nVidia graphics repository out of sync?
Hi updated Tumbleweed a few days ago to build 20210425. Reboot ended in a command prompt rather than firing up KDE. Using Kernel 5.11.4 made it work again. Spent some time today in figuring out what the problem is and it seems the nVidia repository is out of sync - the kernel used there is still at 5.11.11 rather than 5.11.16. br Juergen
On Sunday 02 May 2021, Mailings wrote:
I'm currently typing this from TW 20210427 running 5.11.16-1-default along with the nvidia proprietary 460.73 driver (installed by the "easy" method). If you are also on the 460 driver, my guess would be that your system's nvidia kernel module has not been rebuilt for 5.11.16. The module is normally automatically rebuilt as part of the TW zypper dup, but this sometimes doesn't happen. This has happened to me a few times recently, in these cases I found that a forced reinstall of the driver's kernel module followed by a reboot usually fixes the issue: % zypper in -f nvidia-gfxG05-kmp-default % reboot I've not tracked down exactly why sometimes fails, perhaps it is some chicken and egg circumstance. I did notice on one occasion I seemed to be missing a kernel-source install for my current kernel - but I'm not clear on whether this was the cause of the issue. Cheers, Michael
Hi Michael, Andrei thanks for your swift responses! One correction: I mentioned the wrong kernel version - its 5.11.15 for build 20210425. Michael - I've tried your suggestion but it failed complaining about a "mismatch" in nvidia-gfxG05-kmp-default-460.73.01_k5.11.11_1-37.1.x86_64.rpm [1](The file I've tee'd the output to got overwritten so regrettably I can't provide more details, grmbl) The kernel source install for 5.11.15 looked OK to me. Your comment and Andrei's point stating the nvidia modules are compiled made me try to update to the latest available build 20210429 which employs kernel 5.12.0. (Not knowing much about the "nvidia issue" I assumed nvidia are trying to keep some stuff from others looking into, thus forcing download of some pre-complied objects rather than to compile in situ) After the change to 5.12.0 all is fine again, I'm writing this mail on the system. One observation: I was following the update closely (which I usually don't do ;-)) and was concerned when I saw: ------------------------------------------------------------------------------------------------------------ .......BTF [M] /usr/src/kernel-modules/nvidia-460.73.01-default/nvidia.ko Skipping BTF generation for /usr/src/kernel-modules/nvidia-460.73.01-default/ nvidia.ko due to unavailability of vmlinux make[2]: Verzeichnis „/usr/src/linux-5.12.0-1-obj/x86_64/default“ wird verlassen make[1]: Leaving directory '/usr/src/linux-5.12.0-1' ld -r -o nv-linux.o \ nvidia.mod.o nvidia/nv-interface.o ld: cannot find nvidia/nv-interface.o: No such file or directory make: *** [Makefile:113: nv-linux.o] Error 1 / depmod: WARNING: could not open modules.order at /lib/modules/5.12.0-1-default: No such file or directory depmod: WARNING: could not open modules.builtin at /lib/modules/5.12.0-1- default: No such file or directory Modprobe blacklist files have been created at /etc/modprobe.d to prevent Nouveau from loading. This can be reverted by deleting /etc/modprobe.d/nvidia-*.conf. *** Reboot your computer and verify that the NVIDIA graphics driver can be loaded. *** grep: /etc/sysconfig/kernel: No such file or directory warning: %triggerin(nvidia-gfxG05-kmp-default-460.73.01_k5.11.11_1-37.1.x86_64) scriptlet failed, exit status 1 ........ --------------------------------------------------------------------------------------------------------------- Any idea whether this could turn into something serious or can it be ignored? br Juergen Am Sonntag, 2. Mai 2021, 07:13:02 CEST schrieb Andrei Borzenkov:
On Monday 03 May 2021, J. Merkel wrote:
If I grep /var/log/zypp/history for the above errors I can find all of them from various past updates all the way back to kernel 5.7.7-1-default. I'm not sure whether any of them are associated with past nvidia update issues. I've certainly not been paying any attention to them. In the past should an update cause my desktop to fail to boot to the sddm login screen, I have most often tried a forced reinstall of everything currently installed from the nvidia repo. More recently I have just tried rebuilding the module after someone here suggested that this was all that is necessary - perhaps that is insufficient in some cases, I'm not sure. Force reinstalls have often done the trick. Otherwise I just set my system to boot to the previous kernel - I don't think that has ever failed. I then raise an issue here, and wait for some suggestions or explanation. Often the explanation is along the lines the kernel API's have been changed and we're waiting for nvidia to catchup. It isn't essential that my desktop runs the very latest kernel, so this works fine for me. Michael
Hi Michael, Andrei thanks for your swift responses! One correction: I mentioned the wrong kernel version - its 5.11.15 for build 20210425. Michael - I've tried your suggestion but it failed complaining about a - in my words - "mismatch" with nvidia-gfxG05-kmp-default-460.73.01_k5.11.11_1-37.1.x86_64.rpm (The file I've tee'd the output to got overwritten so regrettably I can't provide more details, grmbl) The kernel-source for 5.11.15 is present. Your comment and also Andrei's point stating the nvidia modules are compiled on the machine made me try to update to the latest available build 20210429 which employs kernel 5.12.0. (Not knowing much about the "nvidia issue" I assumed nvidia are trying to keep some stuff from others looking into, thus forcing download of some pre-complied objects rather than to compile in situ) After the change to 5.12.0 all is fine again - I'm writing this mail on the system. One observation: I was following the update closely (which I usually don't do ;-)) and was concerned when I saw: ------------------------------------------------------------------------------------------------------------ .......BTF [M] /usr/src/kernel-modules/nvidia-460.73.01-default/nvidia.ko Skipping BTF generation for /usr/src/kernel-modules/nvidia-460.73.01-default/ nvidia.ko due to unavailability of vmlinux make[2]: Verzeichnis „/usr/src/linux-5.12.0-1-obj/x86_64/default“ wird verlassen make[1]: Leaving directory '/usr/src/linux-5.12.0-1' ld -r -o nv-linux.o \ nvidia.mod.o nvidia/nv-interface.o ld: cannot find nvidia/nv-interface.o: No such file or directory make: *** [Makefile:113: nv-linux.o] Error 1 / depmod: WARNING: could not open modules.order at /lib/modules/5.12.0-1-default: No such file or directory depmod: WARNING: could not open modules.builtin at /lib/modules/5.12.0-1- default: No such file or directory Modprobe blacklist files have been created at /etc/modprobe.d to prevent Nouveau from loading. This can be reverted by deleting /etc/modprobe.d/nvidia-*.conf. *** Reboot your computer and verify that the NVIDIA graphics driver can be loaded. *** grep: /etc/sysconfig/kernel: No such file or directory warning: %triggerin(nvidia-gfxG05-kmp-default-460.73.01_k5.11.11_1-37.1.x86_64) scriptlet failed, exit status 1 ........ --------------------------------------------------------------------------------------------------------------- Any idea whether this could turn into something serious or can it be ignored? br Juergen Am Sonntag, 2. Mai 2021, 07:13:02 CEST schrieb Andrei Borzenkov:
On 02.05.2021 04:32, Michael Hamilton wrote:
On Sunday 02 May 2021, Mailings wrote:
On 02.05.2021 18:18, Mailings wrote:
Yes, I can reproduce it. It is due to this line of nvidia/nvidia.Kbuild: always += $(NVIDIA_INTERFACE) Changing it to always-m += $(NVIDIA_INTERFACE) allows build to complete. I am not exactly sure what this part does and why it fails for 5.12 and not for 5.11, but you should open bug report on openSUSE bugzilla and possibly report it nvidia.
This is cosmetic.
This is also not a problem, although SPEC file could be improved to avoid this scary output.
On 02.05.2021 20:20, Andrei Borzenkov wrote:
OK, apparently older kernels accepted "always" and 5.12 no more. This is usual case of nVidia build being broken with kernel updates. tw:/usr/src/kernel-modules/nvidia-460.73.01-default # make -f Makefile nv-linux.o SYSSRC=/lib/modules/5.11.15-1-default/source SYSOUT=/usr/src/linux-5.11.15-1-obj/x86_64/default make[1]: Entering directory '/usr/src/linux-5.11.15-1' make[2]: Entering directory '/usr/src/linux-5.11.15-1-obj/x86_64/default' scripts/Makefile.lib:8: 'always' is deprecated. Please use 'always-y' instead Open bug report for openSUSE and possible for nVidia. P.S. I was not sure whether to use always-y or always-m, apparently both work. Some kernel guru needs to chime in. Kernel documentation mentions both.
On Sun, May 02, 2021 at 08:43:30PM +0300, Andrei Borzenkov wrote:
Actually I already fixed this on April 20, but I haven't triggered an update of the repositories yet. I wasn't aware that Kernel 5.12 has already been released. :-( Seems this happened with TW 20210428 on April 30. I'll take care of this tomorrow ... so no need to open a bugreport for this. Thanks, Stefan Public Key available ------------------------------------------------------ Stefan Dirsch (Res. & Dev.) SUSE Software Solutions Germany GmbH Tel: 0911-740 53 0 Maxfeldstraße 5 FAX: 0911-740 53 479 D-90409 Nürnberg http://www.suse.de Germany ---------------------------------------------------------------- (HRB 36809, AG Nürnberg) Geschäftsführer: Felix Imendörffer ----------------------------------------------------------------
On Sun, May 02, 2021 at 09:11:07PM +0200, Stefan Dirsch wrote:
JFYI, repositories have been updated meanwhile and things are supposed to work again with Kernel 5.12. Thanks, Stefan Public Key available ------------------------------------------------------ Stefan Dirsch (Res. & Dev.) SUSE Software Solutions Germany GmbH Tel: 0911-740 53 0 Maxfeldstraße 5 FAX: 0911-740 53 479 D-90409 Nürnberg http://www.suse.de Germany ---------------------------------------------------------------- (HRB 36809, AG Nürnberg) Geschäftsführer: Felix Imendörffer ----------------------------------------------------------------
On Monday 03 May 2021, Andrei Borzenkov wrote:
I've not done a deep dive into any of this stuff, but it's a mystery to me as to why I haven't also experienced any similar issues. I did not experience any nvidia issues for 5.15, 5.16 or 5.12. Looking at /var/log/zypp/history, the last time I had an issue was TW 20210408 5.11.11 (at that time I replied to the factory release noting the nvidia issue). The log /var/log/zypp/history contained lines such as: # CC [M] /usr/src/kernel-modules/nvidia-460.67-default/nvidia/nv-frontend.o # objdump: error while loading shared libraries: libbfd-2.36.1.20210326-3.so: cannot open shared object file: No such file or directory It looks like I did a forced reinstall of all the nvidia packages. Looking at the /var/log/zypp/history subsequent updates for 5.11.12, 5.11.15, 5.11.16 and 5.12.0 seemed to go smoothly: 2021-03-31 15:55:09|install|kernel-default|5.11.6-1.4|x86_64||repo-oss # 2021-04-10 20:08:04 kernel-default-devel-5.11.11-1.2.x86_64.rpm installed ok # install: cannot stat '/usr/src/kernel-modules/nvidia-460.67-default/nvidia*.ko': No such file or directory 2021-04-10 20:08:04|install|kernel-default-devel|5.11.11-1.2|x86_64||repo-oss 2021-04-10 20:10:01|install|kernel-default|5.11.11-1.2|x86_64||repo-oss 2021-04-11 08:26:19|install|nvidia-gfxG05-kmp-default|460.67_k5.11.6_1-36.1|x86_64|root@kosmos1|NVIDIA 2021-04-15 11:23:34|install|nvidia-gfxG05-kmp-default|460.73.01_k5.11.11_1-37.1|x86_64||NVIDIA 2021-04-15 11:23:36|install|nvidia-computeG05|460.73.01-37.1|x86_64||NVIDIA 2021-04-15 11:23:39|install|nvidia-glG05|460.73.01-37.1|x86_64||NVIDIA 2021-04-15 11:23:51|install|x11-video-nvidiaG05|460.73.01-37.1|x86_64||NVIDIA 2021-04-16 11:01:27|install|nvidia-texture-tools|2.1.2-2.1|x86_64||repo-oss # 2021-04-16 11:06:25 kernel-default-devel-5.11.12-1.1.x86_64.rpm installed ok 2021-04-16 11:06:25|install|kernel-default-devel|5.11.12-1.1|x86_64||repo-oss 2021-04-16 11:08:13|install|kernel-default|5.11.12-1.1|x86_64||repo-oss 2021-04-20 12:12:20|install|kernel-default|5.11.15-1.2|x86_64||repo-oss # 2021-04-20 12:17:23 kernel-default-devel-5.11.15-1.2.x86_64.rpm installed ok 2021-04-20 12:17:23|install|kernel-default-devel|5.11.15-1.2|x86_64||repo-oss # 2021-04-28 18:17:18 kernel-default-devel-5.11.16-1.1.x86_64.rpm installed ok 2021-04-28 18:17:18|install|kernel-default-devel|5.11.16-1.1|x86_64||repo-oss 2021-04-28 18:23:07|install|kernel-default|5.11.16-1.1|x86_64||repo-oss # 2021-05-02 13:49:57 kernel-default-devel-5.12.0-1.2.x86_64.rpm installed ok 2021-05-02 13:49:57|install|kernel-default-devel|5.12.0-1.2|x86_64||repo-oss 2021-05-02 13:55:12|install|kernel-default|5.12.0-1.2|x86_64||repo-oss A little curious and a little confused. Michael
Michael Hamilton wrote:
Sort-of similar here. I use nvidia drivers for bumblebee, and compile them via dkms. But that just succeded clean for 5.12, except for a warning for an unused variable 'primary_plane' and notifications it's skipping BTF generation "due to unavailability of vmlinux"
On Sunday 02 May 2021, Mailings wrote:
I'm currently typing this from TW 20210427 running 5.11.16-1-default along with the nvidia proprietary 460.73 driver (installed by the "easy" method). If you are also on the 460 driver, my guess would be that your system's nvidia kernel module has not been rebuilt for 5.11.16. The module is normally automatically rebuilt as part of the TW zypper dup, but this sometimes doesn't happen. This has happened to me a few times recently, in these cases I found that a forced reinstall of the driver's kernel module followed by a reboot usually fixes the issue: % zypper in -f nvidia-gfxG05-kmp-default % reboot I've not tracked down exactly why sometimes fails, perhaps it is some chicken and egg circumstance. I did notice on one occasion I seemed to be missing a kernel-source install for my current kernel - but I'm not clear on whether this was the cause of the issue. Cheers, Michael
Hi Michael, Andrei thanks for your swift responses! One correction: I mentioned the wrong kernel version - its 5.11.15 for build 20210425. Michael - I've tried your suggestion but it failed complaining about a "mismatch" in nvidia-gfxG05-kmp-default-460.73.01_k5.11.11_1-37.1.x86_64.rpm [1](The file I've tee'd the output to got overwritten so regrettably I can't provide more details, grmbl) The kernel source install for 5.11.15 looked OK to me. Your comment and Andrei's point stating the nvidia modules are compiled made me try to update to the latest available build 20210429 which employs kernel 5.12.0. (Not knowing much about the "nvidia issue" I assumed nvidia are trying to keep some stuff from others looking into, thus forcing download of some pre-complied objects rather than to compile in situ) After the change to 5.12.0 all is fine again, I'm writing this mail on the system. One observation: I was following the update closely (which I usually don't do ;-)) and was concerned when I saw: ------------------------------------------------------------------------------------------------------------ .......BTF [M] /usr/src/kernel-modules/nvidia-460.73.01-default/nvidia.ko Skipping BTF generation for /usr/src/kernel-modules/nvidia-460.73.01-default/ nvidia.ko due to unavailability of vmlinux make[2]: Verzeichnis „/usr/src/linux-5.12.0-1-obj/x86_64/default“ wird verlassen make[1]: Leaving directory '/usr/src/linux-5.12.0-1' ld -r -o nv-linux.o \ nvidia.mod.o nvidia/nv-interface.o ld: cannot find nvidia/nv-interface.o: No such file or directory make: *** [Makefile:113: nv-linux.o] Error 1 / depmod: WARNING: could not open modules.order at /lib/modules/5.12.0-1-default: No such file or directory depmod: WARNING: could not open modules.builtin at /lib/modules/5.12.0-1- default: No such file or directory Modprobe blacklist files have been created at /etc/modprobe.d to prevent Nouveau from loading. This can be reverted by deleting /etc/modprobe.d/nvidia-*.conf. *** Reboot your computer and verify that the NVIDIA graphics driver can be loaded. *** grep: /etc/sysconfig/kernel: No such file or directory warning: %triggerin(nvidia-gfxG05-kmp-default-460.73.01_k5.11.11_1-37.1.x86_64) scriptlet failed, exit status 1 ........ --------------------------------------------------------------------------------------------------------------- Any idea whether this could turn into something serious or can it be ignored? br Juergen Am Sonntag, 2. Mai 2021, 07:13:02 CEST schrieb Andrei Borzenkov:
On Monday 03 May 2021, J. Merkel wrote:
If I grep /var/log/zypp/history for the above errors I can find all of them from various past updates all the way back to kernel 5.7.7-1-default. I'm not sure whether any of them are associated with past nvidia update issues. I've certainly not been paying any attention to them. In the past should an update cause my desktop to fail to boot to the sddm login screen, I have most often tried a forced reinstall of everything currently installed from the nvidia repo. More recently I have just tried rebuilding the module after someone here suggested that this was all that is necessary - perhaps that is insufficient in some cases, I'm not sure. Force reinstalls have often done the trick. Otherwise I just set my system to boot to the previous kernel - I don't think that has ever failed. I then raise an issue here, and wait for some suggestions or explanation. Often the explanation is along the lines the kernel API's have been changed and we're waiting for nvidia to catchup. It isn't essential that my desktop runs the very latest kernel, so this works fine for me. Michael
Hi Michael, Andrei thanks for your swift responses! One correction: I mentioned the wrong kernel version - its 5.11.15 for build 20210425. Michael - I've tried your suggestion but it failed complaining about a - in my words - "mismatch" with nvidia-gfxG05-kmp-default-460.73.01_k5.11.11_1-37.1.x86_64.rpm (The file I've tee'd the output to got overwritten so regrettably I can't provide more details, grmbl) The kernel-source for 5.11.15 is present. Your comment and also Andrei's point stating the nvidia modules are compiled on the machine made me try to update to the latest available build 20210429 which employs kernel 5.12.0. (Not knowing much about the "nvidia issue" I assumed nvidia are trying to keep some stuff from others looking into, thus forcing download of some pre-complied objects rather than to compile in situ) After the change to 5.12.0 all is fine again - I'm writing this mail on the system. One observation: I was following the update closely (which I usually don't do ;-)) and was concerned when I saw: ------------------------------------------------------------------------------------------------------------ .......BTF [M] /usr/src/kernel-modules/nvidia-460.73.01-default/nvidia.ko Skipping BTF generation for /usr/src/kernel-modules/nvidia-460.73.01-default/ nvidia.ko due to unavailability of vmlinux make[2]: Verzeichnis „/usr/src/linux-5.12.0-1-obj/x86_64/default“ wird verlassen make[1]: Leaving directory '/usr/src/linux-5.12.0-1' ld -r -o nv-linux.o \ nvidia.mod.o nvidia/nv-interface.o ld: cannot find nvidia/nv-interface.o: No such file or directory make: *** [Makefile:113: nv-linux.o] Error 1 / depmod: WARNING: could not open modules.order at /lib/modules/5.12.0-1-default: No such file or directory depmod: WARNING: could not open modules.builtin at /lib/modules/5.12.0-1- default: No such file or directory Modprobe blacklist files have been created at /etc/modprobe.d to prevent Nouveau from loading. This can be reverted by deleting /etc/modprobe.d/nvidia-*.conf. *** Reboot your computer and verify that the NVIDIA graphics driver can be loaded. *** grep: /etc/sysconfig/kernel: No such file or directory warning: %triggerin(nvidia-gfxG05-kmp-default-460.73.01_k5.11.11_1-37.1.x86_64) scriptlet failed, exit status 1 ........ --------------------------------------------------------------------------------------------------------------- Any idea whether this could turn into something serious or can it be ignored? br Juergen Am Sonntag, 2. Mai 2021, 07:13:02 CEST schrieb Andrei Borzenkov:
On 02.05.2021 04:32, Michael Hamilton wrote:
On Sunday 02 May 2021, Mailings wrote:
On 02.05.2021 18:18, Mailings wrote:
Yes, I can reproduce it. It is due to this line of nvidia/nvidia.Kbuild: always += $(NVIDIA_INTERFACE) Changing it to always-m += $(NVIDIA_INTERFACE) allows build to complete. I am not exactly sure what this part does and why it fails for 5.12 and not for 5.11, but you should open bug report on openSUSE bugzilla and possibly report it nvidia.
This is cosmetic.
This is also not a problem, although SPEC file could be improved to avoid this scary output.
participants (7)
-
Andrei Borzenkov -
J. Merkel -
Mailings -
Michael Hamilton -
Michael Hamilton -
Peter Suetterlin -
Stefan Dirsch