1 /****************************************************************************** 2 * balloon.c 3 * 4 * Xen balloon driver - enables returning/claiming memory to/from Xen. 5 * 6 * Copyright (c) 2003, B Dragovic 7 * Copyright (c) 2003-2004, M Williamson, K Fraser 8 * Copyright (c) 2005 Dan M. Smith, IBM Corporation 9 * 10 * This file may be distributed separately from the Linux kernel, or 11 * incorporated into other software packages, subject to the following license: 12 * 13 * Permission is hereby granted, free of charge, to any person obtaining a copy 14 * of this source file (the "Software"), to deal in the Software without 15 * restriction, including without limitation the rights to use, copy, modify, 16 * merge, publish, distribute, sublicense, and/or sell copies of the Software, 17 * and to permit persons to whom the Software is furnished to do so, subject to 18 * the following conditions: 19 * 20 * The above copyright notice and this permission notice shall be included in 21 * all copies or substantial portions of the Software. 22 * 23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 24 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 25 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 26 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 27 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING 28 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS 29 * IN THE SOFTWARE. 30 */ 31 32 #include <sys/cdefs.h> 33 __FBSDID("$FreeBSD$"); 34 35 #include <sys/param.h> 36 #include <sys/lock.h> 37 #include <sys/kernel.h> 38 #include <sys/kthread.h> 39 #include <sys/malloc.h> 40 #include <sys/mutex.h> 41 #include <sys/sysctl.h> 42 43 #include <machine/xen/xen-os.h> 44 #include <machine/xen/xenvar.h> 45 #include <machine/xen/xenfunc.h> 46 #include <xen/hypervisor.h> 47 #include <xen/xenstore/xenstorevar.h> 48 49 #include <vm/vm.h> 50 #include <vm/vm_page.h> 51 52 static MALLOC_DEFINE(M_BALLOON, "Balloon", "Xen Balloon Driver"); 53 54 struct mtx balloon_mutex; 55 56 /* 57 * Protects atomic reservation decrease/increase against concurrent increases. 58 * Also protects non-atomic updates of current_pages and driver_pages, and 59 * balloon lists. 60 */ 61 struct mtx balloon_lock; 62 63 /* We increase/decrease in batches which fit in a page */ 64 static unsigned long frame_list[PAGE_SIZE / sizeof(unsigned long)]; 65 #define ARRAY_SIZE(A) (sizeof(A) / sizeof(A[0])) 66 67 struct balloon_stats { 68 /* We aim for 'current allocation' == 'target allocation'. */ 69 unsigned long current_pages; 70 unsigned long target_pages; 71 /* We may hit the hard limit in Xen. If we do then we remember it. */ 72 unsigned long hard_limit; 73 /* 74 * Drivers may alter the memory reservation independently, but they 75 * must inform the balloon driver so we avoid hitting the hard limit. 76 */ 77 unsigned long driver_pages; 78 /* Number of pages in high- and low-memory balloons. */ 79 unsigned long balloon_low; 80 unsigned long balloon_high; 81 }; 82 83 static struct balloon_stats balloon_stats; 84 #define bs balloon_stats 85 86 SYSCTL_DECL(_dev_xen); 87 static SYSCTL_NODE(_dev_xen, OID_AUTO, balloon, CTLFLAG_RD, NULL, "Balloon"); 88 SYSCTL_ULONG(_dev_xen_balloon, OID_AUTO, current, CTLFLAG_RD, 89 &bs.current_pages, 0, "Current allocation"); 90 SYSCTL_ULONG(_dev_xen_balloon, OID_AUTO, target, CTLFLAG_RD, 91 &bs.target_pages, 0, "Target allocation"); 92 SYSCTL_ULONG(_dev_xen_balloon, OID_AUTO, driver_pages, CTLFLAG_RD, 93 &bs.driver_pages, 0, "Driver pages"); 94 SYSCTL_ULONG(_dev_xen_balloon, OID_AUTO, hard_limit, CTLFLAG_RD, 95 &bs.hard_limit, 0, "Xen hard limit"); 96 SYSCTL_ULONG(_dev_xen_balloon, OID_AUTO, low_mem, CTLFLAG_RD, 97 &bs.balloon_low, 0, "Low-mem balloon"); 98 SYSCTL_ULONG(_dev_xen_balloon, OID_AUTO, high_mem, CTLFLAG_RD, 99 &bs.balloon_high, 0, "High-mem balloon"); 100 101 struct balloon_entry { 102 vm_page_t page; 103 STAILQ_ENTRY(balloon_entry) list; 104 }; 105 106 /* List of ballooned pages, threaded through the mem_map array. */ 107 static STAILQ_HEAD(,balloon_entry) ballooned_pages; 108 109 /* Main work function, always executed in process context. */ 110 static void balloon_process(void *unused); 111 112 #define IPRINTK(fmt, args...) \ 113 printk(KERN_INFO "xen_mem: " fmt, ##args) 114 #define WPRINTK(fmt, args...) \ 115 printk(KERN_WARNING "xen_mem: " fmt, ##args) 116 117 /* balloon_append: add the given page to the balloon. */ 118 static void 119 balloon_append(vm_page_t page) 120 { 121 struct balloon_entry *entry; 122 123 entry = malloc(sizeof(struct balloon_entry), M_BALLOON, M_WAITOK); 124 entry->page = page; 125 STAILQ_INSERT_HEAD(&ballooned_pages, entry, list); 126 bs.balloon_low++; 127 } 128 129 /* balloon_retrieve: rescue a page from the balloon, if it is not empty. */ 130 static vm_page_t 131 balloon_retrieve(void) 132 { 133 vm_page_t page; 134 struct balloon_entry *entry; 135 136 if (STAILQ_EMPTY(&ballooned_pages)) 137 return NULL; 138 139 entry = STAILQ_FIRST(&ballooned_pages); 140 STAILQ_REMOVE_HEAD(&ballooned_pages, list); 141 142 page = entry->page; 143 free(entry, M_DEVBUF); 144 145 bs.balloon_low--; 146 147 return page; 148 } 149 150 static unsigned long 151 current_target(void) 152 { 153 unsigned long target = min(bs.target_pages, bs.hard_limit); 154 if (target > (bs.current_pages + bs.balloon_low + bs.balloon_high)) 155 target = bs.current_pages + bs.balloon_low + bs.balloon_high; 156 return target; 157 } 158 159 static unsigned long 160 minimum_target(void) 161 { 162 #ifdef XENHVM 163 #define max_pfn physmem 164 #else 165 #define max_pfn HYPERVISOR_shared_info->arch.max_pfn 166 #endif 167 unsigned long min_pages, curr_pages = current_target(); 168 169 #define MB2PAGES(mb) ((mb) << (20 - PAGE_SHIFT)) 170 /* Simple continuous piecewiese linear function: 171 * max MiB -> min MiB gradient 172 * 0 0 173 * 16 16 174 * 32 24 175 * 128 72 (1/2) 176 * 512 168 (1/4) 177 * 2048 360 (1/8) 178 * 8192 552 (1/32) 179 * 32768 1320 180 * 131072 4392 181 */ 182 if (max_pfn < MB2PAGES(128)) 183 min_pages = MB2PAGES(8) + (max_pfn >> 1); 184 else if (max_pfn < MB2PAGES(512)) 185 min_pages = MB2PAGES(40) + (max_pfn >> 2); 186 else if (max_pfn < MB2PAGES(2048)) 187 min_pages = MB2PAGES(104) + (max_pfn >> 3); 188 else 189 min_pages = MB2PAGES(296) + (max_pfn >> 5); 190 #undef MB2PAGES 191 192 /* Don't enforce growth */ 193 return min(min_pages, curr_pages); 194 #ifndef CONFIG_XEN 195 #undef max_pfn 196 #endif 197 } 198 199 static int 200 increase_reservation(unsigned long nr_pages) 201 { 202 unsigned long pfn, i; 203 struct balloon_entry *entry; 204 vm_page_t page; 205 long rc; 206 struct xen_memory_reservation reservation = { 207 .address_bits = 0, 208 .extent_order = 0, 209 .domid = DOMID_SELF 210 }; 211 212 if (nr_pages > ARRAY_SIZE(frame_list)) 213 nr_pages = ARRAY_SIZE(frame_list); 214 215 mtx_lock(&balloon_lock); 216 217 for (entry = STAILQ_FIRST(&ballooned_pages), i = 0; 218 i < nr_pages; i++, entry = STAILQ_NEXT(entry, list)) { 219 KASSERT(entry, ("ballooned_pages list corrupt")); 220 page = entry->page; 221 frame_list[i] = (VM_PAGE_TO_PHYS(page) >> PAGE_SHIFT); 222 } 223 224 set_xen_guest_handle(reservation.extent_start, frame_list); 225 reservation.nr_extents = nr_pages; 226 rc = HYPERVISOR_memory_op( 227 XENMEM_populate_physmap, &reservation); 228 if (rc < nr_pages) { 229 if (rc > 0) { 230 int ret; 231 232 /* We hit the Xen hard limit: reprobe. */ 233 reservation.nr_extents = rc; 234 ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation, 235 &reservation); 236 KASSERT(ret == rc, ("HYPERVISOR_memory_op failed")); 237 } 238 if (rc >= 0) 239 bs.hard_limit = (bs.current_pages + rc - 240 bs.driver_pages); 241 goto out; 242 } 243 244 for (i = 0; i < nr_pages; i++) { 245 page = balloon_retrieve(); 246 KASSERT(page, ("balloon_retrieve failed")); 247 248 pfn = (VM_PAGE_TO_PHYS(page) >> PAGE_SHIFT); 249 KASSERT((xen_feature(XENFEAT_auto_translated_physmap) || 250 !phys_to_machine_mapping_valid(pfn)), 251 ("auto translated physmap but mapping is valid")); 252 253 set_phys_to_machine(pfn, frame_list[i]); 254 255 #if 0 256 #ifndef XENHVM 257 /* Link back into the page tables if not highmem. */ 258 if (pfn < max_low_pfn) { 259 int ret; 260 ret = HYPERVISOR_update_va_mapping( 261 (unsigned long)__va(pfn << PAGE_SHIFT), 262 pfn_pte_ma(frame_list[i], PAGE_KERNEL), 263 0); 264 PASSING(ret == 0, 265 ("HYPERVISOR_update_va_mapping failed")); 266 } 267 #endif 268 #endif 269 270 /* Relinquish the page back to the allocator. */ 271 vm_page_unwire(page, 0); 272 vm_page_free(page); 273 } 274 275 bs.current_pages += nr_pages; 276 //totalram_pages = bs.current_pages; 277 278 out: 279 mtx_unlock(&balloon_lock); 280 281 return 0; 282 } 283 284 static int 285 decrease_reservation(unsigned long nr_pages) 286 { 287 unsigned long pfn, i; 288 vm_page_t page; 289 int need_sleep = 0; 290 int ret; 291 struct xen_memory_reservation reservation = { 292 .address_bits = 0, 293 .extent_order = 0, 294 .domid = DOMID_SELF 295 }; 296 297 if (nr_pages > ARRAY_SIZE(frame_list)) 298 nr_pages = ARRAY_SIZE(frame_list); 299 300 for (i = 0; i < nr_pages; i++) { 301 if ((page = vm_page_alloc(NULL, 0, 302 VM_ALLOC_NORMAL | VM_ALLOC_NOOBJ | 303 VM_ALLOC_WIRED | VM_ALLOC_ZERO)) == NULL) { 304 nr_pages = i; 305 need_sleep = 1; 306 break; 307 } 308 309 pfn = (VM_PAGE_TO_PHYS(page) >> PAGE_SHIFT); 310 frame_list[i] = PFNTOMFN(pfn); 311 312 #if 0 313 if (!PageHighMem(page)) { 314 v = phys_to_virt(pfn << PAGE_SHIFT); 315 scrub_pages(v, 1); 316 #ifdef CONFIG_XEN 317 ret = HYPERVISOR_update_va_mapping( 318 (unsigned long)v, __pte_ma(0), 0); 319 BUG_ON(ret); 320 #endif 321 } 322 #endif 323 #ifdef CONFIG_XEN_SCRUB_PAGES 324 else { 325 v = kmap(page); 326 scrub_pages(v, 1); 327 kunmap(page); 328 } 329 #endif 330 } 331 332 #ifdef CONFIG_XEN 333 /* Ensure that ballooned highmem pages don't have kmaps. */ 334 kmap_flush_unused(); 335 flush_tlb_all(); 336 #endif 337 338 mtx_lock(&balloon_lock); 339 340 /* No more mappings: invalidate P2M and add to balloon. */ 341 for (i = 0; i < nr_pages; i++) { 342 pfn = MFNTOPFN(frame_list[i]); 343 set_phys_to_machine(pfn, INVALID_P2M_ENTRY); 344 balloon_append(PHYS_TO_VM_PAGE(pfn << PAGE_SHIFT)); 345 } 346 347 set_xen_guest_handle(reservation.extent_start, frame_list); 348 reservation.nr_extents = nr_pages; 349 ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation); 350 KASSERT(ret == nr_pages, ("HYPERVISOR_memory_op failed")); 351 352 bs.current_pages -= nr_pages; 353 //totalram_pages = bs.current_pages; 354 355 mtx_unlock(&balloon_lock); 356 357 return (need_sleep); 358 } 359 360 /* 361 * We avoid multiple worker processes conflicting via the balloon mutex. 362 * We may of course race updates of the target counts (which are protected 363 * by the balloon lock), or with changes to the Xen hard limit, but we will 364 * recover from these in time. 365 */ 366 static void 367 balloon_process(void *unused) 368 { 369 int need_sleep = 0; 370 long credit; 371 372 mtx_lock(&balloon_mutex); 373 for (;;) { 374 int sleep_time; 375 376 do { 377 credit = current_target() - bs.current_pages; 378 if (credit > 0) 379 need_sleep = (increase_reservation(credit) != 0); 380 if (credit < 0) 381 need_sleep = (decrease_reservation(-credit) != 0); 382 383 } while ((credit != 0) && !need_sleep); 384 385 /* Schedule more work if there is some still to be done. */ 386 if (current_target() != bs.current_pages) 387 sleep_time = hz; 388 else 389 sleep_time = 0; 390 391 msleep(balloon_process, &balloon_mutex, 0, "balloon", 392 sleep_time); 393 } 394 mtx_unlock(&balloon_mutex); 395 } 396 397 /* Resets the Xen limit, sets new target, and kicks off processing. */ 398 static void 399 set_new_target(unsigned long target) 400 { 401 /* No need for lock. Not read-modify-write updates. */ 402 bs.hard_limit = ~0UL; 403 bs.target_pages = max(target, minimum_target()); 404 wakeup(balloon_process); 405 } 406 407 static struct xs_watch target_watch = 408 { 409 .node = "memory/target" 410 }; 411 412 /* React to a change in the target key */ 413 static void 414 watch_target(struct xs_watch *watch, 415 const char **vec, unsigned int len) 416 { 417 unsigned long long new_target; 418 int err; 419 420 err = xs_scanf(XST_NIL, "memory", "target", NULL, 421 "%llu", &new_target); 422 if (err) { 423 /* This is ok (for domain0 at least) - so just return */ 424 return; 425 } 426 427 /* The given memory/target value is in KiB, so it needs converting to 428 pages. PAGE_SHIFT converts bytes to pages, hence PAGE_SHIFT - 10. 429 */ 430 set_new_target(new_target >> (PAGE_SHIFT - 10)); 431 432 } 433 434 static void 435 balloon_init_watcher(void *arg) 436 { 437 int err; 438 439 err = xs_register_watch(&target_watch); 440 if (err) 441 printf("Failed to set balloon watcher\n"); 442 443 } 444 SYSINIT(balloon_init_watcher, SI_SUB_PSEUDO, SI_ORDER_ANY, 445 balloon_init_watcher, NULL); 446 447 static void 448 balloon_init(void *arg) 449 { 450 #ifndef XENHVM 451 vm_page_t page; 452 unsigned long pfn; 453 454 #define max_pfn HYPERVISOR_shared_info->arch.max_pfn 455 #endif 456 457 if (!is_running_on_xen()) 458 return; 459 460 mtx_init(&balloon_lock, "balloon_lock", NULL, MTX_DEF); 461 mtx_init(&balloon_mutex, "balloon_mutex", NULL, MTX_DEF); 462 463 #ifndef XENHVM 464 bs.current_pages = min(xen_start_info->nr_pages, max_pfn); 465 #else 466 bs.current_pages = physmem; 467 #endif 468 bs.target_pages = bs.current_pages; 469 bs.balloon_low = 0; 470 bs.balloon_high = 0; 471 bs.driver_pages = 0UL; 472 bs.hard_limit = ~0UL; 473 474 kproc_create(balloon_process, NULL, NULL, 0, 0, "balloon"); 475 476 #ifndef XENHVM 477 /* Initialise the balloon with excess memory space. */ 478 for (pfn = xen_start_info->nr_pages; pfn < max_pfn; pfn++) { 479 page = PHYS_TO_VM_PAGE(pfn << PAGE_SHIFT); 480 balloon_append(page); 481 } 482 #undef max_pfn 483 #endif 484 485 target_watch.callback = watch_target; 486 487 return; 488 } 489 SYSINIT(balloon_init, SI_SUB_PSEUDO, SI_ORDER_ANY, balloon_init, NULL); 490 491 void balloon_update_driver_allowance(long delta); 492 493 void 494 balloon_update_driver_allowance(long delta) 495 { 496 mtx_lock(&balloon_lock); 497 bs.driver_pages += delta; 498 mtx_unlock(&balloon_lock); 499 } 500 501 #if 0 502 static int dealloc_pte_fn( 503 pte_t *pte, struct page *pte_page, unsigned long addr, void *data) 504 { 505 unsigned long mfn = pte_mfn(*pte); 506 int ret; 507 struct xen_memory_reservation reservation = { 508 .extent_start = &mfn, 509 .nr_extents = 1, 510 .extent_order = 0, 511 .domid = DOMID_SELF 512 }; 513 set_pte_at(&init_mm, addr, pte, __pte_ma(0)); 514 set_phys_to_machine(__pa(addr) >> PAGE_SHIFT, INVALID_P2M_ENTRY); 515 ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation); 516 KASSERT(ret == 1, ("HYPERVISOR_memory_op failed")); 517 return 0; 518 } 519 520 #endif 521 522 #if 0 523 vm_page_t 524 balloon_alloc_empty_page_range(unsigned long nr_pages) 525 { 526 vm_page_t pages; 527 int i, rc; 528 unsigned long *mfn_list; 529 struct xen_memory_reservation reservation = { 530 .address_bits = 0, 531 .extent_order = 0, 532 .domid = DOMID_SELF 533 }; 534 535 pages = vm_page_alloc_contig(nr_pages, 0, -1, 4, 4) 536 if (pages == NULL) 537 return NULL; 538 539 mfn_list = malloc(nr_pages*sizeof(unsigned long), M_DEVBUF, M_WAITOK); 540 541 for (i = 0; i < nr_pages; i++) { 542 mfn_list[i] = PFNTOMFN(VM_PAGE_TO_PHYS(pages[i]) >> PAGE_SHIFT); 543 PFNTOMFN(i) = INVALID_P2M_ENTRY; 544 reservation.extent_start = mfn_list; 545 reservation.nr_extents = nr_pages; 546 rc = HYPERVISOR_memory_op(XENMEM_decrease_reservation, 547 &reservation); 548 KASSERT(rc == nr_pages, ("HYPERVISOR_memory_op failed")); 549 } 550 551 current_pages -= nr_pages; 552 553 wakeup(balloon_process); 554 555 return pages; 556 } 557 558 void 559 balloon_dealloc_empty_page_range(vm_page_t page, unsigned long nr_pages) 560 { 561 unsigned long i; 562 563 for (i = 0; i < nr_pages; i++) 564 balloon_append(page + i); 565 566 wakeup(balloon_process); 567 } 568 #endif 569