Hamish Moffatt wrote: > > > http://www.linuxtv.org/pipermail/linux-dvb/2005-September/004481.html > > Doesn't this just ignore the errors? Admittedly that might not be a bad > solution (although you will get some corruption when it occurs). It also adds resyncs to the RISC program so that the DMA-engine can recover. And it lowers to fifo-threshold so that the errors are less often. But yes, you get some errors in the stream - the PCI-bridge was to slow to respond to the DMA-request and the data is lost. > The thread that the OP mentioned discussed interaction between the BT878 > and sata drivers. For me the problem was much worse when I started > streaming to SATA disk, but I had seen it on occasion before that. Right, heavy DMA may delay the bus grant for the BT878 long enough that its fifo overflows. > But I do recall that it seemed better in earlier kernels (around 2.6.3?). I had the same impression. 2.6.8 seems to be better. But I've looked left and right and couldn't find a single bit of code in the kernel that would effect the regression :-( > Maybe there is longer interrupt latency for some reason in newer kernels. The problem has nothing to do with interrupt latency. Its the DMA arbitration and that seems to be configured by the BIOS. Ciao, ET.
