Bug#286726: Firmware theory

[Jouni Malinen]

On Mon, Jan 10, 2005 at 11:48:01AM -0500, Andres Salomon wrote:

> Regarding http://hostap.epitest.fi/bugz/show_bug.cgi?id=6

(which, btw, is resolved with WONTFIX)

> On Fri, 2005-01-07 at 11:59 -0500, Ian Gulliver wrote:
> > This card has internal flash and doesn't need firmware uploads on
> > insert; I don't believe that's the issue.
> > 
> 
> Jouni, what are your thoughts?

So far, no one has convinced me that anything needs to be changed in
the driver, so my latest comments in
http://hostap.epitest.fi/bugz/show_bug.cgi?id=6 are still valid:

This looks like a race condition between hotplug script and driver
initialization. The script finds a new interface before the driver has had
enough time to set dev->dev_addr. I was able to reproduce this for wifi0.
However, it did not and should not happen for wlan0 at least with the current
CVS development or stable branch. I tested both and wlan0 had MAC address when
hotplug script was called.

Based on the driver source code, it looks like wifi0 gets register_netdev()
called with all zeroes dev->dev_addr, but wlan0 has dev_addr set when
register_netdev() is called. However, this is not guaranteed for some cases,
e.g., when the card does not have full firmware image in the flash and requires
prism2_srec to be called. For these cases, the workaround would be to delay
ifrename to happen only after the firmware has been successfully downloaded.

-- 
Jouni Malinen                                            PGP id EFC895FA