lkml.org 
[lkml]   [1997]   [Aug]   [8]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: Kernel virtual memory?
On Fri, 8 Aug 1997, Rogier Wolff wrote:

> Benjamin C R LaHaise wrote:
> > shuffling pages around in memory. Any driver that needs to regularly
> > allocate dmable pages should really think about doing that at startup,
> > and just holding onto these flip buffers for its lifetime.
>
> Any driver using DMA for performance reasons is not going to benefit
> from a flip buffer that the CPU needs to copy to provide for userspace
> page fragmentation.

We're only talking about hardware which is generally considered
low-performance and obsolete anyways - aren't most new products PCI? And
most non-intel platforms can DMA the entire address space anyways (again,
except for old, mostly obsolete hardware). But we still need to support
it. Remember users don't like it when their kernel spits out an obscure
message like 'unable to get DMA buffer' - all they know is that they
can't hear the sounds of evil aliens exploding into a million pieces!

> Drivers should in the near future be able to DMA directly into
> userspace memory. We need to lock down the pages involved, and be able
> to send a "list of pages" for the driver to use.....

This is something that I'm keeping in mind while hacking away. There
are a number of other occasions where we need to lock a page to prevent
changes from occuring to it, most notably with shared mapings. Hopefully
what I come up with will be useful.

-ben


\
 
 \ /
  Last update: 2005-03-22 13:40    [W:0.749 / U:0.408 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site