blob: fdf8da929cbea6f3962f19b1ccc6110de22b1adc [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>
K. Y. Srinivasanae339332014-04-23 13:53:39 -070022#include <linux/jiffies.h>
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -080023#include <linux/mman.h>
24#include <linux/delay.h>
25#include <linux/init.h>
26#include <linux/module.h>
27#include <linux/slab.h>
28#include <linux/kthread.h>
29#include <linux/completion.h>
30#include <linux/memory_hotplug.h>
31#include <linux/memory.h>
32#include <linux/notifier.h>
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -080033#include <linux/percpu_counter.h>
34
35#include <linux/hyperv.h>
36
37/*
38 * We begin with definitions supporting the Dynamic Memory protocol
39 * with the host.
40 *
41 * Begin protocol definitions.
42 */
43
44
45
46/*
47 * Protocol versions. The low word is the minor version, the high word the major
48 * version.
49 *
50 * History:
51 * Initial version 1.0
52 * Changed to 0.1 on 2009/03/25
53 * Changes to 0.2 on 2009/05/14
54 * Changes to 0.3 on 2009/12/03
55 * Changed to 1.0 on 2011/04/05
56 */
57
58#define DYNMEM_MAKE_VERSION(Major, Minor) ((__u32)(((Major) << 16) | (Minor)))
59#define DYNMEM_MAJOR_VERSION(Version) ((__u32)(Version) >> 16)
60#define DYNMEM_MINOR_VERSION(Version) ((__u32)(Version) & 0xff)
61
62enum {
63 DYNMEM_PROTOCOL_VERSION_1 = DYNMEM_MAKE_VERSION(0, 3),
64 DYNMEM_PROTOCOL_VERSION_2 = DYNMEM_MAKE_VERSION(1, 0),
Alex Ngb6ddeae2015-08-01 16:08:13 -070065 DYNMEM_PROTOCOL_VERSION_3 = DYNMEM_MAKE_VERSION(2, 0),
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -080066
67 DYNMEM_PROTOCOL_VERSION_WIN7 = DYNMEM_PROTOCOL_VERSION_1,
68 DYNMEM_PROTOCOL_VERSION_WIN8 = DYNMEM_PROTOCOL_VERSION_2,
Alex Ngb6ddeae2015-08-01 16:08:13 -070069 DYNMEM_PROTOCOL_VERSION_WIN10 = DYNMEM_PROTOCOL_VERSION_3,
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -080070
Alex Ngb6ddeae2015-08-01 16:08:13 -070071 DYNMEM_PROTOCOL_VERSION_CURRENT = DYNMEM_PROTOCOL_VERSION_WIN10
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -080072};
73
74
75
76/*
77 * Message Types
78 */
79
80enum dm_message_type {
81 /*
82 * Version 0.3
83 */
84 DM_ERROR = 0,
85 DM_VERSION_REQUEST = 1,
86 DM_VERSION_RESPONSE = 2,
87 DM_CAPABILITIES_REPORT = 3,
88 DM_CAPABILITIES_RESPONSE = 4,
89 DM_STATUS_REPORT = 5,
90 DM_BALLOON_REQUEST = 6,
91 DM_BALLOON_RESPONSE = 7,
92 DM_UNBALLOON_REQUEST = 8,
93 DM_UNBALLOON_RESPONSE = 9,
94 DM_MEM_HOT_ADD_REQUEST = 10,
95 DM_MEM_HOT_ADD_RESPONSE = 11,
96 DM_VERSION_03_MAX = 11,
97 /*
98 * Version 1.0.
99 */
100 DM_INFO_MESSAGE = 12,
101 DM_VERSION_1_MAX = 12
102};
103
104
105/*
106 * Structures defining the dynamic memory management
107 * protocol.
108 */
109
110union dm_version {
111 struct {
112 __u16 minor_version;
113 __u16 major_version;
114 };
115 __u32 version;
116} __packed;
117
118
119union dm_caps {
120 struct {
121 __u64 balloon:1;
122 __u64 hot_add:1;
K. Y. Srinivasan647965a2013-03-29 07:36:11 -0700123 /*
124 * To support guests that may have alignment
125 * limitations on hot-add, the guest can specify
126 * its alignment requirements; a value of n
127 * represents an alignment of 2^n in mega bytes.
128 */
129 __u64 hot_add_alignment:4;
130 __u64 reservedz:58;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800131 } cap_bits;
132 __u64 caps;
133} __packed;
134
135union dm_mem_page_range {
136 struct {
137 /*
138 * The PFN number of the first page in the range.
139 * 40 bits is the architectural limit of a PFN
140 * number for AMD64.
141 */
142 __u64 start_page:40;
143 /*
144 * The number of pages in the range.
145 */
146 __u64 page_cnt:24;
147 } finfo;
148 __u64 page_range;
149} __packed;
150
151
152
153/*
154 * The header for all dynamic memory messages:
155 *
156 * type: Type of the message.
157 * size: Size of the message in bytes; including the header.
158 * trans_id: The guest is responsible for manufacturing this ID.
159 */
160
161struct dm_header {
162 __u16 type;
163 __u16 size;
164 __u32 trans_id;
165} __packed;
166
167/*
168 * A generic message format for dynamic memory.
169 * Specific message formats are defined later in the file.
170 */
171
172struct dm_message {
173 struct dm_header hdr;
174 __u8 data[]; /* enclosed message */
175} __packed;
176
177
178/*
179 * Specific message types supporting the dynamic memory protocol.
180 */
181
182/*
183 * Version negotiation message. Sent from the guest to the host.
184 * The guest is free to try different versions until the host
185 * accepts the version.
186 *
187 * dm_version: The protocol version requested.
188 * is_last_attempt: If TRUE, this is the last version guest will request.
189 * reservedz: Reserved field, set to zero.
190 */
191
192struct dm_version_request {
193 struct dm_header hdr;
194 union dm_version version;
195 __u32 is_last_attempt:1;
196 __u32 reservedz:31;
197} __packed;
198
199/*
200 * Version response message; Host to Guest and indicates
201 * if the host has accepted the version sent by the guest.
202 *
203 * is_accepted: If TRUE, host has accepted the version and the guest
204 * should proceed to the next stage of the protocol. FALSE indicates that
205 * guest should re-try with a different version.
206 *
207 * reservedz: Reserved field, set to zero.
208 */
209
210struct dm_version_response {
211 struct dm_header hdr;
212 __u64 is_accepted:1;
213 __u64 reservedz:63;
214} __packed;
215
216/*
217 * Message reporting capabilities. This is sent from the guest to the
218 * host.
219 */
220
221struct dm_capabilities {
222 struct dm_header hdr;
223 union dm_caps caps;
224 __u64 min_page_cnt;
225 __u64 max_page_number;
226} __packed;
227
228/*
229 * Response to the capabilities message. This is sent from the host to the
230 * guest. This message notifies if the host has accepted the guest's
231 * capabilities. If the host has not accepted, the guest must shutdown
232 * the service.
233 *
234 * is_accepted: Indicates if the host has accepted guest's capabilities.
235 * reservedz: Must be 0.
236 */
237
238struct dm_capabilities_resp_msg {
239 struct dm_header hdr;
240 __u64 is_accepted:1;
241 __u64 reservedz:63;
242} __packed;
243
244/*
245 * This message is used to report memory pressure from the guest.
246 * This message is not part of any transaction and there is no
247 * response to this message.
248 *
249 * num_avail: Available memory in pages.
250 * num_committed: Committed memory in pages.
251 * page_file_size: The accumulated size of all page files
252 * in the system in pages.
253 * zero_free: The nunber of zero and free pages.
254 * page_file_writes: The writes to the page file in pages.
255 * io_diff: An indicator of file cache efficiency or page file activity,
256 * calculated as File Cache Page Fault Count - Page Read Count.
257 * This value is in pages.
258 *
259 * Some of these metrics are Windows specific and fortunately
260 * the algorithm on the host side that computes the guest memory
261 * pressure only uses num_committed value.
262 */
263
264struct dm_status {
265 struct dm_header hdr;
266 __u64 num_avail;
267 __u64 num_committed;
268 __u64 page_file_size;
269 __u64 zero_free;
270 __u32 page_file_writes;
271 __u32 io_diff;
272} __packed;
273
274
275/*
276 * Message to ask the guest to allocate memory - balloon up message.
277 * This message is sent from the host to the guest. The guest may not be
278 * able to allocate as much memory as requested.
279 *
280 * num_pages: number of pages to allocate.
281 */
282
283struct dm_balloon {
284 struct dm_header hdr;
285 __u32 num_pages;
286 __u32 reservedz;
287} __packed;
288
289
290/*
291 * Balloon response message; this message is sent from the guest
292 * to the host in response to the balloon message.
293 *
294 * reservedz: Reserved; must be set to zero.
295 * more_pages: If FALSE, this is the last message of the transaction.
296 * if TRUE there will atleast one more message from the guest.
297 *
298 * range_count: The number of ranges in the range array.
299 *
300 * range_array: An array of page ranges returned to the host.
301 *
302 */
303
304struct dm_balloon_response {
305 struct dm_header hdr;
306 __u32 reservedz;
307 __u32 more_pages:1;
308 __u32 range_count:31;
309 union dm_mem_page_range range_array[];
310} __packed;
311
312/*
313 * Un-balloon message; this message is sent from the host
314 * to the guest to give guest more memory.
315 *
316 * more_pages: If FALSE, this is the last message of the transaction.
317 * if TRUE there will atleast one more message from the guest.
318 *
319 * reservedz: Reserved; must be set to zero.
320 *
321 * range_count: The number of ranges in the range array.
322 *
323 * range_array: An array of page ranges returned to the host.
324 *
325 */
326
327struct dm_unballoon_request {
328 struct dm_header hdr;
329 __u32 more_pages:1;
330 __u32 reservedz:31;
331 __u32 range_count;
332 union dm_mem_page_range range_array[];
333} __packed;
334
335/*
336 * Un-balloon response message; this message is sent from the guest
337 * to the host in response to an unballoon request.
338 *
339 */
340
341struct dm_unballoon_response {
342 struct dm_header hdr;
343} __packed;
344
345
346/*
347 * Hot add request message. Message sent from the host to the guest.
348 *
349 * mem_range: Memory range to hot add.
350 *
351 * On Linux we currently don't support this since we cannot hot add
352 * arbitrary granularity of memory.
353 */
354
355struct dm_hot_add {
356 struct dm_header hdr;
357 union dm_mem_page_range range;
358} __packed;
359
360/*
361 * Hot add response message.
362 * This message is sent by the guest to report the status of a hot add request.
363 * If page_count is less than the requested page count, then the host should
364 * assume all further hot add requests will fail, since this indicates that
365 * the guest has hit an upper physical memory barrier.
366 *
367 * Hot adds may also fail due to low resources; in this case, the guest must
368 * not complete this message until the hot add can succeed, and the host must
369 * not send a new hot add request until the response is sent.
370 * If VSC fails to hot add memory DYNMEM_NUMBER_OF_UNSUCCESSFUL_HOTADD_ATTEMPTS
371 * times it fails the request.
372 *
373 *
374 * page_count: number of pages that were successfully hot added.
375 *
376 * result: result of the operation 1: success, 0: failure.
377 *
378 */
379
380struct dm_hot_add_response {
381 struct dm_header hdr;
382 __u32 page_count;
383 __u32 result;
384} __packed;
385
386/*
387 * Types of information sent from host to the guest.
388 */
389
390enum dm_info_type {
391 INFO_TYPE_MAX_PAGE_CNT = 0,
392 MAX_INFO_TYPE
393};
394
395
396/*
397 * Header for the information message.
398 */
399
400struct dm_info_header {
401 enum dm_info_type type;
402 __u32 data_size;
403} __packed;
404
405/*
406 * This message is sent from the host to the guest to pass
407 * some relevant information (win8 addition).
408 *
409 * reserved: no used.
410 * info_size: size of the information blob.
411 * info: information blob.
412 */
413
414struct dm_info_msg {
K. Y. Srinivasan6427a0d2012-12-06 11:06:54 -0800415 struct dm_header hdr;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800416 __u32 reserved;
417 __u32 info_size;
418 __u8 info[];
419};
420
421/*
422 * End protocol definitions.
423 */
424
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700425/*
426 * State to manage hot adding memory into the guest.
427 * The range start_pfn : end_pfn specifies the range
428 * that the host has asked us to hot add. The range
429 * start_pfn : ha_end_pfn specifies the range that we have
430 * currently hot added. We hot add in multiples of 128M
431 * chunks; it is possible that we may not be able to bring
432 * online all the pages in the region. The range
Vitaly Kuznetsov7cf3b792016-08-24 16:23:09 -0700433 * covered_start_pfn:covered_end_pfn defines the pages that can
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700434 * be brough online.
435 */
436
437struct hv_hotadd_state {
438 struct list_head list;
439 unsigned long start_pfn;
Vitaly Kuznetsov7cf3b792016-08-24 16:23:09 -0700440 unsigned long covered_start_pfn;
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700441 unsigned long covered_end_pfn;
442 unsigned long ha_end_pfn;
443 unsigned long end_pfn;
Vitaly Kuznetsovcb7a5722016-08-24 16:23:10 -0700444 /*
445 * A list of gaps.
446 */
447 struct list_head gap_list;
448};
449
450struct hv_hotadd_gap {
451 struct list_head list;
452 unsigned long start_pfn;
453 unsigned long end_pfn;
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700454};
455
K. Y. Srinivasan6571b2d2013-03-15 12:25:40 -0700456struct balloon_state {
457 __u32 num_pages;
458 struct work_struct wrk;
459};
460
K. Y. Srinivasanc51af822013-03-15 12:25:41 -0700461struct hot_add_wrk {
462 union dm_mem_page_range ha_page_range;
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700463 union dm_mem_page_range ha_region_range;
K. Y. Srinivasanc51af822013-03-15 12:25:41 -0700464 struct work_struct wrk;
465};
466
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700467static bool hot_add = true;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800468static bool do_hot_add;
K. Y. Srinivasane500d152013-02-08 15:57:15 -0800469/*
470 * Delay reporting memory pressure by
471 * the specified number of seconds.
472 */
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700473static uint pressure_report_delay = 45;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800474
K. Y. Srinivasanae339332014-04-23 13:53:39 -0700475/*
476 * The last time we posted a pressure report to host.
477 */
478static unsigned long last_post_time;
479
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800480module_param(hot_add, bool, (S_IRUGO | S_IWUSR));
481MODULE_PARM_DESC(hot_add, "If set attempt memory hot_add");
482
K. Y. Srinivasane500d152013-02-08 15:57:15 -0800483module_param(pressure_report_delay, uint, (S_IRUGO | S_IWUSR));
484MODULE_PARM_DESC(pressure_report_delay, "Delay in secs in reporting pressure");
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800485static atomic_t trans_id = ATOMIC_INIT(0);
486
487static int dm_ring_size = (5 * PAGE_SIZE);
488
489/*
490 * Driver specific state.
491 */
492
493enum hv_dm_state {
494 DM_INITIALIZING = 0,
495 DM_INITIALIZED,
496 DM_BALLOON_UP,
497 DM_BALLOON_DOWN,
498 DM_HOT_ADD,
499 DM_INIT_ERROR
500};
501
502
503static __u8 recv_buffer[PAGE_SIZE];
504static __u8 *send_buffer;
505#define PAGES_IN_2M 512
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700506#define HA_CHUNK (32 * 1024)
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800507
508struct hv_dynmem_device {
509 struct hv_device *dev;
510 enum hv_dm_state state;
511 struct completion host_event;
512 struct completion config_event;
513
514 /*
515 * Number of pages we have currently ballooned out.
516 */
517 unsigned int num_pages_ballooned;
Vitaly Kuznetsov549fd282015-02-28 11:38:59 -0800518 unsigned int num_pages_onlined;
519 unsigned int num_pages_added;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800520
521 /*
K. Y. Srinivasan6571b2d2013-03-15 12:25:40 -0700522 * State to manage the ballooning (up) operation.
523 */
524 struct balloon_state balloon_wrk;
525
526 /*
K. Y. Srinivasanc51af822013-03-15 12:25:41 -0700527 * State to execute the "hot-add" operation.
528 */
529 struct hot_add_wrk ha_wrk;
530
531 /*
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700532 * This state tracks if the host has specified a hot-add
533 * region.
534 */
535 bool host_specified_ha_region;
536
537 /*
538 * State to synchronize hot-add.
539 */
540 struct completion ol_waitevent;
541 bool ha_waiting;
542 /*
K. Y. Srinivasan6571b2d2013-03-15 12:25:40 -0700543 * This thread handles hot-add
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800544 * requests from the host as well as notifying
545 * the host with regards to memory pressure in
546 * the guest.
547 */
548 struct task_struct *thread;
549
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -0700550 /*
551 * Protects ha_region_list, num_pages_onlined counter and individual
552 * regions from ha_region_list.
553 */
554 spinlock_t ha_lock;
K. Y. Srinivasan22f88472015-01-09 23:54:30 -0800555
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800556 /*
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700557 * A list of hot-add regions.
558 */
559 struct list_head ha_region_list;
560
561 /*
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800562 * We start with the highest version we can support
563 * and downgrade based on the host; we save here the
564 * next version to try.
565 */
566 __u32 next_version;
567};
568
569static struct hv_dynmem_device dm_device;
570
K. Y. Srinivasanae339332014-04-23 13:53:39 -0700571static void post_status(struct hv_dynmem_device *dm);
K. Y. Srinivasan22f88472015-01-09 23:54:30 -0800572
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700573#ifdef CONFIG_MEMORY_HOTPLUG
K. Y. Srinivasan22f88472015-01-09 23:54:30 -0800574static int hv_memory_notifier(struct notifier_block *nb, unsigned long val,
575 void *v)
576{
Vitaly Kuznetsov549fd282015-02-28 11:38:59 -0800577 struct memory_notify *mem = (struct memory_notify *)v;
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -0700578 unsigned long flags;
Vitaly Kuznetsov549fd282015-02-28 11:38:59 -0800579
K. Y. Srinivasan22f88472015-01-09 23:54:30 -0800580 switch (val) {
K. Y. Srinivasan22f88472015-01-09 23:54:30 -0800581 case MEM_ONLINE:
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -0700582 spin_lock_irqsave(&dm_device.ha_lock, flags);
Vitaly Kuznetsov549fd282015-02-28 11:38:59 -0800583 dm_device.num_pages_onlined += mem->nr_pages;
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -0700584 spin_unlock_irqrestore(&dm_device.ha_lock, flags);
K. Y. Srinivasan22f88472015-01-09 23:54:30 -0800585 case MEM_CANCEL_ONLINE:
K. Y. Srinivasan22f88472015-01-09 23:54:30 -0800586 if (dm_device.ha_waiting) {
587 dm_device.ha_waiting = false;
588 complete(&dm_device.ol_waitevent);
589 }
590 break;
591
K. Y. Srinivasan22f88472015-01-09 23:54:30 -0800592 case MEM_OFFLINE:
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -0700593 spin_lock_irqsave(&dm_device.ha_lock, flags);
Vitaly Kuznetsov549fd282015-02-28 11:38:59 -0800594 dm_device.num_pages_onlined -= mem->nr_pages;
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -0700595 spin_unlock_irqrestore(&dm_device.ha_lock, flags);
Vitaly Kuznetsov549fd282015-02-28 11:38:59 -0800596 break;
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -0700597 case MEM_GOING_ONLINE:
Vitaly Kuznetsov549fd282015-02-28 11:38:59 -0800598 case MEM_GOING_OFFLINE:
K. Y. Srinivasan22f88472015-01-09 23:54:30 -0800599 case MEM_CANCEL_OFFLINE:
600 break;
601 }
602 return NOTIFY_OK;
603}
604
605static struct notifier_block hv_memory_nb = {
606 .notifier_call = hv_memory_notifier,
607 .priority = 0
608};
609
Vitaly Kuznetsovcb7a5722016-08-24 16:23:10 -0700610/* Check if the particular page is backed and can be onlined and online it. */
611static void hv_page_online_one(struct hv_hotadd_state *has, struct page *pg)
612{
613 unsigned long cur_start_pgp;
614 unsigned long cur_end_pgp;
615 struct hv_hotadd_gap *gap;
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700616
Vitaly Kuznetsovcb7a5722016-08-24 16:23:10 -0700617 cur_start_pgp = (unsigned long)pfn_to_page(has->covered_start_pfn);
618 cur_end_pgp = (unsigned long)pfn_to_page(has->covered_end_pfn);
619
620 /* The page is not backed. */
621 if (((unsigned long)pg < cur_start_pgp) ||
622 ((unsigned long)pg >= cur_end_pgp))
623 return;
624
625 /* Check for gaps. */
626 list_for_each_entry(gap, &has->gap_list, list) {
627 cur_start_pgp = (unsigned long)
628 pfn_to_page(gap->start_pfn);
629 cur_end_pgp = (unsigned long)
630 pfn_to_page(gap->end_pfn);
631 if (((unsigned long)pg >= cur_start_pgp) &&
632 ((unsigned long)pg < cur_end_pgp)) {
633 return;
634 }
635 }
636
637 /* This frame is currently backed; online the page. */
638 __online_page_set_limits(pg);
639 __online_page_increment_counters(pg);
640 __online_page_free(pg);
641}
642
643static void hv_bring_pgs_online(struct hv_hotadd_state *has,
644 unsigned long start_pfn, unsigned long size)
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800645{
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700646 int i;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800647
Vitaly Kuznetsovcb7a5722016-08-24 16:23:10 -0700648 for (i = 0; i < size; i++)
649 hv_page_online_one(has, pfn_to_page(start_pfn + i));
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700650}
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800651
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700652static void hv_mem_hot_add(unsigned long start, unsigned long size,
653 unsigned long pfn_count,
654 struct hv_hotadd_state *has)
655{
656 int ret = 0;
K. Y. Srinivasaned07ec92013-07-14 22:38:11 -0700657 int i, nid;
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700658 unsigned long start_pfn;
659 unsigned long processed_pfn;
660 unsigned long total_pfn = pfn_count;
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -0700661 unsigned long flags;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800662
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700663 for (i = 0; i < (size/HA_CHUNK); i++) {
664 start_pfn = start + (i * HA_CHUNK);
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -0700665
666 spin_lock_irqsave(&dm_device.ha_lock, flags);
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700667 has->ha_end_pfn += HA_CHUNK;
668
669 if (total_pfn > HA_CHUNK) {
670 processed_pfn = HA_CHUNK;
671 total_pfn -= HA_CHUNK;
672 } else {
673 processed_pfn = total_pfn;
674 total_pfn = 0;
675 }
676
677 has->covered_end_pfn += processed_pfn;
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -0700678 spin_unlock_irqrestore(&dm_device.ha_lock, flags);
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700679
680 init_completion(&dm_device.ol_waitevent);
Vitaly Kuznetsova132c542016-08-24 16:23:11 -0700681 dm_device.ha_waiting = !memhp_auto_online;
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700682
683 nid = memory_add_physaddr_to_nid(PFN_PHYS(start_pfn));
684 ret = add_memory(nid, PFN_PHYS((start_pfn)),
685 (HA_CHUNK << PAGE_SHIFT));
686
687 if (ret) {
688 pr_info("hot_add memory failed error is %d\n", ret);
K. Y. Srinivasan7f4f2302013-03-18 13:51:38 -0700689 if (ret == -EEXIST) {
690 /*
691 * This error indicates that the error
692 * is not a transient failure. This is the
693 * case where the guest's physical address map
694 * precludes hot adding memory. Stop all further
695 * memory hot-add.
696 */
697 do_hot_add = false;
698 }
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -0700699 spin_lock_irqsave(&dm_device.ha_lock, flags);
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700700 has->ha_end_pfn -= HA_CHUNK;
701 has->covered_end_pfn -= processed_pfn;
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -0700702 spin_unlock_irqrestore(&dm_device.ha_lock, flags);
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700703 break;
704 }
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800705
706 /*
Vitaly Kuznetsova132c542016-08-24 16:23:11 -0700707 * Wait for the memory block to be onlined when memory onlining
708 * is done outside of kernel (memhp_auto_online). Since the hot
709 * add has succeeded, it is ok to proceed even if the pages in
710 * the hot added region have not been "onlined" within the
711 * allowed time.
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800712 */
Vitaly Kuznetsova132c542016-08-24 16:23:11 -0700713 if (dm_device.ha_waiting)
714 wait_for_completion_timeout(&dm_device.ol_waitevent,
715 5*HZ);
K. Y. Srinivasanae339332014-04-23 13:53:39 -0700716 post_status(&dm_device);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800717 }
718
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700719 return;
720}
721
722static void hv_online_page(struct page *pg)
723{
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700724 struct hv_hotadd_state *has;
725 unsigned long cur_start_pgp;
726 unsigned long cur_end_pgp;
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -0700727 unsigned long flags;
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700728
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -0700729 spin_lock_irqsave(&dm_device.ha_lock, flags);
730 list_for_each_entry(has, &dm_device.ha_region_list, list) {
Vitaly Kuznetsov7cf3b792016-08-24 16:23:09 -0700731 cur_start_pgp = (unsigned long)
Vitaly Kuznetsovcb7a5722016-08-24 16:23:10 -0700732 pfn_to_page(has->start_pfn);
733 cur_end_pgp = (unsigned long)pfn_to_page(has->end_pfn);
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700734
Vitaly Kuznetsovcb7a5722016-08-24 16:23:10 -0700735 /* The page belongs to a different HAS. */
736 if (((unsigned long)pg < cur_start_pgp) ||
737 ((unsigned long)pg >= cur_end_pgp))
738 continue;
739
740 hv_page_online_one(has, pg);
741 break;
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700742 }
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -0700743 spin_unlock_irqrestore(&dm_device.ha_lock, flags);
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700744}
745
Vitaly Kuznetsovcb7a5722016-08-24 16:23:10 -0700746static int pfn_covered(unsigned long start_pfn, unsigned long pfn_cnt)
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700747{
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700748 struct hv_hotadd_state *has;
Vitaly Kuznetsovcb7a5722016-08-24 16:23:10 -0700749 struct hv_hotadd_gap *gap;
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700750 unsigned long residual, new_inc;
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -0700751 int ret = 0;
752 unsigned long flags;
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700753
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -0700754 spin_lock_irqsave(&dm_device.ha_lock, flags);
755 list_for_each_entry(has, &dm_device.ha_region_list, list) {
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700756 /*
757 * If the pfn range we are dealing with is not in the current
758 * "hot add block", move on.
759 */
Vitaly Kuznetsov77c0c972016-04-30 19:21:35 -0700760 if (start_pfn < has->start_pfn || start_pfn >= has->end_pfn)
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700761 continue;
Vitaly Kuznetsovcb7a5722016-08-24 16:23:10 -0700762
763 /*
764 * If the current start pfn is not where the covered_end
765 * is, create a gap and update covered_end_pfn.
766 */
767 if (has->covered_end_pfn != start_pfn) {
768 gap = kzalloc(sizeof(struct hv_hotadd_gap), GFP_ATOMIC);
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -0700769 if (!gap) {
770 ret = -ENOMEM;
771 break;
772 }
Vitaly Kuznetsovcb7a5722016-08-24 16:23:10 -0700773
774 INIT_LIST_HEAD(&gap->list);
775 gap->start_pfn = has->covered_end_pfn;
776 gap->end_pfn = start_pfn;
777 list_add_tail(&gap->list, &has->gap_list);
778
779 has->covered_end_pfn = start_pfn;
780 }
781
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700782 /*
783 * If the current hot add-request extends beyond
784 * our current limit; extend it.
785 */
786 if ((start_pfn + pfn_cnt) > has->end_pfn) {
787 residual = (start_pfn + pfn_cnt - has->end_pfn);
788 /*
789 * Extend the region by multiples of HA_CHUNK.
790 */
791 new_inc = (residual / HA_CHUNK) * HA_CHUNK;
792 if (residual % HA_CHUNK)
793 new_inc += HA_CHUNK;
794
795 has->end_pfn += new_inc;
796 }
797
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -0700798 ret = 1;
799 break;
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700800 }
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -0700801 spin_unlock_irqrestore(&dm_device.ha_lock, flags);
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700802
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -0700803 return ret;
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700804}
805
806static unsigned long handle_pg_range(unsigned long pg_start,
807 unsigned long pg_count)
808{
809 unsigned long start_pfn = pg_start;
810 unsigned long pfn_cnt = pg_count;
811 unsigned long size;
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700812 struct hv_hotadd_state *has;
813 unsigned long pgs_ol = 0;
814 unsigned long old_covered_state;
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -0700815 unsigned long res = 0, flags;
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700816
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -0700817 spin_lock_irqsave(&dm_device.ha_lock, flags);
818 list_for_each_entry(has, &dm_device.ha_region_list, list) {
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700819 /*
820 * If the pfn range we are dealing with is not in the current
821 * "hot add block", move on.
822 */
Vitaly Kuznetsov77c0c972016-04-30 19:21:35 -0700823 if (start_pfn < has->start_pfn || start_pfn >= has->end_pfn)
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700824 continue;
825
826 old_covered_state = has->covered_end_pfn;
827
828 if (start_pfn < has->ha_end_pfn) {
829 /*
830 * This is the case where we are backing pages
831 * in an already hot added region. Bring
832 * these pages online first.
833 */
834 pgs_ol = has->ha_end_pfn - start_pfn;
835 if (pgs_ol > pfn_cnt)
836 pgs_ol = pfn_cnt;
Vitaly Kuznetsovd6cbd2c2015-03-27 09:10:11 -0700837
Vitaly Kuznetsovcb7a5722016-08-24 16:23:10 -0700838 has->covered_end_pfn += pgs_ol;
839 pfn_cnt -= pgs_ol;
Vitaly Kuznetsovd6cbd2c2015-03-27 09:10:11 -0700840 /*
841 * Check if the corresponding memory block is already
842 * online by checking its last previously backed page.
843 * In case it is we need to bring rest (which was not
844 * backed previously) online too.
845 */
846 if (start_pfn > has->start_pfn &&
847 !PageReserved(pfn_to_page(start_pfn - 1)))
Vitaly Kuznetsovcb7a5722016-08-24 16:23:10 -0700848 hv_bring_pgs_online(has, start_pfn, pgs_ol);
Vitaly Kuznetsovd6cbd2c2015-03-27 09:10:11 -0700849
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700850 }
851
852 if ((has->ha_end_pfn < has->end_pfn) && (pfn_cnt > 0)) {
853 /*
854 * We have some residual hot add range
855 * that needs to be hot added; hot add
856 * it now. Hot add a multiple of
857 * of HA_CHUNK that fully covers the pages
858 * we have.
859 */
860 size = (has->end_pfn - has->ha_end_pfn);
861 if (pfn_cnt <= size) {
862 size = ((pfn_cnt / HA_CHUNK) * HA_CHUNK);
863 if (pfn_cnt % HA_CHUNK)
864 size += HA_CHUNK;
865 } else {
866 pfn_cnt = size;
867 }
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -0700868 spin_unlock_irqrestore(&dm_device.ha_lock, flags);
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700869 hv_mem_hot_add(has->ha_end_pfn, size, pfn_cnt, has);
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -0700870 spin_lock_irqsave(&dm_device.ha_lock, flags);
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700871 }
872 /*
873 * If we managed to online any pages that were given to us,
874 * we declare success.
875 */
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -0700876 res = has->covered_end_pfn - old_covered_state;
877 break;
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700878 }
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -0700879 spin_unlock_irqrestore(&dm_device.ha_lock, flags);
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700880
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -0700881 return res;
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700882}
883
884static unsigned long process_hot_add(unsigned long pg_start,
885 unsigned long pfn_cnt,
886 unsigned long rg_start,
887 unsigned long rg_size)
888{
889 struct hv_hotadd_state *ha_region = NULL;
Vitaly Kuznetsovcb7a5722016-08-24 16:23:10 -0700890 int covered;
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -0700891 unsigned long flags;
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700892
893 if (pfn_cnt == 0)
894 return 0;
895
Vitaly Kuznetsovcb7a5722016-08-24 16:23:10 -0700896 if (!dm_device.host_specified_ha_region) {
897 covered = pfn_covered(pg_start, pfn_cnt);
898 if (covered < 0)
899 return 0;
900
901 if (covered)
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700902 goto do_pg_range;
Vitaly Kuznetsovcb7a5722016-08-24 16:23:10 -0700903 }
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700904
905 /*
906 * If the host has specified a hot-add range; deal with it first.
907 */
908
K. Y. Srinivasan647965a2013-03-29 07:36:11 -0700909 if (rg_size != 0) {
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700910 ha_region = kzalloc(sizeof(struct hv_hotadd_state), GFP_KERNEL);
911 if (!ha_region)
912 return 0;
913
914 INIT_LIST_HEAD(&ha_region->list);
Vitaly Kuznetsovcb7a5722016-08-24 16:23:10 -0700915 INIT_LIST_HEAD(&ha_region->gap_list);
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700916
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700917 ha_region->start_pfn = rg_start;
918 ha_region->ha_end_pfn = rg_start;
Vitaly Kuznetsov7cf3b792016-08-24 16:23:09 -0700919 ha_region->covered_start_pfn = pg_start;
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700920 ha_region->covered_end_pfn = pg_start;
921 ha_region->end_pfn = rg_start + rg_size;
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -0700922
923 spin_lock_irqsave(&dm_device.ha_lock, flags);
924 list_add_tail(&ha_region->list, &dm_device.ha_region_list);
925 spin_unlock_irqrestore(&dm_device.ha_lock, flags);
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700926 }
927
928do_pg_range:
929 /*
930 * Process the page range specified; bringing them
931 * online if possible.
932 */
933 return handle_pg_range(pg_start, pfn_cnt);
934}
935
936#endif
937
938static void hot_add_req(struct work_struct *dummy)
939{
940 struct dm_hot_add_response resp;
941#ifdef CONFIG_MEMORY_HOTPLUG
942 unsigned long pg_start, pfn_cnt;
943 unsigned long rg_start, rg_sz;
944#endif
945 struct hv_dynmem_device *dm = &dm_device;
946
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800947 memset(&resp, 0, sizeof(struct dm_hot_add_response));
948 resp.hdr.type = DM_MEM_HOT_ADD_RESPONSE;
949 resp.hdr.size = sizeof(struct dm_hot_add_response);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800950
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700951#ifdef CONFIG_MEMORY_HOTPLUG
952 pg_start = dm->ha_wrk.ha_page_range.finfo.start_page;
953 pfn_cnt = dm->ha_wrk.ha_page_range.finfo.page_cnt;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800954
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700955 rg_start = dm->ha_wrk.ha_region_range.finfo.start_page;
956 rg_sz = dm->ha_wrk.ha_region_range.finfo.page_cnt;
957
958 if ((rg_start == 0) && (!dm->host_specified_ha_region)) {
959 unsigned long region_size;
960 unsigned long region_start;
961
962 /*
963 * The host has not specified the hot-add region.
964 * Based on the hot-add page range being specified,
965 * compute a hot-add region that can cover the pages
966 * that need to be hot-added while ensuring the alignment
967 * and size requirements of Linux as it relates to hot-add.
968 */
969 region_start = pg_start;
970 region_size = (pfn_cnt / HA_CHUNK) * HA_CHUNK;
971 if (pfn_cnt % HA_CHUNK)
972 region_size += HA_CHUNK;
973
974 region_start = (pg_start / HA_CHUNK) * HA_CHUNK;
975
976 rg_start = region_start;
977 rg_sz = region_size;
978 }
979
K. Y. Srinivasan7f4f2302013-03-18 13:51:38 -0700980 if (do_hot_add)
981 resp.page_count = process_hot_add(pg_start, pfn_cnt,
982 rg_start, rg_sz);
Vitaly Kuznetsov549fd282015-02-28 11:38:59 -0800983
984 dm->num_pages_added += resp.page_count;
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -0700985#endif
K. Y. Srinivasan7f4f2302013-03-18 13:51:38 -0700986 /*
987 * The result field of the response structure has the
988 * following semantics:
989 *
990 * 1. If all or some pages hot-added: Guest should return success.
991 *
992 * 2. If no pages could be hot-added:
993 *
994 * If the guest returns success, then the host
995 * will not attempt any further hot-add operations. This
996 * signifies a permanent failure.
997 *
998 * If the guest returns failure, then this failure will be
999 * treated as a transient failure and the host may retry the
1000 * hot-add operation after some delay.
1001 */
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -07001002 if (resp.page_count > 0)
1003 resp.result = 1;
K. Y. Srinivasan7f4f2302013-03-18 13:51:38 -07001004 else if (!do_hot_add)
1005 resp.result = 1;
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -07001006 else
1007 resp.result = 0;
1008
1009 if (!do_hot_add || (resp.page_count == 0))
1010 pr_info("Memory hot add failed\n");
1011
1012 dm->state = DM_INITIALIZED;
K. Y. Srinivasan20138d6c2013-07-17 17:27:27 -07001013 resp.hdr.trans_id = atomic_inc_return(&trans_id);
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -07001014 vmbus_sendpacket(dm->dev->channel, &resp,
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001015 sizeof(struct dm_hot_add_response),
1016 (unsigned long)NULL,
1017 VM_PKT_DATA_INBAND, 0);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001018}
1019
1020static void process_info(struct hv_dynmem_device *dm, struct dm_info_msg *msg)
1021{
K. Y. Srinivasan6427a0d2012-12-06 11:06:54 -08001022 struct dm_info_header *info_hdr;
1023
1024 info_hdr = (struct dm_info_header *)msg->info;
1025
1026 switch (info_hdr->type) {
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001027 case INFO_TYPE_MAX_PAGE_CNT:
1028 pr_info("Received INFO_TYPE_MAX_PAGE_CNT\n");
K. Y. Srinivasan6427a0d2012-12-06 11:06:54 -08001029 pr_info("Data Size is %d\n", info_hdr->data_size);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001030 break;
1031 default:
K. Y. Srinivasan6427a0d2012-12-06 11:06:54 -08001032 pr_info("Received Unknown type: %d\n", info_hdr->type);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001033 }
1034}
1035
Wei Yongjuna6025a22013-03-20 23:25:59 +08001036static unsigned long compute_balloon_floor(void)
K. Y. Srinivasan1c7db962013-02-08 15:57:16 -08001037{
1038 unsigned long min_pages;
1039#define MB2PAGES(mb) ((mb) << (20 - PAGE_SHIFT))
1040 /* Simple continuous piecewiese linear function:
1041 * max MiB -> min MiB gradient
1042 * 0 0
1043 * 16 16
1044 * 32 24
1045 * 128 72 (1/2)
1046 * 512 168 (1/4)
1047 * 2048 360 (1/8)
Vitaly Kuznetsov7fb0e1a2015-03-27 09:10:12 -07001048 * 8192 744 (1/16)
1049 * 32768 1512 (1/32)
K. Y. Srinivasan1c7db962013-02-08 15:57:16 -08001050 */
1051 if (totalram_pages < MB2PAGES(128))
1052 min_pages = MB2PAGES(8) + (totalram_pages >> 1);
1053 else if (totalram_pages < MB2PAGES(512))
1054 min_pages = MB2PAGES(40) + (totalram_pages >> 2);
1055 else if (totalram_pages < MB2PAGES(2048))
1056 min_pages = MB2PAGES(104) + (totalram_pages >> 3);
K. Y. Srinivasan79208c52015-01-09 23:54:29 -08001057 else if (totalram_pages < MB2PAGES(8192))
Vitaly Kuznetsov7fb0e1a2015-03-27 09:10:12 -07001058 min_pages = MB2PAGES(232) + (totalram_pages >> 4);
K. Y. Srinivasan1c7db962013-02-08 15:57:16 -08001059 else
Vitaly Kuznetsov7fb0e1a2015-03-27 09:10:12 -07001060 min_pages = MB2PAGES(488) + (totalram_pages >> 5);
K. Y. Srinivasan1c7db962013-02-08 15:57:16 -08001061#undef MB2PAGES
1062 return min_pages;
1063}
1064
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001065/*
1066 * Post our status as it relates memory pressure to the
1067 * host. Host expects the guests to post this status
1068 * periodically at 1 second intervals.
1069 *
1070 * The metrics specified in this protocol are very Windows
1071 * specific and so we cook up numbers here to convey our memory
1072 * pressure.
1073 */
1074
1075static void post_status(struct hv_dynmem_device *dm)
1076{
1077 struct dm_status status;
K. Y. Srinivasanae339332014-04-23 13:53:39 -07001078 unsigned long now = jiffies;
1079 unsigned long last_post = last_post_time;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001080
K. Y. Srinivasane500d152013-02-08 15:57:15 -08001081 if (pressure_report_delay > 0) {
1082 --pressure_report_delay;
1083 return;
1084 }
K. Y. Srinivasanae339332014-04-23 13:53:39 -07001085
1086 if (!time_after(now, (last_post_time + HZ)))
1087 return;
1088
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001089 memset(&status, 0, sizeof(struct dm_status));
1090 status.hdr.type = DM_STATUS_REPORT;
1091 status.hdr.size = sizeof(struct dm_status);
1092 status.hdr.trans_id = atomic_inc_return(&trans_id);
1093
K. Y. Srinivasan07315722013-01-25 16:18:47 -08001094 /*
Vitaly Kuznetsov549fd282015-02-28 11:38:59 -08001095 * The host expects the guest to report free and committed memory.
1096 * Furthermore, the host expects the pressure information to include
1097 * the ballooned out pages. For a given amount of memory that we are
1098 * managing we need to compute a floor below which we should not
1099 * balloon. Compute this and add it to the pressure report.
1100 * We also need to report all offline pages (num_pages_added -
1101 * num_pages_onlined) as committed to the host, otherwise it can try
1102 * asking us to balloon them out.
K. Y. Srinivasan07315722013-01-25 16:18:47 -08001103 */
Alex Ngb605c2d2016-08-24 16:23:13 -07001104 status.num_avail = si_mem_available();
K. Y. Srinivasan1c7db962013-02-08 15:57:16 -08001105 status.num_committed = vm_memory_committed() +
Vitaly Kuznetsov549fd282015-02-28 11:38:59 -08001106 dm->num_pages_ballooned +
1107 (dm->num_pages_added > dm->num_pages_onlined ?
1108 dm->num_pages_added - dm->num_pages_onlined : 0) +
1109 compute_balloon_floor();
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001110
K. Y. Srinivasanc5e22542013-07-14 22:38:12 -07001111 /*
1112 * If our transaction ID is no longer current, just don't
1113 * send the status. This can happen if we were interrupted
1114 * after we picked our transaction ID.
1115 */
1116 if (status.hdr.trans_id != atomic_read(&trans_id))
1117 return;
1118
K. Y. Srinivasanae339332014-04-23 13:53:39 -07001119 /*
1120 * If the last post time that we sampled has changed,
1121 * we have raced, don't post the status.
1122 */
1123 if (last_post != last_post_time)
1124 return;
1125
1126 last_post_time = jiffies;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001127 vmbus_sendpacket(dm->dev->channel, &status,
1128 sizeof(struct dm_status),
1129 (unsigned long)NULL,
1130 VM_PKT_DATA_INBAND, 0);
1131
1132}
1133
Greg Kroah-Hartman989623c2012-11-21 12:46:40 -08001134static void free_balloon_pages(struct hv_dynmem_device *dm,
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001135 union dm_mem_page_range *range_array)
1136{
1137 int num_pages = range_array->finfo.page_cnt;
1138 __u64 start_frame = range_array->finfo.start_page;
1139 struct page *pg;
1140 int i;
1141
1142 for (i = 0; i < num_pages; i++) {
1143 pg = pfn_to_page(i + start_frame);
1144 __free_page(pg);
1145 dm->num_pages_ballooned--;
1146 }
1147}
1148
1149
1150
Vitaly Kuznetsov797f88c92015-03-31 11:16:41 -07001151static unsigned int alloc_balloon_pages(struct hv_dynmem_device *dm,
1152 unsigned int num_pages,
1153 struct dm_balloon_response *bl_resp,
1154 int alloc_unit)
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001155{
Vitaly Kuznetsov797f88c92015-03-31 11:16:41 -07001156 unsigned int i = 0;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001157 struct page *pg;
1158
1159 if (num_pages < alloc_unit)
1160 return 0;
1161
1162 for (i = 0; (i * alloc_unit) < num_pages; i++) {
1163 if (bl_resp->hdr.size + sizeof(union dm_mem_page_range) >
1164 PAGE_SIZE)
1165 return i * alloc_unit;
1166
1167 /*
1168 * We execute this code in a thread context. Furthermore,
1169 * we don't want the kernel to try too hard.
1170 */
1171 pg = alloc_pages(GFP_HIGHUSER | __GFP_NORETRY |
1172 __GFP_NOMEMALLOC | __GFP_NOWARN,
1173 get_order(alloc_unit << PAGE_SHIFT));
1174
Vitaly Kuznetsov0a1a86a2015-03-27 09:10:13 -07001175 if (!pg)
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001176 return i * alloc_unit;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001177
1178 dm->num_pages_ballooned += alloc_unit;
1179
K. Y. Srinivasanf766dc12013-03-18 13:51:37 -07001180 /*
1181 * If we allocatted 2M pages; split them so we
1182 * can free them in any order we get.
1183 */
1184
1185 if (alloc_unit != 1)
1186 split_page(pg, get_order(alloc_unit << PAGE_SHIFT));
1187
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001188 bl_resp->range_count++;
1189 bl_resp->range_array[i].finfo.start_page =
1190 page_to_pfn(pg);
1191 bl_resp->range_array[i].finfo.page_cnt = alloc_unit;
1192 bl_resp->hdr.size += sizeof(union dm_mem_page_range);
1193
1194 }
1195
1196 return num_pages;
1197}
1198
1199
1200
K. Y. Srinivasan6571b2d2013-03-15 12:25:40 -07001201static void balloon_up(struct work_struct *dummy)
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001202{
Vitaly Kuznetsov797f88c92015-03-31 11:16:41 -07001203 unsigned int num_pages = dm_device.balloon_wrk.num_pages;
1204 unsigned int num_ballooned = 0;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001205 struct dm_balloon_response *bl_resp;
1206 int alloc_unit;
1207 int ret;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001208 bool done = false;
1209 int i;
Alex Ngb605c2d2016-08-24 16:23:13 -07001210 long avail_pages;
Vitaly Kuznetsov530d15b2015-02-28 11:39:00 -08001211 unsigned long floor;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001212
Dexuan Cuif6712232014-11-24 20:32:43 -08001213 /* The host balloons pages in 2M granularity. */
1214 WARN_ON_ONCE(num_pages % PAGES_IN_2M != 0);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001215
1216 /*
K. Y. Srinivasanf766dc12013-03-18 13:51:37 -07001217 * We will attempt 2M allocations. However, if we fail to
1218 * allocate 2M chunks, we will go back to 4k allocations.
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001219 */
K. Y. Srinivasanf766dc12013-03-18 13:51:37 -07001220 alloc_unit = 512;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001221
Alex Ngb605c2d2016-08-24 16:23:13 -07001222 avail_pages = si_mem_available();
Vitaly Kuznetsov530d15b2015-02-28 11:39:00 -08001223 floor = compute_balloon_floor();
1224
1225 /* Refuse to balloon below the floor, keep the 2M granularity. */
Alex Ngb605c2d2016-08-24 16:23:13 -07001226 if (avail_pages < num_pages || avail_pages - num_pages < floor) {
1227 num_pages = avail_pages > floor ? (avail_pages - floor) : 0;
Vitaly Kuznetsov530d15b2015-02-28 11:39:00 -08001228 num_pages -= num_pages % PAGES_IN_2M;
1229 }
1230
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001231 while (!done) {
1232 bl_resp = (struct dm_balloon_response *)send_buffer;
1233 memset(send_buffer, 0, PAGE_SIZE);
1234 bl_resp->hdr.type = DM_BALLOON_RESPONSE;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001235 bl_resp->hdr.size = sizeof(struct dm_balloon_response);
1236 bl_resp->more_pages = 1;
1237
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001238 num_pages -= num_ballooned;
K. Y. Srinivasan6571b2d2013-03-15 12:25:40 -07001239 num_ballooned = alloc_balloon_pages(&dm_device, num_pages,
Vitaly Kuznetsov0a1a86a2015-03-27 09:10:13 -07001240 bl_resp, alloc_unit);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001241
Dexuan Cuif6712232014-11-24 20:32:43 -08001242 if (alloc_unit != 1 && num_ballooned == 0) {
K. Y. Srinivasanf766dc12013-03-18 13:51:37 -07001243 alloc_unit = 1;
1244 continue;
1245 }
1246
Vitaly Kuznetsov0a1a86a2015-03-27 09:10:13 -07001247 if (num_ballooned == 0 || num_ballooned == num_pages) {
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001248 bl_resp->more_pages = 0;
1249 done = true;
K. Y. Srinivasan6571b2d2013-03-15 12:25:40 -07001250 dm_device.state = DM_INITIALIZED;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001251 }
1252
1253 /*
1254 * We are pushing a lot of data through the channel;
1255 * deal with transient failures caused because of the
1256 * lack of space in the ring buffer.
1257 */
1258
1259 do {
K. Y. Srinivasan20138d6c2013-07-17 17:27:27 -07001260 bl_resp->hdr.trans_id = atomic_inc_return(&trans_id);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001261 ret = vmbus_sendpacket(dm_device.dev->channel,
1262 bl_resp,
1263 bl_resp->hdr.size,
1264 (unsigned long)NULL,
1265 VM_PKT_DATA_INBAND, 0);
1266
1267 if (ret == -EAGAIN)
1268 msleep(20);
K. Y. Srinivasanae339332014-04-23 13:53:39 -07001269 post_status(&dm_device);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001270 } while (ret == -EAGAIN);
1271
1272 if (ret) {
1273 /*
1274 * Free up the memory we allocatted.
1275 */
1276 pr_info("Balloon response failed\n");
1277
1278 for (i = 0; i < bl_resp->range_count; i++)
K. Y. Srinivasan6571b2d2013-03-15 12:25:40 -07001279 free_balloon_pages(&dm_device,
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001280 &bl_resp->range_array[i]);
1281
1282 done = true;
1283 }
1284 }
1285
1286}
1287
1288static void balloon_down(struct hv_dynmem_device *dm,
1289 struct dm_unballoon_request *req)
1290{
1291 union dm_mem_page_range *range_array = req->range_array;
1292 int range_count = req->range_count;
1293 struct dm_unballoon_response resp;
1294 int i;
1295
K. Y. Srinivasanae339332014-04-23 13:53:39 -07001296 for (i = 0; i < range_count; i++) {
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001297 free_balloon_pages(dm, &range_array[i]);
K. Y. Srinivasanab3de222015-01-09 23:54:31 -08001298 complete(&dm_device.config_event);
K. Y. Srinivasanae339332014-04-23 13:53:39 -07001299 }
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001300
1301 if (req->more_pages == 1)
1302 return;
1303
1304 memset(&resp, 0, sizeof(struct dm_unballoon_response));
1305 resp.hdr.type = DM_UNBALLOON_RESPONSE;
1306 resp.hdr.trans_id = atomic_inc_return(&trans_id);
1307 resp.hdr.size = sizeof(struct dm_unballoon_response);
1308
1309 vmbus_sendpacket(dm_device.dev->channel, &resp,
1310 sizeof(struct dm_unballoon_response),
1311 (unsigned long)NULL,
1312 VM_PKT_DATA_INBAND, 0);
1313
1314 dm->state = DM_INITIALIZED;
1315}
1316
1317static void balloon_onchannelcallback(void *context);
1318
1319static int dm_thread_func(void *dm_dev)
1320{
1321 struct hv_dynmem_device *dm = dm_dev;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001322
1323 while (!kthread_should_stop()) {
K. Y. Srinivasanab3de222015-01-09 23:54:31 -08001324 wait_for_completion_interruptible_timeout(
K. Y. Srinivasan5dba4c52014-02-13 16:24:33 -08001325 &dm_device.config_event, 1*HZ);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001326 /*
1327 * The host expects us to post information on the memory
1328 * pressure every second.
1329 */
K. Y. Srinivasanab3de222015-01-09 23:54:31 -08001330 reinit_completion(&dm_device.config_event);
1331 post_status(dm);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001332 }
1333
1334 return 0;
1335}
1336
1337
1338static void version_resp(struct hv_dynmem_device *dm,
1339 struct dm_version_response *vresp)
1340{
1341 struct dm_version_request version_req;
1342 int ret;
1343
1344 if (vresp->is_accepted) {
1345 /*
1346 * We are done; wakeup the
1347 * context waiting for version
1348 * negotiation.
1349 */
1350 complete(&dm->host_event);
1351 return;
1352 }
1353 /*
1354 * If there are more versions to try, continue
1355 * with negotiations; if not
1356 * shutdown the service since we are not able
1357 * to negotiate a suitable version number
1358 * with the host.
1359 */
1360 if (dm->next_version == 0)
1361 goto version_error;
1362
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001363 memset(&version_req, 0, sizeof(struct dm_version_request));
1364 version_req.hdr.type = DM_VERSION_REQUEST;
1365 version_req.hdr.size = sizeof(struct dm_version_request);
1366 version_req.hdr.trans_id = atomic_inc_return(&trans_id);
Alex Ngb6ddeae2015-08-01 16:08:13 -07001367 version_req.version.version = dm->next_version;
1368
1369 /*
1370 * Set the next version to try in case current version fails.
1371 * Win7 protocol ought to be the last one to try.
1372 */
1373 switch (version_req.version.version) {
1374 case DYNMEM_PROTOCOL_VERSION_WIN8:
1375 dm->next_version = DYNMEM_PROTOCOL_VERSION_WIN7;
1376 version_req.is_last_attempt = 0;
1377 break;
1378 default:
1379 dm->next_version = 0;
1380 version_req.is_last_attempt = 1;
1381 }
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001382
1383 ret = vmbus_sendpacket(dm->dev->channel, &version_req,
1384 sizeof(struct dm_version_request),
1385 (unsigned long)NULL,
1386 VM_PKT_DATA_INBAND, 0);
1387
1388 if (ret)
1389 goto version_error;
1390
1391 return;
1392
1393version_error:
1394 dm->state = DM_INIT_ERROR;
1395 complete(&dm->host_event);
1396}
1397
1398static void cap_resp(struct hv_dynmem_device *dm,
1399 struct dm_capabilities_resp_msg *cap_resp)
1400{
1401 if (!cap_resp->is_accepted) {
1402 pr_info("Capabilities not accepted by host\n");
1403 dm->state = DM_INIT_ERROR;
1404 }
1405 complete(&dm->host_event);
1406}
1407
1408static void balloon_onchannelcallback(void *context)
1409{
1410 struct hv_device *dev = context;
1411 u32 recvlen;
1412 u64 requestid;
1413 struct dm_message *dm_msg;
1414 struct dm_header *dm_hdr;
1415 struct hv_dynmem_device *dm = hv_get_drvdata(dev);
K. Y. Srinivasan6571b2d2013-03-15 12:25:40 -07001416 struct dm_balloon *bal_msg;
K. Y. Srinivasanc51af822013-03-15 12:25:41 -07001417 struct dm_hot_add *ha_msg;
1418 union dm_mem_page_range *ha_pg_range;
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -07001419 union dm_mem_page_range *ha_region;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001420
1421 memset(recv_buffer, 0, sizeof(recv_buffer));
1422 vmbus_recvpacket(dev->channel, recv_buffer,
1423 PAGE_SIZE, &recvlen, &requestid);
1424
1425 if (recvlen > 0) {
1426 dm_msg = (struct dm_message *)recv_buffer;
1427 dm_hdr = &dm_msg->hdr;
1428
1429 switch (dm_hdr->type) {
1430 case DM_VERSION_RESPONSE:
1431 version_resp(dm,
1432 (struct dm_version_response *)dm_msg);
1433 break;
1434
1435 case DM_CAPABILITIES_RESPONSE:
1436 cap_resp(dm,
1437 (struct dm_capabilities_resp_msg *)dm_msg);
1438 break;
1439
1440 case DM_BALLOON_REQUEST:
K. Y. Srinivasan6571b2d2013-03-15 12:25:40 -07001441 if (dm->state == DM_BALLOON_UP)
1442 pr_warn("Currently ballooning\n");
1443 bal_msg = (struct dm_balloon *)recv_buffer;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001444 dm->state = DM_BALLOON_UP;
K. Y. Srinivasan6571b2d2013-03-15 12:25:40 -07001445 dm_device.balloon_wrk.num_pages = bal_msg->num_pages;
1446 schedule_work(&dm_device.balloon_wrk.wrk);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001447 break;
1448
1449 case DM_UNBALLOON_REQUEST:
1450 dm->state = DM_BALLOON_DOWN;
1451 balloon_down(dm,
1452 (struct dm_unballoon_request *)recv_buffer);
1453 break;
1454
1455 case DM_MEM_HOT_ADD_REQUEST:
K. Y. Srinivasanc51af822013-03-15 12:25:41 -07001456 if (dm->state == DM_HOT_ADD)
1457 pr_warn("Currently hot-adding\n");
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001458 dm->state = DM_HOT_ADD;
K. Y. Srinivasanc51af822013-03-15 12:25:41 -07001459 ha_msg = (struct dm_hot_add *)recv_buffer;
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -07001460 if (ha_msg->hdr.size == sizeof(struct dm_hot_add)) {
1461 /*
1462 * This is a normal hot-add request specifying
1463 * hot-add memory.
1464 */
Vitaly Kuznetsovd19a55d2016-04-30 19:21:36 -07001465 dm->host_specified_ha_region = false;
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -07001466 ha_pg_range = &ha_msg->range;
1467 dm->ha_wrk.ha_page_range = *ha_pg_range;
1468 dm->ha_wrk.ha_region_range.page_range = 0;
1469 } else {
1470 /*
1471 * Host is specifying that we first hot-add
1472 * a region and then partially populate this
1473 * region.
1474 */
1475 dm->host_specified_ha_region = true;
1476 ha_pg_range = &ha_msg->range;
1477 ha_region = &ha_pg_range[1];
1478 dm->ha_wrk.ha_page_range = *ha_pg_range;
1479 dm->ha_wrk.ha_region_range = *ha_region;
1480 }
K. Y. Srinivasanc51af822013-03-15 12:25:41 -07001481 schedule_work(&dm_device.ha_wrk.wrk);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001482 break;
1483
1484 case DM_INFO_MESSAGE:
1485 process_info(dm, (struct dm_info_msg *)dm_msg);
1486 break;
1487
1488 default:
1489 pr_err("Unhandled message: type: %d\n", dm_hdr->type);
1490
1491 }
1492 }
1493
1494}
1495
1496static int balloon_probe(struct hv_device *dev,
1497 const struct hv_vmbus_device_id *dev_id)
1498{
Nicholas Mc Guireb057b3a2015-02-27 11:26:03 -08001499 int ret;
1500 unsigned long t;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001501 struct dm_version_request version_req;
1502 struct dm_capabilities cap_msg;
1503
1504 do_hot_add = hot_add;
1505
1506 /*
1507 * First allocate a send buffer.
1508 */
1509
1510 send_buffer = kmalloc(PAGE_SIZE, GFP_KERNEL);
1511 if (!send_buffer)
1512 return -ENOMEM;
1513
1514 ret = vmbus_open(dev->channel, dm_ring_size, dm_ring_size, NULL, 0,
1515 balloon_onchannelcallback, dev);
1516
1517 if (ret)
K. Y. Srinivasan33080c12012-12-11 11:07:17 -08001518 goto probe_error0;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001519
1520 dm_device.dev = dev;
1521 dm_device.state = DM_INITIALIZING;
Alex Ngb6ddeae2015-08-01 16:08:13 -07001522 dm_device.next_version = DYNMEM_PROTOCOL_VERSION_WIN8;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001523 init_completion(&dm_device.host_event);
1524 init_completion(&dm_device.config_event);
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -07001525 INIT_LIST_HEAD(&dm_device.ha_region_list);
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -07001526 spin_lock_init(&dm_device.ha_lock);
K. Y. Srinivasan6571b2d2013-03-15 12:25:40 -07001527 INIT_WORK(&dm_device.balloon_wrk.wrk, balloon_up);
K. Y. Srinivasanc51af822013-03-15 12:25:41 -07001528 INIT_WORK(&dm_device.ha_wrk.wrk, hot_add_req);
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -07001529 dm_device.host_specified_ha_region = false;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001530
1531 dm_device.thread =
1532 kthread_run(dm_thread_func, &dm_device, "hv_balloon");
1533 if (IS_ERR(dm_device.thread)) {
1534 ret = PTR_ERR(dm_device.thread);
K. Y. Srinivasan33080c12012-12-11 11:07:17 -08001535 goto probe_error1;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001536 }
1537
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -07001538#ifdef CONFIG_MEMORY_HOTPLUG
1539 set_online_page_callback(&hv_online_page);
K. Y. Srinivasan22f88472015-01-09 23:54:30 -08001540 register_memory_notifier(&hv_memory_nb);
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -07001541#endif
1542
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001543 hv_set_drvdata(dev, &dm_device);
1544 /*
1545 * Initiate the hand shake with the host and negotiate
1546 * a version that the host can support. We start with the
1547 * highest version number and go down if the host cannot
1548 * support it.
1549 */
1550 memset(&version_req, 0, sizeof(struct dm_version_request));
1551 version_req.hdr.type = DM_VERSION_REQUEST;
1552 version_req.hdr.size = sizeof(struct dm_version_request);
1553 version_req.hdr.trans_id = atomic_inc_return(&trans_id);
Alex Ngb6ddeae2015-08-01 16:08:13 -07001554 version_req.version.version = DYNMEM_PROTOCOL_VERSION_WIN10;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001555 version_req.is_last_attempt = 0;
1556
1557 ret = vmbus_sendpacket(dev->channel, &version_req,
1558 sizeof(struct dm_version_request),
1559 (unsigned long)NULL,
K. Y. Srinivasan7a64b862013-03-15 12:25:39 -07001560 VM_PKT_DATA_INBAND, 0);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001561 if (ret)
K. Y. Srinivasan33080c12012-12-11 11:07:17 -08001562 goto probe_error2;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001563
1564 t = wait_for_completion_timeout(&dm_device.host_event, 5*HZ);
1565 if (t == 0) {
1566 ret = -ETIMEDOUT;
K. Y. Srinivasan33080c12012-12-11 11:07:17 -08001567 goto probe_error2;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001568 }
1569
1570 /*
1571 * If we could not negotiate a compatible version with the host
1572 * fail the probe function.
1573 */
1574 if (dm_device.state == DM_INIT_ERROR) {
1575 ret = -ETIMEDOUT;
K. Y. Srinivasan33080c12012-12-11 11:07:17 -08001576 goto probe_error2;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001577 }
1578 /*
1579 * Now submit our capabilities to the host.
1580 */
1581 memset(&cap_msg, 0, sizeof(struct dm_capabilities));
1582 cap_msg.hdr.type = DM_CAPABILITIES_REPORT;
1583 cap_msg.hdr.size = sizeof(struct dm_capabilities);
1584 cap_msg.hdr.trans_id = atomic_inc_return(&trans_id);
1585
1586 cap_msg.caps.cap_bits.balloon = 1;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001587 cap_msg.caps.cap_bits.hot_add = 1;
1588
1589 /*
K. Y. Srinivasan647965a2013-03-29 07:36:11 -07001590 * Specify our alignment requirements as it relates
1591 * memory hot-add. Specify 128MB alignment.
1592 */
1593 cap_msg.caps.cap_bits.hot_add_alignment = 7;
1594
1595 /*
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001596 * Currently the host does not use these
1597 * values and we set them to what is done in the
1598 * Windows driver.
1599 */
1600 cap_msg.min_page_cnt = 0;
1601 cap_msg.max_page_number = -1;
1602
1603 ret = vmbus_sendpacket(dev->channel, &cap_msg,
1604 sizeof(struct dm_capabilities),
1605 (unsigned long)NULL,
K. Y. Srinivasan7a64b862013-03-15 12:25:39 -07001606 VM_PKT_DATA_INBAND, 0);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001607 if (ret)
K. Y. Srinivasan33080c12012-12-11 11:07:17 -08001608 goto probe_error2;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001609
1610 t = wait_for_completion_timeout(&dm_device.host_event, 5*HZ);
1611 if (t == 0) {
1612 ret = -ETIMEDOUT;
K. Y. Srinivasan33080c12012-12-11 11:07:17 -08001613 goto probe_error2;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001614 }
1615
1616 /*
1617 * If the host does not like our capabilities,
1618 * fail the probe function.
1619 */
1620 if (dm_device.state == DM_INIT_ERROR) {
1621 ret = -ETIMEDOUT;
K. Y. Srinivasan33080c12012-12-11 11:07:17 -08001622 goto probe_error2;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001623 }
1624
1625 dm_device.state = DM_INITIALIZED;
1626
1627 return 0;
1628
K. Y. Srinivasan33080c12012-12-11 11:07:17 -08001629probe_error2:
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -07001630#ifdef CONFIG_MEMORY_HOTPLUG
1631 restore_online_page_callback(&hv_online_page);
1632#endif
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001633 kthread_stop(dm_device.thread);
1634
K. Y. Srinivasan33080c12012-12-11 11:07:17 -08001635probe_error1:
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001636 vmbus_close(dev->channel);
K. Y. Srinivasan33080c12012-12-11 11:07:17 -08001637probe_error0:
1638 kfree(send_buffer);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001639 return ret;
1640}
1641
1642static int balloon_remove(struct hv_device *dev)
1643{
1644 struct hv_dynmem_device *dm = hv_get_drvdata(dev);
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -07001645 struct hv_hotadd_state *has, *tmp;
Vitaly Kuznetsovcb7a5722016-08-24 16:23:10 -07001646 struct hv_hotadd_gap *gap, *tmp_gap;
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -07001647 unsigned long flags;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001648
1649 if (dm->num_pages_ballooned != 0)
1650 pr_warn("Ballooned pages: %d\n", dm->num_pages_ballooned);
1651
K. Y. Srinivasan6571b2d2013-03-15 12:25:40 -07001652 cancel_work_sync(&dm->balloon_wrk.wrk);
K. Y. Srinivasanc51af822013-03-15 12:25:41 -07001653 cancel_work_sync(&dm->ha_wrk.wrk);
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -07001654
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001655 vmbus_close(dev->channel);
1656 kthread_stop(dm->thread);
K. Y. Srinivasan33080c12012-12-11 11:07:17 -08001657 kfree(send_buffer);
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -07001658#ifdef CONFIG_MEMORY_HOTPLUG
1659 restore_online_page_callback(&hv_online_page);
K. Y. Srinivasan22f88472015-01-09 23:54:30 -08001660 unregister_memory_notifier(&hv_memory_nb);
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -07001661#endif
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -07001662 spin_lock_irqsave(&dm_device.ha_lock, flags);
1663 list_for_each_entry_safe(has, tmp, &dm->ha_region_list, list) {
Vitaly Kuznetsovcb7a5722016-08-24 16:23:10 -07001664 list_for_each_entry_safe(gap, tmp_gap, &has->gap_list, list) {
1665 list_del(&gap->list);
1666 kfree(gap);
1667 }
K. Y. Srinivasan1cac8cd2013-03-15 12:25:43 -07001668 list_del(&has->list);
1669 kfree(has);
1670 }
Vitaly Kuznetsoveece30b2016-08-24 16:23:12 -07001671 spin_unlock_irqrestore(&dm_device.ha_lock, flags);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001672
1673 return 0;
1674}
1675
1676static const struct hv_vmbus_device_id id_table[] = {
1677 /* Dynamic Memory Class ID */
1678 /* 525074DC-8985-46e2-8057-A307DC18A502 */
K. Y. Srinivasand13984e2013-01-23 17:42:41 -08001679 { HV_DM_GUID, },
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001680 { },
1681};
1682
1683MODULE_DEVICE_TABLE(vmbus, id_table);
1684
1685static struct hv_driver balloon_drv = {
1686 .name = "hv_balloon",
1687 .id_table = id_table,
1688 .probe = balloon_probe,
1689 .remove = balloon_remove,
1690};
1691
1692static int __init init_balloon_drv(void)
1693{
1694
1695 return vmbus_driver_register(&balloon_drv);
1696}
1697
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001698module_init(init_balloon_drv);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001699
1700MODULE_DESCRIPTION("Hyper-V Balloon");
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001701MODULE_LICENSE("GPL");