Discussion:
ppc64el Trixie Alpha 1 /boot may be ext2 not ext4
Add Reply
o***@tutanota.com
2025-01-12 10:20:01 UTC
Reply
Permalink
Hello,
installing Trixie Alpha 1 on Tyan TN71-BP012 (POWER8)
(/images/trixie_di_alpha1/ppc64el 2024/12/31)

Installation goes well but installed system isn't bootable with Petitboot.
Installation was made choosing automated partitioning for an entrier disk with LVM and luks.
Doing that unencrypted /boot partition is ext4 and Petitboot can't see it.
When partitioning if I edit /boot partition and set it to ext2 then after finishing installation,
PetitBoot see it and I'm able to boot in the installed system.

Have a nice day.
Olivier
Cyril Brulebois
2025-01-12 10:30:02 UTC
Reply
Permalink
Hi,
Post by o***@tutanota.com
Hello,
installing Trixie Alpha 1 on Tyan TN71-BP012 (POWER8)
(/images/trixie_di_alpha1/ppc64el 2024/12/31)
Installation goes well but installed system isn't bootable with Petitboot.
Installation was made choosing automated partitioning for an entrier disk with LVM and luks.
Doing that unencrypted /boot partition is ext4 and Petitboot can't see it.
When partitioning if I edit /boot partition and set it to ext2 then after finishing installation,
PetitBoot see it and I'm able to boot in the installed system.
Sorry to hear about that. Looping in debian-boot@ who's responsible for
the installer components that implement those choices.


Cheers,
--
Cyril Brulebois (***@debian.org) <https://debamax.com/>
D-I release manager -- Release team member -- Freelance Consultant
Holger Wansing
2025-01-12 11:10:01 UTC
Reply
Permalink
Hi Cyril,
Post by Cyril Brulebois
Post by o***@tutanota.com
Hello,
installing Trixie Alpha 1 on Tyan TN71-BP012 (POWER8)
(/images/trixie_di_alpha1/ppc64el 2024/12/31)
Installation goes well but installed system isn't bootable with Petitboot.
Installation was made choosing automated partitioning for an entrier disk with LVM and luks.
Doing that unencrypted /boot partition is ext4 and Petitboot can't see it.
When partitioning if I edit /boot partition and set it to ext2 then after finishing installation,
PetitBoot see it and I'm able to boot in the installed system.
the installer components that implement those choices.
thanks for pointing me on this.
I have just pushed a fix for this, reverting back to ext2:
https://salsa.debian.org/installer-team/partman-auto/-/commit/a87cccbe0198b4dea2cc46a39d5f5b99501aa095


Regards
Holger
--
Holger Wansing <***@mailbox.org>
PGP-Fingerprint: 496A C6E8 1442 4B34 8508 3529 59F1 87CA 156E B076
Pascal Hambourg
2025-01-12 13:20:02 UTC
Reply
Permalink
Hello,
Post by Holger Wansing
Post by Cyril Brulebois
Post by o***@tutanota.com
Hello,
installing Trixie Alpha 1 on Tyan TN71-BP012 (POWER8)
(/images/trixie_di_alpha1/ppc64el 2024/12/31)
Installation goes well but installed system isn't bootable with Petitboot.
Installation was made choosing automated partitioning for an entrier disk with LVM and luks.
Doing that unencrypted /boot partition is ext4 and Petitboot can't see it.
When partitioning if I edit /boot partition and set it to ext2 then after finishing installation,
PetitBoot see it and I'm able to boot in the installed system.
the installer components that implement those choices.
The move from ext2 to ext4 for /boot was in response to bug #985463
<https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=985463> after it was
considered that bootloaders such as GRUB and U-Boot currently used with
Debian supported ext4.

Inndeed it is unfortunate that a modern boot loader like Petitboot does
not support ext4. It is even unexpected from a boot loader based on a
Linux kernel, as mentioned in
<https://open-power.github.io/petitboot/overview.html>.
Post by Holger Wansing
thanks for pointing me on this.
https://salsa.debian.org/installer-team/partman-auto/-/commit/a87cccbe0198b4dea2cc46a39d5f5b99501aa095
Partitioning recipes for ppc64el create a separate /boot partition only
when using LVM (and this was never changed). When using plain
partitions, /boot is on the ext4 root partition so the above fix will
have no effect.

According to
<https://github.com/open-power/petitboot/blob/master/README.md> and
<https://open-power.github.io/petitboot/platforms.html>, Petitboot is
also available for arm64 platforms. Partitioning recipes for arm64 also
use ext4 for /boot, so it would not work either.
Holger Wansing
2025-01-12 14:10:02 UTC
Reply
Permalink
Hi,
Post by o***@tutanota.com
Hello,
Post by Holger Wansing
Post by Cyril Brulebois
Post by o***@tutanota.com
Hello,
installing Trixie Alpha 1 on Tyan TN71-BP012 (POWER8)
(/images/trixie_di_alpha1/ppc64el 2024/12/31)
Installation goes well but installed system isn't bootable with Petitboot.
Installation was made choosing automated partitioning for an entrier disk with LVM and luks.
Doing that unencrypted /boot partition is ext4 and Petitboot can't see it.
When partitioning if I edit /boot partition and set it to ext2 then after finishing installation,
PetitBoot see it and I'm able to boot in the installed system.
the installer components that implement those choices.
The move from ext2 to ext4 for /boot was in response to bug #985463
<https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=985463> after it was
considered that bootloaders such as GRUB and U-Boot currently used with
Debian supported ext4.
Inndeed it is unfortunate that a modern boot loader like Petitboot does
not support ext4. It is even unexpected from a boot loader based on a
Linux kernel, as mentioned in
<https://open-power.github.io/petitboot/overview.html>.
Post by Holger Wansing
thanks for pointing me on this.
https://salsa.debian.org/installer-team/partman-auto/-/commit/a87cccbe0198b4dea2cc46a39d5f5b99501aa095
Partitioning recipes for ppc64el create a separate /boot partition only
when using LVM (and this was never changed). When using plain
partitions, /boot is on the ext4 root partition so the above fix will
have no effect.
The report was for guided partitioning with encrypted LVM.
So I assumed the above would fit.
Am I missing something?
Post by o***@tutanota.com
According to
<https://github.com/open-power/petitboot/blob/master/README.md> and
<https://open-power.github.io/petitboot/platforms.html>, Petitboot is
also available for arm64 platforms. Partitioning recipes for arm64 also
use ext4 for /boot, so it would not work either.
Hmm, is it widely used on arm64, or only a cornercase?
Is it worth to change the default on arm64 just for petitboot?


Holger
--
Holger Wansing <***@mailbox.org>
PGP-Fingerprint: 496A C6E8 1442 4B34 8508 3529 59F1 87CA 156E B076
Pascal Hambourg
2025-01-12 15:20:01 UTC
Reply
Permalink
Post by Holger Wansing
Post by Pascal Hambourg
Post by Holger Wansing
Post by Cyril Brulebois
Post by o***@tutanota.com
Hello,
installing Trixie Alpha 1 on Tyan TN71-BP012 (POWER8)
(/images/trixie_di_alpha1/ppc64el 2024/12/31)
Installation goes well but installed system isn't bootable with Petitboot.
Installation was made choosing automated partitioning for an entrier disk with LVM and luks.
Doing that unencrypted /boot partition is ext4 and Petitboot can't see it.
When partitioning if I edit /boot partition and set it to ext2 then after finishing installation,
PetitBoot see it and I'm able to boot in the installed system.
the installer components that implement those choices.
The move from ext2 to ext4 for /boot was in response to bug #985463
<https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=985463> after it was
considered that bootloaders such as GRUB and U-Boot currently used with
Debian supported ext4.
Inndeed it is unfortunate that a modern boot loader like Petitboot does
not support ext4. It is even unexpected from a boot loader based on a
Linux kernel, as mentioned in
<https://open-power.github.io/petitboot/overview.html>.
Post by Holger Wansing
thanks for pointing me on this.
https://salsa.debian.org/installer-team/partman-auto/-/commit/a87cccbe0198b4dea2cc46a39d5f5b99501aa095
Partitioning recipes for ppc64el create a separate /boot partition only
when using LVM (and this was never changed). When using plain
partitions, /boot is on the ext4 root partition so the above fix will
have no effect.
The report was for guided partitioning with encrypted LVM.
So I assumed the above would fit.
Am I missing something?
No, I just wanted to make sure everyone is aware that the revert fixes
only this specific use case (guided partitioning using LVM with or
without encryption). The issue already existed and remains when using
guided partitioniong without LVM (ext4 root, no separate /boot).
Post by Holger Wansing
Post by Pascal Hambourg
According to
<https://github.com/open-power/petitboot/blob/master/README.md> and
<https://open-power.github.io/petitboot/platforms.html>, Petitboot is
also available for arm64 platforms. Partitioning recipes for arm64 also
use ext4 for /boot, so it would not work either.
Hmm, is it widely used on arm64, or only a cornercase?
Is it worth to change the default on arm64 just for petitboot?
One could ask the same about ppc64el. Is Petitboot provided by IBM as
part of this platform firmware ? The recipes for ppc64el create a PReP
partition which, as far as I know, is intended for OpenFirmware + GRUB.

I have no clue about Petitboot on arm64, I didn't know anything about
Petitboot before today. A quick web search shows that Petitboot is used
on arm64-based Odroid boards. An article [1] (in french, sorry) suggests
that it supports ext4, so I wonder why it does not on the OP's machine.

[1]
<https://linuxfr.org/users/pied/journaux/petitboot-sur-arm-le-bon-le-bad-et-le-ugly>
Holger Wansing
2025-01-12 15:50:01 UTC
Reply
Permalink
Hi,
Post by Pascal Hambourg
Post by Holger Wansing
Post by Pascal Hambourg
Post by Holger Wansing
Post by Cyril Brulebois
Post by o***@tutanota.com
Hello,
installing Trixie Alpha 1 on Tyan TN71-BP012 (POWER8)
(/images/trixie_di_alpha1/ppc64el 2024/12/31)
Installation goes well but installed system isn't bootable with Petitboot.
Installation was made choosing automated partitioning for an entrier disk with LVM and luks.
Doing that unencrypted /boot partition is ext4 and Petitboot can't see it.
When partitioning if I edit /boot partition and set it to ext2 then after finishing installation,
PetitBoot see it and I'm able to boot in the installed system.
the installer components that implement those choices.
The move from ext2 to ext4 for /boot was in response to bug #985463
<https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=985463> after it was
considered that bootloaders such as GRUB and U-Boot currently used with
Debian supported ext4.
Inndeed it is unfortunate that a modern boot loader like Petitboot does
not support ext4. It is even unexpected from a boot loader based on a
Linux kernel, as mentioned in
<https://open-power.github.io/petitboot/overview.html>.
Post by Holger Wansing
thanks for pointing me on this.
https://salsa.debian.org/installer-team/partman-auto/-/commit/a87cccbe0198b4dea2cc46a39d5f5b99501aa095
Partitioning recipes for ppc64el create a separate /boot partition only
when using LVM (and this was never changed). When using plain
partitions, /boot is on the ext4 root partition so the above fix will
have no effect.
The report was for guided partitioning with encrypted LVM.
So I assumed the above would fit.
Am I missing something?
No, I just wanted to make sure everyone is aware that the revert fixes
only this specific use case (guided partitioning using LVM with or
without encryption). The issue already existed and remains when using
guided partitioniong without LVM (ext4 root, no separate /boot).
Ok, so that's fine.
With this fix in place, there is at least the LVM partitioning scheme
(with or without encryption), which works out of the box with default
choice.
And assuming we got no complains before about this situation on this machine,
I guess it's fine this way.
Post by Pascal Hambourg
Post by Holger Wansing
Post by Pascal Hambourg
According to
<https://github.com/open-power/petitboot/blob/master/README.md> and
<https://open-power.github.io/petitboot/platforms.html>, Petitboot is
also available for arm64 platforms. Partitioning recipes for arm64 also
use ext4 for /boot, so it would not work either.
Hmm, is it widely used on arm64, or only a cornercase?
Is it worth to change the default on arm64 just for petitboot?
One could ask the same about ppc64el. Is Petitboot provided by IBM as
part of this platform firmware ? The recipes for ppc64el create a PReP
partition which, as far as I know, is intended for OpenFirmware + GRUB.
I have no clue about Petitboot on arm64, I didn't know anything about
Petitboot before today. A quick web search shows that Petitboot is used
on arm64-based Odroid boards. An article [1] (in french, sorry) suggests
that it supports ext4, so I wonder why it does not on the OP's machine.
Maybe the OP has some information on this...


Holger
--
Holger Wansing <***@mailbox.org>
PGP-Fingerprint: 496A C6E8 1442 4B34 8508 3529 59F1 87CA 156E B076
Pascal Hambourg
2025-01-12 18:20:01 UTC
Reply
Permalink
Post by Holger Wansing
Post by Pascal Hambourg
I just wanted to make sure everyone is aware that the revert fixes
only this specific use case (guided partitioning using LVM with or
without encryption). The issue already existed and remains when using
guided partitioniong without LVM (ext4 root, no separate /boot).
Ok, so that's fine.
With this fix in place, there is at least the LVM partitioning scheme
(with or without encryption), which works out of the box with default
choice.
And assuming we got no complains before about this situation on this machine,
I guess it's fine this way.
There is too much inconsistency here for me to say it is fine.

On one hand, the recipes for ppc64el create an optional ext2 /boot
partition only with LVM, which means that booting with Petitboot will
fail if the user chooses guided partitioning without LVM.

On the other hand, the recipes for arm64 uselessly create a mandatory
ext4 /boot partition even without LVM.

In between, the recipes for amd64 create an optional ext4 /boot
partition only with LVM and the default recipes create a mandatory ext2
/boot partition in all cases.

Either a common boot loader for a given architecture does not support
/boot on ext4 and then the recipes for this architecture should create a
mandatory ext2 /boot partition in all cases (like default recipes), or
all common boot loaders are known to support /boot on ext4 and then the
recipes should create an optional ext4 /boot partition only with LVM
(like recipes for amd64).
Andrew M.A. Cater
2025-01-12 18:40:02 UTC
Reply
Permalink
Post by Pascal Hambourg
Post by Holger Wansing
Post by Pascal Hambourg
I just wanted to make sure everyone is aware that the revert fixes
only this specific use case (guided partitioning using LVM with or
without encryption). The issue already existed and remains when using
guided partitioniong without LVM (ext4 root, no separate /boot).
Ok, so that's fine.
There is too much inconsistency here for me to say it is fine.
On one hand, the recipes for ppc64el create an optional ext2 /boot partition
only with LVM, which means that booting with Petitboot will fail if the user
chooses guided partitioning without LVM.
ppc64el - I think this is the first time I've heard this mentioned in a
long time? It's one of those architectures that the media releases team
have no hardware for, for example (having just done a point release yesterday).

Also, as noted, Petitboot isn't packaged in Debian as far as I can see
and the Power 8 probably works fine with grub?
Post by Pascal Hambourg
On the other hand, the recipes for arm64 uselessly create a mandatory ext4
/boot partition even without LVM.
In between, the recipes for amd64 create an optional ext4 /boot partition
only with LVM and the default recipes create a mandatory ext2 /boot
partition in all cases.
Either a common boot loader for a given architecture does not support /boot
on ext4 and then the recipes for this architecture should create a mandatory
ext2 /boot partition in all cases (like default recipes), or all common boot
loaders are known to support /boot on ext4 and then the recipes should
create an optional ext4 /boot partition only with LVM (like recipes for
amd64).
Petitboot suggests that it *should* boot on all filesystems supported by
Linux kexec. Although consistency is desirable across architectures,
maybe this really is an edge case?

All the very best, as ever,

Andy
Holger Wansing
2025-01-12 18:50:01 UTC
Reply
Permalink
Hi again,
Post by Pascal Hambourg
Post by Holger Wansing
Post by Pascal Hambourg
I just wanted to make sure everyone is aware that the revert fixes
only this specific use case (guided partitioning using LVM with or
without encryption). The issue already existed and remains when using
guided partitioniong without LVM (ext4 root, no separate /boot).
Ok, so that's fine.
With this fix in place, there is at least the LVM partitioning scheme
(with or without encryption), which works out of the box with default
choice.
And assuming we got no complains before about this situation on this machine,
I guess it's fine this way.
There is too much inconsistency here for me to say it is fine.
"Is fine" was rather meant as "ok, I understand what you meant".
Post by Pascal Hambourg
On one hand, the recipes for ppc64el create an optional ext2 /boot partition only with LVM, which means that booting with Petitboot will fail if the user chooses guided partitioning without LVM.
On the other hand, the recipes for arm64 uselessly create a mandatory ext4 /boot partition even without LVM.
In between, the recipes for amd64 create an optional ext4 /boot partition only with LVM and the default recipes create a mandatory ext2 /boot partition in all cases.
Either a common boot loader for a given architecture does not support /boot on ext4 and then the recipes for this architecture should create a mandatory ext2 /boot partition in all cases (like default recipes), or all common boot loaders are known to support /boot on ext4 and then the recipes should create an optional ext4 /boot partition only with LVM (like recipes for amd64).
What you describe would be the perfect world.
But what we have here leads to a trade-off anyway.
We cannot do it perfectly right for all cases.


Holger
--
Sent from /e/ OS on Fairphone3
Pascal Hambourg
2025-01-12 22:50:01 UTC
Reply
Permalink
Post by Holger Wansing
Post by Pascal Hambourg
Post by Holger Wansing
Post by Pascal Hambourg
I just wanted to make sure everyone is aware that the revert fixes
only this specific use case (guided partitioning using LVM with or
without encryption). The issue already existed and remains when using
guided partitioniong without LVM (ext4 root, no separate /boot).
Ok, so that's fine.
With this fix in place, there is at least the LVM partitioning scheme
(with or without encryption), which works out of the box with default
choice.
And assuming we got no complains before about this situation on this machine,
I guess it's fine this way.
There is too much inconsistency here for me to say it is fine.
"Is fine" was rather meant as "ok, I understand what you meant".
Ah, I read it as "the risk of boot failure when using guided
partitioning without LVM is acceptable while nobody complains".
Post by Holger Wansing
Post by Pascal Hambourg
On one hand, the recipes for ppc64el create an optional ext2 /boot partition only with LVM, which means that booting with Petitboot will fail if the user chooses guided partitioning without LVM.
On the other hand, the recipes for arm64 uselessly create a mandatory ext4 /boot partition even without LVM.
In between, the recipes for amd64 create an optional ext4 /boot partition only with LVM and the default recipes create a mandatory ext2 /boot partition in all cases.
Either a common boot loader for a given architecture does not support /boot on ext4 and then the recipes for this architecture should create a mandatory ext2 /boot partition in all cases (like default recipes), or all common boot loaders are known to support /boot on ext4 and then the recipes should create an optional ext4 /boot partition only with LVM (like recipes for amd64).
What you describe would be the perfect world.
But what we have here leads to a trade-off anyway.
We cannot do it perfectly right for all cases.
In a perfect world, all boot loaders would support /boot on ext4 and we
would not have to make trade-offs. Indeed we must make a trade-off
between robustness (ext4 vs ext2), convenience (/boot in root filesystem
vs small separate /boot) and compatibility with boot loaders. And I
believe we can do it right at least for use cases we know about. You
addressed the case of guided partitioning without LVM on ppc64el, why
not do the same for guided partitioning without LVM ? Do you think that
the robustness or convenience gain of not having a separate ext2 /boot
partition is worth the loss of compatibility with Petitboot ?
Holger Wansing
2025-01-13 06:50:01 UTC
Reply
Permalink
Hi,
In a perfect world, all boot loaders would support /boot on ext4 and we would not have to make trade-offs. Indeed we must make a trade-off between robustness (ext4 vs ext2), convenience (/boot in root filesystem vs small separate /boot) and compatibility with boot loaders. And I believe we can do it right at least for use cases we know about. You addressed the case of guided partitioning without LVM on ppc64el, why not do the same for guided partitioning without LVM ? Do you think that the robustness or convenience gain of not having a separate ext2 /boot partition is worth the loss of compatibility with Petitboot ?
I cannot judge on this, because it depends on, how much Petitboot
is used on that arch.
If it's only some few percent, you would force much other people to
a ext2 /boot without a reason.
And the same for arm64, if we roll back to ext2 there too.

I had hoped that the OP would give some statistics about this...


Holger
--
Sent from /e/ OS on Fairphone3
Pascal Hambourg
2025-01-13 08:40:01 UTC
Reply
Permalink
Post by Holger Wansing
Post by Pascal Hambourg
Indeed we must make a trade-off
between robustness (ext4 vs ext2), convenience (/boot in root filesystem
vs small separate /boot) and compatibility with boot loaders. And I
believe we can do it right at least for use cases we know about. You
addressed the case of guided partitioning without LVM on ppc64el, why
not do the same for guided partitioning without LVM ? Do you think that
the robustness or convenience gain of not having a separate ext2 /boot
partition is worth the loss of compatibility with Petitboot ?
I cannot judge on this, because it depends on, how much Petitboot
is used on that arch.
If it's only some few percent, you would force much other people to
a ext2 /boot without a reason.
I agree. Petitboot is mentioned on IBM's website [1] so I suspect it is
rather standard on IBM POWER machines.

[1]
<https://www.ibm.com/docs/en/linux-on-systems?topic=systems-petitboot-bootloader>
Post by Holger Wansing
And the same for arm64, if we roll back to ext2 there too.
AFAIK the only implementation of Petitboot on arm64 (Odroid) supports
ext4 so there is no need to rollback to ext2. I did not mean to roll
back to ext2 but to create a separate /boot partition only with LVM, as
I do not see the benefit of a separate ext4 /boot partition without LVM.
Post by Holger Wansing
I had hoped that the OP would give some statistics about this...
So do I...
Diederik de Haas
2025-01-13 13:40:01 UTC
Reply
Permalink
Post by Pascal Hambourg
Post by Holger Wansing
Post by Pascal Hambourg
Indeed we must make a trade-off
between robustness (ext4 vs ext2), convenience (/boot in root filesystem
vs small separate /boot) and compatibility with boot loaders. And I
believe we can do it right at least for use cases we know about. You
addressed the case of guided partitioning without LVM on ppc64el, why
not do the same for guided partitioning without LVM ? Do you think that
the robustness or convenience gain of not having a separate ext2 /boot
partition is worth the loss of compatibility with Petitboot ?
I cannot judge on this, because it depends on, how much Petitboot
is used on that arch.
If it's only some few percent, you would force much other people to
a ext2 /boot without a reason.
I agree. Petitboot is mentioned on IBM's website [1] so I suspect it is
rather standard on IBM POWER machines.
[1]
<https://www.ibm.com/docs/en/linux-on-systems?topic=systems-petitboot-bootloader>
Post by Holger Wansing
And the same for arm64, if we roll back to ext2 there too.
AFAIK the only implementation of Petitboot on arm64 (Odroid) supports
ext4 so there is no need to rollback to ext2. I did not mean to roll
back to ext2 but to create a separate /boot partition only with LVM, as
I do not see the benefit of a separate ext4 /boot partition without LVM.
Post by Holger Wansing
I had hoped that the OP would give some statistics about this...
So do I...
FWIW, I hadn't heard about Petitboot before this bug either, so I looked
on their Github repo and found "and ARM64 with ACPI" ... and I think the
"with ACPI" is significant.
So going back to ext2 on arm64 just for Petitboot users seems a bit
heavy handed to me (but I may not fully understand it all).

My 0.02
Pascal Hambourg
2025-01-13 20:20:02 UTC
Reply
Permalink
Post by Diederik de Haas
Post by Pascal Hambourg
AFAIK the only implementation of Petitboot on arm64 (Odroid) supports
ext4 so there is no need to rollback to ext2. I did not mean to roll
back to ext2 but to create a separate /boot partition only with LVM, as
I do not see the benefit of a separate ext4 /boot partition without LVM.
FWIW, I hadn't heard about Petitboot before this bug either, so I looked
on their Github repo and found "and ARM64 with ACPI" ... and I think the
"with ACPI" is significant.
So going back to ext2 on arm64 just for Petitboot users seems a bit
heavy handed to me (but I may not fully understand it all).
Do I need to repeat that I did no suggest to revert the /boot partition
to ext2 on arm64 ? I found other evidence that Petitboot supports ext4:

<https://discourse.nixos.org/t/newbie-installs-nixos-on-an-arm-sbc-or-how-patience-is-a-virtue/35020>
Odroid ARM64
"Additionally when making a new installation, Petitboot supports ext2,
ext3 and ext4 and a couple other filesystems like vfat and iso9660, so
/boot should be on one of those."

<https://forum.odroid.com/viewtopic.php?p=251632>
Odroid ARM64
"OS in the memory card can boot from Petitboot if the type of the root
file system is ext2/3/4."

This one is even more interesting:

<https://www.reddit.com/r/ps3homebrew/comments/18mzyvu/petitboot_ext4/?rdt=46856>
PS3 POWER
"I noticed that petitboot can boot from ext4, but only if it was created
inside petitboot itself. For some reason if I create it using Gentoo,
petitboot is not able to mount it. What could be the difference that
makes it unsupported?"

Maybe some Petiboot instances use older kernels which do not support new
ext4 features enabled by default in more recent e2fsprogs ?
Diederik de Haas
2025-01-13 20:30:01 UTC
Reply
Permalink
Post by Pascal Hambourg
Post by Diederik de Haas
Post by Pascal Hambourg
AFAIK the only implementation of Petitboot on arm64 (Odroid) supports
ext4 so there is no need to rollback to ext2. I did not mean to roll
back to ext2 but to create a separate /boot partition only with LVM, as
I do not see the benefit of a separate ext4 /boot partition without LVM.
FWIW, I hadn't heard about Petitboot before this bug either, so I looked
on their Github repo and found "and ARM64 with ACPI" ... and I think the
"with ACPI" is significant.
So going back to ext2 on arm64 just for Petitboot users seems a bit
heavy handed to me (but I may not fully understand it all).
Do I need to repeat that I did no suggest to revert the /boot partition
Not AFAIC. My post was basically a +1 on yours + some possible extra
arguments to support using ext4 on arm64 and not reverting to ext2.
And FWIW, I also found 'hints' that ext4 not working was to be
considered an anomaly itself.
Holger Wansing
2025-01-14 21:50:01 UTC
Reply
Permalink
Hi,
Post by Pascal Hambourg
In a perfect world, all boot loaders would support /boot on ext4 and we
would not have to make trade-offs. Indeed we must make a trade-off
between robustness (ext4 vs ext2), convenience (/boot in root filesystem
vs small separate /boot) and compatibility with boot loaders. And I
believe we can do it right at least for use cases we know about. You
addressed the case of guided partitioning without LVM on ppc64el, why
without -> with ^^^
Post by Pascal Hambourg
not do the same for guided partitioning without LVM ? Do you think that
the robustness or convenience gain of not having a separate ext2 /boot
partition is worth the loss of compatibility with Petitboot ?
Ok, so I have now changed that, to create a separate ext2 /boot partition
for the non-lvm case as well.


Holger
--
Holger Wansing <***@mailbox.org>
PGP-Fingerprint: 496A C6E8 1442 4B34 8508 3529 59F1 87CA 156E B076
John Paul Adrian Glaubitz
2025-01-13 08:50:01 UTC
Reply
Permalink
Hello,
Post by Pascal Hambourg
Post by Cyril Brulebois
Post by o***@tutanota.com
Hello,
installing Trixie Alpha 1 on Tyan TN71-BP012 (POWER8)
(/images/trixie_di_alpha1/ppc64el 2024/12/31)
Installation goes well but installed system isn't bootable with Petitboot.
Installation was made choosing automated partitioning for an entrier disk with LVM and luks.
Doing that unencrypted /boot partition is ext4 and Petitboot can't see it.
When partitioning if I edit /boot partition and set it to ext2 then after finishing installation,
PetitBoot see it and I'm able to boot in the installed system.
the installer components that implement those choices.
The move from ext2 to ext4 for /boot was in response to bug #985463
<https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=985463> after it was
considered that bootloaders such as GRUB and U-Boot currently used with
Debian supported ext4.
Well, the thing is that a real modern bootloader doesn't require a separate
boot partition in the first place. The main reason for using a dedicated
partition for /boot is because older firmware implementations and bootloaders
have various limitations with regards to the format and size of the partition
where either later stages of the bootloader or the kernel plus initrd are stored.

And since firmware implementations are usually not changing their limitations for
existing hardware, the format and size limitations for the boot partitions can't
just easily changed. Switching the boot partition to a modern filesystem or making
it considerably larger defeats the whole purpose of a boot partition.

This is why I was very skeptical about the switch from ext2 to ext4 in the first
place and the argument mentioned in the bug report was very weak in my opinion
as it basically just the complaint about incorrect file dates beyond 2038 which
I consider a minor annoyance compared to the possible boot breakage for various
supported hardware systems.
Post by Pascal Hambourg
Inndeed it is unfortunate that a modern boot loader like Petitboot does
not support ext4. It is even unexpected from a boot loader based on a
Linux kernel, as mentioned in
<https://open-power.github.io/petitboot/overview.html>.
It might be unexpected but does not excuse us from making such changes without
verifying first that they don't break booting on supported hardware.

Adrian
--
.''`. John Paul Adrian Glaubitz
: :' : Debian Developer
`. `' Physicist
`- GPG: 62FF 8A75 84E0 2956 9546 0006 7426 3B37 F5B5 F913
o***@tutanota.com
2025-01-13 11:30:02 UTC
Reply
Permalink
Hello,

I try generated installation with ext4 /boot in a more modern ppc64el
(Raptor CS Blackbird with Petitboot 0ef84c0 (POWER9)) and it can see and boot the generated installation.
So seems only POWER8 Tyan systems that are unable to boot on ext4,
It seems they (Tyan) stop making POWER soon after so there is not PNOR(bios) update after  march 2016.
For IBM made POWER8, possible better support and being able to boot on /ext4 but don't have one to test.
Sorry for the burden for an old POWER8 by Tyan.

Have a nice day,
Olivier
Loading...