVSC_IDE_MAX_CHANNELS 1
415 * Upper bound on the size of a storvsc packet. vmscsi_size_delta is not
416 * included in the calculation because it is set after STORVSC_MAX_PKT_SIZE
417 * is used in storvsc_connect_to_vsp
419 #define STORVSC_MAX_PKT_SIZE (sizeof(struct vmpacket_descriptor) +\
420 sizeof(struct vstor_packet))
422 struct storvsc_cmd_request {
423 struct scsi_cmnd *cmd;
425 struct hv_device *device;
427 /* Synchronize the request/response if needed */
428 struct completion wait_event;
430 struct vmbus_channel_packet_multipage_buffer mpb;
431 struct vmbus_packet_mpb_array *payload;
434 struct vstor_packet vstor_packet;
438 /* A storvsc device is a device object that contains a vmbus channel */
439 struct storvsc_device {
440 struct hv_device *device;
444 atomic_t num_outstanding_req;
445 struct Scsi_Host *host;
447 wait_queue_head_t waiting_to_drain;
450 * Each unique Port/Path/Target represents 1 channel ie scsi
451 * controller. In reality, the pathid, targetid is always 0
452 * and the port is set by us
454 unsigned int port_number;
455 unsigned char path_id;
456 unsigned char target_id;
459 * The size of the vmscsi_request has changed in win8. The
460 * additional size is because of new elements added to the
461 * structure. These elements are valid only when we are talking
463 * Track the correction to size we need to apply. This value
464 * will likely change during protocol negotiation but it is
465 * valid to start by assuming pre-Win8.
467 int vmscsi_size_delta;
470 * Max I/O, the device can support.
472 u32 max_transfer_bytes;
474 * Number of sub-channels we will open.
477 struct vmbus_channel **stor_chns;
479 * Mask of CPUs bound to subchannels.
481 struct cpumask alloced_cpus;
483 * Serializes modifications of stor_chns[] from storvsc_do_io()
484 * and storvsc_change_target_cpu().
487 /* Used for vsc/vsp channel reset process */
488 struct storvsc_cmd_request init_request;
489 struct storvsc_cmd_request reset_request;
491 * Currently active port and node names for FC devices.
495 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
496 struct fc_rport *rport;
500 struct hv_host_device {
501 struct hv_device *dev;
504 unsigned char target;
505 struct workqueue_struct *handle_error_wq;
506 struct work_struct host_scan_work;
507 struct Scsi_Host *host;
510 struct storvsc_scan_work {
511 struct work_struct work;
512 struct Scsi_Host *host;
517 static void storvsc_device_scan(struct work_struct *work)
519 struct storvsc_scan_work *wrk;
520 struct scsi_device *sdev;
522 wrk = container_of(work, struct storvsc_scan_work, work);
524 sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
527 scsi_rescan_device(&sdev->sdev_gendev);
528 scsi_device_put(sdev);
534 static void storvsc_host_scan(struct work_struct *work)
536 struct Scsi_Host *host;
537 struct scsi_device *sdev;
538 struct hv_host_device *host_device =
539 container_of(work, struct hv_host_device, host_scan_work);
541 host = host_device->host;
543 * Before scanning the host, first check to see if any of the
544 * currrently known devices have been hot removed. We issue a
545 * "unit ready" command against all currently known devices.
546 * This I/O will result in an error for devices that have been
547 * removed. As part of handling the I/O error, we remove the device.
549 * When a LUN is added or removed, the host sends us a signal to
550 * scan the host. Thus we are forced to discover the LUNs that
551 * may have been removed this way.
553 mutex_lock(&host->scan_mutex);
554 shost_for_each_device(sdev, host)
555 scsi_test_unit_ready(sdev, 1, 1, NULL);
556 mutex_unlock(&host->scan_mutex);
558 * Now scan the host to discover LUNs that may have been added.
560 scsi_scan_host(host);
563 static void storvsc_remove_lun(struct work_struct *work)
565 struct storvsc_scan_work *wrk;
566 struct scsi_device *sdev;
568 wrk = container_of(work, struct storvsc_scan_work, work);
569 if (!scsi_host_get(wrk->host))
572 sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
575 scsi_remove_device(sdev);
576 scsi_device_put(sdev);
578 scsi_host_put(wrk->host);
586 * We can get incoming messages from the host that are not in response to
587 * messages that we have sent out. An example of this would be messages
588 * received by the guest to notify dynamic addition/removal of LUNs. To
589 * deal with potential race conditions where the driver may be in the
590 * midst of being unloaded when we might receive an unsolicited message
591 * from the host, we have implemented a mechanism to gurantee sequential
594 * 1) Once the device is marked as being destroyed, we will fail all
596 * 2) We permit incoming messages when the device is being destroyed,
597 * only to properly account for messages already sent out.
600 static inline struct storvsc_device *get_out_stor_device(
601 struct hv_device *device)
603 struct storvsc_device *stor_device;
605 stor_device = hv_get_drvdata(device);
607 if (stor_device && stor_device->destroy)
614 static inline void storvsc_wait_to_drain(struct storvsc_device *dev)
616 dev->drain_notify = true;
617 wait_event(dev->waiting_to_drain,
618 atomic_read(&dev->num_outstanding_req) == 0);
619 dev->drain_notify = false;
622 static inline struct storvsc_device *get_in_stor_device(
623 struct hv_device *device)
625 struct storvsc_device *stor_device;
627 stor_device = hv_get_drvdata(device);
633 * If the device is being destroyed; allow incoming
634 * traffic only to cleanup outstanding requests.
637 if (stor_device->destroy &&
638 (atomic_read(&stor_device->num_outstanding_req) == 0))
646 static void storvsc_change_target_cpu(struct vmbus_channel *channel, u32 old,
649 struct storvsc_device *stor_device;
650 struct vmbus_channel *cur_chn;
651 bool old_is_alloced = false;
652 struct hv_device *device;
656 device = channel->primary_channel ?
657 channel->primary_channel->device_obj
658 : channel->device_obj;
659 stor_device = get_out_stor_device(device);
663 /* See storvsc_do_io() -> get_og_chn(). */
664 spin_lock_irqsave(&stor_device->lock, flags);
667 * Determines if the storvsc device has other channels assigned to
668 * the "old" CPU to update the alloced_cpus mask and the stor_chns
671 if (device->channel != channel && device->channel->target_cpu == old) {
672 cur_chn = device->channel;
673 old_is_alloced = true;
676 list_for_each_entry(cur_chn, &device->channel->sc_list, sc_list) {
677 if (cur_chn == channel)
679 if (cur_chn->target_cpu == old) {
680 old_is_alloced = true;
687 WRITE_ONCE(stor_device->stor_chns[old], cur_chn);
689 cpumask_clear_cpu(old, &stor_device->alloced_cpus);
691 /* "Flush" the stor_chns array. */
692 for_each_possible_cpu(cpu) {
693 if (stor_device->stor_chns[cpu] && !cpumask_test_cpu(
694 cpu, &stor_device->alloced_cpus))
695 WRITE_ONCE(stor_device->stor_chns[cpu], NULL);
698 WRITE_ONCE(stor_device->stor_chns[new], channel);
699 cpumask_set_cpu(new, &stor_device->alloced_cpus);
701 spin_unlock_irqrestore(&stor_device->lock, flags);
704 static u64 storvsc_next_request_id(struct vmbus_channel *channel, u64 rqst_addr)
706 struct storvsc_cmd_request *request =
707 (struct storvsc_cmd_request *)(unsigned long)rqst_addr;
709 if (rqst_addr == VMBUS_RQST_INIT)
710 return VMBUS_RQST_INIT;
711 if (rqst_addr == VMBUS_RQST_RESET)
712 return VMBUS_RQST_RESET;
715 * Cannot return an ID of 0, which is reserved for an unsolicited
716 * message from Hyper-V.
718 return (u64)blk_mq_unique_tag(scsi_cmd_to_rq(request->cmd)) + 1;
721 static void handle_sc_creation(struct vmbus_channel *new_sc)
723 struct hv_device *device = new_sc->primary_channel->device_obj;
724 struct device *dev = &device->device;
725 struct storvsc_device *stor_device;
726 struct vmstorage_channel_properties props;
729 stor_device = get_out_stor_device(device);
733 memset(&props, 0, sizeof(struct vmstorage_channel_properties));
734 new_sc->max_pkt_size = STORVSC_MAX_PKT_SIZE;
736 new_sc->next_request_id_callback = storvsc_next_request_id;
738 ret = vmbus_open(new_sc,
739 storvsc_ringbuffer_size,
740 storvsc_ringbuffer_size,
742 sizeof(struct vmstorage_channel_properties),
743 storvsc_on_channel_callback, new_sc);
745 /* In case vmbus_open() fails, we don't use the sub-channel. */
747 dev_err(dev, "Failed to open sub-channel: err=%d\n", ret);
751 new_sc->change_target_cpu_callback = storvsc_change_target_cpu;
753 /* Add the sub-channel to the array of available channels. */
754 stor_device->stor_chns[new_sc->target_cpu] = new_sc;
755 cpumask_set_cpu(new_sc->target_cpu, &stor_device->alloced_cpus);
758 static void handle_multichannel_storage(struct hv_device *device, int max_chns)
760 struct device *dev = &device->device;
761 struct storvsc_device *stor_device;
763 struct storvsc_cmd_request *request;
764 struct vstor_packet *vstor_packet;
768 * If the number of CPUs is artificially restricted, such as
769 * with maxcpus=1 on the kernel boot line, Hyper-V could offer
770 * sub-channels >= the number of CPUs. These sub-channels
771 * should not be created. The primary channel is already created
772 * and assigned to one CPU, so check against # CPUs - 1.
774 num_sc = min((int)(num_online_cpus() - 1), max_chns);
778 stor_device = get_out_stor_device(device);
782 stor_device->num_sc = num_sc;
783 request = &stor_device->init_request;
784 vstor_packet = &request->vstor_packet;
787 * Establish a handler for dealing with subchannels.
789 vmbus_set_sc_create_callback(device->channel, handle_sc_creation);
792 * Request the host to create sub-channels.
794 memset(request, 0, sizeof(struct storvsc_cmd_request));
795 init_completion(&request->wait_event);
796 vstor_packet->operation = VSTOR_OPERATION_CREATE_SUB_CHANNELS;
797 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
798 vstor_packet->sub_channel_count = num_sc;
800 ret = vmbus_sendpacket(device->channel, vstor_packet,
801 (sizeof(struct vstor_packet) -
802 stor_device->vmscsi_size_delta),
805 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
808 dev_err(dev, "Failed to create sub-channel: err=%d\n", ret);
812 t = wait_for_completion_timeout(&request->wait_event, 10*HZ);
814 dev_err(dev, "Failed to create sub-channel: timed out\n");
818 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
819 vstor_packet->status != 0) {
820 dev_err(dev, "Failed to create sub-channel: op=%d, sts=%d\n",
821 vstor_packet->operation, vstor_packet->status);
826 * We need to do nothing here, because vmbus_process_offer()
827 * invokes channel->sc_creation_callback, which will open and use
828 * the sub-channel(s).
832 static void cache_wwn(struct storvsc_device *stor_device,
833 struct vstor_packet *vstor_packet)
836 * Cache the currently active port and node ww names.
838 if (vstor_packet->wwn_packet.primary_active) {
839 stor_device->node_name =
840 wwn_to_u64(vstor_packet->wwn_packet.primary_node_wwn);
841 stor_device->port_name =
842 wwn_to_u64(vstor_packet->wwn_packet.primary_port_wwn);
844 stor_device->node_name =
845 wwn_to_u64(vstor_packet->wwn_packet.secondary_node_wwn);
846 stor_device->port_name =
847 wwn_to_u64(vstor_packet->wwn_packet.secondary_port_wwn);
852 static int storvsc_execute_vstor_op(struct hv_device *device,
853 struct storvsc_cmd_request *request,
856 struct storvsc_device *stor_device;
857 struct vstor_packet *vstor_packet;
860 stor_device = get_out_stor_device(device);
864 vstor_packet = &request->vstor_packet;
866 init_completion(&request->wait_event);
867 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
869 ret = vmbus_sendpacket(device->channel, vstor_packet,
870 (sizeof(struct vstor_packet) -
871 stor_device->vmscsi_size_delta),
874 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
878 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
885 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
886 vstor_packet->status != 0)
892 static int storvsc_channel_init(struct hv_device *device, bool is_fc)
894 struct storvsc_device *stor_device;
895 struct storvsc_cmd_request *request;
896 struct vstor_packet *vstor_packet;
899 bool process_sub_channels = false;
901 stor_device = get_out_stor_device(device);
905 request = &stor_device->init_request;
906 vstor_packet = &request->vstor_packet;
909 * Now, initiate the vsc/vsp initialization protocol on the open
912 memset(request, 0, sizeof(struct storvsc_cmd_request));
913 vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION;
914 ret = storvsc_execute_vstor_op(device, request, true);
918 * Query host supported protocol version.
921 for (i = 0; i < ARRAY_SIZE(vmstor_protocols); i++) {
922 /* reuse the packet for version range supported */
923 memset(vstor_packet, 0, sizeof(struct vstor_packet));
924 vstor_packet->operation =
925 VSTOR_OPERATION_QUERY_PROTOCOL_VERSION;
927 vstor_packet->version.major_minor =
928 vmstor_protocols[i].protocol_version;
931 * The revision number is only used in Windows; set it to 0.
933 vstor_packet->version.revision = 0;
934 ret = storvsc_execute_vstor_op(device, request, false);
938 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO)
941 if (vstor_packet->status == 0) {
942 vmstor_proto_version =
943 vmstor_protocols[i].protocol_version;
946 vmstor_protocols[i].sense_buffer_size;
948 stor_device->vmscsi_size_delta =
949 vmstor_protocols[i].vmscsi_size_delta;
955 if (vstor_packet->status != 0)
959 memset(vstor_packet, 0, sizeof(struct vstor_packet));
960 vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES;
961 ret = storvsc_execute_vstor_op(device, request, true);
966 * Check to see if multi-channel support is there.
967 * Hosts that implement protocol version of 5.1 and above
968 * support multi-channel.
970 max_chns = vstor_packet->storage_channel_properties.max_channel_cnt;
973 * Allocate state to manage the sub-channels.
974 * We allocate an array based on the numbers of possible CPUs
975 * (Hyper-V does not support cpu online/offline).
976 * This Array will be sparseley populated with unique
977 * channels - primary + sub-channels.
978 * We will however populate all the slots to evenly distribute
981 stor_device->stor_chns = kcalloc(num_possible_cpus(), sizeof(void *),
983 if (stor_device->stor_chns == NULL)
986 device->channel->change_target_cpu_callback = storvsc_change_target_cpu;
988 stor_device->stor_chns[device->channel->target_cpu] = device->channel;
989 cpumask_set_cpu(device->channel->target_cpu,
990 &stor_device->alloced_cpus);
992 if (vmstor_proto_version >= VMSTOR_PROTO_VERSION_WIN8) {
993 if (vstor_packet->storage_channel_properties.flags &
994 STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL)
995 process_sub_channels = true;
997 stor_device->max_transfer_bytes =
998 vstor_packet->storage_channel_properties.max_transfer_bytes;
1004 * For FC devices retrieve FC HBA data.
1006 memset(vstor_packet, 0, sizeof(struct vstor_packet));
1007 vstor_packet->operation = VSTOR_OPERATION_FCHBA_DATA;
1008 ret = storvsc_execute_vstor_op(device, request, true);
1013 * Cache the currently active port and node ww names.
1015 cache_wwn(stor_device, vstor_packet);
1019 memset(vstor_packet, 0, sizeof(struct vstor_packet));
1020 vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION;
1021 ret = storvsc_execute_vstor_op(device, request, true);
1025 if (process_sub_channels)
1026 handle_multichannel_storage(device, max_chns);
1031 static void storvsc_handle_error(struct vmscsi_request *vm_srb,
1032 struct scsi_cmnd *scmnd,
1033 struct Scsi_Host *host,
1036 struct storvsc_scan_work *wrk;
1037 void (*process_err_fn)(struct work_struct *work);
1038 struct hv_host_device *host_dev = shost_priv(host);
1040 switch (SRB_STATUS(vm_srb->srb_status)) {
1041 case SRB_STATUS_ERROR:
1042 case SRB_STATUS_ABORTED:
1043 case SRB_STATUS_INVALID_REQUEST:
1044 if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID) {
1045 /* Check for capacity change */
1046 if ((asc == 0x2a) && (ascq == 0x9)) {
1047 process_err_fn = storvsc_device_scan;
1048 /* Retry the I/O that triggered this. */
1049 set_host_byte(scmnd, DID_REQUEUE);
1054 * Otherwise, let upper layer deal with the
1055 * error when sense message is present
1061 * If there is an error; offline the device since all
1062 * error recovery strategies would have already been
1063 * deployed on the host side. However, if the command
1064 * were a pass-through command deal with it appropriately.
1066 switch (scmnd->cmnd[0]) {
1069 set_host_byte(scmnd, DID_PASSTHROUGH);
1072 * On some Hyper-V hosts TEST_UNIT_READY command can
1073 * return SRB_STATUS_ERROR. Let the upper level code
1074 * deal with it based on the sense information.
1076 case TEST_UNIT_READY:
1079 set_host_byte(scmnd, DID_ERROR);
1083 case SRB_STATUS_INVALID_LUN:
1084 set_host_byte(scmnd, DID_NO_CONNECT);
1085 process_err_fn = storvsc_remove_lun;
1093 * We need to schedule work to process this error; schedule it.
1095 wrk = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC);
1097 set_host_byte(scmnd, DID_TARGET_FAILURE);
1102 wrk->lun = vm_srb->lun;
1103 wrk->tgt_id = vm_srb->target_id;
1104 INIT_WORK(&wrk->work, process_err_fn);
1105 queue_work(host_dev->handle_error_wq, &wrk->work);
1109 static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request,
1110 struct storvsc_device *stor_dev)
1112 struct scsi_cmnd *scmnd = cmd_request->cmd;
1113 struct scsi_sense_hdr sense_hdr;
1114 struct vmscsi_request *vm_srb;
1115 u32 data_transfer_length;
1116 struct Scsi_Host *host;
1117 u32 payload_sz = cmd_request->payload_sz;
1118 void *payload = cmd_request->payload;
1121 host = stor_dev->host;
1123 vm_srb = &cmd_request->vstor_packet.vm_srb;
1124 data_transfer_length = vm_srb->data_transfer_length;
1126 scmnd->result = vm_srb->scsi_status;
1128 if (scmnd->result) {
1129 sense_ok = scsi_normalize_sense(scmnd->sense_buffer,
1130 SCSI_SENSE_BUFFERSIZE, &sense_hdr);
1132 if (sense_ok && do_logging(STORVSC_LOGGING_WARN))
1133 scsi_print_sense_hdr(scmnd->device, "storvsc",
1137 if (vm_srb->srb_status != SRB_STATUS_SUCCESS) {
1138 storvsc_handle_error(vm_srb, scmnd, host, sense_hdr.asc,
1141 * The Windows driver set data_transfer_length on
1142 * SRB_STATUS_DATA_OVERRUN. On other errors, this value
1143 * is untouched. In these cases we set it to 0.
1145 if (vm_srb->srb_status != SRB_STATUS_DATA_OVERRUN)
1146 data_transfer_length = 0;
1149 /* Validate data_transfer_length (from Hyper-V) */
1150 if (data_transfer_length > cmd_request->payload->range.len)
1151 data_transfer_length = cmd_request->payload->range.len;
1153 scsi_set_resid(scmnd,
1154 cmd_request->payload->range.len - data_transfer_length);
1156 scmnd->scsi_done(scmnd);
1159 sizeof(struct vmbus_channel_packet_multipage_buffer))
1163 static void storvsc_on_io_completion(struct storvsc_device *stor_device,
1164 struct vstor_packet *vstor_packet,
1165 struct storvsc_cmd_request *request)
1167 struct vstor_packet *stor_pkt;
1168 struct hv_device *device = stor_device->device;
1170 stor_pkt = &request->vstor_packet;
1173 * The current SCSI handling on the host side does
1174 * not correctly handle:
1175 * INQUIRY command with page code parameter set to 0x80
1176 * MODE_SENSE command with cmd[2] == 0x1c
1178 * Setup srb and scsi status so this won't be fatal.
1179 * We do this so we can distinguish truly fatal failues
1180 * (srb status == 0x4) and off-line the device in that case.
1183 if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) ||
1184 (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) {
1185 vstor_packet->vm_srb.scsi_status = 0;
1186 vstor_packet->vm_srb.srb_status = SRB_STATUS_SUCCESS;
1189 /* Copy over the status...etc */
1190 stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status;
1191 stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status;
1194 * Copy over the sense_info_length, but limit to the known max
1195 * size if Hyper-V returns a bad value.
1197 stor_pkt->vm_srb.sense_info_length = min_t(u8, sense_buffer_size,
1198 vstor_packet->vm_srb.sense_info_length);
1200 if (vstor_packet->vm_srb.scsi_status != 0 ||
1201 vstor_packet->vm_srb.srb_status != SRB_STATUS_SUCCESS) {
1204 * Log TEST_UNIT_READY errors only as warnings. Hyper-V can
1205 * return errors when detecting devices using TEST_UNIT_READY,
1206 * and logging these as errors produces unhelpful noise.
1208 int loglevel = (stor_pkt->vm_srb.cdb[0] == TEST_UNIT_READY) ?
1209 STORVSC_LOGGING_WARN : STORVSC_LOGGING_ERROR;
1211 storvsc_log(device, loglevel,
1212 "tag#%d cmd 0x%x status: scsi 0x%x srb 0x%x hv 0x%x\n",
1213 scsi_cmd_to_rq(request->cmd)->tag,
1214 stor_pkt->vm_srb.cdb[0],
1215 vstor_packet->vm_srb.scsi_status,
1216 vstor_packet->vm_srb.srb_status,
1217 vstor_packet->status);
1220 if (vstor_packet->vm_srb.scsi_status == SAM_STAT_CHECK_CONDITION &&
1221 (vstor_packet->vm_srb.srb_status & SRB_STATUS_AUTOSENSE_VALID))
1222 memcpy(request->cmd->sense_buffer,
1223 vstor_packet->vm_srb.sense_data,
1224 stor_pkt->vm_srb.sense_info_length);
1226 stor_pkt->vm_srb.data_transfer_length =
1227 vstor_packet->vm_srb.data_transfer_length;
1229 storvsc_command_completion(request, stor_device);
1231 if (atomic_dec_and_test(&stor_device->num_outstanding_req) &&
1232 stor_device->drain_notify)
1233 wake_up(&stor_device->waiting_to_drain);
1236 static void storvsc_on_receive(struct storvsc_device *stor_device,
1237 struct vstor_packet *vstor_packet,
1238 struct storvsc_cmd_request *request)
1240 struct hv_host_device *host_dev;
1241 switch (vstor_packet->operation) {
1242 case VSTOR_OPERATION_COMPLETE_IO:
1243 storvsc_on_io_completion(stor_device, vstor_packet, request);
1246 case VSTOR_OPERATION_REMOVE_DEVICE:
1247 case VSTOR_OPERATION_ENUMERATE_BUS:
1248 host_dev = shost_priv(stor_device->host);
1250 host_dev->handle_error_wq, &host_dev->host_scan_work);
1253 case VSTOR_OPERATION_FCHBA_DATA:
1254 cache_wwn(stor_device, vstor_packet);
1255 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1256 fc_host_node_name(stor_device->host) = stor_device->node_name;
1257 fc_host_port_name(stor_device->host) = stor_device->port_name;
1265 static void storvsc_on_channel_callback(void *context)
1267 struct vmbus_channel *channel = (struct vmbus_channel *)context;
1268 const struct vmpacket_descriptor *desc;
1269 struct hv_device *device;
1270 struct storvsc_device *stor_device;
1271 struct Scsi_Host *shost;
1273 if (channel->primary_channel != NULL)
1274 device = channel->primary_channel->device_obj;
1276 device = channel->device_obj;
1278 stor_device = get_in_stor_device(device);
1282 shost = stor_device->host;
1284 foreach_vmbus_pkt(desc, channel) {
1285 struct vstor_packet *packet = hv_pkt_data(desc);
1286 struct storvsc_cmd_request *request = NULL;
1287 u32 pktlen = hv_pkt_datalen(desc);
1288 u64 rqst_id = desc->trans_id;
1289 u32 minlen = rqst_id ? sizeof(struct vstor_packet) -
1290 stor_device->vmscsi_size_delta : sizeof(enum vstor_packet_operation);
1292 if (pktlen < minlen) {
1293 dev_err(&device->device,
1294 "Invalid pkt: id=%llu, len=%u, minlen=%u\n",
1295 rqst_id, pktlen, minlen);
1299 if (rqst_id == VMBUS_RQST_INIT) {
1300 request = &stor_device->init_request;
1301 } else if (rqst_id == VMBUS_RQST_RESET) {
1302 request = &stor_device->reset_request;
1304 /* Hyper-V can send an unsolicited message with ID of 0 */
1307 * storvsc_on_receive() looks at the vstor_packet in the message
1308 * from the ring buffer.
1310 * - If the operation in the vstor_packet is COMPLETE_IO, then
1311 * we call storvsc_on_io_completion(), and dereference the
1312 * guest memory address. Make sure we don't call
1313 * storvsc_on_io_completion() with a guest memory address
1314 * that is zero if Hyper-V were to construct and send such
1317 * - If the operation in the vstor_packet is FCHBA_DATA, then
1318 * we call cache_wwn(), and access the data payload area of
1319 * the packet (wwn_packet); however, there is no guarantee
1320 * that the packet is big enough to contain such area.
1321 * Future-proof the code by rejecting such a bogus packet.
1323 if (packet->operation == VSTOR_OPERATION_COMPLETE_IO ||
1324 packet->operation == VSTOR_OPERATION_FCHBA_DATA) {
1325 dev_err(&device->device, "Invalid packet with ID of 0\n");
1329 struct scsi_cmnd *scmnd;
1331 /* Transaction 'rqst_id' corresponds to tag 'rqst_id - 1' */
1332 scmnd = scsi_host_find_tag(shost, rqst_id - 1);
1333 if (scmnd == NULL) {
1334 dev_err(&device->device, "Incorrect transaction ID\n");
1337 request = (struct storvsc_cmd_request *)scsi_cmd_priv(scmnd);
1340 storvsc_on_receive(stor_device, packet, request);
1344 memcpy(&request->vstor_packet, packet,
1345 (sizeof(struct vstor_packet) - stor_device->vmscsi_size_delta));
1346 complete(&request->wait_event);
1350 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size,
1353 struct vmstorage_channel_properties props;
1356 memset(&props, 0, sizeof(struct vmstorage_channel_properties));
1358 device->channel->max_pkt_size = STORVSC_MAX_PKT_SIZE;
1359 device->channel->next_request_id_callback = storvsc_next_request_id;
1361 ret = vmbus_open(device->channel,
1365 sizeof(struct vmstorage_channel_properties),
1366 storvsc_on_channel_callback, device->channel);
1371 ret = storvsc_channel_init(device, is_fc);
1376 static int storvsc_dev_remove(struct hv_device *device)
1378 struct storvsc_device *stor_device;
1380 stor_device = hv_get_drvdata(device);
1382 stor_device->destroy = true;
1384 /* Make sure flag is set before waiting */
1388 * At this point, all outbound traffic should be disable. We
1389 * only allow inbound traffic (responses) to proceed so that
1390 * outstanding requests can be completed.
1393 storvsc_wait_to_drain(stor_device);
1396 * Since we have already drained, we don't need to busy wait
1397 * as was done in final_release_stor_device()
1398 * Note that we cannot set the ext pointer to NULL until
1399 * we have drained - to drain the outgoing packets, we need to
1400 * allow incoming packets.
1402 hv_set_drvdata(device, NULL);
1404 /* Close the channel */
1405 vmbus_close(device->channel);
1407 kfree(stor_device->stor_chns);
1412 static struct vmbus_channel *get_og_chn(struct storvsc_device *stor_device,
1417 const struct cpumask *node_mask;
1418 int num_channels, tgt_cpu;
1420 if (stor_device->num_sc == 0) {
1421 stor_device->stor_chns[q_num] = stor_device->device->channel;
1422 return stor_device->device->channel;
1426 * Our channel array is sparsley populated and we
1427 * initiated I/O on a processor/hw-q that does not
1428 * currently have a designated channel. Fix this.
1429 * The strategy is simple:
1430 * I. Ensure NUMA locality
1431 * II. Distribute evenly (best effort)
1434 node_mask = cpumask_of_node(cpu_to_node(q_num));
1437 for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1438 if (cpumask_test_cpu(tgt_cpu, node_mask))
1441 if (num_channels == 0) {
1442 stor_device->stor_chns[q_num] = stor_device->device->channel;
1443 return stor_device->device->channel;
1447 while (hash_qnum >= num_channels)
1448 hash_qnum -= num_channels;
1450 for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1451 if (!cpumask_test_cpu(tgt_cpu, node_mask))
1453 if (slot == hash_qnum)
1458 stor_device->stor_chns[q_num] = stor_device->stor_chns[tgt_cpu];
1460 return stor_device->stor_chns[q_num];
1464 static int storvsc_do_io(struct hv_device *device,
1465 struct storvsc_cmd_request *request, u16 q_num)
1467 struct storvsc_device *stor_device;
1468 struct vstor_packet *vstor_packet;
1469 struct vmbus_channel *outgoing_channel, *channel;
1470 unsigned long flags;
1472 const struct cpumask *node_mask;
1475 vstor_packet = &request->vstor_packet;
1476 stor_device = get_out_stor_device(device);
1482 request->device = device;
1484 * Select an appropriate channel to send the request out.
1486 /* See storvsc_change_target_cpu(). */
1487 outgoing_channel = READ_ONCE(stor_device->stor_chns[q_num]);
1488 if (outgoing_channel != NULL) {
1489 if (outgoing_channel->target_cpu == q_num) {
1491 * Ideally, we want to pick a different channel if
1492 * available on the same NUMA node.
1494 node_mask = cpumask_of_node(cpu_to_node(q_num));
1495 for_each_cpu_wrap(tgt_cpu,
1496 &stor_device->alloced_cpus, q_num + 1) {
1497 if (!cpumask_test_cpu(tgt_cpu, node_mask))
1499 if (tgt_cpu == q_num)
1501 channel = READ_ONCE(
1502 stor_device->stor_chns[tgt_cpu]);
1503 if (channel == NULL)
1505 if (hv_get_avail_to_write_percent(
1507 > ring_avail_percent_lowater) {
1508 outgoing_channel = channel;
1514 * All the other channels on the same NUMA node are
1515 * busy. Try to use the channel on the current CPU
1517 if (hv_get_avail_to_write_percent(
1518 &outgoing_channel->outbound)
1519 > ring_avail_percent_lowater)
1523 * If we reach here, all the channels on the current
1524 * NUMA node are busy. Try to find a channel in
1527 for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1528 if (cpumask_test_cpu(tgt_cpu, node_mask))
1530 channel = READ_ONCE(
1531 stor_device->stor_chns[tgt_cpu]);
1532 if (channel == NULL)
1534 if (hv_get_avail_to_write_percent(
1536 > ring_avail_percent_lowater) {
1537 outgoing_channel = channel;
1543 spin_lock_irqsave(&stor_device->lock, flags);
1544 outgoing_channel = stor_device->stor_chns[q_num];
1545 if (outgoing_channel != NULL) {
1546 spin_unlock_irqrestore(&stor_device->lock, flags);
1549 outgoing_channel = get_og_chn(stor_device, q_num);
1550 spin_unlock_irqrestore(&stor_device->lock, flags);
1554 vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
1556 vstor_packet->vm_srb.length = (sizeof(struct vmscsi_request) -
1557 stor_device->vmscsi_size_delta);
1560 vstor_packet->vm_srb.sense_info_length = sense_buffer_size;
1563 vstor_packet->vm_srb.data_transfer_length =
1564 request->payload->range.len;
1566 vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
1568 if (request->payload->range.len) {
1570 ret = vmbus_sendpacket_mpb_desc(outgoing_channel,
1571 request->payload, request->payload_sz,
1573 (sizeof(struct vstor_packet) -
1574 stor_device->vmscsi_size_delta),
1575 (unsigned long)request);
1577 ret = vmbus_sendpacket(outgoing_channel, vstor_packet,
1578 (sizeof(struct vstor_packet) -
1579 stor_device->vmscsi_size_delta),
1580 (unsigned long)request,
1582 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1588 atomic_inc(&stor_device->num_outstanding_req);
1593 static int storvsc_device_alloc(struct scsi_device *sdevice)
1596 * Set blist flag to permit the reading of the VPD pages even when
1597 * the target may claim SPC-2 compliance. MSFT targets currently
1598 * claim SPC-2 compliance while they implement post SPC-2 features.
1599 * With this flag we can correctly handle WRITE_SAME_16 issues.
1601 * Hypervisor reports SCSI_UNKNOWN type for DVD ROM device but
1602 * still supports REPORT LUN.
1604 sdevice->sdev_bflags = BLIST_REPORTLUN2 | BLIST_TRY_VPD_PAGES;
1609 static int storvsc_device_configure(struct scsi_device *sdevice)
1611 blk_queue_rq_timeout(sdevice->request_queue, (storvsc_timeout * HZ));
1613 sdevice->no_write_same = 1;
1616 * If the host is WIN8 or WIN8 R2, claim conformance to SPC-3
1617 * if the device is a MSFT virtual device. If the host is
1618 * WIN10 or newer, allow write_same.
1620 if (!strncmp(sdevice->vendor, "Msft", 4)) {
1621 switch (vmstor_proto_version) {
1622 case VMSTOR_PROTO_VERSION_WIN8:
1623 case VMSTOR_PROTO_VERSION_WIN8_1:
1624 sdevice->scsi_level = SCSI_SPC_3;
1628 if (vmstor_proto_version >= VMSTOR_PROTO_VERSION_WIN10)
1629 sdevice->no_write_same = 0;
1635 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev,
1636 sector_t capacity, int *info)
1638 sector_t nsect = capacity;
1639 sector_t cylinders = nsect;
1640 int heads, sectors_pt;
1643 * We are making up these values; let us keep it simple.
1646 sectors_pt = 0x3f; /* Sectors per track */
1647 sector_div(cylinders, heads * sectors_pt);
1648 if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
1652 info[1] = sectors_pt;
1653 info[2] = (int)cylinders;
1658 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
1660 struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
1661 struct hv_device *device = host_dev->dev;
1663 struct storvsc_device *stor_device;
1664 struct storvsc_cmd_request *request;
1665 struct vstor_packet *vstor_packet;
1668 stor_device = get_out_stor_device(device);
1672 request = &stor_device->reset_request;
1673 vstor_packet = &request->vstor_packet;
1674 memset(vstor_packet, 0, sizeof(struct vstor_packet));
1676 init_completion(&request->wait_event);
1678 vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
1679 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
1680 vstor_packet->vm_srb.path_id = stor_device->path_id;
1682 ret = vmbus_sendpacket(device->channel, vstor_packet,
1683 (sizeof(struct vstor_packet) -
1684 stor_device->vmscsi_size_delta),
1687 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1691 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
1693 return TIMEOUT_ERROR;
1697 * At this point, all outstanding requests in the adapter
1698 * should have been flushed out and return to us
1699 * There is a potential race here where the host may be in
1700 * the process of responding when we return from here.
1701 * Just wait for all in-transit packets to be accounted for
1702 * before we return from here.
1704 storvsc_wait_to_drain(stor_device);
1710 * The host guarantees to respond to each command, although I/O latencies might
1711 * be unbounded on Azure. Reset the timer unconditionally to give the host a
1712 * chance to perform EH.
1714 static enum blk_eh_timer_return storvsc_eh_timed_out(struct scsi_cmnd *scmnd)
1716 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1717 if (scmnd->device->host->transportt == fc_transport_template)
1718 return fc_eh_timed_out(scmnd);
1720 return BLK_EH_RESET_TIMER;
1723 static bool storvsc_scsi_cmd_ok(struct scsi_cmnd *scmnd)
1725 bool allowed = true;
1726 u8 scsi_op = scmnd->cmnd[0];
1729 /* the host does not handle WRITE_SAME, log accident usage */
1732 * smartd sends this command and the host does not handle
1733 * this. So, don't send it.
1736 set_host_byte(scmnd, DID_ERROR);
1745 static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
1748 struct hv_host_device *host_dev = shost_priv(host);
1749 struct hv_device *dev = host_dev->dev;
1750 struct storvsc_cmd_request *cmd_request = scsi_cmd_priv(scmnd);
1752 struct scatterlist *sgl;
1753 unsigned int sg_count;
1754 struct vmscsi_request *vm_srb;
1755 struct vmbus_packet_mpb_array *payload;
1759 if (vmstor_proto_version <= VMSTOR_PROTO_VERSION_WIN8) {
1761 * On legacy hosts filter unimplemented commands.
1762 * Future hosts are expected to correctly handle
1763 * unsupported commands. Furthermore, it is
1764 * possible that some of the currently
1765 * unsupported commands maybe supported in
1766 * future versions of the host.
1768 if (!storvsc_scsi_cmd_ok(scmnd)) {
1769 scmnd->scsi_done(scmnd);
1774 /* Setup the cmd request */
1775 cmd_request->cmd = scmnd;
1777 memset(&cmd_request->vstor_packet, 0, sizeof(struct vstor_packet));
1778 vm_srb = &cmd_request->vstor_packet.vm_srb;
1779 vm_srb->win8_extension.time_out_value = 60;
1781 vm_srb->win8_extension.srb_flags |=
1782 SRB_FLAGS_DISABLE_SYNCH_TRANSFER;
1784 if (scmnd->device->tagged_supported) {
1785 vm_srb->win8_extension.srb_flags |=
1786 (SRB_FLAGS_QUEUE_ACTION_ENABLE | SRB_FLAGS_NO_QUEUE_FREEZE);
1787 vm_srb->win8_extension.queue_tag = SP_UNTAGGED;
1788 vm_srb->win8_extension.queue_action = SRB_SIMPLE_TAG_REQUEST;
1792 switch (scmnd->sc_data_direction) {
1794 vm_srb->data_in = WRITE_TYPE;
1795 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_DATA_OUT;
1797 case DMA_FROM_DEVICE:
1798 vm_srb->data_in = READ_TYPE;
1799 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_DATA_IN;
1802 vm_srb->data_in = UNKNOWN_TYPE;
1803 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_NO_DATA_TRANSFER;
1807 * This is DMA_BIDIRECTIONAL or something else we are never
1808 * supposed to see here.
1810 WARN(1, "Unexpected data direction: %d\n",
1811 scmnd->sc_data_direction);
1816 vm_srb->port_number = host_dev->port;
1817 vm_srb->path_id = scmnd->device->channel;
1818 vm_srb->target_id = scmnd->device->id;
1819 vm_srb->lun = scmnd->device->lun;
1821 vm_srb->cdb_length = scmnd->cmd_len;
1823 memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
1825 sgl = (struct scatterlist *)scsi_sglist(scmnd);
1826 sg_count = scsi_sg_count(scmnd);
1828 length = scsi_bufflen(scmnd);
1829 payload = (struct vmbus_packet_mpb_array *)&cmd_request->mpb;
1830 payload_sz = sizeof(cmd_request->mpb);
1833 unsigned int hvpgoff, hvpfns_to_add;
1834 unsigned long offset_in_hvpg = offset_in_hvpage(sgl->offset);
1835 unsigned int hvpg_count = HVPFN_UP(offset_in_hvpg + length);
1838 if (hvpg_count > MAX_PAGE_BUFFER_COUNT) {
1840 payload_sz = (hvpg_count * sizeof(u64) +
1841 sizeof(struct vmbus_packet_mpb_array));
1842 payload = kzalloc(payload_sz, GFP_ATOMIC);
1844 return SCSI_MLQUEUE_DEVICE_BUSY;
1847 payload->range.len = length;
1848 payload->range.offset = offset_in_hvpg;
1851 for (i = 0; sgl != NULL; sgl = sg_next(sgl)) {
1853 * Init values for the current sgl entry. hvpgoff
1854 * and hvpfns_to_add are in units of Hyper-V size
1855 * pages. Handling the PAGE_SIZE != HV_HYP_PAGE_SIZE
1856 * case also handles values of sgl->offset that are
1857 * larger than PAGE_SIZE. Such offsets are handled
1858 * even on other than the first sgl entry, provided
1859 * they are a multiple of PAGE_SIZE.
1861 hvpgoff = HVPFN_DOWN(sgl->offset);
1862 hvpfn = page_to_hvpfn(sg_page(sgl)) + hvpgoff;
1863 hvpfns_to_add = HVPFN_UP(sgl->offset + sgl->length) -
1867 * Fill the next portion of the PFN array with
1868 * sequential Hyper-V PFNs for the continguous physical
1869 * memory described by the sgl entry. The end of the
1870 * last sgl should be reached at the same time that
1871 * the PFN array is filled.
1873 while (hvpfns_to_add--)
1874 payload->range.pfn_array[i++] = hvpfn++;
1878 cmd_request->payload = payload;
1879 cmd_request->payload_sz = payload_sz;
1881 /* Invokes the vsc to start an IO */
1882 ret = storvsc_do_io(dev, cmd_request, get_cpu());
1885 if (ret == -EAGAIN) {
1886 if (payload_sz > sizeof(cmd_request->mpb))
1889 return SCSI_MLQUEUE_DEVICE_BUSY;
1895 static struct scsi_host_template scsi_driver = {
1896 .module = THIS_MODULE,
1897 .name = "storvsc_host_t",
1898 .cmd_size = sizeof(struct storvsc_cmd_request),
1899 .bios_param = storvsc_get_chs,
1900 .queuecommand = storvsc_queuecommand,
1901 .eh_host_reset_handler = storvsc_host_reset_handler,
1902 .proc_name = "storvsc_host",
1903 .eh_timed_out = storvsc_eh_timed_out,
1904 .slave_alloc = storvsc_device_alloc,
1905 .slave_configure = storvsc_device_configure,
1906 .cmd_per_lun = 2048,
1908 /* Ensure there are no gaps in presented sgls */
1909 .virt_boundary_mask = HV_HYP_PAGE_SIZE - 1,
1911 .track_queue_depth = 1,
1912 .change_queue_depth = storvsc_change_queue_depth,
1921 static const struct hv_vmbus_device_id id_table[] = {
1924 .driver_data = SCSI_GUID
1928 .driver_data = IDE_GUID
1930 /* Fibre Channel GUID */
1933 .driver_data = SFC_GUID
1938 MODULE_DEVICE_TABLE(vmbus, id_table);
1940 static const struct { guid_t guid; } fc_guid = { HV_SYNTHFC_GUID };
1942 static bool hv_dev_is_fc(struct hv_device *hv_dev)
1944 return guid_equal(&fc_guid.guid, &hv_dev->dev_type);
1947 static int storvsc_probe(struct hv_device *device,
1948 const struct hv_vmbus_device_id *dev_id)
1951 int num_cpus = num_online_cpus();
1952 int num_present_cpus = num_present_cpus();
1953 struct Scsi_Host *host;
1954 struct hv_host_device *host_dev;
1955 bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false);
1956 bool is_fc = ((dev_id->driver_data == SFC_GUID) ? true : false);
1958 struct storvsc_device *stor_device;
1959 int max_luns_per_target;
1962 int max_sub_channels = 0;
1966 * Based on the windows host we are running on,
1967 * set state to properly communicate with the host.
1970 if (vmbus_proto_version < VERSION_WIN8) {
1971 max_luns_per_target = STORVSC_IDE_MAX_LUNS_PER_TARGET;
1972 max_targets = STORVSC_IDE_MAX_TARGETS;
1973 max_channels = STORVSC_IDE_MAX_CHANNELS;
1975 max_luns_per_target = STORVSC_MAX_LUNS_PER_TARGET;
1976 max_targets = STORVSC_MAX_TARGETS;
1977 max_channels = STORVSC_MAX_CHANNELS;
1979 * On Windows8 and above, we support sub-channels for storage
1980 * on SCSI and FC controllers.
1981 * The number of sub-channels offerred is based on the number of
1982 * VCPUs in the guest.
1986 (num_cpus - 1) / storvsc_vcpus_per_sub_channel;
1989 scsi_driver.can_queue = max_outstanding_req_per_channel *
1990 (max_sub_channels + 1) *
1991 (100 - ring_avail_percent_lowater) / 100;
1993 host = scsi_host_alloc(&scsi_driver,
1994 sizeof(struct hv_host_device));
1998 host_dev = shost_priv(host);
1999 memset(host_dev, 0, sizeof(struct hv_host_device));
2001 host_dev->port = host->host_no;
2002 host_dev->dev = device;
2003 host_dev->host = host;
2006 stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
2012 stor_device->destroy = false;
2013 init_waitqueue_head(&stor_device->waiting_to_drain);
2014 stor_device->device = device;
2015 stor_device->host = host;
2016 stor_device->vmscsi_size_delta = sizeof(struct vmscsi_win8_extension);
2017 spin_lock_init(&stor_device->lock);
2018 hv_set_drvdata(device, stor_device);
2020 stor_device->port_number = host->host_no;
2021 ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size, is_fc);
2025 host_dev->path = stor_device->path_id;
2026 host_dev->target = stor_device->target_id;
2028 switch (dev_id->driver_data) {
2030 host->max_lun = STORVSC_FC_MAX_LUNS_PER_TARGET;
2031 host->max_id = STORVSC_FC_MAX_TARGETS;
2032 host->max_channel = STORVSC_FC_MAX_CHANNELS - 1;
2033 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2034 host->transportt = fc_transport_template;
2039 host->max_lun = max_luns_per_target;
2040 host->max_id = max_targets;
2041 host->max_channel = max_channels - 1;
2045 host->max_lun = STORVSC_IDE_MAX_LUNS_PER_TARGET;
2046 host->max_id = STORVSC_IDE_MAX_TARGETS;
2047 host->max_channel = STORVSC_IDE_MAX_CHANNELS - 1;
2050 /* max cmd length */
2051 host->max_cmd_len = STORVSC_MAX_CMD_LEN;
2053 * Any reasonable Hyper-V configuration should provide
2054 * max_transfer_bytes value aligning to HV_HYP_PAGE_SIZE,
2055 * protecting it from any weird value.
2057 max_xfer_bytes = round_down(stor_device->max_transfer_bytes, HV_HYP_PAGE_SIZE);
2058 /* max_hw_sectors_kb */
2059 host->max_sectors = max_xfer_bytes >> 9;
2061 * There are 2 requirements for Hyper-V storvsc sgl segments,
2062 * based on which the below calculation for max segments is
2065 * 1. Except for the first and last sgl segment, all sgl segments
2066 * should be align to HV_HYP_PAGE_SIZE, that also means the
2067 * maximum number of segments in a sgl can be calculated by
2068 * dividing the total max transfer length by HV_HYP_PAGE_SIZE.
2070 * 2. Except for the first and last, each entry in the SGL must
2071 * have an offset that is a multiple of HV_HYP_PAGE_SIZE.
2073 host->sg_tablesize = (max_xfer_bytes >> HV_HYP_PAGE_SHIFT) + 1;
2075 * For non-IDE disks, the host supports multiple channels.
2076 * Set the number of HW queues we are supporting.
2079 if (storvsc_max_hw_queues > num_present_cpus) {
2080 storvsc_max_hw_queues = 0;
2081 storvsc_log(device, STORVSC_LOGGING_WARN,
2082 "Resetting invalid storvsc_max_hw_queues value to default.\n");
2084 if (storvsc_max_hw_queues)
2085 host->nr_hw_queues = storvsc_max_hw_queues;
2087 host->nr_hw_queues = num_present_cpus;
2091 * Set the error handler work queue.
2093 host_dev->handle_error_wq =
2094 alloc_ordered_workqueue("storvsc_error_wq_%d",
2097 if (!host_dev->handle_error_wq) {
2101 INIT_WORK(&host_dev->host_scan_work, storvsc_host_scan);
2102 /* Register the HBA and start the scsi bus scan */
2103 ret = scsi_add_host(host, &device->device);
2108 scsi_scan_host(host);
2110 target = (device->dev_instance.b[5] << 8 |
2111 device->dev_instance.b[4]);
2112 ret = scsi_add_device(host, 0, target, 0);
2116 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2117 if (host->transportt == fc_transport_template) {
2118 struct fc_rport_identifiers ids = {
2119 .roles = FC_PORT_ROLE_FCP_DUMMY_INITIATOR,
2122 fc_host_node_name(host) = stor_device->node_name;
2123 fc_host_port_name(host) = stor_device->port_name;
2124 stor_device->rport = fc_remote_port_add(host, 0, &ids);
2125 if (!stor_device->rport) {
2134 scsi_remove_host(host);
2137 destroy_workqueue(host_dev->handle_error_wq);
2141 * Once we have connected with the host, we would need to
2142 * to invoke storvsc_dev_remove() to rollback this state and
2143 * this call also frees up the stor_device; hence the jump around
2146 storvsc_dev_remove(device);
2150 kfree(stor_device->stor_chns);
2154 scsi_host_put(host);
2158 /* Change a scsi target's queue depth */
2159 static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth)
2161 if (queue_depth > scsi_driver.can_queue)
2162 queue_depth = scsi_driver.can_queue;
2164 return scsi_change_queue_depth(sdev, queue_depth);
2167 static int storvsc_remove(struct hv_device *dev)
2169 struct storvsc_device *stor_device = hv_get_drvdata(dev);
2170 struct Scsi_Host *host = stor_device->host;
2171 struct hv_host_device *host_dev = shost_priv(host);
2173 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2174 if (host->transportt == fc_transport_template) {
2175 fc_remote_port_delete(stor_device->rport);
2176 fc_remove_host(host);
2179 destroy_workqueue(host_dev->handle_error_wq);
2180 scsi_remove_host(host);
2181 storvsc_dev_remove(dev);
2182 scsi_host_put(host);
2187 static int storvsc_suspend(struct hv_device *hv_dev)
2189 struct storvsc_device *stor_device = hv_get_drvdata(hv_dev);
2190 struct Scsi_Host *host = stor_device->host;
2191 struct hv_host_device *host_dev = shost_priv(host);
2193 storvsc_wait_to_drain(stor_device);
2195 drain_workqueue(host_dev->handle_error_wq);
2197 vmbus_close(hv_dev->channel);
2199 kfree(stor_device->stor_chns);
2200 stor_device->stor_chns = NULL;
2202 cpumask_clear(&stor_device->alloced_cpus);
2207 static int storvsc_resume(struct hv_device *hv_dev)
2211 ret = storvsc_connect_to_vsp(hv_dev, storvsc_ringbuffer_size,
2212 hv_dev_is_fc(hv_dev));
2216 static struct hv_driver storvsc_drv = {
2217 .name = KBUILD_MODNAME,
2218 .id_table = id_table,
2219 .probe = storvsc_probe,
2220 .remove = storvsc_remove,
2221 .suspend = storvsc_suspend,
2222 .resume = storvsc_resume,
2224 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
2228 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2229 static struct fc_function_template fc_transport_functions = {
2230 .show_host_node_name = 1,
2231 .show_host_port_name = 1,
2235 static int __init storvsc_drv_init(void)
2240 * Divide the ring buffer data size (which is 1 page less
2241 * than the ring buffer size since that page is reserved for
2242 * the ring buffer indices) by the max request size (which is
2243 * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
2245 * The computation underestimates max_outstanding_req_per_channel
2246 * for Win7 and older hosts because it does not take into account
2247 * the vmscsi_size_delta correction to the max request size.
2249 max_outstanding_req_per_channel =
2250 ((storvsc_ringbuffer_size - PAGE_SIZE) /
2251 ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
2252 sizeof(struct vstor_packet) + sizeof(u64),
2255 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2256 fc_transport_template = fc_attach_transport(&fc_transport_functions);
2257 if (!fc_transport_template)
2261 ret = vmbus_driver_register(&storvsc_drv);
2263 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2265 fc_release_transport(fc_transport_template);
2271 static void __exit storvsc_drv_exit(void)
2273 vmbus_driver_unregister(&storvsc_drv);
2274 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2275 fc_release_transport(fc_transport_template);
2279 MODULE_LICENSE("GPL");
2280 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
2281 module_init(storvsc_drv_init);
2282 module_exit(storvsc_drv_exit);