[Cbe-oss-dev] [PATCH 09/13] spufs: block fault handlers in spu_acquire_runnable

[Benjamin Herrenschmidt]

On Mon, 2007-11-19 at 14:36 +1100, Jeremy Kerr wrote:
> From: Arnd Bergmann <arnd.bergmann at de.ibm.com>
> 
> This change disables the logic that under-the-covers faults-in
> spu contexts from the page fault handler. When a fault requires
> a runnable context, the handler will block until the context is
> scheduled by other means.

Is this a interruptible wait or a non-interruptible wait ?

In the former case, what happens if I take a signal ? In the later case,
what happens if I page fault from a thread while kill -9'ing the app at
the same time ?

I'm afraid this solution can cause processes to be stuck in the kernel
with absolutely no way to get them back or kill them, even with kill -9.
It needs to go along with something that at least tests if a fatal
signal happened so that those tasks can at the very least be killed by
root.

There was a patch intoducing a TASK_DEAD state to that effect a while
ago on lkml. Something followed what was the status ?

Cheers,
Ben.