xref: /freebsd/sys/dev/beri/virtio/virtio_block.c (revision fdafd315)
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/param.h>
36 #include <sys/systm.h>
37 #include <sys/bus.h>
38 #include <sys/kernel.h>
39 #include <sys/module.h>
40 #include <sys/rman.h>
41 #include <sys/conf.h>
42 #include <sys/stat.h>
43 #include <sys/endian.h>
44 #include <sys/disk.h>
45 #include <sys/vnode.h>
46 #include <sys/fcntl.h>
47 #include <sys/kthread.h>
48 #include <sys/buf.h>
49 #include <sys/mdioctl.h>
50 #include <sys/namei.h>
51 
52 #include <machine/bus.h>
53 #include <machine/fdt.h>
54 #include <machine/cpu.h>
55 #include <machine/intr.h>
56 
57 #include <dev/fdt/fdt_common.h>
58 #include <dev/ofw/openfirm.h>
59 #include <dev/ofw/ofw_bus.h>
60 #include <dev/ofw/ofw_bus_subr.h>
61 
62 #include <dev/beri/virtio/virtio.h>
63 #include <dev/beri/virtio/virtio_mmio_platform.h>
64 #include <dev/altera/pio/pio.h>
65 #include <dev/virtio/mmio/virtio_mmio.h>
66 #include <dev/virtio/block/virtio_blk.h>
67 #include <dev/virtio/virtio_ids.h>
68 #include <dev/virtio/virtio_config.h>
69 #include <dev/virtio/virtio_ring.h>
70 
71 #include "pio_if.h"
72 
73 #define DPRINTF(fmt, ...)
74 
75 /* We use indirect descriptors */
76 #define	NUM_DESCS	1
77 #define	NUM_QUEUES	1
78 
79 #define	VTBLK_BLK_ID_BYTES	20
80 #define	VTBLK_MAXSEGS		256
81 
82 struct beri_vtblk_softc {
83 	struct resource		*res[1];
84 	bus_space_tag_t		bst;
85 	bus_space_handle_t	bsh;
86 	struct cdev		*cdev;
87 	device_t		dev;
88 	int			opened;
89 	device_t		pio_recv;
90 	device_t		pio_send;
91 	struct vqueue_info	vs_queues[NUM_QUEUES];
92 	char			ident[VTBLK_BLK_ID_BYTES];
93 	struct ucred		*cred;
94 	struct vnode		*vnode;
95 	struct thread		*vtblk_ktd;
96 	struct sx		sc_mtx;
97 	int			beri_mem_offset;
98 	struct md_ioctl		*mdio;
99 	struct virtio_blk_config *cfg;
100 };
101 
102 static struct resource_spec beri_spec[] = {
103 	{ SYS_RES_MEMORY,	0,	RF_ACTIVE },
104 	{ -1, 0 }
105 };
106 
107 static int
vtblk_rdwr(struct beri_vtblk_softc * sc,struct iovec * iov,int cnt,int offset,int operation,int iolen)108 vtblk_rdwr(struct beri_vtblk_softc *sc, struct iovec *iov,
109 	int cnt, int offset, int operation, int iolen)
110 {
111 	struct vnode *vp;
112 	struct mount *mp;
113 	struct uio auio;
114 	int error;
115 
116 	bzero(&auio, sizeof(auio));
117 
118 	vp = sc->vnode;
119 
120 	KASSERT(vp != NULL, ("file not opened"));
121 
122 	auio.uio_iov = iov;
123 	auio.uio_iovcnt = cnt;
124 	auio.uio_offset = offset;
125 	auio.uio_segflg = UIO_SYSSPACE;
126 	auio.uio_rw = operation;
127 	auio.uio_resid = iolen;
128 	auio.uio_td = curthread;
129 
130 	if (operation == 0) {
131 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
132 		error = VOP_READ(vp, &auio, IO_DIRECT, sc->cred);
133 		VOP_UNLOCK(vp);
134 	} else {
135 		(void) vn_start_write(vp, &mp, V_WAIT);
136 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
137 		error = VOP_WRITE(vp, &auio, IO_SYNC, sc->cred);
138 		VOP_UNLOCK(vp);
139 		vn_finished_write(mp);
140 	}
141 
142 	return (error);
143 }
144 
145 static void
vtblk_proc(struct beri_vtblk_softc * sc,struct vqueue_info * vq)146 vtblk_proc(struct beri_vtblk_softc *sc, struct vqueue_info *vq)
147 {
148 	struct iovec iov[VTBLK_MAXSEGS + 2];
149 	uint16_t flags[VTBLK_MAXSEGS + 2];
150 	struct virtio_blk_outhdr *vbh;
151 	struct iovec *tiov;
152 	uint8_t *status;
153 	off_t offset;
154 	int iolen;
155 	int type;
156 	int i, n;
157 	int err;
158 
159 	n = vq_getchain(sc->beri_mem_offset, vq, iov,
160 		VTBLK_MAXSEGS + 2, flags);
161 	KASSERT(n >= 2 && n <= VTBLK_MAXSEGS + 2,
162 		("wrong n value %d", n));
163 
164 	tiov = getcopy(iov, n);
165 	vbh = iov[0].iov_base;
166 
167 	status = iov[n-1].iov_base;
168 	KASSERT(iov[n-1].iov_len == 1,
169 		("iov_len == %d", iov[n-1].iov_len));
170 
171 	type = be32toh(vbh->type) & ~VIRTIO_BLK_T_BARRIER;
172 	offset = be64toh(vbh->sector) * DEV_BSIZE;
173 
174 	iolen = 0;
175 	for (i = 1; i < (n-1); i++) {
176 		iolen += iov[i].iov_len;
177 	}
178 
179 	switch (type) {
180 	case VIRTIO_BLK_T_OUT:
181 	case VIRTIO_BLK_T_IN:
182 		err = vtblk_rdwr(sc, tiov + 1, i - 1,
183 			offset, type, iolen);
184 		break;
185 	case VIRTIO_BLK_T_GET_ID:
186 		/* Assume a single buffer */
187 		strncpy(iov[1].iov_base, sc->ident,
188 		    MIN(iov[1].iov_len, sizeof(sc->ident)));
189 		err = 0;
190 		break;
191 	case VIRTIO_BLK_T_FLUSH:
192 		/* Possible? */
193 	default:
194 		err = -ENOSYS;
195 		break;
196 	}
197 
198 	if (err < 0) {
199 		if (err == -ENOSYS) {
200 			*status = VIRTIO_BLK_S_UNSUPP;
201 		} else
202 			*status = VIRTIO_BLK_S_IOERR;
203 	} else
204 		*status = VIRTIO_BLK_S_OK;
205 
206 	free(tiov, M_DEVBUF);
207 	vq_relchain(vq, iov, n, 1);
208 }
209 
210 static int
close_file(struct beri_vtblk_softc * sc,struct thread * td)211 close_file(struct beri_vtblk_softc *sc, struct thread *td)
212 {
213 	int error;
214 
215 	if (sc->vnode != NULL) {
216 		vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY);
217 		sc->vnode->v_vflag &= ~VV_MD;
218 		VOP_UNLOCK(sc->vnode);
219 		error = vn_close(sc->vnode, (FREAD|FWRITE),
220 				sc->cred, td);
221 		if (error != 0)
222 			return (error);
223 		sc->vnode = NULL;
224 	}
225 
226 	if (sc->cred != NULL)
227 		crfree(sc->cred);
228 
229 	return (0);
230 }
231 
232 static int
open_file(struct beri_vtblk_softc * sc,struct thread * td)233 open_file(struct beri_vtblk_softc *sc, struct thread *td)
234 {
235 	struct nameidata nd;
236 	struct vattr vattr;
237 	int error;
238 	int flags;
239 
240 	flags = (FREAD | FWRITE);
241 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, sc->mdio->md_file);
242 	error = vn_open(&nd, &flags, 0, NULL);
243 	if (error != 0)
244 		return (error);
245 	NDFREE_PNBUF(&nd);
246 
247 	if (nd.ni_vp->v_type != VREG) {
248 		return (EINVAL);
249 	}
250 
251 	error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred);
252 	if (error != 0)
253 		return (error);
254 
255 	if (VOP_ISLOCKED(nd.ni_vp) != LK_EXCLUSIVE) {
256 		vn_lock(nd.ni_vp, LK_UPGRADE | LK_RETRY);
257 		if (VN_IS_DOOMED(nd.ni_vp)) {
258 			return (1);
259 		}
260 	}
261 	nd.ni_vp->v_vflag |= VV_MD;
262 	VOP_UNLOCK(nd.ni_vp);
263 
264 	sc->vnode = nd.ni_vp;
265 	sc->cred = crhold(td->td_ucred);
266 
267 	return (0);
268 }
269 
270 static int
vtblk_notify(struct beri_vtblk_softc * sc)271 vtblk_notify(struct beri_vtblk_softc *sc)
272 {
273 	struct vqueue_info *vq;
274 	int queue;
275 	int reg;
276 
277 	vq = &sc->vs_queues[0];
278 	if (!vq_ring_ready(vq))
279 		return (0);
280 
281 	if (!sc->opened)
282 		return (0);
283 
284 	reg = READ2(sc, VIRTIO_MMIO_QUEUE_NOTIFY);
285 	queue = be16toh(reg);
286 
287 	KASSERT(queue == 0, ("we support single queue only"));
288 
289 	/* Process new descriptors */
290 	vq = &sc->vs_queues[queue];
291 	vq->vq_save_used = be16toh(vq->vq_used->idx);
292 	while (vq_has_descs(vq))
293 		vtblk_proc(sc, vq);
294 
295 	/* Interrupt the other side */
296 	if ((be16toh(vq->vq_avail->flags) & VRING_AVAIL_F_NO_INTERRUPT) == 0) {
297 		reg = htobe32(VIRTIO_MMIO_INT_VRING);
298 		WRITE4(sc, VIRTIO_MMIO_INTERRUPT_STATUS, reg);
299 		PIO_SET(sc->pio_send, Q_INTR, 1);
300 	}
301 
302 	return (0);
303 }
304 
305 static int
vq_init(struct beri_vtblk_softc * sc)306 vq_init(struct beri_vtblk_softc *sc)
307 {
308 	struct vqueue_info *vq;
309 	uint8_t *base;
310 	int size;
311 	int reg;
312 	int pfn;
313 
314 	vq = &sc->vs_queues[0];
315 	vq->vq_qsize = NUM_DESCS;
316 
317 	reg = READ4(sc, VIRTIO_MMIO_QUEUE_PFN);
318 	pfn = be32toh(reg);
319 	vq->vq_pfn = pfn;
320 
321 	size = vring_size(vq->vq_qsize, VRING_ALIGN);
322 	base = paddr_map(sc->beri_mem_offset,
323 		(pfn << PAGE_SHIFT), size);
324 
325 	/* First pages are descriptors */
326 	vq->vq_desc = (struct vring_desc *)base;
327 	base += vq->vq_qsize * sizeof(struct vring_desc);
328 
329 	/* Then avail ring */
330 	vq->vq_avail = (struct vring_avail *)base;
331 	base += (2 + vq->vq_qsize + 1) * sizeof(uint16_t);
332 
333 	/* Then it's rounded up to the next page */
334 	base = (uint8_t *)roundup2((uintptr_t)base, VRING_ALIGN);
335 
336 	/* And the last pages are the used ring */
337 	vq->vq_used = (struct vring_used *)base;
338 
339 	/* Mark queue as allocated, and start at 0 when we use it. */
340 	vq->vq_flags = VQ_ALLOC;
341 	vq->vq_last_avail = 0;
342 
343 	return (0);
344 }
345 
346 static void
vtblk_thread(void * arg)347 vtblk_thread(void *arg)
348 {
349 	struct beri_vtblk_softc *sc;
350 	int err;
351 
352 	sc = arg;
353 
354 	sx_xlock(&sc->sc_mtx);
355 	for (;;) {
356 		err = msleep(sc, &sc->sc_mtx, PCATCH | PZERO, "prd", hz);
357 		vtblk_notify(sc);
358 	}
359 	sx_xunlock(&sc->sc_mtx);
360 
361 	kthread_exit();
362 }
363 
364 static int
backend_info(struct beri_vtblk_softc * sc)365 backend_info(struct beri_vtblk_softc *sc)
366 {
367 	struct virtio_blk_config *cfg;
368 	uint32_t *s;
369 	int reg;
370 	int i;
371 
372 	/* Specify that we provide block device */
373 	reg = htobe32(VIRTIO_ID_BLOCK);
374 	WRITE4(sc, VIRTIO_MMIO_DEVICE_ID, reg);
375 
376 	/* Queue size */
377 	reg = htobe32(NUM_DESCS);
378 	WRITE4(sc, VIRTIO_MMIO_QUEUE_NUM_MAX, reg);
379 
380 	/* Our features */
381 	reg = htobe32(VIRTIO_RING_F_INDIRECT_DESC
382 	    | VIRTIO_BLK_F_BLK_SIZE
383 	    | VIRTIO_BLK_F_SEG_MAX);
384 	WRITE4(sc, VIRTIO_MMIO_HOST_FEATURES, reg);
385 
386 	cfg = sc->cfg;
387 	cfg->capacity = htobe64(sc->mdio->md_mediasize / DEV_BSIZE);
388 	cfg->size_max = 0; /* not negotiated */
389 	cfg->seg_max = htobe32(VTBLK_MAXSEGS);
390 	cfg->blk_size = htobe32(DEV_BSIZE);
391 
392 	s = (uint32_t *)cfg;
393 
394 	for (i = 0; i < sizeof(struct virtio_blk_config); i+=4) {
395 		WRITE4(sc, VIRTIO_MMIO_CONFIG + i, *s);
396 		s+=1;
397 	}
398 
399 	strncpy(sc->ident, "Virtio block backend", sizeof(sc->ident));
400 
401 	return (0);
402 }
403 
404 static void
vtblk_intr(void * arg)405 vtblk_intr(void *arg)
406 {
407 	struct beri_vtblk_softc *sc;
408 	int pending;
409 	int reg;
410 
411 	sc = arg;
412 
413 	reg = PIO_READ(sc->pio_recv);
414 
415 	/* Ack */
416 	PIO_SET(sc->pio_recv, reg, 0);
417 
418 	pending = htobe32(reg);
419 
420 	if (pending & Q_PFN) {
421 		vq_init(sc);
422 	}
423 
424 	if (pending & Q_NOTIFY) {
425 		wakeup(sc);
426 	}
427 }
428 
429 static int
beri_ioctl(struct cdev * dev,u_long cmd,caddr_t addr,int flags,struct thread * td)430 beri_ioctl(struct cdev *dev, u_long cmd, caddr_t addr,
431 		int flags, struct thread *td)
432 {
433 	struct beri_vtblk_softc *sc;
434 	int err;
435 
436 	sc = dev->si_drv1;
437 
438 	switch (cmd) {
439 	case MDIOCATTACH:
440 		/* take file as argument */
441 		if (sc->vnode != NULL) {
442 			/* Already opened */
443 			return (1);
444 		}
445 		sc->mdio = (struct md_ioctl *)addr;
446 		backend_info(sc);
447 		DPRINTF("opening file, td 0x%08x\n", (int)td);
448 		err = open_file(sc, td);
449 		if (err)
450 			return (err);
451 		PIO_SETUP_IRQ(sc->pio_recv, vtblk_intr, sc);
452 		sc->opened = 1;
453 		break;
454 	case MDIOCDETACH:
455 		if (sc->vnode == NULL) {
456 			/* File not opened */
457 			return (1);
458 		}
459 		sc->opened = 0;
460 		DPRINTF("closing file, td 0x%08x\n", (int)td);
461 		err = close_file(sc, td);
462 		if (err)
463 			return (err);
464 		PIO_TEARDOWN_IRQ(sc->pio_recv);
465 		break;
466 	default:
467 		break;
468 	}
469 
470 	return (0);
471 }
472 
473 static struct cdevsw beri_cdevsw = {
474 	.d_version =	D_VERSION,
475 	.d_ioctl =	beri_ioctl,
476 	.d_name =	"virtio block backend",
477 };
478 
479 static int
beri_vtblk_probe(device_t dev)480 beri_vtblk_probe(device_t dev)
481 {
482 
483 	if (!ofw_bus_status_okay(dev))
484 		return (ENXIO);
485 
486 	if (!ofw_bus_is_compatible(dev, "sri-cambridge,beri-vtblk"))
487 		return (ENXIO);
488 
489 	device_set_desc(dev, "SRI-Cambridge BERI block");
490 	return (BUS_PROBE_DEFAULT);
491 }
492 
493 static int
beri_vtblk_attach(device_t dev)494 beri_vtblk_attach(device_t dev)
495 {
496 	struct beri_vtblk_softc *sc;
497 	int error;
498 
499 	sc = device_get_softc(dev);
500 	sc->dev = dev;
501 
502 	if (bus_alloc_resources(dev, beri_spec, sc->res)) {
503 		device_printf(dev, "could not allocate resources\n");
504 		return (ENXIO);
505 	}
506 
507 	/* Memory interface */
508 	sc->bst = rman_get_bustag(sc->res[0]);
509 	sc->bsh = rman_get_bushandle(sc->res[0]);
510 
511 	sc->cfg = malloc(sizeof(struct virtio_blk_config),
512 		M_DEVBUF, M_NOWAIT|M_ZERO);
513 
514 	sx_init(&sc->sc_mtx, device_get_nameunit(sc->dev));
515 
516 	error = kthread_add(vtblk_thread, sc, NULL, &sc->vtblk_ktd,
517 		0, 0, "beri_virtio_block");
518 	if (error) {
519 		device_printf(dev, "cannot create kthread\n");
520 		return (ENXIO);
521 	}
522 
523 	if (setup_offset(dev, &sc->beri_mem_offset) != 0)
524 		return (ENXIO);
525 	if (setup_pio(dev, "pio-send", &sc->pio_send) != 0)
526 		return (ENXIO);
527 	if (setup_pio(dev, "pio-recv", &sc->pio_recv) != 0)
528 		return (ENXIO);
529 
530 	sc->cdev = make_dev(&beri_cdevsw, 0, UID_ROOT, GID_WHEEL,
531 	    S_IRWXU, "beri_vtblk");
532 	if (sc->cdev == NULL) {
533 		device_printf(dev, "Failed to create character device.\n");
534 		return (ENXIO);
535 	}
536 
537 	sc->cdev->si_drv1 = sc;
538 	return (0);
539 }
540 
541 static device_method_t beri_vtblk_methods[] = {
542 	DEVMETHOD(device_probe,		beri_vtblk_probe),
543 	DEVMETHOD(device_attach,	beri_vtblk_attach),
544 	{ 0, 0 }
545 };
546 
547 static driver_t beri_vtblk_driver = {
548 	"beri_vtblk",
549 	beri_vtblk_methods,
550 	sizeof(struct beri_vtblk_softc),
551 };
552 
553 DRIVER_MODULE(beri_vtblk, simplebus, beri_vtblk_driver, 0, 0);
554