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 vmbus_recvpacket(channel
, shut_txf_buf
,
199 HV_HYP_PAGE_SIZE
, &recvlen
, &requestid
);
202 icmsghdrp
= (struct icmsg_hdr
*)&shut_txf_buf
[
203 sizeof(struct vmbuspipe_hdr
)];
205 if (icmsghdrp
->icmsgtype
== ICMSGTYPE_NEGOTIATE
) {
206 if (vmbus_prep_negotiate_resp(icmsghdrp
, shut_txf_buf
,
207 fw_versions
, FW_VER_COUNT
,
208 sd_versions
, SD_VER_COUNT
,
209 NULL
, &sd_srv_version
)) {
210 pr_info("Shutdown IC version %d.%d\n",
211 sd_srv_version
>> 16,
212 sd_srv_version
& 0xFFFF);
216 (struct shutdown_msg_data
*)&shut_txf_buf
[
217 sizeof(struct vmbuspipe_hdr
) +
218 sizeof(struct icmsg_hdr
)];
221 * shutdown_msg->flags can be 0(shut down), 2(reboot),
222 * or 4(hibernate). It may bitwise-OR 1, which means
223 * performing the request by force. Linux always tries
224 * to perform the request by force.
226 switch (shutdown_msg
->flags
) {
229 icmsghdrp
->status
= HV_S_OK
;
230 work
= &shutdown_work
;
231 pr_info("Shutdown request received -"
232 " graceful shutdown initiated\n");
236 icmsghdrp
->status
= HV_S_OK
;
237 work
= &restart_work
;
238 pr_info("Restart request received -"
239 " graceful restart initiated\n");
243 pr_info("Hibernation request received\n");
244 icmsghdrp
->status
= hibernation_supported
?
246 if (hibernation_supported
)
247 work
= &hibernate_context
.work
;
250 icmsghdrp
->status
= HV_E_FAIL
;
251 pr_info("Shutdown request received -"
252 " Invalid request\n");
257 icmsghdrp
->icflags
= ICMSGHDRFLAG_TRANSACTION
258 | ICMSGHDRFLAG_RESPONSE
;
260 vmbus_sendpacket(channel
, shut_txf_buf
,
262 VM_PKT_DATA_INBAND
, 0);
270 * Set the host time in a process context.
272 static struct work_struct adj_time_work
;
275 * The last time sample, received from the host. PTP device responds to
276 * requests by using this data and the current partition-wide time reference
285 static struct timespec64
hv_get_adj_host_time(void)
287 struct timespec64 ts
;
288 u64 newtime
, reftime
;
291 spin_lock_irqsave(&host_ts
.lock
, flags
);
292 reftime
= hv_read_reference_counter();
293 newtime
= host_ts
.host_time
+ (reftime
- host_ts
.ref_time
);
294 ts
= ns_to_timespec64((newtime
- WLTIMEDELTA
) * 100);
295 spin_unlock_irqrestore(&host_ts
.lock
, flags
);
300 static void hv_set_host_time(struct work_struct
*work
)
302 struct timespec64 ts
= hv_get_adj_host_time();
304 do_settimeofday64(&ts
);
308 * Synchronize time with host after reboot, restore, etc.
310 * ICTIMESYNCFLAG_SYNC flag bit indicates reboot, restore events of the VM.
311 * After reboot the flag ICTIMESYNCFLAG_SYNC is included in the first time
312 * message after the timesync channel is opened. Since the hv_utils module is
313 * loaded after hv_vmbus, the first message is usually missed. This bit is
314 * considered a hard request to discipline the clock.
316 * ICTIMESYNCFLAG_SAMPLE bit indicates a time sample from host. This is
317 * typically used as a hint to the guest. The guest is under no obligation
318 * to discipline the clock.
320 static inline void adj_guesttime(u64 hosttime
, u64 reftime
, u8 adj_flags
)
326 * Save the adjusted time sample from the host and the snapshot
327 * of the current system time.
329 spin_lock_irqsave(&host_ts
.lock
, flags
);
331 cur_reftime
= hv_read_reference_counter();
332 host_ts
.host_time
= hosttime
;
333 host_ts
.ref_time
= cur_reftime
;
336 * TimeSync v4 messages contain reference time (guest's Hyper-V
337 * clocksource read when the time sample was generated), we can
338 * improve the precision by adding the delta between now and the
339 * time of generation. For older protocols we set
340 * reftime == cur_reftime on call.
342 host_ts
.host_time
+= (cur_reftime
- reftime
);
344 spin_unlock_irqrestore(&host_ts
.lock
, flags
);
346 /* Schedule work to do do_settimeofday64() */
347 if (adj_flags
& ICTIMESYNCFLAG_SYNC
)
348 schedule_work(&adj_time_work
);
352 * Time Sync Channel message handler.
354 static void timesync_onchannelcallback(void *context
)
356 struct vmbus_channel
*channel
= context
;
359 struct icmsg_hdr
*icmsghdrp
;
360 struct ictimesync_data
*timedatap
;
361 struct ictimesync_ref_data
*refdata
;
362 u8
*time_txf_buf
= util_timesynch
.recv_buffer
;
364 vmbus_recvpacket(channel
, time_txf_buf
,
365 HV_HYP_PAGE_SIZE
, &recvlen
, &requestid
);
368 icmsghdrp
= (struct icmsg_hdr
*)&time_txf_buf
[
369 sizeof(struct vmbuspipe_hdr
)];
371 if (icmsghdrp
->icmsgtype
== ICMSGTYPE_NEGOTIATE
) {
372 if (vmbus_prep_negotiate_resp(icmsghdrp
, time_txf_buf
,
373 fw_versions
, FW_VER_COUNT
,
374 ts_versions
, TS_VER_COUNT
,
375 NULL
, &ts_srv_version
)) {
376 pr_info("TimeSync IC version %d.%d\n",
377 ts_srv_version
>> 16,
378 ts_srv_version
& 0xFFFF);
381 if (ts_srv_version
> TS_VERSION_3
) {
382 refdata
= (struct ictimesync_ref_data
*)
384 sizeof(struct vmbuspipe_hdr
) +
385 sizeof(struct icmsg_hdr
)];
387 adj_guesttime(refdata
->parenttime
,
388 refdata
->vmreferencetime
,
391 timedatap
= (struct ictimesync_data
*)
393 sizeof(struct vmbuspipe_hdr
) +
394 sizeof(struct icmsg_hdr
)];
395 adj_guesttime(timedatap
->parenttime
,
396 hv_read_reference_counter(),
401 icmsghdrp
->icflags
= ICMSGHDRFLAG_TRANSACTION
402 | ICMSGHDRFLAG_RESPONSE
;
404 vmbus_sendpacket(channel
, time_txf_buf
,
406 VM_PKT_DATA_INBAND
, 0);
411 * Heartbeat functionality.
412 * Every two seconds, Hyper-V send us a heartbeat request message.
413 * we respond to this message, and Hyper-V knows we are alive.
415 static void heartbeat_onchannelcallback(void *context
)
417 struct vmbus_channel
*channel
= context
;
420 struct icmsg_hdr
*icmsghdrp
;
421 struct heartbeat_msg_data
*heartbeat_msg
;
422 u8
*hbeat_txf_buf
= util_heartbeat
.recv_buffer
;
426 vmbus_recvpacket(channel
, hbeat_txf_buf
,
427 HV_HYP_PAGE_SIZE
, &recvlen
, &requestid
);
432 icmsghdrp
= (struct icmsg_hdr
*)&hbeat_txf_buf
[
433 sizeof(struct vmbuspipe_hdr
)];
435 if (icmsghdrp
->icmsgtype
== ICMSGTYPE_NEGOTIATE
) {
436 if (vmbus_prep_negotiate_resp(icmsghdrp
,
438 fw_versions
, FW_VER_COUNT
,
439 hb_versions
, HB_VER_COUNT
,
440 NULL
, &hb_srv_version
)) {
442 pr_info("Heartbeat IC version %d.%d\n",
443 hb_srv_version
>> 16,
444 hb_srv_version
& 0xFFFF);
448 (struct heartbeat_msg_data
*)&hbeat_txf_buf
[
449 sizeof(struct vmbuspipe_hdr
) +
450 sizeof(struct icmsg_hdr
)];
452 heartbeat_msg
->seq_num
+= 1;
455 icmsghdrp
->icflags
= ICMSGHDRFLAG_TRANSACTION
456 | ICMSGHDRFLAG_RESPONSE
;
458 vmbus_sendpacket(channel
, hbeat_txf_buf
,
460 VM_PKT_DATA_INBAND
, 0);
464 static int util_probe(struct hv_device
*dev
,
465 const struct hv_vmbus_device_id
*dev_id
)
467 struct hv_util_service
*srv
=
468 (struct hv_util_service
*)dev_id
->driver_data
;
471 srv
->recv_buffer
= kmalloc(HV_HYP_PAGE_SIZE
* 4, GFP_KERNEL
);
472 if (!srv
->recv_buffer
)
474 srv
->channel
= dev
->channel
;
475 if (srv
->util_init
) {
476 ret
= srv
->util_init(srv
);
484 * The set of services managed by the util driver are not performance
485 * critical and do not need batched reading. Furthermore, some services
486 * such as KVP can only handle one message from the host at a time.
487 * Turn off batched reading for all util drivers before we open the
490 set_channel_read_mode(dev
->channel
, HV_CALL_DIRECT
);
492 hv_set_drvdata(dev
, srv
);
494 ret
= vmbus_open(dev
->channel
, 4 * HV_HYP_PAGE_SIZE
,
495 4 * HV_HYP_PAGE_SIZE
, NULL
, 0, srv
->util_cb
,
503 if (srv
->util_deinit
)
506 kfree(srv
->recv_buffer
);
510 static int util_remove(struct hv_device
*dev
)
512 struct hv_util_service
*srv
= hv_get_drvdata(dev
);
514 if (srv
->util_deinit
)
516 vmbus_close(dev
->channel
);
517 kfree(srv
->recv_buffer
);
523 * When we're in util_suspend(), all the userspace processes have been frozen
524 * (refer to hibernate() -> freeze_processes()). The userspace is thawed only
525 * after the whole resume procedure, including util_resume(), finishes.
527 static int util_suspend(struct hv_device
*dev
)
529 struct hv_util_service
*srv
= hv_get_drvdata(dev
);
532 if (srv
->util_pre_suspend
) {
533 ret
= srv
->util_pre_suspend();
538 vmbus_close(dev
->channel
);
543 static int util_resume(struct hv_device
*dev
)
545 struct hv_util_service
*srv
= hv_get_drvdata(dev
);
548 if (srv
->util_pre_resume
) {
549 ret
= srv
->util_pre_resume();
554 ret
= vmbus_open(dev
->channel
, 4 * HV_HYP_PAGE_SIZE
,
555 4 * HV_HYP_PAGE_SIZE
, NULL
, 0, srv
->util_cb
,
560 static const struct hv_vmbus_device_id id_table
[] = {
563 .driver_data
= (unsigned long)&util_shutdown
565 /* Time synch guid */
567 .driver_data
= (unsigned long)&util_timesynch
570 { HV_HEART_BEAT_GUID
,
571 .driver_data
= (unsigned long)&util_heartbeat
575 .driver_data
= (unsigned long)&util_kvp
579 .driver_data
= (unsigned long)&util_vss
583 .driver_data
= (unsigned long)&util_fcopy
588 MODULE_DEVICE_TABLE(vmbus
, id_table
);
590 /* The one and only one */
591 static struct hv_driver util_drv
= {
593 .id_table
= id_table
,
595 .remove
= util_remove
,
596 .suspend
= util_suspend
,
597 .resume
= util_resume
,
599 .probe_type
= PROBE_PREFER_ASYNCHRONOUS
,
603 static int hv_ptp_enable(struct ptp_clock_info
*info
,
604 struct ptp_clock_request
*request
, int on
)
609 static int hv_ptp_settime(struct ptp_clock_info
*p
, const struct timespec64
*ts
)
614 static int hv_ptp_adjfreq(struct ptp_clock_info
*ptp
, s32 delta
)
618 static int hv_ptp_adjtime(struct ptp_clock_info
*ptp
, s64 delta
)
623 static int hv_ptp_gettime(struct ptp_clock_info
*info
, struct timespec64
*ts
)
625 *ts
= hv_get_adj_host_time();
630 static struct ptp_clock_info ptp_hyperv_info
= {
632 .enable
= hv_ptp_enable
,
633 .adjtime
= hv_ptp_adjtime
,
634 .adjfreq
= hv_ptp_adjfreq
,
635 .gettime64
= hv_ptp_gettime
,
636 .settime64
= hv_ptp_settime
,
637 .owner
= THIS_MODULE
,
640 static struct ptp_clock
*hv_ptp_clock
;
642 static int hv_timesync_init(struct hv_util_service
*srv
)
644 /* TimeSync requires Hyper-V clocksource. */
645 if (!hv_read_reference_counter
)
648 spin_lock_init(&host_ts
.lock
);
650 INIT_WORK(&adj_time_work
, hv_set_host_time
);
653 * ptp_clock_register() returns NULL when CONFIG_PTP_1588_CLOCK is
654 * disabled but the driver is still useful without the PTP device
655 * as it still handles the ICTIMESYNCFLAG_SYNC case.
657 hv_ptp_clock
= ptp_clock_register(&ptp_hyperv_info
, NULL
);
658 if (IS_ERR_OR_NULL(hv_ptp_clock
)) {
659 pr_err("cannot register PTP clock: %ld\n",
660 PTR_ERR(hv_ptp_clock
));
667 static void hv_timesync_cancel_work(void)
669 cancel_work_sync(&adj_time_work
);
672 static int hv_timesync_pre_suspend(void)
674 hv_timesync_cancel_work();
678 static void hv_timesync_deinit(void)
681 ptp_clock_unregister(hv_ptp_clock
);
683 hv_timesync_cancel_work();
686 static int __init
init_hyperv_utils(void)
688 pr_info("Registering HyperV Utility Driver\n");
690 return vmbus_driver_register(&util_drv
);
693 static void exit_hyperv_utils(void)
695 pr_info("De-Registered HyperV Utility Driver\n");
697 vmbus_driver_unregister(&util_drv
);
700 module_init(init_hyperv_utils
);
701 module_exit(exit_hyperv_utils
);
703 MODULE_DESCRIPTION("Hyper-V Utilities");
704 MODULE_LICENSE("GPL");