0.9.17.8:
[sbcl.git] / src / runtime / x86-assem.S
index 2069f96..551841a 100644 (file)
@@ -62,7 +62,7 @@
  * that are defined to be no-ops on win32. Hopefully this still works on
  * other platforms.
  */
-#ifndef LISP_FEATURE_WIN32
+#if !defined(LISP_FEATURE_WIN32) && !defined(LISP_FEATURE_DARWIN)
 #define TYPE(name) .type name,@function
 #define SIZE(name) .size name,.-name
 #else
 #define SIZE(name)
 #endif
 
+/*
+ * x86/darwin (as of MacOS X 10.4.5) doesn't reliably file 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 END()
+#define TRAP ud2
+#else
+#define END() .end
+#define TRAP int3
+#endif
+
        .text
-       .global GNAME(foreign_function_call_active)
-       .global GNAME(all_threads)
+       .globl  GNAME(foreign_function_call_active)
+       .globl  GNAME(all_threads)
 \f
 /*
  * A call to call_into_c preserves esi, edi, and ebp.  
  */
        .text
        .align  align_16byte,0x90
-       .global GNAME(call_into_c)
+       .globl GNAME(call_into_c)
        TYPE(GNAME(call_into_c))
 GNAME(call_into_c):
        movl    $1,GNAME(foreign_function_call_active)
@@ -109,6 +125,9 @@ GNAME(call_into_c):
        cld
 #endif
 
+#ifdef LISP_FEATURE_DARWIN
+        andl   $0xfffffff0,%esp        # align stack to 16-byte boundary before calling C
+#endif
        call    *%eax           # normal callout using Lisp stack
 
        movl    %eax,%ecx       # remember integer return value
@@ -160,7 +179,7 @@ Lfp_rtn_value:
 
 \f
        .text   
-       .global GNAME(call_into_lisp_first_time)
+       .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
@@ -178,7 +197,7 @@ GNAME(call_into_lisp_first_time):
        movl    THREAD_CONTROL_STACK_START_OFFSET(%eax) ,%esp
        /* don't think too hard about what happens if we get interrupted
        * here */
-       addl    $THREAD_CONTROL_STACK_SIZE-4,%esp
+       addl    $(THREAD_CONTROL_STACK_SIZE),%esp
 #else
 /* Win32 -really- doesn't like you switching stacks out from under it. */
        movl    GNAME(all_threads),%eax
@@ -186,7 +205,7 @@ GNAME(call_into_lisp_first_time):
        jmp     Lstack
 \f
        .text   
-       .global GNAME(call_into_lisp)
+       .globl GNAME(call_into_lisp)
        TYPE(GNAME(call_into_lisp))
                
 /* The C conventions require that ebx, esi, edi, and ebp be preserved
@@ -264,7 +283,9 @@ Ldone:
        
        /* If the function returned multiple values, it will return to
           this point.  Lose them */
+       jnc     LsingleValue
        mov     %ebx, %esp
+LsingleValue:
        /* A singled value function returns here */
 
 /* Restore the stack, in case there was a stack change. */
@@ -286,7 +307,7 @@ Ldone:
 \f
 /* support for saving and restoring the NPX state from C */
        .text
-       .global GNAME(fpu_save)
+       .globl  GNAME(fpu_save)
        TYPE(GNAME(fpu_save))
        .align  2,0x90
 GNAME(fpu_save):
@@ -295,7 +316,7 @@ GNAME(fpu_save):
        ret
        SIZE(GNAME(fpu_save))
 
-       .global GNAME(fpu_restore)
+       .globl  GNAME(fpu_restore)
        TYPE(GNAME(fpu_restore))
        .align  2,0x90
 GNAME(fpu_restore):
@@ -309,11 +330,11 @@ GNAME(fpu_restore):
  */
        .text
        .align  align_4byte,0x90
-       .global GNAME(undefined_tramp)
+       .globl GNAME(undefined_tramp)
        TYPE(GNAME(undefined_tramp))
         .byte   0, 0, 0, SIMPLE_FUN_HEADER_WIDETAG
 GNAME(undefined_tramp):
-       int3
+       TRAP
        .byte   trap_Error
         .byte   2
         .byte   UNDEFINED_FUN_ERROR
@@ -326,7 +347,7 @@ GNAME(undefined_tramp):
  */
        .text
        .align  align_4byte,0x90
-       .global GNAME(closure_tramp)
+       .globl GNAME(closure_tramp)
        TYPE(GNAME(closure_tramp))
         .byte   0, 0, 0, SIMPLE_FUN_HEADER_WIDETAG
 GNAME(closure_tramp):
@@ -340,41 +361,52 @@ GNAME(closure_tramp):
        jmp     *CLOSURE_FUN_OFFSET(%eax)
        SIZE(GNAME(closure_tramp))
 
+       .text
+       .align  align_4byte,0x90
+       .global GNAME(funcallable_instance_tramp)
+       .type   GNAME(funcallable_instance_tramp),@function
+GNAME(funcallable_instance_tramp):
+       movl    FUNCALLABLE_INSTANCE_FUNCTION_OFFSET(%eax),%eax 
+       /* 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(%eax)
+       .size   GNAME(funcallable_instance_tramp), .-GNAME(funcallable_instance_tramp)
+       
 /*
  * fun-end breakpoint magic
  */
        .text
-       .global GNAME(fun_end_breakpoint_guts)
+       .globl  GNAME(fun_end_breakpoint_guts)
        .align  align_4byte
 GNAME(fun_end_breakpoint_guts):
        /* Multiple Value return */
-       jmp     multiple_value_return
+       jc      multiple_value_return
        /* Single value return: The eventual return will now use the
           multiple values return convention but with a return values
           count of one. */
        movl    %esp,%ebx       # Setup ebx - the ofp.
        subl    $4,%esp         # Allocate one stack slot for the return value
        movl    $4,%ecx         # Setup ecx for one return value.
-       movl    $NIL,%edi       # default second value
-       movl    $NIL,%esi       # default third value
+       movl    $(NIL),%edi     # default second value
+       movl    $(NIL),%esi     # default third value
                
 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)
+       .globl  GNAME(do_pending_interrupt)
        TYPE(GNAME(do_pending_interrupt))
        .align  align_4byte,0x90
 GNAME(do_pending_interrupt):
-       int3
+       TRAP
        .byte   trap_PendingInterrupt
        ret
        SIZE(GNAME(do_pending_interrupt))
@@ -811,6 +843,57 @@ GNAME(alloc_overflow_edi):
         ret
         SIZE(GNAME(alloc_overflow_edi))
 
+
+#ifdef LISP_FEATURE_DARWIN
+        .align align_4byte
+        .globl GNAME(call_into_lisp_tramp)
+        TYPE(GNAME(call_into_lisp_tramp))
+GNAME(call_into_lisp_tramp):
+        /* 1. build the stack frame from the block that's pointed to by ECX
+           2. free the block
+           3. set ECX to 0
+           4. call the function via call_into_lisp
+        */
+        pushl   0(%ecx)          /* return address */
+
+        pushl   %ebp
+        movl    %esp, %ebp
+
+        pushl   32(%ecx)         /* eflags */
+        pushl   28(%ecx)         /* EAX */
+        pushl   20(%ecx)         /* ECX */
+        pushl   16(%ecx)         /* EDX */
+        pushl   24(%ecx)         /* EBX */
+        pushl   $0                /* popal is going to ignore esp */
+        pushl   %ebp              /* is this right?? */
+        pushl   12(%ecx)         /* ESI */
+        pushl   8(%ecx)          /* EDI */
+        pushl   $0                /* args for call_into_lisp */
+        pushl   $0
+        pushl   4(%ecx)          /* function to call */
+
+        /* free our save block */
+        pushl   %ecx              /* reserve sufficient space on stack for args */
+        pushl   %ecx
+        andl    $0xfffffff0, %esp  /* align stack */
+        movl    $0x40, 4(%esp)
+        movl    %ecx, (%esp)
+        call    GNAME(os_invalidate)
+
+        /* call call_into_lisp */
+        leal    -48(%ebp), %esp
+        call    GNAME(call_into_lisp)
+
+        /* Clean up our mess */
+        leal    -36(%ebp), %esp
+        popal
+        popfl
+        leave
+        ret
+        
+        SIZE(call_into_lisp_tramp)
+#endif
+        
        .align  align_4byte,0x90
        .globl  GNAME(post_signal_tramp)
        TYPE(GNAME(post_signal_tramp))
@@ -821,6 +904,10 @@ GNAME(post_signal_tramp):
        addl $12,%esp   /* clear call_into_lisp args from stack */
        popal           /* restore registers */
         popfl
+#ifdef LISP_FEATURE_DARWIN
+        /* skip two padding words */
+        addl $8,%esp
+#endif
        leave
        ret
        SIZE(GNAME(post_signal_tramp))
@@ -830,7 +917,7 @@ GNAME(post_signal_tramp):
         * This is part of the funky magic for exception handling on win32.
         * see sigtrap_emulator() in win32-os.c for details.
         */
-       .global GNAME(sigtrap_trampoline)
+       .globl GNAME(sigtrap_trampoline)
 GNAME(sigtrap_trampoline):
        pushl   %eax
        pushl   %ebp
@@ -838,7 +925,7 @@ GNAME(sigtrap_trampoline):
        call    GNAME(sigtrap_wrapper)
        pop     %eax
        pop     %eax
-       int3
+       TRAP
        .byte   trap_ContextRestore
        hlt                     # We should never return here.
         
@@ -846,7 +933,7 @@ GNAME(sigtrap_trampoline):
         * This is part of the funky magic for exception handling on win32.
         * see handle_exception() in win32-os.c for details.
         */
-       .global GNAME(exception_trampoline)
+       .globl GNAME(exception_trampoline)
 GNAME(exception_trampoline):
        pushl   %eax
        pushl   %ebp
@@ -854,7 +941,7 @@ GNAME(exception_trampoline):
        call    GNAME(handle_win32_exception_wrapper)
        pop     %eax
        pop     %eax
-       int3
+       TRAP
        .byte   trap_ContextRestore
        hlt                     # We should never return here.
 #endif
@@ -863,9 +950,9 @@ GNAME(exception_trampoline):
          * to use.
          */
 \f
-       .global GNAME(fast_bzero_pointer)
+       .globl GNAME(fast_bzero_pointer)
        .data
-        .align 4
+        .align  align_4byte
 GNAME(fast_bzero_pointer):
         /* Variable containing a pointer to the bzero function to use.
          * Initially points to a basic function.  Change this variable
@@ -874,7 +961,7 @@ GNAME(fast_bzero_pointer):
 \f
        .text
        .align  align_8byte,0x90
-       .global GNAME(fast_bzero)
+       .globl GNAME(fast_bzero)
        TYPE(GNAME(fast_bzero))
 GNAME(fast_bzero):        
         /* Indirect function call */
@@ -884,7 +971,7 @@ GNAME(fast_bzero):
 \f      
        .text
        .align  align_8byte,0x90
-       .global GNAME(fast_bzero_detect)
+       .globl GNAME(fast_bzero_detect)
        TYPE(GNAME(fast_bzero_detect))
 GNAME(fast_bzero_detect):
         /* Decide whether to use SSE, MMX or REP version */
@@ -904,10 +991,10 @@ GNAME(fast_bzero_detect):
          * for all non-SSE2 processors.
          */
 Lbase:
-        movl $GNAME(fast_bzero_base), GNAME(fast_bzero_pointer)
+        movl $(GNAME(fast_bzero_base)), GNAME(fast_bzero_pointer)
         jmp Lrestore
 Lsse2:
-        movl $GNAME(fast_bzero_sse), GNAME(fast_bzero_pointer)
+        movl $(GNAME(fast_bzero_sse)), GNAME(fast_bzero_pointer)
         jmp Lrestore
         
 Lrestore:
@@ -922,7 +1009,7 @@ Lrestore:
 \f
        .text
        .align  align_8byte,0x90
-       .global GNAME(fast_bzero_sse)
+       .globl GNAME(fast_bzero_sse)
        TYPE(GNAME(fast_bzero_sse))
         
 GNAME(fast_bzero_sse):
@@ -938,7 +1025,7 @@ GNAME(fast_bzero_sse):
         movups %xmm7, -16(%esp)   /* Save XMM register */
         xorps  %xmm7, %xmm7       /* Zero the XMM register */
         jmp Lloop_sse
-        .align 16
+        .align align_16byte
 Lloop_sse:
 
         /* Copy the 16 zeroes from xmm7 to memory, 4 times. MOVNTDQ is the
@@ -968,7 +1055,7 @@ Lend_sse:
 \f
        .text
        .align  align_8byte,0x90
-       .global GNAME(fast_bzero_base)
+       .globl GNAME(fast_bzero_base)
        TYPE(GNAME(fast_bzero_base))
         
 GNAME(fast_bzero_base):
@@ -984,8 +1071,11 @@ GNAME(fast_bzero_base):
         shr $2, %ecx              /* Amount of 4-byte blocks to copy */
         jz  Lend_base
         cld                       /* Set direction of STOSL to increment */
-        rep stosl                 /* Store EAX to *EDI, ECX times, incrementing
+
+        rep
+        stosl                     /* Store EAX to *EDI, ECX times, incrementing
                                    * EDI by 4 after each store */
+        
 Lend_base:        
         pop %edi                  /* Restore temp registers */
         pop %ecx
@@ -994,4 +1084,5 @@ Lend_base:
        SIZE(GNAME(fast_bzero_base))
         
 \f      
-       .end
+        END()
+        
\ No newline at end of file