x86_64 Linux: Add fallback for CPUs without RDFSBASE/WRFSBASE

src/macros.S

@@ -64,6 +64,52 @@ m1632:
     endbr64
 .endm
 
+.macro READ_FSBASE out_reg teb_reg
+    push rax
+    movzx rax, byte ptr [\teb_reg+TEB_HAS_FSGSBASE]
+    test rax, rax
+    jz 1f
+    rdfsbase \out_reg
+    jmp 2f
+1:
+    push rdi
+    push rsi
+    push rcx
+    sub rsp, 8
+    mov rsi, rsp    # addr
+    mov rdi, 0x1003 # ARCH_GET_FS
+    mov rax, 158    # SYS_arch_prctl
+    syscall
+    pop \out_reg
+    pop rcx
+    pop rsi
+    pop rdi
+2:
+    pop rax
+.endm
+
+.macro WRITE_FSBASE in_reg teb_reg
+    push rax
+    movzx rax, byte ptr [\teb_reg+TEB_HAS_FSGSBASE]
+    test rax, rax
+    jz 1f
+    wrfsbase \in_reg
+    jmp 2f
+1:
+    push rdi
+    push rsi
+    push rcx
+    mov rsi, \in_reg # addr
+    mov rdi, 0x1002  # ARCH_SET_FS
+    mov rax, 158     # SYS_arch_prctl
+    syscall
+    pop rcx
+    pop rsi
+    pop rdi
+2:
+    pop rax
+.endm
+
 #endif // __x86_64__
 
 .macro GET_TEB_HOST reg

src/macros.h

@@ -12,11 +12,12 @@
 
 #ifdef __x86_64__
 
-#define TEB_CS_SEL 0xf9c // CodeSelector
+#define TEB_CS_SEL 0xf9c	 // CodeSelector
-#define TEB_DS_SEL 0xf9e // DataSelector
+#define TEB_DS_SEL 0xf9e	 // DataSelector
-#define TEB_SP 0xfa0	 // CurrentStackPointer
+#define TEB_SP 0xfa0		 // CurrentStackPointer
-#define TEB_FSBASE 0xfa8 // HostFsBase
+#define TEB_FSBASE 0xfa8	 // HostFsBase
-#define TEB_GSBASE 0xfb0 // HostGsBase
+#define TEB_GSBASE 0xfb0	 // HostGsBase
+#define TEB_HAS_FSGSBASE 0xfb8 // HasFsGsBase
 
 #ifdef __linux__
 #define CS_32 0x23 // 32-bit code segment (Linux)

src/setup_linux.cpp

@@ -11,14 +11,16 @@ namespace {
 std::mutex g_tebSetupMutex;
 int g_entryNumber = -1;
 
-} // namespace
-
 constexpr uint16_t createSelector(int entryNumber) {
 	return static_cast<uint16_t>((entryNumber << 3) | USER_PRIVILEGE);
 }
 
+} // namespace
+
 #if defined(__x86_64__)
 
+#include <cpuid.h>
+
 // Implemented in setup.S
 extern "C" int tebThreadSetup64(int entryNumber, TEB *teb);
 
@@ -37,6 +39,15 @@ bool tebThreadSetup(TEB *teb) {
 
 	teb->CurrentFsSelector = createSelector(ret);
 	teb->CurrentGsSelector = 0;
+
+	// Check for FSBASE/GSBASE instruction support
+	unsigned int regs[4];
+	int cpuidMax = __get_cpuid_max(0, nullptr);
+	if (cpuidMax >= 0x7 && __get_cpuid_count(0x7, 0, &regs[0], &regs[1], &regs[2], &regs[3])) {
+		teb->HasFsGsBase = !!(regs[1] & 1);
+	}
+	DEBUG_LOG("setup_linux: FSBASE/GSBASE instruction support: %s\n", teb->HasFsGsBase ? "yes" : "no");
+
 	return true;
 }
 

src/types.h

@@ -549,6 +549,7 @@ typedef struct _TEB {
 #ifdef __x86_64__
 	void *HostFsBase;
 	void *HostGsBase;
+	bool HasFsGsBase;
 #endif
 } TEB;
 typedef GUEST_PTR PTEB;
@@ -575,6 +576,9 @@ static_assert(offsetof(TEB, HostFsBase) == TEB_FSBASE);
 #ifdef TEB_GSBASE
 static_assert(offsetof(TEB, HostGsBase) == TEB_GSBASE);
 #endif
+#ifdef TEB_HAS_FSGSBASE
+static_assert(offsetof(TEB, HasFsGsBase) == TEB_HAS_FSGSBASE);
+#endif
 
 typedef struct _MEMORY_BASIC_INFORMATION {
 	GUEST_PTR BaseAddress;

tools/gen_trampolines.py

@@ -721,7 +721,7 @@ def emit_cc_thunk64(f: FuncInfo | TypedefInfo, lines: List[str]):
 
         if sys.platform != "darwin":
             # Save FS base
-            lines.append("\trdfsbase r9")
+            lines.append("\tREAD_FSBASE r9, rbx")
             lines.append("\tmov qword ptr [rbx+TEB_FSBASE], r9")
 
         # Save RSP and load guest stack
@@ -782,7 +782,7 @@ def emit_cc_thunk64(f: FuncInfo | TypedefInfo, lines: List[str]):
         if sys.platform != "darwin":
             # Restore FS base
             lines.append("\tmov r9, qword ptr [rbx+TEB_FSBASE]")
-            lines.append("\twrfsbase r9")
+            lines.append("\tWRITE_FSBASE r9, rbx")
 
         # Restore host stack
         lines.append("\tmov rsp, qword ptr [rbx+TEB_SP]")