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 <linux/dma-mapping.h>
26 #include <scsi/scsi.h>
27 #include <scsi/scsi_cmnd.h>
28 #include <scsi/scsi_host.h>
29 #include <scsi/scsi_device.h>
30 #include <scsi/scsi_tcq.h>
31 #include <scsi/scsi_eh.h>
32 #include <scsi/scsi_devinfo.h>
33 #include <scsi/scsi_dbg.h>
34 #include <scsi/scsi_transport_fc.h>
35 #include <scsi/scsi_transport.h>
38 * All wire protocol details (storage protocol between the guest and the host)
39 * are consolidated here.
41 * Begin protocol definitions.
47 * V1 RC < 2008/1/31: 1.0
48 * V1 RC > 2008/1/31: 2.0
55 #define VMSTOR_PROTO_VERSION(MAJOR_, MINOR_) ((((MAJOR_) & 0xff) << 8) | \
57 #define VMSTOR_PROTO_VERSION_WIN6 VMSTOR_PROTO_VERSION(2, 0)
58 #define VMSTOR_PROTO_VERSION_WIN7 VMSTOR_PROTO_VERSION(4, 2)
59 #define VMSTOR_PROTO_VERSION_WIN8 VMSTOR_PROTO_VERSION(5, 1)
60 #define VMSTOR_PROTO_VERSION_WIN8_1 VMSTOR_PROTO_VERSION(6, 0)
61 #define VMSTOR_PROTO_VERSION_WIN10 VMSTOR_PROTO_VERSION(6, 2)
63 /* channel callback timeout in ms */
64 #define CALLBACK_TIMEOUT 2
66 /* Packet structure describing virtual storage requests. */
67 enum vstor_packet_operation {
68 VSTOR_OPERATION_COMPLETE_IO = 1,
69 VSTOR_OPERATION_REMOVE_DEVICE = 2,
70 VSTOR_OPERATION_EXECUTE_SRB = 3,
71 VSTOR_OPERATION_RESET_LUN = 4,
72 VSTOR_OPERATION_RESET_ADAPTER = 5,
73 VSTOR_OPERATION_RESET_BUS = 6,
74 VSTOR_OPERATION_BEGIN_INITIALIZATION = 7,
75 VSTOR_OPERATION_END_INITIALIZATION = 8,
76 VSTOR_OPERATION_QUERY_PROTOCOL_VERSION = 9,
77 VSTOR_OPERATION_QUERY_PROPERTIES = 10,
78 VSTOR_OPERATION_ENUMERATE_BUS = 11,
79 VSTOR_OPERATION_FCHBA_DATA = 12,
80 VSTOR_OPERATION_CREATE_SUB_CHANNELS = 13,
81 VSTOR_OPERATION_MAXIMUM = 13
85 * WWN packet for Fibre Channel HBA
88 struct hv_fc_wwn_packet {
91 u8 primary_port_wwn[8];
92 u8 primary_node_wwn[8];
93 u8 secondary_port_wwn[8];
94 u8 secondary_node_wwn[8];
103 #define SRB_FLAGS_QUEUE_ACTION_ENABLE 0x00000002
104 #define SRB_FLAGS_DISABLE_DISCONNECT 0x00000004
105 #define SRB_FLAGS_DISABLE_SYNCH_TRANSFER 0x00000008
106 #define SRB_FLAGS_BYPASS_FROZEN_QUEUE 0x00000010
107 #define SRB_FLAGS_DISABLE_AUTOSENSE 0x00000020
108 #define SRB_FLAGS_DATA_IN 0x00000040
109 #define SRB_FLAGS_DATA_OUT 0x00000080
110 #define SRB_FLAGS_NO_DATA_TRANSFER 0x00000000
111 #define SRB_FLAGS_UNSPECIFIED_DIRECTION (SRB_FLAGS_DATA_IN | SRB_FLAGS_DATA_OUT)
112 #define SRB_FLAGS_NO_QUEUE_FREEZE 0x00000100
113 #define SRB_FLAGS_ADAPTER_CACHE_ENABLE 0x00000200
114 #define SRB_FLAGS_FREE_SENSE_BUFFER 0x00000400
117 * This flag indicates the request is part of the workflow for processing a D3.
119 #define SRB_FLAGS_D3_PROCESSING 0x00000800
120 #define SRB_FLAGS_IS_ACTIVE 0x00010000
121 #define SRB_FLAGS_ALLOCATED_FROM_ZONE 0x00020000
122 #define SRB_FLAGS_SGLIST_FROM_POOL 0x00040000
123 #define SRB_FLAGS_BYPASS_LOCKED_QUEUE 0x00080000
124 #define SRB_FLAGS_NO_KEEP_AWAKE 0x00100000
125 #define SRB_FLAGS_PORT_DRIVER_ALLOCSENSE 0x00200000
126 #define SRB_FLAGS_PORT_DRIVER_SENSEHASPORT 0x00400000
127 #define SRB_FLAGS_DONT_START_NEXT_PACKET 0x00800000
128 #define SRB_FLAGS_PORT_DRIVER_RESERVED 0x0F000000
129 #define SRB_FLAGS_CLASS_DRIVER_RESERVED 0xF0000000
131 #define SP_UNTAGGED ((unsigned char) ~0)
132 #define SRB_SIMPLE_TAG_REQUEST 0x20
135 * Platform neutral description of a scsi request -
136 * this remains the same across the write regardless of 32/64 bit
137 * note: it's patterned off the SCSI_PASS_THROUGH structure
139 #define STORVSC_MAX_CMD_LEN 0x10
141 /* Sense buffer size is the same for all versions since Windows 8 */
142 #define STORVSC_SENSE_BUFFER_SIZE 0x14
143 #define STORVSC_MAX_BUF_LEN_WITH_PADDING 0x14
146 * The storage protocol version is determined during the
147 * initial exchange with the host. It will indicate which
148 * storage functionality is available in the host.
150 static int vmstor_proto_version;
152 #define STORVSC_LOGGING_NONE 0
153 #define STORVSC_LOGGING_ERROR 1
154 #define STORVSC_LOGGING_WARN 2
156 static int logging_level = STORVSC_LOGGING_ERROR;
157 module_param(logging_level, int, S_IRUGO|S_IWUSR);
158 MODULE_PARM_DESC(logging_level,
159 "Logging level, 0 - None, 1 - Error (default), 2 - Warning.");
161 static inline bool do_logging(int level)
163 return logging_level >= level;
166 #define storvsc_log(dev, level, fmt, ...) \
168 if (do_logging(level)) \
169 dev_warn(&(dev)->device, fmt, ##__VA_ARGS__); \
172 struct vmscsi_request {
183 u8 sense_info_length;
187 u32 data_transfer_length;
190 u8 cdb[STORVSC_MAX_CMD_LEN];
191 u8 sense_data[STORVSC_SENSE_BUFFER_SIZE];
192 u8 reserved_array[STORVSC_MAX_BUF_LEN_WITH_PADDING];
195 * The following was added in win8.
204 } __attribute((packed));
207 * The list of windows version in order of preference.
210 static const int protocol_version[] = {
211 VMSTOR_PROTO_VERSION_WIN10,
212 VMSTOR_PROTO_VERSION_WIN8_1,
213 VMSTOR_PROTO_VERSION_WIN8,
218 * This structure is sent during the initialization phase to get the different
219 * properties of the channel.
222 #define STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL 0x1
224 struct vmstorage_channel_properties {
230 u32 max_transfer_bytes;
235 /* This structure is sent during the storage protocol negotiations. */
236 struct vmstorage_protocol_version {
237 /* Major (MSW) and minor (LSW) version numbers. */
241 * Revision number is auto-incremented whenever this file is changed
242 * (See FILL_VMSTOR_REVISION macro above). Mismatch does not
243 * definitely indicate incompatibility--but it does indicate mismatched
245 * This is only used on the windows side. Just set it to 0.
250 /* Channel Property Flags */
251 #define STORAGE_CHANNEL_REMOVABLE_FLAG 0x1
252 #define STORAGE_CHANNEL_EMULATED_IDE_FLAG 0x2
254 struct vstor_packet {
255 /* Requested operation type */
256 enum vstor_packet_operation operation;
258 /* Flags - see below for values */
261 /* Status of the request returned from the server side. */
264 /* Data payload area */
267 * Structure used to forward SCSI commands from the
268 * client to the server.
270 struct vmscsi_request vm_srb;
272 /* Structure used to query channel properties. */
273 struct vmstorage_channel_properties storage_channel_properties;
275 /* Used during version negotiations. */
276 struct vmstorage_protocol_version version;
278 /* Fibre channel address packet */
279 struct hv_fc_wwn_packet wwn_packet;
281 /* Number of sub-channels to create */
282 u16 sub_channel_count;
284 /* This will be the maximum of the union members */
292 * This flag indicates that the server should send back a completion for this
296 #define REQUEST_COMPLETION_FLAG 0x1
298 /* Matches Windows-end */
299 enum storvsc_request_type {
306 * SRB status codes and masks. In the 8-bit field, the two high order bits
307 * are flags, while the remaining 6 bits are an integer status code. The
308 * definitions here include only the subset of the integer status codes that
309 * are tested for in this driver.
311 #define SRB_STATUS_AUTOSENSE_VALID 0x80
312 #define SRB_STATUS_QUEUE_FROZEN 0x40
314 /* SRB status integer codes */
315 #define SRB_STATUS_SUCCESS 0x01
316 #define SRB_STATUS_ABORTED 0x02
317 #define SRB_STATUS_ERROR 0x04
318 #define SRB_STATUS_INVALID_REQUEST 0x06
319 #define SRB_STATUS_DATA_OVERRUN 0x12
320 #define SRB_STATUS_INVALID_LUN 0x20
321 #define SRB_STATUS_INTERNAL_ERROR 0x30
323 #define SRB_STATUS(status) \
324 (status & ~(SRB_STATUS_AUTOSENSE_VALID | SRB_STATUS_QUEUE_FROZEN))
326 * This is the end of Protocol specific defines.
329 static int storvsc_ringbuffer_size = (128 * 1024);
330 static u32 max_outstanding_req_per_channel;
331 static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth);
333 static int storvsc_vcpus_per_sub_channel = 4;
334 static unsigned int storvsc_max_hw_queues;
336 module_param(storvsc_ringbuffer_size, int, S_IRUGO);
337 MODULE_PARM_DESC(storvsc_ringbuffer_size, "Ring buffer size (bytes)");
339 module_param(storvsc_max_hw_queues, uint, 0644);
340 MODULE_PARM_DESC(storvsc_max_hw_queues, "Maximum number of hardware queues");
342 module_param(storvsc_vcpus_per_sub_channel, int, S_IRUGO);
343 MODULE_PARM_DESC(storvsc_vcpus_per_sub_channel, "Ratio of VCPUs to subchannels");
345 static int ring_avail_percent_lowater = 10;
346 module_param(ring_avail_percent_lowater, int, S_IRUGO);
347 MODULE_PARM_DESC(ring_avail_percent_lowater,
348 "Select a channel if available ring size > this in percent");
351 * Timeout in seconds for all devices managed by this driver.
353 static int storvsc_timeout = 180;
355 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
356 static struct scsi_transport_template *fc_transport_template;
359 static struct scsi_host_template scsi_driver;
360 static void storvsc_on_channel_callback(void *context);
362 #define STORVSC_MAX_LUNS_PER_TARGET 255
363 #define STORVSC_MAX_TARGETS 2
364 #define STORVSC_MAX_CHANNELS 8
366 #define STORVSC_FC_MAX_LUNS_PER_TARGET 255
367 #define STORVSC_FC_MAX_TARGETS 128
368 #define STORVSC_FC_MAX_CHANNELS 8
369 #define STORVSC_FC_MAX_XFER_SIZE ((u32)(512 * 1024))
371 #define STORVSC_IDE_MAX_LUNS_PER_TARGET 64
372 #define STORVSC_IDE_MAX_TARGETS 1
373 #define STORVSC_IDE_MAX_CHANNELS 1
376 * Upper bound on the size of a storvsc packet.
378 #define STORVSC_MAX_PKT_SIZE (sizeof(struct vmpacket_descriptor) +\
379 sizeof(struct vstor_packet))
381 struct storvsc_cmd_request {
382 struct scsi_cmnd *cmd;
384 struct hv_device *device;
386 /* Synchronize the request/response if needed */
387 struct completion wait_event;
389 struct vmbus_channel_packet_multipage_buffer mpb;
390 struct vmbus_packet_mpb_array *payload;
393 struct vstor_packet vstor_packet;
397 /* A storvsc device is a device object that contains a vmbus channel */
398 struct storvsc_device {
399 struct hv_device *device;
403 atomic_t num_outstanding_req;
404 struct Scsi_Host *host;
406 wait_queue_head_t waiting_to_drain;
409 * Each unique Port/Path/Target represents 1 channel ie scsi
410 * controller. In reality, the pathid, targetid is always 0
411 * and the port is set by us
413 unsigned int port_number;
414 unsigned char path_id;
415 unsigned char target_id;
418 * Max I/O, the device can support.
420 u32 max_transfer_bytes;
422 * Number of sub-channels we will open.
425 struct vmbus_channel **stor_chns;
427 * Mask of CPUs bound to subchannels.
429 struct cpumask alloced_cpus;
431 * Serializes modifications of stor_chns[] from storvsc_do_io()
432 * and storvsc_change_target_cpu().
435 /* Used for vsc/vsp channel reset process */
436 struct storvsc_cmd_request init_request;
437 struct storvsc_cmd_request reset_request;
439 * Currently active port and node names for FC devices.
443 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
444 struct fc_rport *rport;
448 struct hv_host_device {
449 struct hv_device *dev;
452 unsigned char target;
453 struct workqueue_struct *handle_error_wq;
454 struct work_struct host_scan_work;
455 struct Scsi_Host *host;
458 struct storvsc_scan_work {
459 struct work_struct work;
460 struct Scsi_Host *host;
465 static void storvsc_device_scan(struct work_struct *work)
467 struct storvsc_scan_work *wrk;
468 struct scsi_device *sdev;
470 wrk = container_of(work, struct storvsc_scan_work, work);
472 sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
475 scsi_rescan_device(&sdev->sdev_gendev);
476 scsi_device_put(sdev);
482 static void storvsc_host_scan(struct work_struct *work)
484 struct Scsi_Host *host;
485 struct scsi_device *sdev;
486 struct hv_host_device *host_device =
487 container_of(work, struct hv_host_device, host_scan_work);
489 host = host_device->host;
491 * Before scanning the host, first check to see if any of the
492 * currently known devices have been hot removed. We issue a
493 * "unit ready" command against all currently known devices.
494 * This I/O will result in an error for devices that have been
495 * removed. As part of handling the I/O error, we remove the device.
497 * When a LUN is added or removed, the host sends us a signal to
498 * scan the host. Thus we are forced to discover the LUNs that
499 * may have been removed this way.
501 mutex_lock(&host->scan_mutex);
502 shost_for_each_device(sdev, host)
503 scsi_test_unit_ready(sdev, 1, 1, NULL);
504 mutex_unlock(&host->scan_mutex);
506 * Now scan the host to discover LUNs that may have been added.
508 scsi_scan_host(host);
511 static void storvsc_remove_lun(struct work_struct *work)
513 struct storvsc_scan_work *wrk;
514 struct scsi_device *sdev;
516 wrk = container_of(work, struct storvsc_scan_work, work);
517 if (!scsi_host_get(wrk->host))
520 sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
523 scsi_remove_device(sdev);
524 scsi_device_put(sdev);
526 scsi_host_put(wrk->host);
534 * We can get incoming messages from the host that are not in response to
535 * messages that we have sent out. An example of this would be messages
536 * received by the guest to notify dynamic addition/removal of LUNs. To
537 * deal with potential race conditions where the driver may be in the
538 * midst of being unloaded when we might receive an unsolicited message
539 * from the host, we have implemented a mechanism to gurantee sequential
542 * 1) Once the device is marked as being destroyed, we will fail all
544 * 2) We permit incoming messages when the device is being destroyed,
545 * only to properly account for messages already sent out.
548 static inline struct storvsc_device *get_out_stor_device(
549 struct hv_device *device)
551 struct storvsc_device *stor_device;
553 stor_device = hv_get_drvdata(device);
555 if (stor_device && stor_device->destroy)
562 static inline void storvsc_wait_to_drain(struct storvsc_device *dev)
564 dev->drain_notify = true;
565 wait_event(dev->waiting_to_drain,
566 atomic_read(&dev->num_outstanding_req) == 0);
567 dev->drain_notify = false;
570 static inline struct storvsc_device *get_in_stor_device(
571 struct hv_device *device)
573 struct storvsc_device *stor_device;
575 stor_device = hv_get_drvdata(device);
581 * If the device is being destroyed; allow incoming
582 * traffic only to cleanup outstanding requests.
585 if (stor_device->destroy &&
586 (atomic_read(&stor_device->num_outstanding_req) == 0))
594 static void storvsc_change_target_cpu(struct vmbus_channel *channel, u32 old,
597 struct storvsc_device *stor_device;
598 struct vmbus_channel *cur_chn;
599 bool old_is_alloced = false;
600 struct hv_device *device;
604 device = channel->primary_channel ?
605 channel->primary_channel->device_obj
606 : channel->device_obj;
607 stor_device = get_out_stor_device(device);
611 /* See storvsc_do_io() -> get_og_chn(). */
612 spin_lock_irqsave(&stor_device->lock, flags);
615 * Determines if the storvsc device has other channels assigned to
616 * the "old" CPU to update the alloced_cpus mask and the stor_chns
619 if (device->channel != channel && device->channel->target_cpu == old) {
620 cur_chn = device->channel;
621 old_is_alloced = true;
624 list_for_each_entry(cur_chn, &device->channel->sc_list, sc_list) {
625 if (cur_chn == channel)
627 if (cur_chn->target_cpu == old) {
628 old_is_alloced = true;
635 WRITE_ONCE(stor_device->stor_chns[old], cur_chn);
637 cpumask_clear_cpu(old, &stor_device->alloced_cpus);
639 /* "Flush" the stor_chns array. */
640 for_each_possible_cpu(cpu) {
641 if (stor_device->stor_chns[cpu] && !cpumask_test_cpu(
642 cpu, &stor_device->alloced_cpus))
643 WRITE_ONCE(stor_device->stor_chns[cpu], NULL);
646 WRITE_ONCE(stor_device->stor_chns[new], channel);
647 cpumask_set_cpu(new, &stor_device->alloced_cpus);
649 spin_unlock_irqrestore(&stor_device->lock, flags);
652 static u64 storvsc_next_request_id(struct vmbus_channel *channel, u64 rqst_addr)
654 struct storvsc_cmd_request *request =
655 (struct storvsc_cmd_request *)(unsigned long)rqst_addr;
657 if (rqst_addr == VMBUS_RQST_INIT)
658 return VMBUS_RQST_INIT;
659 if (rqst_addr == VMBUS_RQST_RESET)
660 return VMBUS_RQST_RESET;
663 * Cannot return an ID of 0, which is reserved for an unsolicited
664 * message from Hyper-V.
666 return (u64)blk_mq_unique_tag(scsi_cmd_to_rq(request->cmd)) + 1;
669 static void handle_sc_creation(struct vmbus_channel *new_sc)
671 struct hv_device *device = new_sc->primary_channel->device_obj;
672 struct device *dev = &device->device;
673 struct storvsc_device *stor_device;
674 struct vmstorage_channel_properties props;
677 stor_device = get_out_stor_device(device);
681 memset(&props, 0, sizeof(struct vmstorage_channel_properties));
682 new_sc->max_pkt_size = STORVSC_MAX_PKT_SIZE;
684 new_sc->next_request_id_callback = storvsc_next_request_id;
686 ret = vmbus_open(new_sc,
687 storvsc_ringbuffer_size,
688 storvsc_ringbuffer_size,
690 sizeof(struct vmstorage_channel_properties),
691 storvsc_on_channel_callback, new_sc);
693 /* In case vmbus_open() fails, we don't use the sub-channel. */
695 dev_err(dev, "Failed to open sub-channel: err=%d\n", ret);
699 new_sc->change_target_cpu_callback = storvsc_change_target_cpu;
701 /* Add the sub-channel to the array of available channels. */
702 stor_device->stor_chns[new_sc->target_cpu] = new_sc;
703 cpumask_set_cpu(new_sc->target_cpu, &stor_device->alloced_cpus);
706 static void handle_multichannel_storage(struct hv_device *device, int max_chns)
708 struct device *dev = &device->device;
709 struct storvsc_device *stor_device;
711 struct storvsc_cmd_request *request;
712 struct vstor_packet *vstor_packet;
716 * If the number of CPUs is artificially restricted, such as
717 * with maxcpus=1 on the kernel boot line, Hyper-V could offer
718 * sub-channels >= the number of CPUs. These sub-channels
719 * should not be created. The primary channel is already created
720 * and assigned to one CPU, so check against # CPUs - 1.
722 num_sc = min((int)(num_online_cpus() - 1), max_chns);
726 stor_device = get_out_stor_device(device);
730 stor_device->num_sc = num_sc;
731 request = &stor_device->init_request;
732 vstor_packet = &request->vstor_packet;
735 * Establish a handler for dealing with subchannels.
737 vmbus_set_sc_create_callback(device->channel, handle_sc_creation);
740 * Request the host to create sub-channels.
742 memset(request, 0, sizeof(struct storvsc_cmd_request));
743 init_completion(&request->wait_event);
744 vstor_packet->operation = VSTOR_OPERATION_CREATE_SUB_CHANNELS;
745 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
746 vstor_packet->sub_channel_count = num_sc;
748 ret = vmbus_sendpacket(device->channel, vstor_packet,
749 sizeof(struct vstor_packet),
752 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
755 dev_err(dev, "Failed to create sub-channel: err=%d\n", ret);
759 t = wait_for_completion_timeout(&request->wait_event, 10*HZ);
761 dev_err(dev, "Failed to create sub-channel: timed out\n");
765 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
766 vstor_packet->status != 0) {
767 dev_err(dev, "Failed to create sub-channel: op=%d, sts=%d\n",
768 vstor_packet->operation, vstor_packet->status);
773 * We need to do nothing here, because vmbus_process_offer()
774 * invokes channel->sc_creation_callback, which will open and use
775 * the sub-channel(s).
779 static void cache_wwn(struct storvsc_device *stor_device,
780 struct vstor_packet *vstor_packet)
783 * Cache the currently active port and node ww names.
785 if (vstor_packet->wwn_packet.primary_active) {
786 stor_device->node_name =
787 wwn_to_u64(vstor_packet->wwn_packet.primary_node_wwn);
788 stor_device->port_name =
789 wwn_to_u64(vstor_packet->wwn_packet.primary_port_wwn);
791 stor_device->node_name =
792 wwn_to_u64(vstor_packet->wwn_packet.secondary_node_wwn);
793 stor_device->port_name =
794 wwn_to_u64(vstor_packet->wwn_packet.secondary_port_wwn);
799 static int storvsc_execute_vstor_op(struct hv_device *device,
800 struct storvsc_cmd_request *request,
803 struct storvsc_device *stor_device;
804 struct vstor_packet *vstor_packet;
807 stor_device = get_out_stor_device(device);
811 vstor_packet = &request->vstor_packet;
813 init_completion(&request->wait_event);
814 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
816 ret = vmbus_sendpacket(device->channel, vstor_packet,
817 sizeof(struct vstor_packet),
820 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
824 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
831 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
832 vstor_packet->status != 0)
838 static int storvsc_channel_init(struct hv_device *device, bool is_fc)
840 struct storvsc_device *stor_device;
841 struct storvsc_cmd_request *request;
842 struct vstor_packet *vstor_packet;
845 bool process_sub_channels = false;
847 stor_device = get_out_stor_device(device);
851 request = &stor_device->init_request;
852 vstor_packet = &request->vstor_packet;
855 * Now, initiate the vsc/vsp initialization protocol on the open
858 memset(request, 0, sizeof(struct storvsc_cmd_request));
859 vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION;
860 ret = storvsc_execute_vstor_op(device, request, true);
864 * Query host supported protocol version.
867 for (i = 0; i < ARRAY_SIZE(protocol_version); i++) {
868 /* reuse the packet for version range supported */
869 memset(vstor_packet, 0, sizeof(struct vstor_packet));
870 vstor_packet->operation =
871 VSTOR_OPERATION_QUERY_PROTOCOL_VERSION;
873 vstor_packet->version.major_minor = protocol_version[i];
876 * The revision number is only used in Windows; set it to 0.
878 vstor_packet->version.revision = 0;
879 ret = storvsc_execute_vstor_op(device, request, false);
883 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO)
886 if (vstor_packet->status == 0) {
887 vmstor_proto_version = protocol_version[i];
893 if (vstor_packet->status != 0) {
894 dev_err(&device->device, "Obsolete Hyper-V version\n");
899 memset(vstor_packet, 0, sizeof(struct vstor_packet));
900 vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES;
901 ret = storvsc_execute_vstor_op(device, request, true);
906 * Check to see if multi-channel support is there.
907 * Hosts that implement protocol version of 5.1 and above
908 * support multi-channel.
910 max_chns = vstor_packet->storage_channel_properties.max_channel_cnt;
913 * Allocate state to manage the sub-channels.
914 * We allocate an array based on the numbers of possible CPUs
915 * (Hyper-V does not support cpu online/offline).
916 * This Array will be sparseley populated with unique
917 * channels - primary + sub-channels.
918 * We will however populate all the slots to evenly distribute
921 stor_device->stor_chns = kcalloc(num_possible_cpus(), sizeof(void *),
923 if (stor_device->stor_chns == NULL)
926 device->channel->change_target_cpu_callback = storvsc_change_target_cpu;
928 stor_device->stor_chns[device->channel->target_cpu] = device->channel;
929 cpumask_set_cpu(device->channel->target_cpu,
930 &stor_device->alloced_cpus);
932 if (vstor_packet->storage_channel_properties.flags &
933 STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL)
934 process_sub_channels = true;
936 stor_device->max_transfer_bytes =
937 vstor_packet->storage_channel_properties.max_transfer_bytes;
943 * For FC devices retrieve FC HBA data.
945 memset(vstor_packet, 0, sizeof(struct vstor_packet));
946 vstor_packet->operation = VSTOR_OPERATION_FCHBA_DATA;
947 ret = storvsc_execute_vstor_op(device, request, true);
952 * Cache the currently active port and node ww names.
954 cache_wwn(stor_device, vstor_packet);
958 memset(vstor_packet, 0, sizeof(struct vstor_packet));
959 vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION;
960 ret = storvsc_execute_vstor_op(device, request, true);
964 if (process_sub_channels)
965 handle_multichannel_storage(device, max_chns);
970 static void storvsc_handle_error(struct vmscsi_request *vm_srb,
971 struct scsi_cmnd *scmnd,
972 struct Scsi_Host *host,
975 struct storvsc_scan_work *wrk;
976 void (*process_err_fn)(struct work_struct *work);
977 struct hv_host_device *host_dev = shost_priv(host);
979 switch (SRB_STATUS(vm_srb->srb_status)) {
980 case SRB_STATUS_ERROR:
981 case SRB_STATUS_ABORTED:
982 case SRB_STATUS_INVALID_REQUEST:
983 case SRB_STATUS_INTERNAL_ERROR:
984 if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID) {
985 /* Check for capacity change */
986 if ((asc == 0x2a) && (ascq == 0x9)) {
987 process_err_fn = storvsc_device_scan;
988 /* Retry the I/O that triggered this. */
989 set_host_byte(scmnd, DID_REQUEUE);
994 * Check for "Operating parameters have changed"
995 * due to Hyper-V changing the VHD/VHDX BlockSize
996 * when adding/removing a differencing disk. This
997 * causes discard_granularity to change, so do a
998 * rescan to pick up the new granularity. We don't
999 * want scsi_report_sense() to output a message
1000 * that a sysadmin wouldn't know what to do with.
1002 if ((asc == 0x3f) && (ascq != 0x03) &&
1004 process_err_fn = storvsc_device_scan;
1005 set_host_byte(scmnd, DID_REQUEUE);
1010 * Otherwise, let upper layer deal with the
1011 * error when sense message is present
1017 * If there is an error; offline the device since all
1018 * error recovery strategies would have already been
1019 * deployed on the host side. However, if the command
1020 * were a pass-through command deal with it appropriately.
1022 switch (scmnd->cmnd[0]) {
1025 set_host_byte(scmnd, DID_PASSTHROUGH);
1028 * On some Hyper-V hosts TEST_UNIT_READY command can
1029 * return SRB_STATUS_ERROR. Let the upper level code
1030 * deal with it based on the sense information.
1032 case TEST_UNIT_READY:
1035 set_host_byte(scmnd, DID_ERROR);
1039 case SRB_STATUS_INVALID_LUN:
1040 set_host_byte(scmnd, DID_NO_CONNECT);
1041 process_err_fn = storvsc_remove_lun;
1049 * We need to schedule work to process this error; schedule it.
1051 wrk = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC);
1053 set_host_byte(scmnd, DID_BAD_TARGET);
1058 wrk->lun = vm_srb->lun;
1059 wrk->tgt_id = vm_srb->target_id;
1060 INIT_WORK(&wrk->work, process_err_fn);
1061 queue_work(host_dev->handle_error_wq, &wrk->work);
1065 static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request,
1066 struct storvsc_device *stor_dev)
1068 struct scsi_cmnd *scmnd = cmd_request->cmd;
1069 struct scsi_sense_hdr sense_hdr;
1070 struct vmscsi_request *vm_srb;
1071 u32 data_transfer_length;
1072 struct Scsi_Host *host;
1073 u32 payload_sz = cmd_request->payload_sz;
1074 void *payload = cmd_request->payload;
1077 host = stor_dev->host;
1079 vm_srb = &cmd_request->vstor_packet.vm_srb;
1080 data_transfer_length = vm_srb->data_transfer_length;
1082 scmnd->result = vm_srb->scsi_status;
1084 if (scmnd->result) {
1085 sense_ok = scsi_normalize_sense(scmnd->sense_buffer,
1086 SCSI_SENSE_BUFFERSIZE, &sense_hdr);
1088 if (sense_ok && do_logging(STORVSC_LOGGING_WARN))
1089 scsi_print_sense_hdr(scmnd->device, "storvsc",
1093 if (vm_srb->srb_status != SRB_STATUS_SUCCESS) {
1094 storvsc_handle_error(vm_srb, scmnd, host, sense_hdr.asc,
1097 * The Windows driver set data_transfer_length on
1098 * SRB_STATUS_DATA_OVERRUN. On other errors, this value
1099 * is untouched. In these cases we set it to 0.
1101 if (vm_srb->srb_status != SRB_STATUS_DATA_OVERRUN)
1102 data_transfer_length = 0;
1105 /* Validate data_transfer_length (from Hyper-V) */
1106 if (data_transfer_length > cmd_request->payload->range.len)
1107 data_transfer_length = cmd_request->payload->range.len;
1109 scsi_set_resid(scmnd,
1110 cmd_request->payload->range.len - data_transfer_length);
1115 sizeof(struct vmbus_channel_packet_multipage_buffer))
1119 static void storvsc_on_io_completion(struct storvsc_device *stor_device,
1120 struct vstor_packet *vstor_packet,
1121 struct storvsc_cmd_request *request)
1123 struct vstor_packet *stor_pkt;
1124 struct hv_device *device = stor_device->device;
1126 stor_pkt = &request->vstor_packet;
1129 * The current SCSI handling on the host side does
1130 * not correctly handle:
1131 * INQUIRY command with page code parameter set to 0x80
1132 * MODE_SENSE command with cmd[2] == 0x1c
1134 * Setup srb and scsi status so this won't be fatal.
1135 * We do this so we can distinguish truly fatal failues
1136 * (srb status == 0x4) and off-line the device in that case.
1139 if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) ||
1140 (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) {
1141 vstor_packet->vm_srb.scsi_status = 0;
1142 vstor_packet->vm_srb.srb_status = SRB_STATUS_SUCCESS;
1145 /* Copy over the status...etc */
1146 stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status;
1147 stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status;
1150 * Copy over the sense_info_length, but limit to the known max
1151 * size if Hyper-V returns a bad value.
1153 stor_pkt->vm_srb.sense_info_length = min_t(u8, STORVSC_SENSE_BUFFER_SIZE,
1154 vstor_packet->vm_srb.sense_info_length);
1156 if (vstor_packet->vm_srb.scsi_status != 0 ||
1157 vstor_packet->vm_srb.srb_status != SRB_STATUS_SUCCESS) {
1160 * Log TEST_UNIT_READY errors only as warnings. Hyper-V can
1161 * return errors when detecting devices using TEST_UNIT_READY,
1162 * and logging these as errors produces unhelpful noise.
1164 int loglevel = (stor_pkt->vm_srb.cdb[0] == TEST_UNIT_READY) ?
1165 STORVSC_LOGGING_WARN : STORVSC_LOGGING_ERROR;
1167 storvsc_log(device, loglevel,
1168 "tag#%d cmd 0x%x status: scsi 0x%x srb 0x%x hv 0x%x\n",
1169 scsi_cmd_to_rq(request->cmd)->tag,
1170 stor_pkt->vm_srb.cdb[0],
1171 vstor_packet->vm_srb.scsi_status,
1172 vstor_packet->vm_srb.srb_status,
1173 vstor_packet->status);
1176 if (vstor_packet->vm_srb.scsi_status == SAM_STAT_CHECK_CONDITION &&
1177 (vstor_packet->vm_srb.srb_status & SRB_STATUS_AUTOSENSE_VALID))
1178 memcpy(request->cmd->sense_buffer,
1179 vstor_packet->vm_srb.sense_data,
1180 stor_pkt->vm_srb.sense_info_length);
1182 stor_pkt->vm_srb.data_transfer_length =
1183 vstor_packet->vm_srb.data_transfer_length;
1185 storvsc_command_completion(request, stor_device);
1187 if (atomic_dec_and_test(&stor_device->num_outstanding_req) &&
1188 stor_device->drain_notify)
1189 wake_up(&stor_device->waiting_to_drain);
1192 static void storvsc_on_receive(struct storvsc_device *stor_device,
1193 struct vstor_packet *vstor_packet,
1194 struct storvsc_cmd_request *request)
1196 struct hv_host_device *host_dev;
1197 switch (vstor_packet->operation) {
1198 case VSTOR_OPERATION_COMPLETE_IO:
1199 storvsc_on_io_completion(stor_device, vstor_packet, request);
1202 case VSTOR_OPERATION_REMOVE_DEVICE:
1203 case VSTOR_OPERATION_ENUMERATE_BUS:
1204 host_dev = shost_priv(stor_device->host);
1206 host_dev->handle_error_wq, &host_dev->host_scan_work);
1209 case VSTOR_OPERATION_FCHBA_DATA:
1210 cache_wwn(stor_device, vstor_packet);
1211 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1212 fc_host_node_name(stor_device->host) = stor_device->node_name;
1213 fc_host_port_name(stor_device->host) = stor_device->port_name;
1221 static void storvsc_on_channel_callback(void *context)
1223 struct vmbus_channel *channel = (struct vmbus_channel *)context;
1224 const struct vmpacket_descriptor *desc;
1225 struct hv_device *device;
1226 struct storvsc_device *stor_device;
1227 struct Scsi_Host *shost;
1228 unsigned long time_limit = jiffies + msecs_to_jiffies(CALLBACK_TIMEOUT);
1230 if (channel->primary_channel != NULL)
1231 device = channel->primary_channel->device_obj;
1233 device = channel->device_obj;
1235 stor_device = get_in_stor_device(device);
1239 shost = stor_device->host;
1241 foreach_vmbus_pkt(desc, channel) {
1242 struct vstor_packet *packet = hv_pkt_data(desc);
1243 struct storvsc_cmd_request *request = NULL;
1244 u32 pktlen = hv_pkt_datalen(desc);
1245 u64 rqst_id = desc->trans_id;
1246 u32 minlen = rqst_id ? sizeof(struct vstor_packet) :
1247 sizeof(enum vstor_packet_operation);
1249 if (unlikely(time_after(jiffies, time_limit))) {
1250 hv_pkt_iter_close(channel);
1254 if (pktlen < minlen) {
1255 dev_err(&device->device,
1256 "Invalid pkt: id=%llu, len=%u, minlen=%u\n",
1257 rqst_id, pktlen, minlen);
1261 if (rqst_id == VMBUS_RQST_INIT) {
1262 request = &stor_device->init_request;
1263 } else if (rqst_id == VMBUS_RQST_RESET) {
1264 request = &stor_device->reset_request;
1266 /* Hyper-V can send an unsolicited message with ID of 0 */
1269 * storvsc_on_receive() looks at the vstor_packet in the message
1270 * from the ring buffer.
1272 * - If the operation in the vstor_packet is COMPLETE_IO, then
1273 * we call storvsc_on_io_completion(), and dereference the
1274 * guest memory address. Make sure we don't call
1275 * storvsc_on_io_completion() with a guest memory address
1276 * that is zero if Hyper-V were to construct and send such
1279 * - If the operation in the vstor_packet is FCHBA_DATA, then
1280 * we call cache_wwn(), and access the data payload area of
1281 * the packet (wwn_packet); however, there is no guarantee
1282 * that the packet is big enough to contain such area.
1283 * Future-proof the code by rejecting such a bogus packet.
1285 if (packet->operation == VSTOR_OPERATION_COMPLETE_IO ||
1286 packet->operation == VSTOR_OPERATION_FCHBA_DATA) {
1287 dev_err(&device->device, "Invalid packet with ID of 0\n");
1291 struct scsi_cmnd *scmnd;
1293 /* Transaction 'rqst_id' corresponds to tag 'rqst_id - 1' */
1294 scmnd = scsi_host_find_tag(shost, rqst_id - 1);
1295 if (scmnd == NULL) {
1296 dev_err(&device->device, "Incorrect transaction ID\n");
1299 request = (struct storvsc_cmd_request *)scsi_cmd_priv(scmnd);
1300 scsi_dma_unmap(scmnd);
1303 storvsc_on_receive(stor_device, packet, request);
1307 memcpy(&request->vstor_packet, packet,
1308 sizeof(struct vstor_packet));
1309 complete(&request->wait_event);
1313 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size,
1316 struct vmstorage_channel_properties props;
1319 memset(&props, 0, sizeof(struct vmstorage_channel_properties));
1321 device->channel->max_pkt_size = STORVSC_MAX_PKT_SIZE;
1322 device->channel->next_request_id_callback = storvsc_next_request_id;
1324 ret = vmbus_open(device->channel,
1328 sizeof(struct vmstorage_channel_properties),
1329 storvsc_on_channel_callback, device->channel);
1334 ret = storvsc_channel_init(device, is_fc);
1339 static int storvsc_dev_remove(struct hv_device *device)
1341 struct storvsc_device *stor_device;
1343 stor_device = hv_get_drvdata(device);
1345 stor_device->destroy = true;
1347 /* Make sure flag is set before waiting */
1351 * At this point, all outbound traffic should be disable. We
1352 * only allow inbound traffic (responses) to proceed so that
1353 * outstanding requests can be completed.
1356 storvsc_wait_to_drain(stor_device);
1359 * Since we have already drained, we don't need to busy wait
1360 * as was done in final_release_stor_device()
1361 * Note that we cannot set the ext pointer to NULL until
1362 * we have drained - to drain the outgoing packets, we need to
1363 * allow incoming packets.
1365 hv_set_drvdata(device, NULL);
1367 /* Close the channel */
1368 vmbus_close(device->channel);
1370 kfree(stor_device->stor_chns);
1375 static struct vmbus_channel *get_og_chn(struct storvsc_device *stor_device,
1380 const struct cpumask *node_mask;
1381 int num_channels, tgt_cpu;
1383 if (stor_device->num_sc == 0) {
1384 stor_device->stor_chns[q_num] = stor_device->device->channel;
1385 return stor_device->device->channel;
1389 * Our channel array is sparsley populated and we
1390 * initiated I/O on a processor/hw-q that does not
1391 * currently have a designated channel. Fix this.
1392 * The strategy is simple:
1393 * I. Ensure NUMA locality
1394 * II. Distribute evenly (best effort)
1397 node_mask = cpumask_of_node(cpu_to_node(q_num));
1400 for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1401 if (cpumask_test_cpu(tgt_cpu, node_mask))
1404 if (num_channels == 0) {
1405 stor_device->stor_chns[q_num] = stor_device->device->channel;
1406 return stor_device->device->channel;
1410 while (hash_qnum >= num_channels)
1411 hash_qnum -= num_channels;
1413 for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1414 if (!cpumask_test_cpu(tgt_cpu, node_mask))
1416 if (slot == hash_qnum)
1421 stor_device->stor_chns[q_num] = stor_device->stor_chns[tgt_cpu];
1423 return stor_device->stor_chns[q_num];
1427 static int storvsc_do_io(struct hv_device *device,
1428 struct storvsc_cmd_request *request, u16 q_num)
1430 struct storvsc_device *stor_device;
1431 struct vstor_packet *vstor_packet;
1432 struct vmbus_channel *outgoing_channel, *channel;
1433 unsigned long flags;
1435 const struct cpumask *node_mask;
1438 vstor_packet = &request->vstor_packet;
1439 stor_device = get_out_stor_device(device);
1445 request->device = device;
1447 * Select an appropriate channel to send the request out.
1449 /* See storvsc_change_target_cpu(). */
1450 outgoing_channel = READ_ONCE(stor_device->stor_chns[q_num]);
1451 if (outgoing_channel != NULL) {
1452 if (outgoing_channel->target_cpu == q_num) {
1454 * Ideally, we want to pick a different channel if
1455 * available on the same NUMA node.
1457 node_mask = cpumask_of_node(cpu_to_node(q_num));
1458 for_each_cpu_wrap(tgt_cpu,
1459 &stor_device->alloced_cpus, q_num + 1) {
1460 if (!cpumask_test_cpu(tgt_cpu, node_mask))
1462 if (tgt_cpu == q_num)
1464 channel = READ_ONCE(
1465 stor_device->stor_chns[tgt_cpu]);
1466 if (channel == NULL)
1468 if (hv_get_avail_to_write_percent(
1470 > ring_avail_percent_lowater) {
1471 outgoing_channel = channel;
1477 * All the other channels on the same NUMA node are
1478 * busy. Try to use the channel on the current CPU
1480 if (hv_get_avail_to_write_percent(
1481 &outgoing_channel->outbound)
1482 > ring_avail_percent_lowater)
1486 * If we reach here, all the channels on the current
1487 * NUMA node are busy. Try to find a channel in
1490 for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1491 if (cpumask_test_cpu(tgt_cpu, node_mask))
1493 channel = READ_ONCE(
1494 stor_device->stor_chns[tgt_cpu]);
1495 if (channel == NULL)
1497 if (hv_get_avail_to_write_percent(
1499 > ring_avail_percent_lowater) {
1500 outgoing_channel = channel;
1506 spin_lock_irqsave(&stor_device->lock, flags);
1507 outgoing_channel = stor_device->stor_chns[q_num];
1508 if (outgoing_channel != NULL) {
1509 spin_unlock_irqrestore(&stor_device->lock, flags);
1512 outgoing_channel = get_og_chn(stor_device, q_num);
1513 spin_unlock_irqrestore(&stor_device->lock, flags);
1517 vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
1519 vstor_packet->vm_srb.length = sizeof(struct vmscsi_request);
1522 vstor_packet->vm_srb.sense_info_length = STORVSC_SENSE_BUFFER_SIZE;
1525 vstor_packet->vm_srb.data_transfer_length =
1526 request->payload->range.len;
1528 vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
1530 if (request->payload->range.len) {
1532 ret = vmbus_sendpacket_mpb_desc(outgoing_channel,
1533 request->payload, request->payload_sz,
1535 sizeof(struct vstor_packet),
1536 (unsigned long)request);
1538 ret = vmbus_sendpacket(outgoing_channel, vstor_packet,
1539 sizeof(struct vstor_packet),
1540 (unsigned long)request,
1542 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1548 atomic_inc(&stor_device->num_outstanding_req);
1553 static int storvsc_device_alloc(struct scsi_device *sdevice)
1556 * Set blist flag to permit the reading of the VPD pages even when
1557 * the target may claim SPC-2 compliance. MSFT targets currently
1558 * claim SPC-2 compliance while they implement post SPC-2 features.
1559 * With this flag we can correctly handle WRITE_SAME_16 issues.
1561 * Hypervisor reports SCSI_UNKNOWN type for DVD ROM device but
1562 * still supports REPORT LUN.
1564 sdevice->sdev_bflags = BLIST_REPORTLUN2 | BLIST_TRY_VPD_PAGES;
1569 static int storvsc_device_configure(struct scsi_device *sdevice)
1571 blk_queue_rq_timeout(sdevice->request_queue, (storvsc_timeout * HZ));
1573 /* storvsc devices don't support MAINTENANCE_IN SCSI cmd */
1574 sdevice->no_report_opcodes = 1;
1575 sdevice->no_write_same = 1;
1578 * If the host is WIN8 or WIN8 R2, claim conformance to SPC-3
1579 * if the device is a MSFT virtual device. If the host is
1580 * WIN10 or newer, allow write_same.
1582 if (!strncmp(sdevice->vendor, "Msft", 4)) {
1583 switch (vmstor_proto_version) {
1584 case VMSTOR_PROTO_VERSION_WIN8:
1585 case VMSTOR_PROTO_VERSION_WIN8_1:
1586 sdevice->scsi_level = SCSI_SPC_3;
1590 if (vmstor_proto_version >= VMSTOR_PROTO_VERSION_WIN10)
1591 sdevice->no_write_same = 0;
1597 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev,
1598 sector_t capacity, int *info)
1600 sector_t nsect = capacity;
1601 sector_t cylinders = nsect;
1602 int heads, sectors_pt;
1605 * We are making up these values; let us keep it simple.
1608 sectors_pt = 0x3f; /* Sectors per track */
1609 sector_div(cylinders, heads * sectors_pt);
1610 if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
1614 info[1] = sectors_pt;
1615 info[2] = (int)cylinders;
1620 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
1622 struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
1623 struct hv_device *device = host_dev->dev;
1625 struct storvsc_device *stor_device;
1626 struct storvsc_cmd_request *request;
1627 struct vstor_packet *vstor_packet;
1630 stor_device = get_out_stor_device(device);
1634 request = &stor_device->reset_request;
1635 vstor_packet = &request->vstor_packet;
1636 memset(vstor_packet, 0, sizeof(struct vstor_packet));
1638 init_completion(&request->wait_event);
1640 vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
1641 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
1642 vstor_packet->vm_srb.path_id = stor_device->path_id;
1644 ret = vmbus_sendpacket(device->channel, vstor_packet,
1645 sizeof(struct vstor_packet),
1648 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1652 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
1654 return TIMEOUT_ERROR;
1658 * At this point, all outstanding requests in the adapter
1659 * should have been flushed out and return to us
1660 * There is a potential race here where the host may be in
1661 * the process of responding when we return from here.
1662 * Just wait for all in-transit packets to be accounted for
1663 * before we return from here.
1665 storvsc_wait_to_drain(stor_device);
1671 * The host guarantees to respond to each command, although I/O latencies might
1672 * be unbounded on Azure. Reset the timer unconditionally to give the host a
1673 * chance to perform EH.
1675 static enum scsi_timeout_action storvsc_eh_timed_out(struct scsi_cmnd *scmnd)
1677 return SCSI_EH_RESET_TIMER;
1680 static bool storvsc_scsi_cmd_ok(struct scsi_cmnd *scmnd)
1682 bool allowed = true;
1683 u8 scsi_op = scmnd->cmnd[0];
1686 /* the host does not handle WRITE_SAME, log accident usage */
1689 * smartd sends this command and the host does not handle
1690 * this. So, don't send it.
1693 set_host_byte(scmnd, DID_ERROR);
1702 static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
1705 struct hv_host_device *host_dev = shost_priv(host);
1706 struct hv_device *dev = host_dev->dev;
1707 struct storvsc_cmd_request *cmd_request = scsi_cmd_priv(scmnd);
1708 struct scatterlist *sgl;
1709 struct vmscsi_request *vm_srb;
1710 struct vmbus_packet_mpb_array *payload;
1714 if (vmstor_proto_version <= VMSTOR_PROTO_VERSION_WIN8) {
1716 * On legacy hosts filter unimplemented commands.
1717 * Future hosts are expected to correctly handle
1718 * unsupported commands. Furthermore, it is
1719 * possible that some of the currently
1720 * unsupported commands maybe supported in
1721 * future versions of the host.
1723 if (!storvsc_scsi_cmd_ok(scmnd)) {
1729 /* Setup the cmd request */
1730 cmd_request->cmd = scmnd;
1732 memset(&cmd_request->vstor_packet, 0, sizeof(struct vstor_packet));
1733 vm_srb = &cmd_request->vstor_packet.vm_srb;
1734 vm_srb->time_out_value = 60;
1736 vm_srb->srb_flags |=
1737 SRB_FLAGS_DISABLE_SYNCH_TRANSFER;
1739 if (scmnd->device->tagged_supported) {
1740 vm_srb->srb_flags |=
1741 (SRB_FLAGS_QUEUE_ACTION_ENABLE | SRB_FLAGS_NO_QUEUE_FREEZE);
1742 vm_srb->queue_tag = SP_UNTAGGED;
1743 vm_srb->queue_action = SRB_SIMPLE_TAG_REQUEST;
1747 switch (scmnd->sc_data_direction) {
1749 vm_srb->data_in = WRITE_TYPE;
1750 vm_srb->srb_flags |= SRB_FLAGS_DATA_OUT;
1752 case DMA_FROM_DEVICE:
1753 vm_srb->data_in = READ_TYPE;
1754 vm_srb->srb_flags |= SRB_FLAGS_DATA_IN;
1757 vm_srb->data_in = UNKNOWN_TYPE;
1758 vm_srb->srb_flags |= SRB_FLAGS_NO_DATA_TRANSFER;
1762 * This is DMA_BIDIRECTIONAL or something else we are never
1763 * supposed to see here.
1765 WARN(1, "Unexpected data direction: %d\n",
1766 scmnd->sc_data_direction);
1771 vm_srb->port_number = host_dev->port;
1772 vm_srb->path_id = scmnd->device->channel;
1773 vm_srb->target_id = scmnd->device->id;
1774 vm_srb->lun = scmnd->device->lun;
1776 vm_srb->cdb_length = scmnd->cmd_len;
1778 memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
1780 sgl = (struct scatterlist *)scsi_sglist(scmnd);
1782 length = scsi_bufflen(scmnd);
1783 payload = (struct vmbus_packet_mpb_array *)&cmd_request->mpb;
1786 if (scsi_sg_count(scmnd)) {
1787 unsigned long offset_in_hvpg = offset_in_hvpage(sgl->offset);
1788 unsigned int hvpg_count = HVPFN_UP(offset_in_hvpg + length);
1789 struct scatterlist *sg;
1790 unsigned long hvpfn, hvpfns_to_add;
1791 int j, i = 0, sg_count;
1793 payload_sz = (hvpg_count * sizeof(u64) +
1794 sizeof(struct vmbus_packet_mpb_array));
1796 if (hvpg_count > MAX_PAGE_BUFFER_COUNT) {
1797 payload = kzalloc(payload_sz, GFP_ATOMIC);
1799 return SCSI_MLQUEUE_DEVICE_BUSY;
1802 payload->range.len = length;
1803 payload->range.offset = offset_in_hvpg;
1805 sg_count = scsi_dma_map(scmnd);
1807 ret = SCSI_MLQUEUE_DEVICE_BUSY;
1808 goto err_free_payload;
1811 for_each_sg(sgl, sg, sg_count, j) {
1813 * Init values for the current sgl entry. hvpfns_to_add
1814 * is in units of Hyper-V size pages. Handling the
1815 * PAGE_SIZE != HV_HYP_PAGE_SIZE case also handles
1816 * values of sgl->offset that are larger than PAGE_SIZE.
1817 * Such offsets are handled even on other than the first
1818 * sgl entry, provided they are a multiple of PAGE_SIZE.
1820 hvpfn = HVPFN_DOWN(sg_dma_address(sg));
1821 hvpfns_to_add = HVPFN_UP(sg_dma_address(sg) +
1822 sg_dma_len(sg)) - hvpfn;
1825 * Fill the next portion of the PFN array with
1826 * sequential Hyper-V PFNs for the continguous physical
1827 * memory described by the sgl entry. The end of the
1828 * last sgl should be reached at the same time that
1829 * the PFN array is filled.
1831 while (hvpfns_to_add--)
1832 payload->range.pfn_array[i++] = hvpfn++;
1836 cmd_request->payload = payload;
1837 cmd_request->payload_sz = payload_sz;
1839 /* Invokes the vsc to start an IO */
1840 ret = storvsc_do_io(dev, cmd_request, get_cpu());
1844 scsi_dma_unmap(scmnd);
1846 if (ret == -EAGAIN) {
1848 ret = SCSI_MLQUEUE_DEVICE_BUSY;
1849 goto err_free_payload;
1855 if (payload_sz > sizeof(cmd_request->mpb))
1861 static struct scsi_host_template scsi_driver = {
1862 .module = THIS_MODULE,
1863 .name = "storvsc_host_t",
1864 .cmd_size = sizeof(struct storvsc_cmd_request),
1865 .bios_param = storvsc_get_chs,
1866 .queuecommand = storvsc_queuecommand,
1867 .eh_host_reset_handler = storvsc_host_reset_handler,
1868 .proc_name = "storvsc_host",
1869 .eh_timed_out = storvsc_eh_timed_out,
1870 .slave_alloc = storvsc_device_alloc,
1871 .slave_configure = storvsc_device_configure,
1872 .cmd_per_lun = 2048,
1874 /* Ensure there are no gaps in presented sgls */
1875 .virt_boundary_mask = HV_HYP_PAGE_SIZE - 1,
1877 .track_queue_depth = 1,
1878 .change_queue_depth = storvsc_change_queue_depth,
1887 static const struct hv_vmbus_device_id id_table[] = {
1890 .driver_data = SCSI_GUID
1894 .driver_data = IDE_GUID
1896 /* Fibre Channel GUID */
1899 .driver_data = SFC_GUID
1904 MODULE_DEVICE_TABLE(vmbus, id_table);
1906 static const struct { guid_t guid; } fc_guid = { HV_SYNTHFC_GUID };
1908 static bool hv_dev_is_fc(struct hv_device *hv_dev)
1910 return guid_equal(&fc_guid.guid, &hv_dev->dev_type);
1913 static int storvsc_probe(struct hv_device *device,
1914 const struct hv_vmbus_device_id *dev_id)
1917 int num_cpus = num_online_cpus();
1918 int num_present_cpus = num_present_cpus();
1919 struct Scsi_Host *host;
1920 struct hv_host_device *host_dev;
1921 bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false);
1922 bool is_fc = ((dev_id->driver_data == SFC_GUID) ? true : false);
1924 struct storvsc_device *stor_device;
1925 int max_sub_channels = 0;
1929 * We support sub-channels for storage on SCSI and FC controllers.
1930 * The number of sub-channels offerred is based on the number of
1931 * VCPUs in the guest.
1935 (num_cpus - 1) / storvsc_vcpus_per_sub_channel;
1937 scsi_driver.can_queue = max_outstanding_req_per_channel *
1938 (max_sub_channels + 1) *
1939 (100 - ring_avail_percent_lowater) / 100;
1941 host = scsi_host_alloc(&scsi_driver,
1942 sizeof(struct hv_host_device));
1946 host_dev = shost_priv(host);
1947 memset(host_dev, 0, sizeof(struct hv_host_device));
1949 host_dev->port = host->host_no;
1950 host_dev->dev = device;
1951 host_dev->host = host;
1954 stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
1960 stor_device->destroy = false;
1961 init_waitqueue_head(&stor_device->waiting_to_drain);
1962 stor_device->device = device;
1963 stor_device->host = host;
1964 spin_lock_init(&stor_device->lock);
1965 hv_set_drvdata(device, stor_device);
1966 dma_set_min_align_mask(&device->device, HV_HYP_PAGE_SIZE - 1);
1968 stor_device->port_number = host->host_no;
1969 ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size, is_fc);
1973 host_dev->path = stor_device->path_id;
1974 host_dev->target = stor_device->target_id;
1976 switch (dev_id->driver_data) {
1978 host->max_lun = STORVSC_FC_MAX_LUNS_PER_TARGET;
1979 host->max_id = STORVSC_FC_MAX_TARGETS;
1980 host->max_channel = STORVSC_FC_MAX_CHANNELS - 1;
1981 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1982 host->transportt = fc_transport_template;
1987 host->max_lun = STORVSC_MAX_LUNS_PER_TARGET;
1988 host->max_id = STORVSC_MAX_TARGETS;
1989 host->max_channel = STORVSC_MAX_CHANNELS - 1;
1993 host->max_lun = STORVSC_IDE_MAX_LUNS_PER_TARGET;
1994 host->max_id = STORVSC_IDE_MAX_TARGETS;
1995 host->max_channel = STORVSC_IDE_MAX_CHANNELS - 1;
1998 /* max cmd length */
1999 host->max_cmd_len = STORVSC_MAX_CMD_LEN;
2001 * Any reasonable Hyper-V configuration should provide
2002 * max_transfer_bytes value aligning to HV_HYP_PAGE_SIZE,
2003 * protecting it from any weird value.
2005 max_xfer_bytes = round_down(stor_device->max_transfer_bytes, HV_HYP_PAGE_SIZE);
2007 max_xfer_bytes = min(max_xfer_bytes, STORVSC_FC_MAX_XFER_SIZE);
2009 /* max_hw_sectors_kb */
2010 host->max_sectors = max_xfer_bytes >> 9;
2012 * There are 2 requirements for Hyper-V storvsc sgl segments,
2013 * based on which the below calculation for max segments is
2016 * 1. Except for the first and last sgl segment, all sgl segments
2017 * should be align to HV_HYP_PAGE_SIZE, that also means the
2018 * maximum number of segments in a sgl can be calculated by
2019 * dividing the total max transfer length by HV_HYP_PAGE_SIZE.
2021 * 2. Except for the first and last, each entry in the SGL must
2022 * have an offset that is a multiple of HV_HYP_PAGE_SIZE.
2024 host->sg_tablesize = (max_xfer_bytes >> HV_HYP_PAGE_SHIFT) + 1;
2026 * For non-IDE disks, the host supports multiple channels.
2027 * Set the number of HW queues we are supporting.
2030 if (storvsc_max_hw_queues > num_present_cpus) {
2031 storvsc_max_hw_queues = 0;
2032 storvsc_log(device, STORVSC_LOGGING_WARN,
2033 "Resetting invalid storvsc_max_hw_queues value to default.\n");
2035 if (storvsc_max_hw_queues)
2036 host->nr_hw_queues = storvsc_max_hw_queues;
2038 host->nr_hw_queues = num_present_cpus;
2042 * Set the error handler work queue.
2044 host_dev->handle_error_wq =
2045 alloc_ordered_workqueue("storvsc_error_wq_%d",
2048 if (!host_dev->handle_error_wq) {
2052 INIT_WORK(&host_dev->host_scan_work, storvsc_host_scan);
2053 /* Register the HBA and start the scsi bus scan */
2054 ret = scsi_add_host(host, &device->device);
2059 scsi_scan_host(host);
2061 target = (device->dev_instance.b[5] << 8 |
2062 device->dev_instance.b[4]);
2063 ret = scsi_add_device(host, 0, target, 0);
2067 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2068 if (host->transportt == fc_transport_template) {
2069 struct fc_rport_identifiers ids = {
2070 .roles = FC_PORT_ROLE_FCP_DUMMY_INITIATOR,
2073 fc_host_node_name(host) = stor_device->node_name;
2074 fc_host_port_name(host) = stor_device->port_name;
2075 stor_device->rport = fc_remote_port_add(host, 0, &ids);
2076 if (!stor_device->rport) {
2085 scsi_remove_host(host);
2088 destroy_workqueue(host_dev->handle_error_wq);
2092 * Once we have connected with the host, we would need to
2093 * invoke storvsc_dev_remove() to rollback this state and
2094 * this call also frees up the stor_device; hence the jump around
2097 storvsc_dev_remove(device);
2101 kfree(stor_device->stor_chns);
2105 scsi_host_put(host);
2109 /* Change a scsi target's queue depth */
2110 static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth)
2112 if (queue_depth > scsi_driver.can_queue)
2113 queue_depth = scsi_driver.can_queue;
2115 return scsi_change_queue_depth(sdev, queue_depth);
2118 static void storvsc_remove(struct hv_device *dev)
2120 struct storvsc_device *stor_device = hv_get_drvdata(dev);
2121 struct Scsi_Host *host = stor_device->host;
2122 struct hv_host_device *host_dev = shost_priv(host);
2124 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2125 if (host->transportt == fc_transport_template) {
2126 fc_remote_port_delete(stor_device->rport);
2127 fc_remove_host(host);
2130 destroy_workqueue(host_dev->handle_error_wq);
2131 scsi_remove_host(host);
2132 storvsc_dev_remove(dev);
2133 scsi_host_put(host);
2136 static int storvsc_suspend(struct hv_device *hv_dev)
2138 struct storvsc_device *stor_device = hv_get_drvdata(hv_dev);
2139 struct Scsi_Host *host = stor_device->host;
2140 struct hv_host_device *host_dev = shost_priv(host);
2142 storvsc_wait_to_drain(stor_device);
2144 drain_workqueue(host_dev->handle_error_wq);
2146 vmbus_close(hv_dev->channel);
2148 kfree(stor_device->stor_chns);
2149 stor_device->stor_chns = NULL;
2151 cpumask_clear(&stor_device->alloced_cpus);
2156 static int storvsc_resume(struct hv_device *hv_dev)
2160 ret = storvsc_connect_to_vsp(hv_dev, storvsc_ringbuffer_size,
2161 hv_dev_is_fc(hv_dev));
2165 static struct hv_driver storvsc_drv = {
2166 .name = KBUILD_MODNAME,
2167 .id_table = id_table,
2168 .probe = storvsc_probe,
2169 .remove = storvsc_remove,
2170 .suspend = storvsc_suspend,
2171 .resume = storvsc_resume,
2173 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
2177 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2178 static struct fc_function_template fc_transport_functions = {
2179 .show_host_node_name = 1,
2180 .show_host_port_name = 1,
2184 static int __init storvsc_drv_init(void)
2189 * Divide the ring buffer data size (which is 1 page less
2190 * than the ring buffer size since that page is reserved for
2191 * the ring buffer indices) by the max request size (which is
2192 * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
2194 max_outstanding_req_per_channel =
2195 ((storvsc_ringbuffer_size - PAGE_SIZE) /
2196 ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
2197 sizeof(struct vstor_packet) + sizeof(u64),
2200 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2201 fc_transport_template = fc_attach_transport(&fc_transport_functions);
2202 if (!fc_transport_template)
2206 ret = vmbus_driver_register(&storvsc_drv);
2208 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2210 fc_release_transport(fc_transport_template);
2216 static void __exit storvsc_drv_exit(void)
2218 vmbus_driver_unregister(&storvsc_drv);
2219 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2220 fc_release_transport(fc_transport_template);
2224 MODULE_LICENSE("GPL");
2225 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
2226 module_init(storvsc_drv_init);
2227 module_exit(storvsc_drv_exit);