]>
Commit | Line | Data |
---|---|---|
1775826c | 1 | /****************************************************************************** |
1775826c JF |
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 | |
080e2be7 DK |
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. | |
1775826c JF |
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 | ||
283c0972 JP |
39 | #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt |
40 | ||
cd9151e2 | 41 | #include <linux/cpu.h> |
1775826c | 42 | #include <linux/kernel.h> |
1775826c | 43 | #include <linux/sched.h> |
5b825c3a | 44 | #include <linux/cred.h> |
1775826c JF |
45 | #include <linux/errno.h> |
46 | #include <linux/mm.h> | |
57c8a661 | 47 | #include <linux/memblock.h> |
1775826c JF |
48 | #include <linux/pagemap.h> |
49 | #include <linux/highmem.h> | |
50 | #include <linux/mutex.h> | |
1775826c | 51 | #include <linux/list.h> |
5a0e3ad6 | 52 | #include <linux/gfp.h> |
080e2be7 DK |
53 | #include <linux/notifier.h> |
54 | #include <linux/memory.h> | |
55 | #include <linux/memory_hotplug.h> | |
cd9151e2 | 56 | #include <linux/percpu-defs.h> |
55b3da98 | 57 | #include <linux/slab.h> |
1cf6a6c8 | 58 | #include <linux/sysctl.h> |
1775826c | 59 | |
1775826c JF |
60 | #include <asm/page.h> |
61 | #include <asm/pgalloc.h> | |
62 | #include <asm/pgtable.h> | |
1775826c JF |
63 | #include <asm/tlb.h> |
64 | ||
ecbf29cd JF |
65 | #include <asm/xen/hypervisor.h> |
66 | #include <asm/xen/hypercall.h> | |
1ccbf534 JF |
67 | |
68 | #include <xen/xen.h> | |
ecbf29cd | 69 | #include <xen/interface/xen.h> |
1775826c | 70 | #include <xen/interface/memory.h> |
803eb047 | 71 | #include <xen/balloon.h> |
1775826c JF |
72 | #include <xen/features.h> |
73 | #include <xen/page.h> | |
ae4c51a5 | 74 | #include <xen/mem-reservation.h> |
1775826c | 75 | |
1cf6a6c8 DV |
76 | static int xen_hotplug_unpopulated; |
77 | ||
78 | #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG | |
79 | ||
1cf6a6c8 DV |
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, | |
eec4844f MC |
87 | .extra1 = SYSCTL_ZERO, |
88 | .extra2 = SYSCTL_ONE, | |
1cf6a6c8 DV |
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 | ||
30756c62 JG |
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 | ||
95d2ac4a DK |
119 | /* |
120 | * balloon_process() state: | |
121 | * | |
122 | * BP_DONE: done or nothing to do, | |
b2ac6aa8 | 123 | * BP_WAIT: wait to be rescheduled, |
95d2ac4a DK |
124 | * BP_EAGAIN: error, go to sleep, |
125 | * BP_ECANCELED: error, balloon operation canceled. | |
126 | */ | |
1775826c | 127 | |
95d2ac4a DK |
128 | enum bp_state { |
129 | BP_DONE, | |
b2ac6aa8 | 130 | BP_WAIT, |
95d2ac4a DK |
131 | BP_EAGAIN, |
132 | BP_ECANCELED | |
1775826c JF |
133 | }; |
134 | ||
1775826c | 135 | |
1775826c | 136 | static DEFINE_MUTEX(balloon_mutex); |
1775826c | 137 | |
803eb047 DDG |
138 | struct balloon_stats balloon_stats; |
139 | EXPORT_SYMBOL_GPL(balloon_stats); | |
1775826c JF |
140 | |
141 | /* We increase/decrease in batches which fit in a page */ | |
3990dd27 | 142 | static xen_pfn_t frame_list[PAGE_SIZE / sizeof(xen_pfn_t)]; |
cd9151e2 | 143 | |
1775826c | 144 | |
1775826c JF |
145 | /* List of ballooned pages, threaded through the mem_map array. */ |
146 | static LIST_HEAD(ballooned_pages); | |
1cf6a6c8 | 147 | static DECLARE_WAIT_QUEUE_HEAD(balloon_wq); |
1775826c JF |
148 | |
149 | /* Main work function, always executed in process context. */ | |
150 | static void balloon_process(struct work_struct *work); | |
95170b2e | 151 | static DECLARE_DELAYED_WORK(balloon_worker, balloon_process); |
1775826c JF |
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 | ||
1775826c | 158 | /* balloon_append: add the given page to the balloon. */ |
dde3285f | 159 | static void balloon_append(struct page *page) |
1775826c | 160 | { |
59b52f10 DH |
161 | __SetPageOffline(page); |
162 | ||
1775826c JF |
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++; | |
1775826c JF |
167 | } else { |
168 | list_add(&page->lru, &ballooned_pages); | |
169 | balloon_stats.balloon_low++; | |
170 | } | |
1cf6a6c8 | 171 | wake_up(&balloon_wq); |
9be4d457 | 172 | } |
3d65c948 | 173 | |
1775826c | 174 | /* balloon_retrieve: rescue a page from the balloon, if it is not empty. */ |
81b286e0 | 175 | static struct page *balloon_retrieve(bool require_lowmem) |
1775826c JF |
176 | { |
177 | struct page *page; | |
178 | ||
179 | if (list_empty(&ballooned_pages)) | |
180 | return NULL; | |
181 | ||
81b286e0 DV |
182 | page = list_entry(ballooned_pages.next, struct page, lru); |
183 | if (require_lowmem && PageHighMem(page)) | |
184 | return NULL; | |
1775826c JF |
185 | list_del(&page->lru); |
186 | ||
3dcc0571 | 187 | if (PageHighMem(page)) |
1775826c | 188 | balloon_stats.balloon_high--; |
3dcc0571 | 189 | else |
1775826c JF |
190 | balloon_stats.balloon_low--; |
191 | ||
192 | return page; | |
193 | } | |
194 | ||
1775826c JF |
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 | ||
95d2ac4a | 203 | static enum bp_state update_schedule(enum bp_state state) |
1775826c | 204 | { |
b2ac6aa8 DV |
205 | if (state == BP_WAIT) |
206 | return BP_WAIT; | |
207 | ||
fd8b7951 BO |
208 | if (state == BP_ECANCELED) |
209 | return BP_ECANCELED; | |
210 | ||
95d2ac4a DK |
211 | if (state == BP_DONE) { |
212 | balloon_stats.schedule_delay = 1; | |
213 | balloon_stats.retry_count = 1; | |
214 | return BP_DONE; | |
215 | } | |
216 | ||
95d2ac4a DK |
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) { | |
95d2ac4a DK |
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; | |
1775826c JF |
232 | } |
233 | ||
080e2be7 | 234 | #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG |
842775f1 RL |
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 | ||
55b3da98 DV |
248 | static struct resource *additional_memory_resource(phys_addr_t size) |
249 | { | |
12366410 ID |
250 | struct resource *res; |
251 | int ret; | |
55b3da98 DV |
252 | |
253 | res = kzalloc(sizeof(*res), GFP_KERNEL); | |
254 | if (!res) | |
255 | return NULL; | |
256 | ||
257 | res->name = "System RAM"; | |
782b8664 | 258 | res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY; |
55b3da98 | 259 | |
12366410 ID |
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; | |
55b3da98 DV |
267 | } |
268 | ||
dfd74a1e RL |
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 | ||
55b3da98 DV |
283 | return res; |
284 | } | |
285 | ||
b2ac6aa8 | 286 | static enum bp_state reserve_additional_memory(void) |
080e2be7 | 287 | { |
b2ac6aa8 | 288 | long credit; |
55b3da98 | 289 | struct resource *resource; |
080e2be7 | 290 | int nid, rc; |
55b3da98 | 291 | unsigned long balloon_hotplug; |
080e2be7 | 292 | |
1cf6a6c8 DV |
293 | credit = balloon_stats.target_pages + balloon_stats.target_unpopulated |
294 | - balloon_stats.total_pages; | |
b2ac6aa8 DV |
295 | |
296 | /* | |
297 | * Already hotplugged enough pages? Wait for them to be | |
298 | * onlined. | |
299 | */ | |
300 | if (credit <= 0) | |
301 | return BP_WAIT; | |
302 | ||
55b3da98 DV |
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); | |
080e2be7 | 310 | |
3c56b3a1 | 311 | #ifdef CONFIG_XEN_HAVE_PVMMU |
30756c62 JG |
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 | ||
3c56b3a1 JG |
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 | ||
55b3da98 | 330 | pfn = PFN_DOWN(resource->start); |
3c56b3a1 JG |
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"); | |
55b3da98 | 334 | goto err; |
3c56b3a1 JG |
335 | } |
336 | } | |
337 | } | |
338 | #endif | |
339 | ||
703fc13a VK |
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); | |
8df1d0e4 DH |
347 | /* add_memory_resource() requires the device_hotplug lock */ |
348 | lock_device_hotplug(); | |
f29d8e9c | 349 | rc = add_memory_resource(nid, resource); |
8df1d0e4 | 350 | unlock_device_hotplug(); |
703fc13a VK |
351 | mutex_lock(&balloon_mutex); |
352 | ||
080e2be7 | 353 | if (rc) { |
3dcf6367 | 354 | pr_warn("Cannot add additional memory (%i)\n", rc); |
55b3da98 | 355 | goto err; |
080e2be7 DK |
356 | } |
357 | ||
de5a77d8 | 358 | balloon_stats.total_pages += balloon_hotplug; |
080e2be7 | 359 | |
b2ac6aa8 | 360 | return BP_WAIT; |
55b3da98 DV |
361 | err: |
362 | release_memory_resource(resource); | |
363 | return BP_ECANCELED; | |
080e2be7 DK |
364 | } |
365 | ||
a9cd410a | 366 | static void xen_online_page(struct page *page, unsigned int order) |
080e2be7 | 367 | { |
a9cd410a AK |
368 | unsigned long i, size = (1 << order); |
369 | unsigned long start_pfn = page_to_pfn(page); | |
370 | struct page *p; | |
080e2be7 | 371 | |
a9cd410a | 372 | pr_debug("Online %lu pages starting at pfn 0x%lx\n", size, start_pfn); |
080e2be7 | 373 | mutex_lock(&balloon_mutex); |
a9cd410a AK |
374 | for (i = 0; i < size; i++) { |
375 | p = pfn_to_page(start_pfn + i); | |
376 | __online_page_set_limits(p); | |
dde3285f | 377 | balloon_append(p); |
a9cd410a | 378 | } |
080e2be7 DK |
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 | |
b2ac6aa8 | 395 | static enum bp_state reserve_additional_memory(void) |
1775826c | 396 | { |
de5a77d8 | 397 | balloon_stats.target_pages = balloon_stats.current_pages; |
1cf6a6c8 | 398 | return BP_ECANCELED; |
1775826c | 399 | } |
de5a77d8 | 400 | #endif /* CONFIG_XEN_BALLOON_MEMORY_HOTPLUG */ |
1775826c | 401 | |
de5a77d8 | 402 | static long current_credit(void) |
080e2be7 | 403 | { |
de5a77d8 | 404 | return balloon_stats.target_pages - balloon_stats.current_pages; |
080e2be7 DK |
405 | } |
406 | ||
de5a77d8 | 407 | static bool balloon_is_inflated(void) |
080e2be7 | 408 | { |
de5a77d8 | 409 | return balloon_stats.balloon_low || balloon_stats.balloon_high; |
080e2be7 | 410 | } |
080e2be7 | 411 | |
95d2ac4a | 412 | static enum bp_state increase_reservation(unsigned long nr_pages) |
1775826c | 413 | { |
95d2ac4a | 414 | int rc; |
30756c62 | 415 | unsigned long i; |
1775826c | 416 | struct page *page; |
1775826c JF |
417 | |
418 | if (nr_pages > ARRAY_SIZE(frame_list)) | |
419 | nr_pages = ARRAY_SIZE(frame_list); | |
420 | ||
9346c2a8 | 421 | page = list_first_entry_or_null(&ballooned_pages, struct page, lru); |
1775826c | 422 | for (i = 0; i < nr_pages; i++) { |
95d2ac4a DK |
423 | if (!page) { |
424 | nr_pages = i; | |
425 | break; | |
426 | } | |
30756c62 | 427 | |
30756c62 | 428 | frame_list[i] = page_to_xen_pfn(page); |
1775826c JF |
429 | page = balloon_next_page(page); |
430 | } | |
431 | ||
ae4c51a5 | 432 | rc = xenmem_reservation_increase(nr_pages, frame_list); |
40095de1 | 433 | if (rc <= 0) |
95d2ac4a | 434 | return BP_EAGAIN; |
1775826c | 435 | |
bc2c0303 | 436 | for (i = 0; i < rc; i++) { |
b6f30679 | 437 | page = balloon_retrieve(false); |
1775826c JF |
438 | BUG_ON(page == NULL); |
439 | ||
ae4c51a5 | 440 | xenmem_reservation_va_mapping_update(1, &page, &frame_list[i]); |
1775826c JF |
441 | |
442 | /* Relinquish the page back to the allocator. */ | |
77c4adf6 | 443 | __ClearPageOffline(page); |
709613ad | 444 | free_reserved_page(page); |
1775826c JF |
445 | } |
446 | ||
bc2c0303 | 447 | balloon_stats.current_pages += rc; |
1775826c | 448 | |
95d2ac4a | 449 | return BP_DONE; |
1775826c JF |
450 | } |
451 | ||
b6f30679 | 452 | static enum bp_state decrease_reservation(unsigned long nr_pages, gfp_t gfp) |
1775826c | 453 | { |
95d2ac4a | 454 | enum bp_state state = BP_DONE; |
30756c62 JG |
455 | unsigned long i; |
456 | struct page *page, *tmp; | |
1775826c | 457 | int ret; |
30756c62 | 458 | LIST_HEAD(pages); |
1775826c JF |
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++) { | |
fce92683 LN |
464 | page = alloc_page(gfp); |
465 | if (page == NULL) { | |
1775826c | 466 | nr_pages = i; |
95d2ac4a | 467 | state = BP_EAGAIN; |
1775826c JF |
468 | break; |
469 | } | |
709613ad | 470 | adjust_managed_page_count(page, -1); |
ae4c51a5 | 471 | xenmem_reservation_scrub_page(page); |
30756c62 | 472 | list_add(&page->lru, &pages); |
09ed3d5b | 473 | } |
1775826c | 474 | |
09ed3d5b WL |
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 | ||
30756c62 JG |
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) { | |
30756c62 | 490 | frame_list[i++] = xen_page_to_gfn(page); |
1058a75f | 491 | |
ae4c51a5 | 492 | xenmem_reservation_va_mapping_reset(1, &page); |
30756c62 | 493 | |
30756c62 | 494 | list_del(&page->lru); |
24f69373 | 495 | |
09ed3d5b | 496 | balloon_append(page); |
1775826c JF |
497 | } |
498 | ||
24f69373 | 499 | flush_tlb_all(); |
2bad07ce | 500 | |
ae4c51a5 | 501 | ret = xenmem_reservation_decrease(nr_pages, frame_list); |
1775826c JF |
502 | BUG_ON(ret != nr_pages); |
503 | ||
504 | balloon_stats.current_pages -= nr_pages; | |
1775826c | 505 | |
95d2ac4a | 506 | return state; |
1775826c JF |
507 | } |
508 | ||
509 | /* | |
929423fa | 510 | * As this is a work item it is guaranteed to run as a single instance only. |
1775826c JF |
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 | { | |
95d2ac4a | 517 | enum bp_state state = BP_DONE; |
1775826c JF |
518 | long credit; |
519 | ||
1775826c JF |
520 | |
521 | do { | |
929423fa JG |
522 | mutex_lock(&balloon_mutex); |
523 | ||
83be7e52 | 524 | credit = current_credit(); |
95d2ac4a | 525 | |
080e2be7 DK |
526 | if (credit > 0) { |
527 | if (balloon_is_inflated()) | |
528 | state = increase_reservation(credit); | |
529 | else | |
b2ac6aa8 | 530 | state = reserve_additional_memory(); |
080e2be7 | 531 | } |
95d2ac4a | 532 | |
a1078e82 JG |
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 | } | |
95d2ac4a DK |
542 | |
543 | state = update_schedule(state); | |
1775826c | 544 | |
929423fa JG |
545 | mutex_unlock(&balloon_mutex); |
546 | ||
547 | cond_resched(); | |
548 | ||
95d2ac4a | 549 | } while (credit && state == BP_DONE); |
1775826c JF |
550 | |
551 | /* Schedule more work if there is some still to be done. */ | |
95d2ac4a DK |
552 | if (state == BP_EAGAIN) |
553 | schedule_delayed_work(&balloon_worker, balloon_stats.schedule_delay * HZ); | |
1775826c JF |
554 | } |
555 | ||
556 | /* Resets the Xen limit, sets new target, and kicks off processing. */ | |
803eb047 | 557 | void balloon_set_new_target(unsigned long target) |
1775826c JF |
558 | { |
559 | /* No need for lock. Not read-modify-write updates. */ | |
1775826c | 560 | balloon_stats.target_pages = target; |
95170b2e | 561 | schedule_delayed_work(&balloon_worker, 0); |
1775826c | 562 | } |
803eb047 | 563 | EXPORT_SYMBOL_GPL(balloon_set_new_target); |
1775826c | 564 | |
1cf6a6c8 DV |
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 | ||
a1078e82 JG |
580 | if (si_mem_available() < nr_pages) |
581 | return -ENOMEM; | |
582 | ||
1cf6a6c8 DV |
583 | st = decrease_reservation(nr_pages, GFP_USER); |
584 | if (st != BP_DONE) | |
585 | return -ENOMEM; | |
586 | ||
587 | return 0; | |
588 | } | |
589 | ||
b6f30679 KRW |
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 | */ | |
81b286e0 | 596 | int alloc_xenballooned_pages(int nr_pages, struct page **pages) |
1775826c | 597 | { |
b6f30679 | 598 | int pgno = 0; |
e882dc9c | 599 | struct page *page; |
1cf6a6c8 DV |
600 | int ret; |
601 | ||
b6f30679 | 602 | mutex_lock(&balloon_mutex); |
1cf6a6c8 DV |
603 | |
604 | balloon_stats.target_unpopulated += nr_pages; | |
605 | ||
b6f30679 | 606 | while (pgno < nr_pages) { |
81b286e0 DV |
607 | page = balloon_retrieve(true); |
608 | if (page) { | |
0266def9 | 609 | __ClearPageOffline(page); |
b6f30679 | 610 | pages[pgno++] = page; |
4a69c909 | 611 | #ifdef CONFIG_XEN_HAVE_PVMMU |
30756c62 JG |
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 | ||
b194da25 BO |
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 | } | |
4a69c909 | 623 | #endif |
b6f30679 | 624 | } else { |
1cf6a6c8 DV |
625 | ret = add_ballooned_pages(nr_pages - pgno); |
626 | if (ret < 0) | |
b6f30679 KRW |
627 | goto out_undo; |
628 | } | |
1775826c | 629 | } |
b6f30679 KRW |
630 | mutex_unlock(&balloon_mutex); |
631 | return 0; | |
632 | out_undo: | |
b6f30679 | 633 | mutex_unlock(&balloon_mutex); |
1cf6a6c8 DV |
634 | free_xenballooned_pages(pgno, pages); |
635 | return ret; | |
1775826c | 636 | } |
b6f30679 | 637 | EXPORT_SYMBOL(alloc_xenballooned_pages); |
1775826c | 638 | |
b6f30679 KRW |
639 | /** |
640 | * free_xenballooned_pages - return pages retrieved with get_ballooned_pages | |
641 | * @nr_pages: Number of pages | |
642 | * @pages: pages to return | |
643 | */ | |
e882dc9c | 644 | void free_xenballooned_pages(int nr_pages, struct page **pages) |
1775826c | 645 | { |
b6f30679 | 646 | int i; |
1775826c | 647 | |
b6f30679 | 648 | mutex_lock(&balloon_mutex); |
1775826c | 649 | |
b6f30679 | 650 | for (i = 0; i < nr_pages; i++) { |
59b52f10 | 651 | if (pages[i]) |
b6f30679 KRW |
652 | balloon_append(pages[i]); |
653 | } | |
654 | ||
1cf6a6c8 DV |
655 | balloon_stats.target_unpopulated -= nr_pages; |
656 | ||
b6f30679 | 657 | /* The balloon may be too large now. Shrink it if needed. */ |
83be7e52 | 658 | if (current_credit()) |
b6f30679 | 659 | schedule_delayed_work(&balloon_worker, 0); |
1775826c | 660 | |
b6f30679 KRW |
661 | mutex_unlock(&balloon_mutex); |
662 | } | |
663 | EXPORT_SYMBOL(free_xenballooned_pages); | |
1775826c | 664 | |
4fee9ad8 | 665 | #ifdef CONFIG_XEN_PV |
8b5d44a5 DV |
666 | static void __init balloon_add_region(unsigned long start_pfn, |
667 | unsigned long pages) | |
1775826c | 668 | { |
4dfe22f5 | 669 | unsigned long pfn, extra_pfn_end; |
1775826c | 670 | |
8b5d44a5 DV |
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++) { | |
8b5d44a5 DV |
679 | /* totalram_pages and totalhigh_pages do not |
680 | include the boot-time balloon extension, so | |
681 | don't subtract from it. */ | |
59b52f10 | 682 | balloon_append(pfn_to_page(pfn)); |
8b5d44a5 | 683 | } |
de5a77d8 DV |
684 | |
685 | balloon_stats.total_pages += extra_pfn_end - start_pfn; | |
8b5d44a5 | 686 | } |
4fee9ad8 | 687 | #endif |
8b5d44a5 DV |
688 | |
689 | static int __init balloon_init(void) | |
690 | { | |
53d5522c | 691 | if (!xen_domain()) |
1775826c JF |
692 | return -ENODEV; |
693 | ||
283c0972 | 694 | pr_info("Initialising balloon driver\n"); |
1775826c | 695 | |
4fee9ad8 | 696 | #ifdef CONFIG_XEN_PV |
aa24411b DV |
697 | balloon_stats.current_pages = xen_pv_domain() |
698 | ? min(xen_start_info->nr_pages - xen_released_pages, max_pfn) | |
c275a57f | 699 | : get_num_physpages(); |
4fee9ad8 VK |
700 | #else |
701 | balloon_stats.current_pages = get_num_physpages(); | |
702 | #endif | |
1775826c JF |
703 | balloon_stats.target_pages = balloon_stats.current_pages; |
704 | balloon_stats.balloon_low = 0; | |
705 | balloon_stats.balloon_high = 0; | |
de5a77d8 | 706 | balloon_stats.total_pages = balloon_stats.current_pages; |
1775826c | 707 | |
95d2ac4a DK |
708 | balloon_stats.schedule_delay = 1; |
709 | balloon_stats.max_schedule_delay = 32; | |
710 | balloon_stats.retry_count = 1; | |
a1078e82 | 711 | balloon_stats.max_retry_count = 4; |
1775826c | 712 | |
080e2be7 | 713 | #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG |
080e2be7 DK |
714 | set_online_page_callback(&xen_online_page); |
715 | register_memory_notifier(&xen_memory_nb); | |
1cf6a6c8 | 716 | register_sysctl_table(xen_root); |
080e2be7 DK |
717 | #endif |
718 | ||
4fee9ad8 VK |
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 | |
1775826c | 733 | |
96edd61d JG |
734 | /* Init the xen-balloon driver. */ |
735 | xen_balloon_init(); | |
736 | ||
1775826c JF |
737 | return 0; |
738 | } | |
1775826c | 739 | subsys_initcall(balloon_init); |