Re: Using vdr-dpg package for bug hunting?

Sun, Dec 01, 2024 at 08:55:45PM +0100, schorpp wrote:

>(gdb) print ptsValues[0]
>$1 = 3600
>(gdb) print framesPerPayloadUnit
>$2 = 1
>(gdb) print parser->iFrameTemporalReferenceOffset
>$3 = -1

Okay, this looks like a straightforward division by zero, which in fact 
should have been fixed more than 10 years ago in VDR 2.1.7. Which 
version of VDR are you using again?

commit 57222002b2b48baa9f5c8d0f253184801ba7200a
Author: Klaus Schmidinger <[email protected]>
Date:   Sun Apr 13 13:50:04 2014 +0200

     Fixed a possible division by zero in frame rate detection

As far as I can tell, this fix is also preventing a division by a 
negative number, which would have been another thinkable source of 
SIGFPE.

In your register output, we have a zero divisor in ecx:

>(gdb) info registers
...
>ecx            0x0	0
...

>(gdb) disassemble
...
>   0x08136f4d <+1021>:	mov    0x80(%esp),%ecx
>   0x08136f54 <+1028>:	xor    %edx,%edx
>   0x08136f56 <+1030>:	mov    0x288(%ecx),%esi
>   0x08136f5c <+1036>:	mov    %ecx,%eax
>   0x08136f5e <+1038>:	mov    0xc(%eax),%eax
>   0x08136f61 <+1041>:	mov    0x280(%ecx),%ecx
>   0x08136f67 <+1047>:	add    0x8(%esi),%ecx
>
>=> 0x08136f6a <+1050>:	div    %ecx				 

There is some loading of registers from the stack, as well as zeroing 
out the more significant 32 bits of the dividend (edx). The dividend was 
calculated right before the division.

	Marko