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