Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (c) 2010, Microsoft Corporation.
4 *
5 * Authors:
6 * Haiyang Zhang <haiyangz@microsoft.com>
7 * Hank Janssen <hjanssen@microsoft.com>
8 */
9#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10
11#include <linux/kernel.h>
12#include <linux/init.h>
13#include <linux/module.h>
14#include <linux/slab.h>
15#include <linux/sysctl.h>
16#include <linux/reboot.h>
17#include <linux/hyperv.h>
18#include <linux/clockchips.h>
19#include <linux/ptp_clock_kernel.h>
20#include <asm/mshyperv.h>
21
22#include "hyperv_vmbus.h"
23
24#define SD_MAJOR 3
25#define SD_MINOR 0
26#define SD_MINOR_1 1
27#define SD_MINOR_2 2
28#define SD_VERSION_3_1 (SD_MAJOR << 16 | SD_MINOR_1)
29#define SD_VERSION_3_2 (SD_MAJOR << 16 | SD_MINOR_2)
30#define SD_VERSION (SD_MAJOR << 16 | SD_MINOR)
31
32#define SD_MAJOR_1 1
33#define SD_VERSION_1 (SD_MAJOR_1 << 16 | SD_MINOR)
34
35#define TS_MAJOR 4
36#define TS_MINOR 0
37#define TS_VERSION (TS_MAJOR << 16 | TS_MINOR)
38
39#define TS_MAJOR_1 1
40#define TS_VERSION_1 (TS_MAJOR_1 << 16 | TS_MINOR)
41
42#define TS_MAJOR_3 3
43#define TS_VERSION_3 (TS_MAJOR_3 << 16 | TS_MINOR)
44
45#define HB_MAJOR 3
46#define HB_MINOR 0
47#define HB_VERSION (HB_MAJOR << 16 | HB_MINOR)
48
49#define HB_MAJOR_1 1
50#define HB_VERSION_1 (HB_MAJOR_1 << 16 | HB_MINOR)
51
52static int sd_srv_version;
53static int ts_srv_version;
54static int hb_srv_version;
55
56#define SD_VER_COUNT 4
57static const int sd_versions[] = {
58 SD_VERSION_3_2,
59 SD_VERSION_3_1,
60 SD_VERSION,
61 SD_VERSION_1
62};
63
64#define TS_VER_COUNT 3
65static const int ts_versions[] = {
66 TS_VERSION,
67 TS_VERSION_3,
68 TS_VERSION_1
69};
70
71#define HB_VER_COUNT 2
72static const int hb_versions[] = {
73 HB_VERSION,
74 HB_VERSION_1
75};
76
77#define FW_VER_COUNT 2
78static const int fw_versions[] = {
79 UTIL_FW_VERSION,
80 UTIL_WS2K8_FW_VERSION
81};
82
83/*
84 * Send the "hibernate" udev event in a thread context.
85 */
86struct hibernate_work_context {
87 struct work_struct work;
88 struct hv_device *dev;
89};
90
91static struct hibernate_work_context hibernate_context;
92static bool hibernation_supported;
93
94static void send_hibernate_uevent(struct work_struct *work)
95{
96 char *uevent_env[2] = { "EVENT=hibernate", NULL };
97 struct hibernate_work_context *ctx;
98
99 ctx = container_of(work, struct hibernate_work_context, work);
100
101 kobject_uevent_env(&ctx->dev->device.kobj, KOBJ_CHANGE, uevent_env);
102
103 pr_info("Sent hibernation uevent\n");
104}
105
106static int hv_shutdown_init(struct hv_util_service *srv)
107{
108 struct vmbus_channel *channel = srv->channel;
109
110 INIT_WORK(&hibernate_context.work, send_hibernate_uevent);
111 hibernate_context.dev = channel->device_obj;
112
113 hibernation_supported = hv_is_hibernation_supported();
114
115 return 0;
116}
117
118static void shutdown_onchannelcallback(void *context);
119static struct hv_util_service util_shutdown = {
120 .util_cb = shutdown_onchannelcallback,
121 .util_init = hv_shutdown_init,
122};
123
124static int hv_timesync_init(struct hv_util_service *srv);
125static int hv_timesync_pre_suspend(void);
126static void hv_timesync_deinit(void);
127
128static void timesync_onchannelcallback(void *context);
129static struct hv_util_service util_timesynch = {
130 .util_cb = timesync_onchannelcallback,
131 .util_init = hv_timesync_init,
132 .util_pre_suspend = hv_timesync_pre_suspend,
133 .util_deinit = hv_timesync_deinit,
134};
135
136static void heartbeat_onchannelcallback(void *context);
137static struct hv_util_service util_heartbeat = {
138 .util_cb = heartbeat_onchannelcallback,
139};
140
141static struct hv_util_service util_kvp = {
142 .util_cb = hv_kvp_onchannelcallback,
143 .util_init = hv_kvp_init,
144 .util_pre_suspend = hv_kvp_pre_suspend,
145 .util_pre_resume = hv_kvp_pre_resume,
146 .util_deinit = hv_kvp_deinit,
147};
148
149static struct hv_util_service util_vss = {
150 .util_cb = hv_vss_onchannelcallback,
151 .util_init = hv_vss_init,
152 .util_pre_suspend = hv_vss_pre_suspend,
153 .util_pre_resume = hv_vss_pre_resume,
154 .util_deinit = hv_vss_deinit,
155};
156
157static struct hv_util_service util_fcopy = {
158 .util_cb = hv_fcopy_onchannelcallback,
159 .util_init = hv_fcopy_init,
160 .util_pre_suspend = hv_fcopy_pre_suspend,
161 .util_pre_resume = hv_fcopy_pre_resume,
162 .util_deinit = hv_fcopy_deinit,
163};
164
165static void perform_shutdown(struct work_struct *dummy)
166{
167 orderly_poweroff(true);
168}
169
170static void perform_restart(struct work_struct *dummy)
171{
172 orderly_reboot();
173}
174
175/*
176 * Perform the shutdown operation in a thread context.
177 */
178static DECLARE_WORK(shutdown_work, perform_shutdown);
179
180/*
181 * Perform the restart operation in a thread context.
182 */
183static DECLARE_WORK(restart_work, perform_restart);
184
185static void shutdown_onchannelcallback(void *context)
186{
187 struct vmbus_channel *channel = context;
188 struct work_struct *work = NULL;
189 u32 recvlen;
190 u64 requestid;
191 u8 *shut_txf_buf = util_shutdown.recv_buffer;
192
193 struct shutdown_msg_data *shutdown_msg;
194
195 struct icmsg_hdr *icmsghdrp;
196
197 if (vmbus_recvpacket(channel, shut_txf_buf, HV_HYP_PAGE_SIZE, &recvlen, &requestid)) {
198 pr_err_ratelimited("Shutdown request received. Could not read into shut txf buf\n");
199 return;
200 }
201
202 if (!recvlen)
203 return;
204
205 /* Ensure recvlen is big enough to read header data */
206 if (recvlen < ICMSG_HDR) {
207 pr_err_ratelimited("Shutdown request received. Packet length too small: %d\n",
208 recvlen);
209 return;
210 }
211
212 icmsghdrp = (struct icmsg_hdr *)&shut_txf_buf[sizeof(struct vmbuspipe_hdr)];
213
214 if (icmsghdrp->icmsgtype == ICMSGTYPE_NEGOTIATE) {
215 if (vmbus_prep_negotiate_resp(icmsghdrp,
216 shut_txf_buf, recvlen,
217 fw_versions, FW_VER_COUNT,
218 sd_versions, SD_VER_COUNT,
219 NULL, &sd_srv_version)) {
220 pr_info("Shutdown IC version %d.%d\n",
221 sd_srv_version >> 16,
222 sd_srv_version & 0xFFFF);
223 }
224 } else if (icmsghdrp->icmsgtype == ICMSGTYPE_SHUTDOWN) {
225 /* Ensure recvlen is big enough to contain shutdown_msg_data struct */
226 if (recvlen < ICMSG_HDR + sizeof(struct shutdown_msg_data)) {
227 pr_err_ratelimited("Invalid shutdown msg data. Packet length too small: %u\n",
228 recvlen);
229 return;
230 }
231
232 shutdown_msg = (struct shutdown_msg_data *)&shut_txf_buf[ICMSG_HDR];
233
234 /*
235 * shutdown_msg->flags can be 0(shut down), 2(reboot),
236 * or 4(hibernate). It may bitwise-OR 1, which means
237 * performing the request by force. Linux always tries
238 * to perform the request by force.
239 */
240 switch (shutdown_msg->flags) {
241 case 0:
242 case 1:
243 icmsghdrp->status = HV_S_OK;
244 work = &shutdown_work;
245 pr_info("Shutdown request received - graceful shutdown initiated\n");
246 break;
247 case 2:
248 case 3:
249 icmsghdrp->status = HV_S_OK;
250 work = &restart_work;
251 pr_info("Restart request received - graceful restart initiated\n");
252 break;
253 case 4:
254 case 5:
255 pr_info("Hibernation request received\n");
256 icmsghdrp->status = hibernation_supported ?
257 HV_S_OK : HV_E_FAIL;
258 if (hibernation_supported)
259 work = &hibernate_context.work;
260 break;
261 default:
262 icmsghdrp->status = HV_E_FAIL;
263 pr_info("Shutdown request received - Invalid request\n");
264 break;
265 }
266 } else {
267 icmsghdrp->status = HV_E_FAIL;
268 pr_err_ratelimited("Shutdown request received. Invalid msg type: %d\n",
269 icmsghdrp->icmsgtype);
270 }
271
272 icmsghdrp->icflags = ICMSGHDRFLAG_TRANSACTION
273 | ICMSGHDRFLAG_RESPONSE;
274
275 vmbus_sendpacket(channel, shut_txf_buf,
276 recvlen, requestid,
277 VM_PKT_DATA_INBAND, 0);
278
279 if (work)
280 schedule_work(work);
281}
282
283/*
284 * Set the host time in a process context.
285 */
286static struct work_struct adj_time_work;
287
288/*
289 * The last time sample, received from the host. PTP device responds to
290 * requests by using this data and the current partition-wide time reference
291 * count.
292 */
293static struct {
294 u64 host_time;
295 u64 ref_time;
296 spinlock_t lock;
297} host_ts;
298
299static bool timesync_implicit;
300
301module_param(timesync_implicit, bool, 0644);
302MODULE_PARM_DESC(timesync_implicit, "If set treat SAMPLE as SYNC when clock is behind");
303
304static inline u64 reftime_to_ns(u64 reftime)
305{
306 return (reftime - WLTIMEDELTA) * 100;
307}
308
309/*
310 * Hard coded threshold for host timesync delay: 600 seconds
311 */
312static const u64 HOST_TIMESYNC_DELAY_THRESH = 600 * (u64)NSEC_PER_SEC;
313
314static int hv_get_adj_host_time(struct timespec64 *ts)
315{
316 u64 newtime, reftime, timediff_adj;
317 unsigned long flags;
318 int ret = 0;
319
320 spin_lock_irqsave(&host_ts.lock, flags);
321 reftime = hv_read_reference_counter();
322
323 /*
324 * We need to let the caller know that last update from host
325 * is older than the max allowable threshold. clock_gettime()
326 * and PTP ioctl do not have a documented error that we could
327 * return for this specific case. Use ESTALE to report this.
328 */
329 timediff_adj = reftime - host_ts.ref_time;
330 if (timediff_adj * 100 > HOST_TIMESYNC_DELAY_THRESH) {
331 pr_warn_once("TIMESYNC IC: Stale time stamp, %llu nsecs old\n",
332 (timediff_adj * 100));
333 ret = -ESTALE;
334 }
335
336 newtime = host_ts.host_time + timediff_adj;
337 *ts = ns_to_timespec64(reftime_to_ns(newtime));
338 spin_unlock_irqrestore(&host_ts.lock, flags);
339
340 return ret;
341}
342
343static void hv_set_host_time(struct work_struct *work)
344{
345
346 struct timespec64 ts;
347
348 if (!hv_get_adj_host_time(&ts))
349 do_settimeofday64(&ts);
350}
351
352/*
353 * Due to a bug on Hyper-V hosts, the sync flag may not always be sent on resume.
354 * Force a sync if the guest is behind.
355 */
356static inline bool hv_implicit_sync(u64 host_time)
357{
358 struct timespec64 new_ts;
359 struct timespec64 threshold_ts;
360
361 new_ts = ns_to_timespec64(reftime_to_ns(host_time));
362 ktime_get_real_ts64(&threshold_ts);
363
364 threshold_ts.tv_sec += 5;
365
366 /*
367 * If guest behind the host by 5 or more seconds.
368 */
369 if (timespec64_compare(&new_ts, &threshold_ts) >= 0)
370 return true;
371
372 return false;
373}
374
375/*
376 * Synchronize time with host after reboot, restore, etc.
377 *
378 * ICTIMESYNCFLAG_SYNC flag bit indicates reboot, restore events of the VM.
379 * After reboot the flag ICTIMESYNCFLAG_SYNC is included in the first time
380 * message after the timesync channel is opened. Since the hv_utils module is
381 * loaded after hv_vmbus, the first message is usually missed. This bit is
382 * considered a hard request to discipline the clock.
383 *
384 * ICTIMESYNCFLAG_SAMPLE bit indicates a time sample from host. This is
385 * typically used as a hint to the guest. The guest is under no obligation
386 * to discipline the clock.
387 */
388static inline void adj_guesttime(u64 hosttime, u64 reftime, u8 adj_flags)
389{
390 unsigned long flags;
391 u64 cur_reftime;
392
393 /*
394 * Save the adjusted time sample from the host and the snapshot
395 * of the current system time.
396 */
397 spin_lock_irqsave(&host_ts.lock, flags);
398
399 cur_reftime = hv_read_reference_counter();
400 host_ts.host_time = hosttime;
401 host_ts.ref_time = cur_reftime;
402
403 /*
404 * TimeSync v4 messages contain reference time (guest's Hyper-V
405 * clocksource read when the time sample was generated), we can
406 * improve the precision by adding the delta between now and the
407 * time of generation. For older protocols we set
408 * reftime == cur_reftime on call.
409 */
410 host_ts.host_time += (cur_reftime - reftime);
411
412 spin_unlock_irqrestore(&host_ts.lock, flags);
413
414 /* Schedule work to do do_settimeofday64() */
415 if ((adj_flags & ICTIMESYNCFLAG_SYNC) ||
416 (timesync_implicit && hv_implicit_sync(host_ts.host_time)))
417 schedule_work(&adj_time_work);
418}
419
420/*
421 * Time Sync Channel message handler.
422 */
423static void timesync_onchannelcallback(void *context)
424{
425 struct vmbus_channel *channel = context;
426 u32 recvlen;
427 u64 requestid;
428 struct icmsg_hdr *icmsghdrp;
429 struct ictimesync_data *timedatap;
430 struct ictimesync_ref_data *refdata;
431 u8 *time_txf_buf = util_timesynch.recv_buffer;
432
433 /*
434 * Drain the ring buffer and use the last packet to update
435 * host_ts
436 */
437 while (1) {
438 int ret = vmbus_recvpacket(channel, time_txf_buf,
439 HV_HYP_PAGE_SIZE, &recvlen,
440 &requestid);
441 if (ret) {
442 pr_err_ratelimited("TimeSync IC pkt recv failed (Err: %d)\n",
443 ret);
444 break;
445 }
446
447 if (!recvlen)
448 break;
449
450 /* Ensure recvlen is big enough to read header data */
451 if (recvlen < ICMSG_HDR) {
452 pr_err_ratelimited("Timesync request received. Packet length too small: %d\n",
453 recvlen);
454 break;
455 }
456
457 icmsghdrp = (struct icmsg_hdr *)&time_txf_buf[
458 sizeof(struct vmbuspipe_hdr)];
459
460 if (icmsghdrp->icmsgtype == ICMSGTYPE_NEGOTIATE) {
461 if (vmbus_prep_negotiate_resp(icmsghdrp,
462 time_txf_buf, recvlen,
463 fw_versions, FW_VER_COUNT,
464 ts_versions, TS_VER_COUNT,
465 NULL, &ts_srv_version)) {
466 pr_info("TimeSync IC version %d.%d\n",
467 ts_srv_version >> 16,
468 ts_srv_version & 0xFFFF);
469 }
470 } else if (icmsghdrp->icmsgtype == ICMSGTYPE_TIMESYNC) {
471 if (ts_srv_version > TS_VERSION_3) {
472 /* Ensure recvlen is big enough to read ictimesync_ref_data */
473 if (recvlen < ICMSG_HDR + sizeof(struct ictimesync_ref_data)) {
474 pr_err_ratelimited("Invalid ictimesync ref data. Length too small: %u\n",
475 recvlen);
476 break;
477 }
478 refdata = (struct ictimesync_ref_data *)&time_txf_buf[ICMSG_HDR];
479
480 adj_guesttime(refdata->parenttime,
481 refdata->vmreferencetime,
482 refdata->flags);
483 } else {
484 /* Ensure recvlen is big enough to read ictimesync_data */
485 if (recvlen < ICMSG_HDR + sizeof(struct ictimesync_data)) {
486 pr_err_ratelimited("Invalid ictimesync data. Length too small: %u\n",
487 recvlen);
488 break;
489 }
490 timedatap = (struct ictimesync_data *)&time_txf_buf[ICMSG_HDR];
491
492 adj_guesttime(timedatap->parenttime,
493 hv_read_reference_counter(),
494 timedatap->flags);
495 }
496 } else {
497 icmsghdrp->status = HV_E_FAIL;
498 pr_err_ratelimited("Timesync request received. Invalid msg type: %d\n",
499 icmsghdrp->icmsgtype);
500 }
501
502 icmsghdrp->icflags = ICMSGHDRFLAG_TRANSACTION
503 | ICMSGHDRFLAG_RESPONSE;
504
505 vmbus_sendpacket(channel, time_txf_buf,
506 recvlen, requestid,
507 VM_PKT_DATA_INBAND, 0);
508 }
509}
510
511/*
512 * Heartbeat functionality.
513 * Every two seconds, Hyper-V send us a heartbeat request message.
514 * we respond to this message, and Hyper-V knows we are alive.
515 */
516static void heartbeat_onchannelcallback(void *context)
517{
518 struct vmbus_channel *channel = context;
519 u32 recvlen;
520 u64 requestid;
521 struct icmsg_hdr *icmsghdrp;
522 struct heartbeat_msg_data *heartbeat_msg;
523 u8 *hbeat_txf_buf = util_heartbeat.recv_buffer;
524
525 while (1) {
526
527 if (vmbus_recvpacket(channel, hbeat_txf_buf, HV_HYP_PAGE_SIZE,
528 &recvlen, &requestid)) {
529 pr_err_ratelimited("Heartbeat request received. Could not read into hbeat txf buf\n");
530 return;
531 }
532
533 if (!recvlen)
534 break;
535
536 /* Ensure recvlen is big enough to read header data */
537 if (recvlen < ICMSG_HDR) {
538 pr_err_ratelimited("Heartbeat request received. Packet length too small: %d\n",
539 recvlen);
540 break;
541 }
542
543 icmsghdrp = (struct icmsg_hdr *)&hbeat_txf_buf[
544 sizeof(struct vmbuspipe_hdr)];
545
546 if (icmsghdrp->icmsgtype == ICMSGTYPE_NEGOTIATE) {
547 if (vmbus_prep_negotiate_resp(icmsghdrp,
548 hbeat_txf_buf, recvlen,
549 fw_versions, FW_VER_COUNT,
550 hb_versions, HB_VER_COUNT,
551 NULL, &hb_srv_version)) {
552
553 pr_info("Heartbeat IC version %d.%d\n",
554 hb_srv_version >> 16,
555 hb_srv_version & 0xFFFF);
556 }
557 } else if (icmsghdrp->icmsgtype == ICMSGTYPE_HEARTBEAT) {
558 /*
559 * Ensure recvlen is big enough to read seq_num. Reserved area is not
560 * included in the check as the host may not fill it up entirely
561 */
562 if (recvlen < ICMSG_HDR + sizeof(u64)) {
563 pr_err_ratelimited("Invalid heartbeat msg data. Length too small: %u\n",
564 recvlen);
565 break;
566 }
567 heartbeat_msg = (struct heartbeat_msg_data *)&hbeat_txf_buf[ICMSG_HDR];
568
569 heartbeat_msg->seq_num += 1;
570 } else {
571 icmsghdrp->status = HV_E_FAIL;
572 pr_err_ratelimited("Heartbeat request received. Invalid msg type: %d\n",
573 icmsghdrp->icmsgtype);
574 }
575
576 icmsghdrp->icflags = ICMSGHDRFLAG_TRANSACTION
577 | ICMSGHDRFLAG_RESPONSE;
578
579 vmbus_sendpacket(channel, hbeat_txf_buf,
580 recvlen, requestid,
581 VM_PKT_DATA_INBAND, 0);
582 }
583}
584
585#define HV_UTIL_RING_SEND_SIZE VMBUS_RING_SIZE(3 * HV_HYP_PAGE_SIZE)
586#define HV_UTIL_RING_RECV_SIZE VMBUS_RING_SIZE(3 * HV_HYP_PAGE_SIZE)
587
588static int util_probe(struct hv_device *dev,
589 const struct hv_vmbus_device_id *dev_id)
590{
591 struct hv_util_service *srv =
592 (struct hv_util_service *)dev_id->driver_data;
593 int ret;
594
595 srv->recv_buffer = kmalloc(HV_HYP_PAGE_SIZE * 4, GFP_KERNEL);
596 if (!srv->recv_buffer)
597 return -ENOMEM;
598 srv->channel = dev->channel;
599 if (srv->util_init) {
600 ret = srv->util_init(srv);
601 if (ret) {
602 ret = -ENODEV;
603 goto error1;
604 }
605 }
606
607 /*
608 * The set of services managed by the util driver are not performance
609 * critical and do not need batched reading. Furthermore, some services
610 * such as KVP can only handle one message from the host at a time.
611 * Turn off batched reading for all util drivers before we open the
612 * channel.
613 */
614 set_channel_read_mode(dev->channel, HV_CALL_DIRECT);
615
616 hv_set_drvdata(dev, srv);
617
618 ret = vmbus_open(dev->channel, HV_UTIL_RING_SEND_SIZE,
619 HV_UTIL_RING_RECV_SIZE, NULL, 0, srv->util_cb,
620 dev->channel);
621 if (ret)
622 goto error;
623
624 return 0;
625
626error:
627 if (srv->util_deinit)
628 srv->util_deinit();
629error1:
630 kfree(srv->recv_buffer);
631 return ret;
632}
633
634static void util_remove(struct hv_device *dev)
635{
636 struct hv_util_service *srv = hv_get_drvdata(dev);
637
638 if (srv->util_deinit)
639 srv->util_deinit();
640 vmbus_close(dev->channel);
641 kfree(srv->recv_buffer);
642}
643
644/*
645 * When we're in util_suspend(), all the userspace processes have been frozen
646 * (refer to hibernate() -> freeze_processes()). The userspace is thawed only
647 * after the whole resume procedure, including util_resume(), finishes.
648 */
649static int util_suspend(struct hv_device *dev)
650{
651 struct hv_util_service *srv = hv_get_drvdata(dev);
652 int ret = 0;
653
654 if (srv->util_pre_suspend) {
655 ret = srv->util_pre_suspend();
656 if (ret)
657 return ret;
658 }
659
660 vmbus_close(dev->channel);
661
662 return 0;
663}
664
665static int util_resume(struct hv_device *dev)
666{
667 struct hv_util_service *srv = hv_get_drvdata(dev);
668 int ret = 0;
669
670 if (srv->util_pre_resume) {
671 ret = srv->util_pre_resume();
672 if (ret)
673 return ret;
674 }
675
676 ret = vmbus_open(dev->channel, HV_UTIL_RING_SEND_SIZE,
677 HV_UTIL_RING_RECV_SIZE, NULL, 0, srv->util_cb,
678 dev->channel);
679 return ret;
680}
681
682static const struct hv_vmbus_device_id id_table[] = {
683 /* Shutdown guid */
684 { HV_SHUTDOWN_GUID,
685 .driver_data = (unsigned long)&util_shutdown
686 },
687 /* Time synch guid */
688 { HV_TS_GUID,
689 .driver_data = (unsigned long)&util_timesynch
690 },
691 /* Heartbeat guid */
692 { HV_HEART_BEAT_GUID,
693 .driver_data = (unsigned long)&util_heartbeat
694 },
695 /* KVP guid */
696 { HV_KVP_GUID,
697 .driver_data = (unsigned long)&util_kvp
698 },
699 /* VSS GUID */
700 { HV_VSS_GUID,
701 .driver_data = (unsigned long)&util_vss
702 },
703 /* File copy GUID */
704 { HV_FCOPY_GUID,
705 .driver_data = (unsigned long)&util_fcopy
706 },
707 { },
708};
709
710MODULE_DEVICE_TABLE(vmbus, id_table);
711
712/* The one and only one */
713static struct hv_driver util_drv = {
714 .name = "hv_utils",
715 .id_table = id_table,
716 .probe = util_probe,
717 .remove = util_remove,
718 .suspend = util_suspend,
719 .resume = util_resume,
720 .driver = {
721 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
722 },
723};
724
725static int hv_ptp_enable(struct ptp_clock_info *info,
726 struct ptp_clock_request *request, int on)
727{
728 return -EOPNOTSUPP;
729}
730
731static int hv_ptp_settime(struct ptp_clock_info *p, const struct timespec64 *ts)
732{
733 return -EOPNOTSUPP;
734}
735
736static int hv_ptp_adjfine(struct ptp_clock_info *ptp, long delta)
737{
738 return -EOPNOTSUPP;
739}
740static int hv_ptp_adjtime(struct ptp_clock_info *ptp, s64 delta)
741{
742 return -EOPNOTSUPP;
743}
744
745static int hv_ptp_gettime(struct ptp_clock_info *info, struct timespec64 *ts)
746{
747 return hv_get_adj_host_time(ts);
748}
749
750static struct ptp_clock_info ptp_hyperv_info = {
751 .name = "hyperv",
752 .enable = hv_ptp_enable,
753 .adjtime = hv_ptp_adjtime,
754 .adjfine = hv_ptp_adjfine,
755 .gettime64 = hv_ptp_gettime,
756 .settime64 = hv_ptp_settime,
757 .owner = THIS_MODULE,
758};
759
760static struct ptp_clock *hv_ptp_clock;
761
762static int hv_timesync_init(struct hv_util_service *srv)
763{
764 spin_lock_init(&host_ts.lock);
765
766 INIT_WORK(&adj_time_work, hv_set_host_time);
767
768 /*
769 * ptp_clock_register() returns NULL when CONFIG_PTP_1588_CLOCK is
770 * disabled but the driver is still useful without the PTP device
771 * as it still handles the ICTIMESYNCFLAG_SYNC case.
772 */
773 hv_ptp_clock = ptp_clock_register(&ptp_hyperv_info, NULL);
774 if (IS_ERR_OR_NULL(hv_ptp_clock)) {
775 pr_err("cannot register PTP clock: %d\n",
776 PTR_ERR_OR_ZERO(hv_ptp_clock));
777 hv_ptp_clock = NULL;
778 }
779
780 return 0;
781}
782
783static void hv_timesync_cancel_work(void)
784{
785 cancel_work_sync(&adj_time_work);
786}
787
788static int hv_timesync_pre_suspend(void)
789{
790 hv_timesync_cancel_work();
791 return 0;
792}
793
794static void hv_timesync_deinit(void)
795{
796 if (hv_ptp_clock)
797 ptp_clock_unregister(hv_ptp_clock);
798
799 hv_timesync_cancel_work();
800}
801
802static int __init init_hyperv_utils(void)
803{
804 pr_info("Registering HyperV Utility Driver\n");
805
806 return vmbus_driver_register(&util_drv);
807}
808
809static void exit_hyperv_utils(void)
810{
811 pr_info("De-Registered HyperV Utility Driver\n");
812
813 vmbus_driver_unregister(&util_drv);
814}
815
816module_init(init_hyperv_utils);
817module_exit(exit_hyperv_utils);
818
819MODULE_DESCRIPTION("Hyper-V Utilities");
820MODULE_LICENSE("GPL");
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (c) 2010, Microsoft Corporation.
4 *
5 * Authors:
6 * Haiyang Zhang <haiyangz@microsoft.com>
7 * Hank Janssen <hjanssen@microsoft.com>
8 */
9#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10
11#include <linux/kernel.h>
12#include <linux/init.h>
13#include <linux/module.h>
14#include <linux/slab.h>
15#include <linux/sysctl.h>
16#include <linux/reboot.h>
17#include <linux/hyperv.h>
18#include <linux/clockchips.h>
19#include <linux/ptp_clock_kernel.h>
20#include <clocksource/hyperv_timer.h>
21#include <asm/mshyperv.h>
22
23#include "hyperv_vmbus.h"
24
25#define SD_MAJOR 3
26#define SD_MINOR 0
27#define SD_MINOR_1 1
28#define SD_MINOR_2 2
29#define SD_VERSION_3_1 (SD_MAJOR << 16 | SD_MINOR_1)
30#define SD_VERSION_3_2 (SD_MAJOR << 16 | SD_MINOR_2)
31#define SD_VERSION (SD_MAJOR << 16 | SD_MINOR)
32
33#define SD_MAJOR_1 1
34#define SD_VERSION_1 (SD_MAJOR_1 << 16 | SD_MINOR)
35
36#define TS_MAJOR 4
37#define TS_MINOR 0
38#define TS_VERSION (TS_MAJOR << 16 | TS_MINOR)
39
40#define TS_MAJOR_1 1
41#define TS_VERSION_1 (TS_MAJOR_1 << 16 | TS_MINOR)
42
43#define TS_MAJOR_3 3
44#define TS_VERSION_3 (TS_MAJOR_3 << 16 | TS_MINOR)
45
46#define HB_MAJOR 3
47#define HB_MINOR 0
48#define HB_VERSION (HB_MAJOR << 16 | HB_MINOR)
49
50#define HB_MAJOR_1 1
51#define HB_VERSION_1 (HB_MAJOR_1 << 16 | HB_MINOR)
52
53static int sd_srv_version;
54static int ts_srv_version;
55static int hb_srv_version;
56
57#define SD_VER_COUNT 4
58static const int sd_versions[] = {
59 SD_VERSION_3_2,
60 SD_VERSION_3_1,
61 SD_VERSION,
62 SD_VERSION_1
63};
64
65#define TS_VER_COUNT 3
66static const int ts_versions[] = {
67 TS_VERSION,
68 TS_VERSION_3,
69 TS_VERSION_1
70};
71
72#define HB_VER_COUNT 2
73static const int hb_versions[] = {
74 HB_VERSION,
75 HB_VERSION_1
76};
77
78#define FW_VER_COUNT 2
79static const int fw_versions[] = {
80 UTIL_FW_VERSION,
81 UTIL_WS2K8_FW_VERSION
82};
83
84/*
85 * Send the "hibernate" udev event in a thread context.
86 */
87struct hibernate_work_context {
88 struct work_struct work;
89 struct hv_device *dev;
90};
91
92static struct hibernate_work_context hibernate_context;
93static bool hibernation_supported;
94
95static void send_hibernate_uevent(struct work_struct *work)
96{
97 char *uevent_env[2] = { "EVENT=hibernate", NULL };
98 struct hibernate_work_context *ctx;
99
100 ctx = container_of(work, struct hibernate_work_context, work);
101
102 kobject_uevent_env(&ctx->dev->device.kobj, KOBJ_CHANGE, uevent_env);
103
104 pr_info("Sent hibernation uevent\n");
105}
106
107static int hv_shutdown_init(struct hv_util_service *srv)
108{
109 struct vmbus_channel *channel = srv->channel;
110
111 INIT_WORK(&hibernate_context.work, send_hibernate_uevent);
112 hibernate_context.dev = channel->device_obj;
113
114 hibernation_supported = hv_is_hibernation_supported();
115
116 return 0;
117}
118
119static void shutdown_onchannelcallback(void *context);
120static struct hv_util_service util_shutdown = {
121 .util_cb = shutdown_onchannelcallback,
122 .util_init = hv_shutdown_init,
123};
124
125static int hv_timesync_init(struct hv_util_service *srv);
126static int hv_timesync_pre_suspend(void);
127static void hv_timesync_deinit(void);
128
129static void timesync_onchannelcallback(void *context);
130static struct hv_util_service util_timesynch = {
131 .util_cb = timesync_onchannelcallback,
132 .util_init = hv_timesync_init,
133 .util_pre_suspend = hv_timesync_pre_suspend,
134 .util_deinit = hv_timesync_deinit,
135};
136
137static void heartbeat_onchannelcallback(void *context);
138static struct hv_util_service util_heartbeat = {
139 .util_cb = heartbeat_onchannelcallback,
140};
141
142static struct hv_util_service util_kvp = {
143 .util_cb = hv_kvp_onchannelcallback,
144 .util_init = hv_kvp_init,
145 .util_pre_suspend = hv_kvp_pre_suspend,
146 .util_pre_resume = hv_kvp_pre_resume,
147 .util_deinit = hv_kvp_deinit,
148};
149
150static struct hv_util_service util_vss = {
151 .util_cb = hv_vss_onchannelcallback,
152 .util_init = hv_vss_init,
153 .util_pre_suspend = hv_vss_pre_suspend,
154 .util_pre_resume = hv_vss_pre_resume,
155 .util_deinit = hv_vss_deinit,
156};
157
158static struct hv_util_service util_fcopy = {
159 .util_cb = hv_fcopy_onchannelcallback,
160 .util_init = hv_fcopy_init,
161 .util_pre_suspend = hv_fcopy_pre_suspend,
162 .util_pre_resume = hv_fcopy_pre_resume,
163 .util_deinit = hv_fcopy_deinit,
164};
165
166static void perform_shutdown(struct work_struct *dummy)
167{
168 orderly_poweroff(true);
169}
170
171static void perform_restart(struct work_struct *dummy)
172{
173 orderly_reboot();
174}
175
176/*
177 * Perform the shutdown operation in a thread context.
178 */
179static DECLARE_WORK(shutdown_work, perform_shutdown);
180
181/*
182 * Perform the restart operation in a thread context.
183 */
184static DECLARE_WORK(restart_work, perform_restart);
185
186static void shutdown_onchannelcallback(void *context)
187{
188 struct vmbus_channel *channel = context;
189 struct work_struct *work = NULL;
190 u32 recvlen;
191 u64 requestid;
192 u8 *shut_txf_buf = util_shutdown.recv_buffer;
193
194 struct shutdown_msg_data *shutdown_msg;
195
196 struct icmsg_hdr *icmsghdrp;
197
198 if (vmbus_recvpacket(channel, shut_txf_buf, HV_HYP_PAGE_SIZE, &recvlen, &requestid)) {
199 pr_err_ratelimited("Shutdown request received. Could not read into shut txf buf\n");
200 return;
201 }
202
203 if (!recvlen)
204 return;
205
206 /* Ensure recvlen is big enough to read header data */
207 if (recvlen < ICMSG_HDR) {
208 pr_err_ratelimited("Shutdown request received. Packet length too small: %d\n",
209 recvlen);
210 return;
211 }
212
213 icmsghdrp = (struct icmsg_hdr *)&shut_txf_buf[sizeof(struct vmbuspipe_hdr)];
214
215 if (icmsghdrp->icmsgtype == ICMSGTYPE_NEGOTIATE) {
216 if (vmbus_prep_negotiate_resp(icmsghdrp,
217 shut_txf_buf, recvlen,
218 fw_versions, FW_VER_COUNT,
219 sd_versions, SD_VER_COUNT,
220 NULL, &sd_srv_version)) {
221 pr_info("Shutdown IC version %d.%d\n",
222 sd_srv_version >> 16,
223 sd_srv_version & 0xFFFF);
224 }
225 } else if (icmsghdrp->icmsgtype == ICMSGTYPE_SHUTDOWN) {
226 /* Ensure recvlen is big enough to contain shutdown_msg_data struct */
227 if (recvlen < ICMSG_HDR + sizeof(struct shutdown_msg_data)) {
228 pr_err_ratelimited("Invalid shutdown msg data. Packet length too small: %u\n",
229 recvlen);
230 return;
231 }
232
233 shutdown_msg = (struct shutdown_msg_data *)&shut_txf_buf[ICMSG_HDR];
234
235 /*
236 * shutdown_msg->flags can be 0(shut down), 2(reboot),
237 * or 4(hibernate). It may bitwise-OR 1, which means
238 * performing the request by force. Linux always tries
239 * to perform the request by force.
240 */
241 switch (shutdown_msg->flags) {
242 case 0:
243 case 1:
244 icmsghdrp->status = HV_S_OK;
245 work = &shutdown_work;
246 pr_info("Shutdown request received - graceful shutdown initiated\n");
247 break;
248 case 2:
249 case 3:
250 icmsghdrp->status = HV_S_OK;
251 work = &restart_work;
252 pr_info("Restart request received - graceful restart initiated\n");
253 break;
254 case 4:
255 case 5:
256 pr_info("Hibernation request received\n");
257 icmsghdrp->status = hibernation_supported ?
258 HV_S_OK : HV_E_FAIL;
259 if (hibernation_supported)
260 work = &hibernate_context.work;
261 break;
262 default:
263 icmsghdrp->status = HV_E_FAIL;
264 pr_info("Shutdown request received - Invalid request\n");
265 break;
266 }
267 } else {
268 icmsghdrp->status = HV_E_FAIL;
269 pr_err_ratelimited("Shutdown request received. Invalid msg type: %d\n",
270 icmsghdrp->icmsgtype);
271 }
272
273 icmsghdrp->icflags = ICMSGHDRFLAG_TRANSACTION
274 | ICMSGHDRFLAG_RESPONSE;
275
276 vmbus_sendpacket(channel, shut_txf_buf,
277 recvlen, requestid,
278 VM_PKT_DATA_INBAND, 0);
279
280 if (work)
281 schedule_work(work);
282}
283
284/*
285 * Set the host time in a process context.
286 */
287static struct work_struct adj_time_work;
288
289/*
290 * The last time sample, received from the host. PTP device responds to
291 * requests by using this data and the current partition-wide time reference
292 * count.
293 */
294static struct {
295 u64 host_time;
296 u64 ref_time;
297 spinlock_t lock;
298} host_ts;
299
300static inline u64 reftime_to_ns(u64 reftime)
301{
302 return (reftime - WLTIMEDELTA) * 100;
303}
304
305/*
306 * Hard coded threshold for host timesync delay: 600 seconds
307 */
308static const u64 HOST_TIMESYNC_DELAY_THRESH = 600 * (u64)NSEC_PER_SEC;
309
310static int hv_get_adj_host_time(struct timespec64 *ts)
311{
312 u64 newtime, reftime, timediff_adj;
313 unsigned long flags;
314 int ret = 0;
315
316 spin_lock_irqsave(&host_ts.lock, flags);
317 reftime = hv_read_reference_counter();
318
319 /*
320 * We need to let the caller know that last update from host
321 * is older than the max allowable threshold. clock_gettime()
322 * and PTP ioctl do not have a documented error that we could
323 * return for this specific case. Use ESTALE to report this.
324 */
325 timediff_adj = reftime - host_ts.ref_time;
326 if (timediff_adj * 100 > HOST_TIMESYNC_DELAY_THRESH) {
327 pr_warn_once("TIMESYNC IC: Stale time stamp, %llu nsecs old\n",
328 (timediff_adj * 100));
329 ret = -ESTALE;
330 }
331
332 newtime = host_ts.host_time + timediff_adj;
333 *ts = ns_to_timespec64(reftime_to_ns(newtime));
334 spin_unlock_irqrestore(&host_ts.lock, flags);
335
336 return ret;
337}
338
339static void hv_set_host_time(struct work_struct *work)
340{
341
342 struct timespec64 ts;
343
344 if (!hv_get_adj_host_time(&ts))
345 do_settimeofday64(&ts);
346}
347
348/*
349 * Synchronize time with host after reboot, restore, etc.
350 *
351 * ICTIMESYNCFLAG_SYNC flag bit indicates reboot, restore events of the VM.
352 * After reboot the flag ICTIMESYNCFLAG_SYNC is included in the first time
353 * message after the timesync channel is opened. Since the hv_utils module is
354 * loaded after hv_vmbus, the first message is usually missed. This bit is
355 * considered a hard request to discipline the clock.
356 *
357 * ICTIMESYNCFLAG_SAMPLE bit indicates a time sample from host. This is
358 * typically used as a hint to the guest. The guest is under no obligation
359 * to discipline the clock.
360 */
361static inline void adj_guesttime(u64 hosttime, u64 reftime, u8 adj_flags)
362{
363 unsigned long flags;
364 u64 cur_reftime;
365
366 /*
367 * Save the adjusted time sample from the host and the snapshot
368 * of the current system time.
369 */
370 spin_lock_irqsave(&host_ts.lock, flags);
371
372 cur_reftime = hv_read_reference_counter();
373 host_ts.host_time = hosttime;
374 host_ts.ref_time = cur_reftime;
375
376 /*
377 * TimeSync v4 messages contain reference time (guest's Hyper-V
378 * clocksource read when the time sample was generated), we can
379 * improve the precision by adding the delta between now and the
380 * time of generation. For older protocols we set
381 * reftime == cur_reftime on call.
382 */
383 host_ts.host_time += (cur_reftime - reftime);
384
385 spin_unlock_irqrestore(&host_ts.lock, flags);
386
387 /* Schedule work to do do_settimeofday64() */
388 if (adj_flags & ICTIMESYNCFLAG_SYNC)
389 schedule_work(&adj_time_work);
390}
391
392/*
393 * Time Sync Channel message handler.
394 */
395static void timesync_onchannelcallback(void *context)
396{
397 struct vmbus_channel *channel = context;
398 u32 recvlen;
399 u64 requestid;
400 struct icmsg_hdr *icmsghdrp;
401 struct ictimesync_data *timedatap;
402 struct ictimesync_ref_data *refdata;
403 u8 *time_txf_buf = util_timesynch.recv_buffer;
404
405 /*
406 * Drain the ring buffer and use the last packet to update
407 * host_ts
408 */
409 while (1) {
410 int ret = vmbus_recvpacket(channel, time_txf_buf,
411 HV_HYP_PAGE_SIZE, &recvlen,
412 &requestid);
413 if (ret) {
414 pr_err_ratelimited("TimeSync IC pkt recv failed (Err: %d)\n",
415 ret);
416 break;
417 }
418
419 if (!recvlen)
420 break;
421
422 /* Ensure recvlen is big enough to read header data */
423 if (recvlen < ICMSG_HDR) {
424 pr_err_ratelimited("Timesync request received. Packet length too small: %d\n",
425 recvlen);
426 break;
427 }
428
429 icmsghdrp = (struct icmsg_hdr *)&time_txf_buf[
430 sizeof(struct vmbuspipe_hdr)];
431
432 if (icmsghdrp->icmsgtype == ICMSGTYPE_NEGOTIATE) {
433 if (vmbus_prep_negotiate_resp(icmsghdrp,
434 time_txf_buf, recvlen,
435 fw_versions, FW_VER_COUNT,
436 ts_versions, TS_VER_COUNT,
437 NULL, &ts_srv_version)) {
438 pr_info("TimeSync IC version %d.%d\n",
439 ts_srv_version >> 16,
440 ts_srv_version & 0xFFFF);
441 }
442 } else if (icmsghdrp->icmsgtype == ICMSGTYPE_TIMESYNC) {
443 if (ts_srv_version > TS_VERSION_3) {
444 /* Ensure recvlen is big enough to read ictimesync_ref_data */
445 if (recvlen < ICMSG_HDR + sizeof(struct ictimesync_ref_data)) {
446 pr_err_ratelimited("Invalid ictimesync ref data. Length too small: %u\n",
447 recvlen);
448 break;
449 }
450 refdata = (struct ictimesync_ref_data *)&time_txf_buf[ICMSG_HDR];
451
452 adj_guesttime(refdata->parenttime,
453 refdata->vmreferencetime,
454 refdata->flags);
455 } else {
456 /* Ensure recvlen is big enough to read ictimesync_data */
457 if (recvlen < ICMSG_HDR + sizeof(struct ictimesync_data)) {
458 pr_err_ratelimited("Invalid ictimesync data. Length too small: %u\n",
459 recvlen);
460 break;
461 }
462 timedatap = (struct ictimesync_data *)&time_txf_buf[ICMSG_HDR];
463
464 adj_guesttime(timedatap->parenttime,
465 hv_read_reference_counter(),
466 timedatap->flags);
467 }
468 } else {
469 icmsghdrp->status = HV_E_FAIL;
470 pr_err_ratelimited("Timesync request received. Invalid msg type: %d\n",
471 icmsghdrp->icmsgtype);
472 }
473
474 icmsghdrp->icflags = ICMSGHDRFLAG_TRANSACTION
475 | ICMSGHDRFLAG_RESPONSE;
476
477 vmbus_sendpacket(channel, time_txf_buf,
478 recvlen, requestid,
479 VM_PKT_DATA_INBAND, 0);
480 }
481}
482
483/*
484 * Heartbeat functionality.
485 * Every two seconds, Hyper-V send us a heartbeat request message.
486 * we respond to this message, and Hyper-V knows we are alive.
487 */
488static void heartbeat_onchannelcallback(void *context)
489{
490 struct vmbus_channel *channel = context;
491 u32 recvlen;
492 u64 requestid;
493 struct icmsg_hdr *icmsghdrp;
494 struct heartbeat_msg_data *heartbeat_msg;
495 u8 *hbeat_txf_buf = util_heartbeat.recv_buffer;
496
497 while (1) {
498
499 if (vmbus_recvpacket(channel, hbeat_txf_buf, HV_HYP_PAGE_SIZE,
500 &recvlen, &requestid)) {
501 pr_err_ratelimited("Heartbeat request received. Could not read into hbeat txf buf\n");
502 return;
503 }
504
505 if (!recvlen)
506 break;
507
508 /* Ensure recvlen is big enough to read header data */
509 if (recvlen < ICMSG_HDR) {
510 pr_err_ratelimited("Heartbeat request received. Packet length too small: %d\n",
511 recvlen);
512 break;
513 }
514
515 icmsghdrp = (struct icmsg_hdr *)&hbeat_txf_buf[
516 sizeof(struct vmbuspipe_hdr)];
517
518 if (icmsghdrp->icmsgtype == ICMSGTYPE_NEGOTIATE) {
519 if (vmbus_prep_negotiate_resp(icmsghdrp,
520 hbeat_txf_buf, recvlen,
521 fw_versions, FW_VER_COUNT,
522 hb_versions, HB_VER_COUNT,
523 NULL, &hb_srv_version)) {
524
525 pr_info("Heartbeat IC version %d.%d\n",
526 hb_srv_version >> 16,
527 hb_srv_version & 0xFFFF);
528 }
529 } else if (icmsghdrp->icmsgtype == ICMSGTYPE_HEARTBEAT) {
530 /*
531 * Ensure recvlen is big enough to read seq_num. Reserved area is not
532 * included in the check as the host may not fill it up entirely
533 */
534 if (recvlen < ICMSG_HDR + sizeof(u64)) {
535 pr_err_ratelimited("Invalid heartbeat msg data. Length too small: %u\n",
536 recvlen);
537 break;
538 }
539 heartbeat_msg = (struct heartbeat_msg_data *)&hbeat_txf_buf[ICMSG_HDR];
540
541 heartbeat_msg->seq_num += 1;
542 } else {
543 icmsghdrp->status = HV_E_FAIL;
544 pr_err_ratelimited("Heartbeat request received. Invalid msg type: %d\n",
545 icmsghdrp->icmsgtype);
546 }
547
548 icmsghdrp->icflags = ICMSGHDRFLAG_TRANSACTION
549 | ICMSGHDRFLAG_RESPONSE;
550
551 vmbus_sendpacket(channel, hbeat_txf_buf,
552 recvlen, requestid,
553 VM_PKT_DATA_INBAND, 0);
554 }
555}
556
557#define HV_UTIL_RING_SEND_SIZE VMBUS_RING_SIZE(3 * HV_HYP_PAGE_SIZE)
558#define HV_UTIL_RING_RECV_SIZE VMBUS_RING_SIZE(3 * HV_HYP_PAGE_SIZE)
559
560static int util_probe(struct hv_device *dev,
561 const struct hv_vmbus_device_id *dev_id)
562{
563 struct hv_util_service *srv =
564 (struct hv_util_service *)dev_id->driver_data;
565 int ret;
566
567 srv->recv_buffer = kmalloc(HV_HYP_PAGE_SIZE * 4, GFP_KERNEL);
568 if (!srv->recv_buffer)
569 return -ENOMEM;
570 srv->channel = dev->channel;
571 if (srv->util_init) {
572 ret = srv->util_init(srv);
573 if (ret) {
574 ret = -ENODEV;
575 goto error1;
576 }
577 }
578
579 /*
580 * The set of services managed by the util driver are not performance
581 * critical and do not need batched reading. Furthermore, some services
582 * such as KVP can only handle one message from the host at a time.
583 * Turn off batched reading for all util drivers before we open the
584 * channel.
585 */
586 set_channel_read_mode(dev->channel, HV_CALL_DIRECT);
587
588 hv_set_drvdata(dev, srv);
589
590 ret = vmbus_open(dev->channel, HV_UTIL_RING_SEND_SIZE,
591 HV_UTIL_RING_RECV_SIZE, NULL, 0, srv->util_cb,
592 dev->channel);
593 if (ret)
594 goto error;
595
596 return 0;
597
598error:
599 if (srv->util_deinit)
600 srv->util_deinit();
601error1:
602 kfree(srv->recv_buffer);
603 return ret;
604}
605
606static int util_remove(struct hv_device *dev)
607{
608 struct hv_util_service *srv = hv_get_drvdata(dev);
609
610 if (srv->util_deinit)
611 srv->util_deinit();
612 vmbus_close(dev->channel);
613 kfree(srv->recv_buffer);
614
615 return 0;
616}
617
618/*
619 * When we're in util_suspend(), all the userspace processes have been frozen
620 * (refer to hibernate() -> freeze_processes()). The userspace is thawed only
621 * after the whole resume procedure, including util_resume(), finishes.
622 */
623static int util_suspend(struct hv_device *dev)
624{
625 struct hv_util_service *srv = hv_get_drvdata(dev);
626 int ret = 0;
627
628 if (srv->util_pre_suspend) {
629 ret = srv->util_pre_suspend();
630 if (ret)
631 return ret;
632 }
633
634 vmbus_close(dev->channel);
635
636 return 0;
637}
638
639static int util_resume(struct hv_device *dev)
640{
641 struct hv_util_service *srv = hv_get_drvdata(dev);
642 int ret = 0;
643
644 if (srv->util_pre_resume) {
645 ret = srv->util_pre_resume();
646 if (ret)
647 return ret;
648 }
649
650 ret = vmbus_open(dev->channel, HV_UTIL_RING_SEND_SIZE,
651 HV_UTIL_RING_RECV_SIZE, NULL, 0, srv->util_cb,
652 dev->channel);
653 return ret;
654}
655
656static const struct hv_vmbus_device_id id_table[] = {
657 /* Shutdown guid */
658 { HV_SHUTDOWN_GUID,
659 .driver_data = (unsigned long)&util_shutdown
660 },
661 /* Time synch guid */
662 { HV_TS_GUID,
663 .driver_data = (unsigned long)&util_timesynch
664 },
665 /* Heartbeat guid */
666 { HV_HEART_BEAT_GUID,
667 .driver_data = (unsigned long)&util_heartbeat
668 },
669 /* KVP guid */
670 { HV_KVP_GUID,
671 .driver_data = (unsigned long)&util_kvp
672 },
673 /* VSS GUID */
674 { HV_VSS_GUID,
675 .driver_data = (unsigned long)&util_vss
676 },
677 /* File copy GUID */
678 { HV_FCOPY_GUID,
679 .driver_data = (unsigned long)&util_fcopy
680 },
681 { },
682};
683
684MODULE_DEVICE_TABLE(vmbus, id_table);
685
686/* The one and only one */
687static struct hv_driver util_drv = {
688 .name = "hv_utils",
689 .id_table = id_table,
690 .probe = util_probe,
691 .remove = util_remove,
692 .suspend = util_suspend,
693 .resume = util_resume,
694 .driver = {
695 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
696 },
697};
698
699static int hv_ptp_enable(struct ptp_clock_info *info,
700 struct ptp_clock_request *request, int on)
701{
702 return -EOPNOTSUPP;
703}
704
705static int hv_ptp_settime(struct ptp_clock_info *p, const struct timespec64 *ts)
706{
707 return -EOPNOTSUPP;
708}
709
710static int hv_ptp_adjfreq(struct ptp_clock_info *ptp, s32 delta)
711{
712 return -EOPNOTSUPP;
713}
714static int hv_ptp_adjtime(struct ptp_clock_info *ptp, s64 delta)
715{
716 return -EOPNOTSUPP;
717}
718
719static int hv_ptp_gettime(struct ptp_clock_info *info, struct timespec64 *ts)
720{
721 return hv_get_adj_host_time(ts);
722}
723
724static struct ptp_clock_info ptp_hyperv_info = {
725 .name = "hyperv",
726 .enable = hv_ptp_enable,
727 .adjtime = hv_ptp_adjtime,
728 .adjfreq = hv_ptp_adjfreq,
729 .gettime64 = hv_ptp_gettime,
730 .settime64 = hv_ptp_settime,
731 .owner = THIS_MODULE,
732};
733
734static struct ptp_clock *hv_ptp_clock;
735
736static int hv_timesync_init(struct hv_util_service *srv)
737{
738 /* TimeSync requires Hyper-V clocksource. */
739 if (!hv_read_reference_counter)
740 return -ENODEV;
741
742 spin_lock_init(&host_ts.lock);
743
744 INIT_WORK(&adj_time_work, hv_set_host_time);
745
746 /*
747 * ptp_clock_register() returns NULL when CONFIG_PTP_1588_CLOCK is
748 * disabled but the driver is still useful without the PTP device
749 * as it still handles the ICTIMESYNCFLAG_SYNC case.
750 */
751 hv_ptp_clock = ptp_clock_register(&ptp_hyperv_info, NULL);
752 if (IS_ERR_OR_NULL(hv_ptp_clock)) {
753 pr_err("cannot register PTP clock: %d\n",
754 PTR_ERR_OR_ZERO(hv_ptp_clock));
755 hv_ptp_clock = NULL;
756 }
757
758 return 0;
759}
760
761static void hv_timesync_cancel_work(void)
762{
763 cancel_work_sync(&adj_time_work);
764}
765
766static int hv_timesync_pre_suspend(void)
767{
768 hv_timesync_cancel_work();
769 return 0;
770}
771
772static void hv_timesync_deinit(void)
773{
774 if (hv_ptp_clock)
775 ptp_clock_unregister(hv_ptp_clock);
776
777 hv_timesync_cancel_work();
778}
779
780static int __init init_hyperv_utils(void)
781{
782 pr_info("Registering HyperV Utility Driver\n");
783
784 return vmbus_driver_register(&util_drv);
785}
786
787static void exit_hyperv_utils(void)
788{
789 pr_info("De-Registered HyperV Utility Driver\n");
790
791 vmbus_driver_unregister(&util_drv);
792}
793
794module_init(init_hyperv_utils);
795module_exit(exit_hyperv_utils);
796
797MODULE_DESCRIPTION("Hyper-V Utilities");
798MODULE_LICENSE("GPL");