Re: Help with mapping memory into kernel space?

From: Richard B. Johnson
Date: Wed Aug 18 2004 - 11:01:34 EST


On Wed, 18 Aug 2004, Daniel Gryniewicz wrote:

> On Tue, 2004-08-17 at 22:39 -0700, Brian McGrew wrote:
> > Good day All:
> >
>
> <snip>
>
> > The overall problem is that the more system memory we install,
> > the fewer IBB's we can use. For instance, 256MB lets us use
> > four IBB's; 512MB lets us use three IBB's and so on. Basicly,
> > the kernel blows up trying to map memory. Each IBB has two
> > banks of 64MB of RAM on them which we try and memmap to system
> > memory for speed of addressing. So essentaily, we're sending
> > out four camera systems with only 256MB of memory which is only
> > about one quarter of what we need.
>
> On x86, the kernel has 1 GiB of address space. Try the 2G/2G split
> patches, or the 4G/4G patches, either one of which should increase your
> kernel address space enough to map both memory and your buffers.
>
> Daniel

> out four camera systems with only 256MB of memory which is only
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

You can use as much memory as you can afford if the designer of
this system would only understand that you can't fit 4 GB of
application RAM into 4 GB of address-space and leave anything
available for the system.

It needs to be done like this:

------------|-------------|-----------
Other | N Mbyte | Other address space
address | window | ..................
------------|-------------|-----------
|
|
-----------------
| Page register |
|----------------
|
-----------------------
| Page 0 |
|--|--------------------
| Page 1 |
|---------------------
Page N, etc.

You make a small window in the available address-space, and you
map your RAM into that address-space with a page register. You
can access much, much more RAM than available address-space,
and with things like video and pictures, the occasional switching
of a page-register adds practically no overhead.


Cheers,
Dick Johnson
Penguin : Linux version 2.4.26 on an i686 machine (5570.56 BogoMips).
Note 96.31% of all statistics are fiction.


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/