Add basic support for windows

I wanted to keep it as simple as possible, so didn't even add guard
pages. I'll do it in a separate PR

I tried but just could not get it to work with gcc from mingw64, though
gcc from clang64 works, and clang from mingw64 and clang64 works.
Probably a compiler bug or something. I was getting a segfault when
calling coroutine_restore_context in coroutine_switch_context ..

Author: 0dminnimda

Makefile

@@ -1,5 +1,7 @@
+cc ?= gcc
+
 build/counter: examples/counter.c coroutine.h build/coroutine.a
-	gcc -I. -Wall -Wextra -ggdb -o build/counter examples/counter.c build/coroutine.a
+	$(cc) -I. -Wall -Wextra -ggdb -o build/counter examples/counter.c build/coroutine.a
 
 .PHONY: examples
 examples: build/counter build/counter_cpp build/counter_c3 build/counter_jai build/echo build/lexer
@@ -17,15 +19,15 @@ build/counter_jai: examples/counter.jai build/coroutine.a build/coroutine.so
 	jai-linux examples/counter.jai
 
 build/lexer: examples/lexer.c coroutine.h build/coroutine.a
-	gcc -I. -Wall -Wextra -ggdb -o build/lexer examples/lexer.c build/coroutine.a
+	$(cc) -I. -Wall -Wextra -ggdb -o build/lexer examples/lexer.c build/coroutine.a
 
 build/coroutine.so: coroutine.c
 	mkdir -p build
-	gcc -Wall -Wextra -ggdb -shared -fPIC -o build/coroutine.so coroutine.c
+	$(cc) -Wall -Wextra -ggdb -shared -fPIC -o build/coroutine.so coroutine.c
 
 build/coroutine.a: build/coroutine.o
 	ar -rcs build/coroutine.a build/coroutine.o
 
 build/coroutine.o: coroutine.c coroutine.h
 	mkdir -p build
-	gcc -Wall -Wextra -ggdb -c -o build/coroutine.o coroutine.c
+	$(cc) -Wall -Wextra -ggdb -c -o build/coroutine.o coroutine.c

README.md

@@ -3,7 +3,7 @@
 > [!WARNING]
 > The library is not in production ready state yet
 
-Custom coroutines implementation in GNU C.
+Custom coroutines implementation in C.
 
 ## What is a Coroutine?
 
@@ -26,6 +26,13 @@ $ make
 $ ./build/counter
 ```
 
+For Windows you need to use clang
+
+```console
+$ make cc=clang
+$ ./build/counter.exe
+```
+
 There are actually much more examples in the [./examples/](./examples/) in a variety of languages. To build all of them do:
 
 ```console
@@ -36,7 +43,8 @@ Make sure you have all the corresponding compilers for the languages.
 
 ## Supported platforms
 
-- Linux x86_64
+- Linux   x86_64
+- Windows x86_64 (tested for Msys clang)
 
 *More are planned in the future*
 

coroutine.c

@@ -4,12 +4,26 @@
 #include <stdbool.h>
 #include <string.h>
 
+#if _WIN32
+#define WIN32_LEAN_AND_MEAN
+#include <windows.h>
+#else
 #include <poll.h>
 #include <unistd.h>
 #include <sys/mman.h>
+#endif
 
 #include "coroutine.h"
 
+
+#if _WIN32
+int getpagesize() {
+    SYSTEM_INFO si;
+    GetSystemInfo(&si);
+    return si.dwPageSize;
+}
+#endif
+
 // TODO: make the STACK_CAPACITY customizable by the user
 //#define STACK_CAPACITY (4*1024)
 #define STACK_CAPACITY (1024*getpagesize())
@@ -60,7 +74,11 @@ typedef struct {
 } Indices;
 
 typedef struct {
+#if _WIN32
+    char *items;  // not supported yet
+#else
     struct pollfd *items;
+#endif
     size_t count;
     size_t capacity;
 } Polls;
@@ -82,12 +100,33 @@ typedef enum {
     SM_WRITE,
 } Sleep_Mode;
 
-// Linux x86_64 call convention
+// Linux x86_64 calling convention
 // %rdi, %rsi, %rdx, %rcx, %r8, and %r9
 
+// https://learn.microsoft.com/en-us/cpp/build/x64-software-conventions?view=msvc-170#x64-register-usage
+// Windows x64 calling convention: RCX, RDX, R8, R9
+// Windows x64 ABI considers registers RBX, RBP, RDI, RSI, RSP, R12, R13, R14, R15, and XMM6-XMM15 nonvolatile.
+// They must be saved and restored by a function that uses them.
+
 void __attribute__((naked)) coroutine_yield(void)
 {
     // @arch
+#if _WIN32
+    asm(
+    "    pushq %rcx\n"
+    "    pushq %rbx\n"
+    "    pushq %rbp\n"
+    "    pushq %rdi\n"
+    "    pushq %rsi\n"
+    "    pushq %r12\n"
+    "    pushq %r13\n"
+    "    pushq %r14\n"
+    "    pushq %r15\n"
+    // TODO: push XMM6-XMM15
+    "    movq %rsp, %rcx\n"     // rsp
+    "    movq $0, %rdx\n"       // sm = SM_READ
+    "    jmp coroutine_switch_context\n");
+#else
     asm(
     "    pushq %rdi\n"
     "    pushq %rbp\n"
@@ -99,12 +138,32 @@ void __attribute__((naked)) coroutine_yield(void)
     "    movq %rsp, %rdi\n"     // rsp
     "    movq $0, %rsi\n"       // sm = SM_NONE
     "    jmp coroutine_switch_context\n");
+#endif
 }
 
 void __attribute__((naked)) coroutine_sleep_read(int fd)
 {
+#if !defined(__clang__)
     (void) fd;
+#endif
     // @arch
+#if _WIN32
+    asm(
+    "    pushq %rcx\n"
+    "    pushq %rbx\n"
+    "    pushq %rbp\n"
+    "    pushq %rdi\n"
+    "    pushq %rsi\n"
+    "    pushq %r12\n"
+    "    pushq %r13\n"
+    "    pushq %r14\n"
+    "    pushq %r15\n"
+    // TODO: push XMM6-XMM15
+    "    movq %rcx, %r8\n"      // fd
+    "    movq %rsp, %rcx\n"     // rsp
+    "    movq $1, %rdx\n"       // sm = SM_READ
+    "    jmp coroutine_switch_context\n");
+#else
     asm(
     "    pushq %rdi\n"
     "    pushq %rbp\n"
@@ -117,12 +176,32 @@ void __attribute__((naked)) coroutine_sleep_read(int fd)
     "    movq %rsp, %rdi\n"     // rsp
     "    movq $1, %rsi\n"       // sm = SM_READ
     "    jmp coroutine_switch_context\n");
+#endif
 }
 
 void __attribute__((naked)) coroutine_sleep_write(int fd)
 {
+#if !defined(__clang__)
     (void) fd;
+#endif
     // @arch
+#if _WIN32
+    asm(
+    "    pushq %rcx\n"
+    "    pushq %rbx\n"
+    "    pushq %rbp\n"
+    "    pushq %rdi\n"
+    "    pushq %rsi\n"
+    "    pushq %r12\n"
+    "    pushq %r13\n"
+    "    pushq %r14\n"
+    "    pushq %r15\n"
+    // TODO: push XMM6-XMM15
+    "    movq %rcx, %r8\n"      // fd
+    "    movq %rsp, %rcx\n"     // rsp
+    "    movq $2, %rdx\n"       // sm = SM_READ
+    "    jmp coroutine_switch_context\n");
+#else
     asm(
     "    pushq %rdi\n"
     "    pushq %rbp\n"
@@ -135,12 +214,30 @@ void __attribute__((naked)) coroutine_sleep_write(int fd)
     "    movq %rsp, %rdi\n"     // rsp
     "    movq $2, %rsi\n"       // sm = SM_WRITE
     "    jmp coroutine_switch_context\n");
+#endif
 }
 
 void __attribute__((naked)) coroutine_restore_context(void *rsp)
 {
     // @arch
+#if !defined(__clang__)
     (void)rsp;
+#endif
+#if _WIN32
+    asm(
+    "    movq %rcx, %rsp\n"
+    // TODO: pop XMM15-XMM6
+    "    popq %r15\n"
+    "    popq %r14\n"
+    "    popq %r13\n"
+    "    popq %r12\n"
+    "    popq %rsi\n"
+    "    popq %rdi\n"
+    "    popq %rbp\n"
+    "    popq %rbx\n"
+    "    popq %rcx\n"
+    "    ret\n");
+#else
     asm(
     "    movq %rdi, %rsp\n"
     "    popq %r15\n"
@@ -151,6 +248,7 @@ void __attribute__((naked)) coroutine_restore_context(void *rsp)
     "    popq %rbp\n"
     "    popq %rdi\n"
     "    ret\n");
+#endif
 }
 
 void coroutine_switch_context(void *rsp, Sleep_Mode sm, int fd)
@@ -160,23 +258,36 @@ void coroutine_switch_context(void *rsp, Sleep_Mode sm, int fd)
     switch (sm) {
     case SM_NONE: current += 1; break;
     case SM_READ: {
+#if _WIN32
+        (void)fd;
+        TODO("polling is not implemented for windows");
+#else
         da_append(&asleep, active.items[current]);
         struct pollfd pfd = {.fd = fd, .events = POLLRDNORM,};
         da_append(&polls, pfd);
         da_remove_unordered(&active, current);
+#endif
     } break;
 
     case SM_WRITE: {
+#if _WIN32
+        (void)fd;
+        TODO("polling is not implemented for windows");
+#else
         da_append(&asleep, active.items[current]);
         struct pollfd pfd = {.fd = fd, .events = POLLWRNORM,};
         da_append(&polls, pfd);
         da_remove_unordered(&active, current);
+#endif
     } break;
 
     default: UNREACHABLE("coroutine_switch_context");
     }
 
     if (polls.count > 0) {
+#if _WIN32
+        TODO("polling is not implemented for windows");
+#else
         int timeout = active.count == 0 ? -1 : 0;
         int result = poll(polls.items, polls.count, timeout);
         if (result < 0) TODO("poll");
@@ -191,6 +302,7 @@ void coroutine_switch_context(void *rsp, Sleep_Mode sm, int fd)
                 ++i;
             }
         }
+#endif
     }
 
     assert(active.count > 0);
@@ -216,6 +328,9 @@ void coroutine__finish_current(void)
     da_remove_unordered(&active, current);
 
     if (polls.count > 0) {
+#if _WIN32
+        TODO("polling is not implemented for windows");
+#else
         int timeout = active.count == 0 ? -1 : 0;
         int result = poll(polls.items, polls.count, timeout);
         if (result < 0) TODO("poll");
@@ -230,6 +345,7 @@ void coroutine__finish_current(void)
                 ++i;
             }
         }
+#endif
     }
 
     assert(active.count > 0);
@@ -245,21 +361,40 @@ void coroutine_go(void (*f)(void*), void *arg)
     } else {
         da_append(&contexts, ((Context){0}));
         id = contexts.count-1;
+#if _WIN32
+        void *base = contexts.items[id].stack_base = VirtualAlloc(NULL, STACK_CAPACITY, MEM_COMMIT | MEM_RESERVE, PAGE_READWRITE);
+        assert(base != NULL);
+        // TODO: add VirtualProtect with PAGE_NOACCESS for overflow and underflow
+#else
         contexts.items[id].stack_base = mmap(NULL, STACK_CAPACITY, PROT_WRITE|PROT_READ, MAP_PRIVATE|MAP_STACK|MAP_ANONYMOUS|MAP_GROWSDOWN, -1, 0);
         assert(contexts.items[id].stack_base != MAP_FAILED);
+#endif
     }
 
     void **rsp = (void**)((char*)contexts.items[id].stack_base + STACK_CAPACITY);
     // @arch
     *(--rsp) = coroutine__finish_current;
     *(--rsp) = f;
+#if _WIN32
+    *(--rsp) = arg; // push rcx
+    *(--rsp) = 0;   // push rbx
+    *(--rsp) = 0;   // push rbp
+    *(--rsp) = 0;   // push rdi
+    *(--rsp) = 0;   // push rsi
+    *(--rsp) = 0;   // push r12
+    *(--rsp) = 0;   // push r13
+    *(--rsp) = 0;   // push r14
+    *(--rsp) = 0;   // push r15
+    // TODO: push XMM6-XMM15
+#else
     *(--rsp) = arg; // push rdi
     *(--rsp) = 0;   // push rbx
     *(--rsp) = 0;   // push rbp
     *(--rsp) = 0;   // push r12
     *(--rsp) = 0;   // push r13
     *(--rsp) = 0;   // push r14
     *(--rsp) = 0;   // push r15
+#endif
     contexts.items[id].rsp = rsp;
 
     da_append(&active, id);