ppc64 oops..

Benjamin Herrenschmidt benh at kernel.crashing.org
Tue Nov 15 18:07:12 EST 2005


On Mon, 2005-11-14 at 21:46 -0800, Linus Torvalds wrote:
> 
> On Mon, 14 Nov 2005, Linus Torvalds wrote:
> > 
> > Hmm. I _thought_ I had just 2GB (possibly 4GB) in this machine,
> 
> Yeah. 4GB. My other G5 has just 2GB (I had to upgrade myself, I think it 
> came with 512M from Apple).
> 
> So it seems like it's just the IO hole in the middle that got magically 
> "added" as memory, with real memory at 0-2GB and 4GB-6GB.

Yes, that would be the problem. The hole was not properly accounted
for .

When we setup the zones, we do pass the hole size (but the zone stuff
doesn't seem to have a way to tell _where_ the hole is)

I'm not sure how the memory init is supposed to work any more, that area
had too much changes that I didn't follow. What we do is to set a
bootmem map fo the whole size (6Gb here) and then

free_bootmem() on the actual RAM chunks in there, and memory_present()
too.

free_area_init_node() is called with a zone size of 6Gb but a hole size
of 2Gb.

Ben.





More information about the Linuxppc64-dev mailing list