copy over x86-specific code

src/rt/arch/x64/_context.S

@@ -0,0 +1,76 @@
+	.text
+
+/*
+Callee save registers:
+	ebp, ebx, esi, edi
+
+Caller save registers:
+	eax, ecx, edx
+*/
+	
+/*
+        Stores current registers into arg0/RCX and restores
+        registers found in arg1/RDX. This is used by our
+	    implementation of getcontext.
+*/	
+
+// swap_registers(registers_t *oregs, registers_t *regs)
+.globl swap_registers
+swap_registers:
+    // save the old context
+    movl 4(%esp), %eax
+	//movl %eax, 0(%eax)
+	movl %ebx, 4(%eax)
+	movl %ecx, 8(%eax)
+	movl %edx, 12(%eax)
+	movl %ebp, 16(%eax)
+	movl %esi, 20(%eax)
+	movl %edi, 24(%eax)
+	//movl %cs, 32(%eax)
+	//movl %ds, 34(%eax)
+	//movl %ss, 36(%eax)
+	//movl %es, 38(%eax)
+	//movl %fs, 40(%eax)
+	//movl %gs, 42(%eax)
+
+	// save the flags
+	pushf
+	popl %ecx
+	movl %ecx, 44(%eax)
+
+	// save the return address as the instruction pointer
+    // and save the stack pointer of the caller
+    popl %ecx
+    movl %esp, 28(%eax)
+	movl %ecx, 48(%eax)
+
+    // restore the new context
+    movl 4(%esp), %eax
+
+	movl 4(%eax), %ebx
+	// save ecx for later...
+	movl 12(%eax), %edx
+	movl 16(%eax), %ebp
+	movl 20(%eax), %esi
+	movl 24(%eax), %edi
+	movl 28(%eax), %esp
+	// We can't actually change this...
+	//movl 32(%eax), %cs
+	//movl 34(%eax), %ds
+	//movl 36(%eax), %ss
+	//movl 38(%eax), %es
+	//movl 40(%eax), %fs
+	//movl 42(%eax), %gs
+
+	// restore the flags
+	movl 44(%eax), %ecx
+	push %ecx
+	popf
+
+    // ok, now we can restore ecx
+    movl 8(%eax), %ecx
+	
+    // Return!
+    jmp *48(%eax)
+
+

src/rt/arch/x64/ccall.S

@@ -0,0 +1,22 @@
+    .text
+
+// upcall_call_c_stack(void (*fn)(), void *new_esp)
+//
+// Note that we could use |enter| and |leave| but the manuals tell me they're
+// slower.
+#if defined(__APPLE__) || defined(_WIN32)
+.globl _upcall_call_c_stack
+_upcall_call_c_stack:
+#else
+.globl upcall_call_c_stack
+upcall_call_c_stack:
+#endif
+    pushl %ebp
+    movl %esp,%ebp          // save esp
+    movl 8(%esp),%eax       // eax = callee
+    movl 12(%esp),%esp      // switch stack
+    calll *%eax
+    movl %ebp,%esp          // would like to use "leave" but it's slower
+    popl %ebp
+    ret
+

src/rt/arch/x64/context.cpp

@@ -0,0 +1,73 @@
+#include "context.h"
+
+#include "../../rust.h"
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <assert.h>
+
+extern "C" uint32_t CDECL swap_registers(registers_t *oregs,
+                                         registers_t *regs)
+  asm ("swap_registers");
+
+context::context()
+{
+    assert((void*)&regs == (void*)this);
+}
+
+void context::swap(context &out)
+{
+  swap_registers(&out.regs, &regs);
+}
+
+void context::call(void *f, void *arg, void *stack) {
+  // Get the current context, which we will then modify to call the
+  // given function.
+  swap(*this);
+
+  // set up the trampoline frame
+  uint32_t *sp = (uint32_t *)stack;
+
+  // Shift the stack pointer so the alignment works out right.
+  sp = align_down(sp) - 3;
+  *--sp = (uint32_t)arg;
+  *--sp = 0xdeadbeef;
+
+  regs.esp = (uint32_t)sp;
+  regs.eip = (uint32_t)f;
+}
+
+#if 0
+// This is some useful code to check how the registers struct got
+// layed out in memory.
+int main() {
+  registers_t regs;
+
+  printf("Register offsets\n");
+
+#define REG(r) \
+  printf("  %6s: +%ld\n", #r, (intptr_t)&regs.r - (intptr_t)&regs);
+
+  REG(eax);
+  REG(ebx);
+  REG(ecx);
+  REG(edx);
+  REG(ebp);
+  REG(esi);
+  REG(edi);
+  REG(esp);
+
+  REG(cs);
+  REG(ds);
+  REG(ss);
+  REG(es);
+  REG(fs);
+  REG(gs);
+
+  REG(eflags);
+
+  REG(eip);
+
+  return 0;
+}
+#endif

src/rt/arch/x64/context.h

@@ -0,0 +1,62 @@
+// -*- mode: c++ -*-
+
+#ifndef CONTEXT_H
+#define CONTEXT_H
+
+#include <cstdlib>
+#include <inttypes.h>
+#include <stdint.h>
+
+#ifdef HAVE_VALGRIND
+#include <valgrind/memcheck.h>
+#endif
+
+template<typename T>
+T align_down(T sp)
+{
+    // There is no platform we care about that needs more than a
+    // 16-byte alignment.
+    return (T)((uint32_t)sp & ~(16 - 1));
+}
+
+struct registers_t {
+  // general purpose registers
+  uint32_t eax, ebx, ecx, edx, ebp, esi, edi, esp;
+
+  // segment registers
+  uint16_t cs, ds, ss, es, fs, gs;
+
+  uint32_t eflags;
+
+  uint32_t eip;
+};
+
+class context {
+public:
+  registers_t regs;
+
+  context();
+
+  context *next;
+
+  void swap(context &out);
+  void call(void *f, void *arg, void *sp);
+  void call(void *f, void *sp);
+
+  // Note that this doesn't actually adjust esp. Instead, we adjust esp when
+  // we actually do the call. This is needed for exception safety -- if the
+  // function being called causes the task to fail, then we have to avoid
+  // leaking space on the C stack.
+  inline void *alloc_stack(size_t nbytes) {
+    uint32_t bot = regs.esp;
+    uint32_t top = align_down(bot - nbytes);
+
+#ifdef HAVE_VALGRIND
+    (void)VALGRIND_MAKE_MEM_UNDEFINED(top - 4, bot - top + 4);
+#endif
+
+    return reinterpret_cast<void *>(top);
+  }
+};
+
+#endif