WIP: edk2: Bump to 202308

[mihalicyn]

Fixes canonical/lxd#12166

Not ready yet as I need to solve a build issue:

2023-09-01 14:28:49.855 :: 2023-09-01 12:28:46.988 :: ###
2023-09-01 14:28:49.855 :: 2023-09-01 12:28:46.988 :: Generating DXEFV FV
2023-09-01 14:28:49.855 :: 2023-09-01 12:28:48.970 :: ########### ['GenFv', '-a', '/root/parts/edk2/build/Build/OvmfX64/RELEASE_GCC5/FV/Ffs/FVMAIN_COMPACT.inf', '-o', '/root/parts/edk2/build/Build/OvmfX64/RELEASE_GCC5/FV/FVMAIN_COMPACT.Fv', '-i', '/root/parts/edk2/build/Build/OvmfX64/RELEASE_GCC5/FV/FVMAIN_COMPACT.inf']
2023-09-01 14:28:49.855 :: 2023-09-01 12:28:48.970 :: Return Value = 2
2023-09-01 14:28:49.855 :: 2023-09-01 12:28:48.970 :: GenFv: ERROR 3000: Invalid
2023-09-01 14:28:49.855 :: 2023-09-01 12:28:48.970 ::   the required fv image size 0x1b9180 exceeds the set fv image size 0x1ac000
2023-09-01 14:28:49.855 :: 2023-09-01 12:28:48.970 ::
2023-09-01 14:28:49.855 :: 2023-09-01 12:28:48.971 ::
2023-09-01 14:28:49.855 :: 2023-09-01 12:28:48.971 ::
2023-09-01 14:28:49.855 :: 2023-09-01 12:28:48.971 ::
2023-09-01 14:28:49.855 :: 2023-09-01 12:28:48.971 :: build.py...
2023-09-01 14:28:49.855 :: 2023-09-01 12:28:48.971 ::  : error 7000: Failed to generate FV

[mihalicyn]

Most likely size of the firmware increased after tianocore/edk2@ff3382a
Will need to experiment with this thing more.

0x1ac000 is a standard FV block size for 2MiB firmware:
https://github.com/tianocore/edk2/blob/beafabdae49c873adecdb7511dbebe9d4ff5c8f0/OvmfPkg/Include/Fdf/OvmfPkgDefines.fdf.inc#L48

We fail on 2MiB firmware which does not include CSM blob.

This allows to build 2MiB image:

       build_edk2 \
         "${CRAFT_PART_INSTALL}/share/qemu/OVMF_CODE.2MB.fd" \
         "${CRAFT_PART_INSTALL}/share/qemu/OVMF_VARS.2MB.fd" \
-        -DFD_SIZE_2MB
+        -DFD_SIZE_2MB \
+        -DNETWORK_ISCSI_ENABLE=FALSE \
+        -DBUILD_SHELL=FALSE
       ln -s OVMF_CODE.2MB.fd ${CRAFT_PART_INSTALL}/share/qemu/OVMF_CODE.fd
       ln -s OVMF_VARS.2MB.fd ${CRAFT_PART_INSTALL}/share/qemu/OVMF_VARS.fd

I don't think that we need to build iSCSI support in UEFI firmware (we don't use iSCSI as far as I understand).
But I'm not sure about UEFI Shell. Nobody uses it in 99% cases, but...

As an alternative we can build 4MB blob from the new version of edk2, and 2MB from the old one.

[simondeziel]

IIRC the UEFI shell was sometimes mentioned in forum posts for debugging issues with booting Windows from an ISO.

As for the iSCSI support, it sounds reasonable to drop and I'd vote to drop it from all build variants, not just the 2MiB one.

[tomponline]

Yeah let's drop iscsi from all variants

[tomponline]

Any other features we can drop?

[simondeziel]

We might be able to reclaim some space by disabling TPM1 support (-DTPM1_ENABLE=FALSE) but I'm not sure if (existing) LXD VMs would be affected.

[tomponline]

Whilst replying to

https://discourse.ubuntu.com/t/lxd-vm-agent-fails-to-come-online-with-lxd-5-15-and-large-vm-resource-requirements/37963/12

I noticed that we had some TPM disable flags in the past on edk2 as well:

dadf771

[mihalicyn]

I noticed that we had some TPM disable flags in the past on edk2 as well:

yes, at the same time, both TPM_ENABLE and TPM_CONFIG_ENABLE are not existing in the edk2 code-base anymore.
tianocore/edk2@4de8d61

As far as I can see TPM1_ENABLE is TRUE by default.

[tomponline]

@mihalicyn will you be able to pick this back up after we remove the CSM support as per canonical/lxd#12730