xref: /netbsd-src/sys/dev/pci/viomb.c (revision bdc22b2e01993381dcefeff2bc9b56ca75a4235c)
1 /*	$NetBSD: viomb.c,v 1.9 2018/06/10 14:59:23 jakllsch Exp $	*/
2 
3 /*
4  * Copyright (c) 2010 Minoura Makoto.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
17  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
18  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
19  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
20  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
21  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
22  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
23  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
24  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
25  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
26  */
27 
28 #include <sys/cdefs.h>
29 __KERNEL_RCSID(0, "$NetBSD: viomb.c,v 1.9 2018/06/10 14:59:23 jakllsch Exp $");
30 
31 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/kernel.h>
34 #include <sys/bus.h>
35 #include <sys/condvar.h>
36 #include <sys/device.h>
37 #include <sys/kthread.h>
38 #include <sys/mutex.h>
39 #include <sys/sysctl.h>
40 #include <uvm/uvm_page.h>
41 #include <sys/module.h>
42 
43 #include <dev/pci/virtioreg.h>
44 #include <dev/pci/virtiovar.h>
45 
46 #include "ioconf.h"
47 
48 /* Configuration registers */
49 #define VIRTIO_BALLOON_CONFIG_NUM_PAGES	0 /* 32bit */
50 #define VIRTIO_BALLOON_CONFIG_ACTUAL	4 /* 32bit */
51 
52 /* Feature bits */
53 #define VIRTIO_BALLOON_F_MUST_TELL_HOST (1<<0)
54 #define VIRTIO_BALLOON_F_STATS_VQ	(1<<1)
55 
56 #define VIRTIO_BALLOON_FLAG_BITS \
57 	VIRTIO_COMMON_FLAG_BITS \
58 	"\x02""STATS_VQ" \
59 	"\x01""MUST_TELL_HOST"
60 
61 #define PGS_PER_REQ		(256) /* 1MB, 4KB/page */
62 
63 CTASSERT((PAGE_SIZE) == (VIRTIO_PAGE_SIZE)); /* XXX */
64 
65 struct balloon_req {
66 	bus_dmamap_t			bl_dmamap;
67 	struct pglist			bl_pglist;
68 	int				bl_nentries;
69 	uint32_t			bl_pages[PGS_PER_REQ];
70 };
71 
72 struct viomb_softc {
73 	device_t		sc_dev;
74 
75 	struct virtio_softc	*sc_virtio;
76 	struct virtqueue	sc_vq[2];
77 
78 	unsigned int		sc_npages;
79 	unsigned int		sc_actual;
80 	int			sc_inflight;
81 	struct balloon_req	sc_req;
82 	struct pglist		sc_balloon_pages;
83 
84 	int			sc_inflate_done;
85 	int			sc_deflate_done;
86 
87 	kcondvar_t		sc_wait;
88 	kmutex_t		sc_waitlock;
89 };
90 
91 static int	balloon_initialized = 0; /* multiple balloon is not allowed */
92 
93 static int	viomb_match(device_t, cfdata_t, void *);
94 static void	viomb_attach(device_t, device_t, void *);
95 static void	viomb_read_config(struct viomb_softc *);
96 static int	viomb_config_change(struct virtio_softc *);
97 static int	inflate(struct viomb_softc *);
98 static int	inflateq_done(struct virtqueue *);
99 static int	inflate_done(struct viomb_softc *);
100 static int	deflate(struct viomb_softc *);
101 static int	deflateq_done(struct virtqueue *);
102 static int	deflate_done(struct viomb_softc *);
103 static void	viomb_thread(void *);
104 
105 CFATTACH_DECL_NEW(viomb, sizeof(struct viomb_softc),
106     viomb_match, viomb_attach, NULL, NULL);
107 
108 static int
109 viomb_match(device_t parent, cfdata_t match, void *aux)
110 {
111 	struct virtio_attach_args *va = aux;
112 
113 	if (va->sc_childdevid == PCI_PRODUCT_VIRTIO_BALLOON)
114 		return 1;
115 
116 	return 0;
117 }
118 
119 static void
120 viomb_attach(device_t parent, device_t self, void *aux)
121 {
122 	struct viomb_softc *sc = device_private(self);
123 	struct virtio_softc *vsc = device_private(parent);
124 	const struct sysctlnode *node;
125 
126 	if (virtio_child(vsc) != NULL) {
127 		aprint_normal(": child already attached for %s; "
128 			      "something wrong...\n", device_xname(parent));
129 		return;
130 	}
131 
132 	if (balloon_initialized++) {
133 		aprint_normal(": balloon already exists; something wrong...\n");
134 		goto err_none;
135 	}
136 
137 	sc->sc_dev = self;
138 	sc->sc_virtio = vsc;
139 
140 	if ((virtio_alloc_vq(vsc, &sc->sc_vq[0], 0,
141 			     sizeof(uint32_t)*PGS_PER_REQ, 1,
142 			     "inflate") != 0) ||
143 	    (virtio_alloc_vq(vsc, &sc->sc_vq[1], 1,
144 			     sizeof(uint32_t)*PGS_PER_REQ, 1,
145 			     "deflate") != 0)) {
146 		goto err_none;
147 	}
148 	sc->sc_vq[0].vq_done = inflateq_done;
149 	sc->sc_vq[1].vq_done = deflateq_done;
150 
151 	viomb_read_config(sc);
152 	sc->sc_inflight = 0;
153 	TAILQ_INIT(&sc->sc_balloon_pages);
154 
155 	if (bus_dmamap_create(virtio_dmat(vsc), sizeof(uint32_t)*PGS_PER_REQ,
156 			      1, sizeof(uint32_t)*PGS_PER_REQ, 0,
157 			      BUS_DMA_NOWAIT, &sc->sc_req.bl_dmamap)) {
158 		aprint_error_dev(sc->sc_dev, "dmamap creation failed.\n");
159 		goto err_vq;
160 	}
161 	if (bus_dmamap_load(virtio_dmat(vsc), sc->sc_req.bl_dmamap,
162 			    &sc->sc_req.bl_pages[0],
163 			    sizeof(uint32_t) * PGS_PER_REQ,
164 			    NULL, BUS_DMA_NOWAIT)) {
165 		aprint_error_dev(sc->sc_dev, "dmamap load failed.\n");
166 		goto err_dmamap;
167 	}
168 
169 	sc->sc_inflate_done = sc->sc_deflate_done = 0;
170 	mutex_init(&sc->sc_waitlock, MUTEX_DEFAULT, IPL_VM); /* spin */
171 	cv_init(&sc->sc_wait, "balloon");
172 
173 	virtio_child_attach_start(vsc, self, IPL_VM, sc->sc_vq,
174 	    viomb_config_change, virtio_vq_intr, 0,
175 	    0, VIRTIO_BALLOON_FLAG_BITS);
176 
177 	if (virtio_child_attach_finish(vsc) != 0)
178 		goto err_mutex;
179 
180 	if (kthread_create(PRI_IDLE, KTHREAD_MPSAFE, NULL,
181 			   viomb_thread, sc, NULL, "viomb")) {
182 		aprint_error_dev(sc->sc_dev, "cannot create kthread.\n");
183 		goto err_mutex;
184 	}
185 
186 	sysctl_createv(NULL, 0, NULL, &node, 0, CTLTYPE_NODE,
187 		       "viomb", SYSCTL_DESCR("VirtIO Balloon status"),
188 		       NULL, 0, NULL, 0,
189 		       CTL_HW, CTL_CREATE, CTL_EOL);
190 	sysctl_createv(NULL, 0, NULL, NULL, 0, CTLTYPE_INT,
191 		       "npages", SYSCTL_DESCR("VirtIO Balloon npages value"),
192 		       NULL, 0, &sc->sc_npages, 0,
193 		       CTL_HW, node->sysctl_num, CTL_CREATE, CTL_EOL);
194 	sysctl_createv(NULL, 0, NULL, NULL, 0, CTLTYPE_INT,
195 		       "actual", SYSCTL_DESCR("VirtIO Balloon actual value"),
196 		       NULL, 0, &sc->sc_actual, 0,
197 		       CTL_HW, node->sysctl_num, CTL_CREATE, CTL_EOL);
198 	return;
199 
200 err_mutex:
201 	cv_destroy(&sc->sc_wait);
202 	mutex_destroy(&sc->sc_waitlock);
203 err_dmamap:
204 	bus_dmamap_destroy(virtio_dmat(vsc), sc->sc_req.bl_dmamap);
205 err_vq:
206 	virtio_free_vq(vsc, &sc->sc_vq[1]);
207 	virtio_free_vq(vsc, &sc->sc_vq[0]);
208 err_none:
209 	virtio_child_attach_failed(vsc);
210 	return;
211 }
212 
213 static void
214 viomb_read_config(struct viomb_softc *sc)
215 {
216 	unsigned int reg;
217 
218 	/* these values are explicitly specified as little-endian */
219 	reg = virtio_read_device_config_4(sc->sc_virtio,
220 					  VIRTIO_BALLOON_CONFIG_NUM_PAGES);
221 	sc->sc_npages = le32toh(reg);
222 
223 	reg = virtio_read_device_config_4(sc->sc_virtio,
224 					  VIRTIO_BALLOON_CONFIG_ACTUAL);
225 	sc->sc_actual = le32toh(reg);
226 }
227 
228 /*
229  * Config change callback: wakeup the kthread.
230  */
231 static int
232 viomb_config_change(struct virtio_softc *vsc)
233 {
234 	struct viomb_softc *sc = device_private(virtio_child(vsc));
235 	unsigned int old;
236 
237 	old = sc->sc_npages;
238 	viomb_read_config(sc);
239 	mutex_enter(&sc->sc_waitlock);
240 	cv_signal(&sc->sc_wait);
241 	mutex_exit(&sc->sc_waitlock);
242 	if (sc->sc_npages > old)
243 		printf("%s: inflating balloon from %u to %u.\n",
244 		       device_xname(sc->sc_dev), old, sc->sc_npages);
245 	else if  (sc->sc_npages < old)
246 		printf("%s: deflating balloon from %u to %u.\n",
247 		       device_xname(sc->sc_dev), old, sc->sc_npages);
248 
249 	return 1;
250 }
251 
252 /*
253  * Inflate: consume some amount of physical memory.
254  */
255 static int
256 inflate(struct viomb_softc *sc)
257 {
258 	struct virtio_softc *vsc = sc->sc_virtio;
259 	int i, slot;
260 	uint64_t nvpages, nhpages;
261 	struct balloon_req *b;
262 	struct vm_page *p;
263 	struct virtqueue *vq = &sc->sc_vq[0];
264 
265 	if (sc->sc_inflight)
266 		return 0;
267 	nvpages = sc->sc_npages - sc->sc_actual;
268 	if (nvpages > PGS_PER_REQ)
269 		nvpages = PGS_PER_REQ;
270 	nhpages = nvpages * VIRTIO_PAGE_SIZE / PAGE_SIZE;
271 
272 	b = &sc->sc_req;
273 	if (uvm_pglistalloc(nhpages*PAGE_SIZE, 0, UINT32_MAX*PAGE_SIZE,
274 			    0, 0, &b->bl_pglist, nhpages, 1)) {
275 		printf("%s: %" PRIu64 " pages of physical memory "
276 		       "could not be allocated, retrying...\n",
277 		       device_xname(sc->sc_dev), nhpages);
278 		return 1;	/* sleep longer */
279 	}
280 
281 	b->bl_nentries = nvpages;
282 	i = 0;
283 	TAILQ_FOREACH(p, &b->bl_pglist, pageq.queue) {
284 		b->bl_pages[i++] = p->phys_addr / VIRTIO_PAGE_SIZE;
285 	}
286 	KASSERT(i == nvpages);
287 
288 	if (virtio_enqueue_prep(vsc, vq, &slot) != 0) {
289 		printf("%s: inflate enqueue failed.\n",
290 		       device_xname(sc->sc_dev));
291 		uvm_pglistfree(&b->bl_pglist);
292 		return 0;
293 	}
294 	if (virtio_enqueue_reserve(vsc, vq, slot, 1)) {
295 		printf("%s: inflate enqueue failed.\n",
296 		       device_xname(sc->sc_dev));
297 		uvm_pglistfree(&b->bl_pglist);
298 		return 0;
299 	}
300 	bus_dmamap_sync(virtio_dmat(vsc), b->bl_dmamap, 0,
301 	    sizeof(uint32_t)*nvpages, BUS_DMASYNC_PREWRITE);
302 	virtio_enqueue(vsc, vq, slot, b->bl_dmamap, true);
303 	virtio_enqueue_commit(vsc, vq, slot, true);
304 	sc->sc_inflight += nvpages;
305 
306 	return 0;
307 }
308 
309 static int
310 inflateq_done(struct virtqueue *vq)
311 {
312 	struct virtio_softc *vsc = vq->vq_owner;
313 	struct viomb_softc *sc = device_private(virtio_child(vsc));
314 
315 	mutex_enter(&sc->sc_waitlock);
316 	sc->sc_inflate_done = 1;
317 	cv_signal(&sc->sc_wait);
318 	mutex_exit(&sc->sc_waitlock);
319 
320 	return 1;
321 }
322 
323 static int
324 inflate_done(struct viomb_softc *sc)
325 {
326 	struct virtio_softc *vsc = sc->sc_virtio;
327 	struct virtqueue *vq = &sc->sc_vq[0];
328 	struct balloon_req *b;
329 	int r, slot;
330 	uint64_t nvpages;
331 	struct vm_page *p;
332 
333 	r = virtio_dequeue(vsc, vq, &slot, NULL);
334 	if (r != 0) {
335 		printf("%s: inflate dequeue failed, errno %d.\n",
336 		       device_xname(sc->sc_dev), r);
337 		return 1;
338 	}
339 	virtio_dequeue_commit(vsc, vq, slot);
340 
341 	b = &sc->sc_req;
342 	nvpages = b->bl_nentries;
343 	bus_dmamap_sync(virtio_dmat(vsc), b->bl_dmamap,
344 			offsetof(struct balloon_req, bl_pages),
345 			sizeof(uint32_t)*nvpages,
346 			BUS_DMASYNC_POSTWRITE);
347 	while (!TAILQ_EMPTY(&b->bl_pglist)) {
348 		p = TAILQ_FIRST(&b->bl_pglist);
349 		TAILQ_REMOVE(&b->bl_pglist, p, pageq.queue);
350 		TAILQ_INSERT_TAIL(&sc->sc_balloon_pages, p, pageq.queue);
351 	}
352 
353 	sc->sc_inflight -= nvpages;
354 	virtio_write_device_config_4(vsc,
355 				     VIRTIO_BALLOON_CONFIG_ACTUAL,
356 				     sc->sc_actual + nvpages);
357 	viomb_read_config(sc);
358 
359 	return 1;
360 }
361 
362 /*
363  * Deflate: free previously allocated memory.
364  */
365 static int
366 deflate(struct viomb_softc *sc)
367 {
368 	struct virtio_softc *vsc = sc->sc_virtio;
369 	int i, slot;
370 	uint64_t nvpages, nhpages;
371 	struct balloon_req *b;
372 	struct vm_page *p;
373 	struct virtqueue *vq = &sc->sc_vq[1];
374 
375 	nvpages = (sc->sc_actual + sc->sc_inflight) - sc->sc_npages;
376 	if (nvpages > PGS_PER_REQ)
377 		nvpages = PGS_PER_REQ;
378 	nhpages = nvpages * VIRTIO_PAGE_SIZE / PAGE_SIZE;
379 
380 	b = &sc->sc_req;
381 
382 	b->bl_nentries = nvpages;
383 	TAILQ_INIT(&b->bl_pglist);
384 	for (i = 0; i < nhpages; i++) {
385 		p = TAILQ_FIRST(&sc->sc_balloon_pages);
386 		if (p == NULL)
387 			break;
388 		TAILQ_REMOVE(&sc->sc_balloon_pages, p, pageq.queue);
389 		TAILQ_INSERT_TAIL(&b->bl_pglist, p, pageq.queue);
390 		b->bl_pages[i] = p->phys_addr / VIRTIO_PAGE_SIZE;
391 	}
392 
393 	if (virtio_enqueue_prep(vsc, vq, &slot) != 0) {
394 		printf("%s: deflate enqueue failed.\n",
395 		       device_xname(sc->sc_dev));
396 		TAILQ_FOREACH_REVERSE(p, &b->bl_pglist, pglist, pageq.queue) {
397 			TAILQ_REMOVE(&b->bl_pglist, p, pageq.queue);
398 			TAILQ_INSERT_HEAD(&sc->sc_balloon_pages, p,
399 			    pageq.queue);
400 		}
401 		return 0;
402 	}
403 	if (virtio_enqueue_reserve(vsc, vq, slot, 1) != 0) {
404 		printf("%s: deflate enqueue failed.\n",
405 		       device_xname(sc->sc_dev));
406 		TAILQ_FOREACH_REVERSE(p, &b->bl_pglist, pglist, pageq.queue) {
407 			TAILQ_REMOVE(&b->bl_pglist, p, pageq.queue);
408 			TAILQ_INSERT_HEAD(&sc->sc_balloon_pages, p,
409 			    pageq.queue);
410 		}
411 		return 0;
412 	}
413 	bus_dmamap_sync(virtio_dmat(vsc), b->bl_dmamap, 0,
414 	    sizeof(uint32_t)*nvpages, BUS_DMASYNC_PREWRITE);
415 	virtio_enqueue(vsc, vq, slot, b->bl_dmamap, true);
416 	virtio_enqueue_commit(vsc, vq, slot, true);
417 	sc->sc_inflight -= nvpages;
418 
419 	if (!(virtio_features(vsc) & VIRTIO_BALLOON_F_MUST_TELL_HOST))
420 		uvm_pglistfree(&b->bl_pglist);
421 
422 	return 0;
423 }
424 
425 static int
426 deflateq_done(struct virtqueue *vq)
427 {
428 	struct virtio_softc *vsc = vq->vq_owner;
429 	struct viomb_softc *sc = device_private(virtio_child(vsc));
430 
431 	mutex_enter(&sc->sc_waitlock);
432 	sc->sc_deflate_done = 1;
433 	cv_signal(&sc->sc_wait);
434 	mutex_exit(&sc->sc_waitlock);
435 
436 	return 1;
437 }
438 
439 static int
440 deflate_done(struct viomb_softc *sc)
441 {
442 	struct virtio_softc *vsc = sc->sc_virtio;
443 	struct virtqueue *vq = &sc->sc_vq[1];
444 	struct balloon_req *b;
445 	int r, slot;
446 	uint64_t nvpages;
447 
448 	r = virtio_dequeue(vsc, vq, &slot, NULL);
449 	if (r != 0) {
450 		printf("%s: deflate dequeue failed, errno %d\n",
451 		       device_xname(sc->sc_dev), r);
452 		return 1;
453 	}
454 	virtio_dequeue_commit(vsc, vq, slot);
455 
456 	b = &sc->sc_req;
457 	nvpages = b->bl_nentries;
458 	bus_dmamap_sync(virtio_dmat(vsc), b->bl_dmamap,
459 			offsetof(struct balloon_req, bl_pages),
460 			sizeof(uint32_t)*nvpages,
461 			BUS_DMASYNC_POSTWRITE);
462 
463 	if (virtio_features(vsc) & VIRTIO_BALLOON_F_MUST_TELL_HOST)
464 		uvm_pglistfree(&b->bl_pglist);
465 
466 	sc->sc_inflight += nvpages;
467 	virtio_write_device_config_4(vsc,
468 				     VIRTIO_BALLOON_CONFIG_ACTUAL,
469 				     sc->sc_actual - nvpages);
470 	viomb_read_config(sc);
471 
472 	return 1;
473 }
474 
475 /*
476  * Kthread: sleeps, eventually inflate and deflate.
477  */
478 static void
479 viomb_thread(void *arg)
480 {
481 	struct viomb_softc *sc = arg;
482 	int sleeptime, r;
483 
484 	for ( ; ; ) {
485 		sleeptime = 30000;
486 		if (sc->sc_npages > sc->sc_actual + sc->sc_inflight) {
487 			if (sc->sc_inflight == 0) {
488 				r = inflate(sc);
489 				if (r != 0)
490 					sleeptime = 10000;
491 				else
492 					sleeptime = 1000;
493 			} else
494 				sleeptime = 100;
495 		} else if (sc->sc_npages < sc->sc_actual + sc->sc_inflight) {
496 			if (sc->sc_inflight == 0)
497 				r = deflate(sc);
498 			sleeptime = 100;
499 		}
500 
501 	again:
502 		mutex_enter(&sc->sc_waitlock);
503 		if (sc->sc_inflate_done) {
504 			sc->sc_inflate_done = 0;
505 			mutex_exit(&sc->sc_waitlock);
506 			inflate_done(sc);
507 			goto again;
508 		}
509 		if (sc->sc_deflate_done) {
510 			sc->sc_deflate_done = 0;
511 			mutex_exit(&sc->sc_waitlock);
512 			deflate_done(sc);
513 			goto again;
514 		}
515 		cv_timedwait(&sc->sc_wait, &sc->sc_waitlock,
516 			     mstohz(sleeptime));
517 		mutex_exit(&sc->sc_waitlock);
518 	}
519 }
520 
521 MODULE(MODULE_CLASS_DRIVER, viomb, "virtio");
522 
523 #ifdef _MODULE
524 #include "ioconf.c"
525 #endif
526 
527 static int
528 viomb_modcmd(modcmd_t cmd, void *opaque)
529 {
530 	int error = 0;
531 
532 #ifdef _MODULE
533 	switch (cmd) {
534 	case MODULE_CMD_INIT:
535 		error = config_init_component(cfdriver_ioconf_viomb,
536 		    cfattach_ioconf_viomb, cfdata_ioconf_viomb);
537 		break;
538 	case MODULE_CMD_FINI:
539 		error = config_fini_component(cfdriver_ioconf_viomb,
540 		    cfattach_ioconf_viomb, cfdata_ioconf_viomb);
541 		break;
542 	default:
543 		error = ENOTTY;
544 		break;
545 	}
546 #endif
547 
548 	return error;
549 }
550