On 21.05.14 21:47, Alexander Graf wrote:
Am 21.05.2014 um 18:33 schrieb Paul Janzen <p...@pauljanzen.org>:
On Wed, May 21, 2014 at 12:45 AM, Alexander Graf <ag...@suse.de> wrote:
Or maybe it's safer overall to just call write_IRQreg_idr() instead of
setting idr directly? That would update destmask along the way as well and
we would catch all subtle corner cases.
I considered this initially but decided to go with the simpler
approach. But you have convinced me. Specifically, the current
implementation does not reset src->output or src->nomask, which
write_IRQreg_idr() will. I will test this change and re-submit the
patch.
Do you have a simple test case for this patch? We seem to have the same bug
in the in-kernel KVM MPIC emulation code and I'd like to have it fixed there
as well, but I don't really like to do that change blindly.
I have tested this patch using a proprietary embedded operating system
which doesn't bother to initialize the IDR registers.
Do you think it'd be possible for you to test whether it works with kvm when I
give you a patch? Do you have real hardware to run kvm on? :)
I just checked - we don't properly implement reset at all in the
in-kernel MPIC we do in KVM. So it's merely something to keep in mind
for now until someone sits down and implements proper reset support ;).
Alex
