On Wed, Aug 13, 2008 at 11:12:48AM -0700, Eric W. Biederman wrote: > Linus Torvalds <torvalds at linux-foundation.org> writes: > > > On Wed, 13 Aug 2008, Huang Ying wrote: > >> > >> - xchg(&kexec_lock, 0); > >> + locked = xchg(&kexec_lock, 0); > >> + BUG_ON(!locked); > > > > Why do you want to do this at all? > > > > And why do you implement your locks with xchg() in the first place? That's > > total and utter crap. > > > > Hint: we have _real_ locking primitives in the kernel. > > This part certainly. > > The way the code should work, and the way it has in the past is: > image = xchg(&kexec_image, NULL) > if (!image) > return -EINVAL; > > Very simple and very obvious and very easy to get right, and it has > been that way for years. > Hi Eric, Are there any issues with usage of test_and_set_bit() or usage of spinlock primitives? Thanks Vivek
