early hang in 2.6.24 on au1550 (MIPSLE)

early hang in 2.6.24 on au1550 (MIPSLE)

[Clem Taylor]

A few months ago I switched from 2.6.16.16 to 2.6.24 on an AU1550
(MIPSLE) system. I started rolling this out to more systems, getting
ready for a new software release and discovered that on fresh powerup,
the 2.6.24 kernel sometimes (1 in 10-25 power cycles) fails to start.
The bootloader (uboot) decompresses the kernel from a jffs2 filesystem
and then jumps to it, but I don't get any serial messages from the
kernel.

If I switch back to the 2.6.16.16 kernel, everything is happy. The
annoying thing is that I have been unable to catch the problem with
the JTAG debugger connected, so I'm not sure where it is hanging.

I've been looking at diffs in the arch/mips tree and nothing has
jumped out at me.   I don't think this is a hardware problem, this
hardware platform has been fairly stable and it works just fine with
the older kernel. I was wondering if anyone has any suggestions where
I might look? Also, is anyone using 2.6.24 with a Au1550?

                       Thanks,
                       Clem

Re: early hang in 2.6.24 on au1550 (MIPSLE)

[Ricardo Mendoza]

Clem Taylor wrote:

> A few months ago I switched from 2.6.16.16 to 2.6.24 on an AU1550
> (MIPSLE) system. I started rolling this out to more systems, getting
> ready for a new software release and discovered that on fresh powerup,
> the 2.6.24 kernel sometimes (1 in 10-25 power cycles) fails to start.
> The bootloader (uboot) decompresses the kernel from a jffs2 filesystem
> and then jumps to it, but I don't get any serial messages from the
> kernel.
> 
> If I switch back to the 2.6.16.16 kernel, everything is happy. The
> annoying thing is that I have been unable to catch the problem with
> the JTAG debugger connected, so I'm not sure where it is hanging.
> 
> I've been looking at diffs in the arch/mips tree and nothing has
> jumped out at me.   I don't think this is a hardware problem, this
> hardware platform has been fairly stable and it works just fine with
> the older kernel. I was wondering if anyone has any suggestions where
> I might look? Also, is anyone using 2.6.24 with a Au1550?

Nothing? What about time layer switch to make use of cevt and csrc
devices for example. Gotta keep in mind that  2.6.16 is archaic; you
should write an early printk implementation for au1550 and see where it
dies. With another board once I experienced something similar and it
happened to be timer interrupts getting skipped at times, so, you might
be dying in calibration if your problem is similar.


     Ricardo