Re: [NEEDS-REVIEW] [PATCH v15 03/26] x86/fpu/xstate: Introduce CET MSR XSAVES supervisor states

[Yu, Yu-cheng]

On 11/30/2020 9:45 AM, Dave Hansen wrote:
On 11/10/20 8:21 AM, Yu-cheng Yu wrote:
Control-flow Enforcement Technology (CET) adds five MSRs.  Introduce
them and their XSAVES supervisor states:
     MSR_IA32_U_CET (user-mode CET settings),
     MSR_IA32_PL3_SSP (user-mode Shadow Stack pointer),
     MSR_IA32_PL0_SSP (kernel-mode Shadow Stack pointer),
     MSR_IA32_PL1_SSP (Privilege Level 1 Shadow Stack pointer),
     MSR_IA32_PL2_SSP (Privilege Level 2 Shadow Stack pointer).

This patch goes into a bunch of XSAVE work that this changelog only
briefly touches on.  I think it needs to be beefed up a bit.

I will do that.


@@ -835,8 +843,19 @@ void __init fpu__init_system_xstate(void)
         * Clear XSAVE features that are disabled in the normal CPUID.
         */
        for (i = 0; i < ARRAY_SIZE(xsave_cpuid_features); i++) {
-               if (!boot_cpu_has(xsave_cpuid_features[i]))
-                       xfeatures_mask_all &= ~BIT_ULL(i);
+               if (xsave_cpuid_features[i] == X86_FEATURE_SHSTK) {
+                       /*
+                        * X86_FEATURE_SHSTK and X86_FEATURE_IBT share
+                        * same states, but can be enabled separately.
+                        */
+                       if (!boot_cpu_has(X86_FEATURE_SHSTK) &&
+                           !boot_cpu_has(X86_FEATURE_IBT))
+                               xfeatures_mask_all &= ~BIT_ULL(i);
+               } else {
+                       if ((xsave_cpuid_features[i] == -1) ||

Where did the -1 come from?  Was that introduced earlier in this series?
  I don't see any way a xsave_cpuid_features[] can be -1 in the current tree.


Yes, we used to have a hole in xsave_cpuid_features[] and put -1 there. 
Do we want to keep this in case we again have holes in the future?

+                           !boot_cpu_has(xsave_cpuid_features[i]))
+                               xfeatures_mask_all &= ~BIT_ULL(i);
+               }
        }

Do we have any other spots in the kernel where we care about:

        boot_cpu_has(X86_FEATURE_SHSTK) ||
        boot_cpu_has(X86_FEATURE_IBT)

?  If so, we could also address this by declaring a software-defined
X86_FEATURE_CET and then setting it if SHSTK||IBT is supported, then we
just put that one feature in xsave_cpuid_features[].

That is a better solution.  I will look into that.


I'm also not crazy about the loop as it is.  I'd much rather see this in
a helper like:

bool cpu_supports_xsave_deps(int xfeature)
{
        bool ret;

        ret = boot_cpu_has(xsave_cpuid_features[xfeature])

        /*
         * X86_FEATURE_SHSTK is checked in xsave_cpuid_features()
         * but the CET states are needed if either SHSTK or IBT are
         * available.
         */
        if (xfeature == XFEATURE_CET_USER ||
            xfeature == XFEATURE_CET_KERNEL)
                ret |= boot_cpu_has(X86_FEATURE_IBT)
                
        return ret;
}

See how that's extensible?  You can add as many special cases as you want.


Yes.

Thanks,
Yu-cheng