On Sun, Dec 2, 2012 at 10:26 AM, Chris Wilson <chris at chris-wilson.co.uk> wrote: > > There isn't a race with hardware. So are you concerned about the write > ordering, and so want some smb_mb()? Oops, I've misread the patch, there's no race where I've seen it. Barriers is a good topic though I think we need an smp_wmb before the atomic set in queue_flip to ensure that all stores to the work struct have landed before. Together with the implicit barriers of the spinlock we should be fine. -Daniel -- Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
