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
370 #define STORVSC_IDE_MAX_LUNS_PER_TARGET 64
371 #define STORVSC_IDE_MAX_TARGETS 1
372 #define STORVSC_IDE_MAX_CHANNELS 1
375 * Upper bound on the size of a storvsc packet.
377 #define STORVSC_MAX_PKT_SIZE (sizeof(struct vmpacket_descriptor) +\
378 sizeof(struct vstor_packet))
380 struct storvsc_cmd_request {
381 struct scsi_cmnd *cmd;
383 struct hv_device *device;
385 /* Synchronize the request/response if needed */
386 struct completion wait_event;
388 struct vmbus_channel_packet_multipage_buffer mpb;
389 struct vmbus_packet_mpb_array *payload;
392 struct vstor_packet vstor_packet;
396 /* A storvsc device is a device object that contains a vmbus channel */
397 struct storvsc_device {
398 struct hv_device *device;
402 atomic_t num_outstanding_req;
403 struct Scsi_Host *host;
405 wait_queue_head_t waiting_to_drain;
408 * Each unique Port/Path/Target represents 1 channel ie scsi
409 * controller. In reality, the pathid, targetid is always 0
410 * and the port is set by us
412 unsigned int port_number;
413 unsigned char path_id;
414 unsigned char target_id;
417 * Max I/O, the device can support.
419 u32 max_transfer_bytes;
421 * Number of sub-channels we will open.
424 struct vmbus_channel **stor_chns;
426 * Mask of CPUs bound to subchannels.
428 struct cpumask alloced_cpus;
430 * Serializes modifications of stor_chns[] from storvsc_do_io()
431 * and storvsc_change_target_cpu().
434 /* Used for vsc/vsp channel reset process */
435 struct storvsc_cmd_request init_request;
436 struct storvsc_cmd_request reset_request;
438 * Currently active port and node names for FC devices.
442 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
443 struct fc_rport *rport;
447 struct hv_host_device {
448 struct hv_device *dev;
451 unsigned char target;
452 struct workqueue_struct *handle_error_wq;
453 struct work_struct host_scan_work;
454 struct Scsi_Host *host;
457 struct storvsc_scan_work {
458 struct work_struct work;
459 struct Scsi_Host *host;
464 static void storvsc_device_scan(struct work_struct *work)
466 struct storvsc_scan_work *wrk;
467 struct scsi_device *sdev;
469 wrk = container_of(work, struct storvsc_scan_work, work);
471 sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
474 scsi_rescan_device(&sdev->sdev_gendev);
475 scsi_device_put(sdev);
481 static void storvsc_host_scan(struct work_struct *work)
483 struct Scsi_Host *host;
484 struct scsi_device *sdev;
485 struct hv_host_device *host_device =
486 container_of(work, struct hv_host_device, host_scan_work);
488 host = host_device->host;
490 * Before scanning the host, first check to see if any of the
491 * currently known devices have been hot removed. We issue a
492 * "unit ready" command against all currently known devices.
493 * This I/O will result in an error for devices that have been
494 * removed. As part of handling the I/O error, we remove the device.
496 * When a LUN is added or removed, the host sends us a signal to
497 * scan the host. Thus we are forced to discover the LUNs that
498 * may have been removed this way.
500 mutex_lock(&host->scan_mutex);
501 shost_for_each_device(sdev, host)
502 scsi_test_unit_ready(sdev, 1, 1, NULL);
503 mutex_unlock(&host->scan_mutex);
505 * Now scan the host to discover LUNs that may have been added.
507 scsi_scan_host(host);
510 static void storvsc_remove_lun(struct work_struct *work)
512 struct storvsc_scan_work *wrk;
513 struct scsi_device *sdev;
515 wrk = container_of(work, struct storvsc_scan_work, work);
516 if (!scsi_host_get(wrk->host))
519 sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
522 scsi_remove_device(sdev);
523 scsi_device_put(sdev);
525 scsi_host_put(wrk->host);
533 * We can get incoming messages from the host that are not in response to
534 * messages that we have sent out. An example of this would be messages
535 * received by the guest to notify dynamic addition/removal of LUNs. To
536 * deal with potential race conditions where the driver may be in the
537 * midst of being unloaded when we might receive an unsolicited message
538 * from the host, we have implemented a mechanism to gurantee sequential
541 * 1) Once the device is marked as being destroyed, we will fail all
543 * 2) We permit incoming messages when the device is being destroyed,
544 * only to properly account for messages already sent out.
547 static inline struct storvsc_device *get_out_stor_device(
548 struct hv_device *device)
550 struct storvsc_device *stor_device;
552 stor_device = hv_get_drvdata(device);
554 if (stor_device && stor_device->destroy)
561 static inline void storvsc_wait_to_drain(struct storvsc_device *dev)
563 dev->drain_notify = true;
564 wait_event(dev->waiting_to_drain,
565 atomic_read(&dev->num_outstanding_req) == 0);
566 dev->drain_notify = false;
569 static inline struct storvsc_device *get_in_stor_device(
570 struct hv_device *device)
572 struct storvsc_device *stor_device;
574 stor_device = hv_get_drvdata(device);
580 * If the device is being destroyed; allow incoming
581 * traffic only to cleanup outstanding requests.
584 if (stor_device->destroy &&
585 (atomic_read(&stor_device->num_outstanding_req) == 0))
593 static void storvsc_change_target_cpu(struct vmbus_channel *channel, u32 old,
596 struct storvsc_device *stor_device;
597 struct vmbus_channel *cur_chn;
598 bool old_is_alloced = false;
599 struct hv_device *device;
603 device = channel->primary_channel ?
604 channel->primary_channel->device_obj
605 : channel->device_obj;
606 stor_device = get_out_stor_device(device);
610 /* See storvsc_do_io() -> get_og_chn(). */
611 spin_lock_irqsave(&stor_device->lock, flags);
614 * Determines if the storvsc device has other channels assigned to
615 * the "old" CPU to update the alloced_cpus mask and the stor_chns
618 if (device->channel != channel && device->channel->target_cpu == old) {
619 cur_chn = device->channel;
620 old_is_alloced = true;
623 list_for_each_entry(cur_chn, &device->channel->sc_list, sc_list) {
624 if (cur_chn == channel)
626 if (cur_chn->target_cpu == old) {
627 old_is_alloced = true;
634 WRITE_ONCE(stor_device->stor_chns[old], cur_chn);
636 cpumask_clear_cpu(old, &stor_device->alloced_cpus);
638 /* "Flush" the stor_chns array. */
639 for_each_possible_cpu(cpu) {
640 if (stor_device->stor_chns[cpu] && !cpumask_test_cpu(
641 cpu, &stor_device->alloced_cpus))
642 WRITE_ONCE(stor_device->stor_chns[cpu], NULL);
645 WRITE_ONCE(stor_device->stor_chns[new], channel);
646 cpumask_set_cpu(new, &stor_device->alloced_cpus);
648 spin_unlock_irqrestore(&stor_device->lock, flags);
651 static u64 storvsc_next_request_id(struct vmbus_channel *channel, u64 rqst_addr)
653 struct storvsc_cmd_request *request =
654 (struct storvsc_cmd_request *)(unsigned long)rqst_addr;
656 if (rqst_addr == VMBUS_RQST_INIT)
657 return VMBUS_RQST_INIT;
658 if (rqst_addr == VMBUS_RQST_RESET)
659 return VMBUS_RQST_RESET;
662 * Cannot return an ID of 0, which is reserved for an unsolicited
663 * message from Hyper-V.
665 return (u64)blk_mq_unique_tag(scsi_cmd_to_rq(request->cmd)) + 1;
668 static void handle_sc_creation(struct vmbus_channel *new_sc)
670 struct hv_device *device = new_sc->primary_channel->device_obj;
671 struct device *dev = &device->device;
672 struct storvsc_device *stor_device;
673 struct vmstorage_channel_properties props;
676 stor_device = get_out_stor_device(device);
680 memset(&props, 0, sizeof(struct vmstorage_channel_properties));
681 new_sc->max_pkt_size = STORVSC_MAX_PKT_SIZE;
683 new_sc->next_request_id_callback = storvsc_next_request_id;
685 ret = vmbus_open(new_sc,
686 storvsc_ringbuffer_size,
687 storvsc_ringbuffer_size,
689 sizeof(struct vmstorage_channel_properties),
690 storvsc_on_channel_callback, new_sc);
692 /* In case vmbus_open() fails, we don't use the sub-channel. */
694 dev_err(dev, "Failed to open sub-channel: err=%d\n", ret);
698 new_sc->change_target_cpu_callback = storvsc_change_target_cpu;
700 /* Add the sub-channel to the array of available channels. */
701 stor_device->stor_chns[new_sc->target_cpu] = new_sc;
702 cpumask_set_cpu(new_sc->target_cpu, &stor_device->alloced_cpus);
705 static void handle_multichannel_storage(struct hv_device *device, int max_chns)
707 struct device *dev = &device->device;
708 struct storvsc_device *stor_device;
710 struct storvsc_cmd_request *request;
711 struct vstor_packet *vstor_packet;
715 * If the number of CPUs is artificially restricted, such as
716 * with maxcpus=1 on the kernel boot line, Hyper-V could offer
717 * sub-channels >= the number of CPUs. These sub-channels
718 * should not be created. The primary channel is already created
719 * and assigned to one CPU, so check against # CPUs - 1.
721 num_sc = min((int)(num_online_cpus() - 1), max_chns);
725 stor_device = get_out_stor_device(device);
729 stor_device->num_sc = num_sc;
730 request = &stor_device->init_request;
731 vstor_packet = &request->vstor_packet;
734 * Establish a handler for dealing with subchannels.
736 vmbus_set_sc_create_callback(device->channel, handle_sc_creation);
739 * Request the host to create sub-channels.
741 memset(request, 0, sizeof(struct storvsc_cmd_request));
742 init_completion(&request->wait_event);
743 vstor_packet->operation = VSTOR_OPERATION_CREATE_SUB_CHANNELS;
744 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
745 vstor_packet->sub_channel_count = num_sc;
747 ret = vmbus_sendpacket(device->channel, vstor_packet,
748 sizeof(struct vstor_packet),
751 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
754 dev_err(dev, "Failed to create sub-channel: err=%d\n", ret);
758 t = wait_for_completion_timeout(&request->wait_event, 10*HZ);
760 dev_err(dev, "Failed to create sub-channel: timed out\n");
764 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
765 vstor_packet->status != 0) {
766 dev_err(dev, "Failed to create sub-channel: op=%d, sts=%d\n",
767 vstor_packet->operation, vstor_packet->status);
772 * We need to do nothing here, because vmbus_process_offer()
773 * invokes channel->sc_creation_callback, which will open and use
774 * the sub-channel(s).
778 static void cache_wwn(struct storvsc_device *stor_device,
779 struct vstor_packet *vstor_packet)
782 * Cache the currently active port and node ww names.
784 if (vstor_packet->wwn_packet.primary_active) {
785 stor_device->node_name =
786 wwn_to_u64(vstor_packet->wwn_packet.primary_node_wwn);
787 stor_device->port_name =
788 wwn_to_u64(vstor_packet->wwn_packet.primary_port_wwn);
790 stor_device->node_name =
791 wwn_to_u64(vstor_packet->wwn_packet.secondary_node_wwn);
792 stor_device->port_name =
793 wwn_to_u64(vstor_packet->wwn_packet.secondary_port_wwn);
798 static int storvsc_execute_vstor_op(struct hv_device *device,
799 struct storvsc_cmd_request *request,
802 struct storvsc_device *stor_device;
803 struct vstor_packet *vstor_packet;
806 stor_device = get_out_stor_device(device);
810 vstor_packet = &request->vstor_packet;
812 init_completion(&request->wait_event);
813 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
815 ret = vmbus_sendpacket(device->channel, vstor_packet,
816 sizeof(struct vstor_packet),
819 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
823 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
830 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
831 vstor_packet->status != 0)
837 static int storvsc_channel_init(struct hv_device *device, bool is_fc)
839 struct storvsc_device *stor_device;
840 struct storvsc_cmd_request *request;
841 struct vstor_packet *vstor_packet;
844 bool process_sub_channels = false;
846 stor_device = get_out_stor_device(device);
850 request = &stor_device->init_request;
851 vstor_packet = &request->vstor_packet;
854 * Now, initiate the vsc/vsp initialization protocol on the open
857 memset(request, 0, sizeof(struct storvsc_cmd_request));
858 vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION;
859 ret = storvsc_execute_vstor_op(device, request, true);
863 * Query host supported protocol version.
866 for (i = 0; i < ARRAY_SIZE(protocol_version); i++) {
867 /* reuse the packet for version range supported */
868 memset(vstor_packet, 0, sizeof(struct vstor_packet));
869 vstor_packet->operation =
870 VSTOR_OPERATION_QUERY_PROTOCOL_VERSION;
872 vstor_packet->version.major_minor = protocol_version[i];
875 * The revision number is only used in Windows; set it to 0.
877 vstor_packet->version.revision = 0;
878 ret = storvsc_execute_vstor_op(device, request, false);
882 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO)
885 if (vstor_packet->status == 0) {
886 vmstor_proto_version = protocol_version[i];
892 if (vstor_packet->status != 0) {
893 dev_err(&device->device, "Obsolete Hyper-V version\n");
898 memset(vstor_packet, 0, sizeof(struct vstor_packet));
899 vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES;
900 ret = storvsc_execute_vstor_op(device, request, true);
905 * Check to see if multi-channel support is there.
906 * Hosts that implement protocol version of 5.1 and above
907 * support multi-channel.
909 max_chns = vstor_packet->storage_channel_properties.max_channel_cnt;
912 * Allocate state to manage the sub-channels.
913 * We allocate an array based on the numbers of possible CPUs
914 * (Hyper-V does not support cpu online/offline).
915 * This Array will be sparseley populated with unique
916 * channels - primary + sub-channels.
917 * We will however populate all the slots to evenly distribute
920 stor_device->stor_chns = kcalloc(num_possible_cpus(), sizeof(void *),
922 if (stor_device->stor_chns == NULL)
925 device->channel->change_target_cpu_callback = storvsc_change_target_cpu;
927 stor_device->stor_chns[device->channel->target_cpu] = device->channel;
928 cpumask_set_cpu(device->channel->target_cpu,
929 &stor_device->alloced_cpus);
931 if (vstor_packet->storage_channel_properties.flags &
932 STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL)
933 process_sub_channels = true;
935 stor_device->max_transfer_bytes =
936 vstor_packet->storage_channel_properties.max_transfer_bytes;
942 * For FC devices retrieve FC HBA data.
944 memset(vstor_packet, 0, sizeof(struct vstor_packet));
945 vstor_packet->operation = VSTOR_OPERATION_FCHBA_DATA;
946 ret = storvsc_execute_vstor_op(device, request, true);
951 * Cache the currently active port and node ww names.
953 cache_wwn(stor_device, vstor_packet);
957 memset(vstor_packet, 0, sizeof(struct vstor_packet));
958 vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION;
959 ret = storvsc_execute_vstor_op(device, request, true);
963 if (process_sub_channels)
964 handle_multichannel_storage(device, max_chns);
969 static void storvsc_handle_error(struct vmscsi_request *vm_srb,
970 struct scsi_cmnd *scmnd,
971 struct Scsi_Host *host,
974 struct storvsc_scan_work *wrk;
975 void (*process_err_fn)(struct work_struct *work);
976 struct hv_host_device *host_dev = shost_priv(host);
978 switch (SRB_STATUS(vm_srb->srb_status)) {
979 case SRB_STATUS_ERROR:
980 case SRB_STATUS_ABORTED:
981 case SRB_STATUS_INVALID_REQUEST:
982 case SRB_STATUS_INTERNAL_ERROR:
983 if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID) {
984 /* Check for capacity change */
985 if ((asc == 0x2a) && (ascq == 0x9)) {
986 process_err_fn = storvsc_device_scan;
987 /* Retry the I/O that triggered this. */
988 set_host_byte(scmnd, DID_REQUEUE);
993 * Check for "Operating parameters have changed"
994 * due to Hyper-V changing the VHD/VHDX BlockSize
995 * when adding/removing a differencing disk. This
996 * causes discard_granularity to change, so do a
997 * rescan to pick up the new granularity. We don't
998 * want scsi_report_sense() to output a message
999 * that a sysadmin wouldn't know what to do with.
1001 if ((asc == 0x3f) && (ascq != 0x03) &&
1003 process_err_fn = storvsc_device_scan;
1004 set_host_byte(scmnd, DID_REQUEUE);
1009 * Otherwise, let upper layer deal with the
1010 * error when sense message is present
1016 * If there is an error; offline the device since all
1017 * error recovery strategies would have already been
1018 * deployed on the host side. However, if the command
1019 * were a pass-through command deal with it appropriately.
1021 switch (scmnd->cmnd[0]) {
1024 set_host_byte(scmnd, DID_PASSTHROUGH);
1027 * On some Hyper-V hosts TEST_UNIT_READY command can
1028 * return SRB_STATUS_ERROR. Let the upper level code
1029 * deal with it based on the sense information.
1031 case TEST_UNIT_READY:
1034 set_host_byte(scmnd, DID_ERROR);
1038 case SRB_STATUS_INVALID_LUN:
1039 set_host_byte(scmnd, DID_NO_CONNECT);
1040 process_err_fn = storvsc_remove_lun;
1048 * We need to schedule work to process this error; schedule it.
1050 wrk = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC);
1052 set_host_byte(scmnd, DID_BAD_TARGET);
1057 wrk->lun = vm_srb->lun;
1058 wrk->tgt_id = vm_srb->target_id;
1059 INIT_WORK(&wrk->work, process_err_fn);
1060 queue_work(host_dev->handle_error_wq, &wrk->work);
1064 static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request,
1065 struct storvsc_device *stor_dev)
1067 struct scsi_cmnd *scmnd = cmd_request->cmd;
1068 struct scsi_sense_hdr sense_hdr;
1069 struct vmscsi_request *vm_srb;
1070 u32 data_transfer_length;
1071 struct Scsi_Host *host;
1072 u32 payload_sz = cmd_request->payload_sz;
1073 void *payload = cmd_request->payload;
1076 host = stor_dev->host;
1078 vm_srb = &cmd_request->vstor_packet.vm_srb;
1079 data_transfer_length = vm_srb->data_transfer_length;
1081 scmnd->result = vm_srb->scsi_status;
1083 if (scmnd->result) {
1084 sense_ok = scsi_normalize_sense(scmnd->sense_buffer,
1085 SCSI_SENSE_BUFFERSIZE, &sense_hdr);
1087 if (sense_ok && do_logging(STORVSC_LOGGING_WARN))
1088 scsi_print_sense_hdr(scmnd->device, "storvsc",
1092 if (vm_srb->srb_status != SRB_STATUS_SUCCESS) {
1093 storvsc_handle_error(vm_srb, scmnd, host, sense_hdr.asc,
1096 * The Windows driver set data_transfer_length on
1097 * SRB_STATUS_DATA_OVERRUN. On other errors, this value
1098 * is untouched. In these cases we set it to 0.
1100 if (vm_srb->srb_status != SRB_STATUS_DATA_OVERRUN)
1101 data_transfer_length = 0;
1104 /* Validate data_transfer_length (from Hyper-V) */
1105 if (data_transfer_length > cmd_request->payload->range.len)
1106 data_transfer_length = cmd_request->payload->range.len;
1108 scsi_set_resid(scmnd,
1109 cmd_request->payload->range.len - data_transfer_length);
1114 sizeof(struct vmbus_channel_packet_multipage_buffer))
1118 static void storvsc_on_io_completion(struct storvsc_device *stor_device,
1119 struct vstor_packet *vstor_packet,
1120 struct storvsc_cmd_request *request)
1122 struct vstor_packet *stor_pkt;
1123 struct hv_device *device = stor_device->device;
1125 stor_pkt = &request->vstor_packet;
1128 * The current SCSI handling on the host side does
1129 * not correctly handle:
1130 * INQUIRY command with page code parameter set to 0x80
1131 * MODE_SENSE command with cmd[2] == 0x1c
1133 * Setup srb and scsi status so this won't be fatal.
1134 * We do this so we can distinguish truly fatal failues
1135 * (srb status == 0x4) and off-line the device in that case.
1138 if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) ||
1139 (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) {
1140 vstor_packet->vm_srb.scsi_status = 0;
1141 vstor_packet->vm_srb.srb_status = SRB_STATUS_SUCCESS;
1144 /* Copy over the status...etc */
1145 stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status;
1146 stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status;
1149 * Copy over the sense_info_length, but limit to the known max
1150 * size if Hyper-V returns a bad value.
1152 stor_pkt->vm_srb.sense_info_length = min_t(u8, STORVSC_SENSE_BUFFER_SIZE,
1153 vstor_packet->vm_srb.sense_info_length);
1155 if (vstor_packet->vm_srb.scsi_status != 0 ||
1156 vstor_packet->vm_srb.srb_status != SRB_STATUS_SUCCESS) {
1159 * Log TEST_UNIT_READY errors only as warnings. Hyper-V can
1160 * return errors when detecting devices using TEST_UNIT_READY,
1161 * and logging these as errors produces unhelpful noise.
1163 int loglevel = (stor_pkt->vm_srb.cdb[0] == TEST_UNIT_READY) ?
1164 STORVSC_LOGGING_WARN : STORVSC_LOGGING_ERROR;
1166 storvsc_log(device, loglevel,
1167 "tag#%d cmd 0x%x status: scsi 0x%x srb 0x%x hv 0x%x\n",
1168 scsi_cmd_to_rq(request->cmd)->tag,
1169 stor_pkt->vm_srb.cdb[0],
1170 vstor_packet->vm_srb.scsi_status,
1171 vstor_packet->vm_srb.srb_status,
1172 vstor_packet->status);
1175 if (vstor_packet->vm_srb.scsi_status == SAM_STAT_CHECK_CONDITION &&
1176 (vstor_packet->vm_srb.srb_status & SRB_STATUS_AUTOSENSE_VALID))
1177 memcpy(request->cmd->sense_buffer,
1178 vstor_packet->vm_srb.sense_data,
1179 stor_pkt->vm_srb.sense_info_length);
1181 stor_pkt->vm_srb.data_transfer_length =
1182 vstor_packet->vm_srb.data_transfer_length;
1184 storvsc_command_completion(request, stor_device);
1186 if (atomic_dec_and_test(&stor_device->num_outstanding_req) &&
1187 stor_device->drain_notify)
1188 wake_up(&stor_device->waiting_to_drain);
1191 static void storvsc_on_receive(struct storvsc_device *stor_device,
1192 struct vstor_packet *vstor_packet,
1193 struct storvsc_cmd_request *request)
1195 struct hv_host_device *host_dev;
1196 switch (vstor_packet->operation) {
1197 case VSTOR_OPERATION_COMPLETE_IO:
1198 storvsc_on_io_completion(stor_device, vstor_packet, request);
1201 case VSTOR_OPERATION_REMOVE_DEVICE:
1202 case VSTOR_OPERATION_ENUMERATE_BUS:
1203 host_dev = shost_priv(stor_device->host);
1205 host_dev->handle_error_wq, &host_dev->host_scan_work);
1208 case VSTOR_OPERATION_FCHBA_DATA:
1209 cache_wwn(stor_device, vstor_packet);
1210 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1211 fc_host_node_name(stor_device->host) = stor_device->node_name;
1212 fc_host_port_name(stor_device->host) = stor_device->port_name;
1220 static void storvsc_on_channel_callback(void *context)
1222 struct vmbus_channel *channel = (struct vmbus_channel *)context;
1223 const struct vmpacket_descriptor *desc;
1224 struct hv_device *device;
1225 struct storvsc_device *stor_device;
1226 struct Scsi_Host *shost;
1227 unsigned long time_limit = jiffies + msecs_to_jiffies(CALLBACK_TIMEOUT);
1229 if (channel->primary_channel != NULL)
1230 device = channel->primary_channel->device_obj;
1232 device = channel->device_obj;
1234 stor_device = get_in_stor_device(device);
1238 shost = stor_device->host;
1240 foreach_vmbus_pkt(desc, channel) {
1241 struct vstor_packet *packet = hv_pkt_data(desc);
1242 struct storvsc_cmd_request *request = NULL;
1243 u32 pktlen = hv_pkt_datalen(desc);
1244 u64 rqst_id = desc->trans_id;
1245 u32 minlen = rqst_id ? sizeof(struct vstor_packet) :
1246 sizeof(enum vstor_packet_operation);
1248 if (unlikely(time_after(jiffies, time_limit))) {
1249 hv_pkt_iter_close(channel);
1253 if (pktlen < minlen) {
1254 dev_err(&device->device,
1255 "Invalid pkt: id=%llu, len=%u, minlen=%u\n",
1256 rqst_id, pktlen, minlen);
1260 if (rqst_id == VMBUS_RQST_INIT) {
1261 request = &stor_device->init_request;
1262 } else if (rqst_id == VMBUS_RQST_RESET) {
1263 request = &stor_device->reset_request;
1265 /* Hyper-V can send an unsolicited message with ID of 0 */
1268 * storvsc_on_receive() looks at the vstor_packet in the message
1269 * from the ring buffer.
1271 * - If the operation in the vstor_packet is COMPLETE_IO, then
1272 * we call storvsc_on_io_completion(), and dereference the
1273 * guest memory address. Make sure we don't call
1274 * storvsc_on_io_completion() with a guest memory address
1275 * that is zero if Hyper-V were to construct and send such
1278 * - If the operation in the vstor_packet is FCHBA_DATA, then
1279 * we call cache_wwn(), and access the data payload area of
1280 * the packet (wwn_packet); however, there is no guarantee
1281 * that the packet is big enough to contain such area.
1282 * Future-proof the code by rejecting such a bogus packet.
1284 if (packet->operation == VSTOR_OPERATION_COMPLETE_IO ||
1285 packet->operation == VSTOR_OPERATION_FCHBA_DATA) {
1286 dev_err(&device->device, "Invalid packet with ID of 0\n");
1290 struct scsi_cmnd *scmnd;
1292 /* Transaction 'rqst_id' corresponds to tag 'rqst_id - 1' */
1293 scmnd = scsi_host_find_tag(shost, rqst_id - 1);
1294 if (scmnd == NULL) {
1295 dev_err(&device->device, "Incorrect transaction ID\n");
1298 request = (struct storvsc_cmd_request *)scsi_cmd_priv(scmnd);
1299 scsi_dma_unmap(scmnd);
1302 storvsc_on_receive(stor_device, packet, request);
1306 memcpy(&request->vstor_packet, packet,
1307 sizeof(struct vstor_packet));
1308 complete(&request->wait_event);
1312 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size,
1315 struct vmstorage_channel_properties props;
1318 memset(&props, 0, sizeof(struct vmstorage_channel_properties));
1320 device->channel->max_pkt_size = STORVSC_MAX_PKT_SIZE;
1321 device->channel->next_request_id_callback = storvsc_next_request_id;
1323 ret = vmbus_open(device->channel,
1327 sizeof(struct vmstorage_channel_properties),
1328 storvsc_on_channel_callback, device->channel);
1333 ret = storvsc_channel_init(device, is_fc);
1338 static int storvsc_dev_remove(struct hv_device *device)
1340 struct storvsc_device *stor_device;
1342 stor_device = hv_get_drvdata(device);
1344 stor_device->destroy = true;
1346 /* Make sure flag is set before waiting */
1350 * At this point, all outbound traffic should be disable. We
1351 * only allow inbound traffic (responses) to proceed so that
1352 * outstanding requests can be completed.
1355 storvsc_wait_to_drain(stor_device);
1358 * Since we have already drained, we don't need to busy wait
1359 * as was done in final_release_stor_device()
1360 * Note that we cannot set the ext pointer to NULL until
1361 * we have drained - to drain the outgoing packets, we need to
1362 * allow incoming packets.
1364 hv_set_drvdata(device, NULL);
1366 /* Close the channel */
1367 vmbus_close(device->channel);
1369 kfree(stor_device->stor_chns);
1374 static struct vmbus_channel *get_og_chn(struct storvsc_device *stor_device,
1379 const struct cpumask *node_mask;
1380 int num_channels, tgt_cpu;
1382 if (stor_device->num_sc == 0) {
1383 stor_device->stor_chns[q_num] = stor_device->device->channel;
1384 return stor_device->device->channel;
1388 * Our channel array is sparsley populated and we
1389 * initiated I/O on a processor/hw-q that does not
1390 * currently have a designated channel. Fix this.
1391 * The strategy is simple:
1392 * I. Ensure NUMA locality
1393 * II. Distribute evenly (best effort)
1396 node_mask = cpumask_of_node(cpu_to_node(q_num));
1399 for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1400 if (cpumask_test_cpu(tgt_cpu, node_mask))
1403 if (num_channels == 0) {
1404 stor_device->stor_chns[q_num] = stor_device->device->channel;
1405 return stor_device->device->channel;
1409 while (hash_qnum >= num_channels)
1410 hash_qnum -= num_channels;
1412 for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1413 if (!cpumask_test_cpu(tgt_cpu, node_mask))
1415 if (slot == hash_qnum)
1420 stor_device->stor_chns[q_num] = stor_device->stor_chns[tgt_cpu];
1422 return stor_device->stor_chns[q_num];
1426 static int storvsc_do_io(struct hv_device *device,
1427 struct storvsc_cmd_request *request, u16 q_num)
1429 struct storvsc_device *stor_device;
1430 struct vstor_packet *vstor_packet;
1431 struct vmbus_channel *outgoing_channel, *channel;
1432 unsigned long flags;
1434 const struct cpumask *node_mask;
1437 vstor_packet = &request->vstor_packet;
1438 stor_device = get_out_stor_device(device);
1444 request->device = device;
1446 * Select an appropriate channel to send the request out.
1448 /* See storvsc_change_target_cpu(). */
1449 outgoing_channel = READ_ONCE(stor_device->stor_chns[q_num]);
1450 if (outgoing_channel != NULL) {
1451 if (outgoing_channel->target_cpu == q_num) {
1453 * Ideally, we want to pick a different channel if
1454 * available on the same NUMA node.
1456 node_mask = cpumask_of_node(cpu_to_node(q_num));
1457 for_each_cpu_wrap(tgt_cpu,
1458 &stor_device->alloced_cpus, q_num + 1) {
1459 if (!cpumask_test_cpu(tgt_cpu, node_mask))
1461 if (tgt_cpu == q_num)
1463 channel = READ_ONCE(
1464 stor_device->stor_chns[tgt_cpu]);
1465 if (channel == NULL)
1467 if (hv_get_avail_to_write_percent(
1469 > ring_avail_percent_lowater) {
1470 outgoing_channel = channel;
1476 * All the other channels on the same NUMA node are
1477 * busy. Try to use the channel on the current CPU
1479 if (hv_get_avail_to_write_percent(
1480 &outgoing_channel->outbound)
1481 > ring_avail_percent_lowater)
1485 * If we reach here, all the channels on the current
1486 * NUMA node are busy. Try to find a channel in
1489 for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1490 if (cpumask_test_cpu(tgt_cpu, node_mask))
1492 channel = READ_ONCE(
1493 stor_device->stor_chns[tgt_cpu]);
1494 if (channel == NULL)
1496 if (hv_get_avail_to_write_percent(
1498 > ring_avail_percent_lowater) {
1499 outgoing_channel = channel;
1505 spin_lock_irqsave(&stor_device->lock, flags);
1506 outgoing_channel = stor_device->stor_chns[q_num];
1507 if (outgoing_channel != NULL) {
1508 spin_unlock_irqrestore(&stor_device->lock, flags);
1511 outgoing_channel = get_og_chn(stor_device, q_num);
1512 spin_unlock_irqrestore(&stor_device->lock, flags);
1516 vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
1518 vstor_packet->vm_srb.length = sizeof(struct vmscsi_request);
1521 vstor_packet->vm_srb.sense_info_length = STORVSC_SENSE_BUFFER_SIZE;
1524 vstor_packet->vm_srb.data_transfer_length =
1525 request->payload->range.len;
1527 vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
1529 if (request->payload->range.len) {
1531 ret = vmbus_sendpacket_mpb_desc(outgoing_channel,
1532 request->payload, request->payload_sz,
1534 sizeof(struct vstor_packet),
1535 (unsigned long)request);
1537 ret = vmbus_sendpacket(outgoing_channel, vstor_packet,
1538 sizeof(struct vstor_packet),
1539 (unsigned long)request,
1541 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1547 atomic_inc(&stor_device->num_outstanding_req);
1552 static int storvsc_device_alloc(struct scsi_device *sdevice)
1555 * Set blist flag to permit the reading of the VPD pages even when
1556 * the target may claim SPC-2 compliance. MSFT targets currently
1557 * claim SPC-2 compliance while they implement post SPC-2 features.
1558 * With this flag we can correctly handle WRITE_SAME_16 issues.
1560 * Hypervisor reports SCSI_UNKNOWN type for DVD ROM device but
1561 * still supports REPORT LUN.
1563 sdevice->sdev_bflags = BLIST_REPORTLUN2 | BLIST_TRY_VPD_PAGES;
1568 static int storvsc_device_configure(struct scsi_device *sdevice)
1570 blk_queue_rq_timeout(sdevice->request_queue, (storvsc_timeout * HZ));
1572 /* storvsc devices don't support MAINTENANCE_IN SCSI cmd */
1573 sdevice->no_report_opcodes = 1;
1574 sdevice->no_write_same = 1;
1577 * If the host is WIN8 or WIN8 R2, claim conformance to SPC-3
1578 * if the device is a MSFT virtual device. If the host is
1579 * WIN10 or newer, allow write_same.
1581 if (!strncmp(sdevice->vendor, "Msft", 4)) {
1582 switch (vmstor_proto_version) {
1583 case VMSTOR_PROTO_VERSION_WIN8:
1584 case VMSTOR_PROTO_VERSION_WIN8_1:
1585 sdevice->scsi_level = SCSI_SPC_3;
1589 if (vmstor_proto_version >= VMSTOR_PROTO_VERSION_WIN10)
1590 sdevice->no_write_same = 0;
1596 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev,
1597 sector_t capacity, int *info)
1599 sector_t nsect = capacity;
1600 sector_t cylinders = nsect;
1601 int heads, sectors_pt;
1604 * We are making up these values; let us keep it simple.
1607 sectors_pt = 0x3f; /* Sectors per track */
1608 sector_div(cylinders, heads * sectors_pt);
1609 if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
1613 info[1] = sectors_pt;
1614 info[2] = (int)cylinders;
1619 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
1621 struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
1622 struct hv_device *device = host_dev->dev;
1624 struct storvsc_device *stor_device;
1625 struct storvsc_cmd_request *request;
1626 struct vstor_packet *vstor_packet;
1629 stor_device = get_out_stor_device(device);
1633 request = &stor_device->reset_request;
1634 vstor_packet = &request->vstor_packet;
1635 memset(vstor_packet, 0, sizeof(struct vstor_packet));
1637 init_completion(&request->wait_event);
1639 vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
1640 vstor_packet->flags = REQUEST_COMPLETION_FLAG;
1641 vstor_packet->vm_srb.path_id = stor_device->path_id;
1643 ret = vmbus_sendpacket(device->channel, vstor_packet,
1644 sizeof(struct vstor_packet),
1647 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1651 t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
1653 return TIMEOUT_ERROR;
1657 * At this point, all outstanding requests in the adapter
1658 * should have been flushed out and return to us
1659 * There is a potential race here where the host may be in
1660 * the process of responding when we return from here.
1661 * Just wait for all in-transit packets to be accounted for
1662 * before we return from here.
1664 storvsc_wait_to_drain(stor_device);
1670 * The host guarantees to respond to each command, although I/O latencies might
1671 * be unbounded on Azure. Reset the timer unconditionally to give the host a
1672 * chance to perform EH.
1674 static enum scsi_timeout_action storvsc_eh_timed_out(struct scsi_cmnd *scmnd)
1676 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1677 if (scmnd->device->host->transportt == fc_transport_template)
1678 return fc_eh_timed_out(scmnd);
1680 return SCSI_EH_RESET_TIMER;
1683 static bool storvsc_scsi_cmd_ok(struct scsi_cmnd *scmnd)
1685 bool allowed = true;
1686 u8 scsi_op = scmnd->cmnd[0];
1689 /* the host does not handle WRITE_SAME, log accident usage */
1692 * smartd sends this command and the host does not handle
1693 * this. So, don't send it.
1696 set_host_byte(scmnd, DID_ERROR);
1705 static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
1708 struct hv_host_device *host_dev = shost_priv(host);
1709 struct hv_device *dev = host_dev->dev;
1710 struct storvsc_cmd_request *cmd_request = scsi_cmd_priv(scmnd);
1711 struct scatterlist *sgl;
1712 struct vmscsi_request *vm_srb;
1713 struct vmbus_packet_mpb_array *payload;
1717 if (vmstor_proto_version <= VMSTOR_PROTO_VERSION_WIN8) {
1719 * On legacy hosts filter unimplemented commands.
1720 * Future hosts are expected to correctly handle
1721 * unsupported commands. Furthermore, it is
1722 * possible that some of the currently
1723 * unsupported commands maybe supported in
1724 * future versions of the host.
1726 if (!storvsc_scsi_cmd_ok(scmnd)) {
1732 /* Setup the cmd request */
1733 cmd_request->cmd = scmnd;
1735 memset(&cmd_request->vstor_packet, 0, sizeof(struct vstor_packet));
1736 vm_srb = &cmd_request->vstor_packet.vm_srb;
1737 vm_srb->time_out_value = 60;
1739 vm_srb->srb_flags |=
1740 SRB_FLAGS_DISABLE_SYNCH_TRANSFER;
1742 if (scmnd->device->tagged_supported) {
1743 vm_srb->srb_flags |=
1744 (SRB_FLAGS_QUEUE_ACTION_ENABLE | SRB_FLAGS_NO_QUEUE_FREEZE);
1745 vm_srb->queue_tag = SP_UNTAGGED;
1746 vm_srb->queue_action = SRB_SIMPLE_TAG_REQUEST;
1750 switch (scmnd->sc_data_direction) {
1752 vm_srb->data_in = WRITE_TYPE;
1753 vm_srb->srb_flags |= SRB_FLAGS_DATA_OUT;
1755 case DMA_FROM_DEVICE:
1756 vm_srb->data_in = READ_TYPE;
1757 vm_srb->srb_flags |= SRB_FLAGS_DATA_IN;
1760 vm_srb->data_in = UNKNOWN_TYPE;
1761 vm_srb->srb_flags |= SRB_FLAGS_NO_DATA_TRANSFER;
1765 * This is DMA_BIDIRECTIONAL or something else we are never
1766 * supposed to see here.
1768 WARN(1, "Unexpected data direction: %d\n",
1769 scmnd->sc_data_direction);
1774 vm_srb->port_number = host_dev->port;
1775 vm_srb->path_id = scmnd->device->channel;
1776 vm_srb->target_id = scmnd->device->id;
1777 vm_srb->lun = scmnd->device->lun;
1779 vm_srb->cdb_length = scmnd->cmd_len;
1781 memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
1783 sgl = (struct scatterlist *)scsi_sglist(scmnd);
1785 length = scsi_bufflen(scmnd);
1786 payload = (struct vmbus_packet_mpb_array *)&cmd_request->mpb;
1789 if (scsi_sg_count(scmnd)) {
1790 unsigned long offset_in_hvpg = offset_in_hvpage(sgl->offset);
1791 unsigned int hvpg_count = HVPFN_UP(offset_in_hvpg + length);
1792 struct scatterlist *sg;
1793 unsigned long hvpfn, hvpfns_to_add;
1794 int j, i = 0, sg_count;
1796 payload_sz = (hvpg_count * sizeof(u64) +
1797 sizeof(struct vmbus_packet_mpb_array));
1799 if (hvpg_count > MAX_PAGE_BUFFER_COUNT) {
1800 payload = kzalloc(payload_sz, GFP_ATOMIC);
1802 return SCSI_MLQUEUE_DEVICE_BUSY;
1805 payload->range.len = length;
1806 payload->range.offset = offset_in_hvpg;
1808 sg_count = scsi_dma_map(scmnd);
1810 ret = SCSI_MLQUEUE_DEVICE_BUSY;
1811 goto err_free_payload;
1814 for_each_sg(sgl, sg, sg_count, j) {
1816 * Init values for the current sgl entry. hvpfns_to_add
1817 * is in units of Hyper-V size pages. Handling the
1818 * PAGE_SIZE != HV_HYP_PAGE_SIZE case also handles
1819 * values of sgl->offset that are larger than PAGE_SIZE.
1820 * Such offsets are handled even on other than the first
1821 * sgl entry, provided they are a multiple of PAGE_SIZE.
1823 hvpfn = HVPFN_DOWN(sg_dma_address(sg));
1824 hvpfns_to_add = HVPFN_UP(sg_dma_address(sg) +
1825 sg_dma_len(sg)) - hvpfn;
1828 * Fill the next portion of the PFN array with
1829 * sequential Hyper-V PFNs for the continguous physical
1830 * memory described by the sgl entry. The end of the
1831 * last sgl should be reached at the same time that
1832 * the PFN array is filled.
1834 while (hvpfns_to_add--)
1835 payload->range.pfn_array[i++] = hvpfn++;
1839 cmd_request->payload = payload;
1840 cmd_request->payload_sz = payload_sz;
1842 /* Invokes the vsc to start an IO */
1843 ret = storvsc_do_io(dev, cmd_request, get_cpu());
1847 scsi_dma_unmap(scmnd);
1849 if (ret == -EAGAIN) {
1851 ret = SCSI_MLQUEUE_DEVICE_BUSY;
1852 goto err_free_payload;
1858 if (payload_sz > sizeof(cmd_request->mpb))
1864 static struct scsi_host_template scsi_driver = {
1865 .module = THIS_MODULE,
1866 .name = "storvsc_host_t",
1867 .cmd_size = sizeof(struct storvsc_cmd_request),
1868 .bios_param = storvsc_get_chs,
1869 .queuecommand = storvsc_queuecommand,
1870 .eh_host_reset_handler = storvsc_host_reset_handler,
1871 .proc_name = "storvsc_host",
1872 .eh_timed_out = storvsc_eh_timed_out,
1873 .slave_alloc = storvsc_device_alloc,
1874 .slave_configure = storvsc_device_configure,
1875 .cmd_per_lun = 2048,
1877 /* Ensure there are no gaps in presented sgls */
1878 .virt_boundary_mask = HV_HYP_PAGE_SIZE - 1,
1880 .track_queue_depth = 1,
1881 .change_queue_depth = storvsc_change_queue_depth,
1890 static const struct hv_vmbus_device_id id_table[] = {
1893 .driver_data = SCSI_GUID
1897 .driver_data = IDE_GUID
1899 /* Fibre Channel GUID */
1902 .driver_data = SFC_GUID
1907 MODULE_DEVICE_TABLE(vmbus, id_table);
1909 static const struct { guid_t guid; } fc_guid = { HV_SYNTHFC_GUID };
1911 static bool hv_dev_is_fc(struct hv_device *hv_dev)
1913 return guid_equal(&fc_guid.guid, &hv_dev->dev_type);
1916 static int storvsc_probe(struct hv_device *device,
1917 const struct hv_vmbus_device_id *dev_id)
1920 int num_cpus = num_online_cpus();
1921 int num_present_cpus = num_present_cpus();
1922 struct Scsi_Host *host;
1923 struct hv_host_device *host_dev;
1924 bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false);
1925 bool is_fc = ((dev_id->driver_data == SFC_GUID) ? true : false);
1927 struct storvsc_device *stor_device;
1928 int max_sub_channels = 0;
1932 * We support sub-channels for storage on SCSI and FC controllers.
1933 * The number of sub-channels offerred is based on the number of
1934 * VCPUs in the guest.
1938 (num_cpus - 1) / storvsc_vcpus_per_sub_channel;
1940 scsi_driver.can_queue = max_outstanding_req_per_channel *
1941 (max_sub_channels + 1) *
1942 (100 - ring_avail_percent_lowater) / 100;
1944 host = scsi_host_alloc(&scsi_driver,
1945 sizeof(struct hv_host_device));
1949 host_dev = shost_priv(host);
1950 memset(host_dev, 0, sizeof(struct hv_host_device));
1952 host_dev->port = host->host_no;
1953 host_dev->dev = device;
1954 host_dev->host = host;
1957 stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
1963 stor_device->destroy = false;
1964 init_waitqueue_head(&stor_device->waiting_to_drain);
1965 stor_device->device = device;
1966 stor_device->host = host;
1967 spin_lock_init(&stor_device->lock);
1968 hv_set_drvdata(device, stor_device);
1969 dma_set_min_align_mask(&device->device, HV_HYP_PAGE_SIZE - 1);
1971 stor_device->port_number = host->host_no;
1972 ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size, is_fc);
1976 host_dev->path = stor_device->path_id;
1977 host_dev->target = stor_device->target_id;
1979 switch (dev_id->driver_data) {
1981 host->max_lun = STORVSC_FC_MAX_LUNS_PER_TARGET;
1982 host->max_id = STORVSC_FC_MAX_TARGETS;
1983 host->max_channel = STORVSC_FC_MAX_CHANNELS - 1;
1984 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1985 host->transportt = fc_transport_template;
1990 host->max_lun = STORVSC_MAX_LUNS_PER_TARGET;
1991 host->max_id = STORVSC_MAX_TARGETS;
1992 host->max_channel = STORVSC_MAX_CHANNELS - 1;
1996 host->max_lun = STORVSC_IDE_MAX_LUNS_PER_TARGET;
1997 host->max_id = STORVSC_IDE_MAX_TARGETS;
1998 host->max_channel = STORVSC_IDE_MAX_CHANNELS - 1;
2001 /* max cmd length */
2002 host->max_cmd_len = STORVSC_MAX_CMD_LEN;
2004 * Any reasonable Hyper-V configuration should provide
2005 * max_transfer_bytes value aligning to HV_HYP_PAGE_SIZE,
2006 * protecting it from any weird value.
2008 max_xfer_bytes = round_down(stor_device->max_transfer_bytes, HV_HYP_PAGE_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);