On Sun, Feb 21, 2010 at 02:32:34PM +0100, Harald Welte wrote:

...

I've noticed that hello_world and compal_dsp_dump now crash pretty early. I've traced it down to talloc_zero panicing by running out of free msgb. Augmenting the number to 64 makes the software go a little further but they still crash in the end.

I'm hoping that the person who wrote this part will see an obvious place where the problem lies :)

that would probably be me. I have reproduced the problem and will look at it later today. Interestingly, it doesn't occur with l1test or layer1, who send much more data over the serial port...

it seems like it is really just the normal console printing. Those programs print a number of relatively short lines, and at the moment we use one msgb for each line, even if it is not full yet.

So only dumping the PLL registers twice is already sufficient to use up all the msgb()s.

I will think about how to solve this. Either we introduce some busy-waiting until more space is available, or I will try to fill existing buffers even beyond the end-of-line.