From: Aaron Lewis <aaronlewis@google.com>

In amx_test, userspace requires that XSAVE is supported before running
the test, then the guest checks that it is supported after enabling
AMX.  Remove the redundant check in the guest that XSAVE is supported.

Opportunistically, move the check that OSXSAVE is set to immediately
after the guest sets it, rather than in a separate helper.

Signed-off-by: Aaron Lewis <aaronlewis@google.com>
Signed-off-by: Mingwei Zhang <mizhang@google.com>
---
 tools/testing/selftests/kvm/x86_64/amx_test.c | 8 +-------
 1 file changed, 1 insertion(+), 7 deletions(-)

diff --git a/tools/testing/selftests/kvm/x86_64/amx_test.c b/tools/testing/selftests/kvm/x86_64/amx_test.c
index 1a1565126255..deacd21cf744 100644
--- a/tools/testing/selftests/kvm/x86_64/amx_test.c
+++ b/tools/testing/selftests/kvm/x86_64/amx_test.c
@@ -109,12 +109,6 @@ static inline void __xsavec(struct xstate *xstate, uint64_t rfbm)
 		     : "memory");
 }
 
-static inline void check_cpuid_xsave(void)
-{
-	GUEST_ASSERT(this_cpu_has(X86_FEATURE_XSAVE));
-	GUEST_ASSERT(this_cpu_has(X86_FEATURE_OSXSAVE));
-}
-
 static void check_xtile_info(void)
 {
 	GUEST_ASSERT(this_cpu_has_p(X86_PROPERTY_XSTATE_MAX_SIZE_XCR0));
@@ -157,6 +151,7 @@ static void init_regs(void)
 	cr4 = get_cr4();
 	cr4 |= X86_CR4_OSXSAVE;
 	set_cr4(cr4);
+	GUEST_ASSERT(this_cpu_has(X86_FEATURE_OSXSAVE));
 
 	xcr0 = __xgetbv(0);
 	xcr0 |= XFEATURE_MASK_XTILE;
@@ -169,7 +164,6 @@ static void __attribute__((__flatten__)) guest_code(struct tile_config *amx_cfg,
 						    struct xstate *xstate)
 {
 	init_regs();
-	check_cpuid_xsave();
 	check_xtile_info();
 	GUEST_SYNC(1);
 
-- 
2.39.2.637.g21b0678d19-goog

On Tue, Feb 21, 2023, Mingwei Zhang wrote:
> From: Aaron Lewis <aaronlewis@google.com>
> 
> In amx_test, userspace requires that XSAVE is supported before running
> the test, then the guest checks that it is supported after enabling
> AMX.  Remove the redundant check in the guest that XSAVE is supported.

It's a bit paranoid, but I actually don't mind the extra check.  It's not redundant
per se, just useless in its current location.  If the check is moved _before_
CR4 is set, then it actually provides value, e.g. if something does go sideways,
will fire an assert instead of getting a #GP on set_cr4().