On Mon, 2009-03-30 at 11:57 +0200, Johannes Berg wrote: > > > One thing I was wondering about -- should we just implement polling via > > > a single work struct that iterates all those rfkill drivers that need > > > polling? I've given them each an own right now, but that seems > > > unnecessary. > > > > If it doesn't add new locking problems (global locks, etc...) that'd be OK. > > Hm, indeed, it probably would -- rfkill_register needs a global lock > while being called from the driver, and the polling would need the same > lock while calling into the driver. We can work around that by cancelling the work whenever we want to modify the list -- that way the work doesn't need to take any locks. Not sure it's worth it though, that makes the code harder to follow. johannes
Attachment:
signature.asc
Description: This is a digitally signed message part
