Re: snapper does not work on PowerBook5,6

[Michael <macallan%netbsd.org@localhost>]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hello,

On Oct 22, 2010, at 1:13 PM, Frank Wille wrote:

> Also this driver rarely checks the return values of functions calls.  
> This
> was the main reason we were looking for my audio problem for such a  
> long
> time (read below) ...

There was a time when bus_space_map() on macppc couldn't fail since it  
neither kept track of mappings nor actually mapped anything - PCI  
space is BAT-mapped to bus address == physical address.

> > At least that verifies that the interrupt is enabled and working.
> > Leaves the question why they don't fire on your hardware - I need to
> > stare at the code for a bit, maybe it's something trivial like having
> > to explicitly clear them before use or something.
>
> I found it!
>
> After compiling a kernel with DEBUG I noticed that bus_space_map()  
> failed,
> when mapping the DMA channel registers (one of those return values,  
> which
> should have been checked ;).
>
> The reason that it fails can be found in the OF tree for i2s:
>
> ---8<---
> ff9859b0: /pci@f2000000/mac-io@17/i2s@0
>
> #address-cells          00000001 ........ ........ ........     1
> AAPL,address            80010000 ........ ........ ........   ....
> built-in
> device_type             69327300 ........ ........ ........   "i2s"
> name                    69327300 ........ ........ ........   "i2s"
> ranges
> reg                     00000000 00005000 00000000 00000a00
> ......P.........
>
> --------------------------------------------------------------------------------
>
>
> ff985ba8: /pci@f2000000/mac-io@17/i2s@0/i2s-a@10000
>
> built-in
> compatible              69327362 757300.. ........ ........   "i2sbus"
> device_type             736f756e 64627573 00...... ........    
> "soundbus"
> interrupt-parent        ff981a58 ........ ........ ........   ...X
> interrupts              0000001e 00000001 00000001 00000000
> ................
>            0010:       00000002 00000000 ........ ........   ........
> name                    6932732d 6100.... ........ ........   "i2s-a"
> [...]
> reg                     00010000 00001000 00008000 00000100
> ................
>            0010:       00008100 00000100 ........ ........   ........
>
> ---8<---
>
> The driver is using the "reg" property of "i2s@0", but it should use  
> the
> child's "i2s-a@10000" reg.

OUCH. Thank you Apple for screwing us over once more.

> I have found an OF dump from a MacMini in the archives, and there  
> both "reg"
> properties were identical. So a solution could be to just use the
> reg-property of the child (soundbus) for all models?
>
> Can somebody confirm that this is ok? Or would it break any other  
> model?

On my iBook G4 the properties are not identical but i2s merely  
contains two more ranges than i2s-a, looks like two more sets of DBDMA  
registers.
To be safe we should probably check the reg property on i2s for bogus  
and use i2s-a if it is - I bet there are models out there where always  
using i2s-a breaks stuff.

> BTW, when using the correct property, sound works fine on my  
> PowerBook. I
> get interrupts and everything... :)

Yay :)

have fun
Michael

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.7 (Darwin)

iQEVAwUBTMIHgMpnzkX8Yg2nAQLTeAgAiBGIf6VDZEcBxZJR+naqeKkWuw5oySx5
7wVfgzBPX9V43d1+Dj/sKoL4q9+RDpq6j+iyyLTCXc2I4DcZD4HkSJWw4XepwzfF
oqifrkFopXkOJb4T476y6/fqk02YM7dd2b/cAwI+s+AB+KTuY/MqIGCJxGETTw+K
JKYU88xTeGj6Ww8FbMLwmcGWl6eAzCRzZLIw38dG/t0JO30Ef3ETt3FDyDk88ewx
R6Z6FHcGBgojmwi5Pf+XosZ56sB6A0Dfs786q2Wg4qJZytQl5Znp3KgwOSLxAm5H
8b/kejDGQUND/46ZTx0K/Fa06lJMM5fCLDuHDYGpktIjrH4BMuQZ7A==
=Karb
-----END PGP SIGNATURE-----