On Sun, Apr 10, 2016 at 1:29 PM, Linus Torvalds
<torva...@linux-foundation.org> wrote:
>
> I think the right choice might be to
>
> (a) revert that patch (or just change the signal_pending() into
> fatal_signal_pending())
>
> (b) to get the latency advantage, do something like this:
The attached patch is actually tested and seems to fix the issue.
I do not have a good way to check the latency of signal delivery and I
didn't check if the signal_pending() actually ever triggers, but your
test-case that showed the problem before seems to be fine with it.
Linus
