405 DMA api

405 DMA api

[Armin Kuster]

[-- Attachment #1: Type: text/plain, Size: 787 bytes --]

Hello,

I would like to have the 405 dma share the same dma api as the current
pmac. I propose to make the asm/dma.h _generic_ and move all the power
Mac unique stuff into a seperate file such as "pmac_dma.h".  The dma.h
would then be the gateway for other ppc archs by using "CONFIG_<arch>"
to determine which hearder to pull in.  I made the pmac_pma.h as the
default.  I have tested this on the  walnut and can only compile for the
pmac.  I have included a patch and I  would like some feedback and to
have the pmac_dma verified.

files:
include/asm-ppc/dma.h <- generic defines and config selection
include/asm-ppc/pmac_dma.h     <- most code from dma.h
include/asm-ppc/ppc405_dma.h <- changed func names
arch/ppc/kernel/ppc405_dma.c <- changes to refect new header scheme

TIA
Armin

[-- Attachment #2: pmac_dma.patch.gz --]
[-- Type: application/x-gunzip, Size: 6822 bytes --]

Re: 405 DMA api

[Dan Malek]

Armin Kuster wrote:

> I would like to have the 405 dma share the same dma api as the current
> pmac.

Not a bad idea, since most DMA controllers are learning and becoming
descriptor based.  Please consider that DMA controllers, the devices
to which they are connected, and the integration is going to be unique
to the parts, boards and devices.  Forcing this to be too generic will
likely fail, make a mess of the software, and not benefit anyone.

Have fun!

	-- Dan


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: 405 DMA api

[Benjamin Herrenschmidt]

>Hello,
>
>I would like to have the 405 dma share the same dma api as the current
>pmac. I propose to make the asm/dma.h _generic_ and move all the power
>Mac unique stuff into a seperate file such as "pmac_dma.h".  The dma.h
>would then be the gateway for other ppc archs by using "CONFIG_<arch>"
>to determine which hearder to pull in.  I made the pmac_pma.h as the
>default.  I have tested this on the  walnut and can only compile for the
>pmac.  I have included a patch and I  would like some feedback and to
>have the pmac_dma verified.
>
>files:
>include/asm-ppc/dma.h <- generic defines and config selection
>include/asm-ppc/pmac_dma.h     <- most code from dma.h
>include/asm-ppc/ppc405_dma.h <- changed func names
>arch/ppc/kernel/ppc405_dma.c <- changes to refect new header scheme

Well... except this isn't pmac dma. The pmac has a DBDMA controller
which is much different from the legacy DMA controller used here. What
you see in this dma.h file is, I bet, prep stuff.

Ben.


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: 405 DMA api

[Gabriel Paubert]

On Mon, 25 Feb 2002, Benjamin Herrenschmidt wrote:

>
> >Hello,
> >
> >I would like to have the 405 dma share the same dma api as the current
> >pmac. I propose to make the asm/dma.h _generic_ and move all the power
> >Mac unique stuff into a seperate file such as "pmac_dma.h".  The dma.h
> >would then be the gateway for other ppc archs by using "CONFIG_<arch>"
> >to determine which hearder to pull in.  I made the pmac_pma.h as the
> >default.  I have tested this on the  walnut and can only compile for the
> >pmac.  I have included a patch and I  would like some feedback and to
> >have the pmac_dma verified.
> >
> >files:
> >include/asm-ppc/dma.h <- generic defines and config selection
> >include/asm-ppc/pmac_dma.h     <- most code from dma.h
> >include/asm-ppc/ppc405_dma.h <- changed func names
> >arch/ppc/kernel/ppc405_dma.c <- changes to refect new header scheme
>
> Well... except this isn't pmac dma. The pmac has a DBDMA controller
> which is much different from the legacy DMA controller used here. What
> you see in this dma.h file is, I bet, prep stuff.

True. It´s the legacy ISA DMA controller which can be found on PreP and
CHRP, almost always in the PCI-ISA bridge. However these are not all
created equal: the one in the WinBond 83C553 in my MVME boards, which is
almost identical to some Intel model (82378ZB), is actually capable of 32
bit addressing and can do descripto based scatter-gather DMA. However,
since I never used anything that needed DMA on these boards (the floppy
actually) I did not bother to improve the dma driver.

Some historical note: the 82378ZB, with its improved DMA engine, is AFAICS
the only Intel PCI<->ISA bridge with enhanced DMA capabilities. Also AFAIR
this chip was exactly the one specified in the PreP documentation,
strangely enough, the 82378 production was stopped and the enhanced DMA
capability disappeared from all later Intel PCI<->ISA bridge. My 82378ZB
doc is from March 1996 BTW...

	Regards,
	Gabriel.


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: 405 DMA api

[Armin Kuster]

Gabriel Paubert wrote:
>
> On Mon, 25 Feb 2002, Benjamin Herrenschmidt wrote:
>
> >
> > >Hello,
> > >
> > >I would like to have the 405 dma share the same dma api as the current
> > >pmac. I propose to make the asm/dma.h _generic_ and move all the power
> > >Mac unique stuff into a seperate file such as "pmac_dma.h".  The dma.h
> > >would then be the gateway for other ppc archs by using "CONFIG_<arch>"
> > >to determine which hearder to pull in.  I made the pmac_pma.h as the
> > >default.  I have tested this on the  walnut and can only compile for the
> > >pmac.  I have included a patch and I  would like some feedback and to
> > >have the pmac_dma verified.
> > >
> > >files:
> > >include/asm-ppc/dma.h <- generic defines and config selection
> > >include/asm-ppc/pmac_dma.h     <- most code from dma.h
> > >include/asm-ppc/ppc405_dma.h <- changed func names
> > >arch/ppc/kernel/ppc405_dma.c <- changes to refect new header scheme
> >
> > Well... except this isn't pmac dma. The pmac has a DBDMA controller
> > which is much different from the legacy DMA controller used here. What
> > you see in this dma.h file is, I bet, prep stuff.
>
> True. It´s the legacy ISA DMA controller which can be found on PreP and
> CHRP, almost always in the PCI-ISA bridge. However these are not all
> created equal: the one in the WinBond 83C553 in my MVME boards, which is
> almost identical to some Intel model (82378ZB), is actually capable of 32
> bit addressing and can do descripto based scatter-gather DMA. However,
> since I never used anything that needed DMA on these boards (the floppy
> actually) I did not bother to improve the dma driver.
>
> Some historical note: the 82378ZB, with its improved DMA engine, is AFAICS
> the only Intel PCI<->ISA bridge with enhanced DMA capabilities. Also AFAIR
> this chip was exactly the one specified in the PreP documentation,
> strangely enough, the 82378 production was stopped and the enhanced DMA
> capability disappeared from all later Intel PCI<->ISA bridge. My 82378ZB
> doc is from March 1996 BTW...
>
>         Regards,
>         Gabriel.


Ok ... thanks for the info.  where do we go from here?  Can it get push
in after a better name is selected?

-- armin

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/