"Preston L. Bannister" <[EMAIL PROTECTED]> writes:
> From: Eric W. Biederman
> > - In assembly you initialize and turn on RAM, and you initialize
> > any other hardware you need to assist in initializing the RAM.
> > We don't test the memory because quick memory tests never fail,
> > on ram that has passed even the most cursory quality control.
>
> I think you have a point here - RAM tests do seem mostly a waste
> of time - but this is at least partly a mistake.
>
> Every once in a while you do get bad RAM. It might be worth checking
> a small chunk of memory (say) every 32 megabytes, just to see if it
> at least minimally works. Much less mysterious this way... :)
Well I can see testing at few bytes per dimm. And as we ramp up the
volume of the linuxBIOS deployments this may be an issue. Where we do
something close is to properly initialize ECC SDRAM, we zero all of
memory, a gigabyte takes about a second.
And we do have a memory test, but it is only present as a debugging
assistant to see if we have initialized memory correctly.
If you want to code up and test something more extensive feel free.
> As an aside - shouldn't the mailing list software set the "Reply-to:"
> header to [EMAIL PROTECTED] ?
Nope, most mailers have a reply, and a reply to all feature. And in general
setting a reply-to header is just rude. And generally when I reply
to more people then just the list that is exactly what I intend.
Eric
