CVM: Set GuestCrashRegsAvailable in cpuid

[smalis-msft]

We support these regs on all backends, so no reason not to.

[chris-oo]

Do we need to maybe filter offering this diagnostic on if cvm tracing is enabled?
This is what I see now in a debug build:

[kmsg]: [2.166324] virt_mshv_vtl::processor: INFO  Guest has reported system crash crash=VtlCrash { vp_index: VpIndex(0), last_vtl: Vtl0, control: GuestCrashCtl { pre_os_id: 0, no_crash_dump: false, crash_message: true, crash_notify: true }, parameters: [800000000000001a, 0, 72e, 33d85cb0, 85] }
[kmsg]: [2.166707] underhill_core::dispatch: INFO  Notifying the host of the guest system crash VtlCrash { vp_index: VpIndex(0), last_vtl: Vtl0, control: GuestCrashCtl { pre_os_id: 0, no_crash_dump: false, crash_message: true, crash_notify: true }, parameters: [800000000000001a, 0, 72e, 33d85cb0, 85] }
[kmsg]: [2.167256] firmware_uefi: WARN  unknown uefi write addr=0x2c data=0x1
[kmsg]: [2.167828] vmm_core::partition_unit::vp_set: ERROR  rax=0x800000000000001a rcx=0x72e rdx=0x0 rbx=0x0 rsp=0x33d85c58 rbp=0x33d86d10 rsi=0x66083198 rdi=0x72e r8=0x72e r9=0x0 r10=0x0 r11=0x1b r12=0x80000000 r13=0x33db2458 r14=0x1 r15=0x80000000 rip=0x66af0c34 rflags=0x10206
[kmsg]: [2.168161] vmm_core::partition_unit::vp_set: ERROR  cs=SegmentRegister { base: 0, limit: ffffffff, selector: 38, attributes: a09b } ds=SegmentRegister { base: 0, limit: ffffffff, selector: 30, attributes: c093 } es=SegmentRegister { base: 0, limit: ffffffff, selector: 30, attributes: c093 } fs=SegmentRegister { base: 0, limit: ffffffff, selector: 30, attributes: c093 } gs=SegmentRegister { base: 0, limit: ffffffff, selector: 30, attributes: c093 } ss=SegmentRegister { base: 0, limit: ffffffff, selector: 30, attributes: c093 } tr=SegmentRegister { base: 33dbca30, limit: 67, selector: 20, attributes: 8b } ldtr=SegmentRegister { base: 0, limit: ffff, selector: 0, attributes: 82 } gdtr=TableRegister { base: 66bc7000, limit: 47 } idtr=TableRegister { base: 0, limit: 0 } cr0=0x80010033 cr2=0x0 cr3=0x33801000 cr4=0x660 cr8=0x0 efer=0xd00
[kmsg]: [2.169456] vmm_core::partition_unit::vp_set: ERROR  faulting instruction instruction="ud2" bytes=[f, b, eb, fc, cc, cc, cc, cc, cc, cc, cc, cc, cc, cc, cc, cc]
[kmsg]: [2.169608] vmm_core::partition_unit::vp_set: ERROR  triple fault vtl=Vtl0

versus before:

[kmsg]: [2.186469] underhill_core::dispatch: INFO  starting VM boot_time_ms=0x1b17 boot_time="6.9353505s"
[kmsg]: [2.189205] firmware_uefi: WARN  unknown uefi write addr=0x2b data=0x77000
[kmsg]: [2.230847] vmm_core::partition_unit::vp_set: ERROR  rax=0x800000000000001a rcx=0x72e rdx=0x0 rbx=0x0 rsp=0x33d85c58 rbp=0x33d86d10 rsi=0x66083198 rdi=0x72e r8=0x72e r9=0x0 r10=0x0 r11=0x1b r12=0x80000000 r13=0x33db2458 r14=0x1 r15=0x80000000 rip=0x66af0c34 rflags=0x10206
[kmsg]: [2.231072] vmm_core::partition_unit::vp_set: ERROR  cs=SegmentRegister { base: 0, limit: ffffffff, selector: 38, attributes: a09b } ds=SegmentRegister { base: 0, limit: ffffffff, selector: 30, attributes: c093 } es=SegmentRegister { base: 0, limit: ffffffff, selector: 30, attributes: c093 } fs=SegmentRegister { base: 0, limit: ffffffff, selector: 30, attributes: c093 } gs=SegmentRegister { base: 0, limit: ffffffff, selector: 30, attributes: c093 } ss=SegmentRegister { base: 0, limit: ffffffff, selector: 30, attributes: c093 } tr=SegmentRegister { base: 33dbca30, limit: 67, selector: 20, attributes: 8b } ldtr=SegmentRegister { base: 0, limit: ffff, selector: 0, attributes: 82 } gdtr=TableRegister { base: 66bc7000, limit: 47 } idtr=TableRegister { base: 0, limit: 0 } cr0=0x80010033 cr2=0x0 cr3=0x33801000 cr4=0x660 cr8=0x0 efer=0xd00
[kmsg]: [2.232417] vmm_core::partition_unit::vp_set: ERROR  faulting instruction instruction="ud2" bytes=[f, b, eb, fc, cc, cc, cc, cc, cc, cc, cc, cc, cc, cc, cc, cc]
[kmsg]: [2.232491] vmm_core::partition_unit::vp_set: ERROR  triple fault vtl=Vtl0

[chris-oo]

I think also, we may want to print the crash message within UH itself?

[smmalis37]

We don't need to filter this on cvm tracing, it is the responsibility of the guest to not send secret data to these. However the crash message is a different story, since I believe it's just a pointer into guest memory, so that we would have to be careful about reading and sharing.

[chris-oo]

I guess if we emit the guest message as a normal tracing call, then we'll follow whatever the normal filtering logic is. I think though, we should emit the message because OHCL is may be the only thing that can actually read it, the host can't.

[jstarks]

We don't need to filter this on cvm tracing, it is the responsibility of the guest to not send secret data to these.

I don't think that's true.

[smalis-msft]

I don't think we need to filter the mere availability of the crash registers, and our forwarding of them to the host, on whether CVM debugging is enabled or not. They contain extremely minimal information (5 u64s if no message, 3 if there is a message), and this is already the policy of the legacy HCL.

I do think we need to filter the message itself, if one is included, however that can come in a follow up PR (I agree that tracing the message would be valuable)(#1395). Today the host is unable to read the message anyways, as the memory pointed to by P3 will be encrypted in a CVM.

[benhillis]

Backported in #1403