blob: 5bae515c8e25c0a4f1c90b42e24b271a5777ef84 [file] [log] [blame]
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -07001/******************************************************************************
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -07002 * 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
Daniel Kiper080e2be2011-07-25 17:12:06 -07007 * 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.
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -070013 *
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
Joe Perches283c0972013-06-28 03:21:41 -070039#define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
40
Stefano Stabellinicd9151e2013-08-04 15:39:40 +010041#include <linux/cpu.h>
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -070042#include <linux/kernel.h>
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -070043#include <linux/sched.h>
Ingo Molnar5b825c32017-02-02 17:54:15 +010044#include <linux/cred.h>
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -070045#include <linux/errno.h>
46#include <linux/mm.h>
Mike Rapoport57c8a662018-10-30 15:09:49 -070047#include <linux/memblock.h>
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -070048#include <linux/pagemap.h>
49#include <linux/highmem.h>
50#include <linux/mutex.h>
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -070051#include <linux/list.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090052#include <linux/gfp.h>
Daniel Kiper080e2be2011-07-25 17:12:06 -070053#include <linux/notifier.h>
54#include <linux/memory.h>
55#include <linux/memory_hotplug.h>
Stefano Stabellinicd9151e2013-08-04 15:39:40 +010056#include <linux/percpu-defs.h>
David Vrabel55b3da92015-06-24 15:58:42 +010057#include <linux/slab.h>
David Vrabel1cf6a6c2015-06-25 16:29:18 +010058#include <linux/sysctl.h>
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -070059
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -070060#include <asm/page.h>
61#include <asm/pgalloc.h>
62#include <asm/pgtable.h>
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -070063#include <asm/tlb.h>
64
Jeremy Fitzhardingeecbf29c2008-12-16 12:37:07 -080065#include <asm/xen/hypervisor.h>
66#include <asm/xen/hypercall.h>
Jeremy Fitzhardinge1ccbf532009-10-06 15:11:14 -070067
68#include <xen/xen.h>
Jeremy Fitzhardingeecbf29c2008-12-16 12:37:07 -080069#include <xen/interface/xen.h>
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -070070#include <xen/interface/memory.h>
Daniel De Graaf803eb042011-03-14 11:29:37 -040071#include <xen/balloon.h>
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -070072#include <xen/features.h>
73#include <xen/page.h>
Oleksandr Andrushchenkoae4c51a2018-07-20 12:01:44 +030074#include <xen/mem-reservation.h>
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -070075
David Vrabel1cf6a6c2015-06-25 16:29:18 +010076static int xen_hotplug_unpopulated;
77
78#ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
79
David Vrabel1cf6a6c2015-06-25 16:29:18 +010080static 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,
Matteo Croceeec48442019-07-18 15:58:50 -070087 .extra1 = SYSCTL_ZERO,
88 .extra2 = SYSCTL_ONE,
David Vrabel1cf6a6c2015-06-25 16:29:18 +010089 },
90 { }
91};
92
93static struct ctl_table balloon_root[] = {
94 {
95 .procname = "balloon",
96 .mode = 0555,
97 .child = balloon_table,
98 },
99 { }
100};
101
102static struct ctl_table xen_root[] = {
103 {
104 .procname = "xen",
105 .mode = 0555,
106 .child = balloon_root,
107 },
108 { }
109};
110
111#endif
112
Daniel Kiper95d2ac42011-03-08 22:48:24 +0100113/*
Julien Grall30756c62015-05-04 15:39:08 +0100114 * 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/*
Daniel Kiper95d2ac42011-03-08 22:48:24 +0100120 * balloon_process() state:
121 *
122 * BP_DONE: done or nothing to do,
David Vrabelb2ac6aa2015-06-25 12:10:28 +0100123 * BP_WAIT: wait to be rescheduled,
Daniel Kiper95d2ac42011-03-08 22:48:24 +0100124 * BP_EAGAIN: error, go to sleep,
125 * BP_ECANCELED: error, balloon operation canceled.
126 */
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700127
Daniel Kiper95d2ac42011-03-08 22:48:24 +0100128enum bp_state {
129 BP_DONE,
David Vrabelb2ac6aa2015-06-25 12:10:28 +0100130 BP_WAIT,
Daniel Kiper95d2ac42011-03-08 22:48:24 +0100131 BP_EAGAIN,
132 BP_ECANCELED
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700133};
134
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700135
136static DEFINE_MUTEX(balloon_mutex);
137
Daniel De Graaf803eb042011-03-14 11:29:37 -0400138struct balloon_stats balloon_stats;
139EXPORT_SYMBOL_GPL(balloon_stats);
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700140
141/* We increase/decrease in batches which fit in a page */
Julien Grall3990dd22015-10-07 14:04:33 +0100142static xen_pfn_t frame_list[PAGE_SIZE / sizeof(xen_pfn_t)];
Stefano Stabellinicd9151e2013-08-04 15:39:40 +0100143
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700144
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700145/* List of ballooned pages, threaded through the mem_map array. */
146static LIST_HEAD(ballooned_pages);
David Vrabel1cf6a6c2015-06-25 16:29:18 +0100147static DECLARE_WAIT_QUEUE_HEAD(balloon_wq);
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700148
149/* Main work function, always executed in process context. */
150static void balloon_process(struct work_struct *work);
Daniel Kiper95170b22011-03-08 22:47:39 +0100151static DECLARE_DELAYED_WORK(balloon_worker, balloon_process);
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700152
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
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700158/* balloon_append: add the given page to the balloon. */
David Hildenbranddde32852019-10-01 11:01:50 +0200159static void balloon_append(struct page *page)
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700160{
David Hildenbrand59b52f12019-10-01 11:01:51 +0200161 __SetPageOffline(page);
162
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700163 /* 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++;
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700167 } else {
168 list_add(&page->lru, &ballooned_pages);
169 balloon_stats.balloon_low++;
170 }
David Vrabel1cf6a6c2015-06-25 16:29:18 +0100171 wake_up(&balloon_wq);
Jeremy Fitzhardinge9be4d452010-08-31 15:01:16 -0700172}
Gianluca Guida3d65c942009-07-30 22:54:36 +0100173
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700174/* balloon_retrieve: rescue a page from the balloon, if it is not empty. */
David Vrabel81b286e2015-06-25 13:12:46 +0100175static struct page *balloon_retrieve(bool require_lowmem)
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700176{
177 struct page *page;
178
179 if (list_empty(&ballooned_pages))
180 return NULL;
181
David Vrabel81b286e2015-06-25 13:12:46 +0100182 page = list_entry(ballooned_pages.next, struct page, lru);
183 if (require_lowmem && PageHighMem(page))
184 return NULL;
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700185 list_del(&page->lru);
186
Jiang Liu3dcc0572013-07-03 15:03:21 -0700187 if (PageHighMem(page))
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700188 balloon_stats.balloon_high--;
Jiang Liu3dcc0572013-07-03 15:03:21 -0700189 else
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700190 balloon_stats.balloon_low--;
191
David Hildenbrand04ccbdc2019-10-01 11:01:52 +0200192 __ClearPageOffline(page);
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700193 return page;
194}
195
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700196static struct page *balloon_next_page(struct page *page)
197{
198 struct list_head *next = page->lru.next;
199 if (next == &ballooned_pages)
200 return NULL;
201 return list_entry(next, struct page, lru);
202}
203
Daniel Kiper95d2ac42011-03-08 22:48:24 +0100204static enum bp_state update_schedule(enum bp_state state)
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700205{
David Vrabelb2ac6aa2015-06-25 12:10:28 +0100206 if (state == BP_WAIT)
207 return BP_WAIT;
208
Boris Ostrovskyfd8b7952014-10-07 17:00:07 -0400209 if (state == BP_ECANCELED)
210 return BP_ECANCELED;
211
Daniel Kiper95d2ac42011-03-08 22:48:24 +0100212 if (state == BP_DONE) {
213 balloon_stats.schedule_delay = 1;
214 balloon_stats.retry_count = 1;
215 return BP_DONE;
216 }
217
Daniel Kiper95d2ac42011-03-08 22:48:24 +0100218 ++balloon_stats.retry_count;
219
220 if (balloon_stats.max_retry_count != RETRY_UNLIMITED &&
221 balloon_stats.retry_count > balloon_stats.max_retry_count) {
Daniel Kiper95d2ac42011-03-08 22:48:24 +0100222 balloon_stats.schedule_delay = 1;
223 balloon_stats.retry_count = 1;
224 return BP_ECANCELED;
225 }
226
227 balloon_stats.schedule_delay <<= 1;
228
229 if (balloon_stats.schedule_delay > balloon_stats.max_schedule_delay)
230 balloon_stats.schedule_delay = balloon_stats.max_schedule_delay;
231
232 return BP_EAGAIN;
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700233}
234
Daniel Kiper080e2be2011-07-25 17:12:06 -0700235#ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
Ross Lagerwall842775f2016-05-10 10:27:54 +0100236static void release_memory_resource(struct resource *resource)
237{
238 if (!resource)
239 return;
240
241 /*
242 * No need to reset region to identity mapped since we now
243 * know that no I/O can be in this region
244 */
245 release_resource(resource);
246 kfree(resource);
247}
248
David Vrabel55b3da92015-06-24 15:58:42 +0100249static struct resource *additional_memory_resource(phys_addr_t size)
250{
Igor Druzhinin12366412018-11-27 20:58:21 +0000251 struct resource *res;
252 int ret;
David Vrabel55b3da92015-06-24 15:58:42 +0100253
254 res = kzalloc(sizeof(*res), GFP_KERNEL);
255 if (!res)
256 return NULL;
257
258 res->name = "System RAM";
Toshi Kani782b8662016-01-26 21:57:24 +0100259 res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
David Vrabel55b3da92015-06-24 15:58:42 +0100260
Igor Druzhinin12366412018-11-27 20:58:21 +0000261 ret = allocate_resource(&iomem_resource, res,
262 size, 0, -1,
263 PAGES_PER_SECTION * PAGE_SIZE, NULL, NULL);
264 if (ret < 0) {
265 pr_err("Cannot allocate new System RAM resource\n");
266 kfree(res);
267 return NULL;
David Vrabel55b3da92015-06-24 15:58:42 +0100268 }
269
Ross Lagerwalldfd74a12016-03-17 16:52:00 +0000270#ifdef CONFIG_SPARSEMEM
271 {
272 unsigned long limit = 1UL << (MAX_PHYSMEM_BITS - PAGE_SHIFT);
273 unsigned long pfn = res->start >> PAGE_SHIFT;
274
275 if (pfn > limit) {
276 pr_err("New System RAM resource outside addressable RAM (%lu > %lu)\n",
277 pfn, limit);
278 release_memory_resource(res);
279 return NULL;
280 }
281 }
282#endif
283
David Vrabel55b3da92015-06-24 15:58:42 +0100284 return res;
285}
286
David Vrabelb2ac6aa2015-06-25 12:10:28 +0100287static enum bp_state reserve_additional_memory(void)
Daniel Kiper080e2be2011-07-25 17:12:06 -0700288{
David Vrabelb2ac6aa2015-06-25 12:10:28 +0100289 long credit;
David Vrabel55b3da92015-06-24 15:58:42 +0100290 struct resource *resource;
Daniel Kiper080e2be2011-07-25 17:12:06 -0700291 int nid, rc;
David Vrabel55b3da92015-06-24 15:58:42 +0100292 unsigned long balloon_hotplug;
Daniel Kiper080e2be2011-07-25 17:12:06 -0700293
David Vrabel1cf6a6c2015-06-25 16:29:18 +0100294 credit = balloon_stats.target_pages + balloon_stats.target_unpopulated
295 - balloon_stats.total_pages;
David Vrabelb2ac6aa2015-06-25 12:10:28 +0100296
297 /*
298 * Already hotplugged enough pages? Wait for them to be
299 * onlined.
300 */
301 if (credit <= 0)
302 return BP_WAIT;
303
David Vrabel55b3da92015-06-24 15:58:42 +0100304 balloon_hotplug = round_up(credit, PAGES_PER_SECTION);
305
306 resource = additional_memory_resource(balloon_hotplug * PAGE_SIZE);
307 if (!resource)
308 goto err;
309
310 nid = memory_add_physaddr_to_nid(resource->start);
Daniel Kiper080e2be2011-07-25 17:12:06 -0700311
Juergen Gross3c56b3a12015-03-20 13:55:39 +0100312#ifdef CONFIG_XEN_HAVE_PVMMU
Julien Grall30756c62015-05-04 15:39:08 +0100313 /*
314 * We don't support PV MMU when Linux and Xen is using
315 * different page granularity.
316 */
317 BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
318
Juergen Gross3c56b3a12015-03-20 13:55:39 +0100319 /*
320 * add_memory() will build page tables for the new memory so
321 * the p2m must contain invalid entries so the correct
322 * non-present PTEs will be written.
323 *
324 * If a failure occurs, the original (identity) p2m entries
325 * are not restored since this region is now known not to
326 * conflict with any devices.
327 */
328 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
329 unsigned long pfn, i;
330
David Vrabel55b3da92015-06-24 15:58:42 +0100331 pfn = PFN_DOWN(resource->start);
Juergen Gross3c56b3a12015-03-20 13:55:39 +0100332 for (i = 0; i < balloon_hotplug; i++) {
333 if (!set_phys_to_machine(pfn + i, INVALID_P2M_ENTRY)) {
334 pr_warn("set_phys_to_machine() failed, no memory added\n");
David Vrabel55b3da92015-06-24 15:58:42 +0100335 goto err;
Juergen Gross3c56b3a12015-03-20 13:55:39 +0100336 }
337 }
338 }
339#endif
340
Vitaly Kuznetsov703fc132016-03-15 14:56:52 -0700341 /*
342 * add_memory_resource() will call online_pages() which in its turn
343 * will call xen_online_page() callback causing deadlock if we don't
344 * release balloon_mutex here. Unlocking here is safe because the
345 * callers drop the mutex before trying again.
346 */
347 mutex_unlock(&balloon_mutex);
David Hildenbrand8df1d0e2018-10-30 15:10:24 -0700348 /* add_memory_resource() requires the device_hotplug lock */
349 lock_device_hotplug();
David Hildenbrandf29d8e9c02018-12-28 00:35:36 -0800350 rc = add_memory_resource(nid, resource);
David Hildenbrand8df1d0e2018-10-30 15:10:24 -0700351 unlock_device_hotplug();
Vitaly Kuznetsov703fc132016-03-15 14:56:52 -0700352 mutex_lock(&balloon_mutex);
353
Daniel Kiper080e2be2011-07-25 17:12:06 -0700354 if (rc) {
David Vrabel3dcf6362014-09-01 18:52:44 +0100355 pr_warn("Cannot add additional memory (%i)\n", rc);
David Vrabel55b3da92015-06-24 15:58:42 +0100356 goto err;
Daniel Kiper080e2be2011-07-25 17:12:06 -0700357 }
358
David Vrabelde5a77d2015-06-25 12:08:20 +0100359 balloon_stats.total_pages += balloon_hotplug;
Daniel Kiper080e2be2011-07-25 17:12:06 -0700360
David Vrabelb2ac6aa2015-06-25 12:10:28 +0100361 return BP_WAIT;
David Vrabel55b3da92015-06-24 15:58:42 +0100362 err:
363 release_memory_resource(resource);
364 return BP_ECANCELED;
Daniel Kiper080e2be2011-07-25 17:12:06 -0700365}
366
Arun KSa9cd4102019-03-05 15:42:14 -0800367static void xen_online_page(struct page *page, unsigned int order)
Daniel Kiper080e2be2011-07-25 17:12:06 -0700368{
Arun KSa9cd4102019-03-05 15:42:14 -0800369 unsigned long i, size = (1 << order);
370 unsigned long start_pfn = page_to_pfn(page);
371 struct page *p;
Daniel Kiper080e2be2011-07-25 17:12:06 -0700372
Arun KSa9cd4102019-03-05 15:42:14 -0800373 pr_debug("Online %lu pages starting at pfn 0x%lx\n", size, start_pfn);
Daniel Kiper080e2be2011-07-25 17:12:06 -0700374 mutex_lock(&balloon_mutex);
Arun KSa9cd4102019-03-05 15:42:14 -0800375 for (i = 0; i < size; i++) {
376 p = pfn_to_page(start_pfn + i);
377 __online_page_set_limits(p);
David Hildenbranddde32852019-10-01 11:01:50 +0200378 balloon_append(p);
Arun KSa9cd4102019-03-05 15:42:14 -0800379 }
Daniel Kiper080e2be2011-07-25 17:12:06 -0700380 mutex_unlock(&balloon_mutex);
381}
382
383static int xen_memory_notifier(struct notifier_block *nb, unsigned long val, void *v)
384{
385 if (val == MEM_ONLINE)
386 schedule_delayed_work(&balloon_worker, 0);
387
388 return NOTIFY_OK;
389}
390
391static struct notifier_block xen_memory_nb = {
392 .notifier_call = xen_memory_notifier,
393 .priority = 0
394};
395#else
David Vrabelb2ac6aa2015-06-25 12:10:28 +0100396static enum bp_state reserve_additional_memory(void)
Daniel Kiper080e2be2011-07-25 17:12:06 -0700397{
398 balloon_stats.target_pages = balloon_stats.current_pages;
David Vrabel1cf6a6c2015-06-25 16:29:18 +0100399 return BP_ECANCELED;
Daniel Kiper080e2be2011-07-25 17:12:06 -0700400}
401#endif /* CONFIG_XEN_BALLOON_MEMORY_HOTPLUG */
402
David Vrabelde5a77d2015-06-25 12:08:20 +0100403static long current_credit(void)
404{
405 return balloon_stats.target_pages - balloon_stats.current_pages;
406}
407
408static bool balloon_is_inflated(void)
409{
410 return balloon_stats.balloon_low || balloon_stats.balloon_high;
411}
412
Daniel Kiper95d2ac42011-03-08 22:48:24 +0100413static enum bp_state increase_reservation(unsigned long nr_pages)
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700414{
Daniel Kiper95d2ac42011-03-08 22:48:24 +0100415 int rc;
Julien Grall30756c62015-05-04 15:39:08 +0100416 unsigned long i;
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700417 struct page *page;
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700418
419 if (nr_pages > ARRAY_SIZE(frame_list))
420 nr_pages = ARRAY_SIZE(frame_list);
421
Jie Liu9346c2a2013-11-13 20:59:58 +0800422 page = list_first_entry_or_null(&ballooned_pages, struct page, lru);
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700423 for (i = 0; i < nr_pages; i++) {
Daniel Kiper95d2ac42011-03-08 22:48:24 +0100424 if (!page) {
425 nr_pages = i;
426 break;
427 }
Julien Grall30756c62015-05-04 15:39:08 +0100428
Julien Grall30756c62015-05-04 15:39:08 +0100429 frame_list[i] = page_to_xen_pfn(page);
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700430 page = balloon_next_page(page);
431 }
432
Oleksandr Andrushchenkoae4c51a2018-07-20 12:01:44 +0300433 rc = xenmem_reservation_increase(nr_pages, frame_list);
Konrad Rzeszutek Wilk40095de2011-03-14 11:42:40 -0400434 if (rc <= 0)
Daniel Kiper95d2ac42011-03-08 22:48:24 +0100435 return BP_EAGAIN;
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700436
Ian Campbellbc2c0302009-06-05 11:58:37 +0100437 for (i = 0; i < rc; i++) {
Konrad Rzeszutek Wilkb6f30672011-03-15 10:23:57 -0400438 page = balloon_retrieve(false);
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700439 BUG_ON(page == NULL);
440
Oleksandr Andrushchenkoae4c51a2018-07-20 12:01:44 +0300441 xenmem_reservation_va_mapping_update(1, &page, &frame_list[i]);
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700442
443 /* Relinquish the page back to the allocator. */
Ross Lagerwall709613a2016-12-09 17:10:22 +0000444 free_reserved_page(page);
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700445 }
446
Ian Campbellbc2c0302009-06-05 11:58:37 +0100447 balloon_stats.current_pages += rc;
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700448
Daniel Kiper95d2ac42011-03-08 22:48:24 +0100449 return BP_DONE;
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700450}
451
Konrad Rzeszutek Wilkb6f30672011-03-15 10:23:57 -0400452static enum bp_state decrease_reservation(unsigned long nr_pages, gfp_t gfp)
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700453{
Daniel Kiper95d2ac42011-03-08 22:48:24 +0100454 enum bp_state state = BP_DONE;
Julien Grall30756c62015-05-04 15:39:08 +0100455 unsigned long i;
456 struct page *page, *tmp;
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700457 int ret;
Julien Grall30756c62015-05-04 15:39:08 +0100458 LIST_HEAD(pages);
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700459
460 if (nr_pages > ARRAY_SIZE(frame_list))
461 nr_pages = ARRAY_SIZE(frame_list);
462
463 for (i = 0; i < nr_pages; i++) {
Lisa Nguyenfce92682013-05-15 22:59:40 -0700464 page = alloc_page(gfp);
465 if (page == NULL) {
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700466 nr_pages = i;
Daniel Kiper95d2ac42011-03-08 22:48:24 +0100467 state = BP_EAGAIN;
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700468 break;
469 }
Ross Lagerwall709613a2016-12-09 17:10:22 +0000470 adjust_managed_page_count(page, -1);
Oleksandr Andrushchenkoae4c51a2018-07-20 12:01:44 +0300471 xenmem_reservation_scrub_page(page);
Julien Grall30756c62015-05-04 15:39:08 +0100472 list_add(&page->lru, &pages);
Wei Liu09ed3d52014-03-15 16:11:47 +0000473 }
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
Julien Grall30756c62015-05-04 15:39:08 +0100484 /*
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) {
Julien Grall30756c62015-05-04 15:39:08 +0100490 frame_list[i++] = xen_page_to_gfn(page);
Wei Liu09ed3d52014-03-15 16:11:47 +0000491
Oleksandr Andrushchenkoae4c51a2018-07-20 12:01:44 +0300492 xenmem_reservation_va_mapping_reset(1, &page);
Julien Grall30756c62015-05-04 15:39:08 +0100493
Julien Grall30756c62015-05-04 15:39:08 +0100494 list_del(&page->lru);
David Vrabel24f69372013-09-19 17:14:53 +0100495
Wei Liu09ed3d52014-03-15 16:11:47 +0000496 balloon_append(page);
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700497 }
498
David Vrabel24f69372013-09-19 17:14:53 +0100499 flush_tlb_all();
David Vrabel2bad07c2013-09-11 17:45:44 +0000500
Oleksandr Andrushchenkoae4c51a2018-07-20 12:01:44 +0300501 ret = xenmem_reservation_decrease(nr_pages, frame_list);
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700502 BUG_ON(ret != nr_pages);
503
504 balloon_stats.current_pages -= nr_pages;
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700505
Daniel Kiper95d2ac42011-03-08 22:48:24 +0100506 return state;
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700507}
508
509/*
Juergen Gross929423f2015-07-20 13:49:39 +0200510 * As this is a work item it is guaranteed to run as a single instance only.
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700511 * 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 */
515static void balloon_process(struct work_struct *work)
516{
Daniel Kiper95d2ac42011-03-08 22:48:24 +0100517 enum bp_state state = BP_DONE;
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700518 long credit;
519
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700520
521 do {
Juergen Gross929423f2015-07-20 13:49:39 +0200522 mutex_lock(&balloon_mutex);
523
Daniel Kiper83be7e52011-03-28 11:34:10 +0200524 credit = current_credit();
Daniel Kiper95d2ac42011-03-08 22:48:24 +0100525
Daniel Kiper080e2be2011-07-25 17:12:06 -0700526 if (credit > 0) {
527 if (balloon_is_inflated())
528 state = increase_reservation(credit);
529 else
David Vrabelb2ac6aa2015-06-25 12:10:28 +0100530 state = reserve_additional_memory();
Daniel Kiper080e2be2011-07-25 17:12:06 -0700531 }
Daniel Kiper95d2ac42011-03-08 22:48:24 +0100532
Juergen Grossa1078e82019-06-19 11:00:56 +0200533 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 }
Daniel Kiper95d2ac42011-03-08 22:48:24 +0100542
543 state = update_schedule(state);
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700544
Juergen Gross929423f2015-07-20 13:49:39 +0200545 mutex_unlock(&balloon_mutex);
546
547 cond_resched();
548
Daniel Kiper95d2ac42011-03-08 22:48:24 +0100549 } while (credit && state == BP_DONE);
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700550
551 /* Schedule more work if there is some still to be done. */
Daniel Kiper95d2ac42011-03-08 22:48:24 +0100552 if (state == BP_EAGAIN)
553 schedule_delayed_work(&balloon_worker, balloon_stats.schedule_delay * HZ);
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700554}
555
556/* Resets the Xen limit, sets new target, and kicks off processing. */
Daniel De Graaf803eb042011-03-14 11:29:37 -0400557void balloon_set_new_target(unsigned long target)
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700558{
559 /* No need for lock. Not read-modify-write updates. */
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700560 balloon_stats.target_pages = target;
Daniel Kiper95170b22011-03-08 22:47:39 +0100561 schedule_delayed_work(&balloon_worker, 0);
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700562}
Daniel De Graaf803eb042011-03-14 11:29:37 -0400563EXPORT_SYMBOL_GPL(balloon_set_new_target);
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700564
David Vrabel1cf6a6c2015-06-25 16:29:18 +0100565static 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
Juergen Grossa1078e82019-06-19 11:00:56 +0200580 if (si_mem_available() < nr_pages)
581 return -ENOMEM;
582
David Vrabel1cf6a6c2015-06-25 16:29:18 +0100583 st = decrease_reservation(nr_pages, GFP_USER);
584 if (st != BP_DONE)
585 return -ENOMEM;
586
587 return 0;
588}
589
Konrad Rzeszutek Wilkb6f30672011-03-15 10:23:57 -0400590/**
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 */
David Vrabel81b286e2015-06-25 13:12:46 +0100596int alloc_xenballooned_pages(int nr_pages, struct page **pages)
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700597{
Konrad Rzeszutek Wilkb6f30672011-03-15 10:23:57 -0400598 int pgno = 0;
Ruslan Pisareve882dc92011-07-26 14:15:59 +0300599 struct page *page;
David Vrabel1cf6a6c2015-06-25 16:29:18 +0100600 int ret;
601
Konrad Rzeszutek Wilkb6f30672011-03-15 10:23:57 -0400602 mutex_lock(&balloon_mutex);
David Vrabel1cf6a6c2015-06-25 16:29:18 +0100603
604 balloon_stats.target_unpopulated += nr_pages;
605
Konrad Rzeszutek Wilkb6f30672011-03-15 10:23:57 -0400606 while (pgno < nr_pages) {
David Vrabel81b286e2015-06-25 13:12:46 +0100607 page = balloon_retrieve(true);
608 if (page) {
Konrad Rzeszutek Wilkb6f30672011-03-15 10:23:57 -0400609 pages[pgno++] = page;
David Vrabel4a69c902015-07-22 14:50:37 +0100610#ifdef CONFIG_XEN_HAVE_PVMMU
Julien Grall30756c62015-05-04 15:39:08 +0100611 /*
612 * We don't support PV MMU when Linux and Xen is using
613 * different page granularity.
614 */
615 BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE);
616
Boris Ostrovskyb194da22017-08-23 18:27:04 -0400617 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
618 ret = xen_alloc_p2m_entry(page_to_pfn(page));
619 if (ret < 0)
620 goto out_undo;
621 }
David Vrabel4a69c902015-07-22 14:50:37 +0100622#endif
Konrad Rzeszutek Wilkb6f30672011-03-15 10:23:57 -0400623 } else {
David Vrabel1cf6a6c2015-06-25 16:29:18 +0100624 ret = add_ballooned_pages(nr_pages - pgno);
625 if (ret < 0)
Konrad Rzeszutek Wilkb6f30672011-03-15 10:23:57 -0400626 goto out_undo;
627 }
628 }
629 mutex_unlock(&balloon_mutex);
630 return 0;
631 out_undo:
Konrad Rzeszutek Wilkb6f30672011-03-15 10:23:57 -0400632 mutex_unlock(&balloon_mutex);
David Vrabel1cf6a6c2015-06-25 16:29:18 +0100633 free_xenballooned_pages(pgno, pages);
634 return ret;
Konrad Rzeszutek Wilkb6f30672011-03-15 10:23:57 -0400635}
636EXPORT_SYMBOL(alloc_xenballooned_pages);
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700637
Konrad Rzeszutek Wilkb6f30672011-03-15 10:23:57 -0400638/**
639 * free_xenballooned_pages - return pages retrieved with get_ballooned_pages
640 * @nr_pages: Number of pages
641 * @pages: pages to return
642 */
Ruslan Pisareve882dc92011-07-26 14:15:59 +0300643void free_xenballooned_pages(int nr_pages, struct page **pages)
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700644{
Konrad Rzeszutek Wilkb6f30672011-03-15 10:23:57 -0400645 int i;
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700646
Konrad Rzeszutek Wilkb6f30672011-03-15 10:23:57 -0400647 mutex_lock(&balloon_mutex);
648
649 for (i = 0; i < nr_pages; i++) {
David Hildenbrand59b52f12019-10-01 11:01:51 +0200650 if (pages[i])
Konrad Rzeszutek Wilkb6f30672011-03-15 10:23:57 -0400651 balloon_append(pages[i]);
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700652 }
653
David Vrabel1cf6a6c2015-06-25 16:29:18 +0100654 balloon_stats.target_unpopulated -= nr_pages;
655
Konrad Rzeszutek Wilkb6f30672011-03-15 10:23:57 -0400656 /* The balloon may be too large now. Shrink it if needed. */
Daniel Kiper83be7e52011-03-28 11:34:10 +0200657 if (current_credit())
Konrad Rzeszutek Wilkb6f30672011-03-15 10:23:57 -0400658 schedule_delayed_work(&balloon_worker, 0);
659
660 mutex_unlock(&balloon_mutex);
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700661}
Konrad Rzeszutek Wilkb6f30672011-03-15 10:23:57 -0400662EXPORT_SYMBOL(free_xenballooned_pages);
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700663
Vitaly Kuznetsov4fee9ad2017-03-14 18:35:53 +0100664#ifdef CONFIG_XEN_PV
David Vrabel8b5d44a2011-09-28 17:46:34 +0100665static void __init balloon_add_region(unsigned long start_pfn,
666 unsigned long pages)
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700667{
Daniel Kiper4dfe22f2011-03-28 11:33:18 +0200668 unsigned long pfn, extra_pfn_end;
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700669
David Vrabel8b5d44a2011-09-28 17:46:34 +0100670 /*
671 * If the amount of usable memory has been limited (e.g., with
672 * the 'mem' command line parameter), don't add pages beyond
673 * this limit.
674 */
675 extra_pfn_end = min(max_pfn, start_pfn + pages);
676
677 for (pfn = start_pfn; pfn < extra_pfn_end; pfn++) {
David Vrabel8b5d44a2011-09-28 17:46:34 +0100678 /* totalram_pages and totalhigh_pages do not
679 include the boot-time balloon extension, so
680 don't subtract from it. */
David Hildenbrand59b52f12019-10-01 11:01:51 +0200681 balloon_append(pfn_to_page(pfn));
David Vrabel8b5d44a2011-09-28 17:46:34 +0100682 }
David Vrabelde5a77d2015-06-25 12:08:20 +0100683
684 balloon_stats.total_pages += extra_pfn_end - start_pfn;
David Vrabel8b5d44a2011-09-28 17:46:34 +0100685}
Vitaly Kuznetsov4fee9ad2017-03-14 18:35:53 +0100686#endif
David Vrabel8b5d44a2011-09-28 17:46:34 +0100687
688static int __init balloon_init(void)
689{
Stefano Stabellini53d55222010-12-02 17:55:05 +0000690 if (!xen_domain())
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700691 return -ENODEV;
692
Joe Perches283c0972013-06-28 03:21:41 -0700693 pr_info("Initialising balloon driver\n");
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700694
Vitaly Kuznetsov4fee9ad2017-03-14 18:35:53 +0100695#ifdef CONFIG_XEN_PV
David Vrabelaa244112011-09-28 17:46:32 +0100696 balloon_stats.current_pages = xen_pv_domain()
697 ? min(xen_start_info->nr_pages - xen_released_pages, max_pfn)
Boris Ostrovskyc275a572013-11-06 15:37:40 -0500698 : get_num_physpages();
Vitaly Kuznetsov4fee9ad2017-03-14 18:35:53 +0100699#else
700 balloon_stats.current_pages = get_num_physpages();
701#endif
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700702 balloon_stats.target_pages = balloon_stats.current_pages;
703 balloon_stats.balloon_low = 0;
704 balloon_stats.balloon_high = 0;
David Vrabelde5a77d2015-06-25 12:08:20 +0100705 balloon_stats.total_pages = balloon_stats.current_pages;
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700706
Daniel Kiper95d2ac42011-03-08 22:48:24 +0100707 balloon_stats.schedule_delay = 1;
708 balloon_stats.max_schedule_delay = 32;
709 balloon_stats.retry_count = 1;
Juergen Grossa1078e82019-06-19 11:00:56 +0200710 balloon_stats.max_retry_count = 4;
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700711
Daniel Kiper080e2be2011-07-25 17:12:06 -0700712#ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG
Daniel Kiper080e2be2011-07-25 17:12:06 -0700713 set_online_page_callback(&xen_online_page);
714 register_memory_notifier(&xen_memory_nb);
David Vrabel1cf6a6c2015-06-25 16:29:18 +0100715 register_sysctl_table(xen_root);
Daniel Kiper080e2be2011-07-25 17:12:06 -0700716#endif
717
Vitaly Kuznetsov4fee9ad2017-03-14 18:35:53 +0100718#ifdef CONFIG_XEN_PV
719 {
720 int i;
721
722 /*
723 * Initialize the balloon with pages from the extra memory
724 * regions (see arch/x86/xen/setup.c).
725 */
726 for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++)
727 if (xen_extra_mem[i].n_pfns)
728 balloon_add_region(xen_extra_mem[i].start_pfn,
729 xen_extra_mem[i].n_pfns);
730 }
731#endif
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700732
Juergen Gross96edd612017-07-10 10:10:45 +0200733 /* Init the xen-balloon driver. */
734 xen_balloon_init();
735
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700736 return 0;
737}
Jeremy Fitzhardinge17758262008-04-02 10:54:13 -0700738subsys_initcall(balloon_init);