xref: /freebsd/sys/dev/beri/virtio/virtio_block.c (revision 43a5ec4eb41567cc92586503212743d89686d78f)
1 /*-
2  * Copyright (c) 2014 Ruslan Bukin <br@bsdpad.com>
3  * All rights reserved.
4  *
5  * This software was developed by SRI International and the University of
6  * Cambridge Computer Laboratory under DARPA/AFRL contract (FA8750-10-C-0237)
7  * ("CTSRD"), as part of the DARPA CRASH research programme.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
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 /*
32  * BERI virtio block backend driver
33  */
34 
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/bus.h>
41 #include <sys/kernel.h>
42 #include <sys/module.h>
43 #include <sys/rman.h>
44 #include <sys/conf.h>
45 #include <sys/stat.h>
46 #include <sys/endian.h>
47 #include <sys/disk.h>
48 #include <sys/vnode.h>
49 #include <sys/fcntl.h>
50 #include <sys/kthread.h>
51 #include <sys/buf.h>
52 #include <sys/mdioctl.h>
53 #include <sys/namei.h>
54 
55 #include <machine/bus.h>
56 #include <machine/fdt.h>
57 #include <machine/cpu.h>
58 #include <machine/intr.h>
59 
60 #include <dev/fdt/fdt_common.h>
61 #include <dev/ofw/openfirm.h>
62 #include <dev/ofw/ofw_bus.h>
63 #include <dev/ofw/ofw_bus_subr.h>
64 
65 #include <dev/beri/virtio/virtio.h>
66 #include <dev/beri/virtio/virtio_mmio_platform.h>
67 #include <dev/altera/pio/pio.h>
68 #include <dev/virtio/mmio/virtio_mmio.h>
69 #include <dev/virtio/block/virtio_blk.h>
70 #include <dev/virtio/virtio_ids.h>
71 #include <dev/virtio/virtio_config.h>
72 #include <dev/virtio/virtio_ring.h>
73 
74 #include "pio_if.h"
75 
76 #define DPRINTF(fmt, ...)
77 
78 /* We use indirect descriptors */
79 #define	NUM_DESCS	1
80 #define	NUM_QUEUES	1
81 
82 #define	VTBLK_BLK_ID_BYTES	20
83 #define	VTBLK_MAXSEGS		256
84 
85 struct beri_vtblk_softc {
86 	struct resource		*res[1];
87 	bus_space_tag_t		bst;
88 	bus_space_handle_t	bsh;
89 	struct cdev		*cdev;
90 	device_t		dev;
91 	int			opened;
92 	device_t		pio_recv;
93 	device_t		pio_send;
94 	struct vqueue_info	vs_queues[NUM_QUEUES];
95 	char			ident[VTBLK_BLK_ID_BYTES];
96 	struct ucred		*cred;
97 	struct vnode		*vnode;
98 	struct thread		*vtblk_ktd;
99 	struct sx		sc_mtx;
100 	int			beri_mem_offset;
101 	struct md_ioctl		*mdio;
102 	struct virtio_blk_config *cfg;
103 };
104 
105 static struct resource_spec beri_spec[] = {
106 	{ SYS_RES_MEMORY,	0,	RF_ACTIVE },
107 	{ -1, 0 }
108 };
109 
110 static int
111 vtblk_rdwr(struct beri_vtblk_softc *sc, struct iovec *iov,
112 	int cnt, int offset, int operation, int iolen)
113 {
114 	struct vnode *vp;
115 	struct mount *mp;
116 	struct uio auio;
117 	int error;
118 
119 	bzero(&auio, sizeof(auio));
120 
121 	vp = sc->vnode;
122 
123 	KASSERT(vp != NULL, ("file not opened"));
124 
125 	auio.uio_iov = iov;
126 	auio.uio_iovcnt = cnt;
127 	auio.uio_offset = offset;
128 	auio.uio_segflg = UIO_SYSSPACE;
129 	auio.uio_rw = operation;
130 	auio.uio_resid = iolen;
131 	auio.uio_td = curthread;
132 
133 	if (operation == 0) {
134 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
135 		error = VOP_READ(vp, &auio, IO_DIRECT, sc->cred);
136 		VOP_UNLOCK(vp);
137 	} else {
138 		(void) vn_start_write(vp, &mp, V_WAIT);
139 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
140 		error = VOP_WRITE(vp, &auio, IO_SYNC, sc->cred);
141 		VOP_UNLOCK(vp);
142 		vn_finished_write(mp);
143 	}
144 
145 	return (error);
146 }
147 
148 static void
149 vtblk_proc(struct beri_vtblk_softc *sc, struct vqueue_info *vq)
150 {
151 	struct iovec iov[VTBLK_MAXSEGS + 2];
152 	uint16_t flags[VTBLK_MAXSEGS + 2];
153 	struct virtio_blk_outhdr *vbh;
154 	struct iovec *tiov;
155 	uint8_t *status;
156 	off_t offset;
157 	int iolen;
158 	int type;
159 	int i, n;
160 	int err;
161 
162 	n = vq_getchain(sc->beri_mem_offset, vq, iov,
163 		VTBLK_MAXSEGS + 2, flags);
164 	KASSERT(n >= 2 && n <= VTBLK_MAXSEGS + 2,
165 		("wrong n value %d", n));
166 
167 	tiov = getcopy(iov, n);
168 	vbh = iov[0].iov_base;
169 
170 	status = iov[n-1].iov_base;
171 	KASSERT(iov[n-1].iov_len == 1,
172 		("iov_len == %d", iov[n-1].iov_len));
173 
174 	type = be32toh(vbh->type) & ~VIRTIO_BLK_T_BARRIER;
175 	offset = be64toh(vbh->sector) * DEV_BSIZE;
176 
177 	iolen = 0;
178 	for (i = 1; i < (n-1); i++) {
179 		iolen += iov[i].iov_len;
180 	}
181 
182 	switch (type) {
183 	case VIRTIO_BLK_T_OUT:
184 	case VIRTIO_BLK_T_IN:
185 		err = vtblk_rdwr(sc, tiov + 1, i - 1,
186 			offset, type, iolen);
187 		break;
188 	case VIRTIO_BLK_T_GET_ID:
189 		/* Assume a single buffer */
190 		strncpy(iov[1].iov_base, sc->ident,
191 		    MIN(iov[1].iov_len, sizeof(sc->ident)));
192 		err = 0;
193 		break;
194 	case VIRTIO_BLK_T_FLUSH:
195 		/* Possible? */
196 	default:
197 		err = -ENOSYS;
198 		break;
199 	}
200 
201 	if (err < 0) {
202 		if (err == -ENOSYS) {
203 			*status = VIRTIO_BLK_S_UNSUPP;
204 		} else
205 			*status = VIRTIO_BLK_S_IOERR;
206 	} else
207 		*status = VIRTIO_BLK_S_OK;
208 
209 	free(tiov, M_DEVBUF);
210 	vq_relchain(vq, iov, n, 1);
211 }
212 
213 static int
214 close_file(struct beri_vtblk_softc *sc, struct thread *td)
215 {
216 	int error;
217 
218 	if (sc->vnode != NULL) {
219 		vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY);
220 		sc->vnode->v_vflag &= ~VV_MD;
221 		VOP_UNLOCK(sc->vnode);
222 		error = vn_close(sc->vnode, (FREAD|FWRITE),
223 				sc->cred, td);
224 		if (error != 0)
225 			return (error);
226 		sc->vnode = NULL;
227 	}
228 
229 	if (sc->cred != NULL)
230 		crfree(sc->cred);
231 
232 	return (0);
233 }
234 
235 static int
236 open_file(struct beri_vtblk_softc *sc, struct thread *td)
237 {
238 	struct nameidata nd;
239 	struct vattr vattr;
240 	int error;
241 	int flags;
242 
243 	flags = (FREAD | FWRITE);
244 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, sc->mdio->md_file);
245 	error = vn_open(&nd, &flags, 0, NULL);
246 	if (error != 0)
247 		return (error);
248 	NDFREE(&nd, NDF_ONLY_PNBUF);
249 
250 	if (nd.ni_vp->v_type != VREG) {
251 		return (EINVAL);
252 	}
253 
254 	error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred);
255 	if (error != 0)
256 		return (error);
257 
258 	if (VOP_ISLOCKED(nd.ni_vp) != LK_EXCLUSIVE) {
259 		vn_lock(nd.ni_vp, LK_UPGRADE | LK_RETRY);
260 		if (VN_IS_DOOMED(nd.ni_vp)) {
261 			return (1);
262 		}
263 	}
264 	nd.ni_vp->v_vflag |= VV_MD;
265 	VOP_UNLOCK(nd.ni_vp);
266 
267 	sc->vnode = nd.ni_vp;
268 	sc->cred = crhold(td->td_ucred);
269 
270 	return (0);
271 }
272 
273 static int
274 vtblk_notify(struct beri_vtblk_softc *sc)
275 {
276 	struct vqueue_info *vq;
277 	int queue;
278 	int reg;
279 
280 	vq = &sc->vs_queues[0];
281 	if (!vq_ring_ready(vq))
282 		return (0);
283 
284 	if (!sc->opened)
285 		return (0);
286 
287 	reg = READ2(sc, VIRTIO_MMIO_QUEUE_NOTIFY);
288 	queue = be16toh(reg);
289 
290 	KASSERT(queue == 0, ("we support single queue only"));
291 
292 	/* Process new descriptors */
293 	vq = &sc->vs_queues[queue];
294 	vq->vq_save_used = be16toh(vq->vq_used->idx);
295 	while (vq_has_descs(vq))
296 		vtblk_proc(sc, vq);
297 
298 	/* Interrupt the other side */
299 	if ((be16toh(vq->vq_avail->flags) & VRING_AVAIL_F_NO_INTERRUPT) == 0) {
300 		reg = htobe32(VIRTIO_MMIO_INT_VRING);
301 		WRITE4(sc, VIRTIO_MMIO_INTERRUPT_STATUS, reg);
302 		PIO_SET(sc->pio_send, Q_INTR, 1);
303 	}
304 
305 	return (0);
306 }
307 
308 static int
309 vq_init(struct beri_vtblk_softc *sc)
310 {
311 	struct vqueue_info *vq;
312 	uint8_t *base;
313 	int size;
314 	int reg;
315 	int pfn;
316 
317 	vq = &sc->vs_queues[0];
318 	vq->vq_qsize = NUM_DESCS;
319 
320 	reg = READ4(sc, VIRTIO_MMIO_QUEUE_PFN);
321 	pfn = be32toh(reg);
322 	vq->vq_pfn = pfn;
323 
324 	size = vring_size(vq->vq_qsize, VRING_ALIGN);
325 	base = paddr_map(sc->beri_mem_offset,
326 		(pfn << PAGE_SHIFT), size);
327 
328 	/* First pages are descriptors */
329 	vq->vq_desc = (struct vring_desc *)base;
330 	base += vq->vq_qsize * sizeof(struct vring_desc);
331 
332 	/* Then avail ring */
333 	vq->vq_avail = (struct vring_avail *)base;
334 	base += (2 + vq->vq_qsize + 1) * sizeof(uint16_t);
335 
336 	/* Then it's rounded up to the next page */
337 	base = (uint8_t *)roundup2((uintptr_t)base, VRING_ALIGN);
338 
339 	/* And the last pages are the used ring */
340 	vq->vq_used = (struct vring_used *)base;
341 
342 	/* Mark queue as allocated, and start at 0 when we use it. */
343 	vq->vq_flags = VQ_ALLOC;
344 	vq->vq_last_avail = 0;
345 
346 	return (0);
347 }
348 
349 static void
350 vtblk_thread(void *arg)
351 {
352 	struct beri_vtblk_softc *sc;
353 	int err;
354 
355 	sc = arg;
356 
357 	sx_xlock(&sc->sc_mtx);
358 	for (;;) {
359 		err = msleep(sc, &sc->sc_mtx, PCATCH | PZERO, "prd", hz);
360 		vtblk_notify(sc);
361 	}
362 	sx_xunlock(&sc->sc_mtx);
363 
364 	kthread_exit();
365 }
366 
367 static int
368 backend_info(struct beri_vtblk_softc *sc)
369 {
370 	struct virtio_blk_config *cfg;
371 	uint32_t *s;
372 	int reg;
373 	int i;
374 
375 	/* Specify that we provide block device */
376 	reg = htobe32(VIRTIO_ID_BLOCK);
377 	WRITE4(sc, VIRTIO_MMIO_DEVICE_ID, reg);
378 
379 	/* Queue size */
380 	reg = htobe32(NUM_DESCS);
381 	WRITE4(sc, VIRTIO_MMIO_QUEUE_NUM_MAX, reg);
382 
383 	/* Our features */
384 	reg = htobe32(VIRTIO_RING_F_INDIRECT_DESC
385 	    | VIRTIO_BLK_F_BLK_SIZE
386 	    | VIRTIO_BLK_F_SEG_MAX);
387 	WRITE4(sc, VIRTIO_MMIO_HOST_FEATURES, reg);
388 
389 	cfg = sc->cfg;
390 	cfg->capacity = htobe64(sc->mdio->md_mediasize / DEV_BSIZE);
391 	cfg->size_max = 0; /* not negotiated */
392 	cfg->seg_max = htobe32(VTBLK_MAXSEGS);
393 	cfg->blk_size = htobe32(DEV_BSIZE);
394 
395 	s = (uint32_t *)cfg;
396 
397 	for (i = 0; i < sizeof(struct virtio_blk_config); i+=4) {
398 		WRITE4(sc, VIRTIO_MMIO_CONFIG + i, *s);
399 		s+=1;
400 	}
401 
402 	strncpy(sc->ident, "Virtio block backend", sizeof(sc->ident));
403 
404 	return (0);
405 }
406 
407 static void
408 vtblk_intr(void *arg)
409 {
410 	struct beri_vtblk_softc *sc;
411 	int pending;
412 	int reg;
413 
414 	sc = arg;
415 
416 	reg = PIO_READ(sc->pio_recv);
417 
418 	/* Ack */
419 	PIO_SET(sc->pio_recv, reg, 0);
420 
421 	pending = htobe32(reg);
422 
423 	if (pending & Q_PFN) {
424 		vq_init(sc);
425 	}
426 
427 	if (pending & Q_NOTIFY) {
428 		wakeup(sc);
429 	}
430 }
431 
432 static int
433 beri_ioctl(struct cdev *dev, u_long cmd, caddr_t addr,
434 		int flags, struct thread *td)
435 {
436 	struct beri_vtblk_softc *sc;
437 	int err;
438 
439 	sc = dev->si_drv1;
440 
441 	switch (cmd) {
442 	case MDIOCATTACH:
443 		/* take file as argument */
444 		if (sc->vnode != NULL) {
445 			/* Already opened */
446 			return (1);
447 		}
448 		sc->mdio = (struct md_ioctl *)addr;
449 		backend_info(sc);
450 		DPRINTF("opening file, td 0x%08x\n", (int)td);
451 		err = open_file(sc, td);
452 		if (err)
453 			return (err);
454 		PIO_SETUP_IRQ(sc->pio_recv, vtblk_intr, sc);
455 		sc->opened = 1;
456 		break;
457 	case MDIOCDETACH:
458 		if (sc->vnode == NULL) {
459 			/* File not opened */
460 			return (1);
461 		}
462 		sc->opened = 0;
463 		DPRINTF("closing file, td 0x%08x\n", (int)td);
464 		err = close_file(sc, td);
465 		if (err)
466 			return (err);
467 		PIO_TEARDOWN_IRQ(sc->pio_recv);
468 		break;
469 	default:
470 		break;
471 	}
472 
473 	return (0);
474 }
475 
476 static struct cdevsw beri_cdevsw = {
477 	.d_version =	D_VERSION,
478 	.d_ioctl =	beri_ioctl,
479 	.d_name =	"virtio block backend",
480 };
481 
482 static int
483 beri_vtblk_probe(device_t dev)
484 {
485 
486 	if (!ofw_bus_status_okay(dev))
487 		return (ENXIO);
488 
489 	if (!ofw_bus_is_compatible(dev, "sri-cambridge,beri-vtblk"))
490 		return (ENXIO);
491 
492 	device_set_desc(dev, "SRI-Cambridge BERI block");
493 	return (BUS_PROBE_DEFAULT);
494 }
495 
496 static int
497 beri_vtblk_attach(device_t dev)
498 {
499 	struct beri_vtblk_softc *sc;
500 	int error;
501 
502 	sc = device_get_softc(dev);
503 	sc->dev = dev;
504 
505 	if (bus_alloc_resources(dev, beri_spec, sc->res)) {
506 		device_printf(dev, "could not allocate resources\n");
507 		return (ENXIO);
508 	}
509 
510 	/* Memory interface */
511 	sc->bst = rman_get_bustag(sc->res[0]);
512 	sc->bsh = rman_get_bushandle(sc->res[0]);
513 
514 	sc->cfg = malloc(sizeof(struct virtio_blk_config),
515 		M_DEVBUF, M_NOWAIT|M_ZERO);
516 
517 	sx_init(&sc->sc_mtx, device_get_nameunit(sc->dev));
518 
519 	error = kthread_add(vtblk_thread, sc, NULL, &sc->vtblk_ktd,
520 		0, 0, "beri_virtio_block");
521 	if (error) {
522 		device_printf(dev, "cannot create kthread\n");
523 		return (ENXIO);
524 	}
525 
526 	if (setup_offset(dev, &sc->beri_mem_offset) != 0)
527 		return (ENXIO);
528 	if (setup_pio(dev, "pio-send", &sc->pio_send) != 0)
529 		return (ENXIO);
530 	if (setup_pio(dev, "pio-recv", &sc->pio_recv) != 0)
531 		return (ENXIO);
532 
533 	sc->cdev = make_dev(&beri_cdevsw, 0, UID_ROOT, GID_WHEEL,
534 	    S_IRWXU, "beri_vtblk");
535 	if (sc->cdev == NULL) {
536 		device_printf(dev, "Failed to create character device.\n");
537 		return (ENXIO);
538 	}
539 
540 	sc->cdev->si_drv1 = sc;
541 	return (0);
542 }
543 
544 static device_method_t beri_vtblk_methods[] = {
545 	DEVMETHOD(device_probe,		beri_vtblk_probe),
546 	DEVMETHOD(device_attach,	beri_vtblk_attach),
547 	{ 0, 0 }
548 };
549 
550 static driver_t beri_vtblk_driver = {
551 	"beri_vtblk",
552 	beri_vtblk_methods,
553 	sizeof(struct beri_vtblk_softc),
554 };
555 
556 static devclass_t beri_vtblk_devclass;
557 
558 DRIVER_MODULE(beri_vtblk, simplebus, beri_vtblk_driver,
559     beri_vtblk_devclass, 0, 0);
560