Hi Peter,
On 10/13/26 12:51 PM, Peter Maydell wrote:
> Currently when using KVM on Arm, the normal thing is to use either
> the 'host' or 'max' CPU types, both of which give the guest the same
> CPU as the host. We also for legacy reasons allow 'cortex-a53' and
> 'cortex-a57' CPU types, which will only work when the host CPU is
> exactly that matching CPU. This is because the original idea of the
> KVM ABI was that userspace would request a particular CPU type by
> setting the target field in the KVM_ARM_VCPU_INIT ioctl to an
> appropriate KVM_ARM_TARGET_* constant. However, it quickly became
> clear that this was a bad idea, and now everything uses
> KVM_ARM_TARGET_GENERIC_V8; the tiny number of existing CPU-specific
> target values behave identically to GENERIC_V8 except that they will
> only work if the host CPU matches them.
>
> Future support for "show the KVM guest a CPU type that's not
> identical to the host" is going to work by telling it different ID
> register values. Real Cortex-A53 and A57 CPUs are now all over a
> decade old and not in widespread use. Deprecate the ability to use
> '-cpu cortex-a53' and '-cpu cortex-a57' with KVM, to get rid of this
> odd legacy corner and open up the opportunity in future to support
> named CPU types via ID register settings if we want that.
>
> For any situation where the cortex-a53 and cortex-a57 CPU types
> currently work with KVM, '-cpu host' or '-cpu max' will have
> identical behaviour. (Though note that if you are using "KVM, but
> fall back to TCG" then the fallback under TCG will be different.)
>
> Signed-off-by: Peter Maydell <[email protected]>
Reviewed-by: Eric Auger <[email protected]>
Thanks
Eric
> ---
> Personally I doubt anybody is actually using these named
> CPU types...
> ---
> docs/about/deprecated.rst | 13 +++++++++++++
> target/arm/kvm.c | 7 +++++++
> 2 files changed, 20 insertions(+)
>
> diff --git a/docs/about/deprecated.rst b/docs/about/deprecated.rst
> index ccbc4219497..153fe118151 100644
> --- a/docs/about/deprecated.rst
> +++ b/docs/about/deprecated.rst
> @@ -234,6 +234,19 @@ in the QEMU object model anymore.
> ``Sun-UltraSparc-IIIi+`` and
> but for consistency these will get removed in a future release, too.
> Use ``Sun-UltraSparc-IIIi-plus`` and ``Sun-UltraSparc-IV-plus`` instead.
>
> +``cortex-a53`` and ``cortex-a57`` CPU names with KVM on Arm (since 11.2)
> +''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''
> +
> +QEMU currently allows the CPU name to be set to ``cortex-a53`` or
> +``cortex-a57`` when using KVM; this is identical to using ``host`` or
> +``max``, but it will only work when the host CPU is a Cortex-A53 or
> +Cortex-A57. These CPUs are now over 10 years old, and the ability to
> +use the named CPU type like this is part of a legacy corner of the KVM
> +API, which doesn't work with any other host CPU types. Using these CPU
> +names with KVM enabled is deprecated; use ``host`` or ``max`` instead,
> +which will have identical behaviour on the hosts where using
> +``cortex-a53`` or ``cortex-a57`` works.
> +
> System emulator machines
> ------------------------
>
> diff --git a/target/arm/kvm.c b/target/arm/kvm.c
> index 3d905f8565d..de09d63d2e3 100644
> --- a/target/arm/kvm.c
> +++ b/target/arm/kvm.c
> @@ -2015,6 +2015,13 @@ int kvm_arch_init_vcpu(CPUState *cs)
> return -EINVAL;
> }
>
> + if (cpu->kvm_target == QEMU_KVM_ARM_TARGET_CORTEX_A53 ||
> + cpu->kvm_target == QEMU_KVM_ARM_TARGET_CORTEX_A57) {
> + warn_report_once("Using the cortex-a53 and cortex-a57 CPU types "
> + "with KVM is deprecated. Use the max or host CPU "
> + "types instead.");
> + }
> +
> qemu_add_vm_change_state_handler(kvm_arm_vm_state_change, cpu);
>
> /* Determine init features for this CPU */
