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("$FreeBSD$");
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 STORVSC_WIN7_MAJOR 4
85 #define STORVSC_WIN7_MINOR 2
86 
87 #define STORVSC_WIN8_MAJOR 5
88 #define STORVSC_WIN8_MINOR 1
89 
90 #define VSTOR_PKT_SIZE	(sizeof(struct vstor_packet) - vmscsi_size_delta)
91 
92 #define HV_ALIGN(x, a) roundup2(x, a)
93 
94 struct storvsc_softc;
95 
96 struct hv_sgl_node {
97 	LIST_ENTRY(hv_sgl_node) link;
98 	struct sglist *sgl_data;
99 };
100 
101 struct hv_sgl_page_pool{
102 	LIST_HEAD(, hv_sgl_node) in_use_sgl_list;
103 	LIST_HEAD(, hv_sgl_node) free_sgl_list;
104 	boolean_t                is_init;
105 } g_hv_sgl_page_pool;
106 
107 #define STORVSC_MAX_SG_PAGE_CNT STORVSC_MAX_IO_REQUESTS * HV_MAX_MULTIPAGE_BUFFER_COUNT
108 
109 enum storvsc_request_type {
110 	WRITE_TYPE,
111 	READ_TYPE,
112 	UNKNOWN_TYPE
113 };
114 
115 struct hv_storvsc_request {
116 	LIST_ENTRY(hv_storvsc_request) link;
117 	struct vstor_packet	vstor_packet;
118 	hv_vmbus_multipage_buffer data_buf;
119 	void *sense_data;
120 	uint8_t sense_info_len;
121 	uint8_t retries;
122 	union ccb *ccb;
123 	struct storvsc_softc *softc;
124 	struct callout callout;
125 	struct sema synch_sema; /*Synchronize the request/response if needed */
126 	struct sglist *bounce_sgl;
127 	unsigned int bounce_sgl_count;
128 	uint64_t not_aligned_seg_bits;
129 };
130 
131 struct storvsc_softc {
132 	struct hv_device		*hs_dev;
133 	LIST_HEAD(, hv_storvsc_request)	hs_free_list;
134 	struct mtx			hs_lock;
135 	struct storvsc_driver_props	*hs_drv_props;
136 	int 				hs_unit;
137 	uint32_t			hs_frozen;
138 	struct cam_sim			*hs_sim;
139 	struct cam_path 		*hs_path;
140 	uint32_t			hs_num_out_reqs;
141 	boolean_t			hs_destroy;
142 	boolean_t			hs_drain_notify;
143 	boolean_t			hs_open_multi_channel;
144 	struct sema 			hs_drain_sema;
145 	struct hv_storvsc_request	hs_init_req;
146 	struct hv_storvsc_request	hs_reset_req;
147 };
148 
149 
150 /**
151  * HyperV storvsc timeout testing cases:
152  * a. IO returned after first timeout;
153  * b. IO returned after second timeout and queue freeze;
154  * c. IO returned while timer handler is running
155  * The first can be tested by "sg_senddiag -vv /dev/daX",
156  * and the second and third can be done by
157  * "sg_wr_mode -v -p 08 -c 0,1a -m 0,ff /dev/daX".
158  */
159 #define HVS_TIMEOUT_TEST 0
160 
161 /*
162  * Bus/adapter reset functionality on the Hyper-V host is
163  * buggy and it will be disabled until
164  * it can be further tested.
165  */
166 #define HVS_HOST_RESET 0
167 
168 struct storvsc_driver_props {
169 	char		*drv_name;
170 	char		*drv_desc;
171 	uint8_t		drv_max_luns_per_target;
172 	uint8_t		drv_max_ios_per_target;
173 	uint32_t	drv_ringbuffer_size;
174 };
175 
176 enum hv_storage_type {
177 	DRIVER_BLKVSC,
178 	DRIVER_STORVSC,
179 	DRIVER_UNKNOWN
180 };
181 
182 #define HS_MAX_ADAPTERS 10
183 
184 #define HV_STORAGE_SUPPORTS_MULTI_CHANNEL 0x1
185 
186 /* {ba6163d9-04a1-4d29-b605-72e2ffb1dc7f} */
187 static const hv_guid gStorVscDeviceType={
188 	.data = {0xd9, 0x63, 0x61, 0xba, 0xa1, 0x04, 0x29, 0x4d,
189 		 0xb6, 0x05, 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f}
190 };
191 
192 /* {32412632-86cb-44a2-9b5c-50d1417354f5} */
193 static const hv_guid gBlkVscDeviceType={
194 	.data = {0x32, 0x26, 0x41, 0x32, 0xcb, 0x86, 0xa2, 0x44,
195 		 0x9b, 0x5c, 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5}
196 };
197 
198 static struct storvsc_driver_props g_drv_props_table[] = {
199 	{"blkvsc", "Hyper-V IDE Storage Interface",
200 	 BLKVSC_MAX_IDE_DISKS_PER_TARGET, BLKVSC_MAX_IO_REQUESTS,
201 	 STORVSC_RINGBUFFER_SIZE},
202 	{"storvsc", "Hyper-V SCSI Storage Interface",
203 	 STORVSC_MAX_LUNS_PER_TARGET, STORVSC_MAX_IO_REQUESTS,
204 	 STORVSC_RINGBUFFER_SIZE}
205 };
206 
207 /*
208  * Sense buffer size changed in win8; have a run-time
209  * variable to track the size we should use.
210  */
211 static int sense_buffer_size;
212 
213 /*
214  * The size of the vmscsi_request has changed in win8. The
215  * additional size is for the newly added elements in the
216  * structure. These elements are valid only when we are talking
217  * to a win8 host.
218  * Track the correct size we need to apply.
219  */
220 static int vmscsi_size_delta;
221 
222 static int storvsc_current_major;
223 static int storvsc_current_minor;
224 
225 /* static functions */
226 static int storvsc_probe(device_t dev);
227 static int storvsc_attach(device_t dev);
228 static int storvsc_detach(device_t dev);
229 static void storvsc_poll(struct cam_sim * sim);
230 static void storvsc_action(struct cam_sim * sim, union ccb * ccb);
231 static int create_storvsc_request(union ccb *ccb, struct hv_storvsc_request *reqp);
232 static void storvsc_free_request(struct storvsc_softc *sc, struct hv_storvsc_request *reqp);
233 static enum hv_storage_type storvsc_get_storage_type(device_t dev);
234 static void hv_storvsc_rescan_target(struct storvsc_softc *sc);
235 static void hv_storvsc_on_channel_callback(void *context);
236 static void hv_storvsc_on_iocompletion( struct storvsc_softc *sc,
237 					struct vstor_packet *vstor_packet,
238 					struct hv_storvsc_request *request);
239 static int hv_storvsc_connect_vsp(struct hv_device *device);
240 static void storvsc_io_done(struct hv_storvsc_request *reqp);
241 static void storvsc_copy_sgl_to_bounce_buf(struct sglist *bounce_sgl,
242 				bus_dma_segment_t *orig_sgl,
243 				unsigned int orig_sgl_count,
244 				uint64_t seg_bits);
245 void storvsc_copy_from_bounce_buf_to_sgl(bus_dma_segment_t *dest_sgl,
246 				unsigned int dest_sgl_count,
247 				struct sglist* src_sgl,
248 				uint64_t seg_bits);
249 
250 static device_method_t storvsc_methods[] = {
251 	/* Device interface */
252 	DEVMETHOD(device_probe,		storvsc_probe),
253 	DEVMETHOD(device_attach,	storvsc_attach),
254 	DEVMETHOD(device_detach,	storvsc_detach),
255 	DEVMETHOD(device_shutdown,      bus_generic_shutdown),
256 	DEVMETHOD_END
257 };
258 
259 static driver_t storvsc_driver = {
260 	"storvsc", storvsc_methods, sizeof(struct storvsc_softc),
261 };
262 
263 static devclass_t storvsc_devclass;
264 DRIVER_MODULE(storvsc, vmbus, storvsc_driver, storvsc_devclass, 0, 0);
265 MODULE_VERSION(storvsc, 1);
266 MODULE_DEPEND(storvsc, vmbus, 1, 1, 1);
267 
268 
269 /**
270  * The host is capable of sending messages to us that are
271  * completely unsolicited. So, we need to address the race
272  * condition where we may be in the process of unloading the
273  * driver when the host may send us an unsolicited message.
274  * We address this issue by implementing a sequentially
275  * consistent protocol:
276  *
277  * 1. Channel callback is invoked while holding the the channel lock
278  *    and an unloading driver will reset the channel callback under
279  *    the protection of this channel lock.
280  *
281  * 2. To ensure bounded wait time for unloading a driver, we don't
282  *    permit outgoing traffic once the device is marked as being
283  *    destroyed.
284  *
285  * 3. Once the device is marked as being destroyed, we only
286  *    permit incoming traffic to properly account for
287  *    packets already sent out.
288  */
289 static inline struct storvsc_softc *
290 get_stor_device(struct hv_device *device,
291 				boolean_t outbound)
292 {
293 	struct storvsc_softc *sc;
294 
295 	sc = device_get_softc(device->device);
296 	if (sc == NULL) {
297 		return NULL;
298 	}
299 
300 	if (outbound) {
301 		/*
302 		 * Here we permit outgoing I/O only
303 		 * if the device is not being destroyed.
304 		 */
305 
306 		if (sc->hs_destroy) {
307 			sc = NULL;
308 		}
309 	} else {
310 		/*
311 		 * inbound case; if being destroyed
312 		 * only permit to account for
313 		 * messages already sent out.
314 		 */
315 		if (sc->hs_destroy && (sc->hs_num_out_reqs == 0)) {
316 			sc = NULL;
317 		}
318 	}
319 	return sc;
320 }
321 
322 /**
323  * @brief Callback handler, will be invoked when receive mutil-channel offer
324  *
325  * @param context  new multi-channel
326  */
327 static void
328 storvsc_handle_sc_creation(void *context)
329 {
330 	hv_vmbus_channel *new_channel;
331 	struct hv_device *device;
332 	struct storvsc_softc *sc;
333 	struct vmstor_chan_props props;
334 	int ret = 0;
335 
336 	new_channel = (hv_vmbus_channel *)context;
337 	device = new_channel->primary_channel->device;
338 	sc = get_stor_device(device, TRUE);
339 	if (sc == NULL)
340 		return;
341 
342 	if (FALSE == sc->hs_open_multi_channel)
343 		return;
344 
345 	memset(&props, 0, sizeof(props));
346 
347 	ret = hv_vmbus_channel_open(new_channel,
348 	    sc->hs_drv_props->drv_ringbuffer_size,
349   	    sc->hs_drv_props->drv_ringbuffer_size,
350 	    (void *)&props,
351 	    sizeof(struct vmstor_chan_props),
352 	    hv_storvsc_on_channel_callback,
353 	    new_channel);
354 
355 	return;
356 }
357 
358 /**
359  * @brief Send multi-channel creation request to host
360  *
361  * @param device  a Hyper-V device pointer
362  * @param max_chans  the max channels supported by vmbus
363  */
364 static void
365 storvsc_send_multichannel_request(struct hv_device *dev, int max_chans)
366 {
367 	struct storvsc_softc *sc;
368 	struct hv_storvsc_request *request;
369 	struct vstor_packet *vstor_packet;
370 	int request_channels_cnt = 0;
371 	int ret;
372 
373 	/* get multichannels count that need to create */
374 	request_channels_cnt = MIN(max_chans, mp_ncpus);
375 
376 	sc = get_stor_device(dev, TRUE);
377 	if (sc == NULL) {
378 		printf("Storvsc_error: get sc failed while send mutilchannel "
379 		    "request\n");
380 		return;
381 	}
382 
383 	request = &sc->hs_init_req;
384 
385 	/* Establish a handler for multi-channel */
386 	dev->channel->sc_creation_callback = storvsc_handle_sc_creation;
387 
388 	/* request the host to create multi-channel */
389 	memset(request, 0, sizeof(struct hv_storvsc_request));
390 
391 	sema_init(&request->synch_sema, 0, ("stor_synch_sema"));
392 
393 	vstor_packet = &request->vstor_packet;
394 
395 	vstor_packet->operation = VSTOR_OPERATION_CREATE_MULTI_CHANNELS;
396 	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
397 	vstor_packet->u.multi_channels_cnt = request_channels_cnt;
398 
399 	ret = hv_vmbus_channel_send_packet(
400 	    dev->channel,
401 	    vstor_packet,
402 	    VSTOR_PKT_SIZE,
403 	    (uint64_t)(uintptr_t)request,
404 	    HV_VMBUS_PACKET_TYPE_DATA_IN_BAND,
405 	    HV_VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
406 
407 	/* wait for 5 seconds */
408 	ret = sema_timedwait(&request->synch_sema, 5 * hz);
409 	if (ret != 0) {
410 		printf("Storvsc_error: create multi-channel timeout, %d\n",
411 		    ret);
412 		return;
413 	}
414 
415 	if (vstor_packet->operation != VSTOR_OPERATION_COMPLETEIO ||
416 	    vstor_packet->status != 0) {
417 		printf("Storvsc_error: create multi-channel invalid operation "
418 		    "(%d) or statue (%u)\n",
419 		    vstor_packet->operation, vstor_packet->status);
420 		return;
421 	}
422 
423 	sc->hs_open_multi_channel = TRUE;
424 
425 	if (bootverbose)
426 		printf("Storvsc create multi-channel success!\n");
427 }
428 
429 /**
430  * @brief initialize channel connection to parent partition
431  *
432  * @param dev  a Hyper-V device pointer
433  * @returns  0 on success, non-zero error on failure
434  */
435 static int
436 hv_storvsc_channel_init(struct hv_device *dev)
437 {
438 	int ret = 0;
439 	struct hv_storvsc_request *request;
440 	struct vstor_packet *vstor_packet;
441 	struct storvsc_softc *sc;
442 	uint16_t max_chans = 0;
443 	boolean_t support_multichannel = FALSE;
444 
445 	max_chans = 0;
446 	support_multichannel = FALSE;
447 
448 	sc = get_stor_device(dev, TRUE);
449 	if (sc == NULL)
450 		return (ENODEV);
451 
452 	request = &sc->hs_init_req;
453 	memset(request, 0, sizeof(struct hv_storvsc_request));
454 	vstor_packet = &request->vstor_packet;
455 	request->softc = sc;
456 
457 	/**
458 	 * Initiate the vsc/vsp initialization protocol on the open channel
459 	 */
460 	sema_init(&request->synch_sema, 0, ("stor_synch_sema"));
461 
462 	vstor_packet->operation = VSTOR_OPERATION_BEGININITIALIZATION;
463 	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
464 
465 
466 	ret = hv_vmbus_channel_send_packet(
467 			dev->channel,
468 			vstor_packet,
469 			VSTOR_PKT_SIZE,
470 			(uint64_t)(uintptr_t)request,
471 			HV_VMBUS_PACKET_TYPE_DATA_IN_BAND,
472 			HV_VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
473 
474 	if (ret != 0)
475 		goto cleanup;
476 
477 	/* wait 5 seconds */
478 	ret = sema_timedwait(&request->synch_sema, 5 * hz);
479 	if (ret != 0)
480 		goto cleanup;
481 
482 	if (vstor_packet->operation != VSTOR_OPERATION_COMPLETEIO ||
483 		vstor_packet->status != 0) {
484 		goto cleanup;
485 	}
486 
487 	/* reuse the packet for version range supported */
488 
489 	memset(vstor_packet, 0, sizeof(struct vstor_packet));
490 	vstor_packet->operation = VSTOR_OPERATION_QUERYPROTOCOLVERSION;
491 	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
492 
493 	vstor_packet->u.version.major_minor =
494 	    VMSTOR_PROTOCOL_VERSION(storvsc_current_major, storvsc_current_minor);
495 
496 	/* revision is only significant for Windows guests */
497 	vstor_packet->u.version.revision = 0;
498 
499 	ret = hv_vmbus_channel_send_packet(
500 			dev->channel,
501 			vstor_packet,
502 			VSTOR_PKT_SIZE,
503 			(uint64_t)(uintptr_t)request,
504 			HV_VMBUS_PACKET_TYPE_DATA_IN_BAND,
505 			HV_VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
506 
507 	if (ret != 0)
508 		goto cleanup;
509 
510 	/* wait 5 seconds */
511 	ret = sema_timedwait(&request->synch_sema, 5 * hz);
512 
513 	if (ret)
514 		goto cleanup;
515 
516 	/* TODO: Check returned version */
517 	if (vstor_packet->operation != VSTOR_OPERATION_COMPLETEIO ||
518 		vstor_packet->status != 0)
519 		goto cleanup;
520 
521 	/**
522 	 * Query channel properties
523 	 */
524 	memset(vstor_packet, 0, sizeof(struct vstor_packet));
525 	vstor_packet->operation = VSTOR_OPERATION_QUERYPROPERTIES;
526 	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
527 
528 	ret = hv_vmbus_channel_send_packet(
529 				dev->channel,
530 				vstor_packet,
531 				VSTOR_PKT_SIZE,
532 				(uint64_t)(uintptr_t)request,
533 				HV_VMBUS_PACKET_TYPE_DATA_IN_BAND,
534 				HV_VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
535 
536 	if ( ret != 0)
537 		goto cleanup;
538 
539 	/* wait 5 seconds */
540 	ret = sema_timedwait(&request->synch_sema, 5 * hz);
541 
542 	if (ret != 0)
543 		goto cleanup;
544 
545 	/* TODO: Check returned version */
546 	if (vstor_packet->operation != VSTOR_OPERATION_COMPLETEIO ||
547 	    vstor_packet->status != 0) {
548 		goto cleanup;
549 	}
550 
551 	/* multi-channels feature is supported by WIN8 and above version */
552 	max_chans = vstor_packet->u.chan_props.max_channel_cnt;
553 	if ((hv_vmbus_protocal_version != HV_VMBUS_VERSION_WIN7) &&
554 	    (hv_vmbus_protocal_version != HV_VMBUS_VERSION_WS2008) &&
555 	    (vstor_packet->u.chan_props.flags &
556 	     HV_STORAGE_SUPPORTS_MULTI_CHANNEL)) {
557 		support_multichannel = TRUE;
558 	}
559 
560 	memset(vstor_packet, 0, sizeof(struct vstor_packet));
561 	vstor_packet->operation = VSTOR_OPERATION_ENDINITIALIZATION;
562 	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
563 
564 	ret = hv_vmbus_channel_send_packet(
565 			dev->channel,
566 			vstor_packet,
567 			VSTOR_PKT_SIZE,
568 			(uint64_t)(uintptr_t)request,
569 			HV_VMBUS_PACKET_TYPE_DATA_IN_BAND,
570 			HV_VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
571 
572 	if (ret != 0) {
573 		goto cleanup;
574 	}
575 
576 	/* wait 5 seconds */
577 	ret = sema_timedwait(&request->synch_sema, 5 * hz);
578 
579 	if (ret != 0)
580 		goto cleanup;
581 
582 	if (vstor_packet->operation != VSTOR_OPERATION_COMPLETEIO ||
583 	    vstor_packet->status != 0)
584 		goto cleanup;
585 
586 	/*
587 	 * If multi-channel is supported, send multichannel create
588 	 * request to host.
589 	 */
590 	if (support_multichannel)
591 		storvsc_send_multichannel_request(dev, max_chans);
592 
593 cleanup:
594 	sema_destroy(&request->synch_sema);
595 	return (ret);
596 }
597 
598 /**
599  * @brief Open channel connection to paraent partition StorVSP driver
600  *
601  * Open and initialize channel connection to parent partition StorVSP driver.
602  *
603  * @param pointer to a Hyper-V device
604  * @returns 0 on success, non-zero error on failure
605  */
606 static int
607 hv_storvsc_connect_vsp(struct hv_device *dev)
608 {
609 	int ret = 0;
610 	struct vmstor_chan_props props;
611 	struct storvsc_softc *sc;
612 
613 	sc = device_get_softc(dev->device);
614 
615 	memset(&props, 0, sizeof(struct vmstor_chan_props));
616 
617 	/*
618 	 * Open the channel
619 	 */
620 
621 	ret = hv_vmbus_channel_open(
622 		dev->channel,
623 		sc->hs_drv_props->drv_ringbuffer_size,
624 		sc->hs_drv_props->drv_ringbuffer_size,
625 		(void *)&props,
626 		sizeof(struct vmstor_chan_props),
627 		hv_storvsc_on_channel_callback,
628 		dev->channel);
629 
630 	if (ret != 0) {
631 		return ret;
632 	}
633 
634 	ret = hv_storvsc_channel_init(dev);
635 
636 	return (ret);
637 }
638 
639 #if HVS_HOST_RESET
640 static int
641 hv_storvsc_host_reset(struct hv_device *dev)
642 {
643 	int ret = 0;
644 	struct storvsc_softc *sc;
645 
646 	struct hv_storvsc_request *request;
647 	struct vstor_packet *vstor_packet;
648 
649 	sc = get_stor_device(dev, TRUE);
650 	if (sc == NULL) {
651 		return ENODEV;
652 	}
653 
654 	request = &sc->hs_reset_req;
655 	request->softc = sc;
656 	vstor_packet = &request->vstor_packet;
657 
658 	sema_init(&request->synch_sema, 0, "stor synch sema");
659 
660 	vstor_packet->operation = VSTOR_OPERATION_RESETBUS;
661 	vstor_packet->flags = REQUEST_COMPLETION_FLAG;
662 
663 	ret = hv_vmbus_channel_send_packet(dev->channel,
664 			vstor_packet,
665 			VSTOR_PKT_SIZE,
666 			(uint64_t)(uintptr_t)&sc->hs_reset_req,
667 			HV_VMBUS_PACKET_TYPE_DATA_IN_BAND,
668 			HV_VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
669 
670 	if (ret != 0) {
671 		goto cleanup;
672 	}
673 
674 	ret = sema_timedwait(&request->synch_sema, 5 * hz); /* KYS 5 seconds */
675 
676 	if (ret) {
677 		goto cleanup;
678 	}
679 
680 
681 	/*
682 	 * At this point, all outstanding requests in the adapter
683 	 * should have been flushed out and return to us
684 	 */
685 
686 cleanup:
687 	sema_destroy(&request->synch_sema);
688 	return (ret);
689 }
690 #endif /* HVS_HOST_RESET */
691 
692 /**
693  * @brief Function to initiate an I/O request
694  *
695  * @param device Hyper-V device pointer
696  * @param request pointer to a request structure
697  * @returns 0 on success, non-zero error on failure
698  */
699 static int
700 hv_storvsc_io_request(struct hv_device *device,
701 					  struct hv_storvsc_request *request)
702 {
703 	struct storvsc_softc *sc;
704 	struct vstor_packet *vstor_packet = &request->vstor_packet;
705 	struct hv_vmbus_channel* outgoing_channel = NULL;
706 	int ret = 0;
707 
708 	sc = get_stor_device(device, TRUE);
709 
710 	if (sc == NULL) {
711 		return ENODEV;
712 	}
713 
714 	vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
715 
716 	vstor_packet->u.vm_srb.length = VSTOR_PKT_SIZE;
717 
718 	vstor_packet->u.vm_srb.sense_info_len = sense_buffer_size;
719 
720 	vstor_packet->u.vm_srb.transfer_len = request->data_buf.length;
721 
722 	vstor_packet->operation = VSTOR_OPERATION_EXECUTESRB;
723 
724 	outgoing_channel = vmbus_select_outgoing_channel(device->channel);
725 
726 	mtx_unlock(&request->softc->hs_lock);
727 	if (request->data_buf.length) {
728 		ret = hv_vmbus_channel_send_packet_multipagebuffer(
729 				outgoing_channel,
730 				&request->data_buf,
731 				vstor_packet,
732 				VSTOR_PKT_SIZE,
733 				(uint64_t)(uintptr_t)request);
734 
735 	} else {
736 		ret = hv_vmbus_channel_send_packet(
737 			outgoing_channel,
738 			vstor_packet,
739 			VSTOR_PKT_SIZE,
740 			(uint64_t)(uintptr_t)request,
741 			HV_VMBUS_PACKET_TYPE_DATA_IN_BAND,
742 			HV_VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
743 	}
744 	mtx_lock(&request->softc->hs_lock);
745 
746 	if (ret != 0) {
747 		printf("Unable to send packet %p ret %d", vstor_packet, ret);
748 	} else {
749 		atomic_add_int(&sc->hs_num_out_reqs, 1);
750 	}
751 
752 	return (ret);
753 }
754 
755 
756 /**
757  * Process IO_COMPLETION_OPERATION and ready
758  * the result to be completed for upper layer
759  * processing by the CAM layer.
760  */
761 static void
762 hv_storvsc_on_iocompletion(struct storvsc_softc *sc,
763 			   struct vstor_packet *vstor_packet,
764 			   struct hv_storvsc_request *request)
765 {
766 	struct vmscsi_req *vm_srb;
767 
768 	vm_srb = &vstor_packet->u.vm_srb;
769 
770 	if (((vm_srb->scsi_status & 0xFF) == SCSI_STATUS_CHECK_COND) &&
771 			(vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID)) {
772 		/* Autosense data available */
773 
774 		KASSERT(vm_srb->sense_info_len <= request->sense_info_len,
775 				("vm_srb->sense_info_len <= "
776 				 "request->sense_info_len"));
777 
778 		memcpy(request->sense_data, vm_srb->u.sense_data,
779 			vm_srb->sense_info_len);
780 
781 		request->sense_info_len = vm_srb->sense_info_len;
782 	}
783 
784 	/* Complete request by passing to the CAM layer */
785 	storvsc_io_done(request);
786 	atomic_subtract_int(&sc->hs_num_out_reqs, 1);
787 	if (sc->hs_drain_notify && (sc->hs_num_out_reqs == 0)) {
788 		sema_post(&sc->hs_drain_sema);
789 	}
790 }
791 
792 static void
793 hv_storvsc_rescan_target(struct storvsc_softc *sc)
794 {
795 	path_id_t pathid;
796 	target_id_t targetid;
797 	union ccb *ccb;
798 
799 	pathid = cam_sim_path(sc->hs_sim);
800 	targetid = CAM_TARGET_WILDCARD;
801 
802 	/*
803 	 * Allocate a CCB and schedule a rescan.
804 	 */
805 	ccb = xpt_alloc_ccb_nowait();
806 	if (ccb == NULL) {
807 		printf("unable to alloc CCB for rescan\n");
808 		return;
809 	}
810 
811 	if (xpt_create_path(&ccb->ccb_h.path, NULL, pathid, targetid,
812 	    CAM_LUN_WILDCARD) != CAM_REQ_CMP) {
813 		printf("unable to create path for rescan, pathid: %d,"
814 		    "targetid: %d\n", pathid, targetid);
815 		xpt_free_ccb(ccb);
816 		return;
817 	}
818 
819 	if (targetid == CAM_TARGET_WILDCARD)
820 		ccb->ccb_h.func_code = XPT_SCAN_BUS;
821 	else
822 		ccb->ccb_h.func_code = XPT_SCAN_TGT;
823 
824 	xpt_rescan(ccb);
825 }
826 
827 static void
828 hv_storvsc_on_channel_callback(void *context)
829 {
830 	int ret = 0;
831 	hv_vmbus_channel *channel = (hv_vmbus_channel *)context;
832 	struct hv_device *device = NULL;
833 	struct storvsc_softc *sc;
834 	uint32_t bytes_recvd;
835 	uint64_t request_id;
836 	uint8_t packet[roundup2(sizeof(struct vstor_packet), 8)];
837 	struct hv_storvsc_request *request;
838 	struct vstor_packet *vstor_packet;
839 
840 	if (channel->primary_channel != NULL){
841 		device = channel->primary_channel->device;
842 	} else {
843 		device = channel->device;
844 	}
845 
846 	KASSERT(device, ("device is NULL"));
847 
848 	sc = get_stor_device(device, FALSE);
849 	if (sc == NULL) {
850 		printf("Storvsc_error: get stor device failed.\n");
851 		return;
852 	}
853 
854 	ret = hv_vmbus_channel_recv_packet(
855 			channel,
856 			packet,
857 			roundup2(VSTOR_PKT_SIZE, 8),
858 			&bytes_recvd,
859 			&request_id);
860 
861 	while ((ret == 0) && (bytes_recvd > 0)) {
862 		request = (struct hv_storvsc_request *)(uintptr_t)request_id;
863 
864 		if ((request == &sc->hs_init_req) ||
865 			(request == &sc->hs_reset_req)) {
866 			memcpy(&request->vstor_packet, packet,
867 				   sizeof(struct vstor_packet));
868 			sema_post(&request->synch_sema);
869 		} else {
870 			vstor_packet = (struct vstor_packet *)packet;
871 			switch(vstor_packet->operation) {
872 			case VSTOR_OPERATION_COMPLETEIO:
873 				if (request == NULL)
874 					panic("VMBUS: storvsc received a "
875 					    "packet with NULL request id in "
876 					    "COMPLETEIO operation.");
877 
878 				hv_storvsc_on_iocompletion(sc,
879 							vstor_packet, request);
880 				break;
881 			case VSTOR_OPERATION_REMOVEDEVICE:
882 				printf("VMBUS: storvsc operation %d not "
883 				    "implemented.\n", vstor_packet->operation);
884 				/* TODO: implement */
885 				break;
886 			case VSTOR_OPERATION_ENUMERATE_BUS:
887 				hv_storvsc_rescan_target(sc);
888 				break;
889 			default:
890 				break;
891 			}
892 		}
893 		ret = hv_vmbus_channel_recv_packet(
894 				channel,
895 				packet,
896 				roundup2(VSTOR_PKT_SIZE, 8),
897 				&bytes_recvd,
898 				&request_id);
899 	}
900 }
901 
902 /**
903  * @brief StorVSC probe function
904  *
905  * Device probe function.  Returns 0 if the input device is a StorVSC
906  * device.  Otherwise, a ENXIO is returned.  If the input device is
907  * for BlkVSC (paravirtual IDE) device and this support is disabled in
908  * favor of the emulated ATA/IDE device, return ENXIO.
909  *
910  * @param a device
911  * @returns 0 on success, ENXIO if not a matcing StorVSC device
912  */
913 static int
914 storvsc_probe(device_t dev)
915 {
916 	int ata_disk_enable = 0;
917 	int ret	= ENXIO;
918 
919 	if (hv_vmbus_protocal_version == HV_VMBUS_VERSION_WS2008 ||
920 	    hv_vmbus_protocal_version == HV_VMBUS_VERSION_WIN7) {
921 		sense_buffer_size = PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE;
922 		vmscsi_size_delta = sizeof(struct vmscsi_win8_extension);
923 		storvsc_current_major = STORVSC_WIN7_MAJOR;
924 		storvsc_current_minor = STORVSC_WIN7_MINOR;
925 	} else {
926 		sense_buffer_size = POST_WIN7_STORVSC_SENSE_BUFFER_SIZE;
927 		vmscsi_size_delta = 0;
928 		storvsc_current_major = STORVSC_WIN8_MAJOR;
929 		storvsc_current_minor = STORVSC_WIN8_MINOR;
930 	}
931 
932 	switch (storvsc_get_storage_type(dev)) {
933 	case DRIVER_BLKVSC:
934 		if(bootverbose)
935 			device_printf(dev, "DRIVER_BLKVSC-Emulated ATA/IDE probe\n");
936 		if (!getenv_int("hw.ata.disk_enable", &ata_disk_enable)) {
937 			if(bootverbose)
938 				device_printf(dev,
939 					"Enlightened ATA/IDE detected\n");
940 			ret = BUS_PROBE_DEFAULT;
941 		} else if(bootverbose)
942 			device_printf(dev, "Emulated ATA/IDE set (hw.ata.disk_enable set)\n");
943 		break;
944 	case DRIVER_STORVSC:
945 		if(bootverbose)
946 			device_printf(dev, "Enlightened SCSI device detected\n");
947 		ret = BUS_PROBE_DEFAULT;
948 		break;
949 	default:
950 		ret = ENXIO;
951 	}
952 	return (ret);
953 }
954 
955 /**
956  * @brief StorVSC attach function
957  *
958  * Function responsible for allocating per-device structures,
959  * setting up CAM interfaces and scanning for available LUNs to
960  * be used for SCSI device peripherals.
961  *
962  * @param a device
963  * @returns 0 on success or an error on failure
964  */
965 static int
966 storvsc_attach(device_t dev)
967 {
968 	struct hv_device *hv_dev = vmbus_get_devctx(dev);
969 	enum hv_storage_type stor_type;
970 	struct storvsc_softc *sc;
971 	struct cam_devq *devq;
972 	int ret, i, j;
973 	struct hv_storvsc_request *reqp;
974 	struct root_hold_token *root_mount_token = NULL;
975 	struct hv_sgl_node *sgl_node = NULL;
976 	void *tmp_buff = NULL;
977 
978 	/*
979 	 * We need to serialize storvsc attach calls.
980 	 */
981 	root_mount_token = root_mount_hold("storvsc");
982 
983 	sc = device_get_softc(dev);
984 	if (sc == NULL) {
985 		ret = ENOMEM;
986 		goto cleanup;
987 	}
988 
989 	stor_type = storvsc_get_storage_type(dev);
990 
991 	if (stor_type == DRIVER_UNKNOWN) {
992 		ret = ENODEV;
993 		goto cleanup;
994 	}
995 
996 	bzero(sc, sizeof(struct storvsc_softc));
997 
998 	/* fill in driver specific properties */
999 	sc->hs_drv_props = &g_drv_props_table[stor_type];
1000 
1001 	/* fill in device specific properties */
1002 	sc->hs_unit	= device_get_unit(dev);
1003 	sc->hs_dev	= hv_dev;
1004 	device_set_desc(dev, g_drv_props_table[stor_type].drv_desc);
1005 
1006 	LIST_INIT(&sc->hs_free_list);
1007 	mtx_init(&sc->hs_lock, "hvslck", NULL, MTX_DEF);
1008 
1009 	for (i = 0; i < sc->hs_drv_props->drv_max_ios_per_target; ++i) {
1010 		reqp = malloc(sizeof(struct hv_storvsc_request),
1011 				 M_DEVBUF, M_WAITOK|M_ZERO);
1012 		reqp->softc = sc;
1013 
1014 		LIST_INSERT_HEAD(&sc->hs_free_list, reqp, link);
1015 	}
1016 
1017 	/* create sg-list page pool */
1018 	if (FALSE == g_hv_sgl_page_pool.is_init) {
1019 		g_hv_sgl_page_pool.is_init = TRUE;
1020 		LIST_INIT(&g_hv_sgl_page_pool.in_use_sgl_list);
1021 		LIST_INIT(&g_hv_sgl_page_pool.free_sgl_list);
1022 
1023 		/*
1024 		 * Pre-create SG list, each SG list with
1025 		 * HV_MAX_MULTIPAGE_BUFFER_COUNT segments, each
1026 		 * segment has one page buffer
1027 		 */
1028 		for (i = 0; i < STORVSC_MAX_IO_REQUESTS; i++) {
1029 	        	sgl_node = malloc(sizeof(struct hv_sgl_node),
1030 			    M_DEVBUF, M_WAITOK|M_ZERO);
1031 
1032 			sgl_node->sgl_data =
1033 			    sglist_alloc(HV_MAX_MULTIPAGE_BUFFER_COUNT,
1034 			    M_WAITOK|M_ZERO);
1035 
1036 			for (j = 0; j < HV_MAX_MULTIPAGE_BUFFER_COUNT; j++) {
1037 				tmp_buff = malloc(PAGE_SIZE,
1038 				    M_DEVBUF, M_WAITOK|M_ZERO);
1039 
1040 				sgl_node->sgl_data->sg_segs[j].ss_paddr =
1041 				    (vm_paddr_t)tmp_buff;
1042 			}
1043 
1044 			LIST_INSERT_HEAD(&g_hv_sgl_page_pool.free_sgl_list,
1045 			    sgl_node, link);
1046 		}
1047 	}
1048 
1049 	sc->hs_destroy = FALSE;
1050 	sc->hs_drain_notify = FALSE;
1051 	sc->hs_open_multi_channel = FALSE;
1052 	sema_init(&sc->hs_drain_sema, 0, "Store Drain Sema");
1053 
1054 	ret = hv_storvsc_connect_vsp(hv_dev);
1055 	if (ret != 0) {
1056 		goto cleanup;
1057 	}
1058 
1059 	/*
1060 	 * Create the device queue.
1061 	 * Hyper-V maps each target to one SCSI HBA
1062 	 */
1063 	devq = cam_simq_alloc(sc->hs_drv_props->drv_max_ios_per_target);
1064 	if (devq == NULL) {
1065 		device_printf(dev, "Failed to alloc device queue\n");
1066 		ret = ENOMEM;
1067 		goto cleanup;
1068 	}
1069 
1070 	sc->hs_sim = cam_sim_alloc(storvsc_action,
1071 				storvsc_poll,
1072 				sc->hs_drv_props->drv_name,
1073 				sc,
1074 				sc->hs_unit,
1075 				&sc->hs_lock, 1,
1076 				sc->hs_drv_props->drv_max_ios_per_target,
1077 				devq);
1078 
1079 	if (sc->hs_sim == NULL) {
1080 		device_printf(dev, "Failed to alloc sim\n");
1081 		cam_simq_free(devq);
1082 		ret = ENOMEM;
1083 		goto cleanup;
1084 	}
1085 
1086 	mtx_lock(&sc->hs_lock);
1087 	/* bus_id is set to 0, need to get it from VMBUS channel query? */
1088 	if (xpt_bus_register(sc->hs_sim, dev, 0) != CAM_SUCCESS) {
1089 		cam_sim_free(sc->hs_sim, /*free_devq*/TRUE);
1090 		mtx_unlock(&sc->hs_lock);
1091 		device_printf(dev, "Unable to register SCSI bus\n");
1092 		ret = ENXIO;
1093 		goto cleanup;
1094 	}
1095 
1096 	if (xpt_create_path(&sc->hs_path, /*periph*/NULL,
1097 		 cam_sim_path(sc->hs_sim),
1098 		CAM_TARGET_WILDCARD, CAM_LUN_WILDCARD) != CAM_REQ_CMP) {
1099 		xpt_bus_deregister(cam_sim_path(sc->hs_sim));
1100 		cam_sim_free(sc->hs_sim, /*free_devq*/TRUE);
1101 		mtx_unlock(&sc->hs_lock);
1102 		device_printf(dev, "Unable to create path\n");
1103 		ret = ENXIO;
1104 		goto cleanup;
1105 	}
1106 
1107 	mtx_unlock(&sc->hs_lock);
1108 
1109 	root_mount_rel(root_mount_token);
1110 	return (0);
1111 
1112 
1113 cleanup:
1114 	root_mount_rel(root_mount_token);
1115 	while (!LIST_EMPTY(&sc->hs_free_list)) {
1116 		reqp = LIST_FIRST(&sc->hs_free_list);
1117 		LIST_REMOVE(reqp, link);
1118 		free(reqp, M_DEVBUF);
1119 	}
1120 
1121 	while (!LIST_EMPTY(&g_hv_sgl_page_pool.free_sgl_list)) {
1122 		sgl_node = LIST_FIRST(&g_hv_sgl_page_pool.free_sgl_list);
1123 		LIST_REMOVE(sgl_node, link);
1124 		for (j = 0; j < HV_MAX_MULTIPAGE_BUFFER_COUNT; j++) {
1125 			if (NULL !=
1126 			    (void*)sgl_node->sgl_data->sg_segs[j].ss_paddr) {
1127 				free((void*)sgl_node->sgl_data->sg_segs[j].ss_paddr, M_DEVBUF);
1128 			}
1129 		}
1130 		sglist_free(sgl_node->sgl_data);
1131 		free(sgl_node, M_DEVBUF);
1132 	}
1133 
1134 	return (ret);
1135 }
1136 
1137 /**
1138  * @brief StorVSC device detach function
1139  *
1140  * This function is responsible for safely detaching a
1141  * StorVSC device.  This includes waiting for inbound responses
1142  * to complete and freeing associated per-device structures.
1143  *
1144  * @param dev a device
1145  * returns 0 on success
1146  */
1147 static int
1148 storvsc_detach(device_t dev)
1149 {
1150 	struct storvsc_softc *sc = device_get_softc(dev);
1151 	struct hv_storvsc_request *reqp = NULL;
1152 	struct hv_device *hv_device = vmbus_get_devctx(dev);
1153 	struct hv_sgl_node *sgl_node = NULL;
1154 	int j = 0;
1155 
1156 	mtx_lock(&hv_device->channel->inbound_lock);
1157 	sc->hs_destroy = TRUE;
1158 	mtx_unlock(&hv_device->channel->inbound_lock);
1159 
1160 	/*
1161 	 * At this point, all outbound traffic should be disabled. We
1162 	 * only allow inbound traffic (responses) to proceed so that
1163 	 * outstanding requests can be completed.
1164 	 */
1165 
1166 	sc->hs_drain_notify = TRUE;
1167 	sema_wait(&sc->hs_drain_sema);
1168 	sc->hs_drain_notify = FALSE;
1169 
1170 	/*
1171 	 * Since we have already drained, we don't need to busy wait.
1172 	 * The call to close the channel will reset the callback
1173 	 * under the protection of the incoming channel lock.
1174 	 */
1175 
1176 	hv_vmbus_channel_close(hv_device->channel);
1177 
1178 	mtx_lock(&sc->hs_lock);
1179 	while (!LIST_EMPTY(&sc->hs_free_list)) {
1180 		reqp = LIST_FIRST(&sc->hs_free_list);
1181 		LIST_REMOVE(reqp, link);
1182 
1183 		free(reqp, M_DEVBUF);
1184 	}
1185 	mtx_unlock(&sc->hs_lock);
1186 
1187 	while (!LIST_EMPTY(&g_hv_sgl_page_pool.free_sgl_list)) {
1188 		sgl_node = LIST_FIRST(&g_hv_sgl_page_pool.free_sgl_list);
1189 		LIST_REMOVE(sgl_node, link);
1190 		for (j = 0; j < HV_MAX_MULTIPAGE_BUFFER_COUNT; j++){
1191 			if (NULL !=
1192 			    (void*)sgl_node->sgl_data->sg_segs[j].ss_paddr) {
1193 				free((void*)sgl_node->sgl_data->sg_segs[j].ss_paddr, M_DEVBUF);
1194 			}
1195 		}
1196 		sglist_free(sgl_node->sgl_data);
1197 		free(sgl_node, M_DEVBUF);
1198 	}
1199 
1200 	return (0);
1201 }
1202 
1203 #if HVS_TIMEOUT_TEST
1204 /**
1205  * @brief unit test for timed out operations
1206  *
1207  * This function provides unit testing capability to simulate
1208  * timed out operations.  Recompilation with HV_TIMEOUT_TEST=1
1209  * is required.
1210  *
1211  * @param reqp pointer to a request structure
1212  * @param opcode SCSI operation being performed
1213  * @param wait if 1, wait for I/O to complete
1214  */
1215 static void
1216 storvsc_timeout_test(struct hv_storvsc_request *reqp,
1217 		uint8_t opcode, int wait)
1218 {
1219 	int ret;
1220 	union ccb *ccb = reqp->ccb;
1221 	struct storvsc_softc *sc = reqp->softc;
1222 
1223 	if (reqp->vstor_packet.vm_srb.cdb[0] != opcode) {
1224 		return;
1225 	}
1226 
1227 	if (wait) {
1228 		mtx_lock(&reqp->event.mtx);
1229 	}
1230 	ret = hv_storvsc_io_request(sc->hs_dev, reqp);
1231 	if (ret != 0) {
1232 		if (wait) {
1233 			mtx_unlock(&reqp->event.mtx);
1234 		}
1235 		printf("%s: io_request failed with %d.\n",
1236 				__func__, ret);
1237 		ccb->ccb_h.status = CAM_PROVIDE_FAIL;
1238 		mtx_lock(&sc->hs_lock);
1239 		storvsc_free_request(sc, reqp);
1240 		xpt_done(ccb);
1241 		mtx_unlock(&sc->hs_lock);
1242 		return;
1243 	}
1244 
1245 	if (wait) {
1246 		xpt_print(ccb->ccb_h.path,
1247 				"%u: %s: waiting for IO return.\n",
1248 				ticks, __func__);
1249 		ret = cv_timedwait(&reqp->event.cv, &reqp->event.mtx, 60*hz);
1250 		mtx_unlock(&reqp->event.mtx);
1251 		xpt_print(ccb->ccb_h.path, "%u: %s: %s.\n",
1252 				ticks, __func__, (ret == 0)?
1253 				"IO return detected" :
1254 				"IO return not detected");
1255 		/*
1256 		 * Now both the timer handler and io done are running
1257 		 * simultaneously. We want to confirm the io done always
1258 		 * finishes after the timer handler exits. So reqp used by
1259 		 * timer handler is not freed or stale. Do busy loop for
1260 		 * another 1/10 second to make sure io done does
1261 		 * wait for the timer handler to complete.
1262 		 */
1263 		DELAY(100*1000);
1264 		mtx_lock(&sc->hs_lock);
1265 		xpt_print(ccb->ccb_h.path,
1266 				"%u: %s: finishing, queue frozen %d, "
1267 				"ccb status 0x%x scsi_status 0x%x.\n",
1268 				ticks, __func__, sc->hs_frozen,
1269 				ccb->ccb_h.status,
1270 				ccb->csio.scsi_status);
1271 		mtx_unlock(&sc->hs_lock);
1272 	}
1273 }
1274 #endif /* HVS_TIMEOUT_TEST */
1275 
1276 /**
1277  * @brief timeout handler for requests
1278  *
1279  * This function is called as a result of a callout expiring.
1280  *
1281  * @param arg pointer to a request
1282  */
1283 static void
1284 storvsc_timeout(void *arg)
1285 {
1286 	struct hv_storvsc_request *reqp = arg;
1287 	struct storvsc_softc *sc = reqp->softc;
1288 	union ccb *ccb = reqp->ccb;
1289 
1290 	if (reqp->retries == 0) {
1291 		mtx_lock(&sc->hs_lock);
1292 		xpt_print(ccb->ccb_h.path,
1293 		    "%u: IO timed out (req=0x%p), wait for another %u secs.\n",
1294 		    ticks, reqp, ccb->ccb_h.timeout / 1000);
1295 		cam_error_print(ccb, CAM_ESF_ALL, CAM_EPF_ALL);
1296 		mtx_unlock(&sc->hs_lock);
1297 
1298 		reqp->retries++;
1299 		callout_reset_sbt(&reqp->callout, SBT_1MS * ccb->ccb_h.timeout,
1300 		    0, storvsc_timeout, reqp, 0);
1301 #if HVS_TIMEOUT_TEST
1302 		storvsc_timeout_test(reqp, SEND_DIAGNOSTIC, 0);
1303 #endif
1304 		return;
1305 	}
1306 
1307 	mtx_lock(&sc->hs_lock);
1308 	xpt_print(ccb->ccb_h.path,
1309 		"%u: IO (reqp = 0x%p) did not return for %u seconds, %s.\n",
1310 		ticks, reqp, ccb->ccb_h.timeout * (reqp->retries+1) / 1000,
1311 		(sc->hs_frozen == 0)?
1312 		"freezing the queue" : "the queue is already frozen");
1313 	if (sc->hs_frozen == 0) {
1314 		sc->hs_frozen = 1;
1315 		xpt_freeze_simq(xpt_path_sim(ccb->ccb_h.path), 1);
1316 	}
1317 	mtx_unlock(&sc->hs_lock);
1318 
1319 #if HVS_TIMEOUT_TEST
1320 	storvsc_timeout_test(reqp, MODE_SELECT_10, 1);
1321 #endif
1322 }
1323 
1324 /**
1325  * @brief StorVSC device poll function
1326  *
1327  * This function is responsible for servicing requests when
1328  * interrupts are disabled (i.e when we are dumping core.)
1329  *
1330  * @param sim a pointer to a CAM SCSI interface module
1331  */
1332 static void
1333 storvsc_poll(struct cam_sim *sim)
1334 {
1335 	struct storvsc_softc *sc = cam_sim_softc(sim);
1336 
1337 	mtx_assert(&sc->hs_lock, MA_OWNED);
1338 	mtx_unlock(&sc->hs_lock);
1339 	hv_storvsc_on_channel_callback(sc->hs_dev->channel);
1340 	mtx_lock(&sc->hs_lock);
1341 }
1342 
1343 /**
1344  * @brief StorVSC device action function
1345  *
1346  * This function is responsible for handling SCSI operations which
1347  * are passed from the CAM layer.  The requests are in the form of
1348  * CAM control blocks which indicate the action being performed.
1349  * Not all actions require converting the request to a VSCSI protocol
1350  * message - these actions can be responded to by this driver.
1351  * Requests which are destined for a backend storage device are converted
1352  * to a VSCSI protocol message and sent on the channel connection associated
1353  * with this device.
1354  *
1355  * @param sim pointer to a CAM SCSI interface module
1356  * @param ccb pointer to a CAM control block
1357  */
1358 static void
1359 storvsc_action(struct cam_sim *sim, union ccb *ccb)
1360 {
1361 	struct storvsc_softc *sc = cam_sim_softc(sim);
1362 	int res;
1363 
1364 	mtx_assert(&sc->hs_lock, MA_OWNED);
1365 	switch (ccb->ccb_h.func_code) {
1366 	case XPT_PATH_INQ: {
1367 		struct ccb_pathinq *cpi = &ccb->cpi;
1368 
1369 		cpi->version_num = 1;
1370 		cpi->hba_inquiry = PI_TAG_ABLE|PI_SDTR_ABLE;
1371 		cpi->target_sprt = 0;
1372 		cpi->hba_misc = PIM_NOBUSRESET;
1373 		cpi->hba_eng_cnt = 0;
1374 		cpi->max_target = STORVSC_MAX_TARGETS;
1375 		cpi->max_lun = sc->hs_drv_props->drv_max_luns_per_target;
1376 		cpi->initiator_id = cpi->max_target;
1377 		cpi->bus_id = cam_sim_bus(sim);
1378 		cpi->base_transfer_speed = 300000;
1379 		cpi->transport = XPORT_SAS;
1380 		cpi->transport_version = 0;
1381 		cpi->protocol = PROTO_SCSI;
1382 		cpi->protocol_version = SCSI_REV_SPC2;
1383 		strncpy(cpi->sim_vid, "FreeBSD", SIM_IDLEN);
1384 		strncpy(cpi->hba_vid, sc->hs_drv_props->drv_name, HBA_IDLEN);
1385 		strncpy(cpi->dev_name, cam_sim_name(sim), DEV_IDLEN);
1386 		cpi->unit_number = cam_sim_unit(sim);
1387 
1388 		ccb->ccb_h.status = CAM_REQ_CMP;
1389 		xpt_done(ccb);
1390 		return;
1391 	}
1392 	case XPT_GET_TRAN_SETTINGS: {
1393 		struct  ccb_trans_settings *cts = &ccb->cts;
1394 
1395 		cts->transport = XPORT_SAS;
1396 		cts->transport_version = 0;
1397 		cts->protocol = PROTO_SCSI;
1398 		cts->protocol_version = SCSI_REV_SPC2;
1399 
1400 		/* enable tag queuing and disconnected mode */
1401 		cts->proto_specific.valid = CTS_SCSI_VALID_TQ;
1402 		cts->proto_specific.scsi.valid = CTS_SCSI_VALID_TQ;
1403 		cts->proto_specific.scsi.flags = CTS_SCSI_FLAGS_TAG_ENB;
1404 		cts->xport_specific.valid = CTS_SPI_VALID_DISC;
1405 		cts->xport_specific.spi.flags = CTS_SPI_FLAGS_DISC_ENB;
1406 
1407 		ccb->ccb_h.status = CAM_REQ_CMP;
1408 		xpt_done(ccb);
1409 		return;
1410 	}
1411 	case XPT_SET_TRAN_SETTINGS:	{
1412 		ccb->ccb_h.status = CAM_REQ_CMP;
1413 		xpt_done(ccb);
1414 		return;
1415 	}
1416 	case XPT_CALC_GEOMETRY:{
1417 		cam_calc_geometry(&ccb->ccg, 1);
1418 		xpt_done(ccb);
1419 		return;
1420 	}
1421 	case  XPT_RESET_BUS:
1422 	case  XPT_RESET_DEV:{
1423 #if HVS_HOST_RESET
1424 		if ((res = hv_storvsc_host_reset(sc->hs_dev)) != 0) {
1425 			xpt_print(ccb->ccb_h.path,
1426 				"hv_storvsc_host_reset failed with %d\n", res);
1427 			ccb->ccb_h.status = CAM_PROVIDE_FAIL;
1428 			xpt_done(ccb);
1429 			return;
1430 		}
1431 		ccb->ccb_h.status = CAM_REQ_CMP;
1432 		xpt_done(ccb);
1433 		return;
1434 #else
1435 		xpt_print(ccb->ccb_h.path,
1436 				  "%s reset not supported.\n",
1437 				  (ccb->ccb_h.func_code == XPT_RESET_BUS)?
1438 				  "bus" : "dev");
1439 		ccb->ccb_h.status = CAM_REQ_INVALID;
1440 		xpt_done(ccb);
1441 		return;
1442 #endif	/* HVS_HOST_RESET */
1443 	}
1444 	case XPT_SCSI_IO:
1445 	case XPT_IMMED_NOTIFY: {
1446 		struct hv_storvsc_request *reqp = NULL;
1447 
1448 		if (ccb->csio.cdb_len == 0) {
1449 			panic("cdl_len is 0\n");
1450 		}
1451 
1452 		if (LIST_EMPTY(&sc->hs_free_list)) {
1453 			ccb->ccb_h.status = CAM_REQUEUE_REQ;
1454 			if (sc->hs_frozen == 0) {
1455 				sc->hs_frozen = 1;
1456 				xpt_freeze_simq(sim, /* count*/1);
1457 			}
1458 			xpt_done(ccb);
1459 			return;
1460 		}
1461 
1462 		reqp = LIST_FIRST(&sc->hs_free_list);
1463 		LIST_REMOVE(reqp, link);
1464 
1465 		bzero(reqp, sizeof(struct hv_storvsc_request));
1466 		reqp->softc = sc;
1467 
1468 		ccb->ccb_h.status |= CAM_SIM_QUEUED;
1469 		if ((res = create_storvsc_request(ccb, reqp)) != 0) {
1470 			ccb->ccb_h.status = CAM_REQ_INVALID;
1471 			xpt_done(ccb);
1472 			return;
1473 		}
1474 
1475 		if (ccb->ccb_h.timeout != CAM_TIME_INFINITY) {
1476 			callout_init(&reqp->callout, 1);
1477 			callout_reset_sbt(&reqp->callout,
1478 			    SBT_1MS * ccb->ccb_h.timeout, 0,
1479 			    storvsc_timeout, reqp, 0);
1480 #if HVS_TIMEOUT_TEST
1481 			cv_init(&reqp->event.cv, "storvsc timeout cv");
1482 			mtx_init(&reqp->event.mtx, "storvsc timeout mutex",
1483 					NULL, MTX_DEF);
1484 			switch (reqp->vstor_packet.vm_srb.cdb[0]) {
1485 				case MODE_SELECT_10:
1486 				case SEND_DIAGNOSTIC:
1487 					/* To have timer send the request. */
1488 					return;
1489 				default:
1490 					break;
1491 			}
1492 #endif /* HVS_TIMEOUT_TEST */
1493 		}
1494 
1495 		if ((res = hv_storvsc_io_request(sc->hs_dev, reqp)) != 0) {
1496 			xpt_print(ccb->ccb_h.path,
1497 				"hv_storvsc_io_request failed with %d\n", res);
1498 			ccb->ccb_h.status = CAM_PROVIDE_FAIL;
1499 			storvsc_free_request(sc, reqp);
1500 			xpt_done(ccb);
1501 			return;
1502 		}
1503 		return;
1504 	}
1505 
1506 	default:
1507 		ccb->ccb_h.status = CAM_REQ_INVALID;
1508 		xpt_done(ccb);
1509 		return;
1510 	}
1511 }
1512 
1513 /**
1514  * @brief destroy bounce buffer
1515  *
1516  * This function is responsible for destroy a Scatter/Gather list
1517  * that create by storvsc_create_bounce_buffer()
1518  *
1519  * @param sgl- the Scatter/Gather need be destroy
1520  * @param sg_count- page count of the SG list.
1521  *
1522  */
1523 static void
1524 storvsc_destroy_bounce_buffer(struct sglist *sgl)
1525 {
1526 	struct hv_sgl_node *sgl_node = NULL;
1527 
1528 	sgl_node = LIST_FIRST(&g_hv_sgl_page_pool.in_use_sgl_list);
1529 	LIST_REMOVE(sgl_node, link);
1530 	if (NULL == sgl_node) {
1531 		printf("storvsc error: not enough in use sgl\n");
1532 		return;
1533 	}
1534 	sgl_node->sgl_data = sgl;
1535 	LIST_INSERT_HEAD(&g_hv_sgl_page_pool.free_sgl_list, sgl_node, link);
1536 }
1537 
1538 /**
1539  * @brief create bounce buffer
1540  *
1541  * This function is responsible for create a Scatter/Gather list,
1542  * which hold several pages that can be aligned with page size.
1543  *
1544  * @param seg_count- SG-list segments count
1545  * @param write - if WRITE_TYPE, set SG list page used size to 0,
1546  * otherwise set used size to page size.
1547  *
1548  * return NULL if create failed
1549  */
1550 static struct sglist *
1551 storvsc_create_bounce_buffer(uint16_t seg_count, int write)
1552 {
1553 	int i = 0;
1554 	struct sglist *bounce_sgl = NULL;
1555 	unsigned int buf_len = ((write == WRITE_TYPE) ? 0 : PAGE_SIZE);
1556 	struct hv_sgl_node *sgl_node = NULL;
1557 
1558 	/* get struct sglist from free_sgl_list */
1559 	sgl_node = LIST_FIRST(&g_hv_sgl_page_pool.free_sgl_list);
1560 	LIST_REMOVE(sgl_node, link);
1561 	if (NULL == sgl_node) {
1562 		printf("storvsc error: not enough free sgl\n");
1563 		return NULL;
1564 	}
1565 	bounce_sgl = sgl_node->sgl_data;
1566 	LIST_INSERT_HEAD(&g_hv_sgl_page_pool.in_use_sgl_list, sgl_node, link);
1567 
1568 	bounce_sgl->sg_maxseg = seg_count;
1569 
1570 	if (write == WRITE_TYPE)
1571 		bounce_sgl->sg_nseg = 0;
1572 	else
1573 		bounce_sgl->sg_nseg = seg_count;
1574 
1575 	for (i = 0; i < seg_count; i++)
1576 	        bounce_sgl->sg_segs[i].ss_len = buf_len;
1577 
1578 	return bounce_sgl;
1579 }
1580 
1581 /**
1582  * @brief copy data from SG list to bounce buffer
1583  *
1584  * This function is responsible for copy data from one SG list's segments
1585  * to another SG list which used as bounce buffer.
1586  *
1587  * @param bounce_sgl - the destination SG list
1588  * @param orig_sgl - the segment of the source SG list.
1589  * @param orig_sgl_count - the count of segments.
1590  * @param orig_sgl_count - indicate which segment need bounce buffer,
1591  *  set 1 means need.
1592  *
1593  */
1594 static void
1595 storvsc_copy_sgl_to_bounce_buf(struct sglist *bounce_sgl,
1596 			       bus_dma_segment_t *orig_sgl,
1597 			       unsigned int orig_sgl_count,
1598 			       uint64_t seg_bits)
1599 {
1600 	int src_sgl_idx = 0;
1601 
1602 	for (src_sgl_idx = 0; src_sgl_idx < orig_sgl_count; src_sgl_idx++) {
1603 		if (seg_bits & (1 << src_sgl_idx)) {
1604 			memcpy((void*)bounce_sgl->sg_segs[src_sgl_idx].ss_paddr,
1605 			    (void*)orig_sgl[src_sgl_idx].ds_addr,
1606 			    orig_sgl[src_sgl_idx].ds_len);
1607 
1608 			bounce_sgl->sg_segs[src_sgl_idx].ss_len =
1609 			    orig_sgl[src_sgl_idx].ds_len;
1610 		}
1611 	}
1612 }
1613 
1614 /**
1615  * @brief copy data from SG list which used as bounce to another SG list
1616  *
1617  * This function is responsible for copy data from one SG list with bounce
1618  * buffer to another SG list's segments.
1619  *
1620  * @param dest_sgl - the destination SG list's segments
1621  * @param dest_sgl_count - the count of destination SG list's segment.
1622  * @param src_sgl - the source SG list.
1623  * @param seg_bits - indicate which segment used bounce buffer of src SG-list.
1624  *
1625  */
1626 void
1627 storvsc_copy_from_bounce_buf_to_sgl(bus_dma_segment_t *dest_sgl,
1628 				    unsigned int dest_sgl_count,
1629 				    struct sglist* src_sgl,
1630 				    uint64_t seg_bits)
1631 {
1632 	int sgl_idx = 0;
1633 
1634 	for (sgl_idx = 0; sgl_idx < dest_sgl_count; sgl_idx++) {
1635 		if (seg_bits & (1 << sgl_idx)) {
1636 			memcpy((void*)(dest_sgl[sgl_idx].ds_addr),
1637 			    (void*)(src_sgl->sg_segs[sgl_idx].ss_paddr),
1638 			    src_sgl->sg_segs[sgl_idx].ss_len);
1639 		}
1640 	}
1641 }
1642 
1643 /**
1644  * @brief check SG list with bounce buffer or not
1645  *
1646  * This function is responsible for check if need bounce buffer for SG list.
1647  *
1648  * @param sgl - the SG list's segments
1649  * @param sg_count - the count of SG list's segment.
1650  * @param bits - segmengs number that need bounce buffer
1651  *
1652  * return -1 if SG list needless bounce buffer
1653  */
1654 static int
1655 storvsc_check_bounce_buffer_sgl(bus_dma_segment_t *sgl,
1656 				unsigned int sg_count,
1657 				uint64_t *bits)
1658 {
1659 	int i = 0;
1660 	int offset = 0;
1661 	uint64_t phys_addr = 0;
1662 	uint64_t tmp_bits = 0;
1663 	boolean_t found_hole = FALSE;
1664 	boolean_t pre_aligned = TRUE;
1665 
1666 	if (sg_count < 2){
1667 		return -1;
1668 	}
1669 
1670 	*bits = 0;
1671 
1672 	phys_addr = vtophys(sgl[0].ds_addr);
1673 	offset =  phys_addr - trunc_page(phys_addr);
1674 
1675 	if (offset != 0) {
1676 		pre_aligned = FALSE;
1677 		tmp_bits |= 1;
1678 	}
1679 
1680 	for (i = 1; i < sg_count; i++) {
1681 		phys_addr = vtophys(sgl[i].ds_addr);
1682 		offset =  phys_addr - trunc_page(phys_addr);
1683 
1684 		if (offset == 0) {
1685 			if (FALSE == pre_aligned){
1686 				/*
1687 				 * This segment is aligned, if the previous
1688 				 * one is not aligned, find a hole
1689 				 */
1690 				found_hole = TRUE;
1691 			}
1692 			pre_aligned = TRUE;
1693 		} else {
1694 			tmp_bits |= 1 << i;
1695 			if (!pre_aligned) {
1696 				if (phys_addr != vtophys(sgl[i-1].ds_addr +
1697 				    sgl[i-1].ds_len)) {
1698 					/*
1699 					 * Check whether connect to previous
1700 					 * segment,if not, find the hole
1701 					 */
1702 					found_hole = TRUE;
1703 				}
1704 			} else {
1705 				found_hole = TRUE;
1706 			}
1707 			pre_aligned = FALSE;
1708 		}
1709 	}
1710 
1711 	if (!found_hole) {
1712 		return (-1);
1713 	} else {
1714 		*bits = tmp_bits;
1715 		return 0;
1716 	}
1717 }
1718 
1719 /**
1720  * @brief Fill in a request structure based on a CAM control block
1721  *
1722  * Fills in a request structure based on the contents of a CAM control
1723  * block.  The request structure holds the payload information for
1724  * VSCSI protocol request.
1725  *
1726  * @param ccb pointer to a CAM contorl block
1727  * @param reqp pointer to a request structure
1728  */
1729 static int
1730 create_storvsc_request(union ccb *ccb, struct hv_storvsc_request *reqp)
1731 {
1732 	struct ccb_scsiio *csio = &ccb->csio;
1733 	uint64_t phys_addr;
1734 	uint32_t bytes_to_copy = 0;
1735 	uint32_t pfn_num = 0;
1736 	uint32_t pfn;
1737 	uint64_t not_aligned_seg_bits = 0;
1738 
1739 	/* refer to struct vmscsi_req for meanings of these two fields */
1740 	reqp->vstor_packet.u.vm_srb.port =
1741 		cam_sim_unit(xpt_path_sim(ccb->ccb_h.path));
1742 	reqp->vstor_packet.u.vm_srb.path_id =
1743 		cam_sim_bus(xpt_path_sim(ccb->ccb_h.path));
1744 
1745 	reqp->vstor_packet.u.vm_srb.target_id = ccb->ccb_h.target_id;
1746 	reqp->vstor_packet.u.vm_srb.lun = ccb->ccb_h.target_lun;
1747 
1748 	reqp->vstor_packet.u.vm_srb.cdb_len = csio->cdb_len;
1749 	if(ccb->ccb_h.flags & CAM_CDB_POINTER) {
1750 		memcpy(&reqp->vstor_packet.u.vm_srb.u.cdb, csio->cdb_io.cdb_ptr,
1751 			csio->cdb_len);
1752 	} else {
1753 		memcpy(&reqp->vstor_packet.u.vm_srb.u.cdb, csio->cdb_io.cdb_bytes,
1754 			csio->cdb_len);
1755 	}
1756 
1757 	switch (ccb->ccb_h.flags & CAM_DIR_MASK) {
1758 	case CAM_DIR_OUT:
1759 		reqp->vstor_packet.u.vm_srb.data_in = WRITE_TYPE;
1760 		break;
1761 	case CAM_DIR_IN:
1762 		reqp->vstor_packet.u.vm_srb.data_in = READ_TYPE;
1763 		break;
1764 	case CAM_DIR_NONE:
1765 		reqp->vstor_packet.u.vm_srb.data_in = UNKNOWN_TYPE;
1766 		break;
1767 	default:
1768 		reqp->vstor_packet.u.vm_srb.data_in = UNKNOWN_TYPE;
1769 		break;
1770 	}
1771 
1772 	reqp->sense_data     = &csio->sense_data;
1773 	reqp->sense_info_len = csio->sense_len;
1774 
1775 	reqp->ccb = ccb;
1776 
1777 	if (0 == csio->dxfer_len) {
1778 		return (0);
1779 	}
1780 
1781 	reqp->data_buf.length = csio->dxfer_len;
1782 
1783 	switch (ccb->ccb_h.flags & CAM_DATA_MASK) {
1784 	case CAM_DATA_VADDR:
1785 	{
1786 		bytes_to_copy = csio->dxfer_len;
1787 		phys_addr = vtophys(csio->data_ptr);
1788 		reqp->data_buf.offset = phys_addr & PAGE_MASK;
1789 
1790 		while (bytes_to_copy != 0) {
1791 			int bytes, page_offset;
1792 			phys_addr =
1793 			    vtophys(&csio->data_ptr[reqp->data_buf.length -
1794 			    bytes_to_copy]);
1795 			pfn = phys_addr >> PAGE_SHIFT;
1796 			reqp->data_buf.pfn_array[pfn_num] = pfn;
1797 			page_offset = phys_addr & PAGE_MASK;
1798 
1799 			bytes = min(PAGE_SIZE - page_offset, bytes_to_copy);
1800 
1801 			bytes_to_copy -= bytes;
1802 			pfn_num++;
1803 		}
1804 		break;
1805 	}
1806 
1807 	case CAM_DATA_SG:
1808 	{
1809 		int i = 0;
1810 		int offset = 0;
1811 		int ret;
1812 
1813 		bus_dma_segment_t *storvsc_sglist =
1814 		    (bus_dma_segment_t *)ccb->csio.data_ptr;
1815 		u_int16_t storvsc_sg_count = ccb->csio.sglist_cnt;
1816 
1817 		printf("Storvsc: get SG I/O operation, %d\n",
1818 		    reqp->vstor_packet.u.vm_srb.data_in);
1819 
1820 		if (storvsc_sg_count > HV_MAX_MULTIPAGE_BUFFER_COUNT){
1821 			printf("Storvsc: %d segments is too much, "
1822 			    "only support %d segments\n",
1823 			    storvsc_sg_count, HV_MAX_MULTIPAGE_BUFFER_COUNT);
1824 			return (EINVAL);
1825 		}
1826 
1827 		/*
1828 		 * We create our own bounce buffer function currently. Idealy
1829 		 * we should use BUS_DMA(9) framework. But with current BUS_DMA
1830 		 * code there is no callback API to check the page alignment of
1831 		 * middle segments before busdma can decide if a bounce buffer
1832 		 * is needed for particular segment. There is callback,
1833 		 * "bus_dma_filter_t *filter", but the parrameters are not
1834 		 * sufficient for storvsc driver.
1835 		 * TODO:
1836 		 *	Add page alignment check in BUS_DMA(9) callback. Once
1837 		 *	this is complete, switch the following code to use
1838 		 *	BUS_DMA(9) for storvsc bounce buffer support.
1839 		 */
1840 		/* check if we need to create bounce buffer */
1841 		ret = storvsc_check_bounce_buffer_sgl(storvsc_sglist,
1842 		    storvsc_sg_count, &not_aligned_seg_bits);
1843 		if (ret != -1) {
1844 			reqp->bounce_sgl =
1845 			    storvsc_create_bounce_buffer(storvsc_sg_count,
1846 			    reqp->vstor_packet.u.vm_srb.data_in);
1847 			if (NULL == reqp->bounce_sgl) {
1848 				printf("Storvsc_error: "
1849 				    "create bounce buffer failed.\n");
1850 				return (ENOMEM);
1851 			}
1852 
1853 			reqp->bounce_sgl_count = storvsc_sg_count;
1854 			reqp->not_aligned_seg_bits = not_aligned_seg_bits;
1855 
1856 			/*
1857 			 * if it is write, we need copy the original data
1858 			 *to bounce buffer
1859 			 */
1860 			if (WRITE_TYPE == reqp->vstor_packet.u.vm_srb.data_in) {
1861 				storvsc_copy_sgl_to_bounce_buf(
1862 				    reqp->bounce_sgl,
1863 				    storvsc_sglist,
1864 				    storvsc_sg_count,
1865 				    reqp->not_aligned_seg_bits);
1866 			}
1867 
1868 			/* transfer virtual address to physical frame number */
1869 			if (reqp->not_aligned_seg_bits & 0x1){
1870  				phys_addr =
1871 				    vtophys(reqp->bounce_sgl->sg_segs[0].ss_paddr);
1872 			}else{
1873  				phys_addr =
1874 					vtophys(storvsc_sglist[0].ds_addr);
1875 			}
1876 			reqp->data_buf.offset = phys_addr & PAGE_MASK;
1877 
1878 			pfn = phys_addr >> PAGE_SHIFT;
1879 			reqp->data_buf.pfn_array[0] = pfn;
1880 
1881 			for (i = 1; i < storvsc_sg_count; i++) {
1882 				if (reqp->not_aligned_seg_bits & (1 << i)) {
1883 					phys_addr =
1884 					    vtophys(reqp->bounce_sgl->sg_segs[i].ss_paddr);
1885 				} else {
1886 					phys_addr =
1887 					    vtophys(storvsc_sglist[i].ds_addr);
1888 				}
1889 
1890 				pfn = phys_addr >> PAGE_SHIFT;
1891 				reqp->data_buf.pfn_array[i] = pfn;
1892 			}
1893 		} else {
1894 			phys_addr = vtophys(storvsc_sglist[0].ds_addr);
1895 
1896 			reqp->data_buf.offset = phys_addr & PAGE_MASK;
1897 
1898 			for (i = 0; i < storvsc_sg_count; i++) {
1899 				phys_addr = vtophys(storvsc_sglist[i].ds_addr);
1900 				pfn = phys_addr >> PAGE_SHIFT;
1901 				reqp->data_buf.pfn_array[i] = pfn;
1902 			}
1903 
1904 			/* check the last segment cross boundary or not */
1905 			offset = phys_addr & PAGE_MASK;
1906 			if (offset) {
1907 				phys_addr =
1908 				    vtophys(storvsc_sglist[i-1].ds_addr +
1909 				    PAGE_SIZE - offset);
1910 				pfn = phys_addr >> PAGE_SHIFT;
1911 				reqp->data_buf.pfn_array[i] = pfn;
1912 			}
1913 
1914 			reqp->bounce_sgl_count = 0;
1915 		}
1916 		break;
1917 	}
1918 	default:
1919 		printf("Unknow flags: %d\n", ccb->ccb_h.flags);
1920 		return(EINVAL);
1921 	}
1922 
1923 	return(0);
1924 }
1925 
1926 /**
1927  * @brief completion function before returning to CAM
1928  *
1929  * I/O process has been completed and the result needs
1930  * to be passed to the CAM layer.
1931  * Free resources related to this request.
1932  *
1933  * @param reqp pointer to a request structure
1934  */
1935 static void
1936 storvsc_io_done(struct hv_storvsc_request *reqp)
1937 {
1938 	union ccb *ccb = reqp->ccb;
1939 	struct ccb_scsiio *csio = &ccb->csio;
1940 	struct storvsc_softc *sc = reqp->softc;
1941 	struct vmscsi_req *vm_srb = &reqp->vstor_packet.u.vm_srb;
1942 	bus_dma_segment_t *ori_sglist = NULL;
1943 	int ori_sg_count = 0;
1944 
1945 	/* destroy bounce buffer if it is used */
1946 	if (reqp->bounce_sgl_count) {
1947 		ori_sglist = (bus_dma_segment_t *)ccb->csio.data_ptr;
1948 		ori_sg_count = ccb->csio.sglist_cnt;
1949 
1950 		/*
1951 		 * If it is READ operation, we should copy back the data
1952 		 * to original SG list.
1953 		 */
1954 		if (READ_TYPE == reqp->vstor_packet.u.vm_srb.data_in) {
1955 			storvsc_copy_from_bounce_buf_to_sgl(ori_sglist,
1956 			    ori_sg_count,
1957 			    reqp->bounce_sgl,
1958 			    reqp->not_aligned_seg_bits);
1959 		}
1960 
1961 		storvsc_destroy_bounce_buffer(reqp->bounce_sgl);
1962 		reqp->bounce_sgl_count = 0;
1963 	}
1964 
1965 	if (reqp->retries > 0) {
1966 		mtx_lock(&sc->hs_lock);
1967 #if HVS_TIMEOUT_TEST
1968 		xpt_print(ccb->ccb_h.path,
1969 			"%u: IO returned after timeout, "
1970 			"waking up timer handler if any.\n", ticks);
1971 		mtx_lock(&reqp->event.mtx);
1972 		cv_signal(&reqp->event.cv);
1973 		mtx_unlock(&reqp->event.mtx);
1974 #endif
1975 		reqp->retries = 0;
1976 		xpt_print(ccb->ccb_h.path,
1977 			"%u: IO returned after timeout, "
1978 			"stopping timer if any.\n", ticks);
1979 		mtx_unlock(&sc->hs_lock);
1980 	}
1981 
1982 	/*
1983 	 * callout_drain() will wait for the timer handler to finish
1984 	 * if it is running. So we don't need any lock to synchronize
1985 	 * between this routine and the timer handler.
1986 	 * Note that we need to make sure reqp is not freed when timer
1987 	 * handler is using or will use it.
1988 	 */
1989 	if (ccb->ccb_h.timeout != CAM_TIME_INFINITY) {
1990 		callout_drain(&reqp->callout);
1991 	}
1992 
1993 	ccb->ccb_h.status &= ~CAM_SIM_QUEUED;
1994 	ccb->ccb_h.status &= ~CAM_STATUS_MASK;
1995 	if (vm_srb->scsi_status == SCSI_STATUS_OK) {
1996 		ccb->ccb_h.status |= CAM_REQ_CMP;
1997 	 } else {
1998 		mtx_lock(&sc->hs_lock);
1999 		xpt_print(ccb->ccb_h.path,
2000 			"srovsc scsi_status = %d\n",
2001 			vm_srb->scsi_status);
2002 		mtx_unlock(&sc->hs_lock);
2003 		ccb->ccb_h.status |= CAM_SCSI_STATUS_ERROR;
2004 	}
2005 
2006 	ccb->csio.scsi_status = (vm_srb->scsi_status & 0xFF);
2007 	ccb->csio.resid = ccb->csio.dxfer_len - vm_srb->transfer_len;
2008 
2009 	if (reqp->sense_info_len != 0) {
2010 		csio->sense_resid = csio->sense_len - reqp->sense_info_len;
2011 		ccb->ccb_h.status |= CAM_AUTOSNS_VALID;
2012 	}
2013 
2014 	mtx_lock(&sc->hs_lock);
2015 	if (reqp->softc->hs_frozen == 1) {
2016 		xpt_print(ccb->ccb_h.path,
2017 			"%u: storvsc unfreezing softc 0x%p.\n",
2018 			ticks, reqp->softc);
2019 		ccb->ccb_h.status |= CAM_RELEASE_SIMQ;
2020 		reqp->softc->hs_frozen = 0;
2021 	}
2022 	storvsc_free_request(sc, reqp);
2023 	xpt_done(ccb);
2024 	mtx_unlock(&sc->hs_lock);
2025 }
2026 
2027 /**
2028  * @brief Free a request structure
2029  *
2030  * Free a request structure by returning it to the free list
2031  *
2032  * @param sc pointer to a softc
2033  * @param reqp pointer to a request structure
2034  */
2035 static void
2036 storvsc_free_request(struct storvsc_softc *sc, struct hv_storvsc_request *reqp)
2037 {
2038 
2039 	LIST_INSERT_HEAD(&sc->hs_free_list, reqp, link);
2040 }
2041 
2042 /**
2043  * @brief Determine type of storage device from GUID
2044  *
2045  * Using the type GUID, determine if this is a StorVSC (paravirtual
2046  * SCSI or BlkVSC (paravirtual IDE) device.
2047  *
2048  * @param dev a device
2049  * returns an enum
2050  */
2051 static enum hv_storage_type
2052 storvsc_get_storage_type(device_t dev)
2053 {
2054 	const char *p = vmbus_get_type(dev);
2055 
2056 	if (!memcmp(p, &gBlkVscDeviceType, sizeof(hv_guid))) {
2057 		return DRIVER_BLKVSC;
2058 	} else if (!memcmp(p, &gStorVscDeviceType, sizeof(hv_guid))) {
2059 		return DRIVER_STORVSC;
2060 	}
2061 	return (DRIVER_UNKNOWN);
2062 }
2063 
2064