2 files changed, 142 insertions, 0 deletions
diff --git a/lib/matcha.h b/lib/matcha.h
new file mode 100644
index 0000000..9387bf9
--- /dev/null
+++ b/lib/matcha.h
@@ -0,0 +1,77 @@
+#include <cassert>
+#include <cstdint>
+#include <cstdio>
+#include <sys/mman.h>
+#include <unistd.h>  // sysconf
+
+extern "C" void thread_create();
+extern "C" void yield(void* new_stack, void** old_stack);
+
+long get_pagesize() {
+    return sysconf(_SC_PAGESIZE);
+}
+
+struct Thread {
+    Thread(const Thread&) = delete;
+    Thread& operator=(const Thread&) = delete;
+    Thread(void (*fn)());
+
+    static void entry(void* obj);
+
+    // private:
+    void* mStackPtr;
+    void (*mUserFn)();
+};
+
+Thread::Thread(void (*fn)()) : mUserFn(fn) {
+    const long PAGE_SIZE = get_pagesize();
+    const long STACK_SIZE = 8 * PAGE_SIZE;
+
+    // create new stack
+    void* stack = mmap(nullptr, STACK_SIZE, PROT_WRITE | PROT_READ, MAP_PRIVATE | MAP_ANONYMOUS, -1 /* fd */, 0 /* offset */);
+    assert(stack != MAP_FAILED);
+
+    // protect last stack page (lowest addr)
+    int ret = mprotect(stack, PAGE_SIZE, PROT_NONE);
+    assert(ret == 0);
+
+    // stack grows downwards
+    mStackPtr = static_cast<uint8_t*>(stack) + STACK_SIZE;
+    {
+        uint64_t* stack = static_cast<uint64_t*>(mStackPtr);
+        *(--stack) = (uint64_t)this;
+        *(--stack) = (uint64_t)Thread::entry;
+
+        // initial values for yield epilog
+        *(--stack) = (uint64_t)thread_create;
+        *(--stack) = (uint64_t)0;
+
+        // initial values for callee saved regs
+        *(--stack) = (uint64_t)0;                                        // rbx
+        *(--stack) = (uint64_t)(static_cast<uint64_t*>(mStackPtr) - 4);  // rbp
+        *(--stack) = (uint64_t)0;                                        // r12
+        *(--stack) = (uint64_t)0;                                        // r13
+        *(--stack) = (uint64_t)0;                                        // r14
+        *(--stack) = (uint64_t)0;                                        // r15
+
+        mStackPtr = static_cast<void*>(stack);
+    }
+}
+
+void Thread::entry(void* obj) {
+    Thread* t = static_cast<Thread*>(obj);
+
+    puts("thread entry");
+    t->mUserFn();
+    puts("thread done");
+}
+
+void* gOriginalStack;
+
+void yield_to(const Thread& t) {
+    yield(t.mStackPtr, &gOriginalStack);
+}
+
+void yield_from(Thread& t) {
+    yield(gOriginalStack, &t.mStackPtr);
+}
diff --git a/lib/thread_create.s b/lib/thread_create.s
new file mode 100644
index 0000000..6d6fa96
--- /dev/null
+++ b/lib/thread_create.s
@@ -0,0 +1,65 @@
+# SysV AMD64 ABI
+# int/ptr args : rdi, rsi, rdx, rcx, r8, r9
+# int/ptr ret  : rax
+
+    .intel_syntax noprefix
+    .section .text ,"ax",@progbits
+
+     # extern "C" void thread_create();
+    .global thread_create
+    .type   thread_create, @function
+thread_create:
+    .cfi_startproc
+    mov rdi, qword ptr [rsp+0x8]
+    mov rsi, qword ptr [rsp]
+
+    call rsi
+
+    # FIXME: no return from thread after user fn finished.
+1:
+    jmp 1b
+    .cfi_endproc
+    .size thread_create, .-thread_create
+
+
+
+    # extern "C" void yield(void* new_stack, void** old_stack);
+    #                             ^^^^^^^^^         ^^^^^^^^^
+    #                             rdi               rsi
+    .global yield
+    .type   yield, @function
+yield:
+    .cfi_startproc
+    // prologue
+    push rbp
+    mov rbp, rsp
+
+    // push callee saved registers
+    push rbx
+    push rbp
+    push r12
+    push r13
+    push r14
+    push r15
+
+    // arg0: rdi holds new stack
+    // arg1: rsi holds addr to location current stack must be saved
+    mov [rsi], rsp  # save current stack ptr
+    mov rsp, rdi    # switch to new stack ptr
+
+    // pop callee saved registers
+    pop r15
+    pop r14
+    pop r13
+    pop r12
+    pop rbp
+    pop rbx
+
+    // epilogue
+    mov rsp, rbp
+    pop rbp
+
+    ret
+    .cfi_endproc
+    .size yield, .-yield
+