Re: [PATCH v3 8/8] KVM: selftests: Add XCR0 Test

[Mingwei Zhang <mizhang@google.com>]

On Thu, Mar 02, 2023, Aaron Lewis wrote:
> On Thu, Mar 2, 2023 at 8:34 PM Mingwei Zhang <mizhang@google.com> wrote:
> > > > On Fri, Feb 24, 2023, Aaron Lewis wrote:
> > > Check both architectural rules and KVM's own software-defined rules to
> > > ensure the supported xfeatures[1] don't violate any of them.
> > >
> > > The architectural rules[2] and KVM's rules ensure for a given
> > > feature, e.g. sse, avx, amx, etc... their associated xfeatures are
> > > either all sets or none of them are set, and any dependencies
> > > are enabled if needed.
> > >
> > > [1] EDX:EAX of CPUID.(EAX=0DH,ECX=0)
> > > [2] SDM vol 1, 13.3 ENABLING THE XSAVE FEATURE SET AND XSAVE-ENABLED
> > >     FEATURES
> > >
> > > Signed-off-by: Aaron Lewis <aaronlewis@google.com>
> >
> > Sorry, I did not get the point of this test? I run your test in an old
> > (unpatched) kernel on two machines: 1) one with AMX and 2) one without
> > it. (SPR and Skylake). Neither of them fails. Do you want to clarify a
> > little bit?
> 
> The only known issue exists on newer versions of the kernel when run
> on SPR.  It occurs after the syscall, prctl (to enable XTILEDATA), was
> introduced.  If you run this test without the fix[1] you will get the
> assert below, indicating the XTILECFG is supported by the guest, but
> XTILEDATA isn't.
> 
> ==== Test Assertion Failure ====
>   x86_64/xcr0_cpuid_test.c:116: false
>   pid=18124 tid=18124 errno=4 - Interrupted system call
>      1 0x0000000000401894: main at xcr0_cpuid_test.c:116
>      2 0x0000000000414263: __libc_start_call_main at libc-start.o:?
>      3 0x00000000004158af: __libc_start_main_impl at ??:?
>      4 0x0000000000401660: _start at ??:?
>   Failed guest assert: !__supported || __supported == ((((((1ULL))) <<
> (18)) | ((((1ULL))) << (17)))) at x86_64/xcr0_cpuid_test.c:72
> 0x20000 0x60000 0x0
> 
> [1] KVM: x86: Clear all supported AMX xfeatures if they are not all set
> https://lore.kernel.org/kvm/20230224223607.1580880-6-aaronlewis@google.com/
> 

Understood. Thanks for reminding me. Yes, without the invocation of
vm_xsave_require_permission(XSTATE_XTILE_DATA_BIT); VM guest will see a
partially enabled feature (XTILECFG without XTILEDATA).

Reviewed-by: Mingwei Zhang <mizhang@google.com>
> >
> >
> > Thanks.
> > -Mingwei