nap/dfs on 7448

nap/dfs on 7448

[Leisner, Martin]

I asked this on linuxppc-embedded a week ago (I didn't even know this
list existed until last week -- another reason to get rid of
linuxpcc-embedded).

Has anyone gotten NAP/DFS to reliably work on a 7448?

I'm seeing strange problems with peripherals...(using a ram disk
works fine).

marty

Re: nap/dfs on 7448

[Benjamin Herrenschmidt]

On Mon, 2007-10-01 at 17:32 -0400, Leisner, Martin wrote:
> I asked this on linuxppc-embedded a week ago (I didn't even know this
> list existed until last week -- another reason to get rid of
> linuxpcc-embedded).
> 
> Has anyone gotten NAP/DFS to reliably work on a 7448?
> 
> I'm seeing strange problems with peripherals...(using a ram disk
> works fine).

Could it be that your host bridge isn't properly waking up the CPU to
DOZE state for snooping DMA ? (It might require some delays on QACK in
some cases, I know Apple had workarounds in those areas, maybe something
along those lines need to be configured in the chipset).

Ben.

update: RE: nap/dfs on 7448

[Leisner, Martin]

Turns out the Tundra TSI109 has various "problems" trying
to nap -- its not a 7448 issue...

We're working with tundra to get a workaround (in software, they=20
won't respin the part).  It seems DMA while napping is an issue, we
have to turn off the tsi109 ethernet queues before entering nap mode...

If anyone needs this work, let me know (it will  be covered under GPL).

marty

>   -----Original Message-----
>   From: Benjamin Herrenschmidt [mailto:benh@kernel.crashing.org]
>   Sent: Tuesday, October 02, 2007 2:02 AM
>   To: Leisner, Martin
>   Cc: linuxppc-dev
>   Subject: Re: nap/dfs on 7448
>  =20
>  =20
>   On Mon, 2007-10-01 at 17:32 -0400, Leisner, Martin wrote:
>   > I asked this on linuxppc-embedded a week ago (I didn't even know
this
>   > list existed until last week -- another reason to get rid of
>   > linuxpcc-embedded).
>   >
>   > Has anyone gotten NAP/DFS to reliably work on a 7448?
>   >
>   > I'm seeing strange problems with peripherals...(using a ram disk
>   > works fine).
>  =20
>   Could it be that your host bridge isn't properly waking up the CPU
to
>   DOZE state for snooping DMA ? (It might require some delays on QACK
in
>   some cases, I know Apple had workarounds in those areas, maybe
something
>   along those lines need to be configured in the chipset).
>  =20
>   Ben.
>  =20

Re: update: RE: nap/dfs on 7448

[Benjamin Herrenschmidt]

On Thu, 2007-12-27 at 17:37 -0500, Leisner, Martin wrote:
> Turns out the Tundra TSI109 has various "problems" trying
> to nap -- its not a 7448 issue...
> 
> We're working with tundra to get a workaround (in software, they 
> won't respin the part).  It seems DMA while napping is an issue, we
> have to turn off the tsi109 ethernet queues before entering nap mode...
> 
> If anyone needs this work, let me know (it will  be covered under GPL).

This looks like yet another f*cked up bridge ... I wonder if we'll ever
get a half decent working northbridge for those processors...

Unfortunately, it looks like the market for those is limited to junk
from vendor A and shit from vendor B...

Is the problem limited to the TSI internal ethernet or it happens with
any DMA ?

Looks to me like turning off DMA in NAP mode basically makes NAP mode
unuseable, perdiod. You can't start loosing packets because your
processor is idle. So no power saving, and let's eat the watts since
they won't respin the part. Great !

Ben.