1	// SPDX-License-Identifier: GPL-2.0-only
2	/*
3	* Copyright (c) 2009, Microsoft Corporation.
4	*
5	* Authors:
6	* Haiyang Zhang <haiyangz@microsoft.com>
7	* Hank Janssen <hjanssen@microsoft.com>
8	* K. Y. Srinivasan <kys@microsoft.com>
9	*/
10
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>
16	#include <linux/mm.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>
25
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>
36
37	/*
38	* All wire protocol details (storage protocol between the guest and the host)
39	* are consolidated here.
40	*
41	* Begin protocol definitions.
42	*/
43
44	/*
45	* Version history:
46	* V1 Beta: 0.1
47	* V1 RC < 2008/1/31: 1.0
48	* V1 RC > 2008/1/31: 2.0
49	* Win7: 4.2
50	* Win8: 5.1
51	* Win8.1: 6.0
52	* Win10: 6.2
53	*/
54
55	#define VMSTOR_PROTO_VERSION(MAJOR_, MINOR_) ((((MAJOR_) & 0xff) << 8) | \
56	(((MINOR_) & 0xff)))
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)
62
63	/* channel callback timeout in ms */
64	#define CALLBACK_TIMEOUT 2
65
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
82	};
83
84	/*
85	* WWN packet for Fibre Channel HBA
86	*/
87
88	struct hv_fc_wwn_packet {
89	u8 primary_active;
90	u8 reserved1[3];
91	u8 primary_port_wwn[8];
92	u8 primary_node_wwn[8];
93	u8 secondary_port_wwn[8];
94	u8 secondary_node_wwn[8];
95	};
96
97
98
99	/*
100	* SRB Flag Bits
101	*/
102
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
115
116	/*
117	* This flag indicates the request is part of the workflow for processing a D3.
118	*/
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
130
131	#define SP_UNTAGGED ((unsigned char) ~0)
132	#define SRB_SIMPLE_TAG_REQUEST 0x20
133
134	/*
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
138	*/
139	#define STORVSC_MAX_CMD_LEN 0x10
140
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
144
145	/*
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.
149	*/
150	static int vmstor_proto_version;
151
152	static bool hv_dev_is_fc(struct hv_device *hv_dev);
153
154	#define STORVSC_LOGGING_NONE 0
155	#define STORVSC_LOGGING_ERROR 1
156	#define STORVSC_LOGGING_WARN 2
157
158	static int logging_level = STORVSC_LOGGING_ERROR;
159	module_param(logging_level, int, S_IRUGO|S_IWUSR);
160	MODULE_PARM_DESC(logging_level,
161	"Logging level, 0 - None, 1 - Error (default), 2 - Warning.");
162
163	static inline bool do_logging(int level)
164	{
165	return logging_level >= level;
166	}
167
168	#define storvsc_log(dev, level, fmt, ...) \
169	do { \
170	if (do_logging(level)) \
171	dev_warn(&(dev)->device, fmt, ##__VA_ARGS__); \
172	} while (0)
173
174	#define storvsc_log_ratelimited(dev, level, fmt, ...) \
175	do { \
176	if (do_logging(level)) \
177	dev_warn_ratelimited(&(dev)->device, fmt, ##__VA_ARGS__); \
178	} while (0)
179
180	struct vmscsi_request {
181	u16 length;
182	u8 srb_status;
183	u8 scsi_status;
184
185	u8 port_number;
186	u8 path_id;
187	u8 target_id;
188	u8 lun;
189
190	u8 cdb_length;
191	u8 sense_info_length;
192	u8 data_in;
193	u8 reserved;
194
195	u32 data_transfer_length;
196
197	union {
198	u8 cdb[STORVSC_MAX_CMD_LEN];
199	u8 sense_data[STORVSC_SENSE_BUFFER_SIZE];
200	u8 reserved_array[STORVSC_MAX_BUF_LEN_WITH_PADDING];
201	};
202	/*
203	* The following was added in win8.
204	*/
205	u16 reserve;
206	u8 queue_tag;
207	u8 queue_action;
208	u32 srb_flags;
209	u32 time_out_value;
210	u32 queue_sort_ey;
211
212	} __attribute((packed));
213
214	/*
215	* The list of windows version in order of preference.
216	*/
217
218	static const int protocol_version[] = {
219	VMSTOR_PROTO_VERSION_WIN10,
220	VMSTOR_PROTO_VERSION_WIN8_1,
221	VMSTOR_PROTO_VERSION_WIN8,
222	};
223
224
225	/*
226	* This structure is sent during the initialization phase to get the different
227	* properties of the channel.
228	*/
229
230	#define STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL 0x1
231
232	struct vmstorage_channel_properties {
233	u32 reserved;
234	u16 max_channel_cnt;
235	u16 reserved1;
236
237	u32 flags;
238	u32 max_transfer_bytes;
239
240	u64 reserved2;
241	} __packed;
242
243	/* This structure is sent during the storage protocol negotiations. */
244	struct vmstorage_protocol_version {
245	/* Major (MSW) and minor (LSW) version numbers. */
246	u16 major_minor;
247
248	/*
249	* Revision number is auto-incremented whenever this file is changed
250	* (See FILL_VMSTOR_REVISION macro above). Mismatch does not
251	* definitely indicate incompatibility--but it does indicate mismatched
252	* builds.
253	* This is only used on the windows side. Just set it to 0.
254	*/
255	u16 revision;
256	} __packed;
257
258	/* Channel Property Flags */
259	#define STORAGE_CHANNEL_REMOVABLE_FLAG 0x1
260	#define STORAGE_CHANNEL_EMULATED_IDE_FLAG 0x2
261
262	struct vstor_packet {
263	/* Requested operation type */
264	enum vstor_packet_operation operation;
265
266	/* Flags - see below for values */
267	u32 flags;
268
269	/* Status of the request returned from the server side. */
270	u32 status;
271
272	/* Data payload area */
273	union {
274	/*
275	* Structure used to forward SCSI commands from the
276	* client to the server.
277	*/
278	struct vmscsi_request vm_srb;
279
280	/* Structure used to query channel properties. */
281	struct vmstorage_channel_properties storage_channel_properties;
282
283	/* Used during version negotiations. */
284	struct vmstorage_protocol_version version;
285
286	/* Fibre channel address packet */
287	struct hv_fc_wwn_packet wwn_packet;
288
289	/* Number of sub-channels to create */
290	u16 sub_channel_count;
291
292	/* This will be the maximum of the union members */
293	u8 buffer[0x34];
294	};
295	} __packed;
296
297	/*
298	* Packet Flags:
299	*
300	* This flag indicates that the server should send back a completion for this
301	* packet.
302	*/
303
304	#define REQUEST_COMPLETION_FLAG 0x1
305
306	/* Matches Windows-end */
307	enum storvsc_request_type {
308	WRITE_TYPE = 0,
309	READ_TYPE,
310	UNKNOWN_TYPE,
311	};
312
313	/*
314	* SRB status codes and masks. In the 8-bit field, the two high order bits
315	* are flags, while the remaining 6 bits are an integer status code. The
316	* definitions here include only the subset of the integer status codes that
317	* are tested for in this driver.
318	*/
319	#define SRB_STATUS_AUTOSENSE_VALID 0x80
320	#define SRB_STATUS_QUEUE_FROZEN 0x40
321
322	/* SRB status integer codes */
323	#define SRB_STATUS_SUCCESS 0x01
324	#define SRB_STATUS_ABORTED 0x02
325	#define SRB_STATUS_ERROR 0x04
326	#define SRB_STATUS_INVALID_REQUEST 0x06
327	#define SRB_STATUS_TIMEOUT 0x09
328	#define SRB_STATUS_SELECTION_TIMEOUT 0x0A
329	#define SRB_STATUS_BUS_RESET 0x0E
330	#define SRB_STATUS_DATA_OVERRUN 0x12
331	#define SRB_STATUS_INVALID_LUN 0x20
332	#define SRB_STATUS_INTERNAL_ERROR 0x30
333
334	#define SRB_STATUS(status) \
335	(status & ~(SRB_STATUS_AUTOSENSE_VALID | SRB_STATUS_QUEUE_FROZEN))
336	/*
337	* This is the end of Protocol specific defines.
338	*/
339
340	static int storvsc_ringbuffer_size = (128 * 1024);
341	static int aligned_ringbuffer_size;
342	static u32 max_outstanding_req_per_channel;
343	static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth);
344
345	static int storvsc_vcpus_per_sub_channel = 4;
346	static unsigned int storvsc_max_hw_queues;
347
348	module_param(storvsc_ringbuffer_size, int, S_IRUGO);
349	MODULE_PARM_DESC(storvsc_ringbuffer_size, "Ring buffer size (bytes)");
350
351	module_param(storvsc_max_hw_queues, uint, 0644);
352	MODULE_PARM_DESC(storvsc_max_hw_queues, "Maximum number of hardware queues");
353
354	module_param(storvsc_vcpus_per_sub_channel, int, S_IRUGO);
355	MODULE_PARM_DESC(storvsc_vcpus_per_sub_channel, "Ratio of VCPUs to subchannels");
356
357	static int ring_avail_percent_lowater = 10;
358	module_param(ring_avail_percent_lowater, int, S_IRUGO);
359	MODULE_PARM_DESC(ring_avail_percent_lowater,
360	"Select a channel if available ring size > this in percent");
361
362	/*
363	* Timeout in seconds for all devices managed by this driver.
364	*/
365	static const int storvsc_timeout = 180;
366
367	#if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
368	static struct scsi_transport_template *fc_transport_template;
369	#endif
370
371	static struct scsi_host_template scsi_driver;
372	static void storvsc_on_channel_callback(void *context);
373
374	#define STORVSC_MAX_LUNS_PER_TARGET 255
375	#define STORVSC_MAX_TARGETS 2
376	#define STORVSC_MAX_CHANNELS 8
377
378	#define STORVSC_FC_MAX_LUNS_PER_TARGET 255
379	#define STORVSC_FC_MAX_TARGETS 128
380	#define STORVSC_FC_MAX_CHANNELS 8
381	#define STORVSC_FC_MAX_XFER_SIZE ((u32)(512 * 1024))
382
383	#define STORVSC_IDE_MAX_LUNS_PER_TARGET 64
384	#define STORVSC_IDE_MAX_TARGETS 1
385	#define STORVSC_IDE_MAX_CHANNELS 1
386
387	/*
388	* Upper bound on the size of a storvsc packet.
389	*/
390	#define STORVSC_MAX_PKT_SIZE (sizeof(struct vmpacket_descriptor) +\
391	sizeof(struct vstor_packet))
392
393	struct storvsc_cmd_request {
394	struct scsi_cmnd *cmd;
395
396	struct hv_device *device;
397
398	/* Synchronize the request/response if needed */
399	struct completion wait_event;
400
401	struct vmbus_channel_packet_multipage_buffer mpb;
402	struct vmbus_packet_mpb_array *payload;
403	u32 payload_sz;
404
405	struct vstor_packet vstor_packet;
406	};
407
408
409	/* A storvsc device is a device object that contains a vmbus channel */
410	struct storvsc_device {
411	struct hv_device *device;
412
413	bool destroy;
414	bool drain_notify;
415	atomic_t num_outstanding_req;
416	struct Scsi_Host *host;
417
418	wait_queue_head_t waiting_to_drain;
419
420	/*
421	* Each unique Port/Path/Target represents 1 channel ie scsi
422	* controller. In reality, the pathid, targetid is always 0
423	* and the port is set by us
424	*/
425	unsigned int port_number;
426	unsigned char path_id;
427	unsigned char target_id;
428
429	/*
430	* Max I/O, the device can support.
431	*/
432	u32 max_transfer_bytes;
433	/*
434	* Number of sub-channels we will open.
435	*/
436	u16 num_sc;
437	struct vmbus_channel **stor_chns;
438	/*
439	* Mask of CPUs bound to subchannels.
440	*/
441	struct cpumask alloced_cpus;
442	/*
443	* Serializes modifications of stor_chns[] from storvsc_do_io()
444	* and storvsc_change_target_cpu().
445	*/
446	spinlock_t lock;
447	/* Used for vsc/vsp channel reset process */
448	struct storvsc_cmd_request init_request;
449	struct storvsc_cmd_request reset_request;
450	/*
451	* Currently active port and node names for FC devices.
452	*/
453	u64 node_name;
454	u64 port_name;
455	#if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
456	struct fc_rport *rport;
457	#endif
458	};
459
460	struct hv_host_device {
461	struct hv_device *dev;
462	unsigned int port;
463	unsigned char path;
464	unsigned char target;
465	struct workqueue_struct *handle_error_wq;
466	struct work_struct host_scan_work;
467	struct Scsi_Host *host;
468	};
469
470	struct storvsc_scan_work {
471	struct work_struct work;
472	struct Scsi_Host *host;
473	u8 lun;
474	u8 tgt_id;
475	};
476
477	static void storvsc_device_scan(struct work_struct *work)
478	{
479	struct storvsc_scan_work *wrk;
480	struct scsi_device *sdev;
481
482	wrk = container_of(work, struct storvsc_scan_work, work);
483
484	sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
485	if (!sdev)
486	goto done;
487	scsi_rescan_device(sdev);
488	scsi_device_put(sdev);
489
490	done:
491	kfree(objp: wrk);
492	}
493
494	static void storvsc_host_scan(struct work_struct *work)
495	{
496	struct Scsi_Host *host;
497	struct scsi_device *sdev;
498	struct hv_host_device *host_device =
499	container_of(work, struct hv_host_device, host_scan_work);
500
501	host = host_device->host;
502	/*
503	* Before scanning the host, first check to see if any of the
504	* currently known devices have been hot removed. We issue a
505	* "unit ready" command against all currently known devices.
506	* This I/O will result in an error for devices that have been
507	* removed. As part of handling the I/O error, we remove the device.
508	*
509	* When a LUN is added or removed, the host sends us a signal to
510	* scan the host. Thus we are forced to discover the LUNs that
511	* may have been removed this way.
512	*/
513	mutex_lock(&host->scan_mutex);
514	shost_for_each_device(sdev, host)
515	scsi_test_unit_ready(sdev, timeout: 1, retries: 1, NULL);
516	mutex_unlock(lock: &host->scan_mutex);
517	/*
518	* Now scan the host to discover LUNs that may have been added.
519	*/
520	scsi_scan_host(host);
521	}
522
523	static void storvsc_remove_lun(struct work_struct *work)
524	{
525	struct storvsc_scan_work *wrk;
526	struct scsi_device *sdev;
527
528	wrk = container_of(work, struct storvsc_scan_work, work);
529	if (!scsi_host_get(wrk->host))
530	goto done;
531
532	sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun);
533
534	if (sdev) {
535	scsi_remove_device(sdev);
536	scsi_device_put(sdev);
537	}
538	scsi_host_put(t: wrk->host);
539
540	done:
541	kfree(objp: wrk);
542	}
543
544
545	/*
546	* We can get incoming messages from the host that are not in response to
547	* messages that we have sent out. An example of this would be messages
548	* received by the guest to notify dynamic addition/removal of LUNs. To
549	* deal with potential race conditions where the driver may be in the
550	* midst of being unloaded when we might receive an unsolicited message
551	* from the host, we have implemented a mechanism to gurantee sequential
552	* consistency:
553	*
554	* 1) Once the device is marked as being destroyed, we will fail all
555	* outgoing messages.
556	* 2) We permit incoming messages when the device is being destroyed,
557	* only to properly account for messages already sent out.
558	*/
559
560	static inline struct storvsc_device *get_out_stor_device(
561	struct hv_device *device)
562	{
563	struct storvsc_device *stor_device;
564
565	stor_device = hv_get_drvdata(dev: device);
566
567	if (stor_device && stor_device->destroy)
568	stor_device = NULL;
569
570	return stor_device;
571	}
572
573
574	static inline void storvsc_wait_to_drain(struct storvsc_device *dev)
575	{
576	dev->drain_notify = true;
577	wait_event(dev->waiting_to_drain,
578	atomic_read(&dev->num_outstanding_req) == 0);
579	dev->drain_notify = false;
580	}
581
582	static inline struct storvsc_device *get_in_stor_device(
583	struct hv_device *device)
584	{
585	struct storvsc_device *stor_device;
586
587	stor_device = hv_get_drvdata(dev: device);
588
589	if (!stor_device)
590	goto get_in_err;
591
592	/*
593	* If the device is being destroyed; allow incoming
594	* traffic only to cleanup outstanding requests.
595	*/
596
597	if (stor_device->destroy &&
598	(atomic_read(v: &stor_device->num_outstanding_req) == 0))
599	stor_device = NULL;
600
601	get_in_err:
602	return stor_device;
603
604	}
605
606	static void storvsc_change_target_cpu(struct vmbus_channel *channel, u32 old,
607	u32 new)
608	{
609	struct storvsc_device *stor_device;
610	struct vmbus_channel *cur_chn;
611	bool old_is_alloced = false;
612	struct hv_device *device;
613	unsigned long flags;
614	int cpu;
615
616	device = channel->primary_channel ?
617	channel->primary_channel->device_obj
618	: channel->device_obj;
619	stor_device = get_out_stor_device(device);
620	if (!stor_device)
621	return;
622
623	/* See storvsc_do_io() -> get_og_chn(). */
624	spin_lock_irqsave(&stor_device->lock, flags);
625
626	/*
627	* Determines if the storvsc device has other channels assigned to
628	* the "old" CPU to update the alloced_cpus mask and the stor_chns
629	* array.
630	*/
631	if (device->channel != channel && device->channel->target_cpu == old) {
632	cur_chn = device->channel;
633	old_is_alloced = true;
634	goto old_is_alloced;
635	}
636	list_for_each_entry(cur_chn, &device->channel->sc_list, sc_list) {
637	if (cur_chn == channel)
638	continue;
639	if (cur_chn->target_cpu == old) {
640	old_is_alloced = true;
641	goto old_is_alloced;
642	}
643	}
644
645	old_is_alloced:
646	if (old_is_alloced)
647	WRITE_ONCE(stor_device->stor_chns[old], cur_chn);
648	else
649	cpumask_clear_cpu(cpu: old, dstp: &stor_device->alloced_cpus);
650
651	/* "Flush" the stor_chns array. */
652	for_each_possible_cpu(cpu) {
653	if (stor_device->stor_chns[cpu] && !cpumask_test_cpu(
654	cpu, cpumask: &stor_device->alloced_cpus))
655	WRITE_ONCE(stor_device->stor_chns[cpu], NULL);
656	}
657
658	WRITE_ONCE(stor_device->stor_chns[new], channel);
659	cpumask_set_cpu(cpu: new, dstp: &stor_device->alloced_cpus);
660
661	spin_unlock_irqrestore(lock: &stor_device->lock, flags);
662	}
663
664	static u64 storvsc_next_request_id(struct vmbus_channel *channel, u64 rqst_addr)
665	{
666	struct storvsc_cmd_request *request =
667	(struct storvsc_cmd_request *)(unsigned long)rqst_addr;
668
669	if (rqst_addr == VMBUS_RQST_INIT)
670	return VMBUS_RQST_INIT;
671	if (rqst_addr == VMBUS_RQST_RESET)
672	return VMBUS_RQST_RESET;
673
674	/*
675	* Cannot return an ID of 0, which is reserved for an unsolicited
676	* message from Hyper-V.
677	*/
678	return (u64)blk_mq_unique_tag(rq: scsi_cmd_to_rq(scmd: request->cmd)) + 1;
679	}
680
681	static void handle_sc_creation(struct vmbus_channel *new_sc)
682	{
683	struct hv_device *device = new_sc->primary_channel->device_obj;
684	struct device *dev = &device->device;
685	struct storvsc_device *stor_device;
686	struct vmstorage_channel_properties props;
687	int ret;
688
689	stor_device = get_out_stor_device(device);
690	if (!stor_device)
691	return;
692
693	memset(&props, 0, sizeof(struct vmstorage_channel_properties));
694	new_sc->max_pkt_size = STORVSC_MAX_PKT_SIZE;
695
696	new_sc->next_request_id_callback = storvsc_next_request_id;
697
698	ret = vmbus_open(channel: new_sc,
699	send_ringbuffersize: aligned_ringbuffer_size,
700	recv_ringbuffersize: aligned_ringbuffer_size,
701	userdata: (void *)&props,
702	userdatalen: sizeof(struct vmstorage_channel_properties),
703	onchannel_callback: storvsc_on_channel_callback, context: new_sc);
704
705	/* In case vmbus_open() fails, we don't use the sub-channel. */
706	if (ret != 0) {
707	dev_err(dev, "Failed to open sub-channel: err=%d\n", ret);
708	return;
709	}
710
711	new_sc->change_target_cpu_callback = storvsc_change_target_cpu;
712
713	/* Add the sub-channel to the array of available channels. */
714	stor_device->stor_chns[new_sc->target_cpu] = new_sc;
715	cpumask_set_cpu(cpu: new_sc->target_cpu, dstp: &stor_device->alloced_cpus);
716	}
717
718	static void handle_multichannel_storage(struct hv_device *device, int max_chns)
719	{
720	struct device *dev = &device->device;
721	struct storvsc_device *stor_device;
722	int num_sc;
723	struct storvsc_cmd_request *request;
724	struct vstor_packet *vstor_packet;
725	int ret, t;
726
727	/*
728	* If the number of CPUs is artificially restricted, such as
729	* with maxcpus=1 on the kernel boot line, Hyper-V could offer
730	* sub-channels >= the number of CPUs. These sub-channels
731	* should not be created. The primary channel is already created
732	* and assigned to one CPU, so check against # CPUs - 1.
733	*/
734	num_sc = min((int)(num_online_cpus() - 1), max_chns);
735	if (!num_sc)
736	return;
737
738	stor_device = get_out_stor_device(device);
739	if (!stor_device)
740	return;
741
742	stor_device->num_sc = num_sc;
743	request = &stor_device->init_request;
744	vstor_packet = &request->vstor_packet;
745
746	/*
747	* Establish a handler for dealing with subchannels.
748	*/
749	vmbus_set_sc_create_callback(primary_channel: device->channel, sc_cr_cb: handle_sc_creation);
750
751	/*
752	* Request the host to create sub-channels.
753	*/
754	memset(request, 0, sizeof(struct storvsc_cmd_request));
755	init_completion(x: &request->wait_event);
756	vstor_packet->operation = VSTOR_OPERATION_CREATE_SUB_CHANNELS;
757	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
758	vstor_packet->sub_channel_count = num_sc;
759
760	ret = vmbus_sendpacket(channel: device->channel, buffer: vstor_packet,
761	bufferLen: sizeof(struct vstor_packet),
762	VMBUS_RQST_INIT,
763	type: VM_PKT_DATA_INBAND,
764	VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
765
766	if (ret != 0) {
767	dev_err(dev, "Failed to create sub-channel: err=%d\n", ret);
768	return;
769	}
770
771	t = wait_for_completion_timeout(x: &request->wait_event, timeout: storvsc_timeout * HZ);
772	if (t == 0) {
773	dev_err(dev, "Failed to create sub-channel: timed out\n");
774	return;
775	}
776
777	if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
778	vstor_packet->status != 0) {
779	dev_err(dev, "Failed to create sub-channel: op=%d, host=0x%x\n",
780	vstor_packet->operation, vstor_packet->status);
781	return;
782	}
783
784	/*
785	* We need to do nothing here, because vmbus_process_offer()
786	* invokes channel->sc_creation_callback, which will open and use
787	* the sub-channel(s).
788	*/
789	}
790
791	static void cache_wwn(struct storvsc_device *stor_device,
792	struct vstor_packet *vstor_packet)
793	{
794	/*
795	* Cache the currently active port and node ww names.
796	*/
797	if (vstor_packet->wwn_packet.primary_active) {
798	stor_device->node_name =
799	wwn_to_u64(wwn: vstor_packet->wwn_packet.primary_node_wwn);
800	stor_device->port_name =
801	wwn_to_u64(wwn: vstor_packet->wwn_packet.primary_port_wwn);
802	} else {
803	stor_device->node_name =
804	wwn_to_u64(wwn: vstor_packet->wwn_packet.secondary_node_wwn);
805	stor_device->port_name =
806	wwn_to_u64(wwn: vstor_packet->wwn_packet.secondary_port_wwn);
807	}
808	}
809
810
811	static int storvsc_execute_vstor_op(struct hv_device *device,
812	struct storvsc_cmd_request *request,
813	bool status_check)
814	{
815	struct storvsc_device *stor_device;
816	struct vstor_packet *vstor_packet;
817	int ret, t;
818
819	stor_device = get_out_stor_device(device);
820	if (!stor_device)
821	return -ENODEV;
822
823	vstor_packet = &request->vstor_packet;
824
825	init_completion(x: &request->wait_event);
826	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
827
828	ret = vmbus_sendpacket(channel: device->channel, buffer: vstor_packet,
829	bufferLen: sizeof(struct vstor_packet),
830	VMBUS_RQST_INIT,
831	type: VM_PKT_DATA_INBAND,
832	VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
833	if (ret != 0)
834	return ret;
835
836	t = wait_for_completion_timeout(x: &request->wait_event, timeout: storvsc_timeout * HZ);
837	if (t == 0)
838	return -ETIMEDOUT;
839
840	if (!status_check)
841	return ret;
842
843	if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
844	vstor_packet->status != 0)
845	return -EINVAL;
846
847	return ret;
848	}
849
850	static int storvsc_channel_init(struct hv_device *device, bool is_fc)
851	{
852	struct storvsc_device *stor_device;
853	struct storvsc_cmd_request *request;
854	struct vstor_packet *vstor_packet;
855	int ret, i;
856	int max_chns;
857	bool process_sub_channels = false;
858
859	stor_device = get_out_stor_device(device);
860	if (!stor_device)
861	return -ENODEV;
862
863	request = &stor_device->init_request;
864	vstor_packet = &request->vstor_packet;
865
866	/*
867	* Now, initiate the vsc/vsp initialization protocol on the open
868	* channel
869	*/
870	memset(request, 0, sizeof(struct storvsc_cmd_request));
871	vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION;
872	ret = storvsc_execute_vstor_op(device, request, status_check: true);
873	if (ret)
874	return ret;
875	/*
876	* Query host supported protocol version.
877	*/
878
879	for (i = 0; i < ARRAY_SIZE(protocol_version); i++) {
880	/* reuse the packet for version range supported */
881	memset(vstor_packet, 0, sizeof(struct vstor_packet));
882	vstor_packet->operation =
883	VSTOR_OPERATION_QUERY_PROTOCOL_VERSION;
884
885	vstor_packet->version.major_minor = protocol_version[i];
886
887	/*
888	* The revision number is only used in Windows; set it to 0.
889	*/
890	vstor_packet->version.revision = 0;
891	ret = storvsc_execute_vstor_op(device, request, status_check: false);
892	if (ret != 0)
893	return ret;
894
895	if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO)
896	return -EINVAL;
897
898	if (vstor_packet->status == 0) {
899	vmstor_proto_version = protocol_version[i];
900
901	break;
902	}
903	}
904
905	if (vstor_packet->status != 0) {
906	dev_err(&device->device, "Obsolete Hyper-V version\n");
907	return -EINVAL;
908	}
909
910
911	memset(vstor_packet, 0, sizeof(struct vstor_packet));
912	vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES;
913	ret = storvsc_execute_vstor_op(device, request, status_check: true);
914	if (ret != 0)
915	return ret;
916
917	/*
918	* Check to see if multi-channel support is there.
919	* Hosts that implement protocol version of 5.1 and above
920	* support multi-channel.
921	*/
922	max_chns = vstor_packet->storage_channel_properties.max_channel_cnt;
923
924	/*
925	* Allocate state to manage the sub-channels.
926	* We allocate an array based on the number of CPU ids. This array
927	* is initially sparsely populated for the CPUs assigned to channels:
928	* primary + sub-channels. As I/Os are initiated by different CPUs,
929	* the slots for all online CPUs are populated to evenly distribute
930	* the load across all channels.
931	*/
932	stor_device->stor_chns = kcalloc(nr_cpu_ids, sizeof(void *),
933	GFP_KERNEL);
934	if (stor_device->stor_chns == NULL)
935	return -ENOMEM;
936
937	device->channel->change_target_cpu_callback = storvsc_change_target_cpu;
938
939	stor_device->stor_chns[device->channel->target_cpu] = device->channel;
940	cpumask_set_cpu(cpu: device->channel->target_cpu,
941	dstp: &stor_device->alloced_cpus);
942
943	if (vstor_packet->storage_channel_properties.flags &
944	STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL)
945	process_sub_channels = true;
946
947	stor_device->max_transfer_bytes =
948	vstor_packet->storage_channel_properties.max_transfer_bytes;
949
950	if (!is_fc)
951	goto done;
952
953	/*
954	* For FC devices retrieve FC HBA data.
955	*/
956	memset(vstor_packet, 0, sizeof(struct vstor_packet));
957	vstor_packet->operation = VSTOR_OPERATION_FCHBA_DATA;
958	ret = storvsc_execute_vstor_op(device, request, status_check: true);
959	if (ret != 0)
960	return ret;
961
962	/*
963	* Cache the currently active port and node ww names.
964	*/
965	cache_wwn(stor_device, vstor_packet);
966
967	done:
968
969	memset(vstor_packet, 0, sizeof(struct vstor_packet));
970	vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION;
971	ret = storvsc_execute_vstor_op(device, request, status_check: true);
972	if (ret != 0)
973	return ret;
974
975	if (process_sub_channels)
976	handle_multichannel_storage(device, max_chns);
977
978	return ret;
979	}
980
981	static void storvsc_handle_error(struct vmscsi_request *vm_srb,
982	struct scsi_cmnd *scmnd,
983	struct Scsi_Host *host,
984	u8 asc, u8 ascq)
985	{
986	struct storvsc_scan_work *wrk;
987	void (*process_err_fn)(struct work_struct *work);
988	struct hv_host_device *host_dev = shost_priv(shost: host);
989
990	switch (SRB_STATUS(vm_srb->srb_status)) {
991	case SRB_STATUS_ERROR:
992	case SRB_STATUS_ABORTED:
993	case SRB_STATUS_INVALID_REQUEST:
994	case SRB_STATUS_INTERNAL_ERROR:
995	case SRB_STATUS_TIMEOUT:
996	case SRB_STATUS_SELECTION_TIMEOUT:
997	case SRB_STATUS_BUS_RESET:
998	case SRB_STATUS_DATA_OVERRUN:
999	if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID) {
1000	/* Check for capacity change */
1001	if ((asc == 0x2a) && (ascq == 0x9)) {
1002	process_err_fn = storvsc_device_scan;
1003	/* Retry the I/O that triggered this. */
1004	set_host_byte(cmd: scmnd, status: DID_REQUEUE);
1005	goto do_work;
1006	}
1007
1008	/*
1009	* Check for "Operating parameters have changed"
1010	* due to Hyper-V changing the VHD/VHDX BlockSize
1011	* when adding/removing a differencing disk. This
1012	* causes discard_granularity to change, so do a
1013	* rescan to pick up the new granularity. We don't
1014	* want scsi_report_sense() to output a message
1015	* that a sysadmin wouldn't know what to do with.
1016	*/
1017	if ((asc == 0x3f) && (ascq != 0x03) &&
1018	(ascq != 0x0e)) {
1019	process_err_fn = storvsc_device_scan;
1020	set_host_byte(cmd: scmnd, status: DID_REQUEUE);
1021	goto do_work;
1022	}
1023
1024	/*
1025	* Otherwise, let upper layer deal with the
1026	* error when sense message is present
1027	*/
1028	return;
1029	}
1030
1031	/*
1032	* If there is an error; offline the device since all
1033	* error recovery strategies would have already been
1034	* deployed on the host side. However, if the command
1035	* were a pass-through command deal with it appropriately.
1036	*/
1037	switch (scmnd->cmnd[0]) {
1038	case ATA_16:
1039	case ATA_12:
1040	set_host_byte(cmd: scmnd, status: DID_PASSTHROUGH);
1041	break;
1042	/*
1043	* On some Hyper-V hosts TEST_UNIT_READY command can
1044	* return SRB_STATUS_ERROR. Let the upper level code
1045	* deal with it based on the sense information.
1046	*/
1047	case TEST_UNIT_READY:
1048	break;
1049	default:
1050	set_host_byte(cmd: scmnd, status: DID_ERROR);
1051	}
1052	return;
1053
1054	case SRB_STATUS_INVALID_LUN:
1055	set_host_byte(cmd: scmnd, status: DID_NO_CONNECT);
1056	process_err_fn = storvsc_remove_lun;
1057	goto do_work;
1058
1059	}
1060	return;
1061
1062	do_work:
1063	/*
1064	* We need to schedule work to process this error; schedule it.
1065	*/
1066	wrk = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC);
1067	if (!wrk) {
1068	set_host_byte(cmd: scmnd, status: DID_BAD_TARGET);
1069	return;
1070	}
1071
1072	wrk->host = host;
1073	wrk->lun = vm_srb->lun;
1074	wrk->tgt_id = vm_srb->target_id;
1075	INIT_WORK(&wrk->work, process_err_fn);
1076	queue_work(wq: host_dev->handle_error_wq, work: &wrk->work);
1077	}
1078
1079
1080	static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request,
1081	struct storvsc_device *stor_dev)
1082	{
1083	struct scsi_cmnd *scmnd = cmd_request->cmd;
1084	struct scsi_sense_hdr sense_hdr;
1085	struct vmscsi_request *vm_srb;
1086	u32 data_transfer_length;
1087	struct Scsi_Host *host;
1088	u32 payload_sz = cmd_request->payload_sz;
1089	void *payload = cmd_request->payload;
1090	bool sense_ok;
1091
1092	host = stor_dev->host;
1093
1094	vm_srb = &cmd_request->vstor_packet.vm_srb;
1095	data_transfer_length = vm_srb->data_transfer_length;
1096
1097	scmnd->result = vm_srb->scsi_status;
1098
1099	if (scmnd->result) {
1100	sense_ok = scsi_normalize_sense(sense_buffer: scmnd->sense_buffer,
1101	SCSI_SENSE_BUFFERSIZE, sshdr: &sense_hdr);
1102
1103	if (sense_ok && do_logging(STORVSC_LOGGING_WARN))
1104	scsi_print_sense_hdr(scmnd->device, "storvsc",
1105	&sense_hdr);
1106	}
1107
1108	if (vm_srb->srb_status != SRB_STATUS_SUCCESS) {
1109	storvsc_handle_error(vm_srb, scmnd, host, asc: sense_hdr.asc,
1110	ascq: sense_hdr.ascq);
1111	/*
1112	* The Windows driver set data_transfer_length on
1113	* SRB_STATUS_DATA_OVERRUN. On other errors, this value
1114	* is untouched. In these cases we set it to 0.
1115	*/
1116	if (vm_srb->srb_status != SRB_STATUS_DATA_OVERRUN)
1117	data_transfer_length = 0;
1118	}
1119
1120	/* Validate data_transfer_length (from Hyper-V) */
1121	if (data_transfer_length > cmd_request->payload->range.len)
1122	data_transfer_length = cmd_request->payload->range.len;
1123
1124	scsi_set_resid(cmd: scmnd,
1125	resid: cmd_request->payload->range.len - data_transfer_length);
1126
1127	scsi_done(cmd: scmnd);
1128
1129	if (payload_sz >
1130	sizeof(struct vmbus_channel_packet_multipage_buffer))
1131	kfree(objp: payload);
1132	}
1133
1134	static void storvsc_on_io_completion(struct storvsc_device *stor_device,
1135	struct vstor_packet *vstor_packet,
1136	struct storvsc_cmd_request *request)
1137	{
1138	struct vstor_packet *stor_pkt;
1139	struct hv_device *device = stor_device->device;
1140
1141	stor_pkt = &request->vstor_packet;
1142
1143	/*
1144	* The current SCSI handling on the host side does
1145	* not correctly handle:
1146	* INQUIRY command with page code parameter set to 0x80
1147	* MODE_SENSE and MODE_SENSE_10 command with cmd[2] == 0x1c
1148	* MAINTENANCE_IN is not supported by HyperV FC passthrough
1149	*
1150	* Setup srb and scsi status so this won't be fatal.
1151	* We do this so we can distinguish truly fatal failues
1152	* (srb status == 0x4) and off-line the device in that case.
1153	*/
1154
1155	if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) ||
1156	(stor_pkt->vm_srb.cdb[0] == MODE_SENSE) ||
1157	(stor_pkt->vm_srb.cdb[0] == MODE_SENSE_10) ||
1158	(stor_pkt->vm_srb.cdb[0] == MAINTENANCE_IN &&
1159	hv_dev_is_fc(hv_dev: device))) {
1160	vstor_packet->vm_srb.scsi_status = 0;
1161	vstor_packet->vm_srb.srb_status = SRB_STATUS_SUCCESS;
1162	}
1163
1164	/* Copy over the status...etc */
1165	stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status;
1166	stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status;
1167
1168	/*
1169	* Copy over the sense_info_length, but limit to the known max
1170	* size if Hyper-V returns a bad value.
1171	*/
1172	stor_pkt->vm_srb.sense_info_length = min_t(u8, STORVSC_SENSE_BUFFER_SIZE,
1173	vstor_packet->vm_srb.sense_info_length);
1174
1175	if (vstor_packet->vm_srb.scsi_status != 0 ||
1176	vstor_packet->vm_srb.srb_status != SRB_STATUS_SUCCESS) {
1177
1178	/*
1179	* Log TEST_UNIT_READY errors only as warnings. Hyper-V can
1180	* return errors when detecting devices using TEST_UNIT_READY,
1181	* and logging these as errors produces unhelpful noise.
1182	*/
1183	int loglevel = (stor_pkt->vm_srb.cdb[0] == TEST_UNIT_READY) ?
1184	STORVSC_LOGGING_WARN : STORVSC_LOGGING_ERROR;
1185
1186	storvsc_log_ratelimited(device, loglevel,
1187	"tag#%d cmd 0x%x status: scsi 0x%x srb 0x%x host 0x%x\n",
1188	scsi_cmd_to_rq(request->cmd)->tag,
1189	stor_pkt->vm_srb.cdb[0],
1190	vstor_packet->vm_srb.scsi_status,
1191	vstor_packet->vm_srb.srb_status,
1192	vstor_packet->status);
1193	}
1194
1195	if (vstor_packet->vm_srb.scsi_status == SAM_STAT_CHECK_CONDITION &&
1196	(vstor_packet->vm_srb.srb_status & SRB_STATUS_AUTOSENSE_VALID))
1197	memcpy(request->cmd->sense_buffer,
1198	vstor_packet->vm_srb.sense_data,
1199	stor_pkt->vm_srb.sense_info_length);
1200
1201	stor_pkt->vm_srb.data_transfer_length =
1202	vstor_packet->vm_srb.data_transfer_length;
1203
1204	storvsc_command_completion(cmd_request: request, stor_dev: stor_device);
1205
1206	if (atomic_dec_and_test(v: &stor_device->num_outstanding_req) &&
1207	stor_device->drain_notify)
1208	wake_up(&stor_device->waiting_to_drain);
1209	}
1210
1211	static void storvsc_on_receive(struct storvsc_device *stor_device,
1212	struct vstor_packet *vstor_packet,
1213	struct storvsc_cmd_request *request)
1214	{
1215	struct hv_host_device *host_dev;
1216	switch (vstor_packet->operation) {
1217	case VSTOR_OPERATION_COMPLETE_IO:
1218	storvsc_on_io_completion(stor_device, vstor_packet, request);
1219	break;
1220
1221	case VSTOR_OPERATION_REMOVE_DEVICE:
1222	case VSTOR_OPERATION_ENUMERATE_BUS:
1223	host_dev = shost_priv(shost: stor_device->host);
1224	queue_work(
1225	wq: host_dev->handle_error_wq, work: &host_dev->host_scan_work);
1226	break;
1227
1228	case VSTOR_OPERATION_FCHBA_DATA:
1229	cache_wwn(stor_device, vstor_packet);
1230	#if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
1231	fc_host_node_name(stor_device->host) = stor_device->node_name;
1232	fc_host_port_name(stor_device->host) = stor_device->port_name;
1233	#endif
1234	break;
1235	default:
1236	break;
1237	}
1238	}
1239
1240	static void storvsc_on_channel_callback(void *context)
1241	{
1242	struct vmbus_channel *channel = (struct vmbus_channel *)context;
1243	const struct vmpacket_descriptor *desc;
1244	struct hv_device *device;
1245	struct storvsc_device *stor_device;
1246	struct Scsi_Host *shost;
1247	unsigned long time_limit = jiffies + msecs_to_jiffies(CALLBACK_TIMEOUT);
1248
1249	if (channel->primary_channel != NULL)
1250	device = channel->primary_channel->device_obj;
1251	else
1252	device = channel->device_obj;
1253
1254	stor_device = get_in_stor_device(device);
1255	if (!stor_device)
1256	return;
1257
1258	shost = stor_device->host;
1259
1260	foreach_vmbus_pkt(desc, channel) {
1261	struct vstor_packet *packet = hv_pkt_data(desc);
1262	struct storvsc_cmd_request *request = NULL;
1263	u32 pktlen = hv_pkt_datalen(desc);
1264	u64 rqst_id = desc->trans_id;
1265	u32 minlen = rqst_id ? sizeof(struct vstor_packet) :
1266	sizeof(enum vstor_packet_operation);
1267
1268	if (unlikely(time_after(jiffies, time_limit))) {
1269	hv_pkt_iter_close(channel);
1270	return;
1271	}
1272
1273	if (pktlen < minlen) {
1274	dev_err(&device->device,
1275	"Invalid pkt: id=%llu, len=%u, minlen=%u\n",
1276	rqst_id, pktlen, minlen);
1277	continue;
1278	}
1279
1280	if (rqst_id == VMBUS_RQST_INIT) {
1281	request = &stor_device->init_request;
1282	} else if (rqst_id == VMBUS_RQST_RESET) {
1283	request = &stor_device->reset_request;
1284	} else {
1285	/* Hyper-V can send an unsolicited message with ID of 0 */
1286	if (rqst_id == 0) {
1287	/*
1288	* storvsc_on_receive() looks at the vstor_packet in the message
1289	* from the ring buffer.
1290	*
1291	* - If the operation in the vstor_packet is COMPLETE_IO, then
1292	* we call storvsc_on_io_completion(), and dereference the
1293	* guest memory address. Make sure we don't call
1294	* storvsc_on_io_completion() with a guest memory address
1295	* that is zero if Hyper-V were to construct and send such
1296	* a bogus packet.
1297	*
1298	* - If the operation in the vstor_packet is FCHBA_DATA, then
1299	* we call cache_wwn(), and access the data payload area of
1300	* the packet (wwn_packet); however, there is no guarantee
1301	* that the packet is big enough to contain such area.
1302	* Future-proof the code by rejecting such a bogus packet.
1303	*/
1304	if (packet->operation == VSTOR_OPERATION_COMPLETE_IO ||
1305	packet->operation == VSTOR_OPERATION_FCHBA_DATA) {
1306	dev_err(&device->device, "Invalid packet with ID of 0\n");
1307	continue;
1308	}
1309	} else {
1310	struct scsi_cmnd *scmnd;
1311
1312	/* Transaction 'rqst_id' corresponds to tag 'rqst_id - 1' */
1313	scmnd = scsi_host_find_tag(shost, tag: rqst_id - 1);
1314	if (scmnd == NULL) {
1315	dev_err(&device->device, "Incorrect transaction ID\n");
1316	continue;
1317	}
1318	request = (struct storvsc_cmd_request *)scsi_cmd_priv(cmd: scmnd);
1319	scsi_dma_unmap(cmd: scmnd);
1320	}
1321
1322	storvsc_on_receive(stor_device, vstor_packet: packet, request);
1323	continue;
1324	}
1325
1326	memcpy(&request->vstor_packet, packet,
1327	sizeof(struct vstor_packet));
1328	complete(&request->wait_event);
1329	}
1330	}
1331
1332	static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size,
1333	bool is_fc)
1334	{
1335	struct vmstorage_channel_properties props;
1336	int ret;
1337
1338	memset(&props, 0, sizeof(struct vmstorage_channel_properties));
1339
1340	device->channel->max_pkt_size = STORVSC_MAX_PKT_SIZE;
1341	device->channel->next_request_id_callback = storvsc_next_request_id;
1342
1343	ret = vmbus_open(channel: device->channel,
1344	send_ringbuffersize: ring_size,
1345	recv_ringbuffersize: ring_size,
1346	userdata: (void *)&props,
1347	userdatalen: sizeof(struct vmstorage_channel_properties),
1348	onchannel_callback: storvsc_on_channel_callback, context: device->channel);
1349
1350	if (ret != 0)
1351	return ret;
1352
1353	ret = storvsc_channel_init(device, is_fc);
1354	if (ret)
1355	vmbus_close(channel: device->channel);
1356
1357	return ret;
1358	}
1359
1360	static int storvsc_dev_remove(struct hv_device *device)
1361	{
1362	struct storvsc_device *stor_device;
1363
1364	stor_device = hv_get_drvdata(dev: device);
1365
1366	stor_device->destroy = true;
1367
1368	/* Make sure flag is set before waiting */
1369	wmb();
1370
1371	/*
1372	* At this point, all outbound traffic should be disable. We
1373	* only allow inbound traffic (responses) to proceed so that
1374	* outstanding requests can be completed.
1375	*/
1376
1377	storvsc_wait_to_drain(dev: stor_device);
1378
1379	/*
1380	* Since we have already drained, we don't need to busy wait
1381	* as was done in final_release_stor_device()
1382	* Note that we cannot set the ext pointer to NULL until
1383	* we have drained - to drain the outgoing packets, we need to
1384	* allow incoming packets.
1385	*/
1386	hv_set_drvdata(dev: device, NULL);
1387
1388	/* Close the channel */
1389	vmbus_close(channel: device->channel);
1390
1391	kfree(objp: stor_device->stor_chns);
1392	kfree(objp: stor_device);
1393	return 0;
1394	}
1395
1396	static struct vmbus_channel *get_og_chn(struct storvsc_device *stor_device,
1397	u16 q_num)
1398	{
1399	u16 slot = 0;
1400	u16 hash_qnum;
1401	const struct cpumask *node_mask;
1402	int num_channels, tgt_cpu;
1403
1404	if (stor_device->num_sc == 0) {
1405	stor_device->stor_chns[q_num] = stor_device->device->channel;
1406	return stor_device->device->channel;
1407	}
1408
1409	/*
1410	* Our channel array could be sparsley populated and we
1411	* initiated I/O on a processor/hw-q that does not
1412	* currently have a designated channel. Fix this.
1413	* The strategy is simple:
1414	* I. Prefer the channel associated with the current CPU
1415	* II. Ensure NUMA locality
1416	* III. Distribute evenly (best effort)
1417	*/
1418
1419	/* Prefer the channel on the I/O issuing processor/hw-q */
1420	if (cpumask_test_cpu(cpu: q_num, cpumask: &stor_device->alloced_cpus))
1421	return stor_device->stor_chns[q_num];
1422
1423	node_mask = cpumask_of_node(cpu_to_node(cpu: q_num));
1424
1425	num_channels = 0;
1426	for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1427	if (cpumask_test_cpu(cpu: tgt_cpu, cpumask: node_mask))
1428	num_channels++;
1429	}
1430	if (num_channels == 0) {
1431	stor_device->stor_chns[q_num] = stor_device->device->channel;
1432	return stor_device->device->channel;
1433	}
1434
1435	hash_qnum = q_num;
1436	while (hash_qnum >= num_channels)
1437	hash_qnum -= num_channels;
1438
1439	for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) {
1440	if (!cpumask_test_cpu(cpu: tgt_cpu, cpumask: node_mask))
1441	continue;
1442	if (slot == hash_qnum)
1443	break;
1444	slot++;
1445	}
1446
1447	stor_device->stor_chns[q_num] = stor_device->stor_chns[tgt_cpu];
1448
1449	return stor_device->stor_chns[q_num];
1450	}
1451
1452
1453	static int storvsc_do_io(struct hv_device *device,
1454	struct storvsc_cmd_request *request, u16 q_num)
1455	{
1456	struct storvsc_device *stor_device;
1457	struct vstor_packet *vstor_packet;
1458	struct vmbus_channel *outgoing_channel, *channel;
1459	unsigned long flags;
1460	int ret = 0;
1461	const struct cpumask *node_mask;
1462	int tgt_cpu;
1463
1464	vstor_packet = &request->vstor_packet;
1465	stor_device = get_out_stor_device(device);
1466
1467	if (!stor_device)
1468	return -ENODEV;
1469
1470
1471	request->device = device;
1472	/*
1473	* Select an appropriate channel to send the request out.
1474	*/
1475	/* See storvsc_change_target_cpu(). */
1476	outgoing_channel = READ_ONCE(stor_device->stor_chns[q_num]);
1477	if (outgoing_channel != NULL) {
1478	if (hv_get_avail_to_write_percent(rbi: &outgoing_channel->outbound)
1479	> ring_avail_percent_lowater)
1480	goto found_channel;
1481
1482	/*
1483	* Channel is busy, try to find a channel on the same NUMA node
1484	*/
1485	node_mask = cpumask_of_node(cpu_to_node(cpu: q_num));
1486	for_each_cpu_wrap(tgt_cpu, &stor_device->alloced_cpus,
1487	q_num + 1) {
1488	if (!cpumask_test_cpu(cpu: tgt_cpu, cpumask: node_mask))
1489	continue;
1490	channel = READ_ONCE(stor_device->stor_chns[tgt_cpu]);
1491	if (!channel)
1492	continue;
1493	if (hv_get_avail_to_write_percent(rbi: &channel->outbound)
1494	> ring_avail_percent_lowater) {
1495	outgoing_channel = channel;
1496	goto found_channel;
1497	}
1498	}
1499
1500	/*
1501	* If we reach here, all the channels on the current
1502	* NUMA node are busy. Try to find a channel in
1503	* all NUMA nodes
1504	*/
1505	for_each_cpu_wrap(tgt_cpu, &stor_device->alloced_cpus,
1506	q_num + 1) {
1507	channel = READ_ONCE(stor_device->stor_chns[tgt_cpu]);
1508	if (!channel)
1509	continue;
1510	if (hv_get_avail_to_write_percent(rbi: &channel->outbound)
1511	> ring_avail_percent_lowater) {
1512	outgoing_channel = channel;
1513	goto found_channel;
1514	}
1515	}
1516	/*
1517	* If we reach here, all the channels are busy. Use the
1518	* original channel found.
1519	*/
1520	} else {
1521	spin_lock_irqsave(&stor_device->lock, flags);
1522	outgoing_channel = stor_device->stor_chns[q_num];
1523	if (outgoing_channel != NULL) {
1524	spin_unlock_irqrestore(lock: &stor_device->lock, flags);
1525	goto found_channel;
1526	}
1527	outgoing_channel = get_og_chn(stor_device, q_num);
1528	spin_unlock_irqrestore(lock: &stor_device->lock, flags);
1529	}
1530
1531	found_channel:
1532	vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
1533
1534	vstor_packet->vm_srb.length = sizeof(struct vmscsi_request);
1535
1536
1537	vstor_packet->vm_srb.sense_info_length = STORVSC_SENSE_BUFFER_SIZE;
1538
1539
1540	vstor_packet->vm_srb.data_transfer_length =
1541	request->payload->range.len;
1542
1543	vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
1544
1545	if (request->payload->range.len) {
1546
1547	ret = vmbus_sendpacket_mpb_desc(channel: outgoing_channel,
1548	mpb: request->payload, desc_size: request->payload_sz,
1549	buffer: vstor_packet,
1550	bufferlen: sizeof(struct vstor_packet),
1551	requestid: (unsigned long)request);
1552	} else {
1553	ret = vmbus_sendpacket(channel: outgoing_channel, buffer: vstor_packet,
1554	bufferLen: sizeof(struct vstor_packet),
1555	requestid: (unsigned long)request,
1556	type: VM_PKT_DATA_INBAND,
1557	VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1558	}
1559
1560	if (ret != 0)
1561	return ret;
1562
1563	atomic_inc(v: &stor_device->num_outstanding_req);
1564
1565	return ret;
1566	}
1567
1568	static int storvsc_device_alloc(struct scsi_device *sdevice)
1569	{
1570	/*
1571	* Set blist flag to permit the reading of the VPD pages even when
1572	* the target may claim SPC-2 compliance. MSFT targets currently
1573	* claim SPC-2 compliance while they implement post SPC-2 features.
1574	* With this flag we can correctly handle WRITE_SAME_16 issues.
1575	*
1576	* Hypervisor reports SCSI_UNKNOWN type for DVD ROM device but
1577	* still supports REPORT LUN.
1578	*/
1579	sdevice->sdev_bflags = BLIST_REPORTLUN2 | BLIST_TRY_VPD_PAGES;
1580
1581	return 0;
1582	}
1583
1584	static int storvsc_sdev_configure(struct scsi_device *sdevice,
1585	struct queue_limits *lim)
1586	{
1587	blk_queue_rq_timeout(sdevice->request_queue, (storvsc_timeout * HZ));
1588
1589	/* storvsc devices don't support MAINTENANCE_IN SCSI cmd */
1590	sdevice->no_report_opcodes = 1;
1591	sdevice->no_write_same = 1;
1592
1593	/*
1594	* If the host is WIN8 or WIN8 R2, claim conformance to SPC-3
1595	* if the device is a MSFT virtual device. If the host is
1596	* WIN10 or newer, allow write_same.
1597	*/
1598	if (!strncmp(sdevice->vendor, "Msft", 4)) {
1599	switch (vmstor_proto_version) {
1600	case VMSTOR_PROTO_VERSION_WIN8:
1601	case VMSTOR_PROTO_VERSION_WIN8_1:
1602	sdevice->scsi_level = SCSI_SPC_3;
1603	break;
1604	}
1605
1606	if (vmstor_proto_version >= VMSTOR_PROTO_VERSION_WIN10)
1607	sdevice->no_write_same = 0;
1608	}
1609
1610	return 0;
1611	}
1612
1613	static int storvsc_get_chs(struct scsi_device *sdev, struct gendisk *unused,
1614	sector_t capacity, int *info)
1615	{
1616	sector_t nsect = capacity;
1617	sector_t cylinders = nsect;
1618	int heads, sectors_pt;
1619
1620	/*
1621	* We are making up these values; let us keep it simple.
1622	*/
1623	heads = 0xff;
1624	sectors_pt = 0x3f; /* Sectors per track */
1625	sector_div(cylinders, heads * sectors_pt);
1626	if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
1627	cylinders = 0xffff;
1628
1629	info[0] = heads;
1630	info[1] = sectors_pt;
1631	info[2] = (int)cylinders;
1632
1633	return 0;
1634	}
1635
1636	static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
1637	{
1638	struct hv_host_device *host_dev = shost_priv(shost: scmnd->device->host);
1639	struct hv_device *device = host_dev->dev;
1640
1641	struct storvsc_device *stor_device;
1642	struct storvsc_cmd_request *request;
1643	struct vstor_packet *vstor_packet;
1644	int ret, t;
1645
1646	stor_device = get_out_stor_device(device);
1647	if (!stor_device)
1648	return FAILED;
1649
1650	request = &stor_device->reset_request;
1651	vstor_packet = &request->vstor_packet;
1652	memset(vstor_packet, 0, sizeof(struct vstor_packet));
1653
1654	init_completion(x: &request->wait_event);
1655
1656	vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
1657	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
1658	vstor_packet->vm_srb.path_id = stor_device->path_id;
1659
1660	ret = vmbus_sendpacket(channel: device->channel, buffer: vstor_packet,
1661	bufferLen: sizeof(struct vstor_packet),
1662	VMBUS_RQST_RESET,
1663	type: VM_PKT_DATA_INBAND,
1664	VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1665	if (ret != 0)
1666	return FAILED;
1667
1668	t = wait_for_completion_timeout(x: &request->wait_event, timeout: storvsc_timeout * HZ);
1669	if (t == 0)
1670	return TIMEOUT_ERROR;
1671
1672
1673	/*
1674	* At this point, all outstanding requests in the adapter
1675	* should have been flushed out and return to us
1676	* There is a potential race here where the host may be in
1677	* the process of responding when we return from here.
1678	* Just wait for all in-transit packets to be accounted for
1679	* before we return from here.
1680	*/
1681	storvsc_wait_to_drain(dev: stor_device);
1682
1683	return SUCCESS;
1684	}
1685
1686	/*
1687	* The host guarantees to respond to each command, although I/O latencies might
1688	* be unbounded on Azure. Reset the timer unconditionally to give the host a
1689	* chance to perform EH.
1690	*/
1691	static enum scsi_timeout_action storvsc_eh_timed_out(struct scsi_cmnd *scmnd)
1692	{
1693	return SCSI_EH_RESET_TIMER;
1694	}
1695
1696	static bool storvsc_scsi_cmd_ok(struct scsi_cmnd *scmnd)
1697	{
1698	bool allowed = true;
1699	u8 scsi_op = scmnd->cmnd[0];
1700
1701	switch (scsi_op) {
1702	/* the host does not handle WRITE_SAME, log accident usage */
1703	case WRITE_SAME:
1704	/*
1705	* smartd sends this command and the host does not handle
1706	* this. So, don't send it.
1707	*/
1708	case SET_WINDOW:
1709	set_host_byte(cmd: scmnd, status: DID_ERROR);
1710	allowed = false;
1711	break;
1712	default:
1713	break;
1714	}
1715	return allowed;
1716	}
1717
1718	static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
1719	{
1720	int ret;
1721	struct hv_host_device *host_dev = shost_priv(shost: host);
1722	struct hv_device *dev = host_dev->dev;
1723	struct storvsc_cmd_request *cmd_request = scsi_cmd_priv(cmd: scmnd);
1724	struct scatterlist *sgl;
1725	struct vmscsi_request *vm_srb;
1726	struct vmbus_packet_mpb_array *payload;
1727	u32 payload_sz;
1728	u32 length;
1729
1730	if (vmstor_proto_version <= VMSTOR_PROTO_VERSION_WIN8) {
1731	/*
1732	* On legacy hosts filter unimplemented commands.
1733	* Future hosts are expected to correctly handle
1734	* unsupported commands. Furthermore, it is
1735	* possible that some of the currently
1736	* unsupported commands maybe supported in
1737	* future versions of the host.
1738	*/
1739	if (!storvsc_scsi_cmd_ok(scmnd)) {
1740	scsi_done(cmd: scmnd);
1741	return 0;
1742	}
1743	}
1744
1745	/* Setup the cmd request */
1746	cmd_request->cmd = scmnd;
1747
1748	memset(&cmd_request->vstor_packet, 0, sizeof(struct vstor_packet));
1749	vm_srb = &cmd_request->vstor_packet.vm_srb;
1750	vm_srb->time_out_value = 60;
1751
1752	vm_srb->srb_flags |=
1753	SRB_FLAGS_DISABLE_SYNCH_TRANSFER;
1754
1755	if (scmnd->device->tagged_supported) {
1756	vm_srb->srb_flags |=
1757	(SRB_FLAGS_QUEUE_ACTION_ENABLE | SRB_FLAGS_NO_QUEUE_FREEZE);
1758	vm_srb->queue_tag = SP_UNTAGGED;
1759	vm_srb->queue_action = SRB_SIMPLE_TAG_REQUEST;
1760	}
1761
1762	/* Build the SRB */
1763	switch (scmnd->sc_data_direction) {
1764	case DMA_TO_DEVICE:
1765	vm_srb->data_in = WRITE_TYPE;
1766	vm_srb->srb_flags |= SRB_FLAGS_DATA_OUT;
1767	break;
1768	case DMA_FROM_DEVICE:
1769	vm_srb->data_in = READ_TYPE;
1770	vm_srb->srb_flags |= SRB_FLAGS_DATA_IN;
1771	break;
1772	case DMA_NONE:
1773	vm_srb->data_in = UNKNOWN_TYPE;
1774	vm_srb->srb_flags |= SRB_FLAGS_NO_DATA_TRANSFER;
1775	break;
1776	default:
1777	/*
1778	* This is DMA_BIDIRECTIONAL or something else we are never
1779	* supposed to see here.
1780	*/
1781	WARN(1, "Unexpected data direction: %d\n",
1782	scmnd->sc_data_direction);
1783	return -EINVAL;
1784	}
1785
1786
1787	vm_srb->port_number = host_dev->port;
1788	vm_srb->path_id = scmnd->device->channel;
1789	vm_srb->target_id = scmnd->device->id;
1790	vm_srb->lun = scmnd->device->lun;
1791
1792	vm_srb->cdb_length = scmnd->cmd_len;
1793
1794	memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
1795
1796	sgl = (struct scatterlist *)scsi_sglist(cmd: scmnd);
1797
1798	length = scsi_bufflen(cmd: scmnd);
1799	payload = (struct vmbus_packet_mpb_array *)&cmd_request->mpb;
1800	payload->range.len = 0;
1801	payload_sz = 0;
1802
1803	if (scsi_sg_count(cmd: scmnd)) {
1804	unsigned long offset_in_hvpg = offset_in_hvpage(sgl->offset);
1805	unsigned int hvpg_count = HVPFN_UP(offset_in_hvpg + length);
1806	struct scatterlist *sg;
1807	unsigned long hvpfn, hvpfns_to_add;
1808	int j, i = 0, sg_count;
1809
1810	payload_sz = (hvpg_count * sizeof(u64) +
1811	sizeof(struct vmbus_packet_mpb_array));
1812
1813	if (hvpg_count > MAX_PAGE_BUFFER_COUNT) {
1814	payload = kzalloc(payload_sz, GFP_ATOMIC);
1815	if (!payload)
1816	return SCSI_MLQUEUE_DEVICE_BUSY;
1817	}
1818
1819	payload->rangecount = 1;
1820	payload->range.len = length;
1821	payload->range.offset = offset_in_hvpg;
1822
1823	sg_count = scsi_dma_map(cmd: scmnd);
1824	if (sg_count < 0) {
1825	ret = SCSI_MLQUEUE_DEVICE_BUSY;
1826	goto err_free_payload;
1827	}
1828
1829	for_each_sg(sgl, sg, sg_count, j) {
1830	/*
1831	* Init values for the current sgl entry. hvpfns_to_add
1832	* is in units of Hyper-V size pages. Handling the
1833	* PAGE_SIZE != HV_HYP_PAGE_SIZE case also handles
1834	* values of sgl->offset that are larger than PAGE_SIZE.
1835	* Such offsets are handled even on other than the first
1836	* sgl entry, provided they are a multiple of PAGE_SIZE.
1837	*/
1838	hvpfn = HVPFN_DOWN(sg_dma_address(sg));
1839	hvpfns_to_add = HVPFN_UP(sg_dma_address(sg) +
1840	sg_dma_len(sg)) - hvpfn;
1841
1842	/*
1843	* Fill the next portion of the PFN array with
1844	* sequential Hyper-V PFNs for the continguous physical
1845	* memory described by the sgl entry. The end of the
1846	* last sgl should be reached at the same time that
1847	* the PFN array is filled.
1848	*/
1849	while (hvpfns_to_add--)
1850	payload->range.pfn_array[i++] = hvpfn++;
1851	}
1852	}
1853
1854	cmd_request->payload = payload;
1855	cmd_request->payload_sz = payload_sz;
1856
1857	/* Invokes the vsc to start an IO */
1858	ret = storvsc_do_io(device: dev, request: cmd_request, get_cpu());
1859	put_cpu();
1860
1861	if (ret)
1862	scsi_dma_unmap(cmd: scmnd);
1863
1864	if (ret == -EAGAIN) {
1865	/* no more space */
1866	ret = SCSI_MLQUEUE_DEVICE_BUSY;
1867	goto err_free_payload;
1868	}
1869
1870	return 0;
1871
1872	err_free_payload:
1873	if (payload_sz > sizeof(cmd_request->mpb))
1874	kfree(objp: payload);
1875
1876	return ret;
1877	}
1878
1879	static struct scsi_host_template scsi_driver = {
1880	.module = THIS_MODULE,
1881	.name = "storvsc_host_t",
1882	.cmd_size = sizeof(struct storvsc_cmd_request),
1883	.bios_param = storvsc_get_chs,
1884	.queuecommand = storvsc_queuecommand,
1885	.eh_host_reset_handler = storvsc_host_reset_handler,
1886	.proc_name = "storvsc_host",
1887	.eh_timed_out = storvsc_eh_timed_out,
1888	.sdev_init = storvsc_device_alloc,
1889	.sdev_configure = storvsc_sdev_configure,
1890	.cmd_per_lun = 2048,
1891	.this_id = -1,
1892	/* Ensure there are no gaps in presented sgls */
1893	.virt_boundary_mask = HV_HYP_PAGE_SIZE - 1,
1894	.no_write_same = 1,
1895	.track_queue_depth = 1,
1896	.change_queue_depth = storvsc_change_queue_depth,
1897	};
1898
1899	enum {
1900	SCSI_GUID,
1901	IDE_GUID,
1902	SFC_GUID,
1903	};
1904
1905	static const struct hv_vmbus_device_id id_table[] = {
1906	/* SCSI guid */
1907	{ HV_SCSI_GUID,
1908	.driver_data = SCSI_GUID
1909	},
1910	/* IDE guid */
1911	{ HV_IDE_GUID,
1912	.driver_data = IDE_GUID
1913	},
1914	/* Fibre Channel GUID */
1915	{
1916	HV_SYNTHFC_GUID,
1917	.driver_data = SFC_GUID
1918	},
1919	{ },
1920	};
1921
1922	MODULE_DEVICE_TABLE(vmbus, id_table);
1923
1924	static const struct { guid_t guid; } fc_guid = { HV_SYNTHFC_GUID };
1925
1926	static bool hv_dev_is_fc(struct hv_device *hv_dev)
1927	{
1928	return guid_equal(u1: &fc_guid.guid, u2: &hv_dev->dev_type);
1929	}
1930
1931	static int storvsc_probe(struct hv_device *device,
1932	const struct hv_vmbus_device_id *dev_id)
1933	{
1934	int ret;
1935	int num_cpus = num_online_cpus();
1936	int num_present_cpus = num_present_cpus();
1937	struct Scsi_Host *host;
1938	struct hv_host_device *host_dev;
1939	bool dev_is_ide = dev_id->driver_data == IDE_GUID;
1940	bool is_fc = dev_id->driver_data == SFC_GUID;
1941	int target = 0;
1942	struct storvsc_device *stor_device;
1943	int max_sub_channels = 0;
1944	u32 max_xfer_bytes;
1945
1946	/*
1947	* We support sub-channels for storage on SCSI and FC controllers.
1948	* The number of sub-channels offerred is based on the number of
1949	* VCPUs in the guest.
1950	*/
1951	if (!dev_is_ide)
1952	max_sub_channels =
1953	(num_cpus - 1) / storvsc_vcpus_per_sub_channel;
1954
1955	scsi_driver.can_queue = max_outstanding_req_per_channel *
1956	(max_sub_channels + 1) *
1957	(100 - ring_avail_percent_lowater) / 100;
1958
1959	host = scsi_host_alloc(&scsi_driver,
1960	sizeof(struct hv_host_device));
1961	if (!host)
1962	return -ENOMEM;
1963
1964	host_dev = shost_priv(shost: host);
1965	memset(host_dev, 0, sizeof(struct hv_host_device));
1966
1967	host_dev->port = host->host_no;
1968	host_dev->dev = device;
1969	host_dev->host = host;
1970
1971
1972	stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
1973	if (!stor_device) {
1974	ret = -ENOMEM;
1975	goto err_out0;
1976	}
1977
1978	stor_device->destroy = false;
1979	init_waitqueue_head(&stor_device->waiting_to_drain);
1980	stor_device->device = device;
1981	stor_device->host = host;
1982	spin_lock_init(&stor_device->lock);
1983	hv_set_drvdata(dev: device, data: stor_device);
1984	dma_set_min_align_mask(dev: &device->device, HV_HYP_PAGE_SIZE - 1);
1985
1986	stor_device->port_number = host->host_no;
1987	ret = storvsc_connect_to_vsp(device, ring_size: aligned_ringbuffer_size, is_fc);
1988	if (ret)
1989	goto err_out1;
1990
1991	host_dev->path = stor_device->path_id;
1992	host_dev->target = stor_device->target_id;
1993
1994	switch (dev_id->driver_data) {
1995	case SFC_GUID:
1996	host->max_lun = STORVSC_FC_MAX_LUNS_PER_TARGET;
1997	host->max_id = STORVSC_FC_MAX_TARGETS;
1998	host->max_channel = STORVSC_FC_MAX_CHANNELS - 1;
1999	#if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2000	host->transportt = fc_transport_template;
2001	#endif
2002	break;
2003
2004	case SCSI_GUID:
2005	host->max_lun = STORVSC_MAX_LUNS_PER_TARGET;
2006	host->max_id = STORVSC_MAX_TARGETS;
2007	host->max_channel = STORVSC_MAX_CHANNELS - 1;
2008	break;
2009
2010	default:
2011	host->max_lun = STORVSC_IDE_MAX_LUNS_PER_TARGET;
2012	host->max_id = STORVSC_IDE_MAX_TARGETS;
2013	host->max_channel = STORVSC_IDE_MAX_CHANNELS - 1;
2014	break;
2015	}
2016	/* max cmd length */
2017	host->max_cmd_len = STORVSC_MAX_CMD_LEN;
2018	/*
2019	* Any reasonable Hyper-V configuration should provide
2020	* max_transfer_bytes value aligning to HV_HYP_PAGE_SIZE,
2021	* protecting it from any weird value.
2022	*/
2023	max_xfer_bytes = round_down(stor_device->max_transfer_bytes, HV_HYP_PAGE_SIZE);
2024	if (is_fc)
2025	max_xfer_bytes = min(max_xfer_bytes, STORVSC_FC_MAX_XFER_SIZE);
2026
2027	/* max_hw_sectors_kb */
2028	host->max_sectors = max_xfer_bytes >> 9;
2029	/*
2030	* There are 2 requirements for Hyper-V storvsc sgl segments,
2031	* based on which the below calculation for max segments is
2032	* done:
2033	*
2034	* 1. Except for the first and last sgl segment, all sgl segments
2035	* should be align to HV_HYP_PAGE_SIZE, that also means the
2036	* maximum number of segments in a sgl can be calculated by
2037	* dividing the total max transfer length by HV_HYP_PAGE_SIZE.
2038	*
2039	* 2. Except for the first and last, each entry in the SGL must
2040	* have an offset that is a multiple of HV_HYP_PAGE_SIZE.
2041	*/
2042	host->sg_tablesize = (max_xfer_bytes >> HV_HYP_PAGE_SHIFT) + 1;
2043	/*
2044	* For non-IDE disks, the host supports multiple channels.
2045	* Set the number of HW queues we are supporting.
2046	*/
2047	if (!dev_is_ide) {
2048	if (storvsc_max_hw_queues > num_present_cpus) {
2049	storvsc_max_hw_queues = 0;
2050	storvsc_log(device, STORVSC_LOGGING_WARN,
2051	"Resetting invalid storvsc_max_hw_queues value to default.\n");
2052	}
2053	if (storvsc_max_hw_queues)
2054	host->nr_hw_queues = storvsc_max_hw_queues;
2055	else
2056	host->nr_hw_queues = num_present_cpus;
2057	}
2058
2059	/*
2060	* Set the error handler work queue.
2061	*/
2062	host_dev->handle_error_wq =
2063	alloc_ordered_workqueue("storvsc_error_wq_%d",
2064	0,
2065	host->host_no);
2066	if (!host_dev->handle_error_wq) {
2067	ret = -ENOMEM;
2068	goto err_out2;
2069	}
2070	INIT_WORK(&host_dev->host_scan_work, storvsc_host_scan);
2071	/* Register the HBA and start the scsi bus scan */
2072	ret = scsi_add_host(host, dev: &device->device);
2073	if (ret != 0)
2074	goto err_out3;
2075
2076	if (!dev_is_ide) {
2077	scsi_scan_host(host);
2078	} else {
2079	target = (device->dev_instance.b[5] << 8 |
2080	device->dev_instance.b[4]);
2081	ret = scsi_add_device(host, channel: 0, target, lun: 0);
2082	if (ret)
2083	goto err_out4;
2084	}
2085	#if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2086	if (host->transportt == fc_transport_template) {
2087	struct fc_rport_identifiers ids = {
2088	.roles = FC_PORT_ROLE_FCP_DUMMY_INITIATOR,
2089	};
2090
2091	fc_host_node_name(host) = stor_device->node_name;
2092	fc_host_port_name(host) = stor_device->port_name;
2093	stor_device->rport = fc_remote_port_add(shost: host, channel: 0, ids: &ids);
2094	if (!stor_device->rport) {
2095	ret = -ENOMEM;
2096	goto err_out4;
2097	}
2098	}
2099	#endif
2100	return 0;
2101
2102	err_out4:
2103	scsi_remove_host(host);
2104
2105	err_out3:
2106	destroy_workqueue(wq: host_dev->handle_error_wq);
2107
2108	err_out2:
2109	/*
2110	* Once we have connected with the host, we would need to
2111	* invoke storvsc_dev_remove() to rollback this state and
2112	* this call also frees up the stor_device; hence the jump around
2113	* err_out1 label.
2114	*/
2115	storvsc_dev_remove(device);
2116	goto err_out0;
2117
2118	err_out1:
2119	kfree(objp: stor_device->stor_chns);
2120	kfree(objp: stor_device);
2121
2122	err_out0:
2123	scsi_host_put(t: host);
2124	return ret;
2125	}
2126
2127	/* Change a scsi target's queue depth */
2128	static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth)
2129	{
2130	if (queue_depth > scsi_driver.can_queue)
2131	queue_depth = scsi_driver.can_queue;
2132
2133	return scsi_change_queue_depth(sdev, queue_depth);
2134	}
2135
2136	static void storvsc_remove(struct hv_device *dev)
2137	{
2138	struct storvsc_device *stor_device = hv_get_drvdata(dev);
2139	struct Scsi_Host *host = stor_device->host;
2140	struct hv_host_device *host_dev = shost_priv(shost: host);
2141
2142	#if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2143	if (host->transportt == fc_transport_template) {
2144	fc_remote_port_delete(rport: stor_device->rport);
2145	fc_remove_host(host);
2146	}
2147	#endif
2148	destroy_workqueue(wq: host_dev->handle_error_wq);
2149	scsi_remove_host(host);
2150	storvsc_dev_remove(device: dev);
2151	scsi_host_put(t: host);
2152	}
2153
2154	static int storvsc_suspend(struct hv_device *hv_dev)
2155	{
2156	struct storvsc_device *stor_device = hv_get_drvdata(dev: hv_dev);
2157	struct Scsi_Host *host = stor_device->host;
2158	struct hv_host_device *host_dev = shost_priv(shost: host);
2159
2160	storvsc_wait_to_drain(dev: stor_device);
2161
2162	drain_workqueue(wq: host_dev->handle_error_wq);
2163
2164	vmbus_close(channel: hv_dev->channel);
2165
2166	kfree(objp: stor_device->stor_chns);
2167	stor_device->stor_chns = NULL;
2168
2169	cpumask_clear(dstp: &stor_device->alloced_cpus);
2170
2171	return 0;
2172	}
2173
2174	static int storvsc_resume(struct hv_device *hv_dev)
2175	{
2176	int ret;
2177
2178	ret = storvsc_connect_to_vsp(device: hv_dev, ring_size: aligned_ringbuffer_size,
2179	is_fc: hv_dev_is_fc(hv_dev));
2180	return ret;
2181	}
2182
2183	static struct hv_driver storvsc_drv = {
2184	.name = KBUILD_MODNAME,
2185	.id_table = id_table,
2186	.probe = storvsc_probe,
2187	.remove = storvsc_remove,
2188	.suspend = storvsc_suspend,
2189	.resume = storvsc_resume,
2190	.driver = {
2191	.probe_type = PROBE_PREFER_ASYNCHRONOUS,
2192	},
2193	};
2194
2195	#if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2196	static struct fc_function_template fc_transport_functions = {
2197	.show_host_node_name = 1,
2198	.show_host_port_name = 1,
2199	};
2200	#endif
2201
2202	static int __init storvsc_drv_init(void)
2203	{
2204	int ret;
2205
2206	/*
2207	* Divide the ring buffer data size (which is 1 page less
2208	* than the ring buffer size since that page is reserved for
2209	* the ring buffer indices) by the max request size (which is
2210	* vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
2211	*/
2212	aligned_ringbuffer_size = VMBUS_RING_SIZE(storvsc_ringbuffer_size);
2213	max_outstanding_req_per_channel =
2214	((aligned_ringbuffer_size - PAGE_SIZE) /
2215	ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
2216	sizeof(struct vstor_packet) + sizeof(u64),
2217	sizeof(u64)));
2218
2219	#if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2220	fc_transport_template = fc_attach_transport(&fc_transport_functions);
2221	if (!fc_transport_template)
2222	return -ENODEV;
2223	#endif
2224
2225	ret = vmbus_driver_register(&storvsc_drv);
2226
2227	#if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2228	if (ret)
2229	fc_release_transport(fc_transport_template);
2230	#endif
2231
2232	return ret;
2233	}
2234
2235	static void __exit storvsc_drv_exit(void)
2236	{
2237	vmbus_driver_unregister(hv_driver: &storvsc_drv);
2238	#if IS_ENABLED(CONFIG_SCSI_FC_ATTRS)
2239	fc_release_transport(fc_transport_template);
2240	#endif
2241	}
2242
2243	MODULE_LICENSE("GPL");
2244	MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
2245	module_init(storvsc_drv_init);
2246	module_exit(storvsc_drv_exit);
2247