[Click] Re: click memory questions

[Eddie Kohler]

Hi Erik!

On Oct 7, 2004, at 10:08 AM, Erik VandeKieft wrote:
> Hi Eddie,
>
> I'm the one that was asking you before about kernel mode click memory
> usage (using kmalloc versus vmalloc, etc.) I have to quick questions 
> for
> you:
>
> (1) Why does click currently do all its kmalloc() allocations using the
> GFP_ATOMIC flag? Wouldn't GFP_KERNEL be fine? Why must we not block?

Hmm.  GFP_KERNEL probably would be fine.  Click *may* occasionally 
allocate memory in an interrupt context (FromDevice or some such), but 
it probably shouldn't.  So I think GFP_KERNEL is worth a try.

> (2) I'm told that no matter what I do, because of limitations of linux
> 2.4, I will not be able to use more than 1GB of memory while running in
> kernel mode (even less because of the 128MB for page tables). Has 
> anyone
> run into this limitation before, and is there anything that can be done
> about it?

I don't believe anyone has.

> (3) I'm no longer doing kmalloc() for large memory requests (I use 
> vmalloc
> for everything greater than a certain size), but kmalloc still 
> eventually
> fails. It fails after only 520MB total have been allocated on a 2GB
> machine... do you know if there is some kind of statically defined 
> kmalloc
> pool limit that can be increased?

The relevant Linux source is mm/slab.c.  I don't know the answer to 
this off the top of my head.  Is 520MB including the vmalloc()ed pages?

Eddie


>
> Thanks,
> --Erik