On Tue, Jan 10, 2023, Mingwei Zhang wrote: > Repeat the checking of AMX component in xheader after xsavec when s/xsavec/XSAVEC. Not sure about xheader, though it seems like that should be capitalized too. > IA32_XFD[18] is set. This check calibrates the functionality scope of s/18/XTILEDATA > IA32_XFD: it does not intercept the XSAVE state management. I didn't follow this. Is this basically saying "Verify XTILEDATA is saved by XSAVEC even when disabled via IA32_XFD"? > Regardless of the values in IA32_XFD, AMX component state will still be > managed by XSAVE* and XRSTOR* as long as XCR[18:17] are set. > > Signed-off-by: Mingwei Zhang <mizhang@xxxxxxxxxx> > --- > tools/testing/selftests/kvm/x86_64/amx_test.c | 10 ++++++++++ > 1 file changed, 10 insertions(+) > > diff --git a/tools/testing/selftests/kvm/x86_64/amx_test.c b/tools/testing/selftests/kvm/x86_64/amx_test.c > index 18203e399e9d..9a80a59b64e6 100644 > --- a/tools/testing/selftests/kvm/x86_64/amx_test.c > +++ b/tools/testing/selftests/kvm/x86_64/amx_test.c > @@ -235,6 +235,16 @@ static void __attribute__((__flatten__)) guest_code(struct tile_config *amx_cfg, > > /* xfd=0x40000, disable amx tiledata */ > wrmsr(MSR_IA32_XFD, XFEATURE_MASK_XTILEDATA); > + > + /* > + * Bit 18 is cleared in XSTATE_BV but set in XCOMP_BV, this property s/Bit 18/XTILEDATA > + * remains the same even when IA32_XFD disables amx tiledata. I would phrase this as "even when AMX tile data is disabled via IA32_XFD". Software disables AMX, IA32_XFD is the mechanic by which that is done. > + */ > + set_xstatebv(xsave_data, XFEATURE_MASK_XTILEDATA); > + __xsavec(xsave_data, XFEATURE_MASK_XTILEDATA); > + GUEST_ASSERT((get_xstatebv(xsave_data) & XFEATURE_MASK_XTILEDATA) == 0); > + GUEST_ASSERT((get_xcompbv(xsave_data) & XFEATURE_MASK_XTILEDATA) == XFEATURE_MASK_XTILEDATA); Same feedback about using !(...) and unnecessary "== x". > + > GUEST_SYNC(6); > GUEST_ASSERT(rdmsr(MSR_IA32_XFD) == XFEATURE_MASK_XTILEDATA); > set_tilecfg(amx_cfg); > -- > 2.39.0.314.g84b9a713c41-goog >
