xref: /freebsd/usr.sbin/bhyve/pci_virtio_block.c (revision a66ffea41d7ea4e39a49bc146e6f6decb4fbd02c)
1 /*-
2  * Copyright (c) 2011 NetApp, Inc.
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  * $FreeBSD$
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/linker_set.h>
34 #include <sys/stat.h>
35 #include <sys/uio.h>
36 #include <sys/ioctl.h>
37 #include <sys/disk.h>
38 
39 #include <errno.h>
40 #include <fcntl.h>
41 #include <stdio.h>
42 #include <stdlib.h>
43 #include <stdint.h>
44 #include <string.h>
45 #include <strings.h>
46 #include <unistd.h>
47 #include <assert.h>
48 #include <pthread.h>
49 
50 #include "bhyverun.h"
51 #include "pci_emul.h"
52 #include "virtio.h"
53 
54 #define VTBLK_RINGSZ	64
55 
56 #define VTBLK_CFGSZ	28
57 
58 #define VTBLK_R_CFG		VTCFG_R_CFG1
59 #define VTBLK_R_CFG_END		VTBLK_R_CFG + VTBLK_CFGSZ -1
60 #define VTBLK_R_MAX		VTBLK_R_CFG_END
61 
62 #define VTBLK_REGSZ		VTBLK_R_MAX+1
63 
64 #define VTBLK_MAXSEGS	32
65 
66 #define VTBLK_S_OK	0
67 #define VTBLK_S_IOERR	1
68 
69 /*
70  * Host capabilities
71  */
72 #define VTBLK_S_HOSTCAPS      \
73   ( 0x00000004 |	/* host maximum request segments */ \
74     0x10000000 )	/* supports indirect descriptors */
75 
76 static int use_msix = 1;
77 
78 struct vring_hqueue {
79 	/* Internal state */
80 	uint16_t	hq_size;
81 	uint16_t	hq_cur_aidx;		/* trails behind 'avail_idx' */
82 
83 	 /* Host-context pointers to the queue */
84 	struct virtio_desc *hq_dtable;
85 	uint16_t	*hq_avail_flags;
86 	uint16_t	*hq_avail_idx;		/* monotonically increasing */
87 	uint16_t	*hq_avail_ring;
88 
89 	uint16_t	*hq_used_flags;
90 	uint16_t	*hq_used_idx;		/* monotonically increasing */
91 	struct virtio_used *hq_used_ring;
92 };
93 
94 /*
95  * Config space
96  */
97 struct vtblk_config {
98 	uint64_t	vbc_capacity;
99 	uint32_t	vbc_size_max;
100 	uint32_t	vbc_seg_max;
101 	uint16_t	vbc_geom_c;
102 	uint8_t		vbc_geom_h;
103 	uint8_t		vbc_geom_s;
104 	uint32_t	vbc_blk_size;
105 	uint32_t	vbc_sectors_max;
106 } __packed;
107 CTASSERT(sizeof(struct vtblk_config) == VTBLK_CFGSZ);
108 
109 /*
110  * Fixed-size block header
111  */
112 struct virtio_blk_hdr {
113 #define	VBH_OP_READ		0
114 #define	VBH_OP_WRITE		1
115 #define	VBH_FLAG_BARRIER	0x80000000	/* OR'ed into vbh_type */
116 	uint32_t       	vbh_type;
117 	uint32_t	vbh_ioprio;
118 	uint64_t	vbh_sector;
119 } __packed;
120 
121 /*
122  * Debug printf
123  */
124 static int pci_vtblk_debug;
125 #define DPRINTF(params) if (pci_vtblk_debug) printf params
126 #define WPRINTF(params) printf params
127 
128 /*
129  * Per-device softc
130  */
131 struct pci_vtblk_softc {
132 	struct pci_devinst *vbsc_pi;
133 	int		vbsc_fd;
134 	int		vbsc_status;
135 	int		vbsc_isr;
136 	int		vbsc_lastq;
137 	uint32_t	vbsc_features;
138 	uint64_t	vbsc_pfn;
139 	struct vring_hqueue vbsc_q;
140 	struct vtblk_config vbsc_cfg;
141 	uint16_t	msix_table_idx_req;
142 	uint16_t	msix_table_idx_cfg;
143 };
144 
145 /*
146  * Return the size of IO BAR that maps virtio header and device specific
147  * region. The size would vary depending on whether MSI-X is enabled or
148  * not
149  */
150 static uint64_t
151 pci_vtblk_iosize(struct pci_devinst *pi)
152 {
153 
154 	if (pci_msix_enabled(pi))
155 		return (VTBLK_REGSZ);
156 	else
157 		return (VTBLK_REGSZ - (VTCFG_R_CFG1 - VTCFG_R_MSIX));
158 }
159 
160 /*
161  * Return the number of available descriptors in the vring taking care
162  * of the 16-bit index wraparound.
163  */
164 static int
165 hq_num_avail(struct vring_hqueue *hq)
166 {
167 	int ndesc;
168 
169 	if (*hq->hq_avail_idx >= hq->hq_cur_aidx)
170 		ndesc = *hq->hq_avail_idx - hq->hq_cur_aidx;
171 	else
172 		ndesc = UINT16_MAX - hq->hq_cur_aidx + *hq->hq_avail_idx + 1;
173 
174 	assert(ndesc >= 0 && ndesc <= hq->hq_size);
175 
176 	return (ndesc);
177 }
178 
179 static void
180 pci_vtblk_update_status(struct pci_vtblk_softc *sc, uint32_t value)
181 {
182 	if (value == 0) {
183 		DPRINTF(("vtblk: device reset requested !\n"));
184 	}
185 
186 	sc->vbsc_status = value;
187 }
188 
189 static void
190 pci_vtblk_proc(struct pci_vtblk_softc *sc, struct vring_hqueue *hq)
191 {
192 	struct iovec iov[VTBLK_MAXSEGS];
193 	struct virtio_blk_hdr *vbh;
194 	struct virtio_desc *vd, *vid;
195 	struct virtio_used *vu;
196 	uint8_t *status;
197 	int i;
198 	int err;
199 	int iolen;
200 	int nsegs;
201 	int uidx, aidx, didx;
202 	int writeop, type;
203 	off_t offset;
204 
205 	uidx = *hq->hq_used_idx;
206 	aidx = hq->hq_cur_aidx;
207 	didx = hq->hq_avail_ring[aidx % hq->hq_size];
208 	assert(didx >= 0 && didx < hq->hq_size);
209 
210 	vd = &hq->hq_dtable[didx];
211 
212 	/*
213 	 * Verify that the descriptor is indirect, and obtain
214 	 * the pointer to the indirect descriptor.
215 	 * There has to be space for at least 3 descriptors
216 	 * in the indirect descriptor array: the block header,
217 	 * 1 or more data descriptors, and a status byte.
218 	 */
219 	assert(vd->vd_flags & VRING_DESC_F_INDIRECT);
220 
221 	nsegs = vd->vd_len / sizeof(struct virtio_desc);
222 	assert(nsegs >= 3);
223 	assert(nsegs < VTBLK_MAXSEGS + 2);
224 
225 	vid = paddr_guest2host(vd->vd_addr);
226 	assert((vid->vd_flags & VRING_DESC_F_INDIRECT) == 0);
227 
228 	/*
229 	 * The first descriptor will be the read-only fixed header
230 	 */
231 	vbh = paddr_guest2host(vid[0].vd_addr);
232 	assert(vid[0].vd_len == sizeof(struct virtio_blk_hdr));
233 	assert(vid[0].vd_flags & VRING_DESC_F_NEXT);
234 	assert((vid[0].vd_flags & VRING_DESC_F_WRITE) == 0);
235 
236 	/*
237 	 * XXX
238 	 * The guest should not be setting the BARRIER flag because
239 	 * we don't advertise the capability.
240 	 */
241 	type = vbh->vbh_type & ~VBH_FLAG_BARRIER;
242 	writeop = (type == VBH_OP_WRITE);
243 
244 	offset = vbh->vbh_sector * DEV_BSIZE;
245 
246 	/*
247 	 * Build up the iovec based on the guest's data descriptors
248 	 */
249 	for (i = 1, iolen = 0; i < nsegs - 1; i++) {
250 		iov[i-1].iov_base = paddr_guest2host(vid[i].vd_addr);
251 		iov[i-1].iov_len = vid[i].vd_len;
252 		iolen += vid[i].vd_len;
253 
254 		assert(vid[i].vd_flags & VRING_DESC_F_NEXT);
255 		assert((vid[i].vd_flags & VRING_DESC_F_INDIRECT) == 0);
256 
257 		/*
258 		 * - write op implies read-only descriptor,
259 		 * - read op implies write-only descriptor,
260 		 * therefore test the inverse of the descriptor bit
261 		 * to the op.
262 		 */
263 		assert(((vid[i].vd_flags & VRING_DESC_F_WRITE) == 0) ==
264 		       writeop);
265 	}
266 
267 	/* Lastly, get the address of the status byte */
268 	status = paddr_guest2host(vid[nsegs - 1].vd_addr);
269 	assert(vid[nsegs - 1].vd_len == 1);
270 	assert((vid[nsegs - 1].vd_flags & VRING_DESC_F_NEXT) == 0);
271 	assert(vid[nsegs - 1].vd_flags & VRING_DESC_F_WRITE);
272 
273 	DPRINTF(("virtio-block: %s op, %d bytes, %d segs, offset %ld\n\r",
274 		 writeop ? "write" : "read", iolen, nsegs - 2, offset));
275 
276 	if (writeop){
277 		err = pwritev(sc->vbsc_fd, iov, nsegs - 2, offset);
278 	} else {
279 		err = preadv(sc->vbsc_fd, iov, nsegs - 2, offset);
280 	}
281 
282 	*status = err < 0 ? VTBLK_S_IOERR : VTBLK_S_OK;
283 
284 	/*
285 	 * Return the single indirect descriptor back to the host
286 	 */
287 	vu = &hq->hq_used_ring[uidx % hq->hq_size];
288 	vu->vu_idx = didx;
289 	vu->vu_tlen = 1;
290 	hq->hq_cur_aidx++;
291 	*hq->hq_used_idx += 1;
292 }
293 
294 static void
295 pci_vtblk_qnotify(struct pci_vtblk_softc *sc)
296 {
297 	struct vring_hqueue *hq = &sc->vbsc_q;
298 	int i;
299 	int ndescs;
300 
301 	/*
302 	 * Calculate number of ring entries to process
303 	 */
304 	ndescs = hq_num_avail(hq);
305 
306 	if (ndescs == 0)
307 		return;
308 
309 	/*
310 	 * Run through all the entries, placing them into iovecs and
311 	 * sending when an end-of-packet is found
312 	 */
313 	for (i = 0; i < ndescs; i++)
314 		pci_vtblk_proc(sc, hq);
315 
316 	/*
317 	 * Generate an interrupt if able
318 	 */
319 	if ((*hq->hq_avail_flags & VRING_AVAIL_F_NO_INTERRUPT) == 0) {
320 		if (use_msix) {
321 			pci_generate_msix(sc->vbsc_pi, sc->msix_table_idx_req);
322 		} else if (sc->vbsc_isr == 0) {
323 			sc->vbsc_isr = 1;
324 			pci_generate_msi(sc->vbsc_pi, 0);
325 		}
326 	}
327 
328 }
329 
330 static void
331 pci_vtblk_ring_init(struct pci_vtblk_softc *sc, uint64_t pfn)
332 {
333 	struct vring_hqueue *hq;
334 
335 	sc->vbsc_pfn = pfn << VRING_PFN;
336 
337 	/*
338 	 * Set up host pointers to the various parts of the
339 	 * queue
340 	 */
341 	hq = &sc->vbsc_q;
342 	hq->hq_size = VTBLK_RINGSZ;
343 
344 	hq->hq_dtable = paddr_guest2host(pfn << VRING_PFN);
345 	hq->hq_avail_flags =  (uint16_t *)(hq->hq_dtable + hq->hq_size);
346 	hq->hq_avail_idx = hq->hq_avail_flags + 1;
347 	hq->hq_avail_ring = hq->hq_avail_flags + 2;
348 	hq->hq_used_flags = (uint16_t *)roundup2((uintptr_t)hq->hq_avail_ring,
349 						 VRING_ALIGN);
350 	hq->hq_used_idx = hq->hq_used_flags + 1;
351 	hq->hq_used_ring = (struct virtio_used *)(hq->hq_used_flags + 2);
352 
353 	/*
354 	 * Initialize queue indexes
355 	 */
356 	hq->hq_cur_aidx = 0;
357 }
358 
359 static int
360 pci_vtblk_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
361 {
362 	struct stat sbuf;
363 	struct pci_vtblk_softc *sc;
364 	off_t size;
365 	int fd;
366 	int sectsz;
367 	const char *env_msi;
368 
369 	if (opts == NULL) {
370 		printf("virtio-block: backing device required\n");
371 		return (1);
372 	}
373 
374 	/*
375 	 * Access to guest memory is required. Fail if
376 	 * memory not mapped
377 	 */
378 	if (paddr_guest2host(0) == NULL)
379 		return (1);
380 
381 	/*
382 	 * The supplied backing file has to exist
383 	 */
384 	fd = open(opts, O_RDWR);
385 	if (fd < 0) {
386 		perror("Could not open backing file");
387 		return (1);
388 	}
389 
390 	if (fstat(fd, &sbuf) < 0) {
391 		perror("Could not stat backing file");
392 		close(fd);
393 		return (1);
394 	}
395 
396 	/*
397 	 * Deal with raw devices
398 	 */
399 	size = sbuf.st_size;
400 	sectsz = DEV_BSIZE;
401 	if (S_ISCHR(sbuf.st_mode)) {
402 		if (ioctl(fd, DIOCGMEDIASIZE, &size) < 0 ||
403 		    ioctl(fd, DIOCGSECTORSIZE, &sectsz)) {
404 			perror("Could not fetch dev blk/sector size");
405 			close(fd);
406 			return (1);
407 		}
408 		assert(size != 0);
409 		assert(sectsz != 0);
410 	}
411 
412 	sc = malloc(sizeof(struct pci_vtblk_softc));
413 	memset(sc, 0, sizeof(struct pci_vtblk_softc));
414 
415 	pi->pi_arg = sc;
416 	sc->vbsc_pi = pi;
417 	sc->vbsc_fd = fd;
418 
419 	/* setup virtio block config space */
420 	sc->vbsc_cfg.vbc_capacity = size / sectsz;
421 	sc->vbsc_cfg.vbc_seg_max = VTBLK_MAXSEGS;
422 	sc->vbsc_cfg.vbc_blk_size = sectsz;
423 	sc->vbsc_cfg.vbc_size_max = 0;	/* not negotiated */
424 	sc->vbsc_cfg.vbc_geom_c = 0;	/* no geometry */
425 	sc->vbsc_cfg.vbc_geom_h = 0;
426 	sc->vbsc_cfg.vbc_geom_s = 0;
427 	sc->vbsc_cfg.vbc_sectors_max = 0;
428 
429 	/* initialize config space */
430 	pci_set_cfgdata16(pi, PCIR_DEVICE, VIRTIO_DEV_BLOCK);
431 	pci_set_cfgdata16(pi, PCIR_VENDOR, VIRTIO_VENDOR);
432 	pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_STORAGE);
433 	pci_set_cfgdata16(pi, PCIR_SUBDEV_0, VIRTIO_TYPE_BLOCK);
434 
435 	if ((env_msi = getenv("BHYVE_USE_MSI"))) {
436 		if (strcasecmp(env_msi, "yes") == 0)
437 			use_msix = 0;
438 	}
439 
440 	if (use_msix) {
441 		/* MSI-X Support */
442 		sc->msix_table_idx_req = VIRTIO_MSI_NO_VECTOR;
443 		sc->msix_table_idx_cfg = VIRTIO_MSI_NO_VECTOR;
444 
445 		if (pci_emul_add_msixcap(pi, 2, 1))
446 			return (1);
447 	} else {
448 		/* MSI Support */
449 		pci_emul_add_msicap(pi, 1);
450 	}
451 
452 	pci_emul_alloc_bar(pi, 0, PCIBAR_IO, VTBLK_REGSZ);
453 
454 	return (0);
455 }
456 
457 static uint64_t
458 vtblk_adjust_offset(struct pci_devinst *pi, uint64_t offset)
459 {
460 	/*
461 	 * Device specific offsets used by guest would change
462 	 * based on whether MSI-X capability is enabled or not
463 	 */
464 	if (!pci_msix_enabled(pi)) {
465 		if (offset >= VTCFG_R_MSIX)
466 			return (offset + (VTCFG_R_CFG1 - VTCFG_R_MSIX));
467 	}
468 
469 	return (offset);
470 }
471 
472 static void
473 pci_vtblk_write(struct vmctx *ctx, int vcpu, struct pci_devinst *pi,
474 		int baridx, uint64_t offset, int size, uint64_t value)
475 {
476 	struct pci_vtblk_softc *sc = pi->pi_arg;
477 
478 	if (use_msix) {
479 		if (baridx == pci_msix_table_bar(pi) ||
480 		    baridx == pci_msix_pba_bar(pi)) {
481 			pci_emul_msix_twrite(pi, offset, size, value);
482 			return;
483 		}
484 	}
485 
486 	assert(baridx == 0);
487 
488 	if (offset + size > pci_vtblk_iosize(pi)) {
489 		DPRINTF(("vtblk_write: 2big, offset %ld size %d\n",
490 			 offset, size));
491 		return;
492 	}
493 
494 	offset = vtblk_adjust_offset(pi, offset);
495 
496 	switch (offset) {
497 	case VTCFG_R_GUESTCAP:
498 		assert(size == 4);
499 		sc->vbsc_features = value & VTBLK_S_HOSTCAPS;
500 		break;
501 	case VTCFG_R_PFN:
502 		assert(size == 4);
503 		pci_vtblk_ring_init(sc, value);
504 		break;
505 	case VTCFG_R_QSEL:
506 		assert(size == 2);
507 		sc->vbsc_lastq = value;
508 		break;
509 	case VTCFG_R_QNOTIFY:
510 		assert(size == 2);
511 		assert(value == 0);
512 		pci_vtblk_qnotify(sc);
513 		break;
514 	case VTCFG_R_STATUS:
515 		assert(size == 1);
516 		pci_vtblk_update_status(sc, value);
517 		break;
518 	case VTCFG_R_CFGVEC:
519 		assert(size == 2);
520 		sc->msix_table_idx_cfg = value;
521 		break;
522 	case VTCFG_R_QVEC:
523 		assert(size == 2);
524 		sc->msix_table_idx_req = value;
525 		break;
526 	case VTCFG_R_HOSTCAP:
527 	case VTCFG_R_QNUM:
528 	case VTCFG_R_ISR:
529 	case VTBLK_R_CFG ... VTBLK_R_CFG_END:
530 		DPRINTF(("vtblk: write to readonly reg %ld\n\r", offset));
531 		break;
532 	default:
533 		DPRINTF(("vtblk: unknown i/o write offset %ld\n\r", offset));
534 		value = 0;
535 		break;
536 	}
537 }
538 
539 uint64_t
540 pci_vtblk_read(struct vmctx *ctx, int vcpu, struct pci_devinst *pi,
541 	       int baridx, uint64_t offset, int size)
542 {
543 	struct pci_vtblk_softc *sc = pi->pi_arg;
544 	void *ptr;
545 	uint32_t value;
546 
547 	if (use_msix) {
548 		if (baridx == pci_msix_table_bar(pi) ||
549 		    baridx == pci_msix_pba_bar(pi)) {
550 			return (pci_emul_msix_tread(pi, offset, size));
551 		}
552 	}
553 
554 	assert(baridx == 0);
555 
556 	if (offset + size > pci_vtblk_iosize(pi)) {
557 		DPRINTF(("vtblk_read: 2big, offset %ld size %d\n",
558 			 offset, size));
559 		return (0);
560 	}
561 
562 	offset = vtblk_adjust_offset(pi, offset);
563 
564 	switch (offset) {
565 	case VTCFG_R_HOSTCAP:
566 		assert(size == 4);
567 		value = VTBLK_S_HOSTCAPS;
568 		break;
569 	case VTCFG_R_GUESTCAP:
570 		assert(size == 4);
571 		value = sc->vbsc_features; /* XXX never read ? */
572 		break;
573 	case VTCFG_R_PFN:
574 		assert(size == 4);
575 		value = sc->vbsc_pfn >> VRING_PFN;
576 		break;
577 	case VTCFG_R_QNUM:
578 		value = (sc->vbsc_lastq == 0) ? VTBLK_RINGSZ: 0;
579 		break;
580 	case VTCFG_R_QSEL:
581 		assert(size == 2);
582 		value = sc->vbsc_lastq; /* XXX never read ? */
583 		break;
584 	case VTCFG_R_QNOTIFY:
585 		assert(size == 2);
586 		value = 0; /* XXX never read ? */
587 		break;
588 	case VTCFG_R_STATUS:
589 		assert(size == 1);
590 		value = sc->vbsc_status;
591 		break;
592 	case VTCFG_R_ISR:
593 		assert(size == 1);
594 		value = sc->vbsc_isr;
595 		sc->vbsc_isr = 0;     /* a read clears this flag */
596 		break;
597 	case VTCFG_R_CFGVEC:
598 		assert(size == 2);
599 		value = sc->msix_table_idx_cfg;
600 		break;
601 	case VTCFG_R_QVEC:
602 		assert(size == 2);
603 		value = sc->msix_table_idx_req;
604 		break;
605 	case VTBLK_R_CFG ... VTBLK_R_CFG_END:
606 		assert(size + offset <= (VTBLK_R_CFG_END + 1));
607 		ptr = (uint8_t *)&sc->vbsc_cfg + offset - VTBLK_R_CFG;
608 		if (size == 1) {
609 			value = *(uint8_t *) ptr;
610 		} else if (size == 2) {
611 			value = *(uint16_t *) ptr;
612 		} else {
613 			value = *(uint32_t *) ptr;
614 		}
615 		break;
616 	default:
617 		DPRINTF(("vtblk: unknown i/o read offset %ld\n\r", offset));
618 		value = 0;
619 		break;
620 	}
621 
622 	return (value);
623 }
624 
625 struct pci_devemu pci_de_vblk = {
626 	.pe_emu =	"virtio-blk",
627 	.pe_init =	pci_vtblk_init,
628 	.pe_barwrite =	pci_vtblk_write,
629 	.pe_barread =	pci_vtblk_read
630 };
631 PCI_EMUL_SET(pci_de_vblk);
632