xref: /illumos-gate/usr/src/cmd/bhyve/pci_virtio_scsi.c (revision 8e458de0baeb1fee50643403223bc7e909a48464)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2016 Jakub Klama <jceel@FreeBSD.org>.
5  * Copyright (c) 2018 Marcelo Araujo <araujo@FreeBSD.org>.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer
13  *    in this position and unchanged.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include <sys/param.h>
35 #include <sys/linker_set.h>
36 #include <sys/types.h>
37 #include <sys/uio.h>
38 #include <sys/time.h>
39 #include <sys/queue.h>
40 #include <sys/sbuf.h>
41 
42 #include <errno.h>
43 #include <fcntl.h>
44 #include <stdio.h>
45 #include <stdlib.h>
46 #include <stdbool.h>
47 #include <string.h>
48 #include <unistd.h>
49 #include <assert.h>
50 #include <pthread.h>
51 #include <pthread_np.h>
52 
53 #include <cam/scsi/scsi_all.h>
54 #include <cam/scsi/scsi_message.h>
55 #include <cam/ctl/ctl.h>
56 #include <cam/ctl/ctl_io.h>
57 #include <cam/ctl/ctl_backend.h>
58 #include <cam/ctl/ctl_ioctl.h>
59 #include <cam/ctl/ctl_util.h>
60 #include <cam/ctl/ctl_scsi_all.h>
61 #include <camlib.h>
62 
63 #include "bhyverun.h"
64 #include "config.h"
65 #include "debug.h"
66 #include "pci_emul.h"
67 #include "virtio.h"
68 #include "iov.h"
69 
70 #define VTSCSI_RINGSZ		64
71 #define	VTSCSI_REQUESTQ		1
72 #define	VTSCSI_THR_PER_Q	16
73 #define	VTSCSI_MAXQ		(VTSCSI_REQUESTQ + 2)
74 #define	VTSCSI_MAXSEG		64
75 
76 #define	VTSCSI_IN_HEADER_LEN(_sc)	\
77 	(sizeof(struct pci_vtscsi_req_cmd_rd) + _sc->vss_config.cdb_size)
78 
79 #define	VTSCSI_OUT_HEADER_LEN(_sc) 	\
80 	(sizeof(struct pci_vtscsi_req_cmd_wr) + _sc->vss_config.sense_size)
81 
82 #define	VIRTIO_SCSI_MAX_CHANNEL	0
83 #define	VIRTIO_SCSI_MAX_TARGET	0
84 #define	VIRTIO_SCSI_MAX_LUN	16383
85 
86 #define	VIRTIO_SCSI_F_INOUT	(1 << 0)
87 #define	VIRTIO_SCSI_F_HOTPLUG	(1 << 1)
88 #define	VIRTIO_SCSI_F_CHANGE	(1 << 2)
89 
90 static int pci_vtscsi_debug = 0;
91 #define	DPRINTF(params) if (pci_vtscsi_debug) PRINTLN params
92 #define	WPRINTF(params) PRINTLN params
93 
94 struct pci_vtscsi_config {
95 	uint32_t num_queues;
96 	uint32_t seg_max;
97 	uint32_t max_sectors;
98 	uint32_t cmd_per_lun;
99 	uint32_t event_info_size;
100 	uint32_t sense_size;
101 	uint32_t cdb_size;
102 	uint16_t max_channel;
103 	uint16_t max_target;
104 	uint32_t max_lun;
105 } __attribute__((packed));
106 
107 struct pci_vtscsi_queue {
108 	struct pci_vtscsi_softc *         vsq_sc;
109 	struct vqueue_info *              vsq_vq;
110 	pthread_mutex_t                   vsq_mtx;
111 	pthread_mutex_t                   vsq_qmtx;
112 	pthread_cond_t                    vsq_cv;
113 	STAILQ_HEAD(, pci_vtscsi_request) vsq_requests;
114 	LIST_HEAD(, pci_vtscsi_worker)    vsq_workers;
115 };
116 
117 struct pci_vtscsi_worker {
118 	struct pci_vtscsi_queue *     vsw_queue;
119 	pthread_t                     vsw_thread;
120 	bool                          vsw_exiting;
121 	LIST_ENTRY(pci_vtscsi_worker) vsw_link;
122 };
123 
124 struct pci_vtscsi_request {
125 	struct pci_vtscsi_queue * vsr_queue;
126 	struct iovec              vsr_iov_in[VTSCSI_MAXSEG];
127 	int                       vsr_niov_in;
128 	struct iovec              vsr_iov_out[VTSCSI_MAXSEG];
129 	int                       vsr_niov_out;
130 	uint32_t                  vsr_idx;
131 	STAILQ_ENTRY(pci_vtscsi_request) vsr_link;
132 };
133 
134 /*
135  * Per-device softc
136  */
137 struct pci_vtscsi_softc {
138 	struct virtio_softc      vss_vs;
139 	struct vqueue_info       vss_vq[VTSCSI_MAXQ];
140 	struct pci_vtscsi_queue  vss_queues[VTSCSI_REQUESTQ];
141 	pthread_mutex_t          vss_mtx;
142 	int                      vss_iid;
143 	int                      vss_ctl_fd;
144 	uint32_t                 vss_features;
145 	struct pci_vtscsi_config vss_config;
146 };
147 
148 #define	VIRTIO_SCSI_T_TMF			0
149 #define	VIRTIO_SCSI_T_TMF_ABORT_TASK		0
150 #define	VIRTIO_SCSI_T_TMF_ABORT_TASK_SET	1
151 #define	VIRTIO_SCSI_T_TMF_CLEAR_ACA		2
152 #define	VIRTIO_SCSI_T_TMF_CLEAR_TASK_SET	3
153 #define	VIRTIO_SCSI_T_TMF_I_T_NEXUS_RESET	4
154 #define	VIRTIO_SCSI_T_TMF_LOGICAL_UNIT_RESET	5
155 #define	VIRTIO_SCSI_T_TMF_QUERY_TASK		6
156 #define	VIRTIO_SCSI_T_TMF_QUERY_TASK_SET 	7
157 
158 /* command-specific response values */
159 #define	VIRTIO_SCSI_S_FUNCTION_COMPLETE		0
160 #define	VIRTIO_SCSI_S_FUNCTION_SUCCEEDED	10
161 #define	VIRTIO_SCSI_S_FUNCTION_REJECTED		11
162 
163 struct pci_vtscsi_ctrl_tmf {
164 	uint32_t type;
165 	uint32_t subtype;
166 	uint8_t lun[8];
167 	uint64_t id;
168 	uint8_t response;
169 } __attribute__((packed));
170 
171 #define	VIRTIO_SCSI_T_AN_QUERY			1
172 #define	VIRTIO_SCSI_EVT_ASYNC_OPERATIONAL_CHANGE 2
173 #define	VIRTIO_SCSI_EVT_ASYNC_POWER_MGMT	4
174 #define	VIRTIO_SCSI_EVT_ASYNC_EXTERNAL_REQUEST	8
175 #define	VIRTIO_SCSI_EVT_ASYNC_MEDIA_CHANGE	16
176 #define	VIRTIO_SCSI_EVT_ASYNC_MULTI_HOST	32
177 #define	VIRTIO_SCSI_EVT_ASYNC_DEVICE_BUSY	64
178 
179 struct pci_vtscsi_ctrl_an {
180 	uint32_t type;
181 	uint8_t lun[8];
182 	uint32_t event_requested;
183 	uint32_t event_actual;
184 	uint8_t response;
185 } __attribute__((packed));
186 
187 /* command-specific response values */
188 #define	VIRTIO_SCSI_S_OK 			0
189 #define	VIRTIO_SCSI_S_OVERRUN			1
190 #define	VIRTIO_SCSI_S_ABORTED			2
191 #define	VIRTIO_SCSI_S_BAD_TARGET		3
192 #define	VIRTIO_SCSI_S_RESET			4
193 #define	VIRTIO_SCSI_S_BUSY			5
194 #define	VIRTIO_SCSI_S_TRANSPORT_FAILURE		6
195 #define	VIRTIO_SCSI_S_TARGET_FAILURE		7
196 #define	VIRTIO_SCSI_S_NEXUS_FAILURE		8
197 #define	VIRTIO_SCSI_S_FAILURE			9
198 #define	VIRTIO_SCSI_S_INCORRECT_LUN		12
199 
200 /* task_attr */
201 #define	VIRTIO_SCSI_S_SIMPLE			0
202 #define	VIRTIO_SCSI_S_ORDERED			1
203 #define	VIRTIO_SCSI_S_HEAD			2
204 #define	VIRTIO_SCSI_S_ACA			3
205 
206 struct pci_vtscsi_event {
207 	uint32_t event;
208 	uint8_t lun[8];
209 	uint32_t reason;
210 } __attribute__((packed));
211 
212 struct pci_vtscsi_req_cmd_rd {
213 	uint8_t lun[8];
214 	uint64_t id;
215 	uint8_t task_attr;
216 	uint8_t prio;
217 	uint8_t crn;
218 	uint8_t cdb[];
219 } __attribute__((packed));
220 
221 struct pci_vtscsi_req_cmd_wr {
222 	uint32_t sense_len;
223 	uint32_t residual;
224 	uint16_t status_qualifier;
225 	uint8_t status;
226 	uint8_t response;
227 	uint8_t sense[];
228 } __attribute__((packed));
229 
230 static void *pci_vtscsi_proc(void *);
231 static void pci_vtscsi_reset(void *);
232 static void pci_vtscsi_neg_features(void *, uint64_t);
233 static int pci_vtscsi_cfgread(void *, int, int, uint32_t *);
234 static int pci_vtscsi_cfgwrite(void *, int, int, uint32_t);
235 static inline int pci_vtscsi_get_lun(uint8_t *);
236 static int pci_vtscsi_control_handle(struct pci_vtscsi_softc *, void *, size_t);
237 static int pci_vtscsi_tmf_handle(struct pci_vtscsi_softc *,
238     struct pci_vtscsi_ctrl_tmf *);
239 static int pci_vtscsi_an_handle(struct pci_vtscsi_softc *,
240     struct pci_vtscsi_ctrl_an *);
241 static int pci_vtscsi_request_handle(struct pci_vtscsi_queue *, struct iovec *,
242     int, struct iovec *, int);
243 static void pci_vtscsi_controlq_notify(void *, struct vqueue_info *);
244 static void pci_vtscsi_eventq_notify(void *, struct vqueue_info *);
245 static void pci_vtscsi_requestq_notify(void *, struct vqueue_info *);
246 static int  pci_vtscsi_init_queue(struct pci_vtscsi_softc *,
247     struct pci_vtscsi_queue *, int);
248 static int pci_vtscsi_init(struct vmctx *, struct pci_devinst *, nvlist_t *);
249 
250 static struct virtio_consts vtscsi_vi_consts = {
251 	"vtscsi",				/* our name */
252 	VTSCSI_MAXQ,				/* we support 2+n virtqueues */
253 	sizeof(struct pci_vtscsi_config),	/* config reg size */
254 	pci_vtscsi_reset,			/* reset */
255 	NULL,					/* device-wide qnotify */
256 	pci_vtscsi_cfgread,			/* read virtio config */
257 	pci_vtscsi_cfgwrite,			/* write virtio config */
258 	pci_vtscsi_neg_features,		/* apply negotiated features */
259 	0,					/* our capabilities */
260 };
261 
262 static void *
263 pci_vtscsi_proc(void *arg)
264 {
265 	struct pci_vtscsi_worker *worker = (struct pci_vtscsi_worker *)arg;
266 	struct pci_vtscsi_queue *q = worker->vsw_queue;
267 	struct pci_vtscsi_request *req;
268 	int iolen;
269 
270 	for (;;) {
271 		pthread_mutex_lock(&q->vsq_mtx);
272 
273 		while (STAILQ_EMPTY(&q->vsq_requests)
274 		    && !worker->vsw_exiting)
275 			pthread_cond_wait(&q->vsq_cv, &q->vsq_mtx);
276 
277 		if (worker->vsw_exiting)
278 			break;
279 
280 		req = STAILQ_FIRST(&q->vsq_requests);
281 		STAILQ_REMOVE_HEAD(&q->vsq_requests, vsr_link);
282 
283 		pthread_mutex_unlock(&q->vsq_mtx);
284 		iolen = pci_vtscsi_request_handle(q, req->vsr_iov_in,
285 		    req->vsr_niov_in, req->vsr_iov_out, req->vsr_niov_out);
286 
287 		pthread_mutex_lock(&q->vsq_qmtx);
288 		vq_relchain(q->vsq_vq, req->vsr_idx, iolen);
289 		vq_endchains(q->vsq_vq, 0);
290 		pthread_mutex_unlock(&q->vsq_qmtx);
291 
292 		DPRINTF(("virtio-scsi: request <idx=%d> completed",
293 		    req->vsr_idx));
294 		free(req);
295 	}
296 
297 	pthread_mutex_unlock(&q->vsq_mtx);
298 	return (NULL);
299 }
300 
301 static void
302 pci_vtscsi_reset(void *vsc)
303 {
304 	struct pci_vtscsi_softc *sc;
305 
306 	sc = vsc;
307 
308 	DPRINTF(("vtscsi: device reset requested"));
309 	vi_reset_dev(&sc->vss_vs);
310 
311 	/* initialize config structure */
312 	sc->vss_config = (struct pci_vtscsi_config){
313 		.num_queues = VTSCSI_REQUESTQ,
314 		/* Leave room for the request and the response. */
315 		.seg_max = VTSCSI_MAXSEG - 2,
316 		.max_sectors = 2,
317 		.cmd_per_lun = 1,
318 		.event_info_size = sizeof(struct pci_vtscsi_event),
319 		.sense_size = 96,
320 		.cdb_size = 32,
321 		.max_channel = VIRTIO_SCSI_MAX_CHANNEL,
322 		.max_target = VIRTIO_SCSI_MAX_TARGET,
323 		.max_lun = VIRTIO_SCSI_MAX_LUN
324 	};
325 }
326 
327 static void
328 pci_vtscsi_neg_features(void *vsc, uint64_t negotiated_features)
329 {
330 	struct pci_vtscsi_softc *sc = vsc;
331 
332 	sc->vss_features = negotiated_features;
333 }
334 
335 static int
336 pci_vtscsi_cfgread(void *vsc, int offset, int size, uint32_t *retval)
337 {
338 	struct pci_vtscsi_softc *sc = vsc;
339 	void *ptr;
340 
341 	ptr = (uint8_t *)&sc->vss_config + offset;
342 	memcpy(retval, ptr, size);
343 	return (0);
344 }
345 
346 static int
347 pci_vtscsi_cfgwrite(void *vsc, int offset, int size, uint32_t val)
348 {
349 
350 	return (0);
351 }
352 
353 static inline int
354 pci_vtscsi_get_lun(uint8_t *lun)
355 {
356 
357 	return (((lun[2] << 8) | lun[3]) & 0x3fff);
358 }
359 
360 static int
361 pci_vtscsi_control_handle(struct pci_vtscsi_softc *sc, void *buf,
362     size_t bufsize)
363 {
364 	struct pci_vtscsi_ctrl_tmf *tmf;
365 	struct pci_vtscsi_ctrl_an *an;
366 	uint32_t type;
367 
368 	type = *(uint32_t *)buf;
369 
370 	if (type == VIRTIO_SCSI_T_TMF) {
371 		tmf = (struct pci_vtscsi_ctrl_tmf *)buf;
372 		return (pci_vtscsi_tmf_handle(sc, tmf));
373 	}
374 
375 	if (type == VIRTIO_SCSI_T_AN_QUERY) {
376 		an = (struct pci_vtscsi_ctrl_an *)buf;
377 		return (pci_vtscsi_an_handle(sc, an));
378 	}
379 
380 	return (0);
381 }
382 
383 static int
384 pci_vtscsi_tmf_handle(struct pci_vtscsi_softc *sc,
385     struct pci_vtscsi_ctrl_tmf *tmf)
386 {
387 	union ctl_io *io;
388 	int err;
389 
390 	io = ctl_scsi_alloc_io(sc->vss_iid);
391 	ctl_scsi_zero_io(io);
392 
393 	io->io_hdr.io_type = CTL_IO_TASK;
394 	io->io_hdr.nexus.initid = sc->vss_iid;
395 	io->io_hdr.nexus.targ_lun = pci_vtscsi_get_lun(tmf->lun);
396 	io->taskio.tag_type = CTL_TAG_SIMPLE;
397 	io->taskio.tag_num = (uint32_t)tmf->id;
398 
399 	switch (tmf->subtype) {
400 	case VIRTIO_SCSI_T_TMF_ABORT_TASK:
401 		io->taskio.task_action = CTL_TASK_ABORT_TASK;
402 		break;
403 
404 	case VIRTIO_SCSI_T_TMF_ABORT_TASK_SET:
405 		io->taskio.task_action = CTL_TASK_ABORT_TASK_SET;
406 		break;
407 
408 	case VIRTIO_SCSI_T_TMF_CLEAR_ACA:
409 		io->taskio.task_action = CTL_TASK_CLEAR_ACA;
410 		break;
411 
412 	case VIRTIO_SCSI_T_TMF_CLEAR_TASK_SET:
413 		io->taskio.task_action = CTL_TASK_CLEAR_TASK_SET;
414 		break;
415 
416 	case VIRTIO_SCSI_T_TMF_I_T_NEXUS_RESET:
417 		io->taskio.task_action = CTL_TASK_I_T_NEXUS_RESET;
418 		break;
419 
420 	case VIRTIO_SCSI_T_TMF_LOGICAL_UNIT_RESET:
421 		io->taskio.task_action = CTL_TASK_LUN_RESET;
422 		break;
423 
424 	case VIRTIO_SCSI_T_TMF_QUERY_TASK:
425 		io->taskio.task_action = CTL_TASK_QUERY_TASK;
426 		break;
427 
428 	case VIRTIO_SCSI_T_TMF_QUERY_TASK_SET:
429 		io->taskio.task_action = CTL_TASK_QUERY_TASK_SET;
430 		break;
431 	}
432 
433 	if (pci_vtscsi_debug) {
434 		struct sbuf *sb = sbuf_new_auto();
435 		ctl_io_sbuf(io, sb);
436 		sbuf_finish(sb);
437 		DPRINTF(("pci_virtio_scsi: %s", sbuf_data(sb)));
438 		sbuf_delete(sb);
439 	}
440 
441 	err = ioctl(sc->vss_ctl_fd, CTL_IO, io);
442 	if (err != 0)
443 		WPRINTF(("CTL_IO: err=%d (%s)", errno, strerror(errno)));
444 
445 	tmf->response = io->taskio.task_status;
446 	ctl_scsi_free_io(io);
447 	return (1);
448 }
449 
450 static int
451 pci_vtscsi_an_handle(struct pci_vtscsi_softc *sc,
452     struct pci_vtscsi_ctrl_an *an)
453 {
454 
455 	return (0);
456 }
457 
458 static int
459 pci_vtscsi_request_handle(struct pci_vtscsi_queue *q, struct iovec *iov_in,
460     int niov_in, struct iovec *iov_out, int niov_out)
461 {
462 	struct pci_vtscsi_softc *sc = q->vsq_sc;
463 	struct pci_vtscsi_req_cmd_rd *cmd_rd = NULL;
464 	struct pci_vtscsi_req_cmd_wr *cmd_wr;
465 	struct iovec data_iov_in[VTSCSI_MAXSEG], data_iov_out[VTSCSI_MAXSEG];
466 	union ctl_io *io;
467 	int data_niov_in, data_niov_out;
468 	void *ext_data_ptr = NULL;
469 	uint32_t ext_data_len = 0, ext_sg_entries = 0;
470 	int err, nxferred;
471 
472 	seek_iov(iov_in, niov_in, data_iov_in, &data_niov_in,
473 	    VTSCSI_IN_HEADER_LEN(sc));
474 	seek_iov(iov_out, niov_out, data_iov_out, &data_niov_out,
475 	    VTSCSI_OUT_HEADER_LEN(sc));
476 
477 	truncate_iov(iov_in, &niov_in, VTSCSI_IN_HEADER_LEN(sc));
478 	truncate_iov(iov_out, &niov_out, VTSCSI_OUT_HEADER_LEN(sc));
479 	iov_to_buf(iov_in, niov_in, (void **)&cmd_rd);
480 
481 	cmd_wr = malloc(VTSCSI_OUT_HEADER_LEN(sc));
482 	io = ctl_scsi_alloc_io(sc->vss_iid);
483 	ctl_scsi_zero_io(io);
484 
485 	io->io_hdr.nexus.initid = sc->vss_iid;
486 	io->io_hdr.nexus.targ_lun = pci_vtscsi_get_lun(cmd_rd->lun);
487 
488 	io->io_hdr.io_type = CTL_IO_SCSI;
489 
490 	if (data_niov_in > 0) {
491 		ext_data_ptr = (void *)data_iov_in;
492 		ext_sg_entries = data_niov_in;
493 		ext_data_len = count_iov(data_iov_in, data_niov_in);
494 		io->io_hdr.flags |= CTL_FLAG_DATA_OUT;
495 	} else if (data_niov_out > 0) {
496 		ext_data_ptr = (void *)data_iov_out;
497 		ext_sg_entries = data_niov_out;
498 		ext_data_len = count_iov(data_iov_out, data_niov_out);
499 		io->io_hdr.flags |= CTL_FLAG_DATA_IN;
500 	}
501 
502 	io->scsiio.sense_len = sc->vss_config.sense_size;
503 	io->scsiio.tag_num = (uint32_t)cmd_rd->id;
504 	switch (cmd_rd->task_attr) {
505 	case VIRTIO_SCSI_S_ORDERED:
506 		io->scsiio.tag_type = CTL_TAG_ORDERED;
507 		break;
508 	case VIRTIO_SCSI_S_HEAD:
509 		io->scsiio.tag_type = CTL_TAG_HEAD_OF_QUEUE;
510 		break;
511 	case VIRTIO_SCSI_S_ACA:
512 		io->scsiio.tag_type = CTL_TAG_ACA;
513 		break;
514 	case VIRTIO_SCSI_S_SIMPLE:
515 	default:
516 		io->scsiio.tag_type = CTL_TAG_SIMPLE;
517 		break;
518 	}
519 	io->scsiio.ext_sg_entries = ext_sg_entries;
520 	io->scsiio.ext_data_ptr = ext_data_ptr;
521 	io->scsiio.ext_data_len = ext_data_len;
522 	io->scsiio.ext_data_filled = 0;
523 	io->scsiio.cdb_len = sc->vss_config.cdb_size;
524 	memcpy(io->scsiio.cdb, cmd_rd->cdb, sc->vss_config.cdb_size);
525 
526 	if (pci_vtscsi_debug) {
527 		struct sbuf *sb = sbuf_new_auto();
528 		ctl_io_sbuf(io, sb);
529 		sbuf_finish(sb);
530 		DPRINTF(("pci_virtio_scsi: %s", sbuf_data(sb)));
531 		sbuf_delete(sb);
532 	}
533 
534 	err = ioctl(sc->vss_ctl_fd, CTL_IO, io);
535 	if (err != 0) {
536 		WPRINTF(("CTL_IO: err=%d (%s)", errno, strerror(errno)));
537 		cmd_wr->response = VIRTIO_SCSI_S_FAILURE;
538 	} else {
539 		cmd_wr->sense_len = MIN(io->scsiio.sense_len,
540 		    sc->vss_config.sense_size);
541 		cmd_wr->residual = io->scsiio.residual;
542 		cmd_wr->status = io->scsiio.scsi_status;
543 		cmd_wr->response = VIRTIO_SCSI_S_OK;
544 		memcpy(&cmd_wr->sense, &io->scsiio.sense_data,
545 		    cmd_wr->sense_len);
546 	}
547 
548 	buf_to_iov(cmd_wr, VTSCSI_OUT_HEADER_LEN(sc), iov_out, niov_out, 0);
549 	nxferred = VTSCSI_OUT_HEADER_LEN(sc) + io->scsiio.ext_data_filled;
550 	free(cmd_rd);
551 	free(cmd_wr);
552 	ctl_scsi_free_io(io);
553 	return (nxferred);
554 }
555 
556 static void
557 pci_vtscsi_controlq_notify(void *vsc, struct vqueue_info *vq)
558 {
559 	struct pci_vtscsi_softc *sc;
560 	struct iovec iov[VTSCSI_MAXSEG];
561 	uint16_t idx, n;
562 	void *buf = NULL;
563 	size_t bufsize;
564 	int iolen;
565 
566 	sc = vsc;
567 
568 	while (vq_has_descs(vq)) {
569 		n = vq_getchain(vq, &idx, iov, VTSCSI_MAXSEG, NULL);
570 		bufsize = iov_to_buf(iov, n, &buf);
571 		iolen = pci_vtscsi_control_handle(sc, buf, bufsize);
572 		buf_to_iov(buf + bufsize - iolen, iolen, iov, n,
573 		    bufsize - iolen);
574 
575 		/*
576 		 * Release this chain and handle more
577 		 */
578 		vq_relchain(vq, idx, iolen);
579 	}
580 	vq_endchains(vq, 1);	/* Generate interrupt if appropriate. */
581 	free(buf);
582 }
583 
584 static void
585 pci_vtscsi_eventq_notify(void *vsc, struct vqueue_info *vq)
586 {
587 
588 	vq_kick_disable(vq);
589 }
590 
591 static void
592 pci_vtscsi_requestq_notify(void *vsc, struct vqueue_info *vq)
593 {
594 	struct pci_vtscsi_softc *sc;
595 	struct pci_vtscsi_queue *q;
596 	struct pci_vtscsi_request *req;
597 	struct iovec iov[VTSCSI_MAXSEG];
598 	uint16_t flags[VTSCSI_MAXSEG];
599 	uint16_t idx, n, i;
600 	int readable;
601 
602 	sc = vsc;
603 	q = &sc->vss_queues[vq->vq_num - 2];
604 
605 	while (vq_has_descs(vq)) {
606 		readable = 0;
607 		n = vq_getchain(vq, &idx, iov, VTSCSI_MAXSEG, flags);
608 
609 		/* Count readable descriptors */
610 		for (i = 0; i < n; i++) {
611 			if (flags[i] & VRING_DESC_F_WRITE)
612 				break;
613 
614 			readable++;
615 		}
616 
617 		req = calloc(1, sizeof(struct pci_vtscsi_request));
618 		req->vsr_idx = idx;
619 		req->vsr_queue = q;
620 		req->vsr_niov_in = readable;
621 		req->vsr_niov_out = n - readable;
622 		memcpy(req->vsr_iov_in, iov,
623 		    req->vsr_niov_in * sizeof(struct iovec));
624 		memcpy(req->vsr_iov_out, iov + readable,
625 		    req->vsr_niov_out * sizeof(struct iovec));
626 
627 		pthread_mutex_lock(&q->vsq_mtx);
628 		STAILQ_INSERT_TAIL(&q->vsq_requests, req, vsr_link);
629 		pthread_cond_signal(&q->vsq_cv);
630 		pthread_mutex_unlock(&q->vsq_mtx);
631 
632 		DPRINTF(("virtio-scsi: request <idx=%d> enqueued", idx));
633 	}
634 }
635 
636 static int
637 pci_vtscsi_init_queue(struct pci_vtscsi_softc *sc,
638     struct pci_vtscsi_queue *queue, int num)
639 {
640 	struct pci_vtscsi_worker *worker;
641 	char tname[MAXCOMLEN + 1];
642 	int i;
643 
644 	queue->vsq_sc = sc;
645 	queue->vsq_vq = &sc->vss_vq[num + 2];
646 
647 	pthread_mutex_init(&queue->vsq_mtx, NULL);
648 	pthread_mutex_init(&queue->vsq_qmtx, NULL);
649 	pthread_cond_init(&queue->vsq_cv, NULL);
650 	STAILQ_INIT(&queue->vsq_requests);
651 	LIST_INIT(&queue->vsq_workers);
652 
653 	for (i = 0; i < VTSCSI_THR_PER_Q; i++) {
654 		worker = calloc(1, sizeof(struct pci_vtscsi_worker));
655 		worker->vsw_queue = queue;
656 
657 		pthread_create(&worker->vsw_thread, NULL, &pci_vtscsi_proc,
658 		    (void *)worker);
659 
660 		snprintf(tname, sizeof(tname), "vtscsi:%d-%d", num, i);
661 		pthread_set_name_np(worker->vsw_thread, tname);
662 		LIST_INSERT_HEAD(&queue->vsq_workers, worker, vsw_link);
663 	}
664 
665 	return (0);
666 }
667 
668 static int
669 pci_vtscsi_legacy_config(nvlist_t *nvl, const char *opts)
670 {
671 	char *cp, *devname;
672 
673 	cp = strchr(opts, ',');
674 	if (cp == NULL) {
675 		set_config_value_node(nvl, "dev", opts);
676 		return (0);
677 	}
678 	devname = strndup(opts, cp - opts);
679 	set_config_value_node(nvl, "dev", devname);
680 	free(devname);
681 	return (pci_parse_legacy_config(nvl, cp + 1));
682 }
683 
684 static int
685 pci_vtscsi_init(struct vmctx *ctx, struct pci_devinst *pi, nvlist_t *nvl)
686 {
687 	struct pci_vtscsi_softc *sc;
688 	const char *devname, *value;;
689 	int i;
690 
691 	sc = calloc(1, sizeof(struct pci_vtscsi_softc));
692 	value = get_config_value_node(nvl, "iid");
693 	if (value != NULL)
694 		sc->vss_iid = strtoul(value, NULL, 10);
695 
696 	devname = get_config_value_node(nvl, "dev");
697 	if (devname == NULL)
698 		devname = "/dev/cam/ctl";
699 	sc->vss_ctl_fd = open(devname, O_RDWR);
700 	if (sc->vss_ctl_fd < 0) {
701 		WPRINTF(("cannot open %s: %s", devname, strerror(errno)));
702 		free(sc);
703 		return (1);
704 	}
705 
706 	vi_softc_linkup(&sc->vss_vs, &vtscsi_vi_consts, sc, pi, sc->vss_vq);
707 	sc->vss_vs.vs_mtx = &sc->vss_mtx;
708 
709 	/* controlq */
710 	sc->vss_vq[0].vq_qsize = VTSCSI_RINGSZ;
711 	sc->vss_vq[0].vq_notify = pci_vtscsi_controlq_notify;
712 
713 	/* eventq */
714 	sc->vss_vq[1].vq_qsize = VTSCSI_RINGSZ;
715 	sc->vss_vq[1].vq_notify = pci_vtscsi_eventq_notify;
716 
717 	/* request queues */
718 	for (i = 2; i < VTSCSI_MAXQ; i++) {
719 		sc->vss_vq[i].vq_qsize = VTSCSI_RINGSZ;
720 		sc->vss_vq[i].vq_notify = pci_vtscsi_requestq_notify;
721 		pci_vtscsi_init_queue(sc, &sc->vss_queues[i - 2], i - 2);
722 	}
723 
724 	/* initialize config space */
725 	pci_set_cfgdata16(pi, PCIR_DEVICE, VIRTIO_DEV_SCSI);
726 	pci_set_cfgdata16(pi, PCIR_VENDOR, VIRTIO_VENDOR);
727 	pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_STORAGE);
728 	pci_set_cfgdata16(pi, PCIR_SUBDEV_0, VIRTIO_ID_SCSI);
729 	pci_set_cfgdata16(pi, PCIR_SUBVEND_0, VIRTIO_VENDOR);
730 
731 	if (vi_intr_init(&sc->vss_vs, 1, fbsdrun_virtio_msix()))
732 		return (1);
733 	vi_set_io_bar(&sc->vss_vs, 0);
734 
735 	return (0);
736 }
737 
738 
739 struct pci_devemu pci_de_vscsi = {
740 	.pe_emu =	"virtio-scsi",
741 	.pe_init =	pci_vtscsi_init,
742 	.pe_barwrite =	vi_pci_write,
743 	.pe_barread =	vi_pci_read
744 };
745 PCI_EMUL_SET(pci_de_vscsi);
746