| 1 | |
| 2 |
|
| 3 | Clocks and Timers
|
| 4 | =================
|
| 5 |
|
| 6 | arm64
|
| 7 | -----
|
| 8 | On arm64, Hyper-V virtualizes the ARMv8 architectural system counter
|
| 9 | and timer. Guest VMs use this virtualized hardware as the Linux
|
| 10 | clocksource and clockevents via the standard arm_arch_timer.c
|
| 11 | driver, just as they would on bare metal. Linux vDSO support for the
|
| 12 | architectural system counter is functional in guest VMs on Hyper-V.
|
| 13 | While Hyper-V also provides a synthetic system clock and four synthetic
|
| 14 | per-CPU timers as described in the TLFS, they are not used by the
|
| 15 | Linux kernel in a Hyper-V guest on arm64. However, older versions
|
| 16 | of Hyper-V for arm64 only partially virtualize the ARMv8
|
| 17 | architectural timer, such that the timer does not generate
|
| 18 | interrupts in the VM. Because of this limitation, running current
|
| 19 | Linux kernel versions on these older Hyper-V versions requires an
|
| 20 | out-of-tree patch to use the Hyper-V synthetic clocks/timers instead.
|
| 21 |
|
| 22 | x86/x64
|
| 23 | -------
|
| 24 | On x86/x64, Hyper-V provides guest VMs with a synthetic system clock
|
| 25 | and four synthetic per-CPU timers as described in the TLFS. Hyper-V
|
| 26 | also provides access to the virtualized TSC via the RDTSC and
|
| 27 | related instructions. These TSC instructions do not trap to
|
| 28 | the hypervisor and so provide excellent performance in a VM.
|
| 29 | Hyper-V performs TSC calibration, and provides the TSC frequency
|
| 30 | to the guest VM via a synthetic MSR. Hyper-V initialization code
|
| 31 | in Linux reads this MSR to get the frequency, so it skips TSC
|
| 32 | calibration and sets tsc_reliable. Hyper-V provides virtualized
|
| 33 | versions of the PIT (in Hyper-V Generation 1 VMs only), local
|
| 34 | APIC timer, and RTC. Hyper-V does not provide a virtualized HPET in
|
| 35 | guest VMs.
|
| 36 |
|
| 37 | The Hyper-V synthetic system clock can be read via a synthetic MSR,
|
| 38 | but this access traps to the hypervisor. As a faster alternative,
|
| 39 | the guest can configure a memory page to be shared between the guest
|
| 40 | and the hypervisor. Hyper-V populates this memory page with a
|
| 41 | 64-bit scale value and offset value. To read the synthetic clock
|
| 42 | value, the guest reads the TSC and then applies the scale and offset
|
| 43 | as described in the Hyper-V TLFS. The resulting value advances
|
| 44 | at a constant 10 MHz frequency. In the case of a live migration
|
| 45 | to a host with a different TSC frequency, Hyper-V adjusts the
|
| 46 | scale and offset values in the shared page so that the 10 MHz
|
| 47 | frequency is maintained.
|
| 48 |
|
| 49 | Starting with Windows Server 2022 Hyper-V, Hyper-V uses hardware
|
| 50 | support for TSC frequency scaling to enable live migration of VMs
|
| 51 | across Hyper-V hosts where the TSC frequency may be different.
|
| 52 | When a Linux guest detects that this Hyper-V functionality is
|
| 53 | available, it prefers to use Linux's standard TSC-based clocksource.
|
| 54 | Otherwise, it uses the clocksource for the Hyper-V synthetic system
|
| 55 | clock implemented via the shared page (identified as
|
| 56 | "hyperv_clocksource_tsc_page").
|
| 57 |
|
| 58 | The Hyper-V synthetic system clock is available to user space via
|
| 59 | vDSO, and gettimeofday() and related system calls can execute
|
| 60 | entirely in user space. The vDSO is implemented by mapping the
|
| 61 | shared page with scale and offset values into user space. User
|
| 62 | space code performs the same algorithm of reading the TSC and
|
| 63 | applying the scale and offset to get the constant 10 MHz clock.
|
| 64 |
|
| 65 | Linux clockevents are based on Hyper-V synthetic timer 0 (stimer0).
|
| 66 | While Hyper-V offers 4 synthetic timers for each CPU, Linux only uses
|
| 67 | timer 0. In older versions of Hyper-V, an interrupt from stimer0
|
| 68 | results in a VMBus control message that is demultiplexed by
|
| 69 | vmbus_isr() as described in the Documentation/virt/hyperv/vmbus.rst
|
| 70 | documentation. In newer versions of Hyper-V, stimer0 interrupts can
|
| 71 | be mapped to an architectural interrupt, which is referred to as
|
| 72 | "Direct Mode". Linux prefers to use Direct Mode when available. Since
|
| 73 | x86/x64 doesn't support per-CPU interrupts, Direct Mode statically
|
| 74 | allocates an x86 interrupt vector (HYPERV_STIMER0_VECTOR) across all CPUs
|
| 75 | and explicitly codes it to call the stimer0 interrupt handler. Hence
|
| 76 | interrupts from stimer0 are recorded on the "HVS" line in /proc/interrupts
|
| 77 | rather than being associated with a Linux IRQ. Clockevents based on the
|
| 78 | virtualized PIT and local APIC timer also work, but Hyper-V stimer0
|
| 79 | is preferred.
|
| 80 |
|
| 81 | The driver for the Hyper-V synthetic system clock and timers is
|
| 82 | drivers/clocksource/hyperv_timer.c.
|