[XviD-devel] [BUG] Lot of unitialized reads
Edouard Gomez
ed.gomez at free.fr
Tue Nov 4 15:41:28 CET 2003
Edouard Gomez (ed.gomez at free.fr) wrote:
> valgrinding XviD shows we are reading lot of unitialized data in the
> first frames (i can't tell if the first or the second one). It's all
> happening when using bframes. Using Pframes only shows 0 errors (good
> ;-).
This problem seems to be really weird.
When doing a gdb session at the unitialized value used for jumping, the
backtrace is *strange*:
[...]
#4 0x403f578c in CheckCandidate32I (x=37, y=1, data=0xffffffe0, Direction=31)
at ../../src/motion/vop_type_decision.c:49
#5 0x403f578c in CheckCandidate32I (x=0, y=0, data=0xbffff2b0, Direction=0)
at ../../src/motion/vop_type_decision.c:49
#6 0x403f5fc9 in MEanalyzeMB (...)
[...]
CheckCandidate32I calls itself ... and if you look at the function
parameters, they're weird, specially the data value that points to
nowhere (that is the cause of the unitialized value used in jump)
I can reproduce this with whatever sequences and max_bframes=1, it
happens for the second buffered bframe.
Can someone confirm this unitialized reads/jumps with old glibc and old
valgrind versions ? I suspect a valgrind bug corrupting the stack.
valgrind 20031012-3 A memory debugger for x86-linux
libc6 2.3.2.ds1-9 GNU C Library
latest xvid devapi4 branch
--
Edouard Gomez
More information about the XviD-devel
mailing list