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
7 * Copyright (c) 2010 Daniel Kiper
9 * Memory hotplug support was written by Daniel Kiper. Work on
10 * it was sponsored by Google under Google Summer of Code 2010
11 * program. Jeremy Fitzhardinge from Citrix was the mentor for
14 * This program is free software; you can redistribute it and/or
15 * modify it under the terms of the GNU General Public License version 2
16 * as published by the Free Software Foundation; or, when distributed
17 * separately from the Linux kernel or incorporated into other
18 * software packages, subject to the following license:
20 * Permission is hereby granted, free of charge, to any person obtaining a copy
21 * of this source file (the "Software"), to deal in the Software without
22 * restriction, including without limitation the rights to use, copy, modify,
23 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
24 * and to permit persons to whom the Software is furnished to do so, subject to
25 * the following conditions:
27 * The above copyright notice and this permission notice shall be included in
28 * all copies or substantial portions of the Software.
30 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
31 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
32 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
33 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
34 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
35 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
39 #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
41 #include <linux/cpu.h>
42 #include <linux/kernel.h>
43 #include <linux/sched.h>
44 #include <linux/cred.h>
45 #include <linux/errno.h>
46 #include <linux/freezer.h>
47 #include <linux/kthread.h>
49 #include <linux/memblock.h>
50 #include <linux/pagemap.h>
51 #include <linux/highmem.h>
52 #include <linux/mutex.h>
53 #include <linux/list.h>
54 #include <linux/gfp.h>
55 #include <linux/notifier.h>
56 #include <linux/memory.h>
57 #include <linux/memory_hotplug.h>
58 #include <linux/percpu-defs.h>
59 #include <linux/slab.h>
60 #include <linux/sysctl.h>
61 #include <linux/moduleparam.h>
62 #include <linux/jiffies.h>
67 #include <asm/xen/hypervisor.h>
68 #include <asm/xen/hypercall.h>
71 #include <xen/interface/xen.h>
72 #include <xen/interface/memory.h>
73 #include <xen/balloon.h>
74 #include <xen/features.h>
76 #include <xen/mem-reservation.h>
78 #undef MODULE_PARAM_PREFIX
79 #define MODULE_PARAM_PREFIX "xen."
81 static uint __read_mostly balloon_boot_timeout = 180;
82 module_param(balloon_boot_timeout, uint, 0444);
84 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
85 static int xen_hotplug_unpopulated;
87 static struct ctl_table balloon_table[] = {
89 .procname = "hotplug_unpopulated",
90 .data = &xen_hotplug_unpopulated,
91 .maxlen = sizeof(int),
93 .proc_handler = proc_dointvec_minmax,
94 .extra1 = SYSCTL_ZERO,
100 #define xen_hotplug_unpopulated 0
104 * Use one extent per PAGE_SIZE to avoid to break down the page into
107 #define EXTENT_ORDER (fls(XEN_PFN_PER_PAGE) - 1)
110 * balloon_thread() state:
112 * BP_DONE: done or nothing to do,
113 * BP_WAIT: wait to be rescheduled,
114 * BP_EAGAIN: error, go to sleep,
115 * BP_ECANCELED: error, balloon operation canceled.
118 static enum bp_state {
123 } balloon_state = BP_DONE;
125 /* Main waiting point for xen-balloon thread. */
126 static DECLARE_WAIT_QUEUE_HEAD(balloon_thread_wq);
128 static DEFINE_MUTEX(balloon_mutex);
130 struct balloon_stats balloon_stats;
131 EXPORT_SYMBOL_GPL(balloon_stats);
133 /* We increase/decrease in batches which fit in a page */
134 static xen_pfn_t frame_list[PAGE_SIZE / sizeof(xen_pfn_t)];
137 /* List of ballooned pages, threaded through the mem_map array. */
138 static LIST_HEAD(ballooned_pages);
139 static DECLARE_WAIT_QUEUE_HEAD(balloon_wq);
141 /* When ballooning out (allocating memory to return to Xen) we don't really
142 want the kernel to try too hard since that can trigger the oom killer. */
143 #define GFP_BALLOON \
144 (GFP_HIGHUSER | __GFP_NOWARN | __GFP_NORETRY | __GFP_NOMEMALLOC)
146 /* balloon_append: add the given page to the balloon. */
147 static void balloon_append(struct page *page)
149 __SetPageOffline(page);
151 /* Lowmem is re-populated first, so highmem pages go at list tail. */
152 if (PageHighMem(page)) {
153 list_add_tail(&page->lru, &ballooned_pages);
154 balloon_stats.balloon_high++;
156 list_add(&page->lru, &ballooned_pages);
157 balloon_stats.balloon_low++;
159 wake_up(&balloon_wq);
162 /* balloon_retrieve: rescue a page from the balloon, if it is not empty. */
163 static struct page *balloon_retrieve(bool require_lowmem)
167 if (list_empty(&ballooned_pages))
170 page = list_entry(ballooned_pages.next, struct page, lru);
171 if (require_lowmem && PageHighMem(page))
173 list_del(&page->lru);
175 if (PageHighMem(page))
176 balloon_stats.balloon_high--;
178 balloon_stats.balloon_low--;
180 __ClearPageOffline(page);
184 static struct page *balloon_next_page(struct page *page)
186 struct list_head *next = page->lru.next;
187 if (next == &ballooned_pages)
189 return list_entry(next, struct page, lru);
192 static void update_schedule(void)
194 if (balloon_state == BP_WAIT || balloon_state == BP_ECANCELED)
197 if (balloon_state == BP_DONE) {
198 balloon_stats.schedule_delay = 1;
199 balloon_stats.retry_count = 1;
203 ++balloon_stats.retry_count;
205 if (balloon_stats.max_retry_count != RETRY_UNLIMITED &&
206 balloon_stats.retry_count > balloon_stats.max_retry_count) {
207 balloon_stats.schedule_delay = 1;
208 balloon_stats.retry_count = 1;
209 balloon_state = BP_ECANCELED;
213 balloon_stats.schedule_delay <<= 1;
215 if (balloon_stats.schedule_delay > balloon_stats.max_schedule_delay)
216 balloon_stats.schedule_delay = balloon_stats.max_schedule_delay;
218 balloon_state = BP_EAGAIN;
221 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
222 static void release_memory_resource(struct resource *resource)
228 * No need to reset region to identity mapped since we now
229 * know that no I/O can be in this region
231 release_resource(resource);
235 static struct resource *additional_memory_resource(phys_addr_t size)
237 struct resource *res;
240 res = kzalloc(sizeof(*res), GFP_KERNEL);
244 res->name = "System RAM";
245 res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
247 ret = allocate_resource(&iomem_resource, res,
249 PAGES_PER_SECTION * PAGE_SIZE, NULL, NULL);
251 pr_err("Cannot allocate new System RAM resource\n");
259 static enum bp_state reserve_additional_memory(void)
262 struct resource *resource;
264 unsigned long balloon_hotplug;
266 credit = balloon_stats.target_pages + balloon_stats.target_unpopulated
267 - balloon_stats.total_pages;
270 * Already hotplugged enough pages? Wait for them to be
276 balloon_hotplug = round_up(credit, PAGES_PER_SECTION);
278 resource = additional_memory_resource(balloon_hotplug * PAGE_SIZE);
282 nid = memory_add_physaddr_to_nid(resource->start);
284 #ifdef CONFIG_XEN_HAVE_PVMMU
286 * We don't support PV MMU when Linux and Xen is using
287 * different page granularity.
289 BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
292 * add_memory() will build page tables for the new memory so
293 * the p2m must contain invalid entries so the correct
294 * non-present PTEs will be written.
296 * If a failure occurs, the original (identity) p2m entries
297 * are not restored since this region is now known not to
298 * conflict with any devices.
300 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
301 unsigned long pfn, i;
303 pfn = PFN_DOWN(resource->start);
304 for (i = 0; i < balloon_hotplug; i++) {
305 if (!set_phys_to_machine(pfn + i, INVALID_P2M_ENTRY)) {
306 pr_warn("set_phys_to_machine() failed, no memory added\n");
314 * add_memory_resource() will call online_pages() which in its turn
315 * will call xen_online_page() callback causing deadlock if we don't
316 * release balloon_mutex here. Unlocking here is safe because the
317 * callers drop the mutex before trying again.
319 mutex_unlock(&balloon_mutex);
320 /* add_memory_resource() requires the device_hotplug lock */
321 lock_device_hotplug();
322 rc = add_memory_resource(nid, resource, MHP_MERGE_RESOURCE);
323 unlock_device_hotplug();
324 mutex_lock(&balloon_mutex);
327 pr_warn("Cannot add additional memory (%i)\n", rc);
331 balloon_stats.total_pages += balloon_hotplug;
335 release_memory_resource(resource);
339 static void xen_online_page(struct page *page, unsigned int order)
341 unsigned long i, size = (1 << order);
342 unsigned long start_pfn = page_to_pfn(page);
345 pr_debug("Online %lu pages starting at pfn 0x%lx\n", size, start_pfn);
346 mutex_lock(&balloon_mutex);
347 for (i = 0; i < size; i++) {
348 p = pfn_to_page(start_pfn + i);
351 mutex_unlock(&balloon_mutex);
354 static int xen_memory_notifier(struct notifier_block *nb, unsigned long val, void *v)
356 if (val == MEM_ONLINE)
357 wake_up(&balloon_thread_wq);
362 static struct notifier_block xen_memory_nb = {
363 .notifier_call = xen_memory_notifier,
367 static enum bp_state reserve_additional_memory(void)
369 balloon_stats.target_pages = balloon_stats.current_pages +
370 balloon_stats.target_unpopulated;
373 #endif /* CONFIG_XEN_BALLOON_MEMORY_HOTPLUG */
375 static long current_credit(void)
377 return balloon_stats.target_pages - balloon_stats.current_pages;
380 static bool balloon_is_inflated(void)
382 return balloon_stats.balloon_low || balloon_stats.balloon_high;
385 static enum bp_state increase_reservation(unsigned long nr_pages)
391 if (nr_pages > ARRAY_SIZE(frame_list))
392 nr_pages = ARRAY_SIZE(frame_list);
394 page = list_first_entry_or_null(&ballooned_pages, struct page, lru);
395 for (i = 0; i < nr_pages; i++) {
401 frame_list[i] = page_to_xen_pfn(page);
402 page = balloon_next_page(page);
405 rc = xenmem_reservation_increase(nr_pages, frame_list);
409 for (i = 0; i < rc; i++) {
410 page = balloon_retrieve(false);
411 BUG_ON(page == NULL);
413 xenmem_reservation_va_mapping_update(1, &page, &frame_list[i]);
415 /* Relinquish the page back to the allocator. */
416 free_reserved_page(page);
419 balloon_stats.current_pages += rc;
424 static enum bp_state decrease_reservation(unsigned long nr_pages, gfp_t gfp)
426 enum bp_state state = BP_DONE;
428 struct page *page, *tmp;
432 if (nr_pages > ARRAY_SIZE(frame_list))
433 nr_pages = ARRAY_SIZE(frame_list);
435 for (i = 0; i < nr_pages; i++) {
436 page = alloc_page(gfp);
442 adjust_managed_page_count(page, -1);
443 xenmem_reservation_scrub_page(page);
444 list_add(&page->lru, &pages);
448 * Ensure that ballooned highmem pages don't have kmaps.
450 * Do this before changing the p2m as kmap_flush_unused()
451 * reads PTEs to obtain pages (and hence needs the original
457 * Setup the frame, update direct mapping, invalidate P2M,
458 * and add to balloon.
461 list_for_each_entry_safe(page, tmp, &pages, lru) {
462 frame_list[i++] = xen_page_to_gfn(page);
464 xenmem_reservation_va_mapping_reset(1, &page);
466 list_del(&page->lru);
468 balloon_append(page);
473 ret = xenmem_reservation_decrease(nr_pages, frame_list);
474 BUG_ON(ret != nr_pages);
476 balloon_stats.current_pages -= nr_pages;
482 * Stop waiting if either state is BP_DONE and ballooning action is
483 * needed, or if the credit has changed while state is not BP_DONE.
485 static bool balloon_thread_cond(long credit)
487 if (balloon_state == BP_DONE)
490 return current_credit() != credit || kthread_should_stop();
494 * As this is a kthread it is guaranteed to run as a single instance only.
495 * We may of course race updates of the target counts (which are protected
496 * by the balloon lock), or with changes to the Xen hard limit, but we will
497 * recover from these in time.
499 static int balloon_thread(void *unused)
502 unsigned long timeout;
506 switch (balloon_state) {
512 timeout = balloon_stats.schedule_delay * HZ;
519 credit = current_credit();
521 wait_event_freezable_timeout(balloon_thread_wq,
522 balloon_thread_cond(credit), timeout);
524 if (kthread_should_stop())
527 mutex_lock(&balloon_mutex);
529 credit = current_credit();
532 if (balloon_is_inflated())
533 balloon_state = increase_reservation(credit);
535 balloon_state = reserve_additional_memory();
541 n_pages = min(-credit, si_mem_available());
542 balloon_state = decrease_reservation(n_pages,
544 if (balloon_state == BP_DONE && n_pages != -credit &&
545 n_pages < totalreserve_pages)
546 balloon_state = BP_EAGAIN;
551 mutex_unlock(&balloon_mutex);
557 /* Resets the Xen limit, sets new target, and kicks off processing. */
558 void balloon_set_new_target(unsigned long target)
560 /* No need for lock. Not read-modify-write updates. */
561 balloon_stats.target_pages = target;
562 wake_up(&balloon_thread_wq);
564 EXPORT_SYMBOL_GPL(balloon_set_new_target);
566 static int add_ballooned_pages(unsigned int nr_pages)
570 if (xen_hotplug_unpopulated) {
571 st = reserve_additional_memory();
572 if (st != BP_ECANCELED) {
575 mutex_unlock(&balloon_mutex);
576 rc = wait_event_interruptible(balloon_wq,
577 !list_empty(&ballooned_pages));
578 mutex_lock(&balloon_mutex);
579 return rc ? -ENOMEM : 0;
583 if (si_mem_available() < nr_pages)
586 st = decrease_reservation(nr_pages, GFP_USER);
594 * xen_alloc_ballooned_pages - get pages that have been ballooned out
595 * @nr_pages: Number of pages to get
596 * @pages: pages returned
597 * @return 0 on success, error otherwise
599 int xen_alloc_ballooned_pages(unsigned int nr_pages, struct page **pages)
601 unsigned int pgno = 0;
605 mutex_lock(&balloon_mutex);
607 balloon_stats.target_unpopulated += nr_pages;
609 while (pgno < nr_pages) {
610 page = balloon_retrieve(true);
612 pages[pgno++] = page;
613 #ifdef CONFIG_XEN_HAVE_PVMMU
615 * We don't support PV MMU when Linux and Xen is using
616 * different page granularity.
618 BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
620 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
621 ret = xen_alloc_p2m_entry(page_to_pfn(page));
627 ret = add_ballooned_pages(nr_pages - pgno);
632 mutex_unlock(&balloon_mutex);
635 mutex_unlock(&balloon_mutex);
636 xen_free_ballooned_pages(pgno, pages);
638 * NB: xen_free_ballooned_pages will only subtract pgno pages, but since
639 * target_unpopulated is incremented with nr_pages at the start we need
640 * to remove the remaining ones also, or accounting will be screwed.
642 balloon_stats.target_unpopulated -= nr_pages - pgno;
645 EXPORT_SYMBOL(xen_alloc_ballooned_pages);
648 * xen_free_ballooned_pages - return pages retrieved with get_ballooned_pages
649 * @nr_pages: Number of pages
650 * @pages: pages to return
652 void xen_free_ballooned_pages(unsigned int nr_pages, struct page **pages)
656 mutex_lock(&balloon_mutex);
658 for (i = 0; i < nr_pages; i++) {
660 balloon_append(pages[i]);
663 balloon_stats.target_unpopulated -= nr_pages;
665 /* The balloon may be too large now. Shrink it if needed. */
666 if (current_credit())
667 wake_up(&balloon_thread_wq);
669 mutex_unlock(&balloon_mutex);
671 EXPORT_SYMBOL(xen_free_ballooned_pages);
673 static void __init balloon_add_regions(void)
675 #if defined(CONFIG_XEN_PV)
676 unsigned long start_pfn, pages;
677 unsigned long pfn, extra_pfn_end;
680 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
681 pages = xen_extra_mem[i].n_pfns;
685 start_pfn = xen_extra_mem[i].start_pfn;
688 * If the amount of usable memory has been limited (e.g., with
689 * the 'mem' command line parameter), don't add pages beyond
692 extra_pfn_end = min(max_pfn, start_pfn + pages);
694 for (pfn = start_pfn; pfn < extra_pfn_end; pfn++)
695 balloon_append(pfn_to_page(pfn));
697 balloon_stats.total_pages += extra_pfn_end - start_pfn;
702 static int __init balloon_init(void)
704 struct task_struct *task;
709 pr_info("Initialising balloon driver\n");
712 balloon_stats.current_pages = xen_pv_domain()
713 ? min(xen_start_info->nr_pages - xen_released_pages, max_pfn)
714 : get_num_physpages();
716 balloon_stats.current_pages = get_num_physpages();
718 balloon_stats.target_pages = balloon_stats.current_pages;
719 balloon_stats.balloon_low = 0;
720 balloon_stats.balloon_high = 0;
721 balloon_stats.total_pages = balloon_stats.current_pages;
723 balloon_stats.schedule_delay = 1;
724 balloon_stats.max_schedule_delay = 32;
725 balloon_stats.retry_count = 1;
726 balloon_stats.max_retry_count = 4;
728 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
729 set_online_page_callback(&xen_online_page);
730 register_memory_notifier(&xen_memory_nb);
731 register_sysctl_init("xen/balloon", balloon_table);
734 balloon_add_regions();
736 task = kthread_run(balloon_thread, NULL, "xen-balloon");
738 pr_err("xen-balloon thread could not be started, ballooning will not work!\n");
739 return PTR_ERR(task);
742 /* Init the xen-balloon driver. */
747 subsys_initcall(balloon_init);
749 static int __init balloon_wait_finish(void)
751 long credit, last_credit = 0;
752 unsigned long last_changed = 0;
757 /* PV guests don't need to wait. */
758 if (xen_pv_domain() || !current_credit())
761 pr_notice("Waiting for initial ballooning down having finished.\n");
763 while ((credit = current_credit()) < 0) {
764 if (credit != last_credit) {
765 last_changed = jiffies;
766 last_credit = credit;
768 if (balloon_state == BP_ECANCELED) {
769 pr_warn_once("Initial ballooning failed, %ld pages need to be freed.\n",
771 if (time_is_before_eq_jiffies(last_changed + HZ * balloon_boot_timeout))
772 panic("Initial ballooning failed!\n");
775 schedule_timeout_interruptible(HZ / 10);
778 pr_notice("Initial ballooning down finished.\n");
782 late_initcall_sync(balloon_wait_finish);