Re: NetBSD-current on Quadra 610

[Izumi Tsutsui <tsutsui%ceres.dti.ne.jp@localhost>]

martin@ wrote:

> On Mon, Aug 11, 2014 at 10:42:17PM +0000, John Klos wrote:
> > I don't know whether USE_SSP=yes is default or not, but I haven't changed 
> > anything.
> 
> It is not default. I don't mind switching over, if it is only my machine
> affected - didn't have time to track it down further (and gcc 4.5 miscompiles
> my kernel too, or there is a driver bug somewhere corrupting memory).
> 
> Guess I should test the new kmem poision checks.

Per bsd.sys.mk USE_SSP enables -fstack-protector and
it allocates more stack size, so it might be possible
that extra stack area hides wrong accesses (or out of bounds)
against stack variables in a certain funtion.

If we can specify -fstack-protector per each source
and we can investicate which source requires -fstack-protector,
it would help to track the route cause.

There are also a bit awful comment around rtclock_intr() in locore.s
and I wonder if such old hack still works for -current MI hardclock...
(INTERRUPT_SAVEREG and INTERRUPT_RESTOREREG pair also look weird)

---
Izumi Tsutsui