Re: Looks like the arm 20220805 snapshots are still odd, so probably kern.geom.part.mbr.enforce_chs=0 was still in use

> Am 07.08.2022 um 19:16 schrieb Mark Millard <marklmi@yahoo.com>:
> 
>> On 2022-Aug-7, at 14:51, Dr. Rolf Jansen <freebsd-rj@cyclaero.com> wrote:
>> 
>>> Am 07.08.2022 um 16:50 schrieb Mark Millard <marklmi@yahoo.com>:
>>> 
>>> On 2022-Aug-7, at 12:32, Glen Barber <gjb@freebsd.org> wrote:
>>> 
>>>> Correct, it was set to “0” for these builds.
>>>> 
>>>> I honestly do not have any idea where the problems you are seeing are creeping in.
>>>> 
>>>> Should it be set back to “1”?  I’m not sure how to proceed otherwise.
>>> 
>>> My guess is that if the release/tools/arm.subr line:
>>> 
>>>            chroot ${CHROOTDIR} gpart add -t freebsd-ufs -a 64k ${mddev}s2
>>> 
>>> was instead (note the added -b use):
>>> 
>>>            chroot ${CHROOTDIR} gpart add -t freebsd-ufs -b 64k -a 64k ${mddev}s2
>>> 
>>> then the line:
>>> 
>>>            chroot ${CHROOTDIR} newfs -U -L rootfs /dev/${mddev}s2a
>>> 
>>> would work as expected and things would still be aligned:
>>> no aliasing of BSD vs. freebsd-ufs. (In part this is by
>>> prior steps already having achieved alignment of BSD.)
>> 
>> From a strict mathematical stand point of view, -a is not necessary when using -b with an aligned value.
> 
> "-a" controls more than the start offset: also the size.
> 
> QUOTE
>               -a alignment  If specified, then the gpart utility tries to
>                             align start offset and partition size to be
>                             multiple of alignment value.
> END QUOTE
> 
> I expect your statement would at most apply to the start offset, not to
> everything "-a" does.
> 
>> Personally I don’t use the -a option of gpart anymore since it started to do funny magics in front of face. If I remember correct, gpart of the FreeBSD 9.x-RELEASES was still OK (or was it 8?). Nowadays, I align all my storage media by employing -b with a reasonable value.
>> 
> 
> I've no clue of the specifics that you are referencing and so can not check.

It started at unexpected offsets and then left unexpected space at the end - sorry, I don’t use -a anymore for years, and I don’t remember the very details. 

That said, why then would we use -b in addition to -a, if -a would show the expected behaviour on its own. That is because either -a adds some "funny magic" to the logic, that is stated in the man file, or there is a bug in gpart.

Anyway, using -a and -b together bears the danger of the equation being overestimated, namely in the case base is not a multiple (incl. 1) of alignment. Although, that is not the case with your suggestion. -b 64k -a 64k is OK.

Finally, I still wonder what might be the technical reason for aligning the size.

Best regards

Rolf