[PATCH 2.6.11-rc2 0/3] altix: AGP bridge support for tioca chip

[PATCH 2.6.11-rc2 0/3] altix: AGP bridge support for tioca chip

[Mark Maule]

This patchset provides new hardware support for the SGI TIOCA agp chip.

In order to stitch this into the SGI sn2 pci dma layer, it is necessary
to add an abstraction which is implmented by PATCH 1/3 
(sn2-pci-dma-abstraction.patch).

Additionally, a couple of header files were moved out of arch/ia64/sn
and into include/asm-ia64/sn.  I also took the opportunity to kill some
unused macros.  This is done by PATCH 2/3 (sn2-move-pci-headers.patch).

The TIOCA driver is implemented as an sn2 pci provider with hooks 
externalized for an upcoming agpgart backend.  The TIOCA kernel driver is
provided with PATCH 3/3 (sn2-tioca-driver.patch).

Signed-off-by: Mark Maule <maule@sgi.com>

thanks
Mark

Re: [PATCH 2.6.11-rc2 0/3] altix: AGP bridge support for tioca chip

[Christoph Hellwig]

On Fri, Feb 04, 2005 at 03:31:58PM -0600, Mark Maule wrote:
> This patchset provides new hardware support for the SGI TIOCA agp chip.
> 
> In order to stitch this into the SGI sn2 pci dma layer, it is necessary
> to add an abstraction which is implmented by PATCH 1/3 
> (sn2-pci-dma-abstraction.patch).
> 
> Additionally, a couple of header files were moved out of arch/ia64/sn
> and into include/asm-ia64/sn.  I also took the opportunity to kill some
> unused macros.  This is done by PATCH 2/3 (sn2-move-pci-headers.patch).
> 
> The TIOCA driver is implemented as an sn2 pci provider with hooks 
> externalized for an upcoming agpgart backend.  The TIOCA kernel driver is
> provided with PATCH 3/3 (sn2-tioca-driver.patch).

Wouldn't it be better to allow pluggable IOMMUs at the ia64 architecture
level, replacing the machvecs?

zx1 already has a switch to use swiotl for devices with too small dma
masks (and altix would need that one aswell if it tried to support such
devices).

Re: [PATCH 2.6.11-rc2 0/3] altix: AGP bridge support for tioca chip

[Jesse Barnes]

On Monday, February 7, 2005 2:07 am, Christoph Hellwig wrote:
> Wouldn't it be better to allow pluggable IOMMUs at the ia64 architecture
> level, replacing the machvecs?
>
> zx1 already has a switch to use swiotl for devices with too small dma
> masks (and altix would need that one aswell if it tried to support such
> devices).

Something like that might be nice as part of the bridge driver work that's 
being discussed on lkml.  Machine vectors as they stand right now aren't a 
good fit since they're only initialized once for the platform at boot time, 
not on a per-chip basis.

Jesse

[PATCH 2.6.11-rc2 0/3] altix: AGP bridge support for tioca chip

[Mark Maule]

Resend of "altix: tioca chip driver (agp)" originally sent on 2005-02-04:

This patchset provides new hardware support for the SGI TIOCA agp chip.

In order to stitch this into the SGI sn2 pci dma layer, it is necessary
to add an abstraction which is implmented by PATCH 1/3 
(sn2-pci-dma-abstraction.patch).

Additionally, a couple of header files were moved out of arch/ia64/sn
and into include/asm-ia64/sn.  I also took the opportunity to kill some
unused macros.  This is done by PATCH 2/3 (sn2-move-pci-headers.patch).

The TIOCA driver is implemented as an sn2 pci provider with hooks 
externalized for an upcoming agpgart backend.  The TIOCA kernel driver is
provided with PATCH 3/3 (sn2-tioca-driver.patch).

Changes from first review:

sn2-pci-dma-abstraction.patch
	Use a local provider variable to improve readability

sn2-move-pci-headers.patch
	No changes

sn2-tioca-driver.patch
	Misc. style-related cleanups
	Remove unneeded casts
	Fix error-case memory leak
	Use list_for_each_entry() instead of list_for_each()
	Use list routines for dmamap management
	Move prom version checking to tioca_bus_fixup()

	Note:  I elected to leave the EXPORT_SYMBOL's alone as they
	are being used by our agpgart backend stub currently in
	development.  If we end up merging everything together as it looks
	like alpha does, then I'll yank 'em.  At a minimum, we should be
	able to remove tioca_gart_found and just count the elements
	in tioca_list.

thanks
Mark

Re: [PATCH 2.6.11-rc2 0/3] altix: AGP bridge support for tioca chip

[Jesse Barnes]

On Monday, February 7, 2005 4:08 pm, Mark Maule wrote:
> Resend of "altix: tioca chip driver (agp)" originally sent on 2005-02-04:
>
> This patchset provides new hardware support for the SGI TIOCA agp chip.

Tony, although this is probably safe for 2.6.11, it's probably best to queue 
it up for afterwards unless Linus already told you that we won't see 
2.6.11-final for awhile yet.

Thanks,
Jesse

RE: [PATCH 2.6.11-rc2 0/3] altix: AGP bridge support for tioca chip

[Luck, Tony]

>Tony, although this is probably safe for 2.6.11, it's probably 
>best to queue it up for afterwards unless Linus already told you
>that we won't see 2.6.11-final for awhile yet.

I think I'll queue it for later.

-Tony

[PATCH 2.6.11-rc2 0/3] altix: AGP bridge support for tioca chip

[Mark Maule]

Second resend of "altix: tioca chip driver (agp)" originally sent
on 2005-02-04.  First resend was missing a Signed-off-by line in the 
sn2-pci-dma-abstraction.patch.

This patchset provides new hardware support for the SGI TIOCA agp chip.

In order to stitch this into the SGI sn2 pci dma layer, it is necessary
to add an abstraction which is implmented by PATCH 1/3 
(sn2-pci-dma-abstraction.patch).

Additionally, a couple of header files were moved out of arch/ia64/sn
and into include/asm-ia64/sn.  I also took the opportunity to kill some
unused macros.  This is done by PATCH 2/3 (sn2-move-pci-headers.patch).

The TIOCA driver is implemented as an sn2 pci provider with hooks 
externalized for an upcoming agpgart backend.  The TIOCA kernel driver is
provided with PATCH 3/3 (sn2-tioca-driver.patch).

Changes from first review:

sn2-pci-dma-abstraction.patch
	Use a local provider variable to improve readability

sn2-move-pci-headers.patch
	No changes

sn2-tioca-driver.patch
	Misc. style-related cleanups
	Remove unneeded casts
	Fix error-case memory leak
	Use list_for_each_entry() instead of list_for_each()
	Use list routines for dmamap management
	Move prom version checking to tioca_bus_fixup()

	Note:  I elected to leave the EXPORT_SYMBOL's alone as they
	are being used by our agpgart backend stub currently in
	development.  If we end up merging everything together as it looks
	like alpha does, then I'll yank 'em.  At a minimum, we should be
	able to remove tioca_gart_found and just count the elements
	in tioca_list.

thanks
Mark