Starting with the v430 driver, the new kernel driver callback /proc/driver/nvidia/suspend was added to save/restore video memory on suspend/hibernate/resume to mitigate FBO corruption on resume: https://download.nvidia.com/XFree86/Linux-x86_64/430.09/README/powermanagement.html The driver package contains systemd units and a script.
Created attachment 582756 [details] working (for me) nvidia-sleep.sh The nvidia-sleep.sh script included in the driver package is unusable, attaching a much simpler, working one which using chvt and fgconsole for vt switching.
Comment on attachment 582756 [details] working (for me) nvidia-sleep.sh Comments below the lines they relate to. >#!/bin/bash Nothing in particular should require bash specific features. > >RUN_DIR="/var/run/nvidia-sleep" 1) Shouldn't that be /run/... ? 2) It doesn't seem to make sense to me to create a directory in which a single temporary file will be placed. >XORG_VT_FILE="${RUN_DIR}"/Xorg.vt_number > >case "$1" in > suspend|hibernate) > /bin/mkdir -p "${RUN_DIR}" > /usr/bin/fgconsole > "${XORG_VT_FILE}" > /usr/bin/chvt 63 > if [[ $? -ne 0 ]]; then This should be replaced with a non-bash test. > exit $? > fi > /bin/echo "$1" > /proc/driver/nvidia/suspend > exit $? > ;; > resume) > /bin/echo "$1" > /proc/driver/nvidia/suspend 1) Looks like the line ends with a space. 2) What if this fails? > # > # Check if Xorg was determined to be running at the time > # of suspend, and whether its VT was recorded. If so, > # attempt to switch back to this VT. > # > if [[ -f "${XORG_VT_FILE}" ]]; then > XORG_PID=$(cat "${XORG_VT_FILE}") > /bin/rm "${XORG_VT_FILE}" > /usr/bin/chvt "${XORG_PID}" > fi > exit 0 > ;; > *) > exit 1 >esac
(In reply to Jeroen Roovers from comment #2) > Comment on attachment 582756 [details] > > resume) > > /bin/echo "$1" > /proc/driver/nvidia/suspend > 2) What if this fails? If what fails? echoing? Please clarify. echoing "resume" to that handle is instructing the driver to restore the video memory from tmp file. If that fails, the result would be the same corruption as it is now without that function. But there's no call back to get the result, AFAIK. Would be informational only, anyway, since you can literally see that it didn't work. This is just the original script, I basically only replaced a lot of crap with that fgconsole line to get it to work at all, call it WIP.
On second thought: those handles of course only exist if the nvidia driver is loaded so there has to be a check at the start of the script so it bails out if that isn't the case.
A little ToDo-list for nvidia-sleep.sh besides the changes from post #2: - make sure the handle /proc/driver/nvidia/suspend exists, i.e. the (correct) driver is loaded - check /proc/driver/nvidia/params that the correct module parameter NVreg_PreserveVideoMemoryAllocations=1 is set - make sure no Xorg is started on vt63, otherwise use other vt# and make note of it. - writing to that handle only ever errors out if anything else than suspend,resume or hibernate is written to it, shouldn't happen due to switch case and previous check.
any news? I added a package to my overlay which installs missed stuff right from driver's samples. https://github.com/rion-overlay/rion-overlay/tree/master/sys-apps/nvidia-systemd-pm (layman -a rion) But I believe this stuff has to be installed by nvidia-drivers package.
The bug has been closed via the following commit(s): https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=19c0b0218d3848798c162725167be753c080621e commit 19c0b0218d3848798c162725167be753c080621e Author: Jeroen Roovers <jer@gentoo.org> AuthorDate: 2020-04-20 08:29:46 +0000 Commit: Jeroen Roovers <jer@gentoo.org> CommitDate: 2020-04-20 10:14:42 +0000 x11-drivers/nvidia-drivers: Install nvidia-sleep.sh Package-Manager: Portage-2.3.99, Repoman-2.3.22 Closes: https://bugs.gentoo.org/689588 Signed-off-by: Jeroen Roovers <jer@gentoo.org> .../nvidia-drivers/nvidia-drivers-430.64-r2.ebuild | 556 ++++++++++++++++++++ .../nvidia-drivers/nvidia-drivers-435.21-r2.ebuild | 574 ++++++++++++++++++++ .../nvidia-drivers/nvidia-drivers-440.82-r1.ebuild | 580 +++++++++++++++++++++ 3 files changed, 1710 insertions(+)
I updated my ebuild accordingly. But why not install everything (including systemd service files and one additional helper script) from nvidia-drivers?
(In reply to Sergey Ilinykh from comment #8) > I updated my ebuild accordingly. I have no idea what that means. > But why not install everything (including systemd service files I assumed it installed those already. Reopening accordingly. > and one additional helper script) from nvidia-drivers? I have no idea what that means.
I already referenced my ebuild above (https://github.com/rion-overlay/rion-overlay/tree/master/sys-apps/nvidia-systemd-pm ). Its src_install has exactly what I mean. Thanks in advance.
Comment on attachment 582756 [details] working (for me) nvidia-sleep.sh This is basically the same as the upstream nvidia-sleep.sh.
And well according to the docs it tries to keep nvidia driver state in /tmp which iirc is tmpfs for Gentoo. Particularly my laptop has enough RAM for this. But I can't say for everyone else. So maybe it has to be somehow configurable. Or at least some bold warning has to be shown.
Hi systemd@. nvidia-drivers packages three systemd files which need to be installed. ~/portage/x11-drivers/nvidia-drivers-440.82/work $ find -name '*service' ./nvidia-hibernate.service ./nvidia-resume.service ./nvidia-suspend.service systemd.eclass dislikes those file names. How do I fix that?
(In reply to Sergey Ilinykh from comment #10) > I already referenced my ebuild above > (https://github.com/rion-overlay/rion-overlay/tree/master/sys-apps/nvidia- > systemd-pm ). Its src_install has exactly what I mean. > Thanks in advance. Please attach anything you want to discuss here.
> Please attach anything you want to discuss here. src_install() { insinto /lib/systemd/system doins *.service exeinto /lib/systemd/system-sleep doexe nvidia echo "options nvidia NVreg_PreserveVideoMemoryAllocations=1" > nvidia-pm.conf insinto /etc/modprobe.d/ doins nvidia-pm.conf ewarn "To enable nvidia sleep services execute next commands:" ewarn " systemctl enable nvidia-suspend.service" ewarn " systemctl enable nvidia-hibernate.service" ewarn " systemctl enable nvidia-resume.service" ewarn "Note nvidia kernel module reload (or reboot) is required" ewarn "to accept new driver options." ewarn "More details at /usr/share/doc/nvidia-drivers-${PV}/html/powermanagement.html" }
(In reply to Sergey Ilinykh from comment #12) > And well according to the docs it tries to keep nvidia driver state in /tmp > which iirc is tmpfs for Gentoo. > Particularly my laptop has enough RAM for this. But I can't say for everyone > else. So maybe it has to be somehow configurable. Or at least some bold > warning has to be shown. From Nvidia docs: By default, these files are created in /tmp, but this location can be changed with the TemporaryFilePath kernel module parameter, e.g. TemporaryFilePath=/run Don't know if all those messages about enabling the systemd files and kernel module parameters should be in the emerge messages or rather a link to the doc (see OP) since there are more FS requirements to be met when a different location is set.
(In reply to Jeroen Roovers from comment #13) > systemd.eclass dislikes those file names. How do I fix that? I'm not sure what this means. How are you calling systemd.eclass, and what unexpected behavior do you experience?
(In reply to Sergey Ilinykh from comment #15) > > Please attach anything you want to discuss here. > > src_install() { Thanks! > insinto /lib/systemd/system ^^^There is probably some voodoo in systemd.eclass that should help decide what "/lib/systemd/system" should be. > doins *.service > exeinto /lib/systemd/system-sleep > doexe nvidia > echo "options nvidia NVreg_PreserveVideoMemoryAllocations=1" > > nvidia-pm.conf > insinto /etc/modprobe.d/ > doins nvidia-pm.conf ^^^We already install /etc/modprobe.d/nvidia.conf to sets kernel module options, so I wonder if we should be able to simply add this line to nvidia.conf instead. > ewarn "To enable nvidia sleep services execute next commands:" > ewarn " systemctl enable nvidia-suspend.service" > ewarn " systemctl enable nvidia-hibernate.service" > ewarn " systemctl enable nvidia-resume.service" ^^^I guess there no way to make this easier?
(In reply to Mike Gilbert from comment #17) > (In reply to Jeroen Roovers from comment #13) > > systemd.eclass dislikes those file names. How do I fix that? > > I'm not sure what this means. How are you calling systemd.eclass, and what > unexpected behavior do you experience? >>> Install x11-drivers/nvidia-drivers-430.64-r2 into /home/jer/portage/x11-drivers/nvidia-drivers-430.64-r2/image * ERROR: x11-drivers/nvidia-drivers-430.64-r2::gentoo failed (install phase): * Source file needs .conf suffix * * Call stack: * ebuild.sh, line 125: Called src_install * environment, line 4182: Called systemd_install_serviced 'nvidia-hibernate.service' * environment, line 4464: Called die * The specific snippet of code: * [[ ${src} == *.conf ]] || die "Source file needs .conf suffix"; Apparently this is invalid. :-)
> We already install /etc/modprobe.d/nvidia.conf to sets kernel module options, so I wonder if we should be able to simply add this line to nvidia.conf instead. sounds good to me.
(In reply to Jeroen Roovers from comment #19) Use "systemd_dounit" instead. To use OpenRC as an analogy: systemd_dounit is like "doinitd". It installs a systemd unit to /lib/systemd/system. Said unit may be a service definition (*.service), similar in function to an init script. Other unit types are also possible. systemd_install_serviced is like "doconfd". It installs a unit override config file as /etc/systemd/system/*.service.d/*.conf. This is optional, and only rarely used when there are settings that a sysadmin would typically customize and this is not possible through a domain-specific config file. As an aside: At some point, I might rename "systemd_install_serviced" to "systemd_do_unit.d", to match the portage helper naming convention and to make it obvious that the function can be used for units other than services.
(In reply to Sergey Ilinykh from comment #20) > > We already install /etc/modprobe.d/nvidia.conf to sets kernel module options, so I wonder if we should be able to simply add this line to nvidia.conf instead. > > sounds good to me. OK, does it work to add that separately or does that variable need to be included on the same line as the other variables?
(In reply to Mike Gilbert from comment #21) > (In reply to Jeroen Roovers from comment #19) > > Use "systemd_dounit" instead. Right, that appears to work.
(In reply to Sergey Ilinykh from comment #15) > ewarn "More details at > /usr/share/doc/nvidia-drivers-${PV}/html/powermanagement.html" > } That document also mentions TemporaryFilePath, and yet (using 440.82): [ 13.351291] nvidia: unknown parameter 'TemporaryFilePath' ignored
modinfo nvidia tells the correct name is NVreg_TemporaryFilePath
(In reply to Maik from comment #25) > modinfo nvidia tells the correct name is > NVreg_TemporaryFilePath Yes, I'll try to integrate that.
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=613c3dc0e336ee65e262d082084d9a05da84a849 commit 613c3dc0e336ee65e262d082084d9a05da84a849 Author: Jeroen Roovers <jer@gentoo.org> AuthorDate: 2020-04-29 08:07:49 +0000 Commit: Jeroen Roovers <jer@gentoo.org> CommitDate: 2020-04-29 08:08:19 +0000 x11-drivers/nvidia-drivers: Install systemd service files Package-Manager: Portage-2.3.99, Repoman-2.3.22 Bug: https://bugs.gentoo.org/689588 Signed-off-by: Jeroen Roovers <jer@gentoo.org> x11-drivers/nvidia-drivers/files/nvidia-430.conf | 20 + .../nvidia-drivers/nvidia-drivers-430.64-r3.ebuild | 566 ++++++++++++++++++++ .../nvidia-drivers/nvidia-drivers-435.21-r3.ebuild | 584 ++++++++++++++++++++ .../nvidia-drivers/nvidia-drivers-440.82-r2.ebuild | 590 +++++++++++++++++++++ 4 files changed, 1760 insertions(+)
(In reply to Larry the Git Cow from comment #27) > The bug has been referenced in the following commit(s): > > https://gitweb.gentoo.org/repo/gentoo.git/commit/ > ?id=613c3dc0e336ee65e262d082084d9a05da84a849 > > commit 613c3dc0e336ee65e262d082084d9a05da84a849 > Author: Jeroen Roovers <jer@gentoo.org> > AuthorDate: 2020-04-29 08:07:49 +0000 > Commit: Jeroen Roovers <jer@gentoo.org> > CommitDate: 2020-04-29 08:08:19 +0000 > > x11-drivers/nvidia-drivers: Install systemd service files > > Package-Manager: Portage-2.3.99, Repoman-2.3.22 > Bug: https://bugs.gentoo.org/689588 > Signed-off-by: Jeroen Roovers <jer@gentoo.org> > > x11-drivers/nvidia-drivers/files/nvidia-430.conf | 20 + > .../nvidia-drivers/nvidia-drivers-430.64-r3.ebuild | 566 > ++++++++++++++++++++ > .../nvidia-drivers/nvidia-drivers-435.21-r3.ebuild | 584 > ++++++++++++++++++++ > .../nvidia-drivers/nvidia-drivers-440.82-r2.ebuild | 590 > +++++++++++++++++++++ > 4 files changed, 1760 insertions(+) So does that work? Anyone?
The file /lib/systemd/system-sleep/nvidia seems to be missing Comment #18 referenced it > exeinto /lib/systemd/system-sleep > doexe nvidia but now I can't find it in the ebuild x11-drivers/nvidia-drivers-440.82-r2
Some general oddity: since I had this running since july 2019, I already had /usr/bin/nvidia-sleep.sh. This file now got overwritten with the one from ebuild without any collision warning. Bug in portage?
(In reply to Maik from comment #30) Portage probably warned you it was overwriting the file because the installed file was not owned by any package.
(In reply to Maik from comment #29) > The file /lib/systemd/system-sleep/nvidia seems to be missing > > Comment #18 referenced it > > exeinto /lib/systemd/system-sleep > > doexe nvidia > > but now I can't find it in the ebuild x11-drivers/nvidia-drivers-440.82-r2 Oh I missed that. Sorry.
(In reply to Mike Gilbert from comment #31) > (In reply to Maik from comment #30) > > Portage probably warned you it was overwriting the file because the > installed file was not owned by any package. Nope, no warning at all, nothing. Or does portage now check if contents/hash of unowned collision match?
So that would be $(systemd_get_utildir)/system-sleep then?
(In reply to Jeroen Roovers from comment #34) > So that would be $(systemd_get_utildir)/system-sleep then? Technically, we should add an eclass function to expose the systemdsleepdir pkgconfig variable, but $(systemd_get_utildir)/system-sleep will work for now.
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=b622e2f86a2be4087fd4c6b6444380715ca4375b commit b622e2f86a2be4087fd4c6b6444380715ca4375b Author: Jeroen Roovers <jer@gentoo.org> AuthorDate: 2020-04-30 20:03:48 +0000 Commit: Jeroen Roovers <jer@gentoo.org> CommitDate: 2020-05-01 04:08:22 +0000 x11-drivers/nvidia-drivers: Install /lib/systemd/system-sleep/nvidia Package-Manager: Portage-2.3.99, Repoman-2.3.22 Bug: https://bugs.gentoo.org/689588 Signed-off-by: Jeroen Roovers <jer@gentoo.org> ...{nvidia-drivers-430.64-r3.ebuild => nvidia-drivers-430.64-r4.ebuild} | 2 ++ ...{nvidia-drivers-435.21-r3.ebuild => nvidia-drivers-435.21-r4.ebuild} | 2 ++ ...{nvidia-drivers-440.82-r2.ebuild => nvidia-drivers-440.82-r3.ebuild} | 2 ++ 3 files changed, 6 insertions(+)
Thank you, works now. Closing this.
Thanks.
Apparently calling systemd_get_utildir() this way is incorrect because it prefixes PREFIX. Or at least that is what [0] implies. [0] https://qa-reports.gentoo.org/output/gentoo-ci/88e4371c/output.html#x11-drivers/nvidia-drivers
The bug has been closed via the following commit(s): https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=7e7c112c7d334a19e0048054a9ce888964b50ae3 commit 7e7c112c7d334a19e0048054a9ce888964b50ae3 Author: Jeroen Roovers <jer@gentoo.org> AuthorDate: 2020-06-27 04:14:36 +0000 Commit: Jeroen Roovers <jer@gentoo.org> CommitDate: 2020-06-27 04:22:42 +0000 x11-drivers/nvidia-drivers: Do not use systemd_get_utildir() Package-Manager: Portage-2.3.103, Repoman-2.3.23 Closes: https://bugs.gentoo.org/689588 Signed-off-by: Jeroen Roovers <jer@gentoo.org> .../nvidia-drivers/nvidia-drivers-430.64-r5.ebuild | 564 ++++++++++++++++++++ .../nvidia-drivers/nvidia-drivers-435.21-r5.ebuild | 582 ++++++++++++++++++++ .../nvidia-drivers-440.100-r1.ebuild | 585 ++++++++++++++++++++ .../nvidia-drivers/nvidia-drivers-440.82-r3.ebuild | 2 +- .../nvidia-drivers/nvidia-drivers-440.82-r4.ebuild | 588 +++++++++++++++++++++ .../nvidia-drivers/nvidia-drivers-450.51-r1.ebuild | 586 ++++++++++++++++++++ 6 files changed, 2906 insertions(+), 1 deletion(-)
