Michael Trimarchi wrote: > I see the other patches, > I'm not sure if you just reduce the window or your patch is ok, I must > take a look > to the net part to undestand if it is correct.
The patch should be good. What it does is that it moves all the low-level shutdown into ar6000_close which is called from unregister_netdev. This means that it happens while the rtnl_lock is held. The ioctl path also gets rtnl_lock, so the two can't race anymore. There's another problem if you hammer the system with rfkill. Eventually, the SD/MMC stack gets upset and fails to bring up the device. But that happens regardless of this patch. In my torture testing I found that there's also a slow memory leak on bind/unbind that wants looking at. I've applied it now. Sorry for not realizing earlier where the right place for fixing this was, and making you spend time on cleaning up your solution. - Werner
