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 JF |
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> | |
1775826c | 50 | #include <linux/list.h> |
5a0e3ad6 | 51 | #include <linux/gfp.h> |
080e2be7 DK |
52 | #include <linux/notifier.h> |
53 | #include <linux/memory.h> | |
54 | #include <linux/memory_hotplug.h> | |
cd9151e2 | 55 | #include <linux/percpu-defs.h> |
55b3da98 | 56 | #include <linux/slab.h> |
1cf6a6c8 | 57 | #include <linux/sysctl.h> |
1775826c | 58 | |
1775826c JF |
59 | #include <asm/page.h> |
60 | #include <asm/pgalloc.h> | |
61 | #include <asm/pgtable.h> | |
1775826c JF |
62 | #include <asm/tlb.h> |
63 | ||
ecbf29cd JF |
64 | #include <asm/xen/hypervisor.h> |
65 | #include <asm/xen/hypercall.h> | |
1ccbf534 JF |
66 | |
67 | #include <xen/xen.h> | |
ecbf29cd | 68 | #include <xen/interface/xen.h> |
1775826c | 69 | #include <xen/interface/memory.h> |
803eb047 | 70 | #include <xen/balloon.h> |
1775826c JF |
71 | #include <xen/features.h> |
72 | #include <xen/page.h> | |
73 | ||
1cf6a6c8 DV |
74 | static int xen_hotplug_unpopulated; |
75 | ||
76 | #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG | |
77 | ||
78 | static int zero; | |
79 | static int one = 1; | |
80 | ||
81 | static 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 | ||
94 | static struct ctl_table balloon_root[] = { | |
95 | { | |
96 | .procname = "balloon", | |
97 | .mode = 0555, | |
98 | .child = balloon_table, | |
99 | }, | |
100 | { } | |
101 | }; | |
102 | ||
103 | static struct ctl_table xen_root[] = { | |
104 | { | |
105 | .procname = "xen", | |
106 | .mode = 0555, | |
107 | .child = balloon_root, | |
108 | }, | |
109 | { } | |
110 | }; | |
111 | ||
112 | #endif | |
113 | ||
30756c62 JG |
114 | /* |
115 | * Use one extent per PAGE_SIZE to avoid to break down the page into | |
116 | * multiple frame. | |
117 | */ | |
118 | #define EXTENT_ORDER (fls(XEN_PFN_PER_PAGE) - 1) | |
119 | ||
95d2ac4a DK |
120 | /* |
121 | * balloon_process() state: | |
122 | * | |
123 | * BP_DONE: done or nothing to do, | |
b2ac6aa8 | 124 | * BP_WAIT: wait to be rescheduled, |
95d2ac4a DK |
125 | * BP_EAGAIN: error, go to sleep, |
126 | * BP_ECANCELED: error, balloon operation canceled. | |
127 | */ | |
1775826c | 128 | |
95d2ac4a DK |
129 | enum bp_state { |
130 | BP_DONE, | |
b2ac6aa8 | 131 | BP_WAIT, |
95d2ac4a DK |
132 | BP_EAGAIN, |
133 | BP_ECANCELED | |
1775826c JF |
134 | }; |
135 | ||
1775826c | 136 | |
1775826c | 137 | static DEFINE_MUTEX(balloon_mutex); |
1775826c | 138 | |
803eb047 DDG |
139 | struct balloon_stats balloon_stats; |
140 | EXPORT_SYMBOL_GPL(balloon_stats); | |
1775826c JF |
141 | |
142 | /* We increase/decrease in batches which fit in a page */ | |
3990dd27 | 143 | static xen_pfn_t frame_list[PAGE_SIZE / sizeof(xen_pfn_t)]; |
cd9151e2 | 144 | |
1775826c | 145 | |
1775826c JF |
146 | /* List of ballooned pages, threaded through the mem_map array. */ |
147 | static LIST_HEAD(ballooned_pages); | |
1cf6a6c8 | 148 | static DECLARE_WAIT_QUEUE_HEAD(balloon_wq); |
1775826c JF |
149 | |
150 | /* Main work function, always executed in process context. */ | |
151 | static void balloon_process(struct work_struct *work); | |
95170b2e | 152 | static DECLARE_DELAYED_WORK(balloon_worker, balloon_process); |
1775826c JF |
153 | |
154 | /* When ballooning out (allocating memory to return to Xen) we don't really | |
155 | want the kernel to try too hard since that can trigger the oom killer. */ | |
156 | #define GFP_BALLOON \ | |
157 | (GFP_HIGHUSER | __GFP_NOWARN | __GFP_NORETRY | __GFP_NOMEMALLOC) | |
158 | ||
159 | static void scrub_page(struct page *page) | |
160 | { | |
161 | #ifdef CONFIG_XEN_SCRUB_PAGES | |
26a3e991 | 162 | clear_highpage(page); |
1775826c JF |
163 | #endif |
164 | } | |
165 | ||
166 | /* balloon_append: add the given page to the balloon. */ | |
9be4d457 | 167 | static void __balloon_append(struct page *page) |
1775826c JF |
168 | { |
169 | /* Lowmem is re-populated first, so highmem pages go at list tail. */ | |
170 | if (PageHighMem(page)) { | |
171 | list_add_tail(&page->lru, &ballooned_pages); | |
172 | balloon_stats.balloon_high++; | |
1775826c JF |
173 | } else { |
174 | list_add(&page->lru, &ballooned_pages); | |
175 | balloon_stats.balloon_low++; | |
176 | } | |
1cf6a6c8 | 177 | wake_up(&balloon_wq); |
9be4d457 | 178 | } |
3d65c948 | 179 | |
9be4d457 JF |
180 | static void balloon_append(struct page *page) |
181 | { | |
182 | __balloon_append(page); | |
3dcc0571 | 183 | adjust_managed_page_count(page, -1); |
1775826c JF |
184 | } |
185 | ||
186 | /* balloon_retrieve: rescue a page from the balloon, if it is not empty. */ | |
81b286e0 | 187 | static struct page *balloon_retrieve(bool require_lowmem) |
1775826c JF |
188 | { |
189 | struct page *page; | |
190 | ||
191 | if (list_empty(&ballooned_pages)) | |
192 | return NULL; | |
193 | ||
81b286e0 DV |
194 | page = list_entry(ballooned_pages.next, struct page, lru); |
195 | if (require_lowmem && PageHighMem(page)) | |
196 | return NULL; | |
1775826c JF |
197 | list_del(&page->lru); |
198 | ||
3dcc0571 | 199 | if (PageHighMem(page)) |
1775826c | 200 | balloon_stats.balloon_high--; |
3dcc0571 | 201 | else |
1775826c JF |
202 | balloon_stats.balloon_low--; |
203 | ||
3dcc0571 | 204 | adjust_managed_page_count(page, 1); |
3d65c948 | 205 | |
1775826c JF |
206 | return page; |
207 | } | |
208 | ||
1775826c JF |
209 | static 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 | ||
95d2ac4a | 217 | static enum bp_state update_schedule(enum bp_state state) |
1775826c | 218 | { |
b2ac6aa8 DV |
219 | if (state == BP_WAIT) |
220 | return BP_WAIT; | |
221 | ||
fd8b7951 BO |
222 | if (state == BP_ECANCELED) |
223 | return BP_ECANCELED; | |
224 | ||
95d2ac4a DK |
225 | if (state == BP_DONE) { |
226 | balloon_stats.schedule_delay = 1; | |
227 | balloon_stats.retry_count = 1; | |
228 | return BP_DONE; | |
229 | } | |
230 | ||
95d2ac4a DK |
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) { | |
95d2ac4a DK |
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; | |
1775826c JF |
246 | } |
247 | ||
080e2be7 | 248 | #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG |
842775f1 RL |
249 | static void release_memory_resource(struct resource *resource) |
250 | { | |
251 | if (!resource) | |
252 | return; | |
253 | ||
254 | /* | |
255 | * No need to reset region to identity mapped since we now | |
256 | * know that no I/O can be in this region | |
257 | */ | |
258 | release_resource(resource); | |
259 | kfree(resource); | |
260 | } | |
261 | ||
55b3da98 DV |
262 | static 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"; | |
782b8664 | 272 | res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY; |
55b3da98 DV |
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 | ||
dfd74a1e RL |
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 | ||
55b3da98 DV |
297 | return res; |
298 | } | |
299 | ||
b2ac6aa8 | 300 | static enum bp_state reserve_additional_memory(void) |
080e2be7 | 301 | { |
b2ac6aa8 | 302 | long credit; |
55b3da98 | 303 | struct resource *resource; |
080e2be7 | 304 | int nid, rc; |
55b3da98 | 305 | unsigned long balloon_hotplug; |
080e2be7 | 306 | |
1cf6a6c8 DV |
307 | credit = balloon_stats.target_pages + balloon_stats.target_unpopulated |
308 | - balloon_stats.total_pages; | |
b2ac6aa8 DV |
309 | |
310 | /* | |
311 | * Already hotplugged enough pages? Wait for them to be | |
312 | * onlined. | |
313 | */ | |
314 | if (credit <= 0) | |
315 | return BP_WAIT; | |
316 | ||
55b3da98 DV |
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); | |
080e2be7 | 324 | |
3c56b3a1 | 325 | #ifdef CONFIG_XEN_HAVE_PVMMU |
30756c62 JG |
326 | /* |
327 | * We don't support PV MMU when Linux and Xen is using | |
328 | * different page granularity. | |
329 | */ | |
330 | BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE); | |
331 | ||
3c56b3a1 JG |
332 | /* |
333 | * add_memory() will build page tables for the new memory so | |
334 | * the p2m must contain invalid entries so the correct | |
335 | * non-present PTEs will be written. | |
336 | * | |
337 | * If a failure occurs, the original (identity) p2m entries | |
338 | * are not restored since this region is now known not to | |
339 | * conflict with any devices. | |
340 | */ | |
341 | if (!xen_feature(XENFEAT_auto_translated_physmap)) { | |
342 | unsigned long pfn, i; | |
343 | ||
55b3da98 | 344 | pfn = PFN_DOWN(resource->start); |
3c56b3a1 JG |
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"); | |
55b3da98 | 348 | goto err; |
3c56b3a1 JG |
349 | } |
350 | } | |
351 | } | |
352 | #endif | |
353 | ||
703fc13a VK |
354 | /* |
355 | * add_memory_resource() will call online_pages() which in its turn | |
356 | * will call xen_online_page() callback causing deadlock if we don't | |
357 | * release balloon_mutex here. Unlocking here is safe because the | |
358 | * callers drop the mutex before trying again. | |
359 | */ | |
360 | mutex_unlock(&balloon_mutex); | |
361 | rc = add_memory_resource(nid, resource, memhp_auto_online); | |
362 | mutex_lock(&balloon_mutex); | |
363 | ||
080e2be7 | 364 | if (rc) { |
3dcf6367 | 365 | pr_warn("Cannot add additional memory (%i)\n", rc); |
55b3da98 | 366 | goto err; |
080e2be7 DK |
367 | } |
368 | ||
de5a77d8 | 369 | balloon_stats.total_pages += balloon_hotplug; |
080e2be7 | 370 | |
b2ac6aa8 | 371 | return BP_WAIT; |
55b3da98 DV |
372 | err: |
373 | release_memory_resource(resource); | |
374 | return BP_ECANCELED; | |
080e2be7 DK |
375 | } |
376 | ||
377 | static 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 | ||
080e2be7 DK |
385 | mutex_unlock(&balloon_mutex); |
386 | } | |
387 | ||
388 | static 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 | ||
396 | static struct notifier_block xen_memory_nb = { | |
397 | .notifier_call = xen_memory_notifier, | |
398 | .priority = 0 | |
399 | }; | |
400 | #else | |
b2ac6aa8 | 401 | static enum bp_state reserve_additional_memory(void) |
1775826c | 402 | { |
de5a77d8 | 403 | balloon_stats.target_pages = balloon_stats.current_pages; |
1cf6a6c8 | 404 | return BP_ECANCELED; |
1775826c | 405 | } |
de5a77d8 | 406 | #endif /* CONFIG_XEN_BALLOON_MEMORY_HOTPLUG */ |
1775826c | 407 | |
de5a77d8 | 408 | static long current_credit(void) |
080e2be7 | 409 | { |
de5a77d8 | 410 | return balloon_stats.target_pages - balloon_stats.current_pages; |
080e2be7 DK |
411 | } |
412 | ||
de5a77d8 | 413 | static bool balloon_is_inflated(void) |
080e2be7 | 414 | { |
de5a77d8 | 415 | return balloon_stats.balloon_low || balloon_stats.balloon_high; |
080e2be7 | 416 | } |
080e2be7 | 417 | |
95d2ac4a | 418 | static enum bp_state increase_reservation(unsigned long nr_pages) |
1775826c | 419 | { |
95d2ac4a | 420 | int rc; |
30756c62 | 421 | unsigned long i; |
1775826c | 422 | struct page *page; |
1775826c JF |
423 | struct xen_memory_reservation reservation = { |
424 | .address_bits = 0, | |
30756c62 | 425 | .extent_order = EXTENT_ORDER, |
1775826c JF |
426 | .domid = DOMID_SELF |
427 | }; | |
428 | ||
429 | if (nr_pages > ARRAY_SIZE(frame_list)) | |
430 | nr_pages = ARRAY_SIZE(frame_list); | |
431 | ||
9346c2a8 | 432 | page = list_first_entry_or_null(&ballooned_pages, struct page, lru); |
1775826c | 433 | for (i = 0; i < nr_pages; i++) { |
95d2ac4a DK |
434 | if (!page) { |
435 | nr_pages = i; | |
436 | break; | |
437 | } | |
30756c62 JG |
438 | |
439 | /* XENMEM_populate_physmap requires a PFN based on Xen | |
440 | * granularity. | |
441 | */ | |
442 | frame_list[i] = page_to_xen_pfn(page); | |
1775826c JF |
443 | page = balloon_next_page(page); |
444 | } | |
445 | ||
a90971eb | 446 | set_xen_guest_handle(reservation.extent_start, frame_list); |
fde28e8f JF |
447 | reservation.nr_extents = nr_pages; |
448 | rc = HYPERVISOR_memory_op(XENMEM_populate_physmap, &reservation); | |
40095de1 | 449 | if (rc <= 0) |
95d2ac4a | 450 | return BP_EAGAIN; |
1775826c | 451 | |
bc2c0303 | 452 | for (i = 0; i < rc; i++) { |
b6f30679 | 453 | page = balloon_retrieve(false); |
1775826c JF |
454 | BUG_ON(page == NULL); |
455 | ||
c2374bf5 | 456 | #ifdef CONFIG_XEN_HAVE_PVMMU |
30756c62 JG |
457 | /* |
458 | * We don't support PV MMU when Linux and Xen is using | |
459 | * different page granularity. | |
460 | */ | |
461 | BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE); | |
462 | ||
c1d15f5c | 463 | if (!xen_feature(XENFEAT_auto_translated_physmap)) { |
30756c62 JG |
464 | unsigned long pfn = page_to_pfn(page); |
465 | ||
c1d15f5c SS |
466 | set_phys_to_machine(pfn, frame_list[i]); |
467 | ||
468 | /* Link back into the page tables if not highmem. */ | |
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 | } | |
1775826c | 477 | } |
c2374bf5 | 478 | #endif |
1775826c JF |
479 | |
480 | /* Relinquish the page back to the allocator. */ | |
3dcc0571 | 481 | __free_reserved_page(page); |
1775826c JF |
482 | } |
483 | ||
bc2c0303 | 484 | balloon_stats.current_pages += rc; |
1775826c | 485 | |
95d2ac4a | 486 | return BP_DONE; |
1775826c JF |
487 | } |
488 | ||
b6f30679 | 489 | static enum bp_state decrease_reservation(unsigned long nr_pages, gfp_t gfp) |
1775826c | 490 | { |
95d2ac4a | 491 | enum bp_state state = BP_DONE; |
30756c62 JG |
492 | unsigned long i; |
493 | struct page *page, *tmp; | |
1775826c JF |
494 | int ret; |
495 | struct xen_memory_reservation reservation = { | |
496 | .address_bits = 0, | |
30756c62 | 497 | .extent_order = EXTENT_ORDER, |
1775826c JF |
498 | .domid = DOMID_SELF |
499 | }; | |
30756c62 | 500 | LIST_HEAD(pages); |
1775826c JF |
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++) { | |
fce92683 LN |
506 | page = alloc_page(gfp); |
507 | if (page == NULL) { | |
1775826c | 508 | nr_pages = i; |
95d2ac4a | 509 | state = BP_EAGAIN; |
1775826c JF |
510 | break; |
511 | } | |
09ed3d5b | 512 | scrub_page(page); |
30756c62 | 513 | list_add(&page->lru, &pages); |
09ed3d5b | 514 | } |
1775826c | 515 | |
09ed3d5b WL |
516 | /* |
517 | * Ensure that ballooned highmem pages don't have kmaps. | |
518 | * | |
519 | * Do this before changing the p2m as kmap_flush_unused() | |
520 | * reads PTEs to obtain pages (and hence needs the original | |
521 | * p2m entry). | |
522 | */ | |
523 | kmap_flush_unused(); | |
524 | ||
30756c62 JG |
525 | /* |
526 | * Setup the frame, update direct mapping, invalidate P2M, | |
527 | * and add to balloon. | |
528 | */ | |
529 | i = 0; | |
530 | list_for_each_entry_safe(page, tmp, &pages, lru) { | |
531 | /* XENMEM_decrease_reservation requires a GFN */ | |
532 | frame_list[i++] = xen_page_to_gfn(page); | |
1058a75f | 533 | |
c1d15f5c | 534 | #ifdef CONFIG_XEN_HAVE_PVMMU |
30756c62 JG |
535 | /* |
536 | * We don't support PV MMU when Linux and Xen is using | |
537 | * different page granularity. | |
538 | */ | |
539 | BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE); | |
540 | ||
04660bb5 | 541 | if (!xen_feature(XENFEAT_auto_translated_physmap)) { |
30756c62 JG |
542 | unsigned long pfn = page_to_pfn(page); |
543 | ||
c1d15f5c SS |
544 | if (!PageHighMem(page)) { |
545 | ret = HYPERVISOR_update_va_mapping( | |
546 | (unsigned long)__va(pfn << PAGE_SHIFT), | |
0bb599fd | 547 | __pte_ma(0), 0); |
c1d15f5c | 548 | BUG_ON(ret); |
fb9a0c44 DV |
549 | } |
550 | __set_phys_to_machine(pfn, INVALID_P2M_ENTRY); | |
04660bb5 | 551 | } |
c1d15f5c | 552 | #endif |
30756c62 | 553 | list_del(&page->lru); |
24f69373 | 554 | |
09ed3d5b | 555 | balloon_append(page); |
1775826c JF |
556 | } |
557 | ||
24f69373 | 558 | flush_tlb_all(); |
2bad07ce | 559 | |
a90971eb | 560 | set_xen_guest_handle(reservation.extent_start, frame_list); |
1775826c JF |
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; | |
1775826c | 566 | |
95d2ac4a | 567 | return state; |
1775826c JF |
568 | } |
569 | ||
570 | /* | |
929423fa | 571 | * As this is a work item it is guaranteed to run as a single instance only. |
1775826c JF |
572 | * We may of course race updates of the target counts (which are protected |
573 | * by the balloon lock), or with changes to the Xen hard limit, but we will | |
574 | * recover from these in time. | |
575 | */ | |
576 | static void balloon_process(struct work_struct *work) | |
577 | { | |
95d2ac4a | 578 | enum bp_state state = BP_DONE; |
1775826c JF |
579 | long credit; |
580 | ||
1775826c JF |
581 | |
582 | do { | |
929423fa JG |
583 | mutex_lock(&balloon_mutex); |
584 | ||
83be7e52 | 585 | credit = current_credit(); |
95d2ac4a | 586 | |
080e2be7 DK |
587 | if (credit > 0) { |
588 | if (balloon_is_inflated()) | |
589 | state = increase_reservation(credit); | |
590 | else | |
b2ac6aa8 | 591 | state = reserve_additional_memory(); |
080e2be7 | 592 | } |
95d2ac4a | 593 | |
1775826c | 594 | if (credit < 0) |
b6f30679 | 595 | state = decrease_reservation(-credit, GFP_BALLOON); |
95d2ac4a DK |
596 | |
597 | state = update_schedule(state); | |
1775826c | 598 | |
929423fa JG |
599 | mutex_unlock(&balloon_mutex); |
600 | ||
601 | cond_resched(); | |
602 | ||
95d2ac4a | 603 | } while (credit && state == BP_DONE); |
1775826c JF |
604 | |
605 | /* Schedule more work if there is some still to be done. */ | |
95d2ac4a DK |
606 | if (state == BP_EAGAIN) |
607 | schedule_delayed_work(&balloon_worker, balloon_stats.schedule_delay * HZ); | |
1775826c JF |
608 | } |
609 | ||
610 | /* Resets the Xen limit, sets new target, and kicks off processing. */ | |
803eb047 | 611 | void balloon_set_new_target(unsigned long target) |
1775826c JF |
612 | { |
613 | /* No need for lock. Not read-modify-write updates. */ | |
1775826c | 614 | balloon_stats.target_pages = target; |
95170b2e | 615 | schedule_delayed_work(&balloon_worker, 0); |
1775826c | 616 | } |
803eb047 | 617 | EXPORT_SYMBOL_GPL(balloon_set_new_target); |
1775826c | 618 | |
1cf6a6c8 DV |
619 | static 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 | ||
b6f30679 KRW |
641 | /** |
642 | * alloc_xenballooned_pages - get pages that have been ballooned out | |
643 | * @nr_pages: Number of pages to get | |
644 | * @pages: pages returned | |
645 | * @return 0 on success, error otherwise | |
646 | */ | |
81b286e0 | 647 | int alloc_xenballooned_pages(int nr_pages, struct page **pages) |
1775826c | 648 | { |
b6f30679 | 649 | int pgno = 0; |
e882dc9c | 650 | struct page *page; |
1cf6a6c8 DV |
651 | int ret; |
652 | ||
b6f30679 | 653 | mutex_lock(&balloon_mutex); |
1cf6a6c8 DV |
654 | |
655 | balloon_stats.target_unpopulated += nr_pages; | |
656 | ||
b6f30679 | 657 | while (pgno < nr_pages) { |
81b286e0 DV |
658 | page = balloon_retrieve(true); |
659 | if (page) { | |
b6f30679 | 660 | pages[pgno++] = page; |
4a69c909 | 661 | #ifdef CONFIG_XEN_HAVE_PVMMU |
30756c62 JG |
662 | /* |
663 | * We don't support PV MMU when Linux and Xen is using | |
664 | * different page granularity. | |
665 | */ | |
666 | BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE); | |
667 | ||
4a69c909 DV |
668 | ret = xen_alloc_p2m_entry(page_to_pfn(page)); |
669 | if (ret < 0) | |
670 | goto out_undo; | |
671 | #endif | |
b6f30679 | 672 | } else { |
1cf6a6c8 DV |
673 | ret = add_ballooned_pages(nr_pages - pgno); |
674 | if (ret < 0) | |
b6f30679 KRW |
675 | goto out_undo; |
676 | } | |
1775826c | 677 | } |
b6f30679 KRW |
678 | mutex_unlock(&balloon_mutex); |
679 | return 0; | |
680 | out_undo: | |
b6f30679 | 681 | mutex_unlock(&balloon_mutex); |
1cf6a6c8 DV |
682 | free_xenballooned_pages(pgno, pages); |
683 | return ret; | |
1775826c | 684 | } |
b6f30679 | 685 | EXPORT_SYMBOL(alloc_xenballooned_pages); |
1775826c | 686 | |
b6f30679 KRW |
687 | /** |
688 | * free_xenballooned_pages - return pages retrieved with get_ballooned_pages | |
689 | * @nr_pages: Number of pages | |
690 | * @pages: pages to return | |
691 | */ | |
e882dc9c | 692 | void free_xenballooned_pages(int nr_pages, struct page **pages) |
1775826c | 693 | { |
b6f30679 | 694 | int i; |
1775826c | 695 | |
b6f30679 | 696 | mutex_lock(&balloon_mutex); |
1775826c | 697 | |
b6f30679 KRW |
698 | for (i = 0; i < nr_pages; i++) { |
699 | if (pages[i]) | |
700 | balloon_append(pages[i]); | |
701 | } | |
702 | ||
1cf6a6c8 DV |
703 | balloon_stats.target_unpopulated -= nr_pages; |
704 | ||
b6f30679 | 705 | /* The balloon may be too large now. Shrink it if needed. */ |
83be7e52 | 706 | if (current_credit()) |
b6f30679 | 707 | schedule_delayed_work(&balloon_worker, 0); |
1775826c | 708 | |
b6f30679 KRW |
709 | mutex_unlock(&balloon_mutex); |
710 | } | |
711 | EXPORT_SYMBOL(free_xenballooned_pages); | |
1775826c | 712 | |
8b5d44a5 DV |
713 | static void __init balloon_add_region(unsigned long start_pfn, |
714 | unsigned long pages) | |
1775826c | 715 | { |
4dfe22f5 | 716 | unsigned long pfn, extra_pfn_end; |
1775826c JF |
717 | struct page *page; |
718 | ||
8b5d44a5 DV |
719 | /* |
720 | * If the amount of usable memory has been limited (e.g., with | |
721 | * the 'mem' command line parameter), don't add pages beyond | |
722 | * this limit. | |
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 | /* totalram_pages and totalhigh_pages do not | |
729 | include the boot-time balloon extension, so | |
730 | don't subtract from it. */ | |
731 | __balloon_append(page); | |
732 | } | |
de5a77d8 DV |
733 | |
734 | balloon_stats.total_pages += extra_pfn_end - start_pfn; | |
8b5d44a5 DV |
735 | } |
736 | ||
737 | static int __init balloon_init(void) | |
738 | { | |
0bb599fd | 739 | int i; |
8b5d44a5 | 740 | |
53d5522c | 741 | if (!xen_domain()) |
1775826c JF |
742 | return -ENODEV; |
743 | ||
283c0972 | 744 | pr_info("Initialising balloon driver\n"); |
1775826c | 745 | |
aa24411b DV |
746 | balloon_stats.current_pages = xen_pv_domain() |
747 | ? min(xen_start_info->nr_pages - xen_released_pages, max_pfn) | |
c275a57f | 748 | : get_num_physpages(); |
1775826c JF |
749 | balloon_stats.target_pages = balloon_stats.current_pages; |
750 | balloon_stats.balloon_low = 0; | |
751 | balloon_stats.balloon_high = 0; | |
de5a77d8 | 752 | balloon_stats.total_pages = balloon_stats.current_pages; |
1775826c | 753 | |
95d2ac4a DK |
754 | balloon_stats.schedule_delay = 1; |
755 | balloon_stats.max_schedule_delay = 32; | |
756 | balloon_stats.retry_count = 1; | |
40095de1 | 757 | balloon_stats.max_retry_count = RETRY_UNLIMITED; |
1775826c | 758 | |
080e2be7 | 759 | #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG |
080e2be7 DK |
760 | set_online_page_callback(&xen_online_page); |
761 | register_memory_notifier(&xen_memory_nb); | |
1cf6a6c8 | 762 | register_sysctl_table(xen_root); |
080e2be7 DK |
763 | #endif |
764 | ||
2a4c92fa | 765 | /* |
b1cbf9b1 | 766 | * Initialize the balloon with pages from the extra memory |
8b5d44a5 | 767 | * regions (see arch/x86/xen/setup.c). |
2a4c92fa | 768 | */ |
8b5d44a5 | 769 | for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) |
626d7508 JG |
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); | |
1775826c | 773 | |
1775826c JF |
774 | return 0; |
775 | } | |
1775826c | 776 | subsys_initcall(balloon_init); |