Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 30 Apr 2019 14:04:46 -0700
From:      Mark Millard <marklmi@yahoo.com>
To:        Justin Hibbits <chmeeedalf@gmail.com>, FreeBSD PowerPC ML <freebsd-ppc@freebsd.org>
Subject:   Do non-PowerMac G5 powerpc64's ever get any slb faults  in cpudep_ap_bootstrap?
Message-ID:  <F14FF29E-836E-4926-B432-2662C8951FCD@yahoo.com>

next in thread | raw e-mail | index | archive | help
Justin wrote in a bugzilla comment:

> I'm trying to understand why Mark's SLB hack is necessary,
> since it should logically be necessary on any other AIM-architecture =
platform,
> including my POWER9, so I'm really confused.  Unfortunately, my G5 no =
longer
> works, and actually died shortly before I committed the change that =
reportedly
> broke G5s.  I was/still am hoping that someone could go beyond Mark's =
diving,
> and determine just why the APs can't handle SLB faults at that point =
in the
> boot process, because they really should be able to.


Could you find out if the AIM-ish powerpc64 contexts that you have
access to ever seem get slb-miss faults in the pc_curpcb dereference
in cpudep_ap_bootstrap 's:

sp =3D pcpup->pc_curpcb->pcb_sp

? If they do, they are handled or you would see some effect. But,=20
are they even happening? If they are not happening, that could be
a difference with the G5s.

I ask for such, in part because of . . .

The old VM_MAX_KERNEL_ADDRESS value avoiding the issue on G5s
seems to be simply that the G5s were never getting the slb
misses at that line when the old value was in use. (In other
words, I expect that the change exposed a long-standing
problem that was accidentally avoided until the change.)

Unfortunately, I'm limited to G5s and you are limited to non-G5s.
I can not check if other contexts are getting slb-misses in the
same place or not but you might be able to. (But how?)

I've not figured out anything beyond code-inspection for tracking
down what happens for the slb-miss at that specific place when
such happens on the G5's. So far I've not come up with any
potentially useful ideas.


=3D=3D=3D
Mark Millard
marklmi at yahoo.com
( dsl-only.net went
away in early 2018-Mar)




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?F14FF29E-836E-4926-B432-2662C8951FCD>