From: Pawan Gupta <pawan.kumar.gupta@linux.intel.com>
Date: Thu, 19 May 2022 20:35:15 -0700
Subject: KVM: x86/speculation: Disable Fill buffer clear within guests
Git-commit: 027bbb884be006b05d9c577d6401686053aa789e
Patch-mainline: v5.18-rc7
References: bsc#1199650 CVE-2022-21166 CVE-2022-21127 CVE-2022-21123 CVE-2022-21125 CVE-2022-21180
The enumeration of MD_CLEAR in CPUID(EAX=7,ECX=0).EDX{bit 10} is not an
accurate indicator on all CPUs of whether the VERW instruction will
overwrite fill buffers. FB_CLEAR enumeration in
IA32_ARCH_CAPABILITIES{bit 17} covers the case of CPUs that are not
vulnerable to MDS/TAA, indicating that microcode does overwrite fill
buffers.
Guests running in VMM environments may not be aware of all the
capabilities/vulnerabilities of the host CPU. Specifically, a guest may
apply MDS/TAA mitigations when a virtual CPU is enumerated as vulnerable
to MDS/TAA even when the physical CPU is not. On CPUs that enumerate
FB_CLEAR_CTRL the VMM may set FB_CLEAR_DIS to skip overwriting of fill
buffers by the VERW instruction. This is done by setting FB_CLEAR_DIS
during VMENTER and resetting on VMEXIT. For guests that enumerate
FB_CLEAR (explicitly asking for fill buffer clear capability) the VMM
will not use FB_CLEAR_DIS.
Irrespective of guest state, host overwrites CPU buffers before VMENTER
to protect itself from an MMIO capable guest, as part of mitigation for
MMIO Stale Data vulnerabilities.
Signed-off-by: Pawan Gupta <pawan.kumar.gupta@linux.intel.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
---
arch/x86/include/asm/msr-index.h | 6 +++
arch/x86/kvm/vmx.c | 73 +++++++++++++++++++++++++++++++++++++++
arch/x86/kvm/x86.c | 3 +
3 files changed, 82 insertions(+)
--- a/arch/x86/include/asm/msr-index.h
+++ b/arch/x86/include/asm/msr-index.h
@@ -123,6 +123,11 @@
* VERW clears CPU fill buffer
* even on MDS_NO CPUs.
*/
+#define ARCH_CAP_FB_CLEAR_CTRL BIT(18) /*
+ * MSR_IA32_MCU_OPT_CTRL[FB_CLEAR_DIS]
+ * bit available to control VERW
+ * behavior.
+ */
#define MSR_IA32_FLUSH_CMD 0x0000010b
#define L1D_FLUSH BIT(0) /*
@@ -142,6 +147,7 @@
/* SRBDS support */
#define MSR_IA32_MCU_OPT_CTRL 0x00000123
#define RNGDS_MITG_DIS BIT(0)
+#define FB_CLEAR_DIS BIT(3) /* CPU Fill buffer clear disable */
#define MSR_IA32_SYSENTER_CS 0x00000174
#define MSR_IA32_SYSENTER_ESP 0x00000175
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -209,6 +209,9 @@ static const struct {
#define L1D_CACHE_ORDER 4
static void *vmx_l1d_flush_pages;
+/* Control for disabling CPU Fill buffer clear */
+static bool __read_mostly vmx_fb_clear_ctrl_available;
+
static int vmx_setup_l1d_flush(enum vmx_l1d_flush_state l1tf)
{
struct page *page;
@@ -824,6 +827,9 @@ struct vcpu_vmx {
*/
u64 msr_ia32_feature_control;
u64 msr_ia32_feature_control_valid_bits;
+
+ u64 msr_ia32_mcu_opt_ctrl;
+ bool disable_fb_clear;
};
enum segment_cache_field {
@@ -1693,6 +1699,61 @@ static void vmcs_load(struct vmcs *vmcs)
vmcs, phys_addr);
}
+static void vmx_setup_fb_clear_ctrl(void)
+{
+ u64 msr;
+
+ if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES) &&
+ !boot_cpu_has_bug(X86_BUG_MDS) &&
+ !boot_cpu_has_bug(X86_BUG_TAA)) {
+ rdmsrl(MSR_IA32_ARCH_CAPABILITIES, msr);
+ if (msr & ARCH_CAP_FB_CLEAR_CTRL)
+ vmx_fb_clear_ctrl_available = true;
+ }
+}
+
+static __always_inline void vmx_disable_fb_clear(struct vcpu_vmx *vmx)
+{
+ u64 msr;
+
+ if (!vmx->disable_fb_clear)
+ return;
+
+ rdmsrl(MSR_IA32_MCU_OPT_CTRL, msr);
+ msr |= FB_CLEAR_DIS;
+ wrmsrl(MSR_IA32_MCU_OPT_CTRL, msr);
+ /* Cache the MSR value to avoid reading it later */
+ vmx->msr_ia32_mcu_opt_ctrl = msr;
+}
+
+static __always_inline void vmx_enable_fb_clear(struct vcpu_vmx *vmx)
+{
+ if (!vmx->disable_fb_clear)
+ return;
+
+ vmx->msr_ia32_mcu_opt_ctrl &= ~FB_CLEAR_DIS;
+ wrmsrl(MSR_IA32_MCU_OPT_CTRL, vmx->msr_ia32_mcu_opt_ctrl);
+}
+
+static void vmx_update_fb_clear_dis(struct kvm_vcpu *vcpu, struct vcpu_vmx *vmx)
+{
+ vmx->disable_fb_clear = vmx_fb_clear_ctrl_available;
+
+ /*
+ * If guest will not execute VERW, there is no need to set FB_CLEAR_DIS
+ * at VMEntry. Skip the MSR read/write when a guest has no use case to
+ * execute VERW.
+ */
+ if ((vcpu->arch.arch_capabilities & ARCH_CAP_FB_CLEAR) ||
+ ((vcpu->arch.arch_capabilities & ARCH_CAP_MDS_NO) &&
+ (vcpu->arch.arch_capabilities & ARCH_CAP_TAA_NO) &&
+ (vcpu->arch.arch_capabilities & ARCH_CAP_PSDP_NO) &&
+ (vcpu->arch.arch_capabilities & ARCH_CAP_FBSDP_NO) &&
+ (vcpu->arch.arch_capabilities & ARCH_CAP_SBDR_SSDP_NO)))
+ vmx->disable_fb_clear = false;
+}
+
+
#ifdef CONFIG_KEXEC_CORE
/*
* This bitmap is used to indicate whether the vmclear
@@ -3775,6 +3836,10 @@ static int vmx_set_msr(struct kvm_vcpu *
ret = kvm_set_msr_common(vcpu, msr_info);
}
+ /* FB_CLEAR may have changed, also update the FB_CLEAR_DIS behavior */
+ if (msr_index == MSR_IA32_ARCH_CAPABILITIES)
+ vmx_update_fb_clear_dis(vcpu, vmx);
+
return ret;
}
@@ -6109,6 +6174,8 @@ static void vmx_vcpu_reset(struct kvm_vc
update_exception_bitmap(vcpu);
vpid_sync_context(vmx->vpid);
+
+ vmx_update_fb_clear_dis(vcpu, vmx);
}
/*
@@ -9862,6 +9929,8 @@ static void __noclone vmx_vcpu_run(struc
kvm_arch_has_assigned_device(vcpu->kvm))
mds_clear_cpu_buffers();
+ vmx_disable_fb_clear(vmx);
+
asm(
/* Store host registers */
"push %%" _ASM_DX "; push %%" _ASM_BP ";"
@@ -9991,6 +10060,8 @@ static void __noclone vmx_vcpu_run(struc
#endif
);
+ vmx_enable_fb_clear(vmx);
+
/*
* We do not use IBRS in the kernel. If this vCPU has used the
* SPEC_CTRL MSR it may have left it on; save the value and
@@ -13140,6 +13211,8 @@ static int __init vmx_init(void)
}
}
+ vmx_setup_fb_clear_ctrl();
+
#ifdef CONFIG_KEXEC_CORE
rcu_assign_pointer(crash_vmclear_loaded_vmcss,
crash_vmclear_local_loaded_vmcss);
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -1125,6 +1125,9 @@ u64 kvm_get_arch_capabilities(void)
*/
data |= ARCH_CAP_PSCHANGE_MC_NO;
+ /* Guests don't need to know "Fill buffer clear control" exists */
+ data &= ~ARCH_CAP_FB_CLEAR_CTRL;
+
return data;
}