Hi all,
(since many people are having the same problem with lavrec, I'd better
handle them all together, since they seem all the same)
I suppose everyone here reporting this problem has an Athlon or a
P-III/IV as CPU, which would mean that everyone has the same problem
:-).
On Sun, 2002-01-06 at 18:28, Hakan Lennestal wrote:
> Program received signal SIGSEGV, Segmentation fault.
> [Switching to Thread 3076 (LWP 26316)]
> 0x400b4ccd in jcquant_sse () from /usr/lib/liblavfile-1.6.so.0
> (gdb) bt
> #0 0x400b4ccd in jcquant_sse () from /usr/lib/liblavfile-1.6.so.0
> #1 0x400b3be2 in forward_DCT_x86float32 () from /usr/lib/liblavfile-1.6.so.0
> #2 0x400af1ce in compress_data () from /usr/lib/liblavfile-1.6.so.0
> #3 0x400ac1a2 in jpeg_write_raw_data () from /usr/lib/liblavfile-1.6.so.0
> #4 0x40413622 in encode_jpeg_raw () from /usr/lib/liblavjpeg-1.6.so.0
[...]
This seems as a problem in libjpeg-mmx' SSE optimalizations during JPEG
encoding at first sight. The thing is, I cannot reproduce it when using
encode_jpeg_raw() in yuv2lav (my desktop is a P-II w/o SSE, the only
P-III here is my laptop w/o a TV-card... In other words, I can't test it
with lavrec directly...). Neither could you Andrew...
Could it be a thread problem? Either in libjpeg-mmx or in liblavrec
itself (unproper locking, for example, between the capture and the
encoding thread?). I don't think it's a problem but it could be the
case, of course...
A second possibility: could jcquant_sse() - in some way - be trying to
write data to the YUV buffer? This is not really possible since it's a
kernel buffer and I'm mmap()'ing it read-only (PROT_READ) in liblavrec's
lavrec_software_init(). Of course, this can easily be changed, by
applying the following patch:
==========================================================
--- liblavrec.c Sun Jan 6 21:32:32 2002
+++ liblavrec_quick_hack.c Sun Jan 6 21:32:09 2002
@@ -1032,7 +1032,7 @@
/* Map the buffers */
settings->YUV_buff = mmap(0, settings->softreq.size,
- PROT_READ, MAP_SHARED, settings->video_fd, 0);
+ PROT_READ | PROT_WRITE, MAP_SHARED, settings->video_fd, 0);
if (settings->YUV_buff == MAP_FAILED)
{
lavrec_msg(LAVREC_MSG_ERROR, info,
==========================================================
In the case of yuv2lav, it's not a kernel but a userspace buffer which
can be written on, so that could be an potential problem (for the ones
who have compiled from source, please test the above patch)...
That's about all I can think of... Any other ideas, Andrew?
Ronald
--
- .-.
- /V\ | Ronald Bultje <rbultje@xxxxxxxxxxxxxxxxxxx>
- // \\ | Running: Linux 2.2.19 and OpenBSD 2.8
- /( )\ | http://ronald.bitfreak.net/
- ^^-^^
