Date: Fri, 25 Sep 1998 11:01:47 +0200 (CEST)
Message-Id: <199809250901.LAA17707@login-2.eunet.no>
From: "Gisle Vanem" <giva AT bgnett DOT no>
To: djgpp AT delorie DOT com
Cc: eliz AT is DOT elta DOT co DOT il
Subject: Re: preserving exception context
MIME-Version: 1.0
Content-type: text/plain; charset=cp850
Content-Transfer-Encoding: 8bit
Precedence: bulk

Eli Zaretskii <eliz AT is DOT elta DOT co DOT il> said:

> ..  The only thing that's lost is the
> registers' dump.  If you need that, you will probably need to hack the
> signal-handling code (dpmiexcp.c in djlsr201.zip) to store away the
> exception structure.  But do you really need the registers?

Wild values in EDI/ESI often indicates a pointer error. Using Fortify
(a debug malloc lib), EDI/ESI with value 0xA9A9A9A9 tells me I'm using
a dynamic array after it's freed. So yes, the registers are important.

To preserve the exception structure at `__djgpp_exception_state_ptr',
I simply copied it to a local jmp_buf and used that as exception structure
on a nested crash. Works fine AFAIK.

> raise(SIGABRT) in v2.01 just prints "Abort!" and exits.  So you don't get
> any traceback at all, if shutdown doesn't crash.

This little test program tells me otherwise:

#include <signal.h>

void shutdown (int crash)
{
  if (crash)
    *(char*)-1 = 1;
}

void exc_handler (int sig)
{
  signal (SIGSEGV,SIG_DFL);
  shutdown (0);
  raise (SIGABRT);
}

int main (void)
{
  signal (SIGSEGV,exc_handler);
  *(char*)-1 = 1;
  return (0);
}

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

In djgpp 2.01:

  Exiting due to signal SIGABRT
  General Protection Fault at eip=000015aa
  eax=00000000 ebx=0004db80 ecx=00001564 edx=00000123 esi=00000054 edi=0000a2d4
  ebp=0004a2a8 esp=0004a2a8 program=C:\TEMP\MISC\EXC2.EXE
  cs: sel=00a7  base=10000000  limit=0005ffff
  ds: sel=00af  base=10000000  limit=0005ffff
  es: sel=00af  base=10000000  limit=0005ffff
  fs: sel=008f  base=00004200  limit=0000ffff
  gs: sel=00bf  base=00000000  limit=ffffffff
  ss: sel=00af  base=10000000  limit=0005ffff

  Call frame traceback EIPs:
    0x000015aa   _main+26
    0x00001b0e   ___crt1_startup+138

In djgpp 2.02:

  Exiting due to signal SIGABRT
  General Protection Fault at eip=000015be
  eax=00000000 ebx=00001000 ecx=00000488 edx=00008240 esi=00000054 edi=0000b0f0
  ebp=0008b0d0 esp=0008b0d0 program=C:\TEMP\MISC\EXC2.EXE
  cs: sel=00a7  base=10000000  limit=0009ffff
  ds: sel=00af  base=10000000  limit=0009ffff
  es: sel=00af  base=10000000  limit=0009ffff
  fs: sel=008f  base=00004220  limit=0000ffff
  gs: sel=00bf  base=00000000  limit=0010ffff
  ss: sel=00af  base=10000000  limit=0009ffff
  App stack: [0008b0f0..0000b0f0]  Exceptn stack: [0000afd0..00009090]

  Call frame traceback EIPs:
    0x000015be   _main+46
    0x00001bba   ___djgpp_exception_table+166


I notice the selector bases are the same, but some limits are much larger
(256kB) in 2.02. What the extra space used for?

Gisle V.