Old PowerMac G5 2-socket/2-cores-each: head -r368820 kernel reports: bus_dmamem_alloc failed to align memory properly

Tue Dec 22 06:50:40 UTC 2020

On 2020-Dec-21, at 20:25, Dennis Clarke <dclarke at blastwave.org> wrote:

> On 12/21/20 11:03 PM, Mark Millard wrote:
>> 
>> 
>> On 2020-Dec-21, at 19:02, Dennis Clarke via freebsd-ppc <freebsd-ppc at freebsd.org> wrote:
>> 
>>> On 12/21/20 9:58 PM, Dennis Clarke via freebsd-ppc wrote:
>>>> On 12/21/20 9:27 PM, Mark Millard via freebsd-ppc wrote:
>>>>> I'm upgrading from head -r365932 to -r368820 and on the first
>>>>> boot -s with the new (non-debug) kernel I saw that bus_dmamem_alloc
>>>>> was reporting based on:
>>>>> 
>>>> 
>>>> Do you know if there is a workable installer image for anything recent?
>>>> My powermac quad has been dead for a month and I just want to do a
>>>> fresh re-install with ZFS and all the CURRENT goodness.
>>>> 
>>>> 
>>>> 
>>>> 
>>> 
>>> Sorry that was a dumb question.
>>> 
>>> I see :
>>> 
>>> https://download.freebsd.org/ftp/snapshots/powerpc/powerpc64/ISO-IMAGES/13.0/
>>> 
>>> I will give that a try.
>> 
>> 
>> Not a dumb question. PowerMacs have been problematical for as long as
>> I've been trying FreeBSD on them (FreeBSD 10+ basically). That includes
>> examples of of install time problems.
>> 
> 
> Well, I had to give it a try :
> 
> https://beta.genunix.com/freebsd/ppc64/power_mac_quad_freebsd_13_current_17_dec_2020_fail.png
> 
> So that is the most recent installer image there. Not sure what fails.
> 
> I had this old powermac running FreeBSD 12.x and CURRENT last year. I
> even chased after a pile of little bugs and recall working with Justin
> Hibbits to get boot going smoothly with all the fans being correct etc
> etc. Things have changed a lot in a year.  I guess a Power9 server is
> really needed. Or maybe Power8 perhaps.
> 
>> I'm not aware of any unpatched, modern FreeBSD that well-operates any old
>> PowerMac that I have access to (64 bit or 32 bit). But I've not done a
>> from-scratch install (via a installer image) in years so I do not know what
>> the issues are with that stage these days. My understanding is folks
>> have been disabling SMP support so only 1 thread run in order to avoid some
>> of the problems for multi-socket/multi-core PowerMacs.
> 
> Right. I thought we had that fixed last year?

For https://reviews.freebsd.org/D23376 I never figured out how to do
what Justin was after in his comments (simplified summary). No one else
did the work either, as far as I know. I still run with my patches for
the issue.

> Bug 233863 - Various PowerMac G5 models may require kern.smp.disabled=1
> and must set usefdt=1 which causes net interface reorder
> 
> https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=233863

Most of my other PowerMac-operation related patches are in attachments
to the above, although they have not been updated there as FreeBSD has
progressed so some might not apply directly now. (What I've got does
build.)

>   also
> 
> Bug 238730 - r349985 on ppc64 IBM 970MP PowerMac G5 sys/dev/bge/if_bge.c
> must move the device_get_devclass(bus)
> 
> https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=238730

The above one looks to have been checked-in as head -r350025 .

> Bug 233579 - ppc64 r341455 will panic on boot with usefdt=1
> https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=233579

Some of my patches in 233863 got usefdt=1 to work more generally.
I use it. I even made it the default in my environment and instead
have a means of disabling it from the loader. Making it work in
part involved allowing some Apple OpenFirmware violations to not
break the operation. (So far as I know the code still handles
lack of the violations correctly.)

I'll note that 233579 is classified as closed-fixed but has a later
note from you indicating that it was failing. The status was never
changed to track that so any quick look suggests the problem is/was
gone.

> I think you and I have gone back and forth on these and others and for a
> while there we had it working fairly well. With ZFS also.

Yep: we had lots of exchanges. (While I experimented for a time years
ago with ZFS, I use UFS. So that part I do not know about.)

>> 
>> Other notes:
>> 
>> As far as I know, 32-bit powerpc for old PowerMacs still has the kernel
>> gradually zeroing out user-space pages, even for single-socket/single-core
>> 32-bit PowerMacs. So I run 32-bit via a chroot on a 64-bit system: the
>> 64-bit kernel does not have this specific problem. (I seem to remember
>> that there was a different boot failure last I tried 32-bit, but I do not
>> remember any detail at this time.)
>> 
>> There are other kernel problems as well (64-bit and 32-bit), but I'll not
>> get into them here.
>> 
> 
> Well I guess I have to pick my battles here and maybe move onwards with
> the RISC-V goodness wherein I have been in touch with Mitchel Horne and
> I have CURRENT running really well with ZFS. Well, on qemu of course.
> However I can not do a buildworld. That jsut fails in the LLVM/Clang
> world and it looks to be a LLVM/clang bug. Regardless the old old Apple
> PowerMac may be a doorstop now. A good looking shiney door stop but not
> really workable.  Even Debian Linux fails to install these days.
> 
>> The best of the old PowerMac's that I've had access to finally ended up
>> with overheating from the cooling system not working correctly any more.
> 
> I had not seen that here with mine and I have four of them. I take parts
> from one to the other just to keep one running well.

If you give up on them before things quit working, you might want to send
them to Justin or to someone else that likely could do significant PowerMac
work at some point if they had direct access for testing and such (and
they happened to feel like working on such for a bit on occasion).

>> The previously next-best does not have all the memory slots working. That
>> is the last 2-socket/2-cores-each G5 that I have access to. So one of
>> these day's I'll be joining others that no longer have access to such.
> 
> I feel that day is coming soon for me also.
> 
>> With the degree of my time preferences, I'm not sure if I'd use the old
>> PowerMacs without the faster one to do builds on. (I tend to cross
>> build the systems, only self-hosting on rare occasion. But ports
>> I build on the Fastest PowerMac that I have access to.)
>> 
> 
> Well if there was such a thing as a reasonable serial console on these
> things then I would easily grant you ssh access to a KVM switch and to
> the serial console. Debugging anything via camera and the graphical
> console is a real pain.

I'm very familiar with camera and graphics console use. I touched ddb to
automatically run an internal script (no paging) before it got to the
prompt (unable to take input). I'd adjust the script and rebuild to get
different information that would fit on screen. Lots of attempts to
repeat failures to get more information. (The original problem that I
looked into was intermittent.)

If the PowerMac was getting far enough  along, I did manage to use the
FireWire access some in some later activities. I learned from that to
not believe where things looked to have stopped on screen when things
hung up: FireWire access showed more after what the screen displayed,
indicating a later failure point. (No claim FireWire's access/reporting
lasted to the end.)

> Thank you for the detailed reply and I will flip a coin or two and maybe
> just retire these shiney old monsters.  They use a pile of power also :)
> 
> That just leaves the Sun SPARC servers in my life as the most stable
> long term reliable monsters that I have. Sadly Oracle makes like hell
> and not much runs on the newer Fujitsu machines that I have. Looks like
> some investment in RISC-V may be reasonable for next year.
> 

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