[pve-devel] applied: [PATCH edk2-firmware] add patch to revert addition of EFI memory attributes protocol for x86_64

[Fiona Ebner]

Am 27.03.25 um 17:39 schrieb Thomas Lamprecht:
> Am 27.03.25 um 12:53 schrieb Fiona Ebner:
>> Because of a long-standing bug in shim [0], booting will fail for
>> distibutions that do not include the fix yet, like Rocky Linux 9.5
>> and other CentOS-based distibutions. This is cased by the addition
>> of the EFI_MEMORY_ATTRIBUTE_PROTOCOL in edk2 commit efaa102d00
>> ("UefiCpuPkg: Produce EFI memory attributes protocol") for x86_64.
>>
>> Even with the fix in shim, issues in commonly shipped versions of GRUB
>> remain [1].
>>
>> This is relatively recent, i.e. in the edk2-stable202502 tag, and
>> since current non-minor distributions are still affected, revert the
>> problematic commit for now.
>>
>> Once issues are less common in distributions, an option to support
>> disabling it (via fw_cfg on the QEMU command line) can still be added
>> [1]. Then, it can also be nicely documented as a known issue while
>> giving users guidance.
>>
>> There already is a similar patch for ARM [2] inherited from the Debian
>> upstream version.
>>
>> The problematic commit is EFI_MEMORY_ATTRIBUTE_PROTOCOL was added for x86_64 recently in the
>> edk2-stable202502 tag. Since current non-minor distributions are still
>> affected, a revert is done for now.
>>
>> [0]: https://github.com/rhboot/shim/commit/c7b305152802c8db688605654f75e1195def9fd6
>> [1]: https://github.com/tianocore/edk2/pull/10667
>> [2]: ./debian/patches/ArmVirtPkg-disable-the-EFI_MEMORY_ATTRIBUTE-protocol.patch
>>
>> Signed-off-by: Fiona Ebner <f.ebner at proxmox.com>
>> ---
>>  ...g-Produce-EFI-memory-attributes-prot.patch | 368 ++++++++++++++++++
>>  debian/patches/series                         |   1 +
>>  2 files changed, 369 insertions(+)
>>  create mode 100644 debian/patches/Revert-UefiCpuPkg-Produce-EFI-memory-attributes-prot.patch
>>
>>
> 
> applied, re-exported the revert using git format-patch as the unusual
> DOS line-endings that EDK2 source code uses seem to got lost somewhere
> along the way, and while I could have used quilt I figured there's no
> harm with format-patch, thanks!

Right, there's that issue with mailman (which I forgot about since the
last time this came up was nearly 2 years ago):
https://lore.proxmox.com/pve-devel/c72e637b-2e92-2ff5-0bc5-0c99d55b2ec8@proxmox.com/

CC-ing Stoiko