| 1 | /* |
| 2 | * MSHV partition reference clock |
| 3 | * |
| 4 | * Copyright Microsoft, Corp. 2026 |
| 5 | * |
| 6 | * Authors: Magnus Kulke <magnuskulke@microsoft.com> |
| 7 | * |
| 8 | * SPDX-License-Identifier: GPL-2.0-or-later |
| 9 | */ |
| 10 | |
| 11 | #include "qemu/osdep.h" |
| 12 | #include "qemu/error-report.h" |
| 13 | #include "migration/vmstate.h" |
| 14 | #include "system/runstate.h" |
| 15 | #include "hw/hyperv/hvhdk.h" |
| 16 | #include "hw/hyperv/hvhdk_mini.h" |
| 17 | #include "hw/hyperv/hvgdk.h" |
| 18 | #include "hw/hyperv/hvgdk_mini.h" |
| 19 | #include "linux/mshv.h" |
| 20 | #include "system/mshv.h" |
| 21 | #include "system/mshv_int.h" |
| 22 | |
| 23 | /* |
| 24 | * Partition reference clock (HV_PARTITION_PROPERTY_REFERENCE_TIME), captured |
| 25 | * when the VM is stopped and re-applied when it resumes. |
| 26 | * |
| 27 | * Mirrors hw/i386/kvm/clock.c. |
| 28 | */ |
| 29 | typedef struct MshvClockState { |
| 30 | uint64_t ref_time; |
| 31 | bool ref_time_pending; |
| 32 | } MshvClockState; |
| 33 | |
| 34 | static MshvClockState mshv_clock; |
| 35 | |
| 36 | static int mshv_get_reference_time(int vm_fd, uint64_t *ref_time) |
| 37 | { |
| 38 | struct hv_input_get_partition_property in = { 0 }; |
| 39 | struct hv_output_get_partition_property out = { 0 }; |
| 40 | struct mshv_root_hvcall args = { 0 }; |
| 41 | int ret; |
| 42 | |
| 43 | in.property_code = HV_PARTITION_PROPERTY_REFERENCE_TIME; |
| 44 | |
| 45 | args.code = HVCALL_GET_PARTITION_PROPERTY; |
| 46 | args.in_sz = sizeof(in); |
| 47 | args.in_ptr = (uint64_t)∈ |
| 48 | args.out_sz = sizeof(out); |
| 49 | args.out_ptr = (uint64_t)&out; |
| 50 | |
| 51 | ret = mshv_hvcall(vm_fd, &args); |
| 52 | if (ret < 0) { |
| 53 | error_report("Failed to get reference time"); |
| 54 | return -1; |
| 55 | } |
| 56 | |
| 57 | *ref_time = out.property_value; |
| 58 | return 0; |
| 59 | } |
| 60 | |
| 61 | static int mshv_set_reference_time(int vm_fd, uint64_t ref_time) |
| 62 | { |
| 63 | struct hv_input_set_partition_property in = { 0 }; |
| 64 | struct mshv_root_hvcall args = { 0 }; |
| 65 | int ret; |
| 66 | |
| 67 | in.property_code = HV_PARTITION_PROPERTY_REFERENCE_TIME; |
| 68 | in.property_value = ref_time; |
| 69 | |
| 70 | args.code = HVCALL_SET_PARTITION_PROPERTY; |
| 71 | args.in_sz = sizeof(in); |
| 72 | args.in_ptr = (uint64_t)∈ |
| 73 | |
| 74 | ret = mshv_hvcall(vm_fd, &args); |
| 75 | if (ret < 0) { |
| 76 | error_report("Failed to set reference time"); |
| 77 | return -1; |
| 78 | } |
| 79 | |
| 80 | return 0; |
| 81 | } |
| 82 | |
| 83 | /* |
| 84 | * Freeze (freeze=1) or unfreeze (freeze=0) time for a partition. This will not |
| 85 | * pause/eject vCPU execution. It is assumed that the caller already has stopped |
| 86 | * the partition's vCPUs. |
| 87 | * |
| 88 | * NB: a partition's reference clock can only be written while the time is |
| 89 | * frozen. |
| 90 | */ |
| 91 | static int mshv_set_time_freeze(int vm_fd, int freeze) |
| 92 | { |
| 93 | struct hv_input_set_partition_property in = { 0 }; |
| 94 | struct mshv_root_hvcall args = { 0 }; |
| 95 | int ret; |
| 96 | |
| 97 | in.property_code = HV_PARTITION_PROPERTY_TIME_FREEZE; |
| 98 | in.property_value = freeze; |
| 99 | |
| 100 | args.code = HVCALL_SET_PARTITION_PROPERTY; |
| 101 | args.in_sz = sizeof(in); |
| 102 | args.in_ptr = (uint64_t)∈ |
| 103 | |
| 104 | ret = mshv_hvcall(vm_fd, &args); |
| 105 | if (ret < 0) { |
| 106 | error_report("Failed to set time freeze"); |
| 107 | return -1; |
| 108 | } |
| 109 | |
| 110 | return 0; |
| 111 | } |
| 112 | |
| 113 | static void mshv_clock_vm_state_change(void *opaque, bool running, |
| 114 | RunState state) |
| 115 | { |
| 116 | MshvClockState *s = opaque; |
| 117 | int vm_fd = mshv_state->vm; |
| 118 | int ret; |
| 119 | |
| 120 | if (running) { |
| 121 | /* Skip if we have nothing to restore, e.g. on initial boot. */ |
| 122 | if (!s->ref_time_pending) { |
| 123 | return; |
| 124 | } |
| 125 | |
| 126 | ret = mshv_set_time_freeze(vm_fd, 1); |
| 127 | if (ret < 0) { |
| 128 | error_report("Failed to freeze partition time on resume"); |
| 129 | abort(); |
| 130 | } |
| 131 | |
| 132 | /* INVARIANT: reference time can only be written if time is frozen. */ |
| 133 | ret = mshv_set_reference_time(vm_fd, s->ref_time); |
| 134 | if (ret < 0) { |
| 135 | error_report("Failed to restore reference time on resume"); |
| 136 | abort(); |
| 137 | } |
| 138 | |
| 139 | if (mshv_set_time_freeze(vm_fd, 0) < 0) { |
| 140 | error_report("Failed to unfreeze partition time on resume"); |
| 141 | abort(); |
| 142 | } |
| 143 | |
| 144 | s->ref_time_pending = false; |
| 145 | } else { |
| 146 | /* Skip if we already have a to-be-set ref time */ |
| 147 | if (s->ref_time_pending) { |
| 148 | return; |
| 149 | } |
| 150 | |
| 151 | ret = mshv_get_reference_time(vm_fd, &s->ref_time); |
| 152 | if (ret < 0) { |
| 153 | error_report("Failed to capture reference time on stop"); |
| 154 | abort(); |
| 155 | } |
| 156 | |
| 157 | s->ref_time_pending = true; |
| 158 | } |
| 159 | } |
| 160 | |
| 161 | /* |
| 162 | * The incoming reference time should be applied on the next resume before |
| 163 | * vCPUs start executing. |
| 164 | */ |
| 165 | static int mshv_clock_post_load(void *opaque, int version_id) |
| 166 | { |
| 167 | MshvClockState *s = opaque; |
| 168 | |
| 169 | s->ref_time_pending = true; |
| 170 | |
| 171 | return 0; |
| 172 | } |
| 173 | |
| 174 | static const VMStateDescription vmstate_mshv_clock = { |
| 175 | .name = "mshv-clock", |
| 176 | .version_id = 1, |
| 177 | .minimum_version_id = 1, |
| 178 | .post_load = mshv_clock_post_load, |
| 179 | .fields = (const VMStateField[]) { |
| 180 | VMSTATE_UINT64(ref_time, MshvClockState), |
| 181 | VMSTATE_END_OF_LIST() |
| 182 | }, |
| 183 | }; |
| 184 | |
| 185 | void mshv_clock_init(void) |
| 186 | { |
| 187 | vmstate_register(NULL, 0, &vmstate_mshv_clock, &mshv_clock); |
| 188 | qemu_add_vm_change_state_handler(mshv_clock_vm_state_change, &mshv_clock); |
| 189 | } |