panic: data storage interrupt trap when building world on PowerMac G5

[Mark Millard]

On 2020-Jan-23, at 12:29, Michael Tuexen <tuexen at freebsd.org> wrote:

> On 23. Jan 2020, at 20:41, Justin Hibbits <chmeeedalf at gmail.com> wrote:
> 
>> . . .
>> 
>> Is there a reason you still have kern.smp.disabled=1?  I fixed the bug
> Without that, the fans get very loud, on reboot you get timeouts, and
> I think (haven't tried recently) the network interfaces die. With that,
> no such problems.

My experiments with controlling the mftb value relationships from
cross-core (/cross-socket) contexts avoided the fan behavior and
the shutdown/reboot timeouts in my testing.

But I've no evidence associated with network interfaces being tied to
that. In fact, I've never had ethernet interface problems, other than
usefdt=1 swapping which was 0 vs. which was 1. (The fdt conversion
reverses each sibling list but the code using the nodes is based on
fixed traversals, independent of lack vs. presence of a
usefdt definition or the value it is defined to.)

I've basically only used ethernet, so I can not comment on other
networking interfaces.

(I also run with patched code so that the fdt conversion preserves
the order of sibling lists, leaving the fixed traversals alone.
I also have some OF_xref_from_node and OF_node_from_xref (lack of)
usage corrections involved. I'll avoid listing more here. But I do
not get distinct traversal results from usefdt mode.)

> This is the first time I try to do a buildworld after moving to clang
> (which was done by a reinstall). Before the switch, the machine did multiple
> buildworlds and was running very stable.

(I've switched to ELFv2 for powerpc64, but, prior to that, my context
was based on experimental system-clang ELFv1, either built by
system-clang or, at the time, devel/powerpc64-gcc .)


===
Mark Millard
marklmi at yahoo.com
( dsl-only.net went
away in early 2018-Mar)