[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

R: problems using DMA



Dear Mikael,

Thank you for your reply. here below we go to answer to your requests:

Q) Is this the same problem as in your last email?
A) Yes, it is. However we moved back to a previuos and simpler version of
the drivers, using a single DMA buffer instead of the multi-buffer version.

Q) Is the same callchain reported by ksymoops?
A) No, that callcahin refferred to the multi-buffer version of the driver.
In the simpler version (sigle DMA buffer) the log stops before the kernel
oops trace ends.

Q) Does the same problem occur when you do lots of DMA to other interfaces?
A) The very same problem not, but we report different abnormal behavior in
the following test:
receiving video stream in NON-DMA mode and sending over IP: seldomly random
and very long long sequences of chars come out from the debug port (ttyS0),
while image transmission continues.
Apart video-stream reception, we do not use explicit call to DMA in other
ways.


Q) I think you may be affected by the DMA cache bug (refer to ethernet
driver etc).
A) We did experience the ethernet bug, but after the 2.4.19 kernel upgrade
on the ethernet side the system began working fine with the Etrax 100 LX
board.

Q) Can you send me the code?
A) Yes, we are going to send you the single DMA buffer driver version at
your mail address.

Q) Try the following (on a private network...)

10 simultaneous ping -f
ping -l 65536
ping -f -s 1400
ping -l 65536 -s 1400

A) We tried the test suggested above, and the processor didn't hang.

Q) Which revision of ETRAX 100LX are you using (there is a 5 digit number on
the chip indicating the version e.g. 17854)?
A) Version 2 (number 17854).

Best Regards.

Stefano Vicenzetto

Comerson