Re: [PATCH v2 4/5] Patch the CHRP boot script

[Mathieu Malaterre <malat@debian.org>]

On Tue, Nov 21, 2017 at 3:10 PM Nathan Fontenot
<nfont@linux.vnet.ibm.com> wrote:
>
> On 11/21/2017 03:27 AM, Benjamin Herrenschmidt wrote:
> > On Tue, 2017-11-21 at 09:56 +0100, Mathieu Malaterre wrote:
> >> Hi Ben,
> >>
> >> Sorry for the direct question, but I assumed you would know the
> >> answer. We are currently trying to replace yaboot with grub-ieee1275
> >> in debian-installer, but are facing a very small issue, namely with OF
> >> path handling. It appears that on Apple hardware only `ofpath` from
> >> Yaboot returns the correct path. Are you using grub on any of your
> >> Apple hardware ? Thanks much for comments.
> >
> > +Nathan
> >
> > I have used grub on Apple systems in the past but it's quite possible
> > that I ended up fixing up the path by hand. To be honest I haven't
> > booted any of my Apple systems in a while, but I still have some I
> > can dig out to test if needed.
> >
> > Some more comments below...
> >
> >> See below for full context.
> >>
> >> On Tue, Nov 21, 2017 at 7:51 AM, Frank Scheiner <frank.scheiner@web.de> wrote:
> >>> Hi Rick,
> >>>
> >>> On 11/21/2017 02:48 AM, Rick Thomas wrote:
> >>>>
> >>>> If you can tell me *exactly* what to do, and I don’t have to set up an
> >>>> installation environment to do it, I’ll be happy to test ofpathname vs
> >>>> ofpath vs devalias on my PowerPC test machine farm.
> >>>
> >>>
> >>> Well, to be on the safe side, I think you need to use the latest versions of
> >>> ofpath and ofpathname, hence running Debian Sid could be the easiest way to
> >>> make sure this is the case.
> >>>
> >>> You need to have a disk installed, because ofpath and IIC also ofpathname do
> >>> not or cannot translate non existing device aliases to OF paths.
> >>>
> >>> In general the commands I ran for the p5 520Q below should do. Assuming that
> >>> the output of ofpathname will always be wrong for Power Macs, it should be
> >>> sufficient to check only one partition.
> >>>
> >>> I think devalias does only save/return OF paths up to the disk level, but
> >>> not up to the partition level. But the disk level should already be
> >>> sufficient to detect differences to ofpath and/or ofpathname.
> >>>
> >>> You have to go to OF to run devalias, but with a glass console you won't be
> >>> able to copy the output. There are ways to interact with the OF via telnet
> >>> from another machine (see [1] and possibly [2]), but I haven't tried this
> >>> yet.
> >>>
> >>> [1]:
> >>> https://web.archive.org/web/20040202053614/http://developer.apple.com/technotes/tn/tn2004.html
> >>>
> >>> [2]:
> >>> https://web.archive.org/web/20040202060137/http://developer.apple.com:80/technotes/tn/tn2023.html
> >>>
> >>>>
> >>>> I have the following machines:
> >>>>      Power Mac G5 11,2,
> >>>>      Power Mac G5  7,2  (I think — it’s turned off right now)
> >>>>      Mac mini G4 10,1
> >>>
> >>>
> >>> I have these types available, too, "except" for the 7,3 which currently
> >>> won't start up correctly.
> >>>
> >>> Here's the output of ofpath and ofpathname for /dev/sda2 on the my Mac mini
> >>> (10,1):
> >>> ```
> >>> root@mac-mini:~# ofpath /dev/sda2
> >>> /pci@f4000000/ata-6@d/@0:2
> >>>
> >>> root@mac-mini:~# ofpathname /dev/sda2
> >>> /usr/sbin/ofpathname: line 812: warning: command substitution: ignored null
> >>> byte in input
> >>> /pci@f4000000/ata-6@d/scsi@0/sd@0,0
> >
> > So here, ofpathname gets confused by the fact that Linux shows ATA
> > devices as scsi (which is a Linux'ism). ofpath has workarounds to deal
> > with that which should probably be ported over.
> >
> >>>
> >>> Also interesting, it looks like ofpathname cannot even give a "correct"
> >>> result for an IBM machine, at least not for my p5 520Q:
> >>> ```
> >>> root@p5-520q:~# lsblk
> >>> NAME   MAJ:MIN RM   SIZE RO TYPE MOUNTPOINT
> >>> sda      8:0    0 136.7G  0 disk
> >>> ├─sda1   8:1    0     7M  0 part
> >>> ├─sda2   8:2    0 131.1G  0 part
> >>> ├─sda3   8:3    0     1K  0 part
> >>> └─sda5   8:5    0   5.6G  0 part
> >>> sr0     11:0    1  1024M  0 rom
> >>>
> >>> root@p5-520q:~# ofpath /dev/sda1
> >>> /pci@800000020000003/pci@2,4/pci1069,b166@1/scsi@0/@3:1
> >>>
> >>> root@p5-520q:~# ofpathname /dev/sda1
> >>> /pci@800000020000003/pci@2,4/pci1069,b166@1/scsi@0/sd@3,0
> >>>
> >>> root@p5-520q:~# ofpath /dev/sda2
> >>> /pci@800000020000003/pci@2,4/pci1069,b166@1/scsi@0/@3:2
> >>>
> >>> root@p5-520q:~# ofpathname /dev/sda2
> >>> /pci@800000020000003/pci@2,4/pci1069,b166@1/scsi@0/sd@3,0
> >>> ```
> >>>
> >>> Please notice that ofpathname returns the same result for two different
> >>> partitions. :-/
> >
> > Ouch. I think ofpathname is still maintained by some of our IBM guys
> > isn't it ? Might be worth digging there (or fixing it up !).
> >
> > Nathan, do you know anything about this ?
>
> I haven't seen this before.
>
> I don't have access to a Mac to re-create what you're seeing above, if you can
> send me the output from running the script with 'set -x' at the top I will take a
> look.

Just for reference Adrian prepared a PR:

* https://github.com/ibm-power-utilities/powerpc-utils/pull/39

> You could also do this for the command above that produces the waning on line 812.
>
> -Nathan
>
> >
> >>>>      Mac G4 3,4
> >>>>      Mac G4 dual core 3,6
> >>>
> >>>
> >>> Results for these machines would be helpful. It should work to
> >>> install/upgrade to Debian Sid on one machine only and just move the disk to
> >>> the other machine later on for testing.
> >
> > I can test if you are willing to fix the script :)
> >
> > Cheers,
> > Ben.
> >
> >
>