Merge pull request #963 from zig-lang/atomic-stack-and-queue

Atomic stack and queue

CMakeLists.txt

@@ -415,6 +415,9 @@ set(ZIG_CPP_SOURCES
 
 set(ZIG_STD_FILES
     "array_list.zig"
+    "atomic/index.zig"
+    "atomic/stack.zig"
+    "atomic/queue.zig"
     "base64.zig"
     "buf_map.zig"
     "buf_set.zig"

src/ir.cpp

@@ -18184,6 +18184,11 @@ static TypeTableEntry *ir_analyze_instruction_atomic_rmw(IrAnalyze *ira, IrInstr
     } else {
         if (!ir_resolve_atomic_order(ira, instruction->ordering->other, &ordering))
             return ira->codegen->builtin_types.entry_invalid;
+        if (ordering == AtomicOrderUnordered) {
+            ir_add_error(ira, instruction->ordering,
+                buf_sprintf("@atomicRmw atomic ordering must not be Unordered"));
+            return ira->codegen->builtin_types.entry_invalid;
+        }
     }
 
     if (instr_is_comptime(casted_operand) && instr_is_comptime(casted_ptr) && casted_ptr->value.data.x_ptr.mut == ConstPtrMutComptimeVar)

std/atomic/index.zig

@@ -0,0 +1,7 @@
+pub const Stack = @import("stack.zig").Stack;
+pub const Queue = @import("queue.zig").Queue;
+
+test "std.atomic" {
+    _ = @import("stack.zig").Stack;
+    _ = @import("queue.zig").Queue;
+}

std/atomic/queue.zig

@@ -0,0 +1,120 @@
+const builtin = @import("builtin");
+const AtomicOrder = builtin.AtomicOrder;
+const AtomicRmwOp = builtin.AtomicRmwOp;
+
+/// Many reader, many writer, non-allocating, thread-safe, lock-free
+pub fn Queue(comptime T: type) type {
+    return struct {
+        head: &Node,
+        tail: &Node,
+        root: Node,
+
+        pub const Self = this;
+
+        pub const Node = struct {
+            next: ?&Node,
+            data: T,
+        };
+
+        // TODO: well defined copy elision: https://github.com/zig-lang/zig/issues/287
+        pub fn init(self: &Self) void {
+            self.root.next = null;
+            self.head = &self.root;
+            self.tail = &self.root;
+        }
+
+        pub fn put(self: &Self, node: &Node) void {
+            node.next = null;
+
+            const tail = @atomicRmw(&Node, &self.tail, AtomicRmwOp.Xchg, node, AtomicOrder.SeqCst);
+            _ = @atomicRmw(?&Node, &tail.next, AtomicRmwOp.Xchg, node, AtomicOrder.SeqCst);
+        }
+
+        pub fn get(self: &Self) ?&Node {
+            var head = @atomicLoad(&Node, &self.head, AtomicOrder.Acquire);
+            while (true) {
+                const node = head.next ?? return null;
+                head = @cmpxchgWeak(&Node, &self.head, head, node, AtomicOrder.Release, AtomicOrder.Acquire) ?? return node;
+            }
+        }
+    };
+}
+
+const std = @import("std");
+const Context = struct {
+    allocator: &std.mem.Allocator,
+    queue: &Queue(i32),
+    put_sum: isize,
+    get_sum: isize,
+    get_count: usize,
+    puts_done: u8, // TODO make this a bool
+};
+const puts_per_thread = 10000;
+const put_thread_count = 3;
+
+test "std.atomic.queue" {
+    var direct_allocator = std.heap.DirectAllocator.init();
+    defer direct_allocator.deinit();
+
+    var plenty_of_memory = try direct_allocator.allocator.alloc(u8, 64 * 1024 * 1024);
+    defer direct_allocator.allocator.free(plenty_of_memory);
+
+    var fixed_buffer_allocator = std.heap.ThreadSafeFixedBufferAllocator.init(plenty_of_memory);
+    var a = &fixed_buffer_allocator.allocator;
+
+    var queue: Queue(i32) = undefined;
+    queue.init();
+    var context = Context {
+        .allocator = a,
+        .queue = &queue,
+        .put_sum = 0,
+        .get_sum = 0,
+        .puts_done = 0,
+        .get_count = 0,
+    };
+
+    var putters: [put_thread_count]&std.os.Thread = undefined;
+    for (putters) |*t| {
+        *t = try std.os.spawnThread(&context, startPuts);
+    }
+    var getters: [put_thread_count]&std.os.Thread = undefined;
+    for (getters) |*t| {
+        *t = try std.os.spawnThread(&context, startGets);
+    }
+
+    for (putters) |t| t.wait();
+    _ = @atomicRmw(u8, &context.puts_done, builtin.AtomicRmwOp.Xchg, 1, AtomicOrder.SeqCst);
+    for (getters) |t| t.wait();
+
+    std.debug.assert(context.put_sum == context.get_sum);
+    std.debug.assert(context.get_count == puts_per_thread * put_thread_count);
+}
+
+fn startPuts(ctx: &Context) u8 {
+    var put_count: usize = puts_per_thread;
+    var r = std.rand.DefaultPrng.init(0xdeadbeef);
+    while (put_count != 0) : (put_count -= 1) {
+        std.os.time.sleep(0, 1); // let the os scheduler be our fuzz
+        const x = @bitCast(i32, r.random.scalar(u32));
+        const node = ctx.allocator.create(Queue(i32).Node) catch unreachable;
+        node.data = x;
+        ctx.queue.put(node);
+        _ = @atomicRmw(isize, &ctx.put_sum, builtin.AtomicRmwOp.Add, x, AtomicOrder.SeqCst);
+    }
+    return 0;
+}
+
+fn startGets(ctx: &Context) u8 {
+    while (true) {
+        while (ctx.queue.get()) |node| {
+            std.os.time.sleep(0, 1); // let the os scheduler be our fuzz
+            _ = @atomicRmw(isize, &ctx.get_sum, builtin.AtomicRmwOp.Add, node.data, builtin.AtomicOrder.SeqCst);
+            _ = @atomicRmw(usize, &ctx.get_count, builtin.AtomicRmwOp.Add, 1, builtin.AtomicOrder.SeqCst);
+        }
+
+        if (@atomicLoad(u8, &ctx.puts_done, builtin.AtomicOrder.SeqCst) == 1) {
+            break;
+        }
+    }
+    return 0;
+}

std/atomic/stack.zig

@@ -0,0 +1,126 @@
+const builtin = @import("builtin");
+const AtomicOrder = builtin.AtomicOrder;
+
+/// Many reader, many writer, non-allocating, thread-safe, lock-free
+pub fn Stack(comptime T: type) type {
+    return struct {
+        root: ?&Node,
+
+        pub const Self = this;
+
+        pub const Node = struct {
+            next: ?&Node,
+            data: T,
+        };
+
+        pub fn init() Self {
+            return Self {
+                .root = null,
+            };
+        }
+
+        /// push operation, but only if you are the first item in the stack. if you did not succeed in
+        /// being the first item in the stack, returns the other item that was there.
+        pub fn pushFirst(self: &Self, node: &Node) ?&Node {
+            node.next = null;
+            return @cmpxchgStrong(?&Node, &self.root, null, node, AtomicOrder.SeqCst, AtomicOrder.SeqCst);
+        }
+
+        pub fn push(self: &Self, node: &Node) void {
+            var root = @atomicLoad(?&Node, &self.root, AtomicOrder.SeqCst);
+            while (true) {
+                node.next = root;
+                root = @cmpxchgWeak(?&Node, &self.root, root, node, AtomicOrder.SeqCst, AtomicOrder.SeqCst) ?? break;
+            }
+        }
+
+        pub fn pop(self: &Self) ?&Node {
+            var root = @atomicLoad(?&Node, &self.root, AtomicOrder.Acquire);
+            while (true) {
+                root = @cmpxchgWeak(?&Node, &self.root, root, (root ?? return null).next, AtomicOrder.SeqCst, AtomicOrder.SeqCst) ?? return root;
+            }
+        }
+
+        pub fn isEmpty(self: &Self) bool {
+            return @atomicLoad(?&Node, &self.root, AtomicOrder.SeqCst) == null;
+        }
+    };
+}
+
+const std = @import("std");
+const Context = struct {
+    allocator: &std.mem.Allocator,
+    stack: &Stack(i32),
+    put_sum: isize,
+    get_sum: isize,
+    get_count: usize,
+    puts_done: u8, // TODO make this a bool
+};
+const puts_per_thread = 1000;
+const put_thread_count = 3;
+
+test "std.atomic.stack" {
+    var direct_allocator = std.heap.DirectAllocator.init();
+    defer direct_allocator.deinit();
+
+    var plenty_of_memory = try direct_allocator.allocator.alloc(u8, 64 * 1024 * 1024);
+    defer direct_allocator.allocator.free(plenty_of_memory);
+
+    var fixed_buffer_allocator = std.heap.ThreadSafeFixedBufferAllocator.init(plenty_of_memory);
+    var a = &fixed_buffer_allocator.allocator;
+
+    var stack = Stack(i32).init();
+    var context = Context {
+        .allocator = a,
+        .stack = &stack,
+        .put_sum = 0,
+        .get_sum = 0,
+        .puts_done = 0,
+        .get_count = 0,
+    };
+
+    var putters: [put_thread_count]&std.os.Thread = undefined;
+    for (putters) |*t| {
+        *t = try std.os.spawnThread(&context, startPuts);
+    }
+    var getters: [put_thread_count]&std.os.Thread = undefined;
+    for (getters) |*t| {
+        *t = try std.os.spawnThread(&context, startGets);
+    }
+
+    for (putters) |t| t.wait();
+    _ = @atomicRmw(u8, &context.puts_done, builtin.AtomicRmwOp.Xchg, 1, AtomicOrder.SeqCst);
+    for (getters) |t| t.wait();
+
+    std.debug.assert(context.put_sum == context.get_sum);
+    std.debug.assert(context.get_count == puts_per_thread * put_thread_count);
+}
+
+fn startPuts(ctx: &Context) u8 {
+    var put_count: usize = puts_per_thread;
+    var r = std.rand.DefaultPrng.init(0xdeadbeef);
+    while (put_count != 0) : (put_count -= 1) {
+        std.os.time.sleep(0, 1); // let the os scheduler be our fuzz
+        const x = @bitCast(i32, r.random.scalar(u32));
+        const node = ctx.allocator.create(Stack(i32).Node) catch unreachable;
+        node.data = x;
+        ctx.stack.push(node);
+        _ = @atomicRmw(isize, &ctx.put_sum, builtin.AtomicRmwOp.Add, x, AtomicOrder.SeqCst);
+    }
+    return 0;
+}
+
+fn startGets(ctx: &Context) u8 {
+    while (true) {
+        while (ctx.stack.pop()) |node| {
+            std.os.time.sleep(0, 1); // let the os scheduler be our fuzz
+            _ = @atomicRmw(isize, &ctx.get_sum, builtin.AtomicRmwOp.Add, node.data, builtin.AtomicOrder.SeqCst);
+            _ = @atomicRmw(usize, &ctx.get_count, builtin.AtomicRmwOp.Add, 1, builtin.AtomicOrder.SeqCst);
+        }
+
+        if (@atomicLoad(u8, &ctx.puts_done, builtin.AtomicOrder.SeqCst) == 1) {
+            break;
+        }
+    }
+    return 0;
+}

std/c/darwin.zig

@@ -81,3 +81,8 @@ pub const sockaddr = extern struct {
 };
 
 pub const sa_family_t = u8;
+
+pub const pthread_attr_t = extern struct {
+    __sig: c_long,
+    __opaque: [56]u8,
+};

std/c/index.zig

@@ -53,3 +53,13 @@ pub extern "c" fn malloc(usize) ?&c_void;
 pub extern "c" fn realloc(&c_void, usize) ?&c_void;
 pub extern "c" fn free(&c_void) void;
 pub extern "c" fn posix_memalign(memptr: &&c_void, alignment: usize, size: usize) c_int;
+
+pub extern "pthread" fn pthread_create(noalias newthread: &pthread_t,
+    noalias attr: ?&const pthread_attr_t, start_routine: extern fn(?&c_void) ?&c_void,
+    noalias arg: ?&c_void) c_int;
+pub extern "pthread" fn pthread_attr_init(attr: &pthread_attr_t) c_int;
+pub extern "pthread" fn pthread_attr_setstack(attr: &pthread_attr_t, stackaddr: &c_void, stacksize: usize) c_int;
+pub extern "pthread" fn pthread_attr_destroy(attr: &pthread_attr_t) c_int;
+pub extern "pthread" fn pthread_join(thread: pthread_t, arg_return: ?&?&c_void) c_int;
+
+pub const pthread_t = &@OpaqueType();

std/c/linux.zig

@@ -3,3 +3,8 @@ pub use @import("../os/linux/errno.zig");
 pub extern "c" fn getrandom(buf_ptr: &u8, buf_len: usize, flags: c_uint) c_int;
 extern "c" fn __errno_location() &c_int;
 pub const _errno = __errno_location;
+
+pub const pthread_attr_t = extern struct {
+    __size: [56]u8,
+    __align: c_long,
+};

std/heap.zig

@@ -47,13 +47,6 @@ pub const DirectAllocator = struct {
 
     const HeapHandle = if (builtin.os == Os.windows) os.windows.HANDLE else void;
 
-    //pub const canary_bytes = []u8 {48, 239, 128, 46, 18, 49, 147, 9, 195, 59, 203, 3, 245, 54, 9, 122};
-    //pub const want_safety = switch (builtin.mode) {
-    //    builtin.Mode.Debug => true,
-    //    builtin.Mode.ReleaseSafe => true,
-    //    else => false,
-    //};
-
     pub fn init() DirectAllocator {
         return DirectAllocator {
             .allocator = Allocator {
@@ -98,7 +91,7 @@ pub const DirectAllocator = struct {
                 const unused_start = addr;
                 const unused_len = aligned_addr - 1 - unused_start;
 
-                var err = p.munmap(@intToPtr(&u8, unused_start), unused_len);
+                var err = p.munmap(unused_start, unused_len);
                 debug.assert(p.getErrno(err) == 0);
                 
                 //It is impossible that there is an unoccupied page at the top of our
@@ -139,7 +132,7 @@ pub const DirectAllocator = struct {
                     const rem = @rem(new_addr_end, os.page_size);
                     const new_addr_end_rounded = new_addr_end + if (rem == 0) 0 else (os.page_size - rem);
                     if (old_addr_end > new_addr_end_rounded) {
-                        _ = os.posix.munmap(@intToPtr(&u8, new_addr_end_rounded), old_addr_end - new_addr_end_rounded);
+                        _ = os.posix.munmap(new_addr_end_rounded, old_addr_end - new_addr_end_rounded);
                     }
                     return old_mem[0..new_size];
                 }
@@ -177,7 +170,7 @@ pub const DirectAllocator = struct {
 
         switch (builtin.os) {
             Os.linux, Os.macosx, Os.ios => {
-                _ = os.posix.munmap(bytes.ptr, bytes.len);
+                _ = os.posix.munmap(@ptrToInt(bytes.ptr), bytes.len);
             },
             Os.windows => {
                 const record_addr = @ptrToInt(bytes.ptr) + bytes.len;
@@ -298,7 +291,7 @@ pub const FixedBufferAllocator = struct {
 
     fn alloc(allocator: &Allocator, n: usize, alignment: u29) ![]u8 {
         const self = @fieldParentPtr(FixedBufferAllocator, "allocator", allocator);
-        const addr = @ptrToInt(&self.buffer[self.end_index]);
+        const addr = @ptrToInt(self.buffer.ptr) + self.end_index;
         const rem = @rem(addr, alignment);
         const march_forward_bytes = if (rem == 0) 0 else (alignment - rem);
         const adjusted_index = self.end_index + march_forward_bytes;
@@ -325,6 +318,54 @@ pub const FixedBufferAllocator = struct {
     fn free(allocator: &Allocator, bytes: []u8) void { }
 };
 
+/// lock free
+pub const ThreadSafeFixedBufferAllocator = struct {
+    allocator: Allocator,
+    end_index: usize,
+    buffer: []u8,
+
+    pub fn init(buffer: []u8) ThreadSafeFixedBufferAllocator {
+        return ThreadSafeFixedBufferAllocator {
+            .allocator = Allocator {
+                .allocFn = alloc,
+                .reallocFn = realloc,
+                .freeFn = free,
+            },
+            .buffer = buffer,
+            .end_index = 0,
+        };
+    }
+
+    fn alloc(allocator: &Allocator, n: usize, alignment: u29) ![]u8 {
+        const self = @fieldParentPtr(ThreadSafeFixedBufferAllocator, "allocator", allocator);
+        var end_index = @atomicLoad(usize, &self.end_index, builtin.AtomicOrder.SeqCst);
+        while (true) {
+            const addr = @ptrToInt(self.buffer.ptr) + end_index;
+            const rem = @rem(addr, alignment);
+            const march_forward_bytes = if (rem == 0) 0 else (alignment - rem);
+            const adjusted_index = end_index + march_forward_bytes;
+            const new_end_index = adjusted_index + n;
+            if (new_end_index > self.buffer.len) {
+                return error.OutOfMemory;
+            }
+            end_index = @cmpxchgWeak(usize, &self.end_index, end_index, new_end_index,
+                builtin.AtomicOrder.SeqCst, builtin.AtomicOrder.SeqCst) ?? return self.buffer[adjusted_index .. new_end_index];
+        }
+    }
+
+    fn realloc(allocator: &Allocator, old_mem: []u8, new_size: usize, alignment: u29) ![]u8 {
+        if (new_size <= old_mem.len) {
+            return old_mem[0..new_size];
+        } else {
+            const result = try alloc(allocator, new_size, alignment);
+            mem.copy(u8, result, old_mem);
+            return result;
+        }
+    }
+
+    fn free(allocator: &Allocator, bytes: []u8) void { }
+};
+
 
 
 test "c_allocator" {
@@ -363,6 +404,13 @@ test "FixedBufferAllocator" {
     try testAllocatorLargeAlignment(&fixed_buffer_allocator.allocator);
 }
 
+test "ThreadSafeFixedBufferAllocator" {
+    var fixed_buffer_allocator = ThreadSafeFixedBufferAllocator.init(test_fixed_buffer_allocator_memory[0..]);
+
+    try testAllocator(&fixed_buffer_allocator.allocator);
+    try testAllocatorLargeAlignment(&fixed_buffer_allocator.allocator);
+}
+
 fn testAllocator(allocator: &mem.Allocator) !void {
     var slice = try allocator.alloc(&i32, 100);
 

std/index.zig

@@ -8,6 +8,7 @@ pub const HashMap = @import("hash_map.zig").HashMap;
 pub const LinkedList = @import("linked_list.zig").LinkedList;
 pub const IntrusiveLinkedList = @import("linked_list.zig").IntrusiveLinkedList;
 
+pub const atomic = @import("atomic/index.zig");
 pub const base64 = @import("base64.zig");
 pub const build = @import("build.zig");
 pub const c = @import("c/index.zig");
@@ -34,6 +35,7 @@ pub const zig = @import("zig/index.zig");
 
 test "std" {
     // run tests from these
+    _ = @import("atomic/index.zig");
     _ = @import("array_list.zig");
     _ = @import("buf_map.zig");
     _ = @import("buf_set.zig");

std/mem.zig

@@ -32,6 +32,7 @@ pub const Allocator = struct {
     freeFn: fn (self: &Allocator, old_mem: []u8) void,
 
     fn create(self: &Allocator, comptime T: type) !&T {
+        if (@sizeOf(T) == 0) return &{};
         const slice = try self.alloc(T, 1);
         return &slice[0];
     }

std/os/darwin.zig

@@ -184,7 +184,7 @@ pub fn write(fd: i32, buf: &const u8, nbyte: usize) usize {
     return errnoWrap(c.write(fd, @ptrCast(&const c_void, buf), nbyte));
 }
 
-pub fn mmap(address: ?&u8, length: usize, prot: usize, flags: usize, fd: i32,
+pub fn mmap(address: ?&u8, length: usize, prot: usize, flags: u32, fd: i32,
     offset: isize) usize
 {
     const ptr_result = c.mmap(@ptrCast(&c_void, address), length,
@@ -193,8 +193,8 @@ pub fn mmap(address: ?&u8, length: usize, prot: usize, flags: usize, fd: i32,
     return errnoWrap(isize_result);
 }
 
-pub fn munmap(address: &u8, length: usize) usize {
+pub fn munmap(address: usize, length: usize) usize {
-    return errnoWrap(c.munmap(@ptrCast(&c_void, address), length));
+    return errnoWrap(c.munmap(@intToPtr(&c_void, address), length));
 }
 
 pub fn unlink(path: &const u8) usize {
@@ -341,4 +341,4 @@ pub const timeval = c.timeval;
 pub const mach_timebase_info_data = c.mach_timebase_info_data;
 
 pub const mach_absolute_time = c.mach_absolute_time;
-pub const mach_timebase_info = c.mach_timebase_info;
+pub const mach_timebase_info = c.mach_timebase_info;

std/os/index.zig

@@ -2,6 +2,10 @@ const std = @import("../index.zig");
 const builtin = @import("builtin");
 const Os = builtin.Os;
 const is_windows = builtin.os == Os.windows;
+const is_posix = switch (builtin.os) {
+    builtin.Os.linux, builtin.Os.macosx => true,
+    else => false,
+};
 const os = this;
 
 test "std.os" {
@@ -2343,24 +2347,58 @@ pub fn posixGetSockOptConnectError(sockfd: i32) PosixConnectError!void {
 }
 
 pub const Thread = struct {
-    pid: i32,
+    data: Data,
-    allocator: ?&mem.Allocator,
+
-    stack: []u8,
+    pub const use_pthreads = is_posix and builtin.link_libc;
+    const Data = if (use_pthreads) struct {
+      handle: c.pthread_t,
+      stack_addr: usize,
+      stack_len: usize,
+    } else switch (builtin.os) {
+        builtin.Os.linux => struct {
+            pid: i32,
+            stack_addr: usize,
+            stack_len: usize,
+        },
+        builtin.Os.windows => struct {
+            handle: windows.HANDLE,
+            alloc_start: &c_void,
+            heap_handle: windows.HANDLE,
+        },
+        else => @compileError("Unsupported OS"),
+    };
 
     pub fn wait(self: &const Thread) void {
-        while (true) {
+        if (use_pthreads) {
-            const pid_value = @atomicLoad(i32, &self.pid, builtin.AtomicOrder.SeqCst);
+            const err = c.pthread_join(self.data.handle, null);
-            if (pid_value == 0) break;
+            switch (err) {
-            const rc = linux.futex_wait(@ptrToInt(&self.pid), linux.FUTEX_WAIT, pid_value, null);
+                0 => {},
-            switch (linux.getErrno(rc)) {
+                posix.EINVAL => unreachable,
-                0 => continue,
+                posix.ESRCH => unreachable,
-                posix.EINTR => continue,
+                posix.EDEADLK => unreachable,
-                posix.EAGAIN => continue,
                 else => unreachable,
             }
-        }
+            assert(posix.munmap(self.data.stack_addr, self.data.stack_len) == 0);
-        if (self.allocator) |a| {
+        } else switch (builtin.os) {
-            a.free(self.stack);
+            builtin.Os.linux => {
+                while (true) {
+                    const pid_value = @atomicLoad(i32, &self.data.pid, builtin.AtomicOrder.SeqCst);
+                    if (pid_value == 0) break;
+                    const rc = linux.futex_wait(@ptrToInt(&self.data.pid), linux.FUTEX_WAIT, pid_value, null);
+                    switch (linux.getErrno(rc)) {
+                        0 => continue,
+                        posix.EINTR => continue,
+                        posix.EAGAIN => continue,
+                        else => unreachable,
+                    }
+                }
+                assert(posix.munmap(self.data.stack_addr, self.data.stack_len) == 0);
+            },
+            builtin.Os.windows => {
+                assert(windows.WaitForSingleObject(self.data.handle, windows.INFINITE) == windows.WAIT_OBJECT_0);
+                assert(windows.HeapFree(self.data.heap_handle, 0, self.data.alloc_start) != 0);
+            },
+            else => @compileError("Unsupported OS"),
         }
     }
 };
@@ -2385,38 +2423,94 @@ pub const SpawnThreadError = error {
     /// be copied.
     SystemResources,
 
+    /// Not enough userland memory to spawn the thread.
+    OutOfMemory,
+
     Unexpected,
 };
 
-pub const SpawnThreadAllocatorError = SpawnThreadError || error{OutOfMemory};
-
 /// caller must call wait on the returned thread
 /// fn startFn(@typeOf(context)) T
 /// where T is u8, noreturn, void, or !void
-pub fn spawnThreadAllocator(allocator: &mem.Allocator, context: var, comptime startFn: var) SpawnThreadAllocatorError!&Thread {
+/// caller must call wait on the returned thread
+pub fn spawnThread(context: var, comptime startFn: var) SpawnThreadError!&Thread {
     // TODO compile-time call graph analysis to determine stack upper bound
     // https://github.com/zig-lang/zig/issues/157
     const default_stack_size = 8 * 1024 * 1024;
-    const stack_bytes = try allocator.alloc(u8, default_stack_size);
-    const thread = try spawnThread(stack_bytes, context, startFn);
-    thread.allocator = allocator;
-    return thread;
-}
 
-/// stack must be big enough to store one Thread and one @typeOf(context), each with default alignment, at the end
-/// fn startFn(@typeOf(context)) T
-/// where T is u8, noreturn, void, or !void
-/// caller must call wait on the returned thread
-pub fn spawnThread(stack: []u8, context: var, comptime startFn: var) SpawnThreadError!&Thread {
     const Context = @typeOf(context);
     comptime assert(@ArgType(@typeOf(startFn), 0) == Context);
 
-    var stack_end: usize = @ptrToInt(stack.ptr) + stack.len;
+    if (builtin.os == builtin.Os.windows) {
+        const WinThread = struct {
+            const OuterContext = struct {
+                thread: Thread,
+                inner: Context,
+            };
+            extern fn threadMain(arg: windows.LPVOID) windows.DWORD {
+                if (@sizeOf(Context) == 0) {
+                    return startFn({});
+                } else {
+                    return startFn(*@ptrCast(&Context, @alignCast(@alignOf(Context), arg)));
+                }
+            }
+        };
+
+        const heap_handle = windows.GetProcessHeap() ?? return SpawnThreadError.OutOfMemory;
+        const byte_count = @alignOf(WinThread.OuterContext) + @sizeOf(WinThread.OuterContext);
+        const bytes_ptr = windows.HeapAlloc(heap_handle, 0, byte_count) ?? return SpawnThreadError.OutOfMemory;
+        errdefer assert(windows.HeapFree(heap_handle, 0, bytes_ptr) != 0);
+        const bytes = @ptrCast(&u8, bytes_ptr)[0..byte_count];
+        const outer_context = std.heap.FixedBufferAllocator.init(bytes).allocator.create(WinThread.OuterContext) catch unreachable;
+        outer_context.inner = context;
+        outer_context.thread.data.heap_handle = heap_handle;
+        outer_context.thread.data.alloc_start = bytes_ptr;
+
+        const parameter = if (@sizeOf(Context) == 0) null else @ptrCast(&c_void, &outer_context.inner);
+        outer_context.thread.data.handle = windows.CreateThread(null, default_stack_size, WinThread.threadMain,
+            parameter, 0, null) ??
+        {
+            const err = windows.GetLastError();
+            return switch (err) {
+                else => os.unexpectedErrorWindows(err),
+            };
+        };
+        return &outer_context.thread;
+    }
+
+    const MainFuncs = struct {
+        extern fn linuxThreadMain(ctx_addr: usize) u8 {
+            if (@sizeOf(Context) == 0) {
+                return startFn({});
+            } else {
+                return startFn(*@intToPtr(&const Context, ctx_addr));
+            }
+        }
+        extern fn posixThreadMain(ctx: ?&c_void) ?&c_void {
+            if (@sizeOf(Context) == 0) {
+                _ = startFn({});
+                return null;
+            } else {
+                _ = startFn(*@ptrCast(&const Context, @alignCast(@alignOf(Context), ctx)));
+                return null;
+            }
+        }
+    };
+
+    const MAP_GROWSDOWN = if (builtin.os == builtin.Os.linux) linux.MAP_GROWSDOWN else 0;
+
+    const mmap_len = default_stack_size;
+    const stack_addr = posix.mmap(null, mmap_len, posix.PROT_READ|posix.PROT_WRITE,
+            posix.MAP_PRIVATE|posix.MAP_ANONYMOUS|MAP_GROWSDOWN, -1, 0);
+    if (stack_addr == posix.MAP_FAILED) return error.OutOfMemory;
+    errdefer assert(posix.munmap(stack_addr, mmap_len) == 0);
+
+    var stack_end: usize = stack_addr + mmap_len;
     var arg: usize = undefined;
     if (@sizeOf(Context) != 0) {
         stack_end -= @sizeOf(Context);
         stack_end -= stack_end % @alignOf(Context);
-        assert(stack_end >= @ptrToInt(stack.ptr));
+        assert(stack_end >= stack_addr);
         const context_ptr = @alignCast(@alignOf(Context), @intToPtr(&Context, stack_end));
         *context_ptr = context;
         arg = stack_end;
@@ -2424,36 +2518,53 @@ pub fn spawnThread(stack: []u8, context: var, comptime startFn: var) SpawnThread
 
     stack_end -= @sizeOf(Thread);
     stack_end -= stack_end % @alignOf(Thread);
-    assert(stack_end >= @ptrToInt(stack.ptr));
+    assert(stack_end >= stack_addr);
     const thread_ptr = @alignCast(@alignOf(Thread), @intToPtr(&Thread, stack_end));
-    thread_ptr.stack = stack;
-    thread_ptr.allocator = null;
 
-    const threadMain = struct {
+    thread_ptr.data.stack_addr = stack_addr;
-        extern fn threadMain(ctx_addr: usize) u8 {
+    thread_ptr.data.stack_len = mmap_len;
-            if (@sizeOf(Context) == 0) {
+
-                return startFn({});
+    if (builtin.os == builtin.Os.windows) {
-            } else {
+        // use windows API directly
-                return startFn(*@intToPtr(&const Context, ctx_addr));
+        @compileError("TODO support spawnThread for Windows");
-            }
+    } else if (Thread.use_pthreads) {
+        // use pthreads
+        var attr: c.pthread_attr_t = undefined;
+        if (c.pthread_attr_init(&attr) != 0) return SpawnThreadError.SystemResources;
+        defer assert(c.pthread_attr_destroy(&attr) == 0);
+
+        // align to page
+        stack_end -= stack_end % os.page_size;
+        assert(c.pthread_attr_setstack(&attr, @intToPtr(&c_void, stack_addr), stack_end - stack_addr) == 0);
+
+        const err = c.pthread_create(&thread_ptr.data.handle, &attr, MainFuncs.posixThreadMain, @intToPtr(&c_void, arg));
+        switch (err) {
+            0 => return thread_ptr,
+            posix.EAGAIN => return SpawnThreadError.SystemResources,
+            posix.EPERM => unreachable,
+            posix.EINVAL => unreachable,
+            else => return unexpectedErrorPosix(usize(err)),
         }
-    }.threadMain;
+    } else if (builtin.os == builtin.Os.linux) {
-
+        // use linux API directly
-    const flags = posix.CLONE_VM | posix.CLONE_FS | posix.CLONE_FILES | posix.CLONE_SIGHAND
+        const flags = posix.CLONE_VM | posix.CLONE_FS | posix.CLONE_FILES | posix.CLONE_SIGHAND
-        | posix.CLONE_THREAD | posix.CLONE_SYSVSEM // | posix.CLONE_SETTLS
+            | posix.CLONE_THREAD | posix.CLONE_SYSVSEM // | posix.CLONE_SETTLS
-        | posix.CLONE_PARENT_SETTID | posix.CLONE_CHILD_CLEARTID | posix.CLONE_DETACHED;
+            | posix.CLONE_PARENT_SETTID | posix.CLONE_CHILD_CLEARTID | posix.CLONE_DETACHED;
-    const newtls: usize = 0;
+        const newtls: usize = 0;
-    const rc = posix.clone(threadMain, stack_end, flags, arg, &thread_ptr.pid, newtls, &thread_ptr.pid);
+        const rc = posix.clone(MainFuncs.linuxThreadMain, stack_end, flags, arg, &thread_ptr.data.pid, newtls, &thread_ptr.data.pid);
-    const err = posix.getErrno(rc);
+        const err = posix.getErrno(rc);
-    switch (err) {
+        switch (err) {
-        0 => return thread_ptr,
+            0 => return thread_ptr,
-        posix.EAGAIN => return SpawnThreadError.ThreadQuotaExceeded,
+            posix.EAGAIN => return SpawnThreadError.ThreadQuotaExceeded,
-        posix.EINVAL => unreachable,
+            posix.EINVAL => unreachable,
-        posix.ENOMEM => return SpawnThreadError.SystemResources,
+            posix.ENOMEM => return SpawnThreadError.SystemResources,
-        posix.ENOSPC => unreachable,
+            posix.ENOSPC => unreachable,
-        posix.EPERM => unreachable,
+            posix.EPERM => unreachable,
-        posix.EUSERS => unreachable,
+            posix.EUSERS => unreachable,
-        else => return unexpectedErrorPosix(err),
+            else => return unexpectedErrorPosix(err),
+        }
+    } else {
+        @compileError("Unsupported OS");
     }
 }
 

std/os/linux/index.zig

@@ -706,13 +706,13 @@ pub fn umount2(special: &const u8, flags: u32) usize {
     return syscall2(SYS_umount2, @ptrToInt(special), flags);
 }
 
-pub fn mmap(address: ?&u8, length: usize, prot: usize, flags: usize, fd: i32, offset: isize) usize {
+pub fn mmap(address: ?&u8, length: usize, prot: usize, flags: u32, fd: i32, offset: isize) usize {
     return syscall6(SYS_mmap, @ptrToInt(address), length, prot, flags, usize(fd),
         @bitCast(usize, offset));
 }
 
-pub fn munmap(address: &u8, length: usize) usize {
+pub fn munmap(address: usize, length: usize) usize {
-    return syscall2(SYS_munmap, @ptrToInt(address), length);
+    return syscall2(SYS_munmap, address, length);
 }
 
 pub fn read(fd: i32, buf: &u8, count: usize) usize {

std/os/test.zig

@@ -44,24 +44,12 @@ test "access file" {
 }
 
 test "spawn threads" {
-    if (builtin.os != builtin.Os.linux) {
-        // TODO implement threads on macos and windows
-        return;
-    }
-
-    var direct_allocator = std.heap.DirectAllocator.init();
-    defer direct_allocator.deinit();
-
     var shared_ctx: i32 = 1;
 
-    const thread1 = try std.os.spawnThreadAllocator(&direct_allocator.allocator, {}, start1);
+    const thread1 = try std.os.spawnThread({}, start1);
-    const thread4 = try std.os.spawnThreadAllocator(&direct_allocator.allocator, &shared_ctx, start2);
+    const thread2 = try std.os.spawnThread(&shared_ctx, start2);
-
+    const thread3 = try std.os.spawnThread(&shared_ctx, start2);
-    var stack1: [1024]u8 = undefined;
+    const thread4 = try std.os.spawnThread(&shared_ctx, start2);
-    var stack2: [1024]u8 = undefined;
-
-    const thread2 = try std.os.spawnThread(stack1[0..], &shared_ctx, start2);
-    const thread3 = try std.os.spawnThread(stack2[0..], &shared_ctx, start2);
 
     thread1.wait();
     thread2.wait();

std/os/windows/index.zig

@@ -28,6 +28,9 @@ pub extern "kernel32" stdcallcc fn CreateProcessA(lpApplicationName: ?LPCSTR, lp
 pub extern "kernel32" stdcallcc fn CreateSymbolicLinkA(lpSymlinkFileName: LPCSTR, lpTargetFileName: LPCSTR,
     dwFlags: DWORD) BOOLEAN;
 
+
+pub extern "kernel32" stdcallcc fn CreateThread(lpThreadAttributes: ?LPSECURITY_ATTRIBUTES, dwStackSize: SIZE_T, lpStartAddress: LPTHREAD_START_ROUTINE, lpParameter: ?LPVOID, dwCreationFlags: DWORD, lpThreadId: ?LPDWORD) ?HANDLE;
+
 pub extern "kernel32" stdcallcc fn DeleteFileA(lpFileName: LPCSTR) BOOL;
 
 pub extern "kernel32" stdcallcc fn ExitProcess(exit_code: UINT) noreturn;
@@ -318,6 +321,9 @@ pub const HEAP_CREATE_ENABLE_EXECUTE = 0x00040000;
 pub const HEAP_GENERATE_EXCEPTIONS = 0x00000004;
 pub const HEAP_NO_SERIALIZE = 0x00000001;
 
+pub const PTHREAD_START_ROUTINE = extern fn(LPVOID) DWORD;
+pub const LPTHREAD_START_ROUTINE = PTHREAD_START_ROUTINE;
+
 test "import" {
     _ = @import("util.zig");
 }