Re: LOCK prefix on uni processor has its use

["Michael S. Zick" <lkml@morethan.org>]

On Wed May 27 2009, Andi Kleen wrote:
> Harald Welte <HaraldWelte@viatech.com> writes:
> > * All X86 instructions except rep-strings are atomic wrt interrupts.
> > * The lock prefix has uses on a UP processor: It keeps DMA devices from
> >   interfering with a read-modify-write sequence
> 
> In theory yes, but not in Linux -- normal drivers simply don't use LOCK in any way
> on a UP kernel.
> 
> We discussed exactly this in the earlier subthread :)
> 
> > Now the question is: Is this a valid operation of a driver?  Should the driver
> > do such things, or is such a driver broken? 
> 
> The driver is broken because if it relies on this it will not work on a UP kernel.
> Also it's not portable and in general a bad idea.
> 
> > When would that occur?  I'm trying
> > to come up with a case, but typically you e.g. allocate some DMA buffer and
> > then don't touch it until the hardware has processed it.
> 
> Is it known which driver has this problem?
> 
> -Andi (who finds hpa's "timing theory" to be more believable anyways)
> 

I still have not come up with a solid, testable, theory to explain the
order of magnitude in up-time before the kernel locks with/with-out 'lock'.

But we are definitely pecking around the edges of the problem.  ;)

Today's lockdep build has just passed its previous record by hard-coding
the pci cache line size to be the same as the cpu's cache line size. (a WAFG).
Until we hear back from the VIA-CPU people, I just guessed that since the
chip set was designed for use with the processor...

Mike