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 <asm/mshyperv.h>
22 #include "hyperv_vmbus.h"
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)
33 #define SD_VERSION_1 (SD_MAJOR_1 << 16 | SD_MINOR)
37 #define TS_VERSION (TS_MAJOR << 16 | TS_MINOR)
40 #define TS_VERSION_1 (TS_MAJOR_1 << 16 | TS_MINOR)
43 #define TS_VERSION_3 (TS_MAJOR_3 << 16 | TS_MINOR)
47 #define HB_VERSION (HB_MAJOR << 16 | HB_MINOR)
50 #define HB_VERSION_1 (HB_MAJOR_1 << 16 | HB_MINOR)
52 static int sd_srv_version;
53 static int ts_srv_version;
54 static int hb_srv_version;
56 #define SD_VER_COUNT 4
57 static const int sd_versions[] = {
64 #define TS_VER_COUNT 3
65 static const int ts_versions[] = {
71 #define HB_VER_COUNT 2
72 static const int hb_versions[] = {
77 #define FW_VER_COUNT 2
78 static const int fw_versions[] = {
84 * Send the "hibernate" udev event in a thread context.
86 struct hibernate_work_context {
87 struct work_struct work;
88 struct hv_device *dev;
91 static struct hibernate_work_context hibernate_context;
92 static bool hibernation_supported;
94 static void send_hibernate_uevent(struct work_struct *work)
96 char *uevent_env[2] = { "EVENT=hibernate", NULL };
97 struct hibernate_work_context *ctx;
99 ctx = container_of(work, struct hibernate_work_context, work);
101 kobject_uevent_env(&ctx->dev->device.kobj, KOBJ_CHANGE, uevent_env);
103 pr_info("Sent hibernation uevent\n");
106 static int hv_shutdown_init(struct hv_util_service *srv)
108 struct vmbus_channel *channel = srv->channel;
110 INIT_WORK(&hibernate_context.work, send_hibernate_uevent);
111 hibernate_context.dev = channel->device_obj;
113 hibernation_supported = hv_is_hibernation_supported();
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,
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);
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,
136 static void heartbeat_onchannelcallback(void *context);
137 static struct hv_util_service util_heartbeat = {
138 .util_cb = heartbeat_onchannelcallback,
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,
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,
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,
165 static void perform_shutdown(struct work_struct *dummy)
167 orderly_poweroff(true);
170 static void perform_restart(struct work_struct *dummy)
176 * Perform the shutdown operation in a thread context.
178 static DECLARE_WORK(shutdown_work, perform_shutdown);
181 * Perform the restart operation in a thread context.
183 static DECLARE_WORK(restart_work, perform_restart);
185 static void shutdown_onchannelcallback(void *context)
187 struct vmbus_channel *channel = context;
188 struct work_struct *work = NULL;
191 u8 *shut_txf_buf = util_shutdown.recv_buffer;
193 struct shutdown_msg_data *shutdown_msg;
195 struct icmsg_hdr *icmsghdrp;
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");
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",
212 icmsghdrp = (struct icmsg_hdr *)&shut_txf_buf[sizeof(struct vmbuspipe_hdr)];
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);
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",
232 shutdown_msg = (struct shutdown_msg_data *)&shut_txf_buf[ICMSG_HDR];
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.
240 switch (shutdown_msg->flags) {
243 icmsghdrp->status = HV_S_OK;
244 work = &shutdown_work;
245 pr_info("Shutdown request received - graceful shutdown initiated\n");
249 icmsghdrp->status = HV_S_OK;
250 work = &restart_work;
251 pr_info("Restart request received - graceful restart initiated\n");
255 pr_info("Hibernation request received\n");
256 icmsghdrp->status = hibernation_supported ?
258 if (hibernation_supported)
259 work = &hibernate_context.work;
262 icmsghdrp->status = HV_E_FAIL;
263 pr_info("Shutdown request received - Invalid request\n");
267 icmsghdrp->status = HV_E_FAIL;
268 pr_err_ratelimited("Shutdown request received. Invalid msg type: %d\n",
269 icmsghdrp->icmsgtype);
272 icmsghdrp->icflags = ICMSGHDRFLAG_TRANSACTION
273 | ICMSGHDRFLAG_RESPONSE;
275 vmbus_sendpacket(channel, shut_txf_buf,
277 VM_PKT_DATA_INBAND, 0);
284 * Set the host time in a process context.
286 static struct work_struct adj_time_work;
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
299 static inline u64 reftime_to_ns(u64 reftime)
301 return (reftime - WLTIMEDELTA) * 100;
305 * Hard coded threshold for host timesync delay: 600 seconds
307 static const u64 HOST_TIMESYNC_DELAY_THRESH = 600 * (u64)NSEC_PER_SEC;
309 static int hv_get_adj_host_time(struct timespec64 *ts)
311 u64 newtime, reftime, timediff_adj;
315 spin_lock_irqsave(&host_ts.lock, flags);
316 reftime = hv_read_reference_counter();
319 * We need to let the caller know that last update from host
320 * is older than the max allowable threshold. clock_gettime()
321 * and PTP ioctl do not have a documented error that we could
322 * return for this specific case. Use ESTALE to report this.
324 timediff_adj = reftime - host_ts.ref_time;
325 if (timediff_adj * 100 > HOST_TIMESYNC_DELAY_THRESH) {
326 pr_warn_once("TIMESYNC IC: Stale time stamp, %llu nsecs old\n",
327 (timediff_adj * 100));
331 newtime = host_ts.host_time + timediff_adj;
332 *ts = ns_to_timespec64(reftime_to_ns(newtime));
333 spin_unlock_irqrestore(&host_ts.lock, flags);
338 static void hv_set_host_time(struct work_struct *work)
341 struct timespec64 ts;
343 if (!hv_get_adj_host_time(&ts))
344 do_settimeofday64(&ts);
348 * Synchronize time with host after reboot, restore, etc.
350 * ICTIMESYNCFLAG_SYNC flag bit indicates reboot, restore events of the VM.
351 * After reboot the flag ICTIMESYNCFLAG_SYNC is included in the first time
352 * message after the timesync channel is opened. Since the hv_utils module is
353 * loaded after hv_vmbus, the first message is usually missed. This bit is
354 * considered a hard request to discipline the clock.
356 * ICTIMESYNCFLAG_SAMPLE bit indicates a time sample from host. This is
357 * typically used as a hint to the guest. The guest is under no obligation
358 * to discipline the clock.
360 static inline void adj_guesttime(u64 hosttime, u64 reftime, u8 adj_flags)
366 * Save the adjusted time sample from the host and the snapshot
367 * of the current system time.
369 spin_lock_irqsave(&host_ts.lock, flags);
371 cur_reftime = hv_read_reference_counter();
372 host_ts.host_time = hosttime;
373 host_ts.ref_time = cur_reftime;
376 * TimeSync v4 messages contain reference time (guest's Hyper-V
377 * clocksource read when the time sample was generated), we can
378 * improve the precision by adding the delta between now and the
379 * time of generation. For older protocols we set
380 * reftime == cur_reftime on call.
382 host_ts.host_time += (cur_reftime - reftime);
384 spin_unlock_irqrestore(&host_ts.lock, flags);
386 /* Schedule work to do do_settimeofday64() */
387 if (adj_flags & ICTIMESYNCFLAG_SYNC)
388 schedule_work(&adj_time_work);
392 * Time Sync Channel message handler.
394 static void timesync_onchannelcallback(void *context)
396 struct vmbus_channel *channel = context;
399 struct icmsg_hdr *icmsghdrp;
400 struct ictimesync_data *timedatap;
401 struct ictimesync_ref_data *refdata;
402 u8 *time_txf_buf = util_timesynch.recv_buffer;
405 * Drain the ring buffer and use the last packet to update
409 int ret = vmbus_recvpacket(channel, time_txf_buf,
410 HV_HYP_PAGE_SIZE, &recvlen,
413 pr_err_ratelimited("TimeSync IC pkt recv failed (Err: %d)\n",
421 /* Ensure recvlen is big enough to read header data */
422 if (recvlen < ICMSG_HDR) {
423 pr_err_ratelimited("Timesync request received. Packet length too small: %d\n",
428 icmsghdrp = (struct icmsg_hdr *)&time_txf_buf[
429 sizeof(struct vmbuspipe_hdr)];
431 if (icmsghdrp->icmsgtype == ICMSGTYPE_NEGOTIATE) {
432 if (vmbus_prep_negotiate_resp(icmsghdrp,
433 time_txf_buf, recvlen,
434 fw_versions, FW_VER_COUNT,
435 ts_versions, TS_VER_COUNT,
436 NULL, &ts_srv_version)) {
437 pr_info("TimeSync IC version %d.%d\n",
438 ts_srv_version >> 16,
439 ts_srv_version & 0xFFFF);
441 } else if (icmsghdrp->icmsgtype == ICMSGTYPE_TIMESYNC) {
442 if (ts_srv_version > TS_VERSION_3) {
443 /* Ensure recvlen is big enough to read ictimesync_ref_data */
444 if (recvlen < ICMSG_HDR + sizeof(struct ictimesync_ref_data)) {
445 pr_err_ratelimited("Invalid ictimesync ref data. Length too small: %u\n",
449 refdata = (struct ictimesync_ref_data *)&time_txf_buf[ICMSG_HDR];
451 adj_guesttime(refdata->parenttime,
452 refdata->vmreferencetime,
455 /* Ensure recvlen is big enough to read ictimesync_data */
456 if (recvlen < ICMSG_HDR + sizeof(struct ictimesync_data)) {
457 pr_err_ratelimited("Invalid ictimesync data. Length too small: %u\n",
461 timedatap = (struct ictimesync_data *)&time_txf_buf[ICMSG_HDR];
463 adj_guesttime(timedatap->parenttime,
464 hv_read_reference_counter(),
468 icmsghdrp->status = HV_E_FAIL;
469 pr_err_ratelimited("Timesync request received. Invalid msg type: %d\n",
470 icmsghdrp->icmsgtype);
473 icmsghdrp->icflags = ICMSGHDRFLAG_TRANSACTION
474 | ICMSGHDRFLAG_RESPONSE;
476 vmbus_sendpacket(channel, time_txf_buf,
478 VM_PKT_DATA_INBAND, 0);
483 * Heartbeat functionality.
484 * Every two seconds, Hyper-V send us a heartbeat request message.
485 * we respond to this message, and Hyper-V knows we are alive.
487 static void heartbeat_onchannelcallback(void *context)
489 struct vmbus_channel *channel = context;
492 struct icmsg_hdr *icmsghdrp;
493 struct heartbeat_msg_data *heartbeat_msg;
494 u8 *hbeat_txf_buf = util_heartbeat.recv_buffer;
498 if (vmbus_recvpacket(channel, hbeat_txf_buf, HV_HYP_PAGE_SIZE,
499 &recvlen, &requestid)) {
500 pr_err_ratelimited("Heartbeat request received. Could not read into hbeat txf buf\n");
507 /* Ensure recvlen is big enough to read header data */
508 if (recvlen < ICMSG_HDR) {
509 pr_err_ratelimited("Heartbeat request received. Packet length too small: %d\n",
514 icmsghdrp = (struct icmsg_hdr *)&hbeat_txf_buf[
515 sizeof(struct vmbuspipe_hdr)];
517 if (icmsghdrp->icmsgtype == ICMSGTYPE_NEGOTIATE) {
518 if (vmbus_prep_negotiate_resp(icmsghdrp,
519 hbeat_txf_buf, recvlen,
520 fw_versions, FW_VER_COUNT,
521 hb_versions, HB_VER_COUNT,
522 NULL, &hb_srv_version)) {
524 pr_info("Heartbeat IC version %d.%d\n",
525 hb_srv_version >> 16,
526 hb_srv_version & 0xFFFF);
528 } else if (icmsghdrp->icmsgtype == ICMSGTYPE_HEARTBEAT) {
530 * Ensure recvlen is big enough to read seq_num. Reserved area is not
531 * included in the check as the host may not fill it up entirely
533 if (recvlen < ICMSG_HDR + sizeof(u64)) {
534 pr_err_ratelimited("Invalid heartbeat msg data. Length too small: %u\n",
538 heartbeat_msg = (struct heartbeat_msg_data *)&hbeat_txf_buf[ICMSG_HDR];
540 heartbeat_msg->seq_num += 1;
542 icmsghdrp->status = HV_E_FAIL;
543 pr_err_ratelimited("Heartbeat request received. Invalid msg type: %d\n",
544 icmsghdrp->icmsgtype);
547 icmsghdrp->icflags = ICMSGHDRFLAG_TRANSACTION
548 | ICMSGHDRFLAG_RESPONSE;
550 vmbus_sendpacket(channel, hbeat_txf_buf,
552 VM_PKT_DATA_INBAND, 0);
556 #define HV_UTIL_RING_SEND_SIZE VMBUS_RING_SIZE(3 * HV_HYP_PAGE_SIZE)
557 #define HV_UTIL_RING_RECV_SIZE VMBUS_RING_SIZE(3 * HV_HYP_PAGE_SIZE)
559 static int util_probe(struct hv_device *dev,
560 const struct hv_vmbus_device_id *dev_id)
562 struct hv_util_service *srv =
563 (struct hv_util_service *)dev_id->driver_data;
566 srv->recv_buffer = kmalloc(HV_HYP_PAGE_SIZE * 4, GFP_KERNEL);
567 if (!srv->recv_buffer)
569 srv->channel = dev->channel;
570 if (srv->util_init) {
571 ret = srv->util_init(srv);
579 * The set of services managed by the util driver are not performance
580 * critical and do not need batched reading. Furthermore, some services
581 * such as KVP can only handle one message from the host at a time.
582 * Turn off batched reading for all util drivers before we open the
585 set_channel_read_mode(dev->channel, HV_CALL_DIRECT);
587 hv_set_drvdata(dev, srv);
589 ret = vmbus_open(dev->channel, HV_UTIL_RING_SEND_SIZE,
590 HV_UTIL_RING_RECV_SIZE, NULL, 0, srv->util_cb,
598 if (srv->util_deinit)
601 kfree(srv->recv_buffer);
605 static int util_remove(struct hv_device *dev)
607 struct hv_util_service *srv = hv_get_drvdata(dev);
609 if (srv->util_deinit)
611 vmbus_close(dev->channel);
612 kfree(srv->recv_buffer);
618 * When we're in util_suspend(), all the userspace processes have been frozen
619 * (refer to hibernate() -> freeze_processes()). The userspace is thawed only
620 * after the whole resume procedure, including util_resume(), finishes.
622 static int util_suspend(struct hv_device *dev)
624 struct hv_util_service *srv = hv_get_drvdata(dev);
627 if (srv->util_pre_suspend) {
628 ret = srv->util_pre_suspend();
633 vmbus_close(dev->channel);
638 static int util_resume(struct hv_device *dev)
640 struct hv_util_service *srv = hv_get_drvdata(dev);
643 if (srv->util_pre_resume) {
644 ret = srv->util_pre_resume();
649 ret = vmbus_open(dev->channel, HV_UTIL_RING_SEND_SIZE,
650 HV_UTIL_RING_RECV_SIZE, NULL, 0, srv->util_cb,
655 static const struct hv_vmbus_device_id id_table[] = {
658 .driver_data = (unsigned long)&util_shutdown
660 /* Time synch guid */
662 .driver_data = (unsigned long)&util_timesynch
665 { HV_HEART_BEAT_GUID,
666 .driver_data = (unsigned long)&util_heartbeat
670 .driver_data = (unsigned long)&util_kvp
674 .driver_data = (unsigned long)&util_vss
678 .driver_data = (unsigned long)&util_fcopy
683 MODULE_DEVICE_TABLE(vmbus, id_table);
685 /* The one and only one */
686 static struct hv_driver util_drv = {
688 .id_table = id_table,
690 .remove = util_remove,
691 .suspend = util_suspend,
692 .resume = util_resume,
694 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
698 static int hv_ptp_enable(struct ptp_clock_info *info,
699 struct ptp_clock_request *request, int on)
704 static int hv_ptp_settime(struct ptp_clock_info *p, const struct timespec64 *ts)
709 static int hv_ptp_adjfreq(struct ptp_clock_info *ptp, s32 delta)
713 static int hv_ptp_adjtime(struct ptp_clock_info *ptp, s64 delta)
718 static int hv_ptp_gettime(struct ptp_clock_info *info, struct timespec64 *ts)
720 return hv_get_adj_host_time(ts);
723 static struct ptp_clock_info ptp_hyperv_info = {
725 .enable = hv_ptp_enable,
726 .adjtime = hv_ptp_adjtime,
727 .adjfreq = hv_ptp_adjfreq,
728 .gettime64 = hv_ptp_gettime,
729 .settime64 = hv_ptp_settime,
730 .owner = THIS_MODULE,
733 static struct ptp_clock *hv_ptp_clock;
735 static int hv_timesync_init(struct hv_util_service *srv)
737 spin_lock_init(&host_ts.lock);
739 INIT_WORK(&adj_time_work, hv_set_host_time);
742 * ptp_clock_register() returns NULL when CONFIG_PTP_1588_CLOCK is
743 * disabled but the driver is still useful without the PTP device
744 * as it still handles the ICTIMESYNCFLAG_SYNC case.
746 hv_ptp_clock = ptp_clock_register(&ptp_hyperv_info, NULL);
747 if (IS_ERR_OR_NULL(hv_ptp_clock)) {
748 pr_err("cannot register PTP clock: %d\n",
749 PTR_ERR_OR_ZERO(hv_ptp_clock));
756 static void hv_timesync_cancel_work(void)
758 cancel_work_sync(&adj_time_work);
761 static int hv_timesync_pre_suspend(void)
763 hv_timesync_cancel_work();
767 static void hv_timesync_deinit(void)
770 ptp_clock_unregister(hv_ptp_clock);
772 hv_timesync_cancel_work();
775 static int __init init_hyperv_utils(void)
777 pr_info("Registering HyperV Utility Driver\n");
779 return vmbus_driver_register(&util_drv);
782 static void exit_hyperv_utils(void)
784 pr_info("De-Registered HyperV Utility Driver\n");
786 vmbus_driver_unregister(&util_drv);
789 module_init(init_hyperv_utils);
790 module_exit(exit_hyperv_utils);
792 MODULE_DESCRIPTION("Hyper-V Utilities");
793 MODULE_LICENSE("GPL");