feat(userland): ported the CPython 3.12 interpreter

.github/workflows/kernel-unit-tests.yml

@@ -109,6 +109,19 @@ jobs:
         if: steps.libcxx-cache.outputs.cache-hit != 'true'
         run: make libcxx
 
+      - name: Cache compiler-rt builtins
+        uses: actions/cache@v4
+        id: compiler-rt-cache
+        with:
+          path: |
+            userland/sysroot/x86_64/lib/libclang_rt.builtins-x86_64.a
+            userland/sysroot/aarch64/lib/libclang_rt.builtins-aarch64.a
+          key: compiler-rt-builtins-18.1.8
+
+      - name: Build compiler-rt builtins
+        if: steps.compiler-rt-cache.outputs.cache-hit != 'true'
+        run: make compiler-rt
+
       - name: Build kernel
         run: |
           make kernel ARCH=${{ matrix.arch }} \
@@ -191,6 +204,19 @@ jobs:
         if: steps.libcxx-cache.outputs.cache-hit != 'true'
         run: make libcxx
 
+      - name: Cache compiler-rt builtins
+        uses: actions/cache@v4
+        id: compiler-rt-cache
+        with:
+          path: |
+            userland/sysroot/x86_64/lib/libclang_rt.builtins-x86_64.a
+            userland/sysroot/aarch64/lib/libclang_rt.builtins-aarch64.a
+          key: compiler-rt-builtins-18.1.8
+
+      - name: Build compiler-rt builtins
+        if: steps.compiler-rt-cache.outputs.cache-hit != 'true'
+        run: make compiler-rt
+
       - name: Run unit tests
         run: make test ARCH=${{ matrix.arch }} 2>&1 | tee test-log-${{ matrix.arch }}.txt
 

.gitignore

@@ -32,6 +32,7 @@ userland/musl-*
 userland/llvm-project-*
 userland/build/
 initrd/bin/*
+initrd/usr/
 !initrd/bin/.gitkeep
 
 # OS files

Makefile

@@ -648,6 +648,80 @@ libcxx:
 	@echo ""
 	@echo "libc++ $(LLVM_VERSION) ready for both architectures."
 
+# Build the compiler-rt builtins runtime for both supported architectures and
+# install it into the per-arch sysroot as a flat archive
+# (libclang_rt.builtins-<arch>.a alongside libc.a, libstlx.a, etc.). This frees
+# downstream apps (notably the static CPython link) from depending on the host
+# toolchain shipping cross-arch builtins, which Ubuntu's libclang-rt-dev does
+# not provide for aarch64.
+compiler-rt:
+	@echo "Building compiler-rt builtins $(LLVM_VERSION) for x86_64 and aarch64..."
+	@test -f userland/sysroot/x86_64/lib/libc.a || \
+		(echo "ERROR: musl sysroot not found. Run 'make musl' first." && exit 1)
+	@test -f userland/sysroot/aarch64/lib/libc.a || \
+		(echo "ERROR: musl sysroot not found. Run 'make musl' first." && exit 1)
+	@mkdir -p userland
+	@test -f $(LLVM_TARBALL) || \
+		(echo "Downloading LLVM $(LLVM_VERSION) source (~130MB)..." && \
+		 curl -L -o $(LLVM_TARBALL) $(LLVM_URL))
+	@test -d $(LLVM_DIR) || \
+		(echo "Extracting..." && \
+		 cd userland && tar xf llvm-project-$(LLVM_VERSION).src.tar.xz)
+	@echo ""
+	@echo "Building compiler-rt builtins for x86_64..."
+	@mkdir -p $(LLVM_DIR)/build-compiler-rt-x86_64
+	cd $(LLVM_DIR)/build-compiler-rt-x86_64 && cmake -G "Unix Makefiles" ../compiler-rt \
+		-DCMAKE_C_COMPILER=clang \
+		-DCMAKE_ASM_COMPILER=clang \
+		-DCMAKE_C_COMPILER_TARGET=x86_64-linux-musl \
+		-DCMAKE_ASM_COMPILER_TARGET=x86_64-linux-musl \
+		-DCMAKE_SYSROOT=$(abspath userland/sysroot/x86_64) \
+		-DCMAKE_TRY_COMPILE_TARGET_TYPE=STATIC_LIBRARY \
+		-DCOMPILER_RT_DEFAULT_TARGET_ONLY=ON \
+		-DCOMPILER_RT_BUILD_BUILTINS=ON \
+		-DCOMPILER_RT_BUILD_SANITIZERS=OFF \
+		-DCOMPILER_RT_BUILD_XRAY=OFF \
+		-DCOMPILER_RT_BUILD_LIBFUZZER=OFF \
+		-DCOMPILER_RT_BUILD_PROFILE=OFF \
+		-DCOMPILER_RT_BUILD_MEMPROF=OFF \
+		-DCOMPILER_RT_BUILD_ORC=OFF \
+		-DCOMPILER_RT_INCLUDE_TESTS=OFF \
+		'-DCMAKE_C_FLAGS=--target=x86_64-linux-musl --sysroot=$(abspath userland/sysroot/x86_64) -nostdlibinc -isystem $(abspath userland/sysroot/x86_64)/include' \
+		'-DCMAKE_ASM_FLAGS=--target=x86_64-linux-musl --sysroot=$(abspath userland/sysroot/x86_64) -nostdlibinc -isystem $(abspath userland/sysroot/x86_64)/include' \
+		> /dev/null
+	$(MAKE) -C $(LLVM_DIR)/build-compiler-rt-x86_64 -j$$(nproc) builtins > /dev/null
+	@cp "$$(find $(LLVM_DIR)/build-compiler-rt-x86_64 -name 'libclang_rt.builtins*.a' -print -quit)" \
+		userland/sysroot/x86_64/lib/libclang_rt.builtins-x86_64.a
+	@echo "compiler-rt builtins x86_64 installed to userland/sysroot/x86_64/lib/"
+	@echo ""
+	@echo "Building compiler-rt builtins for aarch64..."
+	@mkdir -p $(LLVM_DIR)/build-compiler-rt-aarch64
+	cd $(LLVM_DIR)/build-compiler-rt-aarch64 && cmake -G "Unix Makefiles" ../compiler-rt \
+		-DCMAKE_C_COMPILER=clang \
+		-DCMAKE_ASM_COMPILER=clang \
+		-DCMAKE_C_COMPILER_TARGET=aarch64-linux-musl \
+		-DCMAKE_ASM_COMPILER_TARGET=aarch64-linux-musl \
+		-DCMAKE_SYSROOT=$(abspath userland/sysroot/aarch64) \
+		-DCMAKE_TRY_COMPILE_TARGET_TYPE=STATIC_LIBRARY \
+		-DCOMPILER_RT_DEFAULT_TARGET_ONLY=ON \
+		-DCOMPILER_RT_BUILD_BUILTINS=ON \
+		-DCOMPILER_RT_BUILD_SANITIZERS=OFF \
+		-DCOMPILER_RT_BUILD_XRAY=OFF \
+		-DCOMPILER_RT_BUILD_LIBFUZZER=OFF \
+		-DCOMPILER_RT_BUILD_PROFILE=OFF \
+		-DCOMPILER_RT_BUILD_MEMPROF=OFF \
+		-DCOMPILER_RT_BUILD_ORC=OFF \
+		-DCOMPILER_RT_INCLUDE_TESTS=OFF \
+		'-DCMAKE_C_FLAGS=--target=aarch64-linux-musl --sysroot=$(abspath userland/sysroot/aarch64) -nostdlibinc -isystem $(abspath userland/sysroot/aarch64)/include' \
+		'-DCMAKE_ASM_FLAGS=--target=aarch64-linux-musl --sysroot=$(abspath userland/sysroot/aarch64) -nostdlibinc -isystem $(abspath userland/sysroot/aarch64)/include' \
+		> /dev/null
+	$(MAKE) -C $(LLVM_DIR)/build-compiler-rt-aarch64 -j$$(nproc) builtins > /dev/null
+	@cp "$$(find $(LLVM_DIR)/build-compiler-rt-aarch64 -name 'libclang_rt.builtins*.a' -print -quit)" \
+		userland/sysroot/aarch64/lib/libclang_rt.builtins-aarch64.a
+	@echo "compiler-rt builtins aarch64 installed to userland/sysroot/aarch64/lib/"
+	@echo ""
+	@echo "compiler-rt builtins $(LLVM_VERSION) ready for both architectures."
+
 limine:
 	@echo "Downloading Limine ($(LIMINE_BRANCH))..."
 	@mkdir -p $(BOOT_DIR)
@@ -728,24 +802,34 @@ toolchain-check:
 	@printf "%-24s" "libc++ (aarch64):" && \
 		(test -f userland/sysroot/aarch64/lib/libc++.a && echo "OK" || echo "NOT FOUND - run 'make libcxx'")
 	@printf "%-24s" "builtins (x86_64):" && \
-		(BRT=$$(clang --target=x86_64-linux-musl --rtlib=compiler-rt -print-libgcc-file-name 2>/dev/null); \
-		 if [ -f "$$BRT" ]; then \
-			echo "$$BRT"; \
-		 elif which x86_64-linux-gnu-gcc > /dev/null 2>&1; then \
-			BGCC=$$(x86_64-linux-gnu-gcc -print-libgcc-file-name 2>/dev/null); \
-			[ -f "$$BGCC" ] && echo "$$BGCC" || echo "NOT FOUND"; \
+		(SR=userland/sysroot/x86_64/lib/libclang_rt.builtins-x86_64.a; \
+		 if [ -f "$$SR" ]; then \
+			echo "$$SR"; \
 		 else \
-			echo "NOT FOUND"; \
+			BRT=$$(clang --target=x86_64-linux-musl --rtlib=compiler-rt -print-libgcc-file-name 2>/dev/null); \
+			if [ -f "$$BRT" ]; then \
+				echo "$$BRT"; \
+			elif which x86_64-linux-gnu-gcc > /dev/null 2>&1; then \
+				BGCC=$$(x86_64-linux-gnu-gcc -print-libgcc-file-name 2>/dev/null); \
+				[ -f "$$BGCC" ] && echo "$$BGCC" || echo "NOT FOUND - run 'make compiler-rt'"; \
+			else \
+				echo "NOT FOUND - run 'make compiler-rt'"; \
+			fi; \
 		 fi)
 	@printf "%-24s" "builtins (aarch64):" && \
-		(BRT=$$(clang --target=aarch64-linux-musl --rtlib=compiler-rt -print-libgcc-file-name 2>/dev/null); \
-		 if [ -f "$$BRT" ]; then \
-			echo "$$BRT"; \
-		 elif which aarch64-linux-gnu-gcc > /dev/null 2>&1; then \
-			BGCC=$$(aarch64-linux-gnu-gcc -print-libgcc-file-name 2>/dev/null); \
-			[ -f "$$BGCC" ] && echo "$$BGCC" || echo "NOT FOUND"; \
+		(SR=userland/sysroot/aarch64/lib/libclang_rt.builtins-aarch64.a; \
+		 if [ -f "$$SR" ]; then \
+			echo "$$SR"; \
 		 else \
-			echo "NOT FOUND"; \
+			BRT=$$(clang --target=aarch64-linux-musl --rtlib=compiler-rt -print-libgcc-file-name 2>/dev/null); \
+			if [ -f "$$BRT" ]; then \
+				echo "$$BRT"; \
+			elif which aarch64-linux-gnu-gcc > /dev/null 2>&1; then \
+				BGCC=$$(aarch64-linux-gnu-gcc -print-libgcc-file-name 2>/dev/null); \
+				[ -f "$$BGCC" ] && echo "$$BGCC" || echo "NOT FOUND - run 'make compiler-rt'"; \
+			else \
+				echo "NOT FOUND - run 'make compiler-rt'"; \
+			fi; \
 		 fi)
 	@echo ""
 	@echo "If anything is NOT FOUND, run 'make deps', 'make limine', 'make musl', 'make libcxx', and/or 'make rpi4-firmware'"
@@ -762,6 +846,7 @@ help:
 	@echo "  make limine                  Download Limine bootloader"
 	@echo "  make musl                    Build musl libc for both architectures"
 	@echo "  make libcxx                  Build libc++ for C++ userland support"
+	@echo "  make compiler-rt             Build compiler-rt builtins (required for cross-arch static linking)"
 	@echo "  make rpi4-firmware           Download RPi4 UEFI firmware"
 	@echo "  make doom-wad                Download DOOM1.WAD shareware"
 	@echo "  make toolchain-check         Verify tools are installed"
@@ -796,6 +881,7 @@ help:
 	@echo "  2. make limine"
 	@echo "  3. make musl"
 	@echo "  4. make libcxx"
+	@echo "  5. make compiler-rt"
 	@echo ""
 	@echo "Examples:"
 	@echo "  make kernel ARCH=x86_64"

kernel/fs/fstypes.h

@@ -26,6 +26,7 @@ constexpr uint32_t O_EXCL   = 0x80;
 constexpr uint32_t O_TRUNC  = 0x200;
 constexpr uint32_t O_APPEND   = 0x400;
 constexpr uint32_t O_NONBLOCK = 0x800;
+constexpr uint32_t O_CLOEXEC  = 0x80000;
 
 constexpr uint32_t ACCESS_MODE_MASK = 0x3;
 

kernel/mm/vma.h

@@ -40,7 +40,7 @@ constexpr uint32_t VMA_FLAG_DEVICE    = (1u << 5);
 
 constexpr uintptr_t MMAP_BASE_DEFAULT = 0x00000080000000ULL;
 constexpr uintptr_t USER_STACK_TOP    = 0x00007FFFFFF00000ULL;
-constexpr size_t    USER_STACK_PAGES  = 8; // 32 KiB
+constexpr size_t    USER_STACK_PAGES  = 128; // 512 KiB
 constexpr size_t    USER_STACK_GUARD_PAGES = 1;
 
 struct vma {

kernel/pty/pty.cpp

@@ -3,12 +3,40 @@
 #include "common/ring_buffer.h"
 #include "fs/fstypes.h"
 #include "mm/heap.h"
+#include "mm/uaccess.h"
 #include "dynpriv/dynpriv.h"
 #include "sync/poll.h"
 #include "sync/wait_queue.h"
+#include "terminal/terminal.h"
 
 namespace pty {
 
+constexpr uint32_t TCGETS      = 0x5401;
+constexpr uint32_t TCSETS      = 0x5402;
+constexpr uint32_t TCSETSW     = 0x5403;
+constexpr uint32_t TCSETSF     = 0x5404;
+constexpr uint32_t TIOCGWINSZ  = 0x5413;
+constexpr uint32_t TIOCSWINSZ  = 0x5414;
+
+constexpr uint32_t LINUX_ECHO   = 0x0008;
+constexpr uint32_t LINUX_ICANON = 0x0002;
+
+struct linux_termios {
+    uint32_t c_iflag;
+    uint32_t c_oflag;
+    uint32_t c_cflag;
+    uint32_t c_lflag;
+    uint8_t  c_line;
+    uint8_t  c_cc[19];
+};
+
+struct linux_winsize {
+    uint16_t ws_row;
+    uint16_t ws_col;
+    uint16_t ws_xpixel;
+    uint16_t ws_ypixel;
+};
+
 __PRIVILEGED_BSS static uint32_t g_next_pty_id;
 
 __PRIVILEGED_CODE void pty_channel::ref_destroy(pty_channel* self) {
@@ -179,15 +207,66 @@ static void pty_slave_close(resource::resource_object* obj) {
     obj->impl = nullptr;
 }
 
+static int32_t do_tcgets(pty_channel* chan, uint64_t arg) {
+    linux_termios t = {};
+
+    if (chan->m_ld.mode != terminal::LD_MODE_RAW) {
+        t.c_lflag = LINUX_ICANON | LINUX_ECHO;
+    }
+
+    int32_t rc = mm::uaccess::copy_to_user(
+        reinterpret_cast<void*>(arg), &t, sizeof(t));
+
+    return (rc == mm::uaccess::OK) ? resource::OK : resource::ERR_INVAL;
+}
+
+static int32_t do_tcsets(pty_channel* chan, uint64_t arg) {
+    linux_termios t = {};
+    int32_t rc = mm::uaccess::copy_from_user(
+        &t, reinterpret_cast<const void*>(arg), sizeof(t));
+
+    if (rc != mm::uaccess::OK) {
+        return resource::ERR_INVAL;
+    }
+
+    uint32_t mode = (t.c_lflag & LINUX_ICANON)
+        ? terminal::STLX_TCSETS_COOKED
+        : terminal::STLX_TCSETS_RAW;
+
+    terminal::ld_set_mode(&chan->m_ld, mode);
+    return resource::OK;
+}
+
+static int32_t do_tiocgwinsz(uint64_t arg) {
+    linux_winsize w = { 24, 80, 0, 0 };
+    int32_t rc = mm::uaccess::copy_to_user(
+        reinterpret_cast<void*>(arg), &w, sizeof(w));
+
+    return (rc == mm::uaccess::OK) ? resource::OK : resource::ERR_INVAL;
+}
+
+static int32_t pty_termios_ioctl(pty_channel* chan, uint32_t cmd, uint64_t arg) {
+    switch (cmd) {
+        case TCGETS:                       return do_tcgets(chan, arg);
+        case TCSETS:
+        case TCSETSW:
+        case TCSETSF:                      return do_tcsets(chan, arg);
+        case TIOCGWINSZ:                   return do_tiocgwinsz(arg);
+        case TIOCSWINSZ:                   return resource::OK;
+        case terminal::STLX_TCSETS_RAW:
+        case terminal::STLX_TCSETS_COOKED: return terminal::ld_set_mode(&chan->m_ld, cmd);
+        default:                           return resource::ERR_INVAL;
+    }
+}
+
 static int32_t pty_slave_ioctl(
     resource::resource_object* obj, uint32_t cmd, uint64_t arg
 ) {
-    (void)arg;
     if (!obj || !obj->impl) {
         return resource::ERR_INVAL;
     }
     auto* ep = static_cast<pty_endpoint*>(obj->impl);
-    return terminal::ld_set_mode(&ep->channel->m_ld, cmd);
+    return pty_termios_ioctl(ep->channel.ptr(), cmd, arg);
 }
 
 static uint32_t pty_master_poll(

kernel/resource/handle_table.h

@@ -10,6 +10,9 @@ struct resource_object;
 
 constexpr uint32_t MAX_TASK_HANDLES = 128;
 
+// Private high bit in handle_entry::flags reserved for FD_CLOEXEC state
+constexpr uint32_t RESOURCE_HANDLE_CLOEXEC = 0x80000000;
+
 struct handle_entry {
     bool used;
     uint16_t generation;

kernel/resource/resource.cpp

@@ -113,6 +113,13 @@ __PRIVILEGED_CODE int32_t open(
         return (rc == HANDLE_ERR_NOSPC) ? ERR_TABLEFULL : ERR_IO;
     }
 
+    if (flags & fs::O_CLOEXEC) {
+        uint32_t handle_flags = 0;
+        get_handle_flags(&owner->handles, *out_handle, &handle_flags);
+        set_handle_flags(&owner->handles, *out_handle,
+                         handle_flags | RESOURCE_HANDLE_CLOEXEC);
+    }
+
     // Table now owns one reference.
     resource_release(obj);
     return OK;