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 program is free software; you can redistribute it and/or 11 * modify it under the terms of the GNU General Public License version 2 12 * as published by the Free Software Foundation; or, when distributed 13 * separately from the Linux kernel or incorporated into other 14 * software packages, subject to the following license: 15 * 16 * Permission is hereby granted, free of charge, to any person obtaining a copy 17 * of this source file (the "Software"), to deal in the Software without 18 * restriction, including without limitation the rights to use, copy, modify, 19 * merge, publish, distribute, sublicense, and/or sell copies of the Software, 20 * and to permit persons to whom the Software is furnished to do so, subject to 21 * the following conditions: 22 * 23 * The above copyright notice and this permission notice shall be included in 24 * all copies or substantial portions of the Software. 25 * 26 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 27 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 28 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 29 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 30 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING 31 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS 32 * IN THE SOFTWARE. 33 */ 34 35 #include <linux/kernel.h> 36 #include <linux/module.h> 37 #include <linux/sched.h> 38 #include <linux/errno.h> 39 #include <linux/mm.h> 40 #include <linux/bootmem.h> 41 #include <linux/pagemap.h> 42 #include <linux/highmem.h> 43 #include <linux/mutex.h> 44 #include <linux/list.h> 45 #include <linux/sysdev.h> 46 47 #include <asm/xen/hypervisor.h> 48 #include <asm/page.h> 49 #include <asm/pgalloc.h> 50 #include <asm/pgtable.h> 51 #include <asm/uaccess.h> 52 #include <asm/tlb.h> 53 54 #include <xen/interface/memory.h> 55 #include <xen/xenbus.h> 56 #include <xen/features.h> 57 #include <xen/page.h> 58 59 #define PAGES2KB(_p) ((_p)<<(PAGE_SHIFT-10)) 60 61 #define BALLOON_CLASS_NAME "xen_memory" 62 63 struct balloon_stats { 64 /* We aim for 'current allocation' == 'target allocation'. */ 65 unsigned long current_pages; 66 unsigned long target_pages; 67 /* We may hit the hard limit in Xen. If we do then we remember it. */ 68 unsigned long hard_limit; 69 /* 70 * Drivers may alter the memory reservation independently, but they 71 * must inform the balloon driver so we avoid hitting the hard limit. 72 */ 73 unsigned long driver_pages; 74 /* Number of pages in high- and low-memory balloons. */ 75 unsigned long balloon_low; 76 unsigned long balloon_high; 77 }; 78 79 static DEFINE_MUTEX(balloon_mutex); 80 81 static struct sys_device balloon_sysdev; 82 83 static int register_balloon(struct sys_device *sysdev); 84 85 /* 86 * Protects atomic reservation decrease/increase against concurrent increases. 87 * Also protects non-atomic updates of current_pages and driver_pages, and 88 * balloon lists. 89 */ 90 static DEFINE_SPINLOCK(balloon_lock); 91 92 static struct balloon_stats balloon_stats; 93 94 /* We increase/decrease in batches which fit in a page */ 95 static unsigned long frame_list[PAGE_SIZE / sizeof(unsigned long)]; 96 97 /* VM /proc information for memory */ 98 extern unsigned long totalram_pages; 99 100 #ifdef CONFIG_HIGHMEM 101 extern unsigned long totalhigh_pages; 102 #define inc_totalhigh_pages() (totalhigh_pages++) 103 #define dec_totalhigh_pages() (totalhigh_pages--) 104 #else 105 #define inc_totalhigh_pages() do {} while(0) 106 #define dec_totalhigh_pages() do {} while(0) 107 #endif 108 109 /* List of ballooned pages, threaded through the mem_map array. */ 110 static LIST_HEAD(ballooned_pages); 111 112 /* Main work function, always executed in process context. */ 113 static void balloon_process(struct work_struct *work); 114 static DECLARE_WORK(balloon_worker, balloon_process); 115 static struct timer_list balloon_timer; 116 117 /* When ballooning out (allocating memory to return to Xen) we don't really 118 want the kernel to try too hard since that can trigger the oom killer. */ 119 #define GFP_BALLOON \ 120 (GFP_HIGHUSER | __GFP_NOWARN | __GFP_NORETRY | __GFP_NOMEMALLOC) 121 122 static void scrub_page(struct page *page) 123 { 124 #ifdef CONFIG_XEN_SCRUB_PAGES 125 clear_highpage(page); 126 #endif 127 } 128 129 /* balloon_append: add the given page to the balloon. */ 130 static void balloon_append(struct page *page) 131 { 132 /* Lowmem is re-populated first, so highmem pages go at list tail. */ 133 if (PageHighMem(page)) { 134 list_add_tail(&page->lru, &ballooned_pages); 135 balloon_stats.balloon_high++; 136 dec_totalhigh_pages(); 137 } else { 138 list_add(&page->lru, &ballooned_pages); 139 balloon_stats.balloon_low++; 140 } 141 } 142 143 /* balloon_retrieve: rescue a page from the balloon, if it is not empty. */ 144 static struct page *balloon_retrieve(void) 145 { 146 struct page *page; 147 148 if (list_empty(&ballooned_pages)) 149 return NULL; 150 151 page = list_entry(ballooned_pages.next, struct page, lru); 152 list_del(&page->lru); 153 154 if (PageHighMem(page)) { 155 balloon_stats.balloon_high--; 156 inc_totalhigh_pages(); 157 } 158 else 159 balloon_stats.balloon_low--; 160 161 return page; 162 } 163 164 static struct page *balloon_first_page(void) 165 { 166 if (list_empty(&ballooned_pages)) 167 return NULL; 168 return list_entry(ballooned_pages.next, struct page, lru); 169 } 170 171 static struct page *balloon_next_page(struct page *page) 172 { 173 struct list_head *next = page->lru.next; 174 if (next == &ballooned_pages) 175 return NULL; 176 return list_entry(next, struct page, lru); 177 } 178 179 static void balloon_alarm(unsigned long unused) 180 { 181 schedule_work(&balloon_worker); 182 } 183 184 static unsigned long current_target(void) 185 { 186 unsigned long target = min(balloon_stats.target_pages, balloon_stats.hard_limit); 187 188 target = min(target, 189 balloon_stats.current_pages + 190 balloon_stats.balloon_low + 191 balloon_stats.balloon_high); 192 193 return target; 194 } 195 196 static int increase_reservation(unsigned long nr_pages) 197 { 198 unsigned long pfn, i, flags; 199 struct page *page; 200 long rc; 201 struct xen_memory_reservation reservation = { 202 .address_bits = 0, 203 .extent_order = 0, 204 .domid = DOMID_SELF 205 }; 206 207 if (nr_pages > ARRAY_SIZE(frame_list)) 208 nr_pages = ARRAY_SIZE(frame_list); 209 210 spin_lock_irqsave(&balloon_lock, flags); 211 212 page = balloon_first_page(); 213 for (i = 0; i < nr_pages; i++) { 214 BUG_ON(page == NULL); 215 frame_list[i] = page_to_pfn(page);; 216 page = balloon_next_page(page); 217 } 218 219 set_xen_guest_handle(reservation.extent_start, frame_list); 220 reservation.nr_extents = nr_pages; 221 rc = HYPERVISOR_memory_op(XENMEM_populate_physmap, &reservation); 222 if (rc < nr_pages) { 223 if (rc > 0) { 224 int ret; 225 226 /* We hit the Xen hard limit: reprobe. */ 227 reservation.nr_extents = rc; 228 ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation, 229 &reservation); 230 BUG_ON(ret != rc); 231 } 232 if (rc >= 0) 233 balloon_stats.hard_limit = (balloon_stats.current_pages + rc - 234 balloon_stats.driver_pages); 235 goto out; 236 } 237 238 for (i = 0; i < nr_pages; i++) { 239 page = balloon_retrieve(); 240 BUG_ON(page == NULL); 241 242 pfn = page_to_pfn(page); 243 BUG_ON(!xen_feature(XENFEAT_auto_translated_physmap) && 244 phys_to_machine_mapping_valid(pfn)); 245 246 set_phys_to_machine(pfn, frame_list[i]); 247 248 /* Link back into the page tables if not highmem. */ 249 if (pfn < max_low_pfn) { 250 int ret; 251 ret = HYPERVISOR_update_va_mapping( 252 (unsigned long)__va(pfn << PAGE_SHIFT), 253 mfn_pte(frame_list[i], PAGE_KERNEL), 254 0); 255 BUG_ON(ret); 256 } 257 258 /* Relinquish the page back to the allocator. */ 259 ClearPageReserved(page); 260 init_page_count(page); 261 __free_page(page); 262 } 263 264 balloon_stats.current_pages += nr_pages; 265 totalram_pages = balloon_stats.current_pages; 266 267 out: 268 spin_unlock_irqrestore(&balloon_lock, flags); 269 270 return 0; 271 } 272 273 static int decrease_reservation(unsigned long nr_pages) 274 { 275 unsigned long pfn, i, flags; 276 struct page *page; 277 int need_sleep = 0; 278 int ret; 279 struct xen_memory_reservation reservation = { 280 .address_bits = 0, 281 .extent_order = 0, 282 .domid = DOMID_SELF 283 }; 284 285 if (nr_pages > ARRAY_SIZE(frame_list)) 286 nr_pages = ARRAY_SIZE(frame_list); 287 288 for (i = 0; i < nr_pages; i++) { 289 if ((page = alloc_page(GFP_BALLOON)) == NULL) { 290 nr_pages = i; 291 need_sleep = 1; 292 break; 293 } 294 295 pfn = page_to_pfn(page); 296 frame_list[i] = pfn_to_mfn(pfn); 297 298 scrub_page(page); 299 } 300 301 /* Ensure that ballooned highmem pages don't have kmaps. */ 302 kmap_flush_unused(); 303 flush_tlb_all(); 304 305 spin_lock_irqsave(&balloon_lock, flags); 306 307 /* No more mappings: invalidate P2M and add to balloon. */ 308 for (i = 0; i < nr_pages; i++) { 309 pfn = mfn_to_pfn(frame_list[i]); 310 set_phys_to_machine(pfn, INVALID_P2M_ENTRY); 311 balloon_append(pfn_to_page(pfn)); 312 } 313 314 set_xen_guest_handle(reservation.extent_start, frame_list); 315 reservation.nr_extents = nr_pages; 316 ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation); 317 BUG_ON(ret != nr_pages); 318 319 balloon_stats.current_pages -= nr_pages; 320 totalram_pages = balloon_stats.current_pages; 321 322 spin_unlock_irqrestore(&balloon_lock, flags); 323 324 return need_sleep; 325 } 326 327 /* 328 * We avoid multiple worker processes conflicting via the balloon mutex. 329 * We may of course race updates of the target counts (which are protected 330 * by the balloon lock), or with changes to the Xen hard limit, but we will 331 * recover from these in time. 332 */ 333 static void balloon_process(struct work_struct *work) 334 { 335 int need_sleep = 0; 336 long credit; 337 338 mutex_lock(&balloon_mutex); 339 340 do { 341 credit = current_target() - balloon_stats.current_pages; 342 if (credit > 0) 343 need_sleep = (increase_reservation(credit) != 0); 344 if (credit < 0) 345 need_sleep = (decrease_reservation(-credit) != 0); 346 347 #ifndef CONFIG_PREEMPT 348 if (need_resched()) 349 schedule(); 350 #endif 351 } while ((credit != 0) && !need_sleep); 352 353 /* Schedule more work if there is some still to be done. */ 354 if (current_target() != balloon_stats.current_pages) 355 mod_timer(&balloon_timer, jiffies + HZ); 356 357 mutex_unlock(&balloon_mutex); 358 } 359 360 /* Resets the Xen limit, sets new target, and kicks off processing. */ 361 static void balloon_set_new_target(unsigned long target) 362 { 363 /* No need for lock. Not read-modify-write updates. */ 364 balloon_stats.hard_limit = ~0UL; 365 balloon_stats.target_pages = target; 366 schedule_work(&balloon_worker); 367 } 368 369 static struct xenbus_watch target_watch = 370 { 371 .node = "memory/target" 372 }; 373 374 /* React to a change in the target key */ 375 static void watch_target(struct xenbus_watch *watch, 376 const char **vec, unsigned int len) 377 { 378 unsigned long long new_target; 379 int err; 380 381 err = xenbus_scanf(XBT_NIL, "memory", "target", "%llu", &new_target); 382 if (err != 1) { 383 /* This is ok (for domain0 at least) - so just return */ 384 return; 385 } 386 387 /* The given memory/target value is in KiB, so it needs converting to 388 * pages. PAGE_SHIFT converts bytes to pages, hence PAGE_SHIFT - 10. 389 */ 390 balloon_set_new_target(new_target >> (PAGE_SHIFT - 10)); 391 } 392 393 static int balloon_init_watcher(struct notifier_block *notifier, 394 unsigned long event, 395 void *data) 396 { 397 int err; 398 399 err = register_xenbus_watch(&target_watch); 400 if (err) 401 printk(KERN_ERR "Failed to set balloon watcher\n"); 402 403 return NOTIFY_DONE; 404 } 405 406 static struct notifier_block xenstore_notifier; 407 408 static int __init balloon_init(void) 409 { 410 unsigned long pfn; 411 struct page *page; 412 413 if (!xen_pv_domain()) 414 return -ENODEV; 415 416 pr_info("xen_balloon: Initialising balloon driver.\n"); 417 418 balloon_stats.current_pages = min(xen_start_info->nr_pages, max_pfn); 419 totalram_pages = balloon_stats.current_pages; 420 balloon_stats.target_pages = balloon_stats.current_pages; 421 balloon_stats.balloon_low = 0; 422 balloon_stats.balloon_high = 0; 423 balloon_stats.driver_pages = 0UL; 424 balloon_stats.hard_limit = ~0UL; 425 426 init_timer(&balloon_timer); 427 balloon_timer.data = 0; 428 balloon_timer.function = balloon_alarm; 429 430 register_balloon(&balloon_sysdev); 431 432 /* Initialise the balloon with excess memory space. */ 433 for (pfn = xen_start_info->nr_pages; pfn < max_pfn; pfn++) { 434 page = pfn_to_page(pfn); 435 if (!PageReserved(page)) 436 balloon_append(page); 437 } 438 439 target_watch.callback = watch_target; 440 xenstore_notifier.notifier_call = balloon_init_watcher; 441 442 register_xenstore_notifier(&xenstore_notifier); 443 444 return 0; 445 } 446 447 subsys_initcall(balloon_init); 448 449 static void balloon_exit(void) 450 { 451 /* XXX - release balloon here */ 452 return; 453 } 454 455 module_exit(balloon_exit); 456 457 #define BALLOON_SHOW(name, format, args...) \ 458 static ssize_t show_##name(struct sys_device *dev, \ 459 struct sysdev_attribute *attr, \ 460 char *buf) \ 461 { \ 462 return sprintf(buf, format, ##args); \ 463 } \ 464 static SYSDEV_ATTR(name, S_IRUGO, show_##name, NULL) 465 466 BALLOON_SHOW(current_kb, "%lu\n", PAGES2KB(balloon_stats.current_pages)); 467 BALLOON_SHOW(low_kb, "%lu\n", PAGES2KB(balloon_stats.balloon_low)); 468 BALLOON_SHOW(high_kb, "%lu\n", PAGES2KB(balloon_stats.balloon_high)); 469 BALLOON_SHOW(hard_limit_kb, 470 (balloon_stats.hard_limit!=~0UL) ? "%lu\n" : "???\n", 471 (balloon_stats.hard_limit!=~0UL) ? PAGES2KB(balloon_stats.hard_limit) : 0); 472 BALLOON_SHOW(driver_kb, "%lu\n", PAGES2KB(balloon_stats.driver_pages)); 473 474 static ssize_t show_target_kb(struct sys_device *dev, struct sysdev_attribute *attr, 475 char *buf) 476 { 477 return sprintf(buf, "%lu\n", PAGES2KB(balloon_stats.target_pages)); 478 } 479 480 static ssize_t store_target_kb(struct sys_device *dev, 481 struct sysdev_attribute *attr, 482 const char *buf, 483 size_t count) 484 { 485 char *endchar; 486 unsigned long long target_bytes; 487 488 if (!capable(CAP_SYS_ADMIN)) 489 return -EPERM; 490 491 target_bytes = memparse(buf, &endchar); 492 493 balloon_set_new_target(target_bytes >> PAGE_SHIFT); 494 495 return count; 496 } 497 498 static SYSDEV_ATTR(target_kb, S_IRUGO | S_IWUSR, 499 show_target_kb, store_target_kb); 500 501 static struct sysdev_attribute *balloon_attrs[] = { 502 &attr_target_kb, 503 }; 504 505 static struct attribute *balloon_info_attrs[] = { 506 &attr_current_kb.attr, 507 &attr_low_kb.attr, 508 &attr_high_kb.attr, 509 &attr_hard_limit_kb.attr, 510 &attr_driver_kb.attr, 511 NULL 512 }; 513 514 static struct attribute_group balloon_info_group = { 515 .name = "info", 516 .attrs = balloon_info_attrs, 517 }; 518 519 static struct sysdev_class balloon_sysdev_class = { 520 .name = BALLOON_CLASS_NAME, 521 }; 522 523 static int register_balloon(struct sys_device *sysdev) 524 { 525 int i, error; 526 527 error = sysdev_class_register(&balloon_sysdev_class); 528 if (error) 529 return error; 530 531 sysdev->id = 0; 532 sysdev->cls = &balloon_sysdev_class; 533 534 error = sysdev_register(sysdev); 535 if (error) { 536 sysdev_class_unregister(&balloon_sysdev_class); 537 return error; 538 } 539 540 for (i = 0; i < ARRAY_SIZE(balloon_attrs); i++) { 541 error = sysdev_create_file(sysdev, balloon_attrs[i]); 542 if (error) 543 goto fail; 544 } 545 546 error = sysfs_create_group(&sysdev->kobj, &balloon_info_group); 547 if (error) 548 goto fail; 549 550 return 0; 551 552 fail: 553 while (--i >= 0) 554 sysdev_remove_file(sysdev, balloon_attrs[i]); 555 sysdev_unregister(sysdev); 556 sysdev_class_unregister(&balloon_sysdev_class); 557 return error; 558 } 559 560 MODULE_LICENSE("GPL"); 561