question about dma-ranges

question about dma-ranges

[Yoder Stuart-B08248]

Mitch,

In the dma-ranges proposal:

http://playground.sun.com/1275/proposals/Closed/Accepted/410-it.txt

...there is the following special case described:

   As a special case, a "dma-ranges" property may be present in the
   root node of the device tree.  In this case, the property value
   describes the ranges of DMA addresses that the system bus can
   accept.  In this case, the length of the parent-address portion
   of each entry in the property value is 0 (because the root node
   has no parent), so the format reduces to...

A few of us in the ePAPR committee were discussing this and
were wondering what the use case for a dma-ranges on the
root is.   Is this special case really needed?

Thanks,
Stuart Yoder

Re: question about dma-ranges

[Mitch Bradley]

  It's probably unnecessary on modern machines, but old PCs were fairly restrictive about DMA addresses due to short 
counters.  The buses on which such restrictions applied are no longer at the root level, but they were once there...

On 10/27/2010 8:48 AM, Yoder Stuart-B08248 wrote:
> Mitch,
>
> In the dma-ranges proposal:
>
> http://playground.sun.com/1275/proposals/Closed/Accepted/410-it.txt
>
> ...there is the following special case described:
>
>     As a special case, a "dma-ranges" property may be present in the
>     root node of the device tree.  In this case, the property value
>     describes the ranges of DMA addresses that the system bus can
>     accept.  In this case, the length of the parent-address portion
>     of each entry in the property value is 0 (because the root node
>     has no parent), so the format reduces to...
>
> A few of us in the ePAPR committee were discussing this and
> were wondering what the use case for a dma-ranges on the
> root is.   Is this special case really needed?
>
> Thanks,
> Stuart Yoder
>
>
>

Re: question about dma-ranges

[Timur Tabi]

On Tue, Oct 26, 2010 at 7:51 PM, Mitch Bradley <wmb-D5eQfiDGL7eakBO8gow8eQ@public.gmane.org> wrote:
>  It's probably unnecessary on modern machines, but old PCs were fairly
> restrictive about DMA addresses due to short counters.  The buses on which
> such restrictions applied are no longer at the root level, but they were
> once there...

It's still necessary.  The QE, which we ship on several of our current
parts, can only DMA to/from 32-bit addresses, even on SOCs that
support 36-bit addressing for everything else.

-- 
Timur Tabi
Linux kernel developer at Freescale

Re: question about dma-ranges

[David Gibson]

On Tue, Oct 26, 2010 at 08:37:55PM -0500, Timur Tabi wrote:
> On Tue, Oct 26, 2010 at 7:51 PM, Mitch Bradley <wmb-D5eQfiDGL7eakBO8gow8eQ@public.gmane.org> wrote:
> >  It's probably unnecessary on modern machines, but old PCs were fairly
> > restrictive about DMA addresses due to short counters.  The buses on which
> > such restrictions applied are no longer at the root level, but they were
> > once there...
> 
> It's still necessary.  The QE, which we ship on several of our current
> parts, can only DMA to/from 32-bit addresses, even on SOCs that
> support 36-bit addressing for everything else.

But the QE is not at the top-level, IIRC, so its restrictions can be
encoded in the dma-ranges on its own bus.  We're talking specifically
about the special case of dma-ranges in the root node, not the utility
of dma-ranges in general which is clear.

-- 
David Gibson			| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you.  NOT _the_ _other_
				| _way_ _around_!
http://www.ozlabs.org/~dgibson

Re: question about dma-ranges

[Scott Wood]

On Wed, 27 Oct 2010 13:42:27 +1100
David Gibson <david@gibson.dropbear.id.au> wrote:

> On Tue, Oct 26, 2010 at 08:37:55PM -0500, Timur Tabi wrote:
> > On Tue, Oct 26, 2010 at 7:51 PM, Mitch Bradley <wmb@firmworks.com> wrote:
> > >  It's probably unnecessary on modern machines, but old PCs were fairly
> > > restrictive about DMA addresses due to short counters.  The buses on which
> > > such restrictions applied are no longer at the root level, but they were
> > > once there...
> > 
> > It's still necessary.  The QE, which we ship on several of our current
> > parts, can only DMA to/from 32-bit addresses, even on SOCs that
> > support 36-bit addressing for everything else.
> 
> But the QE is not at the top-level, IIRC, so its restrictions can be
> encoded in the dma-ranges on its own bus.  We're talking specifically
> about the special case of dma-ranges in the root node, not the utility
> of dma-ranges in general which is clear.

Plus, in this case does that need to be expressed in the device tree?

Or can the QE code just know about that because it only has 32-bit
registers/descriptor fields for DMA addresses?  I.e. it is a limitation
of all instances of QE, not just as integrated in this system.

-Scott

_______________________________________________
devicetree-discuss mailing list
devicetree-discuss@lists.ozlabs.org
https://lists.ozlabs.org/listinfo/devicetree-discuss

Re: question about dma-ranges

[Timur Tabi]

Scott Wood wrote:
> Or can the QE code just know about that because it only has 32-bit
> registers/descriptor fields for DMA addresses?  I.e. it is a limitation
> of all instances of QE, not just as integrated in this system.

It's a limitation of the QE registers -- they're only 32-bit.  If we ever
produce a new QE that has bigger registers, we can use the compatible property
to distinguish between them.

-- 
Timur Tabi
Linux kernel developer at Freescale