blob: 7e17a5495e029cc02b127f4c0cb997b43aa09dfe [file] [log] [blame]
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001/*
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>
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -080032#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
61enum {
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
77enum 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
107union dm_version {
108 struct {
109 __u16 minor_version;
110 __u16 major_version;
111 };
112 __u32 version;
113} __packed;
114
115
116union dm_caps {
117 struct {
118 __u64 balloon:1;
119 __u64 hot_add:1;
K. Y. Srinivasan647965a2013-03-29 07:36:11 -0700120 /*
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;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800128 } cap_bits;
129 __u64 caps;
130} __packed;
131
132union 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
158struct 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
169struct 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
189struct 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
207struct 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
218struct 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
235struct 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
261struct 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
280struct 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
301struct 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
324struct 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
338struct 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
352struct 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
377struct 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
387enum 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
397struct 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
411struct dm_info_msg {
K. Y. Srinivasan6427a0d2012-12-06 11:06:54 -0800412 struct dm_header hdr;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800413 __u32 reserved;
414 __u32 info_size;
415 __u8 info[];
416};
417
418/*
419 * End protocol definitions.
420 */
421
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700422/*
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
434struct 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
K. Y. Srinivasan6571b2d2013-03-15 12:25:40 -0700443struct balloon_state {
444 __u32 num_pages;
445 struct work_struct wrk;
446};
447
K. Y. Srinivasanc51af822013-03-15 12:25:41 -0700448struct hot_add_wrk {
449 union dm_mem_page_range ha_page_range;
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700450 union dm_mem_page_range ha_region_range;
K. Y. Srinivasanc51af822013-03-15 12:25:41 -0700451 struct work_struct wrk;
452};
453
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700454static bool hot_add = true;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800455static bool do_hot_add;
K. Y. Srinivasane500d152013-02-08 15:57:15 -0800456/*
457 * Delay reporting memory pressure by
458 * the specified number of seconds.
459 */
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700460static uint pressure_report_delay = 45;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800461
462module_param(hot_add, bool, (S_IRUGO | S_IWUSR));
463MODULE_PARM_DESC(hot_add, "If set attempt memory hot_add");
464
K. Y. Srinivasane500d152013-02-08 15:57:15 -0800465module_param(pressure_report_delay, uint, (S_IRUGO | S_IWUSR));
466MODULE_PARM_DESC(pressure_report_delay, "Delay in secs in reporting pressure");
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800467static atomic_t trans_id = ATOMIC_INIT(0);
468
469static int dm_ring_size = (5 * PAGE_SIZE);
470
471/*
472 * Driver specific state.
473 */
474
475enum 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
485static __u8 recv_buffer[PAGE_SIZE];
486static __u8 *send_buffer;
487#define PAGES_IN_2M 512
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700488#define HA_CHUNK (32 * 1024)
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800489
490struct 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 /*
K. Y. Srinivasan6571b2d2013-03-15 12:25:40 -0700502 * State to manage the ballooning (up) operation.
503 */
504 struct balloon_state balloon_wrk;
505
506 /*
K. Y. Srinivasanc51af822013-03-15 12:25:41 -0700507 * State to execute the "hot-add" operation.
508 */
509 struct hot_add_wrk ha_wrk;
510
511 /*
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700512 * 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;
522 /*
K. Y. Srinivasan6571b2d2013-03-15 12:25:40 -0700523 * This thread handles hot-add
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800524 * 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
530 /*
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700531 * A list of hot-add regions.
532 */
533 struct list_head ha_region_list;
534
535 /*
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800536 * 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
543static struct hv_dynmem_device dm_device;
544
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700545#ifdef CONFIG_MEMORY_HOTPLUG
546
Wei Yongjuna6025a22013-03-20 23:25:59 +0800547static void hv_bring_pgs_online(unsigned long start_pfn, unsigned long size)
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800548{
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700549 int i;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800550
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700551 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}
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800559
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700560static 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;
K. Y. Srinivasaned07ec92013-07-14 22:38:11 -0700565 int i, nid;
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700566 unsigned long start_pfn;
567 unsigned long processed_pfn;
568 unsigned long total_pfn = pfn_count;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800569
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700570 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;
583
584 init_completion(&dm_device.ol_waitevent);
585 dm_device.ha_waiting = true;
586
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);
K. Y. Srinivasan7f4f2302013-03-18 13:51:38 -0700593 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 }
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700603 has->ha_end_pfn -= HA_CHUNK;
604 has->covered_end_pfn -= processed_pfn;
605 break;
606 }
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800607
608 /*
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700609 * Wait for the memory block to be onlined.
K. Y. Srinivasaned07ec92013-07-14 22:38:11 -0700610 * 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.
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800613 */
K. Y. Srinivasaned07ec92013-07-14 22:38:11 -0700614 wait_for_completion_timeout(&dm_device.ol_waitevent, 5*HZ);
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700615
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800616 }
617
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700618 return;
619}
620
621static 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
653static 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
703static 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
773static 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
K. Y. Srinivasan647965a2013-03-29 07:36:11 -0700791 if (rg_size != 0) {
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700792 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
806do_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
816static 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
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800825 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);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800828
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700829#ifdef CONFIG_MEMORY_HOTPLUG
830 pg_start = dm->ha_wrk.ha_page_range.finfo.start_page;
831 pfn_cnt = dm->ha_wrk.ha_page_range.finfo.page_cnt;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800832
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700833 rg_start = dm->ha_wrk.ha_region_range.finfo.start_page;
834 rg_sz = dm->ha_wrk.ha_region_range.finfo.page_cnt;
835
836 if ((rg_start == 0) && (!dm->host_specified_ha_region)) {
837 unsigned long region_size;
838 unsigned long region_start;
839
840 /*
841 * The host has not specified the hot-add region.
842 * Based on the hot-add page range being specified,
843 * compute a hot-add region that can cover the pages
844 * that need to be hot-added while ensuring the alignment
845 * and size requirements of Linux as it relates to hot-add.
846 */
847 region_start = pg_start;
848 region_size = (pfn_cnt / HA_CHUNK) * HA_CHUNK;
849 if (pfn_cnt % HA_CHUNK)
850 region_size += HA_CHUNK;
851
852 region_start = (pg_start / HA_CHUNK) * HA_CHUNK;
853
854 rg_start = region_start;
855 rg_sz = region_size;
856 }
857
K. Y. Srinivasan7f4f2302013-03-18 13:51:38 -0700858 if (do_hot_add)
859 resp.page_count = process_hot_add(pg_start, pfn_cnt,
860 rg_start, rg_sz);
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700861#endif
K. Y. Srinivasan7f4f2302013-03-18 13:51:38 -0700862 /*
863 * The result field of the response structure has the
864 * following semantics:
865 *
866 * 1. If all or some pages hot-added: Guest should return success.
867 *
868 * 2. If no pages could be hot-added:
869 *
870 * If the guest returns success, then the host
871 * will not attempt any further hot-add operations. This
872 * signifies a permanent failure.
873 *
874 * If the guest returns failure, then this failure will be
875 * treated as a transient failure and the host may retry the
876 * hot-add operation after some delay.
877 */
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700878 if (resp.page_count > 0)
879 resp.result = 1;
K. Y. Srinivasan7f4f2302013-03-18 13:51:38 -0700880 else if (!do_hot_add)
881 resp.result = 1;
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700882 else
883 resp.result = 0;
884
885 if (!do_hot_add || (resp.page_count == 0))
886 pr_info("Memory hot add failed\n");
887
888 dm->state = DM_INITIALIZED;
K. Y. Srinivasan20138d6c2013-07-17 17:27:27 -0700889 resp.hdr.trans_id = atomic_inc_return(&trans_id);
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700890 vmbus_sendpacket(dm->dev->channel, &resp,
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800891 sizeof(struct dm_hot_add_response),
892 (unsigned long)NULL,
893 VM_PKT_DATA_INBAND, 0);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800894}
895
896static void process_info(struct hv_dynmem_device *dm, struct dm_info_msg *msg)
897{
K. Y. Srinivasan6427a0d2012-12-06 11:06:54 -0800898 struct dm_info_header *info_hdr;
899
900 info_hdr = (struct dm_info_header *)msg->info;
901
902 switch (info_hdr->type) {
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800903 case INFO_TYPE_MAX_PAGE_CNT:
904 pr_info("Received INFO_TYPE_MAX_PAGE_CNT\n");
K. Y. Srinivasan6427a0d2012-12-06 11:06:54 -0800905 pr_info("Data Size is %d\n", info_hdr->data_size);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800906 break;
907 default:
K. Y. Srinivasan6427a0d2012-12-06 11:06:54 -0800908 pr_info("Received Unknown type: %d\n", info_hdr->type);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800909 }
910}
911
Wei Yongjuna6025a22013-03-20 23:25:59 +0800912static unsigned long compute_balloon_floor(void)
K. Y. Srinivasan1c7db962013-02-08 15:57:16 -0800913{
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
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800940/*
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
950static void post_status(struct hv_dynmem_device *dm)
951{
952 struct dm_status status;
K. Y. Srinivasan07315722013-01-25 16:18:47 -0800953 struct sysinfo val;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800954
K. Y. Srinivasane500d152013-02-08 15:57:15 -0800955 if (pressure_report_delay > 0) {
956 --pressure_report_delay;
957 return;
958 }
K. Y. Srinivasan07315722013-01-25 16:18:47 -0800959 si_meminfo(&val);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800960 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
K. Y. Srinivasan07315722013-01-25 16:18:47 -0800965 /*
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.
K. Y. Srinivasan1c7db962013-02-08 15:57:16 -0800969 * 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.
K. Y. Srinivasan07315722013-01-25 16:18:47 -0800972 */
973 status.num_avail = val.freeram;
K. Y. Srinivasan1c7db962013-02-08 15:57:16 -0800974 status.num_committed = vm_memory_committed() +
975 dm->num_pages_ballooned +
976 compute_balloon_floor();
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800977
K. Y. Srinivasanc5e22542013-07-14 22:38:12 -0700978 /*
979 * If our transaction ID is no longer current, just don't
980 * send the status. This can happen if we were interrupted
981 * after we picked our transaction ID.
982 */
983 if (status.hdr.trans_id != atomic_read(&trans_id))
984 return;
985
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800986 vmbus_sendpacket(dm->dev->channel, &status,
987 sizeof(struct dm_status),
988 (unsigned long)NULL,
989 VM_PKT_DATA_INBAND, 0);
990
991}
992
Greg Kroah-Hartman989623c2012-11-21 12:46:40 -0800993static void free_balloon_pages(struct hv_dynmem_device *dm,
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800994 union dm_mem_page_range *range_array)
995{
996 int num_pages = range_array->finfo.page_cnt;
997 __u64 start_frame = range_array->finfo.start_page;
998 struct page *pg;
999 int i;
1000
1001 for (i = 0; i < num_pages; i++) {
1002 pg = pfn_to_page(i + start_frame);
1003 __free_page(pg);
1004 dm->num_pages_ballooned--;
1005 }
1006}
1007
1008
1009
1010static int alloc_balloon_pages(struct hv_dynmem_device *dm, int num_pages,
1011 struct dm_balloon_response *bl_resp, int alloc_unit,
1012 bool *alloc_error)
1013{
1014 int i = 0;
1015 struct page *pg;
1016
1017 if (num_pages < alloc_unit)
1018 return 0;
1019
1020 for (i = 0; (i * alloc_unit) < num_pages; i++) {
1021 if (bl_resp->hdr.size + sizeof(union dm_mem_page_range) >
1022 PAGE_SIZE)
1023 return i * alloc_unit;
1024
1025 /*
1026 * We execute this code in a thread context. Furthermore,
1027 * we don't want the kernel to try too hard.
1028 */
1029 pg = alloc_pages(GFP_HIGHUSER | __GFP_NORETRY |
1030 __GFP_NOMEMALLOC | __GFP_NOWARN,
1031 get_order(alloc_unit << PAGE_SHIFT));
1032
1033 if (!pg) {
1034 *alloc_error = true;
1035 return i * alloc_unit;
1036 }
1037
1038
1039 dm->num_pages_ballooned += alloc_unit;
1040
K. Y. Srinivasanf766dc12013-03-18 13:51:37 -07001041 /*
1042 * If we allocatted 2M pages; split them so we
1043 * can free them in any order we get.
1044 */
1045
1046 if (alloc_unit != 1)
1047 split_page(pg, get_order(alloc_unit << PAGE_SHIFT));
1048
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001049 bl_resp->range_count++;
1050 bl_resp->range_array[i].finfo.start_page =
1051 page_to_pfn(pg);
1052 bl_resp->range_array[i].finfo.page_cnt = alloc_unit;
1053 bl_resp->hdr.size += sizeof(union dm_mem_page_range);
1054
1055 }
1056
1057 return num_pages;
1058}
1059
1060
1061
K. Y. Srinivasan6571b2d2013-03-15 12:25:40 -07001062static void balloon_up(struct work_struct *dummy)
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001063{
K. Y. Srinivasan6571b2d2013-03-15 12:25:40 -07001064 int num_pages = dm_device.balloon_wrk.num_pages;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001065 int num_ballooned = 0;
1066 struct dm_balloon_response *bl_resp;
1067 int alloc_unit;
1068 int ret;
1069 bool alloc_error = false;
1070 bool done = false;
1071 int i;
1072
1073
1074 /*
K. Y. Srinivasanf766dc12013-03-18 13:51:37 -07001075 * We will attempt 2M allocations. However, if we fail to
1076 * allocate 2M chunks, we will go back to 4k allocations.
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001077 */
K. Y. Srinivasanf766dc12013-03-18 13:51:37 -07001078 alloc_unit = 512;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001079
1080 while (!done) {
1081 bl_resp = (struct dm_balloon_response *)send_buffer;
1082 memset(send_buffer, 0, PAGE_SIZE);
1083 bl_resp->hdr.type = DM_BALLOON_RESPONSE;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001084 bl_resp->hdr.size = sizeof(struct dm_balloon_response);
1085 bl_resp->more_pages = 1;
1086
1087
1088 num_pages -= num_ballooned;
K. Y. Srinivasan6571b2d2013-03-15 12:25:40 -07001089 num_ballooned = alloc_balloon_pages(&dm_device, num_pages,
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001090 bl_resp, alloc_unit,
1091 &alloc_error);
1092
K. Y. Srinivasanf766dc12013-03-18 13:51:37 -07001093 if ((alloc_error) && (alloc_unit != 1)) {
1094 alloc_unit = 1;
1095 continue;
1096 }
1097
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001098 if ((alloc_error) || (num_ballooned == num_pages)) {
1099 bl_resp->more_pages = 0;
1100 done = true;
K. Y. Srinivasan6571b2d2013-03-15 12:25:40 -07001101 dm_device.state = DM_INITIALIZED;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001102 }
1103
1104 /*
1105 * We are pushing a lot of data through the channel;
1106 * deal with transient failures caused because of the
1107 * lack of space in the ring buffer.
1108 */
1109
1110 do {
K. Y. Srinivasan20138d6c2013-07-17 17:27:27 -07001111 bl_resp->hdr.trans_id = atomic_inc_return(&trans_id);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001112 ret = vmbus_sendpacket(dm_device.dev->channel,
1113 bl_resp,
1114 bl_resp->hdr.size,
1115 (unsigned long)NULL,
1116 VM_PKT_DATA_INBAND, 0);
1117
1118 if (ret == -EAGAIN)
1119 msleep(20);
1120
1121 } while (ret == -EAGAIN);
1122
1123 if (ret) {
1124 /*
1125 * Free up the memory we allocatted.
1126 */
1127 pr_info("Balloon response failed\n");
1128
1129 for (i = 0; i < bl_resp->range_count; i++)
K. Y. Srinivasan6571b2d2013-03-15 12:25:40 -07001130 free_balloon_pages(&dm_device,
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001131 &bl_resp->range_array[i]);
1132
1133 done = true;
1134 }
1135 }
1136
1137}
1138
1139static void balloon_down(struct hv_dynmem_device *dm,
1140 struct dm_unballoon_request *req)
1141{
1142 union dm_mem_page_range *range_array = req->range_array;
1143 int range_count = req->range_count;
1144 struct dm_unballoon_response resp;
1145 int i;
1146
1147 for (i = 0; i < range_count; i++)
1148 free_balloon_pages(dm, &range_array[i]);
1149
1150 if (req->more_pages == 1)
1151 return;
1152
1153 memset(&resp, 0, sizeof(struct dm_unballoon_response));
1154 resp.hdr.type = DM_UNBALLOON_RESPONSE;
1155 resp.hdr.trans_id = atomic_inc_return(&trans_id);
1156 resp.hdr.size = sizeof(struct dm_unballoon_response);
1157
1158 vmbus_sendpacket(dm_device.dev->channel, &resp,
1159 sizeof(struct dm_unballoon_response),
1160 (unsigned long)NULL,
1161 VM_PKT_DATA_INBAND, 0);
1162
1163 dm->state = DM_INITIALIZED;
1164}
1165
1166static void balloon_onchannelcallback(void *context);
1167
1168static int dm_thread_func(void *dm_dev)
1169{
1170 struct hv_dynmem_device *dm = dm_dev;
1171 int t;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001172
1173 while (!kthread_should_stop()) {
1174 t = wait_for_completion_timeout(&dm_device.config_event, 1*HZ);
1175 /*
1176 * The host expects us to post information on the memory
1177 * pressure every second.
1178 */
1179
1180 if (t == 0)
1181 post_status(dm);
1182
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001183 }
1184
1185 return 0;
1186}
1187
1188
1189static void version_resp(struct hv_dynmem_device *dm,
1190 struct dm_version_response *vresp)
1191{
1192 struct dm_version_request version_req;
1193 int ret;
1194
1195 if (vresp->is_accepted) {
1196 /*
1197 * We are done; wakeup the
1198 * context waiting for version
1199 * negotiation.
1200 */
1201 complete(&dm->host_event);
1202 return;
1203 }
1204 /*
1205 * If there are more versions to try, continue
1206 * with negotiations; if not
1207 * shutdown the service since we are not able
1208 * to negotiate a suitable version number
1209 * with the host.
1210 */
1211 if (dm->next_version == 0)
1212 goto version_error;
1213
1214 dm->next_version = 0;
1215 memset(&version_req, 0, sizeof(struct dm_version_request));
1216 version_req.hdr.type = DM_VERSION_REQUEST;
1217 version_req.hdr.size = sizeof(struct dm_version_request);
1218 version_req.hdr.trans_id = atomic_inc_return(&trans_id);
1219 version_req.version.version = DYNMEM_PROTOCOL_VERSION_WIN7;
1220 version_req.is_last_attempt = 1;
1221
1222 ret = vmbus_sendpacket(dm->dev->channel, &version_req,
1223 sizeof(struct dm_version_request),
1224 (unsigned long)NULL,
1225 VM_PKT_DATA_INBAND, 0);
1226
1227 if (ret)
1228 goto version_error;
1229
1230 return;
1231
1232version_error:
1233 dm->state = DM_INIT_ERROR;
1234 complete(&dm->host_event);
1235}
1236
1237static void cap_resp(struct hv_dynmem_device *dm,
1238 struct dm_capabilities_resp_msg *cap_resp)
1239{
1240 if (!cap_resp->is_accepted) {
1241 pr_info("Capabilities not accepted by host\n");
1242 dm->state = DM_INIT_ERROR;
1243 }
1244 complete(&dm->host_event);
1245}
1246
1247static void balloon_onchannelcallback(void *context)
1248{
1249 struct hv_device *dev = context;
1250 u32 recvlen;
1251 u64 requestid;
1252 struct dm_message *dm_msg;
1253 struct dm_header *dm_hdr;
1254 struct hv_dynmem_device *dm = hv_get_drvdata(dev);
K. Y. Srinivasan6571b2d2013-03-15 12:25:40 -07001255 struct dm_balloon *bal_msg;
K. Y. Srinivasanc51af822013-03-15 12:25:41 -07001256 struct dm_hot_add *ha_msg;
1257 union dm_mem_page_range *ha_pg_range;
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -07001258 union dm_mem_page_range *ha_region;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001259
1260 memset(recv_buffer, 0, sizeof(recv_buffer));
1261 vmbus_recvpacket(dev->channel, recv_buffer,
1262 PAGE_SIZE, &recvlen, &requestid);
1263
1264 if (recvlen > 0) {
1265 dm_msg = (struct dm_message *)recv_buffer;
1266 dm_hdr = &dm_msg->hdr;
1267
1268 switch (dm_hdr->type) {
1269 case DM_VERSION_RESPONSE:
1270 version_resp(dm,
1271 (struct dm_version_response *)dm_msg);
1272 break;
1273
1274 case DM_CAPABILITIES_RESPONSE:
1275 cap_resp(dm,
1276 (struct dm_capabilities_resp_msg *)dm_msg);
1277 break;
1278
1279 case DM_BALLOON_REQUEST:
K. Y. Srinivasan6571b2d2013-03-15 12:25:40 -07001280 if (dm->state == DM_BALLOON_UP)
1281 pr_warn("Currently ballooning\n");
1282 bal_msg = (struct dm_balloon *)recv_buffer;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001283 dm->state = DM_BALLOON_UP;
K. Y. Srinivasan6571b2d2013-03-15 12:25:40 -07001284 dm_device.balloon_wrk.num_pages = bal_msg->num_pages;
1285 schedule_work(&dm_device.balloon_wrk.wrk);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001286 break;
1287
1288 case DM_UNBALLOON_REQUEST:
1289 dm->state = DM_BALLOON_DOWN;
1290 balloon_down(dm,
1291 (struct dm_unballoon_request *)recv_buffer);
1292 break;
1293
1294 case DM_MEM_HOT_ADD_REQUEST:
K. Y. Srinivasanc51af822013-03-15 12:25:41 -07001295 if (dm->state == DM_HOT_ADD)
1296 pr_warn("Currently hot-adding\n");
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001297 dm->state = DM_HOT_ADD;
K. Y. Srinivasanc51af822013-03-15 12:25:41 -07001298 ha_msg = (struct dm_hot_add *)recv_buffer;
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -07001299 if (ha_msg->hdr.size == sizeof(struct dm_hot_add)) {
1300 /*
1301 * This is a normal hot-add request specifying
1302 * hot-add memory.
1303 */
1304 ha_pg_range = &ha_msg->range;
1305 dm->ha_wrk.ha_page_range = *ha_pg_range;
1306 dm->ha_wrk.ha_region_range.page_range = 0;
1307 } else {
1308 /*
1309 * Host is specifying that we first hot-add
1310 * a region and then partially populate this
1311 * region.
1312 */
1313 dm->host_specified_ha_region = true;
1314 ha_pg_range = &ha_msg->range;
1315 ha_region = &ha_pg_range[1];
1316 dm->ha_wrk.ha_page_range = *ha_pg_range;
1317 dm->ha_wrk.ha_region_range = *ha_region;
1318 }
K. Y. Srinivasanc51af822013-03-15 12:25:41 -07001319 schedule_work(&dm_device.ha_wrk.wrk);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001320 break;
1321
1322 case DM_INFO_MESSAGE:
1323 process_info(dm, (struct dm_info_msg *)dm_msg);
1324 break;
1325
1326 default:
1327 pr_err("Unhandled message: type: %d\n", dm_hdr->type);
1328
1329 }
1330 }
1331
1332}
1333
1334static int balloon_probe(struct hv_device *dev,
1335 const struct hv_vmbus_device_id *dev_id)
1336{
1337 int ret, t;
1338 struct dm_version_request version_req;
1339 struct dm_capabilities cap_msg;
1340
1341 do_hot_add = hot_add;
1342
1343 /*
1344 * First allocate a send buffer.
1345 */
1346
1347 send_buffer = kmalloc(PAGE_SIZE, GFP_KERNEL);
1348 if (!send_buffer)
1349 return -ENOMEM;
1350
1351 ret = vmbus_open(dev->channel, dm_ring_size, dm_ring_size, NULL, 0,
1352 balloon_onchannelcallback, dev);
1353
1354 if (ret)
K. Y. Srinivasan33080c12012-12-11 11:07:17 -08001355 goto probe_error0;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001356
1357 dm_device.dev = dev;
1358 dm_device.state = DM_INITIALIZING;
1359 dm_device.next_version = DYNMEM_PROTOCOL_VERSION_WIN7;
1360 init_completion(&dm_device.host_event);
1361 init_completion(&dm_device.config_event);
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -07001362 INIT_LIST_HEAD(&dm_device.ha_region_list);
K. Y. Srinivasan6571b2d2013-03-15 12:25:40 -07001363 INIT_WORK(&dm_device.balloon_wrk.wrk, balloon_up);
K. Y. Srinivasanc51af822013-03-15 12:25:41 -07001364 INIT_WORK(&dm_device.ha_wrk.wrk, hot_add_req);
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -07001365 dm_device.host_specified_ha_region = false;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001366
1367 dm_device.thread =
1368 kthread_run(dm_thread_func, &dm_device, "hv_balloon");
1369 if (IS_ERR(dm_device.thread)) {
1370 ret = PTR_ERR(dm_device.thread);
K. Y. Srinivasan33080c12012-12-11 11:07:17 -08001371 goto probe_error1;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001372 }
1373
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -07001374#ifdef CONFIG_MEMORY_HOTPLUG
1375 set_online_page_callback(&hv_online_page);
1376#endif
1377
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001378 hv_set_drvdata(dev, &dm_device);
1379 /*
1380 * Initiate the hand shake with the host and negotiate
1381 * a version that the host can support. We start with the
1382 * highest version number and go down if the host cannot
1383 * support it.
1384 */
1385 memset(&version_req, 0, sizeof(struct dm_version_request));
1386 version_req.hdr.type = DM_VERSION_REQUEST;
1387 version_req.hdr.size = sizeof(struct dm_version_request);
1388 version_req.hdr.trans_id = atomic_inc_return(&trans_id);
1389 version_req.version.version = DYNMEM_PROTOCOL_VERSION_WIN8;
1390 version_req.is_last_attempt = 0;
1391
1392 ret = vmbus_sendpacket(dev->channel, &version_req,
1393 sizeof(struct dm_version_request),
1394 (unsigned long)NULL,
K. Y. Srinivasan7a64b862013-03-15 12:25:39 -07001395 VM_PKT_DATA_INBAND, 0);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001396 if (ret)
K. Y. Srinivasan33080c12012-12-11 11:07:17 -08001397 goto probe_error2;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001398
1399 t = wait_for_completion_timeout(&dm_device.host_event, 5*HZ);
1400 if (t == 0) {
1401 ret = -ETIMEDOUT;
K. Y. Srinivasan33080c12012-12-11 11:07:17 -08001402 goto probe_error2;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001403 }
1404
1405 /*
1406 * If we could not negotiate a compatible version with the host
1407 * fail the probe function.
1408 */
1409 if (dm_device.state == DM_INIT_ERROR) {
1410 ret = -ETIMEDOUT;
K. Y. Srinivasan33080c12012-12-11 11:07:17 -08001411 goto probe_error2;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001412 }
1413 /*
1414 * Now submit our capabilities to the host.
1415 */
1416 memset(&cap_msg, 0, sizeof(struct dm_capabilities));
1417 cap_msg.hdr.type = DM_CAPABILITIES_REPORT;
1418 cap_msg.hdr.size = sizeof(struct dm_capabilities);
1419 cap_msg.hdr.trans_id = atomic_inc_return(&trans_id);
1420
1421 cap_msg.caps.cap_bits.balloon = 1;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001422 cap_msg.caps.cap_bits.hot_add = 1;
1423
1424 /*
K. Y. Srinivasan647965a2013-03-29 07:36:11 -07001425 * Specify our alignment requirements as it relates
1426 * memory hot-add. Specify 128MB alignment.
1427 */
1428 cap_msg.caps.cap_bits.hot_add_alignment = 7;
1429
1430 /*
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001431 * Currently the host does not use these
1432 * values and we set them to what is done in the
1433 * Windows driver.
1434 */
1435 cap_msg.min_page_cnt = 0;
1436 cap_msg.max_page_number = -1;
1437
1438 ret = vmbus_sendpacket(dev->channel, &cap_msg,
1439 sizeof(struct dm_capabilities),
1440 (unsigned long)NULL,
K. Y. Srinivasan7a64b862013-03-15 12:25:39 -07001441 VM_PKT_DATA_INBAND, 0);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001442 if (ret)
K. Y. Srinivasan33080c12012-12-11 11:07:17 -08001443 goto probe_error2;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001444
1445 t = wait_for_completion_timeout(&dm_device.host_event, 5*HZ);
1446 if (t == 0) {
1447 ret = -ETIMEDOUT;
K. Y. Srinivasan33080c12012-12-11 11:07:17 -08001448 goto probe_error2;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001449 }
1450
1451 /*
1452 * If the host does not like our capabilities,
1453 * fail the probe function.
1454 */
1455 if (dm_device.state == DM_INIT_ERROR) {
1456 ret = -ETIMEDOUT;
K. Y. Srinivasan33080c12012-12-11 11:07:17 -08001457 goto probe_error2;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001458 }
1459
1460 dm_device.state = DM_INITIALIZED;
1461
1462 return 0;
1463
K. Y. Srinivasan33080c12012-12-11 11:07:17 -08001464probe_error2:
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -07001465#ifdef CONFIG_MEMORY_HOTPLUG
1466 restore_online_page_callback(&hv_online_page);
1467#endif
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001468 kthread_stop(dm_device.thread);
1469
K. Y. Srinivasan33080c12012-12-11 11:07:17 -08001470probe_error1:
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001471 vmbus_close(dev->channel);
K. Y. Srinivasan33080c12012-12-11 11:07:17 -08001472probe_error0:
1473 kfree(send_buffer);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001474 return ret;
1475}
1476
1477static int balloon_remove(struct hv_device *dev)
1478{
1479 struct hv_dynmem_device *dm = hv_get_drvdata(dev);
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -07001480 struct list_head *cur, *tmp;
1481 struct hv_hotadd_state *has;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001482
1483 if (dm->num_pages_ballooned != 0)
1484 pr_warn("Ballooned pages: %d\n", dm->num_pages_ballooned);
1485
K. Y. Srinivasan6571b2d2013-03-15 12:25:40 -07001486 cancel_work_sync(&dm->balloon_wrk.wrk);
K. Y. Srinivasanc51af822013-03-15 12:25:41 -07001487 cancel_work_sync(&dm->ha_wrk.wrk);
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -07001488
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001489 vmbus_close(dev->channel);
1490 kthread_stop(dm->thread);
K. Y. Srinivasan33080c12012-12-11 11:07:17 -08001491 kfree(send_buffer);
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -07001492#ifdef CONFIG_MEMORY_HOTPLUG
1493 restore_online_page_callback(&hv_online_page);
1494#endif
1495 list_for_each_safe(cur, tmp, &dm->ha_region_list) {
1496 has = list_entry(cur, struct hv_hotadd_state, list);
1497 list_del(&has->list);
1498 kfree(has);
1499 }
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001500
1501 return 0;
1502}
1503
1504static const struct hv_vmbus_device_id id_table[] = {
1505 /* Dynamic Memory Class ID */
1506 /* 525074DC-8985-46e2-8057-A307DC18A502 */
K. Y. Srinivasand13984e2013-01-23 17:42:41 -08001507 { HV_DM_GUID, },
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001508 { },
1509};
1510
1511MODULE_DEVICE_TABLE(vmbus, id_table);
1512
1513static struct hv_driver balloon_drv = {
1514 .name = "hv_balloon",
1515 .id_table = id_table,
1516 .probe = balloon_probe,
1517 .remove = balloon_remove,
1518};
1519
1520static int __init init_balloon_drv(void)
1521{
1522
1523 return vmbus_driver_register(&balloon_drv);
1524}
1525
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001526module_init(init_balloon_drv);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001527
1528MODULE_DESCRIPTION("Hyper-V Balloon");
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001529MODULE_LICENSE("GPL");