/* -----------------------------------------------------------------------------
* Foreign export adjustor thunks
*
* Copyright (c) 1998.
*
* ---------------------------------------------------------------------------*/
/* A little bit of background...
An adjustor thunk is a dynamically allocated code snippet that allows
Haskell closures to be viewed as C function pointers.
Stable pointers provide a way for the outside world to get access to,
and evaluate, Haskell heap objects, with the RTS providing a small
range of ops for doing so. So, assuming we've got a stable pointer in
our hand in C, we can jump into the Haskell world and evaluate a callback
procedure, say. This works OK in some cases where callbacks are used, but
does require the external code to know about stable pointers and how to deal
with them. We'd like to hide the Haskell-nature of a callback and have it
be invoked just like any other C function pointer.
Enter adjustor thunks. An adjustor thunk is a little piece of code
that's generated on-the-fly (one per Haskell closure being exported)
that, when entered using some 'universal' calling convention (e.g., the
C calling convention on platform X), pushes an implicit stable pointer
(to the Haskell callback) before calling another (static) C function stub
which takes care of entering the Haskell code via its stable pointer.
An adjustor thunk is allocated on the C heap, and is called from within
Haskell just before handing out the function pointer to the Haskell (IO)
action. User code should never have to invoke it explicitly.
An adjustor thunk differs from a C function pointer in one respect: when
the code is through with it, it has to be freed in order to release Haskell
and C resources. Failure to do so will result in memory leaks on both the C and
Haskell side.
*/
#include "PosixSource.h"
#include "Rts.h"
#include "RtsUtils.h"
#include "Stable.h"
#if defined(USE_LIBFFI_FOR_ADJUSTORS)
#include "ffi.h"
#include <string.h>
#endif
#if defined(i386_HOST_ARCH)
extern void adjustorCode(void);
#elif defined(powerpc_HOST_ARCH) || defined(powerpc64_HOST_ARCH)
// from AdjustorAsm.s
// not declared as a function so that AIX-style
// fundescs can never get in the way.
extern void *adjustorCode;
#endif
#if defined(USE_LIBFFI_FOR_ADJUSTORS)
/* There are subtle differences between how libffi adjustors work on
* different platforms, and the situation is a little complex.
*
* HOW ADJUSTORS/CLOSURES WORK ON LIBFFI:
* libffi's ffi_closure_alloc() function gives you two pointers to a closure,
* 1. the writable pointer, and 2. the executable pointer. You write the
* closure into the writable pointer (and ffi_prep_closure_loc() will do this
* for you) and you execute it at the executable pointer.
*
* THE PROBLEM:
* The RTS deals only with the executable pointer, but when it comes time to
* free the closure, libffi wants the writable pointer back that it gave you
* when you allocated it.
*
* On Linux we solve this problem by storing the address of the writable
* mapping into itself, then returning both writable and executable pointers
* plus 1 machine word for preparing the closure for use by the RTS (see the
* Linux version of allocateExec() in rts/sm/Storage.c). When we want to
* recover the writable address, we subtract 1 word from the executable
* address and fetch. This works because Linux kernel magic gives us two
* pointers with different addresses that refer to the same memory. Whatever
* you write into the writeable address can be read back at the executable
* address. This method is very efficient.
*
* On iOS this breaks for two reasons: 1. the two pointers do not refer to
* the same memory (so we can't retrieve anything stored into the writable
* pointer if we only have the exec pointer), and 2. libffi's
* ffi_closure_alloc() assumes the pointer it has returned you is a
* ffi_closure structure and treats it as such: It uses that memory to
* communicate with ffi_prep_closure_loc(). On Linux by contrast
* ffi_closure_alloc() is viewed simply as a memory allocation, and only
* ffi_prep_closure_loc() deals in ffi_closure structures. Each of these
* differences is enough make the efficient way used on Linux not work on iOS.
* Instead on iOS we use hash tables to recover the writable address from the
* executable one. This method is conservative and would almost certainly work
* on any platform, but on Linux it makes sense to use the faster method.
*/
void
freeHaskellFunctionPtr(void* ptr)
{
ffi_closure *cl;
#if defined(ios_HOST_OS)
cl = execToWritable(ptr);
#else
cl = (ffi_closure*)ptr;
#endif
freeStablePtr(cl->user_data);
stgFree(cl->cif->arg_types);
stgFree(cl->cif);
freeExec(ptr);
}
static ffi_type * char_to_ffi_type(char c)
{
switch (c) {
case 'v': return &ffi_type_void;
case 'f': return &ffi_type_float;
case 'd': return &ffi_type_double;
case 'L': return &ffi_type_sint64;
case 'l': return &ffi_type_uint64;
case 'W': return &ffi_type_sint32;
case 'w': return &ffi_type_uint32;
case 'S': return &ffi_type_sint16;
case 's': return &ffi_type_uint16;
case 'B': return &ffi_type_sint8;
case 'b': return &ffi_type_uint8;
case 'p': return &ffi_type_pointer;
default: barf("char_to_ffi_type: unknown type '%c'", c);
}
}
void*
createAdjustor (int cconv,
StgStablePtr hptr,
StgFunPtr wptr,
char *typeString)
{
ffi_cif *cif;
ffi_type **arg_types;
uint32_t n_args, i;
ffi_type *result_type;
ffi_closure *cl;
int r, abi;
void *code;
n_args = strlen(typeString) - 1;
cif = stgMallocBytes(sizeof(ffi_cif), "createAdjustor");
arg_types = stgMallocBytes(n_args * sizeof(ffi_type*), "createAdjustor");
result_type = char_to_ffi_type(typeString[0]);
for (i=0; i < n_args; i++) {
arg_types[i] = char_to_ffi_type(typeString[i+1]);
}
switch (cconv) {
#if defined(mingw32_HOST_OS) && defined(i386_HOST_ARCH)
case 0: /* stdcall */
abi = FFI_STDCALL;
break;
#endif
case 1: /* ccall */
abi = FFI_DEFAULT_ABI;
break;
default:
barf("createAdjustor: convention %d not supported on this platform", cconv);
}
r = ffi_prep_cif(cif, abi, n_args, result_type, arg_types);
if (r != FFI_OK) barf("ffi_prep_cif failed: %d", r);
cl = allocateExec(sizeof(ffi_closure), &code);
if (cl == NULL) {
barf("createAdjustor: failed to allocate memory");
}
r = ffi_prep_closure_loc(cl, cif, (void*)wptr, hptr/*userdata*/, code);
if (r != FFI_OK) barf("ffi_prep_closure_loc failed: %d", r);
return (void*)code;
}
#else // To end of file...
#if defined(_WIN32)
#include <windows.h>
#endif
#if defined(powerpc_HOST_ARCH) && defined(linux_HOST_OS)
#include <string.h>
#endif
#ifdef LEADING_UNDERSCORE
#define UNDERSCORE "_"
#else
#define UNDERSCORE ""
#endif
#if defined(x86_64_HOST_ARCH)
/*
Now here's something obscure for you:
When generating an adjustor thunk that uses the C calling
convention, we have to make sure that the thunk kicks off
the process of jumping into Haskell with a tail jump. Why?
Because as a result of jumping in into Haskell we may end
up freeing the very adjustor thunk we came from using
freeHaskellFunctionPtr(). Hence, we better not return to
the adjustor code on our way out, since it could by then
point to junk.
The fix is readily at hand, just include the opcodes
for the C stack fixup code that we need to perform when
returning in some static piece of memory and arrange
to return to it before tail jumping from the adjustor thunk.
*/
static void GNUC3_ATTRIBUTE(used) obscure_ccall_wrapper(void)
{
__asm__ (
".globl " UNDERSCORE "obscure_ccall_ret_code\n"
UNDERSCORE "obscure_ccall_ret_code:\n\t"
"addq $0x8, %rsp\n\t"
#if defined(mingw32_HOST_OS)
/* On Win64, we had to put the original return address after the
arg 1-4 spill slots, ro now we have to move it back */
"movq 0x20(%rsp), %rcx\n"
"movq %rcx, (%rsp)\n"
#endif
"ret"
);
}
extern void obscure_ccall_ret_code(void);
#endif
#if defined(alpha_HOST_ARCH)
/* To get the definition of PAL_imb: */
# if defined(linux_HOST_OS)
# include <asm/pal.h>
# else
# include <machine/pal.h>
# endif
#endif
#if defined(ia64_HOST_ARCH)
/* Layout of a function descriptor */
typedef struct _IA64FunDesc {
StgWord64 ip;
StgWord64 gp;
} IA64FunDesc;
static void *
stgAllocStable(size_t size_in_bytes, StgStablePtr *stable)
{
StgArrBytes* arr;
uint32_t data_size_in_words, total_size_in_words;
/* round up to a whole number of words */
data_size_in_words = ROUNDUP_BYTES_TO_WDS(size_in_bytes);
total_size_in_words = sizeofW(StgArrBytes) + data_size_in_words;
/* allocate and fill it in */
arr = (StgArrBytes *)allocate(total_size_in_words);
SET_ARR_HDR(arr, &stg_ARR_WORDS_info, CCCS, size_in_bytes);
/* obtain a stable ptr */
*stable = getStablePtr((StgPtr)arr);
/* and return a ptr to the goods inside the array */
return(&(arr->payload));
}
#endif
#if defined(powerpc_HOST_ARCH) && defined(linux_HOST_OS)
__asm__("obscure_ccall_ret_code:\n\t"
"lwz 1,0(1)\n\t"
"lwz 0,4(1)\n\t"
"mtlr 0\n\t"
"blr");
extern void obscure_ccall_ret_code(void);
#endif
#if defined(powerpc_HOST_ARCH) || defined(powerpc64_HOST_ARCH)
#if !(defined(powerpc_HOST_ARCH) && defined(linux_HOST_OS))
/* !!! !!! WARNING: !!! !!!
* This structure is accessed from AdjustorAsm.s
* Any changes here have to be mirrored in the offsets there.
*/
typedef struct AdjustorStub {
#if defined(powerpc_HOST_ARCH) && defined(darwin_HOST_OS)
unsigned lis;
unsigned ori;
unsigned lwz;
unsigned mtctr;
unsigned bctr;
StgFunPtr code;
#elif defined(powerpc64_HOST_ARCH) && defined(darwin_HOST_OS)
/* powerpc64-darwin: just guessing that it won't use fundescs. */
unsigned lis;
unsigned ori;
unsigned rldimi;
unsigned oris;
unsigned ori2;
unsigned lwz;
unsigned mtctr;
unsigned bctr;
StgFunPtr code;
#else
/* fundesc-based ABIs */
#define FUNDESCS
StgFunPtr code;
struct AdjustorStub
*toc;
void *env;
#endif
StgStablePtr hptr;
StgFunPtr wptr;
StgInt negative_framesize;
StgInt extrawords_plus_one;
} AdjustorStub;
#endif
#endif
#if defined(i386_HOST_ARCH)
/* !!! !!! WARNING: !!! !!!
* This structure is accessed from AdjustorAsm.s
* Any changes here have to be mirrored in the offsets there.
*/
typedef struct AdjustorStub {
unsigned char call[8];
StgStablePtr hptr;
StgFunPtr wptr;
StgInt frame_size;
StgInt argument_size;
} AdjustorStub;
#endif
#if defined(i386_HOST_ARCH) || defined(powerpc_HOST_ARCH) || defined(powerpc64_HOST_ARCH)
static int totalArgumentSize(char *typeString)
{
int sz = 0;
while(*typeString)
{
char t = *typeString++;
switch(t)
{
// on 32-bit platforms, Double and Int64 occupy two words.
case 'd':
case 'l':
case 'L':
if(sizeof(void*) == 4)
{
sz += 2;
break;
}
// everything else is one word.
default:
sz += 1;
}
}
return sz;
}
#endif
void*
createAdjustor(int cconv, StgStablePtr hptr,
StgFunPtr wptr,
char *typeString
#if !defined(powerpc_HOST_ARCH) && !defined(powerpc64_HOST_ARCH) && !defined(x86_64_HOST_ARCH)
STG_UNUSED
#endif
)
{
void *adjustor = NULL;
void *code = NULL;
switch (cconv)
{
case 0: /* _stdcall */
#if defined(i386_HOST_ARCH) && !defined(darwin_HOST_OS)
/* Magic constant computed by inspecting the code length of
the following assembly language snippet
(offset and machine code prefixed):
<0>: 58 popl %eax # temp. remove ret addr..
<1>: 68 fd fc fe fa pushl 0xfafefcfd # constant is large enough to
# hold a StgStablePtr
<6>: 50 pushl %eax # put back ret. addr
<7>: b8 fa ef ff 00 movl $0x00ffeffa, %eax # load up wptr
<c>: ff e0 jmp %eax # and jump to it.
# the callee cleans up the stack
*/
adjustor = allocateExec(14,&code);
{
unsigned char *const adj_code = (unsigned char *)adjustor;
adj_code[0x00] = (unsigned char)0x58; /* popl %eax */
adj_code[0x01] = (unsigned char)0x68; /* pushl hptr (which is a dword immediate ) */
*((StgStablePtr*)(adj_code + 0x02)) = (StgStablePtr)hptr;
adj_code[0x06] = (unsigned char)0x50; /* pushl %eax */
adj_code[0x07] = (unsigned char)0xb8; /* movl $wptr, %eax */
*((StgFunPtr*)(adj_code + 0x08)) = (StgFunPtr)wptr;
adj_code[0x0c] = (unsigned char)0xff; /* jmp %eax */
adj_code[0x0d] = (unsigned char)0xe0;
}
#endif
break;
case 1: /* _ccall */
#if defined(i386_HOST_ARCH)
{
/*
Most of the trickiness here is due to the need to keep the
stack pointer 16-byte aligned (see #5250). That means we
can't just push another argument on the stack and call the
wrapper, we may have to shuffle the whole argument block.
We offload most of the work to AdjustorAsm.S.
*/
AdjustorStub *adjustorStub = allocateExec(sizeof(AdjustorStub),&code);
adjustor = adjustorStub;
int sz = totalArgumentSize(typeString);
adjustorStub->call[0] = 0xe8;
*(long*)&adjustorStub->call[1] = ((char*)&adjustorCode) - ((char*)code + 5);
adjustorStub->hptr = hptr;
adjustorStub->wptr = wptr;
// The adjustor puts the following things on the stack:
// 1.) %ebp link
// 2.) padding and (a copy of) the arguments
// 3.) a dummy argument
// 4.) hptr
// 5.) return address (for returning to the adjustor)
// All these have to add up to a multiple of 16.
// first, include everything in frame_size
adjustorStub->frame_size = sz * 4 + 16;
// align to 16 bytes
adjustorStub->frame_size = (adjustorStub->frame_size + 15) & ~15;
// only count 2.) and 3.) as part of frame_size
adjustorStub->frame_size -= 12;
adjustorStub->argument_size = sz;
}
#elif defined(x86_64_HOST_ARCH)
# if defined(mingw32_HOST_OS)
/*
stack at call:
argn
...
arg5
return address
%rcx,%rdx,%r8,%r9 = arg1..arg4
if there are <4 integer args, then we can just push the
StablePtr into %rcx and shuffle the other args up.
If there are >=4 integer args, then we have to flush one arg
to the stack, and arrange to adjust the stack ptr on return.
The stack will be rearranged to this:
argn
...
arg5
return address *** <-- dummy arg in stub fn.
arg4
obscure_ccall_ret_code
This unfortunately means that the type of the stub function
must have a dummy argument for the original return address
pointer inserted just after the 4th integer argument.
Code for the simple case:
0: 4d 89 c1 mov %r8,%r9
3: 49 89 d0 mov %rdx,%r8
6: 48 89 ca mov %rcx,%rdx
9: f2 0f 10 da movsd %xmm2,%xmm3
d: f2 0f 10 d1 movsd %xmm1,%xmm2
11: f2 0f 10 c8 movsd %xmm0,%xmm1
15: 48 8b 0d 0c 00 00 00 mov 0xc(%rip),%rcx # 28 <.text+0x28>
1c: ff 25 0e 00 00 00 jmpq *0xe(%rip) # 30 <.text+0x30>
22: 90 nop
[...]
And the version for >=4 integer arguments:
[we want to push the 4th argument (either %r9 or %xmm3, depending on
whether it is a floating arg or not) and the return address onto the
stack. However, slots 1-4 are reserved for code we call to spill its
args 1-4 into, so we can't just push them onto the bottom of the stack.
So first put the 4th argument onto the stack, above what will be the
spill slots.]
0: 48 83 ec 08 sub $0x8,%rsp
[if non-floating arg, then do this:]
4: 90 nop
5: 4c 89 4c 24 20 mov %r9,0x20(%rsp)
[else if floating arg then do this:]
4: f2 0f 11 5c 24 20 movsd %xmm3,0x20(%rsp)
[end if]
[Now push the new return address onto the stack]
a: ff 35 30 00 00 00 pushq 0x30(%rip) # 40 <.text+0x40>
[But the old return address has been moved up into a spill slot, so
we need to move it above them]
10: 4c 8b 4c 24 10 mov 0x10(%rsp),%r9
15: 4c 89 4c 24 30 mov %r9,0x30(%rsp)
[Now we do the normal register shuffle-up etc]
1a: 4d 89 c1 mov %r8,%r9
1d: 49 89 d0 mov %rdx,%r8
20: 48 89 ca mov %rcx,%rdx
23: f2 0f 10 da movsd %xmm2,%xmm3
27: f2 0f 10 d1 movsd %xmm1,%xmm2
2b: f2 0f 10 c8 movsd %xmm0,%xmm1
2f: 48 8b 0d 12 00 00 00 mov 0x12(%rip),%rcx # 48 <.text+0x48>
36: ff 25 14 00 00 00 jmpq *0x14(%rip) # 50 <.text+0x50>
3c: 90 nop
3d: 90 nop
3e: 90 nop
3f: 90 nop
[...]
*/
{
StgWord8 *adj_code;
// determine whether we have 4 or more integer arguments,
// and therefore need to flush one to the stack.
if ((typeString[0] == '\0') ||
(typeString[1] == '\0') ||
(typeString[2] == '\0') ||
(typeString[3] == '\0')) {
adjustor = allocateExec(0x38,&code);
adj_code = (StgWord8*)adjustor;
*(StgInt32 *)adj_code = 0x49c1894d;
*(StgInt32 *)(adj_code+0x4) = 0x8948d089;
*(StgInt32 *)(adj_code+0x8) = 0x100ff2ca;
*(StgInt32 *)(adj_code+0xc) = 0x100ff2da;
*(StgInt32 *)(adj_code+0x10) = 0x100ff2d1;
*(StgInt32 *)(adj_code+0x14) = 0x0d8b48c8;
*(StgInt32 *)(adj_code+0x18) = 0x0000000c;
*(StgInt32 *)(adj_code+0x1c) = 0x000e25ff;
*(StgInt32 *)(adj_code+0x20) = 0x00000000;
*(StgInt64 *)(adj_code+0x28) = (StgInt64)hptr;
*(StgInt64 *)(adj_code+0x30) = (StgInt64)wptr;
}
else
{
int fourthFloating;
fourthFloating = (typeString[3] == 'f' || typeString[3] == 'd');
adjustor = allocateExec(0x58,&code);
adj_code = (StgWord8*)adjustor;
*(StgInt32 *)adj_code = 0x08ec8348;
*(StgInt32 *)(adj_code+0x4) = fourthFloating ? 0x5c110ff2
: 0x4c894c90;
*(StgInt32 *)(adj_code+0x8) = 0x35ff2024;
*(StgInt32 *)(adj_code+0xc) = 0x00000030;
*(StgInt32 *)(adj_code+0x10) = 0x244c8b4c;
*(StgInt32 *)(adj_code+0x14) = 0x4c894c10;
*(StgInt32 *)(adj_code+0x18) = 0x894d3024;
*(StgInt32 *)(adj_code+0x1c) = 0xd08949c1;
*(StgInt32 *)(adj_code+0x20) = 0xf2ca8948;
*(StgInt32 *)(adj_code+0x24) = 0xf2da100f;
*(StgInt32 *)(adj_code+0x28) = 0xf2d1100f;
*(StgInt32 *)(adj_code+0x2c) = 0x48c8100f;
*(StgInt32 *)(adj_code+0x30) = 0x00120d8b;
*(StgInt32 *)(adj_code+0x34) = 0x25ff0000;
*(StgInt32 *)(adj_code+0x38) = 0x00000014;
*(StgInt32 *)(adj_code+0x3c) = 0x90909090;
*(StgInt64 *)(adj_code+0x40) = (StgInt64)obscure_ccall_ret_code;
*(StgInt64 *)(adj_code+0x48) = (StgInt64)hptr;
*(StgInt64 *)(adj_code+0x50) = (StgInt64)wptr;
}
}
# else
/*
stack at call:
argn
...
arg7
return address
%rdi,%rsi,%rdx,%rcx,%r8,%r9 = arg1..arg6
if there are <6 integer args, then we can just push the
StablePtr into %edi and shuffle the other args up.
If there are >=6 integer args, then we have to flush one arg
to the stack, and arrange to adjust the stack ptr on return.
The stack will be rearranged to this:
argn
...
arg7
return address *** <-- dummy arg in stub fn.
arg6
obscure_ccall_ret_code
This unfortunately means that the type of the stub function
must have a dummy argument for the original return address
pointer inserted just after the 6th integer argument.
Code for the simple case:
0: 4d 89 c1 mov %r8,%r9
3: 49 89 c8 mov %rcx,%r8
6: 48 89 d1 mov %rdx,%rcx
9: 48 89 f2 mov %rsi,%rdx
c: 48 89 fe mov %rdi,%rsi
f: 48 8b 3d 0a 00 00 00 mov 10(%rip),%rdi
16: ff 25 0c 00 00 00 jmpq *12(%rip)
...
20: .quad 0 # aligned on 8-byte boundary
28: .quad 0 # aligned on 8-byte boundary
And the version for >=6 integer arguments:
0: 41 51 push %r9
2: ff 35 20 00 00 00 pushq 32(%rip) # 28 <ccall_adjustor+0x28>
8: 4d 89 c1 mov %r8,%r9
b: 49 89 c8 mov %rcx,%r8
e: 48 89 d1 mov %rdx,%rcx
11: 48 89 f2 mov %rsi,%rdx
14: 48 89 fe mov %rdi,%rsi
17: 48 8b 3d 12 00 00 00 mov 18(%rip),%rdi # 30 <ccall_adjustor+0x30>
1e: ff 25 14 00 00 00 jmpq *20(%rip) # 38 <ccall_adjustor+0x38>
...
28: .quad 0 # aligned on 8-byte boundary
30: .quad 0 # aligned on 8-byte boundary
38: .quad 0 # aligned on 8-byte boundary
*/
{
int i = 0;
char *c;
StgWord8 *adj_code;
// determine whether we have 6 or more integer arguments,
// and therefore need to flush one to the stack.
for (c = typeString; *c != '\0'; c++) {
if (*c != 'f' && *c != 'd') i++;
if (i == 6) break;
}
if (i < 6) {
adjustor = allocateExec(0x30,&code);
adj_code = (StgWord8*)adjustor;
*(StgInt32 *)adj_code = 0x49c1894d;
*(StgInt32 *)(adj_code+0x4) = 0x8948c889;
*(StgInt32 *)(adj_code+0x8) = 0xf28948d1;
*(StgInt32 *)(adj_code+0xc) = 0x48fe8948;
*(StgInt32 *)(adj_code+0x10) = 0x000a3d8b;
*(StgInt32 *)(adj_code+0x14) = 0x25ff0000;
*(StgInt32 *)(adj_code+0x18) = 0x0000000c;
*(StgInt64 *)(adj_code+0x20) = (StgInt64)hptr;
*(StgInt64 *)(adj_code+0x28) = (StgInt64)wptr;
}
else
{
adjustor = allocateExec(0x40,&code);
adj_code = (StgWord8*)adjustor;
*(StgInt32 *)adj_code = 0x35ff5141;
*(StgInt32 *)(adj_code+0x4) = 0x00000020;
*(StgInt32 *)(adj_code+0x8) = 0x49c1894d;
*(StgInt32 *)(adj_code+0xc) = 0x8948c889;
*(StgInt32 *)(adj_code+0x10) = 0xf28948d1;
*(StgInt32 *)(adj_code+0x14) = 0x48fe8948;
*(StgInt32 *)(adj_code+0x18) = 0x00123d8b;
*(StgInt32 *)(adj_code+0x1c) = 0x25ff0000;
*(StgInt32 *)(adj_code+0x20) = 0x00000014;
*(StgInt64 *)(adj_code+0x28) = (StgInt64)obscure_ccall_ret_code;
*(StgInt64 *)(adj_code+0x30) = (StgInt64)hptr;
*(StgInt64 *)(adj_code+0x38) = (StgInt64)wptr;
}
}
# endif
#elif defined(sparc_HOST_ARCH)
/* Magic constant computed by inspecting the code length of the following
assembly language snippet (offset and machine code prefixed):
<00>: 9C23A008 sub %sp, 8, %sp ! make room for %o4/%o5 in caller's frame
<04>: DA23A060 st %o5, [%sp + 96] ! shift registers by 2 positions
<08>: D823A05C st %o4, [%sp + 92]
<0C>: 9A10000B mov %o3, %o5
<10>: 9810000A mov %o2, %o4
<14>: 96100009 mov %o1, %o3
<18>: 94100008 mov %o0, %o2
<1C>: 13000000 sethi %hi(wptr), %o1 ! load up wptr (1 of 2)
<20>: 11000000 sethi %hi(hptr), %o0 ! load up hptr (1 of 2)
<24>: 81C26000 jmp %o1 + %lo(wptr) ! jump to wptr (load 2 of 2)
<28>: 90122000 or %o0, %lo(hptr), %o0 ! load up hptr (2 of 2, delay slot)
<2C> 00000000 ! place for getting hptr back easily
ccall'ing on SPARC is easy, because we are quite lucky to push a
multiple of 8 bytes (1 word hptr + 1 word dummy arg) in front of the
existing arguments (note that %sp must stay double-word aligned at
all times, see ABI spec at http://www.sparc.org/standards/psABI3rd.pdf).
To do this, we extend the *caller's* stack frame by 2 words and shift
the output registers used for argument passing (%o0 - %o5, we are a *leaf*
procedure because of the tail-jump) by 2 positions. This makes room in
%o0 and %o1 for the additinal arguments, namely hptr and a dummy (used
for destination addr of jump on SPARC, return address on x86, ...). This
shouldn't cause any problems for a C-like caller: alloca is implemented
similarly, and local variables should be accessed via %fp, not %sp. In a
nutshell: This should work! (Famous last words! :-)
*/
adjustor = allocateExec(4*(11+1),&code);
{
unsigned long *const adj_code = (unsigned long *)adjustor;
adj_code[ 0] = 0x9C23A008UL; /* sub %sp, 8, %sp */
adj_code[ 1] = 0xDA23A060UL; /* st %o5, [%sp + 96] */
adj_code[ 2] = 0xD823A05CUL; /* st %o4, [%sp + 92] */
adj_code[ 3] = 0x9A10000BUL; /* mov %o3, %o5 */
adj_code[ 4] = 0x9810000AUL; /* mov %o2, %o4 */
adj_code[ 5] = 0x96100009UL; /* mov %o1, %o3 */
adj_code[ 6] = 0x94100008UL; /* mov %o0, %o2 */
adj_code[ 7] = 0x13000000UL; /* sethi %hi(wptr), %o1 */
adj_code[ 7] |= ((unsigned long)wptr) >> 10;
adj_code[ 8] = 0x11000000UL; /* sethi %hi(hptr), %o0 */
adj_code[ 8] |= ((unsigned long)hptr) >> 10;
adj_code[ 9] = 0x81C26000UL; /* jmp %o1 + %lo(wptr) */
adj_code[ 9] |= ((unsigned long)wptr) & 0x000003FFUL;
adj_code[10] = 0x90122000UL; /* or %o0, %lo(hptr), %o0 */
adj_code[10] |= ((unsigned long)hptr) & 0x000003FFUL;
adj_code[11] = (unsigned long)hptr;
/* flush cache */
asm("flush %0" : : "r" (adj_code ));
asm("flush %0" : : "r" (adj_code + 2));
asm("flush %0" : : "r" (adj_code + 4));
asm("flush %0" : : "r" (adj_code + 6));
asm("flush %0" : : "r" (adj_code + 10));
/* max. 5 instructions latency, and we need at >= 1 for returning */
asm("nop");
asm("nop");
asm("nop");
asm("nop");
}
#elif defined(alpha_HOST_ARCH)
/* Magic constant computed by inspecting the code length of
the following assembly language snippet
(offset and machine code prefixed; note that the machine code
shown is longwords stored in little-endian order):
<00>: 46520414 mov a2, a4
<04>: 46100412 mov a0, a2
<08>: a61b0020 ldq a0, 0x20(pv) # load up hptr
<0c>: 46730415 mov a3, a5
<10>: a77b0028 ldq pv, 0x28(pv) # load up wptr
<14>: 46310413 mov a1, a3
<18>: 6bfb---- jmp (pv), <hint> # jump to wptr (with hint)
<1c>: 00000000 # padding for alignment
<20>: [8 bytes for hptr quadword]
<28>: [8 bytes for wptr quadword]
The "computed" jump at <08> above is really a jump to a fixed
location. Accordingly, we place an always-correct hint in the
jump instruction, namely the address offset from <0c> to wptr,
divided by 4, taking the lowest 14 bits.
We only support passing 4 or fewer argument words, for the same
reason described under sparc_HOST_ARCH above by JRS, 21 Aug 01.
On the Alpha the first 6 integer arguments are in a0 through a5,
and the rest on the stack. Hence we want to shuffle the original
caller's arguments by two.
On the Alpha the calling convention is so complex and dependent
on the callee's signature -- for example, the stack pointer has
to be a multiple of 16 -- that it seems impossible to me [ccshan]
to handle the general case correctly without changing how the
adjustor is called from C. For now, our solution of shuffling
registers only and ignoring the stack only works if the original
caller passed 4 or fewer argument words.
TODO: Depending on how much allocation overhead stgMallocBytes uses for
header information (more precisely, if the overhead is no more than
4 bytes), we should move the first three instructions above down by
4 bytes (getting rid of the nop), hence saving memory. [ccshan]
*/
ASSERT(((StgWord64)wptr & 3) == 0);
adjustor = allocateExec(48,&code);
{
StgWord64 *const code = (StgWord64 *)adjustor;
code[0] = 0x4610041246520414L;
code[1] = 0x46730415a61b0020L;
code[2] = 0x46310413a77b0028L;
code[3] = 0x000000006bfb0000L
| (((StgWord32*)(wptr) - (StgWord32*)(code) - 3) & 0x3fff);
code[4] = (StgWord64)hptr;
code[5] = (StgWord64)wptr;
/* Ensure that instruction cache is consistent with our new code */
__asm__ volatile("call_pal %0" : : "i" (PAL_imb));
}
#elif defined(powerpc_HOST_ARCH) && defined(linux_HOST_OS)
#define OP_LO(op,lo) ((((unsigned)(op)) << 16) | (((unsigned)(lo)) & 0xFFFF))
#define OP_HI(op,hi) ((((unsigned)(op)) << 16) | (((unsigned)(hi)) >> 16))
{
/* The PowerPC Linux (32-bit) calling convention is annoyingly complex.
We need to calculate all the details of the stack frame layout,
taking into account the types of all the arguments, and then
generate code on the fly. */
int src_gpr = 3, dst_gpr = 5;
int fpr = 3;
int src_offset = 0, dst_offset = 0;
int n = strlen(typeString),i;
int src_locs[n], dst_locs[n];
int frameSize;
unsigned *code;
/* Step 1:
Calculate where the arguments should go.
src_locs[] will contain the locations of the arguments in the
original stack frame passed to the adjustor.
dst_locs[] will contain the locations of the arguments after the
adjustor runs, on entry to the wrapper proc pointed to by wptr.
This algorithm is based on the one described on page 3-19 of the
System V ABI PowerPC Processor Supplement.
*/
for(i=0;typeString[i];i++)
{
char t = typeString[i];
if((t == 'f' || t == 'd') && fpr <= 8)
src_locs[i] = dst_locs[i] = -32-(fpr++);
else
{
if((t == 'l' || t == 'L') && src_gpr <= 9)
{
if((src_gpr & 1) == 0)
src_gpr++;
src_locs[i] = -src_gpr;
src_gpr += 2;
}
else if((t == 'w' || t == 'W') && src_gpr <= 10)
{
src_locs[i] = -(src_gpr++);
}
else
{
if(t == 'l' || t == 'L' || t == 'd')
{
if(src_offset % 8)
src_offset += 4;
}
src_locs[i] = src_offset;
src_offset += (t == 'l' || t == 'L' || t == 'd') ? 8 : 4;
}
if((t == 'l' || t == 'L') && dst_gpr <= 9)
{
if((dst_gpr & 1) == 0)
dst_gpr++;
dst_locs[i] = -dst_gpr;
dst_gpr += 2;
}
else if((t == 'w' || t == 'W') && dst_gpr <= 10)
{
dst_locs[i] = -(dst_gpr++);
}
else
{
if(t == 'l' || t == 'L' || t == 'd')
{
if(dst_offset % 8)
dst_offset += 4;
}
dst_locs[i] = dst_offset;
dst_offset += (t == 'l' || t == 'L' || t == 'd') ? 8 : 4;
}
}
}
frameSize = dst_offset + 8;
frameSize = (frameSize+15) & ~0xF;
/* Step 2:
Build the adjustor.
*/
// allocate space for at most 4 insns per parameter
// plus 14 more instructions.
adjustor = allocateExec(4 * (4*n + 14),&code);
code = (unsigned*)adjustor;
*code++ = 0x48000008; // b *+8
// * Put the hptr in a place where freeHaskellFunctionPtr
// can get at it.
*code++ = (unsigned) hptr;
// * save the link register
*code++ = 0x7c0802a6; // mflr r0;
*code++ = 0x90010004; // stw r0, 4(r1);
// * and build a new stack frame
*code++ = OP_LO(0x9421, -frameSize); // stwu r1, -frameSize(r1)
// * now generate instructions to copy arguments
// from the old stack frame into the new stack frame.
for(i=n-1;i>=0;i--)
{
if(src_locs[i] < -32)
ASSERT(dst_locs[i] == src_locs[i]);
else if(src_locs[i] < 0)
{
// source in GPR.
ASSERT(typeString[i] != 'f' && typeString[i] != 'd');
if(dst_locs[i] < 0)
{
ASSERT(dst_locs[i] > -32);
// dst is in GPR, too.
if(typeString[i] == 'l' || typeString[i] == 'L')
{
// mr dst+1, src+1
*code++ = 0x7c000378
| ((-dst_locs[i]+1) << 16)
| ((-src_locs[i]+1) << 11)
| ((-src_locs[i]+1) << 21);
}
// mr dst, src
*code++ = 0x7c000378
| ((-dst_locs[i]) << 16)
| ((-src_locs[i]) << 11)
| ((-src_locs[i]) << 21);
}
else
{
if(typeString[i] == 'l' || typeString[i] == 'L')
{
// stw src+1, dst_offset+4(r1)
*code++ = 0x90010000
| ((-src_locs[i]+1) << 21)
| (dst_locs[i] + 4);
}
// stw src, dst_offset(r1)
*code++ = 0x90010000
| ((-src_locs[i]) << 21)
| (dst_locs[i] + 8);
}
}
else
{
ASSERT(dst_locs[i] >= 0);
ASSERT(typeString[i] != 'f' && typeString[i] != 'd');
if(typeString[i] == 'l' || typeString[i] == 'L')
{
// lwz r0, src_offset(r1)
*code++ = 0x80010000
| (src_locs[i] + frameSize + 8 + 4);
// stw r0, dst_offset(r1)
*code++ = 0x90010000
| (dst_locs[i] + 8 + 4);
}
// lwz r0, src_offset(r1)
*code++ = 0x80010000
| (src_locs[i] + frameSize + 8);
// stw r0, dst_offset(r1)
*code++ = 0x90010000
| (dst_locs[i] + 8);
}
}
// * hptr will be the new first argument.
// lis r3, hi(hptr)
*code++ = OP_HI(0x3c60, hptr);
// ori r3,r3,lo(hptr)
*code++ = OP_LO(0x6063, hptr);
// * we need to return to a piece of code
// which will tear down the stack frame.
// lis r11,hi(obscure_ccall_ret_code)
*code++ = OP_HI(0x3d60, obscure_ccall_ret_code);
// ori r11,r11,lo(obscure_ccall_ret_code)
*code++ = OP_LO(0x616b, obscure_ccall_ret_code);
// mtlr r11
*code++ = 0x7d6803a6;
// * jump to wptr
// lis r11,hi(wptr)
*code++ = OP_HI(0x3d60, wptr);
// ori r11,r11,lo(wptr)
*code++ = OP_LO(0x616b, wptr);
// mtctr r11
*code++ = 0x7d6903a6;
// bctr
*code++ = 0x4e800420;
// Flush the Instruction cache:
{
unsigned *p = adjustor;
while(p < code)
{
__asm__ volatile ("dcbf 0,%0\n\tsync\n\ticbi 0,%0"
: : "r" (p));
p++;
}
__asm__ volatile ("sync\n\tisync");
}
}
#elif defined(powerpc_HOST_ARCH) || defined(powerpc64_HOST_ARCH)
#define OP_LO(op,lo) ((((unsigned)(op)) << 16) | (((unsigned)(lo)) & 0xFFFF))
#define OP_HI(op,hi) ((((unsigned)(op)) << 16) | (((unsigned)(hi)) >> 16))
{
/* The following code applies to all PowerPC and PowerPC64 platforms
whose stack layout is based on the AIX ABI.
Besides (obviously) AIX, this includes
Mac OS 9 and BeOS/PPC (may they rest in peace),
which use the 32-bit AIX ABI
powerpc64-linux,
which uses the 64-bit AIX ABI
and Darwin (Mac OS X),
which uses the same stack layout as AIX,
but no function descriptors.
The actual stack-frame shuffling is implemented out-of-line
in the function adjustorCode, in AdjustorAsm.S.
Here, we set up an AdjustorStub structure, which
is a function descriptor (on platforms that have function
descriptors) or a short piece of stub code (on Darwin) to call
adjustorCode with a pointer to the AdjustorStub struct loaded
into register r2.
One nice thing about this is that there is _no_ code generated at
runtime on the platforms that have function descriptors.
*/
AdjustorStub *adjustorStub;
int sz = 0, extra_sz, total_sz;
#ifdef FUNDESCS
adjustorStub = stgMallocBytes(sizeof(AdjustorStub), "createAdjustor");
#else
adjustorStub = allocateExec(sizeof(AdjustorStub),&code);
#endif
adjustor = adjustorStub;
adjustorStub->code = (void*) &adjustorCode;
#ifdef FUNDESCS
// function descriptors are a cool idea.
// We don't need to generate any code at runtime.
adjustorStub->toc = adjustorStub;
#else
// no function descriptors :-(
// We need to do things "by hand".
#if defined(powerpc_HOST_ARCH)
// lis r2, hi(adjustorStub)
adjustorStub->lis = OP_HI(0x3c40, adjustorStub);
// ori r2, r2, lo(adjustorStub)
adjustorStub->ori = OP_LO(0x6042, adjustorStub);
// lwz r0, code(r2)
adjustorStub->lwz = OP_LO(0x8002, (char*)(&adjustorStub->code)
- (char*)adjustorStub);
// mtctr r0
adjustorStub->mtctr = 0x7c0903a6;
// bctr
adjustorStub->bctr = 0x4e800420;
#else
barf("adjustor creation not supported on this platform");
#endif
// Flush the Instruction cache:
{
int n = sizeof(AdjustorStub)/sizeof(unsigned);
unsigned *p = (unsigned*)adjustor;
while(n--)
{
__asm__ volatile ("dcbf 0,%0\n\tsync\n\ticbi 0,%0"
: : "r" (p));
p++;
}
__asm__ volatile ("sync\n\tisync");
}
#endif
// Calculate the size of the stack frame, in words.
sz = totalArgumentSize(typeString);
// The first eight words of the parameter area
// are just "backing store" for the parameters passed in
// the GPRs. extra_sz is the number of words beyond those first
// 8 words.
extra_sz = sz - 8;
if(extra_sz < 0)
extra_sz = 0;
// Calculate the total size of the stack frame.
total_sz = (6 /* linkage area */
+ 8 /* minimum parameter area */
+ 2 /* two extra arguments */
+ extra_sz)*sizeof(StgWord);
// align to 16 bytes.
// AIX only requires 8 bytes, but who cares?
total_sz = (total_sz+15) & ~0xF;
// Fill in the information that adjustorCode in AdjustorAsm.S
// will use to create a new stack frame with the additional args.
adjustorStub->hptr = hptr;
adjustorStub->wptr = wptr;
adjustorStub->negative_framesize = -total_sz;
adjustorStub->extrawords_plus_one = extra_sz + 1;
}
#elif defined(ia64_HOST_ARCH)
/*
Up to 8 inputs are passed in registers. We flush the last two inputs to
the stack, initially into the 16-byte scratch region left by the caller.
We then shuffle the others along by 4 (taking 2 registers for ourselves
to save return address and previous function state - we need to come back
here on the way out to restore the stack, so this is a real function
rather than just a trampoline).
The function descriptor we create contains the gp of the target function
so gp is already loaded correctly.
[MLX] alloc r16=ar.pfs,10,2,0
movl r17=wptr
[MII] st8.spill [r12]=r38,8 // spill in6 (out4)
mov r41=r37 // out7 = in5 (out3)
mov r40=r36;; // out6 = in4 (out2)
[MII] st8.spill [r12]=r39 // spill in7 (out5)
mov.sptk b6=r17,50
mov r38=r34;; // out4 = in2 (out0)
[MII] mov r39=r35 // out5 = in3 (out1)
mov r37=r33 // out3 = in1 (loc1)
mov r36=r32 // out2 = in0 (loc0)
[MLX] adds r12=-24,r12 // update sp
movl r34=hptr;; // out0 = hptr
[MIB] mov r33=r16 // loc1 = ar.pfs
mov r32=b0 // loc0 = retaddr
br.call.sptk.many b0=b6;;
[MII] adds r12=-16,r12
mov b0=r32
mov.i ar.pfs=r33
[MFB] nop.m 0x0
nop.f 0x0
br.ret.sptk.many b0;;
*/
/* These macros distribute a long constant into the two words of an MLX bundle */
#define BITS(val,start,count) (((val) >> (start)) & ((1 << (count))-1))
#define MOVL_LOWORD(val) (BITS(val,22,18) << 46)
#define MOVL_HIWORD(val) ( (BITS(val,0,7) << 36) \
| (BITS(val,7,9) << 50) \
| (BITS(val,16,5) << 45) \
| (BITS(val,21,1) << 44) \
| (BITS(val,40,23)) \
| (BITS(val,63,1) << 59))
{
StgStablePtr stable;
IA64FunDesc *wdesc = (IA64FunDesc *)wptr;
StgWord64 wcode = wdesc->ip;
IA64FunDesc *fdesc;
StgWord64 *code;
/* we allocate on the Haskell heap since malloc'd memory isn't
* executable - argh */
/* Allocated memory is word-aligned (8 bytes) but functions on ia64
* must be aligned to 16 bytes. We allocate an extra 8 bytes of
* wiggle room so that we can put the code on a 16 byte boundary. */
adjustor = stgAllocStable(sizeof(IA64FunDesc)+18*8+8, &stable);
fdesc = (IA64FunDesc *)adjustor;
code = (StgWord64 *)(fdesc + 1);
/* add 8 bytes to code if needed to align to a 16-byte boundary */
if ((StgWord64)code & 15) code++;
fdesc->ip = (StgWord64)code;
fdesc->gp = wdesc->gp;
code[0] = 0x0000058004288004 | MOVL_LOWORD(wcode);
code[1] = 0x6000000220000000 | MOVL_HIWORD(wcode);
code[2] = 0x029015d818984001;
code[3] = 0x8401200500420094;
code[4] = 0x886011d8189c0001;
code[5] = 0x84011004c00380c0;
code[6] = 0x0250210046013800;
code[7] = 0x8401000480420084;
code[8] = 0x0000233f19a06005 | MOVL_LOWORD((StgWord64)hptr);
code[9] = 0x6000000440000000 | MOVL_HIWORD((StgWord64)hptr);
code[10] = 0x0200210020010811;
code[11] = 0x1080006800006200;
code[12] = 0x0000210018406000;
code[13] = 0x00aa021000038005;
code[14] = 0x000000010000001d;
code[15] = 0x0084000880000200;
/* save stable pointers in convenient form */
code[16] = (StgWord64)hptr;
code[17] = (StgWord64)stable;
}
#else
barf("adjustor creation not supported on this platform");
#endif
break;
default:
ASSERT(0);
break;
}
/* Have fun! */
return code;
}
void
freeHaskellFunctionPtr(void* ptr)
{
#if defined(i386_HOST_ARCH)
if ( *(unsigned char*)ptr != 0xe8 &&
*(unsigned char*)ptr != 0x58 ) {
errorBelch("freeHaskellFunctionPtr: not for me, guv! %p\n", ptr);
return;
}
if (*(unsigned char*)ptr == 0xe8) { /* Aha, a ccall adjustor! */
freeStablePtr(((AdjustorStub*)ptr)->hptr);
} else {
freeStablePtr(*((StgStablePtr*)((unsigned char*)ptr + 0x02)));
}
#elif defined(x86_64_HOST_ARCH)
if ( *(StgWord16 *)ptr == 0x894d ) {
freeStablePtr(*(StgStablePtr*)((StgWord8*)ptr+
#if defined(mingw32_HOST_OS)
0x28
#else
0x20
#endif
));
#if !defined(mingw32_HOST_OS)
} else if ( *(StgWord16 *)ptr == 0x5141 ) {
freeStablePtr(*(StgStablePtr*)((StgWord8*)ptr+0x30));
#endif
#if defined(mingw32_HOST_OS)
} else if ( *(StgWord16 *)ptr == 0x8348 ) {
freeStablePtr(*(StgStablePtr*)((StgWord8*)ptr+0x48));
#endif
} else {
errorBelch("freeHaskellFunctionPtr: not for me, guv! %p\n", ptr);
return;
}
#elif defined(sparc_HOST_ARCH)
if ( *(unsigned long*)ptr != 0x9C23A008UL ) {
errorBelch("freeHaskellFunctionPtr: not for me, guv! %p\n", ptr);
return;
}
/* Free the stable pointer first..*/
freeStablePtr(*((StgStablePtr*)((unsigned long*)ptr + 11)));
#elif defined(alpha_HOST_ARCH)
if ( *(StgWord64*)ptr != 0xa77b0018a61b0010L ) {
errorBelch("freeHaskellFunctionPtr: not for me, guv! %p\n", ptr);
return;
}
/* Free the stable pointer first..*/
freeStablePtr(*((StgStablePtr*)((unsigned char*)ptr + 0x10)));
#elif defined(powerpc_HOST_ARCH) && defined(linux_HOST_OS)
if ( *(StgWord*)ptr != 0x48000008 ) {
errorBelch("freeHaskellFunctionPtr: not for me, guv! %p\n", ptr);
return;
}
freeStablePtr(((StgStablePtr*)ptr)[1]);
#elif defined(powerpc_HOST_ARCH) || defined(powerpc64_HOST_ARCH)
if ( ((AdjustorStub*)ptr)->code != (StgFunPtr) &adjustorCode ) {
errorBelch("freeHaskellFunctionPtr: not for me, guv! %p\n", ptr);
return;
}
freeStablePtr(((AdjustorStub*)ptr)->hptr);
#elif defined(ia64_HOST_ARCH)
IA64FunDesc *fdesc = (IA64FunDesc *)ptr;
StgWord64 *code = (StgWord64 *)(fdesc+1);
if (fdesc->ip != (StgWord64)code) {
errorBelch("freeHaskellFunctionPtr: not for me, guv! %p\n", ptr);
return;
}
freeStablePtr((StgStablePtr)code[16]);
freeStablePtr((StgStablePtr)code[17]);
return;
#else
ASSERT(0);
#endif
// Can't write to this memory, it is only executable:
// *((unsigned char*)ptr) = '\0';
freeExec(ptr);
}
#endif // !USE_LIBFFI_FOR_ADJUSTORS