Re: Some VM questions

[Joseph A Knapka <jknapka@earthlink.net>]

Pichai Raghavan wrote:
> 
> Hi All,
> 
> I was going thru the VM initialisation on my 2.4.2 kernel and had
> the following doubts:
> 
> (i) If alloc_page is called with HIGHMEM, then it allocates pages
>  from all of the 3 zones; for NORMAL it gets from NORMAL and DMA
> and for DMA specifically from DMA. What is the logic behind this?

The rationale is that we want to allow allocs to succeed as
much as possible, so less-constrained allocs should be permitted
to succeed from more-constrained zones. I had this same
question when I was looking at the zone code, and I
concluded that the allocator does not try hard enough to
get pages from the proper zone - it's possible to alloc
DMA pages on a GFP_HIGHMEM alloc, even when there are
free pages in ZONE_HIGH.

> If a certain zone is marked as DMA and if kernel is stealing pages
> for normal allocation then won't it be unfair to device driver
> that is expecting a DMA page?

Yes, I think so, but that's the way it is.
 
> (ii) The initialisation of zones happens in order DMA, NORMAL and
> HIGHMEM. (struct zone_names etc...) Is there any reason why this
> is being done? The kernel is assuming that pages marked for DMA
> precedes the pages meant for NORMAL.

On PC architectures, that's a hardware constraint: usually
only the first physical 16MB of RAM are DMA-able. I don't know
how this works on other architectures, though.

> What happens if I change this
> order?

I suspect you'd find a lot of device drivers would be
unable to utilize DMA.

Cheers,

-- Joe
--
Kernelnewbies: Help each other learn about the Linux kernel.
Archive:       http://mail.nl.linux.org/kernelnewbies/
FAQ:           http://kernelnewbies.org/faq/