xref: /linux/drivers/hv/hv_util.c (revision 37cb0c76ac6c3bf3d82d25a7c95950f2dac3ca41)
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 
send_hibernate_uevent(struct work_struct * work)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 
hv_shutdown_init(struct hv_util_service * srv)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_init_transport = hv_kvp_init_transport,
145 	.util_pre_suspend = hv_kvp_pre_suspend,
146 	.util_pre_resume = hv_kvp_pre_resume,
147 	.util_deinit = hv_kvp_deinit,
148 };
149 
150 static struct hv_util_service util_vss = {
151 	.util_cb = hv_vss_onchannelcallback,
152 	.util_init = hv_vss_init,
153 	.util_init_transport = hv_vss_init_transport,
154 	.util_pre_suspend = hv_vss_pre_suspend,
155 	.util_pre_resume = hv_vss_pre_resume,
156 	.util_deinit = hv_vss_deinit,
157 };
158 
perform_shutdown(struct work_struct * dummy)159 static void perform_shutdown(struct work_struct *dummy)
160 {
161 	orderly_poweroff(true);
162 }
163 
perform_restart(struct work_struct * dummy)164 static void perform_restart(struct work_struct *dummy)
165 {
166 	orderly_reboot();
167 }
168 
169 /*
170  * Perform the shutdown operation in a thread context.
171  */
172 static DECLARE_WORK(shutdown_work, perform_shutdown);
173 
174 /*
175  * Perform the restart operation in a thread context.
176  */
177 static DECLARE_WORK(restart_work, perform_restart);
178 
shutdown_onchannelcallback(void * context)179 static void shutdown_onchannelcallback(void *context)
180 {
181 	struct vmbus_channel *channel = context;
182 	struct work_struct *work = NULL;
183 	u32 recvlen;
184 	u64 requestid;
185 	u8  *shut_txf_buf = util_shutdown.recv_buffer;
186 
187 	struct shutdown_msg_data *shutdown_msg;
188 
189 	struct icmsg_hdr *icmsghdrp;
190 
191 	if (vmbus_recvpacket(channel, shut_txf_buf, HV_HYP_PAGE_SIZE, &recvlen, &requestid)) {
192 		pr_err_ratelimited("Shutdown request received. Could not read into shut txf buf\n");
193 		return;
194 	}
195 
196 	if (!recvlen)
197 		return;
198 
199 	/* Ensure recvlen is big enough to read header data */
200 	if (recvlen < ICMSG_HDR) {
201 		pr_err_ratelimited("Shutdown request received. Packet length too small: %d\n",
202 				   recvlen);
203 		return;
204 	}
205 
206 	icmsghdrp = (struct icmsg_hdr *)&shut_txf_buf[sizeof(struct vmbuspipe_hdr)];
207 
208 	if (icmsghdrp->icmsgtype == ICMSGTYPE_NEGOTIATE) {
209 		if (vmbus_prep_negotiate_resp(icmsghdrp,
210 				shut_txf_buf, recvlen,
211 				fw_versions, FW_VER_COUNT,
212 				sd_versions, SD_VER_COUNT,
213 				NULL, &sd_srv_version)) {
214 			pr_info("Shutdown IC version %d.%d\n",
215 				sd_srv_version >> 16,
216 				sd_srv_version & 0xFFFF);
217 		}
218 	} else if (icmsghdrp->icmsgtype == ICMSGTYPE_SHUTDOWN) {
219 		/* Ensure recvlen is big enough to contain shutdown_msg_data struct */
220 		if (recvlen < ICMSG_HDR + sizeof(struct shutdown_msg_data)) {
221 			pr_err_ratelimited("Invalid shutdown msg data. Packet length too small: %u\n",
222 					   recvlen);
223 			return;
224 		}
225 
226 		shutdown_msg = (struct shutdown_msg_data *)&shut_txf_buf[ICMSG_HDR];
227 
228 		/*
229 		 * shutdown_msg->flags can be 0(shut down), 2(reboot),
230 		 * or 4(hibernate). It may bitwise-OR 1, which means
231 		 * performing the request by force. Linux always tries
232 		 * to perform the request by force.
233 		 */
234 		switch (shutdown_msg->flags) {
235 		case 0:
236 		case 1:
237 			icmsghdrp->status = HV_S_OK;
238 			work = &shutdown_work;
239 			pr_info("Shutdown request received - graceful shutdown initiated\n");
240 			break;
241 		case 2:
242 		case 3:
243 			icmsghdrp->status = HV_S_OK;
244 			work = &restart_work;
245 			pr_info("Restart request received - graceful restart initiated\n");
246 			break;
247 		case 4:
248 		case 5:
249 			pr_info("Hibernation request received\n");
250 			icmsghdrp->status = hibernation_supported ?
251 				HV_S_OK : HV_E_FAIL;
252 			if (hibernation_supported)
253 				work = &hibernate_context.work;
254 			break;
255 		default:
256 			icmsghdrp->status = HV_E_FAIL;
257 			pr_info("Shutdown request received - Invalid request\n");
258 			break;
259 		}
260 	} else {
261 		icmsghdrp->status = HV_E_FAIL;
262 		pr_err_ratelimited("Shutdown request received. Invalid msg type: %d\n",
263 				   icmsghdrp->icmsgtype);
264 	}
265 
266 	icmsghdrp->icflags = ICMSGHDRFLAG_TRANSACTION
267 		| ICMSGHDRFLAG_RESPONSE;
268 
269 	vmbus_sendpacket(channel, shut_txf_buf,
270 			 recvlen, requestid,
271 			 VM_PKT_DATA_INBAND, 0);
272 
273 	if (work)
274 		schedule_work(work);
275 }
276 
277 /*
278  * Set the host time in a process context.
279  */
280 static struct work_struct adj_time_work;
281 
282 /*
283  * The last time sample, received from the host. PTP device responds to
284  * requests by using this data and the current partition-wide time reference
285  * count.
286  */
287 static struct {
288 	u64				host_time;
289 	u64				ref_time;
290 	spinlock_t			lock;
291 } host_ts;
292 
293 static bool timesync_implicit;
294 
295 module_param(timesync_implicit, bool, 0644);
296 MODULE_PARM_DESC(timesync_implicit, "If set treat SAMPLE as SYNC when clock is behind");
297 
reftime_to_ns(u64 reftime)298 static inline u64 reftime_to_ns(u64 reftime)
299 {
300 	return (reftime - WLTIMEDELTA) * 100;
301 }
302 
303 /*
304  * Hard coded threshold for host timesync delay: 600 seconds
305  */
306 static const u64 HOST_TIMESYNC_DELAY_THRESH = 600 * (u64)NSEC_PER_SEC;
307 
hv_get_adj_host_time(struct timespec64 * ts)308 static int hv_get_adj_host_time(struct timespec64 *ts)
309 {
310 	u64 newtime, reftime, timediff_adj;
311 	unsigned long flags;
312 	int ret = 0;
313 
314 	spin_lock_irqsave(&host_ts.lock, flags);
315 	reftime = hv_read_reference_counter();
316 
317 	/*
318 	 * We need to let the caller know that last update from host
319 	 * is older than the max allowable threshold. clock_gettime()
320 	 * and PTP ioctl do not have a documented error that we could
321 	 * return for this specific case. Use ESTALE to report this.
322 	 */
323 	timediff_adj = reftime - host_ts.ref_time;
324 	if (timediff_adj * 100 > HOST_TIMESYNC_DELAY_THRESH) {
325 		pr_warn_once("TIMESYNC IC: Stale time stamp, %llu nsecs old\n",
326 			     (timediff_adj * 100));
327 		ret = -ESTALE;
328 	}
329 
330 	newtime = host_ts.host_time + timediff_adj;
331 	*ts = ns_to_timespec64(reftime_to_ns(newtime));
332 	spin_unlock_irqrestore(&host_ts.lock, flags);
333 
334 	return ret;
335 }
336 
hv_set_host_time(struct work_struct * work)337 static void hv_set_host_time(struct work_struct *work)
338 {
339 
340 	struct timespec64 ts;
341 
342 	if (!hv_get_adj_host_time(&ts))
343 		do_settimeofday64(&ts);
344 }
345 
346 /*
347  * Due to a bug on Hyper-V hosts, the sync flag may not always be sent on resume.
348  * Force a sync if the guest is behind.
349  */
hv_implicit_sync(u64 host_time)350 static inline bool hv_implicit_sync(u64 host_time)
351 {
352 	struct timespec64 new_ts;
353 	struct timespec64 threshold_ts;
354 
355 	new_ts = ns_to_timespec64(reftime_to_ns(host_time));
356 	ktime_get_real_ts64(&threshold_ts);
357 
358 	threshold_ts.tv_sec += 5;
359 
360 	/*
361 	 * If guest behind the host by 5 or more seconds.
362 	 */
363 	if (timespec64_compare(&new_ts, &threshold_ts) >= 0)
364 		return true;
365 
366 	return false;
367 }
368 
369 /*
370  * Synchronize time with host after reboot, restore, etc.
371  *
372  * ICTIMESYNCFLAG_SYNC flag bit indicates reboot, restore events of the VM.
373  * After reboot the flag ICTIMESYNCFLAG_SYNC is included in the first time
374  * message after the timesync channel is opened. Since the hv_utils module is
375  * loaded after hv_vmbus, the first message is usually missed. This bit is
376  * considered a hard request to discipline the clock.
377  *
378  * ICTIMESYNCFLAG_SAMPLE bit indicates a time sample from host. This is
379  * typically used as a hint to the guest. The guest is under no obligation
380  * to discipline the clock.
381  */
adj_guesttime(u64 hosttime,u64 reftime,u8 adj_flags)382 static inline void adj_guesttime(u64 hosttime, u64 reftime, u8 adj_flags)
383 {
384 	unsigned long flags;
385 	u64 cur_reftime;
386 
387 	/*
388 	 * Save the adjusted time sample from the host and the snapshot
389 	 * of the current system time.
390 	 */
391 	spin_lock_irqsave(&host_ts.lock, flags);
392 
393 	cur_reftime = hv_read_reference_counter();
394 	host_ts.host_time = hosttime;
395 	host_ts.ref_time = cur_reftime;
396 
397 	/*
398 	 * TimeSync v4 messages contain reference time (guest's Hyper-V
399 	 * clocksource read when the time sample was generated), we can
400 	 * improve the precision by adding the delta between now and the
401 	 * time of generation. For older protocols we set
402 	 * reftime == cur_reftime on call.
403 	 */
404 	host_ts.host_time += (cur_reftime - reftime);
405 
406 	spin_unlock_irqrestore(&host_ts.lock, flags);
407 
408 	/* Schedule work to do do_settimeofday64() */
409 	if ((adj_flags & ICTIMESYNCFLAG_SYNC) ||
410 	    (timesync_implicit && hv_implicit_sync(host_ts.host_time)))
411 		schedule_work(&adj_time_work);
412 }
413 
414 /*
415  * Time Sync Channel message handler.
416  */
timesync_onchannelcallback(void * context)417 static void timesync_onchannelcallback(void *context)
418 {
419 	struct vmbus_channel *channel = context;
420 	u32 recvlen;
421 	u64 requestid;
422 	struct icmsg_hdr *icmsghdrp;
423 	struct ictimesync_data *timedatap;
424 	struct ictimesync_ref_data *refdata;
425 	u8 *time_txf_buf = util_timesynch.recv_buffer;
426 
427 	/*
428 	 * Drain the ring buffer and use the last packet to update
429 	 * host_ts
430 	 */
431 	while (1) {
432 		int ret = vmbus_recvpacket(channel, time_txf_buf,
433 					   HV_HYP_PAGE_SIZE, &recvlen,
434 					   &requestid);
435 		if (ret) {
436 			pr_err_ratelimited("TimeSync IC pkt recv failed (Err: %d)\n",
437 					   ret);
438 			break;
439 		}
440 
441 		if (!recvlen)
442 			break;
443 
444 		/* Ensure recvlen is big enough to read header data */
445 		if (recvlen < ICMSG_HDR) {
446 			pr_err_ratelimited("Timesync request received. Packet length too small: %d\n",
447 					   recvlen);
448 			break;
449 		}
450 
451 		icmsghdrp = (struct icmsg_hdr *)&time_txf_buf[
452 				sizeof(struct vmbuspipe_hdr)];
453 
454 		if (icmsghdrp->icmsgtype == ICMSGTYPE_NEGOTIATE) {
455 			if (vmbus_prep_negotiate_resp(icmsghdrp,
456 						time_txf_buf, recvlen,
457 						fw_versions, FW_VER_COUNT,
458 						ts_versions, TS_VER_COUNT,
459 						NULL, &ts_srv_version)) {
460 				pr_info("TimeSync IC version %d.%d\n",
461 					ts_srv_version >> 16,
462 					ts_srv_version & 0xFFFF);
463 			}
464 		} else if (icmsghdrp->icmsgtype == ICMSGTYPE_TIMESYNC) {
465 			if (ts_srv_version > TS_VERSION_3) {
466 				/* Ensure recvlen is big enough to read ictimesync_ref_data */
467 				if (recvlen < ICMSG_HDR + sizeof(struct ictimesync_ref_data)) {
468 					pr_err_ratelimited("Invalid ictimesync ref data. Length too small: %u\n",
469 							   recvlen);
470 					break;
471 				}
472 				refdata = (struct ictimesync_ref_data *)&time_txf_buf[ICMSG_HDR];
473 
474 				adj_guesttime(refdata->parenttime,
475 						refdata->vmreferencetime,
476 						refdata->flags);
477 			} else {
478 				/* Ensure recvlen is big enough to read ictimesync_data */
479 				if (recvlen < ICMSG_HDR + sizeof(struct ictimesync_data)) {
480 					pr_err_ratelimited("Invalid ictimesync data. Length too small: %u\n",
481 							   recvlen);
482 					break;
483 				}
484 				timedatap = (struct ictimesync_data *)&time_txf_buf[ICMSG_HDR];
485 
486 				adj_guesttime(timedatap->parenttime,
487 					      hv_read_reference_counter(),
488 					      timedatap->flags);
489 			}
490 		} else {
491 			icmsghdrp->status = HV_E_FAIL;
492 			pr_err_ratelimited("Timesync request received. Invalid msg type: %d\n",
493 					   icmsghdrp->icmsgtype);
494 		}
495 
496 		icmsghdrp->icflags = ICMSGHDRFLAG_TRANSACTION
497 			| ICMSGHDRFLAG_RESPONSE;
498 
499 		vmbus_sendpacket(channel, time_txf_buf,
500 				 recvlen, requestid,
501 				 VM_PKT_DATA_INBAND, 0);
502 	}
503 }
504 
505 /*
506  * Heartbeat functionality.
507  * Every two seconds, Hyper-V send us a heartbeat request message.
508  * we respond to this message, and Hyper-V knows we are alive.
509  */
heartbeat_onchannelcallback(void * context)510 static void heartbeat_onchannelcallback(void *context)
511 {
512 	struct vmbus_channel *channel = context;
513 	u32 recvlen;
514 	u64 requestid;
515 	struct icmsg_hdr *icmsghdrp;
516 	struct heartbeat_msg_data *heartbeat_msg;
517 	u8 *hbeat_txf_buf = util_heartbeat.recv_buffer;
518 
519 	while (1) {
520 
521 		if (vmbus_recvpacket(channel, hbeat_txf_buf, HV_HYP_PAGE_SIZE,
522 				     &recvlen, &requestid)) {
523 			pr_err_ratelimited("Heartbeat request received. Could not read into hbeat txf buf\n");
524 			return;
525 		}
526 
527 		if (!recvlen)
528 			break;
529 
530 		/* Ensure recvlen is big enough to read header data */
531 		if (recvlen < ICMSG_HDR) {
532 			pr_err_ratelimited("Heartbeat request received. Packet length too small: %d\n",
533 					   recvlen);
534 			break;
535 		}
536 
537 		icmsghdrp = (struct icmsg_hdr *)&hbeat_txf_buf[
538 				sizeof(struct vmbuspipe_hdr)];
539 
540 		if (icmsghdrp->icmsgtype == ICMSGTYPE_NEGOTIATE) {
541 			if (vmbus_prep_negotiate_resp(icmsghdrp,
542 					hbeat_txf_buf, recvlen,
543 					fw_versions, FW_VER_COUNT,
544 					hb_versions, HB_VER_COUNT,
545 					NULL, &hb_srv_version)) {
546 
547 				pr_info("Heartbeat IC version %d.%d\n",
548 					hb_srv_version >> 16,
549 					hb_srv_version & 0xFFFF);
550 			}
551 		} else if (icmsghdrp->icmsgtype == ICMSGTYPE_HEARTBEAT) {
552 			/*
553 			 * Ensure recvlen is big enough to read seq_num. Reserved area is not
554 			 * included in the check as the host may not fill it up entirely
555 			 */
556 			if (recvlen < ICMSG_HDR + sizeof(u64)) {
557 				pr_err_ratelimited("Invalid heartbeat msg data. Length too small: %u\n",
558 						   recvlen);
559 				break;
560 			}
561 			heartbeat_msg = (struct heartbeat_msg_data *)&hbeat_txf_buf[ICMSG_HDR];
562 
563 			heartbeat_msg->seq_num += 1;
564 		} else {
565 			icmsghdrp->status = HV_E_FAIL;
566 			pr_err_ratelimited("Heartbeat request received. Invalid msg type: %d\n",
567 					   icmsghdrp->icmsgtype);
568 		}
569 
570 		icmsghdrp->icflags = ICMSGHDRFLAG_TRANSACTION
571 			| ICMSGHDRFLAG_RESPONSE;
572 
573 		vmbus_sendpacket(channel, hbeat_txf_buf,
574 				 recvlen, requestid,
575 				 VM_PKT_DATA_INBAND, 0);
576 	}
577 }
578 
579 #define HV_UTIL_RING_SEND_SIZE VMBUS_RING_SIZE(3 * HV_HYP_PAGE_SIZE)
580 #define HV_UTIL_RING_RECV_SIZE VMBUS_RING_SIZE(3 * HV_HYP_PAGE_SIZE)
581 
util_probe(struct hv_device * dev,const struct hv_vmbus_device_id * dev_id)582 static int util_probe(struct hv_device *dev,
583 			const struct hv_vmbus_device_id *dev_id)
584 {
585 	struct hv_util_service *srv =
586 		(struct hv_util_service *)dev_id->driver_data;
587 	int ret;
588 
589 	srv->recv_buffer = kmalloc(HV_HYP_PAGE_SIZE * 4, GFP_KERNEL);
590 	if (!srv->recv_buffer)
591 		return -ENOMEM;
592 	srv->channel = dev->channel;
593 	if (srv->util_init) {
594 		ret = srv->util_init(srv);
595 		if (ret)
596 			goto error1;
597 	}
598 
599 	/*
600 	 * The set of services managed by the util driver are not performance
601 	 * critical and do not need batched reading. Furthermore, some services
602 	 * such as KVP can only handle one message from the host at a time.
603 	 * Turn off batched reading for all util drivers before we open the
604 	 * channel.
605 	 */
606 	set_channel_read_mode(dev->channel, HV_CALL_DIRECT);
607 
608 	hv_set_drvdata(dev, srv);
609 
610 	ret = vmbus_open(dev->channel, HV_UTIL_RING_SEND_SIZE,
611 			 HV_UTIL_RING_RECV_SIZE, NULL, 0, srv->util_cb,
612 			 dev->channel);
613 	if (ret)
614 		goto error;
615 
616 	if (srv->util_init_transport) {
617 		ret = srv->util_init_transport();
618 		if (ret) {
619 			vmbus_close(dev->channel);
620 			goto error;
621 		}
622 	}
623 	return 0;
624 
625 error:
626 	if (srv->util_deinit)
627 		srv->util_deinit();
628 error1:
629 	kfree(srv->recv_buffer);
630 	return ret;
631 }
632 
util_remove(struct hv_device * dev)633 static void util_remove(struct hv_device *dev)
634 {
635 	struct hv_util_service *srv = hv_get_drvdata(dev);
636 
637 	if (srv->util_deinit)
638 		srv->util_deinit();
639 	vmbus_close(dev->channel);
640 	kfree(srv->recv_buffer);
641 }
642 
643 /*
644  * When we're in util_suspend(), all the userspace processes have been frozen
645  * (refer to hibernate() -> freeze_processes()). The userspace is thawed only
646  * after the whole resume procedure, including util_resume(), finishes.
647  */
util_suspend(struct hv_device * dev)648 static int util_suspend(struct hv_device *dev)
649 {
650 	struct hv_util_service *srv = hv_get_drvdata(dev);
651 	int ret = 0;
652 
653 	if (srv->util_pre_suspend) {
654 		ret = srv->util_pre_suspend();
655 		if (ret)
656 			return ret;
657 	}
658 
659 	vmbus_close(dev->channel);
660 
661 	return 0;
662 }
663 
util_resume(struct hv_device * dev)664 static int util_resume(struct hv_device *dev)
665 {
666 	struct hv_util_service *srv = hv_get_drvdata(dev);
667 	int ret = 0;
668 
669 	if (srv->util_pre_resume) {
670 		ret = srv->util_pre_resume();
671 		if (ret)
672 			return ret;
673 	}
674 
675 	ret = vmbus_open(dev->channel, HV_UTIL_RING_SEND_SIZE,
676 			 HV_UTIL_RING_RECV_SIZE, NULL, 0, srv->util_cb,
677 			 dev->channel);
678 	return ret;
679 }
680 
681 static const struct hv_vmbus_device_id id_table[] = {
682 	/* Shutdown guid */
683 	{ HV_SHUTDOWN_GUID,
684 	  .driver_data = (unsigned long)&util_shutdown
685 	},
686 	/* Time synch guid */
687 	{ HV_TS_GUID,
688 	  .driver_data = (unsigned long)&util_timesynch
689 	},
690 	/* Heartbeat guid */
691 	{ HV_HEART_BEAT_GUID,
692 	  .driver_data = (unsigned long)&util_heartbeat
693 	},
694 	/* KVP guid */
695 	{ HV_KVP_GUID,
696 	  .driver_data = (unsigned long)&util_kvp
697 	},
698 	/* VSS GUID */
699 	{ HV_VSS_GUID,
700 	  .driver_data = (unsigned long)&util_vss
701 	},
702 	{ },
703 };
704 
705 MODULE_DEVICE_TABLE(vmbus, id_table);
706 
707 /* The one and only one */
708 static  struct hv_driver util_drv = {
709 	.name = "hv_utils",
710 	.id_table = id_table,
711 	.probe =  util_probe,
712 	.remove =  util_remove,
713 	.suspend = util_suspend,
714 	.resume =  util_resume,
715 	.driver = {
716 		.probe_type = PROBE_PREFER_ASYNCHRONOUS,
717 	},
718 };
719 
hv_ptp_enable(struct ptp_clock_info * info,struct ptp_clock_request * request,int on)720 static int hv_ptp_enable(struct ptp_clock_info *info,
721 			 struct ptp_clock_request *request, int on)
722 {
723 	return -EOPNOTSUPP;
724 }
725 
hv_ptp_settime(struct ptp_clock_info * p,const struct timespec64 * ts)726 static int hv_ptp_settime(struct ptp_clock_info *p, const struct timespec64 *ts)
727 {
728 	return -EOPNOTSUPP;
729 }
730 
hv_ptp_adjfine(struct ptp_clock_info * ptp,long delta)731 static int hv_ptp_adjfine(struct ptp_clock_info *ptp, long delta)
732 {
733 	return -EOPNOTSUPP;
734 }
hv_ptp_adjtime(struct ptp_clock_info * ptp,s64 delta)735 static int hv_ptp_adjtime(struct ptp_clock_info *ptp, s64 delta)
736 {
737 	return -EOPNOTSUPP;
738 }
739 
hv_ptp_gettime(struct ptp_clock_info * info,struct timespec64 * ts)740 static int hv_ptp_gettime(struct ptp_clock_info *info, struct timespec64 *ts)
741 {
742 	return hv_get_adj_host_time(ts);
743 }
744 
745 static struct ptp_clock_info ptp_hyperv_info = {
746 	.name		= "hyperv",
747 	.enable         = hv_ptp_enable,
748 	.adjtime        = hv_ptp_adjtime,
749 	.adjfine        = hv_ptp_adjfine,
750 	.gettime64      = hv_ptp_gettime,
751 	.settime64      = hv_ptp_settime,
752 	.owner		= THIS_MODULE,
753 };
754 
755 static struct ptp_clock *hv_ptp_clock;
756 
hv_timesync_init(struct hv_util_service * srv)757 static int hv_timesync_init(struct hv_util_service *srv)
758 {
759 	spin_lock_init(&host_ts.lock);
760 
761 	INIT_WORK(&adj_time_work, hv_set_host_time);
762 
763 	/*
764 	 * ptp_clock_register() returns NULL when CONFIG_PTP_1588_CLOCK is
765 	 * disabled but the driver is still useful without the PTP device
766 	 * as it still handles the ICTIMESYNCFLAG_SYNC case.
767 	 */
768 	hv_ptp_clock = ptp_clock_register(&ptp_hyperv_info, NULL);
769 	if (IS_ERR_OR_NULL(hv_ptp_clock)) {
770 		pr_err("cannot register PTP clock: %d\n",
771 		       PTR_ERR_OR_ZERO(hv_ptp_clock));
772 		hv_ptp_clock = NULL;
773 	}
774 
775 	return 0;
776 }
777 
hv_timesync_cancel_work(void)778 static void hv_timesync_cancel_work(void)
779 {
780 	cancel_work_sync(&adj_time_work);
781 }
782 
hv_timesync_pre_suspend(void)783 static int hv_timesync_pre_suspend(void)
784 {
785 	hv_timesync_cancel_work();
786 	return 0;
787 }
788 
hv_timesync_deinit(void)789 static void hv_timesync_deinit(void)
790 {
791 	if (hv_ptp_clock)
792 		ptp_clock_unregister(hv_ptp_clock);
793 
794 	hv_timesync_cancel_work();
795 }
796 
init_hyperv_utils(void)797 static int __init init_hyperv_utils(void)
798 {
799 	pr_info("Registering HyperV Utility Driver\n");
800 
801 	return vmbus_driver_register(&util_drv);
802 }
803 
exit_hyperv_utils(void)804 static void exit_hyperv_utils(void)
805 {
806 	pr_info("De-Registered HyperV Utility Driver\n");
807 
808 	vmbus_driver_unregister(&util_drv);
809 }
810 
811 module_init(init_hyperv_utils);
812 module_exit(exit_hyperv_utils);
813 
814 MODULE_DESCRIPTION("Hyper-V Utilities");
815 MODULE_LICENSE("GPL");
816