Centris 650 Debian 10 SID Installation

classic Classic list List threaded Threaded
84 messages Options
12345
Reply | Threaded
Open this post in threaded view
|

Re: Centris 650 Debian 10 SID Installation

Finn Thain
On Wed, 12 Jun 2019, [hidden email] wrote:

> On 6/10/19 9:32 PM, [hidden email] wrote:
> > On 6/10/19 7:20 PM, Finn Thain wrote:
> >>
> >> Why not use the serial console?
> >>
>
> I was able to install a text-only installation on the Centris 650 using
> the serial console ("console=ttyS0,9600n8").

Nice work!

> It's not necessary to hit "1"; the installation screen eventually comes
> up.
>
> Here are a few observations and comments:
>
> 1) Approximate times were as follows:
>    a) 30 min - from initial setup to prompting of a Debian mirror.
>    b) 4 hrs  - from start of scan of Debian mirror to "popcon" prompt.
>    c) 5 hrs  - from "popcon" prompt to completion of loading packages.
>    d) 3 hrs  - from selection of "openssh server" and "system utilities"
> to completion of installation.
>

That sure makes debootstrap look good. On fast amd64 system packages are
unpacked a lot faster, of course. Stage 2 still happens on the target m68k
system. The --include and --exclude options might be helpful if you're
changing over to sysvinit or openrc.

> 2) After installation, the system was rebooted after a prompt.  A
> question mark appeared on the main screen.  Booting from an external
> disk, Mac OS volumes could be mounted manually using "Disk Utility", but
> they were still not seen after a reboot.  The problem was fixed by
> running "Apple HD SC Setup 7.3.5p" and updating the Apple driver on the
> affected disk.  So it seems likely that something in the installation
> corrupted the Apple driver.
>

Seems so. Is that a parted bug (which could be reproduced outside of the
installer)? Or is the installer expected to clobber existing partitions
(if you don't skip the repartitioning step)?

> 3) Booting into the new Linux installation using a 5.x kernel, the
> kernel crashed after not finding a valid init.  As it turns out, if a
> separate /usr is specified during the installation, then the system
> won't boot, because /usr won't be mounted yet and the following
> directories are symbolic links:
>
> /bin -> usr/bin
> /lib -> usr/lib
> /sbin -> usr/sbin
>
> I realize that most users will probably just use a single filesystem for
> everything, but there have always been compelling reasons to use
> separate root and usr partitions.  At any rate, if it doesn't work to
> specify separate partitions, then the installer should warn about that
> (or maybe have a "Root & Usr" partition option like NetBSD does and then
> not allow specification of a separate /usr partition).
>

I never use a separate /usr as I don't get any benefit. But it seems to me
that those symlinks would not be a problem when using Debian's initrd.

> 4) After booting into a backup partition (Debian 3.1) and backing up "/"
> and "/usr" from the new installation, and then restoring everything to a
> single "/" filesystem, the new system booted using the kernel (and
> initrd) that were created during the installation.
>

How did you avoid the ADB keymap problem?
 
> 5) The system never reached multiuser mode; the startup sequence looped
> on "Starting Network Time Synchronization" (see attached console log
> "Centris_650-Debian_10.txt").  The systemd timeout for this task was 1
> min 30 sec, but it never succeeded within that limit (I let it try 10
> times before forcing a reboot).  I'll try other things, including single
> user mode and checking whether I can increase the limit somehow, or
> possibly disable network time synchronization temporarily.
>

I've seen that happen in Aranym (with no network interface). I find it
baffling that systemd would cripple the entire system merely because
network time sync failed. I don't know whether it is a bug or a design
decision. I prefer to think it is an old bug.

IMHO these log messages are a bit silly --

[    **] (3 of 3) A start job is running for /dev/sda4 (1min 5s / 1min 36s)
...
[ TIME ] Timed out waiting for device /dev/sda4.
[DEPEND] Dependency failed for /dev/sda4.
[DEPEND] Dependency failed for Swap.
...
         Activating swap /dev/sda4...
...
[  450.650000] Adding 524284k swap on /dev/sda4.  Priority:-2 extents:1 across:524284k FS
...
[  OK  ] Activated swap /dev/sda4.


Years ago when I tried systemd on a Mac LC III, the serial console device
unit timed out, so no getty was started for that console...

It will be interesting to see how much more quickly you can boot this
system using openrc or sysvinit. (I see that the 450 seconds logged here
includes the 100 second delay caused by CONFIG_CRYPTO_DH; you can avoid
that by adding initcall_blacklist=dh_init to the kernel parameters.)

> Thanks to all who are working so hard to keep it possible to install a
> modern GNU/Linux system on 30-year-old hardware!
>

You're welcome. It wouldn't be possible without testers like yourself!

--

> -Stan
>

Reply | Threaded
Open this post in threaded view
|

Re: Centris 650 Debian 10 SID Installation

John Paul Adrian Glaubitz
In reply to this post by userm57
On 6/9/19 10:16 PM, [hidden email] wrote:
> The mailing list seems to have munged my message, adding extra question
> marks.  I'm not sure why, maybe because my mail client formatted the
> message in HTML instead of plain text (sorry about that).
>
> Anyway, here it is again, hopefully this will be more readable.

Thank you!

> on a Centris 650 (136 MB, 4 GB root partition, 2 GB swap partition).

That machine has a 68040 clocked at 25 MHz. While it works general, it's
just very slow. I have the very same machine myself.

> My
> plan was to install and configure
> everything on the 650 (including replacing systemd with SysV init since
> systemd doesn't do well on slow,
> low-memory systems) and then create a filesystem image for my other,
> even slower, mac68k systems.

My Amigas run systemd without any particular problems. There is currently
a regression with 239 and newer that delays the login quite a bit.
Downgrading systemd to 238 fixes the problem. I haven't had the time yet
to debug this problem.

> After copying the kernel and initrd from CD to disk and booting with
> Penguin, the installation proceeded,
> though very slowly; at every re-draw of the screen, I could see each
> line being re-drawn.

You should try one of the 9.0 images and see if that makes any difference
so we can figure out whether there is a regression.

I tried twice

> using the stock CD-ROM drive and once using a more modern CD-ROM drive.
> In all three cases, installation
> of the basic system succeeded after about four hours, then stopped at
> the "Configure the package manager"
> menu -- "Your installation CD or DVD has been scanned ... Scan another
> CD or DVD?".  I selected "No" and
> hit return:  no response, even after several hours.  Hitting return
> repeatedly or trying to select "Yes"
> or "Go Back" also doesn't work, and eventually the arrow keys stop
> responding (though F1 and F2 still
> work to move the selection left or right).  Switching to an alternate
> console didn't seem to work (or I
> was doing it wrong; I tried several combinations of ctrl-alt-f1,
> ctrl-alt-right_arrow, etc.).

I'm not aware of any bugs in this part of the installer.

> After one of the failures, I tried booting into the new system.  Booting
> using the kernel from the CD
> didn't work -- there was a two-minute pause, which ended in a kernel
> panic after not finding a root
> filesystem (or any SCSI devices).  Maybe that kernel only works for
> installation using the accompanying
> initrd?

No. The kernel is 100% identical with that of the installed system. There
are no separate installation kernels. In fact, debian-installer is built
by extracting the kernel from the linux-image Debian package used for the
kernels for installed systems.

? I next booted into the new system using a 5.x kernel, thinking
> that I could use apt-get to
> install everything else, but the keyboard map appeared to be wrong, so I
> couldn't log in at the VGA console.

Yes, dpkg-reconfigure keyboard-configuration should help with that.

> I next tried booting using a serial console (console=ttyS0,9600n8).

My preferred way of hooking up to any m68k machines. Works fine for me.

> After the expected slowness of
> systemd bringing everything up (about eight minutes), I see a login
> prompt on the VGA screen and on
> the serial console.  The keymap on the VGA screen is wrong, but appears
> to be ok on the serial console;
> however, attempting to login using root or the regular user account set
> up during installation fails
> ("Login incorrect").  Apparently users and passwords had not been
> configured yet when the installation
> hung.

You can try booting with init=/bin/bash to get a shell prompt.

>  Next I booted back into Debian 3.1 (using the same 5.x kernel) to see
> whether I could chroot to the
> new installation, add a new user and reset root's password.  That worked.

See above, use init=/bin/bash. Booting a modern system with such an old
kernel can result in unexpected bugs. I wouldn't do that.

> Booting back into the new installation, I'm able to log in as root at
> the serial console.  I don't see
> an installer log (I can send that if someone knows where it is or how to
> access it from inside the
> installer that hung).  The network is up, and apt-get works, but it
> doesn't find anything, even after
> adding these two lines to /etc/apt/sources.list:
>
> deb http://ftp.ports.debian.org/debian-ports/ sid main
> deb-src http://ftp.ports.debian.org/debian-ports/ sid main
http://ftp.ports.debian.org/debian-ports/
> Looking in a normal web browser, packages appear to be available in
> "http://ftp.ports.debian.org/debian-ports/pool-m68k/main/", but apt-get
> doesn't appear to see
> anything.  Maybe there's a simple error in my sources.list?

An error message here would be useful. Do you have the keyring package installed?

Adrian

--
 .''`.  John Paul Adrian Glaubitz
: :' :  Debian Developer - [hidden email]
`. `'   Freie Universitaet Berlin - [hidden email]
  `-    GPG: 62FF 8A75 84E0 2956 9546  0006 7426 3B37 F5B5 F913

Reply | Threaded
Open this post in threaded view
|

Re: Centris 650 Debian 10 SID Installation

John Paul Adrian Glaubitz
In reply to this post by Christian T. Steigies
On 6/10/19 12:01 AM, Christian T. Steigies wrote:
>> I tried leaving out the slash after debian-ports; apt-get still fails to
>> find anything.  Perhaps something failed to get installed or configured
>> after the installation hung.
>
> I doubt that.

Again, actual error messages would be helpful.

>> Alternately, if there is a way to restart the installation, skipping the
>> step where it asks whether any additional CDs or DVDs should be read,
>> that might also work.  Otherwise, I can probably get it working well
>> enough by manually installing what I need, though I'm not sure yet how
>> to fix the keymap issue on the VGA console.
>
> I don't think you can continue an installation, but maybe thats a new
> feature. There are always news things to learn.
>
> I am a bit confused by the layout of the ports mirror.
> It seems the m68k debs are in this directory:
>
> http://ftp.ports.debian.org/debian-ports/pool-m68k/main/
>
> But there is also this directory, that I thought that would be used for
> finding debs (this contains only all.debs?):
>
> http://ftp.ports.debian.org/debian-ports/pool/main/

That's normal. That's how the archive layout of Debian Ports looks like
and is owed to the fact that Debian Ports supports uploading packages
for the unreleased distribution per architecture.

> On the official mirrors, I do not see the arch in the name of the pool
> directory, ie this has debs for different arches:
>
> http://cdn-fastly.deb.debian.org/debian/pool/main/b/bash/

The official mirrors use DAK, Debian Ports uses mini-DAK. The mirrors therefore
aren't comparable.

> Maybe that is a new feature (that your apt does not support) or something is
> broken with the ports server, which I doubt. Adrian will tell us?

No, nothing broken, nothing unusual. It should work.

> Maybe you can try to add this to your sources.list:
>
> deb http://deb.debian.org/debian/ buster main
>
> (yes, with a trailing slash is what I use, works on my notebook).

This will not work.

> You will not find m68k.debs there, but it should give you access to
> all.debs, so you could try to apt install a doc package (bash-doc) or
> debhelper (which will fail because it depends on more packages), but maybe
> it can give you a clue where the problem lies.

The all debs are mirrored on Debian Ports.

Adrian

--
 .''`.  John Paul Adrian Glaubitz
: :' :  Debian Developer - [hidden email]
`. `'   Freie Universitaet Berlin - [hidden email]
  `-    GPG: 62FF 8A75 84E0 2956 9546  0006 7426 3B37 F5B5 F913

Reply | Threaded
Open this post in threaded view
|

Re: Centris 650 Debian 10 SID Installation

John Paul Adrian Glaubitz
In reply to this post by Finn Thain
On 6/10/19 2:11 AM, Finn Thain wrote:
>> I tried twice using the stock CD-ROM drive and once using a more modern
>> CD-ROM drive. In all three cases, installation of the basic system
>> succeeded after about four hours, then stopped at the "Configure the
>> package manager" menu -- "Your installation CD or DVD has been scanned
>> ... Scan another CD or DVD?".
>
> Maybe you need to use a different initrd? There are several on the ISO.

No, there is only one initrd used for cdrom installation. The others are
for netboot, i.e. booting kernel and initrd from a TFTP server. The netboot
initrds don't contain any CD-ROM drivers and won't work for CD-ROM installations.

> I'm no expert with the debian installer, but I suspect you should be using
> the initrd for the netinst method (?)

There is no "netinst" initrd. There are initrds for "cdrom" and "netboot".

NETINST is a normal CD-ROM installation image. It just comes with less
packages than the full CD-ROM sets which provide all packages on the
installation media.

NETINST and netboot are two completely distinct things and not to be
confused.

>> Maybe that kernel only works for installation using the accompanying
>> initrd?
>
> No, the debian kernel works for other purposes too, but you would need to
> generate a suitable initrd if you want to use it for other purposes.

Correct. The initrd for installed systems is always generated:

# update-initramfs -k $(uname -r) -c -v

>> I next tried booting using a serial console (console=ttyS0,9600n8).
>> After the expected slowness of systemd bringing everything up (about
>> eight minutes),
>
> Right, systemd is hopeless on these machines. Not just because the CPU is
> slow, but because systemd sets short timeouts on it's own units which
> those units can't live up to. That means that systemd enforces policy that
> excludes slow hardware.

I haven't seen any particular slowness issues except for the login delay
with systemd on elgar.

Adrian

--
 .''`.  John Paul Adrian Glaubitz
: :' :  Debian Developer - [hidden email]
`. `'   Freie Universitaet Berlin - [hidden email]
  `-    GPG: 62FF 8A75 84E0 2956 9546  0006 7426 3B37 F5B5 F913

Reply | Threaded
Open this post in threaded view
|

Re: Centris 650 Debian 10 SID Installation

John Paul Adrian Glaubitz
In reply to this post by userm57
On 6/10/19 3:14 AM, [hidden email] wrote:
>> Maybe you need to use a different initrd? There are several on the ISO.
>
> I see these:
> ./cdrom/initrd.gz
> ./hd-media/initrd.gz
> ./nativehd/initrd.gz
>
> I used ./cdrom/initrd.gz.

Which is the correct initrd for CD-ROM installations. The other ones are
completely untested, FWIW. I haven't looked, but I think there is also
a netboot initrd in the debian-installer-image tarball.

>> If the installation completed normally, the log would be found in the
>> target filesystem... please see
>> https://www.debian.org/releases/stable/i386/ch06s01.html.en
>
> ok, thanks.  I don't see "/var/log/installer", so apparently it wasn't
> written.

Not surprising when the installation couldn't finish.

>> Did you run "apt-get update" after editting sources.list?
>
> No, I hadn't done that, thanks.  Unfortunately, it starts updating but
> then appears to hang while running; I'll have to try it again later.

apt can't find any packages without running "apt update".

Adrian

--
 .''`.  John Paul Adrian Glaubitz
: :' :  Debian Developer - [hidden email]
`. `'   Freie Universitaet Berlin - [hidden email]
  `-    GPG: 62FF 8A75 84E0 2956 9546  0006 7426 3B37 F5B5 F913

Reply | Threaded
Open this post in threaded view
|

Re: Centris 650 Debian 10 SID Installation

John Paul Adrian Glaubitz
In reply to this post by userm57
On 6/10/19 4:58 AM, [hidden email] wrote:
> The post-installation initrd could be copied to MacOS and then specified
> in Penguin, same as the installation initrd.

Patches are welcome anytime. But currently, the installer for m68k does not have
any support for installing a working bootloader on Atari, Amiga or Mac.

> If this were a user's only GNU/Linux installation, they would need to
> know to copy the initrd at the end of the installation or their new
> system might not be bootable.

Yes, I'm fully aware of that. But Debian/m68k is a hobby project and if a feature
is missing, it means I haven't had the time yet to implement it or someone didn't
send me the patches to fix that.

> It might make sense for the developers to include a post-installation
> initrd on the CD.

No, that doesn't make sense at all as initrds are specific to the computer they
are running on. Also, an Amiga or Atari can have other drivers in its initrd
than a Mac.

> A full installation that completes may fix the problem.

Indeed. I don't expect a half-broken installation to be working properly.

Adrian

--
 .''`.  John Paul Adrian Glaubitz
: :' :  Debian Developer - [hidden email]
`. `'   Freie Universitaet Berlin - [hidden email]
  `-    GPG: 62FF 8A75 84E0 2956 9546  0006 7426 3B37 F5B5 F913

Reply | Threaded
Open this post in threaded view
|

Re: Centris 650 Debian 10 SID Installation

Finn Thain
In reply to this post by John Paul Adrian Glaubitz
On Thu, 13 Jun 2019, John Paul Adrian Glaubitz wrote:

> > I'm no expert with the debian installer, but I suspect you should be
> > using the initrd for the netinst method (?)
>
> There is no "netinst" initrd. There are initrds for "cdrom" and
> "netboot".
>
> NETINST is a normal CD-ROM installation image. It just comes with less
> packages than the full CD-ROM sets which provide all packages on the
> installation media.
>
> NETINST and netboot are two completely distinct things and not to be
> confused.
>

I see. Thanks for the clarification.

>
> >> I next tried booting using a serial console (console=ttyS0,9600n8).
> >> After the expected slowness of systemd bringing everything up (about
> >> eight minutes),
> >
> > Right, systemd is hopeless on these machines. Not just because the CPU
> > is slow, but because systemd sets short timeouts on it's own units
> > which those units can't live up to. That means that systemd enforces
> > policy that excludes slow hardware.
>
> I haven't seen any particular slowness issues except for the login delay
> with systemd on elgar.
>

The timeout problem I was referring to has come up before on m68k systems.
https://lists.debian.org/debian-68k/2016/06/msg00000.html

I don't claim that systemd is "slow". It's just that it does a lot of work
that's not needed on small systems. If it was more modular, like the
kernel, all of that extra functionality wouldn't be a burden. Instead, to
run systemd at all, I had to actually enable extra kernel modules, which
just adds to the bloat problem.
https://lists.debian.org/debian-68k/2016/02/msg00064.html

While I do think alternatives to systemd should be encouraged, I don't
expect the Debian project to provide them. Adding alternatives can only
lead to more work for all debian maintainers whose packages interact with
those alternatives. That seems like a huge amount of effort compared with
a small number of users running Debian on small systems like these.

--

> Adrian
>
>

Reply | Threaded
Open this post in threaded view
|

Re: Centris 650 Debian 10 SID Installation

Andreas Schwab-2
In reply to this post by Finn Thain
On Jun 13 2019, Finn Thain <[hidden email]> wrote:

> IMHO these log messages are a bit silly --
>
> [    **] (3 of 3) A start job is running for /dev/sda4 (1min 5s / 1min 36s)
> ...
> [ TIME ] Timed out waiting for device /dev/sda4.
> [DEPEND] Dependency failed for /dev/sda4.
> [DEPEND] Dependency failed for Swap.
> ...
>          Activating swap /dev/sda4...
> ...
> [  450.650000] Adding 524284k swap on /dev/sda4.  Priority:-2 extents:1 across:524284k FS
> ...
> [  OK  ] Activated swap /dev/sda4.

That just means that the timeout is maybe too short, but if a device is
showing up late it is handled like hotplug and works anyway.

Andreas.

--
Andreas Schwab, [hidden email]
GPG Key fingerprint = 7578 EB47 D4E5 4D69 2510  2552 DF73 E780 A9DA AEC1
"And now for something completely different."

Reply | Threaded
Open this post in threaded view
|

Re: Centris 650 Debian 10 SID Installation

Finn Thain
On Thu, 13 Jun 2019, Andreas Schwab wrote:

> On Jun 13 2019, Finn Thain <[hidden email]> wrote:
>
> > IMHO these log messages are a bit silly --
> >
> > [    **] (3 of 3) A start job is running for /dev/sda4 (1min 5s / 1min 36s)
> > ...
> > [ TIME ] Timed out waiting for device /dev/sda4.
> > [DEPEND] Dependency failed for /dev/sda4.
> > [DEPEND] Dependency failed for Swap.
> > ...
> >          Activating swap /dev/sda4...
> > ...
> > [  450.650000] Adding 524284k swap on /dev/sda4.  Priority:-2 extents:1 across:524284k FS
> > ...
> > [  OK  ] Activated swap /dev/sda4.
>
> That just means that the timeout is maybe too short,

Adding swap isn't that complicated.

If you have you swap partition on MD RAID ROCE, I'd hazard a guess that
you aren't using Linux/m68k for that. This is a use-case that systemd
should probably cater to. But making everyone pay for that complexity is
poor design IMHO.

> but if a device is showing up late it is handled like hotplug and works
> anyway.
>

The device isn't late, swapon() is late (compared to Debian 3 or Debian
4).

[  206.380000]  sda: [mac] sda1 sda2 sda3 sda4 sda5 sda6 sda7 sda8 sda9 sda10
...
[  206.790000] sd 0:0:0:0: [sda] Attached SCSI disk
...
[  243.400000] systemd[1]: systemd 241 running in system mode. (+PAM
+AUDIT +SELINUX +IMA +APPARMOR +SMACK +SYSVINIT +UTMP +LIBCRYPTSETUP
+GCRYPT +GNUTLS +ACL +XZ +LZ4 -SECCOMP +BLKID +ELFUTILS +KMOD -IDN2 +IDN
-PCRE2 default-hierarchy=hybrid)

--

> Andreas.
>

Reply | Threaded
Open this post in threaded view
|

Re: Centris 650 Debian 10 SID Installation

userm57
In reply to this post by Finn Thain
On 6/12/19 11:45 PM, Finn Thain wrote:

> On Wed, 12 Jun 2019, [hidden email] wrote:
>
>> On 6/10/19 9:32 PM, [hidden email] wrote:
>> ...
>> Here are a few observations and comments:
>>
>> 1) Approximate times were as follows:
>>    a) 30 min - from initial setup to prompting of a Debian mirror.
>>    b) 4 hrs  - from start of scan of Debian mirror to "popcon" prompt.
>>    c) 5 hrs  - from "popcon" prompt to completion of loading packages.
>>    d) 3 hrs  - from selection of "openssh server" and "system utilities"
>> to completion of installation.
>>
>
> That sure makes debootstrap look good. On fast amd64 system packages are
> unpacked a lot faster, of course. Stage 2 still happens on the target m68k
> system. The --include and --exclude options might be helpful if you're
> changing over to sysvinit or openrc.

Indeed.  Hopefully, I'll only have to install it once (well, once for
each release, or use debootstrap in the future).

>
>> ...
>> 4) After booting into a backup partition (Debian 3.1) and backing up "/"
>> and "/usr" from the new installation, and then restoring everything to a
>> single "/" filesystem, the new system booted using the kernel (and
>> initrd) that were created during the installation.
>>
>
> How did you avoid the ADB keymap problem?

I didn't.  I just used a serial console.  I'm hoping to fix that later
(see below).

> ...
> It will be interesting to see how much more quickly you can boot this
> system using openrc or sysvinit. (I see that the 450 seconds logged here
> includes the 100 second delay caused by CONFIG_CRYPTO_DH; you can avoid
> that by adding initcall_blacklist=dh_init to the kernel parameters.)
> ...

It was somewhat of a long process to switch from systemd to sysvinit.
Apparently, systemd has to be running for the switch to happen, so it's
not good enough to just boot using "init=/bin/bash" and bring up the
network and run "apt-get install sysvinit-core".

So I kept hitting systemd in the knees until it fell over and I could
log in.  The serial console always timed out, and the getty at the
framebuffer console was useless because of the ADB keymap issue.  But
eventually, after going several times back and forth between the systemd
installation and a backup Debian 3.1 installation to fsck the root
filesystem and kill off systemd parts, it eventually ran openssh server,
and I could log in and run the apt-get command from there.  It was slow,
since parts of systemd were still slugging away.

After a reboot (now into a sysvinit system), dbus seemed to hang, so I
disabled that in the scripts.  Not running
/etc/rc*.d/S01console-setup.sh and /etc/rcS.d/S03keyboard-setup.sh fixes
the ADB keymap issue.  But there's still a problem, and I'll have to
keep investigating.  Logins at the framebuffer console, serial console
and ssh all hang.  Control-C cancels the login.  On the serial console,
I see this:

-----
Debian GNU/Linux 10 calvin ttyS0

calvin login: root
Password:
Last login: Thu Jun 13 12:06:39 MDT 2019 on tty1
Linux calvin 4.19.0-5-m68k #1 Debian 4.19.37-3 (2019-05-15) m68k

The programs included with the Debian GNU/Linux system are free
software; the exact distribution terms for each program are described in
the individual files in /usr/share/doc/*/copyright.

Debian GNU/Linux comes with ABSOLUTELY NO WARRANTY, to the extent
permitted by applicable law.

-----

Reply | Threaded
Open this post in threaded view
|

Re: Centris 650 Debian 10 SID Installation

userm57
On 6/13/19 6:14 PM, [hidden email] wrote:
> ... But there's still a problem, and I'll have to
> keep investigating.  Logins at the framebuffer console, serial console
> and ssh all hang.  Control-C cancels the login.  On the serial console,
> ...

Sorry for the excessive messages. Logins appear to be fixed now, and I
can login at the serial console, the framebuffer console and via ssh.
To fix the problem (and I have no idea why this worked), I chroot'ed
into Debian 10/sid from Debian 3.1; "dpkg -l" showed these packages, or
at least their configuration files, were still installed:

systemd
libnss-systemd
libsystemd0

I could get rid of the first two, but to delete the third (systemd
utility library), I would have had to delete many packages, including
the Linux kernel.  So I left that one alone.  I also tried to delete
udev, and that failed because it also had too many dependencies
(including the Linux kernel).  At least I can use the init scripts to
keep udev from running (it's currently still running, so I may leave it
alone).

BTW, besides the memory needed for the kernel, processes are taking up a
total of 60 MB address space (VSZ) and 22 MB physical memory (RSS), as
reported by "ps aux".

-Stan

Reply | Threaded
Open this post in threaded view
|

Re: Centris 650 Debian 10 SID Installation

Finn Thain
In reply to this post by userm57
On Thu, 13 Jun 2019, [hidden email] wrote:

>
> > ... It will be interesting to see how much more quickly you can boot
> > this system using openrc or sysvinit. (I see that the 450 seconds
> > logged here includes the 100 second delay caused by CONFIG_CRYPTO_DH;
> > you can avoid that by adding initcall_blacklist=dh_init to the kernel
> > parameters.) ...
>
> It was somewhat of a long process to switch from systemd to sysvinit.
> Apparently, systemd has to be running for the switch to happen, so it's
> not good enough to just boot using "init=/bin/bash" and bring up the
> network and run "apt-get install sysvinit-core".
>
> So I kept hitting systemd in the knees until it fell over and I could
> log in.  The serial console always timed out, and the getty at the
> framebuffer console was useless because of the ADB keymap issue.  But
> eventually, after going several times back and forth between the systemd
> installation and a backup Debian 3.1 installation to fsck the root
> filesystem and kill off systemd parts, it eventually ran openssh server,
> and I could log in and run the apt-get command from there.  It was slow,
> since parts of systemd were still slugging away.
>

One way around that might be to boot with init=/bin/sh, start sshd
somehow, then exec /sbin/init.

> After a reboot (now into a sysvinit system), dbus seemed to hang, so I
> disabled that in the scripts.  Not running
> /etc/rc*.d/S01console-setup.sh and /etc/rcS.d/S03keyboard-setup.sh fixes
> the ADB keymap issue.

OK. I guess we'll need to report a bug for the console-setup package.

> But there's still a problem, and I'll have to keep investigating.  
> Logins at the framebuffer console, serial console and ssh all hang.  
> Control-C cancels the login.  On the serial console, I see this:
>
> -----
> Debian GNU/Linux 10 calvin ttyS0
>
> calvin login: root
> Password:
> Last login: Thu Jun 13 12:06:39 MDT 2019 on tty1
> Linux calvin 4.19.0-5-m68k #1 Debian 4.19.37-3 (2019-05-15) m68k
>
> The programs included with the Debian GNU/Linux system are free
> software; the exact distribution terms for each program are described in
> the individual files in /usr/share/doc/*/copyright.
>
> Debian GNU/Linux comes with ABSOLUTELY NO WARRANTY, to the extent
> permitted by applicable law.
>
> -----
>

ISTR that years ago I had to hack the bash profile scripts in /etc to
speed this up.
 
In my QEMU VM this isn't such a problem though there is a noticible delay
here.

On my QEMU system, strace shows that the login shell spends a lot of time
in /etc/profile.d/bash_completion.sh and /etc/bash_completion.d before
producing a prompt.

There are lighter alternatives such as ash, dash, mksh etc.

--

Reply | Threaded
Open this post in threaded view
|

Re: Centris 650 Debian 10 SID Installation

Andreas Schwab-2
In reply to this post by Finn Thain
On Jun 14 2019, Finn Thain <[hidden email]> wrote:

> On Thu, 13 Jun 2019, Andreas Schwab wrote:
>
>> On Jun 13 2019, Finn Thain <[hidden email]> wrote:
>>
>> > IMHO these log messages are a bit silly --
>> >
>> > [    **] (3 of 3) A start job is running for /dev/sda4 (1min 5s / 1min 36s)
>> > ...
>> > [ TIME ] Timed out waiting for device /dev/sda4.
>> > [DEPEND] Dependency failed for /dev/sda4.
>> > [DEPEND] Dependency failed for Swap.
>> > ...
>> >          Activating swap /dev/sda4...
>> > ...
>> > [  450.650000] Adding 524284k swap on /dev/sda4.  Priority:-2 extents:1 across:524284k FS
>> > ...
>> > [  OK  ] Activated swap /dev/sda4.
>>
>> That just means that the timeout is maybe too short,
>
> Adding swap isn't that complicated.

It's not about adding swap, it's about adding the device node.  You
cannot run swapon until the device node exists.

> The device isn't late, swapon() is late (compared to Debian 3 or Debian
> 4).

No, the device node shows up late.  That's a user-space operation,
carried out by udev.

Andreas.

--
Andreas Schwab, [hidden email]
GPG Key fingerprint = 7578 EB47 D4E5 4D69 2510  2552 DF73 E780 A9DA AEC1
"And now for something completely different."

Reply | Threaded
Open this post in threaded view
|

Re: Centris 650 Debian 10 SID Installation

Finn Thain
On Fri, 14 Jun 2019, Andreas Schwab wrote:

> On Jun 14 2019, Finn Thain <[hidden email]> wrote:
>
> > On Thu, 13 Jun 2019, Andreas Schwab wrote:
> >
> >> On Jun 13 2019, Finn Thain <[hidden email]> wrote:
> >>
> >> > IMHO these log messages are a bit silly --
> >> >
> >> > [    **] (3 of 3) A start job is running for /dev/sda4 (1min 5s / 1min 36s)
> >> > ...
> >> > [ TIME ] Timed out waiting for device /dev/sda4.
> >> > [DEPEND] Dependency failed for /dev/sda4.
> >> > [DEPEND] Dependency failed for Swap.
> >> > ...
> >> >          Activating swap /dev/sda4...
> >> > ...
> >> > [  450.650000] Adding 524284k swap on /dev/sda4.  Priority:-2 extents:1 across:524284k FS
> >> > ...
> >> > [  OK  ] Activated swap /dev/sda4.
> >>
> >> That just means that the timeout is maybe too short,
> >
> > Adding swap isn't that complicated.
>
> It's not about adding swap, it's about adding the device node.  You
> cannot run swapon until the device node exists.
>

Being that this is devtmpfs, the kernel creates the device node then
notifies udev, right?

> > The device isn't late, swapon() is late (compared to Debian 3 or Debian
> > 4).
>
> No, the device node shows up late.  That's a user-space operation,
> carried out by udev.
>

I guess udev (which is part of systemd) took too long to notify the
systemd unit? Or the kernel took too long to notify udev?

--

> Andreas.
>
>

Reply | Threaded
Open this post in threaded view
|

Re: Centris 650 Debian 10 SID Installation

Eero Tamminen
In reply to this post by userm57
Hi,

On 6/13/19 6:12 AM, [hidden email] wrote:
...
> 5) The system never reached multiuser mode; the startup sequence looped
> on "Starting Network Time Synchronization" (see attached console log
> "Centris_650-Debian_10.txt").  The systemd timeout for this task was 1
> min 30 sec, but it never succeeded within that limit (I let it try 10
> times before forcing a reboot).  I'll try other things, including single
> user mode and checking whether I can increase the limit somehow, or
> possibly disable network time synchronization temporarily.

You can use "systemctl mask <service-name>" to disable a service
permanently (until you unmask it).

Alternatively, one can specify longer timeout:
https://www.freedesktop.org/software/systemd/man/systemd-system.conf.html#DefaultTimeoutStartSec=


I don't think any service explicitly depends on that particular
service, but if you don't have a (working) battery backed up clock,
there are several things which won't work quite correctly, if your
files are newer that the time in your machine.


        - Eero

<rant>
Systemd journalctl is broken by design.  Although individual service
log messages have counter, so they're always sorted correct, the log
output *between* different logging clients are sorted based only on
timestamps.  Which means that one gets reliable boot service log only
if machine has reliable, battery backed clock, otherwise you get
output which can mix logs from multiple boots...
</rant>

Reply | Threaded
Open this post in threaded view
|

Re: Centris 650 Debian 10 SID Installation

userm57
Hi Eero,

Thanks, I will try changing the timeouts and disabling services.  If I
can disable enough things to make the system not too much slower than
with sysvinit, I may be able to keep systemd, at least on larger-memory
systems such as the Centris 650.

I do have a working hardware clock, with a new battery, in the 650.

-Stan

-----

On 6/14/19 11:01 AM, Eero Tamminen wrote:

> Hi,
>
> On 6/13/19 6:12 AM, [hidden email] wrote:
> ...
>> 5) The system never reached multiuser mode; the startup sequence looped
>> on "Starting Network Time Synchronization" (see attached console log
>> "Centris_650-Debian_10.txt").?? The systemd timeout for this task was 1
>> min 30 sec, but it never succeeded within that limit (I let it try 10
>> times before forcing a reboot).?? I'll try other things, including single
>> user mode and checking whether I can increase the limit somehow, or
>> possibly disable network time synchronization temporarily.
>
> You can use "systemctl mask <service-name>" to disable a service
> permanently (until you unmask it).
>
> Alternatively, one can specify longer timeout:
> https://www.freedesktop.org/software/systemd/man/systemd-system.conf.html#DefaultTimeoutStartSec=
>
>
>
> I don't think any service explicitly depends on that particular
> service, but if you don't have a (working) battery backed up clock,
> there are several things which won't work quite correctly, if your
> files are newer that the time in your machine.
>
>
> ????????- Eero
>
> <rant>
> Systemd journalctl is broken by design.?? Although individual service
> log messages have counter, so they're always sorted correct, the log
> output *between* different logging clients are sorted based only on
> timestamps.?? Which means that one gets reliable boot service log only
> if machine has reliable, battery backed clock, otherwise you get
> output which can mix logs from multiple boots...
> </rant>
>

Reply | Threaded
Open this post in threaded view
|

Re: Centris 650 Debian 10 SID Installation

Finn Thain
In reply to this post by Finn Thain
On Mon, 10 Jun 2019, Finn Thain wrote:

>
> Digging a bit deeper, we find this:
>
> /target # cat /etc/default/keyboard
> # KEYBOARD CONFIGURATION FILE
>
> # Consult the keyboard(5) manual page.
>
> XKBMODEL="macintosh_old"
> XKBLAYOUT="us"
> XKBVARIANT=""
> XKBOPTIONS=""
>
> BACKSPACE="guess"
>
>
> I found the man page here:
> https://manpages.debian.org/stretch/keyboard-configuration/keyboard.5.en.html
>
> It turned out that changing $XKBMODEL to "pc105" didn't actually help
> when I booted into /target. So I've no idea how to fix this.
>

The above config change does help when I run "setupcon" afterwards. The
serial console is needed for this. That way it is possible to finish the
installation using the framebuffer console (i.e. mirror select, popcon,
tasksel, etc.).

After rebooting into the newly installed rootfs, keyboard-setup runs again
and causes the same problem. But during startup the console keymap gets
mangled regardless of the settings in /etc/default/keyboard, which means I
have to run "setupcon" on the serial console after every restart. I was
able to work around that with "systemctl mask keyboard-setup".

--

Reply | Threaded
Open this post in threaded view
|

Re: Centris 650 Debian 10 SID Installation

Finn Thain
In reply to this post by userm57
On Wed, 12 Jun 2019, [hidden email] wrote:

>
> 2) After installation, the system was rebooted after a prompt.  A
> question mark appeared on the main screen.  Booting from an external
> disk, Mac OS volumes could be mounted manually using "Disk Utility", but
> they were still not seen after a reboot.  The problem was fixed by
> running "Apple HD SC Setup 7.3.5p" and updating the Apple driver on the
> affected disk.  So it seems likely that something in the installation
> corrupted the Apple driver.
>

I was unable to reproduce this with "parted" after installation. Same with
"partman" during installation when removing and adding the root partition
using the "manual partitioning" option. The installer may not be at fault
here. Can you send /var/log/installer/partman?

--

Reply | Threaded
Open this post in threaded view
|

Re: Centris 650 Debian 10 SID Installation

userm57
On 6/16/19 6:32 PM, Finn Thain wrote:

> On Wed, 12 Jun 2019, [hidden email] wrote:
>
>>
>> 2) After installation, the system was rebooted after a prompt.  A
>> question mark appeared on the main screen.  Booting from an external
>> disk, Mac OS volumes could be mounted manually using "Disk Utility", but
>> they were still not seen after a reboot.  The problem was fixed by
>> running "Apple HD SC Setup 7.3.5p" and updating the Apple driver on the
>> affected disk.  So it seems likely that something in the installation
>> corrupted the Apple driver.
>>
>
> I was unable to reproduce this with "parted" after installation. Same with
> "partman" during installation when removing and adding the root partition
> using the "manual partitioning" option. The installer may not be at fault
> here. Can you send /var/log/installer/partman?
>
Please see attached.  It's a large file, and I'm not sure what to look
for.  I didn't use the installer to partition the disk, only to format
an existing /dev/sda3 partition as ext3 for root and /dev/sda4 as swap
under "manual partitioning".  The Apple driver is on /dev/sda2.  Here's
the disk partitioning as reported by mac-fdisk in Debian 10:

-----
# mac-fdisk -l
Disk /dev/sda
        #  type name           length   base     ( size )  system
/dev/sda1   map Apple              63 @ 1        ( 31.5k)  Partition map
/dev/sda2  Dr43 Macintosh          32 @ 64       ( 16.0k)  Driver 4.3
/dev/sda3  unix root         12582912 @ 96       (  6.0G)  Linux native
/dev/sda4  unix swap          1048576 @ 12583008 (512.0M)  Linux swap
/dev/sda5  unix test             8192 @ 13631584 (  4.0M)  Linux native
/dev/sda6  Free Extra         5234593 @ 13639776 (  2.5G)  Free space
/dev/sda7  unix data          4427887 @ 18874369 (  2.1G)  Linux native
/dev/sda8  unix root_backup   4194304 @ 23302256 (  2.0G)  Linux native
/dev/sda9   HFS MacOS         4194304 @ 27496560 (  2.0G)  HFS
/dev/sda10  HFS Data          4194304 @ 31690864 (  2.0G)  HFS

Block size=512, Number of Blocks=35885168
-----

This looks suspicious -- 2147483648 looks like something hit a 2 GB
limit.  Also, the sizes are wrong, and there's no usr partition, so
maybe these are defaults of some sort?

parted_server: OUT: 1   512-32767       32256   primary unknown
/dev/sda1       Apple
parted_server: OUT: 2   32768-49151     16384   primary unknown
/dev/sda2       Macintosh
parted_server: OUT: 3   49152-2147532799        2147483648      primary
sun-ufs /dev/sda3       root
parted_server: OUT: 4   2147532800-2684403711   536870912       primary
unknown /dev/sda4       swap
parted_server: OUT: 5   2684403712-6979371007   4294967296      primary
unknown /dev/sda5       usr
parted_server: OUT: -1  6979371008-9663676927   2684305920      primary
free    /dev/sda-1
parted_server: OUT: 7   9663676928-11930755071  2267078144      primary
ext3    /dev/sda7       data

partman.xz (8K) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: Centris 650 Debian 10 SID Installation

Laurent Vivier-5
In reply to this post by Finn Thain
Le 17/06/2019 à 02:32, Finn Thain a écrit :

> On Wed, 12 Jun 2019, [hidden email] wrote:
>
>>
>> 2) After installation, the system was rebooted after a prompt.  A
>> question mark appeared on the main screen.  Booting from an external
>> disk, Mac OS volumes could be mounted manually using "Disk Utility", but
>> they were still not seen after a reboot.  The problem was fixed by
>> running "Apple HD SC Setup 7.3.5p" and updating the Apple driver on the
>> affected disk.  So it seems likely that something in the installation
>> corrupted the Apple driver.
>>
>
> I was unable to reproduce this with "parted" after installation. Same with
> "partman" during installation when removing and adding the root partition
> using the "manual partitioning" option. The installer may not be at fault
> here. Can you send /var/log/installer/partman?
>

There is a bug in parted that corrupts the AppleDriver partition.

It is fixed in git repo now by:

commit 43b061e90dcdab799ecd1e822852de110673bf7e
Author: Laurent Vivier <[hidden email]>
Date:   Fri Dec 9 15:10:53 2016 +0100

    libparted: Fix MacOS boot support
   
    boot_region_length (or BootSize in the MacOS dialect) is the length
    of the driver code in the driver partition. This length is used
    to compute the checksum of the driver.
   
    libparted updates this value by setting the whole size of the partition
    without computing the checksum of the driver using this size.
   
    As the checksum is wrong, the driver is broken and cannot be loaded
    by the MacOS ROM, and thus the disk is not bootable anymore.
   
    Moreover, parted try to update the driver list and makes it disappear.
   
    As parted is not able to insert a driver in a partition,
    the driver is generally inserted by the Apple HD Tool,
    this patch removes the line updating the driver size.
   
    We also simplify the driver list scan and fix endianess use.
   
    This has been tested... and it works, now.
   
    I have updated a bootable disk with parted on x86_64 and
    then been able to boot it (again) on a Mac LC III.
   
    Signed-off-by: Laurent Vivier <[hidden email]>
    Signed-off-by: Brian C. Lane <[hidden email]>

12345