blob: 37873213e24f5f55c6ccf136e5b7fe64cdcfeb3b [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;
120 __u64 reservedz:62;
121 } cap_bits;
122 __u64 caps;
123} __packed;
124
125union dm_mem_page_range {
126 struct {
127 /*
128 * The PFN number of the first page in the range.
129 * 40 bits is the architectural limit of a PFN
130 * number for AMD64.
131 */
132 __u64 start_page:40;
133 /*
134 * The number of pages in the range.
135 */
136 __u64 page_cnt:24;
137 } finfo;
138 __u64 page_range;
139} __packed;
140
141
142
143/*
144 * The header for all dynamic memory messages:
145 *
146 * type: Type of the message.
147 * size: Size of the message in bytes; including the header.
148 * trans_id: The guest is responsible for manufacturing this ID.
149 */
150
151struct dm_header {
152 __u16 type;
153 __u16 size;
154 __u32 trans_id;
155} __packed;
156
157/*
158 * A generic message format for dynamic memory.
159 * Specific message formats are defined later in the file.
160 */
161
162struct dm_message {
163 struct dm_header hdr;
164 __u8 data[]; /* enclosed message */
165} __packed;
166
167
168/*
169 * Specific message types supporting the dynamic memory protocol.
170 */
171
172/*
173 * Version negotiation message. Sent from the guest to the host.
174 * The guest is free to try different versions until the host
175 * accepts the version.
176 *
177 * dm_version: The protocol version requested.
178 * is_last_attempt: If TRUE, this is the last version guest will request.
179 * reservedz: Reserved field, set to zero.
180 */
181
182struct dm_version_request {
183 struct dm_header hdr;
184 union dm_version version;
185 __u32 is_last_attempt:1;
186 __u32 reservedz:31;
187} __packed;
188
189/*
190 * Version response message; Host to Guest and indicates
191 * if the host has accepted the version sent by the guest.
192 *
193 * is_accepted: If TRUE, host has accepted the version and the guest
194 * should proceed to the next stage of the protocol. FALSE indicates that
195 * guest should re-try with a different version.
196 *
197 * reservedz: Reserved field, set to zero.
198 */
199
200struct dm_version_response {
201 struct dm_header hdr;
202 __u64 is_accepted:1;
203 __u64 reservedz:63;
204} __packed;
205
206/*
207 * Message reporting capabilities. This is sent from the guest to the
208 * host.
209 */
210
211struct dm_capabilities {
212 struct dm_header hdr;
213 union dm_caps caps;
214 __u64 min_page_cnt;
215 __u64 max_page_number;
216} __packed;
217
218/*
219 * Response to the capabilities message. This is sent from the host to the
220 * guest. This message notifies if the host has accepted the guest's
221 * capabilities. If the host has not accepted, the guest must shutdown
222 * the service.
223 *
224 * is_accepted: Indicates if the host has accepted guest's capabilities.
225 * reservedz: Must be 0.
226 */
227
228struct dm_capabilities_resp_msg {
229 struct dm_header hdr;
230 __u64 is_accepted:1;
231 __u64 reservedz:63;
232} __packed;
233
234/*
235 * This message is used to report memory pressure from the guest.
236 * This message is not part of any transaction and there is no
237 * response to this message.
238 *
239 * num_avail: Available memory in pages.
240 * num_committed: Committed memory in pages.
241 * page_file_size: The accumulated size of all page files
242 * in the system in pages.
243 * zero_free: The nunber of zero and free pages.
244 * page_file_writes: The writes to the page file in pages.
245 * io_diff: An indicator of file cache efficiency or page file activity,
246 * calculated as File Cache Page Fault Count - Page Read Count.
247 * This value is in pages.
248 *
249 * Some of these metrics are Windows specific and fortunately
250 * the algorithm on the host side that computes the guest memory
251 * pressure only uses num_committed value.
252 */
253
254struct dm_status {
255 struct dm_header hdr;
256 __u64 num_avail;
257 __u64 num_committed;
258 __u64 page_file_size;
259 __u64 zero_free;
260 __u32 page_file_writes;
261 __u32 io_diff;
262} __packed;
263
264
265/*
266 * Message to ask the guest to allocate memory - balloon up message.
267 * This message is sent from the host to the guest. The guest may not be
268 * able to allocate as much memory as requested.
269 *
270 * num_pages: number of pages to allocate.
271 */
272
273struct dm_balloon {
274 struct dm_header hdr;
275 __u32 num_pages;
276 __u32 reservedz;
277} __packed;
278
279
280/*
281 * Balloon response message; this message is sent from the guest
282 * to the host in response to the balloon message.
283 *
284 * reservedz: Reserved; must be set to zero.
285 * more_pages: If FALSE, this is the last message of the transaction.
286 * if TRUE there will atleast one more message from the guest.
287 *
288 * range_count: The number of ranges in the range array.
289 *
290 * range_array: An array of page ranges returned to the host.
291 *
292 */
293
294struct dm_balloon_response {
295 struct dm_header hdr;
296 __u32 reservedz;
297 __u32 more_pages:1;
298 __u32 range_count:31;
299 union dm_mem_page_range range_array[];
300} __packed;
301
302/*
303 * Un-balloon message; this message is sent from the host
304 * to the guest to give guest more memory.
305 *
306 * more_pages: If FALSE, this is the last message of the transaction.
307 * if TRUE there will atleast one more message from the guest.
308 *
309 * reservedz: Reserved; must be set to zero.
310 *
311 * range_count: The number of ranges in the range array.
312 *
313 * range_array: An array of page ranges returned to the host.
314 *
315 */
316
317struct dm_unballoon_request {
318 struct dm_header hdr;
319 __u32 more_pages:1;
320 __u32 reservedz:31;
321 __u32 range_count;
322 union dm_mem_page_range range_array[];
323} __packed;
324
325/*
326 * Un-balloon response message; this message is sent from the guest
327 * to the host in response to an unballoon request.
328 *
329 */
330
331struct dm_unballoon_response {
332 struct dm_header hdr;
333} __packed;
334
335
336/*
337 * Hot add request message. Message sent from the host to the guest.
338 *
339 * mem_range: Memory range to hot add.
340 *
341 * On Linux we currently don't support this since we cannot hot add
342 * arbitrary granularity of memory.
343 */
344
345struct dm_hot_add {
346 struct dm_header hdr;
347 union dm_mem_page_range range;
348} __packed;
349
350/*
351 * Hot add response message.
352 * This message is sent by the guest to report the status of a hot add request.
353 * If page_count is less than the requested page count, then the host should
354 * assume all further hot add requests will fail, since this indicates that
355 * the guest has hit an upper physical memory barrier.
356 *
357 * Hot adds may also fail due to low resources; in this case, the guest must
358 * not complete this message until the hot add can succeed, and the host must
359 * not send a new hot add request until the response is sent.
360 * If VSC fails to hot add memory DYNMEM_NUMBER_OF_UNSUCCESSFUL_HOTADD_ATTEMPTS
361 * times it fails the request.
362 *
363 *
364 * page_count: number of pages that were successfully hot added.
365 *
366 * result: result of the operation 1: success, 0: failure.
367 *
368 */
369
370struct dm_hot_add_response {
371 struct dm_header hdr;
372 __u32 page_count;
373 __u32 result;
374} __packed;
375
376/*
377 * Types of information sent from host to the guest.
378 */
379
380enum dm_info_type {
381 INFO_TYPE_MAX_PAGE_CNT = 0,
382 MAX_INFO_TYPE
383};
384
385
386/*
387 * Header for the information message.
388 */
389
390struct dm_info_header {
391 enum dm_info_type type;
392 __u32 data_size;
393} __packed;
394
395/*
396 * This message is sent from the host to the guest to pass
397 * some relevant information (win8 addition).
398 *
399 * reserved: no used.
400 * info_size: size of the information blob.
401 * info: information blob.
402 */
403
404struct dm_info_msg {
K. Y. Srinivasan6427a0d2012-12-06 11:06:54 -0800405 struct dm_header hdr;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800406 __u32 reserved;
407 __u32 info_size;
408 __u8 info[];
409};
410
411/*
412 * End protocol definitions.
413 */
414
415static bool hot_add;
416static bool do_hot_add;
K. Y. Srinivasane500d152013-02-08 15:57:15 -0800417/*
418 * Delay reporting memory pressure by
419 * the specified number of seconds.
420 */
421static uint pressure_report_delay = 30;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800422
423module_param(hot_add, bool, (S_IRUGO | S_IWUSR));
424MODULE_PARM_DESC(hot_add, "If set attempt memory hot_add");
425
K. Y. Srinivasane500d152013-02-08 15:57:15 -0800426module_param(pressure_report_delay, uint, (S_IRUGO | S_IWUSR));
427MODULE_PARM_DESC(pressure_report_delay, "Delay in secs in reporting pressure");
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800428static atomic_t trans_id = ATOMIC_INIT(0);
429
430static int dm_ring_size = (5 * PAGE_SIZE);
431
432/*
433 * Driver specific state.
434 */
435
436enum hv_dm_state {
437 DM_INITIALIZING = 0,
438 DM_INITIALIZED,
439 DM_BALLOON_UP,
440 DM_BALLOON_DOWN,
441 DM_HOT_ADD,
442 DM_INIT_ERROR
443};
444
445
446static __u8 recv_buffer[PAGE_SIZE];
447static __u8 *send_buffer;
448#define PAGES_IN_2M 512
449
450struct hv_dynmem_device {
451 struct hv_device *dev;
452 enum hv_dm_state state;
453 struct completion host_event;
454 struct completion config_event;
455
456 /*
457 * Number of pages we have currently ballooned out.
458 */
459 unsigned int num_pages_ballooned;
460
461 /*
462 * This thread handles both balloon/hot-add
463 * requests from the host as well as notifying
464 * the host with regards to memory pressure in
465 * the guest.
466 */
467 struct task_struct *thread;
468
469 /*
470 * We start with the highest version we can support
471 * and downgrade based on the host; we save here the
472 * next version to try.
473 */
474 __u32 next_version;
475};
476
477static struct hv_dynmem_device dm_device;
478
479static void hot_add_req(struct hv_dynmem_device *dm, struct dm_hot_add *msg)
480{
481
482 struct dm_hot_add_response resp;
483
484 if (do_hot_add) {
485
486 pr_info("Memory hot add not supported\n");
487
488 /*
489 * Currently we do not support hot add.
490 * Just fail the request.
491 */
492 }
493
494 memset(&resp, 0, sizeof(struct dm_hot_add_response));
495 resp.hdr.type = DM_MEM_HOT_ADD_RESPONSE;
496 resp.hdr.size = sizeof(struct dm_hot_add_response);
497 resp.hdr.trans_id = atomic_inc_return(&trans_id);
498
499 resp.page_count = 0;
500 resp.result = 0;
501
502 dm->state = DM_INITIALIZED;
503 vmbus_sendpacket(dm->dev->channel, &resp,
504 sizeof(struct dm_hot_add_response),
505 (unsigned long)NULL,
506 VM_PKT_DATA_INBAND, 0);
507
508}
509
510static void process_info(struct hv_dynmem_device *dm, struct dm_info_msg *msg)
511{
K. Y. Srinivasan6427a0d2012-12-06 11:06:54 -0800512 struct dm_info_header *info_hdr;
513
514 info_hdr = (struct dm_info_header *)msg->info;
515
516 switch (info_hdr->type) {
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800517 case INFO_TYPE_MAX_PAGE_CNT:
518 pr_info("Received INFO_TYPE_MAX_PAGE_CNT\n");
K. Y. Srinivasan6427a0d2012-12-06 11:06:54 -0800519 pr_info("Data Size is %d\n", info_hdr->data_size);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800520 break;
521 default:
K. Y. Srinivasan6427a0d2012-12-06 11:06:54 -0800522 pr_info("Received Unknown type: %d\n", info_hdr->type);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800523 }
524}
525
K. Y. Srinivasan1c7db962013-02-08 15:57:16 -0800526unsigned long compute_balloon_floor(void)
527{
528 unsigned long min_pages;
529#define MB2PAGES(mb) ((mb) << (20 - PAGE_SHIFT))
530 /* Simple continuous piecewiese linear function:
531 * max MiB -> min MiB gradient
532 * 0 0
533 * 16 16
534 * 32 24
535 * 128 72 (1/2)
536 * 512 168 (1/4)
537 * 2048 360 (1/8)
538 * 8192 552 (1/32)
539 * 32768 1320
540 * 131072 4392
541 */
542 if (totalram_pages < MB2PAGES(128))
543 min_pages = MB2PAGES(8) + (totalram_pages >> 1);
544 else if (totalram_pages < MB2PAGES(512))
545 min_pages = MB2PAGES(40) + (totalram_pages >> 2);
546 else if (totalram_pages < MB2PAGES(2048))
547 min_pages = MB2PAGES(104) + (totalram_pages >> 3);
548 else
549 min_pages = MB2PAGES(296) + (totalram_pages >> 5);
550#undef MB2PAGES
551 return min_pages;
552}
553
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800554/*
555 * Post our status as it relates memory pressure to the
556 * host. Host expects the guests to post this status
557 * periodically at 1 second intervals.
558 *
559 * The metrics specified in this protocol are very Windows
560 * specific and so we cook up numbers here to convey our memory
561 * pressure.
562 */
563
564static void post_status(struct hv_dynmem_device *dm)
565{
566 struct dm_status status;
K. Y. Srinivasan07315722013-01-25 16:18:47 -0800567 struct sysinfo val;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800568
K. Y. Srinivasane500d152013-02-08 15:57:15 -0800569 if (pressure_report_delay > 0) {
570 --pressure_report_delay;
571 return;
572 }
K. Y. Srinivasan07315722013-01-25 16:18:47 -0800573 si_meminfo(&val);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800574 memset(&status, 0, sizeof(struct dm_status));
575 status.hdr.type = DM_STATUS_REPORT;
576 status.hdr.size = sizeof(struct dm_status);
577 status.hdr.trans_id = atomic_inc_return(&trans_id);
578
K. Y. Srinivasan07315722013-01-25 16:18:47 -0800579 /*
580 * The host expects the guest to report free memory.
581 * Further, the host expects the pressure information to
582 * include the ballooned out pages.
K. Y. Srinivasan1c7db962013-02-08 15:57:16 -0800583 * For a given amount of memory that we are managing, we
584 * need to compute a floor below which we should not balloon.
585 * Compute this and add it to the pressure report.
K. Y. Srinivasan07315722013-01-25 16:18:47 -0800586 */
587 status.num_avail = val.freeram;
K. Y. Srinivasan1c7db962013-02-08 15:57:16 -0800588 status.num_committed = vm_memory_committed() +
589 dm->num_pages_ballooned +
590 compute_balloon_floor();
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800591
592 vmbus_sendpacket(dm->dev->channel, &status,
593 sizeof(struct dm_status),
594 (unsigned long)NULL,
595 VM_PKT_DATA_INBAND, 0);
596
597}
598
Greg Kroah-Hartman989623c2012-11-21 12:46:40 -0800599static void free_balloon_pages(struct hv_dynmem_device *dm,
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800600 union dm_mem_page_range *range_array)
601{
602 int num_pages = range_array->finfo.page_cnt;
603 __u64 start_frame = range_array->finfo.start_page;
604 struct page *pg;
605 int i;
606
607 for (i = 0; i < num_pages; i++) {
608 pg = pfn_to_page(i + start_frame);
609 __free_page(pg);
610 dm->num_pages_ballooned--;
611 }
612}
613
614
615
616static int alloc_balloon_pages(struct hv_dynmem_device *dm, int num_pages,
617 struct dm_balloon_response *bl_resp, int alloc_unit,
618 bool *alloc_error)
619{
620 int i = 0;
621 struct page *pg;
622
623 if (num_pages < alloc_unit)
624 return 0;
625
626 for (i = 0; (i * alloc_unit) < num_pages; i++) {
627 if (bl_resp->hdr.size + sizeof(union dm_mem_page_range) >
628 PAGE_SIZE)
629 return i * alloc_unit;
630
631 /*
632 * We execute this code in a thread context. Furthermore,
633 * we don't want the kernel to try too hard.
634 */
635 pg = alloc_pages(GFP_HIGHUSER | __GFP_NORETRY |
636 __GFP_NOMEMALLOC | __GFP_NOWARN,
637 get_order(alloc_unit << PAGE_SHIFT));
638
639 if (!pg) {
640 *alloc_error = true;
641 return i * alloc_unit;
642 }
643
644
645 dm->num_pages_ballooned += alloc_unit;
646
647 bl_resp->range_count++;
648 bl_resp->range_array[i].finfo.start_page =
649 page_to_pfn(pg);
650 bl_resp->range_array[i].finfo.page_cnt = alloc_unit;
651 bl_resp->hdr.size += sizeof(union dm_mem_page_range);
652
653 }
654
655 return num_pages;
656}
657
658
659
660static void balloon_up(struct hv_dynmem_device *dm, struct dm_balloon *req)
661{
662 int num_pages = req->num_pages;
663 int num_ballooned = 0;
664 struct dm_balloon_response *bl_resp;
665 int alloc_unit;
666 int ret;
667 bool alloc_error = false;
668 bool done = false;
669 int i;
670
671
672 /*
673 * Currently, we only support 4k allocations.
674 */
675 alloc_unit = 1;
676
677 while (!done) {
678 bl_resp = (struct dm_balloon_response *)send_buffer;
679 memset(send_buffer, 0, PAGE_SIZE);
680 bl_resp->hdr.type = DM_BALLOON_RESPONSE;
681 bl_resp->hdr.trans_id = atomic_inc_return(&trans_id);
682 bl_resp->hdr.size = sizeof(struct dm_balloon_response);
683 bl_resp->more_pages = 1;
684
685
686 num_pages -= num_ballooned;
687 num_ballooned = alloc_balloon_pages(dm, num_pages,
688 bl_resp, alloc_unit,
689 &alloc_error);
690
691 if ((alloc_error) || (num_ballooned == num_pages)) {
692 bl_resp->more_pages = 0;
693 done = true;
694 dm->state = DM_INITIALIZED;
695 }
696
697 /*
698 * We are pushing a lot of data through the channel;
699 * deal with transient failures caused because of the
700 * lack of space in the ring buffer.
701 */
702
703 do {
704 ret = vmbus_sendpacket(dm_device.dev->channel,
705 bl_resp,
706 bl_resp->hdr.size,
707 (unsigned long)NULL,
708 VM_PKT_DATA_INBAND, 0);
709
710 if (ret == -EAGAIN)
711 msleep(20);
712
713 } while (ret == -EAGAIN);
714
715 if (ret) {
716 /*
717 * Free up the memory we allocatted.
718 */
719 pr_info("Balloon response failed\n");
720
721 for (i = 0; i < bl_resp->range_count; i++)
722 free_balloon_pages(dm,
723 &bl_resp->range_array[i]);
724
725 done = true;
726 }
727 }
728
729}
730
731static void balloon_down(struct hv_dynmem_device *dm,
732 struct dm_unballoon_request *req)
733{
734 union dm_mem_page_range *range_array = req->range_array;
735 int range_count = req->range_count;
736 struct dm_unballoon_response resp;
737 int i;
738
739 for (i = 0; i < range_count; i++)
740 free_balloon_pages(dm, &range_array[i]);
741
742 if (req->more_pages == 1)
743 return;
744
745 memset(&resp, 0, sizeof(struct dm_unballoon_response));
746 resp.hdr.type = DM_UNBALLOON_RESPONSE;
747 resp.hdr.trans_id = atomic_inc_return(&trans_id);
748 resp.hdr.size = sizeof(struct dm_unballoon_response);
749
750 vmbus_sendpacket(dm_device.dev->channel, &resp,
751 sizeof(struct dm_unballoon_response),
752 (unsigned long)NULL,
753 VM_PKT_DATA_INBAND, 0);
754
755 dm->state = DM_INITIALIZED;
756}
757
758static void balloon_onchannelcallback(void *context);
759
760static int dm_thread_func(void *dm_dev)
761{
762 struct hv_dynmem_device *dm = dm_dev;
763 int t;
764 unsigned long scan_start;
765
766 while (!kthread_should_stop()) {
767 t = wait_for_completion_timeout(&dm_device.config_event, 1*HZ);
768 /*
769 * The host expects us to post information on the memory
770 * pressure every second.
771 */
772
773 if (t == 0)
774 post_status(dm);
775
776 scan_start = jiffies;
777 switch (dm->state) {
778 case DM_BALLOON_UP:
779 balloon_up(dm, (struct dm_balloon *)recv_buffer);
780 break;
781
782 case DM_HOT_ADD:
783 hot_add_req(dm, (struct dm_hot_add *)recv_buffer);
784 break;
785 default:
786 break;
787 }
788
789 if (!time_in_range(jiffies, scan_start, scan_start + HZ))
790 post_status(dm);
791
792 }
793
794 return 0;
795}
796
797
798static void version_resp(struct hv_dynmem_device *dm,
799 struct dm_version_response *vresp)
800{
801 struct dm_version_request version_req;
802 int ret;
803
804 if (vresp->is_accepted) {
805 /*
806 * We are done; wakeup the
807 * context waiting for version
808 * negotiation.
809 */
810 complete(&dm->host_event);
811 return;
812 }
813 /*
814 * If there are more versions to try, continue
815 * with negotiations; if not
816 * shutdown the service since we are not able
817 * to negotiate a suitable version number
818 * with the host.
819 */
820 if (dm->next_version == 0)
821 goto version_error;
822
823 dm->next_version = 0;
824 memset(&version_req, 0, sizeof(struct dm_version_request));
825 version_req.hdr.type = DM_VERSION_REQUEST;
826 version_req.hdr.size = sizeof(struct dm_version_request);
827 version_req.hdr.trans_id = atomic_inc_return(&trans_id);
828 version_req.version.version = DYNMEM_PROTOCOL_VERSION_WIN7;
829 version_req.is_last_attempt = 1;
830
831 ret = vmbus_sendpacket(dm->dev->channel, &version_req,
832 sizeof(struct dm_version_request),
833 (unsigned long)NULL,
834 VM_PKT_DATA_INBAND, 0);
835
836 if (ret)
837 goto version_error;
838
839 return;
840
841version_error:
842 dm->state = DM_INIT_ERROR;
843 complete(&dm->host_event);
844}
845
846static void cap_resp(struct hv_dynmem_device *dm,
847 struct dm_capabilities_resp_msg *cap_resp)
848{
849 if (!cap_resp->is_accepted) {
850 pr_info("Capabilities not accepted by host\n");
851 dm->state = DM_INIT_ERROR;
852 }
853 complete(&dm->host_event);
854}
855
856static void balloon_onchannelcallback(void *context)
857{
858 struct hv_device *dev = context;
859 u32 recvlen;
860 u64 requestid;
861 struct dm_message *dm_msg;
862 struct dm_header *dm_hdr;
863 struct hv_dynmem_device *dm = hv_get_drvdata(dev);
864
865 memset(recv_buffer, 0, sizeof(recv_buffer));
866 vmbus_recvpacket(dev->channel, recv_buffer,
867 PAGE_SIZE, &recvlen, &requestid);
868
869 if (recvlen > 0) {
870 dm_msg = (struct dm_message *)recv_buffer;
871 dm_hdr = &dm_msg->hdr;
872
873 switch (dm_hdr->type) {
874 case DM_VERSION_RESPONSE:
875 version_resp(dm,
876 (struct dm_version_response *)dm_msg);
877 break;
878
879 case DM_CAPABILITIES_RESPONSE:
880 cap_resp(dm,
881 (struct dm_capabilities_resp_msg *)dm_msg);
882 break;
883
884 case DM_BALLOON_REQUEST:
885 dm->state = DM_BALLOON_UP;
886 complete(&dm->config_event);
887 break;
888
889 case DM_UNBALLOON_REQUEST:
890 dm->state = DM_BALLOON_DOWN;
891 balloon_down(dm,
892 (struct dm_unballoon_request *)recv_buffer);
893 break;
894
895 case DM_MEM_HOT_ADD_REQUEST:
896 dm->state = DM_HOT_ADD;
897 complete(&dm->config_event);
898 break;
899
900 case DM_INFO_MESSAGE:
901 process_info(dm, (struct dm_info_msg *)dm_msg);
902 break;
903
904 default:
905 pr_err("Unhandled message: type: %d\n", dm_hdr->type);
906
907 }
908 }
909
910}
911
912static int balloon_probe(struct hv_device *dev,
913 const struct hv_vmbus_device_id *dev_id)
914{
915 int ret, t;
916 struct dm_version_request version_req;
917 struct dm_capabilities cap_msg;
918
919 do_hot_add = hot_add;
920
921 /*
922 * First allocate a send buffer.
923 */
924
925 send_buffer = kmalloc(PAGE_SIZE, GFP_KERNEL);
926 if (!send_buffer)
927 return -ENOMEM;
928
929 ret = vmbus_open(dev->channel, dm_ring_size, dm_ring_size, NULL, 0,
930 balloon_onchannelcallback, dev);
931
932 if (ret)
K. Y. Srinivasan33080c12012-12-11 11:07:17 -0800933 goto probe_error0;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800934
935 dm_device.dev = dev;
936 dm_device.state = DM_INITIALIZING;
937 dm_device.next_version = DYNMEM_PROTOCOL_VERSION_WIN7;
938 init_completion(&dm_device.host_event);
939 init_completion(&dm_device.config_event);
940
941 dm_device.thread =
942 kthread_run(dm_thread_func, &dm_device, "hv_balloon");
943 if (IS_ERR(dm_device.thread)) {
944 ret = PTR_ERR(dm_device.thread);
K. Y. Srinivasan33080c12012-12-11 11:07:17 -0800945 goto probe_error1;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800946 }
947
948 hv_set_drvdata(dev, &dm_device);
949 /*
950 * Initiate the hand shake with the host and negotiate
951 * a version that the host can support. We start with the
952 * highest version number and go down if the host cannot
953 * support it.
954 */
955 memset(&version_req, 0, sizeof(struct dm_version_request));
956 version_req.hdr.type = DM_VERSION_REQUEST;
957 version_req.hdr.size = sizeof(struct dm_version_request);
958 version_req.hdr.trans_id = atomic_inc_return(&trans_id);
959 version_req.version.version = DYNMEM_PROTOCOL_VERSION_WIN8;
960 version_req.is_last_attempt = 0;
961
962 ret = vmbus_sendpacket(dev->channel, &version_req,
963 sizeof(struct dm_version_request),
964 (unsigned long)NULL,
965 VM_PKT_DATA_INBAND,
966 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
967 if (ret)
K. Y. Srinivasan33080c12012-12-11 11:07:17 -0800968 goto probe_error2;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800969
970 t = wait_for_completion_timeout(&dm_device.host_event, 5*HZ);
971 if (t == 0) {
972 ret = -ETIMEDOUT;
K. Y. Srinivasan33080c12012-12-11 11:07:17 -0800973 goto probe_error2;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800974 }
975
976 /*
977 * If we could not negotiate a compatible version with the host
978 * fail the probe function.
979 */
980 if (dm_device.state == DM_INIT_ERROR) {
981 ret = -ETIMEDOUT;
K. Y. Srinivasan33080c12012-12-11 11:07:17 -0800982 goto probe_error2;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -0800983 }
984 /*
985 * Now submit our capabilities to the host.
986 */
987 memset(&cap_msg, 0, sizeof(struct dm_capabilities));
988 cap_msg.hdr.type = DM_CAPABILITIES_REPORT;
989 cap_msg.hdr.size = sizeof(struct dm_capabilities);
990 cap_msg.hdr.trans_id = atomic_inc_return(&trans_id);
991
992 cap_msg.caps.cap_bits.balloon = 1;
993 /*
994 * While we currently don't support hot-add,
995 * we still advertise this capability since the
996 * host requires that guests partcipating in the
997 * dynamic memory protocol support hot add.
998 */
999 cap_msg.caps.cap_bits.hot_add = 1;
1000
1001 /*
1002 * Currently the host does not use these
1003 * values and we set them to what is done in the
1004 * Windows driver.
1005 */
1006 cap_msg.min_page_cnt = 0;
1007 cap_msg.max_page_number = -1;
1008
1009 ret = vmbus_sendpacket(dev->channel, &cap_msg,
1010 sizeof(struct dm_capabilities),
1011 (unsigned long)NULL,
1012 VM_PKT_DATA_INBAND,
1013 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1014 if (ret)
K. Y. Srinivasan33080c12012-12-11 11:07:17 -08001015 goto probe_error2;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001016
1017 t = wait_for_completion_timeout(&dm_device.host_event, 5*HZ);
1018 if (t == 0) {
1019 ret = -ETIMEDOUT;
K. Y. Srinivasan33080c12012-12-11 11:07:17 -08001020 goto probe_error2;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001021 }
1022
1023 /*
1024 * If the host does not like our capabilities,
1025 * fail the probe function.
1026 */
1027 if (dm_device.state == DM_INIT_ERROR) {
1028 ret = -ETIMEDOUT;
K. Y. Srinivasan33080c12012-12-11 11:07:17 -08001029 goto probe_error2;
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001030 }
1031
1032 dm_device.state = DM_INITIALIZED;
1033
1034 return 0;
1035
K. Y. Srinivasan33080c12012-12-11 11:07:17 -08001036probe_error2:
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001037 kthread_stop(dm_device.thread);
1038
K. Y. Srinivasan33080c12012-12-11 11:07:17 -08001039probe_error1:
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001040 vmbus_close(dev->channel);
K. Y. Srinivasan33080c12012-12-11 11:07:17 -08001041probe_error0:
1042 kfree(send_buffer);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001043 return ret;
1044}
1045
1046static int balloon_remove(struct hv_device *dev)
1047{
1048 struct hv_dynmem_device *dm = hv_get_drvdata(dev);
1049
1050 if (dm->num_pages_ballooned != 0)
1051 pr_warn("Ballooned pages: %d\n", dm->num_pages_ballooned);
1052
1053 vmbus_close(dev->channel);
1054 kthread_stop(dm->thread);
K. Y. Srinivasan33080c12012-12-11 11:07:17 -08001055 kfree(send_buffer);
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001056
1057 return 0;
1058}
1059
1060static const struct hv_vmbus_device_id id_table[] = {
1061 /* Dynamic Memory Class ID */
1062 /* 525074DC-8985-46e2-8057-A307DC18A502 */
K. Y. Srinivasand13984e2013-01-23 17:42:41 -08001063 { HV_DM_GUID, },
K. Y. Srinivasan9aa8b502012-11-14 01:09:02 -08001064 { },
1065};
1066
1067MODULE_DEVICE_TABLE(vmbus, id_table);
1068
1069static struct hv_driver balloon_drv = {
1070 .name = "hv_balloon",
1071 .id_table = id_table,
1072 .probe = balloon_probe,
1073 .remove = balloon_remove,
1074};
1075
1076static int __init init_balloon_drv(void)
1077{
1078
1079 return vmbus_driver_register(&balloon_drv);
1080}
1081
1082static void exit_balloon_drv(void)
1083{
1084
1085 vmbus_driver_unregister(&balloon_drv);
1086}
1087
1088module_init(init_balloon_drv);
1089module_exit(exit_balloon_drv);
1090
1091MODULE_DESCRIPTION("Hyper-V Balloon");
1092MODULE_VERSION(HV_DRV_VERSION);
1093MODULE_LICENSE("GPL");