On Wed, 29 Nov 2006 00:50:43 +0100 Stefan Richter <[EMAIL PROTECTED]> wrote:
> Alan wrote: > > On Tue, 28 Nov 2006 22:24:11 +0100 (CET) > > Stefan Richter <[EMAIL PROTECTED]> wrote: > >> All MMIO writes which were surrounded by the spinlock as well as the > >> very last MMIO write of the IRQ handler are now explicitly flushed by > >> MMIO reads of the respective register. > > > > MMIO is ordered anyway on the bus, you just need mmiowb() to force > > ordering to the bus controller in case you are on a big numa box. > > The mmiowb is a checkpoint to ensure ordering between different threads > of MMIO writes; i.e. it doesn't halt the thread until the write actually > reached the device like a read would do, right? It guarantees that no other mmio will sneak past it from another thread but doesn't guarantee the previous I/O has hit the hardware. It's a much weaker (and thus far faster) guarantee which is usually sufficient as it can be combined with spin_unlock to enforce I/O ordering matching the lock ordering. Alan - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/