linux-mips
[Top] [All Lists]

Re: Kernel space access to 2GB of physical memory

To: Alex Gonzalez <langabe@gmail.com>
Subject: Re: Kernel space access to 2GB of physical memory
From: Ralf Baechle <ralf@linux-mips.org>
Date: Thu, 2 Aug 2007 23:23:07 +0100
Cc: linux-mips <linux-mips@linux-mips.org>
In-reply-to: <c58a7a270708020647q7a0c55f4l6904e864609c7304@mail.gmail.com>
Original-recipient: rfc822;linux-mips@linux-mips.org
References: <c58a7a270708020647q7a0c55f4l6904e864609c7304@mail.gmail.com>
Sender: linux-mips-bounce@linux-mips.org
User-agent: Mutt/1.5.14 (2007-02-12)
On Thu, Aug 02, 2007 at 02:47:37PM +0100, Alex Gonzalez wrote:

> I have a system booting the 2.6.12-rc3 32bits kernel on an RM9000
> processor and I am trying to get my head around how to access up to
> 2GB of physical memory with a mem=256MB command line booting
> parameter. I read that on the MIPS kernel this is split 2GB/2GB
> between kernel and user space, so the kernel should be able to access
> directly 2GB of physical memory.

No.  You assume there is nothing else in the 2GB of kernel address space.

> Kernel drivers will access the memory above 256MB using ioremaps. I
> need it this way as the memory above 256MB will be used in special
> ways.
>
> What I struggle to understand is whether it would be possible to
> access all the memory up to 2GB using the ioremap method.

Ioremap is meant to be used for MMIO regions only.

> The only way I can think of achieving this would be to use ksseg and
> dynamic TLB entries to access it in 256MB chunks. The fact that I can
> access ksseg must mean that the kernel is not clearing the TLB entries
> that the bootloader sets up before launching the kernel, so I would
> expect to be able to add/remove TLB entries dynamically without
> affecting the kernel's own memory management.
> 
> 1) Is there a simpler mechanism to achieve this?

It's called 64-bit kernel.

> 2) Any ill effect on the kernel from the method described above?

You're basically reinventing high memory in an application optimized
variant and that probably includes a similar set of problems.

  Ralf

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