Re: [PATCH] x86: vmx: Allow direct access to MSR_IA32_SPEC_CTRL

[KarimAllah Ahmed <karahmed@amazon.com>]

On 01/29/2018 08:04 PM, Jim Mattson wrote:
> Can I assume you'll send out a new version with the fixes?

Yes, I am currently doing some tests and once I am done I will send a 
new round.

... and the typo is already fixed in 'ibpb-wip' :)

> 
> On Mon, Jan 29, 2018 at 11:01 AM, David Woodhouse <dwmw2@infradead.org> wrote:
>>
>> (Top-posting; sorry.)
>>
>> Much of that is already fixed during our day, in
>> http://git.infradead.org/linux-retpoline.git/shortlog/refs/heads/ibpb
>>
>> I forgot to fix up the wrong-MSR typo though, and we do still need to address reset.
>>
>> On Mon, 2018-01-29 at 10:43 -0800, Jim Mattson wrote:
>>> On Sun, Jan 28, 2018 at 11:29 AM, KarimAllah Ahmed  wrote:
>>>>
>>>> Add direct access to MSR_IA32_SPEC_CTRL for guests. This is needed for guests
>>>> that will only mitigate Spectre V2 through IBRS+IBPB and will not be using a
>>>> retpoline+IBPB based approach.
>>>>
>>>> To avoid the overhead of atomically saving and restoring the MSR_IA32_SPEC_CTRL
>>>> for guests that do not actually use the MSR, only add_atomic_switch_msr when a
>>>> non-zero is written to it.
>>>>
>>>> Cc: Asit Mallick <asit.k.mallick@intel.com>
>>>> Cc: Arjan Van De Ven <arjan.van.de.ven@intel.com>
>>>> Cc: Dave Hansen <dave.hansen@intel.com>
>>>> Cc: Andi Kleen <ak@linux.intel.com>
>>>> Cc: Andrea Arcangeli <aarcange@redhat.com>
>>>> Cc: Linus Torvalds <torvalds@linux-foundation.org>
>>>> Cc: Tim Chen <tim.c.chen@linux.intel.com>
>>>> Cc: Thomas Gleixner <tglx@linutronix.de>
>>>> Cc: Dan Williams <dan.j.williams@intel.com>
>>>> Cc: Jun Nakajima <jun.nakajima@intel.com>
>>>> Cc: Paolo Bonzini <pbonzini@redhat.com>
>>>> Cc: David Woodhouse <dwmw@amazon.co.uk>
>>>> Cc: Greg KH <gregkh@linuxfoundation.org>
>>>> Cc: Andy Lutomirski <luto@kernel.org>
>>>> Signed-off-by: KarimAllah Ahmed <karahmed@amazon.de>
>>>> Signed-off-by: Ashok Raj <ashok.raj@intel.com>
>>>> ---
>>>>   arch/x86/kvm/cpuid.c |  4 +++-
>>>>   arch/x86/kvm/cpuid.h |  1 +
>>>>   arch/x86/kvm/vmx.c   | 63 ++++++++++++++++++++++++++++++++++++++++++++++++++++
>>>>   3 files changed, 67 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/arch/x86/kvm/cpuid.c b/arch/x86/kvm/cpuid.c
>>>> index 0099e10..dc78095 100644
>>>> --- a/arch/x86/kvm/cpuid.c
>>>> +++ b/arch/x86/kvm/cpuid.c
>>>> @@ -70,6 +70,7 @@ u64 kvm_supported_xcr0(void)
>>>>   /* These are scattered features in cpufeatures.h. */
>>>>   #define KVM_CPUID_BIT_AVX512_4VNNIW     2
>>>>   #define KVM_CPUID_BIT_AVX512_4FMAPS     3
>>>> +#define KVM_CPUID_BIT_SPEC_CTRL         26
>>>>   #define KF(x) bit(KVM_CPUID_BIT_##x)
>>>>
>>>>   int kvm_update_cpuid(struct kvm_vcpu *vcpu)
>>>> @@ -392,7 +393,8 @@ static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
>>>>
>>>>          /* cpuid 7.0.edx*/
>>>>          const u32 kvm_cpuid_7_0_edx_x86_features =
>>>> -               KF(AVX512_4VNNIW) | KF(AVX512_4FMAPS);
>>>> +               KF(AVX512_4VNNIW) | KF(AVX512_4FMAPS) | \
>>>> +               (boot_cpu_has(X86_FEATURE_SPEC_CTRL) ? KF(SPEC_CTRL) : 0);
>>> Isn't 'boot_cpu_has()' superflous here? And aren't there two bits to
>>> pass through for existing CPUs (26 and 27)?
>>>
>>>>
>>>>
>>>>          /* all calls to cpuid_count() should be made on the same cpu */
>>>>          get_cpu();
>>>> diff --git a/arch/x86/kvm/cpuid.h b/arch/x86/kvm/cpuid.h
>>>> index cdc70a3..dcfe227 100644
>>>> --- a/arch/x86/kvm/cpuid.h
>>>> +++ b/arch/x86/kvm/cpuid.h
>>>> @@ -54,6 +54,7 @@ static const struct cpuid_reg reverse_cpuid[] = {
>>>>          [CPUID_8000_000A_EDX] = {0x8000000a, 0, CPUID_EDX},
>>>>          [CPUID_7_ECX]         = {         7, 0, CPUID_ECX},
>>>>          [CPUID_8000_0007_EBX] = {0x80000007, 0, CPUID_EBX},
>>>> +       [CPUID_7_EDX]         = {         7, 0, CPUID_EDX},
>>>>   };
>>>>
>>>>   static __always_inline struct cpuid_reg x86_feature_cpuid(unsigned x86_feature)
>>>> diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
>>>> index aa8638a..1b743a0 100644
>>>> --- a/arch/x86/kvm/vmx.c
>>>> +++ b/arch/x86/kvm/vmx.c
>>>> @@ -920,6 +920,9 @@ static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked);
>>>>   static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
>>>>                                              u16 error_code);
>>>>   static void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu);
>>>> +static void __always_inline vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
>>>> +                                                         u32 msr, int type);
>>>> +
>>>>
>>>>   static DEFINE_PER_CPU(struct vmcs *, vmxarea);
>>>>   static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
>>>> @@ -2007,6 +2010,28 @@ static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
>>>>          m->host[i].value = host_val;
>>>>   }
>>>>
>>>> +/* do not touch guest_val and host_val if the msr is not found */
>>>> +static int read_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
>>>> +                                 u64 *guest_val, u64 *host_val)
>>>> +{
>>>> +       unsigned i;
>>>> +       struct msr_autoload *m = &vmx->msr_autoload;
>>>> +
>>>> +       for (i = 0; i < m->nr; ++i)
>>>> +               if (m->guest[i].index == msr)
>>>> +                       break;
>>>> +
>>>> +       if (i == m->nr)
>>>> +               return 1;
>>>> +
>>>> +       if (guest_val)
>>>> +               *guest_val = m->guest[i].value;
>>>> +       if (host_val)
>>>> +               *host_val = m->host[i].value;
>>>> +
>>>> +       return 0;
>>>> +}
>>>> +
>>>>   static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
>>>>   {
>>>>          u64 guest_efer = vmx->vcpu.arch.efer;
>>>> @@ -3203,7 +3228,9 @@ static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
>>>>    */
>>>>   static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
>>>>   {
>>>> +       u64 spec_ctrl = 0;
>>>>          struct shared_msr_entry *msr;
>>>> +       struct vcpu_vmx *vmx = to_vmx(vcpu);
>>>>
>>>>          switch (msr_info->index) {
>>>>   #ifdef CONFIG_X86_64
>>>> @@ -3223,6 +3250,19 @@ static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
>>>>          case MSR_IA32_TSC:
>>>>                  msr_info->data = guest_read_tsc(vcpu);
>>>>                  break;
>>>> +       case MSR_IA32_SPEC_CTRL:
>>>> +               if (!msr_info->host_initiated &&
>>>> +                   !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
>>> Shouldn't this conjunct be:
>>> !(guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL) ||
>>> guest_cpuid_has(vcpu, X86_FEATURE_STIBP))?
>>>
>>>>
>>>> +                       return 1;
>>> What if !boot_cpu_has(X86_FEATURE_SPEC_CTRL) &&
>>> !boot_cpu_has(X86_FEATURE_STIBP)? That should also return 1, I think.
>>>
>>>>
>>>> +
>>>> +               /*
>>>> +                * If the MSR is not in the atomic list yet, then it was never
>>>> +                * written to. So the MSR value will be '0'.
>>>> +                */
>>>> +               read_atomic_switch_msr(vmx, MSR_IA32_SPEC_CTRL, &spec_ctrl, NULL);
>>> Why not just add msr_ia32_spec_ctrl to struct vcpu_vmx, so that you
>>> don't have to search the atomic switch list?
>>>
>>>>
>>>> +
>>>> +               msr_info->data = spec_ctrl;
>>>> +               break;
>>>>          case MSR_IA32_SYSENTER_CS:
>>>>                  msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
>>>>                  break;
>>>> @@ -3289,6 +3329,13 @@ static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
>>>>          int ret = 0;
>>>>          u32 msr_index = msr_info->index;
>>>>          u64 data = msr_info->data;
>>>> +       unsigned long *msr_bitmap;
>>>> +
>>>> +       /*
>>>> +        * IBRS is not used (yet) to protect the host. Once it does, this
>>>> +        * variable needs to be a bit smarter.
>>>> +        */
>>>> +       u64 host_spec_ctrl = 0;
>>>>
>>>>          switch (msr_index) {
>>>>          case MSR_EFER:
>>>> @@ -3330,6 +3377,22 @@ static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
>>>>          case MSR_IA32_TSC:
>>>>                  kvm_write_tsc(vcpu, msr_info);
>>>>                  break;
>>>> +       case MSR_IA32_SPEC_CTRL:
>>>> +               if (!msr_info->host_initiated &&
>>>> +                   !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
>>>> +                       return 1;
>>> This looks incomplete. As above, what if
>>> !boot_cpu_has(X86_FEATURE_SPEC_CTRL) &&
>>> !boot_cpu_has(X86_FEATURE_STIBP)?
>>> If the host doesn't support MSR_IA32_SPEC_CTRL, you'll get a VMX-abort
>>> on loading the host MSRs from the VM-exit MSR load list.
>>>
>>> Also, what if the value being written is illegal?
>>>
>>> /*
>>> * Processors that support IBRS but not STIBP
>>> * (CPUID.(EAX=07H, ECX=0):EDX[27:26] = 01b) will
>>> * ignore attempts to set STIBP instead of causing an
>>> * exception due to setting that reserved bit.
>>> */
>>> if ((data & ~(u64)(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP)) ||
>>>      ((data & SPEC_CTRL_IBRS) &&
>>>       !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL)))
>>> return 1;
>>>
>>>>
>>>> +
>>>> +               /*
>>>> +                * Now we know that the guest is actually using the MSR, so
>>>> +                * atomically load and save the SPEC_CTRL MSR and pass it
>>>> +                * through to the guest.
>>>> +                */
>>>> +               add_atomic_switch_msr(vmx, MSR_IA32_SPEC_CTRL, msr_info->data,
>>>> +                                     host_spec_ctrl);
>>>> +               msr_bitmap = vmx->vmcs01.msr_bitmap;
>>>> +               vmx_disable_intercept_for_msr(msr_bitmap, MSR_FS_BASE, MSR_TYPE_RW);
>>> I assume you mean MSR_IA32_SPEC_CTRL rather than MSR_FS_BASE.
>>>
>>> Also, what if the host and the guest support a different set of bits
>>> in MSR_IA32_SPEC_CTRL, due to a userspace modification of the guest's
>>> CPUID info?
>>>
>>>>
>>>> +
>>>> +               break;
>>>>          case MSR_IA32_CR_PAT:
>>>>                  if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
>>>>                          if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
>>>> --
>>>> 2.7.4
>>>>
>>> Where do you preserve the guest's MSR_IA32_SPEC_CTRL value on VM-exit,
>>> if the guest has been given permission to write the MSR?
>>>
>>> You also have to clear the guest's MSR_IA32_SPEC_CTRL on
>>> vmx_vcpu_reset, don't you?
>>>
Amazon Development Center Germany GmbH
Berlin - Dresden - Aachen
main office: Krausenstr. 38, 10117 Berlin
Geschaeftsfuehrer: Dr. Ralf Herbrich, Christian Schlaeger
Ust-ID: DE289237879
Eingetragen am Amtsgericht Charlottenburg HRB 149173 B