viomb.c revision 1.4
1 /* $NetBSD: viomb.c,v 1.4 2015/05/04 14:02:13 ozaki-r 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 1.4 2015/05/04 14:02:13 ozaki-r 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 42 #include <dev/pci/pcidevs.h> 43 #include <dev/pci/pcireg.h> 44 #include <dev/pci/pcivar.h> 45 46 #include <dev/pci/virtioreg.h> 47 #include <dev/pci/virtiovar.h> 48 49 /* Configuration registers */ 50 #define VIRTIO_BALLOON_CONFIG_NUM_PAGES 0 /* 32bit */ 51 #define VIRTIO_BALLOON_CONFIG_ACTUAL 4 /* 32bit */ 52 53 /* Feature bits */ 54 #define VIRTIO_BALLOON_F_MUST_TELL_HOST (1<<0) 55 #define VIRTIO_BALLOON_F_STATS_VQ (1<<1) 56 57 #define PGS_PER_REQ (256) /* 1MB, 4KB/page */ 58 59 CTASSERT((PAGE_SIZE) == (VIRTIO_PAGE_SIZE)); /* XXX */ 60 61 struct balloon_req { 62 bus_dmamap_t bl_dmamap; 63 struct pglist bl_pglist; 64 int bl_nentries; 65 uint32_t bl_pages[PGS_PER_REQ]; 66 }; 67 68 struct viomb_softc { 69 device_t sc_dev; 70 71 struct virtio_softc *sc_virtio; 72 struct virtqueue sc_vq[2]; 73 74 unsigned int sc_npages; 75 unsigned int sc_actual; 76 int sc_inflight; 77 struct balloon_req sc_req; 78 struct pglist sc_balloon_pages; 79 80 int sc_inflate_done; 81 int sc_deflate_done; 82 83 kcondvar_t sc_wait; 84 kmutex_t sc_waitlock; 85 }; 86 87 static int balloon_initialized = 0; /* multiple balloon is not allowed */ 88 89 static int viomb_match(device_t, cfdata_t, void *); 90 static void viomb_attach(device_t, device_t, void *); 91 static void viomb_read_config(struct viomb_softc *); 92 static int viomb_config_change(struct virtio_softc *); 93 static int inflate(struct viomb_softc *); 94 static int inflateq_done(struct virtqueue *); 95 static int inflate_done(struct viomb_softc *); 96 static int deflate(struct viomb_softc *); 97 static int deflateq_done(struct virtqueue *); 98 static int deflate_done(struct viomb_softc *); 99 static void viomb_thread(void *); 100 101 CFATTACH_DECL_NEW(viomb, sizeof(struct viomb_softc), 102 viomb_match, viomb_attach, NULL, NULL); 103 104 static int 105 viomb_match(device_t parent, cfdata_t match, void *aux) 106 { 107 struct virtio_softc *vsc = aux; 108 109 if (vsc->sc_childdevid == PCI_PRODUCT_VIRTIO_BALLOON) 110 return 1; 111 112 return 0; 113 } 114 115 static void 116 viomb_attach(device_t parent, device_t self, void *aux) 117 { 118 struct viomb_softc *sc = device_private(self); 119 struct virtio_softc *vsc = device_private(parent); 120 const struct sysctlnode *node; 121 122 if (vsc->sc_child != NULL) { 123 aprint_normal(": child already attached for %s; " 124 "something wrong...\n", 125 device_xname(parent)); 126 return; 127 } 128 if (balloon_initialized++) { 129 aprint_normal(": balloon already exists; something wrong...\n"); 130 goto err_none; 131 } 132 aprint_normal("\n"); 133 134 sc->sc_dev = self; 135 sc->sc_virtio = vsc; 136 137 vsc->sc_child = self; 138 vsc->sc_ipl = IPL_VM; 139 vsc->sc_vqs = &sc->sc_vq[0]; 140 vsc->sc_nvqs = 2; 141 vsc->sc_config_change = viomb_config_change; 142 vsc->sc_intrhand = virtio_vq_intr; 143 vsc->sc_flags = 0; 144 145 virtio_negotiate_features(vsc, 146 VIRTIO_CONFIG_DEVICE_FEATURES); 147 if ((virtio_alloc_vq(vsc, &sc->sc_vq[0], 0, 148 sizeof(uint32_t)*PGS_PER_REQ, 1, 149 "inflate") != 0) || 150 (virtio_alloc_vq(vsc, &sc->sc_vq[1], 1, 151 sizeof(uint32_t)*PGS_PER_REQ, 1, 152 "deflate") != 0)) { 153 goto err_none; 154 } 155 sc->sc_vq[0].vq_done = inflateq_done; 156 sc->sc_vq[1].vq_done = deflateq_done; 157 158 viomb_read_config(sc); 159 sc->sc_inflight = 0; 160 TAILQ_INIT(&sc->sc_balloon_pages); 161 162 if (bus_dmamap_create(vsc->sc_dmat, sizeof(uint32_t)*PGS_PER_REQ, 163 1, sizeof(uint32_t)*PGS_PER_REQ, 0, 164 BUS_DMA_NOWAIT, &sc->sc_req.bl_dmamap)) { 165 aprint_error_dev(sc->sc_dev, "dmamap creation failed.\n"); 166 goto err_vq; 167 } 168 if (bus_dmamap_load(vsc->sc_dmat, sc->sc_req.bl_dmamap, 169 &sc->sc_req.bl_pages[0], 170 sizeof(uint32_t) * PGS_PER_REQ, 171 NULL, BUS_DMA_NOWAIT)) { 172 aprint_error_dev(sc->sc_dev, "dmamap load failed.\n"); 173 goto err_dmamap; 174 } 175 176 sc->sc_inflate_done = sc->sc_deflate_done = 0; 177 mutex_init(&sc->sc_waitlock, MUTEX_DEFAULT, IPL_VM); /* spin */ 178 cv_init(&sc->sc_wait, "balloon"); 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(vsc->sc_dmat, 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 vsc->sc_child = (void*)1; 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(vsc->sc_child); 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(vsc->sc_dmat, b->bl_dmamap, 0, sizeof(uint32_t)*nvpages, 301 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(vsc->sc_child); 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(vsc->sc_dmat, 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, pageq.queue); 399 } 400 return 0; 401 } 402 if (virtio_enqueue_reserve(vsc, vq, slot, 1) != 0) { 403 printf("%s: deflate enqueue failed.\n", 404 device_xname(sc->sc_dev)); 405 TAILQ_FOREACH_REVERSE(p, &b->bl_pglist, pglist, pageq.queue) { 406 TAILQ_REMOVE(&b->bl_pglist, p, pageq.queue); 407 TAILQ_INSERT_HEAD(&sc->sc_balloon_pages, p, pageq.queue); 408 } 409 return 0; 410 } 411 bus_dmamap_sync(vsc->sc_dmat, b->bl_dmamap, 0, sizeof(uint32_t)*nvpages, 412 BUS_DMASYNC_PREWRITE); 413 virtio_enqueue(vsc, vq, slot, b->bl_dmamap, true); 414 virtio_enqueue_commit(vsc, vq, slot, true); 415 sc->sc_inflight -= nvpages; 416 417 if (!(vsc->sc_features & VIRTIO_BALLOON_F_MUST_TELL_HOST)) 418 uvm_pglistfree(&b->bl_pglist); 419 420 return 0; 421 } 422 423 static int 424 deflateq_done(struct virtqueue *vq) 425 { 426 struct virtio_softc *vsc = vq->vq_owner; 427 struct viomb_softc *sc = device_private(vsc->sc_child); 428 429 mutex_enter(&sc->sc_waitlock); 430 sc->sc_deflate_done = 1; 431 cv_signal(&sc->sc_wait); 432 mutex_exit(&sc->sc_waitlock); 433 434 return 1; 435 } 436 437 static int 438 deflate_done(struct viomb_softc *sc) 439 { 440 struct virtio_softc *vsc = sc->sc_virtio; 441 struct virtqueue *vq = &sc->sc_vq[1]; 442 struct balloon_req *b; 443 int r, slot; 444 uint64_t nvpages; 445 446 r = virtio_dequeue(vsc, vq, &slot, NULL); 447 if (r != 0) { 448 printf("%s: deflate dequeue failed, errno %d\n", 449 device_xname(sc->sc_dev), r); 450 return 1; 451 } 452 virtio_dequeue_commit(vsc, vq, slot); 453 454 b = &sc->sc_req; 455 nvpages = b->bl_nentries; 456 bus_dmamap_sync(vsc->sc_dmat, b->bl_dmamap, 457 offsetof(struct balloon_req, bl_pages), 458 sizeof(uint32_t)*nvpages, 459 BUS_DMASYNC_POSTWRITE); 460 461 if (vsc->sc_features & VIRTIO_BALLOON_F_MUST_TELL_HOST) 462 uvm_pglistfree(&b->bl_pglist); 463 464 sc->sc_inflight += nvpages; 465 virtio_write_device_config_4(vsc, 466 VIRTIO_BALLOON_CONFIG_ACTUAL, 467 sc->sc_actual - nvpages); 468 viomb_read_config(sc); 469 470 return 1; 471 } 472 473 /* 474 * Kthread: sleeps, eventually inflate and deflate. 475 */ 476 static void 477 viomb_thread(void *arg) 478 { 479 struct viomb_softc *sc = arg; 480 int sleeptime, r; 481 482 for ( ; ; ) { 483 sleeptime = 30000; 484 if (sc->sc_npages > sc->sc_actual + sc->sc_inflight) { 485 if (sc->sc_inflight == 0) { 486 r = inflate(sc); 487 if (r != 0) 488 sleeptime = 10000; 489 else 490 sleeptime = 1000; 491 } else 492 sleeptime = 100; 493 } else if (sc->sc_npages < sc->sc_actual + sc->sc_inflight) { 494 if (sc->sc_inflight == 0) 495 r = deflate(sc); 496 sleeptime = 100; 497 } 498 499 again: 500 mutex_enter(&sc->sc_waitlock); 501 if (sc->sc_inflate_done) { 502 sc->sc_inflate_done = 0; 503 mutex_exit(&sc->sc_waitlock); 504 inflate_done(sc); 505 goto again; 506 } 507 if (sc->sc_deflate_done) { 508 sc->sc_deflate_done = 0; 509 mutex_exit(&sc->sc_waitlock); 510 deflate_done(sc); 511 goto again; 512 } 513 cv_timedwait(&sc->sc_wait, &sc->sc_waitlock, 514 mstohz(sleeptime)); 515 mutex_exit(&sc->sc_waitlock); 516 } 517 } 518
Indexes created Wed Sep 24 15:09:51 GMT 2025