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#include <linux/gfp.h> 47 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 <asm/xen/hypervisor.h> 55#include <asm/xen/hypercall.h> 56 57#include <xen/xen.h> 58#include <xen/interface/xen.h> 59#include <xen/interface/memory.h> 60#include <xen/xenbus.h> 61#include <xen/features.h> 62#include <xen/page.h> 63 64#define PAGES2KB(_p) ((_p)<<(PAGE_SHIFT-10)) 65 66#define BALLOON_CLASS_NAME "xen_memory" 67 68struct balloon_stats { 69 /* We aim for 'current allocation' == 'target allocation'. */ 70 unsigned long current_pages; 71 unsigned long target_pages; 72 /* 73 * Drivers may alter the memory reservation independently, but they 74 * must inform the balloon driver so we avoid hitting the hard limit. 75 */ 76 unsigned long driver_pages; 77 /* Number of pages in high- and low-memory balloons. */ 78 unsigned long balloon_low; 79 unsigned long balloon_high; 80}; 81 82static DEFINE_MUTEX(balloon_mutex); 83 84static struct sys_device balloon_sysdev; 85 86static int register_balloon(struct sys_device *sysdev); 87 88static struct balloon_stats balloon_stats; 89 90/* We increase/decrease in batches which fit in a page */ 91static unsigned long frame_list[PAGE_SIZE / sizeof(unsigned long)]; 92 93#ifdef CONFIG_HIGHMEM 94#define inc_totalhigh_pages() (totalhigh_pages++) 95#define dec_totalhigh_pages() (totalhigh_pages--) 96#else 97#define inc_totalhigh_pages() do {} while(0) 98#define dec_totalhigh_pages() do {} while(0) 99#endif 100 101/* List of ballooned pages, threaded through the mem_map array. */ 102static LIST_HEAD(ballooned_pages); 103 104/* Main work function, always executed in process context. */ 105static void balloon_process(struct work_struct *work); 106static DECLARE_WORK(balloon_worker, balloon_process); 107static struct timer_list balloon_timer; 108 109/* When ballooning out (allocating memory to return to Xen) we don't really 110 want the kernel to try too hard since that can trigger the oom killer. */ 111#define GFP_BALLOON \ 112 (GFP_HIGHUSER | __GFP_NOWARN | __GFP_NORETRY | __GFP_NOMEMALLOC) 113 114static void scrub_page(struct page *page) 115{ 116#ifdef CONFIG_XEN_SCRUB_PAGES 117 clear_highpage(page); 118#endif 119} 120 121/* balloon_append: add the given page to the balloon. */ 122static void balloon_append(struct page *page) 123{ 124 /* Lowmem is re-populated first, so highmem pages go at list tail. */ 125 if (PageHighMem(page)) { 126 list_add_tail(&page->lru, &ballooned_pages); 127 balloon_stats.balloon_high++; 128 dec_totalhigh_pages(); 129 } else { 130 list_add(&page->lru, &ballooned_pages); 131 balloon_stats.balloon_low++; 132 } 133 134 totalram_pages--; 135} 136 137/* balloon_retrieve: rescue a page from the balloon, if it is not empty. */ 138static struct page *balloon_retrieve(void) 139{ 140 struct page *page; 141 142 if (list_empty(&ballooned_pages)) 143 return NULL; 144 145 page = list_entry(ballooned_pages.next, struct page, lru); 146 list_del(&page->lru); 147 148 if (PageHighMem(page)) { 149 balloon_stats.balloon_high--; 150 inc_totalhigh_pages(); 151 } 152 else 153 balloon_stats.balloon_low--; 154 155 totalram_pages++; 156 157 return page; 158} 159 160static struct page *balloon_first_page(void) 161{ 162 if (list_empty(&ballooned_pages)) 163 return NULL; 164 return list_entry(ballooned_pages.next, struct page, lru); 165} 166 167static struct page *balloon_next_page(struct page *page) 168{ 169 struct list_head *next = page->lru.next; 170 if (next == &ballooned_pages) 171 return NULL; 172 return list_entry(next, struct page, lru); 173} 174 175static void balloon_alarm(unsigned long unused) 176{ 177 schedule_work(&balloon_worker); 178} 179 180static unsigned long current_target(void) 181{ 182 unsigned long target = balloon_stats.target_pages; 183 184 target = min(target, 185 balloon_stats.current_pages + 186 balloon_stats.balloon_low + 187 balloon_stats.balloon_high); 188 189 return target; 190} 191 192static int increase_reservation(unsigned long nr_pages) 193{ 194 unsigned long pfn, i, flags; 195 struct page *page; 196 long rc; 197 struct xen_memory_reservation reservation = { 198 .address_bits = 0, 199 .extent_order = 0, 200 .domid = DOMID_SELF 201 }; 202 203 if (nr_pages > ARRAY_SIZE(frame_list)) 204 nr_pages = ARRAY_SIZE(frame_list); 205 206 spin_lock_irqsave(&xen_reservation_lock, flags); 207 208 page = balloon_first_page(); 209 for (i = 0; i < nr_pages; i++) { 210 BUG_ON(page == NULL); 211 frame_list[i] = page_to_pfn(page); 212 page = balloon_next_page(page); 213 } 214 215 set_xen_guest_handle(reservation.extent_start, frame_list); 216 reservation.nr_extents = nr_pages; 217 rc = HYPERVISOR_memory_op(XENMEM_populate_physmap, &reservation); 218 if (rc < 0) 219 goto out; 220 221 for (i = 0; i < rc; i++) { 222 page = balloon_retrieve(); 223 BUG_ON(page == NULL); 224 225 pfn = page_to_pfn(page); 226 BUG_ON(!xen_feature(XENFEAT_auto_translated_physmap) && 227 phys_to_machine_mapping_valid(pfn)); 228 229 set_phys_to_machine(pfn, frame_list[i]); 230 231 /* Link back into the page tables if not highmem. */ 232 if (pfn < max_low_pfn) { 233 int ret; 234 ret = HYPERVISOR_update_va_mapping( 235 (unsigned long)__va(pfn << PAGE_SHIFT), 236 mfn_pte(frame_list[i], PAGE_KERNEL), 237 0); 238 BUG_ON(ret); 239 } 240 241 /* Relinquish the page back to the allocator. */ 242 ClearPageReserved(page); 243 init_page_count(page); 244 __free_page(page); 245 } 246 247 balloon_stats.current_pages += rc; 248 249 out: 250 spin_unlock_irqrestore(&xen_reservation_lock, flags); 251 252 return rc < 0 ? rc : rc != nr_pages; 253} 254 255static int decrease_reservation(unsigned long nr_pages) 256{ 257 unsigned long pfn, i, flags; 258 struct page *page; 259 int need_sleep = 0; 260 int ret; 261 struct xen_memory_reservation reservation = { 262 .address_bits = 0, 263 .extent_order = 0, 264 .domid = DOMID_SELF 265 }; 266 267 if (nr_pages > ARRAY_SIZE(frame_list)) 268 nr_pages = ARRAY_SIZE(frame_list); 269 270 for (i = 0; i < nr_pages; i++) { 271 if ((page = alloc_page(GFP_BALLOON)) == NULL) { 272 nr_pages = i; 273 need_sleep = 1; 274 break; 275 } 276 277 pfn = page_to_pfn(page); 278 frame_list[i] = pfn_to_mfn(pfn); 279 280 scrub_page(page); 281 282 if (!PageHighMem(page)) { 283 ret = HYPERVISOR_update_va_mapping( 284 (unsigned long)__va(pfn << PAGE_SHIFT), 285 __pte_ma(0), 0); 286 BUG_ON(ret); 287 } 288 289 } 290 291 /* Ensure that ballooned highmem pages don't have kmaps. */ 292 kmap_flush_unused(); 293 flush_tlb_all(); 294 295 spin_lock_irqsave(&xen_reservation_lock, flags); 296 297 /* No more mappings: invalidate P2M and add to balloon. */ 298 for (i = 0; i < nr_pages; i++) { 299 pfn = mfn_to_pfn(frame_list[i]); 300 set_phys_to_machine(pfn, INVALID_P2M_ENTRY); 301 balloon_append(pfn_to_page(pfn)); 302 } 303 304 set_xen_guest_handle(reservation.extent_start, frame_list); 305 reservation.nr_extents = nr_pages; 306 ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation); 307 BUG_ON(ret != nr_pages); 308 309 balloon_stats.current_pages -= nr_pages; 310 311 spin_unlock_irqrestore(&xen_reservation_lock, flags); 312 313 return need_sleep; 314} 315 316/* 317 * We avoid multiple worker processes conflicting via the balloon mutex. 318 * We may of course race updates of the target counts (which are protected 319 * by the balloon lock), or with changes to the Xen hard limit, but we will 320 * recover from these in time. 321 */ 322static void balloon_process(struct work_struct *work) 323{ 324 int need_sleep = 0; 325 long credit; 326 327 mutex_lock(&balloon_mutex); 328 329 do { 330 credit = current_target() - balloon_stats.current_pages; 331 if (credit > 0) 332 need_sleep = (increase_reservation(credit) != 0); 333 if (credit < 0) 334 need_sleep = (decrease_reservation(-credit) != 0); 335 336#ifndef CONFIG_PREEMPT 337 if (need_resched()) 338 schedule(); 339#endif 340 } while ((credit != 0) && !need_sleep); 341 342 /* Schedule more work if there is some still to be done. */ 343 if (current_target() != balloon_stats.current_pages) 344 mod_timer(&balloon_timer, jiffies + HZ); 345 346 mutex_unlock(&balloon_mutex); 347} 348 349/* Resets the Xen limit, sets new target, and kicks off processing. */ 350static void balloon_set_new_target(unsigned long target) 351{ 352 /* No need for lock. Not read-modify-write updates. */ 353 balloon_stats.target_pages = target; 354 schedule_work(&balloon_worker); 355} 356 357static struct xenbus_watch target_watch = 358{ 359 .node = "memory/target" 360}; 361 362/* React to a change in the target key */ 363static void watch_target(struct xenbus_watch *watch, 364 const char **vec, unsigned int len) 365{ 366 unsigned long long new_target; 367 int err; 368 369 err = xenbus_scanf(XBT_NIL, "memory", "target", "%llu", &new_target); 370 if (err != 1) { 371 /* This is ok (for domain0 at least) - so just return */ 372 return; 373 } 374 375 /* The given memory/target value is in KiB, so it needs converting to 376 * pages. PAGE_SHIFT converts bytes to pages, hence PAGE_SHIFT - 10. 377 */ 378 balloon_set_new_target(new_target >> (PAGE_SHIFT - 10)); 379} 380 381static int balloon_init_watcher(struct notifier_block *notifier, 382 unsigned long event, 383 void *data) 384{ 385 int err; 386 387 err = register_xenbus_watch(&target_watch); 388 if (err) 389 printk(KERN_ERR "Failed to set balloon watcher\n"); 390 391 return NOTIFY_DONE; 392} 393 394static struct notifier_block xenstore_notifier; 395 396static int __init balloon_init(void) 397{ 398 unsigned long pfn; 399 struct page *page; 400 401 if (!xen_pv_domain()) 402 return -ENODEV; 403 404 pr_info("xen_balloon: Initialising balloon driver.\n"); 405 406 balloon_stats.current_pages = min(xen_start_info->nr_pages, max_pfn); 407 balloon_stats.target_pages = balloon_stats.current_pages; 408 balloon_stats.balloon_low = 0; 409 balloon_stats.balloon_high = 0; 410 balloon_stats.driver_pages = 0UL; 411 412 init_timer(&balloon_timer); 413 balloon_timer.data = 0; 414 balloon_timer.function = balloon_alarm; 415 416 register_balloon(&balloon_sysdev); 417 418 /* Initialise the balloon with excess memory space. */ 419 for (pfn = xen_start_info->nr_pages; pfn < max_pfn; pfn++) { 420 page = pfn_to_page(pfn); 421 if (!PageReserved(page)) 422 balloon_append(page); 423 } 424 425 target_watch.callback = watch_target; 426 xenstore_notifier.notifier_call = balloon_init_watcher; 427 428 register_xenstore_notifier(&xenstore_notifier); 429 430 return 0; 431} 432 433subsys_initcall(balloon_init); 434 435static void balloon_exit(void) 436{ 437 return; 438} 439 440module_exit(balloon_exit); 441 442#define BALLOON_SHOW(name, format, args...) \ 443 static ssize_t show_##name(struct sys_device *dev, \ 444 struct sysdev_attribute *attr, \ 445 char *buf) \ 446 { \ 447 return sprintf(buf, format, ##args); \ 448 } \ 449 static SYSDEV_ATTR(name, S_IRUGO, show_##name, NULL) 450 451BALLOON_SHOW(current_kb, "%lu\n", PAGES2KB(balloon_stats.current_pages)); 452BALLOON_SHOW(low_kb, "%lu\n", PAGES2KB(balloon_stats.balloon_low)); 453BALLOON_SHOW(high_kb, "%lu\n", PAGES2KB(balloon_stats.balloon_high)); 454BALLOON_SHOW(driver_kb, "%lu\n", PAGES2KB(balloon_stats.driver_pages)); 455 456static ssize_t show_target_kb(struct sys_device *dev, struct sysdev_attribute *attr, 457 char *buf) 458{ 459 return sprintf(buf, "%lu\n", PAGES2KB(balloon_stats.target_pages)); 460} 461 462static ssize_t store_target_kb(struct sys_device *dev, 463 struct sysdev_attribute *attr, 464 const char *buf, 465 size_t count) 466{ 467 char *endchar; 468 unsigned long long target_bytes; 469 470 if (!capable(CAP_SYS_ADMIN)) 471 return -EPERM; 472 473 target_bytes = simple_strtoull(buf, &endchar, 0) * 1024; 474 475 balloon_set_new_target(target_bytes >> PAGE_SHIFT); 476 477 return count; 478} 479 480static SYSDEV_ATTR(target_kb, S_IRUGO | S_IWUSR, 481 show_target_kb, store_target_kb); 482 483 484static ssize_t show_target(struct sys_device *dev, struct sysdev_attribute *attr, 485 char *buf) 486{ 487 return sprintf(buf, "%llu\n", 488 (unsigned long long)balloon_stats.target_pages 489 << PAGE_SHIFT); 490} 491 492static ssize_t store_target(struct sys_device *dev, 493 struct sysdev_attribute *attr, 494 const char *buf, 495 size_t count) 496{ 497 char *endchar; 498 unsigned long long target_bytes; 499 500 if (!capable(CAP_SYS_ADMIN)) 501 return -EPERM; 502 503 target_bytes = memparse(buf, &endchar); 504 505 balloon_set_new_target(target_bytes >> PAGE_SHIFT); 506 507 return count; 508} 509 510static SYSDEV_ATTR(target, S_IRUGO | S_IWUSR, 511 show_target, store_target); 512 513 514static struct sysdev_attribute *balloon_attrs[] = { 515 &attr_target_kb, 516 &attr_target, 517}; 518 519static struct attribute *balloon_info_attrs[] = { 520 &attr_current_kb.attr, 521 &attr_low_kb.attr, 522 &attr_high_kb.attr, 523 &attr_driver_kb.attr, 524 NULL 525}; 526 527static struct attribute_group balloon_info_group = { 528 .name = "info", 529 .attrs = balloon_info_attrs, 530}; 531 532static struct sysdev_class balloon_sysdev_class = { 533 .name = BALLOON_CLASS_NAME, 534}; 535 536static int register_balloon(struct sys_device *sysdev) 537{ 538 int i, error; 539 540 error = sysdev_class_register(&balloon_sysdev_class); 541 if (error) 542 return error; 543 544 sysdev->id = 0; 545 sysdev->cls = &balloon_sysdev_class; 546 547 error = sysdev_register(sysdev); 548 if (error) { 549 sysdev_class_unregister(&balloon_sysdev_class); 550 return error; 551 } 552 553 for (i = 0; i < ARRAY_SIZE(balloon_attrs); i++) { 554 error = sysdev_create_file(sysdev, balloon_attrs[i]); 555 if (error) 556 goto fail; 557 } 558 559 error = sysfs_create_group(&sysdev->kobj, &balloon_info_group); 560 if (error) 561 goto fail; 562 563 return 0; 564 565 fail: 566 while (--i >= 0) 567 sysdev_remove_file(sysdev, balloon_attrs[i]); 568 sysdev_unregister(sysdev); 569 sysdev_class_unregister(&balloon_sysdev_class); 570 return error; 571} 572 573MODULE_LICENSE("GPL"); 574