runtime: check a microarchitecture level at startup

Make Go runtime throw if it's been compiled to assume instruction
set extensions that aren't available on the CPU. This feature is
originally suggested by mdempsky in #45453 and #25489.
Updates #45453

Author: vpachkov

src/cmd/compile/internal/gc/obj.go

@@ -17,8 +17,10 @@ import (
 	"cmd/internal/bio"
 	"cmd/internal/obj"
 	"cmd/internal/objabi"
+	"cmd/internal/sys"
 	"encoding/json"
 	"fmt"
+	"internal/buildcfg"
 )
 
 // These modes say which kind of object file to generate.
@@ -162,6 +164,8 @@ func dumpdata() {
 	if newNumPTabs != numPTabs {
 		base.Fatalf("ptabs changed after compile functions loop")
 	}
+
+	addMicroarch()
 }
 
 func dumpLinkerObj(bout *bio.Writer) {
@@ -264,6 +268,22 @@ func addGCLocals() {
 	}
 }
 
+// addMicroarch creates a special symbol pkgpath + "_microarch"
+// which is used by the linker to recreate information about
+// microarchitecture level that the object was compiled for.
+// see ld/data.go - microarch
+func addMicroarch() {
+	// For now only AMD64 supports microaritecture level configuration.
+	// It's done with GOAMD64 enviromnent variable
+	if base.Ctxt.Arch.Family != sys.AMD64 {
+		return
+	}
+
+	microarchSym := base.Ctxt.Lookup(base.Ctxt.Pkgpath + "._microarch")
+	base.Ctxt.Data = append(base.Ctxt.Data, microarchSym)
+	microarchSym.WriteInt(base.Ctxt, 0, 1, int64(buildcfg.GOAMD64))
+}
+
 func ggloblnod(nam *ir.Name) {
 	s := nam.Linksym()
 	s.Gotype = reflectdata.TypeLinksym(nam.Type())

src/cmd/link/internal/ld/data.go

@@ -2199,6 +2199,34 @@ func (ctxt *Link) buildinfo() {
 	r.SetSym(ldr.LookupOrCreateSym("runtime.modinfo", 0))
 }
 
+// microarch creates a special symbol runtime.microarch that's used by runtime to
+// check a micraarchitecture level that the program was compiled for.
+// see runtime/proc.go - checkmicroarch
+func (ctxt *Link) microarch() {
+	ldr := ctxt.loader
+	s := ldr.CreateSymForUpdate("runtime.microarch", 0)
+
+	var microarchLvl uint8 = 1
+	for pkgname, _ := range ctxt.LibraryByPkg {
+		// find the microarctitecture level that the package was compiled for
+		microarchSym := ldr.Lookup(pkgname+"._microarch", 0)
+		if microarchSym == 0 {
+			continue
+		}
+		d := ldr.Data(microarchSym)
+		pkgMicroarchLvl := d[0]
+		if pkgMicroarchLvl > microarchLvl {
+			microarchLvl = pkgMicroarchLvl
+		}
+	}
+
+	data := make([]byte, 1)
+	data[0] = microarchLvl
+	s.SetData(data)
+	s.SetSize(int64(len(data)))
+	s.SetType(sym.SDATA)
+}
+
 // assign addresses to text
 func (ctxt *Link) textaddress() {
 	addsection(ctxt.loader, ctxt.Arch, &Segtext, ".text", 05)

src/cmd/link/internal/ld/main.go

@@ -309,6 +309,8 @@ func Main(arch *sys.Arch, theArch Arch) {
 	ctxt.typelink()
 	bench.Start("buildinfo")
 	ctxt.buildinfo()
+	bench.Start("microarch")
+	ctxt.microarch()
 	bench.Start("pclntab")
 	containers := ctxt.findContainerSyms()
 	pclnState := ctxt.pclntab(containers)

src/internal/cpu/cpu.go

@@ -24,24 +24,34 @@ var CacheLineSize uintptr = CacheLinePadSize
 // in addition to the cpuid feature bit being set.
 // The struct is padded to avoid false sharing.
 var X86 struct {
-	_            CacheLinePad
-	HasAES       bool
-	HasADX       bool
-	HasAVX       bool
-	HasAVX2      bool
-	HasBMI1      bool
-	HasBMI2      bool
-	HasERMS      bool
-	HasFMA       bool
-	HasOSXSAVE   bool
-	HasPCLMULQDQ bool
-	HasPOPCNT    bool
-	HasRDTSCP    bool
-	HasSSE3      bool
-	HasSSSE3     bool
-	HasSSE41     bool
-	HasSSE42     bool
-	_            CacheLinePad
+	_             CacheLinePad
+	HasABM        bool
+	HasAES        bool
+	HasADX        bool
+	HasAVX        bool
+	HasAVX2       bool
+	HasAVX512F    bool
+	HasAVX512BW   bool
+	HasAVX512CD   bool
+	HasAVX512DQ   bool
+	HasAVX512VL   bool
+	HasBMI1       bool
+	HasBMI2       bool
+	HasCMPXCHG16B bool
+	HasERMS       bool
+	HasFMA        bool
+	HasF16C       bool
+	HasLAHF       bool
+	HasMOVBE      bool
+	HasOSXSAVE    bool
+	HasPCLMULQDQ  bool
+	HasPOPCNT     bool
+	HasRDTSCP     bool
+	HasSSE3       bool
+	HasSSSE3      bool
+	HasSSE41      bool
+	HasSSE42      bool
+	_             CacheLinePad
 }
 
 // The booleans in ARM contain the correspondingly named cpu feature bit.

src/internal/cpu/cpu_x86.go

@@ -20,26 +20,38 @@ const (
 	cpuid_SSE2 = 1 << 26
 
 	// ecx bits
-	cpuid_SSE3      = 1 << 0
-	cpuid_PCLMULQDQ = 1 << 1
-	cpuid_SSSE3     = 1 << 9
-	cpuid_FMA       = 1 << 12
-	cpuid_SSE41     = 1 << 19
-	cpuid_SSE42     = 1 << 20
-	cpuid_POPCNT    = 1 << 23
-	cpuid_AES       = 1 << 25
-	cpuid_OSXSAVE   = 1 << 27
-	cpuid_AVX       = 1 << 28
+	cpuid_SSE3       = 1 << 0
+	cpuid_PCLMULQDQ  = 1 << 1
+	cpuid_SSSE3      = 1 << 9
+	cpuid_FMA        = 1 << 12
+	cpuid_CMPXCHG16B = 1 << 13
+	cpuid_SSE41      = 1 << 19
+	cpuid_SSE42      = 1 << 20
+	cpuid_MOVBE      = 1 << 22
+	cpuid_POPCNT     = 1 << 23
+	cpuid_AES        = 1 << 25
+	cpuid_OSXSAVE    = 1 << 27
+	cpuid_AVX        = 1 << 28
+	cpuid_F16C       = 1 << 29
 
 	// ebx bits
-	cpuid_BMI1 = 1 << 3
-	cpuid_AVX2 = 1 << 5
-	cpuid_BMI2 = 1 << 8
-	cpuid_ERMS = 1 << 9
-	cpuid_ADX  = 1 << 19
+	cpuid_BMI1     = 1 << 3
+	cpuid_AVX2     = 1 << 5
+	cpuid_BMI2     = 1 << 8
+	cpuid_ERMS     = 1 << 9
+	cpuid_ADX      = 1 << 19
+	cpuid_AVX512F  = 1 << 16
+	cpuid_AVX512DQ = 1 << 17
+	cpuid_AVX512CD = 1 << 28
+	cpuid_AVX512BW = 1 << 30
+	cpuid_AVX512VL = 1 << 31
 
 	// edx bits for CPUID 0x80000001
 	cpuid_RDTSCP = 1 << 27
+
+	// ecx bits for CPUID 0x80000001
+	cpuid_LAHF = 1 << 0
+	cpuid_ABM  = 1 << 5
 )
 
 var maxExtendedFunctionInformation uint32
@@ -80,6 +92,9 @@ func doinit() {
 	X86.HasSSE42 = isSet(ecx1, cpuid_SSE42)
 	X86.HasPOPCNT = isSet(ecx1, cpuid_POPCNT)
 	X86.HasAES = isSet(ecx1, cpuid_AES)
+	X86.HasMOVBE = isSet(ecx1, cpuid_MOVBE)
+	X86.HasCMPXCHG16B = isSet(ecx1, cpuid_CMPXCHG16B)
+	X86.HasF16C = isSet(ecx1, cpuid_F16C)
 
 	// OSXSAVE can be false when using older Operating Systems
 	// or when explicitly disabled on newer Operating Systems by
@@ -112,6 +127,11 @@ func doinit() {
 	X86.HasBMI2 = isSet(ebx7, cpuid_BMI2)
 	X86.HasERMS = isSet(ebx7, cpuid_ERMS)
 	X86.HasADX = isSet(ebx7, cpuid_ADX)
+	X86.HasAVX512F = isSet(ebx7, cpuid_AVX512F)
+	X86.HasAVX512BW = isSet(ebx7, cpuid_AVX512BW)
+	X86.HasAVX512CD = isSet(ebx7, cpuid_AVX512CD)
+	X86.HasAVX512DQ = isSet(ebx7, cpuid_AVX512DQ)
+	X86.HasAVX512VL = isSet(ebx7, cpuid_AVX512VL)
 
 	var maxExtendedInformation uint32
 	maxExtendedInformation, _, _, _ = cpuid(0x80000000, 0)
@@ -120,8 +140,11 @@ func doinit() {
 		return
 	}
 
-	_, _, _, edxExt1 := cpuid(0x80000001, 0)
+	_, _, ecxExt1, edxExt1 := cpuid(0x80000001, 0)
 	X86.HasRDTSCP = isSet(edxExt1, cpuid_RDTSCP)
+
+	X86.HasLAHF = isSet(ecxExt1, cpuid_LAHF)
+	X86.HasABM = isSet(ecxExt1, cpuid_ABM)
 }
 
 func isSet(hwc uint32, value uint32) bool {

src/runtime/proc.go

@@ -636,6 +636,51 @@ func cpuinit() {
 	}
 }
 
+//go:linkname microarch runtime.microarch
+var microarch uint8
+
+// checkmicroarch checks if a platform supports a micraarchitecture level that the program was compiled for.
+// https://en.wikipedia.org/wiki/X86-64#Microarchitecture_levels - more about AMD64 micraarchitecture levels
+func checkmicroarch() {
+	missmatchHandler := func(platform int) {
+		println("Could not run this program because it was compiled for microarhitecture level", microarch)
+		println("But platform microarhitecture level is less than", platform)
+		throw("Microarhitecture level missmatch")
+	}
+
+	if goarch.IsAmd64 > 0 {
+		x86 := cpu.X86
+		if microarch >= 2 {
+			if !(x86.HasPOPCNT &&
+				x86.HasCMPXCHG16B &&
+				x86.HasLAHF &&
+				x86.HasPOPCNT &&
+				x86.HasSSE3 &&
+				x86.HasSSE41 &&
+				x86.HasSSE42 &&
+				x86.HasSSSE3) {
+				missmatchHandler(2)
+			}
+		}
+		if microarch >= 3 {
+			if !(x86.HasAVX && x86.HasAVX2 &&
+				x86.HasBMI1 && x86.HasBMI2 &&
+				x86.HasF16C &&
+				x86.HasFMA &&
+				x86.HasABM && // for LZCNT
+				x86.HasMOVBE &&
+				x86.HasOSXSAVE) {
+				missmatchHandler(3)
+			}
+		}
+		if microarch >= 4 {
+			if !(x86.HasAVX512F && x86.HasAVX512BW && x86.HasAVX512CD && x86.HasAVX512DQ && x86.HasAVX512VL) {
+				missmatchHandler(4)
+			}
+		}
+	}
+}
+
 // The bootstrap sequence is:
 //
 //	call osinit
@@ -682,7 +727,8 @@ func schedinit() {
 	mallocinit()
 	fastrandinit() // must run before mcommoninit
 	mcommoninit(_g_.m, -1)
-	cpuinit()       // must run before alginit
+	cpuinit() // must run before alginit
+	checkmicroarch()
 	alginit()       // maps must not be used before this call
 	modulesinit()   // provides activeModules
 	typelinksinit() // uses maps, activeModules