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 Tue, 18 Jun 2019, Laurent Vivier wrote:

> Le 17/06/2019 ? 02:32, Finn Thain a ?crit?:
> > 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]>
>

Thanks for the tip. I was able to confirm that Debian's libparted does not
carry this fix (as of version 3.2-25).

--

Reply | Threaded
Open this post in threaded view
|

Re: Centris 650 Debian 10 SID Installation

John Paul Adrian Glaubitz
On 6/19/19 10:05 AM, Finn Thain wrote:
>> It is fixed in git repo now by:
>>
>> commit 43b061e90dcdab799ecd1e822852de110673bf7e
>> Author: Laurent Vivier <[hidden email]>
>> Date:   Fri Dec 9 15:10:53 2016 +0100
>> (...)
> Thanks for the tip. I was able to confirm that Debian's libparted does not
> carry this fix (as of version 3.2-25).

Ah, I just realized that parted hasn't had a release for 5 years. Someone
needs to file a bug against src:parted then and ask for the patch to be
included.

Or convince upstream to make a new release.

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 userm57
On Tue, 18 Jun 2019, [hidden email] wrote:

> 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.

Thanks.

> 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
> ...

By the end of the log, the partition type seems to have been changed from
sun-ufs to ext3 but the size is still 2GB:

parted_server: OUT: 3   49152-2147532799        2147483648      primary ext3    /dev/sda3       root
parted_server: OUT: 4   2147532800-2684403711   536870912       primary linux-swap      /dev/sda4       swap
parted_server: OUT: 5   2684403712-6979371007   4294967296      primary ext3    /dev/sda5       usr

Did parted initialize a 6 GB filesystem on the sda3 partition, or did you
end up with a 2 GiB filesystem at the start of a 6 GB partition?

--

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 10:14 AM, Finn Thain wrote:
> (...)
> 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.

This is not expected. I'll have to look into this.

However, you can just install using a serial console which will work:

qemu-system-m68k -M q800 -serial none -serial mon:stdio -m 1000M -drive file=m68k-system.img,format=qcow2 -drive file=m68k-swap.img,format=qcow2 -net nic,model=dp83932,addr=08:00:07:12:34:56 -net user -nographic -append "root=/dev/sda2 rw console=ttyS0" -kernel vmlinux-4.19.0-5-m68k -initrd initrd.img-4.19.0-5-m68k

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 10:28 AM, Finn Thain wrote:
> Aranym also stalls here (for about 20 seconds). Adding initcall_debug to
> the kernel parameters reveals this:
>
> [    0.620000] calling  dh_init+0x0/0x10 @ 1
> [    2.560000] random: fast init done
> [   24.030000] initcall dh_init+0x0/0x10 returned 0 after 22861089 usecs
>
> The reason why my kernel builds don't pause here is because they don't
> call dh_init, because I have disabled CONFIG_CRYPTO_DH.

If you have a suggestion for a configuration change, please file a bug against
src:linux with [hidden email] CC'ed so that I am notified about
this.

I don't claim that the kernel configuration in Debian for m68k is perfect, but
in order to be able to sort these issues out, people should avoid building their
own custom kernels and help improving the Debian kernel instead.

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: Aranym installation, was Re: Centris 650 Debian 10 SID Installation

John Paul Adrian Glaubitz
In reply to this post by Finn Thain
On 6/11/19 3:12 AM, Finn Thain wrote:
>
> Using Aranym, I booted the kernel binary from the 2019-05-24 ISO with
> these parameters:
> debian-installer/framebuffer=false initcall_blacklist=dh_init
>
> I found that the arrow keys stopped working after I selected the American
> English keymap (this step can be skipped). It turns out that ctrl-P and
> ctrl-N are substitutes for up-arrow and down-arrow keys.

I'm not aware of this issue and haven't encountered it last time I tried.

> I also found that the old installer bug which prevents CD-ROM drive
> detection on Atari is still there. I had to spawn a shell to run these
> commands:
>
> # modprobe pata-falcon
> # modprobe isofs
> # mount /dev/sr0 /cdrom
>
> Then the "Detect and mount CD-ROM" step then worked. (You can use
> falconide and /dev/hdX instead.)

Yes, this is known and I currently don't know how to fix this. Again, if someone
knows how to make the installer load the modules, please let me know. I assume
the problem is that there is no autodetection here so udev can't load the
modules automatically.

> I had to select "no network card" after network detection failed. There's
> probably some issue with my Aranym config; I never resolved this.

Yes, it's an Aranym configuration issue. But I keep on forgetting how to fix this.

> With those workarounds I was able to "install the base system".
>
> As with the Mac installation, the console-setup package changed the
> framebuffer console font and mangled the window borders. You can see the
> terminus font problem in this screenshot:
> https://lists.debian.org/debian-68k/2019/06/msg00019.html

Again, use a serial console. That's much easier.

> The next steps in the installation ("configure the package manager",
> "select and install software") seemingly aren't possible without a network
> connection, so I skipped to "proceed without bootloader" which completed
> successfully.

Yes. This step sets up the download repositories. This needs a network
connection, of course.

> I was able to use the installed initrd and kernel to boot the new system
> once I figured out that root=/dev/nfhdXpY. But that isn't going to work
> outside of Aranym.
>
> The system should be bootable using root=/dev/sdXY but this doesn't work
> because the initrd fails to load the modules (pata-falcon, atari_scsi
> etc.)

You will need to add the drivers manually to /etc/modules.

> I found that the installer didn't list any modules in
> /etc/initramfs-tools/modules.

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 6/11/19 3:26 AM, [hidden email] wrote:
> I originally tried to use the serial console for the installation, but
> it didn't seem to work (maybe I didn't give it enough time).  I'll try
> again.

Serial console should work fine. You just need to make sure to pass the
proper parameters on either side of the cable.

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
Hello!

On 6/13/19 5:12 AM, [hidden email] wrote:

> I was able to install a text-only installation on the Centris 650 using
> the serial console ("console=ttyS0,9600n8").  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.

Nothing unusual on an 68040 machine clocked at 25 MHz.

> 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.

Yes. As mentioned in another part of the thread, Debian's parted package
needs to be updated to include a patch which fixes this problem. It's
a bug in parted and will eventually be fixed in Debian.

> 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).

Separate /usr hasn't been fully supported throughout the Linux world for
a very long time.

See this explanation, for example: https://freedesktop.org/wiki/Software/systemd/separate-usr-is-broken/

This is also not something I am going to work on because it's a fundamental
decision the Linux world has made years ago and changing this would mean
having to fight a futile fight.

Please note that the Debian packages and debian-installer are not customized
for Debian/m68k. Everything you get is stock Debian and thus anything that is
not supported in stock Debian is also not supported by Debian/m68k. This
includes such invasive changes such as a separate /usr.

> 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.
>
> 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 need to pass "emergency", see: https://freedesktop.org/wiki/Software/systemd/Debugging/

Please note that a lot of BSD and Unix knowledge from the 90s no longer
applies to modern systems running systemd, so there is no init runlevel
for single-user mode (it's called emergency target).

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 6/13/19 12:13 PM, Finn Thain wrote:
> 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

You can mask services you don't need. It's just that Debian's systemd is
using a fully fledged setup by default given the fact it's more designed
for larger systems. We don't have a custom systemd package for m68k
as that would mean we would get out of sync with the rest of Debian.

But again, just turn off what you don't need and systemd is fast.

> 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.

You don't need alternatives to systemd. Just configure systemd towards your
needs.

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
|

Atari Falcon platform drivers (was: Re: Aranym installation, was Re: Centris 650 Debian 10 SID Installation)

Geert Uytterhoeven
In reply to this post by John Paul Adrian Glaubitz
Hi Adrian,

On Wed, Jun 19, 2019 at 10:30 AM John Paul Adrian Glaubitz
<[hidden email]> wrote:

> On 6/11/19 3:12 AM, Finn Thain wrote:
> > I also found that the old installer bug which prevents CD-ROM drive
> > detection on Atari is still there. I had to spawn a shell to run these
> > commands:
> >
> > # modprobe pata-falcon
> > # modprobe isofs
> > # mount /dev/sr0 /cdrom
> >
> > Then the "Detect and mount CD-ROM" step then worked. (You can use
> > falconide and /dev/hdX instead.)
>
> Yes, this is known and I currently don't know how to fix this. Again, if someone
> knows how to make the installer load the modules, please let me know. I assume
> the problem is that there is no autodetection here so udev can't load the
> modules automatically.

drivers/ide/falconide.c and drivers/ata/pata_falcon.c should be converted
to proper platform drivers, matching against a platform device registered
in arch/m68k/atari/config.c.

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [hidden email]

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds

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 6/19/19 2:22 AM, Finn Thain wrote:

> On Tue, 18 Jun 2019, [hidden email] wrote:
>
>> On 6/16/19 6:32 PM, Finn Thain wrote:
>>> On Wed, 12 Jun 2019, [hidden email] wrote:
>>>
>>>>
>>>> ... 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
>> ...
>
> By the end of the log, the partition type seems to have been changed from
> sun-ufs to ext3 but the size is still 2GB:
>
> parted_server: OUT: 3   49152-2147532799        2147483648      primary ext3    /dev/sda3       root
> parted_server: OUT: 4   2147532800-2684403711   536870912       primary linux-swap      /dev/sda4       swap
> parted_server: OUT: 5   2684403712-6979371007   4294967296      primary ext3    /dev/sda5       usr
>
> Did parted initialize a 6 GB filesystem on the sda3 partition, or did you
> end up with a 2 GiB filesystem at the start of a 6 GB partition?
>

As far as I can tell, and as reported by "df", the filesystem size is 6
GB.  Large filesystems (> 2GB) have been supported for a long time, so
maybe it's a printing error that's only in the log, such as might be
caused by printing an unsigned int or a long int as an int?

Reply | Threaded
Open this post in threaded view
|

Re: Centris 650 Debian 10 SID Installation

Finn Thain
On Wed, 19 Jun 2019, [hidden email] wrote:

> On 6/19/19 2:22 AM, Finn Thain wrote:
> > On Tue, 18 Jun 2019, [hidden email] wrote:
> >
> >> On 6/16/19 6:32 PM, Finn Thain wrote:
> >>> On Wed, 12 Jun 2019, [hidden email] wrote:
> >>>
> >>>>
> >>>> ... 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
> >> ...
> >
> > By the end of the log, the partition type seems to have been changed from
> > sun-ufs to ext3 but the size is still 2GB:
> >
> > parted_server: OUT: 3   49152-2147532799        2147483648      primary ext3    /dev/sda3       root
> > parted_server: OUT: 4   2147532800-2684403711   536870912       primary linux-swap      /dev/sda4       swap
> > parted_server: OUT: 5   2684403712-6979371007   4294967296      primary ext3    /dev/sda5       usr
> >
> > Did parted initialize a 6 GB filesystem on the sda3 partition, or did you
> > end up with a 2 GiB filesystem at the start of a 6 GB partition?
> >
>
> As far as I can tell, and as reported by "df", the filesystem size is 6
> GB.

Must be a bug.

> Large filesystems (> 2GB) have been supported for a long time,

I believe that Sun UFS is limited to 2 TB. Assuming that this really was
a Sun UFS filesystem. Was it created by NetBSD perhaps?

> so maybe it's a printing error that's only in the log, such as might be
> caused by printing an unsigned int or a long int as an int?
>

I don't see how you get 2 GB from 6 GB with just a formatting bug...

--

Reply | Threaded
Open this post in threaded view
|

Re: Centris 650 Debian 10 SID Installation

userm57
On 6/19/19 5:09 PM, Finn Thain wrote:

> On Wed, 19 Jun 2019, [hidden email] wrote:
>
>> On 6/19/19 2:22 AM, Finn Thain wrote:
>>> On Tue, 18 Jun 2019, [hidden email] wrote:
>>>
>>>> On 6/16/19 6:32 PM, Finn Thain wrote:
>>>>> On Wed, 12 Jun 2019, [hidden email] wrote:
>>>>>
>>>>>>
>>>>>> ... 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
>>>> ...
>>>
>>> By the end of the log, the partition type seems to have been changed from
>>> sun-ufs to ext3 but the size is still 2GB:
>>>
>>> parted_server: OUT: 3   49152-2147532799        2147483648      primary ext3    /dev/sda3       root
>>> parted_server: OUT: 4   2147532800-2684403711   536870912       primary linux-swap      /dev/sda4       swap
>>> parted_server: OUT: 5   2684403712-6979371007   4294967296      primary ext3    /dev/sda5       usr
>>>
>>> Did parted initialize a 6 GB filesystem on the sda3 partition, or did you
>>> end up with a 2 GiB filesystem at the start of a 6 GB partition?
>>>
>>
>> As far as I can tell, and as reported by "df", the filesystem size is 6
>> GB.
>
> Must be a bug.
>
>> Large filesystems (> 2GB) have been supported for a long time,
>
> I believe that Sun UFS is limited to 2 TB. Assuming that this really was
> a Sun UFS filesystem. Was it created by NetBSD perhaps?

Yes, I used to have NetBSD installed in the first 1024 MB of the disk,
but I had re-partitioned the disk before running the Debian installer.
Maybe it was looking at the partition so it could warn me, for example,
that I was about to overwrite a partition with a NetBSD signature?  On
mac68k, NetBSD uses a partition type of Apple_UNIX_SVR2.

>
>> so maybe it's a printing error that's only in the log, such as might be
>> caused by printing an unsigned int or a long int as an int?
>>
>
> I don't see how you get 2 GB from 6 GB with just a formatting bug...
>

I don't either, especially since it ended up creating filesystems
correctly (except for trashing the Apple Driver partition).

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 Wed, 19 Jun 2019, John Paul Adrian Glaubitz wrote:

> On 6/10/19 10:28 AM, Finn Thain wrote:
> > Aranym also stalls here (for about 20 seconds). Adding initcall_debug
> > to the kernel parameters reveals this:
> >
> > [ 0.620000] calling dh_init+0x0/0x10 @ 1
> > [ 2.560000] random: fast init done
> > [ 24.030000] initcall dh_init+0x0/0x10 returned 0 after 22861089 usecs
> >
> > The reason why my kernel builds don't pause here is because they don't
> > call dh_init, because I have disabled CONFIG_CRYPTO_DH.
>
> If you have a suggestion for a configuration change, please file a bug
> against src:linux with [hidden email] CC'ed so that I am
> notified about this.
>

I would not want to become a Debian/m68k porter, even if there was room
for another one.

Besides that, on a modern host, the emulators run Debian/m68k quite well
despite CONFIG_CRYPTO_DH (console bugs and kernel module bugs
notwithstanding).

> I don't claim that the kernel configuration in Debian for m68k is
> perfect, but in order to be able to sort these issues out, people should
> avoid building their own custom kernels and help improving the Debian
> kernel instead.
>

I certainly would not assert that "people should" do anything. However, I
do encourage all Linux/m68k users to learn to cross-compile small kernels
and other packages. That's because real m68k hardware has limitations --
no FPU, limited RAM etc. -- which Debian doesn't cater to.

--

> Adrian
>

Reply | Threaded
Open this post in threaded view
|

Re: Aranym installation, was Re: Centris 650 Debian 10 SID Installation

Finn Thain
In reply to this post by John Paul Adrian Glaubitz
On Wed, 19 Jun 2019, John Paul Adrian Glaubitz wrote:

> >
> > As with the Mac installation, the console-setup package changed the
> > framebuffer console font and mangled the window borders. You can see
> > the terminus font problem in this screenshot:
> > https://lists.debian.org/debian-68k/2019/06/msg00019.html
>
> Again, use a serial console. That's much easier.
>

Again, this was Aranym.

--

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 Wed, 19 Jun 2019, John Paul Adrian Glaubitz wrote:

> On 6/13/19 12:13 PM, Finn Thain wrote:
> > 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
>
> You can mask services you don't need. It's just that Debian's systemd is
> using a fully fledged setup by default given the fact it's more designed
> for larger systems. We don't have a custom systemd package for m68k as
> that would mean we would get out of sync with the rest of Debian.
>
> But again, just turn off what you don't need and systemd is fast.
>
> > 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.
>
> You don't need alternatives to systemd. Just configure systemd towards your
> needs.
>

I fondly remember the good old days when the HTTPS daemon depended on
init, not the other way around...

https://buildd.debian.org/status/fetch.php?pkg=systemd&arch=m68k&ver=241-5&stamp=1558758041&raw=0

--

> Adrian
>
>

Reply | Threaded
Open this post in threaded view
|

Re: Atari Falcon platform drivers

Michael Schmitz-4
In reply to this post by Geert Uytterhoeven
Hi Geert,

Am 19.06.2019 um 20:53 schrieb Geert Uytterhoeven:

> Hi Adrian,
>
> On Wed, Jun 19, 2019 at 10:30 AM John Paul Adrian Glaubitz
> <[hidden email]> wrote:
>> On 6/11/19 3:12 AM, Finn Thain wrote:
>>> I also found that the old installer bug which prevents CD-ROM drive
>>> detection on Atari is still there. I had to spawn a shell to run these
>>> commands:
>>>
>>> # modprobe pata-falcon
>>> # modprobe isofs
>>> # mount /dev/sr0 /cdrom
>>>
>>> Then the "Detect and mount CD-ROM" step then worked. (You can use
>>> falconide and /dev/hdX instead.)
>>
>> Yes, this is known and I currently don't know how to fix this. Again, if someone
>> knows how to make the installer load the modules, please let me know. I assume
>> the problem is that there is no autodetection here so udev can't load the
>> modules automatically.
>
> drivers/ide/falconide.c and drivers/ata/pata_falcon.c should be converted
> to proper platform drivers, matching against a platform device registered
> in arch/m68k/atari/config.c.

And that's going to help with autoloading modules?

In this particular instance, both drivers would need to bind to the same
platform device - how do we handle this case?

pata_falcon.c is probably easiest to convert, and should be sufficient
to test with the installer. falconide.c is going away in two years
anyway, and I can't test these changes on my hardware, so I'd rather
leave this one alone.

Cheers,

        Michael

>
> Gr{oetje,eeting}s,
>
>                         Geert
>

Reply | Threaded
Open this post in threaded view
|

Re: Atari Falcon platform drivers

Geert Uytterhoeven
Hi Michael,

On Thu, Jun 20, 2019 at 5:58 AM Michael Schmitz <[hidden email]> wrote:

> Am 19.06.2019 um 20:53 schrieb Geert Uytterhoeven:
> > On Wed, Jun 19, 2019 at 10:30 AM John Paul Adrian Glaubitz
> > <[hidden email]> wrote:
> >> On 6/11/19 3:12 AM, Finn Thain wrote:
> >>> I also found that the old installer bug which prevents CD-ROM drive
> >>> detection on Atari is still there. I had to spawn a shell to run these
> >>> commands:
> >>>
> >>> # modprobe pata-falcon
> >>> # modprobe isofs
> >>> # mount /dev/sr0 /cdrom
> >>>
> >>> Then the "Detect and mount CD-ROM" step then worked. (You can use
> >>> falconide and /dev/hdX instead.)
> >>
> >> Yes, this is known and I currently don't know how to fix this. Again, if someone
> >> knows how to make the installer load the modules, please let me know. I assume
> >> the problem is that there is no autodetection here so udev can't load the
> >> modules automatically.
> >
> > drivers/ide/falconide.c and drivers/ata/pata_falcon.c should be converted
> > to proper platform drivers, matching against a platform device registered
> > in arch/m68k/atari/config.c.
>
> And that's going to help with autoloading modules?

Yes, platform drivers have automatic module aliases.
Compare e.g.

    $ modinfo drivers/ata/pata_falcon.ko
    filename:       drivers/ata/pata_falcon.ko
    version:        0.1.0
    license:        GPL v2
    description:    low-level driver for Atari Falcon PATA
    author:         Bartlomiej Zolnierkiewicz
    srcversion:     84A5F6FDA8A93A0BC6F8B9E
    depends:        libata
    intree:         Y
    name:           pata_falcon
    vermagic:       5.2.0-rc5+ mod_unload modversions
    $ modinfo drivers/ata/pata_gayle.ko
    filename:       drivers/ata/pata_gayle.ko
    version:        0.1.0
    alias:          platform:amiga-gayle-ide
    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    license:        GPL v2
    description:    low-level driver for Amiga Gayle PATA
    author:         Bartlomiej Zolnierkiewicz
    srcversion:     DC5461633EF515ACD5AB512
    depends:        libata
    intree:         Y
    name:           pata_gayle
    vermagic:       5.2.0-rc5+ mod_unload modversions

> In this particular instance, both drivers would need to bind to the same
> platform device - how do we handle this case?

That's easy: the first driver bound wins ;-)

> pata_falcon.c is probably easiest to convert, and should be sufficient
> to test with the installer. falconide.c is going away in two years
> anyway, and I can't test these changes on my hardware, so I'd rather
> leave this one alone.

pata_falcon.c alrwady creates its own platform device inside the driver,
so that's indeed the easiest to convert.

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [hidden email]

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds

Reply | Threaded
Open this post in threaded view
|

Re: Atari Falcon platform drivers

Michael Schmitz-4
Hi Geert,

Am 20.06.2019 um 19:19 schrieb Geert Uytterhoeven:

>>> drivers/ide/falconide.c and drivers/ata/pata_falcon.c should be converted
>>> to proper platform drivers, matching against a platform device registered
>>> in arch/m68k/atari/config.c.
>>
>> And that's going to help with autoloading modules?
>
> Yes, platform drivers have automatic module aliases.
> Compare e.g.
>
>     $ modinfo drivers/ata/pata_falcon.ko
>     filename:       drivers/ata/pata_falcon.ko
>     version:        0.1.0
>     license:        GPL v2
>     description:    low-level driver for Atari Falcon PATA
>     author:         Bartlomiej Zolnierkiewicz
>     srcversion:     84A5F6FDA8A93A0BC6F8B9E
>     depends:        libata
>     intree:         Y
>     name:           pata_falcon
>     vermagic:       5.2.0-rc5+ mod_unload modversions
>     $ modinfo drivers/ata/pata_gayle.ko
>     filename:       drivers/ata/pata_gayle.ko
>     version:        0.1.0
>     alias:          platform:amiga-gayle-ide
>     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>     license:        GPL v2
>     description:    low-level driver for Amiga Gayle PATA
>     author:         Bartlomiej Zolnierkiewicz
>     srcversion:     DC5461633EF515ACD5AB512
>     depends:        libata
>     intree:         Y
>     name:           pata_gayle
>     vermagic:       5.2.0-rc5+ mod_unload modversions

Thanks - bit hard to test this without ginning up a modified initrd...

>> In this particular instance, both drivers would need to bind to the same
>> platform device - how do we handle this case?
>
> That's easy: the first driver bound wins ;-)

Both need to reserve the IDE register region, so the second one would
fail anyway.

Just be aware that pata_falcon does poll for IDE command completion
instead of using the interrupt (to avoid the whole issue around
exclusive access to the ST-DMA registers). Didn't seem to have a
catastrophic impact on performance, but less than ideal IMO.

>> pata_falcon.c is probably easiest to convert, and should be sufficient
>> to test with the installer. falconide.c is going away in two years
>> anyway, and I can't test these changes on my hardware, so I'd rather
>> leave this one alone.
>
> pata_falcon.c alrwady creates its own platform device inside the driver,
> so that's indeed the easiest to convert.

Yep - I'll test Falcon IDE platform device creation and pata_falcon
builtin ASAP but won't be able to test pata_falcon as a module. Maybe
someone else can eventually do that.

Cheers,

        Michael


> Gr{oetje,eeting}s,
>
>                         Geert
>

Reply | Threaded
Open this post in threaded view
|

Re: Aranym installation, was Re: Centris 650 Debian 10 SID Installation

John Paul Adrian Glaubitz
In reply to this post by Finn Thain
On 6/20/19 2:28 AM, Finn Thain wrote:
>>> As with the Mac installation, the console-setup package changed the
>>> framebuffer console font and mangled the window borders. You can see
>>> the terminus font problem in this screenshot:
>>> https://lists.debian.org/debian-68k/2019/06/msg00019.html
>>
>> Again, use a serial console. That's much easier.
>>
>
> Again, this was Aranym.

Well, this thread is growing too fast. You need to give me a chance to answer.

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

12345