Bug 1219965

Indeed, the has_efi check is only applied on i386 and x86_64 systems:

{% elif distro_tree is arch('i386', 'x86_64') %}
{% if system and system.has_efi %}
{# x86 EFI #}
part /boot/efi --fstype vfat --size 200 --recommended
{%- if ondisk %} --ondisk={{ ondisk }}{% endif %}

https://git.beaker-project.org/cgit/beaker/tree/Server/bkr/server/snippets/rhts_partitions?h=release-20#n26

Rather than using system.has_efi, it would probably be better to treat aarch64 like ia64:

{% elif distro_tree is arch('aarch64') %}
part /boot/efi --fstype vfat --size 200 --recommended
{%- if ondisk %} --ondisk={{ ondisk }}{% endif %}

Thanks for the report Jeff.

(In reply to Jeff Bastian from comment #3)
> Rather than using system.has_efi, it would probably be better to treat
> aarch64 like ia64:
> 
> {% elif distro_tree is arch('aarch64') %}
> part /boot/efi --fstype vfat --size 200 --recommended
> {%- if ondisk %} --ondisk={{ ondisk }}{% endif %}

This is probably the right approach.

I think Fedora can be installed on some existing hardware using devicetree instead of EFI so in those cases /boot/efi may not be needed? However the future seems to be quickly converging on EFI for aarch64 so we probably don't need to worry about other possibilities.

Yes, Fedora can be run on U-boot systems, however, the ARM SBSA (Server Base System Architecture) requires UEFI + ACPI on server class 64-bit ARM systems, and my guess is Beaker instances are more likely to have server class hardware.  

However, to be friendly to U-boot systems, maybe we could add a flag to disable creating /boot/efi?  Something like this:

{% elif distro_tree is arch('aarch64') %}
{%- if uboot is not defined %}
part /boot/efi --fstype vfat --size 200 --recommended
{%- if ondisk %} --ondisk={{ ondisk }}{% endif %}
{% endif %}

So, it will create /boot/efi by default on aarch64 systems, unless you set ks_meta="uboot" which will disable it.

(In reply to Jeff Bastian from comment #5)
> Yes, Fedora can be run on U-boot systems, however, the ARM SBSA (Server Base
> System Architecture) requires UEFI + ACPI on server class 64-bit ARM
> systems, and my guess is Beaker instances are more likely to have server
> class hardware.  
> 
> However, to be friendly to U-boot systems, maybe we could add a flag to
> disable creating /boot/efi?  Something like this:
> 
> {% elif distro_tree is arch('aarch64') %}
> {%- if uboot is not defined %}
> part /boot/efi --fstype vfat --size 200 --recommended
> {%- if ondisk %} --ondisk={{ ondisk }}{% endif %}
> {% endif %}
> 
> So, it will create /boot/efi by default on aarch64 systems, unless you set
> ks_meta="uboot" which will disable it.

That sounds like a good idea. We can add uboot into the custom kickstart metadata variables.

https://beaker-project.org/docs-develop/user-guide/customizing-installation.html#kickstart-metadata

On Gerrit:

  http://gerrit.beaker-project.org/#/c/4226/

Do we really need to bother adding a uboot ksmeta variable? I am worried that it is adding complexity (not to mention, a somewhat confusingly named variable) for a very small corner case.

If someone has an aarch64 system using UBoot, won't the /boot/efi partition just go unused without causing any harm? Maybe we can just unconditionally define /boot/efi on aarch64 and that will be good enough.

Personally, I'd be ok with creating /boot/efi unconditionally.  Yes, it'll just be an unused partition on uboot systems.  Other than wasting a small bit (200M) of disk space, it won't hurt anything.

I was trying to be "friendly" to uboot systems by adding this variable, but I think this actually is a deeper issue (which will need a new BZ): you can add additional partitions with the <partitions> tag in a recipe, but you cannot change or override the base partitions (/ and /boot) unless you use <kickstart> which requires the full kickstart set of instructions (minus %packages and %pre/%post scripts).

If this is feasible, I'll open a new RFE.  The syntax could be something like
    <partitions type="additional">   <-- default if type is missing
    <partitions type="custom">       <-- must specify all including / and /boot

(In reply to Jeff Bastian from comment #9)
> I was trying to be "friendly" to uboot systems by adding this variable, but
> I think this actually is a deeper issue (which will need a new BZ): you can
> add additional partitions with the <partitions> tag in a recipe, but you
> cannot change or override the base partitions (/ and /boot) unless you use
> <kickstart> which requires the full kickstart set of instructions (minus
> %packages and %pre/%post scripts).

Yes that's a good point. We already have a very similar issue in bug 1198881, the latest idea there was to have a ksmeta "no_autopart" so that you can put custom partitioning commands in <ks_appends/>. Although I just remembered that won't work on older RHELs which lack the %end delimiter in kickstarts, hmm...

Verified steps:

1. create a job on an aarch64 system

2. check if kickstart creates /boot/efi partition

(In reply to matt jia from comment #12)
> 1. create a job on an aarch64 system

Specifically it has to use custom partitioning to trigger this bug, for example ks_meta="fstype=ext4".

Beaker 20.2 has been released.