* Add support for the Win32 hook prologue (try 8)
@ 2009-10-07 19:27 Stefan Dösinger
2009-10-09 13:22 ` [Patch ping] " Stefan Dösinger
2009-10-09 22:38 ` Richard Henderson
0 siblings, 2 replies; 9+ messages in thread
From: Stefan Dösinger @ 2009-10-07 19:27 UTC (permalink / raw)
To: gcc-patches; +Cc: Paolo Bonzini
[-- Attachment #1: Type: text/plain, Size: 2722 bytes --]
Here's another version of the ms_hook_prologue patch.
I noticed that the function context isn't set, and the
current_function_decl isn't updated by the time the attribute
callbacks execute. Thus I use current_function_decl, not
decl_function_context(node) to find out if the function is nested or
not. If this is not correct(I only checked the C frontend so far),
please suggest a better solution. The only other way I found was to
check the nestedness during code generation, which doesn't seem right
to me(and puts the error at the end of the function, not its
declaration).
The development / test platform for this patch is x86_64-apple-
darwin10.0.0 on OSX 10.6.1. I have tested older versions of this on
x86_64-pc-linux-gnu as well, but didn't run the testsuite for this
patch on Linux yet.
Changes from try 7:
Disallowed ms_hook_prologue on nested functions
Attached the correct patch this time
Changes from try 6:
the ms_hook_prologue attribute applies to the function, rather than
its type. Changed the attribute lookup accordingly. However, I still
allow the attribute for nested functions, there's no reason why this
wouldn't work. (The case of static_chain and ms_hook_prologue can
potentially be optimized)
Changes from try 5:
Fixed conflicts with static_chain patches, simple handling for
static_chain+ms_hook_prologue cases.
Improved changelog(hopefully correctly)
Changes from try 4:
Use SIMode instead of Pmode for the registers.
Changes from try 3:
Renamed the attribute name to ms_hook_prologue, since its not always
generated
by msvc
Changes from try 2:
* Use dg-require-effective-target ilp32 instead of dg-options -m32 in
the test
* More code layout fixes
Changes from the first try:
* Moved testcase to gcc.target/i386
* restructured ix86_handle_abi_attribute
* Changed HAVE_AS_IX86_SWAP handling as suggested by rth
* ix86_function_msvc_prologue returns a bool, removed fntype != NULL
check
* fixed indentation in ix86_expand_prologue(hopefully correct now)
2009-09-23: Stefan Dösinger <stefan@codeweavers.com
* config/i386/i386.md: Added vswapmov, a volative DIMode
register->register mov with the swap suffix set
* config/i386/i386.c (ix86_expand_prologue): Add a new function
attribute ms_hook_prologue that starts functions with the same
opcode
sequence used in most Win32 API functions
* config/i386/i386.c (ix86_function_ms_hook_prologue): New
* config/i386/i386.c (ix86_handle_fndecl_attribute): New
* gcc.target/i386/ms_hook_prologue.c: New testcase
* configure.ac: Test for swap suffix support in as
[-- Attachment #2: ms_hook_prologue.diff --]
[-- Type: application/octet-stream, Size: 8033 bytes --]
Index: gcc/doc/extend.texi
===================================================================
--- gcc/doc/extend.texi (revision 152535)
+++ gcc/doc/extend.texi (working copy)
@@ -2679,6 +2679,14 @@ when targeting Windows. On all other systems, the
Note, This feature is currently sorried out for Windows targets trying to
+@item ms_hook_prologue
+@cindex @code{ms_hook_prologue} attribute
+
+On 32 bit i[34567]86-*-* targets, you can use this function attribute to make
+gcc generate the "hot-patching" function prologue used in Win32 API
+functions in Microsoft Windows XP Service Pack 2 and newer. This requires
+support for the swap suffix in the assembler. (GNU Binutils 2.19.51 or later)
+
@item naked
@cindex function without a prologue/epilogue code
Use this attribute on the ARM, AVR, IP2K and SPU ports to indicate that
Index: gcc/configure.ac
===================================================================
--- gcc/configure.ac (revision 152535)
+++ gcc/configure.ac (working copy)
@@ -3047,6 +3047,12 @@ foo: nop
[AC_DEFINE(HAVE_AS_IX86_SAHF, 1,
[Define if your assembler supports the sahf mnemonic.])])
+ gcc_GAS_CHECK_FEATURE([swap suffix],
+ gcc_cv_as_ix86_swap,,,
+ [movl.s %esp, %ebp],,
+ [AC_DEFINE(HAVE_AS_IX86_SWAP, 1,
+ [Define if your assembler supports the swap suffix.])])
+
gcc_GAS_CHECK_FEATURE([different section symbol subtraction],
gcc_cv_as_ix86_diff_sect_delta,,,
[.section .rodata
Index: gcc/config/i386/i386.md
===================================================================
--- gcc/config/i386/i386.md (revision 152535)
+++ gcc/config/i386/i386.md (working copy)
@@ -241,6 +241,7 @@
(UNSPECV_RDTSC 18)
(UNSPECV_RDTSCP 19)
(UNSPECV_RDPMC 20)
+ (UNSPECV_VSWAPMOV 21)
])
;; Constants to represent pcomtrue/pcomfalse variants
@@ -15087,6 +15088,16 @@
(set_attr "length_immediate" "0")
(set_attr "modrm" "0")])
+(define_insn "vswapmov"
+ [(set (match_operand:SI 0 "register_operand" "=r")
+ (match_operand:SI 1 "register_operand" "r"))
+ (unspec_volatile [(const_int 0)] UNSPECV_VSWAPMOV)]
+ ""
+ "movl.s\t{%1, %0|%0, %1}"
+ [(set_attr "length" "2")
+ (set_attr "length_immediate" "0")
+ (set_attr "modrm" "0")])
+
;; Pad to 16-byte boundary, max skip in op0. Used to avoid
;; branch prediction penalty for the third jump in a 16-byte
;; block on K8.
Index: gcc/config/i386/i386.c
===================================================================
--- gcc/config/i386/i386.c (revision 152535)
+++ gcc/config/i386/i386.c (working copy)
@@ -4774,6 +4774,19 @@ ix86_function_type_abi (const_tree fntype)
return ix86_abi;
}
+static bool
+ix86_function_ms_hook_prologue (const_tree fntype)
+{
+ if (!TARGET_64BIT)
+ {
+ if(lookup_attribute ("ms_hook_prologue", DECL_ATTRIBUTES (fntype)))
+ {
+ return true;
+ }
+ }
+ return false;
+}
+
static enum calling_abi
ix86_function_abi (const_tree fndecl)
{
@@ -8295,6 +8308,7 @@ ix86_expand_prologue (void)
bool pic_reg_used;
struct ix86_frame frame;
HOST_WIDE_INT allocate;
+ int gen_frame_pointer = frame_pointer_needed;
ix86_finalize_stack_realign_flags ();
@@ -8307,6 +8321,46 @@ ix86_expand_prologue (void)
ix86_compute_frame_layout (&frame);
+ if (ix86_function_ms_hook_prologue (current_function_decl))
+ {
+ rtx push, mov;
+
+ /* Make sure the function starts with
+ 8b ff movl.s %edi,%edi
+ 55 push %ebp
+ 8b ec movl.s %esp,%ebp
+
+ This matches the hookable function prologue in Win32 API functions in Microsoft Windows
+ XP Service Pack 2 and newer. Wine uses this to enable Windows apps to hook the Win32 API
+ functions provided by Wine */
+ insn = emit_insn (gen_vswapmov (gen_rtx_REG (SImode, DI_REG), gen_rtx_REG (SImode, DI_REG)));
+ push = emit_insn (gen_push (hard_frame_pointer_rtx));
+ mov = emit_insn (gen_vswapmov (hard_frame_pointer_rtx, stack_pointer_rtx));
+
+ if (frame_pointer_needed && !(crtl->drap_reg && crtl->stack_realign_needed))
+ {
+ /* The push %ebp and movl.s %esp, %ebp already set up the frame pointer. No need to do
+ this again. */
+ gen_frame_pointer = 0;
+ RTX_FRAME_RELATED_P (push) = 1;
+ RTX_FRAME_RELATED_P (mov) = 1;
+ if (ix86_cfa_state->reg == stack_pointer_rtx)
+ {
+ ix86_cfa_state->reg = hard_frame_pointer_rtx;
+ }
+ }
+ else
+ {
+ /* If the frame pointer is not needed, pop %ebp again. This could be optimized for cases where
+ ebp needs to be backed up for some other reason.
+
+ If stack realignment is needed, pop the base pointer again, align the stack, and later
+ regenerate the frame pointer setup. The frame pointer generated by the hook prologue
+ is not aligned, so it can't be used */
+ insn = emit_insn ((*ix86_gen_pop1) (hard_frame_pointer_rtx));
+ }
+ }
+
/* The first insn of a function that accepts its static chain on the
stack is to push the register that would be filled in by a direct
call. This insn will be skipped by the trampoline. */
@@ -8378,7 +8432,7 @@ ix86_expand_prologue (void)
/* Note: AT&T enter does NOT have reversed args. Enter is probably
slower on all targets. Also sdb doesn't like it. */
- if (frame_pointer_needed)
+ if (gen_frame_pointer)
{
insn = emit_insn (gen_push (hard_frame_pointer_rtx));
RTX_FRAME_RELATED_P (insn) = 1;
@@ -26458,6 +26512,40 @@ ix86_handle_struct_attribute (tree *node, tree nam
return NULL_TREE;
}
+#include <signal.h>
+
+static tree
+ix86_handle_fndecl_attribute (tree *node, tree name,
+ tree args ATTRIBUTE_UNUSED,
+ int flags ATTRIBUTE_UNUSED, bool *no_add_attrs)
+{
+ if (TREE_CODE (*node) != FUNCTION_DECL)
+ {
+ warning (OPT_Wattributes, "%qE attribute only applies to functions",
+ name);
+ *no_add_attrs = true;
+ return NULL_TREE;
+ }
+
+ if (TARGET_64BIT)
+ {
+ warning (OPT_Wattributes, "%qE attribute only available for 32-bit",
+ name);
+ return NULL_TREE;
+ }
+
+ if(current_function_decl != NULL_TREE)
+ {
+ error("ms_hook_prologue is not compatible with nested functions\n");
+ }
+
+#ifndef HAVE_AS_IX86_SWAP
+ sorry ("ms_hook_prologue attribute needs assembler swap suffix support");
+#endif
+
+ return NULL_TREE;
+}
+
static bool
ix86_ms_bitfield_layout_p (const_tree record_type)
{
@@ -29500,6 +29588,7 @@ static const struct attribute_spec ix86_attribute_
/* ms_abi and sysv_abi calling convention function attributes. */
{ "ms_abi", 0, 0, false, true, true, ix86_handle_abi_attribute },
{ "sysv_abi", 0, 0, false, true, true, ix86_handle_abi_attribute },
+ { "ms_hook_prologue", 0, 0, true, false, false, ix86_handle_fndecl_attribute },
/* End element. */
{ NULL, 0, 0, false, false, false, NULL }
};
--- /dev/null 2009-10-07 20:53:54.000000000 +0200
+++ gcc/testsuite/gcc.target/i386/ms_hook_prologue.c 2009-09-23 20:25:31.000000000 +0200
@@ -0,0 +1,29 @@
+/* Test that the ms_hook_prologue attribute generates the correct code. */
+
+/* { dg-do run } */
+/* { dg-require-effective-target ilp32 } */
+/* { dg-options "-O2 -fomit-frame-pointer" } */
+
+int __attribute__ ((__ms_hook_prologue__)) foo ()
+{
+ unsigned char *ptr = (unsigned char *) foo;
+
+ /* The NOP mov must not be optimized away by optimizations.
+ The push %ebp, mov %esp, %ebp must not be removed by
+ -fomit-frame-pointer */
+
+ /* movl.s %edi, %edi */
+ if(*ptr++ != 0x8b) return 1;
+ if(*ptr++ != 0xff) return 1;
+ /* push %ebp */
+ if(*ptr++ != 0x55) return 1;
+ /* movl.s %esp, %ebp */
+ if(*ptr++ != 0x8b) return 1;
+ if(*ptr++ != 0xec) return 1;
+ return 0;
+}
+
+int main ()
+{
+ return foo();
+}
^ permalink raw reply [flat|nested] 9+ messages in thread
* [Patch ping] Add support for the Win32 hook prologue (try 8)
2009-10-07 19:27 Add support for the Win32 hook prologue (try 8) Stefan Dösinger
@ 2009-10-09 13:22 ` Stefan Dösinger
2009-10-09 13:49 ` Paolo Bonzini
2009-10-09 14:53 ` Dave Korn
2009-10-09 22:38 ` Richard Henderson
1 sibling, 2 replies; 9+ messages in thread
From: Stefan Dösinger @ 2009-10-09 13:22 UTC (permalink / raw)
To: gcc-patches
Hi,
On Wednesday I sent another version of my Windows hooking prologue
patch, but so far the patch hasn't been applied, nor have I received
any comments.
Is this patch still under review? Are there still further suggestions
for improvement?
Here's a link to the archives: http://gcc.gnu.org/ml/gcc-patches/2009-10/msg00488.html
Thank you in advance,
Stefan
PS: I don't know the usual application delays in the gcc project, I am
mostly used to how things work in Wine. So if I am too impatient with
pinging a patch after two days I apologize.
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [Patch ping] Add support for the Win32 hook prologue (try 8)
2009-10-09 13:22 ` [Patch ping] " Stefan Dösinger
@ 2009-10-09 13:49 ` Paolo Bonzini
2009-10-09 14:53 ` Dave Korn
1 sibling, 0 replies; 9+ messages in thread
From: Paolo Bonzini @ 2009-10-09 13:49 UTC (permalink / raw)
To: Stefan Dösinger; +Cc: gcc-patches
> PS: I don't know the usual application delays in the gcc project, I am
> mostly used to how things work in Wine. So if I am too impatient with
> pinging a patch after two days I apologize.
It's quite different from Wine indeed. Usually we ping after 1-2 weeks.
I understand this is awful for a random contributor; often people
develop more patches at the same time to ease pipelining. On the other
hand it is in general easier to get a patch in, as there is no BDFL and
it's relatively rare to have disagreements on the fundamental how's and
why's of a patch (these are more common in Wine).
Thanks for your persistence. You chose quite a complicated patch for
your first one. :-) There has been quite some discussion on it, and if
there is no response next week I'll go ahead and apply it for you.
I would still like to do a quick sanity check on the patch, so can you
send me offlist a diff from v4 to v8?
Paolo
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [Patch ping] Add support for the Win32 hook prologue (try 8)
2009-10-09 13:22 ` [Patch ping] " Stefan Dösinger
2009-10-09 13:49 ` Paolo Bonzini
@ 2009-10-09 14:53 ` Dave Korn
1 sibling, 0 replies; 9+ messages in thread
From: Dave Korn @ 2009-10-09 14:53 UTC (permalink / raw)
To: Stefan Dösinger; +Cc: gcc-patches
Stefan Dösinger wrote:
> Hi,
> On Wednesday I sent another version of my Windows hooking prologue
> patch, but so far the patch hasn't been applied, nor have I received any
> comments.
"Thank you" :-)
cheers,
DaveK
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: Add support for the Win32 hook prologue (try 8)
2009-10-07 19:27 Add support for the Win32 hook prologue (try 8) Stefan Dösinger
2009-10-09 13:22 ` [Patch ping] " Stefan Dösinger
@ 2009-10-09 22:38 ` Richard Henderson
2009-10-11 21:50 ` [PATCH] Add support for the Win32 hook prologue (try 9) Stefan Dösinger
1 sibling, 1 reply; 9+ messages in thread
From: Richard Henderson @ 2009-10-09 22:38 UTC (permalink / raw)
To: Stefan Dösinger; +Cc: gcc-patches, Paolo Bonzini
On 10/07/2009 12:08 PM, Stefan Dösinger wrote:
> I noticed that the function context isn't set, and the
> current_function_decl isn't updated by the time the attribute callbacks
> execute. Thus I use current_function_decl, not
> decl_function_context(node) to find out if the function is nested or
> not. If this is not correct(I only checked the C frontend so far),
> please suggest a better solution. The only other way I found was to
> check the nestedness during code generation, which doesn't seem right to
> me(and puts the error at the end of the function, not its declaration).
Ah, I see what you mean here. Yes, the decl context hasn't been set up
yet. However, testing for current_function_decl isn't correct either.
I think the only thing you'll be able to do is an error during code
generation. You can put the error message at the correct line by using
DECL_SOURCE_LOCATION and the error_at function.
The patch is ok with that fix.
r~
^ permalink raw reply [flat|nested] 9+ messages in thread
* [PATCH] Add support for the Win32 hook prologue (try 9)
2009-10-09 22:38 ` Richard Henderson
@ 2009-10-11 21:50 ` Stefan Dösinger
2009-10-12 16:39 ` Richard Henderson
0 siblings, 1 reply; 9+ messages in thread
From: Stefan Dösinger @ 2009-10-11 21:50 UTC (permalink / raw)
To: gcc-patches
[-- Attachment #1: Type: text/plain, Size: 2627 bytes --]
A number of Windows programs(Steam, Xfire and others) try to hook
Win32 API
functions by replacing the first 5 bytes in the function. This causes
problems for Wine because the functions generated by gcc usually start
with a
different opcode sequence than the one expected by these applications.
Starting with Windows XP SP2, Microsoft starts Win32 functions with this
sequence:
8b ff movl.s %edi, %edi
55 push %ebp
8b ec movl.s %esp, %ebp
The attached patch implements a function attribute that allows Wine to
request
the same 5 bytes at the beginning of a function.
Changes from try 8:
moved the nested function check to ix86_function_ms_hook_prologue,
check the decl context rather than current_function_decl.
Changes from try 7:
Disallowed ms_hook_prologue on nested functions
Attached the correct patch this time
Changes from try 6:
the ms_hook_prologue attribute applies to the function, rather than
its type. Changed the attribute lookup accordingly. However, I still
allow the attribute for nested functions, there's no reason why this
wouldn't work. (The case of static_chain and ms_hook_prologue can
potentially be optimized)
Changes from try 5:
Fixed conflicts with static_chain patches, simple handling for
static_chain+ms_hook_prologue cases.
Improved changelog(hopefully correctly)
Changes from try 4:
Use SIMode instead of Pmode for the registers.
Changes from try 3:
Renamed the attribute name to ms_hook_prologue, since its not always
generated
by msvc
Changes from try 2:
* Use dg-require-effective-target ilp32 instead of dg-options -m32 in
the test
* More code layout fixes
Changes from the first try:
* Moved testcase to gcc.target/i386
* restructured ix86_handle_abi_attribute
* Changed HAVE_AS_IX86_SWAP handling as suggested by rth
* ix86_function_msvc_prologue returns a bool, removed fntype != NULL
check
* fixed indentation in ix86_expand_prologue(hopefully correct now)
2009-09-23: Stefan Dösinger <stefan@codeweavers.com
* config/i386/i386.md: Added vswapmov, a volative DIMode
register->register mov with the swap suffix set
* config/i386/i386.c (ix86_expand_prologue): Add a new function
attribute ms_hook_prologue that starts functions with the same
opcode
sequence used in most Win32 API functions
* config/i386/i386.c (ix86_function_ms_hook_prologue): New
* config/i386/i386.c (ix86_handle_fndecl_attribute): New
* gcc.target/i386/ms_hook_prologue.c: New testcase
* configure.ac: Test for swap suffix support in as
[-- Attachment #2: ms_hook_prologue9.diff --]
[-- Type: application/octet-stream, Size: 8119 bytes --]
Index: gcc/doc/extend.texi
===================================================================
--- gcc/doc/extend.texi (revision 152649)
+++ gcc/doc/extend.texi (working copy)
@@ -2679,6 +2679,14 @@ when targeting Windows. On all other systems, the
Note, This feature is currently sorried out for Windows targets trying to
+@item ms_hook_prologue
+@cindex @code{ms_hook_prologue} attribute
+
+On 32 bit i[34567]86-*-* targets, you can use this function attribute to make
+gcc generate the "hot-patching" function prologue used in Win32 API
+functions in Microsoft Windows XP Service Pack 2 and newer. This requires
+support for the swap suffix in the assembler. (GNU Binutils 2.19.51 or later)
+
@item naked
@cindex function without a prologue/epilogue code
Use this attribute on the ARM, AVR, IP2K and SPU ports to indicate that
Index: gcc/configure.ac
===================================================================
--- gcc/configure.ac (revision 152649)
+++ gcc/configure.ac (working copy)
@@ -3047,6 +3047,12 @@ foo: nop
[AC_DEFINE(HAVE_AS_IX86_SAHF, 1,
[Define if your assembler supports the sahf mnemonic.])])
+ gcc_GAS_CHECK_FEATURE([swap suffix],
+ gcc_cv_as_ix86_swap,,,
+ [movl.s %esp, %ebp],,
+ [AC_DEFINE(HAVE_AS_IX86_SWAP, 1,
+ [Define if your assembler supports the swap suffix.])])
+
gcc_GAS_CHECK_FEATURE([different section symbol subtraction],
gcc_cv_as_ix86_diff_sect_delta,,,
[.section .rodata
Index: gcc/config/i386/i386.md
===================================================================
--- gcc/config/i386/i386.md (revision 152649)
+++ gcc/config/i386/i386.md (working copy)
@@ -241,6 +241,7 @@
(UNSPECV_RDTSC 18)
(UNSPECV_RDTSCP 19)
(UNSPECV_RDPMC 20)
+ (UNSPECV_VSWAPMOV 21)
])
;; Constants to represent pcomtrue/pcomfalse variants
@@ -14888,6 +14889,16 @@
(set_attr "length_immediate" "0")
(set_attr "modrm" "0")])
+(define_insn "vswapmov"
+ [(set (match_operand:SI 0 "register_operand" "=r")
+ (match_operand:SI 1 "register_operand" "r"))
+ (unspec_volatile [(const_int 0)] UNSPECV_VSWAPMOV)]
+ ""
+ "movl.s\t{%1, %0|%0, %1}"
+ [(set_attr "length" "2")
+ (set_attr "length_immediate" "0")
+ (set_attr "modrm" "0")])
+
;; Pad to 16-byte boundary, max skip in op0. Used to avoid
;; branch prediction penalty for the third jump in a 16-byte
;; block on K8.
Index: gcc/config/i386/i386.c
===================================================================
--- gcc/config/i386/i386.c (revision 152649)
+++ gcc/config/i386/i386.c (working copy)
@@ -4774,6 +4774,25 @@ ix86_function_type_abi (const_tree fntype)
return ix86_abi;
}
+static bool
+ix86_function_ms_hook_prologue (const_tree fntype)
+{
+ if (!TARGET_64BIT)
+ {
+ if (lookup_attribute ("ms_hook_prologue", DECL_ATTRIBUTES (fntype)))
+ {
+ if (decl_function_context (fntype) != NULL_TREE)
+ {
+ error_at (DECL_SOURCE_LOCATION (fntype),
+ "ms_hook_prologue is not compatible with nested function");
+ }
+
+ return true;
+ }
+ }
+ return false;
+}
+
static enum calling_abi
ix86_function_abi (const_tree fndecl)
{
@@ -8295,6 +8314,7 @@ ix86_expand_prologue (void)
bool pic_reg_used;
struct ix86_frame frame;
HOST_WIDE_INT allocate;
+ int gen_frame_pointer = frame_pointer_needed;
ix86_finalize_stack_realign_flags ();
@@ -8307,6 +8327,46 @@ ix86_expand_prologue (void)
ix86_compute_frame_layout (&frame);
+ if (ix86_function_ms_hook_prologue (current_function_decl))
+ {
+ rtx push, mov;
+
+ /* Make sure the function starts with
+ 8b ff movl.s %edi,%edi
+ 55 push %ebp
+ 8b ec movl.s %esp,%ebp
+
+ This matches the hookable function prologue in Win32 API functions in Microsoft Windows
+ XP Service Pack 2 and newer. Wine uses this to enable Windows apps to hook the Win32 API
+ functions provided by Wine */
+ insn = emit_insn (gen_vswapmov (gen_rtx_REG (SImode, DI_REG), gen_rtx_REG (SImode, DI_REG)));
+ push = emit_insn (gen_push (hard_frame_pointer_rtx));
+ mov = emit_insn (gen_vswapmov (hard_frame_pointer_rtx, stack_pointer_rtx));
+
+ if (frame_pointer_needed && !(crtl->drap_reg && crtl->stack_realign_needed))
+ {
+ /* The push %ebp and movl.s %esp, %ebp already set up the frame pointer. No need to do
+ this again. */
+ gen_frame_pointer = 0;
+ RTX_FRAME_RELATED_P (push) = 1;
+ RTX_FRAME_RELATED_P (mov) = 1;
+ if (ix86_cfa_state->reg == stack_pointer_rtx)
+ {
+ ix86_cfa_state->reg = hard_frame_pointer_rtx;
+ }
+ }
+ else
+ {
+ /* If the frame pointer is not needed, pop %ebp again. This could be optimized for cases where
+ ebp needs to be backed up for some other reason.
+
+ If stack realignment is needed, pop the base pointer again, align the stack, and later
+ regenerate the frame pointer setup. The frame pointer generated by the hook prologue
+ is not aligned, so it can't be used */
+ insn = emit_insn ((*ix86_gen_pop1) (hard_frame_pointer_rtx));
+ }
+ }
+
/* The first insn of a function that accepts its static chain on the
stack is to push the register that would be filled in by a direct
call. This insn will be skipped by the trampoline. */
@@ -8378,7 +8438,7 @@ ix86_expand_prologue (void)
/* Note: AT&T enter does NOT have reversed args. Enter is probably
slower on all targets. Also sdb doesn't like it. */
- if (frame_pointer_needed)
+ if (gen_frame_pointer)
{
insn = emit_insn (gen_push (hard_frame_pointer_rtx));
RTX_FRAME_RELATED_P (insn) = 1;
@@ -26469,6 +26529,35 @@ ix86_handle_struct_attribute (tree *node, tree nam
return NULL_TREE;
}
+#include <signal.h>
+
+static tree
+ix86_handle_fndecl_attribute (tree *node, tree name,
+ tree args ATTRIBUTE_UNUSED,
+ int flags ATTRIBUTE_UNUSED, bool *no_add_attrs)
+{
+ if (TREE_CODE (*node) != FUNCTION_DECL)
+ {
+ warning (OPT_Wattributes, "%qE attribute only applies to functions",
+ name);
+ *no_add_attrs = true;
+ return NULL_TREE;
+ }
+
+ if (TARGET_64BIT)
+ {
+ warning (OPT_Wattributes, "%qE attribute only available for 32-bit",
+ name);
+ return NULL_TREE;
+ }
+
+#ifndef HAVE_AS_IX86_SWAP
+ sorry ("ms_hook_prologue attribute needs assembler swap suffix support");
+#endif
+
+ return NULL_TREE;
+}
+
static bool
ix86_ms_bitfield_layout_p (const_tree record_type)
{
@@ -29511,6 +29600,7 @@ static const struct attribute_spec ix86_attribute_
/* ms_abi and sysv_abi calling convention function attributes. */
{ "ms_abi", 0, 0, false, true, true, ix86_handle_abi_attribute },
{ "sysv_abi", 0, 0, false, true, true, ix86_handle_abi_attribute },
+ { "ms_hook_prologue", 0, 0, true, false, false, ix86_handle_fndecl_attribute },
/* End element. */
{ NULL, 0, 0, false, false, false, NULL }
};
--- /dev/null 2009-10-11 23:42:17.000000000 +0200
+++ gcc/testsuite/gcc.target/i386/ms_hook_prologue.c 2009-09-23 20:25:31.000000000 +0200
@@ -0,0 +1,29 @@
+/* Test that the ms_hook_prologue attribute generates the correct code. */
+
+/* { dg-do run } */
+/* { dg-require-effective-target ilp32 } */
+/* { dg-options "-O2 -fomit-frame-pointer" } */
+
+int __attribute__ ((__ms_hook_prologue__)) foo ()
+{
+ unsigned char *ptr = (unsigned char *) foo;
+
+ /* The NOP mov must not be optimized away by optimizations.
+ The push %ebp, mov %esp, %ebp must not be removed by
+ -fomit-frame-pointer */
+
+ /* movl.s %edi, %edi */
+ if(*ptr++ != 0x8b) return 1;
+ if(*ptr++ != 0xff) return 1;
+ /* push %ebp */
+ if(*ptr++ != 0x55) return 1;
+ /* movl.s %esp, %ebp */
+ if(*ptr++ != 0x8b) return 1;
+ if(*ptr++ != 0xec) return 1;
+ return 0;
+}
+
+int main ()
+{
+ return foo();
+}
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [PATCH] Add support for the Win32 hook prologue (try 9)
2009-10-11 21:50 ` [PATCH] Add support for the Win32 hook prologue (try 9) Stefan Dösinger
@ 2009-10-12 16:39 ` Richard Henderson
2009-10-12 21:12 ` Stefan Dösinger
0 siblings, 1 reply; 9+ messages in thread
From: Richard Henderson @ 2009-10-12 16:39 UTC (permalink / raw)
To: Stefan Dösinger; +Cc: gcc-patches
Committed r152670.
r~
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [PATCH] Add support for the Win32 hook prologue (try 9)
2009-10-12 16:39 ` Richard Henderson
@ 2009-10-12 21:12 ` Stefan Dösinger
2009-10-12 21:38 ` Richard Henderson
0 siblings, 1 reply; 9+ messages in thread
From: Stefan Dösinger @ 2009-10-12 21:12 UTC (permalink / raw)
To: Richard Henderson; +Cc: gcc-patches
Am 12.10.2009 um 18:37 schrieb Richard Henderson:
> Committed r152670.
Thanks!
Unfortunately, something went wrong with the configure update. I'd
expect gcc/config.in to be regenerated like gcc/configure, but this is
apparently not the case. And since config.in wasn't updated, the
configure regeneration didn't pick up the change either. Do you want
me to send a patch updating gcc/config.in? (Ie, just run autoheader,
then send the gcc/config.in diff)?
macbookpro:gcc stefan$ svn status configure.ac configure config.in
macbookpro:gcc stefan$ autoheader
macbookpro:gcc stefan$ autoconf
macbookpro:gcc stefan$
macbookpro:gcc stefan$ svn diff config.in configure
Index: config.in
===================================================================
--- config.in (revision 152686)
+++ config.in (working copy)
@@ -315,6 +315,12 @@
#endif
+/* Define if your assembler supports the swap suffix. */
+#ifndef USED_FOR_TARGET
+#undef HAVE_AS_IX86_SWAP
+#endif
+
+
/* Define if your assembler supports the lituse_jsrdirect
relocation. */
#ifndef USED_FOR_TARGET
#undef HAVE_AS_JSRDIRECT_RELOCS
Index: configure
===================================================================
--- configure (revision 152686)
+++ configure (working copy)
@@ -22917,6 +22917,37 @@
fi
+ { $as_echo "$as_me:${as_lineno-$LINENO}: checking assembler for
swap suffix" >&5
+$as_echo_n "checking assembler for swap suffix... " >&6; }
+if test "${gcc_cv_as_ix86_swap+set}" = set; then :
+ $as_echo_n "(cached) " >&6
+else
+ gcc_cv_as_ix86_swap=no
+ if test x$gcc_cv_as != x; then
+ echo 'movl.s %esp, %ebp' > conftest.s
+ if { ac_try='$gcc_cv_as -o conftest.o conftest.s >&5'
+ { { eval echo "\"\$as_me\":${as_lineno-$LINENO}: \"$ac_try\""; } >&5
+ (eval $ac_try) 2>&5
+ ac_status=$?
+ $as_echo "$as_me:${as_lineno-$LINENO}: \$? = $ac_status" >&5
+ test $ac_status = 0; }; }
+ then
+ gcc_cv_as_ix86_swap=yes
+ else
+ echo "configure: failed program was" >&5
+ cat conftest.s >&5
+ fi
+ rm -f conftest.o conftest.s
+ fi
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result:
$gcc_cv_as_ix86_swap" >&5
+$as_echo "$gcc_cv_as_ix86_swap" >&6; }
+if test $gcc_cv_as_ix86_swap = yes; then
+
+$as_echo "#define HAVE_AS_IX86_SWAP 1" >>confdefs.h
+
+fi
+
{ $as_echo "$as_me:${as_lineno-$LINENO}: checking assembler for
different section symbol subtraction" >&5
$as_echo_n "checking assembler for different section symbol
subtraction... " >&6; }
if test "${gcc_cv_as_ix86_diff_sect_delta+set}" = set; then :
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [PATCH] Add support for the Win32 hook prologue (try 9)
2009-10-12 21:12 ` Stefan Dösinger
@ 2009-10-12 21:38 ` Richard Henderson
0 siblings, 0 replies; 9+ messages in thread
From: Richard Henderson @ 2009-10-12 21:38 UTC (permalink / raw)
To: Stefan Dösinger; +Cc: gcc-patches
On 10/12/2009 02:09 PM, Stefan Dösinger wrote:
> Unfortunately, something went wrong with the configure update.
Indeed, sorry about that. I've regenerated the files and committed.
r~
^ permalink raw reply [flat|nested] 9+ messages in thread
end of thread, other threads:[~2009-10-12 21:30 UTC | newest]
Thread overview: 9+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2009-10-07 19:27 Add support for the Win32 hook prologue (try 8) Stefan Dösinger
2009-10-09 13:22 ` [Patch ping] " Stefan Dösinger
2009-10-09 13:49 ` Paolo Bonzini
2009-10-09 14:53 ` Dave Korn
2009-10-09 22:38 ` Richard Henderson
2009-10-11 21:50 ` [PATCH] Add support for the Win32 hook prologue (try 9) Stefan Dösinger
2009-10-12 16:39 ` Richard Henderson
2009-10-12 21:12 ` Stefan Dösinger
2009-10-12 21:38 ` Richard Henderson
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).