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> | |
72ee5112 | 45 | #include <linux/module.h> |
1775826c JF |
46 | #include <linux/mm.h> |
47 | #include <linux/bootmem.h> | |
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> | |
74 | ||
1cf6a6c8 DV |
75 | static int xen_hotplug_unpopulated; |
76 | ||
77 | #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG | |
78 | ||
79 | static int zero; | |
80 | static int one = 1; | |
81 | ||
82 | static struct ctl_table balloon_table[] = { | |
83 | { | |
84 | .procname = "hotplug_unpopulated", | |
85 | .data = &xen_hotplug_unpopulated, | |
86 | .maxlen = sizeof(int), | |
87 | .mode = 0644, | |
88 | .proc_handler = proc_dointvec_minmax, | |
89 | .extra1 = &zero, | |
90 | .extra2 = &one, | |
91 | }, | |
92 | { } | |
93 | }; | |
94 | ||
95 | static struct ctl_table balloon_root[] = { | |
96 | { | |
97 | .procname = "balloon", | |
98 | .mode = 0555, | |
99 | .child = balloon_table, | |
100 | }, | |
101 | { } | |
102 | }; | |
103 | ||
104 | static struct ctl_table xen_root[] = { | |
105 | { | |
106 | .procname = "xen", | |
107 | .mode = 0555, | |
108 | .child = balloon_root, | |
109 | }, | |
110 | { } | |
111 | }; | |
112 | ||
113 | #endif | |
114 | ||
30756c62 JG |
115 | /* |
116 | * Use one extent per PAGE_SIZE to avoid to break down the page into | |
117 | * multiple frame. | |
118 | */ | |
119 | #define EXTENT_ORDER (fls(XEN_PFN_PER_PAGE) - 1) | |
120 | ||
95d2ac4a DK |
121 | /* |
122 | * balloon_process() state: | |
123 | * | |
124 | * BP_DONE: done or nothing to do, | |
b2ac6aa8 | 125 | * BP_WAIT: wait to be rescheduled, |
95d2ac4a DK |
126 | * BP_EAGAIN: error, go to sleep, |
127 | * BP_ECANCELED: error, balloon operation canceled. | |
128 | */ | |
1775826c | 129 | |
95d2ac4a DK |
130 | enum bp_state { |
131 | BP_DONE, | |
b2ac6aa8 | 132 | BP_WAIT, |
95d2ac4a DK |
133 | BP_EAGAIN, |
134 | BP_ECANCELED | |
1775826c JF |
135 | }; |
136 | ||
1775826c | 137 | |
1775826c | 138 | static DEFINE_MUTEX(balloon_mutex); |
1775826c | 139 | |
803eb047 DDG |
140 | struct balloon_stats balloon_stats; |
141 | EXPORT_SYMBOL_GPL(balloon_stats); | |
1775826c JF |
142 | |
143 | /* We increase/decrease in batches which fit in a page */ | |
3990dd27 | 144 | static xen_pfn_t frame_list[PAGE_SIZE / sizeof(xen_pfn_t)]; |
cd9151e2 | 145 | |
1775826c | 146 | |
1775826c JF |
147 | /* List of ballooned pages, threaded through the mem_map array. */ |
148 | static LIST_HEAD(ballooned_pages); | |
1cf6a6c8 | 149 | static DECLARE_WAIT_QUEUE_HEAD(balloon_wq); |
1775826c JF |
150 | |
151 | /* Main work function, always executed in process context. */ | |
152 | static void balloon_process(struct work_struct *work); | |
95170b2e | 153 | static DECLARE_DELAYED_WORK(balloon_worker, balloon_process); |
1775826c JF |
154 | |
155 | /* When ballooning out (allocating memory to return to Xen) we don't really | |
156 | want the kernel to try too hard since that can trigger the oom killer. */ | |
157 | #define GFP_BALLOON \ | |
158 | (GFP_HIGHUSER | __GFP_NOWARN | __GFP_NORETRY | __GFP_NOMEMALLOC) | |
159 | ||
160 | static void scrub_page(struct page *page) | |
161 | { | |
162 | #ifdef CONFIG_XEN_SCRUB_PAGES | |
26a3e991 | 163 | clear_highpage(page); |
1775826c JF |
164 | #endif |
165 | } | |
166 | ||
167 | /* balloon_append: add the given page to the balloon. */ | |
9be4d457 | 168 | static void __balloon_append(struct page *page) |
1775826c JF |
169 | { |
170 | /* Lowmem is re-populated first, so highmem pages go at list tail. */ | |
171 | if (PageHighMem(page)) { | |
172 | list_add_tail(&page->lru, &ballooned_pages); | |
173 | balloon_stats.balloon_high++; | |
1775826c JF |
174 | } else { |
175 | list_add(&page->lru, &ballooned_pages); | |
176 | balloon_stats.balloon_low++; | |
177 | } | |
1cf6a6c8 | 178 | wake_up(&balloon_wq); |
9be4d457 | 179 | } |
3d65c948 | 180 | |
9be4d457 JF |
181 | static void balloon_append(struct page *page) |
182 | { | |
183 | __balloon_append(page); | |
3dcc0571 | 184 | adjust_managed_page_count(page, -1); |
1775826c JF |
185 | } |
186 | ||
187 | /* balloon_retrieve: rescue a page from the balloon, if it is not empty. */ | |
81b286e0 | 188 | static struct page *balloon_retrieve(bool require_lowmem) |
1775826c JF |
189 | { |
190 | struct page *page; | |
191 | ||
192 | if (list_empty(&ballooned_pages)) | |
193 | return NULL; | |
194 | ||
81b286e0 DV |
195 | page = list_entry(ballooned_pages.next, struct page, lru); |
196 | if (require_lowmem && PageHighMem(page)) | |
197 | return NULL; | |
1775826c JF |
198 | list_del(&page->lru); |
199 | ||
3dcc0571 | 200 | if (PageHighMem(page)) |
1775826c | 201 | balloon_stats.balloon_high--; |
3dcc0571 | 202 | else |
1775826c JF |
203 | balloon_stats.balloon_low--; |
204 | ||
3dcc0571 | 205 | adjust_managed_page_count(page, 1); |
3d65c948 | 206 | |
1775826c JF |
207 | return page; |
208 | } | |
209 | ||
1775826c JF |
210 | static struct page *balloon_next_page(struct page *page) |
211 | { | |
212 | struct list_head *next = page->lru.next; | |
213 | if (next == &ballooned_pages) | |
214 | return NULL; | |
215 | return list_entry(next, struct page, lru); | |
216 | } | |
217 | ||
95d2ac4a | 218 | static enum bp_state update_schedule(enum bp_state state) |
1775826c | 219 | { |
b2ac6aa8 DV |
220 | if (state == BP_WAIT) |
221 | return BP_WAIT; | |
222 | ||
fd8b7951 BO |
223 | if (state == BP_ECANCELED) |
224 | return BP_ECANCELED; | |
225 | ||
95d2ac4a DK |
226 | if (state == BP_DONE) { |
227 | balloon_stats.schedule_delay = 1; | |
228 | balloon_stats.retry_count = 1; | |
229 | return BP_DONE; | |
230 | } | |
231 | ||
95d2ac4a DK |
232 | ++balloon_stats.retry_count; |
233 | ||
234 | if (balloon_stats.max_retry_count != RETRY_UNLIMITED && | |
235 | balloon_stats.retry_count > balloon_stats.max_retry_count) { | |
95d2ac4a DK |
236 | balloon_stats.schedule_delay = 1; |
237 | balloon_stats.retry_count = 1; | |
238 | return BP_ECANCELED; | |
239 | } | |
240 | ||
241 | balloon_stats.schedule_delay <<= 1; | |
242 | ||
243 | if (balloon_stats.schedule_delay > balloon_stats.max_schedule_delay) | |
244 | balloon_stats.schedule_delay = balloon_stats.max_schedule_delay; | |
245 | ||
246 | return BP_EAGAIN; | |
1775826c JF |
247 | } |
248 | ||
080e2be7 | 249 | #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG |
55b3da98 DV |
250 | static struct resource *additional_memory_resource(phys_addr_t size) |
251 | { | |
252 | struct resource *res; | |
253 | int ret; | |
254 | ||
255 | res = kzalloc(sizeof(*res), GFP_KERNEL); | |
256 | if (!res) | |
257 | return NULL; | |
258 | ||
259 | res->name = "System RAM"; | |
260 | res->flags = IORESOURCE_MEM | IORESOURCE_BUSY; | |
261 | ||
262 | ret = allocate_resource(&iomem_resource, res, | |
263 | size, 0, -1, | |
264 | PAGES_PER_SECTION * PAGE_SIZE, NULL, NULL); | |
265 | if (ret < 0) { | |
266 | pr_err("Cannot allocate new System RAM resource\n"); | |
267 | kfree(res); | |
268 | return NULL; | |
269 | } | |
270 | ||
271 | return res; | |
272 | } | |
273 | ||
274 | static void release_memory_resource(struct resource *resource) | |
275 | { | |
276 | if (!resource) | |
277 | return; | |
278 | ||
279 | /* | |
280 | * No need to reset region to identity mapped since we now | |
281 | * know that no I/O can be in this region | |
282 | */ | |
283 | release_resource(resource); | |
284 | kfree(resource); | |
285 | } | |
080e2be7 | 286 | |
b2ac6aa8 | 287 | static enum bp_state reserve_additional_memory(void) |
080e2be7 | 288 | { |
b2ac6aa8 | 289 | long credit; |
55b3da98 | 290 | struct resource *resource; |
080e2be7 | 291 | int nid, rc; |
55b3da98 | 292 | unsigned long balloon_hotplug; |
080e2be7 | 293 | |
1cf6a6c8 DV |
294 | credit = balloon_stats.target_pages + balloon_stats.target_unpopulated |
295 | - balloon_stats.total_pages; | |
b2ac6aa8 DV |
296 | |
297 | /* | |
298 | * Already hotplugged enough pages? Wait for them to be | |
299 | * onlined. | |
300 | */ | |
301 | if (credit <= 0) | |
302 | return BP_WAIT; | |
303 | ||
55b3da98 DV |
304 | 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); | |
080e2be7 | 311 | |
3c56b3a1 | 312 | #ifdef CONFIG_XEN_HAVE_PVMMU |
30756c62 JG |
313 | /* |
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 | ||
3c56b3a1 JG |
319 | /* |
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 | ||
55b3da98 | 331 | pfn = PFN_DOWN(resource->start); |
3c56b3a1 JG |
332 | 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"); | |
55b3da98 | 335 | goto err; |
3c56b3a1 JG |
336 | } |
337 | } | |
338 | } | |
339 | #endif | |
340 | ||
55b3da98 | 341 | rc = add_memory_resource(nid, resource); |
080e2be7 | 342 | if (rc) { |
3dcf6367 | 343 | pr_warn("Cannot add additional memory (%i)\n", rc); |
55b3da98 | 344 | goto err; |
080e2be7 DK |
345 | } |
346 | ||
de5a77d8 | 347 | balloon_stats.total_pages += balloon_hotplug; |
080e2be7 | 348 | |
b2ac6aa8 | 349 | return BP_WAIT; |
55b3da98 DV |
350 | err: |
351 | release_memory_resource(resource); | |
352 | return BP_ECANCELED; | |
080e2be7 DK |
353 | } |
354 | ||
355 | static void xen_online_page(struct page *page) | |
356 | { | |
357 | __online_page_set_limits(page); | |
358 | ||
359 | mutex_lock(&balloon_mutex); | |
360 | ||
361 | __balloon_append(page); | |
362 | ||
080e2be7 DK |
363 | mutex_unlock(&balloon_mutex); |
364 | } | |
365 | ||
366 | static int xen_memory_notifier(struct notifier_block *nb, unsigned long val, void *v) | |
367 | { | |
368 | if (val == MEM_ONLINE) | |
369 | schedule_delayed_work(&balloon_worker, 0); | |
370 | ||
371 | return NOTIFY_OK; | |
372 | } | |
373 | ||
374 | static struct notifier_block xen_memory_nb = { | |
375 | .notifier_call = xen_memory_notifier, | |
376 | .priority = 0 | |
377 | }; | |
378 | #else | |
b2ac6aa8 | 379 | static enum bp_state reserve_additional_memory(void) |
1775826c | 380 | { |
de5a77d8 | 381 | balloon_stats.target_pages = balloon_stats.current_pages; |
1cf6a6c8 | 382 | return BP_ECANCELED; |
1775826c | 383 | } |
de5a77d8 | 384 | #endif /* CONFIG_XEN_BALLOON_MEMORY_HOTPLUG */ |
1775826c | 385 | |
de5a77d8 | 386 | static long current_credit(void) |
080e2be7 | 387 | { |
de5a77d8 | 388 | return balloon_stats.target_pages - balloon_stats.current_pages; |
080e2be7 DK |
389 | } |
390 | ||
de5a77d8 | 391 | static bool balloon_is_inflated(void) |
080e2be7 | 392 | { |
de5a77d8 | 393 | return balloon_stats.balloon_low || balloon_stats.balloon_high; |
080e2be7 | 394 | } |
080e2be7 | 395 | |
95d2ac4a | 396 | static enum bp_state increase_reservation(unsigned long nr_pages) |
1775826c | 397 | { |
95d2ac4a | 398 | int rc; |
30756c62 | 399 | unsigned long i; |
1775826c | 400 | struct page *page; |
1775826c JF |
401 | struct xen_memory_reservation reservation = { |
402 | .address_bits = 0, | |
30756c62 | 403 | .extent_order = EXTENT_ORDER, |
1775826c JF |
404 | .domid = DOMID_SELF |
405 | }; | |
406 | ||
407 | if (nr_pages > ARRAY_SIZE(frame_list)) | |
408 | nr_pages = ARRAY_SIZE(frame_list); | |
409 | ||
9346c2a8 | 410 | page = list_first_entry_or_null(&ballooned_pages, struct page, lru); |
1775826c | 411 | for (i = 0; i < nr_pages; i++) { |
95d2ac4a DK |
412 | if (!page) { |
413 | nr_pages = i; | |
414 | break; | |
415 | } | |
30756c62 JG |
416 | |
417 | /* XENMEM_populate_physmap requires a PFN based on Xen | |
418 | * granularity. | |
419 | */ | |
420 | frame_list[i] = page_to_xen_pfn(page); | |
1775826c JF |
421 | page = balloon_next_page(page); |
422 | } | |
423 | ||
a90971eb | 424 | set_xen_guest_handle(reservation.extent_start, frame_list); |
fde28e8f JF |
425 | reservation.nr_extents = nr_pages; |
426 | rc = HYPERVISOR_memory_op(XENMEM_populate_physmap, &reservation); | |
40095de1 | 427 | if (rc <= 0) |
95d2ac4a | 428 | return BP_EAGAIN; |
1775826c | 429 | |
bc2c0303 | 430 | for (i = 0; i < rc; i++) { |
b6f30679 | 431 | page = balloon_retrieve(false); |
1775826c JF |
432 | BUG_ON(page == NULL); |
433 | ||
c2374bf5 | 434 | #ifdef CONFIG_XEN_HAVE_PVMMU |
30756c62 JG |
435 | /* |
436 | * We don't support PV MMU when Linux and Xen is using | |
437 | * different page granularity. | |
438 | */ | |
439 | BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE); | |
440 | ||
c1d15f5c | 441 | if (!xen_feature(XENFEAT_auto_translated_physmap)) { |
30756c62 JG |
442 | unsigned long pfn = page_to_pfn(page); |
443 | ||
c1d15f5c SS |
444 | set_phys_to_machine(pfn, frame_list[i]); |
445 | ||
446 | /* Link back into the page tables if not highmem. */ | |
447 | if (!PageHighMem(page)) { | |
448 | int ret; | |
449 | ret = HYPERVISOR_update_va_mapping( | |
450 | (unsigned long)__va(pfn << PAGE_SHIFT), | |
451 | mfn_pte(frame_list[i], PAGE_KERNEL), | |
452 | 0); | |
453 | BUG_ON(ret); | |
454 | } | |
1775826c | 455 | } |
c2374bf5 | 456 | #endif |
1775826c JF |
457 | |
458 | /* Relinquish the page back to the allocator. */ | |
3dcc0571 | 459 | __free_reserved_page(page); |
1775826c JF |
460 | } |
461 | ||
bc2c0303 | 462 | balloon_stats.current_pages += rc; |
1775826c | 463 | |
95d2ac4a | 464 | return BP_DONE; |
1775826c JF |
465 | } |
466 | ||
b6f30679 | 467 | static enum bp_state decrease_reservation(unsigned long nr_pages, gfp_t gfp) |
1775826c | 468 | { |
95d2ac4a | 469 | enum bp_state state = BP_DONE; |
30756c62 JG |
470 | unsigned long i; |
471 | struct page *page, *tmp; | |
1775826c JF |
472 | int ret; |
473 | struct xen_memory_reservation reservation = { | |
474 | .address_bits = 0, | |
30756c62 | 475 | .extent_order = EXTENT_ORDER, |
1775826c JF |
476 | .domid = DOMID_SELF |
477 | }; | |
30756c62 | 478 | LIST_HEAD(pages); |
1775826c JF |
479 | |
480 | if (nr_pages > ARRAY_SIZE(frame_list)) | |
481 | nr_pages = ARRAY_SIZE(frame_list); | |
482 | ||
483 | for (i = 0; i < nr_pages; i++) { | |
fce92683 LN |
484 | page = alloc_page(gfp); |
485 | if (page == NULL) { | |
1775826c | 486 | nr_pages = i; |
95d2ac4a | 487 | state = BP_EAGAIN; |
1775826c JF |
488 | break; |
489 | } | |
09ed3d5b | 490 | scrub_page(page); |
30756c62 | 491 | list_add(&page->lru, &pages); |
09ed3d5b | 492 | } |
1775826c | 493 | |
09ed3d5b WL |
494 | /* |
495 | * Ensure that ballooned highmem pages don't have kmaps. | |
496 | * | |
497 | * Do this before changing the p2m as kmap_flush_unused() | |
498 | * reads PTEs to obtain pages (and hence needs the original | |
499 | * p2m entry). | |
500 | */ | |
501 | kmap_flush_unused(); | |
502 | ||
30756c62 JG |
503 | /* |
504 | * Setup the frame, update direct mapping, invalidate P2M, | |
505 | * and add to balloon. | |
506 | */ | |
507 | i = 0; | |
508 | list_for_each_entry_safe(page, tmp, &pages, lru) { | |
509 | /* XENMEM_decrease_reservation requires a GFN */ | |
510 | frame_list[i++] = xen_page_to_gfn(page); | |
1058a75f | 511 | |
c1d15f5c | 512 | #ifdef CONFIG_XEN_HAVE_PVMMU |
30756c62 JG |
513 | /* |
514 | * We don't support PV MMU when Linux and Xen is using | |
515 | * different page granularity. | |
516 | */ | |
517 | BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE); | |
518 | ||
04660bb5 | 519 | if (!xen_feature(XENFEAT_auto_translated_physmap)) { |
30756c62 JG |
520 | unsigned long pfn = page_to_pfn(page); |
521 | ||
c1d15f5c SS |
522 | if (!PageHighMem(page)) { |
523 | ret = HYPERVISOR_update_va_mapping( | |
524 | (unsigned long)__va(pfn << PAGE_SHIFT), | |
0bb599fd | 525 | __pte_ma(0), 0); |
c1d15f5c | 526 | BUG_ON(ret); |
fb9a0c44 DV |
527 | } |
528 | __set_phys_to_machine(pfn, INVALID_P2M_ENTRY); | |
04660bb5 | 529 | } |
c1d15f5c | 530 | #endif |
30756c62 | 531 | list_del(&page->lru); |
24f69373 | 532 | |
09ed3d5b | 533 | balloon_append(page); |
1775826c JF |
534 | } |
535 | ||
24f69373 | 536 | flush_tlb_all(); |
2bad07ce | 537 | |
a90971eb | 538 | set_xen_guest_handle(reservation.extent_start, frame_list); |
1775826c JF |
539 | reservation.nr_extents = nr_pages; |
540 | ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation, &reservation); | |
541 | BUG_ON(ret != nr_pages); | |
542 | ||
543 | balloon_stats.current_pages -= nr_pages; | |
1775826c | 544 | |
95d2ac4a | 545 | return state; |
1775826c JF |
546 | } |
547 | ||
548 | /* | |
929423fa | 549 | * As this is a work item it is guaranteed to run as a single instance only. |
1775826c JF |
550 | * We may of course race updates of the target counts (which are protected |
551 | * by the balloon lock), or with changes to the Xen hard limit, but we will | |
552 | * recover from these in time. | |
553 | */ | |
554 | static void balloon_process(struct work_struct *work) | |
555 | { | |
95d2ac4a | 556 | enum bp_state state = BP_DONE; |
1775826c JF |
557 | long credit; |
558 | ||
1775826c JF |
559 | |
560 | do { | |
929423fa JG |
561 | mutex_lock(&balloon_mutex); |
562 | ||
83be7e52 | 563 | credit = current_credit(); |
95d2ac4a | 564 | |
080e2be7 DK |
565 | if (credit > 0) { |
566 | if (balloon_is_inflated()) | |
567 | state = increase_reservation(credit); | |
568 | else | |
b2ac6aa8 | 569 | state = reserve_additional_memory(); |
080e2be7 | 570 | } |
95d2ac4a | 571 | |
1775826c | 572 | if (credit < 0) |
b6f30679 | 573 | state = decrease_reservation(-credit, GFP_BALLOON); |
95d2ac4a DK |
574 | |
575 | state = update_schedule(state); | |
1775826c | 576 | |
929423fa JG |
577 | mutex_unlock(&balloon_mutex); |
578 | ||
579 | cond_resched(); | |
580 | ||
95d2ac4a | 581 | } while (credit && state == BP_DONE); |
1775826c JF |
582 | |
583 | /* Schedule more work if there is some still to be done. */ | |
95d2ac4a DK |
584 | if (state == BP_EAGAIN) |
585 | schedule_delayed_work(&balloon_worker, balloon_stats.schedule_delay * HZ); | |
1775826c JF |
586 | } |
587 | ||
588 | /* Resets the Xen limit, sets new target, and kicks off processing. */ | |
803eb047 | 589 | void balloon_set_new_target(unsigned long target) |
1775826c JF |
590 | { |
591 | /* No need for lock. Not read-modify-write updates. */ | |
1775826c | 592 | balloon_stats.target_pages = target; |
95170b2e | 593 | schedule_delayed_work(&balloon_worker, 0); |
1775826c | 594 | } |
803eb047 | 595 | EXPORT_SYMBOL_GPL(balloon_set_new_target); |
1775826c | 596 | |
1cf6a6c8 DV |
597 | static int add_ballooned_pages(int nr_pages) |
598 | { | |
599 | enum bp_state st; | |
600 | ||
601 | if (xen_hotplug_unpopulated) { | |
602 | st = reserve_additional_memory(); | |
603 | if (st != BP_ECANCELED) { | |
604 | mutex_unlock(&balloon_mutex); | |
605 | wait_event(balloon_wq, | |
606 | !list_empty(&ballooned_pages)); | |
607 | mutex_lock(&balloon_mutex); | |
608 | return 0; | |
609 | } | |
610 | } | |
611 | ||
612 | st = decrease_reservation(nr_pages, GFP_USER); | |
613 | if (st != BP_DONE) | |
614 | return -ENOMEM; | |
615 | ||
616 | return 0; | |
617 | } | |
618 | ||
b6f30679 KRW |
619 | /** |
620 | * alloc_xenballooned_pages - get pages that have been ballooned out | |
621 | * @nr_pages: Number of pages to get | |
622 | * @pages: pages returned | |
623 | * @return 0 on success, error otherwise | |
624 | */ | |
81b286e0 | 625 | int alloc_xenballooned_pages(int nr_pages, struct page **pages) |
1775826c | 626 | { |
b6f30679 | 627 | int pgno = 0; |
e882dc9c | 628 | struct page *page; |
1cf6a6c8 DV |
629 | int ret; |
630 | ||
b6f30679 | 631 | mutex_lock(&balloon_mutex); |
1cf6a6c8 DV |
632 | |
633 | balloon_stats.target_unpopulated += nr_pages; | |
634 | ||
b6f30679 | 635 | while (pgno < nr_pages) { |
81b286e0 DV |
636 | page = balloon_retrieve(true); |
637 | if (page) { | |
b6f30679 | 638 | pages[pgno++] = page; |
4a69c909 | 639 | #ifdef CONFIG_XEN_HAVE_PVMMU |
30756c62 JG |
640 | /* |
641 | * We don't support PV MMU when Linux and Xen is using | |
642 | * different page granularity. | |
643 | */ | |
644 | BUILD_BUG_ON(XEN_PAGE_SIZE != PAGE_SIZE); | |
645 | ||
4a69c909 DV |
646 | ret = xen_alloc_p2m_entry(page_to_pfn(page)); |
647 | if (ret < 0) | |
648 | goto out_undo; | |
649 | #endif | |
b6f30679 | 650 | } else { |
1cf6a6c8 DV |
651 | ret = add_ballooned_pages(nr_pages - pgno); |
652 | if (ret < 0) | |
b6f30679 KRW |
653 | goto out_undo; |
654 | } | |
1775826c | 655 | } |
b6f30679 KRW |
656 | mutex_unlock(&balloon_mutex); |
657 | return 0; | |
658 | out_undo: | |
b6f30679 | 659 | mutex_unlock(&balloon_mutex); |
1cf6a6c8 DV |
660 | free_xenballooned_pages(pgno, pages); |
661 | return ret; | |
1775826c | 662 | } |
b6f30679 | 663 | EXPORT_SYMBOL(alloc_xenballooned_pages); |
1775826c | 664 | |
b6f30679 KRW |
665 | /** |
666 | * free_xenballooned_pages - return pages retrieved with get_ballooned_pages | |
667 | * @nr_pages: Number of pages | |
668 | * @pages: pages to return | |
669 | */ | |
e882dc9c | 670 | void free_xenballooned_pages(int nr_pages, struct page **pages) |
1775826c | 671 | { |
b6f30679 | 672 | int i; |
1775826c | 673 | |
b6f30679 | 674 | mutex_lock(&balloon_mutex); |
1775826c | 675 | |
b6f30679 KRW |
676 | for (i = 0; i < nr_pages; i++) { |
677 | if (pages[i]) | |
678 | balloon_append(pages[i]); | |
679 | } | |
680 | ||
1cf6a6c8 DV |
681 | balloon_stats.target_unpopulated -= nr_pages; |
682 | ||
b6f30679 | 683 | /* The balloon may be too large now. Shrink it if needed. */ |
83be7e52 | 684 | if (current_credit()) |
b6f30679 | 685 | schedule_delayed_work(&balloon_worker, 0); |
1775826c | 686 | |
b6f30679 KRW |
687 | mutex_unlock(&balloon_mutex); |
688 | } | |
689 | EXPORT_SYMBOL(free_xenballooned_pages); | |
1775826c | 690 | |
8b5d44a5 DV |
691 | static void __init balloon_add_region(unsigned long start_pfn, |
692 | unsigned long pages) | |
1775826c | 693 | { |
4dfe22f5 | 694 | unsigned long pfn, extra_pfn_end; |
1775826c JF |
695 | struct page *page; |
696 | ||
8b5d44a5 DV |
697 | /* |
698 | * If the amount of usable memory has been limited (e.g., with | |
699 | * the 'mem' command line parameter), don't add pages beyond | |
700 | * this limit. | |
701 | */ | |
702 | extra_pfn_end = min(max_pfn, start_pfn + pages); | |
703 | ||
704 | for (pfn = start_pfn; pfn < extra_pfn_end; pfn++) { | |
705 | page = pfn_to_page(pfn); | |
706 | /* totalram_pages and totalhigh_pages do not | |
707 | include the boot-time balloon extension, so | |
708 | don't subtract from it. */ | |
709 | __balloon_append(page); | |
710 | } | |
de5a77d8 DV |
711 | |
712 | balloon_stats.total_pages += extra_pfn_end - start_pfn; | |
8b5d44a5 DV |
713 | } |
714 | ||
715 | static int __init balloon_init(void) | |
716 | { | |
0bb599fd | 717 | int i; |
8b5d44a5 | 718 | |
53d5522c | 719 | if (!xen_domain()) |
1775826c JF |
720 | return -ENODEV; |
721 | ||
283c0972 | 722 | pr_info("Initialising balloon driver\n"); |
1775826c | 723 | |
aa24411b DV |
724 | balloon_stats.current_pages = xen_pv_domain() |
725 | ? min(xen_start_info->nr_pages - xen_released_pages, max_pfn) | |
c275a57f | 726 | : get_num_physpages(); |
1775826c JF |
727 | balloon_stats.target_pages = balloon_stats.current_pages; |
728 | balloon_stats.balloon_low = 0; | |
729 | balloon_stats.balloon_high = 0; | |
de5a77d8 | 730 | balloon_stats.total_pages = balloon_stats.current_pages; |
1775826c | 731 | |
95d2ac4a DK |
732 | balloon_stats.schedule_delay = 1; |
733 | balloon_stats.max_schedule_delay = 32; | |
734 | balloon_stats.retry_count = 1; | |
40095de1 | 735 | balloon_stats.max_retry_count = RETRY_UNLIMITED; |
1775826c | 736 | |
080e2be7 | 737 | #ifdef CONFIG_XEN_BALLOON_MEMORY_HOTPLUG |
080e2be7 DK |
738 | set_online_page_callback(&xen_online_page); |
739 | register_memory_notifier(&xen_memory_nb); | |
1cf6a6c8 | 740 | register_sysctl_table(xen_root); |
080e2be7 DK |
741 | #endif |
742 | ||
2a4c92fa | 743 | /* |
b1cbf9b1 | 744 | * Initialize the balloon with pages from the extra memory |
8b5d44a5 | 745 | * regions (see arch/x86/xen/setup.c). |
2a4c92fa | 746 | */ |
8b5d44a5 | 747 | for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) |
626d7508 JG |
748 | if (xen_extra_mem[i].n_pfns) |
749 | balloon_add_region(xen_extra_mem[i].start_pfn, | |
750 | xen_extra_mem[i].n_pfns); | |
1775826c | 751 | |
1775826c JF |
752 | return 0; |
753 | } | |
754 | ||
755 | subsys_initcall(balloon_init); | |
756 | ||
1775826c | 757 | MODULE_LICENSE("GPL"); |