[edk2-devel] MdeModulePkg build fails for AARCH64 on Ubuntu 22.04

[Ard Biesheuvel]

On Mon, 22 Aug 2022 at 11:30, Ard Biesheuvel <ardb at kernel.org> wrote:
>
> On Mon, 22 Aug 2022 at 11:11, Ard Biesheuvel <ardb at kernel.org> wrote:
> >
> > (cc Bob)
> >
> > NOTE this affects the stable tag - please see below
> >
> >
> > On Sun, 21 Aug 2022 at 06:34, Rebecca Cran <rebecca at bsdio.com> wrote:
> > >
> > > I noticed that MdeModulePkg fails to build on my Ubuntu 22.04 system
> > > (and previously on the Ubuntu 20.04 installation). Is this a known bug?
> > >
> > > It shouldn't be relevant, but I'm using gcc version 11.2.0 (Ubuntu
> > > 11.2.0-17ubuntu1). But perhaps more relevant, I'm using Python 3.10.4.
> > >
> > > I'm trying to build commit e2ac68a23b4954d5c0399913a1df3dd9fd90315d.
> > >
> > >
> > > bcran at photon:~/src/tmp/edk2$ build -p MdeModulePkg/MdeModulePkg.dsc -a
> > > AARCH64 -t GCC5 -b RELEASE
> > > Build environment: Linux-5.15.0-46-generic-x86_64-with-glibc2.35
> > > Build start time: 22:29:18, Aug.20 2022
> > >
> > > WORKSPACE        = /home/bcran/src/tmp/edk2
> > > EDK_TOOLS_PATH   = /home/bcran/src/tmp/edk2/BaseTools
> > > CONF_PATH        = /home/bcran/src/tmp/edk2/Conf
> > > PYTHON_COMMAND   = /usr/bin/python3
> > >
> > >
> > > Processing meta-data
> > > .Architecture(s)  = AARCH64
> > > Build target     = RELEASE
> > > Toolchain        = GCC5
> > >
> > > Active Platform          =
> > > /home/bcran/src/tmp/edk2/MdeModulePkg/MdeModulePkg.dsc
> > >
> > >
> > > build.py...
> > > /home/bcran/src/tmp/edk2/MdeModulePkg/Library/SmmLockBoxLib/SmmLockBoxPeiLib.inf(42):
> > > error 3000: PCD
> > > [gEfiMdeModulePkgTokenSpaceGuid.PcdDxeIplSwitchToLongMode] in
> > > [/home/bcran/src/tmp/edk2/MdeModulePkg/Library/SmmLockBoxLib/SmmLockBoxPeiLib.inf]
> > > is not found in dependent packages:
> > >                  /home/bcran/src/tmp/edk2/MdePkg/MdePkg.dec
> > >          /home/bcran/src/tmp/edk2/MdeModulePkg/MdeModulePkg.dec
> > >
> > >
> >
> > This looks like a BaseTools regression to me - afaik, these packages
> > all used to build for AARCH64 in the past.
> >
> > The following LockBoxLib resolutions appear in MdeModulePkg.dsc (with
> > line numbers)
> >
> > 115-[LibraryClasses.common.PEIM]
> > 119:  LockBoxLib|MdeModulePkg/Library/SmmLockBoxLib/SmmLockBoxPeiLib.inf
> >
> > 178-[LibraryClasses.ARM, LibraryClasses.AARCH64]
> > 181:  LockBoxLib|MdeModulePkg/Library/LockBoxNullLib/LockBoxNullLib.inf
> >
> > No other [LlibraryClasses] references to SmmLockBoxPeiLib.inf exist in
> > that file, and the [Components] reference does not apply to AARCH64.
> >
> > This means the DSC parser ends up using the earlier definition, which
> > I think is a bug, or at least a change in behavior. Note that doing
> > this silently could potentially break many platforms in subtle ways so
> > I think this should be root caused and preferably fixed before
> > creating the stable tag.
>
> Seems to be deliberate:
>
> commit 039bdb4d3e96f9c9264abf135b8a0eef2e2b4860
> Author: Chen, Christine <Yuwei.Chen at intel.com>
> Date:   Thu Jun 30 17:04:05 2022 +0800
>
>     BaseTools: Fix DSC LibraryClass precedence rule
>
> but I don't think the impact on other platforms and drivers has been
> taken into account here. Also, the document [0] seems ambiguous to me:
>
> """
> The first globally defined library instance, defined in a DSC file,
> that satisfies a module's requirement for a Library Class, unless
> specifically overridden by the module in the [Components] section,
> will be used.
> """
>
> This says that the first occurrence of a library instance definition
> that satisfies a INF dependency will be selected. Then, there is
>
> """
> The Library Instances will be selected using the following rules to
> satisfy a library class for each module listed in the [Components]
> section (in order of highest precedence):
> """
>
> which says that not the *first* occurrence, but the occurrence *with
> the highest precedence* will be selected.
>
> Then, the precedence list has these items:
>
> """
> 2. [LibraryClasses.$(Arch).$(MODULE_TYPE),
> LibraryClasses.$(Arch).$(MODULE_TYPE)]
> 3. [LibraryClasses.$(Arch).$(MODULE_TYPE)]
> """
>
> which seem to suggest that a definition that applies to multiple
> arch/module_type combinations automatically takes precedence over one
> that targets a single arch/module_type combination, which makes no
> sense at all.
>
> So I strongly suggest we revert the patch, fix the document so it is
> clear and ambiguous, and preferably, align it with how it used to

*un*ambiguous ....

sigh


-=-=-=-=-=-=-=-=-=-=-=-
Groups.io Links: You receive all messages sent to this group.
View/Reply Online (#92614): https://edk2.groups.io/g/devel/message/92614
Mute This Topic: https://groups.io/mt/93156882/1813853
Group Owner: devel+owner at edk2.groups.io
Unsubscribe: https://edk2.groups.io/g/devel/unsub [edk2-devel-archive at redhat.com]
-=-=-=-=-=-=-=-=-=-=-=-