On Sun, Mar 03, 2019 at 05:36:45PM +0800, Yang Weijiang wrote:
> On Fri, Mar 01, 2019 at 06:53:23AM -0800, Sean Christopherson wrote:
> > On Thu, Feb 28, 2019 at 04:28:32PM +0800, Yang Weijiang wrote:
> > > On Thu, Feb 28, 2019 at 07:59:40AM -0800, Sean Christopherson wrote:
> > > > On Mon, Feb 25, 2019 at 09:27:11PM +0800, Yang Weijiang wrote:
> > > > > Guest CET SHSTK and IBT capability are reported via
> > > > > CPUID.(EAX=7, ECX=0):ECX[bit 7] and EDX[bit 20] respectively.
> > > > > Guest user mode and supervisor mode xsaves component size
> > > > > is reported via CPUID.(EAX=0xD, ECX=1):ECX[bit 11] and ECX[bit 12]
> > > > > respectively.
> > > > >
> > > > > Signed-off-by: Zhang Yi Z <yi.z.zh...@linux.intel.com>
> > > > > Signed-off-by: Yang Weijiang <weijiang.y...@intel.com>
> > > > > ---
> > > > > arch/x86/kvm/cpuid.c | 60
> > > > > +++++++++++++++++++++++++++++++++-----------
> > > > > arch/x86/kvm/x86.h | 4 +++
> > > > > 2 files changed, 50 insertions(+), 14 deletions(-)
> > > > >
> > > > > diff --git a/arch/x86/kvm/cpuid.c b/arch/x86/kvm/cpuid.c
> > > > > index cb1aece25b17..5e05756cc6db 100644
> > > > > --- a/arch/x86/kvm/cpuid.c
> > > > > +++ b/arch/x86/kvm/cpuid.c
> > > > > @@ -65,6 +65,16 @@ u64 kvm_supported_xcr0(void)
> > > > > return xcr0;
> > > > > }
> > > > >
> > > > > +u64 kvm_supported_xss(void)
> > > > > +{
> > > > > + u64 xss;
> > > > > +
> > > > > + rdmsrl(MSR_IA32_XSS, xss);
> > > > > + xss &= KVM_SUPPORTED_XSS;
> > > > > + return xss;
> > > > > +}
> > > > > +EXPORT_SYMBOL(kvm_supported_xss);
> > > > > +
> > > > > #define F(x) bit(X86_FEATURE_##x)
> > > > >
> > > > > /* For scattered features from cpufeatures.h; we currently expose
> > > > > none */
> > > > > @@ -323,6 +333,7 @@ static inline int __do_cpuid_ent(struct
> > > > > kvm_cpuid_entry2 *entry, u32 function,
> > > > > u32 index, int *nent, int maxnent)
> > > > > {
> > > > > int r;
> > > > > + u32 eax, ebx, ecx, edx;
> > > > > unsigned f_nx = is_efer_nx() ? F(NX) : 0;
> > > > > #ifdef CONFIG_X86_64
> > > > > unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() ==
> > > > > PT_PDPE_LEVEL)
> > > > > @@ -503,6 +514,20 @@ static inline int __do_cpuid_ent(struct
> > > > > kvm_cpuid_entry2 *entry, u32 function,
> > > > > * if the host doesn't support it.
> > > > > */
> > > > > entry->edx |= F(ARCH_CAPABILITIES);
> > > > > +
> > > > > + /*
> > > > > + * Guest OS CET enabling is designed
> > > > > independent to
> > > > > + * host enabling, it only has dependency on
> > > > > Host HW
> > > > > + * capability, if it has, report CET support to
> > > > > + * Guest.
> > > > > + */
> > > > > + cpuid_count(7, 0, &eax, &ebx, &ecx, &edx);
> > > > > + if (ecx & F(SHSTK))
> > > > > + entry->ecx |= F(SHSTK);
> > > > > +
> > > > > + if (edx & F(IBT))
> > > > > + entry->edx |= F(IBT);
> > > >
> > > > There's no need to manually add these flags. They will be automatically
> > > > kept if supported in hardware because your previous patch, 02/08, added
> > > > them to the mask of features that can be exposed to the guest,
> > > > i.e. set them in kvm_cpuid_7_0_e{c,d}x_x86_features.
> > > >
> > > I shared the same thought as you before, but after I took a closer look
> > > at the
> > > kernel code, actually, when host CET feature is disabled by user via
> > > cmdline options(no_cet_shstk and no_cet_ibt), it'll mask out CET feature
> > > bits in
> > > boot_cpu_data.x86_capbility[] array, and cpuid_mask() will make the bits
> > > in previous definition lost, so these lines actually add them back when
> > > host CET is disabled.
> >
> > 'entry' is filled by do_cpuid_1_ent(), which does cpuid_count(), same as
> > your code, i.e. it's not affected by whether or not the host kernel is
> > using each feature.
> >
> I checked CET kernel patch:
> #ifdef CONFIG_X86_INTEL_SHADOW_STACK_USER
> static __init int setup_disable_shstk(char *s) {
> /* require an exact match without trailing characters */
> if (s[0] != '\0')
> return 0;
>
> if (!boot_cpu_has(X86_FEATURE_SHSTK))
> return 1;
>
> setup_clear_cpu_cap(X86_FEATURE_SHSTK);
> pr_info("x86: 'no_cet_shstk' specified, disabling Shadow Stack\n");
> return 1;
> }
> __setup("no_cet_shstk", setup_disable_shstk); #endif
>
> setup_disable_shstk()->setup_clear_cpu_cap(X86_FEATURE_SHSTK)->do_clear_cpu_cap(NULL,
> feature)->clear_feature(c, feature)->clear_cpu_cap(&boot_cpu_data, feature);
>
> this path will clear boot_cpu_data.x86_capability[X86_FEATURE_SHSTK] if
> "no_cet_shstk" is set.
> but in cpuid_mask(), it will "AND" the bit with SHSTK bit set
> in kvm_cpuid_7_0_ecx_x86_features, so the bit in ecx is cleared,
> need to add the bit back according to host cpuid_count().
> the CET kernel patch can be seen in below patch link.
Ah, I see. In this case we need to honor boot_cpu_data. The idea is
that a feature should not be exposed to the guest, i.e. actually used,
if it has been explicitly disabled by the user, e.g. to workaround a
hardware or firmware issue. The cases where a feature is exposed to
the guest even when disabled in host is when said feature is emulated
by KVM in software.
>
> > > please check CET kernel patch here:
> > > https://lkml.org/lkml/2018/11/20/204
