| Message ID | 20230330134214.2772913-1-zboszor@gmail.com |
|---|---|
| State | New |
| Headers | show |
| Series | [1/2] systemd: Allow native build | expand |
Can you please point to what piece of libgudev actually requires systemd? Alex On Thu, 30 Mar 2023 at 15:42, Zoltan Boszormenyi <zboszor@gmail.com> wrote: > > systemd-native is a dependency for libgudev-native, which > in turn is a dependency for libfprint-native, which is needed > to build libfprint. > > The native systemctl binary is removed so it doesn't conflict > with the script from systemd-systemctl-native. > > TODO? The new natively built systemctl binary may replace > the script in systemd-systemctl-native, as it supports > more options and works better than the script. For example, > the current script does not create the symlinks in > /etc/systemd/system for WantedBy= and RequiredBy= settings > and does not support systemctl set-default some.target. > > Signed-off-by: Zoltán Böszörményi <zboszor@gmail.com> > --- > meta/recipes-core/systemd/systemd_253.1.bb | 20 ++++++++++++++------ > 1 file changed, 14 insertions(+), 6 deletions(-) > > diff --git a/meta/recipes-core/systemd/systemd_253.1.bb b/meta/recipes-core/systemd/systemd_253.1.bb > index 9c2b96d3c1..c73b5e05b3 100644 > --- a/meta/recipes-core/systemd/systemd_253.1.bb > +++ b/meta/recipes-core/systemd/systemd_253.1.bb > @@ -6,6 +6,8 @@ PE = "1" > > DEPENDS = "intltool-native gperf-native libcap util-linux python3-jinja2-native" > > +BBCLASSEXTEND = "native" > + > SECTION = "base/shell" > > inherit useradd pkgconfig meson perlnative update-rc.d update-alternatives qemu systemd gettext bash-completion manpages features_check > @@ -60,7 +62,7 @@ PAM_PLUGINS = " \ > pam-plugin-namespace \ > " > > -PACKAGECONFIG ??= " \ > +PACKAGECONFIG:class-target ??= " \ > ${@bb.utils.filter('DISTRO_FEATURES', 'acl audit efi ldconfig pam selinux smack usrmerge polkit seccomp', d)} \ > ${@bb.utils.contains('DISTRO_FEATURES', 'wifi', 'rfkill', '', d)} \ > ${@bb.utils.contains('DISTRO_FEATURES', 'x11', 'xkbcommon', '', d)} \ > @@ -197,7 +199,7 @@ PACKAGECONFIG[selinux] = "-Dselinux=true,-Dselinux=false,libselinux,initscripts- > PACKAGECONFIG[smack] = "-Dsmack=true,-Dsmack=false" > PACKAGECONFIG[sysext] = "-Dsysext=true, -Dsysext=false" > PACKAGECONFIG[sysusers] = "-Dsysusers=true,-Dsysusers=false" > -PACKAGECONFIG[sysvinit] = "-Dsysvinit-path=${sysconfdir}/init.d -Dsysvrcnd-path=${sysconfdir},-Dsysvinit-path= -Dsysvrcnd-path=,,systemd-compat-units update-rc.d" > +PACKAGECONFIG[sysvinit] = "-Dsysvinit-path=${sysconfdir}/init.d -Dsysvrcnd-path=${sysconfdir},-Dsysvinit-path= -Dsysvrcnd-path=,,${@'systemd-compat-units update-rc.d' if d.getVar('PN') == d.getVar('BPN') else ''}" > # When enabled use reproducble build timestamp if set as time epoch, > # or build time if not. When disabled, time epoch is unset. > def build_epoch(d): > @@ -211,7 +213,7 @@ PACKAGECONFIG[sbinmerge] = "-Dsplit-bin=false,-Dsplit-bin=true" > PACKAGECONFIG[userdb] = "-Duserdb=true,-Duserdb=false" > PACKAGECONFIG[utmp] = "-Dutmp=true,-Dutmp=false" > PACKAGECONFIG[valgrind] = "-DVALGRIND=1,,valgrind" > -PACKAGECONFIG[vconsole] = "-Dvconsole=true,-Dvconsole=false,,${PN}-vconsole-setup" > +PACKAGECONFIG[vconsole] = "-Dvconsole=true,-Dvconsole=false,,${BPN}-vconsole-setup" > PACKAGECONFIG[wheel-group] = "-Dwheel-group=true, -Dwheel-group=false" > PACKAGECONFIG[xdg-autostart] = "-Dxdg-autostart=true,-Dxdg-autostart=false" > # Verify keymaps on locale change > @@ -364,6 +366,12 @@ do_install() { > fi > } > > +do_install:append:class-native () { > + rm -f ${D}${bindir}/systemctl > + ln -sf ..${base_bindir_native}/udevadm ${D}${base_sbindir}/udevadm > + ln -sf ..${systemd_unitdir#${rootprefix}}/systemd-udevd ${D}${base_sbindir}/udevd > +} > + > python populate_packages:prepend (){ > systemdlibdir = d.getVar("rootlibdir") > do_split_packages(d, systemdlibdir, r'^lib(.*)\.so\.*', 'lib%s', 'Systemd %s library', extra_depends='', allow_links=True) > @@ -670,10 +678,10 @@ FILES:${PN} = " ${base_bindir}/* \ > FILES:${PN}-dev += "${base_libdir}/security/*.la ${datadir}/dbus-1/interfaces/ ${sysconfdir}/rpm/macros.systemd" > > RDEPENDS:${PN} += "kmod dbus util-linux-mount util-linux-umount udev (= ${EXTENDPKGV}) systemd-udev-rules util-linux-agetty util-linux-fsck" > -RDEPENDS:${PN} += "${@bb.utils.contains('PACKAGECONFIG', 'serial-getty-generator', '', 'systemd-serialgetty', d)}" > -RDEPENDS:${PN} += "volatile-binds" > +RDEPENDS:${PN}:class-target += "${@bb.utils.contains('PACKAGECONFIG', 'serial-getty-generator', '', 'systemd-serialgetty', d)}" > +RDEPENDS:${PN}:class-target += "volatile-binds" > > -RRECOMMENDS:${PN} += "systemd-extra-utils \ > +RRECOMMENDS:${PN}:class-target += "systemd-extra-utils \ > udev-hwdb \ > e2fsprogs-e2fsck \ > kernel-module-autofs4 kernel-module-unix kernel-module-ipv6 kernel-module-sch-fq-codel \ > -- > 2.39.2 > > > -=-=-=-=-=-=-=-=-=-=-=- > Links: You receive all messages sent to this group. > View/Reply Online (#179321): https://lists.openembedded.org/g/openembedded-core/message/179321 > Mute This Topic: https://lists.openembedded.org/mt/97950749/1686489 > Group Owner: openembedded-core+owner@lists.openembedded.org > Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub [alex.kanavin@gmail.com] > -=-=-=-=-=-=-=-=-=-=-=- >
On Thu, 2023-03-30 at 15:42 +0200, Zoltan Boszormenyi wrote: > systemd-native is a dependency for libgudev-native, which > in turn is a dependency for libfprint-native, which is needed > to build libfprint. > > The native systemctl binary is removed so it doesn't conflict > with the script from systemd-systemctl-native. > > TODO? The new natively built systemctl binary may replace > the script in systemd-systemctl-native, as it supports > more options and works better than the script. For example, > the current script does not create the symlinks in > /etc/systemd/system for WantedBy= and RequiredBy= settings > and does not support systemctl set-default some.target. Which piece of systemd is libfprint needing? We've had a request for systemd-native before and my opinion on it hasn't changed, I'd much prefer not to have it. Cheers, Richard
2023. 03. 30. 15:46 keltezéssel, Alexander Kanavin írta: > Can you please point to what piece of libgudev actually requires systemd? It's this line: DEPENDS = "glib-2.0 udev" udev is in systemd: PROVIDES = "udev" > > Alex > > On Thu, 30 Mar 2023 at 15:42, Zoltan Boszormenyi <zboszor@gmail.com> wrote: >> systemd-native is a dependency for libgudev-native, which >> in turn is a dependency for libfprint-native, which is needed >> to build libfprint. >> >> The native systemctl binary is removed so it doesn't conflict >> with the script from systemd-systemctl-native. >> >> TODO? The new natively built systemctl binary may replace >> the script in systemd-systemctl-native, as it supports >> more options and works better than the script. For example, >> the current script does not create the symlinks in >> /etc/systemd/system for WantedBy= and RequiredBy= settings >> and does not support systemctl set-default some.target. >> >> Signed-off-by: Zoltán Böszörményi <zboszor@gmail.com> >> --- >> meta/recipes-core/systemd/systemd_253.1.bb | 20 ++++++++++++++------ >> 1 file changed, 14 insertions(+), 6 deletions(-) >> >> diff --git a/meta/recipes-core/systemd/systemd_253.1.bb b/meta/recipes-core/systemd/systemd_253.1.bb >> index 9c2b96d3c1..c73b5e05b3 100644 >> --- a/meta/recipes-core/systemd/systemd_253.1.bb >> +++ b/meta/recipes-core/systemd/systemd_253.1.bb >> @@ -6,6 +6,8 @@ PE = "1" >> >> DEPENDS = "intltool-native gperf-native libcap util-linux python3-jinja2-native" >> >> +BBCLASSEXTEND = "native" >> + >> SECTION = "base/shell" >> >> inherit useradd pkgconfig meson perlnative update-rc.d update-alternatives qemu systemd gettext bash-completion manpages features_check >> @@ -60,7 +62,7 @@ PAM_PLUGINS = " \ >> pam-plugin-namespace \ >> " >> >> -PACKAGECONFIG ??= " \ >> +PACKAGECONFIG:class-target ??= " \ >> ${@bb.utils.filter('DISTRO_FEATURES', 'acl audit efi ldconfig pam selinux smack usrmerge polkit seccomp', d)} \ >> ${@bb.utils.contains('DISTRO_FEATURES', 'wifi', 'rfkill', '', d)} \ >> ${@bb.utils.contains('DISTRO_FEATURES', 'x11', 'xkbcommon', '', d)} \ >> @@ -197,7 +199,7 @@ PACKAGECONFIG[selinux] = "-Dselinux=true,-Dselinux=false,libselinux,initscripts- >> PACKAGECONFIG[smack] = "-Dsmack=true,-Dsmack=false" >> PACKAGECONFIG[sysext] = "-Dsysext=true, -Dsysext=false" >> PACKAGECONFIG[sysusers] = "-Dsysusers=true,-Dsysusers=false" >> -PACKAGECONFIG[sysvinit] = "-Dsysvinit-path=${sysconfdir}/init.d -Dsysvrcnd-path=${sysconfdir},-Dsysvinit-path= -Dsysvrcnd-path=,,systemd-compat-units update-rc.d" >> +PACKAGECONFIG[sysvinit] = "-Dsysvinit-path=${sysconfdir}/init.d -Dsysvrcnd-path=${sysconfdir},-Dsysvinit-path= -Dsysvrcnd-path=,,${@'systemd-compat-units update-rc.d' if d.getVar('PN') == d.getVar('BPN') else ''}" >> # When enabled use reproducble build timestamp if set as time epoch, >> # or build time if not. When disabled, time epoch is unset. >> def build_epoch(d): >> @@ -211,7 +213,7 @@ PACKAGECONFIG[sbinmerge] = "-Dsplit-bin=false,-Dsplit-bin=true" >> PACKAGECONFIG[userdb] = "-Duserdb=true,-Duserdb=false" >> PACKAGECONFIG[utmp] = "-Dutmp=true,-Dutmp=false" >> PACKAGECONFIG[valgrind] = "-DVALGRIND=1,,valgrind" >> -PACKAGECONFIG[vconsole] = "-Dvconsole=true,-Dvconsole=false,,${PN}-vconsole-setup" >> +PACKAGECONFIG[vconsole] = "-Dvconsole=true,-Dvconsole=false,,${BPN}-vconsole-setup" >> PACKAGECONFIG[wheel-group] = "-Dwheel-group=true, -Dwheel-group=false" >> PACKAGECONFIG[xdg-autostart] = "-Dxdg-autostart=true,-Dxdg-autostart=false" >> # Verify keymaps on locale change >> @@ -364,6 +366,12 @@ do_install() { >> fi >> } >> >> +do_install:append:class-native () { >> + rm -f ${D}${bindir}/systemctl >> + ln -sf ..${base_bindir_native}/udevadm ${D}${base_sbindir}/udevadm >> + ln -sf ..${systemd_unitdir#${rootprefix}}/systemd-udevd ${D}${base_sbindir}/udevd >> +} >> + >> python populate_packages:prepend (){ >> systemdlibdir = d.getVar("rootlibdir") >> do_split_packages(d, systemdlibdir, r'^lib(.*)\.so\.*', 'lib%s', 'Systemd %s library', extra_depends='', allow_links=True) >> @@ -670,10 +678,10 @@ FILES:${PN} = " ${base_bindir}/* \ >> FILES:${PN}-dev += "${base_libdir}/security/*.la ${datadir}/dbus-1/interfaces/ ${sysconfdir}/rpm/macros.systemd" >> >> RDEPENDS:${PN} += "kmod dbus util-linux-mount util-linux-umount udev (= ${EXTENDPKGV}) systemd-udev-rules util-linux-agetty util-linux-fsck" >> -RDEPENDS:${PN} += "${@bb.utils.contains('PACKAGECONFIG', 'serial-getty-generator', '', 'systemd-serialgetty', d)}" >> -RDEPENDS:${PN} += "volatile-binds" >> +RDEPENDS:${PN}:class-target += "${@bb.utils.contains('PACKAGECONFIG', 'serial-getty-generator', '', 'systemd-serialgetty', d)}" >> +RDEPENDS:${PN}:class-target += "volatile-binds" >> >> -RRECOMMENDS:${PN} += "systemd-extra-utils \ >> +RRECOMMENDS:${PN}:class-target += "systemd-extra-utils \ >> udev-hwdb \ >> e2fsprogs-e2fsck \ >> kernel-module-autofs4 kernel-module-unix kernel-module-ipv6 kernel-module-sch-fq-codel \ >> -- >> 2.39.2 >> >> >> -=-=-=-=-=-=-=-=-=-=-=- >> Links: You receive all messages sent to this group. >> View/Reply Online (#179321): https://lists.openembedded.org/g/openembedded-core/message/179321 >> Mute This Topic: https://lists.openembedded.org/mt/97950749/1686489 >> Group Owner: openembedded-core+owner@lists.openembedded.org >> Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub [alex.kanavin@gmail.com] >> -=-=-=-=-=-=-=-=-=-=-=- >>
2023. 03. 30. 15:50 keltezéssel, Richard Purdie írta: > On Thu, 2023-03-30 at 15:42 +0200, Zoltan Boszormenyi wrote: >> systemd-native is a dependency for libgudev-native, which >> in turn is a dependency for libfprint-native, which is needed >> to build libfprint. >> >> The native systemctl binary is removed so it doesn't conflict >> with the script from systemd-systemctl-native. >> >> TODO? The new natively built systemctl binary may replace >> the script in systemd-systemctl-native, as it supports >> more options and works better than the script. For example, >> the current script does not create the symlinks in >> /etc/systemd/system for WantedBy= and RequiredBy= settings >> and does not support systemctl set-default some.target. > Which piece of systemd is libfprint needing? Please see the MR: https://gitlab.freedesktop.org/libfprint/libfprint/-/merge_requests/431 udev (provided by systemd) -> libgudev -> libfprint also: udev-native -> libgudev-native -> libfprint-native -> libfprint Also, please read the TODO part of the patch for systemd and consider it. Thanks. > > We've had a request for systemd-native before and my opinion on it > hasn't changed, I'd much prefer not to have it. > > Cheers, > > Richard
On Thu, 30 Mar 2023 at 16:08, Zoltan Boszormenyi <zboszor@gmail.com> wrote: > Please see the MR: > https://gitlab.freedesktop.org/libfprint/libfprint/-/merge_requests/431 > > udev (provided by systemd) -> libgudev -> libfprint > also: > udev-native -> libgudev-native -> libfprint-native -> libfprint > > Also, please read the TODO part of the patch for systemd > and consider it. We can simply direct libfprint to run its own target pieces under qemu usermode, and avoid all this nasty native stuff. Meson has direct support for it, and even if it doesn't work, libfprint can be patched to use a custom 'wrapper' for executing the binaries. Alex
2023. 03. 30. 16:10 keltezéssel, Alexander Kanavin írta: > On Thu, 30 Mar 2023 at 16:08, Zoltan Boszormenyi <zboszor@gmail.com> wrote: >> Please see the MR: >> https://gitlab.freedesktop.org/libfprint/libfprint/-/merge_requests/431 >> >> udev (provided by systemd) -> libgudev -> libfprint >> also: >> udev-native -> libgudev-native -> libfprint-native -> libfprint >> >> Also, please read the TODO part of the patch for systemd >> and consider it. > We can simply direct libfprint to run its own target pieces under qemu > usermode The mariadb change to use CROSSCOMPILING_EMULATOR instead of building mariadb-native was reverted because qemu is not ominpotent. The same applies to meson running executables via qemu. No custom wrapper would help. > , and avoid all this nasty native stuff. Meson has direct > support for it, and even if it doesn't work, libfprint can be patched > to use a custom 'wrapper' for executing the binaries. You seem to have omitted reading a large part in the commit message for the systemd change. > TODO? The new natively built systemctl binary may replace > the script in systemd-systemctl-native, as it supports > more options and works better than the script. For example, > the current script does not create the symlinks in > /etc/systemd/system for WantedBy= and RequiredBy= settings > and does not support systemctl set-default some.target. This is not relevant to libfprint itself, it's a bug in Yocto's systemctl replacement script. ----8<--- foo.service ----8<--- ... [Install] WantedBy=bar.service RequiredBy=baz.service ... ----8<--- foo.service ----8<--- The symlinks in /etc/systemd/system/bar.service.wants and /etc/systemd/system/baz.service.requires are not created for foo.service. This command is also not supported by the Yocto replacement: systemctl set-default custom.target I have been bitten by both issues.
On Thu, 30 Mar 2023 at 16:31, Böszörményi Zoltán <zboszor@gmail.com> wrote: > The mariadb change to use CROSSCOMPILING_EMULATOR > instead of building mariadb-native was reverted because > qemu is not ominpotent. The same applies to meson > running executables via qemu. No custom wrapper would help. Mariadb's problem was that (for a specific target) it somehow included CPU instructions that qemu usermode doesn't support, and no one looked into what those instructions were and why they were produced in the first place, preferring to revert the whole thing. There are lots of counter-examples in core where qemu usermode works fine, and has been working that way for many years, for example gobject introspection, or all of the scripts in ../scripts/postinst-intercepts/ which run every time you bake an image, including update_udev_hwdb by the way that runs target udevadm. I am not maintaining oe-devel, and I do not keep an eye on things like that, otherwise I would've asked for further investigation into what crashes and why. I do not want to add complex native dependency chains when usermode would work. It's me who's going to fix them when they break, not you. > You seem to have omitted reading a large part in the > commit message for the systemd change. > > > TODO? The new natively built systemctl binary may replace > > the script in systemd-systemctl-native, as it supports > > more options and works better than the script. For example, > > the current script does not create the symlinks in > > /etc/systemd/system for WantedBy= and RequiredBy= settings > > and does not support systemctl set-default some.target. This is a separate concern, and needs to be proposed as such. Alex
2023. 03. 30. 16:31 keltezéssel, Zoltan Boszormenyi via lists.openembedded.org írta: > 2023. 03. 30. 16:10 keltezéssel, Alexander Kanavin írta: >> On Thu, 30 Mar 2023 at 16:08, Zoltan Boszormenyi <zboszor@gmail.com> wrote: >>> Please see the MR: >>> https://gitlab.freedesktop.org/libfprint/libfprint/-/merge_requests/431 >>> >>> udev (provided by systemd) -> libgudev -> libfprint >>> also: >>> udev-native -> libgudev-native -> libfprint-native -> libfprint >>> >>> Also, please read the TODO part of the patch for systemd >>> and consider it. >> We can simply direct libfprint to run its own target pieces under qemu >> usermode > > The mariadb change to use CROSSCOMPILING_EMULATOR > instead of building mariadb-native was reverted because > qemu is not ominpotent. The same applies to meson > running executables via qemu. No custom wrapper would help. > >> , and avoid all this nasty native stuff. Meson has direct >> support for it, and even if it doesn't work, libfprint can be patched >> to use a custom 'wrapper' for executing the binaries. > > You seem to have omitted reading a large part in the > commit message for the systemd change. > >> TODO? The new natively built systemctl binary may replace >> the script in systemd-systemctl-native, as it supports >> more options and works better than the script. For example, >> the current script does not create the symlinks in >> /etc/systemd/system for WantedBy= and RequiredBy= settings >> and does not support systemctl set-default some.target. > > > This is not relevant to libfprint itself, it's a bug in > Yocto's systemctl replacement script. > Here's the fixed ----8<--- foo.service ----8<--- ... [Install] WantedBy=bar.service bar2.service bar3.service RequiredBy=baz.service baz2.service baz3.service ... ----8<--- foo.service ----8<--- > The symlinks in /etc/systemd/system/bar.service.wants > and /etc/systemd/system/baz.service.requires are not > created for foo.service. Or not all of them. This command fixed them on the installed system by creating all symlinks and not just the first: systemctl reenable foo.service but on a PXE booted image it's a bummer. > This command is also not supported by the Yocto replacement: > > systemctl set-default custom.target > > I have been bitten by both issues. > > > -=-=-=-=-=-=-=-=-=-=-=- > Links: You receive all messages sent to this group. > View/Reply Online (#179330): https://lists.openembedded.org/g/openembedded-core/message/179330 > Mute This Topic: https://lists.openembedded.org/mt/97950749/3617728 > Group Owner: openembedded-core+owner@lists.openembedded.org > Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub [zboszor@gmail.com] > -=-=-=-=-=-=-=-=-=-=-=- >
2023. 03. 30. 16:41 keltezéssel, Alexander Kanavin írta: > On Thu, 30 Mar 2023 at 16:31, Böszörményi Zoltán <zboszor@gmail.com> wrote: >> The mariadb change to use CROSSCOMPILING_EMULATOR >> instead of building mariadb-native was reverted because >> qemu is not ominpotent. The same applies to meson >> running executables via qemu. No custom wrapper would help. > Mariadb's problem was that (for a specific target) it somehow included > CPU instructions that qemu usermode doesn't support, and no one looked > into what those instructions were and why they were produced in the > first place, preferring to revert the whole thing. There are lots of > counter-examples in core where qemu usermode works fine, and has been > working that way for many years, for example gobject introspection, or > all of the scripts in ../scripts/postinst-intercepts/ which run every > time you bake an image, including update_udev_hwdb by the way that > runs target udevadm. I am not maintaining oe-devel, and I do not keep > an eye on things like that, otherwise I would've asked for further > investigation into what crashes and why. > > I do not want to add complex native dependency chains when usermode > would work. It's me who's going to fix them when they break, not you. > >> You seem to have omitted reading a large part in the >> commit message for the systemd change. >> >>> TODO? The new natively built systemctl binary may replace >>> the script in systemd-systemctl-native, as it supports >>> more options and works better than the script. For example, >>> the current script does not create the symlinks in >>> /etc/systemd/system for WantedBy= and RequiredBy= settings >>> and does not support systemctl set-default some.target. > This is a separate concern, and needs to be proposed as such. I can extend this patchset to remove systemd-systemctl-native.bb and use the native build fixing these problems. Then libfprint can just use that as an innocent bystander.
On Thu, 30 Mar 2023 at 16:51, Böszörményi Zoltán <zboszor@gmail.com> wrote: > >>> TODO? The new natively built systemctl binary may replace > >>> the script in systemd-systemctl-native, as it supports > >>> more options and works better than the script. For example, > >>> the current script does not create the symlinks in > >>> /etc/systemd/system for WantedBy= and RequiredBy= settings > >>> and does not support systemctl set-default some.target. > > This is a separate concern, and needs to be proposed as such. > > I can extend this patchset to remove systemd-systemctl-native.bb > and use the native build fixing these problems. That would make a stronger case for it. But you'd likely be asked to make systemd-native really minimal, and ideally produce the systemctl executable and nothing else. Alex
On Thu, 2023-03-30 at 16:10 +0200, Alexander Kanavin wrote: > On Thu, 30 Mar 2023 at 16:08, Zoltan Boszormenyi <zboszor@gmail.com> wrote: > > Please see the MR: > > https://gitlab.freedesktop.org/libfprint/libfprint/-/merge_requests/431 > > > > udev (provided by systemd) -> libgudev -> libfprint > > also: > > udev-native -> libgudev-native -> libfprint-native -> libfprint > > > > Also, please read the TODO part of the patch for systemd > > and consider it. > > We can simply direct libfprint to run its own target pieces under qemu > usermode, and avoid all this nasty native stuff. Meson has direct > support for it, and even if it doesn't work, libfprint can be patched > to use a custom 'wrapper' for executing the binaries. Before we jump to using qemu user mode, the first question has to be do we really need a libgudev native to make libfprint-native work enough to build libprintf. I'd hope the answer is no, we don't as we don't want the build systems udev setup/devices leaking into a target build. We intentionally try and keep native dependencies minimal and needing native udev/systemd versions of things sets of warnings in my mind and in others since it often means we're doing things we shouldn't be or don't actually need. I appreciate there are issues with systemctl-native but those are separate issues and should be addressed as such. Complicating the build dependencies unnecessarily is bad and we should minimise them where possible. it isn't that I didn't read that bit of the patch, I just don't see it as a reason to have this dependency for libfprint. So, I come back to the question of what libfprint needs from libfprint- native and whether it really does need udev for that? Cheers, Richard
2023. 03. 30. 18:39 keltezéssel, Richard Purdie írta: > On Thu, 2023-03-30 at 16:10 +0200, Alexander Kanavin wrote: >> On Thu, 30 Mar 2023 at 16:08, Zoltan Boszormenyi <zboszor@gmail.com> wrote: >>> Please see the MR: >>> https://gitlab.freedesktop.org/libfprint/libfprint/-/merge_requests/431 >>> >>> udev (provided by systemd) -> libgudev -> libfprint >>> also: >>> udev-native -> libgudev-native -> libfprint-native -> libfprint >>> >>> Also, please read the TODO part of the patch for systemd >>> and consider it. >> We can simply direct libfprint to run its own target pieces under qemu >> usermode, and avoid all this nasty native stuff. Meson has direct >> support for it, and even if it doesn't work, libfprint can be patched >> to use a custom 'wrapper' for executing the binaries. > Before we jump to using qemu user mode, the first question has to be do > we really need a libgudev native to make libfprint-native work enough > to build libprintf. > > I'd hope the answer is no, we don't as we don't want the build systems > udev setup/devices leaking into a target build. > > We intentionally try and keep native dependencies minimal and needing > native udev/systemd versions of things sets of warnings in my mind and > in others since it often means we're doing things we shouldn't be or > don't actually need. > > I appreciate there are issues with systemctl-native but those are > separate issues and should be addressed as such. Complicating the build > dependencies unnecessarily is bad and we should minimise them where > possible. it isn't that I didn't read that bit of the patch, I just > don't see it as a reason to have this dependency for libfprint. > > So, I come back to the question of what libfprint needs from libfprint- > native and whether it really does need udev for that? libfprint needs two generator executables to create a set of udev rules and another file called autosuspend.hwdb. Both executables are linked with the complete libfprint library which includes the complete set of drivers that are configured and built. They seem to need this to deduce the list of supported device USB IDs for the current build. Both executables use fpi_ prefixed functions to ask the libfprint library for their goals. In turn, the libfprint library depend on libgudev and libgusb for hotplug and device discovery purposes, respectively. There is also a meson bug that Alexander Kanavin brought to my attention where custom_target() doesn't run executables using the exe_wrapper and I seem to be hitting this with libfprint. I am not a meson expert so if there's a way to replace the executable()+custom_target(capture: true) combination with something more cross-compiler friendly, please don't hold this information back. I couldn't fix this for libfprint 1.0 / fprintd 0.9.0 two years ago and I cannot fix it for the 2.0 beta versions either. In my current in-house solution there's a separate systemd-native.bb recipe but I thought maybe cleaning it up for upstream and modifying the systemd recipe may be acceptable.
2023. 03. 31. 11:39 keltezéssel, Zoltan Boszormenyi via lists.openembedded.org írta: > 2023. 03. 30. 18:39 keltezéssel, Richard Purdie írta: >> On Thu, 2023-03-30 at 16:10 +0200, Alexander Kanavin wrote: >>> On Thu, 30 Mar 2023 at 16:08, Zoltan Boszormenyi <zboszor@gmail.com> wrote: >>>> Please see the MR: >>>> https://gitlab.freedesktop.org/libfprint/libfprint/-/merge_requests/431 >>>> >>>> udev (provided by systemd) -> libgudev -> libfprint >>>> also: >>>> udev-native -> libgudev-native -> libfprint-native -> libfprint >>>> >>>> Also, please read the TODO part of the patch for systemd >>>> and consider it. >>> We can simply direct libfprint to run its own target pieces under qemu >>> usermode, and avoid all this nasty native stuff. Meson has direct >>> support for it, and even if it doesn't work, libfprint can be patched >>> to use a custom 'wrapper' for executing the binaries. >> Before we jump to using qemu user mode, the first question has to be do >> we really need a libgudev native to make libfprint-native work enough >> to build libprintf. >> >> I'd hope the answer is no, we don't as we don't want the build systems >> udev setup/devices leaking into a target build. >> >> We intentionally try and keep native dependencies minimal and needing >> native udev/systemd versions of things sets of warnings in my mind and >> in others since it often means we're doing things we shouldn't be or >> don't actually need. >> >> I appreciate there are issues with systemctl-native but those are >> separate issues and should be addressed as such. Complicating the build >> dependencies unnecessarily is bad and we should minimise them where >> possible. it isn't that I didn't read that bit of the patch, I just >> don't see it as a reason to have this dependency for libfprint. >> >> So, I come back to the question of what libfprint needs from libfprint- >> native and whether it really does need udev for that? > > libfprint needs two generator executables to create > a set of udev rules and another file called autosuspend.hwdb. > > Both executables are linked with the complete libfprint library > which includes the complete set of drivers that are configured > and built. They seem to need this to deduce the list of supported > device USB IDs for the current build. Both executables use fpi_ > prefixed functions to ask the libfprint library for their goals. > > In turn, the libfprint library depend on libgudev and libgusb > for hotplug and device discovery purposes, respectively. > > There is also a meson bug that Alexander Kanavin brought > to my attention where custom_target() doesn't run executables > using the exe_wrapper and I seem to be hitting this with > libfprint. > > I am not a meson expert so if there's a way to replace > the executable()+custom_target(capture: true) combination > with something more cross-compiler friendly, please don't hold > this information back. > > I couldn't fix this for libfprint 1.0 / fprintd 0.9.0 two years ago > and I cannot fix it for the 2.0 beta versions either. > In my current in-house solution there's a separate systemd-native.bb > recipe but I thought maybe cleaning it up for upstream and > modifying the systemd recipe may be acceptable. Regarding my attempts to fix this: I have looked into making these executables use "native: true" but it would need duplicating all the driver and library targets with "native: true", too, and it seemed to be too intrusive. Also, it doesn't eliminate the native dependency chain, building these executables would still need systemd-native, libgudev-native and libgusb-native. > > > -=-=-=-=-=-=-=-=-=-=-=- > Links: You receive all messages sent to this group. > View/Reply Online (#179406): https://lists.openembedded.org/g/openembedded-core/message/179406 > Mute This Topic: https://lists.openembedded.org/mt/97950749/3617728 > Group Owner: openembedded-core+owner@lists.openembedded.org > Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub [zboszor@gmail.com] > -=-=-=-=-=-=-=-=-=-=-=- >
On 31 Mar 2023, at 10:46, Zoltan Boszormenyi via lists.openembedded.org <zboszor=gmail.com@lists.openembedded.org> wrote: > Regarding my attempts to fix this: I have looked into > making these executables use "native: true" but it > would need duplicating all the driver and library targets > with "native: true", too, and it seemed to be too intrusive. > Also, it doesn't eliminate the native dependency chain, > building these executables would still need systemd-native, > libgudev-native and libgusb-native. I’d say this is still the correct solution, although the ‘native’ libfprint would be as lean as possible and not the entire library. Ross
2023. 04. 03. 12:59 keltezéssel, Ross Burton írta: > On 31 Mar 2023, at 10:46, Zoltan Boszormenyi via lists.openembedded.org <zboszor=gmail.com@lists.openembedded.org> wrote: >> Regarding my attempts to fix this: I have looked into >> making these executables use "native: true" but it >> would need duplicating all the driver and library targets >> with "native: true", too, and it seemed to be too intrusive. >> Also, it doesn't eliminate the native dependency chain, >> building these executables would still need systemd-native, >> libgudev-native and libgusb-native. > I’d say this is still the correct solution, although the ‘native’ libfprint would be as lean as possible and not the entire library. Hopefully someone can solve this bug soon: https://github.com/mesonbuild/meson/issues/11029 FWIW, I commented about the problem but the meson codebase is over my head so I can't fix it. Then libfrint (and probably other meson bases recipes in Yocto) won't need a native part. Until then, please ignore this patchset and the other one for meta-oe. I will stick to my in-house solution for now.
> On 3 Apr 2023, at 14:00, Böszörményi Zoltán <zboszor@gmail.com> wrote: > > 2023. 04. 03. 12:59 keltezéssel, Ross Burton írta: >> On 31 Mar 2023, at 10:46, Zoltan Boszormenyi via lists.openembedded.org <zboszor=gmail.com@lists.openembedded.org> wrote: >>> Regarding my attempts to fix this: I have looked into >>> making these executables use "native: true" but it >>> would need duplicating all the driver and library targets >>> with "native: true", too, and it seemed to be too intrusive. >>> Also, it doesn't eliminate the native dependency chain, >>> building these executables would still need systemd-native, >>> libgudev-native and libgusb-native. >> I’d say this is still the correct solution, although the ‘native’ libfprint would be as lean as possible and not the entire library. > > Hopefully someone can solve this bug soon: > https://github.com/mesonbuild/meson/issues/11029 > FWIW, I commented about the problem but the meson > codebase is over my head so I can't fix it. > > Then libfrint (and probably other meson bases recipes > in Yocto) won't need a native part. FWIW I just took a stripped down copy of your libfprint recipe and it happily built using qemu’s exe wrapper in oe-core master. So, yes, it would be good if the build used native instead of relying on qemu, but it’s not a hard blocker. We should definitely be doing AB runs of poky with the exe wrapper disabled to exercise those codepaths. Ross
2023. 04. 03. 21:52 keltezéssel, Ross Burton írta: > >> On 3 Apr 2023, at 14:00, Böszörményi Zoltán <zboszor@gmail.com> wrote: >> >> 2023. 04. 03. 12:59 keltezéssel, Ross Burton írta: >>> On 31 Mar 2023, at 10:46, Zoltan Boszormenyi via lists.openembedded.org <zboszor=gmail.com@lists.openembedded.org> wrote: >>>> Regarding my attempts to fix this: I have looked into >>>> making these executables use "native: true" but it >>>> would need duplicating all the driver and library targets >>>> with "native: true", too, and it seemed to be too intrusive. >>>> Also, it doesn't eliminate the native dependency chain, >>>> building these executables would still need systemd-native, >>>> libgudev-native and libgusb-native. >>> I’d say this is still the correct solution, although the ‘native’ libfprint would be as lean as possible and not the entire library. >> Hopefully someone can solve this bug soon: >> https://github.com/mesonbuild/meson/issues/11029 >> FWIW, I commented about the problem but the meson >> codebase is over my head so I can't fix it. >> >> Then libfrint (and probably other meson bases recipes >> in Yocto) won't need a native part. > FWIW I just took a stripped down copy of your libfprint recipe and it happily built using qemu’s exe wrapper in oe-core master. After a fresh "repo sync" my libfprint build still fails with the stripped down copy that doesn't want the native dependency. Please share the changes of yours against the libfprint recipe because clearly I am doing something wrong which is not obvious to me. Or there is something subtle going on. For example, does this stripped down copy of the libfprint recipe still build for you if meta-clang is also used? > So, yes, it would be good if the build used native instead of relying on qemu, but it’s not a hard blocker. We should definitely be doing AB runs of poky with the exe wrapper disabled to exercise those codepaths. > > Ross
diff --git a/meta/recipes-core/systemd/systemd_253.1.bb b/meta/recipes-core/systemd/systemd_253.1.bb index 9c2b96d3c1..c73b5e05b3 100644 --- a/meta/recipes-core/systemd/systemd_253.1.bb +++ b/meta/recipes-core/systemd/systemd_253.1.bb @@ -6,6 +6,8 @@ PE = "1" DEPENDS = "intltool-native gperf-native libcap util-linux python3-jinja2-native" +BBCLASSEXTEND = "native" + SECTION = "base/shell" inherit useradd pkgconfig meson perlnative update-rc.d update-alternatives qemu systemd gettext bash-completion manpages features_check @@ -60,7 +62,7 @@ PAM_PLUGINS = " \ pam-plugin-namespace \ " -PACKAGECONFIG ??= " \ +PACKAGECONFIG:class-target ??= " \ ${@bb.utils.filter('DISTRO_FEATURES', 'acl audit efi ldconfig pam selinux smack usrmerge polkit seccomp', d)} \ ${@bb.utils.contains('DISTRO_FEATURES', 'wifi', 'rfkill', '', d)} \ ${@bb.utils.contains('DISTRO_FEATURES', 'x11', 'xkbcommon', '', d)} \ @@ -197,7 +199,7 @@ PACKAGECONFIG[selinux] = "-Dselinux=true,-Dselinux=false,libselinux,initscripts- PACKAGECONFIG[smack] = "-Dsmack=true,-Dsmack=false" PACKAGECONFIG[sysext] = "-Dsysext=true, -Dsysext=false" PACKAGECONFIG[sysusers] = "-Dsysusers=true,-Dsysusers=false" -PACKAGECONFIG[sysvinit] = "-Dsysvinit-path=${sysconfdir}/init.d -Dsysvrcnd-path=${sysconfdir},-Dsysvinit-path= -Dsysvrcnd-path=,,systemd-compat-units update-rc.d" +PACKAGECONFIG[sysvinit] = "-Dsysvinit-path=${sysconfdir}/init.d -Dsysvrcnd-path=${sysconfdir},-Dsysvinit-path= -Dsysvrcnd-path=,,${@'systemd-compat-units update-rc.d' if d.getVar('PN') == d.getVar('BPN') else ''}" # When enabled use reproducble build timestamp if set as time epoch, # or build time if not. When disabled, time epoch is unset. def build_epoch(d): @@ -211,7 +213,7 @@ PACKAGECONFIG[sbinmerge] = "-Dsplit-bin=false,-Dsplit-bin=true" PACKAGECONFIG[userdb] = "-Duserdb=true,-Duserdb=false" PACKAGECONFIG[utmp] = "-Dutmp=true,-Dutmp=false" PACKAGECONFIG[valgrind] = "-DVALGRIND=1,,valgrind" -PACKAGECONFIG[vconsole] = "-Dvconsole=true,-Dvconsole=false,,${PN}-vconsole-setup" +PACKAGECONFIG[vconsole] = "-Dvconsole=true,-Dvconsole=false,,${BPN}-vconsole-setup" PACKAGECONFIG[wheel-group] = "-Dwheel-group=true, -Dwheel-group=false" PACKAGECONFIG[xdg-autostart] = "-Dxdg-autostart=true,-Dxdg-autostart=false" # Verify keymaps on locale change @@ -364,6 +366,12 @@ do_install() { fi } +do_install:append:class-native () { + rm -f ${D}${bindir}/systemctl + ln -sf ..${base_bindir_native}/udevadm ${D}${base_sbindir}/udevadm + ln -sf ..${systemd_unitdir#${rootprefix}}/systemd-udevd ${D}${base_sbindir}/udevd +} + python populate_packages:prepend (){ systemdlibdir = d.getVar("rootlibdir") do_split_packages(d, systemdlibdir, r'^lib(.*)\.so\.*', 'lib%s', 'Systemd %s library', extra_depends='', allow_links=True) @@ -670,10 +678,10 @@ FILES:${PN} = " ${base_bindir}/* \ FILES:${PN}-dev += "${base_libdir}/security/*.la ${datadir}/dbus-1/interfaces/ ${sysconfdir}/rpm/macros.systemd" RDEPENDS:${PN} += "kmod dbus util-linux-mount util-linux-umount udev (= ${EXTENDPKGV}) systemd-udev-rules util-linux-agetty util-linux-fsck" -RDEPENDS:${PN} += "${@bb.utils.contains('PACKAGECONFIG', 'serial-getty-generator', '', 'systemd-serialgetty', d)}" -RDEPENDS:${PN} += "volatile-binds" +RDEPENDS:${PN}:class-target += "${@bb.utils.contains('PACKAGECONFIG', 'serial-getty-generator', '', 'systemd-serialgetty', d)}" +RDEPENDS:${PN}:class-target += "volatile-binds" -RRECOMMENDS:${PN} += "systemd-extra-utils \ +RRECOMMENDS:${PN}:class-target += "systemd-extra-utils \ udev-hwdb \ e2fsprogs-e2fsck \ kernel-module-autofs4 kernel-module-unix kernel-module-ipv6 kernel-module-sch-fq-codel \
systemd-native is a dependency for libgudev-native, which in turn is a dependency for libfprint-native, which is needed to build libfprint. The native systemctl binary is removed so it doesn't conflict with the script from systemd-systemctl-native. TODO? The new natively built systemctl binary may replace the script in systemd-systemctl-native, as it supports more options and works better than the script. For example, the current script does not create the symlinks in /etc/systemd/system for WantedBy= and RequiredBy= settings and does not support systemctl set-default some.target. Signed-off-by: Zoltán Böszörményi <zboszor@gmail.com> --- meta/recipes-core/systemd/systemd_253.1.bb | 20 ++++++++++++++------ 1 file changed, 14 insertions(+), 6 deletions(-)