On Mon, 2014-06-23 at 14:59 +0200, Michal Kazior wrote:

quoted

quoted

But does this bump up the module refcount in such a way that an
in-progress rmmod will wait/block until the refcount reaches 0?

It should, yes, but you race once you do module_put(), no?

It's still racy, I agree.

I don't see how device_release_driver() can be made work synchronously
at all. It should, however, be perfectly fine if this was an async
request to release a device/driver, right?

Yeah, but the work struct would have to live somewhere that can't be
unloaded, or can more safely wait for it...

I guess we could also have a list of these in the modules and
synchronize on them in module_exit()?

johannes