revision 341104 quite unusable

[Mark Millard]

On 2018-Nov-29, at 11:16, Dennis Clarke <dclarke at blastwave.org> wrote:

> On 11/29/18 2:32 AM, Mark Millard wrote:
>> On 2018-Nov-28, at 14:50, Dennis Clarke <dclarke st blastwave.org> wrote:
>>> . . .
> 
>>> . . .
>>> OKay well, as I was saying at lung exhaustive length, I have yet to see
>>> anything build and boot other than the RC2 dvd.  At least in a natural
>>> way.
>> And you started from a context that was observed to be a problem
>> before you even had started (as far as I can tell). If you want
>> things to be more likely to work directly, you would back up to
>> source vintages from before the problem was observed initially for
>> hardware contexts similar to yours.
>> You could always tackle the later issues later.
> 
> I have been working directly from the RC2 install which does boot and
> run with a single cpu core.  Then dragging in the src and doing a build
> seems to make sense but to skip over all that fuss I can drag down bits
> from https://artifact.ci.freebsd.org/snapshot/head/r341253/powerpc/powerpc64/ and see what happens.

FYI for the time relationships and the versions that I tested
and some related observations:

The observed G5 problems started with head/ -r334498 :
it goes back to 2018-Jun-1 .

stable/12/ started as -r339434 being a copy of head/ as of -r339432 :
2018-Oct-19. So long after the problem started to be observed on G5s.

releng/12.0/ started as -r340470 being a copy of stable/12/ as of -r340462 :
2018-Nov-16.


Any correct fix that makes all powerpc family contexts work correctly likely
should work if applied to -r334498 or later instead of just to stable/12/
or releng/12.0/ or just recent revisions of head.


Below I list the reported testing that I did. The wording is
relative to my G5 use. Here I sort by the revision numbers
instead of the order in which I did the approximate bisection.
This was from an attempt to jump from -r333594 to -r339076
that showed the problem.

head -r333594 kernel always has worked
head -r334298 kernel worked
head -r334474 kernel worked
head -r334494 kernel worked
NOTE: head -r334498 is the significant powerpc families
      change between the above and below.
END NOTE
head -r334528 kernel failed
head -r334554 kernel sometimes worked, later: sometimes failed
head -r334594 kernel failed
head -r334644 kernel failed
head -r335006 kernel sometimes worked, later: sometimes failed
head -r335852 kernel sometimes worked, later: sometimes failed
head -r336706 kernel failed
head -r339076 kernel failed

(All the above are before stable/12/ and releng/12.0/ existed.)

More testing of -r334494 to see if it ever fails could be
interesting.

Note the examples of the same kernel sometimes working and
other times not working for the same G5 machine, almost like
it was dependent on uninitialized memory that was varying.
(Again: long before stable/12/ and releng/12.0/ existed.)

Unload/load sequences from the loader prompt to switch kernels
seemed more likely to induce the "sometimes worked" and
"sometimes failed" variations for the same kernel version.

I've not tried to report above any world vs. kernel ties that
might make some mixes problematical. In general I was not
updating world unless I had to. I do not remember the details
for the mixing any more.

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