Okay I stand corrected the Linux implementation of the locks do give you the current value and not cached, I checked the code for x86. On Wed, 28 Aug 2002, Jan Hudec wrote: > On Wed, Aug 28, 2002 at 02:54:47PM -0600, Tom Bradley wrote: > > > Well, every variable that may change without code being in place must in > > > first place be locked or operated using atomic operations. And these > > > > NO! > > > > Locks are used to keep two pieces of code from accessing a variable at the > > same time, nothing to do with cached values. I.e. the Reader/Writer > > problem. Atomic operations state that this operation is gaurenteed to > > complete without interuption, atomic variables are volatile, and are > > used to perform locks. Locks around a variable DO NOT gaurentee you are > > getting the latest value, you can still be getting a cached value. > > Locks, at least in kernel DO guarantee you are geting the latest value > because they are defined to be optimization barriers. > > Locks or atomic operations are only way to ensure you don't get garbage. > Consider loading sparc register with constant, when you only have 16 > bits for an operand in each instruction. > > ------------------------------------------------------------------------------- > Jan 'Bulb' Hudec <bulb@ucw.cz> > -- Kernelnewbies: Help each other learn about the Linux kernel. Archive: http://mail.nl.linux.org/kernelnewbies/ FAQ: http://kernelnewbies.org/faq/
