xref: /linux/drivers/hv/hv_util.c (revision 4b660dbd9ee2059850fd30e0df420ca7a38a1856)
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 
52 static int sd_srv_version;
53 static int ts_srv_version;
54 static int hb_srv_version;
55 
56 #define SD_VER_COUNT 4
57 static 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
65 static const int ts_versions[] = {
66 	TS_VERSION,
67 	TS_VERSION_3,
68 	TS_VERSION_1
69 };
70 
71 #define HB_VER_COUNT 2
72 static const int hb_versions[] = {
73 	HB_VERSION,
74 	HB_VERSION_1
75 };
76 
77 #define FW_VER_COUNT 2
78 static 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  */
86 struct hibernate_work_context {
87 	struct work_struct work;
88 	struct hv_device *dev;
89 };
90 
91 static struct hibernate_work_context hibernate_context;
92 static bool hibernation_supported;
93 
94 static 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 
106 static 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 
118 static void shutdown_onchannelcallback(void *context);
119 static struct hv_util_service util_shutdown = {
120 	.util_cb = shutdown_onchannelcallback,
121 	.util_init = hv_shutdown_init,
122 };
123 
124 static int hv_timesync_init(struct hv_util_service *srv);
125 static int hv_timesync_pre_suspend(void);
126 static void hv_timesync_deinit(void);
127 
128 static void timesync_onchannelcallback(void *context);
129 static 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 
136 static void heartbeat_onchannelcallback(void *context);
137 static struct hv_util_service util_heartbeat = {
138 	.util_cb = heartbeat_onchannelcallback,
139 };
140 
141 static 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 
149 static 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 
157 static 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 
165 static void perform_shutdown(struct work_struct *dummy)
166 {
167 	orderly_poweroff(true);
168 }
169 
170 static void perform_restart(struct work_struct *dummy)
171 {
172 	orderly_reboot();
173 }
174 
175 /*
176  * Perform the shutdown operation in a thread context.
177  */
178 static DECLARE_WORK(shutdown_work, perform_shutdown);
179 
180 /*
181  * Perform the restart operation in a thread context.
182  */
183 static DECLARE_WORK(restart_work, perform_restart);
184 
185 static 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  */
286 static 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  */
293 static struct {
294 	u64				host_time;
295 	u64				ref_time;
296 	spinlock_t			lock;
297 } host_ts;
298 
299 static bool timesync_implicit;
300 
301 module_param(timesync_implicit, bool, 0644);
302 MODULE_PARM_DESC(timesync_implicit, "If set treat SAMPLE as SYNC when clock is behind");
303 
304 static 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  */
312 static const u64 HOST_TIMESYNC_DELAY_THRESH = 600 * (u64)NSEC_PER_SEC;
313 
314 static 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 
343 static 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  */
356 static 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  */
388 static 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  */
423 static 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  */
516 static 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 
588 static 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 
626 error:
627 	if (srv->util_deinit)
628 		srv->util_deinit();
629 error1:
630 	kfree(srv->recv_buffer);
631 	return ret;
632 }
633 
634 static 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  */
649 static 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 
665 static 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 
682 static 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 
710 MODULE_DEVICE_TABLE(vmbus, id_table);
711 
712 /* The one and only one */
713 static  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 
725 static int hv_ptp_enable(struct ptp_clock_info *info,
726 			 struct ptp_clock_request *request, int on)
727 {
728 	return -EOPNOTSUPP;
729 }
730 
731 static int hv_ptp_settime(struct ptp_clock_info *p, const struct timespec64 *ts)
732 {
733 	return -EOPNOTSUPP;
734 }
735 
736 static int hv_ptp_adjfine(struct ptp_clock_info *ptp, long delta)
737 {
738 	return -EOPNOTSUPP;
739 }
740 static int hv_ptp_adjtime(struct ptp_clock_info *ptp, s64 delta)
741 {
742 	return -EOPNOTSUPP;
743 }
744 
745 static int hv_ptp_gettime(struct ptp_clock_info *info, struct timespec64 *ts)
746 {
747 	return hv_get_adj_host_time(ts);
748 }
749 
750 static 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 
760 static struct ptp_clock *hv_ptp_clock;
761 
762 static 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 
783 static void hv_timesync_cancel_work(void)
784 {
785 	cancel_work_sync(&adj_time_work);
786 }
787 
788 static int hv_timesync_pre_suspend(void)
789 {
790 	hv_timesync_cancel_work();
791 	return 0;
792 }
793 
794 static 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 
802 static 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 
809 static void exit_hyperv_utils(void)
810 {
811 	pr_info("De-Registered HyperV Utility Driver\n");
812 
813 	vmbus_driver_unregister(&util_drv);
814 }
815 
816 module_init(init_hyperv_utils);
817 module_exit(exit_hyperv_utils);
818 
819 MODULE_DESCRIPTION("Hyper-V Utilities");
820 MODULE_LICENSE("GPL");
821