Re: [PATCH] x86/tsx: clear RTM and HLE when MSR_IA32_TSX_CTRL is not supported

[Pawan Gupta]

On 07.09.2021 14:56, Hao Peng wrote:
> On Tue, Sep 7, 2021 at 1:36 PM Pawan Gupta
> <pawan.kumar.gupta@xxxxxxxxxxxxxxx> wrote:
> > On 07.09.2021 12:39, Hao Peng wrote:
> > >On Tue, Sep 7, 2021 at 12:26 PM Xiaoyao Li <xiaoyao.li@xxxxxxxxx> wrote:
> > >>
> > >> On 9/7/2021 11:40 AM, Hao Peng wrote:
> > >> > On Tue, Sep 7, 2021 at 10:56 AM Xiaoyao Li <xiaoyao.li@xxxxxxxxx> wrote:
> > >> >>
> > >> >> On 9/7/2021 10:35 AM, Hao Peng wrote:
> > >> >>> On Tue, Sep 7, 2021 at 10:08 AM Xiaoyao Li <xiaoyao.li@xxxxxxxxx> wrote:
> > >> >>>>
> > >> >>>> On 9/7/2021 9:47 AM, Hao Peng wrote:
> > >> >>>>> On Mon, Sep 6, 2021 at 5:30 PM Borislav Petkov <bp@xxxxxxxxx> wrote:
> > >> >>>>>>
> > >> >>>>>> On Mon, Sep 06, 2021 at 10:46:05AM +0800, Hao Peng wrote:
> > >> >>>>>>> If hypervisor does not support MSR_IA32_TSX_CTRL, but guest supports
> > >> >>>>>>> RTM and HLE features, it will affect TAA mitigation.
> > >> >>>>>>>
> > >> >>>>>>> Signed-off-by: Peng Hao <flyingpeng@xxxxxxxxxxx>
> > >> >>>>>>> ---
> > >> >>>>>>>     arch/x86/kernel/cpu/tsx.c | 7 +++++++
> > >> >>>>>>>     1 file changed, 7 insertions(+)
> > >> >>>>>>>
> > >> >>>>>>> diff --git a/arch/x86/kernel/cpu/tsx.c b/arch/x86/kernel/cpu/tsx.c
> > >> >>>>>>> index 9c7a5f049292..5e852c14fef2 100644
> > >> >>>>>>> --- a/arch/x86/kernel/cpu/tsx.c
> > >> >>>>>>> +++ b/arch/x86/kernel/cpu/tsx.c
> > >> >>>>>>> @@ -122,6 +122,13 @@ void __init tsx_init(void)
> > >> >>>>>>>
> > >> >>>>>>>            if (!tsx_ctrl_is_supported()) {
> > >> >>>>>>>                    tsx_ctrl_state = TSX_CTRL_NOT_SUPPORTED;
> > >> >>>>>>> +
> > >> >>>>>>> +               /* If hypervisor does not support MSR_IA32_TSX_CTRL emulation,
> > >> >>>>>>> +                * but guest supports RTM and HLE features, it will affect TAA
> > >> >>>>>>> +                * （tsx_async_abort）mitigation.
> > >> >>>>>>> +                */
> > >> >>>>>>> +               setup_clear_cpu_cap(X86_FEATURE_RTM);
> > >> >>>>>>> +               setup_clear_cpu_cap(X86_FEATURE_HLE);
> > >> >>>>
> > >> >>>> anyway, IMHO, we shouldn't do anything here for TAA. It should be in
> > >> >>>> taa_select_mitigation()
> > >> >>>>
> > >> >>>>>>>                    return;
> > >> >>>>>>>            }
> > >> >>>>>>
> > >> >>>>>> How does that even happen - the hypervisor does not support the MSR but
> > >> >>>>>> "guest supports" TSX features?!
> > >> >>>>>>
> > >> >>>>>> I guess the guest is detecting it wrong.
> > >> >>>>>>
> > >> >>>>>> What hypervisor, what guest, how do I reproduce?
> > >> >>>>>>
> > >> >>>>> hypervisor is kvm, guest is linux too.
> > >> >>>>>> Please give full details.
> > >> >>>>>>
> > >> >>>>> The host I used is kernel-5.4, and guest is kernel-5.13.
> > >> >>>>> MSR_IA32_TSX_CTRL is exposed
> > >> >>>>> to guest and guest to support RTM and HLE features, no direct
> > >> >>>>> dependence. at the qemu I
> > >> >>>>> started guest with -cpu host-model.
> > >> >>>>> I have viewed the code of kernel-5.4, and MSR_IA32_TSX_CTRL is not
> > >> >>>>> exposed to guest.
> > >> >>>>
> > >> >>>> Does guest see TAA_NO bit?
> > >> >>>>
> > >> >>> Guest can't see taa_no, which requires updating qemu to solve. But I think
> > >> >>> there is a compatibility process here.
> > >> >>
> > >> >> Anyway, there should be some existing code in kernel already to handle
> > >> >> the case that CPUID reports TRM while MSR_IA32_CORE_CAPABILITIES doesn't
> > >> >> report MSR_TSX_CTRL nor TAA_NO.
> > >> >>
> > >> > Can you point out which patches ? At present, guest is kernel-5.13
> > >> > still has this problem.
> > >>
> > >> What's the output of 'cat
> > >> /sys/devices/system/cpu/vulnerabilities/tsx_async_abort' on your guest?
> > >>
> > >Vulnerable: Clear CPU buffers attempted, no microcode; SMT Host state unknown.
> >
> > This suggests that the microcode is old. Can you please share the output
> > of below cmds on host and guest:
> >
> >         $ grep . /sys/devices/system/cpu/vulnerabilities/*
> host : not affected
> guest: Vulnerable: Clear CPU buffers attempted, no microcode; SMT Host
> state unknown.

This would mean TAA_NO is not being exported to guest. Most likely qemu
needs to be patched for the CPU configuration you are using.

Thanks,
Pawan