[linux-dvb] [RFC-final] videobuf tree

Mauro Carvalho Chehab mchehab at infradead.org
Sun Oct 7 03:52:27 CEST 2007

> Yes...  I cloned today's master branch, including your changeset cited above. 
> I was sure to do 'make rminstall' in an older tree, to remove all traces of 
> the older video_buf module before installing the new modules.

I suspect that there's a race condition related with the way we do mmap. I 
got a similar issue with MSI TV @nyware AD NB (saa7134) and a dual core 
AMD 64 notebook. The same device on a single core notebook works like a 
charm. I had this trouble with both old and new videobuf.

I also noticed, on tm6000 driver, that, on my tests with newer kernels, 
the calling order for file .mmap handler and videobuf_iolock has changed. 
I did a workaround at videobuf-vmalloc for this case:

         /* It seems that some kernel versions need to do remap *after*
            the mmap() call
         if (mem->vma) {

I suspect that this bug happens on certain workloads, and depends on HZ 

Maybe adding a wait_event_timeout() will solve. I'll do some tests here.


More information about the linux-dvb mailing list