[Mingw-notify] [ mingw-Bugs-1211187 ] -pg causes wrong-code generation with FASTCALL

[SourceForge.net <no...@so...>]

Bugs item #1211187, was opened at 2005-05-30 19:55
Message generated for change (Comment added) made by dannysmith
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=102435&aid=1211187&group_id=2435

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: None
Group: None
Status: Open
Resolution: None
Priority: 5
Submitted By: Oliver Stoeneberg (kidkat)
Assigned to: Nobody/Anonymous (nobody)
Summary: -pg causes wrong-code generation with FASTCALL

Initial Comment:
The resulting executable will crash when compiled with
"-pg".

typedef struct STRING_POOL
{
  int a;
} STRING_POOL;
typedef struct XML_Memory_Handling_Suite
{
  int t;
} XML_Memory_Handling_Suite;
#define FASTCALL __attribute__((stdcall, regparm(3)))
static void FASTCALL
poolInit(STRING_POOL *pool, const
XML_Memory_Handling_Suite *ms)
{
  pool->a = 1;
}

int main(void)
{
  STRING_POOL a;
  XML_Memory_Handling_Suite d;
  poolInit(&a, &d);
}

Here is the backtrace:

Program received signal SIGSEGV, Segmentation fault.
0x0040e456 in poolInit (pool=0x40e44d, ms=0x0) at
src/expat/xmlparse.c:5874
5874      pool->blocks = NULL;
(gdb) bt
#0  0x0040e456 in poolInit (pool=0x40e44d, ms=0x0) at
src/expat/xmlparse.c:5874
#1  0x0040d339 in dtdCreate (ms=0x3f2554) at
src/expat/xmlparse.c:5428
#2  0x00403193 in parserCreate (encodingName=0x0,
memsuite=0x0, nameSep=0x0,
    dtd=0x0) at src/expat/xmlparse.c:737
#3  0x00402fb7 in XML_ParserCreate_MM
(encodingName=0x0, memsuite=0x0,
    nameSep=0x0) at src/expat/xmlparse.c:671
#4  0x00402f4b in XML_ParserCreate (encodingName=0x0)
    at src/expat/xmlparse.c:648
#5  0x00402b7c in process (is=0x7803a690,
os=0x7803a6b0) at src/xml2info.c:766
#6  0x00402efe in main (argc=1, argv=0x3f24c8) at
src/xml2info.c:835

Taken from GCC bugzilla:
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21810


----------------------------------------------------------------------

>Comment By: Danny Smith (dannysmith)
Date: 2005-05-31 12:03

Message:
Logged In: YES 
user_id=11494

Confirmed,

There are two ways to fix this:

1) Add some inline asm to current code in profile/profile.h to 
save and restore caller-clobbered registers. This hack is a bit 
fragile since it depeds on the the real _mcount in mcount.c 
always being inlined into mcount. That can be ensured with 
new gcc __attribute__ ((always_inline)), but not with 2.95. So, 
it would need some GCC version check quards around the 
new additions.

2) A more robust fix that replaces the macro definintion of 
mscount with an assembly stub that calls an external 
_mcount. This is the way that glibc does it and in fact, the 
assembly code, as simple as it is, would inherit the LGPL 
copyright of the source that I looked at.  Maybe this is not a 
major issue since profiled exe's are already under the Cygwin
license due to the code in profil.h. 

Both of these also require a trivial patch to gcc to prevent 
unnecessary use of %edx to store profile counters

Any comments?

Danny 

----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=102435&aid=1211187&group_id=2435