Re: [tip: ras/core] x86/mce: Enable additional error logging on certain Intel CPUs

[Paolo Bonzini]

> Maybe no contract ... but a bunch of places (many of them in Intel
> specific code) that check for it

Interestingly, quite a few of them are actually checking for HYPERVISOR 
not because of missing hypervisor features, but rather because 
hypervisors don't have to work around certain errata. :)

Full analysis after my sig, but tl;dr: the only case of using HYPERVISOR 
before using MSRs are in arch/x86/events/intel/cstate.c and 
arch/x86/events/intel/uncore.c.  There are some workarounds in 
drivers/gpu that might fall into a similar bucket.  But as far as MSRs 
go, the way to go  overwhelmingly seems to be {rd,wr}msrl_safe.

Thanks,

Paolo

On 10/11/20 00:36, Luck, Tony wrote:
> arch/x86/events/core.c: if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {

Print a slightly less frightening warning.

> arch/x86/events/intel/core.c:   if (!boot_cpu_has(X86_FEATURE_HYPERVISOR))

Working around KVM's ignore_msrs=1 option (and quite ugly; shows that 
the option shouldn't be enabled by default).

> arch/x86/events/intel/core.c:           int assume = 3 * !boot_cpu_has(X86_FEATURE_HYPERVISOR);

Seems unnecessary.

> arch/x86/events/intel/cstate.c: if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
> arch/x86/events/intel/uncore.c: if (boot_cpu_has(X86_FEATURE_HYPERVISOR))

Too complicated. :)

> arch/x86/kernel/apic/apic.c:    if (boot_cpu_has(X86_FEATURE_HYPERVISOR))

Hypervisors don't have errata.

> arch/x86/kernel/cpu/bugs.c:     if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
> arch/x86/kernel/cpu/bugs.c:     else if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
> arch/x86/kernel/cpu/bugs.c:     if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
> arch/x86/kernel/cpu/bugs.c:     if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
> arch/x86/kernel/cpu/intel.c:    if (boot_cpu_has(X86_FEATURE_HYPERVISOR))

Print different vulnerability status in sysfs.

> arch/x86/kernel/cpu/mshyperv.c: if (!boot_cpu_has(X86_FEATURE_HYPERVISOR))
> arch/x86/kernel/cpu/vmware.c: * If !boot_cpu_has(X86_FEATURE_HYPERVISOR), vmware_hypercall_mode
> arch/x86/kernel/cpu/vmware.c:   if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
> arch/x86/kernel/jailhouse.c:        !boot_cpu_has(X86_FEATURE_HYPERVISOR))
> arch/x86/kernel/kvm.c:  if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
> arch/x86/kernel/paravirt.c:     if (!boot_cpu_has(X86_FEATURE_HYPERVISOR))

Obviously needed before using paravirt features of the hypervisor.

> arch/x86/kernel/tsc.c:  if (boot_cpu_has(X86_FEATURE_HYPERVISOR) ||

Disables ART in VMs.  Probably the idea is that ART does not have an 
offset field similar to the TSC's, but it's not necessary.  Looking at 
the hypervisor-provided CPUID should be enough.

> arch/x86/mm/init_64.c:  if (!boot_cpu_has(X86_FEATURE_HYPERVISOR)) {

Tweaks hotplug heuristics, no MSRs involved.

> drivers/acpi/processor_idle.c:  if (boot_cpu_has(X86_FEATURE_HYPERVISOR))

Avoids pointless hypervisor exit on idle (i.e. just an optimization).

> drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h:       return boot_cpu_has(X86_FEATURE_HYPERVISOR);

Work around SR-IOV bugs.

> drivers/gpu/drm/i915/i915_memcpy.c:         !boot_cpu_has(X86_FEATURE_HYPERVISOR))

Work around KVM deficiency.

> drivers/gpu/drm/radeon/radeon_device.c: return boot_cpu_has(X86_FEATURE_HYPERVISOR);

Work around SR-IOV bugs.

> drivers/visorbus/visorchipset.c:        if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {

Needed before using paravirt features of the hypervisor.