xref: /freebsd/usr.sbin/bhyve/pci_virtio_scsi.c (revision 2008043f386721d58158e37e0d7e50df8095942d)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
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 #include <sys/param.h>
33 #include <sys/linker_set.h>
34 #include <sys/types.h>
35 #include <sys/uio.h>
36 #include <sys/time.h>
37 #include <sys/queue.h>
38 #include <sys/sbuf.h>
39 
40 #include <errno.h>
41 #include <fcntl.h>
42 #include <stdio.h>
43 #include <stdlib.h>
44 #include <stdbool.h>
45 #include <string.h>
46 #include <unistd.h>
47 #include <assert.h>
48 #include <pthread.h>
49 #include <pthread_np.h>
50 
51 #include <cam/scsi/scsi_all.h>
52 #include <cam/scsi/scsi_message.h>
53 #include <cam/ctl/ctl.h>
54 #include <cam/ctl/ctl_io.h>
55 #include <cam/ctl/ctl_backend.h>
56 #include <cam/ctl/ctl_ioctl.h>
57 #include <cam/ctl/ctl_util.h>
58 #include <cam/ctl/ctl_scsi_all.h>
59 #include <camlib.h>
60 
61 #include "bhyverun.h"
62 #include "config.h"
63 #include "debug.h"
64 #include "pci_emul.h"
65 #include "virtio.h"
66 #include "iov.h"
67 
68 #define VTSCSI_RINGSZ		64
69 #define	VTSCSI_REQUESTQ		1
70 #define	VTSCSI_THR_PER_Q	16
71 #define	VTSCSI_MAXQ		(VTSCSI_REQUESTQ + 2)
72 #define	VTSCSI_MAXSEG		64
73 
74 #define	VTSCSI_IN_HEADER_LEN(_sc)	\
75 	(sizeof(struct pci_vtscsi_req_cmd_rd) + _sc->vss_config.cdb_size)
76 
77 #define	VTSCSI_OUT_HEADER_LEN(_sc) 	\
78 	(sizeof(struct pci_vtscsi_req_cmd_wr) + _sc->vss_config.sense_size)
79 
80 #define	VIRTIO_SCSI_MAX_CHANNEL	0
81 #define	VIRTIO_SCSI_MAX_TARGET	0
82 #define	VIRTIO_SCSI_MAX_LUN	16383
83 
84 #define	VIRTIO_SCSI_F_INOUT	(1 << 0)
85 #define	VIRTIO_SCSI_F_HOTPLUG	(1 << 1)
86 #define	VIRTIO_SCSI_F_CHANGE	(1 << 2)
87 
88 static int pci_vtscsi_debug = 0;
89 #define	WPRINTF(msg, params...) PRINTLN("virtio-scsi: " msg, ##params)
90 #define	DPRINTF(msg, params...) if (pci_vtscsi_debug) WPRINTF(msg, ##params)
91 
92 struct pci_vtscsi_config {
93 	uint32_t num_queues;
94 	uint32_t seg_max;
95 	uint32_t max_sectors;
96 	uint32_t cmd_per_lun;
97 	uint32_t event_info_size;
98 	uint32_t sense_size;
99 	uint32_t cdb_size;
100 	uint16_t max_channel;
101 	uint16_t max_target;
102 	uint32_t max_lun;
103 } __attribute__((packed));
104 
105 struct pci_vtscsi_queue {
106 	struct pci_vtscsi_softc *         vsq_sc;
107 	struct vqueue_info *              vsq_vq;
108 	pthread_mutex_t                   vsq_mtx;
109 	pthread_mutex_t                   vsq_qmtx;
110 	pthread_cond_t                    vsq_cv;
111 	STAILQ_HEAD(, pci_vtscsi_request) vsq_requests;
112 	LIST_HEAD(, pci_vtscsi_worker)    vsq_workers;
113 };
114 
115 struct pci_vtscsi_worker {
116 	struct pci_vtscsi_queue *     vsw_queue;
117 	pthread_t                     vsw_thread;
118 	bool                          vsw_exiting;
119 	LIST_ENTRY(pci_vtscsi_worker) vsw_link;
120 };
121 
122 struct pci_vtscsi_request {
123 	struct pci_vtscsi_queue * vsr_queue;
124 	struct iovec              vsr_iov_in[VTSCSI_MAXSEG];
125 	int                       vsr_niov_in;
126 	struct iovec              vsr_iov_out[VTSCSI_MAXSEG];
127 	int                       vsr_niov_out;
128 	uint32_t                  vsr_idx;
129 	STAILQ_ENTRY(pci_vtscsi_request) vsr_link;
130 };
131 
132 /*
133  * Per-device softc
134  */
135 struct pci_vtscsi_softc {
136 	struct virtio_softc      vss_vs;
137 	struct vqueue_info       vss_vq[VTSCSI_MAXQ];
138 	struct pci_vtscsi_queue  vss_queues[VTSCSI_REQUESTQ];
139 	pthread_mutex_t          vss_mtx;
140 	int                      vss_iid;
141 	int                      vss_ctl_fd;
142 	uint32_t                 vss_features;
143 	struct pci_vtscsi_config vss_config;
144 };
145 
146 #define	VIRTIO_SCSI_T_TMF			0
147 #define	VIRTIO_SCSI_T_TMF_ABORT_TASK		0
148 #define	VIRTIO_SCSI_T_TMF_ABORT_TASK_SET	1
149 #define	VIRTIO_SCSI_T_TMF_CLEAR_ACA		2
150 #define	VIRTIO_SCSI_T_TMF_CLEAR_TASK_SET	3
151 #define	VIRTIO_SCSI_T_TMF_I_T_NEXUS_RESET	4
152 #define	VIRTIO_SCSI_T_TMF_LOGICAL_UNIT_RESET	5
153 #define	VIRTIO_SCSI_T_TMF_QUERY_TASK		6
154 #define	VIRTIO_SCSI_T_TMF_QUERY_TASK_SET 	7
155 
156 /* command-specific response values */
157 #define	VIRTIO_SCSI_S_FUNCTION_COMPLETE		0
158 #define	VIRTIO_SCSI_S_FUNCTION_SUCCEEDED	10
159 #define	VIRTIO_SCSI_S_FUNCTION_REJECTED		11
160 
161 struct pci_vtscsi_ctrl_tmf {
162 	uint32_t type;
163 	uint32_t subtype;
164 	uint8_t lun[8];
165 	uint64_t id;
166 	uint8_t response;
167 } __attribute__((packed));
168 
169 #define	VIRTIO_SCSI_T_AN_QUERY			1
170 #define	VIRTIO_SCSI_EVT_ASYNC_OPERATIONAL_CHANGE 2
171 #define	VIRTIO_SCSI_EVT_ASYNC_POWER_MGMT	4
172 #define	VIRTIO_SCSI_EVT_ASYNC_EXTERNAL_REQUEST	8
173 #define	VIRTIO_SCSI_EVT_ASYNC_MEDIA_CHANGE	16
174 #define	VIRTIO_SCSI_EVT_ASYNC_MULTI_HOST	32
175 #define	VIRTIO_SCSI_EVT_ASYNC_DEVICE_BUSY	64
176 
177 struct pci_vtscsi_ctrl_an {
178 	uint32_t type;
179 	uint8_t lun[8];
180 	uint32_t event_requested;
181 	uint32_t event_actual;
182 	uint8_t response;
183 } __attribute__((packed));
184 
185 /* command-specific response values */
186 #define	VIRTIO_SCSI_S_OK 			0
187 #define	VIRTIO_SCSI_S_OVERRUN			1
188 #define	VIRTIO_SCSI_S_ABORTED			2
189 #define	VIRTIO_SCSI_S_BAD_TARGET		3
190 #define	VIRTIO_SCSI_S_RESET			4
191 #define	VIRTIO_SCSI_S_BUSY			5
192 #define	VIRTIO_SCSI_S_TRANSPORT_FAILURE		6
193 #define	VIRTIO_SCSI_S_TARGET_FAILURE		7
194 #define	VIRTIO_SCSI_S_NEXUS_FAILURE		8
195 #define	VIRTIO_SCSI_S_FAILURE			9
196 #define	VIRTIO_SCSI_S_INCORRECT_LUN		12
197 
198 /* task_attr */
199 #define	VIRTIO_SCSI_S_SIMPLE			0
200 #define	VIRTIO_SCSI_S_ORDERED			1
201 #define	VIRTIO_SCSI_S_HEAD			2
202 #define	VIRTIO_SCSI_S_ACA			3
203 
204 struct pci_vtscsi_event {
205 	uint32_t event;
206 	uint8_t lun[8];
207 	uint32_t reason;
208 } __attribute__((packed));
209 
210 struct pci_vtscsi_req_cmd_rd {
211 	uint8_t lun[8];
212 	uint64_t id;
213 	uint8_t task_attr;
214 	uint8_t prio;
215 	uint8_t crn;
216 	uint8_t cdb[];
217 } __attribute__((packed));
218 
219 struct pci_vtscsi_req_cmd_wr {
220 	uint32_t sense_len;
221 	uint32_t residual;
222 	uint16_t status_qualifier;
223 	uint8_t status;
224 	uint8_t response;
225 	uint8_t sense[];
226 } __attribute__((packed));
227 
228 static void *pci_vtscsi_proc(void *);
229 static void pci_vtscsi_reset(void *);
230 static void pci_vtscsi_neg_features(void *, uint64_t);
231 static int pci_vtscsi_cfgread(void *, int, int, uint32_t *);
232 static int pci_vtscsi_cfgwrite(void *, int, int, uint32_t);
233 static inline int pci_vtscsi_get_lun(uint8_t *);
234 static int pci_vtscsi_control_handle(struct pci_vtscsi_softc *, void *, size_t);
235 static int pci_vtscsi_tmf_handle(struct pci_vtscsi_softc *,
236     struct pci_vtscsi_ctrl_tmf *);
237 static int pci_vtscsi_an_handle(struct pci_vtscsi_softc *,
238     struct pci_vtscsi_ctrl_an *);
239 static int pci_vtscsi_request_handle(struct pci_vtscsi_queue *, struct iovec *,
240     int, struct iovec *, int);
241 static void pci_vtscsi_controlq_notify(void *, struct vqueue_info *);
242 static void pci_vtscsi_eventq_notify(void *, struct vqueue_info *);
243 static void pci_vtscsi_requestq_notify(void *, struct vqueue_info *);
244 static int  pci_vtscsi_init_queue(struct pci_vtscsi_softc *,
245     struct pci_vtscsi_queue *, int);
246 static int pci_vtscsi_init(struct pci_devinst *, nvlist_t *);
247 
248 static struct virtio_consts vtscsi_vi_consts = {
249 	.vc_name =	"vtscsi",
250 	.vc_nvq =	VTSCSI_MAXQ,
251 	.vc_cfgsize =	sizeof(struct pci_vtscsi_config),
252 	.vc_reset =	pci_vtscsi_reset,
253 	.vc_cfgread =	pci_vtscsi_cfgread,
254 	.vc_cfgwrite =	pci_vtscsi_cfgwrite,
255 	.vc_apply_features = pci_vtscsi_neg_features,
256 	.vc_hv_caps =	0,
257 };
258 
259 static void *
260 pci_vtscsi_proc(void *arg)
261 {
262 	struct pci_vtscsi_worker *worker = (struct pci_vtscsi_worker *)arg;
263 	struct pci_vtscsi_queue *q = worker->vsw_queue;
264 	struct pci_vtscsi_request *req;
265 	int iolen;
266 
267 	for (;;) {
268 		pthread_mutex_lock(&q->vsq_mtx);
269 
270 		while (STAILQ_EMPTY(&q->vsq_requests)
271 		    && !worker->vsw_exiting)
272 			pthread_cond_wait(&q->vsq_cv, &q->vsq_mtx);
273 
274 		if (worker->vsw_exiting)
275 			break;
276 
277 		req = STAILQ_FIRST(&q->vsq_requests);
278 		STAILQ_REMOVE_HEAD(&q->vsq_requests, vsr_link);
279 
280 		pthread_mutex_unlock(&q->vsq_mtx);
281 		iolen = pci_vtscsi_request_handle(q, req->vsr_iov_in,
282 		    req->vsr_niov_in, req->vsr_iov_out, req->vsr_niov_out);
283 
284 		pthread_mutex_lock(&q->vsq_qmtx);
285 		vq_relchain(q->vsq_vq, req->vsr_idx, iolen);
286 		vq_endchains(q->vsq_vq, 0);
287 		pthread_mutex_unlock(&q->vsq_qmtx);
288 
289 		DPRINTF("request <idx=%d> completed", req->vsr_idx);
290 		free(req);
291 	}
292 
293 	pthread_mutex_unlock(&q->vsq_mtx);
294 	return (NULL);
295 }
296 
297 static void
298 pci_vtscsi_reset(void *vsc)
299 {
300 	struct pci_vtscsi_softc *sc;
301 
302 	sc = vsc;
303 
304 	DPRINTF("device reset requested");
305 	vi_reset_dev(&sc->vss_vs);
306 
307 	/* initialize config structure */
308 	sc->vss_config = (struct pci_vtscsi_config){
309 		.num_queues = VTSCSI_REQUESTQ,
310 		/* Leave room for the request and the response. */
311 		.seg_max = VTSCSI_MAXSEG - 2,
312 		.max_sectors = 2,
313 		.cmd_per_lun = 1,
314 		.event_info_size = sizeof(struct pci_vtscsi_event),
315 		.sense_size = 96,
316 		.cdb_size = 32,
317 		.max_channel = VIRTIO_SCSI_MAX_CHANNEL,
318 		.max_target = VIRTIO_SCSI_MAX_TARGET,
319 		.max_lun = VIRTIO_SCSI_MAX_LUN
320 	};
321 }
322 
323 static void
324 pci_vtscsi_neg_features(void *vsc, uint64_t negotiated_features)
325 {
326 	struct pci_vtscsi_softc *sc = vsc;
327 
328 	sc->vss_features = negotiated_features;
329 }
330 
331 static int
332 pci_vtscsi_cfgread(void *vsc, int offset, int size, uint32_t *retval)
333 {
334 	struct pci_vtscsi_softc *sc = vsc;
335 	void *ptr;
336 
337 	ptr = (uint8_t *)&sc->vss_config + offset;
338 	memcpy(retval, ptr, size);
339 	return (0);
340 }
341 
342 static int
343 pci_vtscsi_cfgwrite(void *vsc __unused, int offset __unused, int size __unused,
344     uint32_t val __unused)
345 {
346 	return (0);
347 }
348 
349 static inline int
350 pci_vtscsi_get_lun(uint8_t *lun)
351 {
352 
353 	return (((lun[2] << 8) | lun[3]) & 0x3fff);
354 }
355 
356 static int
357 pci_vtscsi_control_handle(struct pci_vtscsi_softc *sc, void *buf,
358     size_t bufsize)
359 {
360 	struct pci_vtscsi_ctrl_tmf *tmf;
361 	struct pci_vtscsi_ctrl_an *an;
362 	uint32_t type;
363 
364 	if (bufsize < sizeof(uint32_t)) {
365 		WPRINTF("ignoring truncated control request");
366 		return (0);
367 	}
368 
369 	type = *(uint32_t *)buf;
370 
371 	if (type == VIRTIO_SCSI_T_TMF) {
372 		if (bufsize != sizeof(*tmf)) {
373 			WPRINTF("ignoring tmf request with size %zu", bufsize);
374 			return (0);
375 		}
376 		tmf = (struct pci_vtscsi_ctrl_tmf *)buf;
377 		return (pci_vtscsi_tmf_handle(sc, tmf));
378 	}
379 
380 	if (type == VIRTIO_SCSI_T_AN_QUERY) {
381 		if (bufsize != sizeof(*an)) {
382 			WPRINTF("ignoring AN request with size %zu", bufsize);
383 			return (0);
384 		}
385 		an = (struct pci_vtscsi_ctrl_an *)buf;
386 		return (pci_vtscsi_an_handle(sc, an));
387 	}
388 
389 	return (0);
390 }
391 
392 static int
393 pci_vtscsi_tmf_handle(struct pci_vtscsi_softc *sc,
394     struct pci_vtscsi_ctrl_tmf *tmf)
395 {
396 	union ctl_io *io;
397 	int err;
398 
399 	io = ctl_scsi_alloc_io(sc->vss_iid);
400 	ctl_scsi_zero_io(io);
401 
402 	io->io_hdr.io_type = CTL_IO_TASK;
403 	io->io_hdr.nexus.initid = sc->vss_iid;
404 	io->io_hdr.nexus.targ_lun = pci_vtscsi_get_lun(tmf->lun);
405 	io->taskio.tag_type = CTL_TAG_SIMPLE;
406 	io->taskio.tag_num = tmf->id;
407 	io->io_hdr.flags |= CTL_FLAG_USER_TAG;
408 
409 	switch (tmf->subtype) {
410 	case VIRTIO_SCSI_T_TMF_ABORT_TASK:
411 		io->taskio.task_action = CTL_TASK_ABORT_TASK;
412 		break;
413 
414 	case VIRTIO_SCSI_T_TMF_ABORT_TASK_SET:
415 		io->taskio.task_action = CTL_TASK_ABORT_TASK_SET;
416 		break;
417 
418 	case VIRTIO_SCSI_T_TMF_CLEAR_ACA:
419 		io->taskio.task_action = CTL_TASK_CLEAR_ACA;
420 		break;
421 
422 	case VIRTIO_SCSI_T_TMF_CLEAR_TASK_SET:
423 		io->taskio.task_action = CTL_TASK_CLEAR_TASK_SET;
424 		break;
425 
426 	case VIRTIO_SCSI_T_TMF_I_T_NEXUS_RESET:
427 		io->taskio.task_action = CTL_TASK_I_T_NEXUS_RESET;
428 		break;
429 
430 	case VIRTIO_SCSI_T_TMF_LOGICAL_UNIT_RESET:
431 		io->taskio.task_action = CTL_TASK_LUN_RESET;
432 		break;
433 
434 	case VIRTIO_SCSI_T_TMF_QUERY_TASK:
435 		io->taskio.task_action = CTL_TASK_QUERY_TASK;
436 		break;
437 
438 	case VIRTIO_SCSI_T_TMF_QUERY_TASK_SET:
439 		io->taskio.task_action = CTL_TASK_QUERY_TASK_SET;
440 		break;
441 	}
442 
443 	if (pci_vtscsi_debug) {
444 		struct sbuf *sb = sbuf_new_auto();
445 		ctl_io_sbuf(io, sb);
446 		sbuf_finish(sb);
447 		DPRINTF("%s", sbuf_data(sb));
448 		sbuf_delete(sb);
449 	}
450 
451 	err = ioctl(sc->vss_ctl_fd, CTL_IO, io);
452 	if (err != 0)
453 		WPRINTF("CTL_IO: err=%d (%s)", errno, strerror(errno));
454 
455 	tmf->response = io->taskio.task_status;
456 	ctl_scsi_free_io(io);
457 	return (1);
458 }
459 
460 static int
461 pci_vtscsi_an_handle(struct pci_vtscsi_softc *sc __unused,
462     struct pci_vtscsi_ctrl_an *an __unused)
463 {
464 	return (0);
465 }
466 
467 static int
468 pci_vtscsi_request_handle(struct pci_vtscsi_queue *q, struct iovec *iov_in,
469     int niov_in, struct iovec *iov_out, int niov_out)
470 {
471 	struct pci_vtscsi_softc *sc = q->vsq_sc;
472 	struct pci_vtscsi_req_cmd_rd *cmd_rd = NULL;
473 	struct pci_vtscsi_req_cmd_wr *cmd_wr;
474 	struct iovec data_iov_in[VTSCSI_MAXSEG], data_iov_out[VTSCSI_MAXSEG];
475 	union ctl_io *io;
476 	int data_niov_in, data_niov_out;
477 	void *ext_data_ptr = NULL;
478 	uint32_t ext_data_len = 0, ext_sg_entries = 0;
479 	int err, nxferred;
480 
481 	if (count_iov(iov_out, niov_out) < VTSCSI_OUT_HEADER_LEN(sc)) {
482 		WPRINTF("ignoring request with insufficient output");
483 		return (0);
484 	}
485 	if (count_iov(iov_in, niov_in) < VTSCSI_IN_HEADER_LEN(sc)) {
486 		WPRINTF("ignoring request with incomplete header");
487 		return (0);
488 	}
489 
490 	seek_iov(iov_in, niov_in, data_iov_in, &data_niov_in,
491 	    VTSCSI_IN_HEADER_LEN(sc));
492 	seek_iov(iov_out, niov_out, data_iov_out, &data_niov_out,
493 	    VTSCSI_OUT_HEADER_LEN(sc));
494 
495 	truncate_iov(iov_in, &niov_in, VTSCSI_IN_HEADER_LEN(sc));
496 	truncate_iov(iov_out, &niov_out, VTSCSI_OUT_HEADER_LEN(sc));
497 	iov_to_buf(iov_in, niov_in, (void **)&cmd_rd);
498 
499 	cmd_wr = calloc(1, VTSCSI_OUT_HEADER_LEN(sc));
500 	io = ctl_scsi_alloc_io(sc->vss_iid);
501 	ctl_scsi_zero_io(io);
502 
503 	io->io_hdr.nexus.initid = sc->vss_iid;
504 	io->io_hdr.nexus.targ_lun = pci_vtscsi_get_lun(cmd_rd->lun);
505 
506 	io->io_hdr.io_type = CTL_IO_SCSI;
507 
508 	if (data_niov_in > 0) {
509 		ext_data_ptr = (void *)data_iov_in;
510 		ext_sg_entries = data_niov_in;
511 		ext_data_len = count_iov(data_iov_in, data_niov_in);
512 		io->io_hdr.flags |= CTL_FLAG_DATA_OUT;
513 	} else if (data_niov_out > 0) {
514 		ext_data_ptr = (void *)data_iov_out;
515 		ext_sg_entries = data_niov_out;
516 		ext_data_len = count_iov(data_iov_out, data_niov_out);
517 		io->io_hdr.flags |= CTL_FLAG_DATA_IN;
518 	}
519 
520 	io->scsiio.sense_len = sc->vss_config.sense_size;
521 	io->scsiio.tag_num = cmd_rd->id;
522 	io->io_hdr.flags |= CTL_FLAG_USER_TAG;
523 	switch (cmd_rd->task_attr) {
524 	case VIRTIO_SCSI_S_ORDERED:
525 		io->scsiio.tag_type = CTL_TAG_ORDERED;
526 		break;
527 	case VIRTIO_SCSI_S_HEAD:
528 		io->scsiio.tag_type = CTL_TAG_HEAD_OF_QUEUE;
529 		break;
530 	case VIRTIO_SCSI_S_ACA:
531 		io->scsiio.tag_type = CTL_TAG_ACA;
532 		break;
533 	case VIRTIO_SCSI_S_SIMPLE:
534 	default:
535 		io->scsiio.tag_type = CTL_TAG_SIMPLE;
536 		break;
537 	}
538 	io->scsiio.ext_sg_entries = ext_sg_entries;
539 	io->scsiio.ext_data_ptr = ext_data_ptr;
540 	io->scsiio.ext_data_len = ext_data_len;
541 	io->scsiio.ext_data_filled = 0;
542 	io->scsiio.cdb_len = sc->vss_config.cdb_size;
543 	memcpy(io->scsiio.cdb, cmd_rd->cdb, sc->vss_config.cdb_size);
544 
545 	if (pci_vtscsi_debug) {
546 		struct sbuf *sb = sbuf_new_auto();
547 		ctl_io_sbuf(io, sb);
548 		sbuf_finish(sb);
549 		DPRINTF("%s", sbuf_data(sb));
550 		sbuf_delete(sb);
551 	}
552 
553 	err = ioctl(sc->vss_ctl_fd, CTL_IO, io);
554 	if (err != 0) {
555 		WPRINTF("CTL_IO: err=%d (%s)", errno, strerror(errno));
556 		cmd_wr->response = VIRTIO_SCSI_S_FAILURE;
557 	} else {
558 		cmd_wr->sense_len = MIN(io->scsiio.sense_len,
559 		    sc->vss_config.sense_size);
560 		cmd_wr->residual = ext_data_len - io->scsiio.ext_data_filled;
561 		cmd_wr->status = io->scsiio.scsi_status;
562 		cmd_wr->response = VIRTIO_SCSI_S_OK;
563 		memcpy(&cmd_wr->sense, &io->scsiio.sense_data,
564 		    cmd_wr->sense_len);
565 	}
566 
567 	buf_to_iov(cmd_wr, VTSCSI_OUT_HEADER_LEN(sc), iov_out, niov_out, 0);
568 	nxferred = VTSCSI_OUT_HEADER_LEN(sc) + io->scsiio.ext_data_filled;
569 	free(cmd_rd);
570 	free(cmd_wr);
571 	ctl_scsi_free_io(io);
572 	return (nxferred);
573 }
574 
575 static void
576 pci_vtscsi_controlq_notify(void *vsc, struct vqueue_info *vq)
577 {
578 	struct pci_vtscsi_softc *sc;
579 	struct iovec iov[VTSCSI_MAXSEG];
580 	struct vi_req req;
581 	void *buf = NULL;
582 	size_t bufsize;
583 	int iolen, n;
584 
585 	sc = vsc;
586 
587 	while (vq_has_descs(vq)) {
588 		n = vq_getchain(vq, iov, VTSCSI_MAXSEG, &req);
589 		assert(n >= 1 && n <= VTSCSI_MAXSEG);
590 
591 		bufsize = iov_to_buf(iov, n, &buf);
592 		iolen = pci_vtscsi_control_handle(sc, buf, bufsize);
593 		buf_to_iov((uint8_t *)buf + bufsize - iolen, iolen, iov, n,
594 		    bufsize - iolen);
595 
596 		/*
597 		 * Release this chain and handle more
598 		 */
599 		vq_relchain(vq, req.idx, iolen);
600 	}
601 	vq_endchains(vq, 1);	/* Generate interrupt if appropriate. */
602 	free(buf);
603 }
604 
605 static void
606 pci_vtscsi_eventq_notify(void *vsc __unused, struct vqueue_info *vq)
607 {
608 	vq_kick_disable(vq);
609 }
610 
611 static void
612 pci_vtscsi_requestq_notify(void *vsc, struct vqueue_info *vq)
613 {
614 	struct pci_vtscsi_softc *sc;
615 	struct pci_vtscsi_queue *q;
616 	struct pci_vtscsi_request *req;
617 	struct iovec iov[VTSCSI_MAXSEG];
618 	struct vi_req vireq;
619 	int n;
620 
621 	sc = vsc;
622 	q = &sc->vss_queues[vq->vq_num - 2];
623 
624 	while (vq_has_descs(vq)) {
625 		n = vq_getchain(vq, iov, VTSCSI_MAXSEG, &vireq);
626 		assert(n >= 1 && n <= VTSCSI_MAXSEG);
627 
628 		req = calloc(1, sizeof(struct pci_vtscsi_request));
629 		req->vsr_idx = vireq.idx;
630 		req->vsr_queue = q;
631 		req->vsr_niov_in = vireq.readable;
632 		req->vsr_niov_out = vireq.writable;
633 		memcpy(req->vsr_iov_in, iov,
634 		    req->vsr_niov_in * sizeof(struct iovec));
635 		memcpy(req->vsr_iov_out, iov + vireq.readable,
636 		    req->vsr_niov_out * sizeof(struct iovec));
637 
638 		pthread_mutex_lock(&q->vsq_mtx);
639 		STAILQ_INSERT_TAIL(&q->vsq_requests, req, vsr_link);
640 		pthread_cond_signal(&q->vsq_cv);
641 		pthread_mutex_unlock(&q->vsq_mtx);
642 
643 		DPRINTF("request <idx=%d> enqueued", vireq.idx);
644 	}
645 }
646 
647 static int
648 pci_vtscsi_init_queue(struct pci_vtscsi_softc *sc,
649     struct pci_vtscsi_queue *queue, int num)
650 {
651 	struct pci_vtscsi_worker *worker;
652 	char tname[MAXCOMLEN + 1];
653 	int i;
654 
655 	queue->vsq_sc = sc;
656 	queue->vsq_vq = &sc->vss_vq[num + 2];
657 
658 	pthread_mutex_init(&queue->vsq_mtx, NULL);
659 	pthread_mutex_init(&queue->vsq_qmtx, NULL);
660 	pthread_cond_init(&queue->vsq_cv, NULL);
661 	STAILQ_INIT(&queue->vsq_requests);
662 	LIST_INIT(&queue->vsq_workers);
663 
664 	for (i = 0; i < VTSCSI_THR_PER_Q; i++) {
665 		worker = calloc(1, sizeof(struct pci_vtscsi_worker));
666 		worker->vsw_queue = queue;
667 
668 		pthread_create(&worker->vsw_thread, NULL, &pci_vtscsi_proc,
669 		    (void *)worker);
670 
671 		snprintf(tname, sizeof(tname), "vtscsi:%d-%d", num, i);
672 		pthread_set_name_np(worker->vsw_thread, tname);
673 		LIST_INSERT_HEAD(&queue->vsq_workers, worker, vsw_link);
674 	}
675 
676 	return (0);
677 }
678 
679 static int
680 pci_vtscsi_legacy_config(nvlist_t *nvl, const char *opts)
681 {
682 	char *cp, *devname;
683 
684 	if (opts == NULL)
685 		return (0);
686 
687 	cp = strchr(opts, ',');
688 	if (cp == NULL) {
689 		set_config_value_node(nvl, "dev", opts);
690 		return (0);
691 	}
692 	devname = strndup(opts, cp - opts);
693 	set_config_value_node(nvl, "dev", devname);
694 	free(devname);
695 	return (pci_parse_legacy_config(nvl, cp + 1));
696 }
697 
698 static int
699 pci_vtscsi_init(struct pci_devinst *pi, nvlist_t *nvl)
700 {
701 	struct pci_vtscsi_softc *sc;
702 	const char *devname, *value;
703 	int i;
704 
705 	sc = calloc(1, sizeof(struct pci_vtscsi_softc));
706 	value = get_config_value_node(nvl, "iid");
707 	if (value != NULL)
708 		sc->vss_iid = strtoul(value, NULL, 10);
709 
710 	value = get_config_value_node(nvl, "bootindex");
711 	if (value != NULL) {
712 		if (pci_emul_add_boot_device(pi, atoi(value))) {
713 			EPRINTLN("Invalid bootindex %d", atoi(value));
714 			free(sc);
715 			return (-1);
716 		}
717 	}
718 
719 	devname = get_config_value_node(nvl, "dev");
720 	if (devname == NULL)
721 		devname = "/dev/cam/ctl";
722 	sc->vss_ctl_fd = open(devname, O_RDWR);
723 	if (sc->vss_ctl_fd < 0) {
724 		WPRINTF("cannot open %s: %s", devname, strerror(errno));
725 		free(sc);
726 		return (1);
727 	}
728 
729 	pthread_mutex_init(&sc->vss_mtx, NULL);
730 
731 	vi_softc_linkup(&sc->vss_vs, &vtscsi_vi_consts, sc, pi, sc->vss_vq);
732 	sc->vss_vs.vs_mtx = &sc->vss_mtx;
733 
734 	/* controlq */
735 	sc->vss_vq[0].vq_qsize = VTSCSI_RINGSZ;
736 	sc->vss_vq[0].vq_notify = pci_vtscsi_controlq_notify;
737 
738 	/* eventq */
739 	sc->vss_vq[1].vq_qsize = VTSCSI_RINGSZ;
740 	sc->vss_vq[1].vq_notify = pci_vtscsi_eventq_notify;
741 
742 	/* request queues */
743 	for (i = 2; i < VTSCSI_MAXQ; i++) {
744 		sc->vss_vq[i].vq_qsize = VTSCSI_RINGSZ;
745 		sc->vss_vq[i].vq_notify = pci_vtscsi_requestq_notify;
746 		pci_vtscsi_init_queue(sc, &sc->vss_queues[i - 2], i - 2);
747 	}
748 
749 	/* initialize config space */
750 	pci_set_cfgdata16(pi, PCIR_DEVICE, VIRTIO_DEV_SCSI);
751 	pci_set_cfgdata16(pi, PCIR_VENDOR, VIRTIO_VENDOR);
752 	pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_STORAGE);
753 	pci_set_cfgdata16(pi, PCIR_SUBDEV_0, VIRTIO_ID_SCSI);
754 	pci_set_cfgdata16(pi, PCIR_SUBVEND_0, VIRTIO_VENDOR);
755 
756 	if (vi_intr_init(&sc->vss_vs, 1, fbsdrun_virtio_msix()))
757 		return (1);
758 	vi_set_io_bar(&sc->vss_vs, 0);
759 
760 	return (0);
761 }
762 
763 
764 static const struct pci_devemu pci_de_vscsi = {
765 	.pe_emu =	"virtio-scsi",
766 	.pe_init =	pci_vtscsi_init,
767 	.pe_legacy_config = pci_vtscsi_legacy_config,
768 	.pe_barwrite =	vi_pci_write,
769 	.pe_barread =	vi_pci_read
770 };
771 PCI_EMUL_SET(pci_de_vscsi);
772