balloon.c revision 618b2c8db24522ae273d8299c6a936ea13793c4d
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/page.h> 48#include <asm/pgalloc.h> 49#include <asm/pgtable.h> 50#include <asm/uaccess.h> 51#include <asm/tlb.h> 52 53#include <asm/xen/hypervisor.h> 54#include <asm/xen/hypercall.h> 55#include <xen/interface/xen.h> 56#include <xen/interface/memory.h> 57#include <xen/xenbus.h> 58#include <xen/features.h> 59#include <xen/page.h> 60 61#define PAGES2KB(_p) ((_p)<<(PAGE_SHIFT-10)) 62 63#define BALLOON_CLASS_NAME "xen_memory" 64 65struct balloon_stats { 66 /* We aim for 'current allocation' == 'target allocation'. */ 67 unsigned long current_pages; 68 unsigned long target_pages; 69 /* We may hit the hard limit in Xen. If we do then we remember it. */ 70 unsigned long hard_limit; 71 /* 72 * Drivers may alter the memory reservation independently, but they 73 * must inform the balloon driver so we avoid hitting the hard limit. 74 */ 75 unsigned long driver_pages; 76 /* Number of pages in high- and low-memory balloons. */ 77 unsigned long balloon_low; 78 unsigned long balloon_high; 79}; 80 81static DEFINE_MUTEX(balloon_mutex); 82 83static struct sys_device balloon_sysdev; 84 85static int register_balloon(struct sys_device *sysdev); 86 87/* 88 * Protects atomic reservation decrease/increase against concurrent increases. 89 * Also protects non-atomic updates of current_pages and driver_pages, and 90 * balloon lists. 91 */ 92static DEFINE_SPINLOCK(balloon_lock); 93 94static struct balloon_stats balloon_stats; 95 96/* We increase/decrease in batches which fit in a page */ 97static unsigned long frame_list[PAGE_SIZE / sizeof(unsigned long)]; 98 99/* VM /proc information for memory */ 100extern unsigned long totalram_pages; 101 102#ifdef CONFIG_HIGHMEM 103extern unsigned long totalhigh_pages; 104#define inc_totalhigh_pages() (totalhigh_pages++) 105#define dec_totalhigh_pages() (totalhigh_pages--) 106#else 107#define inc_totalhigh_pages() do {} while(0) 108#define dec_totalhigh_pages() do {} while(0) 109#endif 110 111/* List of ballooned pages, threaded through the mem_map array. */ 112static LIST_HEAD(ballooned_pages); 113 114/* Main work function, always executed in process context. */ 115static void balloon_process(struct work_struct *work); 116static DECLARE_WORK(balloon_worker, balloon_process); 117static struct timer_list balloon_timer; 118 119/* When ballooning out (allocating memory to return to Xen) we don't really 120 want the kernel to try too hard since that can trigger the oom killer. */ 121#define GFP_BALLOON \ 122 (GFP_HIGHUSER | __GFP_NOWARN | __GFP_NORETRY | __GFP_NOMEMALLOC) 123 124static void scrub_page(struct page *page) 125{ 126#ifdef CONFIG_XEN_SCRUB_PAGES 127 clear_highpage(page); 128#endif 129} 130 131/* balloon_append: add the given page to the balloon. */ 132static void balloon_append(struct page *page) 133{ 134 /* Lowmem is re-populated first, so highmem pages go at list tail. */ 135 if (PageHighMem(page)) { 136 list_add_tail(&page->lru, &ballooned_pages); 137 balloon_stats.balloon_high++; 138 dec_totalhigh_pages(); 139 } else { 140 list_add(&page->lru, &ballooned_pages); 141 balloon_stats.balloon_low++; 142 } 143} 144 145/* balloon_retrieve: rescue a page from the balloon, if it is not empty. */ 146static struct page *balloon_retrieve(void) 147{ 148 struct page *page; 149 150 if (list_empty(&ballooned_pages)) 151 return NULL; 152 153 page = list_entry(ballooned_pages.next, struct page, lru); 154 list_del(&page->lru); 155 156 if (PageHighMem(page)) { 157 balloon_stats.balloon_high--; 158 inc_totalhigh_pages(); 159 } 160 else 161 balloon_stats.balloon_low--; 162 163 return page; 164} 165 166static struct page *balloon_first_page(void) 167{ 168 if (list_empty(&ballooned_pages)) 169 return NULL; 170 return list_entry(ballooned_pages.next, struct page, lru); 171} 172 173static struct page *balloon_next_page(struct page *page) 174{ 175 struct list_head *next = page->lru.next; 176 if (next == &ballooned_pages) 177 return NULL; 178 return list_entry(next, struct page, lru); 179} 180 181static void balloon_alarm(unsigned long unused) 182{ 183 schedule_work(&balloon_worker); 184} 185 186static unsigned long current_target(void) 187{ 188 unsigned long target = min(balloon_stats.target_pages, balloon_stats.hard_limit); 189 190 target = min(target, 191 balloon_stats.current_pages + 192 balloon_stats.balloon_low + 193 balloon_stats.balloon_high); 194 195 return target; 196} 197 198static int increase_reservation(unsigned long nr_pages) 199{ 200 unsigned long pfn, i, flags; 201 struct page *page; 202 long rc; 203 struct xen_memory_reservation reservation = { 204 .address_bits = 0, 205 .extent_order = 0, 206 .domid = DOMID_SELF 207 }; 208 209 if (nr_pages > ARRAY_SIZE(frame_list)) 210 nr_pages = ARRAY_SIZE(frame_list); 211 212 spin_lock_irqsave(&balloon_lock, flags); 213 214 page = balloon_first_page(); 215 for (i = 0; i < nr_pages; i++) { 216 BUG_ON(page == NULL); 217 frame_list[i] = page_to_pfn(page);; 218 page = balloon_next_page(page); 219 } 220 221 set_xen_guest_handle(reservation.extent_start, frame_list); 222 reservation.nr_extents = nr_pages; 223 rc = HYPERVISOR_memory_op(XENMEM_populate_physmap, &reservation); 224 if (rc < nr_pages) { 225 if (rc > 0) { 226 int ret; 227 228 /* We hit the Xen hard limit: reprobe. */ 229 reservation.nr_extents = rc; 230 ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation, 231 &reservation); 232 BUG_ON(ret != rc); 233 } 234 if (rc >= 0) 235 balloon_stats.hard_limit = (balloon_stats.current_pages + rc - 236 balloon_stats.driver_pages); 237 goto out; 238 } 239 240 for (i = 0; i < nr_pages; i++) { 241 page = balloon_retrieve(); 242 BUG_ON(page == NULL); 243 244 pfn = page_to_pfn(page); 245 BUG_ON(!xen_feature(XENFEAT_auto_translated_physmap) && 246 phys_to_machine_mapping_valid(pfn)); 247 248 set_phys_to_machine(pfn, frame_list[i]); 249 250 /* Link back into the page tables if not highmem. */ 251 if (pfn < max_low_pfn) { 252 int ret; 253 ret = HYPERVISOR_update_va_mapping( 254 (unsigned long)__va(pfn << PAGE_SHIFT), 255 mfn_pte(frame_list[i], PAGE_KERNEL), 256 0); 257 BUG_ON(ret); 258 } 259 260 /* Relinquish the page back to the allocator. */ 261 ClearPageReserved(page); 262 init_page_count(page); 263 __free_page(page); 264 } 265 266 balloon_stats.current_pages += nr_pages; 267 totalram_pages = balloon_stats.current_pages; 268 269 out: 270 spin_unlock_irqrestore(&balloon_lock, flags); 271 272 return 0; 273} 274 275static int decrease_reservation(unsigned long nr_pages) 276{ 277 unsigned long pfn, i, flags; 278 struct page *page; 279 int need_sleep = 0; 280 int ret; 281 struct xen_memory_reservation reservation = { 282 .address_bits = 0, 283 .extent_order = 0, 284 .domid = DOMID_SELF 285 }; 286 287 if (nr_pages > ARRAY_SIZE(frame_list)) 288 nr_pages = ARRAY_SIZE(frame_list); 289 290 for (i = 0; i < nr_pages; i++) { 291 if ((page = alloc_page(GFP_BALLOON)) == NULL) { 292 nr_pages = i; 293 need_sleep = 1; 294 break; 295 } 296 297 pfn = page_to_pfn(page); 298 frame_list[i] = pfn_to_mfn(pfn); 299 300 scrub_page(page); 301 302 if (!PageHighMem(page)) { 303 ret = HYPERVISOR_update_va_mapping( 304 (unsigned long)__va(pfn << PAGE_SHIFT), 305 __pte_ma(0), 0); 306 BUG_ON(ret); 307 } 308 309 } 310 311 /* Ensure that ballooned highmem pages don't have kmaps. */ 312 kmap_flush_unused(); 313 flush_tlb_all(); 314 315 spin_lock_irqsave(&balloon_lock, flags); 316 317 /* No more mappings: invalidate P2M and add to balloon. */ 318 for (i = 0; i < nr_pages; i++) { 319 pfn = mfn_to_pfn(frame_list[i]); 320 set_phys_to_machine(pfn, INVALID_P2M_ENTRY); 321 balloon_append(pfn_to_page(pfn)); 322 } 323 324 set_xen_guest_handle(reservation.extent_start, frame_list); 325 reservation.nr_extents = nr_pages; 326 ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation); 327 BUG_ON(ret != nr_pages); 328 329 balloon_stats.current_pages -= nr_pages; 330 totalram_pages = balloon_stats.current_pages; 331 332 spin_unlock_irqrestore(&balloon_lock, flags); 333 334 return need_sleep; 335} 336 337/* 338 * We avoid multiple worker processes conflicting via the balloon mutex. 339 * We may of course race updates of the target counts (which are protected 340 * by the balloon lock), or with changes to the Xen hard limit, but we will 341 * recover from these in time. 342 */ 343static void balloon_process(struct work_struct *work) 344{ 345 int need_sleep = 0; 346 long credit; 347 348 mutex_lock(&balloon_mutex); 349 350 do { 351 credit = current_target() - balloon_stats.current_pages; 352 if (credit > 0) 353 need_sleep = (increase_reservation(credit) != 0); 354 if (credit < 0) 355 need_sleep = (decrease_reservation(-credit) != 0); 356 357#ifndef CONFIG_PREEMPT 358 if (need_resched()) 359 schedule(); 360#endif 361 } while ((credit != 0) && !need_sleep); 362 363 /* Schedule more work if there is some still to be done. */ 364 if (current_target() != balloon_stats.current_pages) 365 mod_timer(&balloon_timer, jiffies + HZ); 366 367 mutex_unlock(&balloon_mutex); 368} 369 370/* Resets the Xen limit, sets new target, and kicks off processing. */ 371static void balloon_set_new_target(unsigned long target) 372{ 373 /* No need for lock. Not read-modify-write updates. */ 374 balloon_stats.hard_limit = ~0UL; 375 balloon_stats.target_pages = target; 376 schedule_work(&balloon_worker); 377} 378 379static struct xenbus_watch target_watch = 380{ 381 .node = "memory/target" 382}; 383 384/* React to a change in the target key */ 385static void watch_target(struct xenbus_watch *watch, 386 const char **vec, unsigned int len) 387{ 388 unsigned long long new_target; 389 int err; 390 391 err = xenbus_scanf(XBT_NIL, "memory", "target", "%llu", &new_target); 392 if (err != 1) { 393 /* This is ok (for domain0 at least) - so just return */ 394 return; 395 } 396 397 /* The given memory/target value is in KiB, so it needs converting to 398 * pages. PAGE_SHIFT converts bytes to pages, hence PAGE_SHIFT - 10. 399 */ 400 balloon_set_new_target(new_target >> (PAGE_SHIFT - 10)); 401} 402 403static int balloon_init_watcher(struct notifier_block *notifier, 404 unsigned long event, 405 void *data) 406{ 407 int err; 408 409 err = register_xenbus_watch(&target_watch); 410 if (err) 411 printk(KERN_ERR "Failed to set balloon watcher\n"); 412 413 return NOTIFY_DONE; 414} 415 416static struct notifier_block xenstore_notifier; 417 418static int __init balloon_init(void) 419{ 420 unsigned long pfn; 421 struct page *page; 422 423 if (!xen_pv_domain()) 424 return -ENODEV; 425 426 pr_info("xen_balloon: Initialising balloon driver.\n"); 427 428 balloon_stats.current_pages = min(xen_start_info->nr_pages, max_pfn); 429 totalram_pages = balloon_stats.current_pages; 430 balloon_stats.target_pages = balloon_stats.current_pages; 431 balloon_stats.balloon_low = 0; 432 balloon_stats.balloon_high = 0; 433 balloon_stats.driver_pages = 0UL; 434 balloon_stats.hard_limit = ~0UL; 435 436 init_timer(&balloon_timer); 437 balloon_timer.data = 0; 438 balloon_timer.function = balloon_alarm; 439 440 register_balloon(&balloon_sysdev); 441 442 /* Initialise the balloon with excess memory space. */ 443 for (pfn = xen_start_info->nr_pages; pfn < max_pfn; pfn++) { 444 page = pfn_to_page(pfn); 445 if (!PageReserved(page)) 446 balloon_append(page); 447 } 448 449 target_watch.callback = watch_target; 450 xenstore_notifier.notifier_call = balloon_init_watcher; 451 452 register_xenstore_notifier(&xenstore_notifier); 453 454 return 0; 455} 456 457subsys_initcall(balloon_init); 458 459static void balloon_exit(void) 460{ 461 /* XXX - release balloon here */ 462 return; 463} 464 465module_exit(balloon_exit); 466 467#define BALLOON_SHOW(name, format, args...) \ 468 static ssize_t show_##name(struct sys_device *dev, \ 469 struct sysdev_attribute *attr, \ 470 char *buf) \ 471 { \ 472 return sprintf(buf, format, ##args); \ 473 } \ 474 static SYSDEV_ATTR(name, S_IRUGO, show_##name, NULL) 475 476BALLOON_SHOW(current_kb, "%lu\n", PAGES2KB(balloon_stats.current_pages)); 477BALLOON_SHOW(low_kb, "%lu\n", PAGES2KB(balloon_stats.balloon_low)); 478BALLOON_SHOW(high_kb, "%lu\n", PAGES2KB(balloon_stats.balloon_high)); 479BALLOON_SHOW(hard_limit_kb, 480 (balloon_stats.hard_limit!=~0UL) ? "%lu\n" : "???\n", 481 (balloon_stats.hard_limit!=~0UL) ? PAGES2KB(balloon_stats.hard_limit) : 0); 482BALLOON_SHOW(driver_kb, "%lu\n", PAGES2KB(balloon_stats.driver_pages)); 483 484static ssize_t show_target_kb(struct sys_device *dev, struct sysdev_attribute *attr, 485 char *buf) 486{ 487 return sprintf(buf, "%lu\n", PAGES2KB(balloon_stats.target_pages)); 488} 489 490static ssize_t store_target_kb(struct sys_device *dev, 491 struct sysdev_attribute *attr, 492 const char *buf, 493 size_t count) 494{ 495 char *endchar; 496 unsigned long long target_bytes; 497 498 if (!capable(CAP_SYS_ADMIN)) 499 return -EPERM; 500 501 target_bytes = simple_strtoull(buf, &endchar, 0) * 1024; 502 503 balloon_set_new_target(target_bytes >> PAGE_SHIFT); 504 505 return count; 506} 507 508static SYSDEV_ATTR(target_kb, S_IRUGO | S_IWUSR, 509 show_target_kb, store_target_kb); 510 511 512static ssize_t show_target(struct sys_device *dev, struct sysdev_attribute *attr, 513 char *buf) 514{ 515 return sprintf(buf, "%llu\n", 516 (u64)balloon_stats.target_pages << PAGE_SHIFT); 517} 518 519static ssize_t store_target(struct sys_device *dev, 520 struct sysdev_attribute *attr, 521 const char *buf, 522 size_t count) 523{ 524 char *endchar; 525 unsigned long long target_bytes; 526 527 if (!capable(CAP_SYS_ADMIN)) 528 return -EPERM; 529 530 target_bytes = memparse(buf, &endchar); 531 532 balloon_set_new_target(target_bytes >> PAGE_SHIFT); 533 534 return count; 535} 536 537static SYSDEV_ATTR(target, S_IRUGO | S_IWUSR, 538 show_target, store_target); 539 540 541static struct sysdev_attribute *balloon_attrs[] = { 542 &attr_target_kb, 543 &attr_target, 544}; 545 546static struct attribute *balloon_info_attrs[] = { 547 &attr_current_kb.attr, 548 &attr_low_kb.attr, 549 &attr_high_kb.attr, 550 &attr_hard_limit_kb.attr, 551 &attr_driver_kb.attr, 552 NULL 553}; 554 555static struct attribute_group balloon_info_group = { 556 .name = "info", 557 .attrs = balloon_info_attrs, 558}; 559 560static struct sysdev_class balloon_sysdev_class = { 561 .name = BALLOON_CLASS_NAME, 562}; 563 564static int register_balloon(struct sys_device *sysdev) 565{ 566 int i, error; 567 568 error = sysdev_class_register(&balloon_sysdev_class); 569 if (error) 570 return error; 571 572 sysdev->id = 0; 573 sysdev->cls = &balloon_sysdev_class; 574 575 error = sysdev_register(sysdev); 576 if (error) { 577 sysdev_class_unregister(&balloon_sysdev_class); 578 return error; 579 } 580 581 for (i = 0; i < ARRAY_SIZE(balloon_attrs); i++) { 582 error = sysdev_create_file(sysdev, balloon_attrs[i]); 583 if (error) 584 goto fail; 585 } 586 587 error = sysfs_create_group(&sysdev->kobj, &balloon_info_group); 588 if (error) 589 goto fail; 590 591 return 0; 592 593 fail: 594 while (--i >= 0) 595 sysdev_remove_file(sysdev, balloon_attrs[i]); 596 sysdev_unregister(sysdev); 597 sysdev_class_unregister(&balloon_sysdev_class); 598 return error; 599} 600 601MODULE_LICENSE("GPL"); 602