Re: Resetting a Broadcom in software

[Ralf Baechle <ralf@linux-mips.org>]

On Wed, Sep 06, 2006 at 02:41:36PM -0700, Jonathan Day wrote:
> Date:	Wed, 6 Sep 2006 14:41:36 -0700 (PDT)
> From:	Jonathan Day <imipak@yahoo.com>
> Subject: Resetting a Broadcom in software
> To:	linux-mips@linux-mips.org
> Content-Type: multipart/mixed; boundary="0-1897457798-1157578896=:54718"
> 
> Hi,
> 
> A co-worker wrote the following test of the Broadcom's
> maths abilities and discovered that it reboots some
> (but not all) MIPS processors it has been tested on.
> It'll reboot the Sentosa, for example, but NOT the
> Swarm.
> 
> (Apologies for the ugly coding, btw.)
> 
> You just make the first file, the ATL_ file gets
> included into it. The compiler flags I'm using are:
> 
> -march=sb1 -mabi=64 -fomit-frame-pointer -O3 -mips64
> -mfused-madd
> 
> The program doesn't link to anything and no linker
> flags are needed.
> 
> This begs three questions:
> 
> 1) What is happening to cause the CPU to reset? (It's
> not a kernel bug, it's an actual CPU reset)
> 
> 2) What is NOT happening on the Swarm, allowing it to
> work fine?
> 
> 3) Is the problem in the category of "preventable in
> hardware", "preventable in the kernel", or
> "preventable by slowly roasting those coders who write
> like this"?

This is not a problem I know of but given your description it sounds very
much like a hardware issue.  Can you find about the exact versions of the
1250 on the various board?  With the FPU being on chip I would expect
some correlation between the chip revision and this issue.

  Ralf