RB wrote:
> After much debugging, I've uncovered two cases that cause this oops;
> although there are several variants throughout the code, the following
> two are representative:
> 
> 1.  Disabling interrupts - cb_writel(socket, CB_SOCKET_MASK, 0x0);
> 2.  Checking interupts - exca_readb(socket, I365_INTCTL);
> 
> I'm really into uncharted personal waters here, so any driver devs
> wanting to stand up and be counted would be awfully welcome.  IRNOT1.
> 
> That said, this seems counterintuitive - why disable by writing to one
> register (3) and check by reading from another (4)?  Do we have the
> registers wrong on this [probably broken] TI implementation?
I don't know much about that interrupt thing that you mentioned, but for
improved stability, you should switch to trunk when playing with Linux 2.6
on Broadcom. There are some cache coherency / memory management issues in
7.09, which are almost completely fixed in trunk.

- Felix
_______________________________________________
openwrt-devel mailing list
[email protected]
http://lists.openwrt.org/cgi-bin/mailman/listinfo/openwrt-devel

Reply via email to