Re: MPC755 L2 Cache data parity error

Re: MPC755 L2 Cache data parity error

[Chuck Brown]

Did you ever solve this problem?  I have a module with a similar
architecture as the one you described and I can get this same data parity
failure just by making the L2 cache private memory and copying a file into
the L2 cache.

Thanks
Chuck Brown

MPC755 L2 Cache data parity error

[Shenrong]

Hi,
   Anybody has experienced with MPC755 L2 Cache data parity error?
We use XPC755B Rev. E at 400M Hz, and 1M L2 Cache at 200M Hz.
After the Linux is up and with busy disk accessing or network
operations for several hours, 755 will receive L2 Cache data parity checking
error machine check. If we don't enable L2 Cache data parity checking,
the system will still crash for some time silently. But if we don't enable
L2 Cache, this problem won't occur. Reduce the L2 Cache freq to
160M Hz, 133M Hz or even less won't avoid this problem. It seems
the system will be some more stable(still has problem) if we set the L2
Cache
size to 512k or 256k, but don't know why. We use both the PBGA and
CBGA packages, seems CBGA version are some more stable(still has problem),
don't know why too.
   It's more strange that I can simply let the error occur under Linux with
continuous file copying in ram disk. But I can't let it occur under PPCBoot
with whatever memory operations for even one time, whether use the L2 Cache
memory as L2 Cache or as private memory. The L2 Cache has the same setting
in the above two circumstances.
   Any ideas and suggestions? Thanks.

Shenrong


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: MPC755 L2 Cache data parity error

[Chris Zimman]

On Sun, Jun 15, 2003 at 11:06:38PM +0800, Shenrong wrote:
> size to 512k or 256k, but don't know why. We use both the PBGA and
> CBGA packages, seems CBGA version are some more stable(still has problem),
> don't know why too.

The first thing I thought of when I read this is heat issues -- especially because
of the mention of PBGA vs. CBGA.  Is your heat sinking/cooling adequate?

I had a problem once with a custom 405GP board that was not entirely dissimilar.
Running the OpenSSL 'speed' benchmark, the system would lock up doing the MD2 tests.
I couldn't fathom how a userland program could lockup the entire system.  One day,
just by chance, I happened to stand the board on end -- no more crashes.  The
difference had been the airflow underneath the device because it didn't have a
heatsink on it.  Consequently, I put a heatsink on the 405 and that fixed the
problem altogether.

>    It's more strange that I can simply let the error occur under Linux with
> continuous file copying in ram disk. But I can't let it occur under PPCBoot
> with whatever memory operations for even one time, whether use the L2 Cache
> memory as L2 Cache or as private memory. The L2 Cache has the same setting
> in the above two circumstances.

The only difference I could think of right away here is that under
Linux, you're doing burst reads and writes.

--Chris

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/