all: prepare for release 0.38.0

Signed-off-by: deadprogram <ron@hybridgroup.com>

.circleci/config.yml

@@ -73,14 +73,6 @@ commands:
             - go-cache-v4-{{ checksum "go.mod" }}
       - llvm-source-linux
       - run: go install -tags=llvm<<parameters.llvm>> .
-      - restore_cache:
-          keys:
-            - wasi-libc-sysroot-systemclang-v7
-      - run: make wasi-libc
-      - save_cache:
-          key: wasi-libc-sysroot-systemclang-v7
-          paths:
-            - lib/wasi-libc/sysroot
       - when:
           condition: <<parameters.fmt-check>>
           steps:
@@ -100,21 +92,19 @@ commands:
             - /go/pkg/mod
 
 jobs:
-  test-llvm15-go119:
+  test-llvm15-go122:
     docker:
-      - image: golang:1.19-bullseye
+      - image: golang:1.22-bullseye
     steps:
       - test-linux:
           llvm: "15"
-          # "make lint" fails before go 1.21 because internal/tools/go.mod specifies packages that require go 1.21
-          fmt-check: false
     resource_class: large
-  test-llvm19-go124:
+  test-llvm20-go124:
     docker:
       - image: golang:1.24-bullseye
     steps:
       - test-linux:
-          llvm: "19"
+          llvm: "20"
     resource_class: large
 
 workflows:
@@ -122,6 +112,6 @@ workflows:
     jobs:
       # This tests our lowest supported versions of Go and LLVM, to make sure at
       # least the smoke tests still pass.
-      - test-llvm15-go119
+      - test-llvm15-go122
-      # This tests LLVM 19 support when linking against system libraries.
+      # This tests LLVM 20 support when linking against system libraries.
-      - test-llvm19-go124
+      - test-llvm20-go124

.github/workflows/build-macos.yml

@@ -89,19 +89,10 @@ jobs:
         with:
           key: ${{ steps.cache-llvm-build.outputs.cache-primary-key }}
           path: llvm-build
-      - name: Cache wasi-libc sysroot
-        uses: actions/cache@v4
-        id: cache-wasi-libc
-        with:
-          key: wasi-libc-sysroot-${{ matrix.os }}-v1
-          path: lib/wasi-libc/sysroot
-      - name: Build wasi-libc
-        if: steps.cache-wasi-libc.outputs.cache-hit != 'true'
-        run: make wasi-libc
       - name: make gen-device
         run: make -j3 gen-device
       - name: Test TinyGo
-        run: make test GOTESTFLAGS="-short"
+        run: make test GOTESTFLAGS="-only-current-os"
       - name: Build TinyGo release tarball
         run: make release -j3
       - name: Test stdlib packages
@@ -126,7 +117,7 @@ jobs:
     runs-on: macos-latest
     strategy:
       matrix:
-        version: [16, 17, 18, 19]
+        version: [16, 17, 18, 19, 20]
     steps:
       - name: Set up Homebrew
         uses: Homebrew/actions/setup-homebrew@master
@@ -150,8 +141,8 @@ jobs:
       - name: Check binary
         run: tinygo version
       - name: Build TinyGo (default LLVM)
-        if: matrix.version == 19
+        if: matrix.version == 20
         run: go install
       - name: Check binary
-        if: matrix.version == 19
+        if: matrix.version == 20
         run: tinygo version

.github/workflows/linux.yml

@@ -104,15 +104,6 @@ jobs:
         run: |
           apk add cmake samurai python3
           make binaryen STATIC=1
-      - name: Cache wasi-libc
-        uses: actions/cache@v4
-        id: cache-wasi-libc
-        with:
-          key: wasi-libc-sysroot-linux-alpine-v2
-          path: lib/wasi-libc/sysroot
-      - name: Build wasi-libc
-        if: steps.cache-wasi-libc.outputs.cache-hit != 'true'
-        run: make wasi-libc
       - name: Install fpm
         run: |
           gem install --version 4.0.7 public_suffix
@@ -258,15 +249,6 @@ jobs:
       - name: Build Binaryen
         if: steps.cache-binaryen.outputs.cache-hit != 'true'
         run: make binaryen
-      - name: Cache wasi-libc
-        uses: actions/cache@v4
-        id: cache-wasi-libc
-        with:
-          key: wasi-libc-sysroot-linux-asserts-v6
-          path: lib/wasi-libc/sysroot
-      - name: Build wasi-libc
-        if: steps.cache-wasi-libc.outputs.cache-hit != 'true'
-        run: make wasi-libc
       - run: make gen-device -j4
       - name: Test TinyGo
         run: make ASSERT=1 test

.github/workflows/sizediff-install-pkgs.sh

@@ -2,11 +2,11 @@
 # still works after checking out the dev branch (that is, when going from LLVM
 # 16 to LLVM 17 for example, both Clang 16 and Clang 17 are installed).
 
-echo 'deb https://apt.llvm.org/noble/ llvm-toolchain-noble-19 main' | sudo tee /etc/apt/sources.list.d/llvm.list
+echo 'deb https://apt.llvm.org/noble/ llvm-toolchain-noble-20 main' | sudo tee /etc/apt/sources.list.d/llvm.list
 wget -O - https://apt.llvm.org/llvm-snapshot.gpg.key | sudo apt-key add -
 sudo apt-get update
 sudo apt-get install --no-install-recommends -y \
-    llvm-19-dev \
+    llvm-20-dev \
-    clang-19 \
+    clang-20 \
-    libclang-19-dev \
+    libclang-20-dev \
-    lld-19
+    lld-20

.github/workflows/windows.yml

@@ -23,7 +23,7 @@ jobs:
           minimum-size: 8GB
           maximum-size: 24GB
           disk-root: "C:"
-      - uses: brechtm/setup-scoop@v2
+      - uses: MinoruSekine/setup-scoop@v4
         with:
           scoop_update: 'false'
       - name: Install Dependencies
@@ -91,15 +91,6 @@ jobs:
         with:
           key: ${{ steps.cache-llvm-build.outputs.cache-primary-key }}
           path: llvm-build
-      - name: Cache wasi-libc sysroot
-        uses: actions/cache@v4
-        id: cache-wasi-libc
-        with:
-          key: wasi-libc-sysroot-v5
-          path: lib/wasi-libc/sysroot
-      - 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:
@@ -114,7 +105,7 @@ jobs:
         run: make -j3 gen-device
       - name: Test TinyGo
         shell: bash
-        run: make test GOTESTFLAGS="-short"
+        run: make test GOTESTFLAGS="-only-current-os"
       - name: Build TinyGo release tarball
         shell: bash
         run: make build/release -j4
@@ -144,7 +135,7 @@ jobs:
           minimum-size: 8GB
           maximum-size: 24GB
           disk-root: "C:"
-      - uses: brechtm/setup-scoop@v2
+      - uses: MinoruSekine/setup-scoop@v4
         with:
           scoop_update: 'false'
       - name: Install Dependencies
@@ -210,7 +201,7 @@ jobs:
           minimum-size: 8GB
           maximum-size: 24GB
           disk-root: "C:"
-      - uses: brechtm/setup-scoop@v2
+      - uses: MinoruSekine/setup-scoop@v4
         with:
           scoop_update: 'false'
       - name: Install Dependencies

.gitmodules

@@ -35,7 +35,6 @@
 [submodule "src/net"]
 	path = src/net
 	url = https://github.com/tinygo-org/net.git
-	branch = dev
 [submodule "lib/wasi-cli"]
 	path = lib/wasi-cli
 	url = https://github.com/WebAssembly/wasi-cli

CHANGELOG.md

@@ -1,3 +1,101 @@
+0.38.0
+---
+* **general**
+  - `go.*`: upgrade `golang.org/x/tools` to v0.30.0
+  - `all`: add support for LLVM 20
+* **build**
+  - go back to using MinoruSekine/setup-scoop for Windows CI builds
+  - `flake.*`: upgrade to nixpkgs 25.05, LLVM 20
+  - `Makefile`: only detect ccache command when needed
+  - `Makefile`: create random filename inside rule
+  - `Makefile`: don't set GOROOT
+  - `Makefile`: call uname at most once
+  - `Makefile`: only read NodeJS version when it is needed
+* **compiler**
+  - add support for `GODEBUG=gotypesalias=1`
+  - `interp`: fix `copy()` from/to external buffers
+  - add `-nobounds` (similar to `-gcflags=-B`)
+  - `compileopts`: add library version to cached library path
+  - `builder`: build wasi-libc inside TinyGo
+  - `builder`: simplify bdwgc libc dependency
+  - `builder`: don't use precompiled libraries
+  - `compileopts`: enable support for `GOARCH=wasm` in `tinygo test`
+* **fixes**
+  - `rp2350`: Fix DMA to SPI transmits on RP2350 (#4903)
+  - `microbit v2`: use OpenOCD flash method on microbit v2 when using Nordic Semi SoftDevice
+  - `main`: display all of the current GC options for the `-gc` flag
+  - Remove duplicated error handling
+  - `sync`: fix `TestMutexConcurrent` test
+  - fix race condition in `testdata/goroutines.go`
+  - fix build warnings on Windows ARM
+* **machine**
+  - `usb`: add USB mass storage class support
+  - implement usb receive message throttling
+  - declare usb endpoints per-platform
+  - `samd21`: implement watchdog
+  - `samd51`: write to flash memory in 512 byte long chunks
+  - `samd21`: write to flash memory in 64 byte long chunks
+  - don't inline RTT `WriteByte` everywhere
+  - `rp2`: unexport machine-specific errors
+  - `rp2`: discount scheduling delays in I2C timeouts (#4876)
+  - use pointer receiver in simulated PWM peripherals
+  - add simulated PWM/timer peripherals
+  - `rp2`: expose usb endpoint stall handling
+  - `arm`: clear pending interrupts before enabling them
+  - `rp2`: merge common usb code (#4856)
+* **main**
+  - add "cores" and "threads" schedulers to help text
+  - add `StartPos` and `EndPos` to `-json` build output
+  - change `-json` flag to match upstream Go
+* **runtime**
+  - don't lock the print output inside interrupts
+  - don't try to interrupt other cores before they are started
+  - implement `NumCPU` for the multicore scheduler
+  - add support for multicore scheduler
+  - refactor obtaining the system stack
+  - `interrupt`: add `Checkpoint` type
+  - add `exportedFuncPtr`
+  - avoid an allocation in `(*time.Timer).Reset`
+  - stub runtime signal functions for `os/signal` on wasip1
+  - move `timeUnit` to a single place
+  - implement `NumCPU` for `-scheduler=threads`
+  - move `mainExited` boolean
+  - `internal/task`: rename `tinygo_pause` to `tinygo_task_exit`
+  - map every goroutine to a new OS thread
+  - refactor `timerQueue`
+  - make conservative and precise GC MT-safe
+  - `internal/task`: implement atomic primitives for preemptive scheduling
+  - Use diskutil on macOS to extract volume name and path for FAT mounts #4928
+* **standard library**
+  - `net`: update submodule to latest commits
+  - `runtime/debug`: add GC related stubs
+  - `metrics`: flesh out some of the metric types
+  - `reflect`: Chan related stubs
+  - `os`: handle relative and abs paths in `Executable()`
+  - `os`: add `os.Executable()` for Darwin
+  - `sync`: implement `RWMutex` using futexes
+  - `reflect`: Add `SliceOf`, `ArrayOf`, `StructOf`, `MapOf`, `FuncOf`
+* **targets**
+  - `rp2040`: add multicore support
+  - `riscv32`: use `gdb` binary as a fallback
+  - add target for Microbit v2 with SoftDevice S140 support for both peripheral and central
+  - `windows`: use MSVCRT.DLL instead of UCRT on i386
+  - `windows`: add windows/386 support
+  - `arm64`: remove unnecessary `.section` directive
+  - `riscv-qemu`: actually sleep in `time.Sleep()`
+  - `riscv`: define CSR constants and use them where possible
+  - `darwin`: support Boehm GC (and use by default)
+  - `windows`: add support for the Boehm-Demers-Weiser GC
+  - `windows`: fix wrong register for first parameter
+* **wasm**
+  - add Boehm GC support
+  - refactor/modify stub signal handling
+  - don't block `//go:wasmexport` because of running goroutines
+  - use `int64` instead of `float64` for the `timeUnit`
+* **boards**
+  - Add board support for BigTreeTech SKR Pico (#4842)
+
+
 0.37.0
 ---
 * **general**

GNUmakefile

@@ -8,13 +8,20 @@ LLVM_PROJECTDIR ?= llvm-project
 CLANG_SRC ?= $(LLVM_PROJECTDIR)/clang
 LLD_SRC ?= $(LLVM_PROJECTDIR)/lld
 
+ifeq ($(OS),Windows_NT)
+    # avoid calling uname on Windows
+    uname := Windows_NT
+else
+    uname := $(shell uname -s)
+endif
+
 # Try to autodetect LLVM build tools.
 # Versions are listed here in descending priority order.
 LLVM_VERSIONS = 19 18 17 16 15
 errifempty = $(if $(1),$(1),$(error $(2)))
 detect = $(shell which $(call errifempty,$(firstword $(foreach p,$(2),$(shell command -v $(p) 2> /dev/null && echo $(p)))),failed to locate $(1) at any of: $(2)))
 toolSearchPathsVersion = $(1)-$(2)
-ifeq ($(shell uname -s),Darwin)
+ifeq ($(uname),Darwin)
 	# Also explicitly search Brew's copy, which is not in PATH by default.
 	BREW_PREFIX := $(shell brew --prefix)
 	toolSearchPathsVersion += $(BREW_PREFIX)/opt/llvm@$(2)/bin/$(1)-$(2) $(BREW_PREFIX)/opt/llvm@$(2)/bin/$(1)
@@ -27,7 +34,6 @@ LLVM_NM ?= $(call findLLVMTool,llvm-nm)
 
 # Go binary and GOROOT to select
 GO ?= go
-export GOROOT = $(shell $(GO) env GOROOT)
 
 # Flags to pass to go test.
 GOTESTFLAGS ?=
@@ -38,14 +44,10 @@ TINYGO ?= $(call detect,tinygo,tinygo $(CURDIR)/build/tinygo)
 
 # Check for ccache if the user hasn't set it to on or off.
 ifeq (, $(CCACHE))
-    # Use CCACHE for LLVM if possible
+    LLVM_OPTION += '-DLLVM_CCACHE_BUILD=$(if $(shell command -v ccache 2> /dev/null),ON,OFF)'
-    ifneq (, $(shell command -v ccache 2> /dev/null))
+else
-        CCACHE := ON
+    LLVM_OPTION += '-DLLVM_CCACHE_BUILD=$(CCACHE)'
-    else
-        CCACHE := OFF
-    endif
 endif
-LLVM_OPTION += '-DLLVM_CCACHE_BUILD=$(CCACHE)'
 
 # Allow enabling LLVM assertions
 ifeq (1, $(ASSERT))
@@ -127,14 +129,14 @@ ifeq ($(OS),Windows_NT)
 
     USE_SYSTEM_BINARYEN ?= 1
 
-else ifeq ($(shell uname -s),Darwin)
+else ifeq ($(uname),Darwin)
     MD5SUM ?= md5
 
     CGO_LDFLAGS += -lxar
 
     USE_SYSTEM_BINARYEN ?= 1
 
-else ifeq ($(shell uname -s),FreeBSD)
+else ifeq ($(uname),FreeBSD)
     MD5SUM ?= md5
     START_GROUP = -Wl,--start-group
     END_GROUP = -Wl,--end-group
@@ -259,13 +261,6 @@ build/wasm-opt$(EXE):
 	cp lib/binaryen/bin/wasm-opt$(EXE) build/wasm-opt$(EXE)
 endif
 
-# Build wasi-libc sysroot
-.PHONY: wasi-libc
-wasi-libc: lib/wasi-libc/sysroot/lib/wasm32-wasi/libc.a
-lib/wasi-libc/sysroot/lib/wasm32-wasi/libc.a:
-	@if [ ! -e lib/wasi-libc/Makefile ]; then echo "Submodules have not been downloaded. Please download them using:\n  git submodule update --init"; exit 1; fi
-	cd lib/wasi-libc && $(MAKE) -j4 EXTRA_CFLAGS="-O2 -g -DNDEBUG -mnontrapping-fptoint -msign-ext" MALLOC_IMPL=none CC="$(CLANG)" AR=$(LLVM_AR) NM=$(LLVM_NM)
-
 # Generate WASI syscall bindings
 WASM_TOOLS_MODULE=go.bytecodealliance.org
 .PHONY: wasi-syscall
@@ -293,7 +288,7 @@ check-nodejs-version:
 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" .
-test: wasi-libc check-nodejs-version
+test: 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)
 
 # Standard library packages that pass tests on darwin, linux, wasi, and windows, but take over a minute in wasi
@@ -439,8 +434,8 @@ TEST_PACKAGES_NONBAREMETAL = \
 	$(nil)
 
 # Report platforms on which each standard library package is known to pass tests
-jointmp := $(shell echo /tmp/join.$$$$)
 report-stdlib-tests-pass:
+	$(eval jointmp := $(shell echo /tmp/join.$$$$))
 	@for t in $(TEST_PACKAGES_DARWIN); do echo "$$t darwin"; done | sort > $(jointmp).darwin
 	@for t in $(TEST_PACKAGES_LINUX); do echo "$$t linux"; done | sort > $(jointmp).linux
 	@for t in $(TEST_PACKAGES_FAST) $(TEST_PACKAGES_SLOW); do echo "$$t darwin linux wasi windows"; done | sort > $(jointmp).portable
@@ -449,11 +444,11 @@ report-stdlib-tests-pass:
 	@rm $(jointmp).*
 
 # Standard library packages that pass tests quickly on the current platform
-ifeq ($(shell uname),Darwin)
+ifeq ($(uname),Darwin)
 TEST_PACKAGES_HOST := $(TEST_PACKAGES_FAST) $(TEST_PACKAGES_DARWIN)
 TEST_IOFS := true
 endif
-ifeq ($(shell uname),Linux)
+ifeq ($(uname),Linux)
 TEST_PACKAGES_HOST := $(TEST_PACKAGES_FAST) $(TEST_PACKAGES_LINUX)
 TEST_IOFS := true
 endif
@@ -526,9 +521,9 @@ test-corpus:
 	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) test $(GOTESTFLAGS) -timeout=1h -buildmode exe -tags byollvm -run TestCorpus . -corpus=testdata/corpus.yaml
 test-corpus-fast:
 	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) test $(GOTESTFLAGS) -timeout=1h -buildmode exe -tags byollvm -run TestCorpus -short . -corpus=testdata/corpus.yaml
-test-corpus-wasi: wasi-libc
+test-corpus-wasi:
 	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) test $(GOTESTFLAGS) -timeout=1h -buildmode exe -tags byollvm -run TestCorpus . -corpus=testdata/corpus.yaml -target=wasip1
-test-corpus-wasip2: wasi-libc
+test-corpus-wasip2:
 	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) test $(GOTESTFLAGS) -timeout=1h -buildmode exe -tags byollvm -run TestCorpus . -corpus=testdata/corpus.yaml -target=wasip2
 
 .PHONY: testchdir
@@ -631,8 +626,12 @@ endif
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=microbit-v2-s113v7  examples/microbit-blink
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=microbit-v2-s140v7  examples/microbit-blink
+	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=nrf52840-mdk        examples/blinky1
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=btt-skr-pico        examples/uart
+	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=pca10031            examples/blinky1
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=reelboard           examples/blinky1
@@ -795,6 +794,10 @@ endif
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=feather-nrf52840    examples/usb-midi
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pico    			examples/usb-storage
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pico2    			examples/usb-storage
+	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=nrf52840-s140v6-uf2-generic	examples/machinetest
 	@$(MD5SUM) test.hex
 ifneq ($(STM32), 0)
@@ -939,14 +942,15 @@ endif
 wasmtest:
 	$(GO) test ./tests/wasm
 
-build/release: tinygo gen-device wasi-libc $(if $(filter 1,$(USE_SYSTEM_BINARYEN)),,binaryen)
+build/release: tinygo gen-device $(if $(filter 1,$(USE_SYSTEM_BINARYEN)),,binaryen)
 	@mkdir -p build/release/tinygo/bin
 	@mkdir -p build/release/tinygo/lib/bdwgc
 	@mkdir -p build/release/tinygo/lib/clang/include
 	@mkdir -p build/release/tinygo/lib/CMSIS/CMSIS
 	@mkdir -p build/release/tinygo/lib/macos-minimal-sdk
+	@mkdir -p build/release/tinygo/lib/mingw-w64/mingw-w64-crt/crt
+	@mkdir -p build/release/tinygo/lib/mingw-w64/mingw-w64-crt/math
 	@mkdir -p build/release/tinygo/lib/mingw-w64/mingw-w64-crt/lib-common
-	@mkdir -p build/release/tinygo/lib/mingw-w64/mingw-w64-crt/stdio
 	@mkdir -p build/release/tinygo/lib/mingw-w64/mingw-w64-headers/defaults
 	@mkdir -p build/release/tinygo/lib/musl/arch
 	@mkdir -p build/release/tinygo/lib/musl/crt
@@ -954,7 +958,7 @@ build/release: tinygo gen-device wasi-libc $(if $(filter 1,$(USE_SYSTEM_BINARYEN
 	@mkdir -p build/release/tinygo/lib/nrfx
 	@mkdir -p build/release/tinygo/lib/picolibc/newlib/libc
 	@mkdir -p build/release/tinygo/lib/picolibc/newlib/libm
-	@mkdir -p build/release/tinygo/lib/wasi-libc/libc-bottom-half/headers
+	@mkdir -p build/release/tinygo/lib/wasi-libc/libc-bottom-half
 	@mkdir -p build/release/tinygo/lib/wasi-libc/libc-top-half/musl/arch
 	@mkdir -p build/release/tinygo/lib/wasi-libc/libc-top-half/musl/src
 	@mkdir -p build/release/tinygo/lib/wasi-cli/
@@ -977,6 +981,7 @@ endif
 	@cp -rp lib/musl/crt/crt1.c          build/release/tinygo/lib/musl/crt
 	@cp -rp lib/musl/COPYRIGHT           build/release/tinygo/lib/musl
 	@cp -rp lib/musl/include             build/release/tinygo/lib/musl
+	@cp -rp lib/musl/src/conf            build/release/tinygo/lib/musl/src
 	@cp -rp lib/musl/src/ctype           build/release/tinygo/lib/musl/src
 	@cp -rp lib/musl/src/env             build/release/tinygo/lib/musl/src
 	@cp -rp lib/musl/src/errno           build/release/tinygo/lib/musl/src
@@ -1001,10 +1006,17 @@ endif
 	@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/crt/pseudo-reloc.c          build/release/tinygo/lib/mingw-w64/mingw-w64-crt/crt
 	@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/gdtoa                       build/release/tinygo/lib/mingw-w64/mingw-w64-crt
+	@cp -rp lib/mingw-w64/mingw-w64-crt/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/advapi32.def.in  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
-	@cp -rp lib/mingw-w64/mingw-w64-crt/stdio/ucrt_*                build/release/tinygo/lib/mingw-w64/mingw-w64-crt/stdio
+	@cp -rp lib/mingw-w64/mingw-w64-crt/lib-common/msvcrt.def.in    build/release/tinygo/lib/mingw-w64/mingw-w64-crt/lib-common
+	@cp -rp lib/mingw-w64/mingw-w64-crt/math/x86                    build/release/tinygo/lib/mingw-w64/mingw-w64-crt/math
+	@cp -rp lib/mingw-w64/mingw-w64-crt/misc                        build/release/tinygo/lib/mingw-w64/mingw-w64-crt
+	@cp -rp lib/mingw-w64/mingw-w64-crt/stdio                       build/release/tinygo/lib/mingw-w64/mingw-w64-crt
 	@cp -rp lib/mingw-w64/mingw-w64-headers/crt/                    build/release/tinygo/lib/mingw-w64/mingw-w64-headers
 	@cp -rp lib/mingw-w64/mingw-w64-headers/defaults/include        build/release/tinygo/lib/mingw-w64/mingw-w64-headers/defaults
 	@cp -rp lib/mingw-w64/mingw-w64-headers/include                 build/release/tinygo/lib/mingw-w64/mingw-w64-headers
@@ -1017,15 +1029,36 @@ endif
 	@cp -rp lib/picolibc/newlib/libm/common      build/release/tinygo/lib/picolibc/newlib/libm
 	@cp -rp lib/picolibc/newlib/libm/math        build/release/tinygo/lib/picolibc/newlib/libm
 	@cp -rp lib/picolibc-stdio.c         build/release/tinygo/lib
-	@cp -rp lib/wasi-libc/libc-bottom-half/headers/public           build/release/tinygo/lib/wasi-libc/libc-bottom-half/headers
+	@cp -rp lib/wasi-libc/libc-bottom-half/cloudlibc                build/release/tinygo/lib/wasi-libc/libc-bottom-half
+	@cp -rp lib/wasi-libc/libc-bottom-half/headers                  build/release/tinygo/lib/wasi-libc/libc-bottom-half
+	@cp -rp lib/wasi-libc/libc-bottom-half/sources                  build/release/tinygo/lib/wasi-libc/libc-bottom-half
+	@cp -rp lib/wasi-libc/libc-top-half/headers                     build/release/tinygo/lib/wasi-libc/libc-top-half
 	@cp -rp lib/wasi-libc/libc-top-half/musl/arch/generic           build/release/tinygo/lib/wasi-libc/libc-top-half/musl/arch
 	@cp -rp lib/wasi-libc/libc-top-half/musl/arch/wasm32            build/release/tinygo/lib/wasi-libc/libc-top-half/musl/arch
+	@cp -rp lib/wasi-libc/libc-top-half/musl/include                build/release/tinygo/lib/wasi-libc/libc-top-half/musl
+	@cp -rp lib/wasi-libc/libc-top-half/musl/src/conf               build/release/tinygo/lib/wasi-libc/libc-top-half/musl/src
+	@cp -rp lib/wasi-libc/libc-top-half/musl/src/dirent             build/release/tinygo/lib/wasi-libc/libc-top-half/musl/src
+	@cp -rp lib/wasi-libc/libc-top-half/musl/src/env                build/release/tinygo/lib/wasi-libc/libc-top-half/musl/src
+	@cp -rp lib/wasi-libc/libc-top-half/musl/src/errno              build/release/tinygo/lib/wasi-libc/libc-top-half/musl/src
+	@cp -rp lib/wasi-libc/libc-top-half/musl/src/exit               build/release/tinygo/lib/wasi-libc/libc-top-half/musl/src
+	@cp -rp lib/wasi-libc/libc-top-half/musl/src/fcntl              build/release/tinygo/lib/wasi-libc/libc-top-half/musl/src
+	@cp -rp lib/wasi-libc/libc-top-half/musl/src/fenv               build/release/tinygo/lib/wasi-libc/libc-top-half/musl/src
 	@cp -rp lib/wasi-libc/libc-top-half/musl/src/include            build/release/tinygo/lib/wasi-libc/libc-top-half/musl/src
 	@cp -rp lib/wasi-libc/libc-top-half/musl/src/internal           build/release/tinygo/lib/wasi-libc/libc-top-half/musl/src
+	@cp -rp lib/wasi-libc/libc-top-half/musl/src/legacy             build/release/tinygo/lib/wasi-libc/libc-top-half/musl/src
+	@cp -rp lib/wasi-libc/libc-top-half/musl/src/locale             build/release/tinygo/lib/wasi-libc/libc-top-half/musl/src
 	@cp -rp lib/wasi-libc/libc-top-half/musl/src/math               build/release/tinygo/lib/wasi-libc/libc-top-half/musl/src
+	@cp -rp lib/wasi-libc/libc-top-half/musl/src/misc               build/release/tinygo/lib/wasi-libc/libc-top-half/musl/src
+	@cp -rp lib/wasi-libc/libc-top-half/musl/src/multibyte          build/release/tinygo/lib/wasi-libc/libc-top-half/musl/src
+	@cp -rp lib/wasi-libc/libc-top-half/musl/src/network            build/release/tinygo/lib/wasi-libc/libc-top-half/musl/src
+	@cp -rp lib/wasi-libc/libc-top-half/musl/src/stat               build/release/tinygo/lib/wasi-libc/libc-top-half/musl/src
+	@cp -rp lib/wasi-libc/libc-top-half/musl/src/stdio              build/release/tinygo/lib/wasi-libc/libc-top-half/musl/src
+	@cp -rp lib/wasi-libc/libc-top-half/musl/src/stdlib             build/release/tinygo/lib/wasi-libc/libc-top-half/musl/src
 	@cp -rp lib/wasi-libc/libc-top-half/musl/src/string             build/release/tinygo/lib/wasi-libc/libc-top-half/musl/src
-	@cp -rp lib/wasi-libc/libc-top-half/musl/include                build/release/tinygo/lib/wasi-libc/libc-top-half/musl
+	@cp -rp lib/wasi-libc/libc-top-half/musl/src/thread             build/release/tinygo/lib/wasi-libc/libc-top-half/musl/src
-	@cp -rp lib/wasi-libc/sysroot                                   build/release/tinygo/lib/wasi-libc/sysroot
+	@cp -rp lib/wasi-libc/libc-top-half/musl/src/time               build/release/tinygo/lib/wasi-libc/libc-top-half/musl/src
+	@cp -rp lib/wasi-libc/libc-top-half/musl/src/unistd             build/release/tinygo/lib/wasi-libc/libc-top-half/musl/src
+	@cp -rp lib/wasi-libc/libc-top-half/sources                     build/release/tinygo/lib/wasi-libc/libc-top-half
 	@cp -rp lib/wasi-cli/wit                                        build/release/tinygo/lib/wasi-cli/wit
 	@cp -rp llvm-project/compiler-rt/lib/builtins build/release/tinygo/lib/compiler-rt-builtins
 	@cp -rp llvm-project/compiler-rt/LICENSE.TXT  build/release/tinygo/lib/compiler-rt-builtins

builder/bdwgc.go

@@ -14,7 +14,7 @@ var BoehmGC = Library{
 	name: "bdwgc",
 	cflags: func(target, headerPath string) []string {
 		libdir := filepath.Join(goenv.Get("TINYGOROOT"), "lib/bdwgc")
-		return []string{
+		flags := []string{
 			// use a modern environment
 			"-DUSE_MMAP",              // mmap is available
 			"-DUSE_MUNMAP",            // return memory to the OS using munmap
@@ -30,6 +30,7 @@ var BoehmGC = Library{
 			// Use a minimal environment.
 			"-DNO_MSGBOX_ON_ERROR", // don't call MessageBoxA on Windows
 			"-DDONT_USE_ATEXIT",
+			"-DNO_GETENV",
 
 			// Special flag to work around the lack of __data_start in ld.lld.
 			// TODO: try to fix this in LLVM/lld directly so we don't have to
@@ -42,17 +43,20 @@ var BoehmGC = Library{
 			// Assertions can be enabled while debugging GC issues.
 			//"-DGC_ASSERTIONS",
 
-			// Threading is not yet supported, so these are disabled.
+			// We use our own way of dealing with threads (that is a bit hacky).
+			// See src/runtime/gc_boehm.go.
 			//"-DGC_THREADS",
 			//"-DTHREAD_LOCAL_ALLOC",
 
 			"-I" + libdir + "/include",
 		}
+		return flags
 	},
+	needsLibc: true,
 	sourceDir: func() string {
 		return filepath.Join(goenv.Get("TINYGOROOT"), "lib/bdwgc")
 	},
-	librarySources: func(target string) ([]string, error) {
+	librarySources: func(target string, _ bool) ([]string, error) {
 		sources := []string{
 			"allchblk.c",
 			"alloc.c",

builder/build.go

@@ -14,7 +14,6 @@ import (
 	"fmt"
 	"go/types"
 	"hash/crc32"
-	"io/fs"
 	"math/bits"
 	"os"
 	"os/exec"
@@ -147,14 +146,13 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 	// the libc needs them.
 	root := goenv.Get("TINYGOROOT")
 	var libcDependencies []*compileJob
-	var libcJob *compileJob
 	switch config.Target.Libc {
 	case "darwin-libSystem":
-		libcJob = makeDarwinLibSystemJob(config, tmpdir)
+		libcJob := makeDarwinLibSystemJob(config, tmpdir)
 		libcDependencies = append(libcDependencies, libcJob)
 	case "musl":
 		var unlock func()
-		libcJob, unlock, err = libMusl.load(config, tmpdir, nil)
+		libcJob, unlock, err := libMusl.load(config, tmpdir)
 		if err != nil {
 			return BuildResult{}, err
 		}
@@ -162,28 +160,28 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 		libcDependencies = append(libcDependencies, dummyCompileJob(filepath.Join(filepath.Dir(libcJob.result), "crt1.o")))
 		libcDependencies = append(libcDependencies, libcJob)
 	case "picolibc":
-		libcJob, unlock, err := libPicolibc.load(config, tmpdir, nil)
+		libcJob, unlock, err := libPicolibc.load(config, tmpdir)
 		if err != nil {
 			return BuildResult{}, err
 		}
 		defer unlock()
 		libcDependencies = append(libcDependencies, libcJob)
 	case "wasi-libc":
-		path := filepath.Join(root, "lib/wasi-libc/sysroot/lib/wasm32-wasi/libc.a")
+		libcJob, unlock, err := libWasiLibc.load(config, tmpdir)
-		if _, err := os.Stat(path); errors.Is(err, fs.ErrNotExist) {
+		if err != nil {
-			return BuildResult{}, errors.New("could not find wasi-libc, perhaps you need to run `make wasi-libc`?")
+			return BuildResult{}, err
 		}
-		libcDependencies = append(libcDependencies, dummyCompileJob(path))
+		defer unlock()
+		libcDependencies = append(libcDependencies, libcJob)
 	case "wasmbuiltins":
-		libcJob, unlock, err := libWasmBuiltins.load(config, tmpdir, nil)
+		libcJob, unlock, err := libWasmBuiltins.load(config, tmpdir)
 		if err != nil {
 			return BuildResult{}, err
 		}
 		defer unlock()
 		libcDependencies = append(libcDependencies, libcJob)
 	case "mingw-w64":
-		var unlock func()
+		libcJob, unlock, err := libMinGW.load(config, tmpdir)
-		libcJob, unlock, err = libMinGW.load(config, tmpdir, nil)
 		if err != nil {
 			return BuildResult{}, err
 		}
@@ -216,6 +214,7 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 		MaxStackAlloc:      config.MaxStackAlloc(),
 		NeedsStackObjects:  config.NeedsStackObjects(),
 		Debug:              !config.Options.SkipDWARF, // emit DWARF except when -internal-nodwarf is passed
+		Nobounds:           config.Options.Nobounds,
 		PanicStrategy:      config.PanicStrategy(),
 	}
 
@@ -704,7 +703,7 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 	// Add compiler-rt dependency if needed. Usually this is a simple load from
 	// a cache.
 	if config.Target.RTLib == "compiler-rt" {
-		job, unlock, err := libCompilerRT.load(config, tmpdir, nil)
+		job, unlock, err := libCompilerRT.load(config, tmpdir)
 		if err != nil {
 			return result, err
 		}
@@ -714,10 +713,7 @@ func Build(pkgName, outpath, tmpdir string, config *compileopts.Config) (BuildRe
 
 	// The Boehm collector is stored in a separate C library.
 	if config.GC() == "boehm" {
-		if libcJob == nil {
+		job, unlock, err := BoehmGC.load(config, tmpdir)
-			return BuildResult{}, fmt.Errorf("boehm GC isn't supported with libc %s", config.Target.Libc)
-		}
-		job, unlock, err := BoehmGC.load(config, tmpdir, libcJob)
 		if err != nil {
 			return BuildResult{}, err
 		}

builder/builder_test.go

@@ -67,6 +67,7 @@ func TestClangAttributes(t *testing.T) {
 		{GOOS: "linux", GOARCH: "mipsle", GOMIPS: "softfloat"},
 		{GOOS: "darwin", GOARCH: "amd64"},
 		{GOOS: "darwin", GOARCH: "arm64"},
+		{GOOS: "windows", GOARCH: "386"},
 		{GOOS: "windows", GOARCH: "amd64"},
 		{GOOS: "windows", GOARCH: "arm64"},
 	} {

builder/builtins.go

@@ -3,6 +3,7 @@ package builder
 import (
 	"os"
 	"path/filepath"
+	"strings"
 
 	"github.com/tinygo-org/tinygo/compileopts"
 	"github.com/tinygo-org/tinygo/goenv"
@@ -201,6 +202,11 @@ var avrBuiltins = []string{
 	"avr/udivmodqi4.S",
 }
 
+// Builtins needed specifically for windows/386.
+var windowsI386Builtins = []string{
+	"i386/chkstk.S", // also _alloca
+}
+
 // libCompilerRT is a library with symbols required by programs compiled with
 // LLVM. These symbols are for operations that cannot be emitted with a single
 // instruction or a short sequence of instructions for that target.
@@ -220,7 +226,7 @@ var libCompilerRT = Library{
 		// Development build.
 		return filepath.Join(goenv.Get("TINYGOROOT"), "lib/compiler-rt-builtins")
 	},
-	librarySources: func(target string) ([]string, error) {
+	librarySources: func(target string, _ bool) ([]string, error) {
 		builtins := append([]string{}, genericBuiltins...) // copy genericBuiltins
 		switch compileopts.CanonicalArchName(target) {
 		case "arm":
@@ -229,6 +235,10 @@ var libCompilerRT = Library{
 			builtins = append(builtins, avrBuiltins...)
 		case "x86_64", "aarch64", "riscv64": // any 64-bit arch
 			builtins = append(builtins, genericBuiltins128...)
+		case "i386":
+			if strings.Split(target, "-")[2] == "windows" {
+				builtins = append(builtins, windowsI386Builtins...)
+			}
 		}
 		return builtins, nil
 	},

builder/library.go

@@ -15,6 +15,9 @@ import (
 
 // Library is a container for information about a single C library, such as a
 // compiler runtime or libc.
+//
+// Note: whenever a library gets changed, the version in compileopts/config.go
+// probably also needs to be incremented.
 type Library struct {
 	// The library name, such as compiler-rt or picolibc.
 	name string
@@ -25,11 +28,17 @@ type Library struct {
 	// cflags returns the C flags specific to this library
 	cflags func(target, headerPath string) []string
 
+	// cflagsForFile returns additional C flags for a particular source file.
+	cflagsForFile func(path string) []string
+
+	// needsLibc is set to true if this library needs libc headers.
+	needsLibc bool
+
 	// The source directory.
 	sourceDir func() string
 
 	// The source files, relative to sourceDir.
-	librarySources func(target string) ([]string, error)
+	librarySources func(target string, libcNeedsMalloc bool) ([]string, error)
 
 	// The source code for the crt1.o file, relative to sourceDir.
 	crt1Source string
@@ -43,18 +52,9 @@ type Library struct {
 // output archive file, it is expected to be removed after use.
 // As a side effect, this call creates the library header files if they didn't
 // exist yet.
-// The provided libc job (if not null) will cause this libc to be added as a
+func (l *Library) load(config *compileopts.Config, tmpdir string) (job *compileJob, abortLock func(), err error) {
-// dependency for all C compiler jobs, and adds libc headers for the given
+	outdir := config.LibraryPath(l.name)
-// target config. In other words, pass this libc if the library needs a libc to
-// compile.
-func (l *Library) load(config *compileopts.Config, tmpdir string, libc *compileJob) (job *compileJob, abortLock func(), err error) {
-	outdir, precompiled := config.LibcPath(l.name)
 	archiveFilePath := filepath.Join(outdir, "lib.a")
-	if precompiled {
-		// Found a precompiled library for this OS/architecture. Return the path
-		// directly.
-		return dummyCompileJob(archiveFilePath), func() {}, nil
-	}
 
 	// Create a lock on the output (if supported).
 	// This is a bit messy, but avoids a deadlock because it is ordered consistently with other library loads within a build.
@@ -185,7 +185,7 @@ func (l *Library) load(config *compileopts.Config, tmpdir string, libc *compileJ
 			args = append(args, "-mfpu=vfpv2")
 		}
 	}
-	if libc != nil {
+	if l.needsLibc {
 		args = append(args, config.LibcCFlags()...)
 	}
 
@@ -226,12 +226,13 @@ func (l *Library) load(config *compileopts.Config, tmpdir string, libc *compileJ
 
 	// Create jobs to compile all sources. These jobs are depended upon by the
 	// archive job above, so must be run first.
-	paths, err := l.librarySources(target)
+	paths, err := l.librarySources(target, config.LibcNeedsMalloc())
 	if err != nil {
 		return nil, nil, err
 	}
 	for _, path := range paths {
 		// Strip leading "../" parts off the path.
+		path := path
 		cleanpath := path
 		for strings.HasPrefix(cleanpath, "../") {
 			cleanpath = cleanpath[3:]
@@ -245,6 +246,9 @@ func (l *Library) load(config *compileopts.Config, tmpdir string, libc *compileJ
 			run: func(*compileJob) error {
 				var compileArgs []string
 				compileArgs = append(compileArgs, args...)
+				if l.cflagsForFile != nil {
+					compileArgs = append(compileArgs, l.cflagsForFile(path)...)
+				}
 				compileArgs = append(compileArgs, "-o", objpath, srcpath)
 				if config.Options.PrintCommands != nil {
 					config.Options.PrintCommands("clang", compileArgs...)
@@ -256,9 +260,6 @@ func (l *Library) load(config *compileopts.Config, tmpdir string, libc *compileJ
 				return nil
 			},
 		}
-		if libc != nil {
-			objfile.dependencies = append(objfile.dependencies, libc)
-		}
 		job.dependencies = append(job.dependencies, objfile)
 	}
 
@@ -289,9 +290,6 @@ func (l *Library) load(config *compileopts.Config, tmpdir string, libc *compileJ
 				return os.Rename(tmpfile.Name(), filepath.Join(outdir, "crt1.o"))
 			},
 		}
-		if libc != nil {
-			crt1Job.dependencies = append(crt1Job.dependencies, libc)
-		}
 		job.dependencies = append(job.dependencies, crt1Job)
 	}
 

builder/mingw-w64.go

@@ -30,25 +30,62 @@ var libMinGW = Library{
 	sourceDir: func() string { return filepath.Join(goenv.Get("TINYGOROOT"), "lib/mingw-w64") },
 	cflags: func(target, headerPath string) []string {
 		mingwDir := filepath.Join(goenv.Get("TINYGOROOT"), "lib/mingw-w64")
-		return []string{
+		flags := []string{
 			"-nostdlibinc",
+			"-isystem", mingwDir + "/mingw-w64-crt/include",
 			"-isystem", mingwDir + "/mingw-w64-headers/crt",
+			"-isystem", mingwDir + "/mingw-w64-headers/include",
 			"-I", mingwDir + "/mingw-w64-headers/defaults/include",
 			"-I" + headerPath,
 		}
+		if strings.Split(target, "-")[0] == "i386" {
+			flags = append(flags,
+				"-D__MSVCRT_VERSION__=0x700", // Microsoft Visual C++ .NET 2002
+				"-D_WIN32_WINNT=0x0501",      // target Windows XP
+				"-D_CRTBLD",
+				"-Wno-pragma-pack",
+			)
+		}
+		return flags
 	},
-	librarySources: func(target string) ([]string, error) {
+	librarySources: func(target string, _ bool) ([]string, error) {
 		// These files are needed so that printf and the like are supported.
-		sources := []string{
+		var sources []string
-			"mingw-w64-crt/stdio/ucrt_fprintf.c",
+		if strings.Split(target, "-")[0] == "i386" {
-			"mingw-w64-crt/stdio/ucrt_fwprintf.c",
+			// Old 32-bit x86 systems use msvcrt.dll.
-			"mingw-w64-crt/stdio/ucrt_printf.c",
+			sources = []string{
-			"mingw-w64-crt/stdio/ucrt_snprintf.c",
+				"mingw-w64-crt/crt/pseudo-reloc.c",
-			"mingw-w64-crt/stdio/ucrt_sprintf.c",
+				"mingw-w64-crt/gdtoa/dmisc.c",
-			"mingw-w64-crt/stdio/ucrt_vfprintf.c",
+				"mingw-w64-crt/gdtoa/gdtoa.c",
-			"mingw-w64-crt/stdio/ucrt_vprintf.c",
+				"mingw-w64-crt/gdtoa/gmisc.c",
-			"mingw-w64-crt/stdio/ucrt_vsnprintf.c",
+				"mingw-w64-crt/gdtoa/misc.c",
-			"mingw-w64-crt/stdio/ucrt_vsprintf.c",
+				"mingw-w64-crt/math/x86/exp2.S",
+				"mingw-w64-crt/math/x86/trunc.S",
+				"mingw-w64-crt/misc/___mb_cur_max_func.c",
+				"mingw-w64-crt/misc/lc_locale_func.c",
+				"mingw-w64-crt/misc/mbrtowc.c",
+				"mingw-w64-crt/misc/strnlen.c",
+				"mingw-w64-crt/misc/wcrtomb.c",
+				"mingw-w64-crt/misc/wcsnlen.c",
+				"mingw-w64-crt/stdio/acrt_iob_func.c",
+				"mingw-w64-crt/stdio/mingw_lock.c",
+				"mingw-w64-crt/stdio/mingw_pformat.c",
+				"mingw-w64-crt/stdio/mingw_vfprintf.c",
+				"mingw-w64-crt/stdio/mingw_vsnprintf.c",
+			}
+		} else {
+			// Anything somewhat modern (amd64, arm64) uses UCRT.
+			sources = []string{
+				"mingw-w64-crt/stdio/ucrt_fprintf.c",
+				"mingw-w64-crt/stdio/ucrt_fwprintf.c",
+				"mingw-w64-crt/stdio/ucrt_printf.c",
+				"mingw-w64-crt/stdio/ucrt_snprintf.c",
+				"mingw-w64-crt/stdio/ucrt_sprintf.c",
+				"mingw-w64-crt/stdio/ucrt_vfprintf.c",
+				"mingw-w64-crt/stdio/ucrt_vprintf.c",
+				"mingw-w64-crt/stdio/ucrt_vsnprintf.c",
+				"mingw-w64-crt/stdio/ucrt_vsprintf.c",
+			}
 		}
 		return sources, nil
 	},
@@ -63,27 +100,41 @@ var libMinGW = Library{
 func makeMinGWExtraLibs(tmpdir, goarch string) []*compileJob {
 	var jobs []*compileJob
 	root := goenv.Get("TINYGOROOT")
-	// Normally all the api-ms-win-crt-*.def files are all compiled to a single
+	var libs []string
-	// .lib file. But to simplify things, we're going to leave them as separate
+	if goarch == "386" {
-	// files.
+		libs = []string{
-	for _, name := range []string{
+			// x86 uses msvcrt.dll instead of UCRT for compatibility with old
-		"kernel32.def.in",
+			// Windows versions.
-		"api-ms-win-crt-conio-l1-1-0.def",
+			"advapi32.def.in",
-		"api-ms-win-crt-convert-l1-1-0.def.in",
+			"kernel32.def.in",
-		"api-ms-win-crt-environment-l1-1-0.def",
+			"msvcrt.def.in",
-		"api-ms-win-crt-filesystem-l1-1-0.def",
+		}
-		"api-ms-win-crt-heap-l1-1-0.def",
+	} else {
-		"api-ms-win-crt-locale-l1-1-0.def",
+		// Use the modernized UCRT on new systems.
-		"api-ms-win-crt-math-l1-1-0.def.in",
+		// Normally all the api-ms-win-crt-*.def files are all compiled to a
-		"api-ms-win-crt-multibyte-l1-1-0.def",
+		// single .lib file. But to simplify things, we're going to leave them
-		"api-ms-win-crt-private-l1-1-0.def.in",
+		// as separate files.
-		"api-ms-win-crt-process-l1-1-0.def",
+		libs = []string{
-		"api-ms-win-crt-runtime-l1-1-0.def.in",
+			"advapi32.def.in",
-		"api-ms-win-crt-stdio-l1-1-0.def",
+			"kernel32.def.in",
-		"api-ms-win-crt-string-l1-1-0.def",
+			"api-ms-win-crt-conio-l1-1-0.def",
-		"api-ms-win-crt-time-l1-1-0.def",
+			"api-ms-win-crt-convert-l1-1-0.def.in",
-		"api-ms-win-crt-utility-l1-1-0.def",
+			"api-ms-win-crt-environment-l1-1-0.def",
-	} {
+			"api-ms-win-crt-filesystem-l1-1-0.def",
+			"api-ms-win-crt-heap-l1-1-0.def",
+			"api-ms-win-crt-locale-l1-1-0.def",
+			"api-ms-win-crt-math-l1-1-0.def.in",
+			"api-ms-win-crt-multibyte-l1-1-0.def",
+			"api-ms-win-crt-private-l1-1-0.def.in",
+			"api-ms-win-crt-process-l1-1-0.def",
+			"api-ms-win-crt-runtime-l1-1-0.def.in",
+			"api-ms-win-crt-stdio-l1-1-0.def",
+			"api-ms-win-crt-string-l1-1-0.def",
+			"api-ms-win-crt-time-l1-1-0.def",
+			"api-ms-win-crt-utility-l1-1-0.def",
+		}
+	}
+	for _, name := range libs {
 		outpath := filepath.Join(tmpdir, filepath.Base(name)+".lib")
 		inpath := filepath.Join(root, "lib/mingw-w64/mingw-w64-crt/lib-common/"+name)
 		job := &compileJob{
@@ -93,6 +144,9 @@ func makeMinGWExtraLibs(tmpdir, goarch string) []*compileJob {
 				defpath := inpath
 				var archDef, emulation string
 				switch goarch {
+				case "386":
+					archDef = "-DDEF_I386"
+					emulation = "i386pe"
 				case "amd64":
 					archDef = "-DDEF_X64"
 					emulation = "i386pep"

builder/musl.go

@@ -12,6 +12,45 @@ import (
 	"github.com/tinygo-org/tinygo/goenv"
 )
 
+// Create the alltypes.h file from the appropriate alltypes.h.in files.
+func buildMuslAllTypes(arch, muslDir, outputBitsDir string) error {
+	// Create the file alltypes.h.
+	f, err := os.Create(filepath.Join(outputBitsDir, "alltypes.h"))
+	if err != nil {
+		return err
+	}
+	infiles := []string{
+		filepath.Join(muslDir, "arch", arch, "bits", "alltypes.h.in"),
+		filepath.Join(muslDir, "include", "alltypes.h.in"),
+	}
+	for _, infile := range infiles {
+		data, err := os.ReadFile(infile)
+		if err != nil {
+			return err
+		}
+		lines := strings.Split(string(data), "\n")
+		for _, line := range lines {
+			if strings.HasPrefix(line, "TYPEDEF ") {
+				matches := regexp.MustCompile(`TYPEDEF (.*) ([^ ]*);`).FindStringSubmatch(line)
+				value := matches[1]
+				name := matches[2]
+				line = fmt.Sprintf("#if defined(__NEED_%s) && !defined(__DEFINED_%s)\ntypedef %s %s;\n#define __DEFINED_%s\n#endif\n", name, name, value, name, name)
+			}
+			if strings.HasPrefix(line, "STRUCT ") {
+				matches := regexp.MustCompile(`STRUCT * ([^ ]*) (.*);`).FindStringSubmatch(line)
+				name := matches[1]
+				value := matches[2]
+				line = fmt.Sprintf("#if defined(__NEED_struct_%s) && !defined(__DEFINED_struct_%s)\nstruct %s %s;\n#define __DEFINED_struct_%s\n#endif\n", name, name, name, value, name)
+			}
+			_, err := f.WriteString(line + "\n")
+			if err != nil {
+				return err
+			}
+		}
+	}
+	return f.Close()
+}
+
 var libMusl = Library{
 	name: "musl",
 	makeHeaders: func(target, includeDir string) error {
@@ -24,41 +63,13 @@ var libMusl = Library{
 		arch := compileopts.MuslArchitecture(target)
 		muslDir := filepath.Join(goenv.Get("TINYGOROOT"), "lib", "musl")
 
-		// Create the file alltypes.h.
+		err = buildMuslAllTypes(arch, muslDir, bits)
-		f, err := os.Create(filepath.Join(bits, "alltypes.h"))
 		if err != nil {
 			return err
 		}
-		infiles := []string{
-			filepath.Join(muslDir, "arch", arch, "bits", "alltypes.h.in"),
-			filepath.Join(muslDir, "include", "alltypes.h.in"),
-		}
-		for _, infile := range infiles {
-			data, err := os.ReadFile(infile)
-			if err != nil {
-				return err
-			}
-			lines := strings.Split(string(data), "\n")
-			for _, line := range lines {
-				if strings.HasPrefix(line, "TYPEDEF ") {
-					matches := regexp.MustCompile(`TYPEDEF (.*) ([^ ]*);`).FindStringSubmatch(line)
-					value := matches[1]
-					name := matches[2]
-					line = fmt.Sprintf("#if defined(__NEED_%s) && !defined(__DEFINED_%s)\ntypedef %s %s;\n#define __DEFINED_%s\n#endif\n", name, name, value, name, name)
-				}
-				if strings.HasPrefix(line, "STRUCT ") {
-					matches := regexp.MustCompile(`STRUCT * ([^ ]*) (.*);`).FindStringSubmatch(line)
-					name := matches[1]
-					value := matches[2]
-					line = fmt.Sprintf("#if defined(__NEED_struct_%s) && !defined(__DEFINED_struct_%s)\nstruct %s %s;\n#define __DEFINED_struct_%s\n#endif\n", name, name, name, value, name)
-				}
-				f.WriteString(line + "\n")
-			}
-		}
-		f.Close()
 
 		// Create the file syscall.h.
-		f, err = os.Create(filepath.Join(bits, "syscall.h"))
+		f, err := os.Create(filepath.Join(bits, "syscall.h"))
 		if err != nil {
 			return err
 		}
@@ -110,9 +121,10 @@ var libMusl = Library{
 		return cflags
 	},
 	sourceDir: func() string { return filepath.Join(goenv.Get("TINYGOROOT"), "lib/musl/src") },
-	librarySources: func(target string) ([]string, error) {
+	librarySources: func(target string, _ bool) ([]string, error) {
 		arch := compileopts.MuslArchitecture(target)
 		globs := []string{
+			"conf/*.c",
 			"ctype/*.c",
 			"env/*.c",
 			"errno/*.c",

builder/picolibc.go

@@ -43,7 +43,7 @@ var libPicolibc = Library{
 		}
 	},
 	sourceDir: func() string { return filepath.Join(goenv.Get("TINYGOROOT"), "lib/picolibc/newlib") },
-	librarySources: func(target string) ([]string, error) {
+	librarySources: func(target string, _ bool) ([]string, error) {
 		sources := append([]string(nil), picolibcSources...)
 		if !strings.HasPrefix(target, "avr") {
 			// Small chips without long jumps can't compile many files (printf,

builder/sizes.go

@@ -490,7 +490,7 @@ func loadProgramSize(path string, packagePathMap map[string]string) (*programSiz
 				continue
 			}
 			if section.Type == elf.SHT_NOBITS {
-				if section.Name == ".stack" {
+				if strings.HasPrefix(section.Name, ".stack") {
 					// TinyGo emits stack sections on microcontroller using the
 					// ".stack" name.
 					// This is a bit ugly, but I don't think there is a way to

builder/sizes_test.go

@@ -42,9 +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", 4560, 280, 0, 2268},
+		{"hifive1b", "examples/echo", 4556, 280, 0, 2264},
-		{"microbit", "examples/serial", 2916, 388, 8, 2272},
+		{"microbit", "examples/serial", 2920, 388, 8, 2272},
-		{"wioterminal", "examples/pininterrupt", 7359, 1489, 116, 6912},
+		{"wioterminal", "examples/pininterrupt", 7379, 1489, 116, 6912},
 
 		// TODO: also check wasm. Right now this is difficult, because
 		// wasm binaries are run through wasm-opt and therefore the

builder/wasilibc.go

@@ -0,0 +1,284 @@
+package builder
+
+import (
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+
+	"github.com/tinygo-org/tinygo/goenv"
+)
+
+var libWasiLibc = Library{
+	name: "wasi-libc",
+	makeHeaders: func(target, includeDir string) error {
+		bits := filepath.Join(includeDir, "bits")
+		err := os.Mkdir(bits, 0777)
+		if err != nil {
+			return err
+		}
+
+		muslDir := filepath.Join(goenv.Get("TINYGOROOT"), "lib", "wasi-libc/libc-top-half/musl")
+		err = buildMuslAllTypes("wasm32", muslDir, bits)
+		if err != nil {
+			return err
+		}
+
+		// See MUSL_OMIT_HEADERS in the Makefile.
+		omitHeaders := map[string]struct{}{
+			"syslog.h":   {},
+			"wait.h":     {},
+			"ucontext.h": {},
+			"paths.h":    {},
+			"utmp.h":     {},
+			"utmpx.h":    {},
+			"lastlog.h":  {},
+			"elf.h":      {},
+			"link.h":     {},
+			"pwd.h":      {},
+			"shadow.h":   {},
+			"grp.h":      {},
+			"mntent.h":   {},
+			"netdb.h":    {},
+			"resolv.h":   {},
+			"pty.h":      {},
+			"dlfcn.h":    {},
+			"setjmp.h":   {},
+			"ulimit.h":   {},
+			"wordexp.h":  {},
+			"spawn.h":    {},
+			"termios.h":  {},
+			"libintl.h":  {},
+			"aio.h":      {},
+
+			"stdarg.h": {},
+			"stddef.h": {},
+
+			"pthread.h": {},
+		}
+
+		for _, glob := range [][2]string{
+			{"libc-bottom-half/headers/public/*.h", ""},
+			{"libc-bottom-half/headers/public/wasi/*.h", "wasi"},
+			{"libc-top-half/musl/arch/wasm32/bits/*.h", "bits"},
+			{"libc-top-half/musl/include/*.h", ""},
+			{"libc-top-half/musl/include/netinet/*.h", "netinet"},
+			{"libc-top-half/musl/include/sys/*.h", "sys"},
+		} {
+			matches, _ := filepath.Glob(filepath.Join(goenv.Get("TINYGOROOT"), "lib/wasi-libc", glob[0]))
+			outDir := filepath.Join(includeDir, glob[1])
+			os.MkdirAll(outDir, 0o777)
+			for _, match := range matches {
+				name := filepath.Base(match)
+				if _, ok := omitHeaders[name]; ok {
+					continue
+				}
+				data, err := os.ReadFile(match)
+				if err != nil {
+					return err
+				}
+				err = os.WriteFile(filepath.Join(outDir, name), data, 0o666)
+				if err != nil {
+					return err
+				}
+			}
+		}
+
+		return nil
+	},
+	cflags: func(target, headerPath string) []string {
+		libcDir := filepath.Join(goenv.Get("TINYGOROOT"), "lib/wasi-libc")
+		return []string{
+			"-Werror",
+			"-Wall",
+			"-std=gnu11",
+			"-nostdlibinc",
+			"-mnontrapping-fptoint", "-msign-ext", "-mbulk-memory",
+			"-Wno-null-pointer-arithmetic", "-Wno-unused-parameter", "-Wno-sign-compare", "-Wno-unused-variable", "-Wno-unused-function", "-Wno-ignored-attributes", "-Wno-missing-braces", "-Wno-ignored-pragmas", "-Wno-unused-but-set-variable", "-Wno-unknown-warning-option",
+			"-Wno-parentheses", "-Wno-shift-op-parentheses", "-Wno-bitwise-op-parentheses", "-Wno-logical-op-parentheses", "-Wno-string-plus-int", "-Wno-dangling-else", "-Wno-unknown-pragmas",
+			"-DNDEBUG",
+			"-D__wasilibc_printscan_no_long_double",
+			"-D__wasilibc_printscan_full_support_option=\"long double support is disabled\"",
+			"-DBULK_MEMORY_THRESHOLD=32", // default threshold in wasi-libc
+			"-isystem", headerPath,
+			"-I" + libcDir + "/libc-top-half/musl/src/include",
+			"-I" + libcDir + "/libc-top-half/musl/src/internal",
+			"-I" + libcDir + "/libc-top-half/musl/arch/wasm32",
+			"-I" + libcDir + "/libc-top-half/musl/arch/generic",
+			"-I" + libcDir + "/libc-top-half/headers/private",
+		}
+	},
+	cflagsForFile: func(path string) []string {
+		if strings.HasPrefix(path, "libc-bottom-half"+string(os.PathSeparator)) {
+			libcDir := filepath.Join(goenv.Get("TINYGOROOT"), "lib/wasi-libc")
+			return []string{
+				"-I" + libcDir + "/libc-bottom-half/headers/private",
+				"-I" + libcDir + "/libc-bottom-half/cloudlibc/src/include",
+				"-I" + libcDir + "/libc-bottom-half/cloudlibc/src",
+			}
+		}
+		return nil
+	},
+	sourceDir: func() string { return filepath.Join(goenv.Get("TINYGOROOT"), "lib/wasi-libc") },
+	librarySources: func(target string, libcNeedsMalloc bool) ([]string, error) {
+		type filePattern struct {
+			glob    string
+			exclude []string
+		}
+
+		// See: LIBC_TOP_HALF_MUSL_SOURCES in the Makefile
+		globs := []filePattern{
+			// Top half: mostly musl sources.
+			{glob: "libc-top-half/sources/*.c"},
+			{glob: "libc-top-half/musl/src/conf/*.c"},
+			{glob: "libc-top-half/musl/src/internal/*.c", exclude: []string{
+				"procfdname.c", "syscall.c", "syscall_ret.c", "vdso.c", "version.c",
+			}},
+			{glob: "libc-top-half/musl/src/locale/*.c", exclude: []string{
+				"dcngettext.c", "textdomain.c", "bind_textdomain_codeset.c"}},
+			{glob: "libc-top-half/musl/src/math/*.c", exclude: []string{
+				"__signbit.c", "__signbitf.c", "__signbitl.c",
+				"__fpclassify.c", "__fpclassifyf.c", "__fpclassifyl.c",
+				"ceilf.c", "ceil.c",
+				"floorf.c", "floor.c",
+				"truncf.c", "trunc.c",
+				"rintf.c", "rint.c",
+				"nearbyintf.c", "nearbyint.c",
+				"sqrtf.c", "sqrt.c",
+				"fabsf.c", "fabs.c",
+				"copysignf.c", "copysign.c",
+				"fminf.c", "fmaxf.c",
+				"fmin.c", "fmax.c,",
+			}},
+			{glob: "libc-top-half/musl/src/multibyte/*.c"},
+			{glob: "libc-top-half/musl/src/stdio/*.c", exclude: []string{
+				"vfwscanf.c", "vfwprintf.c", // long double is unsupported
+				"__lockfile.c", "flockfile.c", "funlockfile.c", "ftrylockfile.c",
+				"rename.c",
+				"tmpnam.c", "tmpfile.c", "tempnam.c",
+				"popen.c", "pclose.c",
+				"remove.c",
+				"gets.c"}},
+			{glob: "libc-top-half/musl/src/stdlib/*.c"},
+			{glob: "libc-top-half/musl/src/string/*.c", exclude: []string{
+				"strsignal.c"}},
+
+			// Bottom half: connect top half to WASI equivalents.
+			{glob: "libc-bottom-half/cloudlibc/src/libc/*/*.c"},
+			{glob: "libc-bottom-half/cloudlibc/src/libc/sys/*/*.c"},
+			{glob: "libc-bottom-half/sources/*.c"},
+		}
+
+		// We're using the Boehm GC, so we need a heap implementation in the libc.
+		if libcNeedsMalloc {
+			globs = append(globs, filePattern{glob: "dlmalloc/src/dlmalloc.c"})
+		}
+
+		// See: LIBC_TOP_HALF_MUSL_SOURCES in the Makefile
+		sources := []string{
+			"libc-top-half/musl/src/misc/a64l.c",
+			"libc-top-half/musl/src/misc/basename.c",
+			"libc-top-half/musl/src/misc/dirname.c",
+			"libc-top-half/musl/src/misc/ffs.c",
+			"libc-top-half/musl/src/misc/ffsl.c",
+			"libc-top-half/musl/src/misc/ffsll.c",
+			"libc-top-half/musl/src/misc/fmtmsg.c",
+			"libc-top-half/musl/src/misc/getdomainname.c",
+			"libc-top-half/musl/src/misc/gethostid.c",
+			"libc-top-half/musl/src/misc/getopt.c",
+			"libc-top-half/musl/src/misc/getopt_long.c",
+			"libc-top-half/musl/src/misc/getsubopt.c",
+			"libc-top-half/musl/src/misc/uname.c",
+			"libc-top-half/musl/src/misc/nftw.c",
+			"libc-top-half/musl/src/errno/strerror.c",
+			"libc-top-half/musl/src/network/htonl.c",
+			"libc-top-half/musl/src/network/htons.c",
+			"libc-top-half/musl/src/network/ntohl.c",
+			"libc-top-half/musl/src/network/ntohs.c",
+			"libc-top-half/musl/src/network/inet_ntop.c",
+			"libc-top-half/musl/src/network/inet_pton.c",
+			"libc-top-half/musl/src/network/inet_aton.c",
+			"libc-top-half/musl/src/network/in6addr_any.c",
+			"libc-top-half/musl/src/network/in6addr_loopback.c",
+			"libc-top-half/musl/src/fenv/fenv.c",
+			"libc-top-half/musl/src/fenv/fesetround.c",
+			"libc-top-half/musl/src/fenv/feupdateenv.c",
+			"libc-top-half/musl/src/fenv/fesetexceptflag.c",
+			"libc-top-half/musl/src/fenv/fegetexceptflag.c",
+			"libc-top-half/musl/src/fenv/feholdexcept.c",
+			"libc-top-half/musl/src/exit/exit.c",
+			"libc-top-half/musl/src/exit/atexit.c",
+			"libc-top-half/musl/src/exit/assert.c",
+			"libc-top-half/musl/src/exit/quick_exit.c",
+			"libc-top-half/musl/src/exit/at_quick_exit.c",
+			"libc-top-half/musl/src/time/strftime.c",
+			"libc-top-half/musl/src/time/asctime.c",
+			"libc-top-half/musl/src/time/asctime_r.c",
+			"libc-top-half/musl/src/time/ctime.c",
+			"libc-top-half/musl/src/time/ctime_r.c",
+			"libc-top-half/musl/src/time/wcsftime.c",
+			"libc-top-half/musl/src/time/strptime.c",
+			"libc-top-half/musl/src/time/difftime.c",
+			"libc-top-half/musl/src/time/timegm.c",
+			"libc-top-half/musl/src/time/ftime.c",
+			"libc-top-half/musl/src/time/gmtime.c",
+			"libc-top-half/musl/src/time/gmtime_r.c",
+			"libc-top-half/musl/src/time/timespec_get.c",
+			"libc-top-half/musl/src/time/getdate.c",
+			"libc-top-half/musl/src/time/localtime.c",
+			"libc-top-half/musl/src/time/localtime_r.c",
+			"libc-top-half/musl/src/time/mktime.c",
+			"libc-top-half/musl/src/time/__tm_to_secs.c",
+			"libc-top-half/musl/src/time/__month_to_secs.c",
+			"libc-top-half/musl/src/time/__secs_to_tm.c",
+			"libc-top-half/musl/src/time/__year_to_secs.c",
+			"libc-top-half/musl/src/time/__tz.c",
+			"libc-top-half/musl/src/fcntl/creat.c",
+			"libc-top-half/musl/src/dirent/alphasort.c",
+			"libc-top-half/musl/src/dirent/versionsort.c",
+			"libc-top-half/musl/src/env/__stack_chk_fail.c",
+			"libc-top-half/musl/src/env/clearenv.c",
+			"libc-top-half/musl/src/env/getenv.c",
+			"libc-top-half/musl/src/env/putenv.c",
+			"libc-top-half/musl/src/env/setenv.c",
+			"libc-top-half/musl/src/env/unsetenv.c",
+			"libc-top-half/musl/src/unistd/posix_close.c",
+			"libc-top-half/musl/src/stat/futimesat.c",
+			"libc-top-half/musl/src/legacy/getpagesize.c",
+			"libc-top-half/musl/src/thread/thrd_sleep.c",
+		}
+
+		basepath := goenv.Get("TINYGOROOT") + "/lib/wasi-libc/"
+		for _, pattern := range globs {
+			matches, err := filepath.Glob(basepath + pattern.glob)
+			if err != nil {
+				// From the documentation:
+				// > Glob ignores file system errors such as I/O errors reading
+				// > directories. The only possible returned error is
+				// > ErrBadPattern, when pattern is malformed.
+				// So the only possible error is when the (statically defined)
+				// pattern is wrong. In other words, a programming bug.
+				return nil, fmt.Errorf("wasi-libc: could not glob source dirs: %w", err)
+			}
+			if len(matches) == 0 {
+				return nil, fmt.Errorf("wasi-libc: did not find any files for pattern %#v", pattern)
+			}
+			excludeSet := map[string]struct{}{}
+			for _, exclude := range pattern.exclude {
+				excludeSet[exclude] = struct{}{}
+			}
+			for _, match := range matches {
+				if _, ok := excludeSet[filepath.Base(match)]; ok {
+					continue
+				}
+				relpath, err := filepath.Rel(basepath, match)
+				if err != nil {
+					// Not sure if this is even possible.
+					return nil, err
+				}
+				sources = append(sources, relpath)
+			}
+		}
+		return sources, nil
+	},
+}

builder/wasmbuiltins.go

@@ -29,6 +29,8 @@ var libWasmBuiltins = Library{
 			"-Wall",
 			"-std=gnu11",
 			"-nostdlibinc",
+			"-mnontrapping-fptoint", // match wasm-unknown (default on in LLVM 20)
+			"-mno-bulk-memory",      // same here
 			"-isystem", libcDir + "/libc-top-half/musl/arch/wasm32",
 			"-isystem", libcDir + "/libc-top-half/musl/arch/generic",
 			"-isystem", libcDir + "/libc-top-half/musl/src/internal",
@@ -39,7 +41,7 @@ var libWasmBuiltins = Library{
 		}
 	},
 	sourceDir: func() string { return filepath.Join(goenv.Get("TINYGOROOT"), "lib/wasi-libc") },
-	librarySources: func(target string) ([]string, error) {
+	librarySources: func(target string, _ bool) ([]string, error) {
 		return []string{
 			// memory builtins needed for llvm.memcpy.*, llvm.memmove.*, and
 			// llvm.memset.* LLVM intrinsics.

cgo/libclang_config_llvm19.go

@@ -1,4 +1,4 @@
-//go:build !byollvm && !llvm15 && !llvm16 && !llvm17 && !llvm18
+//go:build !byollvm && llvm19
 
 package cgo
 

cgo/libclang_config_llvm20.go

@@ -0,0 +1,15 @@
+//go:build !byollvm && !llvm15 && !llvm16 && !llvm17 && !llvm18 && !llvm19
+
+package cgo
+
+/*
+#cgo linux        CFLAGS:  -I/usr/include/llvm-20 -I/usr/include/llvm-c-20 -I/usr/lib/llvm-20/include
+#cgo darwin,amd64 CFLAGS:  -I/usr/local/opt/llvm@20/include
+#cgo darwin,arm64 CFLAGS:  -I/opt/homebrew/opt/llvm@20/include
+#cgo freebsd      CFLAGS:  -I/usr/local/llvm20/include
+#cgo linux        LDFLAGS: -L/usr/lib/llvm-20/lib -lclang
+#cgo darwin,amd64 LDFLAGS: -L/usr/local/opt/llvm@20/lib -lclang
+#cgo darwin,arm64 LDFLAGS: -L/opt/homebrew/opt/llvm@20/lib -lclang
+#cgo freebsd      LDFLAGS: -L/usr/local/llvm20/lib -lclang
+*/
+import "C"

compileopts/config.go

@@ -8,12 +8,24 @@ import (
 	"os"
 	"path/filepath"
 	"regexp"
+	"strconv"
 	"strings"
 
 	"github.com/google/shlex"
 	"github.com/tinygo-org/tinygo/goenv"
 )
 
+// Library versions. Whenever an existing library is changed, this number should
+// be added/increased so that existing caches are invalidated.
+//
+// (This is a bit of a layering violation, this should really be part of the
+// builder.Library struct but that's hard to do since we want to know the
+// library path in advance in several places).
+var libVersions = map[string]int{
+	"musl":  3,
+	"bdwgc": 2,
+}
+
 // Config keeps all configuration affecting the build in a single struct.
 type Config struct {
 	Options        *Options
@@ -99,6 +111,11 @@ func (c *Config) BuildTags() []string {
 		"math_big_pure_go",                           // to get math/big to work
 		"gc." + c.GC(), "scheduler." + c.Scheduler(), // used inside the runtime package
 		"serial." + c.Serial()}...) // used inside the machine package
+	switch c.Scheduler() {
+	case "threads", "cores":
+	default:
+		tags = append(tags, "tinygo.unicore")
+	}
 	for i := 1; i <= c.GoMinorVersion; i++ {
 		tags = append(tags, fmt.Sprintf("go1.%d", i))
 	}
@@ -122,7 +139,7 @@ func (c *Config) GC() string {
 // that can be traced by the garbage collector.
 func (c *Config) NeedsStackObjects() bool {
 	switch c.GC() {
-	case "conservative", "custom", "precise":
+	case "conservative", "custom", "precise", "boehm":
 		for _, tag := range c.BuildTags() {
 			if tag == "tinygo.wasm" {
 				return true
@@ -247,10 +264,18 @@ func MuslArchitecture(triple string) string {
 	return CanonicalArchName(triple)
 }
 
-// LibcPath returns the path to the libc directory. The libc path will be either
+// Returns true if the libc needs to include malloc, for the libcs where this
-// a precompiled libc shipped with a TinyGo build, or a libc path in the cache
+// matters.
-// directory (which might not yet be built).
+func (c *Config) LibcNeedsMalloc() bool {
-func (c *Config) LibcPath(name string) (path string, precompiled bool) {
+	if c.GC() == "boehm" && c.Target.Libc == "wasi-libc" {
+		return true
+	}
+	return false
+}
+
+// LibraryPath returns the path to the library build directory. The path will be
+// a library path in the cache directory (which might not yet be built).
+func (c *Config) LibraryPath(name string) string {
 	archname := c.Triple()
 	if c.CPU() != "" {
 		archname += "-" + c.CPU()
@@ -266,17 +291,19 @@ func (c *Config) LibcPath(name string) (path string, precompiled bool) {
 		archname += "-" + c.Target.Libc
 	}
 
-	// Try to load a precompiled library.
+	// Append a version string, if this library has a version.
-	precompiledDir := filepath.Join(goenv.Get("TINYGOROOT"), "pkg", archname, name)
+	if v, ok := libVersions[name]; ok {
-	if _, err := os.Stat(precompiledDir); err == nil {
+		archname += "-v" + strconv.Itoa(v)
-		// Found a precompiled library for this OS/architecture. Return the path
+	}
-		// directly.
+
-		return precompiledDir, true
+	options := ""
+	if c.LibcNeedsMalloc() {
+		options += "+malloc"
 	}
 
 	// No precompiled library found. Determine the path name that will be used
 	// in the build cache.
-	return filepath.Join(goenv.Get("GOCACHE"), name+"-"+archname), false
+	return filepath.Join(goenv.Get("GOCACHE"), name+options+"-"+archname)
 }
 
 // DefaultBinaryExtension returns the default extension for binaries, such as
@@ -360,7 +387,7 @@ func (c *Config) LibcCFlags() []string {
 	case "picolibc":
 		root := goenv.Get("TINYGOROOT")
 		picolibcDir := filepath.Join(root, "lib", "picolibc", "newlib", "libc")
-		path, _ := c.LibcPath("picolibc")
+		path := c.LibraryPath("picolibc")
 		return []string{
 			"-nostdlibinc",
 			"-isystem", filepath.Join(path, "include"),
@@ -370,7 +397,7 @@ func (c *Config) LibcCFlags() []string {
 		}
 	case "musl":
 		root := goenv.Get("TINYGOROOT")
-		path, _ := c.LibcPath("musl")
+		path := c.LibraryPath("musl")
 		arch := MuslArchitecture(c.Triple())
 		return []string{
 			"-nostdlibinc",
@@ -380,26 +407,36 @@ func (c *Config) LibcCFlags() []string {
 			"-isystem", filepath.Join(root, "lib", "musl", "include"),
 		}
 	case "wasi-libc":
-		root := goenv.Get("TINYGOROOT")
+		path := c.LibraryPath("wasi-libc")
 		return []string{
 			"-nostdlibinc",
-			"-isystem", root + "/lib/wasi-libc/sysroot/include",
+			"-isystem", filepath.Join(path, "include"),
 		}
 	case "wasmbuiltins":
 		// nothing to add (library is purely for builtins)
 		return nil
 	case "mingw-w64":
 		root := goenv.Get("TINYGOROOT")
-		path, _ := c.LibcPath("mingw-w64")
+		path := c.LibraryPath("mingw-w64")
-		return []string{
+		cflags := []string{
 			"-nostdlibinc",
 			"-isystem", filepath.Join(path, "include"),
 			"-isystem", filepath.Join(root, "lib", "mingw-w64", "mingw-w64-headers", "crt"),
 			"-isystem", filepath.Join(root, "lib", "mingw-w64", "mingw-w64-headers", "include"),
 			"-isystem", filepath.Join(root, "lib", "mingw-w64", "mingw-w64-headers", "defaults", "include"),
-			"-D_UCRT",
-			"-D_WIN32_WINNT=0x0a00", // target Windows 10
 		}
+		if c.GOARCH() == "386" {
+			cflags = append(cflags,
+				"-D__MSVCRT_VERSION__=0x700", // Microsoft Visual C++ .NET 2002
+				"-D_WIN32_WINNT=0x0501",      // target Windows XP
+			)
+		} else {
+			cflags = append(cflags,
+				"-D_UCRT",
+				"-D_WIN32_WINNT=0x0a00", // target Windows 10
+			)
+		}
+		return cflags
 	case "":
 		// No libc specified, nothing to add.
 		return nil

compileopts/options.go

@@ -10,7 +10,7 @@ import (
 var (
 	validBuildModeOptions     = []string{"default", "c-shared", "wasi-legacy"}
 	validGCOptions            = []string{"none", "leaking", "conservative", "custom", "precise", "boehm"}
-	validSchedulerOptions     = []string{"none", "tasks", "asyncify"}
+	validSchedulerOptions     = []string{"none", "tasks", "asyncify", "threads", "cores"}
 	validSerialOptions        = []string{"none", "uart", "usb", "rtt"}
 	validPrintSizeOptions     = []string{"none", "short", "full", "html"}
 	validPanicStrategyOptions = []string{"print", "trap"}
@@ -43,6 +43,7 @@ type Options struct {
 	PrintCommands   func(cmd string, args ...string) `json:"-"`
 	Semaphore       chan struct{}                    `json:"-"` // -p flag controls cap
 	Debug           bool
+	Nobounds        bool
 	PrintSizes      string
 	PrintAllocs     *regexp.Regexp // regexp string
 	PrintStacks     bool
@@ -52,7 +53,6 @@ type Options struct {
 	Programmer      string
 	OpenOCDCommands []string
 	LLVMFeatures    string
-	PrintJSON       bool
 	Monitor         bool
 	BaudRate        int
 	Timeout         time.Duration

compileopts/options_test.go

@@ -10,7 +10,7 @@ import (
 func TestVerifyOptions(t *testing.T) {
 
 	expectedGCError := errors.New(`invalid gc option 'incorrect': valid values are none, leaking, conservative, custom, precise, boehm`)
-	expectedSchedulerError := errors.New(`invalid scheduler option 'incorrect': valid values are none, tasks, asyncify`)
+	expectedSchedulerError := errors.New(`invalid scheduler option 'incorrect': valid values are none, tasks, asyncify, threads, cores`)
 	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`)
 

compileopts/target.go

@@ -262,7 +262,6 @@ func defaultTarget(options *Options) (*TargetSpec, error) {
 		GOOS:             options.GOOS,
 		GOARCH:           options.GOARCH,
 		BuildTags:        []string{options.GOOS, options.GOARCH},
-		Scheduler:        "tasks",
 		Linker:           "cc",
 		DefaultStackSize: 1024 * 64, // 64kB
 		GDB:              []string{"gdb"},
@@ -386,6 +385,7 @@ func defaultTarget(options *Options) (*TargetSpec, error) {
 			platformVersion = "11.0.0" // first macosx platform with arm64 support
 		}
 		llvmvendor = "apple"
+		spec.Scheduler = "tasks"
 		spec.Linker = "ld.lld"
 		spec.Libc = "darwin-libSystem"
 		// Use macosx* instead of darwin, otherwise darwin/arm64 will refer to
@@ -404,6 +404,7 @@ func defaultTarget(options *Options) (*TargetSpec, error) {
 			"src/runtime/signal.c")
 	case "linux":
 		spec.GC = "boehm"
+		spec.Scheduler = "threads"
 		spec.Linker = "ld.lld"
 		spec.RTLib = "compiler-rt"
 		spec.Libc = "musl"
@@ -424,20 +425,30 @@ func defaultTarget(options *Options) (*TargetSpec, error) {
 		}
 		spec.ExtraFiles = append(spec.ExtraFiles,
 			"src/internal/futex/futex_linux.c",
+			"src/internal/task/task_threads.c",
 			"src/runtime/runtime_unix.c",
 			"src/runtime/signal.c")
 	case "windows":
 		spec.GC = "boehm"
+		spec.Scheduler = "tasks"
 		spec.Linker = "ld.lld"
 		spec.Libc = "mingw-w64"
-		// Note: using a medium code model, low image base and no ASLR
-		// because Go doesn't really need those features. ASLR patches
-		// around issues for unsafe languages like C/C++ that are not
-		// normally present in Go (without explicitly opting in).
-		// For more discussion:
-		// https://groups.google.com/g/Golang-nuts/c/Jd9tlNc6jUE/m/Zo-7zIP_m3MJ?pli=1
 		switch options.GOARCH {
+		case "386":
+			spec.LDFlags = append(spec.LDFlags,
+				"-m", "i386pe",
+				"--major-os-version", "4",
+				"--major-subsystem-version", "4",
+			)
+			// __udivdi3 is not present in ucrt it seems.
+			spec.RTLib = "compiler-rt"
 		case "amd64":
+			// Note: using a medium code model, low image base and no ASLR
+			// because Go doesn't really need those features. ASLR patches
+			// around issues for unsafe languages like C/C++ that are not
+			// normally present in Go (without explicitly opting in).
+			// For more discussion:
+			// https://groups.google.com/g/Golang-nuts/c/Jd9tlNc6jUE/m/Zo-7zIP_m3MJ?pli=1
 			spec.LDFlags = append(spec.LDFlags,
 				"-m", "i386pep",
 				"--image-base", "0x400000",
@@ -459,6 +470,12 @@ func defaultTarget(options *Options) (*TargetSpec, error) {
 		return nil, fmt.Errorf("unknown GOOS=%s", options.GOOS)
 	}
 
+	if spec.GC == "boehm" {
+		// Add this file only when needed. This fixes a build failure on
+		// Windows.
+		spec.ExtraFiles = append(spec.ExtraFiles, "src/runtime/gc_boehm.c")
+	}
+
 	// Target triples (which actually have four components, but are called
 	// triples for historical reasons) have the form:
 	//   arch-vendor-os-environment

compiler/calls.go

@@ -76,7 +76,15 @@ func (b *builder) createCall(fnType llvm.Type, fn llvm.Value, args []llvm.Value,
 		fragments := b.expandFormalParam(arg)
 		expanded = append(expanded, fragments...)
 	}
-	return b.CreateCall(fnType, fn, expanded, name)
+	call := b.CreateCall(fnType, fn, expanded, name)
+	if !fn.IsAFunction().IsNil() {
+		if cc := fn.FunctionCallConv(); cc != llvm.CCallConv {
+			// Set a different calling convention if needed.
+			// This is needed for GetModuleHandleExA on Windows, for example.
+			call.SetInstructionCallConv(cc)
+		}
+	}
+	return call
 }
 
 // createInvoke is like createCall but continues execution at the landing pad if

compiler/compiler.go

@@ -58,6 +58,7 @@ type Config struct {
 	MaxStackAlloc      uint64
 	NeedsStackObjects  bool
 	Debug              bool // Whether to emit debug information in the LLVM module.
+	Nobounds           bool // Whether to skip bounds checks
 	PanicStrategy      string
 }
 
@@ -388,7 +389,7 @@ func (c *compilerContext) getLLVMType(goType types.Type) llvm.Type {
 // makeLLVMType creates a LLVM type for a Go type. Don't call this, use
 // getLLVMType instead.
 func (c *compilerContext) makeLLVMType(goType types.Type) llvm.Type {
-	switch typ := goType.(type) {
+	switch typ := types.Unalias(goType).(type) {
 	case *types.Array:
 		elemType := c.getLLVMType(typ.Elem())
 		return llvm.ArrayType(elemType, int(typ.Len()))
@@ -496,6 +497,21 @@ func (c *compilerContext) createDIType(typ types.Type) llvm.Metadata {
 	llvmType := c.getLLVMType(typ)
 	sizeInBytes := c.targetData.TypeAllocSize(llvmType)
 	switch typ := typ.(type) {
+	case *types.Alias:
+		// Implement types.Alias just like types.Named: by treating them like a
+		// C typedef.
+		temporaryMDNode := c.dibuilder.CreateReplaceableCompositeType(llvm.Metadata{}, llvm.DIReplaceableCompositeType{
+			Tag:         dwarf.TagTypedef,
+			SizeInBits:  sizeInBytes * 8,
+			AlignInBits: uint32(c.targetData.ABITypeAlignment(llvmType)) * 8,
+		})
+		c.ditypes[typ] = temporaryMDNode
+		md := c.dibuilder.CreateTypedef(llvm.DITypedef{
+			Type: c.getDIType(types.Unalias(typ)), // TODO: use typ.Rhs in Go 1.23
+			Name: typ.String(),
+		})
+		temporaryMDNode.ReplaceAllUsesWith(md)
+		return md
 	case *types.Array:
 		return c.dibuilder.CreateArrayType(llvm.DIArrayType{
 			SizeInBits:  sizeInBytes * 8,
@@ -858,6 +874,11 @@ func (c *compilerContext) createPackage(irbuilder llvm.Builder, pkg *ssa.Package
 				// Interfaces don't have concrete methods.
 				continue
 			}
+			if _, isalias := member.Type().(*types.Alias); isalias {
+				// Aliases don't need to be redefined, since they just refer to
+				// an already existing type whose methods will be defined.
+				continue
+			}
 
 			// Named type. We should make sure all methods are created.
 			// This includes both functions with pointer receivers and those
@@ -1835,15 +1856,7 @@ func (b *builder) createBuiltin(argTypes []types.Type, argValues []llvm.Value, c
 //
 // This is also where compiler intrinsics are implemented.
 func (b *builder) createFunctionCall(instr *ssa.CallCommon) (llvm.Value, error) {
-	var params []llvm.Value
+	// See if this is an intrinsic function that is handled specially.
-	for _, param := range instr.Args {
-		params = append(params, b.getValue(param, getPos(instr)))
-	}
-
-	// Try to call the function directly for trivially static calls.
-	var callee, context llvm.Value
-	var calleeType llvm.Type
-	exported := false
 	if fn := instr.StaticCallee(); fn != nil {
 		// Direct function call, either to a named or anonymous (directly
 		// applied) function call. If it is anonymous, it may be a closure.
@@ -1879,13 +1892,29 @@ func (b *builder) createFunctionCall(instr *ssa.CallCommon) (llvm.Value, error)
 			return llvm.ConstInt(b.ctx.Int8Type(), panicStrategy, false), nil
 		case name == "runtime/interrupt.New":
 			return b.createInterruptGlobal(instr)
+		case name == "runtime.exportedFuncPtr":
+			_, ptr := b.getFunction(instr.Args[0].(*ssa.Function))
+			return b.CreatePtrToInt(ptr, b.uintptrType, ""), nil
+		case name == "(*runtime/interrupt.Checkpoint).Save":
+			return b.createInterruptCheckpoint(instr.Args[0]), nil
 		case name == "internal/abi.FuncPCABI0":
 			retval := b.createDarwinFuncPCABI0Call(instr)
 			if !retval.IsNil() {
 				return retval, nil
 			}
 		}
+	}
 
+	var params []llvm.Value
+	for _, param := range instr.Args {
+		params = append(params, b.getValue(param, getPos(instr)))
+	}
+
+	// Try to call the function directly for trivially static calls.
+	var callee, context llvm.Value
+	var calleeType llvm.Type
+	exported := false
+	if fn := instr.StaticCallee(); fn != nil {
 		calleeType, callee = b.getFunction(fn)
 		info := b.getFunctionInfo(fn)
 		if callee.IsNil() {

compiler/defer.go

@@ -103,10 +103,11 @@ func (b *builder) createLandingPad() {
 	b.CreateBr(b.blockEntries[b.fn.Recover])
 }
 
-// createInvokeCheckpoint saves the function state at the given point, to
+// Create a checkpoint (similar to setjmp). This emits inline assembly that
-// continue at the landing pad if a panic happened. This is implemented using a
+// stores the current program counter inside the ptr address (actually
-// setjmp-like construct.
+// ptr+sizeof(ptr)) and then returns a boolean indicating whether this is the
-func (b *builder) createInvokeCheckpoint() {
+// normal flow (false) or we jumped here from somewhere else (true).
+func (b *builder) createCheckpoint(ptr llvm.Value) llvm.Value {
 	// Construct inline assembly equivalents of setjmp.
 	// The assembly works as follows:
 	//   * All registers (both callee-saved and caller saved) are clobbered
@@ -217,11 +218,19 @@ li a0, 0
 		// This case should have been handled by b.supportsRecover().
 		b.addError(b.fn.Pos(), "unknown architecture for defer: "+b.archFamily())
 	}
-	asmType := llvm.FunctionType(resultType, []llvm.Type{b.deferFrame.Type()}, false)
+	asmType := llvm.FunctionType(resultType, []llvm.Type{b.dataPtrType}, false)
 	asm := llvm.InlineAsm(asmType, asmString, constraints, false, false, 0, false)
-	result := b.CreateCall(asmType, asm, []llvm.Value{b.deferFrame}, "setjmp")
+	result := b.CreateCall(asmType, asm, []llvm.Value{ptr}, "setjmp")
 	result.AddCallSiteAttribute(-1, b.ctx.CreateEnumAttribute(llvm.AttributeKindID("returns_twice"), 0))
 	isZero := b.CreateICmp(llvm.IntEQ, result, llvm.ConstInt(resultType, 0, false), "setjmp.result")
+	return isZero
+}
+
+// createInvokeCheckpoint saves the function state at the given point, to
+// continue at the landing pad if a panic happened. This is implemented using a
+// setjmp-like construct.
+func (b *builder) createInvokeCheckpoint() {
+	isZero := b.createCheckpoint(b.deferFrame)
 	continueBB := b.insertBasicBlock("")
 	b.CreateCondBr(isZero, continueBB, b.landingpad)
 	b.SetInsertPointAtEnd(continueBB)

compiler/inlineasm.go

@@ -249,3 +249,15 @@ func (b *builder) emitCSROperation(call *ssa.CallCommon) (llvm.Value, error) {
 		return llvm.Value{}, b.makeError(call.Pos(), "unknown CSR operation: "+name)
 	}
 }
+
+// Implement runtime/interrupt.Checkpoint.Save. It needs to be implemented
+// directly at the call site. If it isn't implemented directly at the call site
+// (but instead through a function call), it might result in an overwritten
+// stack in the non-jump return case.
+func (b *builder) createInterruptCheckpoint(ptr ssa.Value) llvm.Value {
+	addr := b.getValue(ptr, ptr.Pos())
+	b.createNilCheck(ptr, addr, "deref")
+	stackPointer := b.readStackPointer()
+	b.CreateStore(stackPointer, addr)
+	return b.createCheckpoint(addr)
+}

compiler/interface.go

@@ -122,6 +122,9 @@ func (c *compilerContext) pkgPathPtr(pkgpath string) llvm.Value {
 // This function returns a pointer to the 'kind' field (which might not be the
 // first field in the struct).
 func (c *compilerContext) getTypeCode(typ types.Type) llvm.Value {
+	// Resolve alias types: alias types are resolved at compile time.
+	typ = types.Unalias(typ)
+
 	ms := c.program.MethodSets.MethodSet(typ)
 	hasMethodSet := ms.Len() != 0
 	_, isInterface := typ.Underlying().(*types.Interface)
@@ -512,7 +515,7 @@ var basicTypeNames = [...]string{
 // interface lowering pass to assign type codes as expected by the reflect
 // package. See getTypeCodeNum.
 func getTypeCodeName(t types.Type) (string, bool) {
-	switch t := t.(type) {
+	switch t := types.Unalias(t).(type) {
 	case *types.Named:
 		if t.Obj().Parent() != t.Obj().Pkg().Scope() {
 			return "named:" + t.String() + "$local", true
@@ -942,7 +945,7 @@ func signature(sig *types.Signature) string {
 // normalization around `byte` vs `uint8` for example.
 func typestring(t types.Type) string {
 	// See: https://github.com/golang/go/blob/master/src/go/types/typestring.go
-	switch t := t.(type) {
+	switch t := types.Unalias(t).(type) {
 	case *types.Array:
 		return "[" + strconv.FormatInt(t.Len(), 10) + "]" + typestring(t.Elem())
 	case *types.Basic:

compiler/llvm.go

@@ -452,6 +452,21 @@ func (b *builder) readStackPointer() llvm.Value {
 	return b.CreateCall(stacksave.GlobalValueType(), stacksave, nil, "")
 }
 
+// writeStackPointer emits a LLVM intrinsic call that updates the current stack
+// pointer.
+func (b *builder) writeStackPointer(sp llvm.Value) {
+	name := "llvm.stackrestore.p0"
+	if llvmutil.Version() < 18 {
+		name = "llvm.stackrestore" // backwards compatibility with LLVM 17 and below
+	}
+	stackrestore := b.mod.NamedFunction(name)
+	if stackrestore.IsNil() {
+		fnType := llvm.FunctionType(b.ctx.VoidType(), []llvm.Type{b.dataPtrType}, false)
+		stackrestore = llvm.AddFunction(b.mod, name, fnType)
+	}
+	b.CreateCall(stackrestore.GlobalValueType(), stackrestore, []llvm.Value{sp}, "")
+}
+
 // createZExtOrTrunc lets the input value fit in the output type bits, by zero
 // extending or truncating the integer.
 func (b *builder) createZExtOrTrunc(value llvm.Value, t llvm.Type) llvm.Value {

compiler/map.go

@@ -248,7 +248,7 @@ func (b *builder) createMapIteratorNext(rangeVal ssa.Value, llvmRangeVal, it llv
 // can be compared with runtime.memequal.  Note that padding bytes are undef
 // and can alter two "equal" structs being equal when compared with memequal.
 func hashmapIsBinaryKey(keyType types.Type) bool {
-	switch keyType := keyType.(type) {
+	switch keyType := keyType.Underlying().(type) {
 	case *types.Basic:
 		return keyType.Info()&(types.IsBoolean|types.IsInteger) != 0
 	case *types.Pointer:
@@ -263,8 +263,6 @@ func hashmapIsBinaryKey(keyType types.Type) bool {
 		return true
 	case *types.Array:
 		return hashmapIsBinaryKey(keyType.Elem())
-	case *types.Named:
-		return hashmapIsBinaryKey(keyType.Underlying())
 	default:
 		return false
 	}

compiler/symbol.go

@@ -208,6 +208,12 @@ func (c *compilerContext) getFunction(fn *ssa.Function) (llvm.Type, llvm.Value)
 			// > circumstances, and should not be exposed to source languages.
 			llvmutil.AppendToGlobal(c.mod, "llvm.compiler.used", llvmFn)
 		}
+	case "GetModuleHandleExA", "GetProcAddress", "GetSystemInfo", "GetSystemTimeAsFileTime", "LoadLibraryExW", "QueryPerformanceCounter", "QueryPerformanceFrequency", "QueryUnbiasedInterruptTime", "SetEnvironmentVariableA", "Sleep", "SystemFunction036", "VirtualAlloc":
+		// On Windows we need to use a special calling convention for some
+		// external calls.
+		if c.GOOS == "windows" && c.GOARCH == "386" {
+			llvmFn.SetFunctionCallConv(llvm.X86StdcallCallConv)
+		}
 	}
 
 	// External/exported functions may not retain pointer values.
@@ -438,6 +444,10 @@ func (c *compilerContext) parsePragmas(info *functionInfo, f *ssa.Function) {
 			}
 		}
 	}
+
+	if c.Nobounds {
+		info.nobounds = true
+	}
 }
 
 // Check whether this function can be used in //go:wasmimport or

compiler/syscall.go

@@ -268,6 +268,8 @@ func (b *builder) createSyscall(call *ssa.CallCommon) (llvm.Value, error) {
 		// The signature looks like this:
 		//   func Syscall(trap, nargs, a1, a2, a3 uintptr) (r1, r2 uintptr, err Errno)
 
+		isI386 := strings.HasPrefix(b.Triple, "i386-")
+
 		// Prepare input values.
 		var paramTypes []llvm.Type
 		var params []llvm.Value
@@ -285,11 +287,17 @@ func (b *builder) createSyscall(call *ssa.CallCommon) (llvm.Value, error) {
 		if setLastError.IsNil() {
 			llvmType := llvm.FunctionType(b.ctx.VoidType(), []llvm.Type{b.ctx.Int32Type()}, false)
 			setLastError = llvm.AddFunction(b.mod, "SetLastError", llvmType)
+			if isI386 {
+				setLastError.SetFunctionCallConv(llvm.X86StdcallCallConv)
+			}
 		}
 		getLastError := b.mod.NamedFunction("GetLastError")
 		if getLastError.IsNil() {
 			llvmType := llvm.FunctionType(b.ctx.Int32Type(), nil, false)
 			getLastError = llvm.AddFunction(b.mod, "GetLastError", llvmType)
+			if isI386 {
+				getLastError.SetFunctionCallConv(llvm.X86StdcallCallConv)
+			}
 		}
 
 		// Now do the actual call. Pseudocode:
@@ -300,9 +308,24 @@ func (b *builder) createSyscall(call *ssa.CallCommon) (llvm.Value, error) {
 		// Note that SetLastError/GetLastError could be replaced with direct
 		// access to the thread control block, which is probably smaller and
 		// faster. The Go runtime does this in assembly.
-		b.CreateCall(setLastError.GlobalValueType(), setLastError, []llvm.Value{llvm.ConstNull(b.ctx.Int32Type())}, "")
+		// On windows/386, we also need to save/restore the stack pointer. I'm
+		// not entirely sure why this is needed, but without it these calls
+		// change the stack pointer leading to a crash soon after.
+		setLastErrorCall := b.CreateCall(setLastError.GlobalValueType(), setLastError, []llvm.Value{llvm.ConstNull(b.ctx.Int32Type())}, "")
+		var sp llvm.Value
+		if isI386 {
+			setLastErrorCall.SetInstructionCallConv(llvm.X86StdcallCallConv)
+			sp = b.readStackPointer()
+		}
 		syscallResult := b.CreateCall(llvmType, fnPtr, params, "")
+		if isI386 {
+			syscallResult.SetInstructionCallConv(llvm.X86StdcallCallConv)
+			b.writeStackPointer(sp)
+		}
 		errResult := b.CreateCall(getLastError.GlobalValueType(), getLastError, nil, "err")
+		if isI386 {
+			errResult.SetInstructionCallConv(llvm.X86StdcallCallConv)
+		}
 		if b.uintptrType != b.ctx.Int32Type() {
 			errResult = b.CreateZExt(errResult, b.uintptrType, "err.uintptr")
 		}

diagnostics/diagnostics.go

@@ -10,6 +10,7 @@ import (
 	"go/types"
 	"io"
 	"path/filepath"
+	"reflect"
 	"sort"
 	"strings"
 
@@ -23,6 +24,11 @@ import (
 type Diagnostic struct {
 	Pos token.Position
 	Msg string
+
+	// Start and end position, if available. For many errors these positions are
+	// not available, but for some they are.
+	StartPos token.Position
+	EndPos   token.Position
 }
 
 // One or multiple errors of a particular package.
@@ -114,12 +120,22 @@ func createPackageDiagnostic(err error) PackageDiagnostic {
 func createDiagnostics(err error) []Diagnostic {
 	switch err := err.(type) {
 	case types.Error:
-		return []Diagnostic{
+		diag := Diagnostic{
-			{
+			Pos: err.Fset.Position(err.Pos),
-				Pos: err.Fset.Position(err.Pos),
+			Msg: err.Msg,
-				Msg: err.Msg,
-			},
 		}
+		// There is a special unexported API since Go 1.16 that provides the
+		// range (start and end position) where the type error exists.
+		// There is no promise of backwards compatibility in future Go versions
+		// so we have to be extra careful here to be resilient.
+		v := reflect.ValueOf(err)
+		start := v.FieldByName("go116start")
+		end := v.FieldByName("go116end")
+		if start.IsValid() && end.IsValid() && start.Int() != end.Int() {
+			diag.StartPos = err.Fset.Position(token.Pos(start.Int()))
+			diag.EndPos = err.Fset.Position(token.Pos(end.Int()))
+		}
+		return []Diagnostic{diag}
 	case scanner.Error:
 		return []Diagnostic{
 			{
@@ -188,25 +204,29 @@ func (diag Diagnostic) WriteTo(w io.Writer, wd string) {
 		fmt.Fprintln(w, diag.Msg)
 		return
 	}
-	pos := diag.Pos // make a copy
+	pos := RelativePosition(diag.Pos, wd)
-	if !strings.HasPrefix(pos.Filename, filepath.Join(goenv.Get("GOROOT"), "src")) && !strings.HasPrefix(pos.Filename, filepath.Join(goenv.Get("TINYGOROOT"), "src")) {
-		// This file is not from the standard library (either the GOROOT or the
-		// TINYGOROOT). Make the path relative, for easier reading.  Ignore any
-		// errors in the process (falling back to the absolute path).
-		pos.Filename = tryToMakePathRelative(pos.Filename, wd)
-	}
 	fmt.Fprintf(w, "%s: %s\n", pos, diag.Msg)
 }
 
-// try to make the path relative to the current working directory. If any error
+// Convert the position in pos (assumed to have an absolute path) into a
-// occurs, this error is ignored and the absolute path is returned instead.
+// relative path if possible. Paths inside GOROOT/TINYGOROOT will remain
-func tryToMakePathRelative(dir, wd string) string {
+// absolute.
+func RelativePosition(pos token.Position, wd string) token.Position {
+	// Check whether we even have a working directory.
 	if wd == "" {
-		return dir // working directory not found
+		return pos
 	}
-	relpath, err := filepath.Rel(wd, dir)
+
-	if err != nil {
+	// Paths inside GOROOT should be printed in full.
-		return dir
+	if strings.HasPrefix(pos.Filename, filepath.Join(goenv.Get("GOROOT"), "src")) || strings.HasPrefix(pos.Filename, filepath.Join(goenv.Get("TINYGOROOT"), "src")) {
+		return pos
 	}
-	return relpath
+
+	// Make the path relative, for easier reading. Ignore any errors in the
+	// process (falling back to the absolute path).
+	relpath, err := filepath.Rel(wd, pos.Filename)
+	if err == nil {
+		pos.Filename = relpath
+	}
+	return pos
 }

errors_test.go

@@ -8,6 +8,7 @@ import (
 	"strings"
 	"testing"
 
+	"github.com/tinygo-org/tinygo/builder"
 	"github.com/tinygo-org/tinygo/compileopts"
 	"github.com/tinygo-org/tinygo/diagnostics"
 )
@@ -64,7 +65,11 @@ func testErrorMessages(t *testing.T, filename string, options *compileopts.Optio
 
 	// Try to build a binary (this should fail with an error).
 	tmpdir := t.TempDir()
-	err := Build(filename, tmpdir+"/out", options)
+	config, err := builder.NewConfig(options)
+	if err != nil {
+		t.Fatal("expected to get a compiler error")
+	}
+	err = Build(filename, tmpdir+"/out", config)
 	if err == nil {
 		t.Fatal("expected to get a compiler error")
 	}

flake.lock

@@ -20,16 +20,16 @@
     },
     "nixpkgs": {
       "locked": {
-        "lastModified": 1728500571,
+        "lastModified": 1747953325,
-        "narHash": "sha256-dOymOQ3AfNI4Z337yEwHGohrVQb4yPODCW9MDUyAc4w=",
+        "narHash": "sha256-y2ZtlIlNTuVJUZCqzZAhIw5rrKP4DOSklev6c8PyCkQ=",
         "owner": "NixOS",
         "repo": "nixpkgs",
-        "rev": "d51c28603def282a24fa034bcb007e2bcb5b5dd0",
+        "rev": "55d1f923c480dadce40f5231feb472e81b0bab48",
         "type": "github"
       },
       "original": {
         "id": "nixpkgs",
-        "ref": "nixos-24.05",
+        "ref": "nixos-25.05",
         "type": "indirect"
       }
     },

flake.nix

@@ -21,7 +21,6 @@
 #   make llvm-source            # fetch compiler-rt
 #   git submodule update --init # fetch lots of other libraries and SVD files
 #   make gen-device -j4         # build src/device/*/*.go files
-#   make wasi-libc              # build support for wasi/wasm
 #
 # With this, you should have an environment that can compile anything - except
 # for the Xtensa architecture (ESP8266/ESP32) because support for that lives in
@@ -35,7 +34,7 @@
   inputs = {
     # Use a recent stable release, but fix the version to make it reproducible.
     # This version should be updated from time to time.
-    nixpkgs.url = "nixpkgs/nixos-24.05";
+    nixpkgs.url = "nixpkgs/nixos-25.05";
     flake-utils.url = "github:numtide/flake-utils";
   };
   outputs = { self, nixpkgs, flake-utils }:
@@ -49,11 +48,11 @@
           buildInputs = [
             # These dependencies are required for building tinygo (go install).
             go
-            llvmPackages_18.llvm
+            llvmPackages_20.llvm
-            llvmPackages_18.libclang
+            llvmPackages_20.libclang
             # Additional dependencies needed at runtime, for building and/or
             # flashing.
-            llvmPackages_18.lld
+            llvmPackages_20.lld
             avrdude
             binaryen
             # Additional dependencies needed for on-chip debugging.
@@ -64,20 +63,12 @@
             #openocd
           ];
           shellHook= ''
-            # Configure CLANG, LLVM_AR, and LLVM_NM for `make wasi-libc`.
-            # Without setting these explicitly, Homebrew versions might be used
-            # or the default `ar` and `nm` tools might be used (which don't
-            # support wasi).
-            export CLANG="clang-18 -resource-dir ${llvmPackages_18.clang.cc.lib}/lib/clang/18"
-            export LLVM_AR=llvm-ar
-            export LLVM_NM=llvm-nm
-
             # Make `make smoketest` work (the default is `md5`, while Nix only
             # has `md5sum`).
             export MD5SUM=md5sum
 
             # Ugly hack to make the Clang resources directory available.
-            export GOFLAGS="\"-ldflags=-X github.com/tinygo-org/tinygo/goenv.clangResourceDir=${llvmPackages_18.clang.cc.lib}/lib/clang/18\" -tags=llvm18"
+            export GOFLAGS="\"-ldflags=-X github.com/tinygo-org/tinygo/goenv.clangResourceDir=${llvmPackages_20.clang.cc.lib}/lib/clang/20\" -tags=llvm20"
           '';
         };
       }

go.mod

@@ -1,9 +1,9 @@
 module github.com/tinygo-org/tinygo
 
-go 1.19
+go 1.22.0
 
 require (
-	github.com/aykevl/go-wasm v0.0.2-0.20240825160117-b76c3f9f0982
+	github.com/aykevl/go-wasm v0.0.2-0.20250317121156-42b86c494139
 	github.com/blakesmith/ar v0.0.0-20150311145944-8bd4349a67f2
 	github.com/chromedp/cdproto v0.0.0-20220113222801-0725d94bb6ee
 	github.com/chromedp/chromedp v0.7.6
@@ -16,11 +16,11 @@ require (
 	github.com/sigurn/crc16 v0.0.0-20211026045750-20ab5afb07e3
 	github.com/tetratelabs/wazero v1.6.0
 	go.bug.st/serial v1.6.0
-	golang.org/x/net v0.26.0
+	golang.org/x/net v0.35.0
-	golang.org/x/sys v0.21.0
+	golang.org/x/sys v0.30.0
-	golang.org/x/tools v0.22.1-0.20240621165957-db513b091504
+	golang.org/x/tools v0.30.0
 	gopkg.in/yaml.v2 v2.4.0
-	tinygo.org/x/go-llvm v0.0.0-20250119132755-9dca92dfb4f9
+	tinygo.org/x/go-llvm v0.0.0-20250422114502-b8f170971e74
 )
 
 require (
@@ -33,5 +33,5 @@ require (
 	github.com/mailru/easyjson v0.7.7 // indirect
 	github.com/mattn/go-isatty v0.0.20 // indirect
 	github.com/stretchr/testify v1.8.4 // indirect
-	golang.org/x/text v0.16.0 // indirect
+	golang.org/x/text v0.22.0 // indirect
 )

go.sum

@@ -1,5 +1,5 @@
-github.com/aykevl/go-wasm v0.0.2-0.20240825160117-b76c3f9f0982 h1:cD7QfvrJdYmBw2tFP/VyKPT8ZESlcrwSwo7SvH9Y4dc=
+github.com/aykevl/go-wasm v0.0.2-0.20250317121156-42b86c494139 h1:2O/WuAt8J5id3khcAtVB90czG80m+v0sfkLE07GrCVg=
-github.com/aykevl/go-wasm v0.0.2-0.20240825160117-b76c3f9f0982/go.mod h1:7sXyiaA0WtSogCu67R2252fQpVmJMh9JWJ9ddtGkpWw=
+github.com/aykevl/go-wasm v0.0.2-0.20250317121156-42b86c494139/go.mod h1:7sXyiaA0WtSogCu67R2252fQpVmJMh9JWJ9ddtGkpWw=
 github.com/blakesmith/ar v0.0.0-20150311145944-8bd4349a67f2 h1:oMCHnXa6CCCafdPDbMh/lWRhRByN0VFLvv+g+ayx1SI=
 github.com/blakesmith/ar v0.0.0-20150311145944-8bd4349a67f2/go.mod h1:PkYb9DJNAwrSvRx5DYA+gUcOIgTGVMNkfSCbZM8cWpI=
 github.com/chromedp/cdproto v0.0.0-20211126220118-81fa0469ad77/go.mod h1:At5TxYYdxkbQL0TSefRjhLE3Q0lgvqKKMSFUglJ7i1U=
@@ -12,6 +12,7 @@ github.com/chromedp/sysutil v1.0.0/go.mod h1:kgWmDdq8fTzXYcKIBqIYvRRTnYb9aNS9moA
 github.com/creack/goselect v0.1.2 h1:2DNy14+JPjRBgPzAd1thbQp4BSIihxcBf0IXhQXDRa0=
 github.com/creack/goselect v0.1.2/go.mod h1:a/NhLweNvqIYMuxcMOuWY516Cimucms3DglDzQP3hKY=
 github.com/davecgh/go-spew v1.1.1 h1:vj9j/u1bqnvCEfJOwUhtlOARqs3+rkHYY13jYWTU97c=
+github.com/davecgh/go-spew v1.1.1/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
 github.com/gobwas/httphead v0.1.0 h1:exrUm0f4YX0L7EBwZHuCF4GDp8aJfVeBrlLQrs6NqWU=
 github.com/gobwas/httphead v0.1.0/go.mod h1:O/RXo79gxV8G+RqlR/otEwx4Q36zl9rqC5u12GKvMCM=
 github.com/gobwas/pool v0.2.1 h1:xfeeEhW7pwmX8nuLVlqbzVc7udMDrwetjEv+TZIz1og=
@@ -44,6 +45,7 @@ github.com/mattn/go-tty v0.0.4/go.mod h1:u5GGXBtZU6RQoKV8gY5W6UhMudbR5vXnUe7j3px
 github.com/orisano/pixelmatch v0.0.0-20210112091706-4fa4c7ba91d5 h1:1SoBaSPudixRecmlHXb/GxmaD3fLMtHIDN13QujwQuc=
 github.com/orisano/pixelmatch v0.0.0-20210112091706-4fa4c7ba91d5/go.mod h1:nZgzbfBr3hhjoZnS66nKrHmduYNpc34ny7RK4z5/HM0=
 github.com/pmezard/go-difflib v1.0.0 h1:4DBwDE0NGyQoBHbLQYPwSUPoCMWR5BEzIk/f1lZbAQM=
+github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
 github.com/sigurn/crc16 v0.0.0-20211026045750-20ab5afb07e3 h1:aQKxg3+2p+IFXXg97McgDGT5zcMrQoi0EICZs8Pgchs=
 github.com/sigurn/crc16 v0.0.0-20211026045750-20ab5afb07e3/go.mod h1:9/etS5gpQq9BJsJMWg1wpLbfuSnkm8dPF6FdW2JXVhA=
 github.com/stretchr/testify v1.8.4 h1:CcVxjf3Q8PM0mHUKJCdn+eZZtm5yQwehR5yeSVQQcUk=
@@ -52,10 +54,12 @@ github.com/tetratelabs/wazero v1.6.0 h1:z0H1iikCdP8t+q341xqepY4EWvHEw8Es7tlqiVzl
 github.com/tetratelabs/wazero v1.6.0/go.mod h1:0U0G41+ochRKoPKCJlh0jMg1CHkyfK8kDqiirMmKY8A=
 go.bug.st/serial v1.6.0 h1:mAbRGN4cKE2J5gMwsMHC2KQisdLRQssO9WSM+rbZJ8A=
 go.bug.st/serial v1.6.0/go.mod h1:UABfsluHAiaNI+La2iESysd9Vetq7VRdpxvjx7CmmOE=
-golang.org/x/mod v0.18.0 h1:5+9lSbEzPSdWkH32vYPBwEpX8KwDbM52Ud9xBUvNlb0=
+golang.org/x/mod v0.23.0 h1:Zb7khfcRGKk+kqfxFaP5tZqCnDZMjC5VtUBs87Hr6QM=
-golang.org/x/net v0.26.0 h1:soB7SVo0PWrY4vPW/+ay0jKDNScG2X9wFeYlXIvJsOQ=
+golang.org/x/mod v0.23.0/go.mod h1:6SkKJ3Xj0I0BrPOZoBy3bdMptDDU9oJrpohJ3eWZ1fY=
-golang.org/x/net v0.26.0/go.mod h1:5YKkiSynbBIh3p6iOc/vibscux0x38BZDkn8sCUPxHE=
+golang.org/x/net v0.35.0 h1:T5GQRQb2y08kTAByq9L4/bz8cipCdA8FbRTXewonqY8=
-golang.org/x/sync v0.7.0 h1:YsImfSBoP9QPYL0xyKJPq0gcaJdG3rInoqxTWbfQu9M=
+golang.org/x/net v0.35.0/go.mod h1:EglIi67kWsHKlRzzVMUD93VMSWGFOMSZgxFjparz1Qk=
+golang.org/x/sync v0.11.0 h1:GGz8+XQP4FvTTrjZPzNKTMFtSXH80RAzG+5ghFPgK9w=
+golang.org/x/sync v0.11.0/go.mod h1:Czt+wKu1gCyEFDUtn0jG5QVvpJ6rzVqr5aXyt9drQfk=
 golang.org/x/sys v0.0.0-20190222072716-a9d3bda3a223/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
 golang.org/x/sys v0.0.0-20191008105621-543471e840be/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20191120155948-bd437916bb0e/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
@@ -63,16 +67,17 @@ golang.org/x/sys v0.0.0-20201207223542-d4d67f95c62d/go.mod h1:h1NjWce9XRLGQEsW7w
 golang.org/x/sys v0.0.0-20211124211545-fe61309f8881/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.0.0-20220811171246-fbc7d0a398ab/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.6.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
-golang.org/x/sys v0.21.0 h1:rF+pYz3DAGSQAxAu1CbC7catZg4ebC4UIeIhKxBZvws=
+golang.org/x/sys v0.30.0 h1:QjkSwP/36a20jFYWkSue1YwXzLmsV5Gfq7Eiy72C1uc=
-golang.org/x/sys v0.21.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
+golang.org/x/sys v0.30.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
-golang.org/x/text v0.16.0 h1:a94ExnEXNtEwYLGJSIUxnWoxoRz/ZcCsV63ROupILh4=
+golang.org/x/text v0.22.0 h1:bofq7m3/HAFvbF51jz3Q9wLg3jkvSPuiZu/pD1XwgtM=
-golang.org/x/text v0.16.0/go.mod h1:GhwF1Be+LQoKShO3cGOHzqOgRrGaYc9AvblQOmPVHnI=
+golang.org/x/text v0.22.0/go.mod h1:YRoo4H8PVmsu+E3Ou7cqLVH8oXWIHVoX0jqUWALQhfY=
-golang.org/x/tools v0.22.1-0.20240621165957-db513b091504 h1:MMsD8mMfluf/578+3wrTn22pjI/Xkzm+gPW47SYfspY=
+golang.org/x/tools v0.30.0 h1:BgcpHewrV5AUp2G9MebG4XPFI1E2W41zU1SaqVA9vJY=
-golang.org/x/tools v0.22.1-0.20240621165957-db513b091504/go.mod h1:aCwcsjqvq7Yqt6TNyX7QMU2enbQ/Gt0bo6krSeEri+c=
+golang.org/x/tools v0.30.0/go.mod h1:c347cR/OJfw5TI+GfX7RUPNMdDRRbjvYTS0jPyvsVtY=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405 h1:yhCVgyC4o1eVCa2tZl7eS0r+SDo693bJlVdllGtEeKM=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
 gopkg.in/yaml.v2 v2.4.0 h1:D8xgwECY7CYvx+Y2n4sBz93Jn9JRvxdiyyo8CTfuKaY=
 gopkg.in/yaml.v2 v2.4.0/go.mod h1:RDklbk79AGWmwhnvt/jBztapEOGDOx6ZbXqjP6csGnQ=
 gopkg.in/yaml.v3 v3.0.1 h1:fxVm/GzAzEWqLHuvctI91KS9hhNmmWOoWu0XTYJS7CA=
-tinygo.org/x/go-llvm v0.0.0-20250119132755-9dca92dfb4f9 h1:rMvEzuCYjyiR+pmdiCVWTQw3L6VqiSIXoL19I3lYufE=
+gopkg.in/yaml.v3 v3.0.1/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=
-tinygo.org/x/go-llvm v0.0.0-20250119132755-9dca92dfb4f9/go.mod h1:GFbusT2VTA4I+l4j80b17KFK+6whv69Wtny5U+T8RR0=
+tinygo.org/x/go-llvm v0.0.0-20250422114502-b8f170971e74 h1:ovavgTdIBWCH8YWlcfq9gkpoyT1+IxMKSn+Df27QwE8=
+tinygo.org/x/go-llvm v0.0.0-20250422114502-b8f170971e74/go.mod h1:GFbusT2VTA4I+l4j80b17KFK+6whv69Wtny5U+T8RR0=

goenv/version.go

@@ -10,7 +10,7 @@ import (
 
 // Version of TinyGo.
 // Update this value before release of new version of software.
-const version = "0.38.0-dev"
+const version = "0.38.0"
 
 // Return TinyGo version, either in the form 0.30.0 or as a development version
 // (like 0.30.0-dev-abcd012).

interp/interpreter.go

@@ -364,9 +364,22 @@ func (r *runner) run(fn *function, params []value, parentMem *memoryView, indent
 						continue
 					}
 					nBytes := uint32(n * elemSize)
+					srcObj := mem.get(src.index())
 					dstObj := mem.getWritable(dst.index())
+					if srcObj.buffer == nil || dstObj.buffer == nil {
+						// If the buffer is nil, it means the slice is external.
+						// This can happen for example when copying data out of
+						// a //go:embed slice, which is not available at interp
+						// time.
+						// See: https://github.com/tinygo-org/tinygo/issues/4895
+						err := r.runAtRuntime(fn, inst, locals, &mem, indent)
+						if err != nil {
+							return nil, mem, err
+						}
+						continue
+					}
 					dstBuf := dstObj.buffer.asRawValue(r)
-					srcBuf := mem.get(src.index()).buffer.asRawValue(r)
+					srcBuf := srcObj.buffer.asRawValue(r)
 					copy(dstBuf.buf[dst.offset():dst.offset()+nBytes], srcBuf.buf[src.offset():])
 					dstObj.buffer = dstBuf
 					mem.put(dst.index(), dstObj)

interp/testdata/slice-copy.ll

@@ -11,6 +11,10 @@ target triple = "x86_64--linux"
 @main.sliceDstUntaint.buf = internal global [2 x i8] zeroinitializer
 @main.sliceSrcTaint.buf = internal global [2 x i8] c"cd"
 @main.sliceDstTaint.buf = internal global [2 x i8] zeroinitializer
+@main.sliceSrcExternal1.buf = external global [2 x i8]
+@main.sliceDstExternal1.buf = internal global [2 x i8] zeroinitializer
+@main.sliceSrcExternal2.buf = internal global [2 x i8] zeroinitializer
+@main.sliceDstExternal2.buf = external global [2 x i8]
 
 declare i64 @runtime.sliceCopy(ptr %dst, ptr %src, i64 %dstLen, i64 %srcLen, i64 %elemSize) unnamed_addr
 
@@ -58,6 +62,14 @@ entry:
   %sliceDstTaint.val = load i8, ptr getelementptr inbounds (i8, ptr @main.sliceDstTaint.buf, i32 0)
   call void @runtime.printuint8(i8 %sliceDstTaint.val)
 
+  ; print(sliceDstExternal1[0])
+  %sliceDstExternal1.val = load i8, ptr getelementptr inbounds (i8, ptr @main.sliceDstExternal1.buf, i32 0)
+  call void @runtime.printuint8(i8 %sliceDstExternal1.val)
+
+  ; print(sliceDstExternal2[0])
+  %sliceDstExternal2.val = load i8, ptr getelementptr inbounds (i8, ptr @main.sliceDstExternal2.buf, i32 0)
+  call void @runtime.printuint8(i8 %sliceDstExternal2.val)
+
   ret void
 }
 
@@ -102,5 +114,11 @@ entry:
   call void @use(ptr @main.sliceSrcTaint.buf)
   %copy.n4 = call i64 @runtime.sliceCopy(ptr @main.sliceDstTaint.buf, ptr @main.sliceSrcTaint.buf, i64 2, i64 2, i64 1)
 
+  ; Test that copying from or into external buffers works correctly.
+  ; These copy operations must be done at runtime.
+  ; https://github.com/tinygo-org/tinygo/issues/4895
+  %copy.n5 = call i64 @runtime.sliceCopy(ptr @main.sliceDstExternal1.buf, ptr @main.sliceSrcExternal1.buf, i64 2, i64 2, i64 1)
+  %copy.n6 = call i64 @runtime.sliceCopy(ptr @main.sliceDstExternal2.buf, ptr @main.sliceSrcExternal2.buf, i64 2, i64 2, i64 1)
+
   ret void
 }

interp/testdata/slice-copy.out.ll

@@ -3,6 +3,10 @@ target triple = "x86_64--linux"
 
 @main.sliceSrcTaint.buf = internal global [2 x i8] c"cd"
 @main.sliceDstTaint.buf = internal global [2 x i8] zeroinitializer
+@main.sliceSrcExternal1.buf = external global [2 x i8]
+@main.sliceDstExternal1.buf = internal global [2 x i8] zeroinitializer
+@main.sliceSrcExternal2.buf = internal global [2 x i8] zeroinitializer
+@main.sliceDstExternal2.buf = external global [2 x i8]
 
 declare i64 @runtime.sliceCopy(ptr, ptr, i64, i64, i64) unnamed_addr
 
@@ -16,6 +20,8 @@ define void @runtime.initAll() unnamed_addr {
 entry:
   call void @use(ptr @main.sliceSrcTaint.buf)
   %copy.n4 = call i64 @runtime.sliceCopy(ptr @main.sliceDstTaint.buf, ptr @main.sliceSrcTaint.buf, i64 2, i64 2, i64 1)
+  %copy.n5 = call i64 @runtime.sliceCopy(ptr @main.sliceDstExternal1.buf, ptr @main.sliceSrcExternal1.buf, i64 2, i64 2, i64 1)
+  %copy.n6 = call i64 @runtime.sliceCopy(ptr @main.sliceDstExternal2.buf, ptr @main.sliceSrcExternal2.buf, i64 2, i64 2, i64 1)
   ret void
 }
 
@@ -28,5 +34,9 @@ entry:
   call void @runtime.printuint8(i8 97)
   %sliceDstTaint.val = load i8, ptr @main.sliceDstTaint.buf, align 1
   call void @runtime.printuint8(i8 %sliceDstTaint.val)
+  %sliceDstExternal1.val = load i8, ptr @main.sliceDstExternal1.buf, align 1
+  call void @runtime.printuint8(i8 %sliceDstExternal1.val)
+  %sliceDstExternal2.val = load i8, ptr @main.sliceDstExternal2.buf, align 1
+  call void @runtime.printuint8(i8 %sliceDstExternal2.val)
   ret void
 }

loader/ssa.go

@@ -8,7 +8,9 @@ import (
 //
 // The program must already be parsed and type-checked with the .Parse() method.
 func (p *Program) LoadSSA() *ssa.Program {
-	prog := ssa.NewProgram(p.fset, ssa.SanityCheckFunctions|ssa.BareInits|ssa.GlobalDebug|ssa.InstantiateGenerics)
+	// TODO: re-enable SanityCheckFunctions when x/tools is upgraded to
+	// a version with a fix for https://golang.org/issues/73594.
+	prog := ssa.NewProgram(p.fset /*ssa.SanityCheckFunctions|*/, ssa.BareInits|ssa.GlobalDebug|ssa.InstantiateGenerics)
 
 	for _, pkg := range p.sorted {
 		prog.CreatePackage(pkg.Pkg, pkg.Files, &pkg.info, true)

main.go

@@ -138,27 +138,13 @@ func printCommand(cmd string, args ...string) {
 }
 
 // Build compiles and links the given package and writes it to outpath.
-func Build(pkgName, outpath string, options *compileopts.Options) error {
+func Build(pkgName, outpath string, config *compileopts.Config) error {
-	config, err := builder.NewConfig(options)
-	if err != nil {
-		return err
-	}
-
-	if options.PrintJSON {
-		b, err := json.MarshalIndent(config, "", "  ")
-		if err != nil {
-			handleCompilerError(err)
-		}
-		fmt.Printf("%s\n", string(b))
-		return nil
-	}
-
 	// Create a temporary directory for intermediary files.
 	tmpdir, err := os.MkdirTemp("", "tinygo")
 	if err != nil {
 		return err
 	}
-	if !options.Work {
+	if !config.Options.Work {
 		defer os.RemoveAll(tmpdir)
 	}
 
@@ -1027,15 +1013,50 @@ func findFATMounts(options *compileopts.Options) ([]mountPoint, error) {
 			return nil, fmt.Errorf("could not list mount points: %w", err)
 		}
 		for _, elem := range list {
-			// TODO: find a way to check for the filesystem type.
+			volumePath := filepath.Join("/Volumes", elem.Name())
-			// (Only return FAT filesystems).
+			if _, err := os.Stat(volumePath); err != nil {
-			points = append(points, mountPoint{
+				continue
-				name: elem.Name(),
+			}
-				path: filepath.Join("/Volumes", elem.Name()),
+
-			})
+			cmd := exec.Command("diskutil", "info", volumePath)
+			var out bytes.Buffer
+			cmd.Stdout = &out
+			if err := cmd.Run(); err != nil {
+				continue // skip if diskutil failed
+			}
+
+			diskInfo := map[string]string{}
+			scanner := bufio.NewScanner(&out)
+			for scanner.Scan() {
+				line := scanner.Text()
+				parts := strings.SplitN(line, ":", 2)
+				if len(parts) != 2 {
+					continue
+				}
+				key := strings.TrimSpace(parts[0])
+				value := strings.TrimSpace(parts[1])
+				diskInfo[key] = value
+			}
+			if err := scanner.Err(); err != nil {
+				continue
+			}
+
+			volName, okv := diskInfo["Volume Name"]
+			fsType, okf := diskInfo["File System Personality"]
+			if !okv || !okf {
+				continue
+			}
+
+			// Check if a filesystem type is FAT-based
+			if strings.Contains(strings.ToUpper(fsType), "FAT") {
+				points = append(points, mountPoint{
+					name: volName,
+					path: volumePath,
+				})
+			}
 		}
 		sort.Slice(points, func(i, j int) bool {
-			return points[i].path < points[j].name
+			return points[i].path < points[j].path
 		})
 		return points, nil
 	case "linux":
@@ -1396,6 +1417,68 @@ func usage(command string) {
 
 }
 
+// Print diagnostics very similar to the -json flag in Go.
+func printBuildOutput(err error, jsonDiagnostics bool) {
+	if err == nil {
+		return // nothing to report
+	}
+
+	if jsonDiagnostics {
+		workingDir, getwdErr := os.Getwd()
+		if getwdErr != nil {
+			workingDir = ""
+		}
+
+		type jsonDiagnosticOutput struct {
+			ImportPath string
+			Action     string
+			Output     string `json:",omitempty"`
+			StartPos   string `json:",omitempty"` // non-standard
+			EndPos     string `json:",omitempty"` // non-standard
+		}
+
+		for _, diags := range diagnostics.CreateDiagnostics(err) {
+			if diags.ImportPath != "" {
+				output, _ := json.Marshal(jsonDiagnosticOutput{
+					ImportPath: diags.ImportPath,
+					Action:     "build-output",
+					Output:     "# " + diags.ImportPath + "\n",
+				})
+				os.Stdout.Write(append(output, '\n'))
+			}
+			for _, diag := range diags.Diagnostics {
+				w := &bytes.Buffer{}
+				diag.WriteTo(w, workingDir)
+				data := jsonDiagnosticOutput{
+					ImportPath: diags.ImportPath,
+					Action:     "build-output",
+					Output:     w.String(),
+				}
+				if diag.StartPos.IsValid() && diag.EndPos.IsValid() {
+					// Include the non-standard StartPos/EndPos values. These
+					// are useful for the TinyGo Playground to show better error
+					// messages.
+					data.StartPos = diagnostics.RelativePosition(diag.StartPos, workingDir).String()
+					data.EndPos = diagnostics.RelativePosition(diag.EndPos, workingDir).String()
+				}
+				output, _ := json.Marshal(data)
+				os.Stdout.Write(append(output, '\n'))
+			}
+
+			// Emit the "Action":"build-fail" JSON.
+			output, _ := json.Marshal(jsonDiagnosticOutput{
+				ImportPath: diags.ImportPath,
+				Action:     "build-fail",
+			})
+			os.Stdout.Write(append(output, '\n'))
+		}
+		os.Exit(1)
+	}
+
+	// Regular diagnostic handling.
+	handleCompilerError(err)
+}
+
 func handleCompilerError(err error) {
 	if err != nil {
 		wd, getwdErr := os.Getwd()
@@ -1492,9 +1575,9 @@ func main() {
 	command := os.Args[1]
 
 	opt := flag.String("opt", "z", "optimization level: 0, 1, 2, s, z")
-	gc := flag.String("gc", "", "garbage collector to use (none, leaking, conservative)")
+	gc := flag.String("gc", "", "garbage collector to use (none, leaking, conservative, custom, precise, boehm)")
 	panicStrategy := flag.String("panic", "print", "panic strategy (print, trap)")
-	scheduler := flag.String("scheduler", "", "which scheduler to use (none, tasks, asyncify)")
+	scheduler := flag.String("scheduler", "", "which scheduler to use (none, tasks, cores, threads, asyncify)")
 	serial := flag.String("serial", "", "which serial output to use (none, uart, usb, rtt)")
 	work := flag.Bool("work", false, "print the name of the temporary build directory and do not delete this directory on exit")
 	interpTimeout := flag.Duration("interp-timeout", 180*time.Second, "interp optimization pass timeout")
@@ -1512,8 +1595,10 @@ func main() {
 	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")
+	flagJSON := flag.Bool("json", false, "print output in JSON format")
 	parallelism := flag.Int("p", runtime.GOMAXPROCS(0), "the number of build jobs that can run in parallel")
 	nodebug := flag.Bool("no-debug", false, "strip debug information")
+	nobounds := flag.Bool("nobounds", false, "do not emit bounds checks")
 	ocdCommandsString := flag.String("ocd-commands", "", "OpenOCD commands, overriding target spec (can specify multiple separated by commas)")
 	ocdOutput := flag.Bool("ocd-output", false, "print OCD daemon output during debug")
 	port := flag.String("port", "", "flash port (can specify multiple candidates separated by commas)")
@@ -1536,10 +1621,7 @@ func main() {
 	// development it can be useful to not emit debug information at all.
 	skipDwarf := flag.Bool("internal-nodwarf", false, "internal flag, use -no-debug instead")
 
-	var flagJSON, flagDeps, flagTest bool
+	var flagDeps, flagTest bool
-	if command == "help" || command == "list" || command == "info" || command == "build" {
-		flag.BoolVar(&flagJSON, "json", false, "print data in JSON format")
-	}
 	if command == "help" || command == "list" {
 		flag.BoolVar(&flagDeps, "deps", false, "supply -deps flag to go list")
 		flag.BoolVar(&flagTest, "test", false, "supply -test flag to go list")
@@ -1625,6 +1707,7 @@ func main() {
 		SkipDWARF:       *skipDwarf,
 		Semaphore:       make(chan struct{}, *parallelism),
 		Debug:           !*nodebug,
+		Nobounds:        *nobounds,
 		PrintSizes:      *printSize,
 		PrintStacks:     *printStacks,
 		PrintAllocs:     printAllocs,
@@ -1634,7 +1717,6 @@ func main() {
 		Programmer:      *programmer,
 		OpenOCDCommands: ocdCommands,
 		LLVMFeatures:    *llvmFeatures,
-		PrintJSON:       flagJSON,
 		Monitor:         *monitor,
 		BaudRate:        *baudrate,
 		Timeout:         *timeout,
@@ -1689,13 +1771,15 @@ func main() {
 			os.Exit(1)
 		}
 
-		err := Build(pkgName, outpath, options)
+		config, err := builder.NewConfig(options)
 		handleCompilerError(err)
+		err = Build(pkgName, outpath, config)
+		printBuildOutput(err, *flagJSON)
 	case "flash", "gdb", "lldb":
 		pkgName := filepath.ToSlash(flag.Arg(0))
 		if command == "flash" {
 			err := Flash(pkgName, *port, options)
-			handleCompilerError(err)
+			printBuildOutput(err, *flagJSON)
 		} else {
 			if !options.Debug {
 				fmt.Fprintln(os.Stderr, "Debug disabled while running debugger?")
@@ -1703,7 +1787,7 @@ func main() {
 				os.Exit(1)
 			}
 			err := Debug(command, pkgName, *ocdOutput, options)
-			handleCompilerError(err)
+			printBuildOutput(err, *flagJSON)
 		}
 	case "run":
 		if flag.NArg() < 1 {
@@ -1713,7 +1797,7 @@ func main() {
 		}
 		pkgName := filepath.ToSlash(flag.Arg(0))
 		err := Run(pkgName, options, flag.Args()[1:])
-		handleCompilerError(err)
+		printBuildOutput(err, *flagJSON)
 	case "test":
 		var pkgNames []string
 		for i := 0; i < flag.NArg(); i++ {
@@ -1847,7 +1931,7 @@ func main() {
 			fmt.Fprintln(os.Stderr, err)
 			os.Exit(1)
 		}
-		if flagJSON {
+		if *flagJSON {
 			json, _ := json.MarshalIndent(struct {
 				Target     *compileopts.TargetSpec `json:"target"`
 				GOROOT     string                  `json:"goroot"`
@@ -1889,7 +1973,7 @@ func main() {
 			os.Exit(1)
 		}
 		var extraArgs []string
-		if flagJSON {
+		if *flagJSON {
 			extraArgs = append(extraArgs, "-json")
 		}
 		if flagDeps {

main_test.go

@@ -35,6 +35,8 @@ const TESTDATA = "testdata"
 
 var testTarget = flag.String("target", "", "override test target")
 
+var testOnlyCurrentOS = flag.Bool("only-current-os", false, "")
+
 var supportedLinuxArches = map[string]string{
 	"AMD64Linux": "linux/amd64",
 	"X86Linux":   "linux/386",
@@ -158,20 +160,51 @@ func TestBuild(t *testing.T) {
 		return
 	}
 
-	t.Run("EmulatedCortexM3", func(t *testing.T) {
+	if !*testOnlyCurrentOS {
-		t.Parallel()
+		t.Run("EmulatedCortexM3", func(t *testing.T) {
-		runPlatTests(optionsFromTarget("cortex-m-qemu", sema), tests, t)
+			t.Parallel()
-	})
+			runPlatTests(optionsFromTarget("cortex-m-qemu", sema), tests, t)
+		})
 
-	t.Run("EmulatedRISCV", func(t *testing.T) {
+		t.Run("EmulatedRISCV", func(t *testing.T) {
-		t.Parallel()
+			t.Parallel()
-		runPlatTests(optionsFromTarget("riscv-qemu", sema), tests, t)
+			runPlatTests(optionsFromTarget("riscv-qemu", sema), tests, t)
-	})
+		})
 
-	t.Run("AVR", func(t *testing.T) {
+		t.Run("AVR", func(t *testing.T) {
-		t.Parallel()
+			t.Parallel()
-		runPlatTests(optionsFromTarget("simavr", sema), tests, t)
+			runPlatTests(optionsFromTarget("simavr", sema), tests, t)
-	})
+		})
+
+		t.Run("WebAssembly", func(t *testing.T) {
+			t.Parallel()
+
+			runPlatTests(optionsFromTarget("wasm", sema), tests, t)
+			// Test with -gc=boehm.
+			t.Run("gc.go-boehm", func(t *testing.T) {
+				t.Parallel()
+				optionsBoehm := optionsFromTarget("wasm", sema)
+				optionsBoehm.GC = "boehm"
+				runTest("gc.go", optionsBoehm, t, nil, nil)
+			})
+		})
+		t.Run("WASIp1", func(t *testing.T) {
+			t.Parallel()
+			runPlatTests(optionsFromTarget("wasip1", sema), tests, t)
+
+			// Test with -gc=boehm.
+			t.Run("gc.go-boehm", func(t *testing.T) {
+				t.Parallel()
+				optionsBoehm := optionsFromTarget("wasip1", sema)
+				optionsBoehm.GC = "boehm"
+				runTest("gc.go", optionsBoehm, t, nil, nil)
+			})
+		})
+		t.Run("WASIp2", func(t *testing.T) {
+			t.Parallel()
+			runPlatTests(optionsFromTarget("wasip2", sema), tests, t)
+		})
+	}
 
 	if runtime.GOOS == "linux" {
 		for name, osArch := range supportedLinuxArches {
@@ -191,18 +224,13 @@ func TestBuild(t *testing.T) {
 			options := optionsFromOSARCH("linux/mipsle/softfloat", sema)
 			runTest("cgo/", options, t, nil, nil)
 		})
-		t.Run("WebAssembly", func(t *testing.T) {
+	} else if runtime.GOOS == "windows" {
-			t.Parallel()
+		if runtime.GOARCH != "386" {
-			runPlatTests(optionsFromTarget("wasm", sema), tests, t)
+			t.Run("Windows386", func(t *testing.T) {
-		})
+				t.Parallel()
-		t.Run("WASI", func(t *testing.T) {
+				runPlatTests(optionsFromOSARCH("windows/386", sema), tests, t)
-			t.Parallel()
+			})
-			runPlatTests(optionsFromTarget("wasip1", sema), tests, t)
+		}
-		})
-		t.Run("WASIp2", func(t *testing.T) {
-			t.Parallel()
-			runPlatTests(optionsFromTarget("wasip2", sema), tests, t)
-		})
 	}
 }
 
@@ -675,6 +703,10 @@ func TestWasmExport(t *testing.T) {
 				// again.
 				checkResult("reentrantCall(2, 3)", mustCall(mod.ExportedFunction("reentrantCall").Call(ctx, 2, 3)), []uint64{5})
 				checkResult("reentrantCall(1, 8)", mustCall(mod.ExportedFunction("reentrantCall").Call(ctx, 1, 8)), []uint64{9})
+
+				// Check that goroutines started inside //go:wasmexport don't
+				// block the called function from returning.
+				checkResult("goroutineExit()", mustCall(mod.ExportedFunction("goroutineExit").Call(ctx)), nil)
 			}
 
 			// Add wasip1 module.

src/crypto/rand/rand_windows.go

@@ -1,6 +1,9 @@
 package rand
 
-import "errors"
+import (
+	"errors"
+	"unsafe"
+)
 
 func init() {
 	Reader = &reader{}
@@ -16,28 +19,22 @@ func (r *reader) Read(b []byte) (n int, err error) {
 		return
 	}
 
-	var randomByte uint32
+	// Use the old RtlGenRandom, introduced in Windows XP.
-	for i := range b {
+	// Even though the documentation says it is deprecated, it is widely used
-		// Call rand_s every four bytes because it's a C int (always 32-bit in
+	// and probably won't go away anytime soon.
-		// Windows).
+	// See for example: https://github.com/golang/go/issues/33542
-		if i%4 == 0 {
+	// For Windows 7 and newer, we might switch to ProcessPrng in the future
-			errCode := libc_rand_s(&randomByte)
+	// (which is a documented function and might be a tiny bit faster).
-			if errCode != 0 {
+	ok := libc_RtlGenRandom(unsafe.Pointer(&b[0]), len(b))
-				// According to the documentation, it can return an error.
+	if !ok {
-				return n, errRandom
+		return 0, errRandom
-			}
-		} else {
-			randomByte >>= 8
-		}
-		b[i] = byte(randomByte)
 	}
-
 	return len(b), nil
 }
 
-// Cryptographically secure random number generator.
+// This function is part of advapi32.dll, and is called SystemFunction036 for
-// https://docs.microsoft.com/en-us/cpp/c-runtime-library/reference/rand-s?view=msvc-170
+// some reason. It's available on Windows XP and newer.
-// errno_t rand_s(unsigned int* randomValue);
+// See: https://learn.microsoft.com/en-us/windows/win32/api/ntsecapi/nf-ntsecapi-rtlgenrandom
 //
-//export rand_s
+//export SystemFunction036
-func libc_rand_s(randomValue *uint32) int32
+func libc_RtlGenRandom(buf unsafe.Pointer, len int) bool

src/device/arm/arm.go

@@ -146,6 +146,11 @@ const (
 	SYST_CALIB_NOREF     = 0x80000000 // Bit NOREF.
 )
 
+// ClearPendingIRQ clears the pending status of the interrupt.
+func ClearPendingIRQ(irq uint32) {
+	NVIC.ICPR[irq>>5].Set(1 << (irq & 0x1F))
+}
+
 // Enable the given interrupt number.
 func EnableIRQ(irq uint32) {
 	NVIC.ISER[irq>>5].Set(1 << (irq & 0x1F))

src/device/riscv/csr.go

@@ -275,3 +275,61 @@ const (
 	DPC      CSR = 0x7B1 // Debug PC.
 	DSCRATCH CSR = 0x7B2 // Debug scratch register.
 )
+
+// Bitfields for the CSR registers above.
+const (
+	// MSTATUS (common bits between RV32 and RV64)
+	MSTATUS_SIE  = 1 << 1
+	MSTATUS_MIE  = 1 << 3
+	MSTATUS_SPIE = 1 << 5
+	MSTATUS_UBE  = 1 << 6
+	MSTATUS_MPIE = 1 << 7
+	MSTATUS_SPP  = 1 << 8
+	MSTATUS_MPRV = 1 << 17
+	MSTATUS_SUM  = 1 << 18
+	MSTATUS_MXR  = 1 << 19
+	MSTATUS_TVM  = 1 << 20
+	MSTATUS_TW   = 1 << 21
+	MSTATUS_TSR  = 1 << 22
+
+	MIE_SSIE = 1 << 1
+	MIE_MSIE = 1 << 3
+	MIE_STIE = 1 << 5
+	MIE_MTIE = 1 << 7
+	MIE_SEIE = 1 << 9
+	MIE_MEIE = 1 << 11
+
+	MIP_SSIP = 1 << 1
+	MIP_MSIP = 1 << 3
+	MIP_STIP = 1 << 5
+	MIP_MTIP = 1 << 7
+	MIP_SEIP = 1 << 9
+	MIP_MEIP = 1 << 11
+)
+
+// Interrupt constants
+const (
+	// MCAUSE values with the topmost bit (interrupt bit) set.
+	SupervisorSoftwareInterrupt = 1
+	MachineSoftwareInterrupt    = 3
+	SupervisorTimerInterrupt    = 5
+	MachineTimerInterrupt       = 7
+	SupervisorExternalInterrupt = 9
+	MachineExternalInterrupt    = 11
+
+	// MCAUSE values with the topmost bit (interrupt bit) clear.
+	InstructionAddressMisaligned = 0
+	InstructionAccessFault       = 1
+	IllegalInstruction           = 2
+	Breakpoint                   = 3
+	LoadAddressMisaligned        = 4
+	LoadAccessFault              = 5
+	StoreOrAMOAddressMisaligned  = 6
+	StoreOrAMOAccessFault        = 7
+	EnvironmentCallFromUMode     = 8
+	EnvironmentCallFromSMode     = 9
+	EnvironmentCallFromMMode     = 11
+	InstructionPageFault         = 12
+	LoadPageFault                = 13
+	StoreOrAMOPageFault          = 15
+)

src/device/riscv/start.S

@@ -3,8 +3,47 @@
 .type _start,@function
 
 _start:
+    // If we're on a multicore system, we need to wait for hart 0 to wake us up.
+#if TINYGO_CORES > 1
+    csrr a0, mhartid
+
+    // Hart 0 stack
+    bnez a0, 1f
+    la sp,      _stack_top
+
+1:
+    // Hart 1 stack
+    li a1, 1
+    bne a0, a1, 2f
+    la sp,      _stack1_top
+
+2:
+    // Hart 2 stack
+    #if TINYGO_CORES >= 3
+    li a1, 2
+    bne a0, a1, 3f
+    la sp,      _stack2_top
+    #endif
+
+3:
+    // Hart 3 stack
+    #if TINYGO_CORES >= 4
+    li a1, 3
+    bne a0, a1, 4f
+    la sp,      _stack3_top
+    #endif
+
+4:
+    // done
+
+#if TINYGO_CORES > 4
+#error only up to 4 cores are supported at the moment!
+#endif
+
+#else
     // Load the stack pointer.
     la sp,      _stack_top
+#endif
 
     // Load the globals pointer. The program will load pointers relative to this
     // register, so it must be set to the right value on startup.

src/device/rp/rp2040-extra.go

@@ -0,0 +1,51 @@
+// Hand created file. DO NOT DELETE.
+// Definitions that are missing in src/device/rp/rp2040.go generated from SVDs
+
+//go:build rp && rp2040
+
+package rp
+
+// DMA: 2.5.3.1. System DREQ Table from RP2040 Datasheet
+// https://datasheets.raspberrypi.com/rp2040/rp2040-datasheet.pdf
+const (
+	DREQ_PIO0_TX0   = 0
+	DREQ_PIO0_TX1   = 1
+	DREQ_PIO0_TX2   = 2
+	DREQ_PIO0_TX3   = 3
+	DREQ_PIO0_RX0   = 4
+	DREQ_PIO0_RX1   = 5
+	DREQ_PIO0_RX2   = 6
+	DREQ_PIO0_RX3   = 7
+	DREQ_PIO1_TX0   = 8
+	DREQ_PIO1_TX1   = 9
+	DREQ_PIO1_TX2   = 10
+	DREQ_PIO1_TX3   = 11
+	DREQ_PIO1_RX0   = 12
+	DREQ_PIO1_RX1   = 13
+	DREQ_PIO1_RX2   = 14
+	DREQ_PIO1_RX3   = 15
+	DREQ_SPI0_TX    = 16
+	DREQ_SPI0_RX    = 17
+	DREQ_SPI1_TX    = 18
+	DREQ_SPI1_RX    = 19
+	DREQ_UART0_TX   = 20
+	DREQ_UART0_RX   = 21
+	DREQ_UART1_TX   = 22
+	DREQ_UART1_RX   = 23
+	DREQ_PWM_WRAP0  = 24
+	DREQ_PWM_WRAP1  = 25
+	DREQ_PWM_WRAP2  = 26
+	DREQ_PWM_WRAP3  = 27
+	DREQ_PWM_WRAP4  = 28
+	DREQ_PWM_WRAP5  = 29
+	DREQ_PWM_WRAP6  = 30
+	DREQ_PWM_WRAP7  = 31
+	DREQ_I2C0_TX    = 32
+	DREQ_I2C0_RX    = 33
+	DREQ_I2C1_TX    = 34
+	DREQ_I2C1_RX    = 35
+	DREQ_ADC        = 36
+	DREQ_XIP_STREAM = 37
+	DREQ_XIP_SSITX  = 38
+	DREQ_XIP_SSIRX  = 39
+)

src/device/rp/rp2350-extra.go

@@ -0,0 +1,66 @@
+// Hand created file. DO NOT DELETE.
+// Definitions that are missing in src/device/rp/rp2350.go generated from SVDs
+
+//go:build rp && rp2350
+
+package rp
+
+// DMA: 12.6.4.1. System DREQ Table from RP2350 Datasheet
+// https://datasheets.raspberrypi.com/rp2350/rp2350-datasheet.pdf
+const (
+	DREQ_PIO0_TX0   = 0
+	DREQ_PIO0_TX1   = 1
+	DREQ_PIO0_TX2   = 2
+	DREQ_PIO0_TX3   = 3
+	DREQ_PIO0_RX0   = 4
+	DREQ_PIO0_RX1   = 5
+	DREQ_PIO0_RX2   = 6
+	DREQ_PIO0_RX3   = 7
+	DREQ_PIO1_TX0   = 8
+	DREQ_PIO1_TX1   = 9
+	DREQ_PIO1_TX2   = 10
+	DREQ_PIO1_TX3   = 11
+	DREQ_PIO1_RX0   = 12
+	DREQ_PIO1_RX1   = 13
+	DREQ_PIO1_RX2   = 14
+	DREQ_PIO1_RX3   = 15
+	DREQ_PIO2_TX0   = 16
+	DREQ_PIO2_TX1   = 17
+	DREQ_PIO2_TX2   = 18
+	DREQ_PIO2_TX3   = 19
+	DREQ_PIO2_RX0   = 20
+	DREQ_PIO2_RX1   = 21
+	DREQ_PIO2_RX2   = 22
+	DREQ_PIO2_RX3   = 23
+	DREQ_SPI0_TX    = 24
+	DREQ_SPI0_RX    = 25
+	DREQ_SPI1_TX    = 26
+	DREQ_SPI1_RX    = 27
+	DREQ_UART0_TX   = 28
+	DREQ_UART0_RX   = 29
+	DREQ_UART1_TX   = 30
+	DREQ_UART1_RX   = 31
+	DREQ_PWM_WRAP0  = 32
+	DREQ_PWM_WRAP1  = 33
+	DREQ_PWM_WRAP2  = 34
+	DREQ_PWM_WRAP3  = 35
+	DREQ_PWM_WRAP4  = 36
+	DREQ_PWM_WRAP5  = 37
+	DREQ_PWM_WRAP6  = 38
+	DREQ_PWM_WRAP7  = 39
+	DREQ_PWM_WRAP8  = 40
+	DREQ_PWM_WRAP9  = 41
+	DREQ_PWM_WRAP10 = 42
+	DREQ_PWM_WRAP11 = 43
+	DREQ_I2C0_TX    = 44
+	DREQ_I2C0_RX    = 45
+	DREQ_I2C1_TX    = 46
+	DREQ_I2C1_RX    = 47
+	DREQ_ADC        = 48
+	DREQ_XIP_STREAM = 49
+	DREQ_XIP_QMITX  = 50
+	DREQ_XIP_QMIRX  = 51
+	DREQ_HSTX       = 52
+	DREQ_CORESIGHT  = 53
+	DREQ_SHA256     = 54
+)

src/examples/usb-storage/main.go

@@ -0,0 +1,15 @@
+package main
+
+import (
+	"machine"
+	"machine/usb/msc"
+	"time"
+)
+
+func main() {
+	msc.Port(machine.Flash)
+
+	for {
+		time.Sleep(2 * time.Second)
+	}
+}

src/internal/task/atomic-cooperative.go

@@ -1,3 +1,5 @@
+//go:build tinygo.unicore
+
 package task
 
 // Atomics implementation for cooperative systems. The atomic types here aren't

src/internal/task/atomic-preemptive.go

@@ -0,0 +1,14 @@
+//go:build !tinygo.unicore
+
+package task
+
+// Atomics implementation for non-cooperative systems (multithreaded, etc).
+// These atomic types use real atomic instructions.
+
+import "sync/atomic"
+
+type (
+	Uintptr = atomic.Uintptr
+	Uint32  = atomic.Uint32
+	Uint64  = atomic.Uint64
+)

src/internal/task/futex-cooperative.go

@@ -1,3 +1,5 @@
+//go:build tinygo.unicore
+
 package task
 
 // A futex is a way for userspace to wait with the pointer as the key, and for

src/internal/task/futex-cores.go

@@ -0,0 +1,64 @@
+//go:build scheduler.cores
+
+package task
+
+import "runtime/interrupt"
+
+// 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) {
+	mask := lockFutex()
+
+	if f.Uint32.Load() != cmp {
+		unlockFutex(mask)
+		return false
+	}
+
+	// Push the current goroutine onto the waiter stack.
+	f.waiters.Push(Current())
+
+	unlockFutex(mask)
+
+	// Pause until this task is 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() {
+	mask := lockFutex()
+	if t := f.waiters.Pop(); t != nil {
+		scheduleTask(t)
+	}
+	unlockFutex(mask)
+}
+
+// Wake all waiters.
+func (f *Futex) WakeAll() {
+	mask := lockFutex()
+	for t := f.waiters.Pop(); t != nil; t = f.waiters.Pop() {
+		scheduleTask(t)
+	}
+	unlockFutex(mask)
+}
+
+//go:linkname lockFutex runtime.lockFutex
+func lockFutex() interrupt.State
+
+//go:linkname unlockFutex runtime.unlockFutex
+func unlockFutex(interrupt.State)

src/internal/task/futex-threads.go

@@ -0,0 +1,7 @@
+//go:build scheduler.threads
+
+package task
+
+import "internal/futex"
+
+type Futex = futex.Futex

src/internal/task/linux.go

@@ -0,0 +1,9 @@
+//go:build linux && !baremetal
+
+package task
+
+import "unsafe"
+
+// Musl uses a pointer (or unsigned long for C++) so unsafe.Pointer should be
+// fine.
+type threadID unsafe.Pointer

src/internal/task/mutex-cooperative.go

@@ -1,3 +1,5 @@
+//go:build tinygo.unicore
+
 package task
 
 type Mutex struct {

src/internal/task/mutex-preemptive.go

@@ -0,0 +1,71 @@
+//go:build !tinygo.unicore
+
+package task
+
+// Futex-based mutex.
+// This is largely based on the paper "Futexes are Tricky" by Ulrich Drepper.
+// It describes a few ways to implement mutexes using a futex, and how some
+// seemingly-obvious implementations don't exactly work as intended.
+// Unfortunately, Go atomic operations work slightly differently so we can't
+// copy the algorithm verbatim.
+//
+// The implementation works like this. The futex can have 3 different values,
+// depending on the state:
+//
+//   - 0: the futex is currently unlocked.
+//   - 1: the futex is locked, but is uncontended. There is one special case: if
+//     a contended futex is unlocked, it is set to 0. It is possible for another
+//     thread to lock the futex before the next waiter is woken. But because a
+//     waiter will be woken (if there is one), it will always change to 2
+//     regardless. So this is not a problem.
+//   - 2: the futex is locked, and is contended. At least one thread is trying
+//     to obtain the lock (and is in the contended loop, see below).
+//
+// For the paper, see:
+// https://dept-info.labri.fr/~denis/Enseignement/2008-IR/Articles/01-futex.pdf)
+
+type Mutex struct {
+	futex Futex
+}
+
+func (m *Mutex) Lock() {
+	// Fast path: try to take an uncontended lock.
+	if m.futex.CompareAndSwap(0, 1) {
+		// We obtained the mutex.
+		return
+	}
+
+	// The futex is contended, so we enter the contended loop.
+	// If we manage to change the futex from 0 to 2, we managed to take the
+	// lock. Else, we have to wait until a call to Unlock unlocks this mutex.
+	// (Unlock will wake one waiter when it finds the futex is set to 2 when
+	// unlocking).
+	for m.futex.Swap(2) != 0 {
+		// Wait until we get resumed in Unlock.
+		m.futex.Wait(2)
+	}
+}
+
+func (m *Mutex) Unlock() {
+	if old := m.futex.Swap(0); old == 0 {
+		// Mutex wasn't locked before.
+		panic("sync: unlock of unlocked Mutex")
+	} else if old == 2 {
+		// Mutex was a contended lock, so we need to wake the next waiter.
+		m.futex.Wake()
+	}
+}
+
+// TryLock tries to lock m and reports whether it succeeded.
+//
+// Note that while correct uses of TryLock do exist, they are rare,
+// and use of TryLock is often a sign of a deeper problem
+// in a particular use of mutexes.
+func (m *Mutex) TryLock() bool {
+	// Fast path: try to take an uncontended lock.
+	if m.futex.CompareAndSwap(0, 1) {
+		// We obtained the mutex.
+		return true
+	}
+	return false
+}

src/internal/task/pmutex-cooperative.go

@@ -1,3 +1,5 @@
+//go:build tinygo.unicore
+
 package task
 
 // PMutex is a real mutex on systems that can be either preemptive or threaded,

src/internal/task/pmutex-preemptive.go

@@ -0,0 +1,11 @@
+//go:build !tinygo.unicore
+
+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 = Mutex

src/internal/task/queue.go

@@ -12,9 +12,9 @@ type Queue struct {
 
 // Push a task onto the queue.
 func (q *Queue) Push(t *Task) {
-	i := interrupt.Disable()
+	mask := lockAtomics()
 	if asserts && t.Next != nil {
-		interrupt.Restore(i)
+		unlockAtomics(mask)
 		panic("runtime: pushing a task to a queue with a non-nil Next pointer")
 	}
 	if q.tail != nil {
@@ -25,15 +25,15 @@ func (q *Queue) Push(t *Task) {
 	if q.head == nil {
 		q.head = t
 	}
-	interrupt.Restore(i)
+	unlockAtomics(mask)
 }
 
 // Pop a task off of the queue.
 func (q *Queue) Pop() *Task {
-	i := interrupt.Disable()
+	mask := lockAtomics()
 	t := q.head
 	if t == nil {
-		interrupt.Restore(i)
+		unlockAtomics(mask)
 		return nil
 	}
 	q.head = t.Next
@@ -41,13 +41,13 @@ func (q *Queue) Pop() *Task {
 		q.tail = nil
 	}
 	t.Next = nil
-	interrupt.Restore(i)
+	unlockAtomics(mask)
 	return t
 }
 
 // Append pops the contents of another queue and pushes them onto the end of this queue.
 func (q *Queue) Append(other *Queue) {
-	i := interrupt.Disable()
+	mask := lockAtomics()
 	if q.head == nil {
 		q.head = other.head
 	} else {
@@ -55,14 +55,14 @@ func (q *Queue) Append(other *Queue) {
 	}
 	q.tail = other.tail
 	other.head, other.tail = nil, nil
-	interrupt.Restore(i)
+	unlockAtomics(mask)
 }
 
 // Empty checks if the queue is empty.
 func (q *Queue) Empty() bool {
-	i := interrupt.Disable()
+	mask := lockAtomics()
 	empty := q.head == nil
-	interrupt.Restore(i)
+	unlockAtomics(mask)
 	return empty
 }
 
@@ -75,24 +75,24 @@ type Stack struct {
 
 // Push a task onto the stack.
 func (s *Stack) Push(t *Task) {
-	i := interrupt.Disable()
+	mask := lockAtomics()
 	if asserts && t.Next != nil {
-		interrupt.Restore(i)
+		unlockAtomics(mask)
 		panic("runtime: pushing a task to a stack with a non-nil Next pointer")
 	}
 	s.top, t.Next = t, s.top
-	interrupt.Restore(i)
+	unlockAtomics(mask)
 }
 
 // Pop a task off of the stack.
 func (s *Stack) Pop() *Task {
-	i := interrupt.Disable()
+	mask := lockAtomics()
 	t := s.top
 	if t != nil {
 		s.top = t.Next
 		t.Next = nil
 	}
-	interrupt.Restore(i)
+	unlockAtomics(mask)
 	return t
 }
 
@@ -112,13 +112,26 @@ func (t *Task) tail() *Task {
 // Queue moves the contents of the stack into a queue.
 // Elements can be popped from the queue in the same order that they would be popped from the stack.
 func (s *Stack) Queue() Queue {
-	i := interrupt.Disable()
+	mask := lockAtomics()
 	head := s.top
 	s.top = nil
 	q := Queue{
 		head: head,
 		tail: head.tail(),
 	}
-	interrupt.Restore(i)
+	unlockAtomics(mask)
 	return q
 }
+
+// Use runtime.lockAtomics and runtime.unlockAtomics so that Queue and Stack
+// work correctly even on multicore systems. These functions are normally used
+// to implement atomic operations, but the same spinlock can also be used for
+// Queue/Stack operations which are very fast.
+// These functions are just plain old interrupt disable/restore on non-multicore
+// systems.
+
+//go:linkname lockAtomics runtime.lockAtomics
+func lockAtomics() interrupt.State
+
+//go:linkname unlockAtomics runtime.unlockAtomics
+func unlockAtomics(mask interrupt.State)

src/internal/task/semaphore.go

@@ -0,0 +1,32 @@
+package task
+
+// Barebones semaphore implementation.
+// The main limitation is that if there are multiple waiters, a single Post()
+// call won't do anything. Only when Post() has been called to awaken all
+// waiters will the waiters proceed.
+// This limitation is not a problem when there will only be a single waiter.
+type Semaphore struct {
+	futex Futex
+}
+
+// Post (unlock) the semaphore, incrementing the value in the semaphore.
+func (s *Semaphore) Post() {
+	newValue := s.futex.Add(1)
+	if newValue == 0 {
+		s.futex.WakeAll()
+	}
+}
+
+// Wait (lock) the semaphore, decrementing the value in the semaphore.
+func (s *Semaphore) Wait() {
+	delta := int32(-1)
+	value := s.futex.Add(uint32(delta))
+	for {
+		if int32(value) >= 0 {
+			// Semaphore unlocked!
+			return
+		}
+		s.futex.Wait(value)
+		value = s.futex.Load()
+	}
+}

src/internal/task/task.go

@@ -24,11 +24,28 @@ type Task struct {
 	// This is needed for some crypto packages.
 	FipsIndicator uint8
 
+	// State of the goroutine: running, paused, or must-resume-next-pause.
+	// This extra field doesn't increase memory usage on 32-bit CPUs and above,
+	// since it falls into the padding of the FipsIndicator bit above.
+	RunState uint8
+
 	// DeferFrame stores a pointer to the (stack allocated) defer frame of the
 	// goroutine that is used for the recover builtin.
 	DeferFrame unsafe.Pointer
 }
 
+const (
+	// Initial state: the goroutine state is saved on the stack.
+	RunStatePaused = iota
+
+	// The goroutine is running right now.
+	RunStateRunning
+
+	// The goroutine is running, but already marked as "can resume".
+	// The next call to Pause() won't actually pause the goroutine.
+	RunStateResuming
+)
+
 // 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 {

src/internal/task/task_none.go

@@ -36,3 +36,9 @@ func OnSystemStack() bool {
 	// This scheduler does not do any stack switching.
 	return true
 }
+
+func SystemStack() uintptr {
+	// System stack is the current stack, so this shouldn't be called.
+	runtimePanic("scheduler is disabled")
+	return 0 // unreachable
+}

src/internal/task/task_stack.go

@@ -1,9 +1,8 @@
-//go:build scheduler.tasks
+//go:build scheduler.tasks || scheduler.cores
 
 package task
 
 import (
-	"runtime/interrupt"
 	"unsafe"
 )
 
@@ -32,43 +31,12 @@ type state struct {
 	canaryPtr *uintptr
 }
 
-// currentTask is the current running task, or nil if currently in the scheduler.
+//export tinygo_task_exit
-var currentTask *Task
+func taskExit() {
-
+	// TODO: explicitly free the stack after switching back to the scheduler.
-// Current returns the current active task.
-func Current() *Task {
-	return currentTask
-}
-
-// Pause suspends the current task and returns to the scheduler.
-// This function may only be called when running on a goroutine stack, not when running on the system stack or in an interrupt.
-func Pause() {
-	// Check whether the canary (the lowest address of the stack) is still
-	// valid. If it is not, a stack overflow has occurred.
-	if *currentTask.state.canaryPtr != stackCanary {
-		runtimePanic("goroutine stack overflow")
-	}
-	if interrupt.In() {
-		runtimePanic("blocked inside interrupt")
-	}
-	currentTask.state.pause()
-}
-
-//export tinygo_pause
-func pause() {
 	Pause()
 }
 
-// Resume the task until it pauses or completes.
-// This may only be called from the scheduler.
-func (t *Task) Resume() {
-	currentTask = t
-	t.gcData.swap()
-	t.state.resume()
-	t.gcData.swap()
-	currentTask = nil
-}
-
 // initialize the state and prepare to call the specified function with the specified argument bundle.
 func (s *state) initialize(fn uintptr, args unsafe.Pointer, stackSize uintptr) {
 	// Create a stack.

src/internal/task/task_stack_386.S

@@ -1,7 +1,12 @@
+#ifdef _WIN32
+.global  _tinygo_startTask
+_tinygo_startTask:
+#else // Linux etc
 .section .text.tinygo_startTask
 .global  tinygo_startTask
 .type    tinygo_startTask, %function
 tinygo_startTask:
+#endif
     .cfi_startproc
     // Small assembly stub for starting a goroutine. This is already run on the
     // new stack, with the callee-saved registers already loaded.
@@ -24,12 +29,21 @@ tinygo_startTask:
     addl $4, %esp
 
     // After return, exit this goroutine. This is a tail call.
-    jmp tinygo_pause
+    #ifdef _WIN32
+    jmp _tinygo_task_exit
+    #else
+    jmp tinygo_task_exit
+    #endif
     .cfi_endproc
 
+#ifdef _WIN32
+.global _tinygo_swapTask
+_tinygo_swapTask:
+#else
 .global tinygo_swapTask
 .type tinygo_swapTask, %function
 tinygo_swapTask:
+#endif
     // This function gets the following parameters:
     movl 4(%esp), %eax // newStack uintptr
     movl 8(%esp), %ecx // oldStack *uintptr

src/internal/task/task_stack_amd64.S

@@ -30,9 +30,9 @@ tinygo_startTask:
 
     // After return, exit this goroutine. This is a tail call.
     #ifdef __MACH__
-    jmp _tinygo_pause
+    jmp _tinygo_task_exit
     #else
-    jmp tinygo_pause
+    jmp tinygo_task_exit
     #endif
     .cfi_endproc
 

src/internal/task/task_stack_amd64_windows.S

@@ -22,7 +22,7 @@ tinygo_startTask:
 
     // After return, exit this goroutine.
     // This has to be a call, not a jump, to keep the stack correctly aligned.
-    callq tinygo_pause
+    callq tinygo_task_exit
 
 .global tinygo_swapTask
 .section .text.tinygo_swapTask,"ax"

src/internal/task/task_stack_arm.S

@@ -28,7 +28,7 @@ tinygo_startTask:
     blx   r4
 
     // After return, exit this goroutine. This is a tail call.
-    bl    tinygo_pause
+    bl    tinygo_task_exit
     .cfi_endproc
 .size tinygo_startTask, .-tinygo_startTask
 

src/internal/task/task_stack_arm64.S

@@ -27,9 +27,9 @@ tinygo_startTask:
 
     // After return, exit this goroutine. This is a tail call.
 #ifdef __MACH__
-    b     _tinygo_pause
+    b     _tinygo_task_exit
 #else
-    b     tinygo_pause
+    b     tinygo_task_exit
 #endif
     .cfi_endproc
 #ifndef __MACH__

src/internal/task/task_stack_avr.S

@@ -32,10 +32,10 @@ tinygo_startTask:
     // Note that they will probably not be able to run more than the main
     // goroutine anyway, but this file is compiled for all AVRs so it needs to
     // compile at least.
-    rcall  tinygo_pause
+    rcall  tinygo_task_exit
 #else
     // Other devices can (and must) use the regular call instruction.
-    call tinygo_pause
+    call tinygo_task_exit
 #endif
 
 .global tinygo_swapTask

src/internal/task/task_stack_cortexm.S

@@ -28,7 +28,7 @@ tinygo_startTask:
     blx   r4
 
     // After return, exit this goroutine. This is a tail call.
-    bl    	tinygo_pause
+    bl    	tinygo_task_exit
     .cfi_endproc
 .size tinygo_startTask, .-tinygo_startTask
 

src/internal/task/task_stack_cortexm.c

@@ -1,4 +1,4 @@
-//go:build scheduler.tasks && cortexm
+//go:build (scheduler.tasks || scheduler.cores) && cortexm
 #include <stdint.h>
 
 uintptr_t SystemStack() {

src/internal/task/task_stack_cortexm.go

@@ -1,4 +1,4 @@
-//go:build scheduler.tasks && cortexm
+//go:build (scheduler.tasks || scheduler.cores) && cortexm
 
 package task
 

src/internal/task/task_stack_esp32.S

@@ -27,7 +27,7 @@ tinygo_startTask:
     callx4 a3
 
     // After return, exit this goroutine. This call never returns.
-    call4  tinygo_pause
+    call4  tinygo_task_exit
 
 .section .text.tinygo_swapTask,"ax",@progbits
 .global tinygo_swapTask

src/internal/task/task_stack_esp8266.S

@@ -18,7 +18,7 @@ tinygo_startTask:
     callx0 a12
 
     // After return, exit this goroutine. This is a tail call.
-    call0  tinygo_pause
+    call0  tinygo_task_exit
 .size tinygo_startTask, .-tinygo_startTask
 
 .global tinygo_swapTask

src/internal/task/task_stack_mipsx.S

@@ -22,7 +22,7 @@ tinygo_startTask:
     nop
 
     // After return, exit this goroutine. This is a tail call.
-    j     tinygo_pause
+    j     tinygo_task_exit
     nop
 
 .section .text.tinygo_swapTask

src/internal/task/task_stack_multicore.go

@@ -0,0 +1,53 @@
+//go:build scheduler.cores
+
+package task
+
+import "runtime/interrupt"
+
+// Current returns the current active task.
+//
+//go:linkname Current runtime.currentTask
+func Current() *Task
+
+// Pause suspends the current task and returns to the scheduler.
+// This function may only be called when running on a goroutine stack, not when running on the system stack or in an interrupt.
+func Pause() {
+	lockScheduler()
+	PauseLocked()
+}
+
+// PauseLocked is the same as Pause, but must be called with the scheduler lock
+// already taken.
+func PauseLocked() {
+	// Check whether the canary (the lowest address of the stack) is still
+	// valid. If it is not, a stack overflow has occurred.
+	current := Current()
+	if *current.state.canaryPtr != stackCanary {
+		runtimePanic("goroutine stack overflow")
+	}
+	if interrupt.In() {
+		runtimePanic("blocked inside interrupt")
+	}
+	if current.RunState == RunStateResuming {
+		// Another core already marked this goroutine as ready to resume.
+		current.RunState = RunStateRunning
+		unlockScheduler()
+		return
+	}
+	current.RunState = RunStatePaused
+	current.state.pause()
+}
+
+// Resume the task until it pauses or completes.
+// This may only be called from the scheduler.
+func (t *Task) Resume() {
+	t.gcData.swap()
+	t.state.resume()
+	t.gcData.swap()
+}
+
+//go:linkname lockScheduler runtime.lockScheduler
+func lockScheduler()
+
+//go:linkname unlockScheduler runtime.unlockScheduler
+func unlockScheduler()

src/internal/task/task_stack_tinygoriscv.S

@@ -19,7 +19,7 @@ tinygo_startTask:
     jalr  s0
 
     // After return, exit this goroutine. This is a tail call.
-    tail  tinygo_pause
+    tail  tinygo_task_exit
 
 .section .text.tinygo_swapTask
 .global  tinygo_swapTask

src/internal/task/task_stack_tinygoriscv.go

@@ -1,10 +1,16 @@
-//go:build scheduler.tasks && tinygo.riscv
+//go:build (scheduler.tasks || scheduler.cores) && tinygo.riscv
 
 package task
 
 import "unsafe"
 
-var systemStack uintptr
+// Returns a pointer where the system stack can be stored.
+// This is a layering violation! We should probably refactor this so that we
+// don't need such gymnastics to store the system stack pointer. (It should
+// probably be moved to the runtime).
+//
+//go:linkname runtime_systemStackPtr runtime.systemStackPtr
+func runtime_systemStackPtr() *uintptr
 
 // calleeSavedRegs is the list of registers that must be saved and restored when
 // switching between tasks. Also see scheduler_riscv.S that relies on the
@@ -50,17 +56,18 @@ func (s *state) archInit(r *calleeSavedRegs, fn uintptr, args unsafe.Pointer) {
 }
 
 func (s *state) resume() {
-	swapTask(s.sp, &systemStack)
+	swapTask(s.sp, runtime_systemStackPtr())
 }
 
 func (s *state) pause() {
-	newStack := systemStack
+	systemStackPtr := runtime_systemStackPtr()
-	systemStack = 0
+	newStack := *systemStackPtr
+	*systemStackPtr = 0
 	swapTask(newStack, &s.sp)
 }
 
 // SystemStack returns the system stack pointer when called from a task stack.
 // When called from the system stack, it returns 0.
 func SystemStack() uintptr {
-	return systemStack
+	return *runtime_systemStackPtr()
 }

src/internal/task/task_stack_unicore.go

@@ -0,0 +1,37 @@
+//go:build scheduler.tasks
+
+package task
+
+import "runtime/interrupt"
+
+// currentTask is the current running task, or nil if currently in the scheduler.
+var currentTask *Task
+
+// Current returns the current active task.
+func Current() *Task {
+	return currentTask
+}
+
+// Pause suspends the current task and returns to the scheduler.
+// This function may only be called when running on a goroutine stack, not when running on the system stack or in an interrupt.
+func Pause() {
+	// Check whether the canary (the lowest address of the stack) is still
+	// valid. If it is not, a stack overflow has occurred.
+	if *currentTask.state.canaryPtr != stackCanary {
+		runtimePanic("goroutine stack overflow")
+	}
+	if interrupt.In() {
+		runtimePanic("blocked inside interrupt")
+	}
+	currentTask.state.pause()
+}
+
+// Resume the task until it pauses or completes.
+// This may only be called from the scheduler.
+func (t *Task) Resume() {
+	currentTask = t
+	t.gcData.swap()
+	t.state.resume()
+	t.gcData.swap()
+	currentTask = nil
+}

src/internal/task/task_threads.c

@@ -0,0 +1,113 @@
+//go:build none
+
+#define _GNU_SOURCE
+#include <pthread.h>
+#include <semaphore.h>
+#include <signal.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <unistd.h>
+
+// BDWGC also uses SIGRTMIN+6 on Linux, which seems like a reasonable choice.
+#ifdef __linux__
+#define taskPauseSignal (SIGRTMIN + 6)
+#endif
+
+// Pointer to the current task.Task structure.
+// Ideally the entire task.Task structure would be a thread-local variable but
+// this also works.
+static __thread void *current_task;
+
+struct state_pass {
+    void      *(*start)(void*);
+    void      *args;
+    void      *task;
+    uintptr_t *stackTop;
+    sem_t     startlock;
+};
+
+// Handle the GC pause in Go.
+void tinygo_task_gc_pause(int sig);
+
+// Initialize the main thread.
+void tinygo_task_init(void *mainTask, pthread_t *thread, int *numCPU, void *context) {
+    // Make sure the current task pointer is set correctly for the main
+    // goroutine as well.
+    current_task = mainTask;
+
+    // Store the thread ID of the main thread.
+    *thread = pthread_self();
+
+    // Register the "GC pause" signal for the entire process.
+    // Using pthread_kill, we can still send the signal to a specific thread.
+    struct sigaction act = { 0 };
+    act.sa_flags = SA_SIGINFO;
+    act.sa_handler = &tinygo_task_gc_pause;
+    sigaction(taskPauseSignal, &act, NULL);
+
+    // Obtain the number of CPUs available on program start (for NumCPU).
+    int num = sysconf(_SC_NPROCESSORS_ONLN);
+    if (num <= 0) {
+        // Fallback in case there is an error.
+        num = 1;
+    }
+    *numCPU = num;
+}
+
+void tinygo_task_exited(void*);
+
+// Helper to start a goroutine while also storing the 'task' structure.
+static void* start_wrapper(void *arg) {
+    struct state_pass *state = arg;
+    void *(*start)(void*) = state->start;
+    void *args = state->args;
+    current_task = state->task;
+
+    // Save the current stack pointer in the goroutine state, for the GC.
+    int stackAddr;
+    *(state->stackTop) = (uintptr_t)(&stackAddr);
+
+    // Notify the caller that the thread has successfully started and
+    // initialized.
+    sem_post(&state->startlock);
+
+    // Run the goroutine function.
+    start(args);
+
+    // Notify the Go side this thread will exit.
+    tinygo_task_exited(current_task);
+
+    return NULL;
+};
+
+// Start a new goroutine in an OS thread.
+int tinygo_task_start(uintptr_t fn, void *args, void *task, pthread_t *thread, uintptr_t *stackTop, void *context) {
+    // Sanity check. Should get optimized away.
+    if (sizeof(pthread_t) != sizeof(void*)) {
+        __builtin_trap();
+    }
+
+    struct state_pass state = {
+        .start     = (void*)fn,
+        .args      = args,
+        .task      = task,
+        .stackTop  = stackTop,
+    };
+    sem_init(&state.startlock, 0, 0);
+    int result = pthread_create(thread, NULL, &start_wrapper, &state);
+
+    // Wait until the thread has been created and read all state_pass variables.
+    sem_wait(&state.startlock);
+
+    return result;
+}
+
+// Return the current task (for task.Current()).
+void* tinygo_task_current(void) {
+    return current_task;
+}
+
+// Send a signal to cause the task to pause for the GC mark phase.
+void tinygo_task_send_gc_signal(pthread_t thread) {
+    pthread_kill(thread, taskPauseSignal);
+}

src/internal/task/task_threads.go

@@ -0,0 +1,281 @@
+//go:build scheduler.threads
+
+package task
+
+import (
+	"sync/atomic"
+	"unsafe"
+)
+
+// If true, print verbose debug logs.
+const verbose = false
+
+// Scheduler-specific state.
+type state struct {
+	// Goroutine ID. The number here is not really significant and after a while
+	// it could wrap around. But it is useful for debugging.
+	id uintptr
+
+	// Thread ID, pthread_t or similar (typically implemented as a pointer).
+	thread threadID
+
+	// Highest address of the stack. It is stored when the goroutine starts, and
+	// is needed to be able to scan the stack.
+	stackTop uintptr
+
+	// Next task in the activeTasks queue.
+	QueueNext *Task
+
+	// Semaphore to pause/resume the thread atomically.
+	pauseSem Semaphore
+
+	// Semaphore used for stack scanning.
+	// We can't reuse pauseSem here since the thread might have been paused for
+	// other reasons (for example, because it was waiting on a channel).
+	gcSem Semaphore
+}
+
+// Goroutine counter, starting at 0 for the main goroutine.
+var goroutineID uintptr
+
+var numCPU int32
+
+var mainTask Task
+
+// Queue of tasks (see QueueNext) that currently exist in the program.
+var activeTasks = &mainTask
+var activeTaskLock PMutex
+
+func OnSystemStack() bool {
+	runtimePanic("todo: task.OnSystemStack")
+	return false
+}
+
+// Initialize the main goroutine state. Must be called by the runtime on
+// startup, before starting any other goroutines.
+func Init(sp uintptr) {
+	mainTask.state.stackTop = sp
+	tinygo_task_init(&mainTask, &mainTask.state.thread, &numCPU)
+}
+
+// Return the task struct for the current thread.
+func Current() *Task {
+	t := (*Task)(tinygo_task_current())
+	if t == nil {
+		runtimePanic("unknown current task")
+	}
+	return t
+}
+
+// Pause pauses the current task, until it is resumed by another task.
+// It is possible that another task has called Resume() on the task before it
+// hits Pause(), in which case the task won't be paused but continues
+// immediately.
+func Pause() {
+	// Wait until resumed
+	t := Current()
+	if verbose {
+		println("*** pause:  ", t.state.id)
+	}
+	t.state.pauseSem.Wait()
+}
+
+// Resume the given task.
+// It is legal to resume a task before it gets paused, it means that the next
+// call to Pause() won't pause but will continue immediately. This happens in
+// practice sometimes in channel operations, where the Resume() might get called
+// between the channel unlock and the call to Pause().
+func (t *Task) Resume() {
+	if verbose {
+		println("*** resume: ", t.state.id)
+	}
+	// Increment the semaphore counter.
+	// If the task is currently paused in Wait(), it will resume.
+	// If the task is not yet paused, the next call to Wait() will continue
+	// immediately.
+	t.state.pauseSem.Post()
+}
+
+// Start a new OS thread.
+func start(fn uintptr, args unsafe.Pointer, stackSize uintptr) {
+	t := &Task{}
+	t.state.id = atomic.AddUintptr(&goroutineID, 1)
+	if verbose {
+		println("*** start:  ", t.state.id, "from", Current().state.id)
+	}
+
+	// Start the new thread, and add it to the list of threads.
+	// Do this with a lock so that only started threads are part of the queue
+	// and the stop-the-world GC won't see threads that haven't started yet or
+	// are not fully started yet.
+	activeTaskLock.Lock()
+	errCode := tinygo_task_start(fn, args, t, &t.state.thread, &t.state.stackTop)
+	if errCode != 0 {
+		runtimePanic("could not start thread")
+	}
+	t.state.QueueNext = activeTasks
+	activeTasks = t
+	activeTaskLock.Unlock()
+}
+
+//export tinygo_task_exited
+func taskExited(t *Task) {
+	if verbose {
+		println("*** exit:", t.state.id)
+	}
+
+	// Remove from the queue.
+	// TODO: this can be made more efficient by using a doubly linked list.
+	activeTaskLock.Lock()
+	found := false
+	for q := &activeTasks; *q != nil; q = &(*q).state.QueueNext {
+		if *q == t {
+			*q = t.state.QueueNext
+			found = true
+			break
+		}
+	}
+	activeTaskLock.Unlock()
+
+	// Sanity check.
+	if !found {
+		runtimePanic("taskExited failed")
+	}
+}
+
+// Futex to wait on until all tasks have finished scanning the stack.
+// This is basically a sync.WaitGroup.
+var scanDoneFutex Futex
+
+// GC scan phase. Because we need to stop the world while scanning, this kinda
+// needs to be done in the tasks package.
+//
+// After calling this function, GCResumeWorld needs to be called once to resume
+// all threads again.
+func GCStopWorldAndScan() {
+	current := Current()
+
+	// Don't allow new goroutines to be started while pausing/resuming threads
+	// in the stop-the-world phase.
+	activeTaskLock.Lock()
+
+	// Pause all other threads.
+	numOtherThreads := uint32(0)
+	for t := activeTasks; t != nil; t = t.state.QueueNext {
+		if t != current {
+			numOtherThreads++
+			tinygo_task_send_gc_signal(t.state.thread)
+		}
+	}
+
+	// Store the number of threads to wait for in the futex.
+	// This is the equivalent of doing an initial wg.Add(numOtherThreads).
+	scanDoneFutex.Store(numOtherThreads)
+
+	// Scan the current stack, and all current registers.
+	scanCurrentStack()
+
+	// Wake each paused thread for the first time so it will scan the stack.
+	for t := activeTasks; t != nil; t = t.state.QueueNext {
+		if t != current {
+			t.state.gcSem.Post()
+		}
+	}
+
+	// Wait until all threads have finished scanning their stack.
+	// This is the equivalent of wg.Wait()
+	for {
+		val := scanDoneFutex.Load()
+		if val == 0 {
+			break
+		}
+		scanDoneFutex.Wait(val)
+	}
+
+	// Scan all globals (implemented in the runtime).
+	gcScanGlobals()
+}
+
+// After the GC is done scanning, resume all other threads.
+//
+// This must only be called after a GCStopWorldAndScan call.
+func GCResumeWorld() {
+	current := Current()
+
+	// Wake each paused thread for the second time, so they will resume normal
+	// operation.
+	for t := activeTasks; t != nil; t = t.state.QueueNext {
+		if t != current {
+			t.state.gcSem.Post()
+		}
+	}
+
+	// Allow goroutines to start and exit again.
+	activeTaskLock.Unlock()
+}
+
+// Scan globals, implemented in the runtime package.
+func gcScanGlobals()
+
+var stackScanLock PMutex
+
+//export tinygo_task_gc_pause
+func tingyo_task_gc_pause() {
+	// Wait until we get the signal to start scanning the stack.
+	Current().state.gcSem.Wait()
+
+	// Scan the thread stack.
+	// Only scan a single thread stack at a time, because the GC marking phase
+	// doesn't support parallelism.
+	// TODO: it may be possible to call markRoots directly (without saving
+	// registers) since we are in a signal handler that already saved a bunch of
+	// registers. This is an optimization left for a future time.
+	stackScanLock.Lock()
+	scanCurrentStack()
+	stackScanLock.Unlock()
+
+	// Equivalent of wg.Done(): subtract one from the futex and if the result is
+	// 0 (meaning we were the last in the waitgroup), wake the waiting thread.
+	n := uint32(1)
+	if scanDoneFutex.Add(-n) == 0 {
+		scanDoneFutex.Wake()
+	}
+
+	// Wait until we get the signal we can resume normally (after the mark phase
+	// has finished).
+	Current().state.gcSem.Wait()
+}
+
+//go:export tinygo_scanCurrentStack
+func scanCurrentStack()
+
+// Return the highest address of the current stack.
+func StackTop() uintptr {
+	return Current().state.stackTop
+}
+
+//go:linkname runtimePanic runtime.runtimePanic
+func runtimePanic(msg string)
+
+// Using //go:linkname instead of //export so that we don't tell the compiler
+// that the 't' parameter won't escape (because it will).
+//
+//go:linkname tinygo_task_init tinygo_task_init
+func tinygo_task_init(t *Task, thread *threadID, numCPU *int32)
+
+// Here same as for tinygo_task_init.
+//
+//go:linkname tinygo_task_start tinygo_task_start
+func tinygo_task_start(fn uintptr, args unsafe.Pointer, t *Task, thread *threadID, stackTop *uintptr) int32
+
+// Pause the thread by sending it a signal.
+//
+//export tinygo_task_send_gc_signal
+func tinygo_task_send_gc_signal(threadID)
+
+//export tinygo_task_current
+func tinygo_task_current() unsafe.Pointer
+
+func NumCPU() int {
+	return int(numCPU)
+}

src/machine/board_btt_skr_pico.go

@@ -0,0 +1,121 @@
+//go:build btt_skr_pico
+
+// This contains the pin mappings for the BigTreeTech SKR Pico.
+//
+// Purchase link: https://biqu.equipment/products/btt-skr-pico-v1-0
+// Board schematic: https://github.com/bigtreetech/SKR-Pico/blob/master/Hardware/BTT%20SKR%20Pico%20V1.0-SCH.pdf
+// Pin diagram: https://github.com/bigtreetech/SKR-Pico/blob/master/Hardware/BTT%20SKR%20Pico%20V1.0-PIN.pdf
+
+package machine
+
+// TMC stepper driver motor direction.
+// X/Y/Z/E refers to motors for X/Y/Z and the extruder.
+const (
+	X_DIR = GPIO10
+	Y_DIR = GPIO5
+	Z_DIR = GPIO28
+	E_DIR = GPIO13
+)
+
+// TMC stepper driver motor step
+const (
+	X_STEP = GPIO11
+	Y_STEP = GPIO6
+	Z_STEP = GPIO19
+	E_STEP = GPIO14
+)
+
+// TMC stepper driver enable
+const (
+	X_ENABLE = GPIO12
+	Y_ENABLE = GPIO7
+	Z_ENABLE = GPIO2
+	E_ENABLE = GPIO15
+)
+
+// TMC stepper driver UART
+const (
+	TMC_UART_TX = UART1_TX_PIN
+	TMC_UART_RX = UART1_RX_PIN
+)
+
+// Endstops
+const (
+	X_ENDSTOP = GPIO4
+	Y_ENDSTOP = GPIO3
+	Z_ENDSTOP = GPIO25
+	E_ENDSTOP = GPIO16
+)
+
+// Fan PWM
+const (
+	FAN1_PWM = GPIO17
+	FAN2_PWM = GPIO18
+	FAN3_PWM = GPIO20
+)
+
+// Heater PWM
+const (
+	HEATER_BED_PWM      = GPIO21
+	HEATER_EXTRUDER_PWM = GPIO23
+)
+
+// Thermistors
+const (
+	THERM_BED      = GPIO26 // Bed heater
+	THERM_EXTRUDER = GPIO27 // Toolhead heater
+)
+
+// Misc
+const (
+	RGB   = GPIO24 // Neopixel
+	SERVO = GPIO29 // Servo
+	PROBE = GPIO22 // Probe
+)
+
+// Onboard crystal oscillator frequency, in MHz.
+const (
+	xoscFreq = 12 // MHz
+)
+
+// I2C. We don't have this available
+const (
+	I2C0_SDA_PIN = NoPin
+	I2C0_SCL_PIN = NoPin
+
+	I2C1_SDA_PIN = NoPin
+	I2C1_SCL_PIN = NoPin
+)
+
+// SPI. We don't have this available
+const (
+	SPI0_SCK_PIN = NoPin
+	SPI0_SDO_PIN = NoPin
+	SPI0_SDI_PIN = NoPin
+	SPI1_SCK_PIN = NoPin
+	SPI1_SDO_PIN = NoPin
+	SPI1_SDI_PIN = NoPin
+)
+
+// USB CDC identifiers
+const (
+	usb_STRING_PRODUCT      = "SKR Pico"
+	usb_STRING_MANUFACTURER = "BigTreeTech"
+)
+
+var (
+	usb_VID uint16 = 0x2e8a
+	usb_PID uint16 = 0x0003
+)
+
+// 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

src/machine/machine_atsamd21.go

@@ -1904,7 +1904,7 @@ func (f flashBlockDevice) ReadAt(p []byte, off int64) (n int, err error) {
 }
 
 // WriteAt writes the given number of bytes to the block device.
-// Only word (32 bits) length data can be programmed.
+// Data is written to the page buffer in 4-byte chunks, then saved to flash memory.
 // See Atmel-42181G–SAM-D21_Datasheet–09/2015 page 359.
 // If the length of p is not long enough it will be padded with 0xFF bytes.
 // This method assumes that the destination is already erased.
@@ -1921,8 +1921,10 @@ func (f flashBlockDevice) WriteAt(p []byte, off int64) (n int, err error) {
 	waitWhileFlashBusy()
 
 	for j := 0; j < len(padded); j += int(f.WriteBlockSize()) {
-		// write word
+		// page buffer is 64 bytes long, but only 4 bytes can be written at once
-		*(*uint32)(unsafe.Pointer(address)) = binary.LittleEndian.Uint32(padded[j : j+int(f.WriteBlockSize())])
+		for k := 0; k < int(f.WriteBlockSize()); k += 4 {
+			*(*uint32)(unsafe.Pointer(address + uintptr(k))) = binary.LittleEndian.Uint32(padded[j+k : j+k+4])
+		}
 
 		sam.NVMCTRL.SetADDR(uint32(address >> 1))
 		sam.NVMCTRL.CTRLA.Set(sam.NVMCTRL_CTRLA_CMD_WP | (sam.NVMCTRL_CTRLA_CMDEX_KEY << sam.NVMCTRL_CTRLA_CMDEX_Pos))
@@ -1944,7 +1946,7 @@ func (f flashBlockDevice) Size() int64 {
 	return int64(FlashDataEnd() - FlashDataStart())
 }
 
-const writeBlockSize = 4
+const writeBlockSize = 64
 
 // WriteBlockSize returns the block size in which data can be written to
 // memory. It can be used by a client to optimize writes, non-aligned writes
@@ -2027,3 +2029,62 @@ func checkFlashError() error {
 
 	return nil
 }
+
+// Watchdog provides access to the hardware watchdog available
+// in the SAMD21.
+var Watchdog = &watchdogImpl{}
+
+const (
+	// WatchdogMaxTimeout in milliseconds (16s)
+	WatchdogMaxTimeout = (16384 * 1000) / 1024
+)
+
+type watchdogImpl struct{}
+
+// Configure the watchdog.
+//
+// This method should not be called after the watchdog is started and on
+// some platforms attempting to reconfigure after starting the watchdog
+// is explicitly forbidden / will not work.
+func (wd *watchdogImpl) Configure(config WatchdogConfig) error {
+	// Use OSCULP32K as source for Generic Clock Generator 8, divided by 32 to get 1.024kHz
+	sam.GCLK.GENDIV.Set(sam.GCLK_CLKCTRL_GEN_GCLK8 | (32 << sam.GCLK_GENDIV_DIV_Pos))
+	sam.GCLK.GENCTRL.Set(sam.GCLK_CLKCTRL_GEN_GCLK8 | (sam.GCLK_GENCTRL_SRC_OSCULP32K << sam.GCLK_GENCTRL_SRC_Pos) | sam.GCLK_GENCTRL_GENEN)
+	waitForSync()
+
+	// Use GCLK8 for watchdog
+	sam.GCLK.CLKCTRL.Set(sam.GCLK_CLKCTRL_ID_WDT | (sam.GCLK_CLKCTRL_GEN_GCLK8 << sam.GCLK_CLKCTRL_GEN_Pos) | sam.GCLK_CLKCTRL_CLKEN)
+
+	// Power on the watchdog peripheral
+	sam.PM.APBAMASK.SetBits(sam.PM_APBAMASK_WDT_)
+
+	// 1.024kHz clock
+	cycles := int((int64(config.TimeoutMillis) * 1024) / 1000)
+
+	// period is expressed as a power-of-two, starting at 8 / 1024ths of a second
+	period := uint8(0)
+	cfgCycles := 8
+	for cfgCycles < cycles {
+		period++
+		cfgCycles <<= 1
+
+		if period >= 0xB {
+			break
+		}
+	}
+
+	sam.WDT.CONFIG.Set(period << sam.WDT_CONFIG_PER_Pos)
+
+	return nil
+}
+
+// Starts the watchdog.
+func (wd *watchdogImpl) Start() error {
+	sam.WDT.CTRL.SetBits(sam.WDT_CTRL_ENABLE)
+	return nil
+}
+
+// Update the watchdog, indicating that `source` is healthy.
+func (wd *watchdogImpl) Update() {
+	sam.WDT.CLEAR.Set(sam.WDT_CLEAR_CLEAR_KEY)
+}

src/machine/machine_atsamd21_simulator.go

@@ -0,0 +1,45 @@
+//go:build !baremetal && (gemma_m0 || qtpy || trinket_m0 || arduino_mkr1000 || arduino_mkrwifi1010 || arduino_nano33 || arduino_zero || circuitplay_express || feather_m0_express || feather_m0 || itsybitsy_m0 || p1am_100 || xiao)
+
+// Simulated atsamd21 chips.
+
+package machine
+
+// The timer channels/pins match the hardware, and encode the same information
+// as pinTimerMapping but in a more generic (less efficient) way.
+
+var TCC0 = &timerType{
+	instance:   0,
+	frequency:  48e6,
+	bits:       24,
+	prescalers: []int{1, 2, 4, 8, 16, 64, 256, 1024},
+	channelPins: [][]Pin{
+		{PA04, PA08, PB10, PA14, PB16, PA22, PB30}, // channel 0
+		{PA05, PA09, PB11, PA15, PB17, PA23, PB31}, // channel 1
+		{PA10, PB12, PA12, PA16, PA18, PA20},       // channel 2
+		{PA11, PB13, PA13, PA17, PA19, PA21},       // channel 3
+	},
+}
+
+var TCC1 = &timerType{
+	instance:   1,
+	frequency:  48e6,
+	bits:       24,
+	prescalers: []int{1, 2, 4, 8, 16, 64, 256, 1024},
+	channelPins: [][]Pin{
+		{PA06, PA10, PA30}, // channel 0
+		{PA07, PA11, PA31}, // channel 1
+		{PA08, PA24, PB30}, // channel 2
+		{PA09, PA25, PB31}, // channel 3
+	},
+}
+
+var TCC2 = &timerType{
+	instance:   2,
+	frequency:  48e6,
+	bits:       16,
+	prescalers: []int{1, 2, 4, 8, 16, 64, 256, 1024},
+	channelPins: [][]Pin{
+		{PA00, PA12, PA16}, // channel 0
+		{PA01, PA13, PA17}, // channel 1
+	},
+}

src/machine/machine_atsamd21_usb.go

@@ -19,6 +19,21 @@ const (
 
 	usb_DEVICE_PCKSIZE_MULTI_PACKET_SIZE_Pos  = 14
 	usb_DEVICE_PCKSIZE_MULTI_PACKET_SIZE_Mask = 0x3FFF
+
+	NumberOfUSBEndpoints = 8
+)
+
+var (
+	endPoints = []uint32{
+		usb.CONTROL_ENDPOINT:  usb.ENDPOINT_TYPE_CONTROL,
+		usb.CDC_ENDPOINT_ACM:  (usb.ENDPOINT_TYPE_INTERRUPT | usb.EndpointIn),
+		usb.CDC_ENDPOINT_OUT:  (usb.ENDPOINT_TYPE_BULK | usb.EndpointOut),
+		usb.CDC_ENDPOINT_IN:   (usb.ENDPOINT_TYPE_BULK | usb.EndpointIn),
+		usb.HID_ENDPOINT_IN:   (usb.ENDPOINT_TYPE_DISABLE), // Interrupt In
+		usb.HID_ENDPOINT_OUT:  (usb.ENDPOINT_TYPE_DISABLE), // Interrupt Out
+		usb.MIDI_ENDPOINT_IN:  (usb.ENDPOINT_TYPE_DISABLE), // Bulk In
+		usb.MIDI_ENDPOINT_OUT: (usb.ENDPOINT_TYPE_DISABLE), // Bulk Out
+	}
 )
 
 // Configure the USB peripheral. The config is here for compatibility with the UART interface.
@@ -179,10 +194,9 @@ func handleUSBIRQ(intr interrupt.Interrupt) {
 		setEPINTFLAG(i, epFlags)
 		if (epFlags & sam.USB_DEVICE_EPINTFLAG_TRCPT0) > 0 {
 			buf := handleEndpointRx(i)
-			if usbRxHandler[i] != nil {
+			if usbRxHandler[i] == nil || usbRxHandler[i](buf) {
-				usbRxHandler[i](buf)
+				AckUsbOutTransfer(i)
 			}
-			handleEndpointRxComplete(i)
 		} else if (epFlags & sam.USB_DEVICE_EPINTFLAG_TRCPT1) > 0 {
 			if usbTxHandler[i] != nil {
 				usbTxHandler[i]()
@@ -334,6 +348,8 @@ func SendUSBInPacket(ep uint32, data []byte) bool {
 	return true
 }
 
+// Prevent file size increases: https://github.com/tinygo-org/tinygo/pull/998
+//
 //go:noinline
 func sendUSBPacket(ep uint32, data []byte, maxsize uint16) {
 	l := uint16(len(data))
@@ -400,7 +416,8 @@ func handleEndpointRx(ep uint32) []byte {
 	return udd_ep_out_cache_buffer[ep][:count]
 }
 
-func handleEndpointRxComplete(ep uint32) {
+// AckUsbOutTransfer is called to acknowledge the completion of a USB OUT transfer.
+func AckUsbOutTransfer(ep uint32) {
 	// set byte count to zero
 	usbEndpointDescriptors[ep].DeviceDescBank[0].PCKSIZE.ClearBits(usb_DEVICE_PCKSIZE_BYTE_COUNT_Mask << usb_DEVICE_PCKSIZE_BYTE_COUNT_Pos)
 

src/machine/machine_atsamd51.go

@@ -2163,7 +2163,7 @@ func (f flashBlockDevice) ReadAt(p []byte, off int64) (n int, err error) {
 }
 
 // WriteAt writes the given number of bytes to the block device.
-// Only word (32 bits) length data can be programmed.
+// Data is written to the page buffer in 4-byte chunks, then saved to flash memory.
 // See SAM-D5x-E5x-Family-Data-Sheet-DS60001507.pdf page 591-592.
 // If the length of p is not long enough it will be padded with 0xFF bytes.
 // This method assumes that the destination is already erased.
@@ -2185,16 +2185,13 @@ func (f flashBlockDevice) WriteAt(p []byte, off int64) (n int, err error) {
 	waitWhileFlashBusy()
 
 	for j := 0; j < len(padded); j += int(f.WriteBlockSize()) {
-		// write first word using double-word low order word
+		// page buffer is 512 bytes long, but only 4 bytes can be written at once
-		*(*uint32)(unsafe.Pointer(address)) = binary.LittleEndian.Uint32(padded[j : j+int(f.WriteBlockSize()/2)])
+		for k := 0; k < int(f.WriteBlockSize()); k += 4 {
-
+			*(*uint32)(unsafe.Pointer(address + uintptr(k))) = binary.LittleEndian.Uint32(padded[j+k : j+k+4])
-		// write second word using double-word high order word
+		}
-		*(*uint32)(unsafe.Add(unsafe.Pointer(address), uintptr(f.WriteBlockSize())/2)) = binary.LittleEndian.Uint32(padded[j+int(f.WriteBlockSize()/2) : j+int(f.WriteBlockSize())])
-
-		waitWhileFlashBusy()
 
 		sam.NVMCTRL.SetADDR(uint32(address))
-		sam.NVMCTRL.CTRLB.Set(sam.NVMCTRL_CTRLB_CMD_WQW | (sam.NVMCTRL_CTRLB_CMDEX_KEY << sam.NVMCTRL_CTRLB_CMDEX_Pos))
+		sam.NVMCTRL.CTRLB.Set(sam.NVMCTRL_CTRLB_CMD_WP | (sam.NVMCTRL_CTRLB_CMDEX_KEY << sam.NVMCTRL_CTRLB_CMDEX_Pos))
 
 		waitWhileFlashBusy()
 
@@ -2213,7 +2210,7 @@ func (f flashBlockDevice) Size() int64 {
 	return int64(FlashDataEnd() - FlashDataStart())
 }
 
-const writeBlockSize = 8
+const writeBlockSize = 512
 
 // WriteBlockSize returns the block size in which data can be written to
 // memory. It can be used by a client to optimize writes, non-aligned writes
@@ -2357,6 +2354,5 @@ func (wd *watchdogImpl) Start() error {
 
 // Update the watchdog, indicating that `source` is healthy.
 func (wd *watchdogImpl) Update() {
-	// 0xA5 = magic value (see datasheet)
+	sam.WDT.CLEAR.Set(sam.WDT_CLEAR_CLEAR_KEY)
-	sam.WDT.CLEAR.Set(0xA5)
 }

src/machine/machine_atsamd51_usb.go

@@ -19,6 +19,21 @@ const (
 
 	usb_DEVICE_PCKSIZE_MULTI_PACKET_SIZE_Pos  = 14
 	usb_DEVICE_PCKSIZE_MULTI_PACKET_SIZE_Mask = 0x3FFF
+
+	NumberOfUSBEndpoints = 8
+)
+
+var (
+	endPoints = []uint32{
+		usb.CONTROL_ENDPOINT:  usb.ENDPOINT_TYPE_CONTROL,
+		usb.CDC_ENDPOINT_ACM:  (usb.ENDPOINT_TYPE_INTERRUPT | usb.EndpointIn),
+		usb.CDC_ENDPOINT_OUT:  (usb.ENDPOINT_TYPE_BULK | usb.EndpointOut),
+		usb.CDC_ENDPOINT_IN:   (usb.ENDPOINT_TYPE_BULK | usb.EndpointIn),
+		usb.HID_ENDPOINT_IN:   (usb.ENDPOINT_TYPE_DISABLE), // Interrupt In
+		usb.HID_ENDPOINT_OUT:  (usb.ENDPOINT_TYPE_DISABLE), // Interrupt Out
+		usb.MIDI_ENDPOINT_IN:  (usb.ENDPOINT_TYPE_DISABLE), // Bulk In
+		usb.MIDI_ENDPOINT_OUT: (usb.ENDPOINT_TYPE_DISABLE), // Bulk Out
+	}
 )
 
 // Configure the USB peripheral. The config is here for compatibility with the UART interface.
@@ -182,10 +197,9 @@ func handleUSBIRQ(intr interrupt.Interrupt) {
 		setEPINTFLAG(i, epFlags)
 		if (epFlags & sam.USB_DEVICE_ENDPOINT_EPINTFLAG_TRCPT0) > 0 {
 			buf := handleEndpointRx(i)
-			if usbRxHandler[i] != nil {
+			if usbRxHandler[i] == nil || usbRxHandler[i](buf) {
-				usbRxHandler[i](buf)
+				AckUsbOutTransfer(i)
 			}
-			handleEndpointRxComplete(i)
 		} else if (epFlags & sam.USB_DEVICE_ENDPOINT_EPINTFLAG_TRCPT1) > 0 {
 			if usbTxHandler[i] != nil {
 				usbTxHandler[i]()
@@ -337,6 +351,8 @@ func SendUSBInPacket(ep uint32, data []byte) bool {
 	return true
 }
 
+// Prevent file size increases: https://github.com/tinygo-org/tinygo/pull/998
+//
 //go:noinline
 func sendUSBPacket(ep uint32, data []byte, maxsize uint16) {
 	l := uint16(len(data))
@@ -403,7 +419,8 @@ func handleEndpointRx(ep uint32) []byte {
 	return udd_ep_out_cache_buffer[ep][:count]
 }
 
-func handleEndpointRxComplete(ep uint32) {
+// AckUsbOutTransfer is called to acknowledge the completion of a USB OUT transfer.
+func AckUsbOutTransfer(ep uint32) {
 	// set byte count to zero
 	usbEndpointDescriptors[ep].DeviceDescBank[0].PCKSIZE.ClearBits(usb_DEVICE_PCKSIZE_BYTE_COUNT_Mask << usb_DEVICE_PCKSIZE_BYTE_COUNT_Pos)
 

src/machine/machine_generic.go

@@ -48,6 +48,103 @@ func gpioSet(pin Pin, value bool)
 //export __tinygo_gpio_get
 func gpioGet(pin Pin) bool
 
+// Generic PWM/timer peripheral. Properties can be configured depending on the
+// hardware.
+type timerType struct {
+	// Static properties.
+	instance    int32
+	frequency   uint64
+	bits        int
+	prescalers  []int
+	channelPins [][]Pin
+
+	// Configured 'top' value.
+	top uint32
+}
+
+// Configure the PWM/timer peripheral.
+func (t *timerType) Configure(config PWMConfig) error {
+	// Note: for very large period values, this multiplication will overflow.
+	top := config.Period * t.frequency / 1e9
+	if config.Period == 0 {
+		top = 0xffff // default for LEDs
+	}
+
+	// The maximum value that can be stored with the given number of bits in
+	// this timer.
+	maxTop := uint64(1)<<uint64(t.bits) - 1
+
+	// Look for an appropriate prescaler value.
+	var prescaler int
+	for _, div := range t.prescalers {
+		if top/uint64(div) <= maxTop {
+			prescaler = div
+			top = top / uint64(div)
+			break
+		}
+	}
+	if prescaler == 0 {
+		return ErrPWMPeriodTooLong
+	}
+
+	// Set these values as the configuration.
+	t.top = uint32(top)
+	pwmConfigure(t.instance, float64(t.frequency)/float64(prescaler), uint32(top))
+
+	return nil
+}
+
+// Channel returns a PWM channel for the given pin. Note that one channel may be
+// shared between multiple pins, and so will have the same duty cycle. If this
+// is not desirable, look for a different PWM/timer peripheral or consider using
+// a different pin.
+func (t *timerType) Channel(pin Pin) (uint8, error) {
+	for ch, pins := range t.channelPins {
+		// For nrf52xxx chips specifically we can assign any channel to any pin.
+		// We use a similar (identical?) logic to the hardware implementation,
+		// and pick the first empty channel.
+		if pins == nil {
+			t.channelPins[ch] = []Pin{pin}
+			pwmChannelConfigure(t.instance, int32(ch), pin)
+			return uint8(ch), nil
+		}
+
+		// Check whether the pin can be used on this channel.
+		for _, p := range pins {
+			if p == pin {
+				pwmChannelConfigure(t.instance, int32(ch), pin)
+				return uint8(ch), nil
+			}
+		}
+	}
+
+	return 0, ErrInvalidOutputPin
+}
+
+func (t *timerType) Set(channel uint8, value uint32) {
+	pwmChannelSet(t.instance, channel, value)
+}
+
+// Top returns the current counter top, for use in duty cycle calculation. It
+// will only change with a call to Configure or SetPeriod, otherwise it is
+// constant.
+//
+// The value returned here is hardware dependent. In general, it's best to treat
+// it as an opaque value that can be divided by some number and passed to Set
+// (see Set documentation for more information).
+func (t *timerType) Top() uint32 {
+	return t.top
+}
+
+//export __tinygo_pwm_configure
+func pwmConfigure(instance int32, frequency float64, top uint32)
+
+//export __tinygo_pwm_channel_configure
+func pwmChannelConfigure(instance, channel int32, pin Pin)
+
+//export __tinygo_pwm_channel_set
+func pwmChannelSet(instance int32, channel uint8, value uint32)
+
 type SPI struct {
 	Bus uint8
 }

src/machine/machine_nrf52840_simulator.go

@@ -0,0 +1,60 @@
+//go:build !baremetal && (bluemicro840 || circuitplay_bluefruit || clue_alpha || feather_nrf52840_sense || feather_nrf52840 || itsybitsy_nrf52840 || mdbt50qrx || nano_33_ble || nicenano || nrf52840_mdk || particle_3rd_gen || pca10056 || pca10059 || rak4631 || reelboard || xiao_ble)
+
+// Simulator support for nrf52840 based boards.
+
+package machine
+
+// Channel values below are nil, so that they get filled in on the first use.
+// This is the same as what happens on baremetal.
+
+var PWM0 = &timerType{
+	instance:   0,
+	frequency:  16e6,
+	bits:       15,
+	prescalers: []int{1, 2, 4, 8, 16, 32, 64, 128},
+	channelPins: [][]Pin{
+		nil, // channel 0
+		nil, // channel 1
+		nil, // channel 2
+		nil, // channel 3
+	},
+}
+
+var PWM1 = &timerType{
+	instance:   1,
+	frequency:  16e6,
+	bits:       15,
+	prescalers: []int{1, 2, 4, 8, 16, 32, 64, 128},
+	channelPins: [][]Pin{
+		nil, // channel 0
+		nil, // channel 1
+		nil, // channel 2
+		nil, // channel 3
+	},
+}
+
+var PWM2 = &timerType{
+	instance:   2,
+	frequency:  16e6,
+	bits:       15,
+	prescalers: []int{1, 2, 4, 8, 16, 32, 64, 128},
+	channelPins: [][]Pin{
+		nil, // channel 0
+		nil, // channel 1
+		nil, // channel 2
+		nil, // channel 3
+	},
+}
+
+var PWM3 = &timerType{
+	instance:   3,
+	frequency:  16e6,
+	bits:       15,
+	prescalers: []int{1, 2, 4, 8, 16, 32, 64, 128},
+	channelPins: [][]Pin{
+		nil, // channel 0
+		nil, // channel 1
+		nil, // channel 2
+		nil, // channel 3
+	},
+}

src/machine/machine_nrf52840_usb.go

@@ -11,6 +11,8 @@ import (
 	"unsafe"
 )
 
+const NumberOfUSBEndpoints = 8
+
 var (
 	sendOnEP0DATADONE struct {
 		ptr    *byte
@@ -20,6 +22,17 @@ var (
 	epinen      uint32
 	epouten     uint32
 	easyDMABusy volatile.Register8
+
+	endPoints = []uint32{
+		usb.CONTROL_ENDPOINT:  usb.ENDPOINT_TYPE_CONTROL,
+		usb.CDC_ENDPOINT_ACM:  (usb.ENDPOINT_TYPE_INTERRUPT | usb.EndpointIn),
+		usb.CDC_ENDPOINT_OUT:  (usb.ENDPOINT_TYPE_BULK | usb.EndpointOut),
+		usb.CDC_ENDPOINT_IN:   (usb.ENDPOINT_TYPE_BULK | usb.EndpointIn),
+		usb.HID_ENDPOINT_IN:   (usb.ENDPOINT_TYPE_DISABLE), // Interrupt In
+		usb.HID_ENDPOINT_OUT:  (usb.ENDPOINT_TYPE_DISABLE), // Interrupt Out
+		usb.MIDI_ENDPOINT_IN:  (usb.ENDPOINT_TYPE_DISABLE), // Bulk In
+		usb.MIDI_ENDPOINT_OUT: (usb.ENDPOINT_TYPE_DISABLE), // Bulk Out
+	}
 )
 
 // enterCriticalSection is used to protect access to easyDMA - only one thing
@@ -193,10 +206,9 @@ func handleUSBIRQ(interrupt.Interrupt) {
 		if nrf.USBD.EVENTS_ENDEPOUT[i].Get() > 0 {
 			nrf.USBD.EVENTS_ENDEPOUT[i].Set(0)
 			buf := handleEndpointRx(uint32(i))
-			if usbRxHandler[i] != nil {
+			if usbRxHandler[i] == nil || usbRxHandler[i](buf) {
-				usbRxHandler[i](buf)
+				AckUsbOutTransfer(uint32(i))
 			}
-			handleEndpointRxComplete(uint32(i))
 			exitCriticalSection()
 		}
 	}
@@ -252,6 +264,8 @@ func SendUSBInPacket(ep uint32, data []byte) bool {
 	return true
 }
 
+// Prevent file size increases: https://github.com/tinygo-org/tinygo/pull/998
+//
 //go:noinline
 func sendUSBPacket(ep uint32, data []byte, maxsize uint16) {
 	count := len(data)
@@ -289,7 +303,8 @@ func handleEndpointRx(ep uint32) []byte {
 	return udd_ep_out_cache_buffer[ep][:count]
 }
 
-func handleEndpointRxComplete(ep uint32) {
+// AckUsbOutTransfer is called to acknowledge the completion of a USB OUT transfer.
+func AckUsbOutTransfer(ep uint32) {
 	// set ready for next data
 	nrf.USBD.SIZE.EPOUT[ep].Set(0)
 }

src/machine/machine_rp2040_simulator.go

@@ -0,0 +1,96 @@
+//go:build !baremetal && (ae_rp2040 || badger2040_w || badger2040 || challenger_rp2040 || elecrow_rp2040 || feather_rp2040 || gopher_badge || kb2040 || macropad_rp2040 || nano_rp2040 || pico || qtpy_rp2040 || thingplus_rp2040 || thumby || trinkey_qt2040 || tufty2040 || waveshare_rp2040_tiny || waveshare_rp2040_zero || xiao_rp2040)
+
+// Simulator support for the RP2040.
+//
+// This is *only* for the RP2040. RP2350 is a different chip with slightly
+// different characteristics.
+
+package machine
+
+var PWM0 = &timerType{
+	instance:   0,
+	frequency:  200e6,
+	bits:       16,
+	prescalers: []int{1, 2, 4, 8, 16, 32, 64, 128, 256}, // actually a continuing range, TODO
+	channelPins: [][]Pin{
+		{GPIO0, GPIO16}, // channel A (0)
+		{GPIO1, GPIO17}, // channel B (1)
+	},
+}
+
+var PWM1 = &timerType{
+	instance:   0,
+	frequency:  200e6,
+	bits:       16,
+	prescalers: []int{1, 2, 4, 8, 16, 32, 64, 128, 256},
+	channelPins: [][]Pin{
+		{GPIO2, GPIO18}, // channel A (0)
+		{GPIO3, GPIO19}, // channel B (1)
+	},
+}
+
+var PWM2 = &timerType{
+	instance:   0,
+	frequency:  200e6,
+	bits:       16,
+	prescalers: []int{1, 2, 4, 8, 16, 32, 64, 128, 256},
+	channelPins: [][]Pin{
+		{GPIO4, GPIO20}, // channel A (0)
+		{GPIO5, GPIO21}, // channel B (1)
+	},
+}
+
+var PWM3 = &timerType{
+	instance:   0,
+	frequency:  200e6,
+	bits:       16,
+	prescalers: []int{1, 2, 4, 8, 16, 32, 64, 128, 256},
+	channelPins: [][]Pin{
+		{GPIO6, GPIO22}, // channel A (0)
+		{GPIO7, GPIO23}, // channel B (1)
+	},
+}
+
+var PWM4 = &timerType{
+	instance:   0,
+	frequency:  200e6,
+	bits:       16,
+	prescalers: []int{1, 2, 4, 8, 16, 32, 64, 128, 256},
+	channelPins: [][]Pin{
+		{GPIO8, GPIO24}, // channel A (0)
+		{GPIO9, GPIO25}, // channel B (1)
+	},
+}
+
+var PWM5 = &timerType{
+	instance:   0,
+	frequency:  200e6,
+	bits:       16,
+	prescalers: []int{1, 2, 4, 8, 16, 32, 64, 128, 256},
+	channelPins: [][]Pin{
+		{GPIO10, GPIO26}, // channel A (0)
+		{GPIO11, GPIO27}, // channel B (1)
+	},
+}
+
+var PWM6 = &timerType{
+	instance:   0,
+	frequency:  200e6,
+	bits:       16,
+	prescalers: []int{1, 2, 4, 8, 16, 32, 64, 128, 256},
+	channelPins: [][]Pin{
+		{GPIO12, GPIO28}, // channel A (0)
+		{GPIO13, GPIO29}, // channel B (1)
+	},
+}
+
+var PWM7 = &timerType{
+	instance:   0,
+	frequency:  200e6,
+	bits:       16,
+	prescalers: []int{1, 2, 4, 8, 16, 32, 64, 128, 256},
+	channelPins: [][]Pin{
+		{GPIO14}, // channel A (0)
+		{GPIO15}, // channel B (1)
+	},
+}

src/machine/machine_rp2040_usb.go

@@ -6,16 +6,6 @@ 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.
@@ -77,7 +67,7 @@ func handleUSBIRQ(intr interrupt.Interrupt) {
 
 		if !ok {
 			// Stall endpoint?
-			sendStallViaEPIn(0)
+			USBDev.SetStallEPIn(0)
 		}
 
 	}
@@ -105,10 +95,9 @@ func handleUSBIRQ(intr interrupt.Interrupt) {
 		for i := 0; i < 16; i++ {
 			if s2&(1<<(i*2+1)) > 0 {
 				buf := handleEndpointRx(uint32(i))
-				if usbRxHandler[i] != nil {
+				if usbRxHandler[i] == nil || usbRxHandler[i](buf) {
-					usbRxHandler[i](buf)
+					AckUsbOutTransfer(uint32(i))
 				}
-				handleEndpointRxComplete(uint32(i))
 			}
 		}
 
@@ -134,40 +123,6 @@ func handleUSBIRQ(intr interrupt.Interrupt) {
 	}
 }
 
-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 {
 	// Using 570μs timeout which is exactly the same as SAMD21.
 	const ackTimeout = 570
@@ -188,191 +143,6 @@ func handleUSBSetAddress(setup usb.Setup) bool {
 	return true
 }
 
-// SendUSBInPacket sends a packet for USB (interrupt in / bulk in).
+func armEPZeroStall() {
-func SendUSBInPacket(ep uint32, data []byte) bool {
+	rp.USBCTRL_REGS.EP_STALL_ARM.Set(rp.USBCTRL_REGS_EP_STALL_ARM_EP0_IN)
-	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.USBCTRL_REGS.EP_STALL_ARM.Set(rp.USBCTRL_REGS_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
-)