On Wed, 05 Nov 2008 12:29:52 GMT Richard Porter <[EMAIL PROTECTED]> wrote:
> > His point is that in a PMT system, such deadlocks cannot occur, but > > in a CMT system, enormous care must be taken to stop them being > > commonplace. > > As I recall, and it's a long time since I worked on a CMT kernel, the > golden rule was always to grab and release resources in a set > sequence so that you avoid deadly embraces, but this case isn't quite > that simple! It also isn't a solution :) > The other rule is that you give up control to the > scheduler every so often in case any higher priority task is waiting > to run. In general I/O bound tasks are high priority and CPU > intensive tasks are low priority. Well, there are no priorities in CMT, really. And for complex systems, it can be difficult to yield while in the middle of a complex process. Especially if the OS doesn't actually let you efficiently check and yield if there's nothing to do.
