On Tue, Nov 14, 2023 at 12:43:38AM +0000, Li, Xin3 wrote: > No. tglx asked for it: > https://lkml.kernel.org/kvm/87y1h81ht4.ffs@tglx/ Aha "According to the CPU folks FRED systems are guaranteed to have WRMSRNS - I asked for that :). It's just not yet documented." so I'm going to expect that to appear in the next FRED spec revision... > Because we are doing > wrmsrns(MSR_IA32_FRED_RSP0, ...) > here, and X86_FEATURE_WRMSRNS doesn't guarantee MSR_IA32_FRED_RSP0 exists. > > Or I missed something? Well, according to what I'm hearing and reading so far: FRED means WRMSRNS FRED means MSR_IA32_FRED_RSP0 and if you had to be precise, the code should do: if (cpu_feature_enabled(X86_FEATURE_FRED)) { if (cpu_feature_enabled(X86_FEATURE_WRMSRNS)) wrmsrns(MSR_IA32_FRED_RSP0, (unsigned long)task_stack_page(task) + THREAD_SIZE); else wrmsr(MSR_IA32_FRED_RSP0, (unsigned long)task_stack_page(task) + THREAD_SIZE); } but apparently FRED implies WRMSRNS - not documented anywhere currently - so you can save yourself one check. But your version checks FRED if it can do WRMSRNS while there's a separate WRMSRNS flag and that made me wonder... > Another patch set should replace WRMSR with WRMSRNS, with SERIALIZE added > when needed. I sense someone wants to optimize MSR writes ... :-) Thx. -- Regards/Gruss, Boris. https://people.kernel.org/tglx/notes-about-netiquette
