linux-mips
[Top] [All Lists]

Re: Generic DISCONTIGMEM Support on 32bit MIPS

To: Jason Gunthorpe <jgg@debian.org>
Subject: Re: Generic DISCONTIGMEM Support on 32bit MIPS
From: William Lee Irwin III <wli@holomorphy.com>
Date: Fri, 25 Jan 2002 13:52:01 -0800
Cc: Phil Thompson <phil@river-bank.demon.co.uk>, linux-mips@oss.sgi.com
In-reply-to: <Pine.LNX.3.96.1020125141828.5657B-100000@wakko.deltatee.com>; from jgg@debian.org on Fri, Jan 25, 2002 at 02:20:33PM -0700
Organization: The Domain of Holomorphy
References: <3C51838A.174F8712@river-bank.demon.co.uk> <Pine.LNX.3.96.1020125141828.5657B-100000@wakko.deltatee.com>
Sender: owner-linux-mips@oss.sgi.com
User-agent: Mutt/1.3.17i
On Fri, 25 Jan 2002, Phil Thompson wrote:
>> The first question is: has anybody already done this? Particularly as,
>> once you've identified where the holes are, the code isn't board
>> specific.

On Fri, Jan 25, 2002 at 02:20:33PM -0700, Jason Gunthorpe wrote:
> Is this of any help?
> http://kt.zork.net/kernel-traffic/kt20011112_141.html#6
> William Irwin [*] announced:
> A number of people have expressed a wish to replace the bitmap-based
> bootmem allocator with one that tracks ranges explicitly. I have written
> such a replacement in order to deal with some of the situations I have
> encountered. 
> [...]

I ran into some code acceptance issues in three places:
(1) I used trees
(2) I didn't go about changing the arch-specific code to
        actually simplify the calling sequence as it appeared
        in arch-specific code.
(3) it is a whole-hog rewrite of bootmem.c, which perhaps attracted
        flak from the original author

The last bits of this I released are in:
        ftp://ftp.kernel.org/pub/linux/kernel/people/wli/bootmem/

I'm not sure it addresses all the issues that arise here -- largely
it just avoids some code complexity in laying out the bootmem bitmaps.

DISCONTIGMEM as I understand it just minimally adjusts the core bootmem
so it can handle things at all, and then focuses on the actual hard
parts needed for things to work well on larger systems.

(Of course, that's an extremely vague description of the difference, but
I won't go about reciting featuresets aside from this high-level stuff.)

Cheers,
Bill

<Prev in Thread] Current Thread [Next in Thread>