On 4/29/25 14:49, Marc Zyngier wrote:
On Tue, 29 Apr 2025 14:08:27 +0100,
Ben Horgan <ben.horgan@xxxxxxx> wrote:
- __deactivate_fgt(hctxt, vcpu, kvm, HFGRTR_EL2);
- if (cpus_have_final_cap(ARM64_WORKAROUND_AMPERE_AC03_CPU_38))
Don't we need to continue considering the ampere errata here? Or, at
least worth a mention in the commit message.
The FGT registers are always context switched, so whatever was saved
*before* the workaround was applied in __activate_traps_hfgxtr() is
blindly restored...
- write_sysreg_s(ctxt_sys_reg(hctxt, HFGWTR_EL2), SYS_HFGWTR_EL2);
... and this write always happens.
Thanks for the explanation. I now agree this code is correct.
M.
Thanks,
Ben
