1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (c) 2010, Microsoft Corporation.
6 * Haiyang Zhang <haiyangz@microsoft.com>
7 * Hank Janssen <hjanssen@microsoft.com>
9 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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>
23 #include "hyperv_vmbus.h"
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)
34 #define SD_VERSION_1 (SD_MAJOR_1 << 16 | SD_MINOR)
38 #define TS_VERSION (TS_MAJOR << 16 | TS_MINOR)
41 #define TS_VERSION_1 (TS_MAJOR_1 << 16 | TS_MINOR)
44 #define TS_VERSION_3 (TS_MAJOR_3 << 16 | TS_MINOR)
48 #define HB_VERSION (HB_MAJOR << 16 | HB_MINOR)
51 #define HB_VERSION_1 (HB_MAJOR_1 << 16 | HB_MINOR)
53 static int sd_srv_version;
54 static int ts_srv_version;
55 static int hb_srv_version;
57 #define SD_VER_COUNT 4
58 static const int sd_versions[] = {
65 #define TS_VER_COUNT 3
66 static const int ts_versions[] = {
72 #define HB_VER_COUNT 2
73 static const int hb_versions[] = {
78 #define FW_VER_COUNT 2
79 static const int fw_versions[] = {
85 * Send the "hibernate" udev event in a thread context.
87 struct hibernate_work_context {
88 struct work_struct work;
89 struct hv_device *dev;
92 static struct hibernate_work_context hibernate_context;
93 static bool hibernation_supported;
95 static void send_hibernate_uevent(struct work_struct *work)
97 char *uevent_env[2] = { "EVENT=hibernate", NULL };
98 struct hibernate_work_context *ctx;
100 ctx = container_of(work, struct hibernate_work_context, work);
102 kobject_uevent_env(&ctx->dev->device.kobj, KOBJ_CHANGE, uevent_env);
104 pr_info("Sent hibernation uevent\n");
107 static int hv_shutdown_init(struct hv_util_service *srv)
109 struct vmbus_channel *channel = srv->channel;
111 INIT_WORK(&hibernate_context.work, send_hibernate_uevent);
112 hibernate_context.dev = channel->device_obj;
114 hibernation_supported = hv_is_hibernation_supported();
119 static void shutdown_onchannelcallback(void *context);
120 static struct hv_util_service util_shutdown = {
121 .util_cb = shutdown_onchannelcallback,
122 .util_init = hv_shutdown_init,
125 static int hv_timesync_init(struct hv_util_service *srv);
126 static int hv_timesync_pre_suspend(void);
127 static void hv_timesync_deinit(void);
129 static void timesync_onchannelcallback(void *context);
130 static 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,
137 static void heartbeat_onchannelcallback(void *context);
138 static struct hv_util_service util_heartbeat = {
139 .util_cb = heartbeat_onchannelcallback,
142 static 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,
150 static 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,
158 static 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,
166 static void perform_shutdown(struct work_struct *dummy)
168 orderly_poweroff(true);
171 static void perform_restart(struct work_struct *dummy)
177 * Perform the shutdown operation in a thread context.
179 static DECLARE_WORK(shutdown_work, perform_shutdown);
182 * Perform the restart operation in a thread context.
184 static DECLARE_WORK(restart_work, perform_restart);
186 static void shutdown_onchannelcallback(void *context)
188 struct vmbus_channel *channel = context;
189 struct work_struct *work = NULL;
192 u8 *shut_txf_buf = util_shutdown.recv_buffer;
194 struct shutdown_msg_data *shutdown_msg;
196 struct icmsg_hdr *icmsghdrp;
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");
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",
213 icmsghdrp = (struct icmsg_hdr *)&shut_txf_buf[sizeof(struct vmbuspipe_hdr)];
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);
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",
233 shutdown_msg = (struct shutdown_msg_data *)&shut_txf_buf[ICMSG_HDR];
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.
241 switch (shutdown_msg->flags) {
244 icmsghdrp->status = HV_S_OK;
245 work = &shutdown_work;
246 pr_info("Shutdown request received - graceful shutdown initiated\n");
250 icmsghdrp->status = HV_S_OK;
251 work = &restart_work;
252 pr_info("Restart request received - graceful restart initiated\n");
256 pr_info("Hibernation request received\n");
257 icmsghdrp->status = hibernation_supported ?
259 if (hibernation_supported)
260 work = &hibernate_context.work;
263 icmsghdrp->status = HV_E_FAIL;
264 pr_info("Shutdown request received - Invalid request\n");
268 icmsghdrp->status = HV_E_FAIL;
269 pr_err_ratelimited("Shutdown request received. Invalid msg type: %d\n",
270 icmsghdrp->icmsgtype);
273 icmsghdrp->icflags = ICMSGHDRFLAG_TRANSACTION
274 | ICMSGHDRFLAG_RESPONSE;
276 vmbus_sendpacket(channel, shut_txf_buf,
278 VM_PKT_DATA_INBAND, 0);
285 * Set the host time in a process context.
287 static struct work_struct adj_time_work;
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
300 static inline u64 reftime_to_ns(u64 reftime)
302 return (reftime - WLTIMEDELTA) * 100;
306 * Hard coded threshold for host timesync delay: 600 seconds
308 static const u64 HOST_TIMESYNC_DELAY_THRESH = 600 * (u64)NSEC_PER_SEC;
310 static int hv_get_adj_host_time(struct timespec64 *ts)
312 u64 newtime, reftime, timediff_adj;
316 spin_lock_irqsave(&host_ts.lock, flags);
317 reftime = hv_read_reference_counter();
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.
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));
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);
339 static void hv_set_host_time(struct work_struct *work)
342 struct timespec64 ts;
344 if (!hv_get_adj_host_time(&ts))
345 do_settimeofday64(&ts);
349 * Synchronize time with host after reboot, restore, etc.
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.
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.
361 static inline void adj_guesttime(u64 hosttime, u64 reftime, u8 adj_flags)
367 * Save the adjusted time sample from the host and the snapshot
368 * of the current system time.
370 spin_lock_irqsave(&host_ts.lock, flags);
372 cur_reftime = hv_read_reference_counter();
373 host_ts.host_time = hosttime;
374 host_ts.ref_time = cur_reftime;
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.
383 host_ts.host_time += (cur_reftime - reftime);
385 spin_unlock_irqrestore(&host_ts.lock, flags);
387 /* Schedule work to do do_settimeofday64() */
388 if (adj_flags & ICTIMESYNCFLAG_SYNC)
389 schedule_work(&adj_time_work);
393 * Time Sync Channel message handler.
395 static void timesync_onchannelcallback(void *context)
397 struct vmbus_channel *channel = context;
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;
406 * Drain the ring buffer and use the last packet to update
410 int ret = vmbus_recvpacket(channel, time_txf_buf,
411 HV_HYP_PAGE_SIZE, &recvlen,
414 pr_err_ratelimited("TimeSync IC pkt recv failed (Err: %d)\n",
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",
429 icmsghdrp = (struct icmsg_hdr *)&time_txf_buf[
430 sizeof(struct vmbuspipe_hdr)];
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);
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",
450 refdata = (struct ictimesync_ref_data *)&time_txf_buf[ICMSG_HDR];
452 adj_guesttime(refdata->parenttime,
453 refdata->vmreferencetime,
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",
462 timedatap = (struct ictimesync_data *)&time_txf_buf[ICMSG_HDR];
464 adj_guesttime(timedatap->parenttime,
465 hv_read_reference_counter(),
469 icmsghdrp->status = HV_E_FAIL;
470 pr_err_ratelimited("Timesync request received. Invalid msg type: %d\n",
471 icmsghdrp->icmsgtype);
474 icmsghdrp->icflags = ICMSGHDRFLAG_TRANSACTION
475 | ICMSGHDRFLAG_RESPONSE;
477 vmbus_sendpacket(channel, time_txf_buf,
479 VM_PKT_DATA_INBAND, 0);
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.
488 static void heartbeat_onchannelcallback(void *context)
490 struct vmbus_channel *channel = context;
493 struct icmsg_hdr *icmsghdrp;
494 struct heartbeat_msg_data *heartbeat_msg;
495 u8 *hbeat_txf_buf = util_heartbeat.recv_buffer;
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");
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",
515 icmsghdrp = (struct icmsg_hdr *)&hbeat_txf_buf[
516 sizeof(struct vmbuspipe_hdr)];
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)) {
525 pr_info("Heartbeat IC version %d.%d\n",
526 hb_srv_version >> 16,
527 hb_srv_version & 0xFFFF);
529 } else if (icmsghdrp->icmsgtype == ICMSGTYPE_HEARTBEAT) {
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
534 if (recvlen < ICMSG_HDR + sizeof(u64)) {
535 pr_err_ratelimited("Invalid heartbeat msg data. Length too small: %u\n",
539 heartbeat_msg = (struct heartbeat_msg_data *)&hbeat_txf_buf[ICMSG_HDR];
541 heartbeat_msg->seq_num += 1;
543 icmsghdrp->status = HV_E_FAIL;
544 pr_err_ratelimited("Heartbeat request received. Invalid msg type: %d\n",
545 icmsghdrp->icmsgtype);
548 icmsghdrp->icflags = ICMSGHDRFLAG_TRANSACTION
549 | ICMSGHDRFLAG_RESPONSE;
551 vmbus_sendpacket(channel, hbeat_txf_buf,
553 VM_PKT_DATA_INBAND, 0);
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)
560 static int util_probe(struct hv_device *dev,
561 const struct hv_vmbus_device_id *dev_id)
563 struct hv_util_service *srv =
564 (struct hv_util_service *)dev_id->driver_data;
567 srv->recv_buffer = kmalloc(HV_HYP_PAGE_SIZE * 4, GFP_KERNEL);
568 if (!srv->recv_buffer)
570 srv->channel = dev->channel;
571 if (srv->util_init) {
572 ret = srv->util_init(srv);
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
586 set_channel_read_mode(dev->channel, HV_CALL_DIRECT);
588 hv_set_drvdata(dev, srv);
590 ret = vmbus_open(dev->channel, HV_UTIL_RING_SEND_SIZE,
591 HV_UTIL_RING_RECV_SIZE, NULL, 0, srv->util_cb,
599 if (srv->util_deinit)
602 kfree(srv->recv_buffer);
606 static int util_remove(struct hv_device *dev)
608 struct hv_util_service *srv = hv_get_drvdata(dev);
610 if (srv->util_deinit)
612 vmbus_close(dev->channel);
613 kfree(srv->recv_buffer);
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.
623 static int util_suspend(struct hv_device *dev)
625 struct hv_util_service *srv = hv_get_drvdata(dev);
628 if (srv->util_pre_suspend) {
629 ret = srv->util_pre_suspend();
634 vmbus_close(dev->channel);
639 static int util_resume(struct hv_device *dev)
641 struct hv_util_service *srv = hv_get_drvdata(dev);
644 if (srv->util_pre_resume) {
645 ret = srv->util_pre_resume();
650 ret = vmbus_open(dev->channel, HV_UTIL_RING_SEND_SIZE,
651 HV_UTIL_RING_RECV_SIZE, NULL, 0, srv->util_cb,
656 static const struct hv_vmbus_device_id id_table[] = {
659 .driver_data = (unsigned long)&util_shutdown
661 /* Time synch guid */
663 .driver_data = (unsigned long)&util_timesynch
666 { HV_HEART_BEAT_GUID,
667 .driver_data = (unsigned long)&util_heartbeat
671 .driver_data = (unsigned long)&util_kvp
675 .driver_data = (unsigned long)&util_vss
679 .driver_data = (unsigned long)&util_fcopy
684 MODULE_DEVICE_TABLE(vmbus, id_table);
686 /* The one and only one */
687 static struct hv_driver util_drv = {
689 .id_table = id_table,
691 .remove = util_remove,
692 .suspend = util_suspend,
693 .resume = util_resume,
695 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
699 static int hv_ptp_enable(struct ptp_clock_info *info,
700 struct ptp_clock_request *request, int on)
705 static int hv_ptp_settime(struct ptp_clock_info *p, const struct timespec64 *ts)
710 static int hv_ptp_adjfreq(struct ptp_clock_info *ptp, s32 delta)
714 static int hv_ptp_adjtime(struct ptp_clock_info *ptp, s64 delta)
719 static int hv_ptp_gettime(struct ptp_clock_info *info, struct timespec64 *ts)
721 return hv_get_adj_host_time(ts);
724 static struct ptp_clock_info ptp_hyperv_info = {
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,
734 static struct ptp_clock *hv_ptp_clock;
736 static int hv_timesync_init(struct hv_util_service *srv)
738 /* TimeSync requires Hyper-V clocksource. */
739 if (!hv_read_reference_counter)
742 spin_lock_init(&host_ts.lock);
744 INIT_WORK(&adj_time_work, hv_set_host_time);
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.
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: %ld\n",
754 PTR_ERR(hv_ptp_clock));
761 static void hv_timesync_cancel_work(void)
763 cancel_work_sync(&adj_time_work);
766 static int hv_timesync_pre_suspend(void)
768 hv_timesync_cancel_work();
772 static void hv_timesync_deinit(void)
775 ptp_clock_unregister(hv_ptp_clock);
777 hv_timesync_cancel_work();
780 static int __init init_hyperv_utils(void)
782 pr_info("Registering HyperV Utility Driver\n");
784 return vmbus_driver_register(&util_drv);
787 static void exit_hyperv_utils(void)
789 pr_info("De-Registered HyperV Utility Driver\n");
791 vmbus_driver_unregister(&util_drv);
794 module_init(init_hyperv_utils);
795 module_exit(exit_hyperv_utils);
797 MODULE_DESCRIPTION("Hyper-V Utilities");
798 MODULE_LICENSE("GPL");