Network drivers with direct io memory accesses

Network drivers with direct io memory accesses

[Greg KH]

Hi all,

I'm currently sitting on two network drivers:
 Alacritech gigabit driver that supports 3 different classes of cards:
  http://www.kernel.org/pub/linux/kernel/people/gregkh/gregkh-2.6/patches/ldp/net-add-alacritech-slicoss-driver.patch

 ET1310 network driver
  http://www.kernel.org/pub/linux/kernel/people/gregkh/gregkh-2.6/patches/ldp/net-add-et131x-driver.patch

Both of these drivers are accessing memory on the devices directly,
without using the proper kernel abstractions for it, making them usable
only on x86-based boxes probably.

My question is, does this have to be fixed in order to get them
accepted into the tree?  It's going to be a very big effort to fix them
up from what I can tell.  I don't mind, it's just going to take a while,
and I'm wondering if people want to use these drivers sooner than the
month or so it's going to take me (at the least.)

Yes, I know the Alacritech driver still has the firmware in the patch
itself, that will be moved out, I know, and there are probably other
minor issues left to clean up in each of these drivers, that's not the
point here...

Any thoughts?

thanks,

greg k-h

Re: Network drivers with direct io memory accesses

[David Miller]

From: Greg KH <greg@kroah.com>
Date: Wed, 14 May 2008 17:10:44 -0700

> Both of these drivers are accessing memory on the devices directly,
> without using the proper kernel abstractions for it, making them usable
> only on x86-based boxes probably.
> 
> My question is, does this have to be fixed in order to get them
> accepted into the tree?

I'd say yes.

If that aspect isn't fixed I can't even build test them on my
computers.

Re: Network drivers with direct io memory accesses

[Greg KH]

On Wed, May 14, 2008 at 05:13:15PM -0700, David Miller wrote:
> From: Greg KH <greg@kroah.com>
> Date: Wed, 14 May 2008 17:10:44 -0700
> 
> > Both of these drivers are accessing memory on the devices directly,
> > without using the proper kernel abstractions for it, making them usable
> > only on x86-based boxes probably.
> > 
> > My question is, does this have to be fixed in order to get them
> > accepted into the tree?
> 
> I'd say yes.

{sigh}

I figured that, but thought it would be worth it to at least ask. :)

> If that aspect isn't fixed I can't even build test them on my
> computers.

Hm, why?  They should still build I would think, just not run properly,
right?

Damm, I really gotta get that cross-build system set up here, I'll go
start that right now...

thanks,

greg k-h

Re: Network drivers with direct io memory accesses

[Greg KH]

On Wed, May 14, 2008 at 05:28:24PM -0700, Greg KH wrote:
> On Wed, May 14, 2008 at 05:13:15PM -0700, David Miller wrote:
> > From: Greg KH <greg@kroah.com>
> > Date: Wed, 14 May 2008 17:10:44 -0700
> > 
> > > Both of these drivers are accessing memory on the devices directly,
> > > without using the proper kernel abstractions for it, making them usable
> > > only on x86-based boxes probably.
> > > 
> > > My question is, does this have to be fixed in order to get them
> > > accepted into the tree?
> > 
> > I'd say yes.
> 
> {sigh}
> 
> I figured that, but thought it would be worth it to at least ask. :)
> 
> > If that aspect isn't fixed I can't even build test them on my
> > computers.
> 
> Hm, why?  They should still build I would think, just not run properly,
> right?
> 
> Damm, I really gotta get that cross-build system set up here, I'll go
> start that right now...

Ok, I did this, and I don't see any build errors or warnings on sparc64.
Maybe the code isn't as bad as I imagined, I'll go through it all and
submit it for review.

thanks,

greg k-h