Serial interrupt overrun and hard deadlock on 3.3.2 & 3.5, freescale 8347.

[Christian Melki]

Hi.

I've spent a couple of hours lifting our freescale 8347 platform from 2.6.32 to 3.3.2 and just now to 3.5. The 3.x kernels are giving me some problems.
I get serial input overruns especially when cpu is loded, but I can reproduce overruns even at 0 other load. I can't even remember when I saw overruns like this the last time, must have been in 2.2.x or early 2.4.x.
Also I can reliably trigger a hard kernel lock when using a USB-camera and acessing sequential blocks from the mtd device (read to verify flash). I'm thinking that this maybe has something to do with the reworks of IRQ handling in the powerpc arch lately.

I've tried most things I can think of to see if I can get a symptom change somewhere (incl, preemption, smp-kernel etc etc) but I get nothing.
Hard lock detection seems to be NMI bound on x86 only (?), so I can't use that.
A jtag-debugger does not give me much either, since I can't record the instruction stream in realtime anyway.

I'm kind of starved for ideas, so any help or tip as to what might be wrong would be appreciated.

Regards,
Christian