Re: Memory tester for powerpc?
ozymandias G desid writes:
> On Sat, Jan 26, 2002 at 03:46:04AM -0500, Michael D. Crawford wrote:
>> I have an article that, in part, discusses some memory and burn-in
>> tools, including memtester. See:
memtester does this:
while(/* whatever */){
/* p1 and p2 are pointers */
*p1++ /= foo; /* foo is from rand() */
*p2++ /= foo; /* foo is from rand() */
}
Then the buffers are compared. This same basic idea is
then repeated for other math operations! The above was
division; also you get multiplication, addition, etc.
I spent a few hours with Google, looking for info on how
memory chips are laid out. The picture is pretty grim,
but not hopeless. For a N*M pattern, you could write out
M rows of N*Q bits where Q is a power of 2 from 64 to 16k
divided by N. (assuming N is a power of 2) At some point
you will most likely get your desired pattern on the chip.
Better luck might be had with patterns that don't divide
powers of two. I'd go with 5, 7, 9, 11, and 13 bits in
the patterns. This also helps handle the case of memory
getting the data lines in a funny order.
> BTW, do any of you have a .config for a Cube that's been
> stripped down to the minimum necessary to actually support all of the
> Cube's built-in hardware (along with USB and Firewire support) that's
> tailored for benh kernels? I've been trying to do it myself, but I
> apparently keep exercising some weirdness, because I get huge numbers
> of "IN from blah at blah" errors in the kernel message buffer.
Disable UHCI USB, normal 16x50 serial ports, the PC clock thing,
the plain MDA/CGA/EGA/VGA text-mode console, and laptop stuff.
The USB audio driver is pretty broken. I can often get it to play
something after a reboot. Sometimes processes get stuck. My bug
report (with disassembly and C source even!) got ignored.
Reply to: