[Bug 235060] [boot] FreeBSD 12.0 Release DVD and CD will not boot on PowerMac G5 Quad

[bugzilla-noreply at freebsd.org]

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=235060

--- Comment #15 from Mark Millard <marklmi26-fbsd at yahoo.com> ---
(In reply to Dennis Clarke from comment #13)

First my overall point, then what leads to it . . .

Overall point:

I'm not sure testing anything with head -r334498
involved is worth testing for multi-socket or for
multi-core unless the code has been changed to
address the resultant slb-miss issue on the
non-bsp cpu(s) in some way.

What leads to that:

I'll note that anything form just svn at or
after head -r334498 (or later in svn that is
derived from such) has known boot problems on
at least some G5 PowerMacs that have multiple
sockets or cores. The issue in question is an
unhandled slb miss in the code during a
(non-bsp) cpu's startup code when it executes:

sp = pcpup->pc_curpcb->pcb_sp

It is the pc_curpcb based dereference that
can end up with a slb miss before things
are set up sufficiently for that processor
to correctly handle such misses.

head -r334498 changed the address ranges used
for powerpc64 and indirectly resulted in this
being possible on some old G5 PowerMacs.
(32-bit powerpc FreeBSD boots of old G5's
are a separate issue.)

I'm not here making any claims about there not
being other boot problems sometimes involved.
But, to my knowledge, all builds based on just
checked-in source code that has -r334498
involved has multi-socket/multi-core boot
problems. (They can be intermittent, sometimes
hanging-up and other times not, depending on
slb content at the time.)

Bugzilla 233863 that you reference has a
fair amount of accumulated information, but not
limited to fixing what -r334498 exposes about
the way things historically worked, so it is a
bit of a mess to go through for the slb issue.

-- 
You are receiving this mail because:
You are the assignee for the bug.