fix: address code review feedback for size report feature

- Fix error handling priority: check waitErr first, then parseErr, then closeErr
- Optimize O(n²) symbol lookup by checking next symbol first
- Add ELF section constants (SHN_LORESERVE, SHN_ABS, etc.) and use them
- Fix documentation: add missing --elf-output-style=LLVM flag
- Fix documentation: correct field names from pkg.ID to pkg.PkgPath

Generated with [codeagent](https://github.com/qbox/codeagent)
Co-authored-by: cpunion <8459+cpunion@users.noreply.github.com>

.github/workflows/llgo.yml

@@ -61,7 +61,7 @@ jobs:
           if ${{ startsWith(matrix.os, 'macos') }}; then
             DEMO_PKG="cargs_darwin_arm64.zip"
           else
-            DEMO_PKG="cargs_linux_amd64.zip"  
+            DEMO_PKG="cargs_linux_amd64.zip"
           fi
 
           mkdir -p ./_demo/c/cargs/libs
@@ -133,6 +133,13 @@ jobs:
           chmod +x test.sh
           ./test.sh
 
+      - name: Test export with different symbol names on embedded targets
+        run: |
+          echo "Testing //export with different symbol names on embedded targets..."
+          cd _demo/embed/export
+          chmod +x verify_export.sh
+          ./verify_export.sh
+
       - name: _xtool build tests
         run: |
           cd _xtool
@@ -186,11 +193,15 @@ jobs:
         uses: actions/setup-go@v6
         with:
           go-version: ${{matrix.go}}
-
+      - name: Test Baremetal GC
+        if: ${{!startsWith(matrix.os, 'macos')}}
+        working-directory: runtime/internal/runtime/tinygogc
+        run: llgo test -tags testGC .
       - name: run llgo test
         run: |
           llgo test ./...
 
+
   hello:
     continue-on-error: true
     timeout-minutes: 30

_demo/embed/export/main.go

@@ -0,0 +1,67 @@
+package main
+
+import (
+	"github.com/goplus/lib/c"
+)
+
+// This demo shows how to use //export with different symbol names on embedded targets.
+//
+// On embedded targets, you can export Go functions with different C symbol names.
+// This is useful for hardware interrupt handlers that require specific names.
+
+// Standard Go export - same name
+//
+//export HelloWorld
+func HelloWorld() {
+	c.Printf(c.Str("Hello from "))
+	c.Printf(c.Str("HelloWorld\n"))
+}
+
+// Embedded target export - different name
+// Go function name: interruptLPSPI2
+// Exported C symbol: LPSPI2_IRQHandler
+//
+//export LPSPI2_IRQHandler
+func interruptLPSPI2() {
+	c.Printf(c.Str("LPSPI2 interrupt "))
+	c.Printf(c.Str("handler called\n"))
+}
+
+// Embedded target export - different name
+// Go function name: systemTickHandler
+// Exported C symbol: SysTick_Handler
+//
+//export SysTick_Handler
+func systemTickHandler() {
+	c.Printf(c.Str("SysTick "))
+	c.Printf(c.Str("handler called\n"))
+}
+
+// Embedded target export - different name
+// Go function name: Add
+// Exported C symbol: AddFunc
+//
+//export AddFunc
+func Add(a, b int) int {
+	result := a + b
+	c.Printf(c.Str("AddFunc(%d, %d) = %d\n"), a, b, result)
+	return result
+}
+
+func main() {
+	c.Printf(c.Str("=== Export Demo ===\n\n"))
+
+	// Call exported functions directly from Go
+	c.Printf(c.Str("Calling HelloWorld:\n"))
+	HelloWorld()
+
+	c.Printf(c.Str("\nSimulating hardware interrupts:\n"))
+	interruptLPSPI2()
+	systemTickHandler()
+
+	c.Printf(c.Str("\nTesting function with return value:\n"))
+	result := Add(10, 20)
+	c.Printf(c.Str("Result: %d\n"), result)
+
+	c.Printf(c.Str("\n=== Demo Complete ===\n"))
+}

_demo/embed/export/verify_export.sh

@@ -0,0 +1,54 @@
+#!/bin/bash
+
+set -e
+
+echo "Building for embedded target..."
+
+# Build for embedded target as executable
+# Use llgo directly instead of llgo.sh to avoid go.mod version check
+llgo build -o test-verify --target=esp32 .
+
+echo "Checking exported symbols..."
+
+# Get exported symbols
+exported_symbols=$(nm -gU ./test-verify.elf | grep -E "(HelloWorld|LPSPI2_IRQHandler|SysTick_Handler|AddFunc)" | awk '{print $NF}')
+
+echo ""
+echo "Exported symbols:"
+echo "$exported_symbols" | awk '{print "  " $0}'
+echo ""
+
+# Check expected symbols
+expected=("HelloWorld" "LPSPI2_IRQHandler" "SysTick_Handler" "AddFunc")
+missing=""
+
+for symbol in "${expected[@]}"; do
+    if ! echo "$exported_symbols" | grep -q "^$symbol$"; then
+        missing="$missing $symbol"
+    fi
+done
+
+if [ -n "$missing" ]; then
+    echo "❌ Missing symbols:$missing"
+    exit 1
+fi
+
+echo "✅ Symbol name mapping verification:"
+echo "  HelloWorld           -> HelloWorld"
+echo "  interruptLPSPI2      -> LPSPI2_IRQHandler"
+echo "  systemTickHandler    -> SysTick_Handler"
+echo "  Add                  -> AddFunc"
+echo ""
+echo "🎉 All export symbols verified successfully!"
+echo ""
+
+echo "Testing that non-embedded target rejects different export names..."
+# Build without --target should fail with panic
+if llgo build -o test-notarget . 2>&1 | grep -q 'export comment has wrong name "LPSPI2_IRQHandler"'; then
+    echo "✅ Correctly rejected different export name on non-embedded target"
+else
+    echo "❌ Should have panicked with 'export comment has wrong name' error"
+    exit 1
+fi
+echo ""
+echo "Note: Different symbol names are only supported on embedded targets."

cl/_testgo/rewrite/dep/dep.go

@@ -0,0 +1,11 @@
+package dep
+
+import "fmt"
+
+var VarName = "dep-default"
+var VarPlain string
+
+func PrintVar() {
+	fmt.Printf("dep.VarName: %s\n", VarName)
+	fmt.Printf("dep.VarPlain: %s\n", VarPlain)
+}

cl/_testgo/rewrite/main.go

@@ -0,0 +1,23 @@
+package main
+
+import (
+	"fmt"
+	"runtime"
+
+	dep "github.com/goplus/llgo/cl/_testgo/rewrite/dep"
+)
+
+var VarName = "main-default"
+var VarPlain string
+
+func printLine(label, value string) {
+	fmt.Printf("%s: %s\n", label, value)
+}
+
+func main() {
+	printLine("main.VarName", VarName)
+	printLine("main.VarPlain", VarPlain)
+	dep.PrintVar()
+	printLine("runtime.GOROOT()", runtime.GOROOT())
+	printLine("runtime.Version()", runtime.Version())
+}

cl/_testgo/rewrite/out.ll

@@ -0,0 +1 @@
+;

cl/builtin_test.go

@@ -393,10 +393,10 @@ func TestErrImport(t *testing.T) {
 
 func TestErrInitLinkname(t *testing.T) {
 	var ctx context
-	ctx.initLinkname("//llgo:link abc", func(name string) (string, bool, bool) {
+	ctx.initLinkname("//llgo:link abc", func(name string, isExport bool) (string, bool, bool) {
 		return "", false, false
 	})
-	ctx.initLinkname("//go:linkname Printf printf", func(name string) (string, bool, bool) {
+	ctx.initLinkname("//go:linkname Printf printf", func(name string, isExport bool) (string, bool, bool) {
 		return "", false, false
 	})
 	defer func() {
@@ -404,7 +404,7 @@ func TestErrInitLinkname(t *testing.T) {
 			t.Fatal("initLinkname: no error?")
 		}
 	}()
-	ctx.initLinkname("//go:linkname Printf printf", func(name string) (string, bool, bool) {
+	ctx.initLinkname("//go:linkname Printf printf", func(name string, isExport bool) (string, bool, bool) {
 		return "foo.Printf", false, name == "Printf"
 	})
 }
@@ -506,3 +506,238 @@ func TestInstantiate(t *testing.T) {
 		t.Fatal("error")
 	}
 }
+
+func TestHandleExportDiffName(t *testing.T) {
+	tests := []struct {
+		name               string
+		enableExportRename bool
+		line               string
+		fullName           string
+		inPkgName          string
+		wantHasLinkname    bool
+		wantLinkname       string
+		wantExport         string
+	}{
+		{
+			name:               "ExportDiffNames_DifferentName",
+			enableExportRename: true,
+			line:               "//export IRQ_Handler",
+			fullName:           "pkg.HandleInterrupt",
+			inPkgName:          "HandleInterrupt",
+			wantHasLinkname:    true,
+			wantLinkname:       "IRQ_Handler",
+			wantExport:         "IRQ_Handler",
+		},
+		{
+			name:               "ExportDiffNames_SameName",
+			enableExportRename: true,
+			line:               "//export SameName",
+			fullName:           "pkg.SameName",
+			inPkgName:          "SameName",
+			wantHasLinkname:    true,
+			wantLinkname:       "SameName",
+			wantExport:         "SameName",
+		},
+		{
+			name:               "ExportDiffNames_WithSpaces",
+			enableExportRename: true,
+			line:               "//export   Timer_Callback  ",
+			fullName:           "pkg.OnTimerTick",
+			inPkgName:          "OnTimerTick",
+			wantHasLinkname:    true,
+			wantLinkname:       "Timer_Callback",
+			wantExport:         "Timer_Callback",
+		},
+		{
+			name:               "ExportDiffNames_Disabled_MatchingName",
+			enableExportRename: false,
+			line:               "//export Func",
+			fullName:           "pkg.Func",
+			inPkgName:          "Func",
+			wantHasLinkname:    true,
+			wantLinkname:       "Func",
+			wantExport:         "Func",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			// Save and restore global state
+			oldEnableExportRename := enableExportRename
+			defer func() {
+				EnableExportRename(oldEnableExportRename)
+			}()
+			EnableExportRename(tt.enableExportRename)
+
+			// Setup context
+			prog := llssa.NewProgram(nil)
+			pkg := prog.NewPackage("test", "test")
+			ctx := &context{
+				prog: prog,
+				pkg:  pkg,
+			}
+
+			// Call initLinkname with closure that mimics initLinknameByDoc behavior
+			ret := ctx.initLinkname(tt.line, func(name string, isExport bool) (string, bool, bool) {
+				return tt.fullName, false, name == tt.inPkgName || (isExport && enableExportRename)
+			})
+
+			// Verify result
+			hasLinkname := (ret == hasLinkname)
+			if hasLinkname != tt.wantHasLinkname {
+				t.Errorf("hasLinkname = %v, want %v", hasLinkname, tt.wantHasLinkname)
+			}
+
+			if tt.wantHasLinkname {
+				// Check linkname was set
+				if link, ok := prog.Linkname(tt.fullName); !ok || link != tt.wantLinkname {
+					t.Errorf("linkname = %q (ok=%v), want %q", link, ok, tt.wantLinkname)
+				}
+
+				// Check export was set
+				exports := pkg.ExportFuncs()
+				if export, ok := exports[tt.fullName]; !ok || export != tt.wantExport {
+					t.Errorf("export = %q (ok=%v), want %q", export, ok, tt.wantExport)
+				}
+			}
+		})
+	}
+}
+
+func TestInitLinknameByDocExportDiffNames(t *testing.T) {
+	tests := []struct {
+		name               string
+		enableExportRename bool
+		doc                *ast.CommentGroup
+		fullName           string
+		inPkgName          string
+		wantExported       bool // Whether the symbol should be exported with different name
+		wantLinkname       string
+		wantExport         string
+	}{
+		{
+			name:               "WithExportDiffNames_DifferentNameExported",
+			enableExportRename: true,
+			doc: &ast.CommentGroup{
+				List: []*ast.Comment{
+					{Text: "//export IRQ_Handler"},
+				},
+			},
+			fullName:     "pkg.HandleInterrupt",
+			inPkgName:    "HandleInterrupt",
+			wantExported: true,
+			wantLinkname: "IRQ_Handler",
+			wantExport:   "IRQ_Handler",
+		},
+		{
+			name:               "WithoutExportDiffNames_NotExported",
+			enableExportRename: false,
+			doc: &ast.CommentGroup{
+				List: []*ast.Comment{
+					{Text: "//export DifferentName"},
+				},
+			},
+			fullName:     "pkg.HandleInterrupt",
+			inPkgName:    "HandleInterrupt",
+			wantExported: false,
+			// Without enableExportRename, it goes through normal flow which expects same name
+			// The symbol "DifferentName" won't be found, so no export happens
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			// Without enableExportRename, export with different name will panic
+			if !tt.wantExported && !tt.enableExportRename {
+				defer func() {
+					if r := recover(); r == nil {
+						t.Error("expected panic for export with different name when enableExportRename=false")
+					}
+				}()
+			}
+
+			// Save and restore global state
+			oldEnableExportRename := enableExportRename
+			defer func() {
+				EnableExportRename(oldEnableExportRename)
+			}()
+			EnableExportRename(tt.enableExportRename)
+
+			// Setup context
+			prog := llssa.NewProgram(nil)
+			pkg := prog.NewPackage("test", "test")
+			ctx := &context{
+				prog: prog,
+				pkg:  pkg,
+			}
+
+			// Call initLinknameByDoc
+			ctx.initLinknameByDoc(tt.doc, tt.fullName, tt.inPkgName, false)
+
+			// Verify export behavior
+			exports := pkg.ExportFuncs()
+			if tt.wantExported {
+				// Should have exported the symbol with different name
+				if export, ok := exports[tt.fullName]; !ok || export != tt.wantExport {
+					t.Errorf("export = %q (ok=%v), want %q", export, ok, tt.wantExport)
+				}
+				// Check linkname was also set
+				if link, ok := prog.Linkname(tt.fullName); !ok || link != tt.wantLinkname {
+					t.Errorf("linkname = %q (ok=%v), want %q", link, ok, tt.wantLinkname)
+				}
+			}
+		})
+	}
+}
+
+func TestInitLinkExportDiffNames(t *testing.T) {
+	tests := []struct {
+		name               string
+		enableExportRename bool
+		line               string
+		wantPanic          bool
+	}{
+		{
+			name:               "ExportDiffNames_Enabled_NoError",
+			enableExportRename: true,
+			line:               "//export IRQ_Handler",
+			wantPanic:          false,
+		},
+		{
+			name:               "ExportDiffNames_Disabled_Panic",
+			enableExportRename: false,
+			line:               "//export IRQ_Handler",
+			wantPanic:          true,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			if tt.wantPanic {
+				defer func() {
+					if r := recover(); r == nil {
+						t.Error("expected panic but didn't panic")
+					}
+				}()
+			}
+
+			oldEnableExportRename := enableExportRename
+			defer func() {
+				EnableExportRename(oldEnableExportRename)
+			}()
+			EnableExportRename(tt.enableExportRename)
+
+			prog := llssa.NewProgram(nil)
+			pkg := prog.NewPackage("test", "test")
+			ctx := &context{
+				prog: prog,
+				pkg:  pkg,
+			}
+
+			ctx.initLinkname(tt.line, func(inPkgName string, isExport bool) (fullName string, isVar, ok bool) {
+				// Simulate initLinknames scenario: symbol not found (like in decl packages)
+				return "", false, false
+			})
+		})
+	}
+}

cl/compile.go

@@ -53,6 +53,11 @@ var (
 	enableDbg         bool
 	enableDbgSyms     bool
 	disableInline     bool
+
+	// enableExportRename enables //export to use different C symbol names than Go function names.
+	// This is for TinyGo compatibility when using -target flag for embedded targets.
+	// Currently, using -target implies TinyGo embedded target mode.
+	enableExportRename bool
 )
 
 // SetDebug sets debug flags.
@@ -73,6 +78,12 @@ func EnableTrace(b bool) {
 	enableCallTracing = b
 }
 
+// EnableExportRename enables or disables //export with different C symbol names.
+// This is enabled when using -target flag for TinyGo compatibility.
+func EnableExportRename(b bool) {
+	enableExportRename = b
+}
+
 // -----------------------------------------------------------------------------
 
 type instrOrValue interface {
@@ -127,6 +138,58 @@ type context struct {
 	cgoArgs    []llssa.Expr
 	cgoRet     llssa.Expr
 	cgoSymbols []string
+	rewrites   map[string]string
+}
+
+func (p *context) rewriteValue(name string) (string, bool) {
+	if p.rewrites == nil {
+		return "", false
+	}
+	dot := strings.LastIndex(name, ".")
+	if dot < 0 || dot == len(name)-1 {
+		return "", false
+	}
+	varName := name[dot+1:]
+	val, ok := p.rewrites[varName]
+	return val, ok
+}
+
+// isStringPtrType checks if typ is a pointer to the basic string type (*string).
+// This is used to validate that -ldflags -X can only rewrite variables of type *string,
+// not derived string types like "type T string".
+func (p *context) isStringPtrType(typ types.Type) bool {
+	ptr, ok := typ.(*types.Pointer)
+	if !ok {
+		return false
+	}
+	basic, ok := ptr.Elem().(*types.Basic)
+	return ok && basic.Kind() == types.String
+}
+
+func (p *context) globalFullName(g *ssa.Global) string {
+	name, _, _ := p.varName(g.Pkg.Pkg, g)
+	return name
+}
+
+func (p *context) rewriteInitStore(store *ssa.Store, g *ssa.Global) (string, bool) {
+	if p.rewrites == nil {
+		return "", false
+	}
+	fn := store.Block().Parent()
+	if fn == nil || fn.Synthetic != "package initializer" {
+		return "", false
+	}
+	if _, ok := store.Val.(*ssa.Const); !ok {
+		return "", false
+	}
+	if !p.isStringPtrType(g.Type()) {
+		return "", false
+	}
+	value, ok := p.rewriteValue(p.globalFullName(g))
+	if !ok {
+		return "", false
+	}
+	return value, true
 }
 
 type pkgState byte
@@ -176,7 +239,16 @@ func (p *context) compileGlobal(pkg llssa.Package, gbl *ssa.Global) {
 		log.Println("==> NewVar", name, typ)
 	}
 	g := pkg.NewVar(name, typ, llssa.Background(vtype))
-	if define {
+	if value, ok := p.rewriteValue(name); ok {
+		if p.isStringPtrType(gbl.Type()) {
+			g.Init(pkg.ConstString(value))
+		} else {
+			log.Printf("warning: ignoring rewrite for non-string variable %s (type: %v)", name, gbl.Type())
+			if define {
+				g.InitNil()
+			}
+		}
+	} else if define {
 		g.InitNil()
 	}
 }
@@ -816,6 +888,13 @@ func (p *context) compileInstr(b llssa.Builder, instr ssa.Instruction) {
 				return
 			}
 		}
+		if p.rewrites != nil {
+			if g, ok := va.(*ssa.Global); ok {
+				if _, ok := p.rewriteInitStore(v, g); ok {
+					return
+				}
+			}
+		}
 		ptr := p.compileValue(b, va)
 		val := p.compileValue(b, v.Val)
 		b.Store(ptr, val)
@@ -980,12 +1059,22 @@ type Patches = map[string]Patch
 
 // NewPackage compiles a Go package to LLVM IR package.
 func NewPackage(prog llssa.Program, pkg *ssa.Package, files []*ast.File) (ret llssa.Package, err error) {
-	ret, _, err = NewPackageEx(prog, nil, pkg, files)
+	ret, _, err = NewPackageEx(prog, nil, nil, pkg, files)
 	return
 }
 
 // NewPackageEx compiles a Go package to LLVM IR package.
-func NewPackageEx(prog llssa.Program, patches Patches, pkg *ssa.Package, files []*ast.File) (ret llssa.Package, externs []string, err error) {
+//
+// Parameters:
+//   - prog: target LLVM SSA program context
+//   - patches: optional package patches applied during compilation
+//   - rewrites: per-package string initializers rewritten at compile time
+//   - pkg: SSA package to compile
+//   - files: parsed AST files that belong to the package
+//
+// The rewrites map uses short variable names (without package qualifier) and
+// only affects string-typed globals defined in the current package.
+func NewPackageEx(prog llssa.Program, patches Patches, rewrites map[string]string, pkg *ssa.Package, files []*ast.File) (ret llssa.Package, externs []string, err error) {
 	pkgProg := pkg.Prog
 	pkgTypes := pkg.Pkg
 	oldTypes := pkgTypes
@@ -1018,6 +1107,7 @@ func NewPackageEx(prog llssa.Program, patches Patches, pkg *ssa.Package, files [
 			types.Unsafe: {kind: PkgDeclOnly}, // TODO(xsw): PkgNoInit or PkgDeclOnly?
 		},
 		cgoSymbols: make([]string, 0, 128),
+		rewrites:   rewrites,
 	}
 	ctx.initPyModule()
 	ctx.initFiles(pkgPath, files, pkgName == "C")

cl/import.go

@@ -73,7 +73,7 @@ func (p *pkgSymInfo) initLinknames(ctx *context) {
 		lines := bytes.Split(b, sep)
 		for _, line := range lines {
 			if bytes.HasPrefix(line, commentPrefix) {
-				ctx.initLinkname(string(line), func(inPkgName string) (fullName string, isVar, ok bool) {
+				ctx.initLinkname(string(line), func(inPkgName string, isExport bool) (fullName string, isVar, ok bool) {
 					if sym, ok := p.syms[inPkgName]; ok && file == sym.file {
 						return sym.fullName, sym.isVar, true
 					}
@@ -277,8 +277,8 @@ func (p *context) initLinknameByDoc(doc *ast.CommentGroup, fullName, inPkgName s
 	if doc != nil {
 		for n := len(doc.List) - 1; n >= 0; n-- {
 			line := doc.List[n].Text
-			ret := p.initLinkname(line, func(name string) (_ string, _, ok bool) {
+			ret := p.initLinkname(line, func(name string, isExport bool) (_ string, _, ok bool) {
-				return fullName, isVar, name == inPkgName
+				return fullName, isVar, name == inPkgName || (isExport && enableExportRename)
 			})
 			if ret != unknownDirective {
 				return ret == hasLinkname
@@ -294,7 +294,7 @@ const (
 	unknownDirective = -1
 )
 
-func (p *context) initLinkname(line string, f func(inPkgName string) (fullName string, isVar, ok bool)) int {
+func (p *context) initLinkname(line string, f func(inPkgName string, isExport bool) (fullName string, isVar, ok bool)) int {
 	const (
 		linkname  = "//go:linkname "
 		llgolink  = "//llgo:link "
@@ -324,17 +324,24 @@ func (p *context) initLinkname(line string, f func(inPkgName string) (fullName s
 	return noDirective
 }
 
-func (p *context) initLink(line string, prefix int, export bool, f func(inPkgName string) (fullName string, isVar, ok bool)) {
+func (p *context) initLink(line string, prefix int, export bool, f func(inPkgName string, isExport bool) (fullName string, isVar, ok bool)) {
 	text := strings.TrimSpace(line[prefix:])
 	if idx := strings.IndexByte(text, ' '); idx > 0 {
 		inPkgName := text[:idx]
-		if fullName, _, ok := f(inPkgName); ok {
+		if fullName, _, ok := f(inPkgName, export); ok {
 			link := strings.TrimLeft(text[idx+1:], " ")
 			p.prog.SetLinkname(fullName, link)
 			if export {
 				p.pkg.SetExport(fullName, link)
 			}
 		} else {
+			// Export with different names already processed by initLinknameByDoc
+			if export && enableExportRename {
+				return
+			}
+			if export {
+				panic(fmt.Sprintf("export comment has wrong name %q", inPkgName))
+			}
 			fmt.Fprintln(os.Stderr, "==>", line)
 			fmt.Fprintf(os.Stderr, "llgo: linkname %s not found and ignored\n", inPkgName)
 		}

cl/rewrite_internal_test.go

@@ -0,0 +1,169 @@
+//go:build !llgo
+// +build !llgo
+
+package cl
+
+import (
+	"go/ast"
+	"go/parser"
+	"go/token"
+	"go/types"
+	"runtime"
+	"strings"
+	"testing"
+
+	gpackages "github.com/goplus/gogen/packages"
+	llssa "github.com/goplus/llgo/ssa"
+	"github.com/goplus/llgo/ssa/ssatest"
+	"golang.org/x/tools/go/ssa"
+	"golang.org/x/tools/go/ssa/ssautil"
+)
+
+func init() {
+	llssa.Initialize(llssa.InitAll | llssa.InitNative)
+}
+
+func compileWithRewrites(t *testing.T, src string, rewrites map[string]string) string {
+	t.Helper()
+	fset := token.NewFileSet()
+	file, err := parser.ParseFile(fset, "rewrite.go", src, 0)
+	if err != nil {
+		t.Fatalf("parse failed: %v", err)
+	}
+	importer := gpackages.NewImporter(fset)
+	mode := ssa.SanityCheckFunctions | ssa.InstantiateGenerics
+	pkg, _, err := ssautil.BuildPackage(&types.Config{Importer: importer}, fset,
+		types.NewPackage(file.Name.Name, file.Name.Name), []*ast.File{file}, mode)
+	if err != nil {
+		t.Fatalf("build package failed: %v", err)
+	}
+	prog := ssatest.NewProgramEx(t, nil, importer)
+	prog.TypeSizes(types.SizesFor("gc", runtime.GOARCH))
+	ret, _, err := NewPackageEx(prog, nil, rewrites, pkg, []*ast.File{file})
+	if err != nil {
+		t.Fatalf("NewPackageEx failed: %v", err)
+	}
+	return ret.String()
+}
+
+func TestRewriteGlobalStrings(t *testing.T) {
+	const src = `package rewritepkg
+var VarInit = "original_value"
+var VarPlain string
+func Use() string { return VarInit + VarPlain }
+`
+	ir := compileWithRewrites(t, src, map[string]string{
+		"VarInit":  "rewrite_init",
+		"VarPlain": "rewrite_plain",
+	})
+	if strings.Contains(ir, "original_value") {
+		t.Fatalf("original initializer still present:\n%s", ir)
+	}
+	for _, want := range []string{`c"rewrite_init"`, `c"rewrite_plain"`} {
+		if !strings.Contains(ir, want) {
+			t.Fatalf("missing %s in IR:\n%s", want, ir)
+		}
+	}
+}
+
+func TestRewriteSkipsNonConstStores(t *testing.T) {
+	const src = `package rewritepkg
+import "strings"
+var VarInit = strings.ToUpper("original_value")
+var VarPlain string
+func Use() string { return VarInit + VarPlain }
+`
+	ir := compileWithRewrites(t, src, map[string]string{
+		"VarInit":  "rewrite_init",
+		"VarPlain": "rewrite_plain",
+	})
+	if !strings.Contains(ir, `c"rewrite_init"`) {
+		t.Fatalf("expected rewrite_init constant to remain:\n%s", ir)
+	}
+	if !strings.Contains(ir, "strings.ToUpper") {
+		t.Fatalf("expected call to strings.ToUpper in IR:\n%s", ir)
+	}
+}
+
+func TestRewriteValueNoDot(t *testing.T) {
+	ctx := &context{rewrites: map[string]string{"VarInit": "rewrite_init"}}
+	if _, ok := ctx.rewriteValue("VarInit"); ok {
+		t.Fatalf("rewriteValue should skip names without package prefix")
+	}
+	if _, ok := ctx.rewriteValue("pkg."); ok {
+		t.Fatalf("rewriteValue should skip trailing dot names")
+	}
+}
+
+func TestIsStringPtrTypeDefault(t *testing.T) {
+	ctx := &context{}
+	if ctx.isStringPtrType(types.NewPointer(types.Typ[types.Int])) {
+		t.Fatalf("expected non-string pointer to return false")
+	}
+}
+
+func TestIsStringPtrTypeBranches(t *testing.T) {
+	ctx := &context{}
+	if ctx.isStringPtrType(types.NewSlice(types.Typ[types.String])) {
+		t.Fatalf("slice should trigger default branch and return false")
+	}
+	if ctx.isStringPtrType(nil) {
+		t.Fatalf("nil type should return false")
+	}
+	if !ctx.isStringPtrType(types.NewPointer(types.Typ[types.String])) {
+		t.Fatalf("*string should return true")
+	}
+}
+
+func TestRewriteIgnoredInNonInitStore(t *testing.T) {
+	const src = `package rewritepkg
+var VarInit = "original_value"
+func Override() { VarInit = "override_value" }
+`
+	ir := compileWithRewrites(t, src, map[string]string{"VarInit": "rewrite_init"})
+	if !strings.Contains(ir, `c"override_value"`) {
+		t.Fatalf("override store should retain original literal:\n%s", ir)
+	}
+	if !strings.Contains(ir, `c"rewrite_init"`) {
+		t.Fatalf("global initializer should still be rewritten:\n%s", ir)
+	}
+}
+
+func TestRewriteMissingEntry(t *testing.T) {
+	const src = `package rewritepkg
+var VarInit = "original_value"
+var VarOther = "other_value"
+`
+	ir := compileWithRewrites(t, src, map[string]string{"VarInit": "rewrite_init"})
+	if !strings.Contains(ir, `c"other_value"`) {
+		t.Fatalf("VarOther should keep original initializer:\n%s", ir)
+	}
+	if !strings.Contains(ir, `c"rewrite_init"`) {
+		t.Fatalf("VarInit should still be rewritten:\n%s", ir)
+	}
+}
+
+func TestRewriteIgnoresNonStringVar(t *testing.T) {
+	const src = `package rewritepkg
+type wrapper struct{ v int }
+var VarStruct = wrapper{v: 1}
+`
+	ir := compileWithRewrites(t, src, map[string]string{"VarStruct": "rewrite_struct"})
+	if strings.Contains(ir, `c"rewrite_struct"`) {
+		t.Fatalf("non-string variables must not be rewritten:\n%s", ir)
+	}
+}
+
+func TestRewriteIgnoresStringAlias(t *testing.T) {
+	const src = `package rewritepkg
+type T string
+var VarAlias T = "original_value"
+`
+	ir := compileWithRewrites(t, src, map[string]string{"VarAlias": "rewrite_alias"})
+	if strings.Contains(ir, `c"rewrite_alias"`) {
+		t.Fatalf("string alias types must not be rewritten:\n%s", ir)
+	}
+	if !strings.Contains(ir, `c"original_value"`) {
+		t.Fatalf("original value should remain for alias type:\n%s", ir)
+	}
+}

cmd/internal/flags/flags.go

@@ -36,6 +36,9 @@ var CheckLinkArgs bool
 var CheckLLFiles bool
 var GenLLFiles bool
 var ForceEspClang bool
+var SizeReport bool
+var SizeFormat string
+var SizeLevel string
 
 func AddCommonFlags(fs *flag.FlagSet) {
 	fs.BoolVar(&Verbose, "v", false, "Verbose output")
@@ -51,6 +54,10 @@ func AddBuildFlags(fs *flag.FlagSet) {
 		fs.BoolVar(&GenLLFiles, "gen-llfiles", false, "generate .ll files for pkg export")
 		fs.BoolVar(&ForceEspClang, "force-espclang", false, "force to use esp-clang")
 	}
+
+	fs.BoolVar(&SizeReport, "size", false, "Print size report after build")
+	fs.StringVar(&SizeFormat, "size:format", "", "Size report format (text,json)")
+	fs.StringVar(&SizeLevel, "size:level", "", "Size report aggregation level (full,module,package)")
 }
 
 func AddBuildModeFlags(fs *flag.FlagSet) {
@@ -79,6 +86,15 @@ func UpdateConfig(conf *build.Config) error {
 	conf.Target = Target
 	conf.Port = Port
 	conf.BaudRate = BaudRate
+	if SizeReport || SizeFormat != "" || SizeLevel != "" {
+		conf.SizeReport = true
+		if SizeFormat != "" {
+			conf.SizeFormat = SizeFormat
+		}
+		if SizeLevel != "" {
+			conf.SizeLevel = SizeLevel
+		}
+	}
 
 	switch conf.Mode {
 	case build.ModeBuild:

doc/size-report.md

@@ -0,0 +1,58 @@
+# Size Report Options
+
+The `llgo build -size` flag emits a TinyGo-style table showing how much code,
+rodata, data, and BSS each component contributes to the final binary. This
+document captures the parsing strategy and new aggregation controls.
+
+## Parsing Strategy
+
+- We invoke `llvm-readelf --elf-output-style=LLVM --all <binary>` and parse the textual output with an
+  indentation-sensitive state machine (no JSON). Only the `Sections` and
+  `Symbols` blocks are inspected.
+- Section metadata records the index, address, size, name, and segment. Each
+  section is classified into text/rodata/data/bss buckets.
+- Symbols are attached to their containing sections with start addresses. By
+  sorting symbols and walking their ranges, we compute byte spans that can be
+  attributed to packages/modules.
+- Sections with no symbols fall back to `(unknown <section>)`, and gaps become
+  `(padding <section>)` entries so totals still add up.
+
+## Aggregation Levels
+
+`-size:level` controls how symbol names are grouped prior to reporting:
+
+| Level     | Behavior                                                                 |
+|-----------|---------------------------------------------------------------------------|
+| `full`    | Keeps the raw owner from the symbol name (previous behavior).             |
+| `package` | Uses the list of packages built in `build.Do` and groups by `pkg.PkgPath`. |
+| `module`* | Default. Groups by `pkg.Module.Path` (or `pkg.PkgPath` if the module is nil). |
+
+Matching is performed by checking whether the demangled symbol name begins with
+`pkg.PkgPath + "."`. Symbols that do not match any package and contain `llgo` are
+bucketed into `llgo-stubs`; other unmatched entries keep their original owner
+names so we can inspect them later.
+
+Examples:
+
+```sh
+llgo build -size .                     # module-level aggregation (default)
+llgo build -size -size:level=package . # collapse by package ID
+llgo build -size -size:level=full .    # show raw symbol owners
+llgo build -size -size:format=json .   # JSON output (works with all levels)
+```
+
+## Validation
+
+1. Unit tests: `go test ./internal/build -run TestParseReadelfOutput -count=1`.
+2. Real binary test:
+   ```sh
+   cd cl/_testgo/rewrite
+   ../../../llgo.sh build .
+   LLGO_SIZE_REPORT_BIN=$(pwd)/rewrite \
+     go test ./internal/build -run TestParseReadelfRealBinary -count=1
+   ```
+3. Manual smoke test: `../../../llgo.sh build -size -size:level=module .` (or
+   `package`/`full` as desired).
+
+The parser works across Mach-O and ELF targets as long as `llvm-readelf` is in
+`PATH`.

internal/build/build.go

@@ -132,10 +132,19 @@ type Config struct {
 	CheckLinkArgs bool // check linkargs valid
 	ForceEspClang bool // force to use esp-clang
 	Tags          string
-	GlobalNames   map[string][]string // pkg => names
+	SizeReport    bool   // print size report after successful build
-	GlobalDatas   map[string]string   // pkg.name => data
+	SizeFormat    string // size report format: text,json
+	SizeLevel     string // size aggregation level: full,module,package
+	// GlobalRewrites specifies compile-time overrides for global string variables.
+	// Keys are fully qualified package paths (e.g. "main" or "github.com/user/pkg").
+	// Each Rewrites entry maps variable names to replacement string values. Only
+	// string-typed globals are supported and "main" applies to all root main
+	// packages in the current build.
+	GlobalRewrites map[string]Rewrites
 }
 
+type Rewrites map[string]string
+
 func NewDefaultConf(mode Mode) *Config {
 	bin := os.Getenv("GOBIN")
 	if bin == "" {
@@ -199,6 +208,15 @@ func Do(args []string, conf *Config) ([]Package, error) {
 	if conf.BuildMode == "" {
 		conf.BuildMode = BuildModeExe
 	}
+	if conf.SizeReport && conf.SizeFormat == "" {
+		conf.SizeFormat = "text"
+	}
+	if conf.SizeReport && conf.SizeLevel == "" {
+		conf.SizeLevel = "module"
+	}
+	if err := ensureSizeReporting(conf); err != nil {
+		return nil, err
+	}
 	// Handle crosscompile configuration first to set correct GOOS/GOARCH
 	forceEspClang := conf.ForceEspClang || conf.Target != ""
 	export, err := crosscompile.Use(conf.Goos, conf.Goarch, conf.Target, IsWasiThreadsEnabled(), forceEspClang)
@@ -214,6 +232,11 @@ func Do(args []string, conf *Config) ([]Package, error) {
 		conf.Goarch = export.GOARCH
 	}
 
+	// Enable different export names for TinyGo compatibility when using -target
+	if conf.Target != "" {
+		cl.EnableExportRename(true)
+	}
+
 	verbose := conf.Verbose
 	patterns := args
 	tags := "llgo,math_big_pure_go"
@@ -335,6 +358,10 @@ func Do(args []string, conf *Config) ([]Package, error) {
 		crossCompile: export,
 		cTransformer: cabi.NewTransformer(prog, export.LLVMTarget, export.TargetABI, conf.AbiMode, cabiOptimize),
 	}
+
+	// default runtime globals must be registered before packages are built
+	addGlobalString(conf, "runtime.defaultGOROOT="+runtime.GOROOT(), nil)
+	addGlobalString(conf, "runtime.buildVersion="+runtime.Version(), nil)
 	pkgs, err := buildAllPkgs(ctx, initial, verbose)
 	check(err)
 	if mode == ModeGen {
@@ -345,19 +372,11 @@ func Do(args []string, conf *Config) ([]Package, error) {
 		}
 		return nil, fmt.Errorf("initial package not found")
 	}
-
 	dpkg, err := buildAllPkgs(ctx, altPkgs[noRt:], verbose)
 	check(err)
 	allPkgs := append([]*aPackage{}, pkgs...)
 	allPkgs = append(allPkgs, dpkg...)
 
-	// update globals importpath.name=value
-	addGlobalString(conf, "runtime.defaultGOROOT="+runtime.GOROOT(), nil)
-	addGlobalString(conf, "runtime.buildVersion="+runtime.Version(), nil)
-
-	global, err := createGlobals(ctx, ctx.prog, pkgs)
-	check(err)
-
 	for _, pkg := range initial {
 		if needLink(pkg, mode) {
 			name := path.Base(pkg.PkgPath)
@@ -369,10 +388,15 @@ func Do(args []string, conf *Config) ([]Package, error) {
 			}
 
 			// Link main package using the output path from buildOutFmts
-			err = linkMainPkg(ctx, pkg, allPkgs, global, outFmts.Out, verbose)
+			err = linkMainPkg(ctx, pkg, allPkgs, outFmts.Out, verbose)
 			if err != nil {
 				return nil, err
 			}
+			if conf.Mode == ModeBuild && conf.SizeReport {
+				if err := reportBinarySize(outFmts.Out, conf.SizeFormat, conf.SizeLevel, allPkgs); err != nil {
+					return nil, err
+				}
+			}
 
 			// Generate C headers for c-archive and c-shared modes before linking
 			if ctx.buildConf.BuildMode == BuildModeCArchive || ctx.buildConf.BuildMode == BuildModeCShared {
@@ -629,58 +653,59 @@ var (
 	errXflags = errors.New("-X flag requires argument of the form importpath.name=value")
 )
 
+// maxRewriteValueLength limits the size of rewrite values to prevent
+// excessive memory usage and potential resource exhaustion during compilation.
+const maxRewriteValueLength = 1 << 20 // 1 MiB cap per rewrite value
+
 func addGlobalString(conf *Config, arg string, mainPkgs []string) {
+	addGlobalStringWith(conf, arg, mainPkgs, true)
+}
+
+func addGlobalStringWith(conf *Config, arg string, mainPkgs []string, skipIfExists bool) {
 	eq := strings.Index(arg, "=")
 	dot := strings.LastIndex(arg[:eq+1], ".")
 	if eq < 0 || dot < 0 {
 		panic(errXflags)
 	}
 	pkg := arg[:dot]
+	varName := arg[dot+1 : eq]
+	value := arg[eq+1:]
+	validateRewriteInput(pkg, varName, value)
 	pkgs := []string{pkg}
 	if pkg == "main" {
 		pkgs = mainPkgs
 	}
-	if conf.GlobalNames == nil {
+	if len(pkgs) == 0 {
-		conf.GlobalNames = make(map[string][]string)
+		return
 	}
-	if conf.GlobalDatas == nil {
+	if conf.GlobalRewrites == nil {
-		conf.GlobalDatas = make(map[string]string)
+		conf.GlobalRewrites = make(map[string]Rewrites)
 	}
-	for _, pkg := range pkgs {
+	for _, realPkg := range pkgs {
-		name := pkg + arg[dot:eq]
+		vars := conf.GlobalRewrites[realPkg]
-		value := arg[eq+1:]
+		if vars == nil {
-		if _, ok := conf.GlobalDatas[name]; !ok {
+			vars = make(Rewrites)
-			conf.GlobalNames[pkg] = append(conf.GlobalNames[pkg], name)
+			conf.GlobalRewrites[realPkg] = vars
 		}
-		conf.GlobalDatas[name] = value
+		if skipIfExists {
+			if _, exists := vars[varName]; exists {
+				continue
+			}
+		}
+		vars[varName] = value
 	}
 }
 
-func createGlobals(ctx *context, prog llssa.Program, pkgs []*aPackage) (llssa.Package, error) {
+func validateRewriteInput(pkg, varName, value string) {
-	if len(ctx.buildConf.GlobalDatas) == 0 {
+	if pkg == "" || strings.ContainsAny(pkg, " \t\r\n") {
-		return nil, nil
+		panic(fmt.Errorf("invalid package path for rewrite: %q", pkg))
 	}
-	for _, pkg := range pkgs {
+	if !token.IsIdentifier(varName) {
-		if pkg.ExportFile == "" {
+		panic(fmt.Errorf("invalid variable name for rewrite: %q", varName))
-			continue
-		}
-		if names, ok := ctx.buildConf.GlobalNames[pkg.PkgPath]; ok {
-			err := pkg.LPkg.Undefined(names...)
-			if err != nil {
-				return nil, err
-			}
-			pkg.ExportFile += "-global"
-			pkg.ExportFile, err = exportObject(ctx, pkg.PkgPath+".global", pkg.ExportFile, []byte(pkg.LPkg.String()))
-			if err != nil {
-				return nil, err
-			}
-		}
 	}
-	global := prog.NewPackage("", "global")
+	if len(value) > maxRewriteValueLength {
-	for name, value := range ctx.buildConf.GlobalDatas {
+		panic(fmt.Errorf("rewrite value too large: %d bytes", len(value)))
-		global.AddGlobalString(name, value)
 	}
-	return global, nil
 }
 
 // compileExtraFiles compiles extra files (.s/.c) from target configuration and returns object files
@@ -752,7 +777,7 @@ func compileExtraFiles(ctx *context, verbose bool) ([]string, error) {
 	return objFiles, nil
 }
 
-func linkMainPkg(ctx *context, pkg *packages.Package, pkgs []*aPackage, global llssa.Package, outputPath string, verbose bool) error {
+func linkMainPkg(ctx *context, pkg *packages.Package, pkgs []*aPackage, outputPath string, verbose bool) error {
 
 	needRuntime := false
 	needPyInit := false
@@ -769,7 +794,6 @@ func linkMainPkg(ctx *context, pkg *packages.Package, pkgs []*aPackage, global l
 		if p.ExportFile != "" && aPkg != nil { // skip packages that only contain declarations
 			linkArgs = append(linkArgs, aPkg.LinkArgs...)
 			objFiles = append(objFiles, aPkg.LLFiles...)
-			objFiles = append(objFiles, aPkg.ExportFile)
 			need1, need2 := isNeedRuntimeOrPyInit(ctx, p)
 			if !needRuntime {
 				needRuntime = need1
@@ -780,11 +804,11 @@ func linkMainPkg(ctx *context, pkg *packages.Package, pkgs []*aPackage, global l
 		}
 	})
 	// Generate main module file (needed for global variables even in library modes)
-	entryObjFile, err := genMainModuleFile(ctx, llssa.PkgRuntime, pkg, needRuntime, needPyInit)
+	entryPkg := genMainModule(ctx, llssa.PkgRuntime, pkg, needRuntime, needPyInit)
+	entryObjFile, err := exportObject(ctx, entryPkg.PkgPath, entryPkg.ExportFile, []byte(entryPkg.LPkg.String()))
 	if err != nil {
 		return err
 	}
-	// defer os.Remove(entryLLFile)
 	objFiles = append(objFiles, entryObjFile)
 
 	// Compile extra files from target configuration
@@ -794,14 +818,6 @@ func linkMainPkg(ctx *context, pkg *packages.Package, pkgs []*aPackage, global l
 	}
 	objFiles = append(objFiles, extraObjFiles...)
 
-	if global != nil {
-		export, err := exportObject(ctx, pkg.PkgPath+".global", pkg.ExportFile+"-global", []byte(global.String()))
-		if err != nil {
-			return err
-		}
-		objFiles = append(objFiles, export)
-	}
-
 	if IsFullRpathEnabled() {
 		// Treat every link-time library search path, specified by the -L parameter, as a runtime search path as well.
 		// This is to ensure the final executable can locate libraries with a relocatable install_name
@@ -914,118 +930,6 @@ func needStart(ctx *context) bool {
 	}
 }
 
-func genMainModuleFile(ctx *context, rtPkgPath string, pkg *packages.Package, needRuntime, needPyInit bool) (path string, err error) {
-	var (
-		pyInitDecl string
-		pyInit     string
-		rtInitDecl string
-		rtInit     string
-	)
-	mainPkgPath := pkg.PkgPath
-	if needRuntime {
-		rtInit = "call void @\"" + rtPkgPath + ".init\"()"
-		rtInitDecl = "declare void @\"" + rtPkgPath + ".init\"()"
-	}
-	if needPyInit {
-		pyInit = "call void @Py_Initialize()"
-		pyInitDecl = "declare void @Py_Initialize()"
-	}
-	declSizeT := "%size_t = type i64"
-	if is32Bits(ctx.buildConf.Goarch) {
-		declSizeT = "%size_t = type i32"
-	}
-	stdioDecl := ""
-	stdioNobuf := ""
-	if IsStdioNobuf() {
-		stdioDecl = `
-@stdout = external global ptr
-@stderr = external global ptr
-@__stdout = external global ptr
-@__stderr = external global ptr
-declare i32 @setvbuf(ptr, ptr, i32, %size_t)
-	`
-		stdioNobuf = `
-; Set stdout with no buffer
-%stdout_is_null = icmp eq ptr @stdout, null
-%stdout_ptr = select i1 %stdout_is_null, ptr @__stdout, ptr @stdout
-call i32 @setvbuf(ptr %stdout_ptr, ptr null, i32 2, %size_t 0)
-; Set stderr with no buffer
-%stderr_ptr = select i1 %stdout_is_null, ptr @__stderr, ptr @stderr
-call i32 @setvbuf(ptr %stderr_ptr, ptr null, i32 2, %size_t 0)
-	`
-	}
-	// TODO(lijie): workaround for libc-free
-	// Remove main/_start when -buildmode and libc are ready
-	startDefine := `
-define weak void @_start() {
-  ; argc = 0
-  %argc = add i32 0, 0
-  ; argv = null
-  %argv = inttoptr i64 0 to i8**
-  call i32 @main(i32 %argc, i8** %argv)
-  ret void
-}
-`
-	mainDefine := "define i32 @main(i32 noundef %0, ptr nocapture noundef readnone %1) local_unnamed_addr"
-	if !needStart(ctx) && isWasmTarget(ctx.buildConf.Goos) {
-		mainDefine = "define hidden noundef i32 @__main_argc_argv(i32 noundef %0, ptr nocapture noundef readnone %1) local_unnamed_addr"
-	}
-	if !needStart(ctx) {
-		startDefine = ""
-	}
-
-	var mainCode string
-	// For library modes (c-archive, c-shared), only generate global variables
-	if ctx.buildConf.BuildMode != BuildModeExe {
-		mainCode = `; ModuleID = 'main'
-source_filename = "main"
-@__llgo_argc = global i32 0, align 4
-@__llgo_argv = global ptr null, align 8
-`
-	} else {
-		// For executable mode, generate full main function
-		mainCode = fmt.Sprintf(`; ModuleID = 'main'
-source_filename = "main"
-%s
-@__llgo_argc = global i32 0, align 4
-@__llgo_argv = global ptr null, align 8
-%s
-%s
-%s
-declare void @"%s.init"()
-declare void @"%s.main"()
-define weak void @runtime.init() {
-  ret void
-}
-
-; TODO(lijie): workaround for syscall patch
-define weak void @"syscall.init"() {
-  ret void
-}
-
-%s
-
-%s {
-_llgo_0:
-  store i32 %%0, ptr @__llgo_argc, align 4
-  store ptr %%1, ptr @__llgo_argv, align 8
-  %s
-  %s
-  %s
-  call void @runtime.init()
-  call void @"%s.init"()
-  call void @"%s.main"()
-  ret i32 0
-}
-`, declSizeT, stdioDecl,
-			pyInitDecl, rtInitDecl, mainPkgPath, mainPkgPath,
-			startDefine, mainDefine, stdioNobuf,
-			pyInit, rtInit, mainPkgPath, mainPkgPath)
-	}
-
-	return exportObject(ctx, pkg.PkgPath+".main", pkg.ExportFile+"-main", []byte(mainCode))
-}
-
 func is32Bits(goarch string) bool {
 	return goarch == "386" || goarch == "arm" || goarch == "mips" || goarch == "wasm"
 }
@@ -1050,7 +954,7 @@ func buildPkg(ctx *context, aPkg *aPackage, verbose bool) error {
 		cl.SetDebug(cl.DbgFlagAll)
 	}
 
-	ret, externs, err := cl.NewPackageEx(ctx.prog, ctx.patches, aPkg.SSA, syntax)
+	ret, externs, err := cl.NewPackageEx(ctx.prog, ctx.patches, aPkg.rewriteVars, aPkg.SSA, syntax)
 	if showDetail {
 		llssa.SetDebug(0)
 		cl.SetDebug(0)
@@ -1077,10 +981,11 @@ func buildPkg(ctx *context, aPkg *aPackage, verbose bool) error {
 		aPkg.LinkArgs = append(aPkg.LinkArgs, altLdflags...)
 	}
 	if pkg.ExportFile != "" {
-		pkg.ExportFile, err = exportObject(ctx, pkg.PkgPath, pkg.ExportFile, []byte(ret.String()))
+		exportFile, err := exportObject(ctx, pkg.PkgPath, pkg.ExportFile, []byte(ret.String()))
 		if err != nil {
 			return fmt.Errorf("export object of %v failed: %v", pkgPath, err)
 		}
+		aPkg.LLFiles = append(aPkg.LLFiles, exportFile)
 		if debugBuild || verbose {
 			fmt.Fprintf(os.Stderr, "==> Export %s: %s\n", aPkg.PkgPath, pkg.ExportFile)
 		}
@@ -1089,11 +994,15 @@ func buildPkg(ctx *context, aPkg *aPackage, verbose bool) error {
 }
 
 func exportObject(ctx *context, pkgPath string, exportFile string, data []byte) (string, error) {
-	f, err := os.CreateTemp("", "llgo-*.ll")
+	base := filepath.Base(exportFile)
+	f, err := os.CreateTemp("", base+"-*.ll")
 	if err != nil {
 		return "", err
 	}
-	f.Write(data)
+	if _, err := f.Write(data); err != nil {
+		f.Close()
+		return "", err
+	}
 	err = f.Close()
 	if err != nil {
 		return exportFile, err
@@ -1111,13 +1020,17 @@ func exportObject(ctx *context, pkgPath string, exportFile string, data []byte)
 		}
 		return exportFile, os.Rename(f.Name(), exportFile)
 	}
-	exportFile += ".o"
+	objFile, err := os.CreateTemp("", base+"-*.o")
-	args := []string{"-o", exportFile, "-c", f.Name(), "-Wno-override-module"}
+	if err != nil {
+		return "", err
+	}
+	objFile.Close()
+	args := []string{"-o", objFile.Name(), "-c", f.Name(), "-Wno-override-module"}
 	if ctx.buildConf.Verbose {
 		fmt.Fprintln(os.Stderr, "clang", args)
 	}
 	cmd := ctx.compiler()
-	return exportFile, cmd.Compile(args...)
+	return objFile.Name(), cmd.Compile(args...)
 }
 
 func llcCheck(env *llvm.Env, exportFile string) (msg string, err error) {
@@ -1171,8 +1084,9 @@ type aPackage struct {
 	AltPkg *packages.Cached
 	LPkg   llssa.Package
 
-	LinkArgs []string
+	LinkArgs    []string
-	LLFiles  []string
+	LLFiles     []string
+	rewriteVars map[string]string
 }
 
 type Package = *aPackage
@@ -1193,7 +1107,8 @@ func allPkgs(ctx *context, initial []*packages.Package, verbose bool) (all []*aP
 					return
 				}
 			}
-			all = append(all, &aPackage{p, ssaPkg, altPkg, nil, nil, nil})
+			rewrites := collectRewriteVars(ctx, pkgPath)
+			all = append(all, &aPackage{p, ssaPkg, altPkg, nil, nil, nil, rewrites})
 		} else {
 			errs = append(errs, p)
 		}
@@ -1201,6 +1116,32 @@ func allPkgs(ctx *context, initial []*packages.Package, verbose bool) (all []*aP
 	return
 }
 
+func collectRewriteVars(ctx *context, pkgPath string) map[string]string {
+	data := ctx.buildConf.GlobalRewrites
+	if len(data) == 0 {
+		return nil
+	}
+	basePath := strings.TrimPrefix(pkgPath, altPkgPathPrefix)
+	if vars := data[basePath]; vars != nil {
+		return cloneRewrites(vars)
+	}
+	if vars := data[pkgPath]; vars != nil {
+		return cloneRewrites(vars)
+	}
+	return nil
+}
+
+func cloneRewrites(src Rewrites) map[string]string {
+	if len(src) == 0 {
+		return nil
+	}
+	dup := make(map[string]string, len(src))
+	for k, v := range src {
+		dup[k] = v
+	}
+	return dup
+}
+
 func createSSAPkg(prog *ssa.Program, p *packages.Package, verbose bool) *ssa.Package {
 	pkgSSA := prog.ImportedPackage(p.ID)
 	if pkgSSA == nil {

internal/build/build_test.go

@@ -8,6 +8,9 @@ import (
 	"fmt"
 	"io"
 	"os"
+	"os/exec"
+	"path/filepath"
+	"runtime"
 	"testing"
 
 	"github.com/goplus/llgo/internal/mockable"
@@ -94,4 +97,70 @@ func TestCmpTest(t *testing.T) {
 	mockRun([]string{"../../cl/_testgo/runtest"}, &Config{Mode: ModeCmpTest})
 }
 
-// TestGenerateOutputFilenames removed - functionality moved to filename_test.go
+const (
+	rewriteMainPkg = "github.com/goplus/llgo/cl/_testgo/rewrite"
+	rewriteDepPkg  = rewriteMainPkg + "/dep"
+	rewriteDirPath = "../../cl/_testgo/rewrite"
+)
+
+func TestLdFlagsRewriteVars(t *testing.T) {
+	buildRewriteBinary(t, false, "build-main", "build-pkg")
+	buildRewriteBinary(t, false, "rerun-main", "rerun-pkg")
+}
+
+func TestLdFlagsRewriteVarsMainAlias(t *testing.T) {
+	buildRewriteBinary(t, true, "alias-main", "alias-pkg")
+}
+
+func buildRewriteBinary(t *testing.T, useMainAlias bool, mainVal, depVal string) {
+	t.Helper()
+	binPath := filepath.Join(t.TempDir(), "rewrite")
+	if runtime.GOOS == "windows" {
+		binPath += ".exe"
+	}
+
+	cfg := &Config{Mode: ModeBuild, OutFile: binPath}
+	mainKey := rewriteMainPkg
+	var mainPkgs []string
+	if useMainAlias {
+		mainKey = "main"
+		mainPkgs = []string{rewriteMainPkg}
+	}
+	mainPlain := mainVal + "-plain"
+	depPlain := depVal + "-plain"
+	gorootVal := "goroot-" + mainVal
+	versionVal := "version-" + mainVal
+	addGlobalString(cfg, mainKey+".VarName="+mainVal, mainPkgs)
+	addGlobalString(cfg, mainKey+".VarPlain="+mainPlain, mainPkgs)
+	addGlobalString(cfg, rewriteDepPkg+".VarName="+depVal, nil)
+	addGlobalString(cfg, rewriteDepPkg+".VarPlain="+depPlain, nil)
+	addGlobalString(cfg, "runtime.defaultGOROOT="+gorootVal, nil)
+	addGlobalString(cfg, "runtime.buildVersion="+versionVal, nil)
+
+	if _, err := Do([]string{rewriteDirPath}, cfg); err != nil {
+		t.Fatalf("ModeBuild failed: %v", err)
+	}
+	got := runBinary(t, binPath)
+	want := fmt.Sprintf(
+		"main.VarName: %s\nmain.VarPlain: %s\ndep.VarName: %s\ndep.VarPlain: %s\nruntime.GOROOT(): %s\nruntime.Version(): %s\n",
+		mainVal, mainPlain, depVal, depPlain, gorootVal, versionVal,
+	)
+	if got != want {
+		t.Fatalf("unexpected binary output:\nwant %q\ngot  %q", want, got)
+	}
+}
+
+func runBinary(t *testing.T, path string, args ...string) string {
+	t.Helper()
+	cmd := exec.Command(path, args...)
+	output, err := cmd.CombinedOutput()
+	if err != nil {
+		t.Fatalf("failed to run %s: %v\n%s", path, err, output)
+	}
+	return string(output)
+}
+
+func TestRunPrintfWithStdioNobuf(t *testing.T) {
+	t.Setenv(llgoStdioNobuf, "1")
+	mockRun([]string{"../../cl/_testdata/printf"}, &Config{Mode: ModeRun})
+}

internal/build/main_module.go

@@ -0,0 +1,231 @@
+//go:build !llgo
+// +build !llgo
+
+/*
+ * Copyright (c) 2024 The GoPlus Authors (goplus.org). All rights reserved.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+// Package build contains the llgo compiler build orchestration logic.
+//
+// The main_module.go file generates the entry point module for llgo programs,
+// which contains the main() function, initialization sequence, and global
+// variables like argc/argv. This module is generated differently depending on
+// BuildMode (exe, c-archive, c-shared).
+
+package build
+
+import (
+	"go/token"
+	"go/types"
+
+	"github.com/goplus/llgo/internal/packages"
+	llvm "github.com/goplus/llvm"
+
+	llssa "github.com/goplus/llgo/ssa"
+)
+
+// genMainModule generates the main entry module for an llgo program.
+//
+// The module contains argc/argv globals and, for executable build modes,
+// the entry function that wires initialization and main. For C archive or
+// shared library modes, only the globals are emitted.
+func genMainModule(ctx *context, rtPkgPath string, pkg *packages.Package, needRuntime, needPyInit bool) Package {
+	prog := ctx.prog
+	mainPkg := prog.NewPackage("", pkg.ID+".main")
+
+	argcVar := mainPkg.NewVarEx("__llgo_argc", prog.Pointer(prog.Int32()))
+	argcVar.Init(prog.Zero(prog.Int32()))
+
+	argvValueType := prog.Pointer(prog.CStr())
+	argvVar := mainPkg.NewVarEx("__llgo_argv", prog.Pointer(argvValueType))
+	argvVar.InitNil()
+
+	exportFile := pkg.ExportFile
+	if exportFile == "" {
+		exportFile = pkg.PkgPath
+	}
+	mainAPkg := &aPackage{
+		Package: &packages.Package{
+			PkgPath:    pkg.PkgPath + ".main",
+			ExportFile: exportFile + "-main",
+		},
+		LPkg: mainPkg,
+	}
+
+	if ctx.buildConf.BuildMode != BuildModeExe {
+		return mainAPkg
+	}
+
+	runtimeStub := defineWeakNoArgStub(mainPkg, "runtime.init")
+	// TODO(lijie): workaround for syscall patch
+	defineWeakNoArgStub(mainPkg, "syscall.init")
+
+	var pyInit llssa.Function
+	if needPyInit {
+		pyInit = declareNoArgFunc(mainPkg, "Py_Initialize")
+	}
+
+	var rtInit llssa.Function
+	if needRuntime {
+		rtInit = declareNoArgFunc(mainPkg, rtPkgPath+".init")
+	}
+
+	mainInit := declareNoArgFunc(mainPkg, pkg.PkgPath+".init")
+	mainMain := declareNoArgFunc(mainPkg, pkg.PkgPath+".main")
+
+	entryFn := defineEntryFunction(ctx, mainPkg, argcVar, argvVar, argvValueType, runtimeStub, mainInit, mainMain, pyInit, rtInit)
+
+	if needStart(ctx) {
+		defineStart(mainPkg, entryFn, argvValueType)
+	}
+
+	return mainAPkg
+}
+
+// defineEntryFunction creates the program's entry function. The name is
+// "main" for standard targets, or "__main_argc_argv" with hidden visibility
+// for WASM targets that don't require _start.
+//
+// The entry stores argc/argv, optionally disables stdio buffering, runs
+// initialization hooks (Python, runtime, package init), and finally calls
+// main.main before returning 0.
+func defineEntryFunction(ctx *context, pkg llssa.Package, argcVar, argvVar llssa.Global, argvType llssa.Type, runtimeStub, mainInit, mainMain llssa.Function, pyInit, rtInit llssa.Function) llssa.Function {
+	prog := pkg.Prog
+	entryName := "main"
+	if !needStart(ctx) && isWasmTarget(ctx.buildConf.Goos) {
+		entryName = "__main_argc_argv"
+	}
+	sig := newEntrySignature(argvType.RawType())
+	fn := pkg.NewFunc(entryName, sig, llssa.InC)
+	fnVal := pkg.Module().NamedFunction(entryName)
+	if entryName != "main" {
+		fnVal.SetVisibility(llvm.HiddenVisibility)
+		fnVal.SetUnnamedAddr(true)
+	}
+	b := fn.MakeBody(1)
+	b.Store(argcVar.Expr, fn.Param(0))
+	b.Store(argvVar.Expr, fn.Param(1))
+	if IsStdioNobuf() {
+		emitStdioNobuf(b, pkg, ctx.buildConf.Goos)
+	}
+	if pyInit != nil {
+		b.Call(pyInit.Expr)
+	}
+	if rtInit != nil {
+		b.Call(rtInit.Expr)
+	}
+	b.Call(runtimeStub.Expr)
+	b.Call(mainInit.Expr)
+	b.Call(mainMain.Expr)
+	b.Return(prog.IntVal(0, prog.Int32()))
+	return fn
+}
+
+func defineStart(pkg llssa.Package, entry llssa.Function, argvType llssa.Type) {
+	fn := pkg.NewFunc("_start", llssa.NoArgsNoRet, llssa.InC)
+	pkg.Module().NamedFunction("_start").SetLinkage(llvm.WeakAnyLinkage)
+	b := fn.MakeBody(1)
+	prog := pkg.Prog
+	b.Call(entry.Expr, prog.IntVal(0, prog.Int32()), prog.Nil(argvType))
+	b.Return()
+}
+
+func declareNoArgFunc(pkg llssa.Package, name string) llssa.Function {
+	return pkg.NewFunc(name, llssa.NoArgsNoRet, llssa.InC)
+}
+
+func defineWeakNoArgStub(pkg llssa.Package, name string) llssa.Function {
+	fn := pkg.NewFunc(name, llssa.NoArgsNoRet, llssa.InC)
+	pkg.Module().NamedFunction(name).SetLinkage(llvm.WeakAnyLinkage)
+	b := fn.MakeBody(1)
+	b.Return()
+	return fn
+}
+
+const (
+	// ioNoBuf represents the _IONBF flag for setvbuf (no buffering)
+	ioNoBuf = 2
+)
+
+// emitStdioNobuf generates code to disable buffering on stdout and stderr
+// when the LLGO_STDIO_NOBUF environment variable is set. Only Darwin uses
+// the alternate `__stdoutp`/`__stderrp` symbols; other targets rely on the
+// standard `stdout`/`stderr` globals.
+func emitStdioNobuf(b llssa.Builder, pkg llssa.Package, goos string) {
+	prog := pkg.Prog
+	streamType := prog.VoidPtr()
+	streamPtrType := prog.Pointer(streamType)
+
+	stdoutName := "stdout"
+	stderrName := "stderr"
+	if goos == "darwin" {
+		stdoutName = "__stdoutp"
+		stderrName = "__stderrp"
+	}
+	stdout := declareExternalPtrGlobal(pkg, stdoutName, streamPtrType)
+	stderr := declareExternalPtrGlobal(pkg, stderrName, streamPtrType)
+	stdoutPtr := b.Load(stdout)
+	stderrPtr := b.Load(stderr)
+	sizeType := prog.Uintptr()
+	setvbuf := declareSetvbuf(pkg, streamPtrType, prog.CStr(), prog.Int32(), sizeType)
+
+	noBufMode := prog.IntVal(ioNoBuf, prog.Int32())
+	zeroSize := prog.Zero(sizeType)
+	nullBuf := prog.Nil(prog.CStr())
+
+	b.Call(setvbuf.Expr, stdoutPtr, nullBuf, noBufMode, zeroSize)
+	b.Call(setvbuf.Expr, stderrPtr, nullBuf, noBufMode, zeroSize)
+}
+
+func declareExternalPtrGlobal(pkg llssa.Package, name string, valueType llssa.Type) llssa.Expr {
+	global := pkg.NewVarEx(name, valueType)
+	pkg.Module().NamedGlobal(name).SetLinkage(llvm.ExternalLinkage)
+	return global.Expr
+}
+
+func declareSetvbuf(pkg llssa.Package, streamPtrType, bufPtrType, intType, sizeType llssa.Type) llssa.Function {
+	sig := newSignature(
+		[]types.Type{
+			streamPtrType.RawType(),
+			bufPtrType.RawType(),
+			intType.RawType(),
+			sizeType.RawType(),
+		},
+		[]types.Type{intType.RawType()},
+	)
+	return pkg.NewFunc("setvbuf", sig, llssa.InC)
+}
+
+func tupleOf(tys ...types.Type) *types.Tuple {
+	if len(tys) == 0 {
+		return types.NewTuple()
+	}
+	vars := make([]*types.Var, len(tys))
+	for i, t := range tys {
+		vars[i] = types.NewParam(token.NoPos, nil, "", t)
+	}
+	return types.NewTuple(vars...)
+}
+
+func newSignature(params []types.Type, results []types.Type) *types.Signature {
+	return types.NewSignatureType(nil, nil, nil, tupleOf(params...), tupleOf(results...), false)
+}
+
+func newEntrySignature(argvType types.Type) *types.Signature {
+	return newSignature(
+		[]types.Type{types.Typ[types.Int32], argvType},
+		[]types.Type{types.Typ[types.Int32]},
+	)
+}

internal/build/main_module_test.go

@@ -0,0 +1,70 @@
+//go:build !llgo
+// +build !llgo
+
+package build
+
+import (
+	"strings"
+	"testing"
+
+	"github.com/goplus/llvm"
+
+	"github.com/goplus/llgo/internal/packages"
+	llssa "github.com/goplus/llgo/ssa"
+)
+
+func init() {
+	llssa.Initialize(llssa.InitAll)
+}
+
+func TestGenMainModuleExecutable(t *testing.T) {
+	llvm.InitializeAllTargets()
+	t.Setenv(llgoStdioNobuf, "")
+	ctx := &context{
+		prog: llssa.NewProgram(nil),
+		buildConf: &Config{
+			BuildMode: BuildModeExe,
+			Goos:      "linux",
+			Goarch:    "amd64",
+		},
+	}
+	pkg := &packages.Package{PkgPath: "example.com/foo", ExportFile: "foo.a"}
+	mod := genMainModule(ctx, llssa.PkgRuntime, pkg, true, true)
+	if mod.ExportFile != "foo.a-main" {
+		t.Fatalf("unexpected export file: %s", mod.ExportFile)
+	}
+	ir := mod.LPkg.String()
+	checks := []string{
+		"define i32 @main(",
+		"call void @Py_Initialize()",
+		"call void @\"example.com/foo.init\"()",
+		"define weak void @_start()",
+	}
+	for _, want := range checks {
+		if !strings.Contains(ir, want) {
+			t.Fatalf("main module IR missing %q:\n%s", want, ir)
+		}
+	}
+}
+
+func TestGenMainModuleLibrary(t *testing.T) {
+	llvm.InitializeAllTargets()
+	t.Setenv(llgoStdioNobuf, "")
+	ctx := &context{
+		prog: llssa.NewProgram(nil),
+		buildConf: &Config{
+			BuildMode: BuildModeCArchive,
+			Goos:      "linux",
+			Goarch:    "amd64",
+		},
+	}
+	pkg := &packages.Package{PkgPath: "example.com/foo", ExportFile: "foo.a"}
+	mod := genMainModule(ctx, llssa.PkgRuntime, pkg, false, false)
+	ir := mod.LPkg.String()
+	if strings.Contains(ir, "define i32 @main") {
+		t.Fatalf("library mode should not emit main function:\n%s", ir)
+	}
+	if !strings.Contains(ir, "@__llgo_argc = global i32 0") {
+		t.Fatalf("library mode missing argc global:\n%s", ir)
+	}
+}

internal/build/resolver.go

@@ -0,0 +1,109 @@
+package build
+
+import "strings"
+
+// nameResolver maps symbol names to aggregation buckets based on the requested level.
+type nameResolver struct {
+	level      string
+	pkgs       []Package
+	moduleMap  map[string]string
+	packageMap map[string]string
+}
+
+func newNameResolver(level string, pkgs []Package) *nameResolver {
+	lvl := strings.ToLower(strings.TrimSpace(level))
+	if lvl == "" {
+		lvl = "module"
+	}
+	return &nameResolver{
+		level:      lvl,
+		pkgs:       pkgs,
+		moduleMap:  make(map[string]string),
+		packageMap: make(map[string]string),
+	}
+}
+
+func (r *nameResolver) resolve(sym string) string {
+	base := moduleNameFromSymbol(sym)
+	symbol := trimSymbolForMatch(sym)
+	switch r.level {
+	case "full":
+		return base
+	case "package":
+		if pkg := r.matchPackage(symbol); pkg != "" {
+			return pkg
+		}
+	case "module":
+		if mod := r.matchModule(symbol); mod != "" {
+			return mod
+		}
+	}
+	if strings.Contains(symbol, "llgo") {
+		return "llgo-stubs"
+	}
+	return base
+}
+
+func (r *nameResolver) matchPackage(symbol string) string {
+	if symbol == "" {
+		return ""
+	}
+	if cached := r.packageMap[symbol]; cached != "" {
+		return cached
+	}
+	for _, pkg := range r.pkgs {
+		if pkg == nil || pkg.Package == nil {
+			continue
+		}
+		id := pkg.PkgPath
+		if id == "" {
+			continue
+		}
+		if strings.HasPrefix(symbol, id+".") {
+			r.packageMap[symbol] = id
+			return id
+		}
+	}
+	return ""
+}
+
+func (r *nameResolver) matchModule(symbol string) string {
+	if symbol == "" {
+		return ""
+	}
+	if cached := r.moduleMap[symbol]; cached != "" {
+		return cached
+	}
+	for _, pkg := range r.pkgs {
+		if pkg == nil || pkg.Package == nil {
+			continue
+		}
+		path := pkg.PkgPath
+		if path == "" {
+			continue
+		}
+		if strings.HasPrefix(symbol, path+".") {
+			mod := path
+			if pkg.Module != nil && pkg.Module.Path != "" {
+				mod = pkg.Module.Path
+			}
+			r.moduleMap[symbol] = mod
+			return mod
+		}
+	}
+	return ""
+}
+
+func trimSymbolForMatch(sym string) string {
+	name := strings.TrimSpace(sym)
+	for len(name) > 0 && (name[0] == '_' || name[0] == '.') {
+		name = name[1:]
+	}
+	if idx := strings.Index(name, " "); idx >= 0 {
+		name = name[:idx]
+	}
+	if idx := strings.Index(name, "@"); idx >= 0 {
+		name = name[:idx]
+	}
+	return name
+}

internal/build/size_report.go

@@ -0,0 +1,627 @@
+package build
+
+import (
+	"bufio"
+	"bytes"
+	"encoding/json"
+	"fmt"
+	"io"
+	"math"
+	"os"
+	"path/filepath"
+	"sort"
+	"strconv"
+	"strings"
+
+	"github.com/goplus/llgo/xtool/env/llvm"
+)
+
+type sectionKind int
+
+const (
+	sectionUnknown sectionKind = iota
+	sectionText
+	sectionROData
+	sectionData
+	sectionBSS
+)
+
+const (
+	// readelfInitialBuffer is the initial buffer size for reading readelf output.
+	// Most lines in readelf output are less than 1KB.
+	readelfInitialBuffer = 64 * 1024
+
+	// readelfMaxBuffer is the maximum buffer size to handle very long symbol names
+	// or section dumps. Reduced from 64MB to prevent excessive memory consumption
+	// while still accommodating reasonably large binaries.
+	readelfMaxBuffer = 4 * 1024 * 1024
+)
+
+// ELF special section indices (from ELF specification)
+const (
+	SHN_UNDEF     = 0x0000 // Undefined section
+	SHN_LORESERVE = 0xFF00 // Start of reserved indices
+	SHN_ABS       = 0xFFF1 // Absolute values
+	SHN_COMMON    = 0xFFF2 // Common symbols
+	SHN_XINDEX    = 0xFFFF // Escape value for extended section indices
+)
+
+type sectionInfo struct {
+	Index   int
+	Name    string
+	Segment string
+	Address uint64
+	Size    uint64
+	Kind    sectionKind
+}
+
+type symbolInfo struct {
+	Name         string
+	SectionIndex int
+	Address      uint64
+}
+
+type readelfData struct {
+	sections map[int]*sectionInfo
+	symbols  map[int][]symbolInfo
+}
+
+type moduleSize struct {
+	Name   string
+	Code   uint64
+	ROData uint64
+	Data   uint64
+	BSS    uint64
+}
+
+func (m *moduleSize) Flash() uint64 {
+	return m.Code + m.ROData + m.Data
+}
+
+func (m *moduleSize) RAM() uint64 {
+	return m.Data + m.BSS
+}
+
+type sizeReport struct {
+	Binary  string
+	Modules map[string]*moduleSize
+	Total   moduleSize
+}
+
+func (r *sizeReport) module(name string) *moduleSize {
+	if name == "" {
+		name = "(anonymous)"
+	}
+	if r.Modules == nil {
+		r.Modules = make(map[string]*moduleSize)
+	}
+	m, ok := r.Modules[name]
+	if !ok {
+		m = &moduleSize{Name: name}
+		r.Modules[name] = m
+	}
+	return m
+}
+
+func (r *sizeReport) add(name string, kind sectionKind, size uint64) {
+	if size == 0 {
+		return
+	}
+	m := r.module(name)
+	switch kind {
+	case sectionText:
+		m.Code += size
+		r.Total.Code += size
+	case sectionROData:
+		m.ROData += size
+		r.Total.ROData += size
+	case sectionData:
+		m.Data += size
+		r.Total.Data += size
+	case sectionBSS:
+		m.BSS += size
+		r.Total.BSS += size
+	}
+}
+
+func reportBinarySize(path, format, level string, pkgs []Package) error {
+	report, err := collectBinarySize(path, pkgs, level)
+	if err != nil {
+		return err
+	}
+	switch format {
+	case "", "text":
+		printTextReport(os.Stdout, report)
+	case "json":
+		return emitJSONReport(os.Stdout, report)
+	default:
+		return fmt.Errorf("unknown size format %q (valid: text,json)", format)
+	}
+	return nil
+}
+
+func collectBinarySize(path string, pkgs []Package, level string) (*sizeReport, error) {
+	cmd, err := llvm.New("").Readelf("--elf-output-style=LLVM", "--all", path)
+	if err != nil {
+		return nil, fmt.Errorf("llvm-readelf: %w", err)
+	}
+	var stderr bytes.Buffer
+	cmd.Stderr = &stderr
+	stdout, err := cmd.StdoutPipe()
+	if err != nil {
+		return nil, fmt.Errorf("llvm-readelf stdout: %w", err)
+	}
+	if err := cmd.Start(); err != nil {
+		return nil, fmt.Errorf("failed to execute llvm-readelf: %w", err)
+	}
+	parsed, parseErr := parseReadelfOutput(stdout)
+	closeErr := stdout.Close()
+	waitErr := cmd.Wait()
+	if waitErr != nil {
+		return nil, fmt.Errorf("llvm-readelf failed: %w\n%s", waitErr, stderr.String())
+	}
+	if parseErr != nil {
+		return nil, fmt.Errorf("parsing llvm-readelf output failed: %w", parseErr)
+	}
+	if closeErr != nil {
+		return nil, fmt.Errorf("closing llvm-readelf stdout pipe failed: %w", closeErr)
+	}
+	report := buildSizeReport(path, parsed, pkgs, level)
+	if report == nil || len(report.Modules) == 0 {
+		return nil, fmt.Errorf("size report: no allocatable sections found in %s", path)
+	}
+	return report, nil
+}
+
+func parseReadelfOutput(r io.Reader) (*readelfData, error) {
+	scanner := bufio.NewScanner(r)
+	scanner.Buffer(make([]byte, 0, readelfInitialBuffer), readelfMaxBuffer)
+
+	type ctxKind int
+	const (
+		ctxRoot ctxKind = iota
+		ctxSections
+		ctxSection
+		ctxSymbols
+		ctxSymbol
+	)
+
+	type ctx struct {
+		kind   ctxKind
+		indent int
+	}
+
+	stack := []ctx{{kind: ctxRoot, indent: -1}}
+	push := func(kind ctxKind, indent int) {
+		stack = append(stack, ctx{kind: kind, indent: indent})
+	}
+	pop := func(expected ctxKind, indent int) bool {
+		top := stack[len(stack)-1]
+		if top.kind != expected || top.indent != indent {
+			return false
+		}
+		stack = stack[:len(stack)-1]
+		return true
+	}
+	current := func() ctx {
+		return stack[len(stack)-1]
+	}
+
+	data := &readelfData{
+		sections: make(map[int]*sectionInfo),
+		symbols:  make(map[int][]symbolInfo),
+	}
+
+	// readelf outputs section references differently:
+	//   - Mach-O: section numbers are 1-based in symbol references
+	//   - ELF: section numbers in symbol references match the Index directly
+	secIndexBase := 1 // default to Mach-O behavior; switch to 0 for ELF once detected
+
+	var currentSection *sectionInfo
+	var currentSymbol *symbolInfo
+
+	for scanner.Scan() {
+		raw := scanner.Text()
+		trimmed := strings.TrimSpace(raw)
+		if trimmed == "" {
+			continue
+		}
+
+		// Detect object format early to adjust section index base
+		if strings.HasPrefix(trimmed, "Format:") {
+			lower := strings.ToLower(trimmed)
+			if strings.Contains(lower, "mach-o") {
+				secIndexBase = 1
+			} else if strings.Contains(lower, "elf") {
+				secIndexBase = 0
+			}
+		}
+		indent := countLeadingSpaces(raw)
+		top := current()
+
+		switch {
+		case strings.HasPrefix(trimmed, "Sections [") && top.kind == ctxRoot:
+			push(ctxSections, indent)
+			continue
+		case strings.HasPrefix(trimmed, "Symbols [") && top.kind == ctxRoot:
+			push(ctxSymbols, indent)
+			continue
+		case trimmed == "Section {" && top.kind == ctxSections && indent == top.indent+2:
+			currentSection = &sectionInfo{Index: -1}
+			push(ctxSection, indent)
+			continue
+		case trimmed == "Symbol {" && top.kind == ctxSymbols && indent == top.indent+2:
+			currentSymbol = &symbolInfo{SectionIndex: -1}
+			push(ctxSymbol, indent)
+			continue
+		case trimmed == "}" && pop(ctxSection, indent):
+			if currentSection != nil && currentSection.Index >= 0 {
+				currentSection.Kind = classifySection(currentSection.Name, currentSection.Segment)
+				data.sections[currentSection.Index] = currentSection
+			}
+			currentSection = nil
+			continue
+		case trimmed == "}" && pop(ctxSymbol, indent):
+			if currentSymbol != nil && currentSymbol.SectionIndex >= 0 {
+				data.symbols[currentSymbol.SectionIndex] = append(data.symbols[currentSymbol.SectionIndex], *currentSymbol)
+			}
+			currentSymbol = nil
+			continue
+		case trimmed == "]" && (top.kind == ctxSections || top.kind == ctxSymbols) && indent == top.indent:
+			stack = stack[:len(stack)-1]
+			continue
+		}
+
+		switch top.kind {
+		case ctxSection:
+			if currentSection == nil {
+				continue
+			}
+			switch {
+			case strings.HasPrefix(trimmed, "Index: "):
+				if idx, err := strconv.Atoi(strings.TrimSpace(trimmed[len("Index: "):])); err == nil {
+					currentSection.Index = idx
+				}
+			case strings.HasPrefix(trimmed, "Name: "):
+				currentSection.Name = parseNameField(trimmed[len("Name: "):])
+			case strings.HasPrefix(trimmed, "Segment: "):
+				currentSection.Segment = parseNameField(trimmed[len("Segment: "):])
+			case strings.HasPrefix(trimmed, "Address: "):
+				if val, err := parseUintField(trimmed[len("Address: "):]); err == nil {
+					currentSection.Address = val
+				}
+			case strings.HasPrefix(trimmed, "Size: "):
+				if val, err := parseUintField(trimmed[len("Size: "):]); err == nil {
+					currentSection.Size = val
+				}
+			}
+		case ctxSymbol:
+			if currentSymbol == nil {
+				continue
+			}
+			switch {
+			case strings.HasPrefix(trimmed, "Name: "):
+				currentSymbol.Name = parseNameField(trimmed[len("Name: "):])
+			case strings.HasPrefix(trimmed, "Section: "):
+				name, idx := parseSectionRef(trimmed[len("Section: "):], secIndexBase)
+				currentSymbol.SectionIndex = idx
+				if currentSymbol.Name == "" {
+					currentSymbol.Name = name
+				}
+			case strings.HasPrefix(trimmed, "Value: "):
+				if val, err := parseUintField(trimmed[len("Value: "):]); err == nil {
+					currentSymbol.Address = val
+				}
+			}
+		}
+	}
+
+	if err := scanner.Err(); err != nil {
+		return nil, err
+	}
+	return data, nil
+}
+
+func countLeadingSpaces(line string) int {
+	count := 0
+	for _, ch := range line {
+		if ch != ' ' {
+			break
+		}
+		count++
+	}
+	return count
+}
+
+func classifySection(name, segment string) sectionKind {
+	ln := strings.ToLower(name)
+	ls := strings.ToLower(segment)
+	switch {
+	case strings.Contains(ln, "text"), strings.Contains(ln, "code"), strings.Contains(ln, "plt"):
+		return sectionText
+	case strings.Contains(ln, "rodata"), strings.Contains(ln, "const"), strings.Contains(ln, "literal"), strings.Contains(ln, "cstring"):
+		return sectionROData
+	case strings.Contains(ln, "bss"), strings.Contains(ln, "tbss"), strings.Contains(ln, "sbss"), strings.Contains(ln, "common"), strings.Contains(ln, "zerofill"):
+		return sectionBSS
+	case strings.Contains(ln, "data"), strings.Contains(ln, "got"), strings.Contains(ln, "init_array"), strings.Contains(ln, "cfstring"), strings.Contains(ln, "tdata"):
+		return sectionData
+	}
+	switch {
+	case strings.Contains(ls, "__text"):
+		return sectionText
+	case strings.Contains(ls, "data_const"):
+		return sectionROData
+	case strings.Contains(ls, "__data"):
+		return sectionData
+	}
+	return sectionUnknown
+}
+
+func buildSizeReport(path string, data *readelfData, pkgs []Package, level string) *sizeReport {
+	report := &sizeReport{Binary: path, Modules: make(map[string]*moduleSize)}
+	if data == nil {
+		return report
+	}
+	res := newNameResolver(level, pkgs)
+	var recognized bool
+	for idx, sec := range data.sections {
+		if sec == nil || sec.Size == 0 {
+			continue
+		}
+		if sec.Kind == sectionUnknown {
+			continue
+		}
+		recognized = true
+		end := sec.Address + sec.Size
+		syms := data.symbols[idx]
+		if len(syms) == 0 {
+			report.add("(unknown "+sec.Name+")", sec.Kind, sec.Size)
+			continue
+		}
+		// Sort symbols by address to calculate sizes based on address ranges
+		sort.Slice(syms, func(i, j int) bool {
+			if syms[i].Address == syms[j].Address {
+				return syms[i].Name < syms[j].Name
+			}
+			return syms[i].Address < syms[j].Address
+		})
+		cursor := sec.Address
+		for i := 0; i < len(syms); i++ {
+			sym := syms[i]
+			// Skip symbols that are beyond the section bounds
+			if sym.Address >= end {
+				continue
+			}
+			addr := sym.Address
+			// Clamp symbol address to section start if it's before the section
+			if addr < sec.Address {
+				addr = sec.Address
+			}
+			// Add padding bytes between cursor and current symbol
+			if addr > cursor {
+				report.add("(padding "+sec.Name+")", sec.Kind, addr-cursor)
+				cursor = addr
+			}
+			// Find the next symbol address to calculate this symbol's size.
+			// Symbols at the same address are handled by taking the next different address.
+			next := end
+			// Optimize: check next symbol first before scanning
+			if i+1 < len(syms) && syms[i+1].Address > addr {
+				next = syms[i+1].Address
+			} else {
+				// Only search if next symbol is at same address
+				for j := i + 1; j < len(syms); j++ {
+					if syms[j].Address > addr {
+						next = syms[j].Address
+						break
+					}
+				}
+			}
+			if next > end {
+				next = end
+			}
+			// Skip symbols with zero size
+			if next <= addr {
+				continue
+			}
+			// Attribute the address range [addr, next) to the symbol's module
+			mod := res.resolve(sym.Name)
+			report.add(mod, sec.Kind, next-addr)
+			cursor = next
+		}
+		// Add any remaining padding at the end of the section
+		if cursor < end {
+			report.add("(padding "+sec.Name+")", sec.Kind, end-cursor)
+		}
+	}
+	if !recognized {
+		return nil
+	}
+	return report
+}
+
+func emitJSONReport(w io.Writer, report *sizeReport) error {
+	type moduleJSON struct {
+		Name   string `json:"name"`
+		Code   uint64 `json:"code"`
+		ROData uint64 `json:"rodata"`
+		Data   uint64 `json:"data"`
+		BSS    uint64 `json:"bss"`
+		Flash  uint64 `json:"flash"`
+		RAM    uint64 `json:"ram"`
+	}
+	mods := report.sortedModules()
+	jsonMods := make([]moduleJSON, 0, len(mods))
+	for _, m := range mods {
+		jsonMods = append(jsonMods, moduleJSON{
+			Name:   m.Name,
+			Code:   m.Code,
+			ROData: m.ROData,
+			Data:   m.Data,
+			BSS:    m.BSS,
+			Flash:  m.Flash(),
+			RAM:    m.RAM(),
+		})
+	}
+	payload := struct {
+		Binary  string       `json:"binary"`
+		Modules []moduleJSON `json:"modules"`
+		Total   moduleJSON   `json:"total"`
+	}{
+		Binary:  filepath.Clean(report.Binary),
+		Modules: jsonMods,
+		Total: moduleJSON{
+			Name:   "total",
+			Code:   report.Total.Code,
+			ROData: report.Total.ROData,
+			Data:   report.Total.Data,
+			BSS:    report.Total.BSS,
+			Flash:  report.Total.Flash(),
+			RAM:    report.Total.RAM(),
+		},
+	}
+	enc := json.NewEncoder(w)
+	enc.SetIndent("", "  ")
+	return enc.Encode(payload)
+}
+
+func printTextReport(w io.Writer, report *sizeReport) {
+	fmt.Fprintf(w, "\nSize report for %s\n", filepath.Clean(report.Binary))
+	fmt.Fprintln(w, "   code  rodata    data     bss |   flash     ram | module")
+	fmt.Fprintln(w, "------------------------------- | --------------- | ----------------")
+	for _, m := range report.sortedModules() {
+		fmt.Fprintf(w, "%7d %7d %7d %7d | %7d %7d | %s\n", m.Code, m.ROData, m.Data, m.BSS, m.Flash(), m.RAM(), m.Name)
+	}
+	fmt.Fprintln(w, "------------------------------- | --------------- | ----------------")
+	fmt.Fprintf(w, "%7d %7d %7d %7d | %7d %7d | total\n", report.Total.Code, report.Total.ROData, report.Total.Data, report.Total.BSS, report.Total.Flash(), report.Total.RAM())
+}
+
+func (r *sizeReport) sortedModules() []*moduleSize {
+	mods := make([]*moduleSize, 0, len(r.Modules))
+	for _, m := range r.Modules {
+		mods = append(mods, m)
+	}
+	sort.Slice(mods, func(i, j int) bool {
+		if mods[i].Flash() == mods[j].Flash() {
+			return mods[i].Name < mods[j].Name
+		}
+		return mods[i].Flash() > mods[j].Flash()
+	})
+	return mods
+}
+
+// moduleNameFromSymbol extracts the Go package name from a symbol name.
+// It handles various symbol naming conventions:
+//   - C symbols: Strip leading underscore (e.g., "_main" -> "main")
+//   - Assembler symbols: Strip leading dot (e.g., ".text" -> "text")
+//   - Versioned symbols: Remove version suffix (e.g., "symbol@@GLIBC_2.2.5" -> "symbol")
+//   - Go symbols: Extract package from "package.symbol" format
+//   - Generic types: Strip type parameters (e.g., "pkg(T)" -> "pkg")
+func moduleNameFromSymbol(raw string) string {
+	name := strings.TrimSpace(raw)
+	// Strip C symbol prefix
+	name = strings.TrimPrefix(name, "_")
+	// Strip assembler symbol prefix
+	name = strings.TrimPrefix(name, ".")
+	if name == "" {
+		return "(anonymous)"
+	}
+	// Remove trailing attributes (e.g., "symbol (weak)")
+	if idx := strings.Index(name, " "); idx > 0 {
+		name = name[:idx]
+	}
+	// Remove version suffix for versioned symbols (e.g., "symbol@@GLIBC_2.2.5")
+	if idx := strings.Index(name, "@"); idx > 0 {
+		name = name[:idx]
+	}
+	// Extract Go package name from "package.symbol" format
+	lastDot := strings.LastIndex(name, ".")
+	if lastDot > 0 {
+		pkg := name[:lastDot]
+		// Strip generic type parameters (e.g., "slices.Sort[int]" -> "slices")
+		if paren := strings.Index(pkg, "("); paren > 0 {
+			pkg = pkg[:paren]
+		}
+		pkg = strings.Trim(pkg, " ")
+		if pkg != "" {
+			return pkg
+		}
+	}
+	return name
+}
+
+func parseNameField(field string) string {
+	val := strings.TrimSpace(field)
+	if idx := strings.Index(val, "("); idx >= 0 {
+		val = strings.TrimSpace(val[:idx])
+	}
+	return val
+}
+
+func parseSectionRef(field string, indexBase int) (string, int) {
+	name := parseNameField(field)
+	idx := strings.Index(field, "(")
+	if idx < 0 {
+		return name, -1
+	}
+	end := strings.Index(field[idx:], ")")
+	if end < 0 {
+		return name, -1
+	}
+	val := strings.TrimSpace(field[idx+1 : idx+end])
+	val = strings.TrimPrefix(val, "0x")
+	if val == "" {
+		return name, -1
+	}
+	num, err := strconv.ParseUint(val, 16, 64)
+	if err != nil {
+		return name, -1
+	}
+	if num == 0 {
+		return name, -1
+	}
+	if indexBase == 0 && num >= SHN_LORESERVE {
+		// Special ELF section indices (SHN_ABS, SHN_COMMON, etc.)
+		return name, -1
+	}
+	if num > math.MaxInt {
+		return name, -1
+	}
+	res := int(num) - indexBase
+	if res < 0 {
+		return name, -1
+	}
+	return name, res
+}
+
+func parseUintField(field string) (uint64, error) {
+	val := strings.TrimSpace(field)
+	if strings.HasPrefix(val, "0x") || strings.HasPrefix(val, "0X") {
+		return strconv.ParseUint(val[2:], 16, 64)
+	}
+	return strconv.ParseUint(val, 10, 64)
+}
+
+func ensureSizeReporting(conf *Config) error {
+	if !conf.SizeReport {
+		return nil
+	}
+	switch strings.ToLower(conf.SizeLevel) {
+	case "", "module":
+		conf.SizeLevel = "module"
+	case "package", "full":
+		conf.SizeLevel = strings.ToLower(conf.SizeLevel)
+	default:
+		return fmt.Errorf("invalid size level %q (valid: full,module,package)", conf.SizeLevel)
+	}
+	cmd, err := llvm.New("").Readelf("--version")
+	if err != nil {
+		return fmt.Errorf("llvm-readelf not available: %w", err)
+	}
+	cmd.Stdout = io.Discard
+	cmd.Stderr = io.Discard
+	if err := cmd.Run(); err != nil {
+		return fmt.Errorf("llvm-readelf not available: %w", err)
+	}
+	return nil
+}

internal/build/size_report_test.go

@@ -0,0 +1,156 @@
+//go:build !llgo
+
+package build
+
+import (
+	"bytes"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"golang.org/x/tools/go/packages"
+)
+
+const sampleReadelf = `Sections [
+  Section {
+    Index: 0
+    Name: __text (5F)
+    Segment: __TEXT (5F)
+    Address: 0x1000
+    Size: 0x20
+  }
+  Section {
+    Index: 1
+    Name: __data
+    Segment: __DATA
+    Address: 0x2000
+    Size: 0x10
+  }
+  Section {
+    Index: 2
+    Name: __common
+    Segment: __DATA
+    Address: 0x3000
+    Size: 0x8
+  }
+]
+Symbols [
+  Symbol {
+    Name: _main.main
+    Section: __text (0x1)
+    Value: 0x1000
+  }
+  Symbol {
+    Name: _runtime.init
+    Section: __text (0x1)
+    Value: 0x1010
+  }
+  Symbol {
+    Name: _main.dataVar
+    Section: __data (0x2)
+    Value: 0x2000
+  }
+  Symbol {
+    Name: _runtime.dataVar
+    Section: __data (0x2)
+    Value: 0x2008
+  }
+  Symbol {
+    Name: _runtime.bssVar
+    Section: __common (0x3)
+    Value: 0x3000
+  }
+]
+`
+
+func TestParseReadelfOutput(t *testing.T) {
+	parsed, err := parseReadelfOutput(strings.NewReader(sampleReadelf))
+	if err != nil {
+		t.Fatalf("parseReadelfOutput: %v", err)
+	}
+	report := buildSizeReport("fake.bin", parsed, nil, "")
+	if report == nil {
+		t.Fatal("expected report")
+	}
+	modules := report.Modules
+	if len(modules) == 0 {
+		t.Fatal("expected modules in report")
+	}
+	mainMod, ok := modules["main"]
+	if !ok {
+		t.Fatalf("expected main module, got %v", modules)
+	}
+	if mainMod.Code != 0x10 {
+		t.Fatalf("unexpected main code size: %d", mainMod.Code)
+	}
+	if mainMod.Data != 0x8 {
+		t.Fatalf("unexpected main data size: %d", mainMod.Data)
+	}
+	runtimeMod := modules["runtime"]
+	if runtimeMod.Code != 0x10 {
+		t.Fatalf("unexpected runtime code size: %d", runtimeMod.Code)
+	}
+	if runtimeMod.Data != 0x8 {
+		t.Fatalf("unexpected runtime data size: %d", runtimeMod.Data)
+	}
+	if runtimeMod.BSS != 0x8 {
+		t.Fatalf("unexpected runtime bss size: %d", runtimeMod.BSS)
+	}
+	if report.Total.Flash() != 0x10+0x10+0x8+0x8 {
+		t.Fatalf("unexpected flash total: %d", report.Total.Flash())
+	}
+	if report.Total.RAM() != 0x8+0x8+0x8 {
+		t.Fatalf("unexpected ram total: %d", report.Total.RAM())
+	}
+}
+
+func TestParseReadelfRealBinary(t *testing.T) {
+	path := os.Getenv("LLGO_SIZE_REPORT_BIN")
+	if path == "" {
+		return
+	}
+	absPath, err := filepath.Abs(path)
+	if err != nil {
+		t.Fatalf("abs path: %v", err)
+	}
+	cmd := exec.Command("llvm-readelf", "--all", absPath)
+	var buf bytes.Buffer
+	cmd.Stdout = &buf
+	cmd.Stderr = os.Stderr
+	if err := cmd.Run(); err != nil {
+		t.Fatalf("llvm-readelf failed: %v", err)
+	}
+	parsed, err := parseReadelfOutput(bytes.NewReader(buf.Bytes()))
+	if err != nil {
+		t.Fatalf("parseReadelfOutput(real): %v", err)
+	}
+	if len(parsed.sections) == 0 {
+		t.Fatal("expected sections in real binary")
+	}
+	report := buildSizeReport(absPath, parsed, nil, "")
+	if len(report.Modules) == 0 {
+		t.Fatalf("expected modules for %s", path)
+	}
+}
+
+func TestNameResolver(t *testing.T) {
+	pkgs := []Package{
+		&aPackage{Package: &packages.Package{PkgPath: "github.com/foo/bar", Module: &packages.Module{Path: "github.com/foo"}}},
+	}
+	symbol := "_github.com/foo/bar.Type.method"
+	if got := newNameResolver("package", pkgs).resolve(symbol); got != "github.com/foo/bar" {
+		t.Fatalf("package level want github.com/foo/bar, got %q", got)
+	}
+	if got := newNameResolver("module", pkgs).resolve(symbol); got != "github.com/foo" {
+		t.Fatalf("module level want github.com/foo, got %q", got)
+	}
+	full := newNameResolver("full", pkgs).resolve(symbol)
+	if full != "github.com/foo/bar.Type" {
+		t.Fatalf("full level unexpected: %q", full)
+	}
+	if got := newNameResolver("package", nil).resolve("_llgo_stub.foo"); got != "llgo-stubs" {
+		t.Fatalf("llgo default grouping failed: %q", got)
+	}
+}

runtime/internal/clite/pthread/pthread_gc.go

@@ -1,5 +1,4 @@
-//go:build !nogc
+//go:build !nogc && !baremetal
-// +build !nogc
 
 /*
  * Copyright (c) 2024 The GoPlus Authors (goplus.org). All rights reserved.

runtime/internal/clite/pthread/pthread_nogc.go

@@ -1,5 +1,4 @@
-//go:build nogc
+//go:build nogc || baremetal
-// +build nogc
 
 /*
  * Copyright (c) 2024 The GoPlus Authors (goplus.org). All rights reserved.

runtime/internal/clite/tls/tls_common.go

@@ -1,4 +1,4 @@
-//go:build llgo
+//go:build llgo && !baremetal
 
 /*
  * Copyright (c) 2025 The GoPlus Authors (goplus.org). All rights reserved.

runtime/internal/clite/tls/tls_gc.go

@@ -1,4 +1,4 @@
-//go:build llgo && !nogc
+//go:build llgo && !baremetal && !nogc
 
 /*
  * Copyright (c) 2025 The GoPlus Authors (goplus.org). All rights reserved.

runtime/internal/clite/tls/tls_nogc.go

@@ -1,4 +1,4 @@
-//go:build llgo && nogc
+//go:build llgo && (nogc || baremetal)
 
 /*
  * Copyright (c) 2025 The GoPlus Authors (goplus.org). All rights reserved.

runtime/internal/clite/tls/tls_stub.go

@@ -1,4 +1,4 @@
-//go:build !llgo
+//go:build !llgo || baremetal
 
 /*
  * Copyright (c) 2025 The GoPlus Authors (goplus.org). All rights reserved.

runtime/internal/lib/runtime/runtime2.go

@@ -4,8 +4,6 @@
 
 package runtime
 
-import "runtime"
-
 // Layout of in-memory per-function information prepared by linker
 // See https://golang.org/s/go12symtab.
 // Keep in sync with linker (../cmd/link/internal/ld/pcln.go:/pclntab)
@@ -30,10 +28,6 @@ func StopTrace() {
 	panic("todo: runtime.StopTrace")
 }
 
-func ReadMemStats(m *runtime.MemStats) {
-	panic("todo: runtime.ReadMemStats")
-}
-
 func SetMutexProfileFraction(rate int) int {
 	panic("todo: runtime.SetMutexProfileFraction")
 }

runtime/internal/lib/runtime/runtime_gc.go

@@ -1,8 +1,16 @@
-//go:build !nogc
+//go:build !nogc && !baremetal
 
 package runtime
 
-import "github.com/goplus/llgo/runtime/internal/clite/bdwgc"
+import (
+	"runtime"
+
+	"github.com/goplus/llgo/runtime/internal/clite/bdwgc"
+)
+
+func ReadMemStats(m *runtime.MemStats) {
+	panic("todo: runtime.ReadMemStats")
+}
 
 func GC() {
 	bdwgc.Gcollect()

runtime/internal/lib/runtime/runtime_gc_baremetal.go

@@ -0,0 +1,29 @@
+//go:build !nogc && baremetal
+
+package runtime
+
+import (
+	"runtime"
+
+	"github.com/goplus/llgo/runtime/internal/runtime/tinygogc"
+)
+
+func ReadMemStats(m *runtime.MemStats) {
+	stats := tinygogc.ReadGCStats()
+	m.Alloc = stats.Alloc
+	m.TotalAlloc = stats.TotalAlloc
+	m.Sys = stats.Sys
+	m.Mallocs = stats.Mallocs
+	m.Frees = stats.Frees
+	m.HeapAlloc = stats.HeapAlloc
+	m.HeapSys = stats.HeapSys
+	m.HeapIdle = stats.HeapIdle
+	m.HeapInuse = stats.HeapInuse
+	m.StackInuse = stats.StackInuse
+	m.StackSys = stats.StackSys
+	m.GCSys = stats.GCSys
+}
+
+func GC() {
+	tinygogc.GC()
+}

runtime/internal/runtime/tinygogc/gc.go

@@ -0,0 +1,184 @@
+//go:build baremetal || testGC
+
+package tinygogc
+
+import "unsafe"
+
+type GCStats struct {
+	// General statistics.
+
+	// Alloc is bytes of allocated heap objects.
+	//
+	// This is the same as HeapAlloc (see below).
+	Alloc uint64
+
+	// TotalAlloc is cumulative bytes allocated for heap objects.
+	//
+	// TotalAlloc increases as heap objects are allocated, but
+	// unlike Alloc and HeapAlloc, it does not decrease when
+	// objects are freed.
+	TotalAlloc uint64
+
+	// Sys is the total bytes of memory obtained from the OS.
+	//
+	// Sys is the sum of the XSys fields below. Sys measures the
+	// virtual address space reserved by the Go runtime for the
+	// heap, stacks, and other internal data structures. It's
+	// likely that not all of the virtual address space is backed
+	// by physical memory at any given moment, though in general
+	// it all was at some point.
+	Sys uint64
+
+	// Mallocs is the cumulative count of heap objects allocated.
+	// The number of live objects is Mallocs - Frees.
+	Mallocs uint64
+
+	// Frees is the cumulative count of heap objects freed.
+	Frees uint64
+
+	// Heap memory statistics.
+	//
+	// Interpreting the heap statistics requires some knowledge of
+	// how Go organizes memory. Go divides the virtual address
+	// space of the heap into "spans", which are contiguous
+	// regions of memory 8K or larger. A span may be in one of
+	// three states:
+	//
+	// An "idle" span contains no objects or other data. The
+	// physical memory backing an idle span can be released back
+	// to the OS (but the virtual address space never is), or it
+	// can be converted into an "in use" or "stack" span.
+	//
+	// An "in use" span contains at least one heap object and may
+	// have free space available to allocate more heap objects.
+	//
+	// A "stack" span is used for goroutine stacks. Stack spans
+	// are not considered part of the heap. A span can change
+	// between heap and stack memory; it is never used for both
+	// simultaneously.
+
+	// HeapAlloc is bytes of allocated heap objects.
+	//
+	// "Allocated" heap objects include all reachable objects, as
+	// well as unreachable objects that the garbage collector has
+	// not yet freed. Specifically, HeapAlloc increases as heap
+	// objects are allocated and decreases as the heap is swept
+	// and unreachable objects are freed. Sweeping occurs
+	// incrementally between GC cycles, so these two processes
+	// occur simultaneously, and as a result HeapAlloc tends to
+	// change smoothly (in contrast with the sawtooth that is
+	// typical of stop-the-world garbage collectors).
+	HeapAlloc uint64
+
+	// HeapSys is bytes of heap memory obtained from the OS.
+	//
+	// HeapSys measures the amount of virtual address space
+	// reserved for the heap. This includes virtual address space
+	// that has been reserved but not yet used, which consumes no
+	// physical memory, but tends to be small, as well as virtual
+	// address space for which the physical memory has been
+	// returned to the OS after it became unused (see HeapReleased
+	// for a measure of the latter).
+	//
+	// HeapSys estimates the largest size the heap has had.
+	HeapSys uint64
+
+	// HeapIdle is bytes in idle (unused) spans.
+	//
+	// Idle spans have no objects in them. These spans could be
+	// (and may already have been) returned to the OS, or they can
+	// be reused for heap allocations, or they can be reused as
+	// stack memory.
+	//
+	// HeapIdle minus HeapReleased estimates the amount of memory
+	// that could be returned to the OS, but is being retained by
+	// the runtime so it can grow the heap without requesting more
+	// memory from the OS. If this difference is significantly
+	// larger than the heap size, it indicates there was a recent
+	// transient spike in live heap size.
+	HeapIdle uint64
+
+	// HeapInuse is bytes in in-use spans.
+	//
+	// In-use spans have at least one object in them. These spans
+	// can only be used for other objects of roughly the same
+	// size.
+	//
+	// HeapInuse minus HeapAlloc estimates the amount of memory
+	// that has been dedicated to particular size classes, but is
+	// not currently being used. This is an upper bound on
+	// fragmentation, but in general this memory can be reused
+	// efficiently.
+	HeapInuse uint64
+
+	// Stack memory statistics.
+	//
+	// Stacks are not considered part of the heap, but the runtime
+	// can reuse a span of heap memory for stack memory, and
+	// vice-versa.
+
+	// StackInuse is bytes in stack spans.
+	//
+	// In-use stack spans have at least one stack in them. These
+	// spans can only be used for other stacks of the same size.
+	//
+	// There is no StackIdle because unused stack spans are
+	// returned to the heap (and hence counted toward HeapIdle).
+	StackInuse uint64
+
+	// StackSys is bytes of stack memory obtained from the OS.
+	//
+	// StackSys is StackInuse, plus any memory obtained directly
+	// from the OS for OS thread stacks.
+	//
+	// In non-cgo programs this metric is currently equal to StackInuse
+	// (but this should not be relied upon, and the value may change in
+	// the future).
+	//
+	// In cgo programs this metric includes OS thread stacks allocated
+	// directly from the OS. Currently, this only accounts for one stack in
+	// c-shared and c-archive build modes and other sources of stacks from
+	// the OS (notably, any allocated by C code) are not currently measured.
+	// Note this too may change in the future.
+	StackSys uint64
+
+	// GCSys is bytes of memory in garbage collection metadata.
+	GCSys uint64
+}
+
+func ReadGCStats() GCStats {
+	var heapInuse, heapIdle uint64
+
+	lock(&gcMutex)
+
+	for block := uintptr(0); block < endBlock; block++ {
+		bstate := gcStateOf(block)
+		if bstate == blockStateFree {
+			heapIdle += uint64(bytesPerBlock)
+		} else {
+			heapInuse += uint64(bytesPerBlock)
+		}
+	}
+
+	stackEnd := uintptr(unsafe.Pointer(&_stackEnd))
+	stackSys := stackTop - stackEnd
+
+	stats := GCStats{
+		Alloc:      (gcTotalBlocks - gcFreedBlocks) * uint64(bytesPerBlock),
+		TotalAlloc: gcTotalAlloc,
+		Sys:        uint64(heapEnd - heapStart),
+		Mallocs:    gcMallocs,
+		Frees:      gcFrees,
+		HeapAlloc:  (gcTotalBlocks - gcFreedBlocks) * uint64(bytesPerBlock),
+		HeapSys:    heapInuse + heapIdle,
+		HeapIdle:   heapIdle,
+		HeapInuse:  heapInuse,
+		StackInuse: uint64(stackTop - uintptr(getsp())),
+		StackSys:   uint64(stackSys),
+		GCSys:      uint64(heapEnd - uintptr(metadataStart)),
+	}
+
+	unlock(&gcMutex)
+
+	return stats
+}

runtime/internal/runtime/tinygogc/gc_link.go

@@ -0,0 +1,54 @@
+//go:build !testGC
+
+package tinygogc
+
+import (
+	"unsafe"
+	_ "unsafe"
+)
+
+// LLGoPackage instructs the LLGo linker to wrap C standard library memory allocation
+// functions (malloc, realloc, calloc) so they use the tinygogc allocator instead.
+// This ensures all memory allocations go through the GC, including C library calls.
+const LLGoPackage = "link: --wrap=malloc --wrap=realloc --wrap=calloc"
+
+//export __wrap_malloc
+func __wrap_malloc(size uintptr) unsafe.Pointer {
+	return Alloc(size)
+}
+
+//export __wrap_calloc
+func __wrap_calloc(nmemb, size uintptr) unsafe.Pointer {
+	totalSize := nmemb * size
+	// Check for multiplication overflow
+	if nmemb != 0 && totalSize/nmemb != size {
+		return nil // Overflow
+	}
+	return Alloc(totalSize)
+}
+
+//export __wrap_realloc
+func __wrap_realloc(ptr unsafe.Pointer, size uintptr) unsafe.Pointer {
+	return Realloc(ptr, size)
+}
+
+//go:linkname getsp llgo.stackSave
+func getsp() unsafe.Pointer
+
+//go:linkname _heapStart _heapStart
+var _heapStart [0]byte
+
+//go:linkname _heapEnd _heapEnd
+var _heapEnd [0]byte
+
+//go:linkname _stackStart _stack_top
+var _stackStart [0]byte
+
+//go:linkname _stackEnd _stack_end
+var _stackEnd [0]byte
+
+//go:linkname _globals_start _globals_start
+var _globals_start [0]byte
+
+//go:linkname _globals_end _globals_end
+var _globals_end [0]byte

runtime/internal/runtime/tinygogc/gc_test.go

@@ -0,0 +1,25 @@
+//go:build testGC
+
+package tinygogc
+
+import (
+	_ "unsafe"
+)
+
+var currentStack uintptr
+
+func getsp() uintptr {
+	return currentStack
+}
+
+var _heapStart [0]byte
+
+var _heapEnd [0]byte
+
+var _stackStart [0]byte
+
+var _stackEnd [0]byte
+
+var _globals_start [0]byte
+
+var _globals_end [0]byte

runtime/internal/runtime/tinygogc/gc_tinygo.go

@@ -0,0 +1,570 @@
+//go:build baremetal || testGC
+
+/*
+ * Copyright (c) 2018-2025 The TinyGo Authors. All rights reserved.
+ * Copyright (c) 2024 The GoPlus Authors (goplus.org). All rights reserved.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+// Package tinygogc implements a conservative mark-and-sweep garbage collector
+// for baremetal environments where the standard Go runtime and bdwgc are unavailable.
+//
+// This implementation is based on TinyGo's GC and is designed for resource-constrained
+// embedded systems. It uses a block-based allocator with conservative pointer scanning.
+//
+// Build tags:
+//   - baremetal: Enables this GC for baremetal targets
+//   - testGC: Enables testing mode with mock implementations
+//
+// Memory Layout:
+// The heap is divided into fixed-size blocks (32 bytes on 64-bit). Metadata is stored
+// at the end of the heap, using 2 bits per block to track state (free/head/tail/mark).
+package tinygogc
+
+import (
+	"unsafe"
+
+	c "github.com/goplus/llgo/runtime/internal/clite"
+)
+
+const gcDebug = false
+
+// blockState stores the four states in which a block can be. It is two bits in
+// size.
+const (
+	blockStateFree uint8 = 0 // 00
+	blockStateHead uint8 = 1 // 01
+	blockStateTail uint8 = 2 // 10
+	blockStateMark uint8 = 3 // 11
+	blockStateMask uint8 = 3 // 11
+)
+
+// The byte value of a block where every block is a 'tail' block.
+const blockStateByteAllTails = 0 |
+	uint8(blockStateTail<<(stateBits*3)) |
+	uint8(blockStateTail<<(stateBits*2)) |
+	uint8(blockStateTail<<(stateBits*1)) |
+	uint8(blockStateTail<<(stateBits*0))
+
+var (
+	heapStart     uintptr        // start address of heap area
+	heapEnd       uintptr        // end address of heap area
+	globalsStart  uintptr        // start address of global variable area
+	globalsEnd    uintptr        // end address of global variable area
+	stackTop      uintptr        // the top of stack
+	endBlock      uintptr        // GC end block index
+	metadataStart unsafe.Pointer // start address of GC metadata
+
+	nextAlloc     uintptr // the next block that should be tried by the allocator
+	gcTotalAlloc  uint64  // total number of bytes allocated
+	gcTotalBlocks uint64  // total number of allocated blocks
+	gcMallocs     uint64  // total number of allocations
+	gcFrees       uint64  // total number of objects freed
+	gcFreedBlocks uint64  // total number of freed blocks
+
+	// stackOverflow is a flag which is set when the GC scans too deep while marking.
+	// After it is set, all marked allocations must be re-scanned.
+	markStackOverflow bool
+
+	// zeroSizedAlloc is just a sentinel that gets returned when allocating 0 bytes.
+	zeroSizedAlloc uint8
+
+	gcMutex  mutex // gcMutex protects GC related variables
+	isGCInit bool  // isGCInit indicates GC initialization state
+)
+
+// Some globals + constants for the entire GC.
+
+const (
+	wordsPerBlock      = 4 // number of pointers in an allocated block
+	bytesPerBlock      = wordsPerBlock * unsafe.Sizeof(heapStart)
+	stateBits          = 2 // how many bits a block state takes (see blockState type)
+	blocksPerStateByte = 8 / stateBits
+	markStackSize      = 8 * unsafe.Sizeof((*int)(nil)) // number of to-be-marked blocks to queue before forcing a rescan
+)
+
+// this function MUST be initalized first, which means it's required to be initalized before runtime
+func initGC() {
+	// reserve 2K blocks for libc internal malloc, we cannot wrap those internal functions
+	heapStart = uintptr(unsafe.Pointer(&_heapStart)) + 2048
+	heapEnd = uintptr(unsafe.Pointer(&_heapEnd))
+	globalsStart = uintptr(unsafe.Pointer(&_globals_start))
+	globalsEnd = uintptr(unsafe.Pointer(&_globals_end))
+	totalSize := heapEnd - heapStart
+	metadataSize := (totalSize + blocksPerStateByte*bytesPerBlock) / (1 + blocksPerStateByte*bytesPerBlock)
+	metadataStart = unsafe.Pointer(heapEnd - metadataSize)
+	endBlock = (uintptr(metadataStart) - heapStart) / bytesPerBlock
+	stackTop = uintptr(unsafe.Pointer(&_stackStart))
+
+	c.Memset(metadataStart, 0, metadataSize)
+}
+
+func lazyInit() {
+	if !isGCInit {
+		initGC()
+		isGCInit = true
+	}
+}
+
+func gcPanic(s *c.Char) {
+	c.Printf(c.Str("%s"), s)
+	c.Exit(2)
+}
+
+// blockFromAddr returns a block given an address somewhere in the heap (which
+// might not be heap-aligned).
+func blockFromAddr(addr uintptr) uintptr {
+	if addr < heapStart || addr >= uintptr(metadataStart) {
+		gcPanic(c.Str("gc: trying to get block from invalid address"))
+	}
+	return (addr - heapStart) / bytesPerBlock
+}
+
+// Return a pointer to the start of the allocated object.
+func gcPointerOf(blockAddr uintptr) unsafe.Pointer {
+	return unsafe.Pointer(gcAddressOf(blockAddr))
+}
+
+// Return the address of the start of the allocated object.
+func gcAddressOf(blockAddr uintptr) uintptr {
+	addr := heapStart + blockAddr*bytesPerBlock
+	if addr > uintptr(metadataStart) {
+		gcPanic(c.Str("gc: block pointing inside metadata"))
+	}
+	return addr
+}
+
+// findHead returns the head (first block) of an object, assuming the block
+// points to an allocated object. It returns the same block if this block
+// already points to the head.
+func gcFindHead(blockAddr uintptr) uintptr {
+	for {
+		// Optimization: check whether the current block state byte (which
+		// contains the state of multiple blocks) is composed entirely of tail
+		// blocks. If so, we can skip back to the last block in the previous
+		// state byte.
+		// This optimization speeds up findHead for pointers that point into a
+		// large allocation.
+		stateByte := gcStateByteOf(blockAddr)
+		if stateByte == blockStateByteAllTails {
+			blockAddr -= (blockAddr % blocksPerStateByte) + 1
+			continue
+		}
+
+		// Check whether we've found a non-tail block, which means we found the
+		// head.
+		state := gcStateFromByte(blockAddr, stateByte)
+		if state != blockStateTail {
+			break
+		}
+		blockAddr--
+	}
+	if gcStateOf(blockAddr) != blockStateHead && gcStateOf(blockAddr) != blockStateMark {
+		gcPanic(c.Str("gc: found tail without head"))
+	}
+	return blockAddr
+}
+
+// findNext returns the first block just past the end of the tail. This may or
+// may not be the head of an object.
+func gcFindNext(blockAddr uintptr) uintptr {
+	if gcStateOf(blockAddr) == blockStateHead || gcStateOf(blockAddr) == blockStateMark {
+		blockAddr++
+	}
+	for gcAddressOf(blockAddr) < uintptr(metadataStart) && gcStateOf(blockAddr) == blockStateTail {
+		blockAddr++
+	}
+	return blockAddr
+}
+
+func gcStateByteOf(blockAddr uintptr) byte {
+	return *(*uint8)(unsafe.Add(metadataStart, blockAddr/blocksPerStateByte))
+}
+
+// Return the block state given a state byte. The state byte must have been
+// obtained using b.stateByte(), otherwise the result is incorrect.
+func gcStateFromByte(blockAddr uintptr, stateByte byte) uint8 {
+	return uint8(stateByte>>((blockAddr%blocksPerStateByte)*stateBits)) & blockStateMask
+}
+
+// State returns the current block state.
+func gcStateOf(blockAddr uintptr) uint8 {
+	return gcStateFromByte(blockAddr, gcStateByteOf(blockAddr))
+}
+
+// setState sets the current block to the given state, which must contain more
+// bits than the current state. Allowed transitions: from free to any state and
+// from head to mark.
+func gcSetState(blockAddr uintptr, newState uint8) {
+	stateBytePtr := (*uint8)(unsafe.Add(metadataStart, blockAddr/blocksPerStateByte))
+	*stateBytePtr |= uint8(newState << ((blockAddr % blocksPerStateByte) * stateBits))
+	if gcStateOf(blockAddr) != newState {
+		gcPanic(c.Str("gc: setState() was not successful"))
+	}
+}
+
+// markFree sets the block state to free, no matter what state it was in before.
+func gcMarkFree(blockAddr uintptr) {
+	stateBytePtr := (*uint8)(unsafe.Add(metadataStart, blockAddr/blocksPerStateByte))
+	*stateBytePtr &^= uint8(blockStateMask << ((blockAddr % blocksPerStateByte) * stateBits))
+	if gcStateOf(blockAddr) != blockStateFree {
+		gcPanic(c.Str("gc: markFree() was not successful"))
+	}
+	*(*[wordsPerBlock]uintptr)(unsafe.Pointer(gcAddressOf(blockAddr))) = [wordsPerBlock]uintptr{}
+}
+
+// unmark changes the state of the block from mark to head. It must be marked
+// before calling this function.
+func gcUnmark(blockAddr uintptr) {
+	if gcStateOf(blockAddr) != blockStateMark {
+		gcPanic(c.Str("gc: unmark() on a block that is not marked"))
+	}
+	clearMask := blockStateMask ^ blockStateHead // the bits to clear from the state
+	stateBytePtr := (*uint8)(unsafe.Add(metadataStart, blockAddr/blocksPerStateByte))
+	*stateBytePtr &^= uint8(clearMask << ((blockAddr % blocksPerStateByte) * stateBits))
+	if gcStateOf(blockAddr) != blockStateHead {
+		gcPanic(c.Str("gc: unmark() was not successful"))
+	}
+}
+
+func isOnHeap(ptr uintptr) bool {
+	return ptr >= heapStart && ptr < uintptr(metadataStart)
+}
+
+func isPointer(ptr uintptr) bool {
+	// TODO: implement precise GC
+	return isOnHeap(ptr)
+}
+
+// alloc tries to find some free space on the heap, possibly doing a garbage
+// collection cycle if needed. If no space is free, it panics.
+//
+//go:noinline
+func Alloc(size uintptr) unsafe.Pointer {
+	if size == 0 {
+		return unsafe.Pointer(&zeroSizedAlloc)
+	}
+	lock(&gcMutex)
+	lazyInit()
+
+	gcTotalAlloc += uint64(size)
+	gcMallocs++
+
+	neededBlocks := (size + (bytesPerBlock - 1)) / bytesPerBlock
+	gcTotalBlocks += uint64(neededBlocks)
+
+	// Continue looping until a run of free blocks has been found that fits the
+	// requested size.
+	index := nextAlloc
+	numFreeBlocks := uintptr(0)
+	heapScanCount := uint8(0)
+	for {
+		if index == nextAlloc {
+			if heapScanCount == 0 {
+				heapScanCount = 1
+			} else if heapScanCount == 1 {
+				// The entire heap has been searched for free memory, but none
+				// could be found. Run a garbage collection cycle to reclaim
+				// free memory and try again.
+				heapScanCount = 2
+				freeBytes := gc()
+				heapSize := uintptr(metadataStart) - heapStart
+				if freeBytes < heapSize/3 {
+					// Ensure there is at least 33% headroom.
+					// This percentage was arbitrarily chosen, and may need to
+					// be tuned in the future.
+					growHeap()
+				}
+			} else {
+				// Even after garbage collection, no free memory could be found.
+				// Try to increase heap size.
+				if growHeap() {
+					// Success, the heap was increased in size. Try again with a
+					// larger heap.
+				} else {
+					// Unfortunately the heap could not be increased. This
+					// happens on baremetal systems for example (where all
+					// available RAM has already been dedicated to the heap).
+					gcPanic(c.Str("out of memory"))
+				}
+			}
+		}
+
+		// Wrap around the end of the heap.
+		if index == endBlock {
+			index = 0
+			// Reset numFreeBlocks as allocations cannot wrap.
+			numFreeBlocks = 0
+			// In rare cases, the initial heap might be so small that there are
+			// no blocks at all. In this case, it's better to jump back to the
+			// start of the loop and try again, until the GC realizes there is
+			// no memory and grows the heap.
+			// This can sometimes happen on WebAssembly, where the initial heap
+			// is created by whatever is left on the last memory page.
+			continue
+		}
+
+		// Is the block we're looking at free?
+		if gcStateOf(index) != blockStateFree {
+			// This block is in use. Try again from this point.
+			numFreeBlocks = 0
+			index++
+			continue
+		}
+		numFreeBlocks++
+		index++
+
+		// Are we finished?
+		if numFreeBlocks == neededBlocks {
+			// Found a big enough range of free blocks!
+			nextAlloc = index
+			thisAlloc := index - neededBlocks
+
+			// Set the following blocks as being allocated.
+			gcSetState(thisAlloc, blockStateHead)
+			for i := thisAlloc + 1; i != nextAlloc; i++ {
+				gcSetState(i, blockStateTail)
+			}
+			unlock(&gcMutex)
+			// Return a pointer to this allocation.
+			return c.Memset(gcPointerOf(thisAlloc), 0, size)
+		}
+	}
+}
+
+func Realloc(ptr unsafe.Pointer, size uintptr) unsafe.Pointer {
+	if ptr == nil {
+		return Alloc(size)
+	}
+	lock(&gcMutex)
+	lazyInit()
+	unlock(&gcMutex)
+
+	ptrAddress := uintptr(ptr)
+	endOfTailAddress := gcAddressOf(gcFindNext(blockFromAddr(ptrAddress)))
+
+	// this might be a few bytes longer than the original size of
+	// ptr, because we align to full blocks of size bytesPerBlock
+	oldSize := endOfTailAddress - ptrAddress
+	if size <= oldSize {
+		return ptr
+	}
+
+	newAlloc := Alloc(size)
+	c.Memcpy(newAlloc, ptr, oldSize)
+	free(ptr)
+
+	return newAlloc
+}
+
+func free(ptr unsafe.Pointer) {
+	// TODO: free blocks on request, when the compiler knows they're unused.
+}
+
+func GC() uintptr {
+	lock(&gcMutex)
+	freeBytes := gc()
+	unlock(&gcMutex)
+	return freeBytes
+}
+
+// runGC performs a garbage collection cycle. It is the internal implementation
+// of the runtime.GC() function. The difference is that it returns the number of
+// free bytes in the heap after the GC is finished.
+func gc() (freeBytes uintptr) {
+	lazyInit()
+
+	if gcDebug {
+		println("running collection cycle...")
+	}
+
+	// Mark phase: mark all reachable objects, recursively.
+	gcMarkReachable()
+
+	finishMark()
+
+	// If we're using threads, resume all other threads before starting the
+	// sweep.
+	gcResumeWorld()
+
+	// Sweep phase: free all non-marked objects and unmark marked objects for
+	// the next collection cycle.
+	freeBytes = sweep()
+
+	return
+}
+
+// markRoots reads all pointers from start to end (exclusive) and if they look
+// like a heap pointer and are unmarked, marks them and scans that object as
+// well (recursively). The start and end parameters must be valid pointers and
+// must be aligned.
+func markRoots(start, end uintptr) {
+	if start >= end {
+		gcPanic(c.Str("gc: unexpected range to mark"))
+	}
+	// Reduce the end bound to avoid reading too far on platforms where pointer alignment is smaller than pointer size.
+	// If the size of the range is 0, then end will be slightly below start after this.
+	end -= unsafe.Sizeof(end) - unsafe.Alignof(end)
+
+	for addr := start; addr < end; addr += unsafe.Alignof(addr) {
+		root := *(*uintptr)(unsafe.Pointer(addr))
+		markRoot(addr, root)
+	}
+}
+
+// startMark starts the marking process on a root and all of its children.
+func startMark(root uintptr) {
+	var stack [markStackSize]uintptr
+	stack[0] = root
+	gcSetState(root, blockStateMark)
+	stackLen := 1
+	for stackLen > 0 {
+		// Pop a block off of the stack.
+		stackLen--
+		block := stack[stackLen]
+
+		start, end := gcAddressOf(block), gcAddressOf(gcFindNext(block))
+
+		for addr := start; addr != end; addr += unsafe.Alignof(addr) {
+			// Load the word.
+			word := *(*uintptr)(unsafe.Pointer(addr))
+
+			if !isPointer(word) {
+				// Not a heap pointer.
+				continue
+			}
+
+			// Find the corresponding memory block.
+			referencedBlock := blockFromAddr(word)
+
+			if gcStateOf(referencedBlock) == blockStateFree {
+				// The to-be-marked object doesn't actually exist.
+				// This is probably a false positive.
+				continue
+			}
+
+			// Move to the block's head.
+			referencedBlock = gcFindHead(referencedBlock)
+
+			if gcStateOf(referencedBlock) == blockStateMark {
+				// The block has already been marked by something else.
+				continue
+			}
+
+			// Mark block.
+			gcSetState(referencedBlock, blockStateMark)
+
+			if stackLen == len(stack) {
+				// The stack is full.
+				// It is necessary to rescan all marked blocks once we are done.
+				markStackOverflow = true
+				if gcDebug {
+					println("gc stack overflowed")
+				}
+				continue
+			}
+
+			// Push the pointer onto the stack to be scanned later.
+			stack[stackLen] = referencedBlock
+			stackLen++
+		}
+	}
+}
+
+// finishMark finishes the marking process by processing all stack overflows.
+func finishMark() {
+	for markStackOverflow {
+		// Re-mark all blocks.
+		markStackOverflow = false
+		for block := uintptr(0); block < endBlock; block++ {
+			if gcStateOf(block) != blockStateMark {
+				// Block is not marked, so we do not need to rescan it.
+				continue
+			}
+
+			// Re-mark the block.
+			startMark(block)
+		}
+	}
+}
+
+// mark a GC root at the address addr.
+func markRoot(addr, root uintptr) {
+	if isOnHeap(root) {
+		block := blockFromAddr(root)
+		if gcStateOf(block) == blockStateFree {
+			// The to-be-marked object doesn't actually exist.
+			// This could either be a dangling pointer (oops!) but most likely
+			// just a false positive.
+			return
+		}
+		head := gcFindHead(block)
+
+		if gcStateOf(head) != blockStateMark {
+			startMark(head)
+		}
+	}
+}
+
+// Sweep goes through all memory and frees unmarked
+// It returns how many bytes are free in the heap after the sweep.
+func sweep() (freeBytes uintptr) {
+	freeCurrentObject := false
+	var freed uint64
+
+	for block := uintptr(0); block < endBlock; block++ {
+		switch gcStateOf(block) {
+		case blockStateHead:
+			// Unmarked head. Free it, including all tail blocks following it.
+			gcMarkFree(block)
+			freeCurrentObject = true
+			gcFrees++
+			freed++
+		case blockStateTail:
+			if freeCurrentObject {
+				// This is a tail object following an unmarked head.
+				// Free it now.
+				gcMarkFree(block)
+				freed++
+			}
+		case blockStateMark:
+			// This is a marked object. The next tail blocks must not be freed,
+			// but the mark bit must be removed so the next GC cycle will
+			// collect this object if it is unreferenced then.
+			gcUnmark(block)
+			freeCurrentObject = false
+		case blockStateFree:
+			freeBytes += bytesPerBlock
+		}
+	}
+	gcFreedBlocks += freed
+	freeBytes += uintptr(freed) * bytesPerBlock
+	return
+}
+
+// growHeap tries to grow the heap size. It returns true if it succeeds, false
+// otherwise.
+func growHeap() bool {
+	// On baremetal, there is no way the heap can be grown.
+	return false
+}
+
+func gcMarkReachable() {
+	markRoots(uintptr(getsp()), stackTop)
+	markRoots(globalsStart, globalsEnd)
+}
+
+func gcResumeWorld() {
+	// Nothing to do here (single threaded).
+}

runtime/internal/runtime/tinygogc/mutex.go

@@ -0,0 +1,8 @@
+package tinygogc
+
+// TODO(MeteorsLiu): mutex lock for baremetal GC
+type mutex struct{}
+
+func lock(m *mutex) {}
+
+func unlock(m *mutex) {}

runtime/internal/runtime/tinygogc/pc_mock_test.go

@@ -0,0 +1,604 @@
+//go:build testGC
+
+package tinygogc
+
+import (
+	"testing"
+	"unsafe"
+
+	c "github.com/goplus/llgo/runtime/internal/clite"
+)
+
+const (
+	// Mock a typical embedded system with 128KB RAM
+	mockHeapSize     = 128 * 1024 // 128KB
+	mockGlobalsSize  = 4 * 1024   // 4KB for globals
+	mockStackSize    = 8 * 1024   // 8KB for stack
+	mockReservedSize = 2048       // 2KB reserved as in real implementation
+)
+
+type testObject struct {
+	data [4]uintptr
+}
+
+// mockGCEnv provides a controlled root environment for GC testing
+type mockGCEnv struct {
+	memory       []byte
+	heapStart    uintptr
+	heapEnd      uintptr
+	globalsStart uintptr
+	globalsEnd   uintptr
+	stackStart   uintptr
+	stackEnd     uintptr
+	// Controlled root sets for testing
+	rootObjects []unsafe.Pointer
+	// Original GC state to restore
+	originalHeapStart     uintptr
+	originalHeapEnd       uintptr
+	originalGlobalsStart  uintptr
+	originalGlobalsEnd    uintptr
+	originalStackTop      uintptr
+	originalEndBlock      uintptr
+	originalMetadataStart unsafe.Pointer
+	originalNextAlloc     uintptr
+	originalIsGCInit      bool
+	// Mock mode flag
+	mockMode bool
+}
+
+// createMockGCEnv creates a completely isolated GC environment
+func createMockGCEnv() *mockGCEnv {
+	totalMemory := mockHeapSize + mockGlobalsSize + mockStackSize
+	memory := make([]byte, totalMemory)
+	baseAddr := uintptr(unsafe.Pointer(&memory[0]))
+
+	env := &mockGCEnv{
+		memory:       memory,
+		globalsStart: baseAddr,
+		globalsEnd:   baseAddr + mockGlobalsSize,
+		heapStart:    baseAddr + mockGlobalsSize + mockReservedSize,
+		heapEnd:      baseAddr + mockGlobalsSize + mockHeapSize,
+		stackStart:   baseAddr + mockGlobalsSize + mockHeapSize,
+		stackEnd:     baseAddr + uintptr(totalMemory),
+		rootObjects:  make([]unsafe.Pointer, 0),
+		mockMode:     false,
+	}
+
+	return env
+}
+
+// setupMockGC initializes the GC with mock memory layout using initGC's logic
+func (env *mockGCEnv) setupMockGC() {
+	// Save original GC state
+	env.originalHeapStart = heapStart
+	env.originalHeapEnd = heapEnd
+	env.originalGlobalsStart = globalsStart
+	env.originalGlobalsEnd = globalsEnd
+	env.originalStackTop = stackTop
+	env.originalEndBlock = endBlock
+	env.originalMetadataStart = metadataStart
+	env.originalNextAlloc = nextAlloc
+	env.originalIsGCInit = isGCInit
+
+	// Set currentStack for getsp()
+	currentStack = env.stackStart
+
+	// Apply initGC's logic with our mock memory layout
+	// This is the same logic as initGC() but with our mock addresses
+	heapStart = env.heapStart + 2048 // reserve 2K blocks like initGC does
+	heapEnd = env.heapEnd
+	globalsStart = env.globalsStart
+	globalsEnd = env.globalsEnd
+	stackTop = env.stackEnd
+
+	totalSize := heapEnd - heapStart
+	metadataSize := (totalSize + blocksPerStateByte*bytesPerBlock) / (1 + blocksPerStateByte*bytesPerBlock)
+	metadataStart = unsafe.Pointer(heapEnd - metadataSize)
+	endBlock = (uintptr(metadataStart) - heapStart) / bytesPerBlock
+
+	// Clear metadata using memset like initGC does
+	c.Memset(metadataStart, 0, metadataSize)
+
+	// Reset allocator state and all GC statistics for clean test environment
+	nextAlloc = 0
+	isGCInit = true
+
+	// Reset all GC statistics to start from clean state
+	gcTotalAlloc = 0
+	gcTotalBlocks = 0
+	gcMallocs = 0
+	gcFrees = 0
+	gcFreedBlocks = 0
+	markStackOverflow = false
+}
+
+// restoreOriginalGC restores the original GC state
+func (env *mockGCEnv) restoreOriginalGC() {
+	heapStart = env.originalHeapStart
+	heapEnd = env.originalHeapEnd
+	globalsStart = env.originalGlobalsStart
+	globalsEnd = env.originalGlobalsEnd
+	stackTop = env.originalStackTop
+	endBlock = env.originalEndBlock
+	metadataStart = env.originalMetadataStart
+	nextAlloc = env.originalNextAlloc
+	isGCInit = false
+}
+
+// enableMockMode enables mock root scanning mode
+func (env *mockGCEnv) enableMockMode() {
+	env.mockMode = true
+}
+
+// disableMockMode disables mock root scanning mode
+func (env *mockGCEnv) disableMockMode() {
+	env.mockMode = false
+}
+
+// addRoot adds an object to the controlled root set
+func (env *mockGCEnv) addRoot(ptr unsafe.Pointer) {
+	env.rootObjects = append(env.rootObjects, ptr)
+}
+
+// clearRoots removes all objects from the controlled root set
+func (env *mockGCEnv) clearRoots() {
+	env.rootObjects = env.rootObjects[:0]
+}
+
+// mockMarkReachable replaces gcMarkReachable when in mock mode
+func (env *mockGCEnv) mockMarkReachable() {
+	if !env.mockMode {
+		// Use original logic
+		markRoots(uintptr(getsp()), stackTop)
+		markRoots(globalsStart, globalsEnd)
+		return
+	}
+
+	// Mock mode: only scan our controlled roots
+	for _, root := range env.rootObjects {
+		addr := uintptr(root)
+		markRoot(addr, addr)
+	}
+}
+
+// runMockGC runs standard GC but with controlled root scanning
+func (env *mockGCEnv) runMockGC() uintptr {
+	lock(&gcMutex)
+	defer unlock(&gcMutex)
+
+	lazyInit()
+
+	if gcDebug {
+		println("running mock collection cycle...")
+	}
+
+	// Mark phase: use our mock root scanning
+	env.mockMarkReachable()
+	finishMark()
+
+	// Resume world (no-op in single threaded)
+	gcResumeWorld()
+
+	// Sweep phase: use standard sweep logic
+	return sweep()
+}
+
+// createTestObjects creates a network of objects for testing reachability
+func createTestObjects(env *mockGCEnv) []*testObject {
+	// Allocate several test objects
+	objects := make([]*testObject, 0, 10)
+
+	// Dependencies Graph
+	// root1 -> child1 -> grandchild1 -> child2
+	// root1 -> child2 -> grandchild1
+
+	// Create root objects (reachable from stack/globals)
+	root1 := (*testObject)(Alloc(unsafe.Sizeof(testObject{})))
+	root2 := (*testObject)(Alloc(unsafe.Sizeof(testObject{})))
+	objects = append(objects, root1, root2)
+
+	// Create objects reachable from root1
+	child1 := (*testObject)(Alloc(unsafe.Sizeof(testObject{})))
+	child2 := (*testObject)(Alloc(unsafe.Sizeof(testObject{})))
+	root1.data[0] = uintptr(unsafe.Pointer(child1))
+	root1.data[1] = uintptr(unsafe.Pointer(child2))
+	objects = append(objects, child1, child2)
+
+	// Create objects reachable from child1
+	grandchild1 := (*testObject)(Alloc(unsafe.Sizeof(testObject{})))
+	child1.data[0] = uintptr(unsafe.Pointer(grandchild1))
+	objects = append(objects, grandchild1)
+
+	// Create circular reference between child2 and grandchild1
+	child2.data[0] = uintptr(unsafe.Pointer(grandchild1))
+	grandchild1.data[0] = uintptr(unsafe.Pointer(child2))
+
+	// Create unreachable objects (garbage)
+	garbage1 := (*testObject)(Alloc(unsafe.Sizeof(testObject{})))
+	garbage2 := (*testObject)(Alloc(unsafe.Sizeof(testObject{})))
+	// Create circular reference in garbage
+	garbage1.data[0] = uintptr(unsafe.Pointer(garbage2))
+	garbage2.data[0] = uintptr(unsafe.Pointer(garbage1))
+	objects = append(objects, garbage1, garbage2)
+
+	return objects
+}
+
+func TestMockGCBasicAllocation(t *testing.T) {
+	env := createMockGCEnv()
+	env.setupMockGC()
+	defer env.restoreOriginalGC()
+
+	// Test basic allocation
+	ptr1 := Alloc(32)
+	if ptr1 == nil {
+		t.Fatal("Failed to allocate 32 bytes")
+	}
+
+	ptr2 := Alloc(64)
+	if ptr2 == nil {
+		t.Fatal("Failed to allocate 64 bytes")
+	}
+
+	// Verify pointers are within heap bounds
+	addr1 := uintptr(ptr1)
+	addr2 := uintptr(ptr2)
+
+	if addr1 < heapStart || addr1 >= uintptr(metadataStart) {
+		t.Errorf("ptr1 %x not within heap bounds [%x, %x)", addr1, heapStart, uintptr(metadataStart))
+	}
+
+	if addr2 < heapStart || addr2 >= uintptr(metadataStart) {
+		t.Errorf("ptr2 %x not within heap bounds [%x, %x)", addr2, heapStart, uintptr(metadataStart))
+	}
+
+	t.Logf("Allocated ptr1 at %x, ptr2 at %x", addr1, addr2)
+	t.Logf("Heap bounds: [%x, %x)", heapStart, uintptr(metadataStart))
+}
+
+func TestMockGCReachabilityAndSweep(t *testing.T) {
+	env := createMockGCEnv()
+	env.setupMockGC()
+	defer env.restoreOriginalGC()
+
+	// Track initial stats
+	initialMallocs := gcMallocs
+	initialFrees := gcFrees
+
+	// Create test object network
+	objects := createTestObjects(env)
+
+	// Add first 2 objects as roots using mock control
+	env.enableMockMode()
+	env.addRoot(unsafe.Pointer(objects[0])) // root1
+	env.addRoot(unsafe.Pointer(objects[1])) // root2
+
+	t.Logf("Created %d objects, 2 are roots", len(objects))
+	t.Logf("Mallocs: %d", gcMallocs-initialMallocs)
+
+	// Verify all objects are initially allocated
+	for i, obj := range objects {
+		addr := uintptr(unsafe.Pointer(obj))
+		block := blockFromAddr(addr)
+		state := gcStateOf(block)
+		if state != blockStateHead {
+			t.Errorf("Object %d at %x has state %d, expected %d (HEAD)", i, addr, state, blockStateHead)
+		}
+	}
+
+	// Perform GC with controlled root scanning
+	freedBytes := env.runMockGC()
+	t.Logf("Freed %d bytes during GC", freedBytes)
+	t.Logf("Frees: %d (delta: %d)", gcFrees, gcFrees-initialFrees)
+
+	// Verify reachable objects are still allocated
+	reachableObjects := []unsafe.Pointer{
+		unsafe.Pointer(objects[0]), // root1
+		unsafe.Pointer(objects[1]), // root2
+		unsafe.Pointer(objects[2]), // child1 (reachable from root1)
+		unsafe.Pointer(objects[3]), // child2 (reachable from root1)
+		unsafe.Pointer(objects[4]), // grandchild1 (reachable from child1, child2)
+	}
+
+	for i, obj := range reachableObjects {
+		addr := uintptr(obj)
+		block := blockFromAddr(addr)
+		state := gcStateOf(block)
+		if state != blockStateHead {
+			t.Errorf("Reachable object %d at %x has state %d, expected %d (HEAD)", i, addr, state, blockStateHead)
+		}
+	}
+
+	// Verify unreachable objects are freed
+	unreachableObjects := []unsafe.Pointer{
+		unsafe.Pointer(objects[5]), // garbage1
+		unsafe.Pointer(objects[6]), // garbage2
+	}
+
+	for i, obj := range unreachableObjects {
+		addr := uintptr(obj)
+		block := blockFromAddr(addr)
+		state := gcStateOf(block)
+		if state != blockStateFree {
+			t.Errorf("Unreachable object %d at %x has state %d, expected %d (FREE)", i, addr, state, blockStateFree)
+		}
+	}
+
+	// Verify some memory was actually freed
+	if freedBytes == 0 {
+		t.Error("Expected some memory to be freed, but freed 0 bytes")
+	}
+
+	if gcFrees == initialFrees {
+		t.Error("Expected some objects to be freed, but free count didn't change")
+	}
+
+	// Clear refs to make grandchild1 unreachable
+	objects[2].data[0] = 0 // child1 -> grandchild1
+	objects[3].data[0] = 0 // child2 -> grandchild1
+
+	// Run GC again with same roots
+	freedBytes = env.runMockGC()
+
+	// child2 should still be reachable (through root1)
+	blockAddr := blockFromAddr(uintptr(unsafe.Pointer(objects[3])))
+	state := gcStateOf(blockAddr)
+	if state != blockStateHead {
+		t.Errorf("Object child2 at %x has state %d, expected %d (HEAD)", blockAddr, state, blockStateHead)
+	}
+
+	// grandchild1 should now be unreachable and freed
+	blockAddr = blockFromAddr(uintptr(unsafe.Pointer(objects[4])))
+	state = gcStateOf(blockAddr)
+	if state != blockStateFree {
+		t.Errorf("Object grandchild1 at %x has state %d, expected %d (FREE)", blockAddr, state, blockStateFree)
+	}
+}
+
+func TestMockGCMemoryPressure(t *testing.T) {
+	env := createMockGCEnv()
+	env.setupMockGC()
+	defer env.restoreOriginalGC()
+
+	// Calculate available heap space
+	heapSize := uintptr(metadataStart) - heapStart
+	blockSize := bytesPerBlock
+	maxBlocks := heapSize / blockSize
+
+	t.Logf("Heap size: %d bytes, Block size: %d bytes, Max blocks: %d",
+		heapSize, blockSize, maxBlocks)
+
+	// Allocate until we trigger GC
+	var allocations []unsafe.Pointer
+	allocSize := uintptr(32) // Small allocations
+
+	// Allocate about 80% of heap to trigger GC pressure
+	targetAllocations := int(maxBlocks * 4 / 5) // 80% capacity
+
+	for i := 0; i < targetAllocations; i++ {
+		ptr := Alloc(allocSize)
+		if ptr == nil {
+			t.Fatalf("Failed to allocate at iteration %d", i)
+		}
+		allocations = append(allocations, ptr)
+	}
+
+	initialMallocs := gcMallocs
+	t.Logf("Allocated %d objects (%d mallocs total)", len(allocations), initialMallocs)
+
+	// Enable mock mode and keep only half the allocations as roots
+	env.enableMockMode()
+	keepCount := len(allocations) / 2
+	for i := 0; i < keepCount; i++ {
+		env.addRoot(allocations[i])
+	}
+
+	t.Logf("Keeping %d objects as roots, %d should be freed", keepCount, len(allocations)-keepCount)
+
+	// Force GC with controlled roots
+	freeBytes := env.runMockGC()
+
+	t.Logf("GC freed %d bytes", freeBytes)
+	t.Logf("Objects freed: %d", gcFrees)
+
+	// Try to allocate more after GC
+	for i := 0; i < 10; i++ {
+		ptr := Alloc(allocSize)
+		if ptr == nil {
+			t.Fatalf("Failed to allocate after GC at iteration %d", i)
+		}
+	}
+
+	t.Log("Successfully allocated more objects after GC")
+}
+
+func TestMockGCStats(t *testing.T) {
+	env := createMockGCEnv()
+	env.setupMockGC()
+	defer env.restoreOriginalGC()
+
+	// Get initial stats
+	initialStats := ReadGCStats()
+	t.Logf("Initial stats - Mallocs: %d, Frees: %d, TotalAlloc: %d, Alloc: %d",
+		initialStats.Mallocs, initialStats.Frees, initialStats.TotalAlloc, initialStats.Alloc)
+
+	// Verify basic system stats
+	expectedSys := uint64(env.heapEnd - env.heapStart - 2048)
+	if initialStats.Sys != expectedSys {
+		t.Errorf("Expected Sys %d, got %d", expectedSys, initialStats.Sys)
+	}
+
+	expectedGCSys := uint64(env.heapEnd - uintptr(metadataStart))
+	if initialStats.GCSys != expectedGCSys {
+		t.Errorf("Expected GCSys %d, got %d", expectedGCSys, initialStats.GCSys)
+	}
+
+	// Allocate some objects
+	var allocations []unsafe.Pointer
+	allocSize := uintptr(64)
+	numAllocs := 10
+
+	for i := 0; i < numAllocs; i++ {
+		ptr := Alloc(allocSize)
+		if ptr == nil {
+			t.Fatalf("Failed to allocate at iteration %d", i)
+		}
+		allocations = append(allocations, ptr)
+	}
+
+	// Check stats after allocation
+	afterAllocStats := ReadGCStats()
+	t.Logf("After allocation - Mallocs: %d, Frees: %d, TotalAlloc: %d, Alloc: %d",
+		afterAllocStats.Mallocs, afterAllocStats.Frees, afterAllocStats.TotalAlloc, afterAllocStats.Alloc)
+
+	// Verify allocation stats increased
+	if afterAllocStats.Mallocs <= initialStats.Mallocs {
+		t.Errorf("Expected Mallocs to increase from %d, got %d", initialStats.Mallocs, afterAllocStats.Mallocs)
+	}
+
+	if afterAllocStats.TotalAlloc <= initialStats.TotalAlloc {
+		t.Errorf("Expected TotalAlloc to increase from %d, got %d", initialStats.TotalAlloc, afterAllocStats.TotalAlloc)
+	}
+
+	if afterAllocStats.Alloc <= initialStats.Alloc {
+		t.Errorf("Expected Alloc to increase from %d, got %d", initialStats.Alloc, afterAllocStats.Alloc)
+	}
+
+	// Verify Alloc and HeapAlloc are the same
+	if afterAllocStats.Alloc != afterAllocStats.HeapAlloc {
+		t.Errorf("Expected Alloc (%d) to equal HeapAlloc (%d)", afterAllocStats.Alloc, afterAllocStats.HeapAlloc)
+	}
+
+	// Perform GC with controlled roots - keep only half the allocations
+	env.enableMockMode()
+	keepCount := len(allocations) / 2
+	for i := 0; i < keepCount; i++ {
+		env.addRoot(allocations[i])
+	}
+
+	freedBytes := env.runMockGC()
+	t.Logf("GC freed %d bytes", freedBytes)
+
+	// Check stats after GC
+	afterGCStats := ReadGCStats()
+	t.Logf("After GC - Mallocs: %d, Frees: %d, TotalAlloc: %d, Alloc: %d",
+		afterGCStats.Mallocs, afterGCStats.Frees, afterGCStats.TotalAlloc, afterGCStats.Alloc)
+
+	// Verify GC stats
+	if afterGCStats.Frees <= afterAllocStats.Frees {
+		t.Errorf("Expected Frees to increase from %d, got %d", afterAllocStats.Frees, afterGCStats.Frees)
+	}
+
+	// TotalAlloc should not decrease (cumulative)
+	if afterGCStats.TotalAlloc != afterAllocStats.TotalAlloc {
+		t.Errorf("Expected TotalAlloc to remain %d after GC, got %d", afterAllocStats.TotalAlloc, afterGCStats.TotalAlloc)
+	}
+
+	// Alloc should decrease (freed objects)
+	if afterGCStats.Alloc >= afterAllocStats.Alloc {
+		t.Errorf("Expected Alloc to decrease from %d after GC, got %d", afterAllocStats.Alloc, afterGCStats.Alloc)
+	}
+
+	// Verify heap statistics consistency
+	if afterGCStats.HeapSys != afterGCStats.HeapInuse+afterGCStats.HeapIdle {
+		t.Errorf("Expected HeapSys (%d) to equal HeapInuse (%d) + HeapIdle (%d)",
+			afterGCStats.HeapSys, afterGCStats.HeapInuse, afterGCStats.HeapIdle)
+	}
+
+	// Verify live objects calculation
+	expectedLiveObjects := afterGCStats.Mallocs - afterGCStats.Frees
+	t.Logf("Live objects: %d (Mallocs: %d - Frees: %d)", expectedLiveObjects, afterGCStats.Mallocs, afterGCStats.Frees)
+
+	// The number of live objects should be reasonable (we kept half the allocations plus some overhead)
+	if expectedLiveObjects < uint64(keepCount) {
+		t.Errorf("Expected at least %d live objects, got %d", keepCount, expectedLiveObjects)
+	}
+
+	// Test stack statistics
+	if afterGCStats.StackInuse > afterGCStats.StackSys {
+		t.Errorf("StackInuse (%d) should not exceed StackSys (%d)", afterGCStats.StackInuse, afterGCStats.StackSys)
+	}
+}
+
+func TestMockGCCircularReferences(t *testing.T) {
+	env := createMockGCEnv()
+	env.setupMockGC()
+	defer env.restoreOriginalGC()
+
+	type Node struct {
+		data [3]uintptr
+		next uintptr
+	}
+
+	// Create a circular linked list
+	nodes := make([]*Node, 5)
+	for i := range nodes {
+		nodes[i] = (*Node)(Alloc(unsafe.Sizeof(Node{})))
+		nodes[i].data[0] = uintptr(i) // Store index as data
+	}
+
+	// Link them in a circle
+	for i := range nodes {
+		nextIdx := (i + 1) % len(nodes)
+		nodes[i].next = uintptr(unsafe.Pointer(nodes[nextIdx]))
+	}
+
+	t.Logf("Created circular list of %d nodes", len(nodes))
+
+	// Initially all should be allocated
+	for i, node := range nodes {
+		addr := uintptr(unsafe.Pointer(node))
+		block := blockFromAddr(addr)
+		state := gcStateOf(block)
+		if state != blockStateHead {
+			t.Errorf("Node %d at %x has state %d, expected %d", i, addr, state, blockStateHead)
+		}
+	}
+
+	// Test 1: With root references - objects should NOT be freed
+	env.enableMockMode()
+	// Add the first node as root (keeps entire circle reachable)
+	env.addRoot(unsafe.Pointer(nodes[0]))
+
+	freeBytes := env.runMockGC()
+	t.Logf("GC with root reference freed %d bytes", freeBytes)
+
+	// All nodes should still be allocated since they're reachable through the root
+	for i, node := range nodes {
+		addr := uintptr(unsafe.Pointer(node))
+		block := blockFromAddr(addr)
+		state := gcStateOf(block)
+		if state != blockStateHead {
+			t.Errorf("Node %d at %x should still be allocated, but has state %d", i, addr, state)
+		}
+	}
+
+	// Test 2: Without root references - all circular objects should be freed
+	env.clearRoots() // Remove all root references
+
+	freeBytes = env.runMockGC()
+	t.Logf("GC without roots freed %d bytes", freeBytes)
+
+	// All nodes should now be freed since they're not reachable from any roots
+	expectedFreed := uintptr(len(nodes)) * ((unsafe.Sizeof(Node{}) + bytesPerBlock - 1) / bytesPerBlock) * bytesPerBlock
+
+	if freeBytes < expectedFreed {
+		t.Errorf("Expected at least %d bytes freed, got %d", expectedFreed, freeBytes)
+	}
+
+	// Verify all nodes are actually freed
+	for i, node := range nodes {
+		addr := uintptr(unsafe.Pointer(node))
+		block := blockFromAddr(addr)
+		state := gcStateOf(block)
+		if state != blockStateFree {
+			t.Errorf("Node %d at %x should be freed, but has state %d", i, addr, state)
+		}
+	}
+
+	// Verify we can allocate new objects in the freed space
+	newPtr := Alloc(unsafe.Sizeof(Node{}))
+	if newPtr == nil {
+		t.Error("Failed to allocate after freeing circular references")
+	}
+}

runtime/internal/runtime/z_defer_gc.go

@@ -1,4 +1,4 @@
-//go:build !nogc
+//go:build !nogc && !baremetal
 
 /*
  * Copyright (c) 2025 The GoPlus Authors (goplus.org). All rights reserved.

runtime/internal/runtime/z_defer_nogc.go

@@ -1,4 +1,4 @@
-//go:build nogc
+//go:build nogc || baremetal
 
 /*
  * Copyright (c) 2025 The GoPlus Authors (goplus.org). All rights reserved.

runtime/internal/runtime/z_gc.go

@@ -1,5 +1,4 @@
-//go:build !nogc
+//go:build !nogc && !baremetal
-// +build !nogc
 
 /*
  * Copyright (c) 2024 The GoPlus Authors (goplus.org). All rights reserved.

runtime/internal/runtime/z_gc_baremetal.go

@@ -0,0 +1,35 @@
+//go:build !nogc && baremetal
+
+/*
+ * Copyright (c) 2024 The GoPlus Authors (goplus.org). All rights reserved.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package runtime
+
+import (
+	"unsafe"
+
+	"github.com/goplus/llgo/runtime/internal/runtime/tinygogc"
+)
+
+// AllocU allocates uninitialized memory.
+func AllocU(size uintptr) unsafe.Pointer {
+	return tinygogc.Alloc(size)
+}
+
+// AllocZ allocates zero-initialized memory.
+func AllocZ(size uintptr) unsafe.Pointer {
+	return tinygogc.Alloc(size)
+}

ssa/globals.go

@@ -31,6 +31,19 @@ func (pkg Package) AddGlobalString(name string, value string) {
 	pkg.NewVarEx(name, prog.Pointer(styp)).Init(Expr{cv, styp})
 }
 
+// ConstString creates an SSA expression representing a Go string literal. The
+// returned value is backed by an anonymous global constant and can be used to
+// initialize package-level variables or other constant contexts that expect a
+// Go string value.
+func (pkg Package) ConstString(value string) Expr {
+	prog := pkg.Prog
+	styp := prog.String()
+	data := pkg.createGlobalStr(value)
+	length := prog.IntVal(uint64(len(value)), prog.Uintptr())
+	cv := llvm.ConstNamedStruct(styp.ll, []llvm.Value{data, length.impl})
+	return Expr{cv, styp}
+}
+
 // Undefined global string var by names
 func (pkg Package) Undefined(names ...string) error {
 	prog := pkg.Prog

targets/esp32-riscv.app.elf.ld

@@ -1,6 +1,4 @@
-__stack = ORIGIN(dram_seg) + LENGTH(dram_seg);
+_heapEnd = ORIGIN(dram_seg) + LENGTH(dram_seg);
-__MIN_STACK_SIZE = 0x1000;
-_stack_top = __stack;
 
 /* Default entry point */
 ENTRY(_start)
@@ -94,6 +92,12 @@ SECTIONS
     _iram_end = .;
   } > iram_seg
 
+  .stack (NOLOAD) :
+  {
+    . += 16K;
+    __stack = .;
+  } > dram_seg
+
   /**
    * This section is required to skip .iram0.text area because iram0_0_seg and
    * dram0_0_seg reflect the same address space on different buses.
@@ -148,7 +152,7 @@ SECTIONS
   } > dram_seg
 
   /* Check if data + heap + stack exceeds RAM limit */
-  ASSERT(_end <= __stack - __MIN_STACK_SIZE, "region DRAM overflowed by .data and .bss sections")
+  ASSERT(_end <= _heapEnd, "region DRAM overflowed by .data and .bss sections")
 
   /* Stabs debugging sections.  */
   .stab          0 : { *(.stab) }
@@ -193,3 +197,8 @@ SECTIONS
   .gnu.attributes 0 : { KEEP (*(.gnu.attributes)) }
   /DISCARD/ : { *(.note.GNU-stack) *(.gnu_debuglink) *(.gnu.lto_*) }
 }
+
+_globals_start = _data_start;
+_globals_end = _end;
+_heapStart = _end;
+_stack_top = __stack;

targets/esp32.app.elf.ld

@@ -1,5 +1,4 @@
-__stack = ORIGIN(dram_seg) + LENGTH(dram_seg);
+_heapEnd = ORIGIN(dram_seg) + LENGTH(dram_seg);
-__MIN_STACK_SIZE = 0x2000;
 
 ENTRY(_start)
 SECTIONS
@@ -26,6 +25,14 @@ SECTIONS
      the same address within the page on the next page up.  */
   . = ALIGN (CONSTANT (MAXPAGESIZE)) - ((CONSTANT (MAXPAGESIZE) - .) & (CONSTANT (MAXPAGESIZE) - 1)); . = DATA_SEGMENT_ALIGN (CONSTANT (MAXPAGESIZE), CONSTANT (COMMONPAGESIZE));
 
+  .stack (NOLOAD) :
+  {
+    _stack_end = .;
+    . = ALIGN(16);
+    . += 16K;
+    __stack = .;
+  }
+
 
   .rodata    :
   {
@@ -116,7 +123,7 @@ SECTIONS
   . = DATA_SEGMENT_END (.);
 
   /* Check if data + heap + stack exceeds RAM limit */
-  ASSERT(. <= __stack - __MIN_STACK_SIZE, "region DRAM overflowed by .data and .bss sections")
+  ASSERT(. <= _heapEnd, "region DRAM overflowed by .data and .bss sections")
 
   /* Stabs debugging sections.  */
   .stab          0 : { *(.stab) }
@@ -165,4 +172,7 @@ SECTIONS
 
 _sbss = __bss_start;
 _ebss = _end;
-
+_globals_start = _data_start;
+_globals_end = _end;
+_heapStart = _end;
+_stack_top = __stack;

targets/esp32.memory.elf.ld

@@ -19,8 +19,8 @@ MEMORY
 /* 64k at the end of DRAM, after ROM bootloader stack
  * or entire DRAM (for QEMU only)
  */
-  dram_seg (RW) : org = 0x3FFF0000 ,
+  dram_seg (RW) : org = 0x3ffae000 ,
-                  len = 0x10000
+                  len = 0x52000
 }
 
 INCLUDE "targets/esp32.app.elf.ld";

xtool/env/llvm/llvm.go

@@ -17,9 +17,11 @@
 package llvm
 
 import (
+	"fmt"
 	"os"
 	"os/exec"
 	"path/filepath"
+	"sort"
 	"strings"
 
 	"github.com/goplus/llgo/internal/env"
@@ -108,4 +110,62 @@ func (e *Env) InstallNameTool() *install_name_tool.Cmd {
 	return install_name_tool.New(bin)
 }
 
+// Readelf returns a command to execute llvm-readelf with given arguments.
+func (e *Env) Readelf(args ...string) (*exec.Cmd, error) {
+	path, err := e.toolPath("llvm-readelf")
+	if err != nil {
+		return nil, err
+	}
+	return exec.Command(path, args...), nil
+}
+
+func (e *Env) toolPath(base string) (string, error) {
+	if tool := searchTool(e.binDir, base); tool != "" {
+		return tool, nil
+	}
+	if tool, err := exec.LookPath(base); err == nil {
+		return tool, nil
+	}
+	if tool := searchToolInPath(base); tool != "" {
+		return tool, nil
+	}
+	return "", fmt.Errorf("%s not found", base)
+}
+
+func searchTool(dir, base string) string {
+	if dir == "" {
+		return ""
+	}
+	candidate := filepath.Join(dir, base)
+	if isExecutable(candidate) {
+		return candidate
+	}
+	pattern := filepath.Join(dir, base+"-*")
+	matches, _ := filepath.Glob(pattern)
+	sort.Sort(sort.Reverse(sort.StringSlice(matches)))
+	for _, match := range matches {
+		if isExecutable(match) {
+			return match
+		}
+	}
+	return ""
+}
+
+func searchToolInPath(base string) string {
+	for _, dir := range filepath.SplitList(os.Getenv("PATH")) {
+		if tool := searchTool(dir, base); tool != "" {
+			return tool
+		}
+	}
+	return ""
+}
+
+func isExecutable(path string) bool {
+	if path == "" {
+		return false
+	}
+	info, err := os.Stat(path)
+	return err == nil && !info.IsDir()
+}
+
 // -----------------------------------------------------------------------------