re: Nouveau driver on nvidia 8500GT (amd64)

[matthew green <mrg%eterna.com.au@localhost>]

Taylor R Campbell writes:
>    Date: Mon, 26 Oct 2015 14:35:29 +0000 (UTC)
>    From: Mariusz Wawrzyniak <rowerek%gmail.com@localhost>
> 
>    Today I built the kernel from the latest sources with nouveau, LOCKDEBUG,
>    DEBUG and KMEM_GUARD_DEPTH=30000 enabled. It gave me no console
>    and the relevant fragment of /var/log/messages looks like this:
> 
> FYI, nouveau support is totally unfinished and experimental at the
> moment.  I think mrg@ has gotten it to boot and control the display on
> one device, with wscons text drawn to the nvidia device's framebuffer
> by software -- but that's it so far.

actually -- lately it's now crashing when trying to load the
firmware.  seems that the pci_dev that ends up being passed
to request_firmware() is being intepreted as a (linux-style)
struct device, and it isn't set, and we deref NULL and crash.

this didn't used to happen, and i mostly suspect the changes
Jared made to support non-pci, but i don't know how this was
supposed to work previously anyway...  this code from
request_firmware():

     49 static inline int
     50 request_firmware(const struct firmware **fwp, const char *image_name,
     51     struct device *dev)
[ ... ]
     65         drvname = device_cfdriver(dev)->cd_name;

fails because dev is NULL.  i suspect there is confusion
between linux and netbsd 'struct device'.

i haven't yet really looked closer at what changed to make this
break and why it seems so confusing/wrong.


.mrg.