[Cbe-oss-dev] spufs poll issues

[Christoph Hellwig]

On Tue, Nov 27, 2007 at 09:23:58AM -0200, Luke Browning wrote:
> poll is not supposed to wait for events (i/o) to happen, but it is
> required to atomically access kernel data.  It therefore needs to take
> locks and may block for a lock. 

The primary poll/select wait is handled in kernelspace in Linux.  If you
look at other poll implementations you'll see that they all use
spinlocks or lockless algorithms only.