On Tue, Jan 10, 2023, Mingwei Zhang wrote:
> When #NM is triggered, the handler needs to ensure the exception is
State what the patch does (and explain why), don't say ABC needs/should do XYZ.
The #NM handler doesn't _need_ to ensure the #NM wasn't due to CR0.TS
> triggered by AMX by checking IA32_XFD_ERR and not because of CR0.TS[bit 3]
CR0.TS is a single bit, using square braces makes it look like an index into CR0.TS.
I would drop the "bit 3" part altogether, it's not relevant
> is 1. Note that the value of IA32_XFD_ERR comes from "the logical AND of
> the IA32_XFD MSR and the bitmap corresponding to the state components
> required by the faulting instruction." (Intel SDM vol 1. Section 13.14)
>
> Add the missing check of CR0.TS before checking the value of IA32_XFD_ERR.
> In addition, add an extra check to IA32_XFD to ensure the behavior is
> consistent with the AMX archtecture. In addition, repeat the checks across
> context switch to ensure the values of IA32_XFD and IA32_XFD_ERR are well
> preserved.
Split the MSR_IA32_XFD checks to a separate patch. Or I guess given the shortlog
is about IA32_XFD, split the CR0.TS check to a separate patch.
>
> Signed-off-by: Mingwei Zhang <mizhang@xxxxxxxxxx>
> ---
> tools/testing/selftests/kvm/x86_64/amx_test.c | 3 +++
> 1 file changed, 3 insertions(+)
>
> diff --git a/tools/testing/selftests/kvm/x86_64/amx_test.c b/tools/testing/selftests/kvm/x86_64/amx_test.c
> index 16533949a189..b2369f956fea 100644
> --- a/tools/testing/selftests/kvm/x86_64/amx_test.c
> +++ b/tools/testing/selftests/kvm/x86_64/amx_test.c
> @@ -226,9 +226,12 @@ void guest_nm_handler(struct ex_regs *regs)
> {
> /* Check if #NM is triggered by XFEATURE_MASK_XTILEDATA */
> GUEST_SYNC(7);
> + GUEST_ASSERT((get_cr0() & X86_CR0_TS) == 0);
GUEST_ASSERT(!(get_cr0() & X86_CR0_TS));
> GUEST_ASSERT(rdmsr(MSR_IA32_XFD_ERR) == XFEATURE_MASK_XTILEDATA);
> + GUEST_ASSERT((rdmsr(MSR_IA32_XFD) & XFEATURE_MASK_XTILEDATA) == XFEATURE_MASK_XTILEDATA);
Isn't this just
GUEST_ASSERT(rdmsr(MSR_IA32_XFD) & XFEATURE_MASK_XTILEDATA);
or am I horribly misreading the code?
> GUEST_SYNC(8);
> GUEST_ASSERT(rdmsr(MSR_IA32_XFD_ERR) == XFEATURE_MASK_XTILEDATA);
> + GUEST_ASSERT((rdmsr(MSR_IA32_XFD) & XFEATURE_MASK_XTILEDATA) == XFEATURE_MASK_XTILEDATA);
Same here.
> /* Clear xfd_err */
> wrmsr(MSR_IA32_XFD_ERR, 0);
> /* xfd=0, enable amx */
> --
> 2.39.0.314.g84b9a713c41-goog
>
