Help about Freescale p2020ds problem under cpu load test

[Benjamin Herrenschmidt]

On Mon, 2010-03-29 at 12:11 +0800, Li.Wang wrote:
> Hi Benjamin Herrenschmidt,

Well, I'm sure you'll have more chance with this one asking somebody
from Freescale :-)

It does look like there may be a coherency problem but I don't know
those chips well enough to tell

Cheers,
Ben.

> I use Freescale p2020ds board and kernel linux 2.6.33.
> Under cpu load test which test case is in attachment
> (#./cpu_load_multi.sh 500),
> the follow issues happen after a long time:
> =======================
> find: md5sum terminated by signal 6
> 
> find: md5sum terminated by signal 11
> =======================
> 
> If use kernel linux 2.6.27(porting p2020ds bsp to it), it's easy to 
> reproduce the issue:
> =======================
> find: ../nptl/sysdeps/unix/sysv/linux/powerpc/../i386/../fork.c:138:
> __libc_fork: Assertion `((void)(self), (((struct pthread *)
> (__thread_register - 0x7000 - (sizeof (struct pthread) + ((sizeof
> (tcbhead_t) + __alignof__ (struct pthread) - 1)&   ~(__alignof__ (struct
> pthread) - 1))))))->tid) != ppid' failed.
> 
> find: md5sum terminated by signal 6
> 
> find: md5sum terminated by signal 11
> 
> ./cpu_load.sh: line 6:  7944 Segmentation fault      find /usr -type f
> -exec md5sum {} \;>    /dev/null
> =======================
> 
> If turn off CONFIG_SMP, the issue disappear.
> 
> If use Freescale 8572ds board, the issue disappear, too.
> I know the two board is same about cpu(e500v2), but their version have a 
> little difference:
> ===================
> fsl_8572ds:
> cpu        : e500v2
> revision    : 3.0 (pvr 8021 0030)
> 
> fsl_p2020ds:
> cpu        : e500v2
> revision    : 4.0 (pvr 8021 1040)
> ===================
> 
> Please help me to check the problem.
> Thanks,
> LiWang.