[sf-lug] memory integrity checking / excluding bad memory

[Asheesh Laroia]

On Sat, 2 Feb 2008, Kristian Erik Hermansen wrote:

> On Feb 2, 2008 2:54 PM, Alex Kleider <a_kleider at yahoo.com> wrote:
>> Also, please, what's the command to exclude the memory that is
>> discovered to be bad?
>
> Sameer already told you the other answer, but here's this one.  There
> is a neat trick that Asheesh came up with, that I didn't know of, that
> allows you to restrict the running kernel memory mapping to a selected
> allocated chunk.  Basically, we passed the kernel the mem=512M option
> so that we could exclude the dim module that we thought was bad, and
> thus prevent it form being utilized in the OS during installation.

(-:

> However, we are not sure how that option is implemented.  Does it grab
> the first 512 megs of RAM, distribute it across both sticks, just grab
> randomly from a chunk pool?

It should do the lower 512 physical megabytes of RAM.

> We didn't investigate.  However, it
> didn't work (perhaps the first DIMM was bad?), but I;m sure if anyone
> has info on how that is implemented we would love to hear about it :-)
> Not gonna go dig the sources myself right now...heh

Another more reasonable option is to use the badmem patch - 
http://badmem.sourceforge.net/ - that lets you specify exactly which parts 
of RAM are bad.  (You can get this list from memtest86.)

But in general, such hacks are frowned-upon - if you have bad hardware, 
who knows if more of it will fail in the next ten minutes?

But if it works for you well enough, then so be it....

-- Asheesh.

-- 
Just because your doctor has a name for your condition doesn't mean he
knows what it is.