]> asedeno.scripts.mit.edu Git - linux.git/blob - drivers/xen/balloon.c
xen/balloon: Mark pages PG_offline in balloon_append()
[linux.git] / drivers / xen / balloon.c
1 /******************************************************************************
2  * Xen balloon driver - enables returning/claiming memory to/from Xen.
3  *
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
8  *
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
12  * this project.
13  *
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:
19  *
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:
26  *
27  * The above copyright notice and this permission notice shall be included in
28  * all copies or substantial portions of the Software.
29  *
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
36  * IN THE SOFTWARE.
37  */
38
39 #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
40
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/mm.h>
47 #include <linux/memblock.h>
48 #include <linux/pagemap.h>
49 #include <linux/highmem.h>
50 #include <linux/mutex.h>
51 #include <linux/list.h>
52 #include <linux/gfp.h>
53 #include <linux/notifier.h>
54 #include <linux/memory.h>
55 #include <linux/memory_hotplug.h>
56 #include <linux/percpu-defs.h>
57 #include <linux/slab.h>
58 #include <linux/sysctl.h>
59
60 #include <asm/page.h>
61 #include <asm/pgalloc.h>
62 #include <asm/pgtable.h>
63 #include <asm/tlb.h>
64
65 #include <asm/xen/hypervisor.h>
66 #include <asm/xen/hypercall.h>
67
68 #include <xen/xen.h>
69 #include <xen/interface/xen.h>
70 #include <xen/interface/memory.h>
71 #include <xen/balloon.h>
72 #include <xen/features.h>
73 #include <xen/page.h>
74 #include <xen/mem-reservation.h>
75
76 static int xen_hotplug_unpopulated;
77
78 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
79
80 static struct ctl_table balloon_table[] = {
81         {
82                 .procname       = "hotplug_unpopulated",
83                 .data           = &xen_hotplug_unpopulated,
84                 .maxlen         = sizeof(int),
85                 .mode           = 0644,
86                 .proc_handler   = proc_dointvec_minmax,
87                 .extra1         = SYSCTL_ZERO,
88                 .extra2         = SYSCTL_ONE,
89         },
90         { }
91 };
92
93 static struct ctl_table balloon_root[] = {
94         {
95                 .procname       = "balloon",
96                 .mode           = 0555,
97                 .child          = balloon_table,
98         },
99         { }
100 };
101
102 static struct ctl_table xen_root[] = {
103         {
104                 .procname       = "xen",
105                 .mode           = 0555,
106                 .child          = balloon_root,
107         },
108         { }
109 };
110
111 #endif
112
113 /*
114  * Use one extent per PAGE_SIZE to avoid to break down the page into
115  * multiple frame.
116  */
117 #define EXTENT_ORDER (fls(XEN_PFN_PER_PAGE) - 1)
118
119 /*
120  * balloon_process() state:
121  *
122  * BP_DONE: done or nothing to do,
123  * BP_WAIT: wait to be rescheduled,
124  * BP_EAGAIN: error, go to sleep,
125  * BP_ECANCELED: error, balloon operation canceled.
126  */
127
128 enum bp_state {
129         BP_DONE,
130         BP_WAIT,
131         BP_EAGAIN,
132         BP_ECANCELED
133 };
134
135
136 static DEFINE_MUTEX(balloon_mutex);
137
138 struct balloon_stats balloon_stats;
139 EXPORT_SYMBOL_GPL(balloon_stats);
140
141 /* We increase/decrease in batches which fit in a page */
142 static xen_pfn_t frame_list[PAGE_SIZE / sizeof(xen_pfn_t)];
143
144
145 /* List of ballooned pages, threaded through the mem_map array. */
146 static LIST_HEAD(ballooned_pages);
147 static DECLARE_WAIT_QUEUE_HEAD(balloon_wq);
148
149 /* Main work function, always executed in process context. */
150 static void balloon_process(struct work_struct *work);
151 static DECLARE_DELAYED_WORK(balloon_worker, balloon_process);
152
153 /* When ballooning out (allocating memory to return to Xen) we don't really
154    want the kernel to try too hard since that can trigger the oom killer. */
155 #define GFP_BALLOON \
156         (GFP_HIGHUSER | __GFP_NOWARN | __GFP_NORETRY | __GFP_NOMEMALLOC)
157
158 /* balloon_append: add the given page to the balloon. */
159 static void balloon_append(struct page *page)
160 {
161         __SetPageOffline(page);
162
163         /* Lowmem is re-populated first, so highmem pages go at list tail. */
164         if (PageHighMem(page)) {
165                 list_add_tail(&page->lru, &ballooned_pages);
166                 balloon_stats.balloon_high++;
167         } else {
168                 list_add(&page->lru, &ballooned_pages);
169                 balloon_stats.balloon_low++;
170         }
171         wake_up(&balloon_wq);
172 }
173
174 /* balloon_retrieve: rescue a page from the balloon, if it is not empty. */
175 static struct page *balloon_retrieve(bool require_lowmem)
176 {
177         struct page *page;
178
179         if (list_empty(&ballooned_pages))
180                 return NULL;
181
182         page = list_entry(ballooned_pages.next, struct page, lru);
183         if (require_lowmem && PageHighMem(page))
184                 return NULL;
185         list_del(&page->lru);
186
187         if (PageHighMem(page))
188                 balloon_stats.balloon_high--;
189         else
190                 balloon_stats.balloon_low--;
191
192         return page;
193 }
194
195 static struct page *balloon_next_page(struct page *page)
196 {
197         struct list_head *next = page->lru.next;
198         if (next == &ballooned_pages)
199                 return NULL;
200         return list_entry(next, struct page, lru);
201 }
202
203 static enum bp_state update_schedule(enum bp_state state)
204 {
205         if (state == BP_WAIT)
206                 return BP_WAIT;
207
208         if (state == BP_ECANCELED)
209                 return BP_ECANCELED;
210
211         if (state == BP_DONE) {
212                 balloon_stats.schedule_delay = 1;
213                 balloon_stats.retry_count = 1;
214                 return BP_DONE;
215         }
216
217         ++balloon_stats.retry_count;
218
219         if (balloon_stats.max_retry_count != RETRY_UNLIMITED &&
220                         balloon_stats.retry_count > balloon_stats.max_retry_count) {
221                 balloon_stats.schedule_delay = 1;
222                 balloon_stats.retry_count = 1;
223                 return BP_ECANCELED;
224         }
225
226         balloon_stats.schedule_delay <<= 1;
227
228         if (balloon_stats.schedule_delay > balloon_stats.max_schedule_delay)
229                 balloon_stats.schedule_delay = balloon_stats.max_schedule_delay;
230
231         return BP_EAGAIN;
232 }
233
234 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
235 static void release_memory_resource(struct resource *resource)
236 {
237         if (!resource)
238                 return;
239
240         /*
241          * No need to reset region to identity mapped since we now
242          * know that no I/O can be in this region
243          */
244         release_resource(resource);
245         kfree(resource);
246 }
247
248 static struct resource *additional_memory_resource(phys_addr_t size)
249 {
250         struct resource *res;
251         int ret;
252
253         res = kzalloc(sizeof(*res), GFP_KERNEL);
254         if (!res)
255                 return NULL;
256
257         res->name = "System RAM";
258         res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
259
260         ret = allocate_resource(&iomem_resource, res,
261                                 size, 0, -1,
262                                 PAGES_PER_SECTION * PAGE_SIZE, NULL, NULL);
263         if (ret < 0) {
264                 pr_err("Cannot allocate new System RAM resource\n");
265                 kfree(res);
266                 return NULL;
267         }
268
269 #ifdef CONFIG_SPARSEMEM
270         {
271                 unsigned long limit = 1UL << (MAX_PHYSMEM_BITS - PAGE_SHIFT);
272                 unsigned long pfn = res->start >> PAGE_SHIFT;
273
274                 if (pfn > limit) {
275                         pr_err("New System RAM resource outside addressable RAM (%lu > %lu)\n",
276                                pfn, limit);
277                         release_memory_resource(res);
278                         return NULL;
279                 }
280         }
281 #endif
282
283         return res;
284 }
285
286 static enum bp_state reserve_additional_memory(void)
287 {
288         long credit;
289         struct resource *resource;
290         int nid, rc;
291         unsigned long balloon_hotplug;
292
293         credit = balloon_stats.target_pages + balloon_stats.target_unpopulated
294                 - balloon_stats.total_pages;
295
296         /*
297          * Already hotplugged enough pages?  Wait for them to be
298          * onlined.
299          */
300         if (credit <= 0)
301                 return BP_WAIT;
302
303         balloon_hotplug = round_up(credit, PAGES_PER_SECTION);
304
305         resource = additional_memory_resource(balloon_hotplug * PAGE_SIZE);
306         if (!resource)
307                 goto err;
308
309         nid = memory_add_physaddr_to_nid(resource->start);
310
311 #ifdef CONFIG_XEN_HAVE_PVMMU
312         /*
313          * We don't support PV MMU when Linux and Xen is using
314          * different page granularity.
315          */
316         BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
317
318         /*
319          * add_memory() will build page tables for the new memory so
320          * the p2m must contain invalid entries so the correct
321          * non-present PTEs will be written.
322          *
323          * If a failure occurs, the original (identity) p2m entries
324          * are not restored since this region is now known not to
325          * conflict with any devices.
326          */ 
327         if (!xen_feature(XENFEAT_auto_translated_physmap)) {
328                 unsigned long pfn, i;
329
330                 pfn = PFN_DOWN(resource->start);
331                 for (i = 0; i < balloon_hotplug; i++) {
332                         if (!set_phys_to_machine(pfn + i, INVALID_P2M_ENTRY)) {
333                                 pr_warn("set_phys_to_machine() failed, no memory added\n");
334                                 goto err;
335                         }
336                 }
337         }
338 #endif
339
340         /*
341          * add_memory_resource() will call online_pages() which in its turn
342          * will call xen_online_page() callback causing deadlock if we don't
343          * release balloon_mutex here. Unlocking here is safe because the
344          * callers drop the mutex before trying again.
345          */
346         mutex_unlock(&balloon_mutex);
347         /* add_memory_resource() requires the device_hotplug lock */
348         lock_device_hotplug();
349         rc = add_memory_resource(nid, resource);
350         unlock_device_hotplug();
351         mutex_lock(&balloon_mutex);
352
353         if (rc) {
354                 pr_warn("Cannot add additional memory (%i)\n", rc);
355                 goto err;
356         }
357
358         balloon_stats.total_pages += balloon_hotplug;
359
360         return BP_WAIT;
361   err:
362         release_memory_resource(resource);
363         return BP_ECANCELED;
364 }
365
366 static void xen_online_page(struct page *page, unsigned int order)
367 {
368         unsigned long i, size = (1 << order);
369         unsigned long start_pfn = page_to_pfn(page);
370         struct page *p;
371
372         pr_debug("Online %lu pages starting at pfn 0x%lx\n", size, start_pfn);
373         mutex_lock(&balloon_mutex);
374         for (i = 0; i < size; i++) {
375                 p = pfn_to_page(start_pfn + i);
376                 __online_page_set_limits(p);
377                 balloon_append(p);
378         }
379         mutex_unlock(&balloon_mutex);
380 }
381
382 static int xen_memory_notifier(struct notifier_block *nb, unsigned long val, void *v)
383 {
384         if (val == MEM_ONLINE)
385                 schedule_delayed_work(&balloon_worker, 0);
386
387         return NOTIFY_OK;
388 }
389
390 static struct notifier_block xen_memory_nb = {
391         .notifier_call = xen_memory_notifier,
392         .priority = 0
393 };
394 #else
395 static enum bp_state reserve_additional_memory(void)
396 {
397         balloon_stats.target_pages = balloon_stats.current_pages;
398         return BP_ECANCELED;
399 }
400 #endif /* CONFIG_XEN_BALLOON_MEMORY_HOTPLUG */
401
402 static long current_credit(void)
403 {
404         return balloon_stats.target_pages - balloon_stats.current_pages;
405 }
406
407 static bool balloon_is_inflated(void)
408 {
409         return balloon_stats.balloon_low || balloon_stats.balloon_high;
410 }
411
412 static enum bp_state increase_reservation(unsigned long nr_pages)
413 {
414         int rc;
415         unsigned long i;
416         struct page   *page;
417
418         if (nr_pages > ARRAY_SIZE(frame_list))
419                 nr_pages = ARRAY_SIZE(frame_list);
420
421         page = list_first_entry_or_null(&ballooned_pages, struct page, lru);
422         for (i = 0; i < nr_pages; i++) {
423                 if (!page) {
424                         nr_pages = i;
425                         break;
426                 }
427
428                 frame_list[i] = page_to_xen_pfn(page);
429                 page = balloon_next_page(page);
430         }
431
432         rc = xenmem_reservation_increase(nr_pages, frame_list);
433         if (rc <= 0)
434                 return BP_EAGAIN;
435
436         for (i = 0; i < rc; i++) {
437                 page = balloon_retrieve(false);
438                 BUG_ON(page == NULL);
439
440                 xenmem_reservation_va_mapping_update(1, &page, &frame_list[i]);
441
442                 /* Relinquish the page back to the allocator. */
443                 __ClearPageOffline(page);
444                 free_reserved_page(page);
445         }
446
447         balloon_stats.current_pages += rc;
448
449         return BP_DONE;
450 }
451
452 static enum bp_state decrease_reservation(unsigned long nr_pages, gfp_t gfp)
453 {
454         enum bp_state state = BP_DONE;
455         unsigned long i;
456         struct page *page, *tmp;
457         int ret;
458         LIST_HEAD(pages);
459
460         if (nr_pages > ARRAY_SIZE(frame_list))
461                 nr_pages = ARRAY_SIZE(frame_list);
462
463         for (i = 0; i < nr_pages; i++) {
464                 page = alloc_page(gfp);
465                 if (page == NULL) {
466                         nr_pages = i;
467                         state = BP_EAGAIN;
468                         break;
469                 }
470                 adjust_managed_page_count(page, -1);
471                 xenmem_reservation_scrub_page(page);
472                 list_add(&page->lru, &pages);
473         }
474
475         /*
476          * Ensure that ballooned highmem pages don't have kmaps.
477          *
478          * Do this before changing the p2m as kmap_flush_unused()
479          * reads PTEs to obtain pages (and hence needs the original
480          * p2m entry).
481          */
482         kmap_flush_unused();
483
484         /*
485          * Setup the frame, update direct mapping, invalidate P2M,
486          * and add to balloon.
487          */
488         i = 0;
489         list_for_each_entry_safe(page, tmp, &pages, lru) {
490                 frame_list[i++] = xen_page_to_gfn(page);
491
492                 xenmem_reservation_va_mapping_reset(1, &page);
493
494                 list_del(&page->lru);
495
496                 balloon_append(page);
497         }
498
499         flush_tlb_all();
500
501         ret = xenmem_reservation_decrease(nr_pages, frame_list);
502         BUG_ON(ret != nr_pages);
503
504         balloon_stats.current_pages -= nr_pages;
505
506         return state;
507 }
508
509 /*
510  * As this is a work item it is guaranteed to run as a single instance only.
511  * We may of course race updates of the target counts (which are protected
512  * by the balloon lock), or with changes to the Xen hard limit, but we will
513  * recover from these in time.
514  */
515 static void balloon_process(struct work_struct *work)
516 {
517         enum bp_state state = BP_DONE;
518         long credit;
519
520
521         do {
522                 mutex_lock(&balloon_mutex);
523
524                 credit = current_credit();
525
526                 if (credit > 0) {
527                         if (balloon_is_inflated())
528                                 state = increase_reservation(credit);
529                         else
530                                 state = reserve_additional_memory();
531                 }
532
533                 if (credit < 0) {
534                         long n_pages;
535
536                         n_pages = min(-credit, si_mem_available());
537                         state = decrease_reservation(n_pages, GFP_BALLOON);
538                         if (state == BP_DONE && n_pages != -credit &&
539                             n_pages < totalreserve_pages)
540                                 state = BP_EAGAIN;
541                 }
542
543                 state = update_schedule(state);
544
545                 mutex_unlock(&balloon_mutex);
546
547                 cond_resched();
548
549         } while (credit && state == BP_DONE);
550
551         /* Schedule more work if there is some still to be done. */
552         if (state == BP_EAGAIN)
553                 schedule_delayed_work(&balloon_worker, balloon_stats.schedule_delay * HZ);
554 }
555
556 /* Resets the Xen limit, sets new target, and kicks off processing. */
557 void balloon_set_new_target(unsigned long target)
558 {
559         /* No need for lock. Not read-modify-write updates. */
560         balloon_stats.target_pages = target;
561         schedule_delayed_work(&balloon_worker, 0);
562 }
563 EXPORT_SYMBOL_GPL(balloon_set_new_target);
564
565 static int add_ballooned_pages(int nr_pages)
566 {
567         enum bp_state st;
568
569         if (xen_hotplug_unpopulated) {
570                 st = reserve_additional_memory();
571                 if (st != BP_ECANCELED) {
572                         mutex_unlock(&balloon_mutex);
573                         wait_event(balloon_wq,
574                                    !list_empty(&ballooned_pages));
575                         mutex_lock(&balloon_mutex);
576                         return 0;
577                 }
578         }
579
580         if (si_mem_available() < nr_pages)
581                 return -ENOMEM;
582
583         st = decrease_reservation(nr_pages, GFP_USER);
584         if (st != BP_DONE)
585                 return -ENOMEM;
586
587         return 0;
588 }
589
590 /**
591  * alloc_xenballooned_pages - get pages that have been ballooned out
592  * @nr_pages: Number of pages to get
593  * @pages: pages returned
594  * @return 0 on success, error otherwise
595  */
596 int alloc_xenballooned_pages(int nr_pages, struct page **pages)
597 {
598         int pgno = 0;
599         struct page *page;
600         int ret;
601
602         mutex_lock(&balloon_mutex);
603
604         balloon_stats.target_unpopulated += nr_pages;
605
606         while (pgno < nr_pages) {
607                 page = balloon_retrieve(true);
608                 if (page) {
609                         __ClearPageOffline(page);
610                         pages[pgno++] = page;
611 #ifdef CONFIG_XEN_HAVE_PVMMU
612                         /*
613                          * We don't support PV MMU when Linux and Xen is using
614                          * different page granularity.
615                          */
616                         BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
617
618                         if (!xen_feature(XENFEAT_auto_translated_physmap)) {
619                                 ret = xen_alloc_p2m_entry(page_to_pfn(page));
620                                 if (ret < 0)
621                                         goto out_undo;
622                         }
623 #endif
624                 } else {
625                         ret = add_ballooned_pages(nr_pages - pgno);
626                         if (ret < 0)
627                                 goto out_undo;
628                 }
629         }
630         mutex_unlock(&balloon_mutex);
631         return 0;
632  out_undo:
633         mutex_unlock(&balloon_mutex);
634         free_xenballooned_pages(pgno, pages);
635         return ret;
636 }
637 EXPORT_SYMBOL(alloc_xenballooned_pages);
638
639 /**
640  * free_xenballooned_pages - return pages retrieved with get_ballooned_pages
641  * @nr_pages: Number of pages
642  * @pages: pages to return
643  */
644 void free_xenballooned_pages(int nr_pages, struct page **pages)
645 {
646         int i;
647
648         mutex_lock(&balloon_mutex);
649
650         for (i = 0; i < nr_pages; i++) {
651                 if (pages[i])
652                         balloon_append(pages[i]);
653         }
654
655         balloon_stats.target_unpopulated -= nr_pages;
656
657         /* The balloon may be too large now. Shrink it if needed. */
658         if (current_credit())
659                 schedule_delayed_work(&balloon_worker, 0);
660
661         mutex_unlock(&balloon_mutex);
662 }
663 EXPORT_SYMBOL(free_xenballooned_pages);
664
665 #ifdef CONFIG_XEN_PV
666 static void __init balloon_add_region(unsigned long start_pfn,
667                                       unsigned long pages)
668 {
669         unsigned long pfn, extra_pfn_end;
670
671         /*
672          * If the amount of usable memory has been limited (e.g., with
673          * the 'mem' command line parameter), don't add pages beyond
674          * this limit.
675          */
676         extra_pfn_end = min(max_pfn, start_pfn + pages);
677
678         for (pfn = start_pfn; pfn < extra_pfn_end; pfn++) {
679                 /* totalram_pages and totalhigh_pages do not
680                    include the boot-time balloon extension, so
681                    don't subtract from it. */
682                 balloon_append(pfn_to_page(pfn));
683         }
684
685         balloon_stats.total_pages += extra_pfn_end - start_pfn;
686 }
687 #endif
688
689 static int __init balloon_init(void)
690 {
691         if (!xen_domain())
692                 return -ENODEV;
693
694         pr_info("Initialising balloon driver\n");
695
696 #ifdef CONFIG_XEN_PV
697         balloon_stats.current_pages = xen_pv_domain()
698                 ? min(xen_start_info->nr_pages - xen_released_pages, max_pfn)
699                 : get_num_physpages();
700 #else
701         balloon_stats.current_pages = get_num_physpages();
702 #endif
703         balloon_stats.target_pages  = balloon_stats.current_pages;
704         balloon_stats.balloon_low   = 0;
705         balloon_stats.balloon_high  = 0;
706         balloon_stats.total_pages   = balloon_stats.current_pages;
707
708         balloon_stats.schedule_delay = 1;
709         balloon_stats.max_schedule_delay = 32;
710         balloon_stats.retry_count = 1;
711         balloon_stats.max_retry_count = 4;
712
713 #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
714         set_online_page_callback(&xen_online_page);
715         register_memory_notifier(&xen_memory_nb);
716         register_sysctl_table(xen_root);
717 #endif
718
719 #ifdef CONFIG_XEN_PV
720         {
721                 int i;
722
723                 /*
724                  * Initialize the balloon with pages from the extra memory
725                  * regions (see arch/x86/xen/setup.c).
726                  */
727                 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++)
728                         if (xen_extra_mem[i].n_pfns)
729                                 balloon_add_region(xen_extra_mem[i].start_pfn,
730                                                    xen_extra_mem[i].n_pfns);
731         }
732 #endif
733
734         /* Init the xen-balloon driver. */
735         xen_balloon_init();
736
737         return 0;
738 }
739 subsys_initcall(balloon_init);