Commit | Line | Data |
---|---|---|
9aa8b50b S |
1 | /* |
2 | * Copyright (c) 2012, Microsoft Corporation. | |
3 | * | |
4 | * Author: | |
5 | * K. Y. Srinivasan <kys@microsoft.com> | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or modify it | |
8 | * under the terms of the GNU General Public License version 2 as published | |
9 | * by the Free Software Foundation. | |
10 | * | |
11 | * This program is distributed in the hope that it will be useful, but | |
12 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 | * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or | |
14 | * NON INFRINGEMENT. See the GNU General Public License for more | |
15 | * details. | |
16 | * | |
17 | */ | |
18 | ||
19 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt | |
20 | ||
21 | #include <linux/kernel.h> | |
ae339336 | 22 | #include <linux/jiffies.h> |
9aa8b50b S |
23 | #include <linux/mman.h> |
24 | #include <linux/delay.h> | |
25 | #include <linux/init.h> | |
26 | #include <linux/module.h> | |
27 | #include <linux/slab.h> | |
28 | #include <linux/kthread.h> | |
29 | #include <linux/completion.h> | |
30 | #include <linux/memory_hotplug.h> | |
31 | #include <linux/memory.h> | |
32 | #include <linux/notifier.h> | |
9aa8b50b S |
33 | #include <linux/percpu_counter.h> |
34 | ||
35 | #include <linux/hyperv.h> | |
36 | ||
37 | /* | |
38 | * We begin with definitions supporting the Dynamic Memory protocol | |
39 | * with the host. | |
40 | * | |
41 | * Begin protocol definitions. | |
42 | */ | |
43 | ||
44 | ||
45 | ||
46 | /* | |
47 | * Protocol versions. The low word is the minor version, the high word the major | |
48 | * version. | |
49 | * | |
50 | * History: | |
51 | * Initial version 1.0 | |
52 | * Changed to 0.1 on 2009/03/25 | |
53 | * Changes to 0.2 on 2009/05/14 | |
54 | * Changes to 0.3 on 2009/12/03 | |
55 | * Changed to 1.0 on 2011/04/05 | |
56 | */ | |
57 | ||
58 | #define DYNMEM_MAKE_VERSION(Major, Minor) ((__u32)(((Major) << 16) | (Minor))) | |
59 | #define DYNMEM_MAJOR_VERSION(Version) ((__u32)(Version) >> 16) | |
60 | #define DYNMEM_MINOR_VERSION(Version) ((__u32)(Version) & 0xff) | |
61 | ||
62 | enum { | |
63 | DYNMEM_PROTOCOL_VERSION_1 = DYNMEM_MAKE_VERSION(0, 3), | |
64 | DYNMEM_PROTOCOL_VERSION_2 = DYNMEM_MAKE_VERSION(1, 0), | |
65 | ||
66 | DYNMEM_PROTOCOL_VERSION_WIN7 = DYNMEM_PROTOCOL_VERSION_1, | |
67 | DYNMEM_PROTOCOL_VERSION_WIN8 = DYNMEM_PROTOCOL_VERSION_2, | |
68 | ||
69 | DYNMEM_PROTOCOL_VERSION_CURRENT = DYNMEM_PROTOCOL_VERSION_WIN8 | |
70 | }; | |
71 | ||
72 | ||
73 | ||
74 | /* | |
75 | * Message Types | |
76 | */ | |
77 | ||
78 | enum dm_message_type { | |
79 | /* | |
80 | * Version 0.3 | |
81 | */ | |
82 | DM_ERROR = 0, | |
83 | DM_VERSION_REQUEST = 1, | |
84 | DM_VERSION_RESPONSE = 2, | |
85 | DM_CAPABILITIES_REPORT = 3, | |
86 | DM_CAPABILITIES_RESPONSE = 4, | |
87 | DM_STATUS_REPORT = 5, | |
88 | DM_BALLOON_REQUEST = 6, | |
89 | DM_BALLOON_RESPONSE = 7, | |
90 | DM_UNBALLOON_REQUEST = 8, | |
91 | DM_UNBALLOON_RESPONSE = 9, | |
92 | DM_MEM_HOT_ADD_REQUEST = 10, | |
93 | DM_MEM_HOT_ADD_RESPONSE = 11, | |
94 | DM_VERSION_03_MAX = 11, | |
95 | /* | |
96 | * Version 1.0. | |
97 | */ | |
98 | DM_INFO_MESSAGE = 12, | |
99 | DM_VERSION_1_MAX = 12 | |
100 | }; | |
101 | ||
102 | ||
103 | /* | |
104 | * Structures defining the dynamic memory management | |
105 | * protocol. | |
106 | */ | |
107 | ||
108 | union dm_version { | |
109 | struct { | |
110 | __u16 minor_version; | |
111 | __u16 major_version; | |
112 | }; | |
113 | __u32 version; | |
114 | } __packed; | |
115 | ||
116 | ||
117 | union dm_caps { | |
118 | struct { | |
119 | __u64 balloon:1; | |
120 | __u64 hot_add:1; | |
647965a2 S |
121 | /* |
122 | * To support guests that may have alignment | |
123 | * limitations on hot-add, the guest can specify | |
124 | * its alignment requirements; a value of n | |
125 | * represents an alignment of 2^n in mega bytes. | |
126 | */ | |
127 | __u64 hot_add_alignment:4; | |
128 | __u64 reservedz:58; | |
9aa8b50b S |
129 | } cap_bits; |
130 | __u64 caps; | |
131 | } __packed; | |
132 | ||
133 | union dm_mem_page_range { | |
134 | struct { | |
135 | /* | |
136 | * The PFN number of the first page in the range. | |
137 | * 40 bits is the architectural limit of a PFN | |
138 | * number for AMD64. | |
139 | */ | |
140 | __u64 start_page:40; | |
141 | /* | |
142 | * The number of pages in the range. | |
143 | */ | |
144 | __u64 page_cnt:24; | |
145 | } finfo; | |
146 | __u64 page_range; | |
147 | } __packed; | |
148 | ||
149 | ||
150 | ||
151 | /* | |
152 | * The header for all dynamic memory messages: | |
153 | * | |
154 | * type: Type of the message. | |
155 | * size: Size of the message in bytes; including the header. | |
156 | * trans_id: The guest is responsible for manufacturing this ID. | |
157 | */ | |
158 | ||
159 | struct dm_header { | |
160 | __u16 type; | |
161 | __u16 size; | |
162 | __u32 trans_id; | |
163 | } __packed; | |
164 | ||
165 | /* | |
166 | * A generic message format for dynamic memory. | |
167 | * Specific message formats are defined later in the file. | |
168 | */ | |
169 | ||
170 | struct dm_message { | |
171 | struct dm_header hdr; | |
172 | __u8 data[]; /* enclosed message */ | |
173 | } __packed; | |
174 | ||
175 | ||
176 | /* | |
177 | * Specific message types supporting the dynamic memory protocol. | |
178 | */ | |
179 | ||
180 | /* | |
181 | * Version negotiation message. Sent from the guest to the host. | |
182 | * The guest is free to try different versions until the host | |
183 | * accepts the version. | |
184 | * | |
185 | * dm_version: The protocol version requested. | |
186 | * is_last_attempt: If TRUE, this is the last version guest will request. | |
187 | * reservedz: Reserved field, set to zero. | |
188 | */ | |
189 | ||
190 | struct dm_version_request { | |
191 | struct dm_header hdr; | |
192 | union dm_version version; | |
193 | __u32 is_last_attempt:1; | |
194 | __u32 reservedz:31; | |
195 | } __packed; | |
196 | ||
197 | /* | |
198 | * Version response message; Host to Guest and indicates | |
199 | * if the host has accepted the version sent by the guest. | |
200 | * | |
201 | * is_accepted: If TRUE, host has accepted the version and the guest | |
202 | * should proceed to the next stage of the protocol. FALSE indicates that | |
203 | * guest should re-try with a different version. | |
204 | * | |
205 | * reservedz: Reserved field, set to zero. | |
206 | */ | |
207 | ||
208 | struct dm_version_response { | |
209 | struct dm_header hdr; | |
210 | __u64 is_accepted:1; | |
211 | __u64 reservedz:63; | |
212 | } __packed; | |
213 | ||
214 | /* | |
215 | * Message reporting capabilities. This is sent from the guest to the | |
216 | * host. | |
217 | */ | |
218 | ||
219 | struct dm_capabilities { | |
220 | struct dm_header hdr; | |
221 | union dm_caps caps; | |
222 | __u64 min_page_cnt; | |
223 | __u64 max_page_number; | |
224 | } __packed; | |
225 | ||
226 | /* | |
227 | * Response to the capabilities message. This is sent from the host to the | |
228 | * guest. This message notifies if the host has accepted the guest's | |
229 | * capabilities. If the host has not accepted, the guest must shutdown | |
230 | * the service. | |
231 | * | |
232 | * is_accepted: Indicates if the host has accepted guest's capabilities. | |
233 | * reservedz: Must be 0. | |
234 | */ | |
235 | ||
236 | struct dm_capabilities_resp_msg { | |
237 | struct dm_header hdr; | |
238 | __u64 is_accepted:1; | |
239 | __u64 reservedz:63; | |
240 | } __packed; | |
241 | ||
242 | /* | |
243 | * This message is used to report memory pressure from the guest. | |
244 | * This message is not part of any transaction and there is no | |
245 | * response to this message. | |
246 | * | |
247 | * num_avail: Available memory in pages. | |
248 | * num_committed: Committed memory in pages. | |
249 | * page_file_size: The accumulated size of all page files | |
250 | * in the system in pages. | |
251 | * zero_free: The nunber of zero and free pages. | |
252 | * page_file_writes: The writes to the page file in pages. | |
253 | * io_diff: An indicator of file cache efficiency or page file activity, | |
254 | * calculated as File Cache Page Fault Count - Page Read Count. | |
255 | * This value is in pages. | |
256 | * | |
257 | * Some of these metrics are Windows specific and fortunately | |
258 | * the algorithm on the host side that computes the guest memory | |
259 | * pressure only uses num_committed value. | |
260 | */ | |
261 | ||
262 | struct dm_status { | |
263 | struct dm_header hdr; | |
264 | __u64 num_avail; | |
265 | __u64 num_committed; | |
266 | __u64 page_file_size; | |
267 | __u64 zero_free; | |
268 | __u32 page_file_writes; | |
269 | __u32 io_diff; | |
270 | } __packed; | |
271 | ||
272 | ||
273 | /* | |
274 | * Message to ask the guest to allocate memory - balloon up message. | |
275 | * This message is sent from the host to the guest. The guest may not be | |
276 | * able to allocate as much memory as requested. | |
277 | * | |
278 | * num_pages: number of pages to allocate. | |
279 | */ | |
280 | ||
281 | struct dm_balloon { | |
282 | struct dm_header hdr; | |
283 | __u32 num_pages; | |
284 | __u32 reservedz; | |
285 | } __packed; | |
286 | ||
287 | ||
288 | /* | |
289 | * Balloon response message; this message is sent from the guest | |
290 | * to the host in response to the balloon message. | |
291 | * | |
292 | * reservedz: Reserved; must be set to zero. | |
293 | * more_pages: If FALSE, this is the last message of the transaction. | |
294 | * if TRUE there will atleast one more message from the guest. | |
295 | * | |
296 | * range_count: The number of ranges in the range array. | |
297 | * | |
298 | * range_array: An array of page ranges returned to the host. | |
299 | * | |
300 | */ | |
301 | ||
302 | struct dm_balloon_response { | |
303 | struct dm_header hdr; | |
304 | __u32 reservedz; | |
305 | __u32 more_pages:1; | |
306 | __u32 range_count:31; | |
307 | union dm_mem_page_range range_array[]; | |
308 | } __packed; | |
309 | ||
310 | /* | |
311 | * Un-balloon message; this message is sent from the host | |
312 | * to the guest to give guest more memory. | |
313 | * | |
314 | * more_pages: If FALSE, this is the last message of the transaction. | |
315 | * if TRUE there will atleast one more message from the guest. | |
316 | * | |
317 | * reservedz: Reserved; must be set to zero. | |
318 | * | |
319 | * range_count: The number of ranges in the range array. | |
320 | * | |
321 | * range_array: An array of page ranges returned to the host. | |
322 | * | |
323 | */ | |
324 | ||
325 | struct dm_unballoon_request { | |
326 | struct dm_header hdr; | |
327 | __u32 more_pages:1; | |
328 | __u32 reservedz:31; | |
329 | __u32 range_count; | |
330 | union dm_mem_page_range range_array[]; | |
331 | } __packed; | |
332 | ||
333 | /* | |
334 | * Un-balloon response message; this message is sent from the guest | |
335 | * to the host in response to an unballoon request. | |
336 | * | |
337 | */ | |
338 | ||
339 | struct dm_unballoon_response { | |
340 | struct dm_header hdr; | |
341 | } __packed; | |
342 | ||
343 | ||
344 | /* | |
345 | * Hot add request message. Message sent from the host to the guest. | |
346 | * | |
347 | * mem_range: Memory range to hot add. | |
348 | * | |
349 | * On Linux we currently don't support this since we cannot hot add | |
350 | * arbitrary granularity of memory. | |
351 | */ | |
352 | ||
353 | struct dm_hot_add { | |
354 | struct dm_header hdr; | |
355 | union dm_mem_page_range range; | |
356 | } __packed; | |
357 | ||
358 | /* | |
359 | * Hot add response message. | |
360 | * This message is sent by the guest to report the status of a hot add request. | |
361 | * If page_count is less than the requested page count, then the host should | |
362 | * assume all further hot add requests will fail, since this indicates that | |
363 | * the guest has hit an upper physical memory barrier. | |
364 | * | |
365 | * Hot adds may also fail due to low resources; in this case, the guest must | |
366 | * not complete this message until the hot add can succeed, and the host must | |
367 | * not send a new hot add request until the response is sent. | |
368 | * If VSC fails to hot add memory DYNMEM_NUMBER_OF_UNSUCCESSFUL_HOTADD_ATTEMPTS | |
369 | * times it fails the request. | |
370 | * | |
371 | * | |
372 | * page_count: number of pages that were successfully hot added. | |
373 | * | |
374 | * result: result of the operation 1: success, 0: failure. | |
375 | * | |
376 | */ | |
377 | ||
378 | struct dm_hot_add_response { | |
379 | struct dm_header hdr; | |
380 | __u32 page_count; | |
381 | __u32 result; | |
382 | } __packed; | |
383 | ||
384 | /* | |
385 | * Types of information sent from host to the guest. | |
386 | */ | |
387 | ||
388 | enum dm_info_type { | |
389 | INFO_TYPE_MAX_PAGE_CNT = 0, | |
390 | MAX_INFO_TYPE | |
391 | }; | |
392 | ||
393 | ||
394 | /* | |
395 | * Header for the information message. | |
396 | */ | |
397 | ||
398 | struct dm_info_header { | |
399 | enum dm_info_type type; | |
400 | __u32 data_size; | |
401 | } __packed; | |
402 | ||
403 | /* | |
404 | * This message is sent from the host to the guest to pass | |
405 | * some relevant information (win8 addition). | |
406 | * | |
407 | * reserved: no used. | |
408 | * info_size: size of the information blob. | |
409 | * info: information blob. | |
410 | */ | |
411 | ||
412 | struct dm_info_msg { | |
6427a0d7 | 413 | struct dm_header hdr; |
9aa8b50b S |
414 | __u32 reserved; |
415 | __u32 info_size; | |
416 | __u8 info[]; | |
417 | }; | |
418 | ||
419 | /* | |
420 | * End protocol definitions. | |
421 | */ | |
422 | ||
1cac8cd4 S |
423 | /* |
424 | * State to manage hot adding memory into the guest. | |
425 | * The range start_pfn : end_pfn specifies the range | |
426 | * that the host has asked us to hot add. The range | |
427 | * start_pfn : ha_end_pfn specifies the range that we have | |
428 | * currently hot added. We hot add in multiples of 128M | |
429 | * chunks; it is possible that we may not be able to bring | |
430 | * online all the pages in the region. The range | |
431 | * covered_start_pfn : covered_end_pfn defines the pages that can | |
432 | * be brough online. | |
433 | */ | |
434 | ||
435 | struct hv_hotadd_state { | |
436 | struct list_head list; | |
437 | unsigned long start_pfn; | |
438 | unsigned long covered_start_pfn; | |
439 | unsigned long covered_end_pfn; | |
440 | unsigned long ha_end_pfn; | |
441 | unsigned long end_pfn; | |
442 | }; | |
443 | ||
6571b2da S |
444 | struct balloon_state { |
445 | __u32 num_pages; | |
446 | struct work_struct wrk; | |
447 | }; | |
448 | ||
c51af826 S |
449 | struct hot_add_wrk { |
450 | union dm_mem_page_range ha_page_range; | |
1cac8cd4 | 451 | union dm_mem_page_range ha_region_range; |
c51af826 S |
452 | struct work_struct wrk; |
453 | }; | |
454 | ||
1cac8cd4 | 455 | static bool hot_add = true; |
9aa8b50b | 456 | static bool do_hot_add; |
e500d158 S |
457 | /* |
458 | * Delay reporting memory pressure by | |
459 | * the specified number of seconds. | |
460 | */ | |
1cac8cd4 | 461 | static uint pressure_report_delay = 45; |
9aa8b50b | 462 | |
ae339336 S |
463 | /* |
464 | * The last time we posted a pressure report to host. | |
465 | */ | |
466 | static unsigned long last_post_time; | |
467 | ||
9aa8b50b S |
468 | module_param(hot_add, bool, (S_IRUGO | S_IWUSR)); |
469 | MODULE_PARM_DESC(hot_add, "If set attempt memory hot_add"); | |
470 | ||
e500d158 S |
471 | module_param(pressure_report_delay, uint, (S_IRUGO | S_IWUSR)); |
472 | MODULE_PARM_DESC(pressure_report_delay, "Delay in secs in reporting pressure"); | |
9aa8b50b S |
473 | static atomic_t trans_id = ATOMIC_INIT(0); |
474 | ||
475 | static int dm_ring_size = (5 * PAGE_SIZE); | |
476 | ||
477 | /* | |
478 | * Driver specific state. | |
479 | */ | |
480 | ||
481 | enum hv_dm_state { | |
482 | DM_INITIALIZING = 0, | |
483 | DM_INITIALIZED, | |
484 | DM_BALLOON_UP, | |
485 | DM_BALLOON_DOWN, | |
486 | DM_HOT_ADD, | |
487 | DM_INIT_ERROR | |
488 | }; | |
489 | ||
490 | ||
491 | static __u8 recv_buffer[PAGE_SIZE]; | |
492 | static __u8 *send_buffer; | |
493 | #define PAGES_IN_2M 512 | |
1cac8cd4 | 494 | #define HA_CHUNK (32 * 1024) |
9aa8b50b S |
495 | |
496 | struct hv_dynmem_device { | |
497 | struct hv_device *dev; | |
498 | enum hv_dm_state state; | |
499 | struct completion host_event; | |
500 | struct completion config_event; | |
501 | ||
502 | /* | |
503 | * Number of pages we have currently ballooned out. | |
504 | */ | |
505 | unsigned int num_pages_ballooned; | |
506 | ||
507 | /* | |
6571b2da S |
508 | * State to manage the ballooning (up) operation. |
509 | */ | |
510 | struct balloon_state balloon_wrk; | |
511 | ||
c51af826 S |
512 | /* |
513 | * State to execute the "hot-add" operation. | |
514 | */ | |
515 | struct hot_add_wrk ha_wrk; | |
516 | ||
1cac8cd4 S |
517 | /* |
518 | * This state tracks if the host has specified a hot-add | |
519 | * region. | |
520 | */ | |
521 | bool host_specified_ha_region; | |
522 | ||
523 | /* | |
524 | * State to synchronize hot-add. | |
525 | */ | |
526 | struct completion ol_waitevent; | |
527 | bool ha_waiting; | |
6571b2da S |
528 | /* |
529 | * This thread handles hot-add | |
9aa8b50b S |
530 | * requests from the host as well as notifying |
531 | * the host with regards to memory pressure in | |
532 | * the guest. | |
533 | */ | |
534 | struct task_struct *thread; | |
535 | ||
1cac8cd4 S |
536 | /* |
537 | * A list of hot-add regions. | |
538 | */ | |
539 | struct list_head ha_region_list; | |
540 | ||
9aa8b50b S |
541 | /* |
542 | * We start with the highest version we can support | |
543 | * and downgrade based on the host; we save here the | |
544 | * next version to try. | |
545 | */ | |
546 | __u32 next_version; | |
547 | }; | |
548 | ||
549 | static struct hv_dynmem_device dm_device; | |
550 | ||
ae339336 | 551 | static void post_status(struct hv_dynmem_device *dm); |
1cac8cd4 S |
552 | #ifdef CONFIG_MEMORY_HOTPLUG |
553 | ||
a6025a2a | 554 | static void hv_bring_pgs_online(unsigned long start_pfn, unsigned long size) |
9aa8b50b | 555 | { |
1cac8cd4 | 556 | int i; |
9aa8b50b | 557 | |
1cac8cd4 S |
558 | for (i = 0; i < size; i++) { |
559 | struct page *pg; | |
560 | pg = pfn_to_page(start_pfn + i); | |
561 | __online_page_set_limits(pg); | |
562 | __online_page_increment_counters(pg); | |
563 | __online_page_free(pg); | |
564 | } | |
565 | } | |
566 | ||
567 | static void hv_mem_hot_add(unsigned long start, unsigned long size, | |
568 | unsigned long pfn_count, | |
569 | struct hv_hotadd_state *has) | |
570 | { | |
571 | int ret = 0; | |
ed07ec93 | 572 | int i, nid; |
1cac8cd4 S |
573 | unsigned long start_pfn; |
574 | unsigned long processed_pfn; | |
575 | unsigned long total_pfn = pfn_count; | |
576 | ||
577 | for (i = 0; i < (size/HA_CHUNK); i++) { | |
578 | start_pfn = start + (i * HA_CHUNK); | |
579 | has->ha_end_pfn += HA_CHUNK; | |
580 | ||
581 | if (total_pfn > HA_CHUNK) { | |
582 | processed_pfn = HA_CHUNK; | |
583 | total_pfn -= HA_CHUNK; | |
584 | } else { | |
585 | processed_pfn = total_pfn; | |
586 | total_pfn = 0; | |
587 | } | |
588 | ||
589 | has->covered_end_pfn += processed_pfn; | |
9aa8b50b | 590 | |
1cac8cd4 S |
591 | init_completion(&dm_device.ol_waitevent); |
592 | dm_device.ha_waiting = true; | |
9aa8b50b | 593 | |
1cac8cd4 S |
594 | nid = memory_add_physaddr_to_nid(PFN_PHYS(start_pfn)); |
595 | ret = add_memory(nid, PFN_PHYS((start_pfn)), | |
596 | (HA_CHUNK << PAGE_SHIFT)); | |
597 | ||
598 | if (ret) { | |
599 | pr_info("hot_add memory failed error is %d\n", ret); | |
7f4f2302 S |
600 | if (ret == -EEXIST) { |
601 | /* | |
602 | * This error indicates that the error | |
603 | * is not a transient failure. This is the | |
604 | * case where the guest's physical address map | |
605 | * precludes hot adding memory. Stop all further | |
606 | * memory hot-add. | |
607 | */ | |
608 | do_hot_add = false; | |
609 | } | |
1cac8cd4 S |
610 | has->ha_end_pfn -= HA_CHUNK; |
611 | has->covered_end_pfn -= processed_pfn; | |
612 | break; | |
613 | } | |
9aa8b50b S |
614 | |
615 | /* | |
1cac8cd4 | 616 | * Wait for the memory block to be onlined. |
ed07ec93 S |
617 | * Since the hot add has succeeded, it is ok to |
618 | * proceed even if the pages in the hot added region | |
619 | * have not been "onlined" within the allowed time. | |
9aa8b50b | 620 | */ |
ed07ec93 | 621 | wait_for_completion_timeout(&dm_device.ol_waitevent, 5*HZ); |
ae339336 | 622 | post_status(&dm_device); |
9aa8b50b S |
623 | } |
624 | ||
1cac8cd4 S |
625 | return; |
626 | } | |
627 | ||
628 | static void hv_online_page(struct page *pg) | |
629 | { | |
630 | struct list_head *cur; | |
631 | struct hv_hotadd_state *has; | |
632 | unsigned long cur_start_pgp; | |
633 | unsigned long cur_end_pgp; | |
634 | ||
635 | if (dm_device.ha_waiting) { | |
636 | dm_device.ha_waiting = false; | |
637 | complete(&dm_device.ol_waitevent); | |
638 | } | |
639 | ||
640 | list_for_each(cur, &dm_device.ha_region_list) { | |
641 | has = list_entry(cur, struct hv_hotadd_state, list); | |
642 | cur_start_pgp = (unsigned long) | |
643 | pfn_to_page(has->covered_start_pfn); | |
644 | cur_end_pgp = (unsigned long)pfn_to_page(has->covered_end_pfn); | |
645 | ||
646 | if (((unsigned long)pg >= cur_start_pgp) && | |
647 | ((unsigned long)pg < cur_end_pgp)) { | |
648 | /* | |
649 | * This frame is currently backed; online the | |
650 | * page. | |
651 | */ | |
652 | __online_page_set_limits(pg); | |
653 | __online_page_increment_counters(pg); | |
654 | __online_page_free(pg); | |
655 | has->covered_start_pfn++; | |
656 | } | |
657 | } | |
658 | } | |
659 | ||
660 | static bool pfn_covered(unsigned long start_pfn, unsigned long pfn_cnt) | |
661 | { | |
662 | struct list_head *cur; | |
663 | struct hv_hotadd_state *has; | |
664 | unsigned long residual, new_inc; | |
665 | ||
666 | if (list_empty(&dm_device.ha_region_list)) | |
667 | return false; | |
668 | ||
669 | list_for_each(cur, &dm_device.ha_region_list) { | |
670 | has = list_entry(cur, struct hv_hotadd_state, list); | |
671 | ||
672 | /* | |
673 | * If the pfn range we are dealing with is not in the current | |
674 | * "hot add block", move on. | |
675 | */ | |
676 | if ((start_pfn >= has->end_pfn)) | |
677 | continue; | |
678 | /* | |
679 | * If the current hot add-request extends beyond | |
680 | * our current limit; extend it. | |
681 | */ | |
682 | if ((start_pfn + pfn_cnt) > has->end_pfn) { | |
683 | residual = (start_pfn + pfn_cnt - has->end_pfn); | |
684 | /* | |
685 | * Extend the region by multiples of HA_CHUNK. | |
686 | */ | |
687 | new_inc = (residual / HA_CHUNK) * HA_CHUNK; | |
688 | if (residual % HA_CHUNK) | |
689 | new_inc += HA_CHUNK; | |
690 | ||
691 | has->end_pfn += new_inc; | |
692 | } | |
693 | ||
694 | /* | |
695 | * If the current start pfn is not where the covered_end | |
696 | * is, update it. | |
697 | */ | |
698 | ||
699 | if (has->covered_end_pfn != start_pfn) { | |
700 | has->covered_end_pfn = start_pfn; | |
701 | has->covered_start_pfn = start_pfn; | |
702 | } | |
703 | return true; | |
704 | ||
705 | } | |
706 | ||
707 | return false; | |
708 | } | |
709 | ||
710 | static unsigned long handle_pg_range(unsigned long pg_start, | |
711 | unsigned long pg_count) | |
712 | { | |
713 | unsigned long start_pfn = pg_start; | |
714 | unsigned long pfn_cnt = pg_count; | |
715 | unsigned long size; | |
716 | struct list_head *cur; | |
717 | struct hv_hotadd_state *has; | |
718 | unsigned long pgs_ol = 0; | |
719 | unsigned long old_covered_state; | |
720 | ||
721 | if (list_empty(&dm_device.ha_region_list)) | |
722 | return 0; | |
723 | ||
724 | list_for_each(cur, &dm_device.ha_region_list) { | |
725 | has = list_entry(cur, struct hv_hotadd_state, list); | |
726 | ||
727 | /* | |
728 | * If the pfn range we are dealing with is not in the current | |
729 | * "hot add block", move on. | |
730 | */ | |
731 | if ((start_pfn >= has->end_pfn)) | |
732 | continue; | |
733 | ||
734 | old_covered_state = has->covered_end_pfn; | |
735 | ||
736 | if (start_pfn < has->ha_end_pfn) { | |
737 | /* | |
738 | * This is the case where we are backing pages | |
739 | * in an already hot added region. Bring | |
740 | * these pages online first. | |
741 | */ | |
742 | pgs_ol = has->ha_end_pfn - start_pfn; | |
743 | if (pgs_ol > pfn_cnt) | |
744 | pgs_ol = pfn_cnt; | |
745 | hv_bring_pgs_online(start_pfn, pgs_ol); | |
746 | has->covered_end_pfn += pgs_ol; | |
747 | has->covered_start_pfn += pgs_ol; | |
748 | pfn_cnt -= pgs_ol; | |
749 | } | |
750 | ||
751 | if ((has->ha_end_pfn < has->end_pfn) && (pfn_cnt > 0)) { | |
752 | /* | |
753 | * We have some residual hot add range | |
754 | * that needs to be hot added; hot add | |
755 | * it now. Hot add a multiple of | |
756 | * of HA_CHUNK that fully covers the pages | |
757 | * we have. | |
758 | */ | |
759 | size = (has->end_pfn - has->ha_end_pfn); | |
760 | if (pfn_cnt <= size) { | |
761 | size = ((pfn_cnt / HA_CHUNK) * HA_CHUNK); | |
762 | if (pfn_cnt % HA_CHUNK) | |
763 | size += HA_CHUNK; | |
764 | } else { | |
765 | pfn_cnt = size; | |
766 | } | |
767 | hv_mem_hot_add(has->ha_end_pfn, size, pfn_cnt, has); | |
768 | } | |
769 | /* | |
770 | * If we managed to online any pages that were given to us, | |
771 | * we declare success. | |
772 | */ | |
773 | return has->covered_end_pfn - old_covered_state; | |
774 | ||
775 | } | |
776 | ||
777 | return 0; | |
778 | } | |
779 | ||
780 | static unsigned long process_hot_add(unsigned long pg_start, | |
781 | unsigned long pfn_cnt, | |
782 | unsigned long rg_start, | |
783 | unsigned long rg_size) | |
784 | { | |
785 | struct hv_hotadd_state *ha_region = NULL; | |
786 | ||
787 | if (pfn_cnt == 0) | |
788 | return 0; | |
789 | ||
790 | if (!dm_device.host_specified_ha_region) | |
791 | if (pfn_covered(pg_start, pfn_cnt)) | |
792 | goto do_pg_range; | |
793 | ||
794 | /* | |
795 | * If the host has specified a hot-add range; deal with it first. | |
796 | */ | |
797 | ||
647965a2 | 798 | if (rg_size != 0) { |
1cac8cd4 S |
799 | ha_region = kzalloc(sizeof(struct hv_hotadd_state), GFP_KERNEL); |
800 | if (!ha_region) | |
801 | return 0; | |
802 | ||
803 | INIT_LIST_HEAD(&ha_region->list); | |
804 | ||
805 | list_add_tail(&ha_region->list, &dm_device.ha_region_list); | |
806 | ha_region->start_pfn = rg_start; | |
807 | ha_region->ha_end_pfn = rg_start; | |
808 | ha_region->covered_start_pfn = pg_start; | |
809 | ha_region->covered_end_pfn = pg_start; | |
810 | ha_region->end_pfn = rg_start + rg_size; | |
811 | } | |
812 | ||
813 | do_pg_range: | |
814 | /* | |
815 | * Process the page range specified; bringing them | |
816 | * online if possible. | |
817 | */ | |
818 | return handle_pg_range(pg_start, pfn_cnt); | |
819 | } | |
820 | ||
821 | #endif | |
822 | ||
823 | static void hot_add_req(struct work_struct *dummy) | |
824 | { | |
825 | struct dm_hot_add_response resp; | |
826 | #ifdef CONFIG_MEMORY_HOTPLUG | |
827 | unsigned long pg_start, pfn_cnt; | |
828 | unsigned long rg_start, rg_sz; | |
829 | #endif | |
830 | struct hv_dynmem_device *dm = &dm_device; | |
831 | ||
9aa8b50b S |
832 | memset(&resp, 0, sizeof(struct dm_hot_add_response)); |
833 | resp.hdr.type = DM_MEM_HOT_ADD_RESPONSE; | |
834 | resp.hdr.size = sizeof(struct dm_hot_add_response); | |
9aa8b50b | 835 | |
1cac8cd4 S |
836 | #ifdef CONFIG_MEMORY_HOTPLUG |
837 | pg_start = dm->ha_wrk.ha_page_range.finfo.start_page; | |
838 | pfn_cnt = dm->ha_wrk.ha_page_range.finfo.page_cnt; | |
9aa8b50b | 839 | |
1cac8cd4 S |
840 | rg_start = dm->ha_wrk.ha_region_range.finfo.start_page; |
841 | rg_sz = dm->ha_wrk.ha_region_range.finfo.page_cnt; | |
842 | ||
843 | if ((rg_start == 0) && (!dm->host_specified_ha_region)) { | |
844 | unsigned long region_size; | |
845 | unsigned long region_start; | |
846 | ||
847 | /* | |
848 | * The host has not specified the hot-add region. | |
849 | * Based on the hot-add page range being specified, | |
850 | * compute a hot-add region that can cover the pages | |
851 | * that need to be hot-added while ensuring the alignment | |
852 | * and size requirements of Linux as it relates to hot-add. | |
853 | */ | |
854 | region_start = pg_start; | |
855 | region_size = (pfn_cnt / HA_CHUNK) * HA_CHUNK; | |
856 | if (pfn_cnt % HA_CHUNK) | |
857 | region_size += HA_CHUNK; | |
858 | ||
859 | region_start = (pg_start / HA_CHUNK) * HA_CHUNK; | |
860 | ||
861 | rg_start = region_start; | |
862 | rg_sz = region_size; | |
863 | } | |
864 | ||
7f4f2302 S |
865 | if (do_hot_add) |
866 | resp.page_count = process_hot_add(pg_start, pfn_cnt, | |
867 | rg_start, rg_sz); | |
1cac8cd4 | 868 | #endif |
7f4f2302 S |
869 | /* |
870 | * The result field of the response structure has the | |
871 | * following semantics: | |
872 | * | |
873 | * 1. If all or some pages hot-added: Guest should return success. | |
874 | * | |
875 | * 2. If no pages could be hot-added: | |
876 | * | |
877 | * If the guest returns success, then the host | |
878 | * will not attempt any further hot-add operations. This | |
879 | * signifies a permanent failure. | |
880 | * | |
881 | * If the guest returns failure, then this failure will be | |
882 | * treated as a transient failure and the host may retry the | |
883 | * hot-add operation after some delay. | |
884 | */ | |
1cac8cd4 S |
885 | if (resp.page_count > 0) |
886 | resp.result = 1; | |
7f4f2302 S |
887 | else if (!do_hot_add) |
888 | resp.result = 1; | |
1cac8cd4 S |
889 | else |
890 | resp.result = 0; | |
891 | ||
892 | if (!do_hot_add || (resp.page_count == 0)) | |
893 | pr_info("Memory hot add failed\n"); | |
894 | ||
895 | dm->state = DM_INITIALIZED; | |
20138d6c | 896 | resp.hdr.trans_id = atomic_inc_return(&trans_id); |
1cac8cd4 | 897 | vmbus_sendpacket(dm->dev->channel, &resp, |
9aa8b50b S |
898 | sizeof(struct dm_hot_add_response), |
899 | (unsigned long)NULL, | |
900 | VM_PKT_DATA_INBAND, 0); | |
9aa8b50b S |
901 | } |
902 | ||
903 | static void process_info(struct hv_dynmem_device *dm, struct dm_info_msg *msg) | |
904 | { | |
6427a0d7 S |
905 | struct dm_info_header *info_hdr; |
906 | ||
907 | info_hdr = (struct dm_info_header *)msg->info; | |
908 | ||
909 | switch (info_hdr->type) { | |
9aa8b50b S |
910 | case INFO_TYPE_MAX_PAGE_CNT: |
911 | pr_info("Received INFO_TYPE_MAX_PAGE_CNT\n"); | |
6427a0d7 | 912 | pr_info("Data Size is %d\n", info_hdr->data_size); |
9aa8b50b S |
913 | break; |
914 | default: | |
6427a0d7 | 915 | pr_info("Received Unknown type: %d\n", info_hdr->type); |
9aa8b50b S |
916 | } |
917 | } | |
918 | ||
a6025a2a | 919 | static unsigned long compute_balloon_floor(void) |
1c7db96f S |
920 | { |
921 | unsigned long min_pages; | |
922 | #define MB2PAGES(mb) ((mb) << (20 - PAGE_SHIFT)) | |
923 | /* Simple continuous piecewiese linear function: | |
924 | * max MiB -> min MiB gradient | |
925 | * 0 0 | |
926 | * 16 16 | |
927 | * 32 24 | |
928 | * 128 72 (1/2) | |
929 | * 512 168 (1/4) | |
930 | * 2048 360 (1/8) | |
931 | * 8192 552 (1/32) | |
932 | * 32768 1320 | |
933 | * 131072 4392 | |
934 | */ | |
935 | if (totalram_pages < MB2PAGES(128)) | |
936 | min_pages = MB2PAGES(8) + (totalram_pages >> 1); | |
937 | else if (totalram_pages < MB2PAGES(512)) | |
938 | min_pages = MB2PAGES(40) + (totalram_pages >> 2); | |
939 | else if (totalram_pages < MB2PAGES(2048)) | |
940 | min_pages = MB2PAGES(104) + (totalram_pages >> 3); | |
941 | else | |
942 | min_pages = MB2PAGES(296) + (totalram_pages >> 5); | |
943 | #undef MB2PAGES | |
944 | return min_pages; | |
945 | } | |
946 | ||
9aa8b50b S |
947 | /* |
948 | * Post our status as it relates memory pressure to the | |
949 | * host. Host expects the guests to post this status | |
950 | * periodically at 1 second intervals. | |
951 | * | |
952 | * The metrics specified in this protocol are very Windows | |
953 | * specific and so we cook up numbers here to convey our memory | |
954 | * pressure. | |
955 | */ | |
956 | ||
957 | static void post_status(struct hv_dynmem_device *dm) | |
958 | { | |
959 | struct dm_status status; | |
0731572b | 960 | struct sysinfo val; |
ae339336 S |
961 | unsigned long now = jiffies; |
962 | unsigned long last_post = last_post_time; | |
9aa8b50b | 963 | |
e500d158 S |
964 | if (pressure_report_delay > 0) { |
965 | --pressure_report_delay; | |
966 | return; | |
967 | } | |
ae339336 S |
968 | |
969 | if (!time_after(now, (last_post_time + HZ))) | |
970 | return; | |
971 | ||
0731572b | 972 | si_meminfo(&val); |
9aa8b50b S |
973 | memset(&status, 0, sizeof(struct dm_status)); |
974 | status.hdr.type = DM_STATUS_REPORT; | |
975 | status.hdr.size = sizeof(struct dm_status); | |
976 | status.hdr.trans_id = atomic_inc_return(&trans_id); | |
977 | ||
0731572b S |
978 | /* |
979 | * The host expects the guest to report free memory. | |
980 | * Further, the host expects the pressure information to | |
981 | * include the ballooned out pages. | |
1c7db96f S |
982 | * For a given amount of memory that we are managing, we |
983 | * need to compute a floor below which we should not balloon. | |
984 | * Compute this and add it to the pressure report. | |
0731572b S |
985 | */ |
986 | status.num_avail = val.freeram; | |
1c7db96f S |
987 | status.num_committed = vm_memory_committed() + |
988 | dm->num_pages_ballooned + | |
989 | compute_balloon_floor(); | |
9aa8b50b | 990 | |
c5e2254f S |
991 | /* |
992 | * If our transaction ID is no longer current, just don't | |
993 | * send the status. This can happen if we were interrupted | |
994 | * after we picked our transaction ID. | |
995 | */ | |
996 | if (status.hdr.trans_id != atomic_read(&trans_id)) | |
997 | return; | |
998 | ||
ae339336 S |
999 | /* |
1000 | * If the last post time that we sampled has changed, | |
1001 | * we have raced, don't post the status. | |
1002 | */ | |
1003 | if (last_post != last_post_time) | |
1004 | return; | |
1005 | ||
1006 | last_post_time = jiffies; | |
9aa8b50b S |
1007 | vmbus_sendpacket(dm->dev->channel, &status, |
1008 | sizeof(struct dm_status), | |
1009 | (unsigned long)NULL, | |
1010 | VM_PKT_DATA_INBAND, 0); | |
1011 | ||
1012 | } | |
1013 | ||
989623c7 | 1014 | static void free_balloon_pages(struct hv_dynmem_device *dm, |
9aa8b50b S |
1015 | union dm_mem_page_range *range_array) |
1016 | { | |
1017 | int num_pages = range_array->finfo.page_cnt; | |
1018 | __u64 start_frame = range_array->finfo.start_page; | |
1019 | struct page *pg; | |
1020 | int i; | |
1021 | ||
1022 | for (i = 0; i < num_pages; i++) { | |
1023 | pg = pfn_to_page(i + start_frame); | |
1024 | __free_page(pg); | |
1025 | dm->num_pages_ballooned--; | |
1026 | } | |
1027 | } | |
1028 | ||
1029 | ||
1030 | ||
1031 | static int alloc_balloon_pages(struct hv_dynmem_device *dm, int num_pages, | |
1032 | struct dm_balloon_response *bl_resp, int alloc_unit, | |
1033 | bool *alloc_error) | |
1034 | { | |
1035 | int i = 0; | |
1036 | struct page *pg; | |
1037 | ||
1038 | if (num_pages < alloc_unit) | |
1039 | return 0; | |
1040 | ||
1041 | for (i = 0; (i * alloc_unit) < num_pages; i++) { | |
1042 | if (bl_resp->hdr.size + sizeof(union dm_mem_page_range) > | |
1043 | PAGE_SIZE) | |
1044 | return i * alloc_unit; | |
1045 | ||
1046 | /* | |
1047 | * We execute this code in a thread context. Furthermore, | |
1048 | * we don't want the kernel to try too hard. | |
1049 | */ | |
1050 | pg = alloc_pages(GFP_HIGHUSER | __GFP_NORETRY | | |
1051 | __GFP_NOMEMALLOC | __GFP_NOWARN, | |
1052 | get_order(alloc_unit << PAGE_SHIFT)); | |
1053 | ||
1054 | if (!pg) { | |
1055 | *alloc_error = true; | |
1056 | return i * alloc_unit; | |
1057 | } | |
1058 | ||
1059 | ||
1060 | dm->num_pages_ballooned += alloc_unit; | |
1061 | ||
f766dc1e S |
1062 | /* |
1063 | * If we allocatted 2M pages; split them so we | |
1064 | * can free them in any order we get. | |
1065 | */ | |
1066 | ||
1067 | if (alloc_unit != 1) | |
1068 | split_page(pg, get_order(alloc_unit << PAGE_SHIFT)); | |
1069 | ||
9aa8b50b S |
1070 | bl_resp->range_count++; |
1071 | bl_resp->range_array[i].finfo.start_page = | |
1072 | page_to_pfn(pg); | |
1073 | bl_resp->range_array[i].finfo.page_cnt = alloc_unit; | |
1074 | bl_resp->hdr.size += sizeof(union dm_mem_page_range); | |
1075 | ||
1076 | } | |
1077 | ||
1078 | return num_pages; | |
1079 | } | |
1080 | ||
1081 | ||
1082 | ||
6571b2da | 1083 | static void balloon_up(struct work_struct *dummy) |
9aa8b50b | 1084 | { |
6571b2da | 1085 | int num_pages = dm_device.balloon_wrk.num_pages; |
9aa8b50b S |
1086 | int num_ballooned = 0; |
1087 | struct dm_balloon_response *bl_resp; | |
1088 | int alloc_unit; | |
1089 | int ret; | |
1090 | bool alloc_error = false; | |
1091 | bool done = false; | |
1092 | int i; | |
1093 | ||
1094 | ||
1095 | /* | |
f766dc1e S |
1096 | * We will attempt 2M allocations. However, if we fail to |
1097 | * allocate 2M chunks, we will go back to 4k allocations. | |
9aa8b50b | 1098 | */ |
f766dc1e | 1099 | alloc_unit = 512; |
9aa8b50b S |
1100 | |
1101 | while (!done) { | |
1102 | bl_resp = (struct dm_balloon_response *)send_buffer; | |
1103 | memset(send_buffer, 0, PAGE_SIZE); | |
1104 | bl_resp->hdr.type = DM_BALLOON_RESPONSE; | |
9aa8b50b S |
1105 | bl_resp->hdr.size = sizeof(struct dm_balloon_response); |
1106 | bl_resp->more_pages = 1; | |
1107 | ||
1108 | ||
1109 | num_pages -= num_ballooned; | |
6571b2da | 1110 | num_ballooned = alloc_balloon_pages(&dm_device, num_pages, |
9aa8b50b S |
1111 | bl_resp, alloc_unit, |
1112 | &alloc_error); | |
1113 | ||
f766dc1e S |
1114 | if ((alloc_error) && (alloc_unit != 1)) { |
1115 | alloc_unit = 1; | |
1116 | continue; | |
1117 | } | |
1118 | ||
9aa8b50b S |
1119 | if ((alloc_error) || (num_ballooned == num_pages)) { |
1120 | bl_resp->more_pages = 0; | |
1121 | done = true; | |
6571b2da | 1122 | dm_device.state = DM_INITIALIZED; |
9aa8b50b S |
1123 | } |
1124 | ||
1125 | /* | |
1126 | * We are pushing a lot of data through the channel; | |
1127 | * deal with transient failures caused because of the | |
1128 | * lack of space in the ring buffer. | |
1129 | */ | |
1130 | ||
1131 | do { | |
20138d6c | 1132 | bl_resp->hdr.trans_id = atomic_inc_return(&trans_id); |
9aa8b50b S |
1133 | ret = vmbus_sendpacket(dm_device.dev->channel, |
1134 | bl_resp, | |
1135 | bl_resp->hdr.size, | |
1136 | (unsigned long)NULL, | |
1137 | VM_PKT_DATA_INBAND, 0); | |
1138 | ||
1139 | if (ret == -EAGAIN) | |
1140 | msleep(20); | |
ae339336 | 1141 | post_status(&dm_device); |
9aa8b50b S |
1142 | } while (ret == -EAGAIN); |
1143 | ||
1144 | if (ret) { | |
1145 | /* | |
1146 | * Free up the memory we allocatted. | |
1147 | */ | |
1148 | pr_info("Balloon response failed\n"); | |
1149 | ||
1150 | for (i = 0; i < bl_resp->range_count; i++) | |
6571b2da | 1151 | free_balloon_pages(&dm_device, |
9aa8b50b S |
1152 | &bl_resp->range_array[i]); |
1153 | ||
1154 | done = true; | |
1155 | } | |
1156 | } | |
1157 | ||
1158 | } | |
1159 | ||
1160 | static void balloon_down(struct hv_dynmem_device *dm, | |
1161 | struct dm_unballoon_request *req) | |
1162 | { | |
1163 | union dm_mem_page_range *range_array = req->range_array; | |
1164 | int range_count = req->range_count; | |
1165 | struct dm_unballoon_response resp; | |
1166 | int i; | |
1167 | ||
ae339336 | 1168 | for (i = 0; i < range_count; i++) { |
9aa8b50b | 1169 | free_balloon_pages(dm, &range_array[i]); |
ae339336 S |
1170 | post_status(&dm_device); |
1171 | } | |
9aa8b50b S |
1172 | |
1173 | if (req->more_pages == 1) | |
1174 | return; | |
1175 | ||
1176 | memset(&resp, 0, sizeof(struct dm_unballoon_response)); | |
1177 | resp.hdr.type = DM_UNBALLOON_RESPONSE; | |
1178 | resp.hdr.trans_id = atomic_inc_return(&trans_id); | |
1179 | resp.hdr.size = sizeof(struct dm_unballoon_response); | |
1180 | ||
1181 | vmbus_sendpacket(dm_device.dev->channel, &resp, | |
1182 | sizeof(struct dm_unballoon_response), | |
1183 | (unsigned long)NULL, | |
1184 | VM_PKT_DATA_INBAND, 0); | |
1185 | ||
1186 | dm->state = DM_INITIALIZED; | |
1187 | } | |
1188 | ||
1189 | static void balloon_onchannelcallback(void *context); | |
1190 | ||
1191 | static int dm_thread_func(void *dm_dev) | |
1192 | { | |
1193 | struct hv_dynmem_device *dm = dm_dev; | |
1194 | int t; | |
9aa8b50b S |
1195 | |
1196 | while (!kthread_should_stop()) { | |
5dba4c56 S |
1197 | t = wait_for_completion_interruptible_timeout( |
1198 | &dm_device.config_event, 1*HZ); | |
9aa8b50b S |
1199 | /* |
1200 | * The host expects us to post information on the memory | |
1201 | * pressure every second. | |
1202 | */ | |
1203 | ||
1204 | if (t == 0) | |
1205 | post_status(dm); | |
1206 | ||
9aa8b50b S |
1207 | } |
1208 | ||
1209 | return 0; | |
1210 | } | |
1211 | ||
1212 | ||
1213 | static void version_resp(struct hv_dynmem_device *dm, | |
1214 | struct dm_version_response *vresp) | |
1215 | { | |
1216 | struct dm_version_request version_req; | |
1217 | int ret; | |
1218 | ||
1219 | if (vresp->is_accepted) { | |
1220 | /* | |
1221 | * We are done; wakeup the | |
1222 | * context waiting for version | |
1223 | * negotiation. | |
1224 | */ | |
1225 | complete(&dm->host_event); | |
1226 | return; | |
1227 | } | |
1228 | /* | |
1229 | * If there are more versions to try, continue | |
1230 | * with negotiations; if not | |
1231 | * shutdown the service since we are not able | |
1232 | * to negotiate a suitable version number | |
1233 | * with the host. | |
1234 | */ | |
1235 | if (dm->next_version == 0) | |
1236 | goto version_error; | |
1237 | ||
1238 | dm->next_version = 0; | |
1239 | memset(&version_req, 0, sizeof(struct dm_version_request)); | |
1240 | version_req.hdr.type = DM_VERSION_REQUEST; | |
1241 | version_req.hdr.size = sizeof(struct dm_version_request); | |
1242 | version_req.hdr.trans_id = atomic_inc_return(&trans_id); | |
1243 | version_req.version.version = DYNMEM_PROTOCOL_VERSION_WIN7; | |
1244 | version_req.is_last_attempt = 1; | |
1245 | ||
1246 | ret = vmbus_sendpacket(dm->dev->channel, &version_req, | |
1247 | sizeof(struct dm_version_request), | |
1248 | (unsigned long)NULL, | |
1249 | VM_PKT_DATA_INBAND, 0); | |
1250 | ||
1251 | if (ret) | |
1252 | goto version_error; | |
1253 | ||
1254 | return; | |
1255 | ||
1256 | version_error: | |
1257 | dm->state = DM_INIT_ERROR; | |
1258 | complete(&dm->host_event); | |
1259 | } | |
1260 | ||
1261 | static void cap_resp(struct hv_dynmem_device *dm, | |
1262 | struct dm_capabilities_resp_msg *cap_resp) | |
1263 | { | |
1264 | if (!cap_resp->is_accepted) { | |
1265 | pr_info("Capabilities not accepted by host\n"); | |
1266 | dm->state = DM_INIT_ERROR; | |
1267 | } | |
1268 | complete(&dm->host_event); | |
1269 | } | |
1270 | ||
1271 | static void balloon_onchannelcallback(void *context) | |
1272 | { | |
1273 | struct hv_device *dev = context; | |
1274 | u32 recvlen; | |
1275 | u64 requestid; | |
1276 | struct dm_message *dm_msg; | |
1277 | struct dm_header *dm_hdr; | |
1278 | struct hv_dynmem_device *dm = hv_get_drvdata(dev); | |
6571b2da | 1279 | struct dm_balloon *bal_msg; |
c51af826 S |
1280 | struct dm_hot_add *ha_msg; |
1281 | union dm_mem_page_range *ha_pg_range; | |
1cac8cd4 | 1282 | union dm_mem_page_range *ha_region; |
9aa8b50b S |
1283 | |
1284 | memset(recv_buffer, 0, sizeof(recv_buffer)); | |
1285 | vmbus_recvpacket(dev->channel, recv_buffer, | |
1286 | PAGE_SIZE, &recvlen, &requestid); | |
1287 | ||
1288 | if (recvlen > 0) { | |
1289 | dm_msg = (struct dm_message *)recv_buffer; | |
1290 | dm_hdr = &dm_msg->hdr; | |
1291 | ||
1292 | switch (dm_hdr->type) { | |
1293 | case DM_VERSION_RESPONSE: | |
1294 | version_resp(dm, | |
1295 | (struct dm_version_response *)dm_msg); | |
1296 | break; | |
1297 | ||
1298 | case DM_CAPABILITIES_RESPONSE: | |
1299 | cap_resp(dm, | |
1300 | (struct dm_capabilities_resp_msg *)dm_msg); | |
1301 | break; | |
1302 | ||
1303 | case DM_BALLOON_REQUEST: | |
6571b2da S |
1304 | if (dm->state == DM_BALLOON_UP) |
1305 | pr_warn("Currently ballooning\n"); | |
1306 | bal_msg = (struct dm_balloon *)recv_buffer; | |
9aa8b50b | 1307 | dm->state = DM_BALLOON_UP; |
6571b2da S |
1308 | dm_device.balloon_wrk.num_pages = bal_msg->num_pages; |
1309 | schedule_work(&dm_device.balloon_wrk.wrk); | |
9aa8b50b S |
1310 | break; |
1311 | ||
1312 | case DM_UNBALLOON_REQUEST: | |
1313 | dm->state = DM_BALLOON_DOWN; | |
1314 | balloon_down(dm, | |
1315 | (struct dm_unballoon_request *)recv_buffer); | |
1316 | break; | |
1317 | ||
1318 | case DM_MEM_HOT_ADD_REQUEST: | |
c51af826 S |
1319 | if (dm->state == DM_HOT_ADD) |
1320 | pr_warn("Currently hot-adding\n"); | |
9aa8b50b | 1321 | dm->state = DM_HOT_ADD; |
c51af826 | 1322 | ha_msg = (struct dm_hot_add *)recv_buffer; |
1cac8cd4 S |
1323 | if (ha_msg->hdr.size == sizeof(struct dm_hot_add)) { |
1324 | /* | |
1325 | * This is a normal hot-add request specifying | |
1326 | * hot-add memory. | |
1327 | */ | |
1328 | ha_pg_range = &ha_msg->range; | |
1329 | dm->ha_wrk.ha_page_range = *ha_pg_range; | |
1330 | dm->ha_wrk.ha_region_range.page_range = 0; | |
1331 | } else { | |
1332 | /* | |
1333 | * Host is specifying that we first hot-add | |
1334 | * a region and then partially populate this | |
1335 | * region. | |
1336 | */ | |
1337 | dm->host_specified_ha_region = true; | |
1338 | ha_pg_range = &ha_msg->range; | |
1339 | ha_region = &ha_pg_range[1]; | |
1340 | dm->ha_wrk.ha_page_range = *ha_pg_range; | |
1341 | dm->ha_wrk.ha_region_range = *ha_region; | |
1342 | } | |
c51af826 | 1343 | schedule_work(&dm_device.ha_wrk.wrk); |
9aa8b50b S |
1344 | break; |
1345 | ||
1346 | case DM_INFO_MESSAGE: | |
1347 | process_info(dm, (struct dm_info_msg *)dm_msg); | |
1348 | break; | |
1349 | ||
1350 | default: | |
1351 | pr_err("Unhandled message: type: %d\n", dm_hdr->type); | |
1352 | ||
1353 | } | |
1354 | } | |
1355 | ||
1356 | } | |
1357 | ||
1358 | static int balloon_probe(struct hv_device *dev, | |
1359 | const struct hv_vmbus_device_id *dev_id) | |
1360 | { | |
1361 | int ret, t; | |
1362 | struct dm_version_request version_req; | |
1363 | struct dm_capabilities cap_msg; | |
1364 | ||
1365 | do_hot_add = hot_add; | |
1366 | ||
1367 | /* | |
1368 | * First allocate a send buffer. | |
1369 | */ | |
1370 | ||
1371 | send_buffer = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
1372 | if (!send_buffer) | |
1373 | return -ENOMEM; | |
1374 | ||
1375 | ret = vmbus_open(dev->channel, dm_ring_size, dm_ring_size, NULL, 0, | |
1376 | balloon_onchannelcallback, dev); | |
1377 | ||
1378 | if (ret) | |
33080c1c | 1379 | goto probe_error0; |
9aa8b50b S |
1380 | |
1381 | dm_device.dev = dev; | |
1382 | dm_device.state = DM_INITIALIZING; | |
1383 | dm_device.next_version = DYNMEM_PROTOCOL_VERSION_WIN7; | |
1384 | init_completion(&dm_device.host_event); | |
1385 | init_completion(&dm_device.config_event); | |
1cac8cd4 | 1386 | INIT_LIST_HEAD(&dm_device.ha_region_list); |
6571b2da | 1387 | INIT_WORK(&dm_device.balloon_wrk.wrk, balloon_up); |
c51af826 | 1388 | INIT_WORK(&dm_device.ha_wrk.wrk, hot_add_req); |
1cac8cd4 | 1389 | dm_device.host_specified_ha_region = false; |
9aa8b50b S |
1390 | |
1391 | dm_device.thread = | |
1392 | kthread_run(dm_thread_func, &dm_device, "hv_balloon"); | |
1393 | if (IS_ERR(dm_device.thread)) { | |
1394 | ret = PTR_ERR(dm_device.thread); | |
33080c1c | 1395 | goto probe_error1; |
9aa8b50b S |
1396 | } |
1397 | ||
1cac8cd4 S |
1398 | #ifdef CONFIG_MEMORY_HOTPLUG |
1399 | set_online_page_callback(&hv_online_page); | |
1400 | #endif | |
1401 | ||
9aa8b50b S |
1402 | hv_set_drvdata(dev, &dm_device); |
1403 | /* | |
1404 | * Initiate the hand shake with the host and negotiate | |
1405 | * a version that the host can support. We start with the | |
1406 | * highest version number and go down if the host cannot | |
1407 | * support it. | |
1408 | */ | |
1409 | memset(&version_req, 0, sizeof(struct dm_version_request)); | |
1410 | version_req.hdr.type = DM_VERSION_REQUEST; | |
1411 | version_req.hdr.size = sizeof(struct dm_version_request); | |
1412 | version_req.hdr.trans_id = atomic_inc_return(&trans_id); | |
1413 | version_req.version.version = DYNMEM_PROTOCOL_VERSION_WIN8; | |
1414 | version_req.is_last_attempt = 0; | |
1415 | ||
1416 | ret = vmbus_sendpacket(dev->channel, &version_req, | |
1417 | sizeof(struct dm_version_request), | |
1418 | (unsigned long)NULL, | |
7a64b864 | 1419 | VM_PKT_DATA_INBAND, 0); |
9aa8b50b | 1420 | if (ret) |
33080c1c | 1421 | goto probe_error2; |
9aa8b50b S |
1422 | |
1423 | t = wait_for_completion_timeout(&dm_device.host_event, 5*HZ); | |
1424 | if (t == 0) { | |
1425 | ret = -ETIMEDOUT; | |
33080c1c | 1426 | goto probe_error2; |
9aa8b50b S |
1427 | } |
1428 | ||
1429 | /* | |
1430 | * If we could not negotiate a compatible version with the host | |
1431 | * fail the probe function. | |
1432 | */ | |
1433 | if (dm_device.state == DM_INIT_ERROR) { | |
1434 | ret = -ETIMEDOUT; | |
33080c1c | 1435 | goto probe_error2; |
9aa8b50b S |
1436 | } |
1437 | /* | |
1438 | * Now submit our capabilities to the host. | |
1439 | */ | |
1440 | memset(&cap_msg, 0, sizeof(struct dm_capabilities)); | |
1441 | cap_msg.hdr.type = DM_CAPABILITIES_REPORT; | |
1442 | cap_msg.hdr.size = sizeof(struct dm_capabilities); | |
1443 | cap_msg.hdr.trans_id = atomic_inc_return(&trans_id); | |
1444 | ||
1445 | cap_msg.caps.cap_bits.balloon = 1; | |
9aa8b50b S |
1446 | cap_msg.caps.cap_bits.hot_add = 1; |
1447 | ||
647965a2 S |
1448 | /* |
1449 | * Specify our alignment requirements as it relates | |
1450 | * memory hot-add. Specify 128MB alignment. | |
1451 | */ | |
1452 | cap_msg.caps.cap_bits.hot_add_alignment = 7; | |
1453 | ||
9aa8b50b S |
1454 | /* |
1455 | * Currently the host does not use these | |
1456 | * values and we set them to what is done in the | |
1457 | * Windows driver. | |
1458 | */ | |
1459 | cap_msg.min_page_cnt = 0; | |
1460 | cap_msg.max_page_number = -1; | |
1461 | ||
1462 | ret = vmbus_sendpacket(dev->channel, &cap_msg, | |
1463 | sizeof(struct dm_capabilities), | |
1464 | (unsigned long)NULL, | |
7a64b864 | 1465 | VM_PKT_DATA_INBAND, 0); |
9aa8b50b | 1466 | if (ret) |
33080c1c | 1467 | goto probe_error2; |
9aa8b50b S |
1468 | |
1469 | t = wait_for_completion_timeout(&dm_device.host_event, 5*HZ); | |
1470 | if (t == 0) { | |
1471 | ret = -ETIMEDOUT; | |
33080c1c | 1472 | goto probe_error2; |
9aa8b50b S |
1473 | } |
1474 | ||
1475 | /* | |
1476 | * If the host does not like our capabilities, | |
1477 | * fail the probe function. | |
1478 | */ | |
1479 | if (dm_device.state == DM_INIT_ERROR) { | |
1480 | ret = -ETIMEDOUT; | |
33080c1c | 1481 | goto probe_error2; |
9aa8b50b S |
1482 | } |
1483 | ||
1484 | dm_device.state = DM_INITIALIZED; | |
1485 | ||
1486 | return 0; | |
1487 | ||
33080c1c | 1488 | probe_error2: |
1cac8cd4 S |
1489 | #ifdef CONFIG_MEMORY_HOTPLUG |
1490 | restore_online_page_callback(&hv_online_page); | |
1491 | #endif | |
9aa8b50b S |
1492 | kthread_stop(dm_device.thread); |
1493 | ||
33080c1c | 1494 | probe_error1: |
9aa8b50b | 1495 | vmbus_close(dev->channel); |
33080c1c S |
1496 | probe_error0: |
1497 | kfree(send_buffer); | |
9aa8b50b S |
1498 | return ret; |
1499 | } | |
1500 | ||
1501 | static int balloon_remove(struct hv_device *dev) | |
1502 | { | |
1503 | struct hv_dynmem_device *dm = hv_get_drvdata(dev); | |
1cac8cd4 S |
1504 | struct list_head *cur, *tmp; |
1505 | struct hv_hotadd_state *has; | |
9aa8b50b S |
1506 | |
1507 | if (dm->num_pages_ballooned != 0) | |
1508 | pr_warn("Ballooned pages: %d\n", dm->num_pages_ballooned); | |
1509 | ||
6571b2da | 1510 | cancel_work_sync(&dm->balloon_wrk.wrk); |
c51af826 | 1511 | cancel_work_sync(&dm->ha_wrk.wrk); |
1cac8cd4 | 1512 | |
9aa8b50b S |
1513 | vmbus_close(dev->channel); |
1514 | kthread_stop(dm->thread); | |
33080c1c | 1515 | kfree(send_buffer); |
1cac8cd4 S |
1516 | #ifdef CONFIG_MEMORY_HOTPLUG |
1517 | restore_online_page_callback(&hv_online_page); | |
1518 | #endif | |
1519 | list_for_each_safe(cur, tmp, &dm->ha_region_list) { | |
1520 | has = list_entry(cur, struct hv_hotadd_state, list); | |
1521 | list_del(&has->list); | |
1522 | kfree(has); | |
1523 | } | |
9aa8b50b S |
1524 | |
1525 | return 0; | |
1526 | } | |
1527 | ||
1528 | static const struct hv_vmbus_device_id id_table[] = { | |
1529 | /* Dynamic Memory Class ID */ | |
1530 | /* 525074DC-8985-46e2-8057-A307DC18A502 */ | |
d13984e5 | 1531 | { HV_DM_GUID, }, |
9aa8b50b S |
1532 | { }, |
1533 | }; | |
1534 | ||
1535 | MODULE_DEVICE_TABLE(vmbus, id_table); | |
1536 | ||
1537 | static struct hv_driver balloon_drv = { | |
1538 | .name = "hv_balloon", | |
1539 | .id_table = id_table, | |
1540 | .probe = balloon_probe, | |
1541 | .remove = balloon_remove, | |
1542 | }; | |
1543 | ||
1544 | static int __init init_balloon_drv(void) | |
1545 | { | |
1546 | ||
1547 | return vmbus_driver_register(&balloon_drv); | |
1548 | } | |
1549 | ||
9aa8b50b | 1550 | module_init(init_balloon_drv); |
9aa8b50b S |
1551 | |
1552 | MODULE_DESCRIPTION("Hyper-V Balloon"); | |
9aa8b50b | 1553 | MODULE_LICENSE("GPL"); |