Andi Kleen wrote: > > Can you explain the rationale behind that running on the BKL? What type > > of > > It used to always run with the BKL because everything used to > and originally nobody wanted to review all ioctl handlers in tree to see if > they can run with more fine grained locking. A lot probably can though. > > > things needs to be protected so that this huge hammer is needed? What > > would be an earlier point to release the BKL? > > That depends on the driver. A lot don't need BKL at all and > in others it can be easily eliminated. But it needs case-by-case > review of the locking situation. > > The goal of the proposal here is just to make it more visible. So if I write my own driver and have never heard of ioctls running on BKL before I can rather be sure that I just can change the interface of the ioctl function, put it in unlocked_ioctl and are fine? Cool. Eike
Attachment:
signature.asc
Description: This is a digitally signed message part.