[Qemu-devel][PATCH 0/4] This patch-set is to enable Guest CET support.

[Qemu-devel][PATCH 0/4] This patch-set is to enable Guest CET support.

[Yang Weijiang]

Control-flow Enforcement Technology (CET) provides protection against
return/jump-oriented programming (ROP) attacks. To make kvm Guest OS
own the capability, this patch-set is required. It enables CET related
CPUID report and xsaves/xrstors support etc in qemu.

Yang Weijiang (4):
  Add CET xsaves/xrstors related macros and structures.
  Add CET SHSTK and IBT CPUID feature-word definitions.
  Add hepler functions for CPUID xsave area size calculation.
  Report CPUID xsave area support for CET.

 target/i386/cpu.c | 67 ++++++++++++++++++++++++++++++++++++++++++++---
 target/i386/cpu.h | 36 ++++++++++++++++++++++++-
 2 files changed, 99 insertions(+), 4 deletions(-)

-- 
2.17.1

[Qemu-devel][PATCH 1/4] Add CET xsaves/xrstors related macros and structures.

[Yang Weijiang]

CET protection in user mode and kernel mode relies on
specific MSRs, these MSRs' contents are automatically
saved/restored by xsaves/xrstors instructions.

Signed-off-by: Zhang Yi <yi.z.zhang@linux.intel.com>
Signed-off-by: Yang Weijiang <weijiang.yang@intel.com>
---
 target/i386/cpu.h | 36 +++++++++++++++++++++++++++++++++++-
 1 file changed, 35 insertions(+), 1 deletion(-)

diff --git a/target/i386/cpu.h b/target/i386/cpu.h
index 9c52d0cbeb..f3f724d8e6 100644
--- a/target/i386/cpu.h
+++ b/target/i386/cpu.h
@@ -469,6 +469,9 @@ typedef enum X86Seg {
 #define XSTATE_ZMM_Hi256_BIT            6
 #define XSTATE_Hi16_ZMM_BIT             7
 #define XSTATE_PKRU_BIT                 9
+#define XSTATE_RESERVED_BIT             10
+#define XSTATE_CET_U_BIT                11
+#define XSTATE_CET_S_BIT                12
 
 #define XSTATE_FP_MASK                  (1ULL << XSTATE_FP_BIT)
 #define XSTATE_SSE_MASK                 (1ULL << XSTATE_SSE_BIT)
@@ -479,6 +482,19 @@ typedef enum X86Seg {
 #define XSTATE_ZMM_Hi256_MASK           (1ULL << XSTATE_ZMM_Hi256_BIT)
 #define XSTATE_Hi16_ZMM_MASK            (1ULL << XSTATE_Hi16_ZMM_BIT)
 #define XSTATE_PKRU_MASK                (1ULL << XSTATE_PKRU_BIT)
+#define XSTATE_RESERVED_MASK            (1ULL << XSTATE_RESERVED_BIT)
+#define XSTATE_CET_U_MASK               (1ULL << XSTATE_CET_U_BIT)
+#define XSTATE_CET_S_MASK               (1ULL << XSTATE_CET_S_BIT)
+
+/* CPUID feature bits available in XCR0 */
+#define CPUID_XSTATE_USER_MASK  (XSTATE_FP_MASK | XSTATE_SSE_MASK \
+                                | XSTATE_YMM_MASK | XSTATE_BNDREGS_MASK \
+                                | XSTATE_BNDCSR_MASK | XSTATE_OPMASK_MASK \
+                                | XSTATE_ZMM_Hi256_MASK \
+                                | XSTATE_Hi16_ZMM_MASK | XSTATE_PKRU_MASK)
+
+/* CPUID feature bits available in XSS */
+#define CPUID_XSTATE_KERNEL_MASK    (XSTATE_CET_U_MASK | XSTATE_CET_S_MASK)
 
 /* CPUID feature words */
 typedef enum FeatureWord {
@@ -503,6 +519,8 @@ typedef enum FeatureWord {
     FEAT_XSAVE_COMP_LO, /* CPUID[EAX=0xd,ECX=0].EAX */
     FEAT_XSAVE_COMP_HI, /* CPUID[EAX=0xd,ECX=0].EDX */
     FEAT_ARCH_CAPABILITIES,
+    FEAT_XSAVE_SV_LO,   /* CPUID[EAX=0xd,ECX=1].ECX */
+    FEAT_XSAVE_SV_HI,   /* CPUID[EAX=0xd,ECX=1].EDX */
     FEATURE_WORDS,
 } FeatureWord;
 
@@ -687,7 +705,7 @@ typedef uint32_t FeatureWordArray[FEATURE_WORDS];
 #define CPUID_7_0_ECX_LA57     (1U << 16)
 #define CPUID_7_0_ECX_RDPID    (1U << 22)
 #define CPUID_7_0_ECX_CLDEMOTE (1U << 25)  /* CLDEMOTE Instruction */
-
+#define CPUID_7_0_ECX_CET_SHSTK     (1U << 7)  /* CET SHSTK feature bit */
 #define CPUID_7_0_EDX_AVX512_4VNNIW (1U << 2) /* AVX512 Neural Network Instructions */
 #define CPUID_7_0_EDX_AVX512_4FMAPS (1U << 3) /* AVX512 Multiply Accumulation Single Precision */
 #define CPUID_7_0_EDX_PCONFIG (1U << 18)       /* Platform Configuration */
@@ -1021,6 +1039,19 @@ typedef struct XSavePKRU {
     uint32_t padding;
 } XSavePKRU;
 
+/* Ext. save area 11: User mode CET state */
+typedef struct XSaveCETU {
+    uint64_t u_cet;
+    uint64_t user_ssp;
+} XSaveCETU;
+
+/* Ext. save area 12: Supervisor mode CET state */
+typedef struct XSaveCETS {
+    uint64_t kernel_ssp;
+    uint64_t pl1_ssp;
+    uint64_t pl2_ssp;
+} XSaveCETS;
+
 typedef struct X86XSaveArea {
     X86LegacyXSaveArea legacy;
     X86XSaveHeader header;
@@ -1039,6 +1070,9 @@ typedef struct X86XSaveArea {
     XSaveHi16_ZMM hi16_zmm_state;
     /* PKRU State: */
     XSavePKRU pkru_state;
+    /* CET State: */
+    XSaveCETU cet_u;
+    XSaveCETS cet_s;
 } X86XSaveArea;
 
 QEMU_BUILD_BUG_ON(offsetof(X86XSaveArea, avx_state) != 0x240);
-- 
2.17.1

[Qemu-devel][PATCH 2/4] Add CET SHSTK and IBT CPUID feature-word definitions.

[Yang Weijiang]

XSS[bit 11] and XSS[bit 12] correspond to CET
user mode area and supervisor mode area respectively.

Signed-off-by: Zhang Yi <yi.z.zhang@linux.intel.com>
Signed-off-by: Yang Weijiang <weijiang.yang@intel.com>
---
 target/i386/cpu.c | 31 +++++++++++++++++++++++++++++--
 1 file changed, 29 insertions(+), 2 deletions(-)

diff --git a/target/i386/cpu.c b/target/i386/cpu.c
index f81d35e1f9..3630c688d6 100644
--- a/target/i386/cpu.c
+++ b/target/i386/cpu.c
@@ -1018,7 +1018,7 @@ static FeatureWordInfo feature_word_info[FEATURE_WORDS] = {
         .type = CPUID_FEATURE_WORD,
         .feat_names = {
             NULL, "avx512vbmi", "umip", "pku",
-            NULL /* ospke */, NULL, "avx512vbmi2", NULL,
+            NULL /* ospke */, NULL, "avx512vbmi2", "shstk",
             "gfni", "vaes", "vpclmulqdq", "avx512vnni",
             "avx512bitalg", NULL, "avx512-vpopcntdq", NULL,
             "la57", NULL, NULL, NULL,
@@ -1041,7 +1041,7 @@ static FeatureWordInfo feature_word_info[FEATURE_WORDS] = {
             NULL, NULL, NULL, NULL,
             NULL, NULL, NULL, NULL,
             NULL, NULL, "pconfig", NULL,
-            NULL, NULL, NULL, NULL,
+            "ibt", NULL, NULL, NULL,
             NULL, NULL, "spec-ctrl", NULL,
             NULL, "arch-capabilities", NULL, "ssbd",
         },
@@ -1162,6 +1162,25 @@ static FeatureWordInfo feature_word_info[FEATURE_WORDS] = {
             }
         },
     },
+    /* Below are CET supervisor xsave features */
+    [FEAT_XSAVE_SV_LO] = {
+        .type = CPUID_FEATURE_WORD,
+        .cpuid = {
+            .eax = 0xD,
+            .needs_ecx = true,
+            .ecx = 1,
+            .reg = R_ECX,
+        },
+    },
+    [FEAT_XSAVE_SV_HI] = {
+        .type = CPUID_FEATURE_WORD,
+        .cpuid = {
+            .eax = 0xD,
+            .needs_ecx = true,
+            .ecx = 1,
+            .reg = R_EDX
+        },
+    }
 };
 
 typedef struct X86RegisterInfo32 {
@@ -1233,6 +1252,14 @@ static const ExtSaveArea x86_ext_save_areas[] = {
           { .feature = FEAT_7_0_ECX, .bits = CPUID_7_0_ECX_PKU,
             .offset = offsetof(X86XSaveArea, pkru_state),
             .size = sizeof(XSavePKRU) },
+    [XSTATE_CET_U_BIT] = {
+            .feature = FEAT_7_0_ECX, .bits = CPUID_7_0_ECX_CET_SHSTK,
+            .offset = offsetof(X86XSaveArea, cet_u),
+            .size = sizeof(XSaveCETU) },
+    [XSTATE_CET_S_BIT] = {
+            .feature = FEAT_7_0_ECX, .bits = CPUID_7_0_ECX_CET_SHSTK,
+            .offset = offsetof(X86XSaveArea, cet_s),
+            .size = sizeof(XSaveCETS) },
 };
 
 static uint32_t xsave_area_size(uint64_t mask)
-- 
2.17.1

[Qemu-devel][PATCH 3/4] Add hepler functions for CPUID xsave area size calculation.

[Yang Weijiang]

These functions are called when return CPUID xsave area
size information.

Signed-off-by: Zhang Yi <yi.z.zhang@linux.intel.com>
Signed-off-by: Yang Weijiang <weijiang.yang@intel.com>
---
 target/i386/cpu.c | 26 +++++++++++++++++++++++++-
 1 file changed, 25 insertions(+), 1 deletion(-)

diff --git a/target/i386/cpu.c b/target/i386/cpu.c
index 3630c688d6..cf4f2798dc 100644
--- a/target/i386/cpu.c
+++ b/target/i386/cpu.c
@@ -1281,12 +1281,34 @@ static inline bool accel_uses_host_cpuid(void)
     return kvm_enabled() || hvf_enabled();
 }
 
+static uint32_t xsave_area_size_compat(uint64_t mask)
+{
+    int i;
+    uint64_t ret = 0;
+    uint32_t offset;
+
+    for (i = 0; i < ARRAY_SIZE(x86_ext_save_areas); i++) {
+        const ExtSaveArea *esa = &x86_ext_save_areas[i];
+        offset = i > 1 ? ret : esa->offset;
+        if ((mask >> i) & 1) {
+            ret = MAX(ret, offset + esa->size);
+        }
+    }
+    return ret;
+}
+
 static inline uint64_t x86_cpu_xsave_components(X86CPU *cpu)
 {
     return ((uint64_t)cpu->env.features[FEAT_XSAVE_COMP_HI]) << 32 |
            cpu->env.features[FEAT_XSAVE_COMP_LO];
 }
 
+static inline uint64_t x86_cpu_xsave_sv_components(X86CPU *cpu)
+{
+    return ((uint64_t)cpu->env.features[FEAT_XSAVE_SV_HI]) << 32 |
+           cpu->env.features[FEAT_XSAVE_SV_LO];
+}
+
 const char *get_register_name_32(unsigned int reg)
 {
     if (reg >= CPU_NB_REGS32) {
@@ -4913,8 +4935,10 @@ static void x86_cpu_enable_xsave_components(X86CPU *cpu)
         }
     }
 
-    env->features[FEAT_XSAVE_COMP_LO] = mask;
+    env->features[FEAT_XSAVE_COMP_LO] = mask & CPUID_XSTATE_USER_MASK;
     env->features[FEAT_XSAVE_COMP_HI] = mask >> 32;
+    env->features[FEAT_XSAVE_SV_LO] = mask & CPUID_XSTATE_KERNEL_MASK;
+    env->features[FEAT_XSAVE_SV_HI] = mask >> 32;
 }
 
 /***** Steps involved on loading and filtering CPUID data
-- 
2.17.1

[Qemu-devel][PATCH 4/4] Report CPUID xsave area support for CET.

[Yang Weijiang]

CPUID bit definition as below:
CPUID.(EAX=d, ECX=1):ECX.CET_U(bit 11): user mode state
CPUID.(EAX=d, ECX=1):ECX.CET_S(bit 12): kernel mode state

Signed-off-by: Zhang Yi <yi.z.zhang@linux.intel.com>
Signed-off-by: Yang Weijiang <weijiang.yang@intel.com>
---
 target/i386/cpu.c | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/target/i386/cpu.c b/target/i386/cpu.c
index cf4f2798dc..78994bfa1d 100644
--- a/target/i386/cpu.c
+++ b/target/i386/cpu.c
@@ -4396,12 +4396,22 @@ void cpu_x86_cpuid(CPUX86State *env, uint32_t index, uint32_t count,
             *ebx = xsave_area_size(env->xcr0);
         } else if (count == 1) {
             *eax = env->features[FEAT_XSAVE];
+            *ecx = env->features[FEAT_XSAVE_SV_LO];
+            *edx = env->features[FEAT_XSAVE_SV_HI];
+            *ebx = xsave_area_size_compat(x86_cpu_xsave_components(cpu) |
+                    x86_cpu_xsave_sv_components(cpu));
         } else if (count < ARRAY_SIZE(x86_ext_save_areas)) {
             if ((x86_cpu_xsave_components(cpu) >> count) & 1) {
                 const ExtSaveArea *esa = &x86_ext_save_areas[count];
                 *eax = esa->size;
                 *ebx = esa->offset;
             }
+            if ((x86_cpu_xsave_sv_components(cpu) >> count) & 1) {
+                const ExtSaveArea *esa_sv = &x86_ext_save_areas[count];
+                *eax = esa_sv->size;
+                *ebx = esa_sv->offset;
+                *ecx = 1;
+            }
         }
         break;
     }
-- 
2.17.1

Re: [Qemu-devel][PATCH 2/4] Add CET SHSTK and IBT CPUID feature-word definitions.

[Paolo Bonzini]

On 26/12/18 09:25, Yang Weijiang wrote:
> @@ -1233,6 +1252,14 @@ static const ExtSaveArea x86_ext_save_areas[] = {
>            { .feature = FEAT_7_0_ECX, .bits = CPUID_7_0_ECX_PKU,
>              .offset = offsetof(X86XSaveArea, pkru_state),
>              .size = sizeof(XSavePKRU) },
> +    [XSTATE_CET_U_BIT] = {
> +            .feature = FEAT_7_0_ECX, .bits = CPUID_7_0_ECX_CET_SHSTK,
> +            .offset = offsetof(X86XSaveArea, cet_u),

These offsets are incorrect, since supervisor states are only stored in
the compacted format.  In fact, in patch 4, supervisor states should
return 0 in CPUID(EAX=0Dh,ECX=n).EBX.

You can use offset == 0 to distinguish supervisor and user states, so
that supervisor states are skipped in xsave_area_size and x86_cpu_reset.

Thanks,

Paolo

> +            .size = sizeof(XSaveCETU) },
> +    [XSTATE_CET_S_BIT] = {
> +            .feature = FEAT_7_0_ECX, .bits = CPUID_7_0_ECX_CET_SHSTK,
> +            .offset = offsetof(X86XSaveArea, cet_s),
> +            .size = sizeof(XSaveCETS) },
>  };
>  
>  static uint32_t xsave_area_size(uint64_t mask)

Re: [Qemu-devel][PATCH 2/4] Add CET SHSTK and IBT CPUID feature-word definitions.

[Yang Weijiang]

On Fri, Dec 28, 2018 at 03:25:10PM +0100, Paolo Bonzini wrote:
Thanks a lot Paolo for the comments!

I'll fix the issue in next version.

> On 26/12/18 09:25, Yang Weijiang wrote:
> > @@ -1233,6 +1252,14 @@ static const ExtSaveArea x86_ext_save_areas[] = {
> >            { .feature = FEAT_7_0_ECX, .bits = CPUID_7_0_ECX_PKU,
> >              .offset = offsetof(X86XSaveArea, pkru_state),
> >              .size = sizeof(XSavePKRU) },
> > +    [XSTATE_CET_U_BIT] = {
> > +            .feature = FEAT_7_0_ECX, .bits = CPUID_7_0_ECX_CET_SHSTK,
> > +            .offset = offsetof(X86XSaveArea, cet_u),
> 
> These offsets are incorrect, since supervisor states are only stored in
> the compacted format.  In fact, in patch 4, supervisor states should
> return 0 in CPUID(EAX=0Dh,ECX=n).EBX.
> 
> You can use offset == 0 to distinguish supervisor and user states, so
> that supervisor states are skipped in xsave_area_size and x86_cpu_reset.
> 
> Thanks,
> 
> Paolo
> 
> > +            .size = sizeof(XSaveCETU) },
> > +    [XSTATE_CET_S_BIT] = {
> > +            .feature = FEAT_7_0_ECX, .bits = CPUID_7_0_ECX_CET_SHSTK,
> > +            .offset = offsetof(X86XSaveArea, cet_s),
> > +            .size = sizeof(XSaveCETS) },
> >  };
> >  
> >  static uint32_t xsave_area_size(uint64_t mask)

Re: [Qemu-devel][PATCH 3/4] Add hepler functions for CPUID xsave area size calculation.

[Christophe de Dinechin]

Typo in subject line (helper)

> On 26 Dec 2018, at 09:25, Yang Weijiang <weijiang.yang@intel.com> wrote:
> 
> These functions are called when return CPUID xsave area
> size information.
> 
> Signed-off-by: Zhang Yi <yi.z.zhang@linux.intel.com>
> Signed-off-by: Yang Weijiang <weijiang.yang@intel.com>
> ---
> target/i386/cpu.c | 26 +++++++++++++++++++++++++-
> 1 file changed, 25 insertions(+), 1 deletion(-)
> 
> diff --git a/target/i386/cpu.c b/target/i386/cpu.c
> index 3630c688d6..cf4f2798dc 100644
> --- a/target/i386/cpu.c
> +++ b/target/i386/cpu.c
> @@ -1281,12 +1281,34 @@ static inline bool accel_uses_host_cpuid(void)
>     return kvm_enabled() || hvf_enabled();
> }
> 
> +static uint32_t xsave_area_size_compat(uint64_t mask)

Just curious, why “compat”?

> +{
> +    int i;
> +    uint64_t ret = 0;
> +    uint32_t offset;
> +
> +    for (i = 0; i < ARRAY_SIZE(x86_ext_save_areas); i++) {
> +        const ExtSaveArea *esa = &x86_ext_save_areas[i];
> +        offset = i > 1 ? ret : esa->offset;

What about a named constant instead of ‘1’?

(note that a loop around line 4657 starts at 2, so I guess it’s OK to use 1 here, or at least >= 2 ;-)


> +        if ((mask >> i) & 1) {
> +            ret = MAX(ret, offset + esa->size);
> +        }
> +    }
> +    return ret;
> +}
> +
> static inline uint64_t x86_cpu_xsave_components(X86CPU *cpu)
> {
>     return ((uint64_t)cpu->env.features[FEAT_XSAVE_COMP_HI]) << 32 |
>            cpu->env.features[FEAT_XSAVE_COMP_LO];
> }
> 
> +static inline uint64_t x86_cpu_xsave_sv_components(X86CPU *cpu)
> +{
> +    return ((uint64_t)cpu->env.features[FEAT_XSAVE_SV_HI]) << 32 |
> +           cpu->env.features[FEAT_XSAVE_SV_LO];
> +}
> +
> const char *get_register_name_32(unsigned int reg)
> {
>     if (reg >= CPU_NB_REGS32) {
> @@ -4913,8 +4935,10 @@ static void x86_cpu_enable_xsave_components(X86CPU *cpu)
>         }
>     }
> 
> -    env->features[FEAT_XSAVE_COMP_LO] = mask;
> +    env->features[FEAT_XSAVE_COMP_LO] = mask & CPUID_XSTATE_USER_MASK;
>     env->features[FEAT_XSAVE_COMP_HI] = mask >> 32;
> +    env->features[FEAT_XSAVE_SV_LO] = mask & CPUID_XSTATE_KERNEL_MASK;
> +    env->features[FEAT_XSAVE_SV_HI] = mask >> 32;
> }
> 
> /***** Steps involved on loading and filtering CPUID data
> -- 
> 2.17.1
>