1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
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/errno.h>
45#include <linux/mm.h>
46#include <linux/bootmem.h>
47#include <linux/pagemap.h>
48#include <linux/highmem.h>
49#include <linux/mutex.h>
50#include <linux/list.h>
51#include <linux/gfp.h>
52#include <linux/notifier.h>
53#include <linux/memory.h>
54#include <linux/memory_hotplug.h>
55#include <linux/percpu-defs.h>
56#include <linux/slab.h>
57#include <linux/sysctl.h>
58
59#include <asm/page.h>
60#include <asm/pgalloc.h>
61#include <asm/pgtable.h>
62#include <asm/tlb.h>
63
64#include <asm/xen/hypervisor.h>
65#include <asm/xen/hypercall.h>
66
67#include <xen/xen.h>
68#include <xen/interface/xen.h>
69#include <xen/interface/memory.h>
70#include <xen/balloon.h>
71#include <xen/features.h>
72#include <xen/page.h>
73
74static int xen_hotplug_unpopulated;
75
76#ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
77
78static int zero;
79static int one = 1;
80
81static struct ctl_table balloon_table[] = {
82 {
83 .procname = "hotplug_unpopulated",
84 .data = &xen_hotplug_unpopulated,
85 .maxlen = sizeof(int),
86 .mode = 0644,
87 .proc_handler = proc_dointvec_minmax,
88 .extra1 = &zero,
89 .extra2 = &one,
90 },
91 { }
92};
93
94static struct ctl_table balloon_root[] = {
95 {
96 .procname = "balloon",
97 .mode = 0555,
98 .child = balloon_table,
99 },
100 { }
101};
102
103static struct ctl_table xen_root[] = {
104 {
105 .procname = "xen",
106 .mode = 0555,
107 .child = balloon_root,
108 },
109 { }
110};
111
112#endif
113
114
115
116
117
118#define EXTENT_ORDER (fls(XEN_PFN_PER_PAGE) - 1)
119
120
121
122
123
124
125
126
127
128
129enum bp_state {
130 BP_DONE,
131 BP_WAIT,
132 BP_EAGAIN,
133 BP_ECANCELED
134};
135
136
137static DEFINE_MUTEX(balloon_mutex);
138
139struct balloon_stats balloon_stats;
140EXPORT_SYMBOL_GPL(balloon_stats);
141
142
143static xen_pfn_t frame_list[PAGE_SIZE / sizeof(xen_pfn_t)];
144
145
146
147static LIST_HEAD(ballooned_pages);
148static DECLARE_WAIT_QUEUE_HEAD(balloon_wq);
149
150
151static void balloon_process(struct work_struct *work);
152static DECLARE_DELAYED_WORK(balloon_worker, balloon_process);
153
154
155
156#define GFP_BALLOON \
157 (GFP_HIGHUSER | __GFP_NOWARN | __GFP_NORETRY | __GFP_NOMEMALLOC)
158
159static void scrub_page(struct page *page)
160{
161#ifdef CONFIG_XEN_SCRUB_PAGES
162 clear_highpage(page);
163#endif
164}
165
166
167static void __balloon_append(struct page *page)
168{
169
170 if (PageHighMem(page)) {
171 list_add_tail(&page->lru, &ballooned_pages);
172 balloon_stats.balloon_high++;
173 } else {
174 list_add(&page->lru, &ballooned_pages);
175 balloon_stats.balloon_low++;
176 }
177 wake_up(&balloon_wq);
178}
179
180static void balloon_append(struct page *page)
181{
182 __balloon_append(page);
183 adjust_managed_page_count(page, -1);
184}
185
186
187static struct page *balloon_retrieve(bool require_lowmem)
188{
189 struct page *page;
190
191 if (list_empty(&ballooned_pages))
192 return NULL;
193
194 page = list_entry(ballooned_pages.next, struct page, lru);
195 if (require_lowmem && PageHighMem(page))
196 return NULL;
197 list_del(&page->lru);
198
199 if (PageHighMem(page))
200 balloon_stats.balloon_high--;
201 else
202 balloon_stats.balloon_low--;
203
204 adjust_managed_page_count(page, 1);
205
206 return page;
207}
208
209static struct page *balloon_next_page(struct page *page)
210{
211 struct list_head *next = page->lru.next;
212 if (next == &ballooned_pages)
213 return NULL;
214 return list_entry(next, struct page, lru);
215}
216
217static enum bp_state update_schedule(enum bp_state state)
218{
219 if (state == BP_WAIT)
220 return BP_WAIT;
221
222 if (state == BP_ECANCELED)
223 return BP_ECANCELED;
224
225 if (state == BP_DONE) {
226 balloon_stats.schedule_delay = 1;
227 balloon_stats.retry_count = 1;
228 return BP_DONE;
229 }
230
231 ++balloon_stats.retry_count;
232
233 if (balloon_stats.max_retry_count != RETRY_UNLIMITED &&
234 balloon_stats.retry_count > balloon_stats.max_retry_count) {
235 balloon_stats.schedule_delay = 1;
236 balloon_stats.retry_count = 1;
237 return BP_ECANCELED;
238 }
239
240 balloon_stats.schedule_delay <<= 1;
241
242 if (balloon_stats.schedule_delay > balloon_stats.max_schedule_delay)
243 balloon_stats.schedule_delay = balloon_stats.max_schedule_delay;
244
245 return BP_EAGAIN;
246}
247
248#ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
249static void release_memory_resource(struct resource *resource)
250{
251 if (!resource)
252 return;
253
254
255
256
257
258 release_resource(resource);
259 kfree(resource);
260}
261
262static struct resource *additional_memory_resource(phys_addr_t size)
263{
264 struct resource *res;
265 int ret;
266
267 res = kzalloc(sizeof(*res), GFP_KERNEL);
268 if (!res)
269 return NULL;
270
271 res->name = "System RAM";
272 res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
273
274 ret = allocate_resource(&iomem_resource, res,
275 size, 0, -1,
276 PAGES_PER_SECTION * PAGE_SIZE, NULL, NULL);
277 if (ret < 0) {
278 pr_err("Cannot allocate new System RAM resource\n");
279 kfree(res);
280 return NULL;
281 }
282
283#ifdef CONFIG_SPARSEMEM
284 {
285 unsigned long limit = 1UL << (MAX_PHYSMEM_BITS - PAGE_SHIFT);
286 unsigned long pfn = res->start >> PAGE_SHIFT;
287
288 if (pfn > limit) {
289 pr_err("New System RAM resource outside addressable RAM (%lu > %lu)\n",
290 pfn, limit);
291 release_memory_resource(res);
292 return NULL;
293 }
294 }
295#endif
296
297 return res;
298}
299
300static enum bp_state reserve_additional_memory(void)
301{
302 long credit;
303 struct resource *resource;
304 int nid, rc;
305 unsigned long balloon_hotplug;
306
307 credit = balloon_stats.target_pages + balloon_stats.target_unpopulated
308 - balloon_stats.total_pages;
309
310
311
312
313
314 if (credit <= 0)
315 return BP_WAIT;
316
317 balloon_hotplug = round_up(credit, PAGES_PER_SECTION);
318
319 resource = additional_memory_resource(balloon_hotplug * PAGE_SIZE);
320 if (!resource)
321 goto err;
322
323 nid = memory_add_physaddr_to_nid(resource->start);
324
325#ifdef CONFIG_XEN_HAVE_PVMMU
326
327
328
329
330 BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
331
332
333
334
335
336
337
338
339
340
341 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
342 unsigned long pfn, i;
343
344 pfn = PFN_DOWN(resource->start);
345 for (i = 0; i < balloon_hotplug; i++) {
346 if (!set_phys_to_machine(pfn + i, INVALID_P2M_ENTRY)) {
347 pr_warn("set_phys_to_machine() failed, no memory added\n");
348 goto err;
349 }
350 }
351 }
352#endif
353
354
355
356
357
358
359
360 mutex_unlock(&balloon_mutex);
361 rc = add_memory_resource(nid, resource, memhp_auto_online);
362 mutex_lock(&balloon_mutex);
363
364 if (rc) {
365 pr_warn("Cannot add additional memory (%i)\n", rc);
366 goto err;
367 }
368
369 balloon_stats.total_pages += balloon_hotplug;
370
371 return BP_WAIT;
372 err:
373 release_memory_resource(resource);
374 return BP_ECANCELED;
375}
376
377static void xen_online_page(struct page *page)
378{
379 __online_page_set_limits(page);
380
381 mutex_lock(&balloon_mutex);
382
383 __balloon_append(page);
384
385 mutex_unlock(&balloon_mutex);
386}
387
388static int xen_memory_notifier(struct notifier_block *nb, unsigned long val, void *v)
389{
390 if (val == MEM_ONLINE)
391 schedule_delayed_work(&balloon_worker, 0);
392
393 return NOTIFY_OK;
394}
395
396static struct notifier_block xen_memory_nb = {
397 .notifier_call = xen_memory_notifier,
398 .priority = 0
399};
400#else
401static enum bp_state reserve_additional_memory(void)
402{
403 balloon_stats.target_pages = balloon_stats.current_pages;
404 return BP_ECANCELED;
405}
406#endif
407
408static long current_credit(void)
409{
410 return balloon_stats.target_pages - balloon_stats.current_pages;
411}
412
413static bool balloon_is_inflated(void)
414{
415 return balloon_stats.balloon_low || balloon_stats.balloon_high;
416}
417
418static enum bp_state increase_reservation(unsigned long nr_pages)
419{
420 int rc;
421 unsigned long i;
422 struct page *page;
423 struct xen_memory_reservation reservation = {
424 .address_bits = 0,
425 .extent_order = EXTENT_ORDER,
426 .domid = DOMID_SELF
427 };
428
429 if (nr_pages > ARRAY_SIZE(frame_list))
430 nr_pages = ARRAY_SIZE(frame_list);
431
432 page = list_first_entry_or_null(&ballooned_pages, struct page, lru);
433 for (i = 0; i < nr_pages; i++) {
434 if (!page) {
435 nr_pages = i;
436 break;
437 }
438
439
440
441
442 frame_list[i] = page_to_xen_pfn(page);
443 page = balloon_next_page(page);
444 }
445
446 set_xen_guest_handle(reservation.extent_start, frame_list);
447 reservation.nr_extents = nr_pages;
448 rc = HYPERVISOR_memory_op(XENMEM_populate_physmap, &reservation);
449 if (rc <= 0)
450 return BP_EAGAIN;
451
452 for (i = 0; i < rc; i++) {
453 page = balloon_retrieve(false);
454 BUG_ON(page == NULL);
455
456#ifdef CONFIG_XEN_HAVE_PVMMU
457
458
459
460
461 BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
462
463 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
464 unsigned long pfn = page_to_pfn(page);
465
466 set_phys_to_machine(pfn, frame_list[i]);
467
468
469 if (!PageHighMem(page)) {
470 int ret;
471 ret = HYPERVISOR_update_va_mapping(
472 (unsigned long)__va(pfn << PAGE_SHIFT),
473 mfn_pte(frame_list[i], PAGE_KERNEL),
474 0);
475 BUG_ON(ret);
476 }
477 }
478#endif
479
480
481 __free_reserved_page(page);
482 }
483
484 balloon_stats.current_pages += rc;
485
486 return BP_DONE;
487}
488
489static enum bp_state decrease_reservation(unsigned long nr_pages, gfp_t gfp)
490{
491 enum bp_state state = BP_DONE;
492 unsigned long i;
493 struct page *page, *tmp;
494 int ret;
495 struct xen_memory_reservation reservation = {
496 .address_bits = 0,
497 .extent_order = EXTENT_ORDER,
498 .domid = DOMID_SELF
499 };
500 LIST_HEAD(pages);
501
502 if (nr_pages > ARRAY_SIZE(frame_list))
503 nr_pages = ARRAY_SIZE(frame_list);
504
505 for (i = 0; i < nr_pages; i++) {
506 page = alloc_page(gfp);
507 if (page == NULL) {
508 nr_pages = i;
509 state = BP_EAGAIN;
510 break;
511 }
512 scrub_page(page);
513 list_add(&page->lru, &pages);
514 }
515
516
517
518
519
520
521
522
523 kmap_flush_unused();
524
525
526
527
528
529 i = 0;
530 list_for_each_entry_safe(page, tmp, &pages, lru) {
531
532 frame_list[i++] = xen_page_to_gfn(page);
533
534#ifdef CONFIG_XEN_HAVE_PVMMU
535
536
537
538
539 BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
540
541 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
542 unsigned long pfn = page_to_pfn(page);
543
544 if (!PageHighMem(page)) {
545 ret = HYPERVISOR_update_va_mapping(
546 (unsigned long)__va(pfn << PAGE_SHIFT),
547 __pte_ma(0), 0);
548 BUG_ON(ret);
549 }
550 __set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
551 }
552#endif
553 list_del(&page->lru);
554
555 balloon_append(page);
556 }
557
558 flush_tlb_all();
559
560 set_xen_guest_handle(reservation.extent_start, frame_list);
561 reservation.nr_extents = nr_pages;
562 ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation);
563 BUG_ON(ret != nr_pages);
564
565 balloon_stats.current_pages -= nr_pages;
566
567 return state;
568}
569
570
571
572
573
574
575
576static void balloon_process(struct work_struct *work)
577{
578 enum bp_state state = BP_DONE;
579 long credit;
580
581
582 do {
583 mutex_lock(&balloon_mutex);
584
585 credit = current_credit();
586
587 if (credit > 0) {
588 if (balloon_is_inflated())
589 state = increase_reservation(credit);
590 else
591 state = reserve_additional_memory();
592 }
593
594 if (credit < 0)
595 state = decrease_reservation(-credit, GFP_BALLOON);
596
597 state = update_schedule(state);
598
599 mutex_unlock(&balloon_mutex);
600
601 cond_resched();
602
603 } while (credit && state == BP_DONE);
604
605
606 if (state == BP_EAGAIN)
607 schedule_delayed_work(&balloon_worker, balloon_stats.schedule_delay * HZ);
608}
609
610
611void balloon_set_new_target(unsigned long target)
612{
613
614 balloon_stats.target_pages = target;
615 schedule_delayed_work(&balloon_worker, 0);
616}
617EXPORT_SYMBOL_GPL(balloon_set_new_target);
618
619static int add_ballooned_pages(int nr_pages)
620{
621 enum bp_state st;
622
623 if (xen_hotplug_unpopulated) {
624 st = reserve_additional_memory();
625 if (st != BP_ECANCELED) {
626 mutex_unlock(&balloon_mutex);
627 wait_event(balloon_wq,
628 !list_empty(&ballooned_pages));
629 mutex_lock(&balloon_mutex);
630 return 0;
631 }
632 }
633
634 st = decrease_reservation(nr_pages, GFP_USER);
635 if (st != BP_DONE)
636 return -ENOMEM;
637
638 return 0;
639}
640
641
642
643
644
645
646
647int alloc_xenballooned_pages(int nr_pages, struct page **pages)
648{
649 int pgno = 0;
650 struct page *page;
651 int ret;
652
653 mutex_lock(&balloon_mutex);
654
655 balloon_stats.target_unpopulated += nr_pages;
656
657 while (pgno < nr_pages) {
658 page = balloon_retrieve(true);
659 if (page) {
660 pages[pgno++] = page;
661#ifdef CONFIG_XEN_HAVE_PVMMU
662
663
664
665
666 BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
667
668 ret = xen_alloc_p2m_entry(page_to_pfn(page));
669 if (ret < 0)
670 goto out_undo;
671#endif
672 } else {
673 ret = add_ballooned_pages(nr_pages - pgno);
674 if (ret < 0)
675 goto out_undo;
676 }
677 }
678 mutex_unlock(&balloon_mutex);
679 return 0;
680 out_undo:
681 mutex_unlock(&balloon_mutex);
682 free_xenballooned_pages(pgno, pages);
683 return ret;
684}
685EXPORT_SYMBOL(alloc_xenballooned_pages);
686
687
688
689
690
691
692void free_xenballooned_pages(int nr_pages, struct page **pages)
693{
694 int i;
695
696 mutex_lock(&balloon_mutex);
697
698 for (i = 0; i < nr_pages; i++) {
699 if (pages[i])
700 balloon_append(pages[i]);
701 }
702
703 balloon_stats.target_unpopulated -= nr_pages;
704
705
706 if (current_credit())
707 schedule_delayed_work(&balloon_worker, 0);
708
709 mutex_unlock(&balloon_mutex);
710}
711EXPORT_SYMBOL(free_xenballooned_pages);
712
713static void __init balloon_add_region(unsigned long start_pfn,
714 unsigned long pages)
715{
716 unsigned long pfn, extra_pfn_end;
717 struct page *page;
718
719
720
721
722
723
724 extra_pfn_end = min(max_pfn, start_pfn + pages);
725
726 for (pfn = start_pfn; pfn < extra_pfn_end; pfn++) {
727 page = pfn_to_page(pfn);
728
729
730
731 __balloon_append(page);
732 }
733
734 balloon_stats.total_pages += extra_pfn_end - start_pfn;
735}
736
737static int __init balloon_init(void)
738{
739 int i;
740
741 if (!xen_domain())
742 return -ENODEV;
743
744 pr_info("Initialising balloon driver\n");
745
746 balloon_stats.current_pages = xen_pv_domain()
747 ? min(xen_start_info->nr_pages - xen_released_pages, max_pfn)
748 : get_num_physpages();
749 balloon_stats.target_pages = balloon_stats.current_pages;
750 balloon_stats.balloon_low = 0;
751 balloon_stats.balloon_high = 0;
752 balloon_stats.total_pages = balloon_stats.current_pages;
753
754 balloon_stats.schedule_delay = 1;
755 balloon_stats.max_schedule_delay = 32;
756 balloon_stats.retry_count = 1;
757 balloon_stats.max_retry_count = RETRY_UNLIMITED;
758
759#ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
760 set_online_page_callback(&xen_online_page);
761 register_memory_notifier(&xen_memory_nb);
762 register_sysctl_table(xen_root);
763#endif
764
765
766
767
768
769 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++)
770 if (xen_extra_mem[i].n_pfns)
771 balloon_add_region(xen_extra_mem[i].start_pfn,
772 xen_extra_mem[i].n_pfns);
773
774 return 0;
775}
776subsys_initcall(balloon_init);
777