1.0.3.16: experimental x86-64/darwin suport
[sbcl.git] / src / runtime / x86-64-assem.S
index 92e0716..d0cc742 100644 (file)
  */
 \f
 #define LANGUAGE_ASSEMBLY
+#include "genesis/config.h"
 #include "validate.h"
 #include "sbcl.h"
 #include "genesis/closure.h"
+#include "genesis/funcallable-instance.h"
 #include "genesis/fdefn.h"
 #include "genesis/static-symbols.h"
 #include "genesis/symbol.h"
 #define        align_16byte    4       
 #endif                 
 
+/*
+ * The assembler used for win32 doesn't like .type or .size directives,
+ * so we want to conditionally kill them out. So let's wrap them in macros
+ * that are defined to be no-ops on win32. Hopefully this still works on
+ * other platforms.
+ */
+#if !defined(LISP_FEATURE_WIN32) && !defined(LISP_FEATURE_DARWIN)
+#define TYPE(name) .type name,@function
+#define SIZE(name) .size name,.-name
+#define DOLLAR(name) $(name)
+#else
+#define TYPE(name)
+#define SIZE(name)
+#endif
+
+/*
+ * x86/darwin (as of MacOS X 10.4.5) doesn't reliably fire signal
+ * handlers (SIGTRAP or Mach exception handlers) for 0xCC, wo we have
+ * to use ud2 instead. ud2 is an undefined opcode, #x0b0f, or
+ * 0F 0B in low-endian notation, that causes SIGILL to fire. We check
+ * for this instruction in the SIGILL handler and if we see it, we
+ * advance the EIP by two bytes to skip over ud2 instruction and
+ * call sigtrap_handler. */
+#if defined(LISP_FEATURE_DARWIN)
+#define TRAP ud2
+#else
+#define TRAP int3
+#endif
+
+/*
+ * More Apple assembler hacks
+ */
+
+#if defined(LISP_FEATURE_DARWIN)
+/* global symbol x86-64 sym(%rip) hack:*/
+#define GSYM(name) name(%rip)
+#define END()
+#else
+#define GSYM(name) $name
+#define END() .end
+#endif
+
+        
        .text
-       .global GNAME(foreign_function_call_active)
-       .global GNAME(all_threads)
+        .globl  GNAME(all_threads)
+        
        
 \f
 /* From lower to higher-numbered addresses, the stack contains 
  */
        .text
        .align  align_16byte,0x90
-       .global GNAME(call_into_c)
-       .type   GNAME(call_into_c),@function
+       .globl  GNAME(call_into_c)
+       TYPE(GNAME(call_into_c))
 GNAME(call_into_c):
         /* ABI requires that the direction flag be clear on function
          * entry and exit. */
         cld
-        
        push    %rbp            # Save old frame pointer.
        mov     %rsp,%rbp       # Establish new frame.
 
@@ -80,12 +124,12 @@ GNAME(call_into_c):
        mov     %rbp,%rsp
        pop     %rbp
        ret
-       .size   GNAME(call_into_c), . - GNAME(call_into_c)
+       SIZE(GNAME(call_into_c))
 
 \f
        .text   
-       .global GNAME(call_into_lisp_first_time)
-       .type  GNAME(call_into_lisp_first_time),@function
+       .globl  GNAME(call_into_lisp_first_time)
+       TYPE(GNAME(call_into_lisp_first_time))
                
 /* The *ALIEN-STACK* pointer is set up on the first call_into_lisp when
  * the stack changes.  We don't worry too much about saving registers 
@@ -96,17 +140,17 @@ GNAME(call_into_c):
 GNAME(call_into_lisp_first_time):
        push    %rbp            # Save old frame pointer.
        mov     %rsp,%rbp       # Establish new frame.
-       mov    %rsp,ALIEN_STACK + SYMBOL_VALUE_OFFSET
-       mov    GNAME(all_threads),%rax
-       mov    THREAD_CONTROL_STACK_START_OFFSET(%rax) ,%rsp
+       mov     %rsp,ALIEN_STACK + SYMBOL_VALUE_OFFSET
+       movq    GSYM(GNAME(all_threads)),%rax
+       mov     THREAD_CONTROL_STACK_START_OFFSET(%rax) ,%rsp
        /* don't think too hard about what happens if we get interrupted
        * here */
-       add     $THREAD_CONTROL_STACK_SIZE-16,%rsp
+        add    $(THREAD_CONTROL_STACK_SIZE)-16,%rsp
        jmp     Lstack
 \f
        .text   
-       .global GNAME(call_into_lisp)
-       .type  GNAME(call_into_lisp),@function
+       .globl  GNAME(call_into_lisp)
+       TYPE(GNAME(call_into_lisp))
                
 /*
  * amd64 calling convention: C expects that
@@ -193,49 +237,49 @@ LsingleValue:
        /* return value is already in rax where lisp expects it */
        leave
        ret
-       .size   GNAME(call_into_lisp), . - GNAME(call_into_lisp)
+       SIZE(GNAME(call_into_lisp))
 \f
 /* support for saving and restoring the NPX state from C */
        .text
-       .global GNAME(fpu_save)
-       .type   GNAME(fpu_save),@function
+       .globl  GNAME(fpu_save)
+       TYPE(GNAME(fpu_save))
        .align  2,0x90
 GNAME(fpu_save):
        mov     4(%rsp),%rax
        fnsave  (%rax)          # Save the NPX state. (resets NPX)
        ret
-       .size   GNAME(fpu_save),.-GNAME(fpu_save)
+       SIZE(GNAME(fpu_save))
 
-       .global GNAME(fpu_restore)
-       .type   GNAME(fpu_restore),@function
+       .globl  GNAME(fpu_restore)
+       TYPE(GNAME(fpu_restore))
        .align  2,0x90
 GNAME(fpu_restore):
        mov     4(%rsp),%rax
        frstor  (%rax)          # Restore the NPX state.
        ret
-       .size   GNAME(fpu_restore),.-GNAME(fpu_restore)
+       SIZE(GNAME(fpu_restore))
 \f
 /*
  * the undefined-function trampoline
  */
        .text
        .align  align_8byte,0x90
-       .global GNAME(undefined_tramp)
-       .type   GNAME(undefined_tramp),@function
+       .globl  GNAME(undefined_tramp)
+       TYPE(GNAME(undefined_tramp))
 GNAME(undefined_tramp):
-       int3
+       TRAP
        .byte   trap_Error
         .byte   2
         .byte   UNDEFINED_FUN_ERROR
         .byte   sc_DescriptorReg # eax in the Descriptor-reg SC
        ret
-       .size   GNAME(undefined_tramp), .-GNAME(undefined_tramp)
+       SIZE(GNAME(undefined_tramp))
 
 
        .text
        .align  align_8byte,0x90
-       .global GNAME(alloc_tramp)
-       .type   GNAME(alloc_tramp),@function
+       .globl  GNAME(alloc_tramp)
+       TYPE(GNAME(alloc_tramp))
 GNAME(alloc_tramp):
        push    %rbp            # Save old frame pointer.
        mov     %rsp,%rbp       # Establish new frame.
@@ -249,7 +293,7 @@ GNAME(alloc_tramp):
        push    %r10
        push    %r11
        mov     16(%rbp),%rdi   
-       call    alloc
+       call    GNAME(alloc)
        mov     %rax,16(%rbp)
        pop     %r11
        pop     %r10
@@ -262,7 +306,7 @@ GNAME(alloc_tramp):
        pop     %rax
        pop     %rbp
        ret
-       .size   GNAME(alloc_tramp),.-GNAME(alloc_tramp)
+       SIZE(GNAME(alloc_tramp))
 
                
 /*
@@ -270,8 +314,8 @@ GNAME(alloc_tramp):
  */
        .text
        .align  align_8byte,0x90
-       .global GNAME(closure_tramp)
-       .type   GNAME(closure_tramp),@function
+       .globl  GNAME(closure_tramp)
+       TYPE(GNAME(closure_tramp))
 GNAME(closure_tramp):
        mov     FDEFN_FUN_OFFSET(%rax),%rax
        /* FIXME: The '*' after "jmp" in the next line is from PVE's
@@ -281,13 +325,27 @@ GNAME(closure_tramp):
         * right. It would be good to find a way to force the flow of
         * control through here to test it. */
        jmp     *CLOSURE_FUN_OFFSET(%rax)
-       .size   GNAME(closure_tramp), .-GNAME(closure_tramp)
+       SIZE(GNAME(closure_tramp))
 
+       .text
+       .align  align_8byte,0x90
+       .globl  GNAME(funcallable_instance_tramp)
+#if !defined(LISP_FEATURE_DARWIN)
+        .type  GNAME(funcallable_instance_tramp),@function
+#endif
+        GNAME(funcallable_instance_tramp):
+       mov     FUNCALLABLE_INSTANCE_FUNCTION_OFFSET(%rax),%rax
+       /* KLUDGE: on this platform, whatever kind of function is in %rax
+        * now, the first word of it contains the address to jump to. */
+       jmp     *CLOSURE_FUN_OFFSET(%rax)
+#if !defined(LISP_FEATURE_DARWIN)
+       .size   GNAME(funcallable_instance_tramp), .-GNAME(funcallable_instance_tramp)
+#endif
 /*
  * fun-end breakpoint magic
  */
        .text
-       .global GNAME(fun_end_breakpoint_guts)
+       .globl  GNAME(fun_end_breakpoint_guts)
        .align  align_8byte
 GNAME(fun_end_breakpoint_guts):
        /* Multiple Value return */
@@ -298,32 +356,36 @@ GNAME(fun_end_breakpoint_guts):
        mov     %rsp,%rbx       # Setup ebx - the ofp.
        sub     $8,%rsp         # Allocate one stack slot for the return value
        mov     $8,%rcx         # Setup ecx for one return value.
+#if defined(LISP_FEATURE_DARWIN)
+       mov     GSYM(NIL),%rdi  # default second value
+       mov     GSYM(NIL),%rsi  # default third value
+#else
        mov     $NIL,%rdi       # default second value
        mov     $NIL,%rsi       # default third value
-               
+#endif
 multiple_value_return:
        
-       .global GNAME(fun_end_breakpoint_trap)
+       .globl  GNAME(fun_end_breakpoint_trap)
 GNAME(fun_end_breakpoint_trap):
-       int3
+       TRAP
        .byte   trap_FunEndBreakpoint
        hlt                     # We should never return here.
 
-       .global GNAME(fun_end_breakpoint_end)
+       .globl  GNAME(fun_end_breakpoint_end)
 GNAME(fun_end_breakpoint_end):
 
 \f
-       .global GNAME(do_pending_interrupt)
-       .type   GNAME(do_pending_interrupt),@function
+       .globl  GNAME(do_pending_interrupt)
+       TYPE(GNAME(do_pending_interrupt))
        .align  align_8byte,0x90
 GNAME(do_pending_interrupt):
-       int3
+       TRAP
        .byte   trap_PendingInterrupt
        ret
-       .size   GNAME(do_pending_interrupt),.-GNAME(do_pending_interrupt)
+       SIZE(GNAME(do_pending_interrupt))
 \f
        .globl  GNAME(post_signal_tramp)
-       .type   GNAME(post_signal_tramp),@function
+       TYPE(GNAME(post_signal_tramp))
        .align  align_8byte,0x90
 GNAME(post_signal_tramp):
        /* this is notionally the second half of a function whose first half
@@ -347,12 +409,12 @@ GNAME(post_signal_tramp):
         popfq
        leave
        ret
-       .size GNAME(post_signal_tramp),.-GNAME(post_signal_tramp)
+       SIZE(GNAME(post_signal_tramp))
 \f
        .text
        .align  align_8byte,0x90
-       .global GNAME(fast_bzero)
-       .type   GNAME(fast_bzero),@function
+       .globl  GNAME(fast_bzero)
+       TYPE(GNAME(fast_bzero))
         
 GNAME(fast_bzero):
         /* A fast routine for zero-filling blocks of memory that are
@@ -364,7 +426,7 @@ GNAME(fast_bzero):
         movups %xmm7, -16(%rsp)   /* Save XMM register */
         xorps  %xmm7, %xmm7       /* Zero the XMM register */
         jmp Lloop
-        .align 16                 
+        .align align_16byte                 
 Lloop:
 
         /* Copy the 16 zeroes from xmm7 to memory, 4 times. MOVNTDQ is the
@@ -387,6 +449,6 @@ Lloop:
                                  * since it's likely to be used immediately. */
 Lend:        
        ret
-       .size   GNAME(fast_bzero), .-GNAME(fast_bzero)
+       SIZE(GNAME(fast_bzero))
 
-        .end
+        END()