1 /******************************************************************************
2 * Xen balloon driver - enables returning/claiming memory to/from Xen.
4 * Copyright (c) 2003, B Dragovic
5 * Copyright (c) 2003-2004, M Williamson, K Fraser
6 * Copyright (c) 2005 Dan M. Smith, IBM Corporation
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License version 2
10 * as published by the Free Software Foundation; or, when distributed
11 * separately from the Linux kernel or incorporated into other
12 * software packages, subject to the following license:
14 * Permission is hereby granted, free of charge, to any person obtaining a copy
15 * of this source file (the "Software"), to deal in the Software without
16 * restriction, including without limitation the rights to use, copy, modify,
17 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
18 * and to permit persons to whom the Software is furnished to do so, subject to
19 * the following conditions:
21 * The above copyright notice and this permission notice shall be included in
22 * all copies or substantial portions of the Software.
24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
25 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
26 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
27 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
28 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
29 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
33 #include <linux/kernel.h>
34 #include <linux/sched.h>
35 #include <linux/errno.h>
37 #include <linux/bootmem.h>
38 #include <linux/pagemap.h>
39 #include <linux/highmem.h>
40 #include <linux/mutex.h>
41 #include <linux/list.h>
42 #include <linux/gfp.h>
45 #include <asm/pgalloc.h>
46 #include <asm/pgtable.h>
50 #include <asm/xen/hypervisor.h>
51 #include <asm/xen/hypercall.h>
54 #include <xen/interface/xen.h>
55 #include <xen/interface/memory.h>
56 #include <xen/balloon.h>
57 #include <xen/features.h>
61 * balloon_process() state:
63 * BP_DONE: done or nothing to do,
64 * BP_EAGAIN: error, go to sleep,
65 * BP_ECANCELED: error, balloon operation canceled.
75 static DEFINE_MUTEX(balloon_mutex);
77 struct balloon_stats balloon_stats;
78 EXPORT_SYMBOL_GPL(balloon_stats);
80 /* We increase/decrease in batches which fit in a page */
81 static unsigned long frame_list[PAGE_SIZE / sizeof(unsigned long)];
84 #define inc_totalhigh_pages() (totalhigh_pages++)
85 #define dec_totalhigh_pages() (totalhigh_pages--)
87 #define inc_totalhigh_pages() do {} while(0)
88 #define dec_totalhigh_pages() do {} while(0)
91 /* List of ballooned pages, threaded through the mem_map array. */
92 static LIST_HEAD(ballooned_pages);
94 /* Main work function, always executed in process context. */
95 static void balloon_process(struct work_struct *work);
96 static DECLARE_DELAYED_WORK(balloon_worker, balloon_process);
98 /* When ballooning out (allocating memory to return to Xen) we don't really
99 want the kernel to try too hard since that can trigger the oom killer. */
100 #define GFP_BALLOON \
101 (GFP_HIGHUSER | __GFP_NOWARN | __GFP_NORETRY | __GFP_NOMEMALLOC)
103 static void scrub_page(struct page *page)
105 #ifdef CONFIG_XEN_SCRUB_PAGES
106 clear_highpage(page);
110 /* balloon_append: add the given page to the balloon. */
111 static void __balloon_append(struct page *page)
113 /* Lowmem is re-populated first, so highmem pages go at list tail. */
114 if (PageHighMem(page)) {
115 list_add_tail(&page->lru, &ballooned_pages);
116 balloon_stats.balloon_high++;
117 dec_totalhigh_pages();
119 list_add(&page->lru, &ballooned_pages);
120 balloon_stats.balloon_low++;
124 static void balloon_append(struct page *page)
126 __balloon_append(page);
130 /* balloon_retrieve: rescue a page from the balloon, if it is not empty. */
131 static struct page *balloon_retrieve(bool prefer_highmem)
135 if (list_empty(&ballooned_pages))
139 page = list_entry(ballooned_pages.prev, struct page, lru);
141 page = list_entry(ballooned_pages.next, struct page, lru);
142 list_del(&page->lru);
144 if (PageHighMem(page)) {
145 balloon_stats.balloon_high--;
146 inc_totalhigh_pages();
149 balloon_stats.balloon_low--;
156 static struct page *balloon_first_page(void)
158 if (list_empty(&ballooned_pages))
160 return list_entry(ballooned_pages.next, struct page, lru);
163 static struct page *balloon_next_page(struct page *page)
165 struct list_head *next = page->lru.next;
166 if (next == &ballooned_pages)
168 return list_entry(next, struct page, lru);
171 static enum bp_state update_schedule(enum bp_state state)
173 if (state == BP_DONE) {
174 balloon_stats.schedule_delay = 1;
175 balloon_stats.retry_count = 1;
179 ++balloon_stats.retry_count;
181 if (balloon_stats.max_retry_count != RETRY_UNLIMITED &&
182 balloon_stats.retry_count > balloon_stats.max_retry_count) {
183 balloon_stats.schedule_delay = 1;
184 balloon_stats.retry_count = 1;
188 balloon_stats.schedule_delay <<= 1;
190 if (balloon_stats.schedule_delay > balloon_stats.max_schedule_delay)
191 balloon_stats.schedule_delay = balloon_stats.max_schedule_delay;
196 static unsigned long current_target(void)
198 unsigned long target = balloon_stats.target_pages;
201 balloon_stats.current_pages +
202 balloon_stats.balloon_low +
203 balloon_stats.balloon_high);
208 static enum bp_state increase_reservation(unsigned long nr_pages)
211 unsigned long pfn, i;
213 struct xen_memory_reservation reservation = {
219 if (nr_pages > ARRAY_SIZE(frame_list))
220 nr_pages = ARRAY_SIZE(frame_list);
222 page = balloon_first_page();
223 for (i = 0; i < nr_pages; i++) {
228 frame_list[i] = page_to_pfn(page);
229 page = balloon_next_page(page);
232 set_xen_guest_handle(reservation.extent_start, frame_list);
233 reservation.nr_extents = nr_pages;
234 rc = HYPERVISOR_memory_op(XENMEM_populate_physmap, &reservation);
238 for (i = 0; i < rc; i++) {
239 page = balloon_retrieve(false);
240 BUG_ON(page == NULL);
242 pfn = page_to_pfn(page);
243 BUG_ON(!xen_feature(XENFEAT_auto_translated_physmap) &&
244 phys_to_machine_mapping_valid(pfn));
246 set_phys_to_machine(pfn, frame_list[i]);
248 /* Link back into the page tables if not highmem. */
249 if (!xen_hvm_domain() && pfn < max_low_pfn) {
251 ret = HYPERVISOR_update_va_mapping(
252 (unsigned long)__va(pfn << PAGE_SHIFT),
253 mfn_pte(frame_list[i], PAGE_KERNEL),
258 /* Relinquish the page back to the allocator. */
259 ClearPageReserved(page);
260 init_page_count(page);
264 balloon_stats.current_pages += rc;
269 static enum bp_state decrease_reservation(unsigned long nr_pages, gfp_t gfp)
271 enum bp_state state = BP_DONE;
272 unsigned long pfn, i;
275 struct xen_memory_reservation reservation = {
281 if (nr_pages > ARRAY_SIZE(frame_list))
282 nr_pages = ARRAY_SIZE(frame_list);
284 for (i = 0; i < nr_pages; i++) {
285 if ((page = alloc_page(gfp)) == NULL) {
291 pfn = page_to_pfn(page);
292 frame_list[i] = pfn_to_mfn(pfn);
296 if (!xen_hvm_domain() && !PageHighMem(page)) {
297 ret = HYPERVISOR_update_va_mapping(
298 (unsigned long)__va(pfn << PAGE_SHIFT),
305 /* Ensure that ballooned highmem pages don't have kmaps. */
309 /* No more mappings: invalidate P2M and add to balloon. */
310 for (i = 0; i < nr_pages; i++) {
311 pfn = mfn_to_pfn(frame_list[i]);
312 __set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
313 balloon_append(pfn_to_page(pfn));
316 set_xen_guest_handle(reservation.extent_start, frame_list);
317 reservation.nr_extents = nr_pages;
318 ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
319 BUG_ON(ret != nr_pages);
321 balloon_stats.current_pages -= nr_pages;
327 * We avoid multiple worker processes conflicting via the balloon mutex.
328 * We may of course race updates of the target counts (which are protected
329 * by the balloon lock), or with changes to the Xen hard limit, but we will
330 * recover from these in time.
332 static void balloon_process(struct work_struct *work)
334 enum bp_state state = BP_DONE;
337 mutex_lock(&balloon_mutex);
340 credit = current_target() - balloon_stats.current_pages;
343 state = increase_reservation(credit);
346 state = decrease_reservation(-credit, GFP_BALLOON);
348 state = update_schedule(state);
350 #ifndef CONFIG_PREEMPT
354 } while (credit && state == BP_DONE);
356 /* Schedule more work if there is some still to be done. */
357 if (state == BP_EAGAIN)
358 schedule_delayed_work(&balloon_worker, balloon_stats.schedule_delay * HZ);
360 mutex_unlock(&balloon_mutex);
363 /* Resets the Xen limit, sets new target, and kicks off processing. */
364 void balloon_set_new_target(unsigned long target)
366 /* No need for lock. Not read-modify-write updates. */
367 balloon_stats.target_pages = target;
368 schedule_delayed_work(&balloon_worker, 0);
370 EXPORT_SYMBOL_GPL(balloon_set_new_target);
373 * alloc_xenballooned_pages - get pages that have been ballooned out
374 * @nr_pages: Number of pages to get
375 * @pages: pages returned
376 * @return 0 on success, error otherwise
378 int alloc_xenballooned_pages(int nr_pages, struct page** pages)
382 mutex_lock(&balloon_mutex);
383 while (pgno < nr_pages) {
384 page = balloon_retrieve(true);
386 pages[pgno++] = page;
389 st = decrease_reservation(nr_pages - pgno, GFP_HIGHUSER);
394 mutex_unlock(&balloon_mutex);
398 balloon_append(pages[--pgno]);
399 /* Free the memory back to the kernel soon */
400 schedule_delayed_work(&balloon_worker, 0);
401 mutex_unlock(&balloon_mutex);
404 EXPORT_SYMBOL(alloc_xenballooned_pages);
407 * free_xenballooned_pages - return pages retrieved with get_ballooned_pages
408 * @nr_pages: Number of pages
409 * @pages: pages to return
411 void free_xenballooned_pages(int nr_pages, struct page** pages)
415 mutex_lock(&balloon_mutex);
417 for (i = 0; i < nr_pages; i++) {
419 balloon_append(pages[i]);
422 /* The balloon may be too large now. Shrink it if needed. */
423 if (current_target() != balloon_stats.current_pages)
424 schedule_delayed_work(&balloon_worker, 0);
426 mutex_unlock(&balloon_mutex);
428 EXPORT_SYMBOL(free_xenballooned_pages);
430 static int __init balloon_init(void)
432 unsigned long pfn, nr_pages, extra_pfn_end;
438 pr_info("xen/balloon: Initialising balloon driver.\n");
441 nr_pages = xen_start_info->nr_pages;
444 balloon_stats.current_pages = min(nr_pages, max_pfn);
445 balloon_stats.target_pages = balloon_stats.current_pages;
446 balloon_stats.balloon_low = 0;
447 balloon_stats.balloon_high = 0;
449 balloon_stats.schedule_delay = 1;
450 balloon_stats.max_schedule_delay = 32;
451 balloon_stats.retry_count = 1;
452 balloon_stats.max_retry_count = RETRY_UNLIMITED;
455 * Initialise the balloon with excess memory space. We need
456 * to make sure we don't add memory which doesn't exist or
457 * logically exist. The E820 map can be trimmed to be smaller
458 * than the amount of physical memory due to the mem= command
459 * line parameter. And if this is a 32-bit non-HIGHMEM kernel
460 * on a system with memory which requires highmem to access,
461 * don't try to use it.
463 extra_pfn_end = min(min(max_pfn, e820_end_of_ram_pfn()),
464 (unsigned long)PFN_DOWN(xen_extra_mem_start + xen_extra_mem_size));
465 for (pfn = PFN_UP(xen_extra_mem_start);
468 page = pfn_to_page(pfn);
469 /* totalram_pages doesn't include the boot-time
470 balloon extension, so don't subtract from it. */
471 __balloon_append(page);
477 subsys_initcall(balloon_init);
479 MODULE_LICENSE("GPL");