hv_storvsc_drv_freebsd.c revision 301866
1/*-
2 * Copyright (c) 2009-2012 Microsoft Corp.
3 * Copyright (c) 2012 NetApp Inc.
4 * Copyright (c) 2012 Citrix Inc.
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 *    notice unmodified, this list of conditions, and the following
12 *    disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 *    notice, this list of conditions and the following disclaimer in the
15 *    documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 */
28
29/**
30 * StorVSC driver for Hyper-V.  This driver presents a SCSI HBA interface
31 * to the Comman Access Method (CAM) layer.  CAM control blocks (CCBs) are
32 * converted into VSCSI protocol messages which are delivered to the parent
33 * partition StorVSP driver over the Hyper-V VMBUS.
34 */
35#include <sys/cdefs.h>
36__FBSDID("$FreeBSD: stable/10/sys/dev/hyperv/storvsc/hv_storvsc_drv_freebsd.c 301866 2016-06-13 09:02:08Z sephe $");
37
38#include <sys/param.h>
39#include <sys/proc.h>
40#include <sys/condvar.h>
41#include <sys/time.h>
42#include <sys/systm.h>
43#include <sys/sockio.h>
44#include <sys/mbuf.h>
45#include <sys/malloc.h>
46#include <sys/module.h>
47#include <sys/kernel.h>
48#include <sys/queue.h>
49#include <sys/lock.h>
50#include <sys/sx.h>
51#include <sys/taskqueue.h>
52#include <sys/bus.h>
53#include <sys/mutex.h>
54#include <sys/callout.h>
55#include <vm/vm.h>
56#include <vm/pmap.h>
57#include <vm/uma.h>
58#include <sys/lock.h>
59#include <sys/sema.h>
60#include <sys/sglist.h>
61#include <machine/bus.h>
62#include <sys/bus_dma.h>
63
64#include <cam/cam.h>
65#include <cam/cam_ccb.h>
66#include <cam/cam_periph.h>
67#include <cam/cam_sim.h>
68#include <cam/cam_xpt_sim.h>
69#include <cam/cam_xpt_internal.h>
70#include <cam/cam_debug.h>
71#include <cam/scsi/scsi_all.h>
72#include <cam/scsi/scsi_message.h>
73
74#include <dev/hyperv/include/hyperv.h>
75#include "hv_vstorage.h"
76
77#define STORVSC_RINGBUFFER_SIZE		(20*PAGE_SIZE)
78#define STORVSC_MAX_LUNS_PER_TARGET	(64)
79#define STORVSC_MAX_IO_REQUESTS		(STORVSC_MAX_LUNS_PER_TARGET * 2)
80#define BLKVSC_MAX_IDE_DISKS_PER_TARGET	(1)
81#define BLKVSC_MAX_IO_REQUESTS		STORVSC_MAX_IO_REQUESTS
82#define STORVSC_MAX_TARGETS		(2)
83
84#define VSTOR_PKT_SIZE	(sizeof(struct vstor_packet) - vmscsi_size_delta)
85
86#define HV_ALIGN(x, a) roundup2(x, a)
87
88struct storvsc_softc;
89
90struct hv_sgl_node {
91	LIST_ENTRY(hv_sgl_node) link;
92	struct sglist *sgl_data;
93};
94
95struct hv_sgl_page_pool{
96	LIST_HEAD(, hv_sgl_node) in_use_sgl_list;
97	LIST_HEAD(, hv_sgl_node) free_sgl_list;
98	boolean_t                is_init;
99} g_hv_sgl_page_pool;
100
101#define STORVSC_MAX_SG_PAGE_CNT STORVSC_MAX_IO_REQUESTS * HV_MAX_MULTIPAGE_BUFFER_COUNT
102
103enum storvsc_request_type {
104	WRITE_TYPE,
105	READ_TYPE,
106	UNKNOWN_TYPE
107};
108
109struct hv_storvsc_request {
110	LIST_ENTRY(hv_storvsc_request) link;
111	struct vstor_packet	vstor_packet;
112	hv_vmbus_multipage_buffer data_buf;
113	void *sense_data;
114	uint8_t sense_info_len;
115	uint8_t retries;
116	union ccb *ccb;
117	struct storvsc_softc *softc;
118	struct callout callout;
119	struct sema synch_sema; /*Synchronize the request/response if needed */
120	struct sglist *bounce_sgl;
121	unsigned int bounce_sgl_count;
122	uint64_t not_aligned_seg_bits;
123};
124
125struct storvsc_softc {
126	struct hv_device		*hs_dev;
127	LIST_HEAD(, hv_storvsc_request)	hs_free_list;
128	struct mtx			hs_lock;
129	struct storvsc_driver_props	*hs_drv_props;
130	int 				hs_unit;
131	uint32_t			hs_frozen;
132	struct cam_sim			*hs_sim;
133	struct cam_path 		*hs_path;
134	uint32_t			hs_num_out_reqs;
135	boolean_t			hs_destroy;
136	boolean_t			hs_drain_notify;
137	boolean_t			hs_open_multi_channel;
138	struct sema 			hs_drain_sema;
139	struct hv_storvsc_request	hs_init_req;
140	struct hv_storvsc_request	hs_reset_req;
141};
142
143
144/**
145 * HyperV storvsc timeout testing cases:
146 * a. IO returned after first timeout;
147 * b. IO returned after second timeout and queue freeze;
148 * c. IO returned while timer handler is running
149 * The first can be tested by "sg_senddiag -vv /dev/daX",
150 * and the second and third can be done by
151 * "sg_wr_mode -v -p 08 -c 0,1a -m 0,ff /dev/daX".
152 */
153#define HVS_TIMEOUT_TEST 0
154
155/*
156 * Bus/adapter reset functionality on the Hyper-V host is
157 * buggy and it will be disabled until
158 * it can be further tested.
159 */
160#define HVS_HOST_RESET 0
161
162struct storvsc_driver_props {
163	char		*drv_name;
164	char		*drv_desc;
165	uint8_t		drv_max_luns_per_target;
166	uint8_t		drv_max_ios_per_target;
167	uint32_t	drv_ringbuffer_size;
168};
169
170enum hv_storage_type {
171	DRIVER_BLKVSC,
172	DRIVER_STORVSC,
173	DRIVER_UNKNOWN
174};
175
176#define HS_MAX_ADAPTERS 10
177
178#define HV_STORAGE_SUPPORTS_MULTI_CHANNEL 0x1
179
180/* {ba6163d9-04a1-4d29-b605-72e2ffb1dc7f} */
181static const hv_guid gStorVscDeviceType={
182	.data = {0xd9, 0x63, 0x61, 0xba, 0xa1, 0x04, 0x29, 0x4d,
183		 0xb6, 0x05, 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f}
184};
185
186/* {32412632-86cb-44a2-9b5c-50d1417354f5} */
187static const hv_guid gBlkVscDeviceType={
188	.data = {0x32, 0x26, 0x41, 0x32, 0xcb, 0x86, 0xa2, 0x44,
189		 0x9b, 0x5c, 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5}
190};
191
192static struct storvsc_driver_props g_drv_props_table[] = {
193	{"blkvsc", "Hyper-V IDE Storage Interface",
194	 BLKVSC_MAX_IDE_DISKS_PER_TARGET, BLKVSC_MAX_IO_REQUESTS,
195	 STORVSC_RINGBUFFER_SIZE},
196	{"storvsc", "Hyper-V SCSI Storage Interface",
197	 STORVSC_MAX_LUNS_PER_TARGET, STORVSC_MAX_IO_REQUESTS,
198	 STORVSC_RINGBUFFER_SIZE}
199};
200
201/*
202 * Sense buffer size changed in win8; have a run-time
203 * variable to track the size we should use.
204 */
205static int sense_buffer_size = PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE;
206
207/*
208 * The size of the vmscsi_request has changed in win8. The
209 * additional size is for the newly added elements in the
210 * structure. These elements are valid only when we are talking
211 * to a win8 host.
212 * Track the correct size we need to apply.
213 */
214static int vmscsi_size_delta;
215/*
216 * The storage protocol version is determined during the
217 * initial exchange with the host.  It will indicate which
218 * storage functionality is available in the host.
219*/
220static int vmstor_proto_version;
221
222struct vmstor_proto {
223        int proto_version;
224        int sense_buffer_size;
225        int vmscsi_size_delta;
226};
227
228static const struct vmstor_proto vmstor_proto_list[] = {
229        {
230                VMSTOR_PROTOCOL_VERSION_WIN10,
231                POST_WIN7_STORVSC_SENSE_BUFFER_SIZE,
232                0
233        },
234        {
235                VMSTOR_PROTOCOL_VERSION_WIN8_1,
236                POST_WIN7_STORVSC_SENSE_BUFFER_SIZE,
237                0
238        },
239        {
240                VMSTOR_PROTOCOL_VERSION_WIN8,
241                POST_WIN7_STORVSC_SENSE_BUFFER_SIZE,
242                0
243        },
244        {
245                VMSTOR_PROTOCOL_VERSION_WIN7,
246                PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE,
247                sizeof(struct vmscsi_win8_extension),
248        },
249        {
250                VMSTOR_PROTOCOL_VERSION_WIN6,
251                PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE,
252                sizeof(struct vmscsi_win8_extension),
253        }
254};
255
256/* static functions */
257static int storvsc_probe(device_t dev);
258static int storvsc_attach(device_t dev);
259static int storvsc_detach(device_t dev);
260static void storvsc_poll(struct cam_sim * sim);
261static void storvsc_action(struct cam_sim * sim, union ccb * ccb);
262static int create_storvsc_request(union ccb *ccb, struct hv_storvsc_request *reqp);
263static void storvsc_free_request(struct storvsc_softc *sc, struct hv_storvsc_request *reqp);
264static enum hv_storage_type storvsc_get_storage_type(device_t dev);
265static void hv_storvsc_rescan_target(struct storvsc_softc *sc);
266static void hv_storvsc_on_channel_callback(void *context);
267static void hv_storvsc_on_iocompletion( struct storvsc_softc *sc,
268					struct vstor_packet *vstor_packet,
269					struct hv_storvsc_request *request);
270static int hv_storvsc_connect_vsp(struct hv_device *device);
271static void storvsc_io_done(struct hv_storvsc_request *reqp);
272static void storvsc_copy_sgl_to_bounce_buf(struct sglist *bounce_sgl,
273				bus_dma_segment_t *orig_sgl,
274				unsigned int orig_sgl_count,
275				uint64_t seg_bits);
276void storvsc_copy_from_bounce_buf_to_sgl(bus_dma_segment_t *dest_sgl,
277				unsigned int dest_sgl_count,
278				struct sglist* src_sgl,
279				uint64_t seg_bits);
280
281static device_method_t storvsc_methods[] = {
282	/* Device interface */
283	DEVMETHOD(device_probe,		storvsc_probe),
284	DEVMETHOD(device_attach,	storvsc_attach),
285	DEVMETHOD(device_detach,	storvsc_detach),
286	DEVMETHOD(device_shutdown,      bus_generic_shutdown),
287	DEVMETHOD_END
288};
289
290static driver_t storvsc_driver = {
291	"storvsc", storvsc_methods, sizeof(struct storvsc_softc),
292};
293
294static devclass_t storvsc_devclass;
295DRIVER_MODULE(storvsc, vmbus, storvsc_driver, storvsc_devclass, 0, 0);
296MODULE_VERSION(storvsc, 1);
297MODULE_DEPEND(storvsc, vmbus, 1, 1, 1);
298
299
300/**
301 * The host is capable of sending messages to us that are
302 * completely unsolicited. So, we need to address the race
303 * condition where we may be in the process of unloading the
304 * driver when the host may send us an unsolicited message.
305 * We address this issue by implementing a sequentially
306 * consistent protocol:
307 *
308 * 1. Channel callback is invoked while holding the the channel lock
309 *    and an unloading driver will reset the channel callback under
310 *    the protection of this channel lock.
311 *
312 * 2. To ensure bounded wait time for unloading a driver, we don't
313 *    permit outgoing traffic once the device is marked as being
314 *    destroyed.
315 *
316 * 3. Once the device is marked as being destroyed, we only
317 *    permit incoming traffic to properly account for
318 *    packets already sent out.
319 */
320static inline struct storvsc_softc *
321get_stor_device(struct hv_device *device,
322				boolean_t outbound)
323{
324	struct storvsc_softc *sc;
325
326	sc = device_get_softc(device->device);
327	if (sc == NULL) {
328		return NULL;
329	}
330
331	if (outbound) {
332		/*
333		 * Here we permit outgoing I/O only
334		 * if the device is not being destroyed.
335		 */
336
337		if (sc->hs_destroy) {
338			sc = NULL;
339		}
340	} else {
341		/*
342		 * inbound case; if being destroyed
343		 * only permit to account for
344		 * messages already sent out.
345		 */
346		if (sc->hs_destroy && (sc->hs_num_out_reqs == 0)) {
347			sc = NULL;
348		}
349	}
350	return sc;
351}
352
353/**
354 * @brief Callback handler, will be invoked when receive mutil-channel offer
355 *
356 * @param context  new multi-channel
357 */
358static void
359storvsc_handle_sc_creation(void *context)
360{
361	hv_vmbus_channel *new_channel;
362	struct hv_device *device;
363	struct storvsc_softc *sc;
364	struct vmstor_chan_props props;
365	int ret = 0;
366
367	new_channel = (hv_vmbus_channel *)context;
368	device = new_channel->primary_channel->device;
369	sc = get_stor_device(device, TRUE);
370	if (sc == NULL)
371		return;
372
373	if (FALSE == sc->hs_open_multi_channel)
374		return;
375
376	memset(&props, 0, sizeof(props));
377
378	ret = hv_vmbus_channel_open(new_channel,
379	    sc->hs_drv_props->drv_ringbuffer_size,
380  	    sc->hs_drv_props->drv_ringbuffer_size,
381	    (void *)&props,
382	    sizeof(struct vmstor_chan_props),
383	    hv_storvsc_on_channel_callback,
384	    new_channel);
385
386	return;
387}
388
389/**
390 * @brief Send multi-channel creation request to host
391 *
392 * @param device  a Hyper-V device pointer
393 * @param max_chans  the max channels supported by vmbus
394 */
395static void
396storvsc_send_multichannel_request(struct hv_device *dev, int max_chans)
397{
398	struct storvsc_softc *sc;
399	struct hv_storvsc_request *request;
400	struct vstor_packet *vstor_packet;
401	int request_channels_cnt = 0;
402	int ret;
403
404	/* get multichannels count that need to create */
405	request_channels_cnt = MIN(max_chans, mp_ncpus);
406
407	sc = get_stor_device(dev, TRUE);
408	if (sc == NULL) {
409		printf("Storvsc_error: get sc failed while send mutilchannel "
410		    "request\n");
411		return;
412	}
413
414	request = &sc->hs_init_req;
415
416	/* Establish a handler for multi-channel */
417	dev->channel->sc_creation_callback = storvsc_handle_sc_creation;
418
419	/* request the host to create multi-channel */
420	memset(request, 0, sizeof(struct hv_storvsc_request));
421
422	sema_init(&request->synch_sema, 0, ("stor_synch_sema"));
423
424	vstor_packet = &request->vstor_packet;
425
426	vstor_packet->operation = VSTOR_OPERATION_CREATE_MULTI_CHANNELS;
427	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
428	vstor_packet->u.multi_channels_cnt = request_channels_cnt;
429
430	ret = hv_vmbus_channel_send_packet(
431	    dev->channel,
432	    vstor_packet,
433	    VSTOR_PKT_SIZE,
434	    (uint64_t)(uintptr_t)request,
435	    HV_VMBUS_PACKET_TYPE_DATA_IN_BAND,
436	    HV_VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
437
438	/* wait for 5 seconds */
439	ret = sema_timedwait(&request->synch_sema, 5 * hz);
440	if (ret != 0) {
441		printf("Storvsc_error: create multi-channel timeout, %d\n",
442		    ret);
443		return;
444	}
445
446	if (vstor_packet->operation != VSTOR_OPERATION_COMPLETEIO ||
447	    vstor_packet->status != 0) {
448		printf("Storvsc_error: create multi-channel invalid operation "
449		    "(%d) or statue (%u)\n",
450		    vstor_packet->operation, vstor_packet->status);
451		return;
452	}
453
454	sc->hs_open_multi_channel = TRUE;
455
456	if (bootverbose)
457		printf("Storvsc create multi-channel success!\n");
458}
459
460/**
461 * @brief initialize channel connection to parent partition
462 *
463 * @param dev  a Hyper-V device pointer
464 * @returns  0 on success, non-zero error on failure
465 */
466static int
467hv_storvsc_channel_init(struct hv_device *dev)
468{
469	int ret = 0, i;
470	struct hv_storvsc_request *request;
471	struct vstor_packet *vstor_packet;
472	struct storvsc_softc *sc;
473	uint16_t max_chans = 0;
474	boolean_t support_multichannel = FALSE;
475
476	max_chans = 0;
477	support_multichannel = FALSE;
478
479	sc = get_stor_device(dev, TRUE);
480	if (sc == NULL)
481		return (ENODEV);
482
483	request = &sc->hs_init_req;
484	memset(request, 0, sizeof(struct hv_storvsc_request));
485	vstor_packet = &request->vstor_packet;
486	request->softc = sc;
487
488	/**
489	 * Initiate the vsc/vsp initialization protocol on the open channel
490	 */
491	sema_init(&request->synch_sema, 0, ("stor_synch_sema"));
492
493	vstor_packet->operation = VSTOR_OPERATION_BEGININITIALIZATION;
494	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
495
496
497	ret = hv_vmbus_channel_send_packet(
498			dev->channel,
499			vstor_packet,
500			VSTOR_PKT_SIZE,
501			(uint64_t)(uintptr_t)request,
502			HV_VMBUS_PACKET_TYPE_DATA_IN_BAND,
503			HV_VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
504
505	if (ret != 0)
506		goto cleanup;
507
508	/* wait 5 seconds */
509	ret = sema_timedwait(&request->synch_sema, 5 * hz);
510	if (ret != 0)
511		goto cleanup;
512
513	if (vstor_packet->operation != VSTOR_OPERATION_COMPLETEIO ||
514		vstor_packet->status != 0) {
515		goto cleanup;
516	}
517
518	for (i = 0; i < nitems(vmstor_proto_list); i++) {
519		/* reuse the packet for version range supported */
520
521		memset(vstor_packet, 0, sizeof(struct vstor_packet));
522		vstor_packet->operation = VSTOR_OPERATION_QUERYPROTOCOLVERSION;
523		vstor_packet->flags = REQUEST_COMPLETION_FLAG;
524
525		vstor_packet->u.version.major_minor =
526			vmstor_proto_list[i].proto_version;
527
528		/* revision is only significant for Windows guests */
529		vstor_packet->u.version.revision = 0;
530
531		ret = hv_vmbus_channel_send_packet(
532			dev->channel,
533			vstor_packet,
534			VSTOR_PKT_SIZE,
535			(uint64_t)(uintptr_t)request,
536			HV_VMBUS_PACKET_TYPE_DATA_IN_BAND,
537			HV_VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
538
539		if (ret != 0)
540			goto cleanup;
541
542		/* wait 5 seconds */
543		ret = sema_timedwait(&request->synch_sema, 5 * hz);
544
545		if (ret)
546			goto cleanup;
547
548		if (vstor_packet->operation != VSTOR_OPERATION_COMPLETEIO) {
549			ret = EINVAL;
550			goto cleanup;
551		}
552		if (vstor_packet->status == 0) {
553			vmstor_proto_version =
554				vmstor_proto_list[i].proto_version;
555			sense_buffer_size =
556				vmstor_proto_list[i].sense_buffer_size;
557			vmscsi_size_delta =
558				vmstor_proto_list[i].vmscsi_size_delta;
559			break;
560		}
561	}
562
563	if (vstor_packet->status != 0) {
564		ret = EINVAL;
565		goto cleanup;
566	}
567	/**
568	 * Query channel properties
569	 */
570	memset(vstor_packet, 0, sizeof(struct vstor_packet));
571	vstor_packet->operation = VSTOR_OPERATION_QUERYPROPERTIES;
572	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
573
574	ret = hv_vmbus_channel_send_packet(
575				dev->channel,
576				vstor_packet,
577				VSTOR_PKT_SIZE,
578				(uint64_t)(uintptr_t)request,
579				HV_VMBUS_PACKET_TYPE_DATA_IN_BAND,
580				HV_VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
581
582	if ( ret != 0)
583		goto cleanup;
584
585	/* wait 5 seconds */
586	ret = sema_timedwait(&request->synch_sema, 5 * hz);
587
588	if (ret != 0)
589		goto cleanup;
590
591	/* TODO: Check returned version */
592	if (vstor_packet->operation != VSTOR_OPERATION_COMPLETEIO ||
593	    vstor_packet->status != 0) {
594		goto cleanup;
595	}
596
597	/* multi-channels feature is supported by WIN8 and above version */
598	max_chans = vstor_packet->u.chan_props.max_channel_cnt;
599	if ((hv_vmbus_protocal_version != HV_VMBUS_VERSION_WIN7) &&
600	    (hv_vmbus_protocal_version != HV_VMBUS_VERSION_WS2008) &&
601	    (vstor_packet->u.chan_props.flags &
602	     HV_STORAGE_SUPPORTS_MULTI_CHANNEL)) {
603		support_multichannel = TRUE;
604	}
605
606	memset(vstor_packet, 0, sizeof(struct vstor_packet));
607	vstor_packet->operation = VSTOR_OPERATION_ENDINITIALIZATION;
608	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
609
610	ret = hv_vmbus_channel_send_packet(
611			dev->channel,
612			vstor_packet,
613			VSTOR_PKT_SIZE,
614			(uint64_t)(uintptr_t)request,
615			HV_VMBUS_PACKET_TYPE_DATA_IN_BAND,
616			HV_VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
617
618	if (ret != 0) {
619		goto cleanup;
620	}
621
622	/* wait 5 seconds */
623	ret = sema_timedwait(&request->synch_sema, 5 * hz);
624
625	if (ret != 0)
626		goto cleanup;
627
628	if (vstor_packet->operation != VSTOR_OPERATION_COMPLETEIO ||
629	    vstor_packet->status != 0)
630		goto cleanup;
631
632	/*
633	 * If multi-channel is supported, send multichannel create
634	 * request to host.
635	 */
636	if (support_multichannel)
637		storvsc_send_multichannel_request(dev, max_chans);
638
639cleanup:
640	sema_destroy(&request->synch_sema);
641	return (ret);
642}
643
644/**
645 * @brief Open channel connection to paraent partition StorVSP driver
646 *
647 * Open and initialize channel connection to parent partition StorVSP driver.
648 *
649 * @param pointer to a Hyper-V device
650 * @returns 0 on success, non-zero error on failure
651 */
652static int
653hv_storvsc_connect_vsp(struct hv_device *dev)
654{
655	int ret = 0;
656	struct vmstor_chan_props props;
657	struct storvsc_softc *sc;
658
659	sc = device_get_softc(dev->device);
660
661	memset(&props, 0, sizeof(struct vmstor_chan_props));
662
663	/*
664	 * Open the channel
665	 */
666
667	ret = hv_vmbus_channel_open(
668		dev->channel,
669		sc->hs_drv_props->drv_ringbuffer_size,
670		sc->hs_drv_props->drv_ringbuffer_size,
671		(void *)&props,
672		sizeof(struct vmstor_chan_props),
673		hv_storvsc_on_channel_callback,
674		dev->channel);
675
676	if (ret != 0) {
677		return ret;
678	}
679
680	ret = hv_storvsc_channel_init(dev);
681
682	return (ret);
683}
684
685#if HVS_HOST_RESET
686static int
687hv_storvsc_host_reset(struct hv_device *dev)
688{
689	int ret = 0;
690	struct storvsc_softc *sc;
691
692	struct hv_storvsc_request *request;
693	struct vstor_packet *vstor_packet;
694
695	sc = get_stor_device(dev, TRUE);
696	if (sc == NULL) {
697		return ENODEV;
698	}
699
700	request = &sc->hs_reset_req;
701	request->softc = sc;
702	vstor_packet = &request->vstor_packet;
703
704	sema_init(&request->synch_sema, 0, "stor synch sema");
705
706	vstor_packet->operation = VSTOR_OPERATION_RESETBUS;
707	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
708
709	ret = hv_vmbus_channel_send_packet(dev->channel,
710			vstor_packet,
711			VSTOR_PKT_SIZE,
712			(uint64_t)(uintptr_t)&sc->hs_reset_req,
713			HV_VMBUS_PACKET_TYPE_DATA_IN_BAND,
714			HV_VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
715
716	if (ret != 0) {
717		goto cleanup;
718	}
719
720	ret = sema_timedwait(&request->synch_sema, 5 * hz); /* KYS 5 seconds */
721
722	if (ret) {
723		goto cleanup;
724	}
725
726
727	/*
728	 * At this point, all outstanding requests in the adapter
729	 * should have been flushed out and return to us
730	 */
731
732cleanup:
733	sema_destroy(&request->synch_sema);
734	return (ret);
735}
736#endif /* HVS_HOST_RESET */
737
738/**
739 * @brief Function to initiate an I/O request
740 *
741 * @param device Hyper-V device pointer
742 * @param request pointer to a request structure
743 * @returns 0 on success, non-zero error on failure
744 */
745static int
746hv_storvsc_io_request(struct hv_device *device,
747					  struct hv_storvsc_request *request)
748{
749	struct storvsc_softc *sc;
750	struct vstor_packet *vstor_packet = &request->vstor_packet;
751	struct hv_vmbus_channel* outgoing_channel = NULL;
752	int ret = 0;
753
754	sc = get_stor_device(device, TRUE);
755
756	if (sc == NULL) {
757		return ENODEV;
758	}
759
760	vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
761
762	vstor_packet->u.vm_srb.length = VSTOR_PKT_SIZE;
763
764	vstor_packet->u.vm_srb.sense_info_len = sense_buffer_size;
765
766	vstor_packet->u.vm_srb.transfer_len = request->data_buf.length;
767
768	vstor_packet->operation = VSTOR_OPERATION_EXECUTESRB;
769
770	outgoing_channel = vmbus_select_outgoing_channel(device->channel);
771
772	mtx_unlock(&request->softc->hs_lock);
773	if (request->data_buf.length) {
774		ret = hv_vmbus_channel_send_packet_multipagebuffer(
775				outgoing_channel,
776				&request->data_buf,
777				vstor_packet,
778				VSTOR_PKT_SIZE,
779				(uint64_t)(uintptr_t)request);
780
781	} else {
782		ret = hv_vmbus_channel_send_packet(
783			outgoing_channel,
784			vstor_packet,
785			VSTOR_PKT_SIZE,
786			(uint64_t)(uintptr_t)request,
787			HV_VMBUS_PACKET_TYPE_DATA_IN_BAND,
788			HV_VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
789	}
790	mtx_lock(&request->softc->hs_lock);
791
792	if (ret != 0) {
793		printf("Unable to send packet %p ret %d", vstor_packet, ret);
794	} else {
795		atomic_add_int(&sc->hs_num_out_reqs, 1);
796	}
797
798	return (ret);
799}
800
801
802/**
803 * Process IO_COMPLETION_OPERATION and ready
804 * the result to be completed for upper layer
805 * processing by the CAM layer.
806 */
807static void
808hv_storvsc_on_iocompletion(struct storvsc_softc *sc,
809			   struct vstor_packet *vstor_packet,
810			   struct hv_storvsc_request *request)
811{
812	struct vmscsi_req *vm_srb;
813
814	vm_srb = &vstor_packet->u.vm_srb;
815
816	if (((vm_srb->scsi_status & 0xFF) == SCSI_STATUS_CHECK_COND) &&
817			(vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID)) {
818		/* Autosense data available */
819
820		KASSERT(vm_srb->sense_info_len <= request->sense_info_len,
821				("vm_srb->sense_info_len <= "
822				 "request->sense_info_len"));
823
824		memcpy(request->sense_data, vm_srb->u.sense_data,
825			vm_srb->sense_info_len);
826
827		request->sense_info_len = vm_srb->sense_info_len;
828	}
829
830	/* Complete request by passing to the CAM layer */
831	storvsc_io_done(request);
832	atomic_subtract_int(&sc->hs_num_out_reqs, 1);
833	if (sc->hs_drain_notify && (sc->hs_num_out_reqs == 0)) {
834		sema_post(&sc->hs_drain_sema);
835	}
836}
837
838static void
839hv_storvsc_rescan_target(struct storvsc_softc *sc)
840{
841	path_id_t pathid;
842	target_id_t targetid;
843	union ccb *ccb;
844
845	pathid = cam_sim_path(sc->hs_sim);
846	targetid = CAM_TARGET_WILDCARD;
847
848	/*
849	 * Allocate a CCB and schedule a rescan.
850	 */
851	ccb = xpt_alloc_ccb_nowait();
852	if (ccb == NULL) {
853		printf("unable to alloc CCB for rescan\n");
854		return;
855	}
856
857	if (xpt_create_path(&ccb->ccb_h.path, NULL, pathid, targetid,
858	    CAM_LUN_WILDCARD) != CAM_REQ_CMP) {
859		printf("unable to create path for rescan, pathid: %u,"
860		    "targetid: %u\n", pathid, targetid);
861		xpt_free_ccb(ccb);
862		return;
863	}
864
865	if (targetid == CAM_TARGET_WILDCARD)
866		ccb->ccb_h.func_code = XPT_SCAN_BUS;
867	else
868		ccb->ccb_h.func_code = XPT_SCAN_TGT;
869
870	xpt_rescan(ccb);
871}
872
873static void
874hv_storvsc_on_channel_callback(void *context)
875{
876	int ret = 0;
877	hv_vmbus_channel *channel = (hv_vmbus_channel *)context;
878	struct hv_device *device = NULL;
879	struct storvsc_softc *sc;
880	uint32_t bytes_recvd;
881	uint64_t request_id;
882	uint8_t packet[roundup2(sizeof(struct vstor_packet), 8)];
883	struct hv_storvsc_request *request;
884	struct vstor_packet *vstor_packet;
885
886	if (channel->primary_channel != NULL){
887		device = channel->primary_channel->device;
888	} else {
889		device = channel->device;
890	}
891
892	KASSERT(device, ("device is NULL"));
893
894	sc = get_stor_device(device, FALSE);
895	if (sc == NULL) {
896		printf("Storvsc_error: get stor device failed.\n");
897		return;
898	}
899
900	ret = hv_vmbus_channel_recv_packet(
901			channel,
902			packet,
903			roundup2(VSTOR_PKT_SIZE, 8),
904			&bytes_recvd,
905			&request_id);
906
907	while ((ret == 0) && (bytes_recvd > 0)) {
908		request = (struct hv_storvsc_request *)(uintptr_t)request_id;
909
910		if ((request == &sc->hs_init_req) ||
911			(request == &sc->hs_reset_req)) {
912			memcpy(&request->vstor_packet, packet,
913				   sizeof(struct vstor_packet));
914			sema_post(&request->synch_sema);
915		} else {
916			vstor_packet = (struct vstor_packet *)packet;
917			switch(vstor_packet->operation) {
918			case VSTOR_OPERATION_COMPLETEIO:
919				if (request == NULL)
920					panic("VMBUS: storvsc received a "
921					    "packet with NULL request id in "
922					    "COMPLETEIO operation.");
923
924				hv_storvsc_on_iocompletion(sc,
925							vstor_packet, request);
926				break;
927			case VSTOR_OPERATION_REMOVEDEVICE:
928				printf("VMBUS: storvsc operation %d not "
929				    "implemented.\n", vstor_packet->operation);
930				/* TODO: implement */
931				break;
932			case VSTOR_OPERATION_ENUMERATE_BUS:
933				hv_storvsc_rescan_target(sc);
934				break;
935			default:
936				break;
937			}
938		}
939		ret = hv_vmbus_channel_recv_packet(
940				channel,
941				packet,
942				roundup2(VSTOR_PKT_SIZE, 8),
943				&bytes_recvd,
944				&request_id);
945	}
946}
947
948/**
949 * @brief StorVSC probe function
950 *
951 * Device probe function.  Returns 0 if the input device is a StorVSC
952 * device.  Otherwise, a ENXIO is returned.  If the input device is
953 * for BlkVSC (paravirtual IDE) device and this support is disabled in
954 * favor of the emulated ATA/IDE device, return ENXIO.
955 *
956 * @param a device
957 * @returns 0 on success, ENXIO if not a matcing StorVSC device
958 */
959static int
960storvsc_probe(device_t dev)
961{
962	int ata_disk_enable = 0;
963	int ret	= ENXIO;
964
965	switch (storvsc_get_storage_type(dev)) {
966	case DRIVER_BLKVSC:
967		if(bootverbose)
968			device_printf(dev, "DRIVER_BLKVSC-Emulated ATA/IDE probe\n");
969		if (!getenv_int("hw.ata.disk_enable", &ata_disk_enable)) {
970			if(bootverbose)
971				device_printf(dev,
972					"Enlightened ATA/IDE detected\n");
973			ret = BUS_PROBE_DEFAULT;
974		} else if(bootverbose)
975			device_printf(dev, "Emulated ATA/IDE set (hw.ata.disk_enable set)\n");
976		break;
977	case DRIVER_STORVSC:
978		if(bootverbose)
979			device_printf(dev, "Enlightened SCSI device detected\n");
980		ret = BUS_PROBE_DEFAULT;
981		break;
982	default:
983		ret = ENXIO;
984	}
985	return (ret);
986}
987
988/**
989 * @brief StorVSC attach function
990 *
991 * Function responsible for allocating per-device structures,
992 * setting up CAM interfaces and scanning for available LUNs to
993 * be used for SCSI device peripherals.
994 *
995 * @param a device
996 * @returns 0 on success or an error on failure
997 */
998static int
999storvsc_attach(device_t dev)
1000{
1001	struct hv_device *hv_dev = vmbus_get_devctx(dev);
1002	enum hv_storage_type stor_type;
1003	struct storvsc_softc *sc;
1004	struct cam_devq *devq;
1005	int ret, i, j;
1006	struct hv_storvsc_request *reqp;
1007	struct root_hold_token *root_mount_token = NULL;
1008	struct hv_sgl_node *sgl_node = NULL;
1009	void *tmp_buff = NULL;
1010
1011	/*
1012	 * We need to serialize storvsc attach calls.
1013	 */
1014	root_mount_token = root_mount_hold("storvsc");
1015
1016	sc = device_get_softc(dev);
1017	if (sc == NULL) {
1018		ret = ENOMEM;
1019		goto cleanup;
1020	}
1021
1022	stor_type = storvsc_get_storage_type(dev);
1023
1024	if (stor_type == DRIVER_UNKNOWN) {
1025		ret = ENODEV;
1026		goto cleanup;
1027	}
1028
1029	bzero(sc, sizeof(struct storvsc_softc));
1030
1031	/* fill in driver specific properties */
1032	sc->hs_drv_props = &g_drv_props_table[stor_type];
1033
1034	/* fill in device specific properties */
1035	sc->hs_unit	= device_get_unit(dev);
1036	sc->hs_dev	= hv_dev;
1037	device_set_desc(dev, g_drv_props_table[stor_type].drv_desc);
1038
1039	LIST_INIT(&sc->hs_free_list);
1040	mtx_init(&sc->hs_lock, "hvslck", NULL, MTX_DEF);
1041
1042	for (i = 0; i < sc->hs_drv_props->drv_max_ios_per_target; ++i) {
1043		reqp = malloc(sizeof(struct hv_storvsc_request),
1044				 M_DEVBUF, M_WAITOK|M_ZERO);
1045		reqp->softc = sc;
1046
1047		LIST_INSERT_HEAD(&sc->hs_free_list, reqp, link);
1048	}
1049
1050	/* create sg-list page pool */
1051	if (FALSE == g_hv_sgl_page_pool.is_init) {
1052		g_hv_sgl_page_pool.is_init = TRUE;
1053		LIST_INIT(&g_hv_sgl_page_pool.in_use_sgl_list);
1054		LIST_INIT(&g_hv_sgl_page_pool.free_sgl_list);
1055
1056		/*
1057		 * Pre-create SG list, each SG list with
1058		 * HV_MAX_MULTIPAGE_BUFFER_COUNT segments, each
1059		 * segment has one page buffer
1060		 */
1061		for (i = 0; i < STORVSC_MAX_IO_REQUESTS; i++) {
1062	        	sgl_node = malloc(sizeof(struct hv_sgl_node),
1063			    M_DEVBUF, M_WAITOK|M_ZERO);
1064
1065			sgl_node->sgl_data =
1066			    sglist_alloc(HV_MAX_MULTIPAGE_BUFFER_COUNT,
1067			    M_WAITOK|M_ZERO);
1068
1069			for (j = 0; j < HV_MAX_MULTIPAGE_BUFFER_COUNT; j++) {
1070				tmp_buff = malloc(PAGE_SIZE,
1071				    M_DEVBUF, M_WAITOK|M_ZERO);
1072
1073				sgl_node->sgl_data->sg_segs[j].ss_paddr =
1074				    (vm_paddr_t)tmp_buff;
1075			}
1076
1077			LIST_INSERT_HEAD(&g_hv_sgl_page_pool.free_sgl_list,
1078			    sgl_node, link);
1079		}
1080	}
1081
1082	sc->hs_destroy = FALSE;
1083	sc->hs_drain_notify = FALSE;
1084	sc->hs_open_multi_channel = FALSE;
1085	sema_init(&sc->hs_drain_sema, 0, "Store Drain Sema");
1086
1087	ret = hv_storvsc_connect_vsp(hv_dev);
1088	if (ret != 0) {
1089		goto cleanup;
1090	}
1091
1092	/*
1093	 * Create the device queue.
1094	 * Hyper-V maps each target to one SCSI HBA
1095	 */
1096	devq = cam_simq_alloc(sc->hs_drv_props->drv_max_ios_per_target);
1097	if (devq == NULL) {
1098		device_printf(dev, "Failed to alloc device queue\n");
1099		ret = ENOMEM;
1100		goto cleanup;
1101	}
1102
1103	sc->hs_sim = cam_sim_alloc(storvsc_action,
1104				storvsc_poll,
1105				sc->hs_drv_props->drv_name,
1106				sc,
1107				sc->hs_unit,
1108				&sc->hs_lock, 1,
1109				sc->hs_drv_props->drv_max_ios_per_target,
1110				devq);
1111
1112	if (sc->hs_sim == NULL) {
1113		device_printf(dev, "Failed to alloc sim\n");
1114		cam_simq_free(devq);
1115		ret = ENOMEM;
1116		goto cleanup;
1117	}
1118
1119	mtx_lock(&sc->hs_lock);
1120	/* bus_id is set to 0, need to get it from VMBUS channel query? */
1121	if (xpt_bus_register(sc->hs_sim, dev, 0) != CAM_SUCCESS) {
1122		cam_sim_free(sc->hs_sim, /*free_devq*/TRUE);
1123		mtx_unlock(&sc->hs_lock);
1124		device_printf(dev, "Unable to register SCSI bus\n");
1125		ret = ENXIO;
1126		goto cleanup;
1127	}
1128
1129	if (xpt_create_path(&sc->hs_path, /*periph*/NULL,
1130		 cam_sim_path(sc->hs_sim),
1131		CAM_TARGET_WILDCARD, CAM_LUN_WILDCARD) != CAM_REQ_CMP) {
1132		xpt_bus_deregister(cam_sim_path(sc->hs_sim));
1133		cam_sim_free(sc->hs_sim, /*free_devq*/TRUE);
1134		mtx_unlock(&sc->hs_lock);
1135		device_printf(dev, "Unable to create path\n");
1136		ret = ENXIO;
1137		goto cleanup;
1138	}
1139
1140	mtx_unlock(&sc->hs_lock);
1141
1142	root_mount_rel(root_mount_token);
1143	return (0);
1144
1145
1146cleanup:
1147	root_mount_rel(root_mount_token);
1148	while (!LIST_EMPTY(&sc->hs_free_list)) {
1149		reqp = LIST_FIRST(&sc->hs_free_list);
1150		LIST_REMOVE(reqp, link);
1151		free(reqp, M_DEVBUF);
1152	}
1153
1154	while (!LIST_EMPTY(&g_hv_sgl_page_pool.free_sgl_list)) {
1155		sgl_node = LIST_FIRST(&g_hv_sgl_page_pool.free_sgl_list);
1156		LIST_REMOVE(sgl_node, link);
1157		for (j = 0; j < HV_MAX_MULTIPAGE_BUFFER_COUNT; j++) {
1158			if (NULL !=
1159			    (void*)sgl_node->sgl_data->sg_segs[j].ss_paddr) {
1160				free((void*)sgl_node->sgl_data->sg_segs[j].ss_paddr, M_DEVBUF);
1161			}
1162		}
1163		sglist_free(sgl_node->sgl_data);
1164		free(sgl_node, M_DEVBUF);
1165	}
1166
1167	return (ret);
1168}
1169
1170/**
1171 * @brief StorVSC device detach function
1172 *
1173 * This function is responsible for safely detaching a
1174 * StorVSC device.  This includes waiting for inbound responses
1175 * to complete and freeing associated per-device structures.
1176 *
1177 * @param dev a device
1178 * returns 0 on success
1179 */
1180static int
1181storvsc_detach(device_t dev)
1182{
1183	struct storvsc_softc *sc = device_get_softc(dev);
1184	struct hv_storvsc_request *reqp = NULL;
1185	struct hv_device *hv_device = vmbus_get_devctx(dev);
1186	struct hv_sgl_node *sgl_node = NULL;
1187	int j = 0;
1188
1189	mtx_lock(&hv_device->channel->inbound_lock);
1190	sc->hs_destroy = TRUE;
1191	mtx_unlock(&hv_device->channel->inbound_lock);
1192
1193	/*
1194	 * At this point, all outbound traffic should be disabled. We
1195	 * only allow inbound traffic (responses) to proceed so that
1196	 * outstanding requests can be completed.
1197	 */
1198
1199	sc->hs_drain_notify = TRUE;
1200	sema_wait(&sc->hs_drain_sema);
1201	sc->hs_drain_notify = FALSE;
1202
1203	/*
1204	 * Since we have already drained, we don't need to busy wait.
1205	 * The call to close the channel will reset the callback
1206	 * under the protection of the incoming channel lock.
1207	 */
1208
1209	hv_vmbus_channel_close(hv_device->channel);
1210
1211	mtx_lock(&sc->hs_lock);
1212	while (!LIST_EMPTY(&sc->hs_free_list)) {
1213		reqp = LIST_FIRST(&sc->hs_free_list);
1214		LIST_REMOVE(reqp, link);
1215
1216		free(reqp, M_DEVBUF);
1217	}
1218	mtx_unlock(&sc->hs_lock);
1219
1220	while (!LIST_EMPTY(&g_hv_sgl_page_pool.free_sgl_list)) {
1221		sgl_node = LIST_FIRST(&g_hv_sgl_page_pool.free_sgl_list);
1222		LIST_REMOVE(sgl_node, link);
1223		for (j = 0; j < HV_MAX_MULTIPAGE_BUFFER_COUNT; j++){
1224			if (NULL !=
1225			    (void*)sgl_node->sgl_data->sg_segs[j].ss_paddr) {
1226				free((void*)sgl_node->sgl_data->sg_segs[j].ss_paddr, M_DEVBUF);
1227			}
1228		}
1229		sglist_free(sgl_node->sgl_data);
1230		free(sgl_node, M_DEVBUF);
1231	}
1232
1233	return (0);
1234}
1235
1236#if HVS_TIMEOUT_TEST
1237/**
1238 * @brief unit test for timed out operations
1239 *
1240 * This function provides unit testing capability to simulate
1241 * timed out operations.  Recompilation with HV_TIMEOUT_TEST=1
1242 * is required.
1243 *
1244 * @param reqp pointer to a request structure
1245 * @param opcode SCSI operation being performed
1246 * @param wait if 1, wait for I/O to complete
1247 */
1248static void
1249storvsc_timeout_test(struct hv_storvsc_request *reqp,
1250		uint8_t opcode, int wait)
1251{
1252	int ret;
1253	union ccb *ccb = reqp->ccb;
1254	struct storvsc_softc *sc = reqp->softc;
1255
1256	if (reqp->vstor_packet.vm_srb.cdb[0] != opcode) {
1257		return;
1258	}
1259
1260	if (wait) {
1261		mtx_lock(&reqp->event.mtx);
1262	}
1263	ret = hv_storvsc_io_request(sc->hs_dev, reqp);
1264	if (ret != 0) {
1265		if (wait) {
1266			mtx_unlock(&reqp->event.mtx);
1267		}
1268		printf("%s: io_request failed with %d.\n",
1269				__func__, ret);
1270		ccb->ccb_h.status = CAM_PROVIDE_FAIL;
1271		mtx_lock(&sc->hs_lock);
1272		storvsc_free_request(sc, reqp);
1273		xpt_done(ccb);
1274		mtx_unlock(&sc->hs_lock);
1275		return;
1276	}
1277
1278	if (wait) {
1279		xpt_print(ccb->ccb_h.path,
1280				"%u: %s: waiting for IO return.\n",
1281				ticks, __func__);
1282		ret = cv_timedwait(&reqp->event.cv, &reqp->event.mtx, 60*hz);
1283		mtx_unlock(&reqp->event.mtx);
1284		xpt_print(ccb->ccb_h.path, "%u: %s: %s.\n",
1285				ticks, __func__, (ret == 0)?
1286				"IO return detected" :
1287				"IO return not detected");
1288		/*
1289		 * Now both the timer handler and io done are running
1290		 * simultaneously. We want to confirm the io done always
1291		 * finishes after the timer handler exits. So reqp used by
1292		 * timer handler is not freed or stale. Do busy loop for
1293		 * another 1/10 second to make sure io done does
1294		 * wait for the timer handler to complete.
1295		 */
1296		DELAY(100*1000);
1297		mtx_lock(&sc->hs_lock);
1298		xpt_print(ccb->ccb_h.path,
1299				"%u: %s: finishing, queue frozen %d, "
1300				"ccb status 0x%x scsi_status 0x%x.\n",
1301				ticks, __func__, sc->hs_frozen,
1302				ccb->ccb_h.status,
1303				ccb->csio.scsi_status);
1304		mtx_unlock(&sc->hs_lock);
1305	}
1306}
1307#endif /* HVS_TIMEOUT_TEST */
1308
1309#ifdef notyet
1310/**
1311 * @brief timeout handler for requests
1312 *
1313 * This function is called as a result of a callout expiring.
1314 *
1315 * @param arg pointer to a request
1316 */
1317static void
1318storvsc_timeout(void *arg)
1319{
1320	struct hv_storvsc_request *reqp = arg;
1321	struct storvsc_softc *sc = reqp->softc;
1322	union ccb *ccb = reqp->ccb;
1323
1324	if (reqp->retries == 0) {
1325		mtx_lock(&sc->hs_lock);
1326		xpt_print(ccb->ccb_h.path,
1327		    "%u: IO timed out (req=0x%p), wait for another %u secs.\n",
1328		    ticks, reqp, ccb->ccb_h.timeout / 1000);
1329		cam_error_print(ccb, CAM_ESF_ALL, CAM_EPF_ALL);
1330		mtx_unlock(&sc->hs_lock);
1331
1332		reqp->retries++;
1333		callout_reset_sbt(&reqp->callout, SBT_1MS * ccb->ccb_h.timeout,
1334		    0, storvsc_timeout, reqp, 0);
1335#if HVS_TIMEOUT_TEST
1336		storvsc_timeout_test(reqp, SEND_DIAGNOSTIC, 0);
1337#endif
1338		return;
1339	}
1340
1341	mtx_lock(&sc->hs_lock);
1342	xpt_print(ccb->ccb_h.path,
1343		"%u: IO (reqp = 0x%p) did not return for %u seconds, %s.\n",
1344		ticks, reqp, ccb->ccb_h.timeout * (reqp->retries+1) / 1000,
1345		(sc->hs_frozen == 0)?
1346		"freezing the queue" : "the queue is already frozen");
1347	if (sc->hs_frozen == 0) {
1348		sc->hs_frozen = 1;
1349		xpt_freeze_simq(xpt_path_sim(ccb->ccb_h.path), 1);
1350	}
1351	mtx_unlock(&sc->hs_lock);
1352
1353#if HVS_TIMEOUT_TEST
1354	storvsc_timeout_test(reqp, MODE_SELECT_10, 1);
1355#endif
1356}
1357#endif
1358
1359/**
1360 * @brief StorVSC device poll function
1361 *
1362 * This function is responsible for servicing requests when
1363 * interrupts are disabled (i.e when we are dumping core.)
1364 *
1365 * @param sim a pointer to a CAM SCSI interface module
1366 */
1367static void
1368storvsc_poll(struct cam_sim *sim)
1369{
1370	struct storvsc_softc *sc = cam_sim_softc(sim);
1371
1372	mtx_assert(&sc->hs_lock, MA_OWNED);
1373	mtx_unlock(&sc->hs_lock);
1374	hv_storvsc_on_channel_callback(sc->hs_dev->channel);
1375	mtx_lock(&sc->hs_lock);
1376}
1377
1378/**
1379 * @brief StorVSC device action function
1380 *
1381 * This function is responsible for handling SCSI operations which
1382 * are passed from the CAM layer.  The requests are in the form of
1383 * CAM control blocks which indicate the action being performed.
1384 * Not all actions require converting the request to a VSCSI protocol
1385 * message - these actions can be responded to by this driver.
1386 * Requests which are destined for a backend storage device are converted
1387 * to a VSCSI protocol message and sent on the channel connection associated
1388 * with this device.
1389 *
1390 * @param sim pointer to a CAM SCSI interface module
1391 * @param ccb pointer to a CAM control block
1392 */
1393static void
1394storvsc_action(struct cam_sim *sim, union ccb *ccb)
1395{
1396	struct storvsc_softc *sc = cam_sim_softc(sim);
1397	int res;
1398
1399	mtx_assert(&sc->hs_lock, MA_OWNED);
1400	switch (ccb->ccb_h.func_code) {
1401	case XPT_PATH_INQ: {
1402		struct ccb_pathinq *cpi = &ccb->cpi;
1403
1404		cpi->version_num = 1;
1405		cpi->hba_inquiry = PI_TAG_ABLE|PI_SDTR_ABLE;
1406		cpi->target_sprt = 0;
1407		cpi->hba_misc = PIM_NOBUSRESET;
1408		cpi->hba_eng_cnt = 0;
1409		cpi->max_target = STORVSC_MAX_TARGETS;
1410		cpi->max_lun = sc->hs_drv_props->drv_max_luns_per_target;
1411		cpi->initiator_id = cpi->max_target;
1412		cpi->bus_id = cam_sim_bus(sim);
1413		cpi->base_transfer_speed = 300000;
1414		cpi->transport = XPORT_SAS;
1415		cpi->transport_version = 0;
1416		cpi->protocol = PROTO_SCSI;
1417		cpi->protocol_version = SCSI_REV_SPC2;
1418		strncpy(cpi->sim_vid, "FreeBSD", SIM_IDLEN);
1419		strncpy(cpi->hba_vid, sc->hs_drv_props->drv_name, HBA_IDLEN);
1420		strncpy(cpi->dev_name, cam_sim_name(sim), DEV_IDLEN);
1421		cpi->unit_number = cam_sim_unit(sim);
1422
1423		ccb->ccb_h.status = CAM_REQ_CMP;
1424		xpt_done(ccb);
1425		return;
1426	}
1427	case XPT_GET_TRAN_SETTINGS: {
1428		struct  ccb_trans_settings *cts = &ccb->cts;
1429
1430		cts->transport = XPORT_SAS;
1431		cts->transport_version = 0;
1432		cts->protocol = PROTO_SCSI;
1433		cts->protocol_version = SCSI_REV_SPC2;
1434
1435		/* enable tag queuing and disconnected mode */
1436		cts->proto_specific.valid = CTS_SCSI_VALID_TQ;
1437		cts->proto_specific.scsi.valid = CTS_SCSI_VALID_TQ;
1438		cts->proto_specific.scsi.flags = CTS_SCSI_FLAGS_TAG_ENB;
1439		cts->xport_specific.valid = CTS_SPI_VALID_DISC;
1440		cts->xport_specific.spi.flags = CTS_SPI_FLAGS_DISC_ENB;
1441
1442		ccb->ccb_h.status = CAM_REQ_CMP;
1443		xpt_done(ccb);
1444		return;
1445	}
1446	case XPT_SET_TRAN_SETTINGS:	{
1447		ccb->ccb_h.status = CAM_REQ_CMP;
1448		xpt_done(ccb);
1449		return;
1450	}
1451	case XPT_CALC_GEOMETRY:{
1452		cam_calc_geometry(&ccb->ccg, 1);
1453		xpt_done(ccb);
1454		return;
1455	}
1456	case  XPT_RESET_BUS:
1457	case  XPT_RESET_DEV:{
1458#if HVS_HOST_RESET
1459		if ((res = hv_storvsc_host_reset(sc->hs_dev)) != 0) {
1460			xpt_print(ccb->ccb_h.path,
1461				"hv_storvsc_host_reset failed with %d\n", res);
1462			ccb->ccb_h.status = CAM_PROVIDE_FAIL;
1463			xpt_done(ccb);
1464			return;
1465		}
1466		ccb->ccb_h.status = CAM_REQ_CMP;
1467		xpt_done(ccb);
1468		return;
1469#else
1470		xpt_print(ccb->ccb_h.path,
1471				  "%s reset not supported.\n",
1472				  (ccb->ccb_h.func_code == XPT_RESET_BUS)?
1473				  "bus" : "dev");
1474		ccb->ccb_h.status = CAM_REQ_INVALID;
1475		xpt_done(ccb);
1476		return;
1477#endif	/* HVS_HOST_RESET */
1478	}
1479	case XPT_SCSI_IO:
1480	case XPT_IMMED_NOTIFY: {
1481		struct hv_storvsc_request *reqp = NULL;
1482
1483		if (ccb->csio.cdb_len == 0) {
1484			panic("cdl_len is 0\n");
1485		}
1486
1487		if (LIST_EMPTY(&sc->hs_free_list)) {
1488			ccb->ccb_h.status = CAM_REQUEUE_REQ;
1489			if (sc->hs_frozen == 0) {
1490				sc->hs_frozen = 1;
1491				xpt_freeze_simq(sim, /* count*/1);
1492			}
1493			xpt_done(ccb);
1494			return;
1495		}
1496
1497		reqp = LIST_FIRST(&sc->hs_free_list);
1498		LIST_REMOVE(reqp, link);
1499
1500		bzero(reqp, sizeof(struct hv_storvsc_request));
1501		reqp->softc = sc;
1502
1503		ccb->ccb_h.status |= CAM_SIM_QUEUED;
1504		if ((res = create_storvsc_request(ccb, reqp)) != 0) {
1505			ccb->ccb_h.status = CAM_REQ_INVALID;
1506			xpt_done(ccb);
1507			return;
1508		}
1509
1510#ifdef notyet
1511		if (ccb->ccb_h.timeout != CAM_TIME_INFINITY) {
1512			callout_init(&reqp->callout, CALLOUT_MPSAFE);
1513			callout_reset_sbt(&reqp->callout,
1514			    SBT_1MS * ccb->ccb_h.timeout, 0,
1515			    storvsc_timeout, reqp, 0);
1516#if HVS_TIMEOUT_TEST
1517			cv_init(&reqp->event.cv, "storvsc timeout cv");
1518			mtx_init(&reqp->event.mtx, "storvsc timeout mutex",
1519					NULL, MTX_DEF);
1520			switch (reqp->vstor_packet.vm_srb.cdb[0]) {
1521				case MODE_SELECT_10:
1522				case SEND_DIAGNOSTIC:
1523					/* To have timer send the request. */
1524					return;
1525				default:
1526					break;
1527			}
1528#endif /* HVS_TIMEOUT_TEST */
1529		}
1530#endif
1531
1532		if ((res = hv_storvsc_io_request(sc->hs_dev, reqp)) != 0) {
1533			xpt_print(ccb->ccb_h.path,
1534				"hv_storvsc_io_request failed with %d\n", res);
1535			ccb->ccb_h.status = CAM_PROVIDE_FAIL;
1536			storvsc_free_request(sc, reqp);
1537			xpt_done(ccb);
1538			return;
1539		}
1540		return;
1541	}
1542
1543	default:
1544		ccb->ccb_h.status = CAM_REQ_INVALID;
1545		xpt_done(ccb);
1546		return;
1547	}
1548}
1549
1550/**
1551 * @brief destroy bounce buffer
1552 *
1553 * This function is responsible for destroy a Scatter/Gather list
1554 * that create by storvsc_create_bounce_buffer()
1555 *
1556 * @param sgl- the Scatter/Gather need be destroy
1557 * @param sg_count- page count of the SG list.
1558 *
1559 */
1560static void
1561storvsc_destroy_bounce_buffer(struct sglist *sgl)
1562{
1563	struct hv_sgl_node *sgl_node = NULL;
1564	if (LIST_EMPTY(&g_hv_sgl_page_pool.in_use_sgl_list)) {
1565		printf("storvsc error: not enough in use sgl\n");
1566		return;
1567	}
1568	sgl_node = LIST_FIRST(&g_hv_sgl_page_pool.in_use_sgl_list);
1569	LIST_REMOVE(sgl_node, link);
1570	sgl_node->sgl_data = sgl;
1571	LIST_INSERT_HEAD(&g_hv_sgl_page_pool.free_sgl_list, sgl_node, link);
1572}
1573
1574/**
1575 * @brief create bounce buffer
1576 *
1577 * This function is responsible for create a Scatter/Gather list,
1578 * which hold several pages that can be aligned with page size.
1579 *
1580 * @param seg_count- SG-list segments count
1581 * @param write - if WRITE_TYPE, set SG list page used size to 0,
1582 * otherwise set used size to page size.
1583 *
1584 * return NULL if create failed
1585 */
1586static struct sglist *
1587storvsc_create_bounce_buffer(uint16_t seg_count, int write)
1588{
1589	int i = 0;
1590	struct sglist *bounce_sgl = NULL;
1591	unsigned int buf_len = ((write == WRITE_TYPE) ? 0 : PAGE_SIZE);
1592	struct hv_sgl_node *sgl_node = NULL;
1593
1594	/* get struct sglist from free_sgl_list */
1595	if (LIST_EMPTY(&g_hv_sgl_page_pool.free_sgl_list)) {
1596		printf("storvsc error: not enough free sgl\n");
1597		return NULL;
1598	}
1599	sgl_node = LIST_FIRST(&g_hv_sgl_page_pool.free_sgl_list);
1600	LIST_REMOVE(sgl_node, link);
1601	bounce_sgl = sgl_node->sgl_data;
1602	LIST_INSERT_HEAD(&g_hv_sgl_page_pool.in_use_sgl_list, sgl_node, link);
1603
1604	bounce_sgl->sg_maxseg = seg_count;
1605
1606	if (write == WRITE_TYPE)
1607		bounce_sgl->sg_nseg = 0;
1608	else
1609		bounce_sgl->sg_nseg = seg_count;
1610
1611	for (i = 0; i < seg_count; i++)
1612	        bounce_sgl->sg_segs[i].ss_len = buf_len;
1613
1614	return bounce_sgl;
1615}
1616
1617/**
1618 * @brief copy data from SG list to bounce buffer
1619 *
1620 * This function is responsible for copy data from one SG list's segments
1621 * to another SG list which used as bounce buffer.
1622 *
1623 * @param bounce_sgl - the destination SG list
1624 * @param orig_sgl - the segment of the source SG list.
1625 * @param orig_sgl_count - the count of segments.
1626 * @param orig_sgl_count - indicate which segment need bounce buffer,
1627 *  set 1 means need.
1628 *
1629 */
1630static void
1631storvsc_copy_sgl_to_bounce_buf(struct sglist *bounce_sgl,
1632			       bus_dma_segment_t *orig_sgl,
1633			       unsigned int orig_sgl_count,
1634			       uint64_t seg_bits)
1635{
1636	int src_sgl_idx = 0;
1637
1638	for (src_sgl_idx = 0; src_sgl_idx < orig_sgl_count; src_sgl_idx++) {
1639		if (seg_bits & (1 << src_sgl_idx)) {
1640			memcpy((void*)bounce_sgl->sg_segs[src_sgl_idx].ss_paddr,
1641			    (void*)orig_sgl[src_sgl_idx].ds_addr,
1642			    orig_sgl[src_sgl_idx].ds_len);
1643
1644			bounce_sgl->sg_segs[src_sgl_idx].ss_len =
1645			    orig_sgl[src_sgl_idx].ds_len;
1646		}
1647	}
1648}
1649
1650/**
1651 * @brief copy data from SG list which used as bounce to another SG list
1652 *
1653 * This function is responsible for copy data from one SG list with bounce
1654 * buffer to another SG list's segments.
1655 *
1656 * @param dest_sgl - the destination SG list's segments
1657 * @param dest_sgl_count - the count of destination SG list's segment.
1658 * @param src_sgl - the source SG list.
1659 * @param seg_bits - indicate which segment used bounce buffer of src SG-list.
1660 *
1661 */
1662void
1663storvsc_copy_from_bounce_buf_to_sgl(bus_dma_segment_t *dest_sgl,
1664				    unsigned int dest_sgl_count,
1665				    struct sglist* src_sgl,
1666				    uint64_t seg_bits)
1667{
1668	int sgl_idx = 0;
1669
1670	for (sgl_idx = 0; sgl_idx < dest_sgl_count; sgl_idx++) {
1671		if (seg_bits & (1 << sgl_idx)) {
1672			memcpy((void*)(dest_sgl[sgl_idx].ds_addr),
1673			    (void*)(src_sgl->sg_segs[sgl_idx].ss_paddr),
1674			    src_sgl->sg_segs[sgl_idx].ss_len);
1675		}
1676	}
1677}
1678
1679/**
1680 * @brief check SG list with bounce buffer or not
1681 *
1682 * This function is responsible for check if need bounce buffer for SG list.
1683 *
1684 * @param sgl - the SG list's segments
1685 * @param sg_count - the count of SG list's segment.
1686 * @param bits - segmengs number that need bounce buffer
1687 *
1688 * return -1 if SG list needless bounce buffer
1689 */
1690static int
1691storvsc_check_bounce_buffer_sgl(bus_dma_segment_t *sgl,
1692				unsigned int sg_count,
1693				uint64_t *bits)
1694{
1695	int i = 0;
1696	int offset = 0;
1697	uint64_t phys_addr = 0;
1698	uint64_t tmp_bits = 0;
1699	boolean_t found_hole = FALSE;
1700	boolean_t pre_aligned = TRUE;
1701
1702	if (sg_count < 2){
1703		return -1;
1704	}
1705
1706	*bits = 0;
1707
1708	phys_addr = vtophys(sgl[0].ds_addr);
1709	offset =  phys_addr - trunc_page(phys_addr);
1710
1711	if (offset != 0) {
1712		pre_aligned = FALSE;
1713		tmp_bits |= 1;
1714	}
1715
1716	for (i = 1; i < sg_count; i++) {
1717		phys_addr = vtophys(sgl[i].ds_addr);
1718		offset =  phys_addr - trunc_page(phys_addr);
1719
1720		if (offset == 0) {
1721			if (FALSE == pre_aligned){
1722				/*
1723				 * This segment is aligned, if the previous
1724				 * one is not aligned, find a hole
1725				 */
1726				found_hole = TRUE;
1727			}
1728			pre_aligned = TRUE;
1729		} else {
1730			tmp_bits |= 1 << i;
1731			if (!pre_aligned) {
1732				if (phys_addr != vtophys(sgl[i-1].ds_addr +
1733				    sgl[i-1].ds_len)) {
1734					/*
1735					 * Check whether connect to previous
1736					 * segment,if not, find the hole
1737					 */
1738					found_hole = TRUE;
1739				}
1740			} else {
1741				found_hole = TRUE;
1742			}
1743			pre_aligned = FALSE;
1744		}
1745	}
1746
1747	if (!found_hole) {
1748		return (-1);
1749	} else {
1750		*bits = tmp_bits;
1751		return 0;
1752	}
1753}
1754
1755/**
1756 * @brief Fill in a request structure based on a CAM control block
1757 *
1758 * Fills in a request structure based on the contents of a CAM control
1759 * block.  The request structure holds the payload information for
1760 * VSCSI protocol request.
1761 *
1762 * @param ccb pointer to a CAM contorl block
1763 * @param reqp pointer to a request structure
1764 */
1765static int
1766create_storvsc_request(union ccb *ccb, struct hv_storvsc_request *reqp)
1767{
1768	struct ccb_scsiio *csio = &ccb->csio;
1769	uint64_t phys_addr;
1770	uint32_t bytes_to_copy = 0;
1771	uint32_t pfn_num = 0;
1772	uint32_t pfn;
1773	uint64_t not_aligned_seg_bits = 0;
1774
1775	/* refer to struct vmscsi_req for meanings of these two fields */
1776	reqp->vstor_packet.u.vm_srb.port =
1777		cam_sim_unit(xpt_path_sim(ccb->ccb_h.path));
1778	reqp->vstor_packet.u.vm_srb.path_id =
1779		cam_sim_bus(xpt_path_sim(ccb->ccb_h.path));
1780
1781	reqp->vstor_packet.u.vm_srb.target_id = ccb->ccb_h.target_id;
1782	reqp->vstor_packet.u.vm_srb.lun = ccb->ccb_h.target_lun;
1783
1784	reqp->vstor_packet.u.vm_srb.cdb_len = csio->cdb_len;
1785	if(ccb->ccb_h.flags & CAM_CDB_POINTER) {
1786		memcpy(&reqp->vstor_packet.u.vm_srb.u.cdb, csio->cdb_io.cdb_ptr,
1787			csio->cdb_len);
1788	} else {
1789		memcpy(&reqp->vstor_packet.u.vm_srb.u.cdb, csio->cdb_io.cdb_bytes,
1790			csio->cdb_len);
1791	}
1792
1793	switch (ccb->ccb_h.flags & CAM_DIR_MASK) {
1794	case CAM_DIR_OUT:
1795		reqp->vstor_packet.u.vm_srb.data_in = WRITE_TYPE;
1796		break;
1797	case CAM_DIR_IN:
1798		reqp->vstor_packet.u.vm_srb.data_in = READ_TYPE;
1799		break;
1800	case CAM_DIR_NONE:
1801		reqp->vstor_packet.u.vm_srb.data_in = UNKNOWN_TYPE;
1802		break;
1803	default:
1804		reqp->vstor_packet.u.vm_srb.data_in = UNKNOWN_TYPE;
1805		break;
1806	}
1807
1808	reqp->sense_data     = &csio->sense_data;
1809	reqp->sense_info_len = csio->sense_len;
1810
1811	reqp->ccb = ccb;
1812
1813	if (0 == csio->dxfer_len) {
1814		return (0);
1815	}
1816
1817	reqp->data_buf.length = csio->dxfer_len;
1818
1819	switch (ccb->ccb_h.flags & CAM_DATA_MASK) {
1820	case CAM_DATA_VADDR:
1821	{
1822		bytes_to_copy = csio->dxfer_len;
1823		phys_addr = vtophys(csio->data_ptr);
1824		reqp->data_buf.offset = phys_addr & PAGE_MASK;
1825
1826		while (bytes_to_copy != 0) {
1827			int bytes, page_offset;
1828			phys_addr =
1829			    vtophys(&csio->data_ptr[reqp->data_buf.length -
1830			    bytes_to_copy]);
1831			pfn = phys_addr >> PAGE_SHIFT;
1832			reqp->data_buf.pfn_array[pfn_num] = pfn;
1833			page_offset = phys_addr & PAGE_MASK;
1834
1835			bytes = min(PAGE_SIZE - page_offset, bytes_to_copy);
1836
1837			bytes_to_copy -= bytes;
1838			pfn_num++;
1839		}
1840		break;
1841	}
1842
1843	case CAM_DATA_SG:
1844	{
1845		int i = 0;
1846		int offset = 0;
1847		int ret;
1848
1849		bus_dma_segment_t *storvsc_sglist =
1850		    (bus_dma_segment_t *)ccb->csio.data_ptr;
1851		u_int16_t storvsc_sg_count = ccb->csio.sglist_cnt;
1852
1853		printf("Storvsc: get SG I/O operation, %d\n",
1854		    reqp->vstor_packet.u.vm_srb.data_in);
1855
1856		if (storvsc_sg_count > HV_MAX_MULTIPAGE_BUFFER_COUNT){
1857			printf("Storvsc: %d segments is too much, "
1858			    "only support %d segments\n",
1859			    storvsc_sg_count, HV_MAX_MULTIPAGE_BUFFER_COUNT);
1860			return (EINVAL);
1861		}
1862
1863		/*
1864		 * We create our own bounce buffer function currently. Idealy
1865		 * we should use BUS_DMA(9) framework. But with current BUS_DMA
1866		 * code there is no callback API to check the page alignment of
1867		 * middle segments before busdma can decide if a bounce buffer
1868		 * is needed for particular segment. There is callback,
1869		 * "bus_dma_filter_t *filter", but the parrameters are not
1870		 * sufficient for storvsc driver.
1871		 * TODO:
1872		 *	Add page alignment check in BUS_DMA(9) callback. Once
1873		 *	this is complete, switch the following code to use
1874		 *	BUS_DMA(9) for storvsc bounce buffer support.
1875		 */
1876		/* check if we need to create bounce buffer */
1877		ret = storvsc_check_bounce_buffer_sgl(storvsc_sglist,
1878		    storvsc_sg_count, &not_aligned_seg_bits);
1879		if (ret != -1) {
1880			reqp->bounce_sgl =
1881			    storvsc_create_bounce_buffer(storvsc_sg_count,
1882			    reqp->vstor_packet.u.vm_srb.data_in);
1883			if (NULL == reqp->bounce_sgl) {
1884				printf("Storvsc_error: "
1885				    "create bounce buffer failed.\n");
1886				return (ENOMEM);
1887			}
1888
1889			reqp->bounce_sgl_count = storvsc_sg_count;
1890			reqp->not_aligned_seg_bits = not_aligned_seg_bits;
1891
1892			/*
1893			 * if it is write, we need copy the original data
1894			 *to bounce buffer
1895			 */
1896			if (WRITE_TYPE == reqp->vstor_packet.u.vm_srb.data_in) {
1897				storvsc_copy_sgl_to_bounce_buf(
1898				    reqp->bounce_sgl,
1899				    storvsc_sglist,
1900				    storvsc_sg_count,
1901				    reqp->not_aligned_seg_bits);
1902			}
1903
1904			/* transfer virtual address to physical frame number */
1905			if (reqp->not_aligned_seg_bits & 0x1){
1906 				phys_addr =
1907				    vtophys(reqp->bounce_sgl->sg_segs[0].ss_paddr);
1908			}else{
1909 				phys_addr =
1910					vtophys(storvsc_sglist[0].ds_addr);
1911			}
1912			reqp->data_buf.offset = phys_addr & PAGE_MASK;
1913
1914			pfn = phys_addr >> PAGE_SHIFT;
1915			reqp->data_buf.pfn_array[0] = pfn;
1916
1917			for (i = 1; i < storvsc_sg_count; i++) {
1918				if (reqp->not_aligned_seg_bits & (1 << i)) {
1919					phys_addr =
1920					    vtophys(reqp->bounce_sgl->sg_segs[i].ss_paddr);
1921				} else {
1922					phys_addr =
1923					    vtophys(storvsc_sglist[i].ds_addr);
1924				}
1925
1926				pfn = phys_addr >> PAGE_SHIFT;
1927				reqp->data_buf.pfn_array[i] = pfn;
1928			}
1929		} else {
1930			phys_addr = vtophys(storvsc_sglist[0].ds_addr);
1931
1932			reqp->data_buf.offset = phys_addr & PAGE_MASK;
1933
1934			for (i = 0; i < storvsc_sg_count; i++) {
1935				phys_addr = vtophys(storvsc_sglist[i].ds_addr);
1936				pfn = phys_addr >> PAGE_SHIFT;
1937				reqp->data_buf.pfn_array[i] = pfn;
1938			}
1939
1940			/* check the last segment cross boundary or not */
1941			offset = phys_addr & PAGE_MASK;
1942			if (offset) {
1943				phys_addr =
1944				    vtophys(storvsc_sglist[i-1].ds_addr +
1945				    PAGE_SIZE - offset);
1946				pfn = phys_addr >> PAGE_SHIFT;
1947				reqp->data_buf.pfn_array[i] = pfn;
1948			}
1949
1950			reqp->bounce_sgl_count = 0;
1951		}
1952		break;
1953	}
1954	default:
1955		printf("Unknow flags: %d\n", ccb->ccb_h.flags);
1956		return(EINVAL);
1957	}
1958
1959	return(0);
1960}
1961
1962/*
1963 * Modified based on scsi_print_inquiry which is responsible to
1964 * print the detail information for scsi_inquiry_data.
1965 *
1966 * Return 1 if it is valid, 0 otherwise.
1967 */
1968static inline int
1969is_inquiry_valid(const struct scsi_inquiry_data *inq_data)
1970{
1971	uint8_t type;
1972	char vendor[16], product[48], revision[16];
1973
1974	/*
1975	 * Check device type and qualifier
1976	 */
1977	if (!(SID_QUAL_IS_VENDOR_UNIQUE(inq_data) ||
1978	    SID_QUAL(inq_data) == SID_QUAL_LU_CONNECTED))
1979		return (0);
1980
1981	type = SID_TYPE(inq_data);
1982	switch (type) {
1983	case T_DIRECT:
1984	case T_SEQUENTIAL:
1985	case T_PRINTER:
1986	case T_PROCESSOR:
1987	case T_WORM:
1988	case T_CDROM:
1989	case T_SCANNER:
1990	case T_OPTICAL:
1991	case T_CHANGER:
1992	case T_COMM:
1993	case T_STORARRAY:
1994	case T_ENCLOSURE:
1995	case T_RBC:
1996	case T_OCRW:
1997	case T_OSD:
1998	case T_ADC:
1999		break;
2000	case T_NODEVICE:
2001	default:
2002		return (0);
2003	}
2004
2005	/*
2006	 * Check vendor, product, and revision
2007	 */
2008	cam_strvis(vendor, inq_data->vendor, sizeof(inq_data->vendor),
2009	    sizeof(vendor));
2010	cam_strvis(product, inq_data->product, sizeof(inq_data->product),
2011	    sizeof(product));
2012	cam_strvis(revision, inq_data->revision, sizeof(inq_data->revision),
2013	    sizeof(revision));
2014	if (strlen(vendor) == 0  ||
2015	    strlen(product) == 0 ||
2016	    strlen(revision) == 0)
2017		return (0);
2018
2019	return (1);
2020}
2021
2022/**
2023 * @brief completion function before returning to CAM
2024 *
2025 * I/O process has been completed and the result needs
2026 * to be passed to the CAM layer.
2027 * Free resources related to this request.
2028 *
2029 * @param reqp pointer to a request structure
2030 */
2031static void
2032storvsc_io_done(struct hv_storvsc_request *reqp)
2033{
2034	union ccb *ccb = reqp->ccb;
2035	struct ccb_scsiio *csio = &ccb->csio;
2036	struct storvsc_softc *sc = reqp->softc;
2037	struct vmscsi_req *vm_srb = &reqp->vstor_packet.u.vm_srb;
2038	bus_dma_segment_t *ori_sglist = NULL;
2039	int ori_sg_count = 0;
2040
2041	/* destroy bounce buffer if it is used */
2042	if (reqp->bounce_sgl_count) {
2043		ori_sglist = (bus_dma_segment_t *)ccb->csio.data_ptr;
2044		ori_sg_count = ccb->csio.sglist_cnt;
2045
2046		/*
2047		 * If it is READ operation, we should copy back the data
2048		 * to original SG list.
2049		 */
2050		if (READ_TYPE == reqp->vstor_packet.u.vm_srb.data_in) {
2051			storvsc_copy_from_bounce_buf_to_sgl(ori_sglist,
2052			    ori_sg_count,
2053			    reqp->bounce_sgl,
2054			    reqp->not_aligned_seg_bits);
2055		}
2056
2057		storvsc_destroy_bounce_buffer(reqp->bounce_sgl);
2058		reqp->bounce_sgl_count = 0;
2059	}
2060
2061	if (reqp->retries > 0) {
2062		mtx_lock(&sc->hs_lock);
2063#if HVS_TIMEOUT_TEST
2064		xpt_print(ccb->ccb_h.path,
2065			"%u: IO returned after timeout, "
2066			"waking up timer handler if any.\n", ticks);
2067		mtx_lock(&reqp->event.mtx);
2068		cv_signal(&reqp->event.cv);
2069		mtx_unlock(&reqp->event.mtx);
2070#endif
2071		reqp->retries = 0;
2072		xpt_print(ccb->ccb_h.path,
2073			"%u: IO returned after timeout, "
2074			"stopping timer if any.\n", ticks);
2075		mtx_unlock(&sc->hs_lock);
2076	}
2077
2078#ifdef notyet
2079	/*
2080	 * callout_drain() will wait for the timer handler to finish
2081	 * if it is running. So we don't need any lock to synchronize
2082	 * between this routine and the timer handler.
2083	 * Note that we need to make sure reqp is not freed when timer
2084	 * handler is using or will use it.
2085	 */
2086	if (ccb->ccb_h.timeout != CAM_TIME_INFINITY) {
2087		callout_drain(&reqp->callout);
2088	}
2089#endif
2090
2091	ccb->ccb_h.status &= ~CAM_SIM_QUEUED;
2092	ccb->ccb_h.status &= ~CAM_STATUS_MASK;
2093	if (vm_srb->scsi_status == SCSI_STATUS_OK) {
2094		const struct scsi_generic *cmd;
2095
2096		/*
2097		 * Check whether the data for INQUIRY cmd is valid or
2098		 * not.  Windows 10 and Windows 2016 send all zero
2099		 * inquiry data to VM even for unpopulated slots.
2100		 */
2101		cmd = (const struct scsi_generic *)
2102		    ((ccb->ccb_h.flags & CAM_CDB_POINTER) ?
2103		     csio->cdb_io.cdb_ptr : csio->cdb_io.cdb_bytes);
2104		if (cmd->opcode == INQUIRY &&
2105		    /*
2106		     * XXX: Temporary work around disk hot plugin on win2k12r2,
2107		     * only filtering the invalid disk on win10 or 2016 server.
2108		     * So, the hot plugin on win10 and 2016 server needs
2109		     * to be fixed.
2110		     */
2111		    vmstor_proto_version == VMSTOR_PROTOCOL_VERSION_WIN10 &&
2112		    is_inquiry_valid(
2113		    (const struct scsi_inquiry_data *)csio->data_ptr) == 0) {
2114			ccb->ccb_h.status |= CAM_DEV_NOT_THERE;
2115			if (bootverbose) {
2116				mtx_lock(&sc->hs_lock);
2117				xpt_print(ccb->ccb_h.path,
2118				    "storvsc uninstalled device\n");
2119				mtx_unlock(&sc->hs_lock);
2120			}
2121		} else {
2122			ccb->ccb_h.status |= CAM_REQ_CMP;
2123		}
2124	} else {
2125		mtx_lock(&sc->hs_lock);
2126		xpt_print(ccb->ccb_h.path,
2127			"storvsc scsi_status = %d\n",
2128			vm_srb->scsi_status);
2129		mtx_unlock(&sc->hs_lock);
2130		ccb->ccb_h.status |= CAM_SCSI_STATUS_ERROR;
2131	}
2132
2133	ccb->csio.scsi_status = (vm_srb->scsi_status & 0xFF);
2134	ccb->csio.resid = ccb->csio.dxfer_len - vm_srb->transfer_len;
2135
2136	if (reqp->sense_info_len != 0) {
2137		csio->sense_resid = csio->sense_len - reqp->sense_info_len;
2138		ccb->ccb_h.status |= CAM_AUTOSNS_VALID;
2139	}
2140
2141	mtx_lock(&sc->hs_lock);
2142	if (reqp->softc->hs_frozen == 1) {
2143		xpt_print(ccb->ccb_h.path,
2144			"%u: storvsc unfreezing softc 0x%p.\n",
2145			ticks, reqp->softc);
2146		ccb->ccb_h.status |= CAM_RELEASE_SIMQ;
2147		reqp->softc->hs_frozen = 0;
2148	}
2149	storvsc_free_request(sc, reqp);
2150	xpt_done(ccb);
2151	mtx_unlock(&sc->hs_lock);
2152}
2153
2154/**
2155 * @brief Free a request structure
2156 *
2157 * Free a request structure by returning it to the free list
2158 *
2159 * @param sc pointer to a softc
2160 * @param reqp pointer to a request structure
2161 */
2162static void
2163storvsc_free_request(struct storvsc_softc *sc, struct hv_storvsc_request *reqp)
2164{
2165
2166	LIST_INSERT_HEAD(&sc->hs_free_list, reqp, link);
2167}
2168
2169/**
2170 * @brief Determine type of storage device from GUID
2171 *
2172 * Using the type GUID, determine if this is a StorVSC (paravirtual
2173 * SCSI or BlkVSC (paravirtual IDE) device.
2174 *
2175 * @param dev a device
2176 * returns an enum
2177 */
2178static enum hv_storage_type
2179storvsc_get_storage_type(device_t dev)
2180{
2181	const char *p = vmbus_get_type(dev);
2182
2183	if (!memcmp(p, &gBlkVscDeviceType, sizeof(hv_guid))) {
2184		return DRIVER_BLKVSC;
2185	} else if (!memcmp(p, &gStorVscDeviceType, sizeof(hv_guid))) {
2186		return DRIVER_STORVSC;
2187	}
2188	return (DRIVER_UNKNOWN);
2189}
2190
2191