all: version 0.35.0

builder: show files in size report table
Show which files cause a binary size increase. This makes it easier to see where the size is going: for example, this makes it easy to see how much the GC contributes to code size compared to other runtime parts.
2026-07-28 15:48:40 +00:00 · 2024-12-20 12:09:22 +01:00 · 2024-12-19 15:08:37 +01:00 · 2024-12-19 15:08:37 +01:00 · 2024-12-19 13:44:11 +01:00 · 2024-12-19 12:54:25 +01:00
231 changed files with 6183 additions and 2355 deletions
@@ -90,6 +90,8 @@ commands:
              name: Check Go code formatting
              command: make fmt-check lint
      - run: make gen-device -j4
+      # TODO: change this to -skip='TestErrors|TestWasm' with Go 1.20
+      - run: go test -tags=llvm<<parameters.llvm>> -short -run='TestBuild|TestTest|TestGetList|TestTraceback'
      - run: make smoketest XTENSA=0
      - save_cache:
          key: go-cache-v4-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_BUILD_NUM }}
@@ -33,6 +33,9 @@ jobs:
        uses: actions/checkout@v4
        with:
          submodules: true
+      - name: Extract TinyGo version
+        id: version
+        run: ./.github/workflows/tinygo-extract-version.sh | tee -a "$GITHUB_OUTPUT"
      - name: Install Go
        uses: actions/setup-go@v5
        with:
@@ -104,7 +107,7 @@ jobs:
      - name: Test stdlib packages
        run: make tinygo-test
      - name: Make release artifact
-        run: cp -p build/release.tar.gz build/tinygo.darwin-${{ matrix.goarch }}.tar.gz
+        run: cp -p build/release.tar.gz build/tinygo${{ steps.version.outputs.version }}.darwin-${{ matrix.goarch }}.tar.gz
      - name: Publish release artifact
        # Note: this release artifact is double-zipped, see:
        # https://github.com/actions/upload-artifact/issues/39
@@ -114,8 +117,8 @@ jobs:
        # We're doing the former here, to keep artifact uploads fast.
        uses: actions/upload-artifact@v4
        with:
-          name: darwin-${{ matrix.goarch }}-double-zipped
-          path: build/tinygo.darwin-${{ matrix.goarch }}.tar.gz
+          name: darwin-${{ matrix.goarch }}-double-zipped-${{ steps.version.outputs.version }}
+          path: build/tinygo${{ steps.version.outputs.version }}.darwin-${{ matrix.goarch }}.tar.gz
      - name: Smoke tests
        run: make smoketest TINYGO=$(PWD)/build/tinygo
  test-macos-homebrew:
@@ -19,6 +19,8 @@ jobs:
    runs-on: ubuntu-latest
    container:
      image: golang:1.23-alpine
+    outputs:
+      version: ${{ steps.version.outputs.version }}
    steps:
      - name: Install apk dependencies
        # tar: needed for actions/cache@v4
@@ -32,6 +34,9 @@ jobs:
        uses: actions/checkout@v4
        with:
          submodules: true
+      - name: Extract TinyGo version
+        id: version
+        run: ./.github/workflows/tinygo-extract-version.sh | tee -a "$GITHUB_OUTPUT"
      - name: Cache Go
        uses: actions/cache@v4
        with:
@@ -120,15 +125,15 @@ jobs:
      - name: Build TinyGo release
        run: |
          make release deb -j3 STATIC=1
-          cp -p build/release.tar.gz /tmp/tinygo.linux-amd64.tar.gz
-          cp -p build/release.deb    /tmp/tinygo_amd64.deb
+          cp -p build/release.tar.gz /tmp/tinygo${{ steps.version.outputs.version }}.linux-amd64.tar.gz
+          cp -p build/release.deb    /tmp/tinygo_${{ steps.version.outputs.version }}_amd64.deb
      - name: Publish release artifact
        uses: actions/upload-artifact@v4
        with:
-          name: linux-amd64-double-zipped
+          name: linux-amd64-double-zipped-${{ steps.version.outputs.version }}
          path: |
-            /tmp/tinygo.linux-amd64.tar.gz
-            /tmp/tinygo_amd64.deb
+            /tmp/tinygo${{ steps.version.outputs.version }}.linux-amd64.tar.gz
+            /tmp/tinygo_${{ steps.version.outputs.version }}_amd64.deb
  test-linux-build:
    # Test the binaries built in the build-linux job by running the smoke tests.
    runs-on: ubuntu-latest
@@ -152,11 +157,11 @@ jobs:
      - name: Download release artifact
        uses: actions/download-artifact@v4
        with:
-          name: linux-amd64-double-zipped
+          name: linux-amd64-double-zipped-${{ needs.build-linux.outputs.version }}
      - name: Extract release tarball
        run: |
          mkdir -p ~/lib
-          tar -C ~/lib -xf tinygo.linux-amd64.tar.gz
+          tar -C ~/lib -xf tinygo${{ needs.build-linux.outputs.version }}.linux-amd64.tar.gz
          ln -s ~/lib/tinygo/bin/tinygo ~/go/bin/tinygo
      - run: make tinygo-test-wasip1-fast
      - run: make tinygo-test-wasip2-fast
@@ -164,7 +169,7 @@ jobs:
  assert-test-linux:
    # Run all tests that can run on Linux, with LLVM assertions enabled to catch
    # potential bugs.
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-22.04
    steps:
      - name: Checkout
        uses: actions/checkout@v4
@@ -297,6 +302,9 @@ jobs:
    steps:
      - name: Checkout
        uses: actions/checkout@v4
+      - name: Get TinyGo version
+        id: version
+        run: ./.github/workflows/tinygo-extract-version.sh | tee -a "$GITHUB_OUTPUT"
      - name: Install apt dependencies
        run: |
          sudo apt-get update
@@ -381,11 +389,11 @@ jobs:
      - name: Download amd64 release
        uses: actions/download-artifact@v4
        with:
-          name: linux-amd64-double-zipped
+          name: linux-amd64-double-zipped-${{ needs.build-linux.outputs.version }}
      - name: Extract amd64 release
        run: |
          mkdir -p build/release
-          tar -xf tinygo.linux-amd64.tar.gz -C build/release tinygo
+          tar -xf tinygo${{ needs.build-linux.outputs.version }}.linux-amd64.tar.gz -C build/release tinygo
      - name: Modify release
        run: |
          cp -p build/tinygo   build/release/tinygo/bin
@@ -393,12 +401,12 @@ jobs:
      - name: Create ${{ matrix.goarch }} release
        run: |
          make release deb RELEASEONLY=1 DEB_ARCH=${{ matrix.libc }}
-          cp -p build/release.tar.gz /tmp/tinygo.linux-${{ matrix.goarch }}.tar.gz
-          cp -p build/release.deb    /tmp/tinygo_${{ matrix.libc }}.deb
+          cp -p build/release.tar.gz /tmp/tinygo${{ steps.version.outputs.version }}.linux-${{ matrix.goarch }}.tar.gz
+          cp -p build/release.deb    /tmp/tinygo_${{ steps.version.outputs.version }}_${{ matrix.libc }}.deb
      - name: Publish release artifact
        uses: actions/upload-artifact@v4
        with:
-          name: linux-${{ matrix.goarch }}-double-zipped
+          name: linux-${{ matrix.goarch }}-double-zipped-${{ steps.version.outputs.version }}
          path: |
-            /tmp/tinygo.linux-${{ matrix.goarch }}.tar.gz
-            /tmp/tinygo_${{ matrix.libc }}.deb
+            /tmp/tinygo${{ steps.version.outputs.version }}.linux-${{ matrix.goarch }}.tar.gz
+            /tmp/tinygo_${{ steps.version.outputs.version }}_${{ matrix.libc }}.deb
@@ -0,0 +1,4 @@
+#!/bin/sh
+
+# Extract the version string from the source code, to be stored in a variable.
+grep 'const version' goenv/version.go | sed 's/^const version = "\(.*\)"$/version=\1/g'
@@ -14,6 +14,8 @@ concurrency:
 jobs:
  build-windows:
    runs-on: windows-2022
+    outputs:
+      version: ${{ steps.version.outputs.version }}
    steps:
      - name: Configure pagefile
        uses: al-cheb/configure-pagefile-action@v1.4
@@ -32,6 +34,10 @@ jobs:
        uses: actions/checkout@v4
        with:
          submodules: true
+      - name: Extract TinyGo version
+        id: version
+        shell: bash
+        run: ./.github/workflows/tinygo-extract-version.sh | tee -a "$GITHUB_OUTPUT"
      - name: Install Go
        uses: actions/setup-go@v5
        with:
@@ -94,6 +100,13 @@ jobs:
      - name: Build wasi-libc
        if: steps.cache-wasi-libc.outputs.cache-hit != 'true'
        run: make wasi-libc
+      - name: Cache Go cache
+        uses: actions/cache@v4
+        with:
+          key: go-cache-windows-v1-${{ hashFiles('go.mod') }}
+          path: |
+            C:/Users/runneradmin/AppData/Local/go-build
+            C:/Users/runneradmin/go/pkg/mod
      - name: Install wasmtime
        run: |
          scoop install wasmtime@14.0.4
@@ -108,7 +121,7 @@ jobs:
      - name: Make release artifact
        shell: bash
        working-directory: build/release
-        run: 7z -tzip a release.zip tinygo
+        run: 7z -tzip a tinygo${{ steps.version.outputs.version }}.windows-amd64.zip tinygo
      - name: Publish release artifact
        # Note: this release artifact is double-zipped, see:
        # https://github.com/actions/upload-artifact/issues/39
@@ -118,8 +131,8 @@ jobs:
        # We're doing the former here, to keep artifact uploads fast.
        uses: actions/upload-artifact@v4
        with:
-          name: windows-amd64-double-zipped
-          path: build/release/release.zip
+          name: windows-amd64-double-zipped-${{ steps.version.outputs.version }}
+          path: build/release/tinygo${{ steps.version.outputs.version }}.windows-amd64.zip

  smoke-test-windows:
    runs-on: windows-2022
@@ -148,12 +161,12 @@ jobs:
      - name: Download TinyGo build
        uses: actions/download-artifact@v4
        with:
-          name: windows-amd64-double-zipped
+          name: windows-amd64-double-zipped-${{ needs.build-windows.outputs.version }}
          path: build/
      - name: Unzip TinyGo build
        shell: bash
        working-directory: build
-        run: 7z x release.zip -r
+        run: 7z x tinygo*.windows-amd64.zip -r
      - name: Smoke tests
        shell: bash
        run: make smoketest TINYGO=$(PWD)/build/tinygo/bin/tinygo
@@ -178,12 +191,12 @@ jobs:
      - name: Download TinyGo build
        uses: actions/download-artifact@v4
        with:
-          name: windows-amd64-double-zipped
+          name: windows-amd64-double-zipped-${{ needs.build-windows.outputs.version }}
          path: build/
      - name: Unzip TinyGo build
        shell: bash
        working-directory: build
-        run: 7z x release.zip -r
+        run: 7z x tinygo*.windows-amd64.zip -r
      - name: Test stdlib packages
        run: make tinygo-test TINYGO=$(PWD)/build/tinygo/bin/tinygo

@@ -214,11 +227,11 @@ jobs:
      - name: Download TinyGo build
        uses: actions/download-artifact@v4
        with:
-          name: windows-amd64-double-zipped
+          name: windows-amd64-double-zipped-${{ needs.build-windows.outputs.version }}
          path: build/
      - name: Unzip TinyGo build
        shell: bash
        working-directory: build
-        run: 7z x release.zip -r
+        run: 7z x tinygo*.windows-amd64.zip -r
      - name: Test stdlib packages on wasip1
        run: make tinygo-test-wasip1-fast TINYGO=$(PWD)/build/tinygo/bin/tinygo
@@ -37,5 +37,9 @@ test.exe
 test.gba
 test.hex
 test.nro
+test.uf2
 test.wasm
 wasm.wasm
+
+*.uf2
+*.elf
@@ -1,3 +1,71 @@
+0.35.0
+---
+* **general**
+  - update cmsis-svd library
+  - use default UART settings in the echo example
+  - `goenv`: also show git hash with custom build of TinyGo
+  - `goenv`: support parsing development versions of Go
+  - `main`: parse extldflags early so we can report the error message
+* **compiler**
+  - `builder`: whitelist temporary directory env var for Clang invocation to fix Windows bug
+  - `builder`: fix cache paths in `-size=full` output
+  - `builder`: work around incorrectly escaped DWARF paths on Windows (Clang bug)
+  - `builder`: fix wasi-libc path names on Windows with `-size=full`
+  - `builder`: write HTML size report
+  - `cgo`: support C identifiers only referred to from within macros
+  - `cgo`: support function-like macros
+  - `cgo`: support errno value as second return parameter
+  - `cgo`: add support for `#cgo noescape` lines
+  - `compiler`: fix bug in interrupt lowering
+  - `compiler`: allow panic directly in `defer`
+  - `compiler`: fix wasmimport -> wasmexport in error message
+  - `compiler`: support `//go:noescape` pragma
+  - `compiler`: report error instead of crashing when instantiating a generic function without body
+  - `interp`: align created globals
+* **standard library**
+  - `machine`: modify i2s interface/implementation to better match specification
+  - `os`: implement `StartProcess`
+  - `reflect`: add `Value.Clear`
+  - `reflect`: add interface support to `NumMethods`
+  - `reflect`: fix `AssignableTo` for named + non-named types
+  - `reflect`: implement `CanConvert`
+  - `reflect`: handle more cases in `Convert`
+  - `reflect`: fix Copy of non-pointer array with size > 64bits
+  - `runtime`: don't call sleepTicks with a negative duration
+  - `runtime`: optimize GC scanning (findHead)
+  - `runtime`: move constants into shared package
+  - `runtime`: add `runtime.fcntl` function for internal/syscall/unix
+  - `runtime`: heapptr only needs to be initialized once
+  - `runtime`: refactor scheduler (this fixes a few bugs with `-scheduler=none`)
+  - `runtime`: rewrite channel implementation to be smaller and more flexible
+  - `runtime`: use `SA_RESTART` when registering a signal for os/signal
+  - `runtime`: implement race-free signals using futexes
+  - `runtime`: run deferred functions in `Goexit`
+  - `runtime`: remove `Cond` which seems to be unused
+  - `runtime`: properly handle unix read on directory
+  - `runtime/trace`: stub all public methods
+  - `sync`: don't use volatile in `Mutex`
+  - `sync`: implement `WaitGroup` using a (pseudo)futex
+  - `sync`: make `Cond` parallelism-safe
+  - `syscall`: use wasi-libc tables for wasm/js target
+* **targets**
+  - `mips`: fix a bug when scanning the stack
+  - `nintendoswitch`: get this target to compile again
+  - `rp2350`: add support for the new RP2350
+  - `rp2040/rp2350` : make I2C implementation shared for rp2040/rp2350
+  - `rp2040/rp2350` : make SPI implementation shared for rp2040/rp2350
+  - `rp2040/rp2350` : make RNG implementation shared for rp2040/rp2350
+  - `wasm`: revise and simplify wasmtime argument handling
+  - `wasm`: support `//go:wasmexport` functions after a call to `time.Sleep`
+  - `wasm`: correctly return from run() in wasm_exec.js
+  - `wasm`: call process.exit() when go.run() returns
+  - `windows`: don't return, exit via exit(0) instead to flush stdout buffer
+* **boards**
+  - add support for the Tillitis TKey
+  - add support for the Raspberry Pi Pico2 (based on the RP2040)
+  - add support for Pimoroni Tiny2350
+
+
 0.34.0
 ---
 * **general**
@@ -291,7 +291,7 @@ endif

 tinygo: ## Build the TinyGo compiler
 	@if [ ! -f "$(LLVM_BUILDDIR)/bin/llvm-config" ]; then echo "Fetch and build LLVM first by running:"; echo "  $(MAKE) llvm-source"; echo "  $(MAKE) $(LLVM_BUILDDIR)"; exit 1; fi
-	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GOENVFLAGS) $(GO) build -buildmode exe -o build/tinygo$(EXE) -tags "byollvm osusergo" -ldflags="-X github.com/tinygo-org/tinygo/goenv.GitSha1=`git rev-parse --short HEAD`" .
+	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GOENVFLAGS) $(GO) build -buildmode exe -o build/tinygo$(EXE) -tags "byollvm osusergo" .
 test: wasi-libc check-nodejs-version
 	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) test $(GOTESTFLAGS) -timeout=1h -buildmode exe -tags "byollvm osusergo" $(GOTESTPKGS)

@@ -741,6 +741,10 @@ endif
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=thumby              examples/echo
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pico2               examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=tiny2350            examples/blinky1
+	@$(MD5SUM) test.hex
 	# test pwm
 	$(TINYGO) build -size short -o test.hex -target=itsybitsy-m0        examples/pwm
 	@$(MD5SUM) test.hex
@@ -853,6 +857,8 @@ endif
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=maixbit             examples/blinky1
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=tkey                examples/blinky1
+	@$(MD5SUM) test.hex
 ifneq ($(WASM), 0)
 	$(TINYGO) build -size short -o wasm.wasm -target=wasm               examples/wasm/export
 	$(TINYGO) build -size short -o wasm.wasm -target=wasm               examples/wasm/main
@@ -867,7 +873,7 @@ endif
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=pca10040 -serial=rtt examples/echo
 	@$(MD5SUM) test.hex
-	$(TINYGO) build             -o test.nro -target=nintendoswitch      examples/serial
+	$(TINYGO) build             -o test.nro -target=nintendoswitch      examples/echo2
 	@$(MD5SUM) test.nro
 	$(TINYGO) build -size short -o test.hex -target=pca10040 -opt=0     ./testdata/stdlib.go
 	@$(MD5SUM) test.hex
@@ -926,6 +932,7 @@ endif
 	@cp -rp lib/musl/src/env             build/release/tinygo/lib/musl/src
 	@cp -rp lib/musl/src/errno           build/release/tinygo/lib/musl/src
 	@cp -rp lib/musl/src/exit            build/release/tinygo/lib/musl/src
+	@cp -rp lib/musl/src/fcntl           build/release/tinygo/lib/musl/src
 	@cp -rp lib/musl/src/include         build/release/tinygo/lib/musl/src
 	@cp -rp lib/musl/src/internal        build/release/tinygo/lib/musl/src
 	@cp -rp lib/musl/src/legacy          build/release/tinygo/lib/musl/src
@@ -934,6 +941,7 @@ endif
 	@cp -rp lib/musl/src/malloc          build/release/tinygo/lib/musl/src
 	@cp -rp lib/musl/src/mman            build/release/tinygo/lib/musl/src
 	@cp -rp lib/musl/src/math            build/release/tinygo/lib/musl/src
+	@cp -rp lib/musl/src/misc            build/release/tinygo/lib/musl/src
 	@cp -rp lib/musl/src/multibyte       build/release/tinygo/lib/musl/src
 	@cp -rp lib/musl/src/signal          build/release/tinygo/lib/musl/src
 	@cp -rp lib/musl/src/stdio           build/release/tinygo/lib/musl/src
@@ -941,6 +949,7 @@ endif
 	@cp -rp lib/musl/src/thread          build/release/tinygo/lib/musl/src
 	@cp -rp lib/musl/src/time            build/release/tinygo/lib/musl/src
 	@cp -rp lib/musl/src/unistd          build/release/tinygo/lib/musl/src
+	@cp -rp lib/musl/src/process         build/release/tinygo/lib/musl/src
 	@cp -rp lib/mingw-w64/mingw-w64-crt/def-include                 build/release/tinygo/lib/mingw-w64/mingw-w64-crt
 	@cp -rp lib/mingw-w64/mingw-w64-crt/lib-common/api-ms-win-crt-* build/release/tinygo/lib/mingw-w64/mingw-w64-crt/lib-common
 	@cp -rp lib/mingw-w64/mingw-w64-crt/lib-common/kernel32.def.in  build/release/tinygo/lib/mingw-w64/mingw-w64-crt/lib-common
@@ -61,6 +61,10 @@ type BuildResult struct {
 	// correctly printing test results: the import path isn't always the same as
 	// the path listed on the command line.
 	ImportPath string
+
+	// Map from path to package name. It is needed to attribute binary size to
+	// the right Go package.
+	PackagePathMap map[string]string
 }

 // packageAction is the struct that is serialized to JSON and hashed, to work as
@@ -242,6 +246,12 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 		return result, err
 	}

+	// Store which filesystem paths map to which package name.
+	result.PackagePathMap = make(map[string]string, len(lprogram.Packages))
+	for _, pkg := range lprogram.Sorted() {
+		result.PackagePathMap[pkg.OriginalDir()] = pkg.Pkg.Path()
+	}
+
 	// Create the *ssa.Program. This does not yet build the entire SSA of the
 	// program so it's pretty fast and doesn't need to be parallelized.
 	program := lprogram.LoadSSA()
@@ -690,7 +700,7 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 	for _, pkg := range lprogram.Sorted() {
 		pkg := pkg
 		for _, filename := range pkg.CFiles {
-			abspath := filepath.Join(pkg.Dir, filename)
+			abspath := filepath.Join(pkg.OriginalDir(), filename)
 			job := &compileJob{
 				description: "compile CGo file " + abspath,
 				run: func(job *compileJob) error {
@@ -812,6 +822,12 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 					return fmt.Errorf("could not modify stack sizes: %w", err)
 				}
 			}
+
+			// Apply patches of bootloader in the order they appear.
+			if len(config.Target.BootPatches) > 0 {
+				err = applyPatches(result.Executable, config.Target.BootPatches)
+			}
+
 			if config.RP2040BootPatch() {
 				// Patch the second stage bootloader CRC into the .boot2 section
 				err = patchRP2040BootCRC(result.Executable)
@@ -915,19 +931,16 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 			}

 			// Print code size if requested.
-			if config.Options.PrintSizes == "short" || config.Options.PrintSizes == "full" {
-				packagePathMap := make(map[string]string, len(lprogram.Packages))
-				for _, pkg := range lprogram.Sorted() {
-					packagePathMap[pkg.OriginalDir()] = pkg.Pkg.Path()
-				}
-				sizes, err := loadProgramSize(result.Executable, packagePathMap)
+			if config.Options.PrintSizes != "" {
+				sizes, err := loadProgramSize(result.Executable, result.PackagePathMap)
 				if err != nil {
 					return err
 				}
-				if config.Options.PrintSizes == "short" {
+				switch config.Options.PrintSizes {
+				case "short":
 					fmt.Printf("   code    data     bss |   flash     ram\n")
 					fmt.Printf("%7d %7d %7d | %7d %7d\n", sizes.Code+sizes.ROData, sizes.Data, sizes.BSS, sizes.Flash(), sizes.RAM())
-				} else {
+				case "full":
 					if !config.Debug() {
 						fmt.Println("warning: data incomplete, remove the -no-debug flag for more detail")
 					}
@@ -939,6 +952,13 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 					}
 					fmt.Printf("------------------------------- | --------------- | -------\n")
 					fmt.Printf("%7d %7d %7d %7d | %7d %7d | total\n", sizes.Code, sizes.ROData, sizes.Data, sizes.BSS, sizes.Code+sizes.ROData+sizes.Data, sizes.Data+sizes.BSS)
+				case "html":
+					const filename = "size-report.html"
+					err := writeSizeReport(sizes, filename, pkgName)
+					if err != nil {
+						return err
+					}
+					fmt.Println("Wrote size report to", filename)
 				}
 			}

@@ -1428,6 +1448,23 @@ func printStacks(calculatedStacks []string, stackSizes map[string]functionStackS
 	}
 }

+func applyPatches(executable string, bootPatches []string) (err error) {
+	for _, patch := range bootPatches {
+		switch patch {
+		case "rp2040":
+			err = patchRP2040BootCRC(executable)
+		// case "rp2350":
+		// 	err = patchRP2350BootIMAGE_DEF(executable)
+		default:
+			err = errors.New("undefined boot patch name")
+		}
+		if err != nil {
+			return fmt.Errorf("apply boot patch %q: %w", patch, err)
+		}
+	}
+	return nil
+}
+
 // RP2040 second stage bootloader CRC32 calculation
 //
 // Spec: https://datasheets.raspberrypi.org/rp2040/rp2040-datasheet.pdf
@@ -1439,7 +1476,7 @@ func patchRP2040BootCRC(executable string) error {
 	}

 	if len(bytes) != 256 {
-		return fmt.Errorf("rp2040 .boot2 section must be exactly 256 bytes")
+		return fmt.Errorf("rp2040 .boot2 section must be exactly 256 bytes, got %d", len(bytes))
 	}

 	// From the 'official' RP2040 checksum script:
@@ -1478,3 +1515,10 @@ func lock(path string) func() {

 	return func() { flock.Close() }
 }
+
+func b2u8(b bool) uint8 {
+	if b {
+		return 1
+	}
+	return 0
+}
@@ -33,6 +33,7 @@ func TestClangAttributes(t *testing.T) {
 		"k210",
 		"nintendoswitch",
 		"riscv-qemu",
+		"tkey",
 		"wasip1",
 		"wasip2",
 		"wasm",
@@ -116,6 +116,7 @@ var libMusl = Library{
 			"env/*.c",
 			"errno/*.c",
 			"exit/*.c",
+			"fcntl/*.c",
 			"internal/defsysinfo.c",
 			"internal/libc.c",
 			"internal/syscall_ret.c",
@@ -127,6 +128,7 @@ var libMusl = Library{
 			"malloc/mallocng/*.c",
 			"mman/*.c",
 			"math/*.c",
+			"misc/*.c",
 			"multibyte/*.c",
 			"signal/" + arch + "/*.s",
 			"signal/*.c",
@@ -136,12 +138,20 @@ var libMusl = Library{
 			"thread/*.c",
 			"time/*.c",
 			"unistd/*.c",
+			"process/*.c",
 		}
+
 		if arch == "arm" {
 			// These files need to be added to the start for some reason.
 			globs = append([]string{"thread/arm/*.c"}, globs...)
 		}

+		if arch != "aarch64" && arch != "mips" {
+			//aarch64 and mips have no architecture specific code, either they
+			// are not supported or don't need any?
+			globs = append([]string{"process/" + arch + "/*.s"}, globs...)
+		}
+
 		var sources []string
 		seenSources := map[string]struct{}{}
 		basepath := goenv.Get("TINYGOROOT") + "/lib/musl/src/"
@@ -34,6 +34,7 @@ var libPicolibc = Library{
 			"-D__OBSOLETE_MATH_FLOAT=1", // use old math code that doesn't expect a FPU
 			"-D__OBSOLETE_MATH_DOUBLE=0",
 			"-D_WANT_IO_C99_FORMATS",
+			"-D__PICOLIBC_ERRNO_FUNCTION=__errno_location",
 			"-nostdlibinc",
 			"-isystem", newlibDir + "/libc/include",
 			"-I" + newlibDir + "/libc/tinystdio",
@@ -0,0 +1,56 @@
+package builder
+
+import (
+	_ "embed"
+	"fmt"
+	"html/template"
+	"os"
+)
+
+//go:embed size-report.html
+var sizeReportBase string
+
+func writeSizeReport(sizes *programSize, filename, pkgName string) error {
+	tmpl, err := template.New("report").Parse(sizeReportBase)
+	if err != nil {
+		return err
+	}
+
+	f, err := os.Create(filename)
+	if err != nil {
+		return fmt.Errorf("could not open report file: %w", err)
+	}
+	defer f.Close()
+
+	// Prepare data for the report.
+	type sizeLine struct {
+		Name string
+		Size *packageSize
+	}
+	programData := []sizeLine{}
+	for _, name := range sizes.sortedPackageNames() {
+		pkgSize := sizes.Packages[name]
+		programData = append(programData, sizeLine{
+			Name: name,
+			Size: pkgSize,
+		})
+	}
+	sizeTotal := map[string]uint64{
+		"code":   sizes.Code,
+		"rodata": sizes.ROData,
+		"data":   sizes.Data,
+		"bss":    sizes.BSS,
+		"flash":  sizes.Flash(),
+	}
+
+	// Write the report.
+	err = tmpl.Execute(f, map[string]any{
+		"pkgName":   pkgName,
+		"sizes":     programData,
+		"sizeTotal": sizeTotal,
+	})
+	if err != nil {
+		return fmt.Errorf("could not create report file: %w", err)
+	}
+	return nil
+}
@@ -0,0 +1,109 @@
+<!DOCTYPE html>
+<html lang="en">
+  <head>
+    <title>Size Report for {{.pkgName}}</title>
+    <meta charset="utf-8"/>
+    <meta name="viewport" content="width=device-width, initial-scale=1">
+    <link href="https://cdn.jsdelivr.net/npm/bootstrap@5.3.3/dist/css/bootstrap.min.css" rel="stylesheet" integrity="sha384-QWTKZyjpPEjISv5WaRU9OFeRpok6YctnYmDr5pNlyT2bRjXh0JMhjY6hW+ALEwIH" crossorigin="anonymous">
+    <style>
+
+.table-vertical-border {
+  border-left: calc(var(--bs-border-width) * 2) solid currentcolor;
+}
+
+/* Hover on only the rows that are clickable. */
+.row-package:hover > * {
+  --bs-table-color-state: var(--bs-table-hover-color);
+  --bs-table-bg-state: var(--bs-table-hover-bg);
+}
+
+    </style>
+  </head>
+  <body>
+    <div class="container-xxl">
+      <h1>Size Report for {{.pkgName}}</h1>
+
+      <p>How much space is used by Go packages, C libraries, and other bits to set up the program environment.</p>
+
+      <ul>
+        <li><strong>Code</strong> is the actual program code (machine code instructions).</li>
+        <li><strong>Read-only data</strong> are read-only global variables. On most microcontrollers, these are stored in flash and do not take up any RAM.</li>
+        <li><strong>Data</strong> are writable global variables with a non-zero initializer. On microcontrollers, they are copied from flash to RAM on reset.</li>
+        <li><strong>BSS</strong> are writable global variables that are zero initialized. They do not take up any space in the binary, but do take up RAM. On microcontrollers, this area is zeroed on reset.</li>
+      </ul>
+
+      <p>The binary size consists of code, read-only data, and data. On microcontrollers, this is exactly the size of the firmware image. On other systems, there is some extra overhead: binary metadata (headers of the ELF/MachO/COFF file), debug information, exception tables, symbol names, etc. Using <code>-no-debug</code> strips most of those.</p>
+
+      <h2>Program breakdown</h2>
+
+      <p>You can click on the rows below to see which files contribute to the binary size.</p>
+
+      <div class="table-responsive">
+        <table class="table w-auto">
+          <thead>
+            <tr>
+              <th>Package</th>
+              <th class="table-vertical-border">Code</th>
+              <th>Read-only data</th>
+              <th>Data</th>
+              <th title="zero-initialized data">BSS</th>
+              <th class="table-vertical-border" style="min-width: 16em">Binary size</th>
+            </tr>
+          </thead>
+          <tbody class="table-group-divider">
+            {{range $i, $pkg := .sizes}}
+            <tr class="row-package" data-collapse=".collapse-row-{{$i}}">
+              <td>{{.Name}}</td>
+              <td class="table-vertical-border">{{.Size.Code}}</td>
+              <td>{{.Size.ROData}}</td>
+              <td>{{.Size.Data}}</td>
+              <td>{{.Size.BSS}}</td>
+              <td class="table-vertical-border" style="background: linear-gradient(to right, var(--bs-info-bg-subtle) {{.Size.FlashPercent}}%, var(--bs-table-bg) {{.Size.FlashPercent}}%)">
+                {{.Size.Flash}}
+              </td>
+            </tr>
+            {{range $filename, $sizes := .Size.Sub}}
+            <tr class="table-secondary collapse collapse-row-{{$i}}">
+              <td class="ps-4">
+                {{if eq $filename ""}}
+                  (unknown file)
+                {{else}}
+                  {{$filename}}
+                {{end}}
+              </td>
+              <td class="table-vertical-border">{{$sizes.Code}}</td>
+              <td>{{$sizes.ROData}}</td>
+              <td>{{$sizes.Data}}</td>
+              <td>{{$sizes.BSS}}</td>
+              <td class="table-vertical-border" style="background: linear-gradient(to right, var(--bs-info-bg-subtle) {{$sizes.FlashPercent}}%, var(--bs-table-bg) {{$sizes.FlashPercent}}%)">
+                {{$sizes.Flash}}
+              </td>
+            </tr>
+            {{end}}
+            {{end}}
+          </tbody>
+          <tfoot class="table-group-divider">
+            <tr>
+              <th>Total</th>
+              <td class="table-vertical-border">{{.sizeTotal.code}}</td>
+              <td>{{.sizeTotal.rodata}}</td>
+              <td>{{.sizeTotal.data}}</td>
+              <td>{{.sizeTotal.bss}}</td>
+              <td class="table-vertical-border">{{.sizeTotal.flash}}</td>
+            </tr>
+          </tfoot>
+        </table>
+      </div>
+    </div>
+    <script>
+// Make table rows toggleable to show filenames.
+for (let clickable of document.querySelectorAll('.row-package')) {
+  clickable.addEventListener('click', e => {
+    for (let row of document.querySelectorAll(clickable.dataset.collapse)) {
+      row.classList.toggle('show');
+    }
+  });
+}
+    </script>
+  </body>
+</html>
@@ -12,6 +12,7 @@ import (
 	"os"
 	"path/filepath"
 	"regexp"
+	"runtime"
 	"sort"
 	"strings"

@@ -24,7 +25,7 @@ const sizesDebug = false

 // programSize contains size statistics per package of a compiled program.
 type programSize struct {
-	Packages map[string]packageSize
+	Packages map[string]*packageSize
 	Code     uint64
 	ROData   uint64
 	Data     uint64
@@ -52,13 +53,29 @@ func (ps *programSize) RAM() uint64 {
 	return ps.Data + ps.BSS
 }

+// Return the package size information for a given package path, creating it if
+// it doesn't exist yet.
+func (ps *programSize) getPackage(path string) *packageSize {
+	if field, ok := ps.Packages[path]; ok {
+		return field
+	}
+	field := &packageSize{
+		Program: ps,
+		Sub:     map[string]*packageSize{},
+	}
+	ps.Packages[path] = field
+	return field
+}
+
 // packageSize contains the size of a package, calculated from the linked object
 // file.
 type packageSize struct {
-	Code   uint64
-	ROData uint64
-	Data   uint64
-	BSS    uint64
+	Program *programSize
+	Code    uint64
+	ROData  uint64
+	Data    uint64
+	BSS     uint64
+	Sub     map[string]*packageSize
 }

 // Flash usage in regular microcontrollers.
@@ -71,6 +88,31 @@ func (ps *packageSize) RAM() uint64 {
 	return ps.Data + ps.BSS
 }

+// Flash usage in regular microcontrollers, as a percentage of the total flash
+// usage of the program.
+func (ps *packageSize) FlashPercent() float64 {
+	return float64(ps.Flash()) / float64(ps.Program.Flash()) * 100
+}
+
+// Add a single size data point to this package.
+// This must only be called while calculating package size, not afterwards.
+func (ps *packageSize) addSize(getField func(*packageSize, bool) *uint64, filename string, size uint64, isVariable bool) {
+	if size == 0 {
+		return
+	}
+
+	// Add size for the package.
+	*getField(ps, isVariable) += size
+
+	// Add size for file inside package.
+	sub, ok := ps.Sub[filename]
+	if !ok {
+		sub = &packageSize{Program: ps.Program}
+		ps.Sub[filename] = sub
+	}
+	*getField(sub, isVariable) += size
+}
+
 // A mapping of a single chunk of code or data to a file path.
 type addressLine struct {
 	Address    uint64
@@ -194,11 +236,22 @@ func readProgramSizeFromDWARF(data *dwarf.Data, codeOffset, codeAlignment uint64
 				if !prevLineEntry.EndSequence {
 					// The chunk describes the code from prevLineEntry to
 					// lineEntry.
+					path := prevLineEntry.File.Name
+					if runtime.GOOS == "windows" {
+						// Work around a Clang bug on Windows:
+						// https://github.com/llvm/llvm-project/issues/117317
+						path = strings.ReplaceAll(path, "\\\\", "\\")
+
+						// wasi-libc likes to use forward slashes, but we
+						// canonicalize everything to use backwards slashes as
+						// is common on Windows.
+						path = strings.ReplaceAll(path, "/", "\\")
+					}
 					line := addressLine{
 						Address: prevLineEntry.Address + codeOffset,
 						Length:  lineEntry.Address - prevLineEntry.Address,
 						Align:   codeAlignment,
-						File:    prevLineEntry.File.Name,
+						File:    path,
 					}
 					if line.Length != 0 {
 						addresses = append(addresses, line)
@@ -773,49 +826,40 @@ func loadProgramSize(path string, packagePathMap map[string]string) (*programSiz

 	// Now finally determine the binary/RAM size usage per package by going
 	// through each allocated section.
-	sizes := make(map[string]packageSize)
+	sizes := make(map[string]*packageSize)
+	program := &programSize{
+		Packages: sizes,
+	}
 	for _, section := range sections {
 		switch section.Type {
 		case memoryCode:
-			readSection(section, addresses, func(path string, size uint64, isVariable bool) {
-				field := sizes[path]
+			readSection(section, addresses, program, func(ps *packageSize, isVariable bool) *uint64 {
 				if isVariable {
-					field.ROData += size
-				} else {
-					field.Code += size
+					return &ps.ROData
 				}
-				sizes[path] = field
+				return &ps.Code
 			}, packagePathMap)
 		case memoryROData:
-			readSection(section, addresses, func(path string, size uint64, isVariable bool) {
-				field := sizes[path]
-				field.ROData += size
-				sizes[path] = field
+			readSection(section, addresses, program, func(ps *packageSize, isVariable bool) *uint64 {
+				return &ps.ROData
 			}, packagePathMap)
 		case memoryData:
-			readSection(section, addresses, func(path string, size uint64, isVariable bool) {
-				field := sizes[path]
-				field.Data += size
-				sizes[path] = field
+			readSection(section, addresses, program, func(ps *packageSize, isVariable bool) *uint64 {
+				return &ps.Data
 			}, packagePathMap)
 		case memoryBSS:
-			readSection(section, addresses, func(path string, size uint64, isVariable bool) {
-				field := sizes[path]
-				field.BSS += size
-				sizes[path] = field
+			readSection(section, addresses, program, func(ps *packageSize, isVariable bool) *uint64 {
+				return &ps.BSS
 			}, packagePathMap)
 		case memoryStack:
 			// We store the C stack as a pseudo-package.
-			sizes["C stack"] = packageSize{
-				BSS: section.Size,
-			}
+			program.getPackage("C stack").addSize(func(ps *packageSize, isVariable bool) *uint64 {
+				return &ps.BSS
+			}, "", section.Size, false)
 		}
 	}

 	// ...and summarize the results.
-	program := &programSize{
-		Packages: sizes,
-	}
 	for _, pkg := range sizes {
 		program.Code += pkg.Code
 		program.ROData += pkg.ROData
@@ -826,8 +870,8 @@ func loadProgramSize(path string, packagePathMap map[string]string) (*programSiz
 }

 // readSection determines for each byte in this section to which package it
-// belongs. It reports this usage through the addSize callback.
-func readSection(section memorySection, addresses []addressLine, addSize func(string, uint64, bool), packagePathMap map[string]string) {
+// belongs.
+func readSection(section memorySection, addresses []addressLine, program *programSize, getField func(*packageSize, bool) *uint64, packagePathMap map[string]string) {
 	// The addr variable tracks at which address we are while going through this
 	// section. We start at the beginning.
 	addr := section.Address
@@ -849,9 +893,9 @@ func readSection(section memorySection, addresses []addressLine, addSize func(st
 			addrAligned := (addr + line.Align - 1) &^ (line.Align - 1)
 			if line.Align > 1 && addrAligned >= line.Address {
 				// It is, assume that's what causes the gap.
-				addSize("(padding)", line.Address-addr, true)
+				program.getPackage("(padding)").addSize(getField, "", line.Address-addr, true)
 			} else {
-				addSize("(unknown)", line.Address-addr, false)
+				program.getPackage("(unknown)").addSize(getField, "", line.Address-addr, false)
 				if sizesDebug {
 					fmt.Printf("%08x..%08x %5d:  unknown (gap), alignment=%d\n", addr, line.Address, line.Address-addr, line.Align)
 				}
@@ -873,7 +917,8 @@ func readSection(section memorySection, addresses []addressLine, addSize func(st
 			length = line.Length - (addr - line.Address)
 		}
 		// Finally, mark this chunk of memory as used by the given package.
-		addSize(findPackagePath(line.File, packagePathMap), length, line.IsVariable)
+		packagePath, filename := findPackagePath(line.File, packagePathMap)
+		program.getPackage(packagePath).addSize(getField, filename, length, line.IsVariable)
 		addr = line.Address + line.Length
 	}
 	if addr < sectionEnd {
@@ -882,9 +927,9 @@ func readSection(section memorySection, addresses []addressLine, addSize func(st
 		if section.Align > 1 && addrAligned >= sectionEnd {
 			// The gap is caused by the section alignment.
 			// For example, if a .rodata section ends with a non-aligned string.
-			addSize("(padding)", sectionEnd-addr, true)
+			program.getPackage("(padding)").addSize(getField, "", sectionEnd-addr, true)
 		} else {
-			addSize("(unknown)", sectionEnd-addr, false)
+			program.getPackage("(unknown)").addSize(getField, "", sectionEnd-addr, false)
 			if sizesDebug {
 				fmt.Printf("%08x..%08x %5d:  unknown (end), alignment=%d\n", addr, sectionEnd, sectionEnd-addr, section.Align)
 			}
@@ -894,17 +939,25 @@ func readSection(section memorySection, addresses []addressLine, addSize func(st

 // findPackagePath returns the Go package (or a pseudo package) for the given
 // path. It uses some heuristics, for example for some C libraries.
-func findPackagePath(path string, packagePathMap map[string]string) string {
+func findPackagePath(path string, packagePathMap map[string]string) (packagePath, filename string) {
 	// Check whether this path is part of one of the compiled packages.
 	packagePath, ok := packagePathMap[filepath.Dir(path)]
-	if !ok {
+	if ok {
+		// Directory is known as a Go package.
+		// Add the file itself as well.
+		filename = filepath.Base(path)
+	} else {
 		if strings.HasPrefix(path, filepath.Join(goenv.Get("TINYGOROOT"), "lib")) {
 			// Emit C libraries (in the lib subdirectory of TinyGo) as a single
-			// package, with a "C" prefix. For example: "C compiler-rt" for the
-			// compiler runtime library from LLVM.
-			packagePath = "C " + strings.Split(strings.TrimPrefix(path, filepath.Join(goenv.Get("TINYGOROOT"), "lib")), string(os.PathSeparator))[1]
-		} else if strings.HasPrefix(path, filepath.Join(goenv.Get("TINYGOROOT"), "llvm-project")) {
+			// package, with a "C" prefix. For example: "C picolibc" for the
+			// baremetal libc.
+			libPath := strings.TrimPrefix(path, filepath.Join(goenv.Get("TINYGOROOT"), "lib")+string(os.PathSeparator))
+			parts := strings.SplitN(libPath, string(os.PathSeparator), 2)
+			packagePath = "C " + parts[0]
+			filename = parts[1]
+		} else if prefix := filepath.Join(goenv.Get("TINYGOROOT"), "llvm-project", "compiler-rt"); strings.HasPrefix(path, prefix) {
 			packagePath = "C compiler-rt"
+			filename = strings.TrimPrefix(path, prefix+string(os.PathSeparator))
 		} else if packageSymbolRegexp.MatchString(path) {
 			// Parse symbol names like main$alloc or runtime$string.
 			packagePath = path[:strings.LastIndex(path, "$")]
@@ -927,9 +980,11 @@ func findPackagePath(path string, packagePathMap map[string]string) string {
 			// fixed in the compiler.
 			packagePath = "-"
 		} else {
-			// This is some other path. Not sure what it is, so just emit its directory.
-			packagePath = filepath.Dir(path) // fallback
+			// This is some other path. Not sure what it is, so just emit its
+			// directory as a fallback.
+			packagePath = filepath.Dir(path)
+			filename = filepath.Base(path)
 		}
 	}
-	return packagePath
+	return
 }
@@ -1,6 +1,7 @@
 package builder

 import (
+	"regexp"
 	"runtime"
 	"testing"
 	"time"
@@ -41,9 +42,9 @@ func TestBinarySize(t *testing.T) {
 	// This is a small number of very diverse targets that we want to test.
 	tests := []sizeTest{
 		// microcontrollers
-		{"hifive1b", "examples/echo", 4568, 280, 0, 2268},
-		{"microbit", "examples/serial", 2868, 388, 8, 2272},
-		{"wioterminal", "examples/pininterrupt", 6104, 1484, 116, 6832},
+		{"hifive1b", "examples/echo", 4600, 280, 0, 2268},
+		{"microbit", "examples/serial", 2908, 388, 8, 2272},
+		{"wioterminal", "examples/pininterrupt", 6140, 1484, 116, 6824},

 		// TODO: also check wasm. Right now this is difficult, because
 		// wasm binaries are run through wasm-opt and therefore the
@@ -55,26 +56,7 @@ func TestBinarySize(t *testing.T) {
 			t.Parallel()

 			// Build the binary.
-			options := compileopts.Options{
-				Target:        tc.target,
-				Opt:           "z",
-				Semaphore:     sema,
-				InterpTimeout: 60 * time.Second,
-				Debug:         true,
-				VerifyIR:      true,
-			}
-			target, err := compileopts.LoadTarget(&options)
-			if err != nil {
-				t.Fatal("could not load target:", err)
-			}
-			config := &compileopts.Config{
-				Options: &options,
-				Target:  target,
-			}
-			result, err := Build(tc.path, "", t.TempDir(), config)
-			if err != nil {
-				t.Fatal("could not build:", err)
-			}
+			result := buildBinary(t, tc.target, tc.path)

 			// Check whether the size of the binary matches the expected size.
 			sizes, err := loadProgramSize(result.Executable, nil)
@@ -90,3 +72,69 @@ func TestBinarySize(t *testing.T) {
 		})
 	}
 }
+
+// Check that the -size=full flag attributes binary size to the correct package
+// without filesystem paths and things like that.
+func TestSizeFull(t *testing.T) {
+	tests := []string{
+		"microbit",
+		"wasip1",
+	}
+
+	libMatch := regexp.MustCompile(`^C [a-z -]+$`) // example: "C interrupt vector"
+	pkgMatch := regexp.MustCompile(`^[a-z/]+$`)    // example: "internal/task"
+
+	for _, target := range tests {
+		target := target
+		t.Run(target, func(t *testing.T) {
+			t.Parallel()
+
+			// Build the binary.
+			result := buildBinary(t, target, "examples/serial")
+
+			// Check whether the binary doesn't contain any unexpected package
+			// names.
+			sizes, err := loadProgramSize(result.Executable, result.PackagePathMap)
+			if err != nil {
+				t.Fatal("could not read program size:", err)
+			}
+			for _, pkg := range sizes.sortedPackageNames() {
+				if pkg == "(padding)" || pkg == "(unknown)" {
+					// TODO: correctly attribute all unknown binary size.
+					continue
+				}
+				if libMatch.MatchString(pkg) {
+					continue
+				}
+				if pkgMatch.MatchString(pkg) {
+					continue
+				}
+				t.Error("unexpected package name in size output:", pkg)
+			}
+		})
+	}
+}
+
+func buildBinary(t *testing.T, targetString, pkgName string) BuildResult {
+	options := compileopts.Options{
+		Target:        targetString,
+		Opt:           "z",
+		Semaphore:     sema,
+		InterpTimeout: 60 * time.Second,
+		Debug:         true,
+		VerifyIR:      true,
+	}
+	target, err := compileopts.LoadTarget(&options)
+	if err != nil {
+		t.Fatal("could not load target:", err)
+	}
+	config := &compileopts.Config{
+		Options: &options,
+		Target:  target,
+	}
+	result, err := Build(pkgName, "", t.TempDir(), config)
+	if err != nil {
+		t.Fatal("could not build:", err)
+	}
+	return result
+}
@@ -32,10 +32,26 @@ func runCCompiler(flags ...string) error {
 	cmd.Stderr = os.Stderr

 	// Make sure the command doesn't use any environmental variables.
-	// Most importantly, it should not use C_INCLUDE_PATH and the like. But
-	// removing all environmental variables also works.
+	// Most importantly, it should not use C_INCLUDE_PATH and the like.
 	cmd.Env = []string{}

+	// Let some environment variables through. One important one is the
+	// temporary directory, especially on Windows it looks like Clang breaks if
+	// the temporary directory has not been set.
+	// See: https://github.com/tinygo-org/tinygo/issues/4557
+	// Also see: https://github.com/llvm/llvm-project/blob/release/18.x/llvm/lib/Support/Unix/Path.inc#L1435
+	for _, env := range os.Environ() {
+		// We could parse the key and look it up in a map, but since there are
+		// only a few keys iterating through them is easier and maybe even
+		// faster.
+		for _, prefix := range []string{"TMPDIR=", "TMP=", "TEMP=", "TEMPDIR="} {
+			if strings.HasPrefix(env, prefix) {
+				cmd.Env = append(cmd.Env, env)
+				break
+			}
+		}
+	}
+
 	return cmd.Run()
 }

@@ -18,6 +18,7 @@ import (
 	"go/scanner"
 	"go/token"
 	"path/filepath"
+	"sort"
 	"strconv"
 	"strings"

@@ -42,6 +43,7 @@ type cgoPackage struct {
 	fset            *token.FileSet
 	tokenFiles      map[string]*token.File
 	definedGlobally map[string]ast.Node
+	noescapingFuncs map[string]*noescapingFunc // #cgo noescape lines
 	anonDecls       map[interface{}]string
 	cflags          []string // CFlags from #cgo lines
 	ldflags         []string // LDFlags from #cgo lines
@@ -80,6 +82,13 @@ type bitfieldInfo struct {
 	endBit   int64 // may be 0 meaning "until the end of the field"
 }

+// Information about a #cgo noescape line in the source code.
+type noescapingFunc struct {
+	name string
+	pos  token.Pos
+	used bool // true if used somewhere in the source (for proper error reporting)
+}
+
 // cgoAliases list type aliases between Go and C, for types that are equivalent
 // in both languages. See addTypeAliases.
 var cgoAliases = map[string]string{
@@ -138,7 +147,8 @@ typedef unsigned long long  _Cgo_ulonglong;
 // first.
 // These functions will be modified to get a "C." prefix, so the source below
 // doesn't reflect the final AST.
-const generatedGoFilePrefix = `
+const generatedGoFilePrefixBase = `
+import "syscall"
 import "unsafe"

 var _ unsafe.Pointer
@@ -169,6 +179,74 @@ func __CBytes([]byte) unsafe.Pointer
 func CBytes(b []byte) unsafe.Pointer {
 	return C.__CBytes(b)
 }
+
+//go:linkname C.__get_errno_num runtime.cgo_errno
+func __get_errno_num() uintptr
+`
+
+const generatedGoFilePrefixOther = generatedGoFilePrefixBase + `
+func __get_errno() error {
+	return syscall.Errno(C.__get_errno_num())
+}
+`
+
+// Windows uses fake errno values in the syscall package.
+// See for example: https://github.com/golang/go/issues/23468
+// TinyGo uses mingw-w64 though, which does have defined errno values. Since the
+// syscall package is the standard library one we can't change it, but we can
+// map the errno values to match the values in the syscall package.
+// Source of the errno values: lib/mingw-w64/mingw-w64-headers/crt/errno.h
+const generatedGoFilePrefixWindows = generatedGoFilePrefixBase + `
+var __errno_mapping = [...]syscall.Errno{
+	1:  syscall.EPERM,
+	2:  syscall.ENOENT,
+	3:  syscall.ESRCH,
+	4:  syscall.EINTR,
+	5:  syscall.EIO,
+	6:  syscall.ENXIO,
+	7:  syscall.E2BIG,
+	8:  syscall.ENOEXEC,
+	9:  syscall.EBADF,
+	10: syscall.ECHILD,
+	11: syscall.EAGAIN,
+	12: syscall.ENOMEM,
+	13: syscall.EACCES,
+	14: syscall.EFAULT,
+	16: syscall.EBUSY,
+	17: syscall.EEXIST,
+	18: syscall.EXDEV,
+	19: syscall.ENODEV,
+	20: syscall.ENOTDIR,
+	21: syscall.EISDIR,
+	22: syscall.EINVAL,
+	23: syscall.ENFILE,
+	24: syscall.EMFILE,
+	25: syscall.ENOTTY,
+	27: syscall.EFBIG,
+	28: syscall.ENOSPC,
+	29: syscall.ESPIPE,
+	30: syscall.EROFS,
+	31: syscall.EMLINK,
+	32: syscall.EPIPE,
+	33: syscall.EDOM,
+	34: syscall.ERANGE,
+	36: syscall.EDEADLK,
+	38: syscall.ENAMETOOLONG,
+	39: syscall.ENOLCK,
+	40: syscall.ENOSYS,
+	41: syscall.ENOTEMPTY,
+	42: syscall.EILSEQ,
+}
+
+func __get_errno() error {
+	num := C.__get_errno_num()
+	if num < uintptr(len(__errno_mapping)) {
+		if mapped := __errno_mapping[num]; mapped != 0 {
+			return mapped
+		}
+	}
+	return syscall.Errno(num)
+}
 `

 // Process extracts `import "C"` statements from the AST, parses the comment
@@ -178,7 +256,7 @@ func CBytes(b []byte) unsafe.Pointer {
 // functions), the CFLAGS and LDFLAGS found in #cgo lines, and a map of file
 // hashes of the accessed C header files. If there is one or more error, it
 // returns these in the []error slice but still modifies the AST.
-func Process(files []*ast.File, dir, importPath string, fset *token.FileSet, cflags []string) ([]*ast.File, []string, []string, []string, map[string][]byte, []error) {
+func Process(files []*ast.File, dir, importPath string, fset *token.FileSet, cflags []string, goos string) ([]*ast.File, []string, []string, []string, map[string][]byte, []error) {
 	p := &cgoPackage{
 		packageName:     files[0].Name.Name,
 		currentDir:      dir,
@@ -186,6 +264,7 @@ func Process(files []*ast.File, dir, importPath string, fset *token.FileSet, cfl
 		fset:            fset,
 		tokenFiles:      map[string]*token.File{},
 		definedGlobally: map[string]ast.Node{},
+		noescapingFuncs: map[string]*noescapingFunc{},
 		anonDecls:       map[interface{}]string{},
 		visitedFiles:    map[string][]byte{},
 	}
@@ -210,7 +289,12 @@ func Process(files []*ast.File, dir, importPath string, fset *token.FileSet, cfl
 	// Construct a new in-memory AST for CGo declarations of this package.
 	// The first part is written as Go code that is then parsed, but more code
 	// is added later to the AST to declare functions, globals, etc.
-	goCode := "package " + files[0].Name.Name + "\n\n" + generatedGoFilePrefix
+	goCode := "package " + files[0].Name.Name + "\n\n"
+	if goos == "windows" {
+		goCode += generatedGoFilePrefixWindows
+	} else {
+		goCode += generatedGoFilePrefixOther
+	}
 	p.generated, err = parser.ParseFile(fset, dir+"/!cgo.go", goCode, parser.ParseComments)
 	if err != nil {
 		// This is always a bug in the cgo package.
@@ -225,7 +309,7 @@ func Process(files []*ast.File, dir, importPath string, fset *token.FileSet, cfl
 		switch decl := decl.(type) {
 		case *ast.FuncDecl:
 			switch decl.Name.Name {
-			case "CString", "GoString", "GoStringN", "__GoStringN", "GoBytes", "__GoBytes", "CBytes", "__CBytes":
+			case "CString", "GoString", "GoStringN", "__GoStringN", "GoBytes", "__GoBytes", "CBytes", "__CBytes", "__get_errno_num", "__get_errno", "__errno_mapping":
 				// Adjust the name to have a "C." prefix so it is correctly
 				// resolved.
 				decl.Name.Name = "C." + decl.Name.Name
@@ -344,6 +428,22 @@ func Process(files []*ast.File, dir, importPath string, fset *token.FileSet, cfl
 		})
 	}

+	// Show an error when a #cgo noescape line isn't used in practice.
+	// This matches upstream Go. I think the goal is to avoid issues with
+	// misspelled function names, which seems very useful.
+	var unusedNoescapeLines []*noescapingFunc
+	for _, value := range p.noescapingFuncs {
+		if !value.used {
+			unusedNoescapeLines = append(unusedNoescapeLines, value)
+		}
+	}
+	sort.SliceStable(unusedNoescapeLines, func(i, j int) bool {
+		return unusedNoescapeLines[i].pos < unusedNoescapeLines[j].pos
+	})
+	for _, value := range unusedNoescapeLines {
+		p.addError(value.pos, fmt.Sprintf("function %#v in #cgo noescape line is not used", value.name))
+	}
+
 	// Print the newly generated in-memory AST, for debugging.
 	//ast.Print(fset, p.generated)

@@ -409,6 +509,33 @@ func (p *cgoPackage) parseCGoPreprocessorLines(text string, pos token.Pos) strin
 		}
 		text = text[:lineStart] + string(spaces) + text[lineEnd:]

+		allFields := strings.Fields(line[4:])
+		switch allFields[0] {
+		case "noescape":
+			// The code indicates that pointer parameters will not be captured
+			// by the called C function.
+			if len(allFields) < 2 {
+				p.addErrorAfter(pos, text[:lineStart], "missing function name in #cgo noescape line")
+				continue
+			}
+			if len(allFields) > 2 {
+				p.addErrorAfter(pos, text[:lineStart], "multiple function names in #cgo noescape line")
+				continue
+			}
+			name := allFields[1]
+			p.noescapingFuncs[name] = &noescapingFunc{
+				name: name,
+				pos:  pos,
+				used: false,
+			}
+			continue
+		case "nocallback":
+			// We don't do anything special when calling a C function, so there
+			// appears to be no optimization that we can do here.
+			// Accept, but ignore the parameter for compatibility.
+			continue
+		}
+
 		// Get the text before the colon in the #cgo directive.
 		colon := strings.IndexByte(line, ':')
 		if colon < 0 {
@@ -1148,7 +1275,7 @@ func (f *cgoFile) getASTDeclName(name string, found clangCursor, iscall bool) st
 	if alias := cgoAliases["C."+name]; alias != "" {
 		return alias
 	}
-	node := f.getASTDeclNode(name, found, iscall)
+	node := f.getASTDeclNode(name, found)
 	if node, ok := node.(*ast.FuncDecl); ok {
 		if !iscall {
 			return node.Name.Name + "$funcaddr"
@@ -1160,7 +1287,7 @@ func (f *cgoFile) getASTDeclName(name string, found clangCursor, iscall bool) st

 // getASTDeclNode will declare the given C AST node (if not already defined) and
 // returns it.
-func (f *cgoFile) getASTDeclNode(name string, found clangCursor, iscall bool) ast.Node {
+func (f *cgoFile) getASTDeclNode(name string, found clangCursor) ast.Node {
 	if node, ok := f.defined[name]; ok {
 		// Declaration was found in the current file, so return it immediately.
 		return node
@@ -1279,6 +1406,45 @@ extern __typeof(%s) %s __attribute__((alias(%#v)));
 // separate namespace (no _Cgo_ hacks like in gc).
 func (f *cgoFile) walker(cursor *astutil.Cursor, names map[string]clangCursor) bool {
 	switch node := cursor.Node().(type) {
+	case *ast.AssignStmt:
+		// An assign statement could be something like this:
+		//
+		//   val, errno := C.some_func()
+		//
+		// Check whether it looks like that, and if so, read the errno value and
+		// return it as the second return value. The call will be transformed
+		// into something like this:
+		//
+		//   val, errno := C.some_func(), C.__get_errno()
+		if len(node.Lhs) != 2 || len(node.Rhs) != 1 {
+			return true
+		}
+		rhs, ok := node.Rhs[0].(*ast.CallExpr)
+		if !ok {
+			return true
+		}
+		fun, ok := rhs.Fun.(*ast.SelectorExpr)
+		if !ok {
+			return true
+		}
+		x, ok := fun.X.(*ast.Ident)
+		if !ok {
+			return true
+		}
+		if found, ok := names[fun.Sel.Name]; ok && x.Name == "C" {
+			// Replace "C"."some_func" into "C.somefunc".
+			rhs.Fun = &ast.Ident{
+				NamePos: x.NamePos,
+				Name:    f.getASTDeclName(fun.Sel.Name, found, true),
+			}
+			// Add the errno value as the second value in the statement.
+			node.Rhs = append(node.Rhs, &ast.CallExpr{
+				Fun: &ast.Ident{
+					NamePos: node.Lhs[1].End(),
+					Name:    "C.__get_errno",
+				},
+			})
+		}
 	case *ast.CallExpr:
 		fun, ok := node.Fun.(*ast.SelectorExpr)
 		if !ok {
@@ -56,7 +56,7 @@ func TestCGo(t *testing.T) {
 			}

 			// Process the AST with CGo.
-			cgoFiles, _, _, _, _, cgoErrors := Process([]*ast.File{f}, "testdata", "main", fset, cflags)
+			cgoFiles, _, _, _, _, cgoErrors := Process([]*ast.File{f}, "testdata", "main", fset, cflags, "linux")

 			// Check the AST for type errors.
 			var typecheckErrors []error
@@ -64,7 +64,7 @@ func TestCGo(t *testing.T) {
 				Error: func(err error) {
 					typecheckErrors = append(typecheckErrors, err)
 				},
-				Importer: simpleImporter{},
+				Importer: newSimpleImporter(),
 				Sizes:    types.SizesFor("gccgo", "arm"),
 			}
 			_, err = config.Check("", fset, append([]*ast.File{f}, cgoFiles...), nil)
@@ -202,14 +202,33 @@ func Test_cgoPackage_isEquivalentAST(t *testing.T) {
 }

 // simpleImporter implements the types.Importer interface, but only allows
-// importing the unsafe package.
+// importing the syscall and unsafe packages.
 type simpleImporter struct {
+	syscallPkg *types.Package
+}
+
+func newSimpleImporter() *simpleImporter {
+	i := &simpleImporter{}
+
+	// Implement a dummy syscall package with the Errno type.
+	i.syscallPkg = types.NewPackage("syscall", "syscall")
+	obj := types.NewTypeName(token.NoPos, i.syscallPkg, "Errno", nil)
+	named := types.NewNamed(obj, nil, nil)
+	i.syscallPkg.Scope().Insert(obj)
+	named.SetUnderlying(types.Typ[types.Uintptr])
+	sig := types.NewSignatureType(nil, nil, nil, types.NewTuple(), types.NewTuple(types.NewParam(token.NoPos, i.syscallPkg, "", types.Typ[types.String])), false)
+	named.AddMethod(types.NewFunc(token.NoPos, i.syscallPkg, "Error", sig))
+	i.syscallPkg.MarkComplete()
+
+	return i
 }

 // Import implements the Importer interface. For testing usage only: it only
 // supports importing the unsafe package.
-func (i simpleImporter) Import(path string) (*types.Package, error) {
+func (i *simpleImporter) Import(path string) (*types.Package, error) {
 	switch path {
+	case "syscall":
+		return i.syscallPkg, nil
 	case "unsafe":
 		return types.Unsafe, nil
 	default:
@@ -54,8 +54,72 @@ func init() {
 }

 // parseConst parses the given string as a C constant.
-func parseConst(pos token.Pos, fset *token.FileSet, value string) (ast.Expr, *scanner.Error) {
-	t := newTokenizer(pos, fset, value)
+func parseConst(pos token.Pos, fset *token.FileSet, value string, params []ast.Expr, callerPos token.Pos, f *cgoFile) (ast.Expr, *scanner.Error) {
+	t := newTokenizer(pos, fset, value, f)
+
+	// If params is non-nil (could be a zero length slice), this const is
+	// actually a function-call like expression from another macro.
+	// This means we have to parse a string like "(a, b) (a+b)".
+	// We do this by parsing the parameters at the start and then treating the
+	// following like a normal constant expression.
+	if params != nil {
+		// Parse opening paren.
+		if t.curToken != token.LPAREN {
+			return nil, unexpectedToken(t, token.LPAREN)
+		}
+		t.Next()
+
+		// Parse parameters (identifiers) and closing paren.
+		var paramIdents []string
+		for i := 0; ; i++ {
+			if i == 0 && t.curToken == token.RPAREN {
+				// No parameters, break early.
+				t.Next()
+				break
+			}
+
+			// Read the parameter name.
+			if t.curToken != token.IDENT {
+				return nil, unexpectedToken(t, token.IDENT)
+			}
+			paramIdents = append(paramIdents, t.curValue)
+			t.Next()
+
+			// Read the next token: either a continuation (comma) or end of list
+			// (rparen).
+			if t.curToken == token.RPAREN {
+				// End of parameter list.
+				t.Next()
+				break
+			} else if t.curToken == token.COMMA {
+				// Comma, so there will be another parameter name.
+				t.Next()
+			} else {
+				return nil, &scanner.Error{
+					Pos: t.fset.Position(t.curPos),
+					Msg: "unexpected token " + t.curToken.String() + " inside macro parameters, expected ',' or ')'",
+				}
+			}
+		}
+
+		// Report an error if there is a mismatch in parameter length.
+		// The error is reported at the location of the closing paren from the
+		// caller location.
+		if len(params) != len(paramIdents) {
+			return nil, &scanner.Error{
+				Pos: t.fset.Position(callerPos),
+				Msg: fmt.Sprintf("unexpected number of parameters: expected %d, got %d", len(paramIdents), len(params)),
+			}
+		}
+
+		// Assign values to the parameters.
+		// These parameter names are closer in 'scope' than other identifiers so
+		// will be used first when parsing an identifier.
+		for i, name := range paramIdents {
+			t.params[name] = params[i]
+		}
+	}
+
 	expr, err := parseConstExpr(t, precedenceLowest)
 	t.Next()
 	if t.curToken != token.EOF {
@@ -96,6 +160,68 @@ func parseConstExpr(t *tokenizer, precedence int) (ast.Expr, *scanner.Error) {
 }

 func parseIdent(t *tokenizer) (ast.Expr, *scanner.Error) {
+	// If the identifier is one of the parameters of this function-like macro,
+	// use the parameter value.
+	if val, ok := t.params[t.curValue]; ok {
+		return val, nil
+	}
+
+	if t.f != nil {
+		// Check whether this identifier is actually a macro "call" with
+		// parameters. In that case, we should parse the parameters and pass it
+		// on to a new invocation of parseConst.
+		if t.peekToken == token.LPAREN {
+			if cursor, ok := t.f.names[t.curValue]; ok && t.f.isFunctionLikeMacro(cursor) {
+				// We know the current and peek tokens (the peek one is the '('
+				// token). So skip ahead until the current token is the first
+				// unknown token.
+				t.Next()
+				t.Next()
+
+				// Parse the list of parameters until ')' (rparen) is found.
+				params := []ast.Expr{}
+				for i := 0; ; i++ {
+					if i == 0 && t.curToken == token.RPAREN {
+						break
+					}
+					x, err := parseConstExpr(t, precedenceLowest)
+					if err != nil {
+						return nil, err
+					}
+					params = append(params, x)
+					t.Next()
+					if t.curToken == token.COMMA {
+						t.Next()
+					} else if t.curToken == token.RPAREN {
+						break
+					} else {
+						return nil, &scanner.Error{
+							Pos: t.fset.Position(t.curPos),
+							Msg: "unexpected token " + t.curToken.String() + ", ',' or ')'",
+						}
+					}
+				}
+
+				// Evaluate the macro value and use it as the identifier value.
+				rparen := t.curPos
+				pos, text := t.f.getMacro(cursor)
+				return parseConst(pos, t.fset, text, params, rparen, t.f)
+			}
+		}
+
+		// Normally the name is something defined in the file (like another
+		// macro) which we get the declaration from using getASTDeclName.
+		// This ensures that names that are only referenced inside a macro are
+		// still getting defined.
+		if cursor, ok := t.f.names[t.curValue]; ok {
+			return &ast.Ident{
+				NamePos: t.curPos,
+				Name:    t.f.getASTDeclName(t.curValue, cursor, false),
+			}, nil
+		}
+	}
+
+	// t.f is nil during testing. This is a fallback.
 	return &ast.Ident{
 		NamePos: t.curPos,
 		Name:    "C." + t.curValue,
@@ -164,21 +290,25 @@ func unexpectedToken(t *tokenizer, expected token.Token) *scanner.Error {

 // tokenizer reads C source code and converts it to Go tokens.
 type tokenizer struct {
+	f                   *cgoFile
 	curPos, peekPos     token.Pos
 	fset                *token.FileSet
 	curToken, peekToken token.Token
 	curValue, peekValue string
 	buf                 string
+	params              map[string]ast.Expr
 }

 // newTokenizer initializes a new tokenizer, positioned at the first token in
 // the string.
-func newTokenizer(start token.Pos, fset *token.FileSet, buf string) *tokenizer {
+func newTokenizer(start token.Pos, fset *token.FileSet, buf string, f *cgoFile) *tokenizer {
 	t := &tokenizer{
+		f:         f,
 		peekPos:   start,
 		fset:      fset,
 		buf:       buf,
 		peekToken: token.ILLEGAL,
+		params:    make(map[string]ast.Expr),
 	}
 	// Parse the first two tokens (cur and peek).
 	t.Next()
@@ -230,7 +360,7 @@ func (t *tokenizer) Next() {
 			t.peekValue = t.buf[:2]
 			t.buf = t.buf[2:]
 			return
-		case c == '(' || c == ')' || c == '+' || c == '-' || c == '*' || c == '/' || c == '%' || c == '&' || c == '|' || c == '^':
+		case c == '(' || c == ')' || c == ',' || c == '+' || c == '-' || c == '*' || c == '/' || c == '%' || c == '&' || c == '|' || c == '^':
 			// Single-character tokens.
 			// TODO: ++ (increment) and -- (decrement) operators.
 			switch c {
@@ -238,6 +368,8 @@ func (t *tokenizer) Next() {
 				t.peekToken = token.LPAREN
 			case ')':
 				t.peekToken = token.RPAREN
+			case ',':
+				t.peekToken = token.COMMA
 			case '+':
 				t.peekToken = token.ADD
 			case '-':
@@ -59,7 +59,7 @@ func TestParseConst(t *testing.T) {
 	} {
 		fset := token.NewFileSet()
 		startPos := fset.AddFile("", -1, 1000).Pos(0)
-		expr, err := parseConst(startPos, fset, tc.C)
+		expr, err := parseConst(startPos, fset, tc.C, nil, token.NoPos, nil)
 		s := "<invalid>"
 		if err != nil {
 			if !strings.HasPrefix(tc.Go, "error: ") {
@@ -63,10 +63,24 @@ long long tinygo_clang_getEnumConstantDeclValue(GoCXCursor c);
 CXType tinygo_clang_getEnumDeclIntegerType(GoCXCursor c);
 unsigned tinygo_clang_Cursor_isAnonymous(GoCXCursor c);
 unsigned tinygo_clang_Cursor_isBitField(GoCXCursor c);
+unsigned tinygo_clang_Cursor_isMacroFunctionLike(GoCXCursor c);

+// Fix some warnings on Windows ARM. Without the __declspec(dllexport), it gives warnings like this:
+//     In file included from _cgo_export.c:4:
+//     cgo-gcc-export-header-prolog:49:34: warning: redeclaration of 'tinygo_clang_globals_visitor' should not add 'dllexport' attribute [-Wdll-attribute-on-redeclaration]
+//     libclang.go:68:5: note: previous declaration is here
+// See: https://github.com/golang/go/issues/49721
+#if defined(_WIN32)
+#define CGO_DECL // __declspec(dllexport)
+#else
+#define CGO_DECL
+#endif
+
+CGO_DECL
 int tinygo_clang_globals_visitor(GoCXCursor c, GoCXCursor parent, CXClientData client_data);
+CGO_DECL
 int tinygo_clang_struct_visitor(GoCXCursor c, GoCXCursor parent, CXClientData client_data);
-int tinygo_clang_enum_visitor(GoCXCursor c, GoCXCursor parent, CXClientData client_data);
+CGO_DECL
 void tinygo_clang_inclusion_visitor(CXFile included_file, CXSourceLocation *inclusion_stack, unsigned include_len, CXClientData client_data);
 */
 import "C"
@@ -255,10 +269,18 @@ func (f *cgoFile) createASTNode(name string, c clangCursor) (ast.Node, any) {
 				},
 			},
 		}
+		var doc []string
 		if C.clang_isFunctionTypeVariadic(cursorType) != 0 {
+			doc = append(doc, "//go:variadic")
+		}
+		if _, ok := f.noescapingFuncs[name]; ok {
+			doc = append(doc, "//go:noescape")
+			f.noescapingFuncs[name].used = true
+		}
+		if len(doc) != 0 {
 			decl.Doc.List = append(decl.Doc.List, &ast.Comment{
 				Slash: pos - 1,
-				Text:  "//go:variadic",
+				Text:  strings.Join(doc, "\n"),
 			})
 		}
 		for i := 0; i < numArgs; i++ {
@@ -370,45 +392,8 @@ func (f *cgoFile) createASTNode(name string, c clangCursor) (ast.Node, any) {
 		gen.Specs = append(gen.Specs, valueSpec)
 		return gen, nil
 	case C.CXCursor_MacroDefinition:
-		// Extract tokens from the Clang tokenizer.
-		// See: https://stackoverflow.com/a/19074846/559350
-		sourceRange := C.tinygo_clang_getCursorExtent(c)
-		tu := C.tinygo_clang_Cursor_getTranslationUnit(c)
-		var rawTokens *C.CXToken
-		var numTokens C.unsigned
-		C.clang_tokenize(tu, sourceRange, &rawTokens, &numTokens)
-		tokens := unsafe.Slice(rawTokens, numTokens)
-		// Convert this range of tokens back to source text.
-		// Ugly, but it works well enough.
-		sourceBuf := &bytes.Buffer{}
-		var startOffset int
-		for i, token := range tokens {
-			spelling := getString(C.clang_getTokenSpelling(tu, token))
-			location := C.clang_getTokenLocation(tu, token)
-			var tokenOffset C.unsigned
-			C.clang_getExpansionLocation(location, nil, nil, nil, &tokenOffset)
-			if i == 0 {
-				// The first token is the macro name itself.
-				// Skip it (after using its location).
-				startOffset = int(tokenOffset) + len(name)
-			} else {
-				// Later tokens are the macro contents.
-				for int(tokenOffset) > (startOffset + sourceBuf.Len()) {
-					// Pad the source text with whitespace (that must have been
-					// present in the original source as well).
-					sourceBuf.WriteByte(' ')
-				}
-				sourceBuf.WriteString(spelling)
-			}
-		}
-		C.clang_disposeTokens(tu, rawTokens, numTokens)
-		value := sourceBuf.String()
-		// Try to convert this #define into a Go constant expression.
-		tokenPos := token.NoPos
-		if pos != token.NoPos {
-			tokenPos = pos + token.Pos(len(name))
-		}
-		expr, scannerError := parseConst(tokenPos, f.fset, value)
+		tokenPos, value := f.getMacro(c)
+		expr, scannerError := parseConst(tokenPos, f.fset, value, nil, token.NoPos, f)
 		if scannerError != nil {
 			f.errors = append(f.errors, *scannerError)
 			return nil, nil
@@ -488,6 +473,62 @@ func (f *cgoFile) createASTNode(name string, c clangCursor) (ast.Node, any) {
 	}
 }

+// Return whether this is a macro that's also function-like, like this:
+//
+//	#define add(a, b) (a+b)
+func (f *cgoFile) isFunctionLikeMacro(c clangCursor) bool {
+	if C.tinygo_clang_getCursorKind(c) != C.CXCursor_MacroDefinition {
+		return false
+	}
+	return C.tinygo_clang_Cursor_isMacroFunctionLike(c) != 0
+}
+
+// Get the macro value: the position in the source file and the string value of
+// the macro.
+func (f *cgoFile) getMacro(c clangCursor) (pos token.Pos, value string) {
+	// Extract tokens from the Clang tokenizer.
+	// See: https://stackoverflow.com/a/19074846/559350
+	sourceRange := C.tinygo_clang_getCursorExtent(c)
+	tu := C.tinygo_clang_Cursor_getTranslationUnit(c)
+	var rawTokens *C.CXToken
+	var numTokens C.unsigned
+	C.clang_tokenize(tu, sourceRange, &rawTokens, &numTokens)
+	tokens := unsafe.Slice(rawTokens, numTokens)
+	defer C.clang_disposeTokens(tu, rawTokens, numTokens)
+
+	// Convert this range of tokens back to source text.
+	// Ugly, but it works well enough.
+	sourceBuf := &bytes.Buffer{}
+	var startOffset int
+	for i, token := range tokens {
+		spelling := getString(C.clang_getTokenSpelling(tu, token))
+		location := C.clang_getTokenLocation(tu, token)
+		var tokenOffset C.unsigned
+		C.clang_getExpansionLocation(location, nil, nil, nil, &tokenOffset)
+		if i == 0 {
+			// The first token is the macro name itself.
+			// Skip it (after using its location).
+			startOffset = int(tokenOffset)
+		} else {
+			// Later tokens are the macro contents.
+			for int(tokenOffset) > (startOffset + sourceBuf.Len()) {
+				// Pad the source text with whitespace (that must have been
+				// present in the original source as well).
+				sourceBuf.WriteByte(' ')
+			}
+			sourceBuf.WriteString(spelling)
+		}
+	}
+	value = sourceBuf.String()
+
+	// Obtain the position of this token. This is the position of the first
+	// character in the 'value' string and is used to report errors at the
+	// correct location in the source file.
+	pos = f.getCursorPosition(c)
+
+	return
+}
+
 func getString(clangString C.CXString) (s string) {
 	rawString := C.clang_getCString(clangString)
 	s = C.GoString(rawString)
@@ -84,3 +84,7 @@ unsigned tinygo_clang_Cursor_isAnonymous(CXCursor c) {
 unsigned tinygo_clang_Cursor_isBitField(CXCursor c) {
 	return clang_Cursor_isBitField(c);
 }
+
+unsigned tinygo_clang_Cursor_isMacroFunctionLike(CXCursor c) {
+	return clang_Cursor_isMacroFunctionLike(c);
+}
@@ -1,5 +1,6 @@
 package main

+import "syscall"
 import "unsafe"

 var _ unsafe.Pointer
@@ -31,6 +32,13 @@ func C.CBytes(b []byte) unsafe.Pointer {
 	return C.__CBytes(b)
 }

+//go:linkname C.__get_errno_num runtime.cgo_errno
+func C.__get_errno_num() uintptr
+
+func C.__get_errno() error {
+	return syscall.Errno(C.__get_errno_num())
+}
+
 type (
 	C.char      uint8
 	C.schar     int8
@@ -3,10 +3,26 @@ package main
 /*
 #define foo 3
 #define bar foo
+
+#define unreferenced 4
+#define referenced unreferenced
+
+#define fnlike() 5
+#define fnlike_val fnlike()
+#define square(n) (n*n)
+#define square_val square(20)
+#define add(a, b) (a + b)
+#define add_val add(3, 5)
 */
 import "C"

 const (
 	Foo = C.foo
 	Bar = C.bar
+
+	Baz = C.referenced
+
+	fnlike = C.fnlike_val
+	square = C.square_val
+	add    = C.add_val
 )
@@ -1,5 +1,6 @@
 package main

+import "syscall"
 import "unsafe"

 var _ unsafe.Pointer
@@ -31,6 +32,13 @@ func C.CBytes(b []byte) unsafe.Pointer {
 	return C.__CBytes(b)
 }

+//go:linkname C.__get_errno_num runtime.cgo_errno
+func C.__get_errno_num() uintptr
+
+func C.__get_errno() error {
+	return syscall.Errno(C.__get_errno_num())
+}
+
 type (
 	C.char      uint8
 	C.schar     int8
@@ -47,3 +55,8 @@ type (

 const C.foo = 3
 const C.bar = C.foo
+const C.unreferenced = 4
+const C.referenced = C.unreferenced
+const C.fnlike_val = 5
+const C.square_val = (20 * 20)
+const C.add_val = (3 + 5)
@@ -10,6 +10,11 @@ typedef struct {

 typedef someType noType; // undefined type

+// Some invalid noescape lines
+#cgo noescape
+#cgo noescape foo bar
+#cgo noescape unusedFunction
+
 #define SOME_CONST_1 5) // invalid const syntax
 #define SOME_CONST_2 6) // const not used (so no error)
 #define SOME_CONST_3 1234 // const too large for byte
@@ -26,6 +31,11 @@ import "C"
 // #warning another warning
 import "C"

+// #define add(a, b) (a+b)
+// #define add_toomuch add(1, 2, 3)
+// #define add_toolittle add(1)
+import "C"
+
 // Make sure that errors for the following lines won't change with future
 // additions to the CGo preamble.
 //
@@ -51,4 +61,7 @@ var (
 	// constants passed by a command line parameter
 	_ = C.SOME_PARAM_CONST_invalid
 	_ = C.SOME_PARAM_CONST_valid
+
+	_ = C.add_toomuch
+	_ = C.add_toolittle
 )
@@ -1,12 +1,17 @@
 // CGo errors:
+//     testdata/errors.go:14:1: missing function name in #cgo noescape line
+//     testdata/errors.go:15:1: multiple function names in #cgo noescape line
 //     testdata/errors.go:4:2: warning: some warning
 //     testdata/errors.go:11:9: error: unknown type name 'someType'
-//     testdata/errors.go:26:5: warning: another warning
-//     testdata/errors.go:13:23: unexpected token ), expected end of expression
-//     testdata/errors.go:21:26: unexpected token ), expected end of expression
-//     testdata/errors.go:16:33: unexpected token ), expected end of expression
-//     testdata/errors.go:17:34: unexpected token ), expected end of expression
+//     testdata/errors.go:31:5: warning: another warning
+//     testdata/errors.go:18:23: unexpected token ), expected end of expression
+//     testdata/errors.go:26:26: unexpected token ), expected end of expression
+//     testdata/errors.go:21:33: unexpected token ), expected end of expression
+//     testdata/errors.go:22:34: unexpected token ), expected end of expression
 //     -: unexpected token INT, expected end of expression
+//     testdata/errors.go:35:35: unexpected number of parameters: expected 2, got 3
+//     testdata/errors.go:36:31: unexpected number of parameters: expected 2, got 1
+//     testdata/errors.go:3:1: function "unusedFunction" in #cgo noescape line is not used

 // Type checking errors after CGo processing:
 //     testdata/errors.go:102: cannot use 2 << 10 (untyped int constant 2048) as C.char value in variable declaration (overflows)
@@ -17,9 +22,12 @@
 //     testdata/errors.go:114: undefined: C.SOME_CONST_b
 //     testdata/errors.go:116: undefined: C.SOME_CONST_startspace
 //     testdata/errors.go:119: undefined: C.SOME_PARAM_CONST_invalid
+//     testdata/errors.go:122: undefined: C.add_toomuch
+//     testdata/errors.go:123: undefined: C.add_toolittle

 package main

+import "syscall"
 import "unsafe"

 var _ unsafe.Pointer
@@ -51,6 +59,13 @@ func C.CBytes(b []byte) unsafe.Pointer {
 	return C.__CBytes(b)
 }

+//go:linkname C.__get_errno_num runtime.cgo_errno
+func C.__get_errno_num() uintptr
+
+func C.__get_errno() error {
+	return syscall.Errno(C.__get_errno_num())
+}
+
 type (
 	C.char      uint8
 	C.schar     int8
@@ -5,6 +5,7 @@

 package main

+import "syscall"
 import "unsafe"

 var _ unsafe.Pointer
@@ -36,6 +37,13 @@ func C.CBytes(b []byte) unsafe.Pointer {
 	return C.__CBytes(b)
 }

+//go:linkname C.__get_errno_num runtime.cgo_errno
+func C.__get_errno_num() uintptr
+
+func C.__get_errno() error {
+	return syscall.Errno(C.__get_errno_num())
+}
+
 type (
 	C.char      uint8
 	C.schar     int8
@@ -9,6 +9,10 @@ static void staticfunc(int x);

 // Global variable signatures.
 extern int someValue;
+
+void notEscapingFunction(int *a);
+
+#cgo noescape notEscapingFunction
 */
 import "C"

@@ -18,6 +22,7 @@ func accessFunctions() {
 	C.variadic0()
 	C.variadic2(3, 5)
 	C.staticfunc(3)
+	C.notEscapingFunction(nil)
 }

 func accessGlobals() {
@@ -1,5 +1,6 @@
 package main

+import "syscall"
 import "unsafe"

 var _ unsafe.Pointer
@@ -31,6 +32,13 @@ func C.CBytes(b []byte) unsafe.Pointer {
 	return C.__CBytes(b)
 }

+//go:linkname C.__get_errno_num runtime.cgo_errno
+func C.__get_errno_num() uintptr
+
+func C.__get_errno() error {
+	return syscall.Errno(C.__get_errno_num())
+}
+
 type (
 	C.char      uint8
 	C.schar     int8
@@ -67,5 +75,10 @@ func C.staticfunc!symbols.go(x C.int)

 var C.staticfunc!symbols.go$funcaddr unsafe.Pointer

+//export notEscapingFunction
+//go:noescape
+func C.notEscapingFunction(a *C.int)
+
+var C.notEscapingFunction$funcaddr unsafe.Pointer
 //go:extern someValue
 var C.someValue C.int
@@ -1,5 +1,6 @@
 package main

+import "syscall"
 import "unsafe"

 var _ unsafe.Pointer
@@ -31,6 +32,13 @@ func C.CBytes(b []byte) unsafe.Pointer {
 	return C.__CBytes(b)
 }

+//go:linkname C.__get_errno_num runtime.cgo_errno
+func C.__get_errno_num() uintptr
+
+func C.__get_errno() error {
+	return syscall.Errno(C.__get_errno_num())
+}
+
 type (
 	C.char      uint8
 	C.schar     int8
@@ -331,6 +331,7 @@ func (c *Config) CFlags(libclang bool) []string {
 			"-isystem", filepath.Join(path, "include"),
 			"-isystem", filepath.Join(picolibcDir, "include"),
 			"-isystem", filepath.Join(picolibcDir, "tinystdio"),
+			"-D__PICOLIBC_ERRNO_FUNCTION=__errno_location",
 		)
 	case "musl":
 		root := goenv.Get("TINYGOROOT")
@@ -340,6 +341,7 @@ func (c *Config) CFlags(libclang bool) []string {
 			"-nostdlibinc",
 			"-isystem", filepath.Join(path, "include"),
 			"-isystem", filepath.Join(root, "lib", "musl", "arch", arch),
+			"-isystem", filepath.Join(root, "lib", "musl", "arch", "generic"),
 			"-isystem", filepath.Join(root, "lib", "musl", "include"),
 		)
 	case "wasi-libc":
@@ -406,15 +408,7 @@ func (c *Config) LDFlags() []string {
 	if c.Target.LinkerScript != "" {
 		ldflags = append(ldflags, "-T", c.Target.LinkerScript)
 	}
-
-	if c.Options.ExtLDFlags != "" {
-		ext, err := shlex.Split(c.Options.ExtLDFlags)
-		if err != nil {
-			// if shlex can't split it, pass it as-is and let the external linker complain
-			ext = []string{c.Options.ExtLDFlags}
-		}
-		ldflags = append(ldflags, ext...)
-	}
+	ldflags = append(ldflags, c.Options.ExtLDFlags...)

 	return ldflags
 }
@@ -12,7 +12,7 @@ var (
 	validGCOptions            = []string{"none", "leaking", "conservative", "custom", "precise"}
 	validSchedulerOptions     = []string{"none", "tasks", "asyncify"}
 	validSerialOptions        = []string{"none", "uart", "usb", "rtt"}
-	validPrintSizeOptions     = []string{"none", "short", "full"}
+	validPrintSizeOptions     = []string{"none", "short", "full", "html"}
 	validPanicStrategyOptions = []string{"print", "trap"}
 	validOptOptions           = []string{"none", "0", "1", "2", "s", "z"}
 )
@@ -58,7 +58,7 @@ type Options struct {
 	Timeout         time.Duration
 	WITPackage      string // pass through to wasm-tools component embed invocation
 	WITWorld        string // pass through to wasm-tools component embed -w option
-	ExtLDFlags      string
+	ExtLDFlags      []string
 }

 // Verify performs a validation on the given options, raising an error if options are not valid.
@@ -11,7 +11,7 @@ func TestVerifyOptions(t *testing.T) {

 	expectedGCError := errors.New(`invalid gc option 'incorrect': valid values are none, leaking, conservative, custom, precise`)
 	expectedSchedulerError := errors.New(`invalid scheduler option 'incorrect': valid values are none, tasks, asyncify`)
-	expectedPrintSizeError := errors.New(`invalid size option 'incorrect': valid values are none, short, full`)
+	expectedPrintSizeError := errors.New(`invalid size option 'incorrect': valid values are none, short, full, html`)
 	expectedPanicStrategyError := errors.New(`invalid panic option 'incorrect': valid values are print, trap`)

 	testCases := []struct {
@@ -46,6 +46,7 @@ type TargetSpec struct {
 	LinkerScript     string   `json:"linkerscript,omitempty"`
 	ExtraFiles       []string `json:"extra-files,omitempty"`
 	RP2040BootPatch  *bool    `json:"rp2040-boot-patch,omitempty"` // Patch RP2040 2nd stage bootloader checksum
+	BootPatches      []string `json:"boot-patches,omitempty"`      // Bootloader patches to be applied in the order they appear.
 	Emulator         string   `json:"emulator,omitempty"`
 	FlashCommand     string   `json:"flash-command,omitempty"`
 	GDB              []string `json:"gdb,omitempty"`
@@ -390,6 +391,7 @@ func defaultTarget(options *Options) (*TargetSpec, error) {
 			"-platform_version", "macos", platformVersion, platformVersion,
 		)
 		spec.ExtraFiles = append(spec.ExtraFiles,
+			"src/internal/futex/futex_darwin.c",
 			"src/runtime/os_darwin.c",
 			"src/runtime/runtime_unix.c",
 			"src/runtime/signal.c")
@@ -413,6 +415,7 @@ func defaultTarget(options *Options) (*TargetSpec, error) {
 			spec.CFlags = append(spec.CFlags, "-mno-outline-atomics")
 		}
 		spec.ExtraFiles = append(spec.ExtraFiles,
+			"src/internal/futex/futex_linux.c",
 			"src/runtime/runtime_unix.c",
 			"src/runtime/signal.c")
 	case "windows":
@@ -19,8 +19,9 @@ const maxFieldsPerParam = 3
 // useful while declaring or defining a function.
 type paramInfo struct {
 	llvmType llvm.Type
-	name     string // name, possibly with suffixes for e.g. struct fields
-	elemSize uint64 // size of pointer element type, or 0 if this isn't a pointer
+	name     string     // name, possibly with suffixes for e.g. struct fields
+	elemSize uint64     // size of pointer element type, or 0 if this isn't a pointer
+	flags    paramFlags // extra flags for this parameter
 }

 // paramFlags identifies parameter attributes for flags. Most importantly, it
@@ -28,9 +29,9 @@ type paramInfo struct {
 type paramFlags uint8

 const (
-	// Parameter may have the deferenceable_or_null attribute. This attribute
-	// cannot be applied to unsafe.Pointer and to the data pointer of slices.
-	paramIsDeferenceableOrNull = 1 << iota
+	// Whether this is a full or partial Go parameter (int, slice, etc).
+	// The extra context parameter is not a Go parameter.
+	paramIsGoParam = 1 << iota
 )

 // createRuntimeCallCommon creates a runtime call. Use createRuntimeCall or
@@ -195,6 +196,7 @@ func (c *compilerContext) getParamInfo(t llvm.Type, name string, goType types.Ty
 	info := paramInfo{
 		llvmType: t,
 		name:     name,
+		flags:    paramIsGoParam,
 	}
 	if goType != nil {
 		switch underlying := goType.Underlying().(type) {
@@ -4,7 +4,9 @@ package compiler
 // or pseudo-operations that are lowered during goroutine lowering.

 import (
+	"fmt"
 	"go/types"
+	"math"

 	"github.com/tinygo-org/tinygo/compiler/llvmutil"
 	"golang.org/x/tools/go/ssa"
@@ -41,17 +43,17 @@ func (b *builder) createChanSend(instr *ssa.Send) {
 		b.CreateStore(chanValue, valueAlloca)
 	}

-	// Allocate blockedlist buffer.
-	channelBlockedList := b.getLLVMRuntimeType("channelBlockedList")
-	channelBlockedListAlloca, channelBlockedListAllocaSize := b.createTemporaryAlloca(channelBlockedList, "chan.blockedList")
+	// Allocate buffer for the channel operation.
+	channelOp := b.getLLVMRuntimeType("channelOp")
+	channelOpAlloca, channelOpAllocaSize := b.createTemporaryAlloca(channelOp, "chan.op")

 	// Do the send.
-	b.createRuntimeCall("chanSend", []llvm.Value{ch, valueAlloca, channelBlockedListAlloca}, "")
+	b.createRuntimeCall("chanSend", []llvm.Value{ch, valueAlloca, channelOpAlloca}, "")

 	// End the lifetime of the allocas.
 	// This also works around a bug in CoroSplit, at least in LLVM 8:
 	// https://bugs.llvm.org/show_bug.cgi?id=41742
-	b.emitLifetimeEnd(channelBlockedListAlloca, channelBlockedListAllocaSize)
+	b.emitLifetimeEnd(channelOpAlloca, channelOpAllocaSize)
 	if !isZeroSize {
 		b.emitLifetimeEnd(valueAlloca, valueAllocaSize)
 	}
@@ -72,12 +74,12 @@ func (b *builder) createChanRecv(unop *ssa.UnOp) llvm.Value {
 		valueAlloca, valueAllocaSize = b.createTemporaryAlloca(valueType, "chan.value")
 	}

-	// Allocate blockedlist buffer.
-	channelBlockedList := b.getLLVMRuntimeType("channelBlockedList")
-	channelBlockedListAlloca, channelBlockedListAllocaSize := b.createTemporaryAlloca(channelBlockedList, "chan.blockedList")
+	// Allocate buffer for the channel operation.
+	channelOp := b.getLLVMRuntimeType("channelOp")
+	channelOpAlloca, channelOpAllocaSize := b.createTemporaryAlloca(channelOp, "chan.op")

 	// Do the receive.
-	commaOk := b.createRuntimeCall("chanRecv", []llvm.Value{ch, valueAlloca, channelBlockedListAlloca}, "")
+	commaOk := b.createRuntimeCall("chanRecv", []llvm.Value{ch, valueAlloca, channelOpAlloca}, "")
 	var received llvm.Value
 	if isZeroSize {
 		received = llvm.ConstNull(valueType)
@@ -85,7 +87,7 @@ func (b *builder) createChanRecv(unop *ssa.UnOp) llvm.Value {
 		received = b.CreateLoad(valueType, valueAlloca, "chan.received")
 		b.emitLifetimeEnd(valueAlloca, valueAllocaSize)
 	}
-	b.emitLifetimeEnd(channelBlockedListAlloca, channelBlockedListAllocaSize)
+	b.emitLifetimeEnd(channelOpAlloca, channelOpAllocaSize)

 	if unop.CommaOk {
 		tuple := llvm.Undef(b.ctx.StructType([]llvm.Type{valueType, b.ctx.Int1Type()}, false))
@@ -124,6 +126,20 @@ func (b *builder) createSelect(expr *ssa.Select) llvm.Value {
 		}
 	}

+	const maxSelectStates = math.MaxUint32 >> 2
+	if len(expr.States) > maxSelectStates {
+		// The runtime code assumes that the number of state must fit in 30 bits
+		// (so the select index can be stored in a uint32 with two bits reserved
+		// for other purposes). It seems unlikely that a real program would have
+		// that many states, but we check for this case anyway to be sure.
+		// We use a uint32 (and not a uintptr or uint64) to avoid 64-bit atomic
+		// operations which aren't available everywhere.
+		b.addError(expr.Pos(), fmt.Sprintf("too many select states: got %d but the maximum supported number is %d", len(expr.States), maxSelectStates))
+
+		// Continue as usual (we'll generate broken code but the error will
+		// prevent the compilation to complete).
+	}
+
 	// This code create a (stack-allocated) slice containing all the select
 	// cases and then calls runtime.chanSelect to perform the actual select
 	// statement.
@@ -198,10 +214,10 @@ func (b *builder) createSelect(expr *ssa.Select) llvm.Value {
 	if expr.Blocking {
 		// Stack-allocate operation structures.
 		// If these were simply created as a slice, they would heap-allocate.
-		chBlockAllocaType := llvm.ArrayType(b.getLLVMRuntimeType("channelBlockedList"), len(selectStates))
-		chBlockAlloca, chBlockSize := b.createTemporaryAlloca(chBlockAllocaType, "select.block.alloca")
-		chBlockLen := llvm.ConstInt(b.uintptrType, uint64(len(selectStates)), false)
-		chBlockPtr := b.CreateGEP(chBlockAllocaType, chBlockAlloca, []llvm.Value{
+		opsAllocaType := llvm.ArrayType(b.getLLVMRuntimeType("channelOp"), len(selectStates))
+		opsAlloca, opsSize := b.createTemporaryAlloca(opsAllocaType, "select.block.alloca")
+		opsLen := llvm.ConstInt(b.uintptrType, uint64(len(selectStates)), false)
+		opsPtr := b.CreateGEP(opsAllocaType, opsAlloca, []llvm.Value{
 			llvm.ConstInt(b.ctx.Int32Type(), 0, false),
 			llvm.ConstInt(b.ctx.Int32Type(), 0, false),
 		}, "select.block")
@@ -209,15 +225,18 @@ func (b *builder) createSelect(expr *ssa.Select) llvm.Value {
 		results = b.createRuntimeCall("chanSelect", []llvm.Value{
 			recvbuf,
 			statesPtr, statesLen, statesLen, // []chanSelectState
-			chBlockPtr, chBlockLen, chBlockLen, // []channelBlockList
+			opsPtr, opsLen, opsLen, // []channelOp
 		}, "select.result")

 		// Terminate the lifetime of the operation structures.
-		b.emitLifetimeEnd(chBlockAlloca, chBlockSize)
+		b.emitLifetimeEnd(opsAlloca, opsSize)
 	} else {
-		results = b.createRuntimeCall("tryChanSelect", []llvm.Value{
+		opsPtr := llvm.ConstNull(b.dataPtrType)
+		opsLen := llvm.ConstInt(b.uintptrType, 0, false)
+		results = b.createRuntimeCall("chanSelect", []llvm.Value{
 			recvbuf,
 			statesPtr, statesLen, statesLen, // []chanSelectState
+			opsPtr, opsLen, opsLen, // []channelOp (nil slice)
 		}, "select.result")
 	}

@@ -17,6 +17,7 @@ import (

 	"github.com/tinygo-org/tinygo/compiler/llvmutil"
 	"github.com/tinygo-org/tinygo/loader"
+	"github.com/tinygo-org/tinygo/src/tinygo"
 	"golang.org/x/tools/go/ssa"
 	"golang.org/x/tools/go/types/typeutil"
 	"tinygo.org/x/go-llvm"
@@ -1680,7 +1681,12 @@ func (b *builder) createBuiltin(argTypes []types.Type, argValues []llvm.Value, c
 			result = b.CreateSelect(cmp, result, arg, "")
 		}
 		return result, nil
+	case "panic":
+		// This is rare, but happens in "defer panic()".
+		b.createRuntimeInvoke("_panic", argValues, "")
+		return llvm.Value{}, nil
 	case "print", "println":
+		b.createRuntimeCall("printlock", nil, "")
 		for i, value := range argValues {
 			if i >= 1 && callName == "println" {
 				b.createRuntimeCall("printspace", nil, "")
@@ -1741,6 +1747,7 @@ func (b *builder) createBuiltin(argTypes []types.Type, argValues []llvm.Value, c
 		if callName == "println" {
 			b.createRuntimeCall("printnl", nil, "")
 		}
+		b.createRuntimeCall("printunlock", nil, "")
 		return llvm.Value{}, nil // print() or println() returns void
 	case "real":
 		cplx := argValues[0]
@@ -1865,10 +1872,9 @@ func (b *builder) createFunctionCall(instr *ssa.CallCommon) (llvm.Value, error)
 			}
 			return llvm.ConstInt(b.ctx.Int1Type(), supportsRecover, false), nil
 		case name == "runtime.panicStrategy":
-			// These constants are defined in src/runtime/panic.go.
 			panicStrategy := map[string]uint64{
-				"print": 1, // panicStrategyPrint
-				"trap":  2, // panicStrategyTrap
+				"print": tinygo.PanicStrategyPrint,
+				"trap":  tinygo.PanicStrategyTrap,
 			}[b.Config.PanicStrategy]
 			return llvm.ConstInt(b.ctx.Int8Type(), panicStrategy, false), nil
 		case name == "runtime/interrupt.New":
@@ -41,6 +41,8 @@ func (b *builder) createInterruptGlobal(instr *ssa.CallCommon) (llvm.Value, erro

 	// Create a new global of type runtime/interrupt.handle. Globals of this
 	// type are lowered in the interrupt lowering pass.
+	// It must have an alignment of 1, otherwise LLVM thinks a ptrtoint of the
+	// global has the lower bits unset.
 	globalType := b.program.ImportedPackage("runtime/interrupt").Type("handle").Type()
 	globalLLVMType := b.getLLVMType(globalType)
 	globalName := b.fn.Package().Pkg.Path() + "$interrupt" + strconv.FormatInt(id.Int64(), 10)
@@ -48,6 +50,7 @@ func (b *builder) createInterruptGlobal(instr *ssa.CallCommon) (llvm.Value, erro
 	global.SetVisibility(llvm.HiddenVisibility)
 	global.SetGlobalConstant(true)
 	global.SetUnnamedAddr(true)
+	global.SetAlignment(1)
 	initializer := llvm.ConstNull(globalLLVMType)
 	initializer = b.CreateInsertValue(initializer, funcContext, 0, "")
 	initializer = b.CreateInsertValue(initializer, funcPtr, 1, "")
@@ -6,17 +6,11 @@ import (
 	"go/token"
 	"go/types"

+	"github.com/tinygo-org/tinygo/src/tinygo"
 	"golang.org/x/tools/go/ssa"
 	"tinygo.org/x/go-llvm"
 )

-// constants for hashmap algorithms; must match src/runtime/hashmap.go
-const (
-	hashmapAlgorithmBinary = iota
-	hashmapAlgorithmString
-	hashmapAlgorithmInterface
-)
-
 // createMakeMap creates a new map object (runtime.hashmap) by allocating and
 // initializing an appropriately sized object.
 func (b *builder) createMakeMap(expr *ssa.MakeMap) (llvm.Value, error) {
@@ -24,20 +18,20 @@ func (b *builder) createMakeMap(expr *ssa.MakeMap) (llvm.Value, error) {
 	keyType := mapType.Key().Underlying()
 	llvmValueType := b.getLLVMType(mapType.Elem().Underlying())
 	var llvmKeyType llvm.Type
-	var alg uint64 // must match values in src/runtime/hashmap.go
+	var alg uint64
 	if t, ok := keyType.(*types.Basic); ok && t.Info()&types.IsString != 0 {
 		// String keys.
 		llvmKeyType = b.getLLVMType(keyType)
-		alg = hashmapAlgorithmString
+		alg = uint64(tinygo.HashmapAlgorithmString)
 	} else if hashmapIsBinaryKey(keyType) {
 		// Trivially comparable keys.
 		llvmKeyType = b.getLLVMType(keyType)
-		alg = hashmapAlgorithmBinary
+		alg = uint64(tinygo.HashmapAlgorithmBinary)
 	} else {
 		// All other keys. Implemented as map[interface{}]valueType for ease of
 		// implementation.
 		llvmKeyType = b.getLLVMRuntimeType("_interface")
-		alg = hashmapAlgorithmInterface
+		alg = uint64(tinygo.HashmapAlgorithmInterface)
 	}
 	keySize := b.targetData.TypeAllocSize(llvmKeyType)
 	valueSize := b.targetData.TypeAllocSize(llvmValueType)
@@ -33,6 +33,7 @@ type functionInfo struct {
 	exported      bool       // go:export, CGo
 	interrupt     bool       // go:interrupt
 	nobounds      bool       // go:nobounds
+	noescape      bool       // go:noescape
 	variadic      bool       // go:variadic (CGo only)
 	inline        inlineType // go:inline
 }
@@ -127,11 +128,20 @@ func (c *compilerContext) getFunction(fn *ssa.Function) (llvm.Type, llvm.Value)
 	c.addStandardDeclaredAttributes(llvmFn)

 	dereferenceableOrNullKind := llvm.AttributeKindID("dereferenceable_or_null")
-	for i, info := range paramInfos {
-		if info.elemSize != 0 {
-			dereferenceableOrNull := c.ctx.CreateEnumAttribute(dereferenceableOrNullKind, info.elemSize)
+	for i, paramInfo := range paramInfos {
+		if paramInfo.elemSize != 0 {
+			dereferenceableOrNull := c.ctx.CreateEnumAttribute(dereferenceableOrNullKind, paramInfo.elemSize)
 			llvmFn.AddAttributeAtIndex(i+1, dereferenceableOrNull)
 		}
+		if info.noescape && paramInfo.flags&paramIsGoParam != 0 && paramInfo.llvmType.TypeKind() == llvm.PointerTypeKind {
+			// Parameters to functions with a //go:noescape parameter should get
+			// the nocapture attribute. However, the context parameter should
+			// not.
+			// (It may be safe to add the nocapture parameter to the context
+			// parameter, but I'd like to stay on the safe side here).
+			nocapture := c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nocapture"), 0)
+			llvmFn.AddAttributeAtIndex(i+1, nocapture)
+		}
 	}

 	// Set a number of function or parameter attributes, depending on the
@@ -221,6 +231,15 @@ func (c *compilerContext) getFunction(fn *ssa.Function) (llvm.Type, llvm.Value)
 		}
 	}

+	// Build the function if needed.
+	c.maybeCreateSyntheticFunction(fn, llvmFn)
+
+	return fnType, llvmFn
+}
+
+// If this is a synthetic function (such as a generic function or a wrapper),
+// create it now.
+func (c *compilerContext) maybeCreateSyntheticFunction(fn *ssa.Function, llvmFn llvm.Value) {
 	// Synthetic functions are functions that do not appear in the source code,
 	// they are artificially constructed. Usually they are wrapper functions
 	// that are not referenced anywhere except in a SSA call instruction so
@@ -228,6 +247,10 @@ func (c *compilerContext) getFunction(fn *ssa.Function) (llvm.Type, llvm.Value)
 	// The exception is the package initializer, which does appear in the
 	// *ssa.Package members and so shouldn't be created here.
 	if fn.Synthetic != "" && fn.Synthetic != "package initializer" && fn.Synthetic != "generic function" && fn.Synthetic != "range-over-func yield" {
+		if len(fn.Blocks) == 0 {
+			c.addError(fn.Pos(), "missing function body")
+			return
+		}
 		irbuilder := c.ctx.NewBuilder()
 		b := newBuilder(c, irbuilder, fn)
 		b.createFunction()
@@ -235,8 +258,6 @@ func (c *compilerContext) getFunction(fn *ssa.Function) (llvm.Type, llvm.Value)
 		llvmFn.SetLinkage(llvm.LinkOnceODRLinkage)
 		llvmFn.SetUnnamedAddr(true)
 	}
-
-	return fnType, llvmFn
 }

 // getFunctionInfo returns information about a function that is not directly
@@ -345,7 +366,7 @@ func (c *compilerContext) parsePragmas(info *functionInfo, f *ssa.Function) {
 				continue
 			}
 			if len(parts) != 2 {
-				c.addError(f.Pos(), fmt.Sprintf("expected one parameter to //go:wasmimport, not %d", len(parts)-1))
+				c.addError(f.Pos(), fmt.Sprintf("expected one parameter to //go:wasmexport, not %d", len(parts)-1))
 				continue
 			}
 			name := parts[1]
@@ -394,6 +415,13 @@ func (c *compilerContext) parsePragmas(info *functionInfo, f *ssa.Function) {
 			if hasUnsafeImport(f.Pkg.Pkg) {
 				info.nobounds = true
 			}
+		case "//go:noescape":
+			// Don't let pointer parameters escape.
+			// Following the upstream Go implementation, we only do this for
+			// declarations, not definitions.
+			if len(f.Blocks) == 0 {
+				info.noescape = true
+			}
 		case "//go:variadic":
 			// The //go:variadic pragma is emitted by the CGo preprocessing
 			// pass for C variadic functions. This includes both explicit
@@ -3,7 +3,7 @@ source_filename = "channel.go"
 target datalayout = "e-m:e-p:32:32-p10:8:8-p20:8:8-i64:64-n32:64-S128-ni:1:10:20"
 target triple = "wasm32-unknown-wasi"

-%runtime.channelBlockedList = type { ptr, ptr, ptr, { ptr, i32, i32 } }
+%runtime.channelOp = type { ptr, ptr, i32, ptr }
 %runtime.chanSelectState = type { ptr, ptr }

 ; Function Attrs: allockind("alloc,zeroed") allocsize(0)
@@ -18,15 +18,15 @@ entry:
 }

 ; Function Attrs: nounwind
-define hidden void @main.chanIntSend(ptr dereferenceable_or_null(32) %ch, ptr %context) unnamed_addr #2 {
+define hidden void @main.chanIntSend(ptr dereferenceable_or_null(36) %ch, ptr %context) unnamed_addr #2 {
 entry:
-  %chan.blockedList = alloca %runtime.channelBlockedList, align 8
+  %chan.op = alloca %runtime.channelOp, align 8
  %chan.value = alloca i32, align 4
  call void @llvm.lifetime.start.p0(i64 4, ptr nonnull %chan.value)
  store i32 3, ptr %chan.value, align 4
-  call void @llvm.lifetime.start.p0(i64 24, ptr nonnull %chan.blockedList)
-  call void @runtime.chanSend(ptr %ch, ptr nonnull %chan.value, ptr nonnull %chan.blockedList, ptr undef) #4
-  call void @llvm.lifetime.end.p0(i64 24, ptr nonnull %chan.blockedList)
+  call void @llvm.lifetime.start.p0(i64 16, ptr nonnull %chan.op)
+  call void @runtime.chanSend(ptr %ch, ptr nonnull %chan.value, ptr nonnull %chan.op, ptr undef) #4
+  call void @llvm.lifetime.end.p0(i64 16, ptr nonnull %chan.op)
  call void @llvm.lifetime.end.p0(i64 4, ptr nonnull %chan.value)
  ret void
 }
@@ -34,48 +34,48 @@ entry:
 ; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
 declare void @llvm.lifetime.start.p0(i64 immarg, ptr nocapture) #3

-declare void @runtime.chanSend(ptr dereferenceable_or_null(32), ptr, ptr dereferenceable_or_null(24), ptr) #1
+declare void @runtime.chanSend(ptr dereferenceable_or_null(36), ptr, ptr dereferenceable_or_null(16), ptr) #1

 ; Function Attrs: nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
 declare void @llvm.lifetime.end.p0(i64 immarg, ptr nocapture) #3

 ; Function Attrs: nounwind
-define hidden void @main.chanIntRecv(ptr dereferenceable_or_null(32) %ch, ptr %context) unnamed_addr #2 {
+define hidden void @main.chanIntRecv(ptr dereferenceable_or_null(36) %ch, ptr %context) unnamed_addr #2 {
 entry:
-  %chan.blockedList = alloca %runtime.channelBlockedList, align 8
+  %chan.op = alloca %runtime.channelOp, align 8
  %chan.value = alloca i32, align 4
  call void @llvm.lifetime.start.p0(i64 4, ptr nonnull %chan.value)
-  call void @llvm.lifetime.start.p0(i64 24, ptr nonnull %chan.blockedList)
-  %0 = call i1 @runtime.chanRecv(ptr %ch, ptr nonnull %chan.value, ptr nonnull %chan.blockedList, ptr undef) #4
+  call void @llvm.lifetime.start.p0(i64 16, ptr nonnull %chan.op)
+  %0 = call i1 @runtime.chanRecv(ptr %ch, ptr nonnull %chan.value, ptr nonnull %chan.op, ptr undef) #4
  call void @llvm.lifetime.end.p0(i64 4, ptr nonnull %chan.value)
-  call void @llvm.lifetime.end.p0(i64 24, ptr nonnull %chan.blockedList)
+  call void @llvm.lifetime.end.p0(i64 16, ptr nonnull %chan.op)
  ret void
 }

-declare i1 @runtime.chanRecv(ptr dereferenceable_or_null(32), ptr, ptr dereferenceable_or_null(24), ptr) #1
+declare i1 @runtime.chanRecv(ptr dereferenceable_or_null(36), ptr, ptr dereferenceable_or_null(16), ptr) #1

 ; Function Attrs: nounwind
-define hidden void @main.chanZeroSend(ptr dereferenceable_or_null(32) %ch, ptr %context) unnamed_addr #2 {
+define hidden void @main.chanZeroSend(ptr dereferenceable_or_null(36) %ch, ptr %context) unnamed_addr #2 {
 entry:
-  %chan.blockedList = alloca %runtime.channelBlockedList, align 8
-  call void @llvm.lifetime.start.p0(i64 24, ptr nonnull %chan.blockedList)
-  call void @runtime.chanSend(ptr %ch, ptr null, ptr nonnull %chan.blockedList, ptr undef) #4
-  call void @llvm.lifetime.end.p0(i64 24, ptr nonnull %chan.blockedList)
+  %chan.op = alloca %runtime.channelOp, align 8
+  call void @llvm.lifetime.start.p0(i64 16, ptr nonnull %chan.op)
+  call void @runtime.chanSend(ptr %ch, ptr null, ptr nonnull %chan.op, ptr undef) #4
+  call void @llvm.lifetime.end.p0(i64 16, ptr nonnull %chan.op)
  ret void
 }

 ; Function Attrs: nounwind
-define hidden void @main.chanZeroRecv(ptr dereferenceable_or_null(32) %ch, ptr %context) unnamed_addr #2 {
+define hidden void @main.chanZeroRecv(ptr dereferenceable_or_null(36) %ch, ptr %context) unnamed_addr #2 {
 entry:
-  %chan.blockedList = alloca %runtime.channelBlockedList, align 8
-  call void @llvm.lifetime.start.p0(i64 24, ptr nonnull %chan.blockedList)
-  %0 = call i1 @runtime.chanRecv(ptr %ch, ptr null, ptr nonnull %chan.blockedList, ptr undef) #4
-  call void @llvm.lifetime.end.p0(i64 24, ptr nonnull %chan.blockedList)
+  %chan.op = alloca %runtime.channelOp, align 8
+  call void @llvm.lifetime.start.p0(i64 16, ptr nonnull %chan.op)
+  %0 = call i1 @runtime.chanRecv(ptr %ch, ptr null, ptr nonnull %chan.op, ptr undef) #4
+  call void @llvm.lifetime.end.p0(i64 16, ptr nonnull %chan.op)
  ret void
 }

 ; Function Attrs: nounwind
-define hidden void @main.selectZeroRecv(ptr dereferenceable_or_null(32) %ch1, ptr dereferenceable_or_null(32) %ch2, ptr %context) unnamed_addr #2 {
+define hidden void @main.selectZeroRecv(ptr dereferenceable_or_null(36) %ch1, ptr dereferenceable_or_null(36) %ch2, ptr %context) unnamed_addr #2 {
 entry:
  %select.states.alloca = alloca [2 x %runtime.chanSelectState], align 8
  %select.send.value = alloca i32, align 4
@@ -88,7 +88,7 @@ entry:
  store ptr %ch2, ptr %0, align 4
  %.repack3 = getelementptr inbounds [2 x %runtime.chanSelectState], ptr %select.states.alloca, i32 0, i32 1, i32 1
  store ptr null, ptr %.repack3, align 4
-  %select.result = call { i32, i1 } @runtime.tryChanSelect(ptr undef, ptr nonnull %select.states.alloca, i32 2, i32 2, ptr undef) #4
+  %select.result = call { i32, i1 } @runtime.chanSelect(ptr undef, ptr nonnull %select.states.alloca, i32 2, i32 2, ptr null, i32 0, i32 0, ptr undef) #4
  call void @llvm.lifetime.end.p0(i64 16, ptr nonnull %select.states.alloca)
  %1 = extractvalue { i32, i1 } %select.result, 0
  %2 = icmp eq i32 %1, 0
@@ -105,7 +105,7 @@ select.body:                                      ; preds = %select.next
  br label %select.done
 }

-declare { i32, i1 } @runtime.tryChanSelect(ptr, ptr, i32, i32, ptr) #1
+declare { i32, i1 } @runtime.chanSelect(ptr, ptr, i32, i32, ptr, i32, i32, ptr) #1

 attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+mutable-globals,+nontrapping-fptoint,+sign-ext" }
 attributes #1 = { "target-features"="+bulk-memory,+mutable-globals,+nontrapping-fptoint,+sign-ext" }
@@ -3,7 +3,7 @@ source_filename = "defer.go"
 target datalayout = "e-m:e-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64"
 target triple = "thumbv7m-unknown-unknown-eabi"

-%runtime.deferFrame = type { ptr, ptr, [0 x ptr], ptr, i1, %runtime._interface }
+%runtime.deferFrame = type { ptr, ptr, [0 x ptr], ptr, i8, %runtime._interface }
 %runtime._interface = type { ptr, ptr }
 %runtime._defer = type { i32, ptr }

@@ -122,12 +122,18 @@ declare void @runtime.destroyDeferFrame(ptr dereferenceable_or_null(24), ptr) #2
 ; Function Attrs: nounwind
 define internal void @"main.deferSimple$1"(ptr %context) unnamed_addr #1 {
 entry:
+  call void @runtime.printlock(ptr undef) #4
  call void @runtime.printint32(i32 3, ptr undef) #4
+  call void @runtime.printunlock(ptr undef) #4
  ret void
 }

+declare void @runtime.printlock(ptr) #2
+
 declare void @runtime.printint32(i32, ptr) #2

+declare void @runtime.printunlock(ptr) #2
+
 ; Function Attrs: nounwind
 define hidden void @main.deferMultiple(ptr %context) unnamed_addr #1 {
 entry:
@@ -250,14 +256,18 @@ rundefers.end7:                                   ; preds = %rundefers.loophead1
 ; Function Attrs: nounwind
 define internal void @"main.deferMultiple$1"(ptr %context) unnamed_addr #1 {
 entry:
+  call void @runtime.printlock(ptr undef) #4
  call void @runtime.printint32(i32 3, ptr undef) #4
+  call void @runtime.printunlock(ptr undef) #4
  ret void
 }

 ; Function Attrs: nounwind
 define internal void @"main.deferMultiple$2"(ptr %context) unnamed_addr #1 {
 entry:
+  call void @runtime.printlock(ptr undef) #4
  call void @runtime.printint32(i32 5, ptr undef) #4
+  call void @runtime.printunlock(ptr undef) #4
  ret void
 }

@@ -70,7 +70,9 @@ entry:
  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (ptr @"main.closureFunctionGoroutine$1$gowrapper" to i32), ptr undef) #9
  call void @"internal/task.start"(i32 ptrtoint (ptr @"main.closureFunctionGoroutine$1$gowrapper" to i32), ptr nonnull %0, i32 %stacksize, ptr undef) #9
  %2 = load i32, ptr %n, align 4
+  call void @runtime.printlock(ptr undef) #9
  call void @runtime.printint32(i32 %2, ptr undef) #9
+  call void @runtime.printunlock(ptr undef) #9
  ret void
 }

@@ -91,8 +93,12 @@ entry:
  ret void
 }

+declare void @runtime.printlock(ptr) #2
+
 declare void @runtime.printint32(i32, ptr) #2

+declare void @runtime.printunlock(ptr) #2
+
 ; Function Attrs: nounwind
 define hidden void @main.funcGoroutine(ptr %fn.context, ptr %fn.funcptr, ptr %context) unnamed_addr #1 {
 entry:
@@ -135,13 +141,13 @@ entry:
 declare i32 @runtime.sliceCopy(ptr nocapture writeonly, ptr nocapture readonly, i32, i32, i32, ptr) #2

 ; Function Attrs: nounwind
-define hidden void @main.closeBuiltinGoroutine(ptr dereferenceable_or_null(32) %ch, ptr %context) unnamed_addr #1 {
+define hidden void @main.closeBuiltinGoroutine(ptr dereferenceable_or_null(36) %ch, ptr %context) unnamed_addr #1 {
 entry:
  call void @runtime.chanClose(ptr %ch, ptr undef) #9
  ret void
 }

-declare void @runtime.chanClose(ptr dereferenceable_or_null(32), ptr) #2
+declare void @runtime.chanClose(ptr dereferenceable_or_null(36), ptr) #2

 ; Function Attrs: nounwind
 define hidden void @main.startInterfaceMethod(ptr %itf.typecode, ptr %itf.value, ptr %context) unnamed_addr #1 {
@@ -76,7 +76,9 @@ entry:
  store ptr %n, ptr %1, align 4
  call void @"internal/task.start"(i32 ptrtoint (ptr @"main.closureFunctionGoroutine$1$gowrapper" to i32), ptr nonnull %0, i32 65536, ptr undef) #9
  %2 = load i32, ptr %n, align 4
+  call void @runtime.printlock(ptr undef) #9
  call void @runtime.printint32(i32 %2, ptr undef) #9
+  call void @runtime.printunlock(ptr undef) #9
  ret void
 }

@@ -98,8 +100,12 @@ entry:
  unreachable
 }

+declare void @runtime.printlock(ptr) #1
+
 declare void @runtime.printint32(i32, ptr) #1

+declare void @runtime.printunlock(ptr) #1
+
 ; Function Attrs: nounwind
 define hidden void @main.funcGoroutine(ptr %fn.context, ptr %fn.funcptr, ptr %context) unnamed_addr #2 {
 entry:
@@ -144,13 +150,13 @@ entry:
 declare i32 @runtime.sliceCopy(ptr nocapture writeonly, ptr nocapture readonly, i32, i32, i32, ptr) #1

 ; Function Attrs: nounwind
-define hidden void @main.closeBuiltinGoroutine(ptr dereferenceable_or_null(32) %ch, ptr %context) unnamed_addr #2 {
+define hidden void @main.closeBuiltinGoroutine(ptr dereferenceable_or_null(36) %ch, ptr %context) unnamed_addr #2 {
 entry:
  call void @runtime.chanClose(ptr %ch, ptr undef) #9
  ret void
 }

-declare void @runtime.chanClose(ptr dereferenceable_or_null(32), ptr) #1
+declare void @runtime.chanClose(ptr dereferenceable_or_null(36), ptr) #1

 ; Function Attrs: nounwind
 define hidden void @main.startInterfaceMethod(ptr %itf.typecode, ptr %itf.value, ptr %context) unnamed_addr #2 {
@@ -106,3 +106,12 @@ var undefinedGlobalNotInSection uint32
 //go:align 1024
 //go:section .global_section
 var multipleGlobalPragmas uint32
+
+//go:noescape
+func doesNotEscapeParam(a *int, b []int, c chan int, d *[0]byte)
+
+// The //go:noescape pragma only works on declarations, not definitions.
+//
+//go:noescape
+func stillEscapes(a *int, b []int, c chan int, d *[0]byte) {
+}
@@ -85,6 +85,14 @@ entry:

 declare void @main.undefinedFunctionNotInSection(ptr) #1

+declare void @main.doesNotEscapeParam(ptr nocapture dereferenceable_or_null(4), ptr nocapture, i32, i32, ptr nocapture dereferenceable_or_null(36), ptr nocapture, ptr) #1
+
+; Function Attrs: nounwind
+define hidden void @main.stillEscapes(ptr dereferenceable_or_null(4) %a, ptr %b.data, i32 %b.len, i32 %b.cap, ptr dereferenceable_or_null(36) %c, ptr %d, ptr %context) unnamed_addr #2 {
+entry:
+  ret void
+}
+
 attributes #0 = { allockind("alloc,zeroed") allocsize(0) "alloc-family"="runtime.alloc" "target-features"="+bulk-memory,+mutable-globals,+nontrapping-fptoint,+sign-ext" }
 attributes #1 = { "target-features"="+bulk-memory,+mutable-globals,+nontrapping-fptoint,+sign-ext" }
 attributes #2 = { nounwind "target-features"="+bulk-memory,+mutable-globals,+nontrapping-fptoint,+sign-ext" }
@@ -4,29 +4,37 @@ import (
 	"errors"
 	"fmt"
 	"io"
+	"runtime/debug"
 	"strings"
 )

 // Version of TinyGo.
 // Update this value before release of new version of software.
-const version = "0.34.0"
-
-var (
-	// This variable is set at build time using -ldflags parameters.
-	// See: https://stackoverflow.com/a/11355611
-	GitSha1 string
-)
+const version = "0.35.0"

 // Return TinyGo version, either in the form 0.30.0 or as a development version
 // (like 0.30.0-dev-abcd012).
 func Version() string {
 	v := version
-	if strings.HasSuffix(version, "-dev") && GitSha1 != "" {
-		v += "-" + GitSha1
+	if strings.HasSuffix(version, "-dev") {
+		if hash := readGitHash(); hash != "" {
+			v += "-" + hash
+		}
 	}
 	return v
 }

+func readGitHash() string {
+	if info, ok := debug.ReadBuildInfo(); ok {
+		for _, setting := range info.Settings {
+			if setting.Key == "vcs.revision" {
+				return setting.Value[:8]
+			}
+		}
+	}
+	return ""
+}
+
 // GetGorootVersion returns the major and minor version for a given GOROOT path.
 // If the goroot cannot be determined, (0, 0) is returned.
 func GetGorootVersion() (major, minor int, err error) {
@@ -42,6 +50,9 @@ func GetGorootVersion() (major, minor int, err error) {
 // major, minor, and patch version: 1, 3, and 2 in this example.
 // If there is an error, (0, 0, 0) and an error will be returned.
 func Parse(version string) (major, minor, patch int, err error) {
+	if strings.HasPrefix(version, "devel ") {
+		version = strings.Split(strings.TrimPrefix(version, "devel "), version)[0]
+	}
 	if version == "" || version[:2] != "go" {
 		return 0, 0, 0, errors.New("could not parse Go version: version does not start with 'go' prefix")
 	}
@@ -21,6 +21,7 @@ func TestParse(t *testing.T) {
 		{"go1.23.5-rc6", 1, 23, 5, false},
 		{"go2.0", 2, 0, 0, false},
 		{"go2.0.15", 2, 0, 15, false},
+		{"devel go1.24-f99f5da18f Thu Nov 14 22:29:26 2024 +0000 darwin/arm64", 1, 24, 0, false},
 	}
 	for _, tt := range tests {
 		t.Run(tt.v, func(t *testing.T) {
@@ -2,7 +2,7 @@ module github.com/tinygo-org/tinygo/internal/tools

 go 1.22.4

-require github.com/bytecodealliance/wasm-tools-go v0.3.0
+require github.com/bytecodealliance/wasm-tools-go v0.3.1

 require (
 	github.com/coreos/go-semver v0.3.1 // indirect
@@ -12,8 +12,7 @@ require (
 	github.com/regclient/regclient v0.7.1 // indirect
 	github.com/sirupsen/logrus v1.9.3 // indirect
 	github.com/ulikunitz/xz v0.5.12 // indirect
-	github.com/urfave/cli/v3 v3.0.0-alpha9 // indirect
-	github.com/xrash/smetrics v0.0.0-20201216005158-039620a65673 // indirect
+	github.com/urfave/cli/v3 v3.0.0-alpha9.2 // indirect
 	golang.org/x/mod v0.21.0 // indirect
 	golang.org/x/sys v0.26.0 // indirect
 )
@@ -1,5 +1,5 @@
-github.com/bytecodealliance/wasm-tools-go v0.3.0 h1:9aeDFYpbi3gtIW/nJCH+P+LhFMqezGoOfzqbUZLadho=
-github.com/bytecodealliance/wasm-tools-go v0.3.0/go.mod h1:VY+9FlpLi6jnhCrZLkyJjF9rjU4aEekgaRTk28MS2JE=
+github.com/bytecodealliance/wasm-tools-go v0.3.1 h1:9Q9PjSzkbiVmkUvZ7nYCfJ02mcQDBalxycA3s8g7kR4=
+github.com/bytecodealliance/wasm-tools-go v0.3.1/go.mod h1:vNAQ8DAEp6xvvk+TUHah5DslLEa76f4H6e737OeaxuY=
 github.com/coreos/go-semver v0.3.1 h1:yi21YpKnrx1gt5R+la8n5WgS0kCrsPp33dmEyHReZr4=
 github.com/coreos/go-semver v0.3.1/go.mod h1:irMmmIw/7yzSRPWryHsK7EYSg09caPQL03VsM8rvUec=
 github.com/davecgh/go-spew v1.1.0/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
@@ -23,14 +23,12 @@ github.com/sirupsen/logrus v1.9.3 h1:dueUQJ1C2q9oE3F7wvmSGAaVtTmUizReu6fjN8uqzbQ
 github.com/sirupsen/logrus v1.9.3/go.mod h1:naHLuLoDiP4jHNo9R0sCBMtWGeIprob74mVsIT4qYEQ=
 github.com/stretchr/objx v0.1.0/go.mod h1:HFkY916IF+rwdDfMAkV7OtwuqBVzrE8GR6GFx+wExME=
 github.com/stretchr/testify v1.7.0/go.mod h1:6Fq8oRcR53rry900zMqJjRRixrwX3KX962/h/Wwjteg=
-github.com/stretchr/testify v1.8.4 h1:CcVxjf3Q8PM0mHUKJCdn+eZZtm5yQwehR5yeSVQQcUk=
-github.com/stretchr/testify v1.8.4/go.mod h1:sz/lmYIOXD/1dqDmKjjqLyZ2RngseejIcXlSw2iwfAo=
+github.com/stretchr/testify v1.9.0 h1:HtqpIVDClZ4nwg75+f6Lvsy/wHu+3BoSGCbBAcpTsTg=
+github.com/stretchr/testify v1.9.0/go.mod h1:r2ic/lqez/lEtzL7wO/rwa5dbSLXVDPFyf8C91i36aY=
 github.com/ulikunitz/xz v0.5.12 h1:37Nm15o69RwBkXM0J6A5OlE67RZTfzUxTj8fB3dfcsc=
 github.com/ulikunitz/xz v0.5.12/go.mod h1:nbz6k7qbPmH4IRqmfOplQw/tblSgqTqBwxkY0oWt/14=
-github.com/urfave/cli/v3 v3.0.0-alpha9 h1:P0RMy5fQm1AslQS+XCmy9UknDXctOmG/q/FZkUFnJSo=
-github.com/urfave/cli/v3 v3.0.0-alpha9/go.mod h1:0kK/RUFHyh+yIKSfWxwheGndfnrvYSmYFVeKCh03ZUc=
-github.com/xrash/smetrics v0.0.0-20201216005158-039620a65673 h1:bAn7/zixMGCfxrRTfdpNzjtPYqr8smhKouy9mxVdGPU=
-github.com/xrash/smetrics v0.0.0-20201216005158-039620a65673/go.mod h1:N3UwUGtsrSj3ccvlPHLoLsHnpR27oXr4ZE984MbSER8=
+github.com/urfave/cli/v3 v3.0.0-alpha9.2 h1:CL8llQj3dGRLVQQzHxS+ZYRLanOuhyK1fXgLKD+qV+Y=
+github.com/urfave/cli/v3 v3.0.0-alpha9.2/go.mod h1:FnIeEMYu+ko8zP1F9Ypr3xkZMIDqW3DR92yUtY39q1Y=
 golang.org/x/mod v0.21.0 h1:vvrHzRwRfVKSiLrG+d4FMl/Qi4ukBCE6kZlTUkDYRT0=
 golang.org/x/mod v0.21.0/go.mod h1:6SkKJ3Xj0I0BrPOZoBy3bdMptDDU9oJrpohJ3eWZ1fY=
 golang.org/x/sync v0.8.0 h1:3NFvSEYkUoMifnESzZl15y791HH1qU2xm6eCJU5ZPXQ=
@@ -287,9 +287,17 @@ func (r *runner) run(fn *function, params []value, parentMem *memoryView, indent
 				// Get the object layout, if it is available.
 				llvmLayoutType := r.getLLVMTypeFromLayout(operands[2])

+				// Get the alignment of the memory to be allocated.
+				alignment := 0 // use default alignment if unset
+				alignAttr := inst.llvmInst.GetCallSiteEnumAttribute(0, llvm.AttributeKindID("align"))
+				if !alignAttr.IsNil() {
+					alignment = int(alignAttr.GetEnumValue())
+				}
+
 				// Create the object.
 				alloc := object{
 					globalName:     r.pkgName + "$alloc",
+					align:          alignment,
 					llvmLayoutType: llvmLayoutType,
 					buffer:         newRawValue(uint32(size)),
 					size:           uint32(size),
@@ -646,6 +654,7 @@ func (r *runner) run(fn *function, params []value, parentMem *memoryView, indent
 				globalName: r.pkgName + "$alloca",
 				buffer:     newRawValue(uint32(size)),
 				size:       uint32(size),
+				align:      inst.llvmInst.Alignment(),
 			}
 			index := len(r.objects)
 			r.objects = append(r.objects, alloca)
@@ -42,6 +42,7 @@ type object struct {
 	globalName     string    // name, if not yet created (not guaranteed to be the final name)
 	buffer         value     // buffer with value as given by interp, nil if external
 	size           uint32    // must match buffer.len(), if available
+	align          int       // alignment of the object (may be 0 if unknown)
 	constant       bool      // true if this is a constant global
 	marked         uint8     // 0 means unmarked, 1 means external read, 2 means external write
 }
@@ -593,6 +594,12 @@ func (v pointerValue) toLLVMValue(llvmType llvm.Type, mem *memoryView) (llvm.Val
 		// runtime.alloc.
 		// First allocate a new global for this object.
 		obj := mem.get(v.index())
+		alignment := obj.align
+		if alignment == 0 {
+			// Unknown alignment, perhaps from a direct call to runtime.alloc in
+			// the runtime. Use a conservative default instead.
+			alignment = mem.r.maxAlign
+		}
 		if obj.llvmType.IsNil() && obj.llvmLayoutType.IsNil() {
 			// Create an initializer without knowing the global type.
 			// This is probably the result of a runtime.alloc call.
@@ -603,7 +610,7 @@ func (v pointerValue) toLLVMValue(llvmType llvm.Type, mem *memoryView) (llvm.Val
 			globalType := initializer.Type()
 			llvmValue = llvm.AddGlobal(mem.r.mod, globalType, obj.globalName)
 			llvmValue.SetInitializer(initializer)
-			llvmValue.SetAlignment(mem.r.maxAlign)
+			llvmValue.SetAlignment(alignment)
 			obj.llvmGlobal = llvmValue
 			mem.put(v.index(), obj)
 		} else {
@@ -642,11 +649,7 @@ func (v pointerValue) toLLVMValue(llvmType llvm.Type, mem *memoryView) (llvm.Val
 				return llvm.Value{}, errors.New("interp: allocated value does not match allocated type")
 			}
 			llvmValue.SetInitializer(initializer)
-			if obj.llvmType.IsNil() {
-				// The exact type isn't known (only the layout), so use the
-				// alignment that would normally be expected from runtime.alloc.
-				llvmValue.SetAlignment(mem.r.maxAlign)
-			}
+			llvmValue.SetAlignment(alignment)
 		}

 		// It should be included in r.globals because otherwise markExternal
@@ -243,6 +243,7 @@ func pathsToOverride(goMinor int, needsSyscallPackage bool) map[string]bool {
 		"internal/binary/":            false,
 		"internal/bytealg/":           false,
 		"internal/cm/":                false,
+		"internal/futex/":             false,
 		"internal/fuzz/":              false,
 		"internal/reflectlite/":       false,
 		"internal/gclayout":           false,
@@ -256,6 +257,7 @@ func pathsToOverride(goMinor int, needsSyscallPackage bool) map[string]bool {
 		"runtime/":                    false,
 		"sync/":                       true,
 		"testing/":                    true,
+		"tinygo/":                     false,
 		"unique/":                     false,
 	}

@@ -485,7 +485,7 @@ func (p *Package) parseFiles() ([]*ast.File, error) {
 		var initialCFlags []string
 		initialCFlags = append(initialCFlags, p.program.config.CFlags(true)...)
 		initialCFlags = append(initialCFlags, "-I"+p.Dir)
-		generated, headerCode, cflags, ldflags, accessedFiles, errs := cgo.Process(files, p.program.workingDir, p.ImportPath, p.program.fset, initialCFlags)
+		generated, headerCode, cflags, ldflags, accessedFiles, errs := cgo.Process(files, p.program.workingDir, p.ImportPath, p.program.fset, initialCFlags, p.program.config.GOOS())
 		p.CFlags = append(initialCFlags, cflags...)
 		p.CGoHeaders = headerCode
 		for path, hash := range accessedFiles {
@@ -1509,7 +1509,7 @@ func main() {
 		stackSize = uint64(size)
 		return err
 	})
-	printSize := flag.String("size", "", "print sizes (none, short, full)")
+	printSize := flag.String("size", "", "print sizes (none, short, full, html)")
 	printStacks := flag.Bool("print-stacks", false, "print stack sizes of goroutines")
 	printAllocsString := flag.String("print-allocs", "", "regular expression of functions for which heap allocations should be printed")
 	printCommands := flag.Bool("x", false, "Print commands")
@@ -1641,12 +1641,19 @@ func main() {
 		Timeout:         *timeout,
 		WITPackage:      witPackage,
 		WITWorld:        witWorld,
-		ExtLDFlags:      extLDFlags,
 	}
 	if *printCommands {
 		options.PrintCommands = printCommand
 	}

+	if extLDFlags != "" {
+		options.ExtLDFlags, err = shlex.Split(extLDFlags)
+		if err != nil {
+			fmt.Fprintln(os.Stderr, "could not parse -extldflags:", err)
+			os.Exit(1)
+		}
+	}
+
 	err = options.Verify()
 	if err != nil {
 		fmt.Fprintln(os.Stderr, err.Error())
@@ -15,7 +15,6 @@ import (
 	"reflect"
 	"regexp"
 	"runtime"
-	"slices"
 	"strings"
 	"sync"
 	"testing"
@@ -25,6 +24,7 @@ import (
 	"github.com/tetratelabs/wazero"
 	"github.com/tetratelabs/wazero/api"
 	"github.com/tetratelabs/wazero/imports/wasi_snapshot_preview1"
+	"github.com/tetratelabs/wazero/sys"
 	"github.com/tinygo-org/tinygo/builder"
 	"github.com/tinygo-org/tinygo/compileopts"
 	"github.com/tinygo-org/tinygo/diagnostics"
@@ -429,6 +429,9 @@ func runTestWithConfig(name string, t *testing.T, options compileopts.Options, c
 		for _, line := range strings.Split(strings.TrimRight(w.String(), "\n"), "\n") {
 			t.Log(line)
 		}
+		if stdout.Len() != 0 {
+			t.Logf("output:\n%s", stdout.String())
+		}
 		t.Fail()
 		return
 	}
@@ -517,7 +520,7 @@ func TestWebAssembly(t *testing.T) {
 						}
 					}
 				}
-				if !slices.Equal(imports, tc.imports) {
+				if !stringSlicesEqual(imports, tc.imports) {
 					t.Errorf("import list not as expected!\nexpected: %v\nactual:   %v", tc.imports, imports)
 				}
 			}
@@ -525,6 +528,20 @@ func TestWebAssembly(t *testing.T) {
 	}
 }

+func stringSlicesEqual(s1, s2 []string) bool {
+	// We can use slices.Equal once we drop support for Go 1.20 (it was added in
+	// Go 1.21).
+	if len(s1) != len(s2) {
+		return false
+	}
+	for i, s := range s1 {
+		if s != s2[i] {
+			return false
+		}
+	}
+	return true
+}
+
 func TestWasmExport(t *testing.T) {
 	t.Parallel()

@@ -683,7 +700,14 @@ func TestWasmExport(t *testing.T) {
 			if tc.command {
 				// Call _start (the entry point), which calls
 				// tester.callTestMain, which then runs all the tests.
-				mustCall(mod.ExportedFunction("_start").Call(ctx))
+				_, err := mod.ExportedFunction("_start").Call(ctx)
+				if err != nil {
+					if exitErr, ok := err.(*sys.ExitError); ok && exitErr.ExitCode() == 0 {
+						// Exited with code 0. Nothing to worry about.
+					} else {
+						t.Error("failed to run _start:", err)
+					}
+				}
 			} else {
 				// Run the _initialize call, because this is reactor mode wasm.
 				mustCall(mod.ExportedFunction("_initialize").Call(ctx))
@@ -728,6 +752,7 @@ func TestWasmFuncOf(t *testing.T) {

 // Test //go:wasmexport in JavaScript (using NodeJS).
 func TestWasmExportJS(t *testing.T) {
+	t.Parallel()
 	type testCase struct {
 		name      string
 		buildMode string
@@ -738,7 +763,9 @@ func TestWasmExportJS(t *testing.T) {
 		{name: "c-shared", buildMode: "c-shared"},
 	}
 	for _, tc := range tests {
+		tc := tc
 		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
 			// Build the wasm binary.
 			tmpdir := t.TempDir()
 			options := optionsFromTarget("wasm", sema)
@@ -766,12 +793,56 @@ func TestWasmExportJS(t *testing.T) {
 	}
 }

+// Test whether Go.run() (in wasm_exec.js) normally returns and returns the
+// right exit code.
+func TestWasmExit(t *testing.T) {
+	t.Parallel()
+
+	type testCase struct {
+		name   string
+		output string
+	}
+
+	tests := []testCase{
+		{name: "normal", output: "exit code: 0\n"},
+		{name: "exit-0", output: "exit code: 0\n"},
+		{name: "exit-0-sleep", output: "slept\nexit code: 0\n"},
+		{name: "exit-1", output: "exit code: 1\n"},
+		{name: "exit-1-sleep", output: "slept\nexit code: 1\n"},
+	}
+	for _, tc := range tests {
+		tc := tc
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			options := optionsFromTarget("wasm", sema)
+			buildConfig, err := builder.NewConfig(&options)
+			if err != nil {
+				t.Fatal(err)
+			}
+			buildConfig.Target.Emulator = "node testdata/wasmexit.js {}"
+			output := &bytes.Buffer{}
+			_, err = buildAndRun("testdata/wasmexit.go", buildConfig, output, []string{tc.name}, nil, time.Minute, func(cmd *exec.Cmd, result builder.BuildResult) error {
+				return cmd.Run()
+			})
+			if err != nil {
+				t.Error(err)
+			}
+			expected := "wasmexit test: " + tc.name + "\n" + tc.output
+			checkOutputData(t, []byte(expected), output.Bytes())
+		})
+	}
+}
+
 // Check whether the output of a test equals the expected output.
 func checkOutput(t *testing.T, filename string, actual []byte) {
 	expectedOutput, err := os.ReadFile(filename)
 	if err != nil {
 		t.Fatal("could not read output file:", err)
 	}
+	checkOutputData(t, expectedOutput, actual)
+}
+
+func checkOutputData(t *testing.T, expectedOutput, actual []byte) {
 	expectedOutput = bytes.ReplaceAll(expectedOutput, []byte("\r\n"), []byte("\n"))
 	actual = bytes.ReplaceAll(actual, []byte("\r\n"), []byte("\n"))

@@ -1,4 +1,4 @@
-//go:build nrf || (stm32 && !(stm32f103 || stm32l0x1)) || (sam && atsamd51) || (sam && atsame5x) || esp32c3
+//go:build nrf || (stm32 && !(stm32f103 || stm32l0x1)) || (sam && atsamd51) || (sam && atsame5x) || esp32c3 || tkey

 // If you update the above build constraint, you'll probably also need to update
 // src/runtime/rand_hwrng.go.
@@ -0,0 +1,139 @@
+//go:build tkey
+
+// Hand written file based on https://github.com/tillitis/tkey-libs/blob/main/include/tkey/tk1_mem.h
+
+package tkey
+
+import (
+	"runtime/volatile"
+	"unsafe"
+)
+
+// Peripherals
+var (
+	TRNG = (*TRNG_Type)(unsafe.Pointer(TK1_MMIO_TRNG_BASE))
+
+	TIMER = (*TIMER_Type)(unsafe.Pointer(TK1_MMIO_TIMER_BASE))
+
+	UDS = (*UDS_Type)(unsafe.Pointer(TK1_MMIO_UDS_BASE))
+
+	UART = (*UART_Type)(unsafe.Pointer(TK1_MMIO_UART_BASE))
+
+	TOUCH = (*TOUCH_Type)(unsafe.Pointer(TK1_MMIO_TOUCH_BASE))
+
+	TK1 = (*TK1_Type)(unsafe.Pointer(TK1_MMIO_TK1_BASE))
+)
+
+// Memory sections
+const (
+	TK1_ROM_BASE uintptr = 0x00000000
+
+	TK1_RAM_BASE uintptr = 0x40000000
+
+	TK1_MMIO_BASE uintptr = 0xc0000000
+
+	TK1_MMIO_TRNG_BASE uintptr = 0xc0000000
+
+	TK1_MMIO_TIMER_BASE uintptr = 0xc1000000
+
+	TK1_MMIO_UDS_BASE uintptr = 0xc2000000
+
+	TK1_MMIO_UART_BASE uintptr = 0xc3000000
+
+	TK1_MMIO_TOUCH_BASE uintptr = 0xc4000000
+
+	TK1_MMIO_FW_RAM_BASE uintptr = 0xd0000000
+
+	TK1_MMIO_TK1_BASE uintptr = 0xff000000
+)
+
+// Memory section sizes
+const (
+	TK1_RAM_SIZE uintptr = 0x20000
+
+	TK1_MMIO_SIZE uintptr = 0x3fffffff
+)
+
+type TRNG_Type struct {
+	_       [36]byte
+	STATUS  volatile.Register32
+	_       [88]byte
+	ENTROPY volatile.Register32
+}
+
+type TIMER_Type struct {
+	_         [32]byte
+	CTRL      volatile.Register32
+	STATUS    volatile.Register32
+	PRESCALER volatile.Register32
+	TIMER     volatile.Register32
+}
+
+type UDS_Type struct {
+	_    [64]byte
+	DATA [8]volatile.Register32
+}
+
+type UART_Type struct {
+	_         [128]byte
+	RX_STATUS volatile.Register32
+	RX_DATA   volatile.Register32
+	RX_BYTES  volatile.Register32
+	_         [116]byte
+	TX_STATUS volatile.Register32
+	TX_DATA   volatile.Register32
+}
+
+type TOUCH_Type struct {
+	_      [36]byte
+	STATUS volatile.Register32
+}
+
+type TK1_Type struct {
+	NAME0         volatile.Register32
+	NAME1         volatile.Register32
+	VERSION       volatile.Register32
+	_             [16]byte
+	SWITCH_APP    volatile.Register32
+	_             [4]byte
+	LED           volatile.Register32
+	GPIO          volatile.Register16
+	APP_ADDR      volatile.Register32
+	APP_SIZE      volatile.Register32
+	BLAKE2S       volatile.Register32
+	_             [72]byte
+	CDI_FIRST     [8]volatile.Register32
+	_             [32]byte
+	UDI_FIRST     [2]volatile.Register32
+	_             [62]byte
+	RAM_ADDR_RAND volatile.Register16
+	_             [2]byte
+	RAM_DATA_RAND volatile.Register16
+	_             [126]byte
+	CPU_MON_CTRL  volatile.Register16
+	_             [2]byte
+	CPU_MON_FIRST volatile.Register32
+	CPU_MON_LAST  volatile.Register32
+	_             [60]byte
+	SYSTEM_RESET  volatile.Register16
+	_             [66]byte
+	SPI_EN        volatile.Register32
+	SPI_XFER      volatile.Register32
+	SPI_DATA      volatile.Register32
+}
+
+const (
+	TK1_MMIO_TIMER_CTRL_START_BIT = 0
+	TK1_MMIO_TIMER_CTRL_STOP_BIT  = 1
+	TK1_MMIO_TIMER_CTRL_START     = 1 << TK1_MMIO_TIMER_CTRL_START_BIT
+	TK1_MMIO_TIMER_CTRL_STOP      = 1 << TK1_MMIO_TIMER_CTRL_STOP_BIT
+
+	TK1_MMIO_TK1_LED_R_BIT = 2
+	TK1_MMIO_TK1_LED_G_BIT = 1
+	TK1_MMIO_TK1_LED_B_BIT = 0
+
+	TK1_MMIO_TK1_GPIO1_BIT = 0
+	TK1_MMIO_TK1_GPIO2_BIT = 1
+	TK1_MMIO_TK1_GPIO3_BIT = 2
+	TK1_MMIO_TK1_GPIO4_BIT = 3
+)
@@ -7,15 +7,13 @@ import (
 	"time"
 )

-// change these to test a different UART or pins if available
 var (
 	uart = machine.Serial
-	tx   = machine.UART_TX_PIN
-	rx   = machine.UART_RX_PIN
 )

 func main() {
-	uart.Configure(machine.UARTConfig{TX: tx, RX: rx})
+	// use default settings for UART
+	uart.Configure(machine.UARTConfig{})
 	uart.Write([]byte("Echo console enabled. Type something then press enter:\r\n"))

 	input := make([]byte, 64)
@@ -13,11 +13,11 @@ func main() {
 		Stereo:      true,
 	})

-	data := make([]uint32, 64)
+	data := make([]uint16, 64)

 	for {
 		// get the next group of samples
-		machine.I2S0.Read(data)
+		machine.I2S0.ReadMono(data)

 		println("data", data[0], data[1], data[2], data[4], "...")
 	}
@@ -71,6 +71,16 @@ func (r *result[Shape, OK, Err]) Err() *Err {
 	return (*Err)(unsafe.Pointer(&r.data))
 }

+// Result returns (OK, zero value of Err, false) if r represents the OK case,
+// or (zero value of OK, Err, true) if r represents the error case.
+// This does not have a pointer receiver, so it can be chained.
+func (r result[Shape, OK, Err]) Result() (ok OK, err Err, isErr bool) {
+	if r.isErr {
+		return ok, *(*Err)(unsafe.Pointer(&r.data)), true
+	}
+	return *(*OK)(unsafe.Pointer(&r.data)), err, false
+}
+
 // This function is sized so it can be inlined and optimized away.
 func (r *result[Shape, OK, Err]) validate() {
 	var shape Shape
@@ -0,0 +1,72 @@
+package futex
+
+// Cross platform futex implementation.
+// Futexes are supported on all major operating systems and on WebAssembly.
+//
+// For more information, see: https://outerproduct.net/futex-dictionary.html
+
+import (
+	"sync/atomic"
+	"unsafe"
+)
+
+// A futex is a way for userspace to wait with the pointer as the key, and for
+// another thread to wake one or all waiting threads keyed on the same pointer.
+//
+// A futex does not change the underlying value, it only reads it before going
+// to sleep (atomically) to prevent lost wake-ups.
+type Futex struct {
+	atomic.Uint32
+}
+
+// Atomically check for cmp to still be equal to the futex value and if so, go
+// to sleep. Return true if we were definitely awoken by a call to Wake or
+// WakeAll, and false if we can't be sure of that.
+func (f *Futex) Wait(cmp uint32) bool {
+	tinygo_futex_wait((*uint32)(unsafe.Pointer(&f.Uint32)), cmp)
+
+	// We *could* detect a zero return value from the futex system call which
+	// would indicate we got awoken by a Wake or WakeAll call. However, this is
+	// what the manual page has to say:
+	//
+	// > Note that a wake-up can also be caused by common futex usage patterns
+	// > in unrelated code that happened to have previously used the futex
+	// > word's memory location (e.g., typical futex-based implementations of
+	// > Pthreads mutexes can cause this under some conditions). Therefore,
+	// > callers should always conservatively assume that a return value of 0
+	// > can mean a spurious wake-up, and use the futex word's value (i.e., the
+	// > user-space synchronization scheme) to decide whether to continue to
+	// > block or not.
+	//
+	// I'm not sure whether we do anything like pthread does, so to be on the
+	// safe side we say we don't know whether the wakeup was spurious or not and
+	// return false.
+	return false
+}
+
+// Like Wait, but times out after the number of nanoseconds in timeout.
+func (f *Futex) WaitUntil(cmp uint32, timeout uint64) {
+	tinygo_futex_wait_timeout((*uint32)(unsafe.Pointer(&f.Uint32)), cmp, timeout)
+}
+
+// Wake a single waiter.
+func (f *Futex) Wake() {
+	tinygo_futex_wake((*uint32)(unsafe.Pointer(&f.Uint32)))
+}
+
+// Wake all waiters.
+func (f *Futex) WakeAll() {
+	tinygo_futex_wake_all((*uint32)(unsafe.Pointer(&f.Uint32)))
+}
+
+//export tinygo_futex_wait
+func tinygo_futex_wait(addr *uint32, cmp uint32)
+
+//export tinygo_futex_wait_timeout
+func tinygo_futex_wait_timeout(addr *uint32, cmp uint32, timeout uint64)
+
+//export tinygo_futex_wake
+func tinygo_futex_wake(addr *uint32)
+
+//export tinygo_futex_wake_all
+func tinygo_futex_wake_all(addr *uint32)
@@ -0,0 +1,49 @@
+//go:build none
+
+// This file is manually included, to avoid CGo which would cause a circular
+// import.
+
+#include <stdint.h>
+
+// This API isn't documented by Apple, but it is used by LLVM libc++ (so should
+// be stable) and has been documented extensively here:
+// https://outerproduct.net/futex-dictionary.html
+
+int __ulock_wait(uint32_t operation, void *addr, uint64_t value, uint32_t timeout_us);
+int __ulock_wait2(uint32_t operation, void *addr, uint64_t value, uint64_t timeout_ns, uint64_t value2);
+int __ulock_wake(uint32_t operation, void *addr, uint64_t wake_value);
+
+// Operation code.
+#define UL_COMPARE_AND_WAIT 1
+
+// Flags to the operation value.
+#define ULF_WAKE_ALL 0x00000100
+#define ULF_NO_ERRNO 0x01000000
+
+void tinygo_futex_wait(uint32_t *addr, uint32_t cmp) {
+    __ulock_wait(UL_COMPARE_AND_WAIT|ULF_NO_ERRNO, addr, (uint64_t)cmp, 0);
+}
+
+void tinygo_futex_wait_timeout(uint32_t *addr, uint32_t cmp, uint64_t timeout) {
+    // Make sure that an accidental use of a zero timeout is not treated as an
+    // infinite timeout. Return if it's zero since it wouldn't be waiting for
+    // any significant time anyway.
+    // Probably unnecessary, but guards against potential bugs.
+    if (timeout == 0) {
+        return;
+    }
+
+    // Note: __ulock_wait2 is available since MacOS 11.
+    // I think that's fine, since the version before that (MacOS 10.15) is EOL
+    // since 2022. Though if needed, we could certainly use __ulock_wait instead
+    // and deal with the smaller timeout value.
+    __ulock_wait2(UL_COMPARE_AND_WAIT|ULF_NO_ERRNO, addr, (uint64_t)cmp, timeout, 0);
+}
+
+void tinygo_futex_wake(uint32_t *addr) {
+    __ulock_wake(UL_COMPARE_AND_WAIT|ULF_NO_ERRNO, addr, 0);
+}
+
+void tinygo_futex_wake_all(uint32_t *addr) {
+    __ulock_wake(UL_COMPARE_AND_WAIT|ULF_NO_ERRNO|ULF_WAKE_ALL, addr, 0);
+}
@@ -0,0 +1,33 @@
+//go:build none
+
+// This file is manually included, to avoid CGo which would cause a circular
+// import.
+
+#include <limits.h>
+#include <stdint.h>
+#include <sys/syscall.h>
+#include <time.h>
+#include <unistd.h>
+
+#define FUTEX_WAIT         0
+#define FUTEX_WAKE         1
+#define FUTEX_PRIVATE_FLAG 128
+
+void tinygo_futex_wait(uint32_t *addr, uint32_t cmp) {
+    syscall(SYS_futex, addr, FUTEX_WAIT|FUTEX_PRIVATE_FLAG, cmp, NULL, NULL, 0);
+}
+
+void tinygo_futex_wait_timeout(uint32_t *addr, uint32_t cmp, uint64_t timeout) {
+    struct timespec ts = {0};
+    ts.tv_sec = timeout / 1000000000;
+    ts.tv_nsec = timeout % 1000000000;
+    syscall(SYS_futex, addr, FUTEX_WAIT|FUTEX_PRIVATE_FLAG, cmp, &ts, NULL, 0);
+}
+
+void tinygo_futex_wake(uint32_t *addr) {
+    syscall(SYS_futex, addr, FUTEX_WAKE|FUTEX_PRIVATE_FLAG, 1, NULL, NULL, 0);
+}
+
+void tinygo_futex_wake_all(uint32_t *addr) {
+    syscall(SYS_futex, addr, FUTEX_WAKE|FUTEX_PRIVATE_FLAG, INT_MAX, NULL, NULL, 0);
+}
@@ -0,0 +1,41 @@
+package task
+
+// Atomics implementation for cooperative systems. The atomic types here aren't
+// actually atomic, they assume that accesses cannot be interrupted by a
+// different goroutine or interrupt happening at the same time.
+
+type atomicIntegerType interface {
+	uintptr | uint32 | uint64
+}
+
+type pseudoAtomic[T atomicIntegerType] struct {
+	v T
+}
+
+func (x *pseudoAtomic[T]) Add(delta T) T { x.v += delta; return x.v }
+func (x *pseudoAtomic[T]) Load() T       { return x.v }
+func (x *pseudoAtomic[T]) Store(val T)   { x.v = val }
+func (x *pseudoAtomic[T]) CompareAndSwap(old, new T) (swapped bool) {
+	if x.v != old {
+		return false
+	}
+	x.v = new
+	return true
+}
+func (x *pseudoAtomic[T]) Swap(new T) (old T) {
+	old = x.v
+	x.v = new
+	return
+}
+
+// Uintptr is an atomic uintptr when multithreading is enabled, and a plain old
+// uintptr otherwise.
+type Uintptr = pseudoAtomic[uintptr]
+
+// Uint32 is an atomic uint32 when multithreading is enabled, and a plain old
+// uint32 otherwise.
+type Uint32 = pseudoAtomic[uint32]
+
+// Uint64 is an atomic uint64 when multithreading is enabled, and a plain old
+// uint64 otherwise.
+type Uint64 = pseudoAtomic[uint64]
@@ -0,0 +1,44 @@
+package task
+
+// A futex is a way for userspace to wait with the pointer as the key, and for
+// another thread to wake one or all waiting threads keyed on the same pointer.
+//
+// A futex does not change the underlying value, it only reads it before to prevent
+// lost wake-ups.
+type Futex struct {
+	Uint32
+	waiters Stack
+}
+
+// Atomically check for cmp to still be equal to the futex value and if so, go
+// to sleep. Return true if we were definitely awoken by a call to Wake or
+// WakeAll, and false if we can't be sure of that.
+func (f *Futex) Wait(cmp uint32) (awoken bool) {
+	if f.Uint32.v != cmp {
+		return false
+	}
+
+	// Push the current goroutine onto the waiter stack.
+	f.waiters.Push(Current())
+
+	// Pause until the waiters are awoken by Wake/WakeAll.
+	Pause()
+
+	// We were awoken by a call to Wake or WakeAll. There is no chance for
+	// spurious wakeups.
+	return true
+}
+
+// Wake a single waiter.
+func (f *Futex) Wake() {
+	if t := f.waiters.Pop(); t != nil {
+		scheduleTask(t)
+	}
+}
+
+// Wake all waiters.
+func (f *Futex) WakeAll() {
+	for t := f.waiters.Pop(); t != nil; t = f.waiters.Pop() {
+		scheduleTask(t)
+	}
+}
@@ -0,0 +1,16 @@
+package task
+
+// PMutex is a real mutex on systems that can be either preemptive or threaded,
+// and a dummy lock on other (purely cooperative) systems.
+//
+// It is mainly useful for short operations that need a lock when threading may
+// be involved, but which do not need a lock with a purely cooperative
+// scheduler.
+type PMutex struct {
+}
+
+func (m *PMutex) Lock() {
+}
+
+func (m *PMutex) Unlock() {
+}
@@ -26,6 +26,23 @@ type Task struct {
 	DeferFrame unsafe.Pointer
 }

+// DataUint32 returns the Data field as a uint32. The value is only valid after
+// setting it through SetDataUint32 or by storing to it using DataAtomicUint32.
+func (t *Task) DataUint32() uint32 {
+	return *(*uint32)(unsafe.Pointer(&t.Data))
+}
+
+// SetDataUint32 updates the uint32 portion of the Data field (which could be
+// the first 4 or last 4 bytes depending on the architecture endianness).
+func (t *Task) SetDataUint32(val uint32) {
+	*(*uint32)(unsafe.Pointer(&t.Data)) = val
+}
+
+// DataAtomicUint32 returns the Data field as an atomic-if-needed Uint32 value.
+func (t *Task) DataAtomicUint32() *Uint32 {
+	return (*Uint32)(unsafe.Pointer(&t.Data))
+}
+
 // getGoroutineStackSize is a compiler intrinsic that returns the stack size for
 // the given function and falls back to the default stack size. It is replaced
 // with a load from a special section just before codegen.
@@ -33,3 +50,6 @@ func getGoroutineStackSize(fn uintptr) uintptr

 //go:linkname runtime_alloc runtime.alloc
 func runtime_alloc(size uintptr, layout unsafe.Pointer) unsafe.Pointer
+
+//go:linkname scheduleTask runtime.scheduleTask
+func scheduleTask(*Task)
@@ -52,7 +52,7 @@ type stackState struct {
 func start(fn uintptr, args unsafe.Pointer, stackSize uintptr) {
 	t := &Task{}
 	t.state.initialize(fn, args, stackSize)
-	runqueuePushBack(t)
+	scheduleTask(t)
 }

 //export tinygo_launch
@@ -82,9 +82,6 @@ func (s *state) initialize(fn uintptr, args unsafe.Pointer, stackSize uintptr) {
 	s.csp = unsafe.Add(stack, stackSize)
 }

-//go:linkname runqueuePushBack runtime.runqueuePushBack
-func runqueuePushBack(*Task)
-
 // currentTask is the current running task, or nil if currently in the scheduler.
 var currentTask *Task

@@ -101,15 +101,12 @@ func swapTask(oldStack uintptr, newStack *uintptr)
 //go:extern tinygo_startTask
 var startTask [0]uint8

-//go:linkname runqueuePushBack runtime.runqueuePushBack
-func runqueuePushBack(*Task)
-
 // start creates and starts a new goroutine with the given function and arguments.
 // The new goroutine is scheduled to run later.
 func start(fn uintptr, args unsafe.Pointer, stackSize uintptr) {
 	t := &Task{}
 	t.state.initialize(fn, args, stackSize)
-	runqueuePushBack(t)
+	scheduleTask(t)
 }

 // OnSystemStack returns whether the caller is running on the system stack.
@@ -1,3 +1,7 @@
+// Do not reorder instructions to insert a branch delay slot.
+// We know what we're doing, and will manually fill the branch delay slot.
+.set noreorder
+
 .section .text.tinygo_startTask
 .global  tinygo_startTask
 .type    tinygo_startTask, %function
@@ -74,8 +74,9 @@ const (
 // I2S pins
 const (
 	I2S_SCK_PIN Pin = PA10
-	I2S_SD_PIN  Pin = PA07
-	I2S_WS_PIN      = NoPin // TODO: figure out what this is on Arduino Nano 33.
+	I2S_SDO_PIN Pin = PA07
+	I2S_SDI_PIN     = NoPin
+	I2S_WS_PIN      = NoPin // TODO: figure out what this is on Arduino MKR1000
 )

 // USB CDC identifiers
@@ -74,7 +74,8 @@ const (
 // I2S pins
 const (
 	I2S_SCK_PIN Pin = PA10
-	I2S_SD_PIN  Pin = PA07
+	I2S_SDO_PIN Pin = PA07
+	I2S_SDI_PIN     = NoPin
 	I2S_WS_PIN      = NoPin // TODO: figure out what this is on Arduino MKR WiFi 1010.
 )

@@ -118,7 +118,8 @@ const (
 // I2S pins
 const (
 	I2S_SCK_PIN Pin = PA10
-	I2S_SD_PIN  Pin = PA08
+	I2S_SDO_PIN Pin = PA08
+	I2S_SDI_PIN     = NoPin
 	I2S_WS_PIN      = NoPin // TODO: figure out what this is on Arduino Nano 33.
 )

@@ -69,7 +69,8 @@ const (
 // I2S pins - might not be exposed
 const (
 	I2S_SCK_PIN Pin = PA10
-	I2S_SD_PIN  Pin = PA07
+	I2S_SDO_PIN Pin = PA07
+	I2S_SDI_PIN     = NoPin
 	I2S_WS_PIN  Pin = PA11
 )

@@ -102,7 +102,8 @@ var SPI0 = sercomSPIM3
 // I2S pins
 const (
 	I2S_SCK_PIN = PA10
-	I2S_SD_PIN  = PA08
+	I2S_SDO_PIN = PA08
+	I2S_SDI_PIN = NoPin
 	I2S_WS_PIN  = NoPin // no WS, instead uses SCK to sync
 )

@@ -81,7 +81,8 @@ var SPI0 = sercomSPIM4
 // I2S pins
 const (
 	I2S_SCK_PIN = PA10
-	I2S_SD_PIN  = PA07
+	I2S_SDO_PIN = PA07
+	I2S_SDI_PIN = NoPin
 	I2S_WS_PIN  = NoPin // TODO: figure out what this is on Feather M0 Express.
 )

@@ -76,7 +76,8 @@ var SPI0 = sercomSPIM4
 // I2S pins
 const (
 	I2S_SCK_PIN = PA10
-	I2S_SD_PIN  = PA08
+	I2S_SDO_PIN = PA08
+	I2S_SDI_PIN = NoPin
 	I2S_WS_PIN  = NoPin // TODO: figure out what this is on Feather M0.
 )

@@ -76,7 +76,8 @@ var (
 // I2S (not connected, needed for atsamd21).
 const (
 	I2S_SCK_PIN = NoPin
-	I2S_SD_PIN  = NoPin
+	I2S_SDO_PIN = NoPin
+	I2S_SDI_PIN = NoPin
 	I2S_WS_PIN  = NoPin
 )

@@ -224,7 +224,8 @@ const (

 	I2S_SCK_PIN = I2S0_SCK_PIN // default pins
 	I2S_WS_PIN  = I2S0_FS_PIN  //
-	I2S_SD_PIN  = I2S0_SDO_PIN //
+	I2S_SDO_PIN = I2S0_SDO_PIN
+	I2S_SDI_PIN = NoPin
 )

 // SD card pins
@@ -89,7 +89,8 @@ var SPI1 = sercomSPIM5
 // I2S pins
 const (
 	I2S_SCK_PIN = PA10
-	I2S_SD_PIN  = PA08
+	I2S_SDO_PIN = PA08
+	I2S_SDI_PIN = NoPin
 	I2S_WS_PIN  = NoPin // TODO: figure out what this is on ItsyBitsy M0.
 )

@@ -123,7 +123,8 @@ var (
 // I2S pins
 const (
 	I2S_SCK_PIN Pin = D2
-	I2S_SD_PIN  Pin = A6
+	I2S_SDO_PIN Pin = A6
+	I2S_SDI_PIN     = NoPin
 	I2S_WS_PIN      = D3
 )

@@ -0,0 +1,88 @@
+//go:build pico2
+
+package machine
+
+// GPIO pins
+const (
+	GP0  Pin = GPIO0
+	GP1  Pin = GPIO1
+	GP2  Pin = GPIO2
+	GP3  Pin = GPIO3
+	GP4  Pin = GPIO4
+	GP5  Pin = GPIO5
+	GP6  Pin = GPIO6
+	GP7  Pin = GPIO7
+	GP8  Pin = GPIO8
+	GP9  Pin = GPIO9
+	GP10 Pin = GPIO10
+	GP11 Pin = GPIO11
+	GP12 Pin = GPIO12
+	GP13 Pin = GPIO13
+	GP14 Pin = GPIO14
+	GP15 Pin = GPIO15
+	GP16 Pin = GPIO16
+	GP17 Pin = GPIO17
+	GP18 Pin = GPIO18
+	GP19 Pin = GPIO19
+	GP20 Pin = GPIO20
+	GP21 Pin = GPIO21
+	GP22 Pin = GPIO22
+	GP26 Pin = GPIO26
+	GP27 Pin = GPIO27
+	GP28 Pin = GPIO28
+
+	// Onboard LED
+	LED Pin = GPIO25
+
+	// Onboard crystal oscillator frequency, in MHz.
+	xoscFreq = 12 // MHz
+)
+
+// I2C Default pins on Raspberry Pico.
+const (
+	I2C0_SDA_PIN = GP4
+	I2C0_SCL_PIN = GP5
+
+	I2C1_SDA_PIN = GP2
+	I2C1_SCL_PIN = GP3
+)
+
+// SPI default pins
+const (
+	// Default Serial Clock Bus 0 for SPI communications
+	SPI0_SCK_PIN = GPIO18
+	// Default Serial Out Bus 0 for SPI communications
+	SPI0_SDO_PIN = GPIO19 // Tx
+	// Default Serial In Bus 0 for SPI communications
+	SPI0_SDI_PIN = GPIO16 // Rx
+
+	// Default Serial Clock Bus 1 for SPI communications
+	SPI1_SCK_PIN = GPIO10
+	// Default Serial Out Bus 1 for SPI communications
+	SPI1_SDO_PIN = GPIO11 // Tx
+	// Default Serial In Bus 1 for SPI communications
+	SPI1_SDI_PIN = GPIO12 // Rx
+)
+
+// UART pins
+const (
+	UART0_TX_PIN = GPIO0
+	UART0_RX_PIN = GPIO1
+	UART1_TX_PIN = GPIO8
+	UART1_RX_PIN = GPIO9
+	UART_TX_PIN  = UART0_TX_PIN
+	UART_RX_PIN  = UART0_RX_PIN
+)
+
+var DefaultUART = UART0
+
+// USB identifiers
+const (
+	usb_STRING_PRODUCT      = "Pico2"
+	usb_STRING_MANUFACTURER = "Raspberry Pi"
+)
+
+var (
+	usb_VID uint16 = 0x2E8A
+	usb_PID uint16 = 0x000A
+)
@@ -81,7 +81,8 @@ var (
 // I2S pins
 const (
 	I2S_SCK_PIN = PA10
-	I2S_SD_PIN  = PA08
+	I2S_SDO_PIN = PA08
+	I2S_SDI_PIN = NoPin // TODO: figure out what this is on QT Py M0.
 	I2S_WS_PIN  = NoPin // TODO: figure out what this is on QT Py M0.
 )

@@ -0,0 +1,82 @@
+//go:build tiny2350
+
+package machine
+
+// GPIO pins
+const (
+	GP0  Pin = GPIO0
+	GP1  Pin = GPIO1
+	GP2  Pin = GPIO2
+	GP3  Pin = GPIO3
+	GP4  Pin = GPIO4
+	GP5  Pin = GPIO5
+	GP6  Pin = GPIO6
+	GP7  Pin = GPIO7
+	GP12 Pin = GPIO12
+	GP13 Pin = GPIO13
+	GP18 Pin = GPIO18
+	GP19 Pin = GPIO19
+	GP20 Pin = GPIO20
+	GP26 Pin = GPIO26
+	GP27 Pin = GPIO27
+	GP28 Pin = GPIO28
+	GP29 Pin = GPIO29
+
+	// Onboard LED
+	LED_RED   Pin = GPIO18
+	LED_GREEN Pin = GPIO19
+	LED_BLUE  Pin = GPIO20
+	LED           = LED_RED
+
+	// Onboard crystal oscillator frequency, in MHz.
+	xoscFreq = 12 // MHz
+)
+
+// I2C Default pins on Tiny2350.
+const (
+	I2C0_SDA_PIN = GP12
+	I2C0_SCL_PIN = GP13
+
+	I2C1_SDA_PIN = GP2
+	I2C1_SCL_PIN = GP3
+)
+
+// SPI default pins
+const (
+	// Default Serial Clock Bus 0 for SPI communications
+	SPI0_SCK_PIN = GPIO6
+	// Default Serial Out Bus 0 for SPI communications
+	SPI0_SDO_PIN = GPIO7 // Tx
+	// Default Serial In Bus 0 for SPI communications
+	SPI0_SDI_PIN = GPIO4 // Rx
+
+	// Default Serial Clock Bus 1 for SPI communications
+	SPI1_SCK_PIN = GPIO26
+	// Default Serial Out Bus 1 for SPI communications
+	SPI1_SDO_PIN = GPIO27 // Tx
+	// Default Serial In Bus 1 for SPI communications
+	SPI1_SDI_PIN = GPIO28 // Rx
+)
+
+// UART pins
+const (
+	UART0_TX_PIN = GPIO0
+	UART0_RX_PIN = GPIO1
+	UART1_TX_PIN = GPIO4
+	UART1_RX_PIN = GPIO5
+	UART_TX_PIN  = UART0_TX_PIN
+	UART_RX_PIN  = UART0_RX_PIN
+)
+
+var DefaultUART = UART0
+
+// USB identifiers
+const (
+	usb_STRING_PRODUCT      = "Tiny2350"
+	usb_STRING_MANUFACTURER = "Pimoroni"
+)
+
+var (
+	usb_VID uint16 = 0x2E8A
+	usb_PID uint16 = 0x000F
+)
@@ -67,7 +67,8 @@ var (
 // I2S pins
 const (
 	I2S_SCK_PIN = PA10
-	I2S_SD_PIN  = PA08
+	I2S_SDO_PIN = PA08
+	I2S_SDI_PIN = NoPin // TODO: figure out what this is on Trinket M0.
 	I2S_WS_PIN  = NoPin // TODO: figure out what this is on Trinket M0.
 )

@@ -376,6 +376,14 @@ var (
 	I2C1 = sercomI2CM3
 )

+// I2S pins
+const (
+	I2S_SCK_PIN = BCM18
+	I2S_SDO_PIN = BCM21
+	I2S_SDI_PIN = BCM20
+	I2S_WS_PIN  = BCM19
+)
+
 // SPI pins
 const (
 	SPI0_SCK_PIN = SCK // SCK:  SERCOM5/PAD[1]
@@ -82,7 +82,8 @@ var SPI0 = sercomSPIM0
 // I2S pins
 const (
 	I2S_SCK_PIN = PA10
-	I2S_SD_PIN  = PA08
+	I2S_SDO_PIN = PA08
+	I2S_SDI_PIN = NoPin // TODO: figure out what this is on Xiao
 	I2S_WS_PIN  = NoPin // TODO: figure out what this is on Xiao
 )

@@ -1,4 +1,4 @@
-//go:build !baremetal || atmega || nrf || sam || stm32 || fe310 || k210 || rp2040 || mimxrt1062 || (esp32c3 && !m5stamp_c3) || esp32
+//go:build !baremetal || atmega || nrf || sam || stm32 || fe310 || k210 || rp2040 || rp2350 || mimxrt1062 || (esp32c3 && !m5stamp_c3) || esp32

 package machine

@@ -1,4 +1,4 @@
-//go:build sam
+//go:build sam && atsamd21

 // This is the definition for I2S bus functions.
 // Actual implementations if available for any given hardware
@@ -9,6 +9,22 @@

 package machine

+import "errors"
+
+// If you are getting a compile error on this line please check to see you've
+// correctly implemented the methods on the I2S type. They must match
+// the interface method signatures type to type perfectly.
+// If not implementing the I2S type please remove your target from the build tags
+// at the top of this file.
+var _ interface {
+	SetSampleFrequency(freq uint32) error
+	ReadMono(b []uint16) (int, error)
+	ReadStereo(b []uint32) (int, error)
+	WriteMono(b []uint16) (int, error)
+	WriteStereo(b []uint32) (int, error)
+	Enable(enabled bool)
+} = (*I2S)(nil)
+
 type I2SMode uint8
 type I2SStandard uint8
 type I2SClockSource uint8
@@ -18,6 +34,7 @@ const (
 	I2SModeSource I2SMode = iota
 	I2SModeReceiver
 	I2SModePDM
+	I2SModeSourceReceiver
 )

 const (
@@ -39,11 +56,20 @@ const (
 	I2SDataFormat32bit                 = 32
 )

+var (
+	ErrInvalidSampleFrequency = errors.New("i2s: invalid sample frequency")
+)
+
 // All fields are optional and may not be required or used on a particular platform.
 type I2SConfig struct {
-	SCK             Pin
-	WS              Pin
-	SD              Pin
+	// clock
+	SCK Pin
+	// word select
+	WS Pin
+	// data out
+	SDO Pin
+	// data in
+	SDI             Pin
 	Mode            I2SMode
 	Standard        I2SStandard
 	ClockSource     I2SClockSource
@@ -926,23 +926,26 @@ func (i2c *I2C) readByte() byte {

 // I2S
 type I2S struct {
-	Bus *sam.I2S_Type
+	Bus        *sam.I2S_Type
+	Frequency  uint32
+	DataFormat I2SDataFormat
 }

 var I2S0 = I2S{Bus: sam.I2S}

 // Configure is used to configure the I2S interface. You must call this
 // before you can use the I2S bus.
-func (i2s I2S) Configure(config I2SConfig) {
+func (i2s *I2S) Configure(config I2SConfig) error {
 	// handle defaults
 	if config.SCK == 0 {
 		config.SCK = I2S_SCK_PIN
 		config.WS = I2S_WS_PIN
-		config.SD = I2S_SD_PIN
+		config.SDO = I2S_SDO_PIN
+		config.SDI = I2S_SDI_PIN
 	}

 	if config.AudioFrequency == 0 {
-		config.AudioFrequency = 48000
+		config.AudioFrequency = 44100
 	}

 	if config.DataFormat == I2SDataFormatDefault {
@@ -952,39 +955,17 @@ func (i2s I2S) Configure(config I2SConfig) {
 			config.DataFormat = I2SDataFormat32bit
 		}
 	}
+	i2s.DataFormat = config.DataFormat

 	// Turn on clock for I2S
 	sam.PM.APBCMASK.SetBits(sam.PM_APBCMASK_I2S_)

-	// setting clock rate for sample.
-	division_factor := CPUFrequency() / (config.AudioFrequency * uint32(config.DataFormat))
-
-	// Switch Generic Clock Generator 3 to DFLL48M.
-	sam.GCLK.GENDIV.Set((sam.GCLK_CLKCTRL_GEN_GCLK3 << sam.GCLK_GENDIV_ID_Pos) |
-		(division_factor << sam.GCLK_GENDIV_DIV_Pos))
-	waitForSync()
-
-	sam.GCLK.GENCTRL.Set((sam.GCLK_CLKCTRL_GEN_GCLK3 << sam.GCLK_GENCTRL_ID_Pos) |
-		(sam.GCLK_GENCTRL_SRC_DFLL48M << sam.GCLK_GENCTRL_SRC_Pos) |
-		sam.GCLK_GENCTRL_IDC |
-		sam.GCLK_GENCTRL_GENEN)
-	waitForSync()
-
-	// Use Generic Clock Generator 3 as source for I2S.
-	sam.GCLK.CLKCTRL.Set((sam.GCLK_CLKCTRL_ID_I2S_0 << sam.GCLK_CLKCTRL_ID_Pos) |
-		(sam.GCLK_CLKCTRL_GEN_GCLK3 << sam.GCLK_CLKCTRL_GEN_Pos) |
-		sam.GCLK_CLKCTRL_CLKEN)
-	waitForSync()
-
-	// reset the device
-	i2s.Bus.CTRLA.SetBits(sam.I2S_CTRLA_SWRST)
-	for i2s.Bus.SYNCBUSY.HasBits(sam.I2S_SYNCBUSY_SWRST) {
+	if err := i2s.SetSampleFrequency(config.AudioFrequency); err != nil {
+		return err
 	}

 	// disable device before continuing
-	for i2s.Bus.SYNCBUSY.HasBits(sam.I2S_SYNCBUSY_ENABLE) {
-	}
-	i2s.Bus.CTRLA.ClearBits(sam.I2S_CTRLA_ENABLE)
+	i2s.Enable(false)

 	// setup clock
 	if config.ClockSource == I2SClockSourceInternal {
@@ -1067,19 +1048,25 @@ func (i2s I2S) Configure(config I2SConfig) {
 	}

 	// set serializer mode.
-	if config.Mode == I2SModePDM {
+	switch config.Mode {
+	case I2SModePDM:
 		i2s.Bus.SERCTRL1.SetBits(sam.I2S_SERCTRL_SERMODE_PDM2)
-	} else {
+	case I2SModeSource:
+		i2s.Bus.SERCTRL1.SetBits(sam.I2S_SERCTRL_SERMODE_TX)
+	case I2SModeReceiver:
 		i2s.Bus.SERCTRL1.SetBits(sam.I2S_SERCTRL_SERMODE_RX)
 	}

-	// configure data pin
-	config.SD.Configure(PinConfig{Mode: PinCom})
+	// configure data pins
+	if config.SDO != NoPin {
+		config.SDO.Configure(PinConfig{Mode: PinCom})
+	}
+	if config.SDI != NoPin {
+		config.SDI.Configure(PinConfig{Mode: PinCom})
+	}

 	// re-enable
-	i2s.Bus.CTRLA.SetBits(sam.I2S_CTRLA_ENABLE)
-	for i2s.Bus.SYNCBUSY.HasBits(sam.I2S_SYNCBUSY_ENABLE) {
-	}
+	i2s.Enable(true)

 	// enable i2s clock
 	i2s.Bus.CTRLA.SetBits(sam.I2S_CTRLA_CKEN0)
@@ -1090,11 +1077,23 @@ func (i2s I2S) Configure(config I2SConfig) {
 	i2s.Bus.CTRLA.SetBits(sam.I2S_CTRLA_SEREN1)
 	for i2s.Bus.SYNCBUSY.HasBits(sam.I2S_SYNCBUSY_SEREN1) {
 	}
+
+	return nil
 }

-// Read data from the I2S bus into the provided slice.
+// Read mono data from the I2S bus into the provided slice.
 // The I2S bus must already have been configured correctly.
-func (i2s I2S) Read(p []uint32) (n int, err error) {
+func (i2s *I2S) ReadMono(p []uint16) (n int, err error) {
+	return i2sRead(i2s, p)
+}
+
+// Read stereo data from the I2S bus into the provided slice.
+// The I2S bus must already have been configured correctly.
+func (i2s *I2S) ReadStereo(p []uint32) (n int, err error) {
+	return i2sRead(i2s, p)
+}
+
+func i2sRead[T uint16 | uint32](i2s *I2S, p []T) (int, error) {
 	i := 0
 	for i = 0; i < len(p); i++ {
 		// Wait until ready
@@ -1105,7 +1104,7 @@ func (i2s I2S) Read(p []uint32) (n int, err error) {
 		}

 		// read data
-		p[i] = i2s.Bus.DATA1.Get()
+		p[i] = T(i2s.Bus.DATA1.Get())

 		// indicate read complete
 		i2s.Bus.INTFLAG.Set(sam.I2S_INTFLAG_RXRDY1)
@@ -1114,9 +1113,19 @@ func (i2s I2S) Read(p []uint32) (n int, err error) {
 	return i, nil
 }

-// Write data to the I2S bus from the provided slice.
+// Write mono data to the I2S bus from the provided slice.
 // The I2S bus must already have been configured correctly.
-func (i2s I2S) Write(p []uint32) (n int, err error) {
+func (i2s *I2S) WriteMono(p []uint16) (n int, err error) {
+	return i2sWrite(i2s, p)
+}
+
+// Write stereo data to the I2S bus from the provided slice.
+// The I2S bus must already have been configured correctly.
+func (i2s *I2S) WriteStereo(p []uint32) (n int, err error) {
+	return i2sWrite(i2s, p)
+}
+
+func i2sWrite[T uint16 | uint32](i2s *I2S, p []T) (int, error) {
 	i := 0
 	for i = 0; i < len(p); i++ {
 		// Wait until ready
@@ -1127,7 +1136,7 @@ func (i2s I2S) Write(p []uint32) (n int, err error) {
 		}

 		// write data
-		i2s.Bus.DATA1.Set(p[i])
+		i2s.Bus.DATA1.Set(uint32(p[i]))

 		// indicate write complete
 		i2s.Bus.INTFLAG.Set(sam.I2S_INTFLAG_TXRDY1)
@@ -1136,18 +1145,64 @@ func (i2s I2S) Write(p []uint32) (n int, err error) {
 	return i, nil
 }

-// Close the I2S bus.
-func (i2s I2S) Close() error {
-	// Sync wait
-	for i2s.Bus.SYNCBUSY.HasBits(sam.I2S_SYNCBUSY_ENABLE) {
+// SetSampleFrequency is used to set the sample frequency for the I2S bus.
+func (i2s *I2S) SetSampleFrequency(freq uint32) error {
+	if freq == 0 {
+		return ErrInvalidSampleFrequency
 	}

-	// disable I2S
-	i2s.Bus.CTRLA.ClearBits(sam.I2S_CTRLA_ENABLE)
+	if i2s.Frequency == freq {
+		return nil
+	}
+
+	i2s.Frequency = freq
+
+	// setting clock rate for sample.
+	division_factor := CPUFrequency() / (i2s.Frequency * uint32(i2s.DataFormat))
+
+	// Switch Generic Clock Generator 3 to DFLL48M.
+	sam.GCLK.GENDIV.Set((sam.GCLK_CLKCTRL_GEN_GCLK3 << sam.GCLK_GENDIV_ID_Pos) |
+		(division_factor << sam.GCLK_GENDIV_DIV_Pos))
+	waitForSync()
+
+	sam.GCLK.GENCTRL.Set((sam.GCLK_CLKCTRL_GEN_GCLK3 << sam.GCLK_GENCTRL_ID_Pos) |
+		(sam.GCLK_GENCTRL_SRC_DFLL48M << sam.GCLK_GENCTRL_SRC_Pos) |
+		sam.GCLK_GENCTRL_IDC |
+		sam.GCLK_GENCTRL_GENEN)
+	waitForSync()
+
+	// Use Generic Clock Generator 3 as source for I2S.
+	sam.GCLK.CLKCTRL.Set((sam.GCLK_CLKCTRL_ID_I2S_0 << sam.GCLK_CLKCTRL_ID_Pos) |
+		(sam.GCLK_CLKCTRL_GEN_GCLK3 << sam.GCLK_CLKCTRL_GEN_Pos) |
+		sam.GCLK_CLKCTRL_CLKEN)
+	waitForSync()
+
+	// reset the device
+	i2s.Bus.CTRLA.SetBits(sam.I2S_CTRLA_SWRST)
+	for i2s.Bus.SYNCBUSY.HasBits(sam.I2S_SYNCBUSY_SWRST) {
+	}

 	return nil
 }

+// Enabled is used to enable or disable the I2S bus.
+func (i2s *I2S) Enable(enabled bool) {
+	if enabled {
+		i2s.Bus.CTRLA.SetBits(sam.I2S_CTRLA_ENABLE)
+		for i2s.Bus.SYNCBUSY.HasBits(sam.I2S_SYNCBUSY_ENABLE) {
+		}
+
+		return
+	}
+
+	// disable
+	for i2s.Bus.SYNCBUSY.HasBits(sam.I2S_SYNCBUSY_ENABLE) {
+	}
+	i2s.Bus.CTRLA.ClearBits(sam.I2S_CTRLA_ENABLE)
+
+	return
+}
+
 func waitForSync() {
 	for sam.GCLK.STATUS.HasBits(sam.GCLK_STATUS_SYNCBUSY) {
 	}
@@ -1,38 +1,55 @@
-//go:build rp2040
+//go:build rp2040 || rp2350

 package machine

 import (
 	"device/rp"
+	"runtime/interrupt"
 	"runtime/volatile"
 	"unsafe"
 )

 const deviceName = rp.Device

+const (
+	// Number of spin locks available
+	// Note: On RP2350, most spinlocks are unusable due to Errata 2
+	_NUMSPINLOCKS         = 32
+	_PICO_SPINLOCK_ID_IRQ = 9
+)
+
+// UART on the RP2040
+var (
+	UART0  = &_UART0
+	_UART0 = UART{
+		Buffer: NewRingBuffer(),
+		Bus:    rp.UART0,
+	}
+
+	UART1  = &_UART1
+	_UART1 = UART{
+		Buffer: NewRingBuffer(),
+		Bus:    rp.UART1,
+	}
+)
+
+func init() {
+	UART0.Interrupt = interrupt.New(rp.IRQ_UART0_IRQ, _UART0.handleInterrupt)
+	UART1.Interrupt = interrupt.New(rp.IRQ_UART1_IRQ, _UART1.handleInterrupt)
+}
+
 //go:linkname machineInit runtime.machineInit
 func machineInit() {
 	// Reset all peripherals to put system into a known state,
 	// except for QSPI pads and the XIP IO bank, as this is fatal if running from flash
 	// and the PLLs, as this is fatal if clock muxing has not been reset on this boot
 	// and USB, syscfg, as this disturbs USB-to-SWD on core 1
-	bits := ^uint32(rp.RESETS_RESET_IO_QSPI |
-		rp.RESETS_RESET_PADS_QSPI |
-		rp.RESETS_RESET_PLL_USB |
-		rp.RESETS_RESET_USBCTRL |
-		rp.RESETS_RESET_SYSCFG |
-		rp.RESETS_RESET_PLL_SYS)
+	bits := ^uint32(initDontReset)
 	resetBlock(bits)

 	// Remove reset from peripherals which are clocked only by clkSys and
 	// clkRef. Other peripherals stay in reset until we've configured clocks.
-	bits = ^uint32(rp.RESETS_RESET_ADC |
-		rp.RESETS_RESET_RTC |
-		rp.RESETS_RESET_SPI0 |
-		rp.RESETS_RESET_SPI1 |
-		rp.RESETS_RESET_UART0 |
-		rp.RESETS_RESET_UART1 |
-		rp.RESETS_RESET_USBCTRL)
+	bits = ^uint32(initUnreset)
 	unresetBlockWait(bits)

 	clocks.init()
@@ -94,4 +111,25 @@ const (
 )

 // DMA channels usable on the RP2040.
-var dmaChannels = (*[12]dmaChannel)(unsafe.Pointer(rp.DMA))
+var dmaChannels = (*[12 + 4*rp2350ExtraReg]dmaChannel)(unsafe.Pointer(rp.DMA))
+
+//go:inline
+func boolToBit(a bool) uint32 {
+	if a {
+		return 1
+	}
+	return 0
+}
+
+//go:inline
+func u32max(a, b uint32) uint32 {
+	if a > b {
+		return a
+	}
+	return b
+}
+
+//go:inline
+func isReservedI2CAddr(addr uint8) bool {
+	return (addr&0x78) == 0 || (addr&0x78) == 0x78
+}
@@ -97,7 +97,7 @@ func toAlarmTime(delay uint32) rtcTime {

 func (rtc *rtcType) setDivider() {
 	// Get clk_rtc freq and make sure it is running
-	rtcFreq := configuredFreq[clkRTC]
+	rtcFreq := configuredFreq[ClkRTC]
 	if rtcFreq == 0 {
 		panic("can not set RTC divider, clock is not running")
 	}
@@ -0,0 +1,153 @@
+//go:build tinygo && rp2350
+
+package machine
+
+import ()
+
+/*
+typedef unsigned char uint8_t;
+typedef unsigned short uint16_t;
+typedef unsigned long uint32_t;
+typedef unsigned long size_t;
+typedef unsigned long uintptr_t;
+
+#define false 0
+#define true 1
+typedef int bool;
+
+// https://github.com/raspberrypi/pico-sdk
+// src/rp2_common/pico_platform_compiler/include/pico/platform/compiler.h
+
+#define pico_default_asm_volatile(...) __asm volatile (".syntax unified\n" __VA_ARGS__)
+
+
+// https://github.com/raspberrypi/pico-sdk
+// src/rp2350/pico_platform/include/pico/platform.h
+
+static bool pico_processor_state_is_nonsecure(void) {
+//    // todo add a define to disable NS checking at all?
+//    // IDAU-Exempt addresses return S=1 when tested in the Secure state,
+//    // whereas executing a tt in the NonSecure state will always return S=0.
+//    uint32_t tt;
+//    pico_default_asm_volatile (
+//        "movs %0, #0\n"
+//        "tt %0, %0\n"
+//        : "=r" (tt) : : "cc"
+//    );
+//    return !(tt & (1u << 22));
+
+    return false;
+}
+
+
+// https://github.com/raspberrypi/pico-sdk
+// src/rp2_common/pico_bootrom/include/pico/bootrom_constants.h
+
+// RP2040 & RP2350
+#define ROM_DATA_SOFTWARE_GIT_REVISION          ROM_TABLE_CODE('G', 'R')
+#define ROM_FUNC_FLASH_ENTER_CMD_XIP            ROM_TABLE_CODE('C', 'X')
+#define ROM_FUNC_FLASH_EXIT_XIP                 ROM_TABLE_CODE('E', 'X')
+#define ROM_FUNC_FLASH_FLUSH_CACHE              ROM_TABLE_CODE('F', 'C')
+#define ROM_FUNC_CONNECT_INTERNAL_FLASH         ROM_TABLE_CODE('I', 'F')
+#define ROM_FUNC_FLASH_RANGE_ERASE              ROM_TABLE_CODE('R', 'E')
+#define ROM_FUNC_FLASH_RANGE_PROGRAM            ROM_TABLE_CODE('R', 'P')
+
+// RP2350 only
+#define ROM_FUNC_PICK_AB_PARTITION              ROM_TABLE_CODE('A', 'B')
+#define ROM_FUNC_CHAIN_IMAGE                    ROM_TABLE_CODE('C', 'I')
+#define ROM_FUNC_EXPLICIT_BUY                   ROM_TABLE_CODE('E', 'B')
+#define ROM_FUNC_FLASH_RUNTIME_TO_STORAGE_ADDR  ROM_TABLE_CODE('F', 'A')
+#define ROM_DATA_FLASH_DEVINFO16_PTR            ROM_TABLE_CODE('F', 'D')
+#define ROM_FUNC_FLASH_OP                       ROM_TABLE_CODE('F', 'O')
+#define ROM_FUNC_GET_B_PARTITION                ROM_TABLE_CODE('G', 'B')
+#define ROM_FUNC_GET_PARTITION_TABLE_INFO       ROM_TABLE_CODE('G', 'P')
+#define ROM_FUNC_GET_SYS_INFO                   ROM_TABLE_CODE('G', 'S')
+#define ROM_FUNC_GET_UF2_TARGET_PARTITION       ROM_TABLE_CODE('G', 'U')
+#define ROM_FUNC_LOAD_PARTITION_TABLE           ROM_TABLE_CODE('L', 'P')
+#define ROM_FUNC_OTP_ACCESS                     ROM_TABLE_CODE('O', 'A')
+#define ROM_DATA_PARTITION_TABLE_PTR            ROM_TABLE_CODE('P', 'T')
+#define ROM_FUNC_FLASH_RESET_ADDRESS_TRANS      ROM_TABLE_CODE('R', 'A')
+#define ROM_FUNC_REBOOT                         ROM_TABLE_CODE('R', 'B')
+#define ROM_FUNC_SET_ROM_CALLBACK               ROM_TABLE_CODE('R', 'C')
+#define ROM_FUNC_SECURE_CALL                    ROM_TABLE_CODE('S', 'C')
+#define ROM_FUNC_SET_NS_API_PERMISSION          ROM_TABLE_CODE('S', 'P')
+#define ROM_FUNC_BOOTROM_STATE_RESET            ROM_TABLE_CODE('S', 'R')
+#define ROM_FUNC_SET_BOOTROM_STACK              ROM_TABLE_CODE('S', 'S')
+#define ROM_DATA_SAVED_XIP_SETUP_FUNC_PTR       ROM_TABLE_CODE('X', 'F')
+#define ROM_FUNC_FLASH_SELECT_XIP_READ_MODE     ROM_TABLE_CODE('X', 'M')
+#define ROM_FUNC_VALIDATE_NS_BUFFER             ROM_TABLE_CODE('V', 'B')
+
+#define BOOTSEL_FLAG_GPIO_PIN_SPECIFIED         0x20
+
+#define BOOTROM_FUNC_TABLE_OFFSET 0x14
+
+// todo remove this (or #ifdef it for A1/A2)
+#define BOOTROM_IS_A2() ((*(volatile uint8_t *)0x13) == 2)
+#define BOOTROM_WELL_KNOWN_PTR_SIZE (BOOTROM_IS_A2() ? 2 : 4)
+
+#define BOOTROM_VTABLE_OFFSET 0x00
+#define BOOTROM_TABLE_LOOKUP_OFFSET     (BOOTROM_FUNC_TABLE_OFFSET + BOOTROM_WELL_KNOWN_PTR_SIZE)
+
+// https://github.com/raspberrypi/pico-sdk
+// src/common/boot_picoboot_headers/include/boot/picoboot_constants.h
+
+// values 0-7 are secure/non-secure
+#define REBOOT2_FLAG_REBOOT_TYPE_NORMAL       0x0 // param0 = diagnostic partition
+#define REBOOT2_FLAG_REBOOT_TYPE_BOOTSEL      0x2 // param0 = bootsel_flags, param1 = gpio_config
+#define REBOOT2_FLAG_REBOOT_TYPE_RAM_IMAGE    0x3 // param0 = image_base, param1 = image_end
+#define REBOOT2_FLAG_REBOOT_TYPE_FLASH_UPDATE 0x4 // param0 = update_base
+
+#define REBOOT2_FLAG_NO_RETURN_ON_SUCCESS    0x100
+
+#define RT_FLAG_FUNC_ARM_SEC    0x0004
+#define RT_FLAG_FUNC_ARM_NONSEC 0x0010
+
+// https://github.com/raspberrypi/pico-sdk
+// src/rp2_common/pico_bootrom/include/pico/bootrom.h
+
+#define ROM_TABLE_CODE(c1, c2) ((c1) | ((c2) << 8))
+
+typedef void *(*rom_table_lookup_fn)(uint32_t code, uint32_t mask);
+
+__attribute__((always_inline))
+static void *rom_func_lookup_inline(uint32_t code) {
+    rom_table_lookup_fn rom_table_lookup = (rom_table_lookup_fn) (uintptr_t)*(uint16_t*)(BOOTROM_TABLE_LOOKUP_OFFSET);
+    if (pico_processor_state_is_nonsecure()) {
+        return rom_table_lookup(code, RT_FLAG_FUNC_ARM_NONSEC);
+    } else {
+        return rom_table_lookup(code, RT_FLAG_FUNC_ARM_SEC);
+    }
+}
+
+
+typedef int (*rom_reboot_fn)(uint32_t flags, uint32_t delay_ms, uint32_t p0, uint32_t p1);
+
+__attribute__((always_inline))
+int rom_reboot(uint32_t flags, uint32_t delay_ms, uint32_t p0, uint32_t p1) {
+    rom_reboot_fn func = (rom_reboot_fn) rom_func_lookup_inline(ROM_FUNC_REBOOT);
+    return func(flags, delay_ms, p0, p1);
+}
+
+
+// https://github.com/raspberrypi/pico-sdk
+// src/rp2_common/pico_bootrom/bootrom.c
+
+
+void reset_usb_boot(uint32_t usb_activity_gpio_pin_mask, uint32_t disable_interface_mask) {
+    uint32_t flags = disable_interface_mask;
+    if (usb_activity_gpio_pin_mask) {
+        flags |= BOOTSEL_FLAG_GPIO_PIN_SPECIFIED;
+        // the parameter is actually the gpio number, but we only care if BOOTSEL_FLAG_GPIO_PIN_SPECIFIED
+        usb_activity_gpio_pin_mask = (uint32_t)__builtin_ctz(usb_activity_gpio_pin_mask);
+    }
+    rom_reboot(REBOOT2_FLAG_REBOOT_TYPE_BOOTSEL | REBOOT2_FLAG_NO_RETURN_ON_SUCCESS, 10, flags, usb_activity_gpio_pin_mask);
+    __builtin_unreachable();
+}
+
+
+*/
+import "C"
+
+func enterBootloader() {
+	C.reset_usb_boot(0, 0)
+}
@@ -0,0 +1,380 @@
+//go:build rp2350
+
+package machine
+
+import (
+	"device/rp"
+	"machine/usb"
+	"runtime/interrupt"
+	"runtime/volatile"
+	"unsafe"
+)
+
+var (
+	sendOnEP0DATADONE struct {
+		offset int
+		data   []byte
+		pid    uint32
+	}
+)
+
+// Configure the USB peripheral. The config is here for compatibility with the UART interface.
+func (dev *USBDevice) Configure(config UARTConfig) {
+	// Reset usb controller
+	resetBlock(rp.RESETS_RESET_USBCTRL)
+	unresetBlockWait(rp.RESETS_RESET_USBCTRL)
+
+	// Clear any previous state in dpram just in case
+	usbDPSRAM.clear()
+
+	// Enable USB interrupt at processor
+	rp.USB.INTE.Set(0)
+	intr := interrupt.New(rp.IRQ_USBCTRL_IRQ, handleUSBIRQ)
+	intr.SetPriority(0x00)
+	intr.Enable()
+	irqSet(rp.IRQ_USBCTRL_IRQ, true)
+
+	// Mux the controller to the onboard usb phy
+	rp.USB.USB_MUXING.Set(rp.USB_USB_MUXING_TO_PHY | rp.USB_USB_MUXING_SOFTCON)
+
+	// Force VBUS detect so the device thinks it is plugged into a host
+	rp.USB.USB_PWR.Set(rp.USB_USB_PWR_VBUS_DETECT | rp.USB_USB_PWR_VBUS_DETECT_OVERRIDE_EN)
+
+	// Enable the USB controller in device mode.
+	rp.USB.MAIN_CTRL.Set(rp.USB_MAIN_CTRL_CONTROLLER_EN)
+
+	// Enable an interrupt per EP0 transaction
+	rp.USB.SIE_CTRL.Set(rp.USB_SIE_CTRL_EP0_INT_1BUF)
+
+	// Enable interrupts for when a buffer is done, when the bus is reset,
+	// and when a setup packet is received
+	rp.USB.INTE.Set(rp.USB_INTE_BUFF_STATUS |
+		rp.USB_INTE_BUS_RESET |
+		rp.USB_INTE_SETUP_REQ)
+
+	// Present full speed device by enabling pull up on DP
+	rp.USB.SIE_CTRL.SetBits(rp.USB_SIE_CTRL_PULLUP_EN)
+
+	// 12.7.2 Disable phy isolation
+	rp.USB.SetMAIN_CTRL_PHY_ISO(0x0)
+}
+
+func handleUSBIRQ(intr interrupt.Interrupt) {
+	status := rp.USB.INTS.Get()
+
+	// Setup packet received
+	if (status & rp.USB_INTS_SETUP_REQ) > 0 {
+		rp.USB.SIE_STATUS.Set(rp.USB_SIE_STATUS_SETUP_REC)
+		setup := usb.NewSetup(usbDPSRAM.setupBytes())
+
+		ok := false
+		if (setup.BmRequestType & usb.REQUEST_TYPE) == usb.REQUEST_STANDARD {
+			// Standard Requests
+			ok = handleStandardSetup(setup)
+		} else {
+			// Class Interface Requests
+			if setup.WIndex < uint16(len(usbSetupHandler)) && usbSetupHandler[setup.WIndex] != nil {
+				ok = usbSetupHandler[setup.WIndex](setup)
+			}
+		}
+
+		if !ok {
+			// Stall endpoint?
+			sendStallViaEPIn(0)
+		}
+
+	}
+
+	// Buffer status, one or more buffers have completed
+	if (status & rp.USB_INTS_BUFF_STATUS) > 0 {
+		if sendOnEP0DATADONE.offset > 0 {
+			ep := uint32(0)
+			data := sendOnEP0DATADONE.data
+			count := len(data) - sendOnEP0DATADONE.offset
+			if ep == 0 && count > usb.EndpointPacketSize {
+				count = usb.EndpointPacketSize
+			}
+
+			sendViaEPIn(ep, data[sendOnEP0DATADONE.offset:], count)
+			sendOnEP0DATADONE.offset += count
+			if sendOnEP0DATADONE.offset == len(data) {
+				sendOnEP0DATADONE.offset = 0
+			}
+		}
+
+		s2 := rp.USB.BUFF_STATUS.Get()
+
+		// OUT (PC -> rp2040)
+		for i := 0; i < 16; i++ {
+			if s2&(1<<(i*2+1)) > 0 {
+				buf := handleEndpointRx(uint32(i))
+				if usbRxHandler[i] != nil {
+					usbRxHandler[i](buf)
+				}
+				handleEndpointRxComplete(uint32(i))
+			}
+		}
+
+		// IN (rp2040 -> PC)
+		for i := 0; i < 16; i++ {
+			if s2&(1<<(i*2)) > 0 {
+				if usbTxHandler[i] != nil {
+					usbTxHandler[i]()
+				}
+			}
+		}
+
+		rp.USB.BUFF_STATUS.Set(s2)
+	}
+
+	// Bus is reset
+	if (status & rp.USB_INTS_BUS_RESET) > 0 {
+		rp.USB.SIE_STATUS.Set(rp.USB_SIE_STATUS_BUS_RESET)
+		//fixRP2040UsbDeviceEnumeration()
+
+		rp.USB.ADDR_ENDP.Set(0)
+		initEndpoint(0, usb.ENDPOINT_TYPE_CONTROL)
+	}
+}
+
+func initEndpoint(ep, config uint32) {
+	val := uint32(usbEpControlEnable) | uint32(usbEpControlInterruptPerBuff)
+	offset := ep*2*USBBufferLen + 0x100
+	val |= offset
+
+	switch config {
+	case usb.ENDPOINT_TYPE_INTERRUPT | usb.EndpointIn:
+		val |= usbEpControlEndpointTypeInterrupt
+		usbDPSRAM.EPxControl[ep].In.Set(val)
+
+	case usb.ENDPOINT_TYPE_BULK | usb.EndpointOut:
+		val |= usbEpControlEndpointTypeBulk
+		usbDPSRAM.EPxControl[ep].Out.Set(val)
+		usbDPSRAM.EPxBufferControl[ep].Out.Set(USBBufferLen & usbBuf0CtrlLenMask)
+		usbDPSRAM.EPxBufferControl[ep].Out.SetBits(usbBuf0CtrlAvail)
+
+	case usb.ENDPOINT_TYPE_INTERRUPT | usb.EndpointOut:
+		val |= usbEpControlEndpointTypeInterrupt
+		usbDPSRAM.EPxControl[ep].Out.Set(val)
+		usbDPSRAM.EPxBufferControl[ep].Out.Set(USBBufferLen & usbBuf0CtrlLenMask)
+		usbDPSRAM.EPxBufferControl[ep].Out.SetBits(usbBuf0CtrlAvail)
+
+	case usb.ENDPOINT_TYPE_BULK | usb.EndpointIn:
+		val |= usbEpControlEndpointTypeBulk
+		usbDPSRAM.EPxControl[ep].In.Set(val)
+
+	case usb.ENDPOINT_TYPE_CONTROL:
+		val |= usbEpControlEndpointTypeControl
+		usbDPSRAM.EPxBufferControl[ep].Out.Set(usbBuf0CtrlData1Pid)
+		usbDPSRAM.EPxBufferControl[ep].Out.SetBits(usbBuf0CtrlAvail)
+
+	}
+}
+
+func handleUSBSetAddress(setup usb.Setup) bool {
+	sendUSBPacket(0, []byte{}, 0)
+
+	// last, set the device address to that requested by host
+	// wait for transfer to complete
+	timeout := 3000
+	rp.USB.SIE_STATUS.Set(rp.USB_SIE_STATUS_ACK_REC)
+	for (rp.USB.SIE_STATUS.Get() & rp.USB_SIE_STATUS_ACK_REC) == 0 {
+		timeout--
+		if timeout == 0 {
+			return true
+		}
+	}
+
+	rp.USB.ADDR_ENDP.Set(uint32(setup.WValueL) & rp.USB_ADDR_ENDP_ADDRESS_Msk)
+
+	return true
+}
+
+// SendUSBInPacket sends a packet for USB (interrupt in / bulk in).
+func SendUSBInPacket(ep uint32, data []byte) bool {
+	sendUSBPacket(ep, data, 0)
+	return true
+}
+
+//go:noinline
+func sendUSBPacket(ep uint32, data []byte, maxsize uint16) {
+	count := len(data)
+	if 0 < int(maxsize) && int(maxsize) < count {
+		count = int(maxsize)
+	}
+
+	if ep == 0 {
+		if count > usb.EndpointPacketSize {
+			count = usb.EndpointPacketSize
+
+			sendOnEP0DATADONE.offset = count
+			sendOnEP0DATADONE.data = data
+		} else {
+			sendOnEP0DATADONE.offset = 0
+		}
+		epXdata0[ep] = true
+	}
+
+	sendViaEPIn(ep, data, count)
+}
+
+func ReceiveUSBControlPacket() ([cdcLineInfoSize]byte, error) {
+	var b [cdcLineInfoSize]byte
+	ep := 0
+
+	for !usbDPSRAM.EPxBufferControl[ep].Out.HasBits(usbBuf0CtrlFull) {
+		// TODO: timeout
+	}
+
+	ctrl := usbDPSRAM.EPxBufferControl[ep].Out.Get()
+	usbDPSRAM.EPxBufferControl[ep].Out.Set(USBBufferLen & usbBuf0CtrlLenMask)
+	sz := ctrl & usbBuf0CtrlLenMask
+
+	copy(b[:], usbDPSRAM.EPxBuffer[ep].Buffer0[:sz])
+
+	usbDPSRAM.EPxBufferControl[ep].Out.SetBits(usbBuf0CtrlData1Pid)
+	usbDPSRAM.EPxBufferControl[ep].Out.SetBits(usbBuf0CtrlAvail)
+
+	return b, nil
+}
+
+func handleEndpointRx(ep uint32) []byte {
+	ctrl := usbDPSRAM.EPxBufferControl[ep].Out.Get()
+	usbDPSRAM.EPxBufferControl[ep].Out.Set(USBBufferLen & usbBuf0CtrlLenMask)
+	sz := ctrl & usbBuf0CtrlLenMask
+
+	return usbDPSRAM.EPxBuffer[ep].Buffer0[:sz]
+}
+
+func handleEndpointRxComplete(ep uint32) {
+	epXdata0[ep] = !epXdata0[ep]
+	if epXdata0[ep] || ep == 0 {
+		usbDPSRAM.EPxBufferControl[ep].Out.SetBits(usbBuf0CtrlData1Pid)
+	}
+
+	usbDPSRAM.EPxBufferControl[ep].Out.SetBits(usbBuf0CtrlAvail)
+}
+
+func SendZlp() {
+	sendUSBPacket(0, []byte{}, 0)
+}
+
+func sendViaEPIn(ep uint32, data []byte, count int) {
+	// Prepare buffer control register value
+	val := uint32(count) | usbBuf0CtrlAvail
+
+	// DATA0 or DATA1
+	epXdata0[ep&0x7F] = !epXdata0[ep&0x7F]
+	if !epXdata0[ep&0x7F] {
+		val |= usbBuf0CtrlData1Pid
+	}
+
+	// Mark as full
+	val |= usbBuf0CtrlFull
+
+	copy(usbDPSRAM.EPxBuffer[ep&0x7F].Buffer0[:], data[:count])
+	usbDPSRAM.EPxBufferControl[ep&0x7F].In.Set(val)
+}
+
+func sendStallViaEPIn(ep uint32) {
+	// Prepare buffer control register value
+	if ep == 0 {
+		rp.USB.EP_STALL_ARM.Set(rp.USB_EP_STALL_ARM_EP0_IN)
+	}
+	val := uint32(usbBuf0CtrlFull)
+	usbDPSRAM.EPxBufferControl[ep&0x7F].In.Set(val)
+	val |= uint32(usbBuf0CtrlStall)
+	usbDPSRAM.EPxBufferControl[ep&0x7F].In.Set(val)
+}
+
+type USBDPSRAM struct {
+	// Note that EPxControl[0] is not EP0Control but 8-byte setup data.
+	EPxControl [16]USBEndpointControlRegister
+
+	EPxBufferControl [16]USBBufferControlRegister
+
+	EPxBuffer [16]USBBuffer
+}
+
+type USBEndpointControlRegister struct {
+	In  volatile.Register32
+	Out volatile.Register32
+}
+type USBBufferControlRegister struct {
+	In  volatile.Register32
+	Out volatile.Register32
+}
+
+type USBBuffer struct {
+	Buffer0 [USBBufferLen]byte
+	Buffer1 [USBBufferLen]byte
+}
+
+var (
+	usbDPSRAM  = (*USBDPSRAM)(unsafe.Pointer(uintptr(0x50100000)))
+	epXdata0   [16]bool
+	setupBytes [8]byte
+)
+
+func (d *USBDPSRAM) setupBytes() []byte {
+
+	data := d.EPxControl[usb.CONTROL_ENDPOINT].In.Get()
+	setupBytes[0] = byte(data)
+	setupBytes[1] = byte(data >> 8)
+	setupBytes[2] = byte(data >> 16)
+	setupBytes[3] = byte(data >> 24)
+
+	data = d.EPxControl[usb.CONTROL_ENDPOINT].Out.Get()
+	setupBytes[4] = byte(data)
+	setupBytes[5] = byte(data >> 8)
+	setupBytes[6] = byte(data >> 16)
+	setupBytes[7] = byte(data >> 24)
+
+	return setupBytes[:]
+}
+
+func (d *USBDPSRAM) clear() {
+	for i := 0; i < len(d.EPxControl); i++ {
+		d.EPxControl[i].In.Set(0)
+		d.EPxControl[i].Out.Set(0)
+		d.EPxBufferControl[i].In.Set(0)
+		d.EPxBufferControl[i].Out.Set(0)
+	}
+}
+
+const (
+	// DPRAM : Endpoint control register
+	usbEpControlEnable                 = 0x80000000
+	usbEpControlDoubleBuffered         = 0x40000000
+	usbEpControlInterruptPerBuff       = 0x20000000
+	usbEpControlInterruptPerDoubleBuff = 0x10000000
+	usbEpControlEndpointType           = 0x0c000000
+	usbEpControlInterruptOnStall       = 0x00020000
+	usbEpControlInterruptOnNak         = 0x00010000
+	usbEpControlBufferAddress          = 0x0000ffff
+
+	usbEpControlEndpointTypeControl   = 0x00000000
+	usbEpControlEndpointTypeISO       = 0x04000000
+	usbEpControlEndpointTypeBulk      = 0x08000000
+	usbEpControlEndpointTypeInterrupt = 0x0c000000
+
+	// Endpoint buffer control bits
+	usbBuf1CtrlFull     = 0x80000000
+	usbBuf1CtrlLast     = 0x40000000
+	usbBuf1CtrlData0Pid = 0x20000000
+	usbBuf1CtrlData1Pid = 0x00000000
+	usbBuf1CtrlSel      = 0x10000000
+	usbBuf1CtrlStall    = 0x08000000
+	usbBuf1CtrlAvail    = 0x04000000
+	usbBuf1CtrlLenMask  = 0x03FF0000
+	usbBuf0CtrlFull     = 0x00008000
+	usbBuf0CtrlLast     = 0x00004000
+	usbBuf0CtrlData0Pid = 0x00000000
+	usbBuf0CtrlData1Pid = 0x00002000
+	usbBuf0CtrlSel      = 0x00001000
+	usbBuf0CtrlStall    = 0x00000800
+	usbBuf0CtrlAvail    = 0x00000400
+	usbBuf0CtrlLenMask  = 0x000003FF
+
+	USBBufferLen = 64
+)
@@ -0,0 +1,199 @@
+//go:build rp2040
+
+package machine
+
+import (
+	"device/rp"
+	"runtime/volatile"
+	"unsafe"
+)
+
+const (
+	_NUMBANK0_GPIOS  = 30
+	_NUMBANK0_IRQS   = 4
+	_NUMIRQ          = 32
+	rp2350ExtraReg   = 0
+	RESETS_RESET_Msk = 0x01ffffff
+	initUnreset      = rp.RESETS_RESET_ADC |
+		rp.RESETS_RESET_RTC |
+		rp.RESETS_RESET_SPI0 |
+		rp.RESETS_RESET_SPI1 |
+		rp.RESETS_RESET_UART0 |
+		rp.RESETS_RESET_UART1 |
+		rp.RESETS_RESET_USBCTRL
+	initDontReset = rp.RESETS_RESET_IO_QSPI |
+		rp.RESETS_RESET_PADS_QSPI |
+		rp.RESETS_RESET_PLL_USB |
+		rp.RESETS_RESET_USBCTRL |
+		rp.RESETS_RESET_SYSCFG |
+		rp.RESETS_RESET_PLL_SYS
+	padEnableMask = rp.PADS_BANK0_GPIO0_IE_Msk |
+		rp.PADS_BANK0_GPIO0_OD_Msk
+)
+
+const (
+	PinOutput PinMode = iota
+	PinInput
+	PinInputPulldown
+	PinInputPullup
+	PinAnalog
+	PinUART
+	PinPWM
+	PinI2C
+	PinSPI
+	PinPIO0
+	PinPIO1
+)
+
+const (
+	ClkGPOUT0 clockIndex = iota // GPIO Muxing 0
+	ClkGPOUT1                   // GPIO Muxing 1
+	ClkGPOUT2                   // GPIO Muxing 2
+	ClkGPOUT3                   // GPIO Muxing 3
+	ClkRef                      // Watchdog and timers reference clock
+	ClkSys                      // Processors, bus fabric, memory, memory mapped registers
+	ClkPeri                     // Peripheral clock for UART and SPI
+	ClkUSB                      // USB clock
+	ClkADC                      // ADC clock
+	ClkRTC                      // Real time clock
+	NumClocks
+)
+
+func CalcClockDiv(srcFreq, freq uint32) uint32 {
+	// Div register is 24.8 int.frac divider so multiply by 2^8 (left shift by 8)
+	return uint32((uint64(srcFreq) << 8) / uint64(freq))
+}
+
+type clocksType struct {
+	clk   [NumClocks]clockType
+	resus struct {
+		ctrl   volatile.Register32
+		status volatile.Register32
+	}
+	fc0      fc
+	wakeEN0  volatile.Register32
+	wakeEN1  volatile.Register32
+	sleepEN0 volatile.Register32
+	sleepEN1 volatile.Register32
+	enabled0 volatile.Register32
+	enabled1 volatile.Register32
+	intR     volatile.Register32
+	intE     volatile.Register32
+	intF     volatile.Register32
+	intS     volatile.Register32
+}
+
+// GPIO function selectors
+const (
+	fnJTAG pinFunc = 0
+	fnSPI  pinFunc = 1 // Connect one of the internal PL022 SPI peripherals to GPIO
+	fnUART pinFunc = 2
+	fnI2C  pinFunc = 3
+	// Connect a PWM slice to GPIO. There are eight PWM slices,
+	// each with two outputchannels (A/B). The B pin can also be used as an input,
+	// for frequency and duty cyclemeasurement
+	fnPWM pinFunc = 4
+	// Software control of GPIO, from the single-cycle IO (SIO) block.
+	// The SIO function (F5)must be selected for the processors to drive a GPIO,
+	// but the input is always connected,so software can check the state of GPIOs at any time.
+	fnSIO pinFunc = 5
+	// Connect one of the programmable IO blocks (PIO) to GPIO. PIO can implement a widevariety of interfaces,
+	// and has its own internal pin mapping hardware, allowing flexibleplacement of digital interfaces on bank 0 GPIOs.
+	// The PIO function (F6, F7) must beselected for PIO to drive a GPIO, but the input is always connected,
+	// so the PIOs canalways see the state of all pins.
+	fnPIO0, fnPIO1 pinFunc = 6, 7
+	// General purpose clock inputs/outputs. Can be routed to a number of internal clock domains onRP2040,
+	// e.g. Input: to provide a 1 Hz clock for the RTC, or can be connected to an internalfrequency counter.
+	// e.g. Output: optional integer divide
+	fnGPCK pinFunc = 8
+	// USB power control signals to/from the internal USB controller
+	fnUSB  pinFunc = 9
+	fnNULL pinFunc = 0x1f
+
+	fnXIP pinFunc = 0
+)
+
+// Configure configures the gpio pin as per mode.
+func (p Pin) Configure(config PinConfig) {
+	if p == NoPin {
+		return
+	}
+	p.init()
+	mask := uint32(1) << p
+	switch config.Mode {
+	case PinOutput:
+		p.setFunc(fnSIO)
+		rp.SIO.GPIO_OE_SET.Set(mask)
+	case PinInput:
+		p.setFunc(fnSIO)
+		p.pulloff()
+	case PinInputPulldown:
+		p.setFunc(fnSIO)
+		p.pulldown()
+	case PinInputPullup:
+		p.setFunc(fnSIO)
+		p.pullup()
+	case PinAnalog:
+		p.setFunc(fnNULL)
+		p.pulloff()
+	case PinUART:
+		p.setFunc(fnUART)
+	case PinPWM:
+		p.setFunc(fnPWM)
+	case PinI2C:
+		// IO config according to 4.3.1.3 of rp2040 datasheet.
+		p.setFunc(fnI2C)
+		p.pullup()
+		p.setSchmitt(true)
+		p.setSlew(false)
+	case PinSPI:
+		p.setFunc(fnSPI)
+	case PinPIO0:
+		p.setFunc(fnPIO0)
+	case PinPIO1:
+		p.setFunc(fnPIO1)
+	}
+}
+
+var (
+	timer = (*timerType)(unsafe.Pointer(rp.TIMER))
+)
+
+// Enable or disable a specific interrupt on the executing core.
+// num is the interrupt number which must be in [0,31].
+func irqSet(num uint32, enabled bool) {
+	if num >= _NUMIRQ {
+		return
+	}
+	irqSetMask(1<<num, enabled)
+}
+
+func irqSetMask(mask uint32, enabled bool) {
+	if enabled {
+		// Clear pending before enable
+		// (if IRQ is actually asserted, it will immediately re-pend)
+		rp.PPB.NVIC_ICPR.Set(mask)
+		rp.PPB.NVIC_ISER.Set(mask)
+	} else {
+		rp.PPB.NVIC_ICER.Set(mask)
+	}
+}
+
+func (clks *clocksType) initRTC() {
+	// ClkRTC = pllUSB (48MHz) / 1024 = 46875Hz
+	clkrtc := clks.clock(ClkRTC)
+	clkrtc.configure(0, // No GLMUX
+		rp.CLOCKS_CLK_RTC_CTRL_AUXSRC_CLKSRC_PLL_USB,
+		48*MHz,
+		46875)
+}
+
+func (clks *clocksType) initTicks() {} // No ticks on RP2040
+
+// startTick starts the watchdog tick.
+// cycles needs to be a divider that when applied to the xosc input,
+// produces a 1MHz clock. So if the xosc frequency is 12MHz,
+// this will need to be 12.
+func (wd *watchdogImpl) startTick(cycles uint32) {
+	rp.WATCHDOG.TICK.Set(cycles | rp.WATCHDOG_TICK_ENABLE)
+}
--- a/Show More
+++ b/Show More