On Fri, Nov 27, 2020 at 04:53:59PM +0000, Peter Maydell wrote: > On Fri, 27 Nov 2020 at 16:47, Eduardo Habkost <ehabk...@redhat.com> wrote: > > Do you know how the implicitly-accelerator-specific code is > > implemented? PARange is in id_aa64mmfr0, correct? I don't see > > any accel-specific code for initializing id_aa64mmfr0. > > For TCG, the value of id_aa64mmfr0 is set by the per-cpu > init functions aarch64_a57_initfn(), aarch64_a72_initfn(), etc. > > For KVM, if we're using "-cpu cortex-a53" or "-cpu cortex-a57" > these only work if the host CPU really is an A53 or A57, in > which case the reset value set by the initfn is correct. > In the more usual case of "-cpu host", we ask the kernel for > the ID register values in kvm_arm_get_host_cpu_features(), > which is part of the implementation of > kvm_arm_set_cpu_features_from_host(), which gets called > in aarch64_max_initfn() (inside a kvm_enabled() conditional). > > So there is a *_enabled() check involved, which I hadn't > realised until I worked back up to where this stuff is called. >
Thanks! Is the data returned by kvm_arm_get_host_cpu_features() supposed to eventually affect the value of id_aa64mmfr0? I don't see how that could happen. -- Eduardo
