[systemd-devel] udev 182: response timeout for request_firmware in module_probe path

[Benjamin Herrenschmidt]

On Thu, 2012-08-23 at 17:46 +0100, Alan Cox wrote:
> > IMO, the driver probing path is allowed to sleep, so looks request
> firmware
> > should be allowed inside .probe().
> 
> I'm not convinced about that. It can sleep but its holding various
> locks
> in most cases, and it looks like that can end up in a complete heap.
> 
> By all means *request* the firmware asynchronously in the probe, but
> there needs to be a seperate method somewhere after the probe to
> finish
> the job once the firmware appears. 

Not necessarily enough in the general case, for example some stacks will
cause a driver open to be call back from somewhere within the
register_foo() the driver did to register itself with it's subsystem
inside probe. For example register_framebuffer(), register_netdev(), ...

This is clearly a udev bug :-)

Cheers,
Ben.