On Fri, Feb 28, 2003 at 08:44:07PM +0100, Frank A. Uepping wrote: > On Friday 28 February 2003 11:41, you wrote: > > On Fri, Feb 28, 2003 at 12:29:54AM +0100, Rok Papež wrote: > > > Hello! > > > > > > Dne četrtek 27. februar 2003 20:47 je Frank A. Uepping napisal(a): > > > > Is that true that a spinlock gets undefined when compiled > > > > on a uniprocessor system? > > > > > > No. > > > > YES, IT DOES. Except _irqsave/_irqrestore variants (they reduce to > > cli/sti). And for 2.4, that is. With preempt patched (IIRC they are > > included in late 2.5) they become magic. > > > > > > Thus spinlocks can't be used to provide mutual exclusion between > > > > concurrently executing processes on a uniprocessor system, isn't is? > > > > > > No. > > > > There is a limitation. You must not, under penalty of complete machine > > freeze, call schedule() with spinlock locked. You can never be > > interrupted except by interrupt unless you call schedule() [except with > > preept patches. Think of a preempt kernel as always running SMP]. > > > > You however must always write code for SMP and on UP spinlocks reduce so > > that they guarantee the same as on SMP taking into account that running > > on UP guarantees some things itself. > > > > > I use them on UP as mutexes.. heavily. You use spinlock if structures you > > > are protecting are being accessed from interrupt/timer/softirq context. > > > > Only _irqsave/_irqrestore variants! Unsing the normal variant to lock > > against interrupt is a bug under penalty of complete machine freeze. > > > > > Mutexes are to be used when structures are accessed *only* from process > > > context. > > > > Yes, semaphores can only be manipulated in process context. > > > > --------------------------------------------------------------------------- > >---- Jan 'Bulb' Hudec <bulb@ucw.cz> > > For clarification. > When I have a struct that I access from proc and int context I have > to use the _irqsave() variant in the proc context (in order to prevent > an irq to be happen) but in the int context I stick to the common variant. > Further a spinlock in not allowed to hold while the current proc sleeps. > Semaphores are not allowed to use in int context because they might > sleep/block. > /FAU Yes. Except you should probably use _irqsave/_irqrestore in int context too (it saves flags and restores them un unlock so it does not matter if interrupts were enabled befor). IIRC in bottomhalves/tasklets interrupts are enabled, so you need irqsave there. ------------------------------------------------------------------------------- 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/
