[opensuse-kernel] bug in kernel-source-2.6.27.13-2.1 from http://download.opensuse.org/repositories/Kernel:/HEAD/openSUSE_Factory/
hi list, i think, there is a bug in the latest factory-kernel 2.6.27.13-2 from http://download.opensuse.org/repositories/Kernel:/HEAD/openSUSE_Factory/. at least the kernel source has a bug, i think. i tried to install latest nvidia-driver 180.22 on an opensuse11.1 x86_64 system. former kernel 2.6.27.12-1 had no problems with this. but this time, it was not possible to load nvidia kernel module. the installation-script stopped, and told me that the built kernel module can't be loaded. there were also warnings about missing symbols at compile time. this is the relevant part of dmesg-output when loading the driver: nvidia: Unknown symbol pv_mmu_ops nvidia: Unknown symbol pv_cpu_ops nvidia: Unknown symbol pv_irq_ops nvidia: Unknown symbol pv_lock_ops it seems, the above symbols are missing. this was not the case with 2.6.27.12-1. i had to downgrade kernel to latest available 2.6.27.7-9 from oss-repo. with this version building the nvidia driver also was no problem. right now i am writing this mail with this setup. there must have been some changes between 2.6.27.12-1 and 2.6.27.13-2 which have something to do with these missing symbols. does anybody know anything about this error? -- einen schönen Tag noch DI Rainer Klier Und bitte keine TOFU-Mails (http://de.wikipedia.org/wiki/TOFU) -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
On Mon, Jan 26, 2009 at 05:25:06PM +0100, Rainer Klier wrote:
i tried to install latest nvidia-driver 180.22 on an opensuse11.1 x86_64 system.
Please contact nvidia about their driver, there is nothing we can do about it due to their licensing decision. good luck, greg k-h -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
Am Monday 26 January 2009 17:43:41 schrieb Greg KH:
On Mon, Jan 26, 2009 at 05:25:06PM +0100, Rainer Klier wrote:
i tried to install latest nvidia-driver 180.22 on an opensuse11.1 x86_64 system.
Please contact nvidia about their driver, there is nothing we can do about it due to their licensing decision.
i don't think, the problem comes from the driver. because, i was able to build the same nvidia-driver with previous version of factory-kernel, without any problems. so there must be some changes in the factory-kernel causing the problems. something about the symbols symbol pv_mmu_ops, pv_cpu_ops, pv_irq_ops and pv_lock_ops. -- einen schönen Tag noch DI Rainer Klier Und bitte keine TOFU-Mails (http://de.wikipedia.org/wiki/TOFU) -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
hi list,
i think, there is a bug in the latest factory-kernel 2.6.27.13-2 from http://download.opensuse.org/repositories/Kernel:/HEAD/openSUSE_Factory/.
at least the kernel source has a bug, i think.
i tried to install latest nvidia-driver 180.22 on an opensuse11.1 x86_64 system. former kernel 2.6.27.12-1 had no problems with this. but this time, it was not possible to load nvidia kernel module.
the installation-script stopped, and told me that the built kernel module can't be loaded.
there were also warnings about missing symbols at compile time.
this is the relevant part of dmesg-output when loading the driver: nvidia: Unknown symbol pv_mmu_ops nvidia: Unknown symbol pv_cpu_ops nvidia: Unknown symbol pv_irq_ops nvidia: Unknown symbol pv_lock_ops
it seems, the above symbols are missing.
this was not the case with 2.6.27.12-1.
i had to downgrade kernel to latest available 2.6.27.7-9 from oss-repo. with this version building the nvidia driver also was no problem. right now i am writing this mail with this setup.
there must have been some changes between 2.6.27.12-1 and 2.6.27.13-2 which have something to do with these missing symbols. Hmm , I'am running 2.6.27.13-HEAD_20090125010735_8d0b1ab5-default x86_64 from http://ftp.suse.com/pub/projects/kernel/kotd/HEAD/x86_64/ and I had no
Am Montag, 26. Januar 2009 schrieb Rainer Klier: problems to build 180.22. However I didn't clone the configuration from the previous version , but used the SUSE default configuration from linux-2.6.27.13- HEAD_20090125010735_8d0b1ab5-obj/x86_64/default/.config
-- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
Am Monday 26 January 2009 18:22:51 schrieb Markus Koßmann:
Am Montag, 26. Januar 2009 schrieb Rainer Klier:
i had to downgrade kernel to latest available 2.6.27.7-9 from oss-repo. with this version building the nvidia driver also was no problem. right now i am writing this mail with this setup.
there must have been some changes between 2.6.27.12-1 and 2.6.27.13-2 which have something to do with these missing symbols. Hmm , I'am running 2.6.27.13-HEAD_20090125010735_8d0b1ab5-default x86_64 from http://ftp.suse.com/pub/projects/kernel/kotd/HEAD/x86_64/ and I had no problems to build 180.22. However I didn't clone the configuration from the previous version , but used the SUSE default configuration from linux-2.6.27.13- HEAD_20090125010735_8d0b1ab5-obj/x86_64/default/.config
interesting approach! i will try that. normally i always do "make cloneconfig". what is the difference between doing "make cloneconfig" and using a downloaded .config? i always thought "make cloneconfig" gives me the config from the kernel i am running right now, and this is the best possible option. -- einen schönen Tag noch DI Rainer Klier Und bitte keine TOFU-Mails (http://de.wikipedia.org/wiki/TOFU) -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
-------- Original-Nachricht --------
Datum: Tue, 27 Jan 2009 08:55:58 +0100 Von: Rainer Klier <r.klier@eta.co.at>
used the SUSE default configuration from linux-2.6.27.13- HEAD_20090125010735_8d0b1ab5-obj/x86_64/default/.config
interesting approach! i will try that. normally i always do "make cloneconfig". what is the difference between doing "make cloneconfig" and using a downloaded .config? i always thought "make cloneconfig" gives me the config from the kernel i am running right now, and this is the best possible option.
Well, think of new options. How do you set them when running cloneconfig ? That .config is used by SUSE to build their binary kernels. So it should reflect such changes. -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
* Markus Koßmann [2009-01-27 09:11:22]:
-------- Original-Nachricht --------
Datum: Tue, 27 Jan 2009 08:55:58 +0100 Von: Rainer Klier <r.klier@eta.co.at>
used the SUSE default configuration from linux-2.6.27.13- HEAD_20090125010735_8d0b1ab5-obj/x86_64/default/.config
interesting approach! i will try that. normally i always do "make cloneconfig". what is the difference between doing "make cloneconfig" and using a downloaded .config? i always thought "make cloneconfig" gives me the config from the kernel i am running right now, and this is the best possible option.
Well, think of new options. How do you set them when running cloneconfig ?
The process will ask you. Bernhard -- Bernhard Walle, SUSE LINUX Products GmbH, Architecture Development
Am Tuesday 27 January 2009 09:11:22 schrieb Markus Koßmann:
-------- Original-Nachricht --------
Datum: Tue, 27 Jan 2009 08:55:58 +0100 Von: Rainer Klier <r.klier@eta.co.at>
used the SUSE default configuration from linux-2.6.27.13- HEAD_20090125010735_8d0b1ab5-obj/x86_64/default/.config
interesting approach! i will try that. normally i always do "make cloneconfig". what is the difference between doing "make cloneconfig" and using a downloaded .config? i always thought "make cloneconfig" gives me the config from the kernel i am running right now, and this is the best possible option.
Well, think of new options. How do you set them when running cloneconfig ?
when i do "make cloneconfig", then the .config is cloned from /proc/config.gz which comes from the kernel currently running the machine, which is the same kernel i am using the source from for building the nvidia-module. -- einen schönen Tag noch DI Rainer Klier Und bitte keine TOFU-Mails (http://de.wikipedia.org/wiki/TOFU) -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
At Tue, 27 Jan 2009 08:55:58 +0100, Rainer Klier wrote:
Am Monday 26 January 2009 18:22:51 schrieb Markus Koßmann:
Am Montag, 26. Januar 2009 schrieb Rainer Klier:
i had to downgrade kernel to latest available 2.6.27.7-9 from oss-repo. with this version building the nvidia driver also was no problem. right now i am writing this mail with this setup.
there must have been some changes between 2.6.27.12-1 and 2.6.27.13-2 which have something to do with these missing symbols. Hmm , I'am running 2.6.27.13-HEAD_20090125010735_8d0b1ab5-default x86_64 from http://ftp.suse.com/pub/projects/kernel/kotd/HEAD/x86_64/ and I had no problems to build 180.22. However I didn't clone the configuration from the previous version , but used the SUSE default configuration from linux-2.6.27.13- HEAD_20090125010735_8d0b1ab5-obj/x86_64/default/.config
interesting approach! i will try that. normally i always do "make cloneconfig". what is the difference between doing "make cloneconfig" and using a downloaded .config? i always thought "make cloneconfig" gives me the config from the kernel i am running right now, and this is the best possible option.
Basically it works so. I wonder what got broken on your setup, too. Anyway, check whether you have CONFIG_UNUSED_SYMBOLS=y. Without this option, the symbols needed for nvidia / frglx won't be exported to modules. Takashi -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
Am Tuesday 27 January 2009 09:13:38 schrieb Takashi Iwai:
At Tue, 27 Jan 2009 08:55:58 +0100, Rainer Klier wrote:
Am Monday 26 January 2009 18:22:51 schrieb Markus Koßmann:
Am Montag, 26. Januar 2009 schrieb Rainer Klier:
i had to downgrade kernel to latest available 2.6.27.7-9 from oss-repo. with this version building the nvidia driver also was no problem. right now i am writing this mail with this setup.
there must have been some changes between 2.6.27.12-1 and 2.6.27.13-2 which have something to do with these missing symbols. Hmm , I'am running 2.6.27.13-HEAD_20090125010735_8d0b1ab5-default x86_64 from http://ftp.suse.com/pub/projects/kernel/kotd/HEAD/x86_64/ and I had no problems to build 180.22. However I didn't clone the configuration from the previous version , but used the SUSE default configuration from linux-2.6.27.13- HEAD_20090125010735_8d0b1ab5-obj/x86_64/default/.config
interesting approach! i will try that. normally i always do "make cloneconfig". what is the difference between doing "make cloneconfig" and using a downloaded .config? i always thought "make cloneconfig" gives me the config from the kernel i am running right now, and this is the best possible option.
Basically it works so. I wonder what got broken on your setup, too.
yes, me too. i always do it this way. it never failed in the past. and i did nothing special this time. installed new kernel with the following packages: kernel-default-base-2.6.27.13-2.1 kernel-default-extra-2.6.27.13-2.1 kernel-source-2.6.27.13-2.1 kernel-default-2.6.27.13-2.1 kernel-syms-2.6.27.13-2.1
Anyway, check whether you have CONFIG_UNUSED_SYMBOLS=y. Without this option, the symbols needed for nvidia / frglx won't be exported to modules.
i will try that. what i found out today is, that the package "kernel-syms" contains the file /usr/src/linux-2.6.27.xxx-obj/x86_64/default/Module.symvers in this file i found the following lines: 0xf16b27a5 pv_mmu_ops arch/x86/kernel/built-in EXPORT_SYMBOL 0xf97a76a6 pv_cpu_ops arch/x86/kernel/built-in EXPORT_SYMBOL 0x522f5573 pv_irq_ops arch/x86/kernel/built-in EXPORT_SYMBOL 0x60400b7f pv_lock_ops arch/x86/kernel/built-in EXPORT_SYMBOL maybe this file is broken in version 2.6.27.13-2.1 and the above lines are not there. what would happen, if i uninstall the package "kernel-syms"? will it be possible to make nvidia-driver without it? -- einen schönen Tag noch DI Rainer Klier Und bitte keine TOFU-Mails (http://de.wikipedia.org/wiki/TOFU) -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
At Tue, 27 Jan 2009 09:30:04 +0100, Rainer Klier wrote:
Am Tuesday 27 January 2009 09:13:38 schrieb Takashi Iwai:
At Tue, 27 Jan 2009 08:55:58 +0100, Rainer Klier wrote:
Am Monday 26 January 2009 18:22:51 schrieb Markus Koßmann:
Am Montag, 26. Januar 2009 schrieb Rainer Klier:
i had to downgrade kernel to latest available 2.6.27.7-9 from oss-repo. with this version building the nvidia driver also was no problem. right now i am writing this mail with this setup.
there must have been some changes between 2.6.27.12-1 and 2.6.27.13-2 which have something to do with these missing symbols. Hmm , I'am running 2.6.27.13-HEAD_20090125010735_8d0b1ab5-default x86_64 from http://ftp.suse.com/pub/projects/kernel/kotd/HEAD/x86_64/ and I had no problems to build 180.22. However I didn't clone the configuration from the previous version , but used the SUSE default configuration from linux-2.6.27.13- HEAD_20090125010735_8d0b1ab5-obj/x86_64/default/.config
interesting approach! i will try that. normally i always do "make cloneconfig". what is the difference between doing "make cloneconfig" and using a downloaded .config? i always thought "make cloneconfig" gives me the config from the kernel i am running right now, and this is the best possible option.
Basically it works so. I wonder what got broken on your setup, too.
yes, me too. i always do it this way. it never failed in the past. and i did nothing special this time. installed new kernel with the following packages: kernel-default-base-2.6.27.13-2.1 kernel-default-extra-2.6.27.13-2.1 kernel-source-2.6.27.13-2.1 kernel-default-2.6.27.13-2.1 kernel-syms-2.6.27.13-2.1
Anyway, check whether you have CONFIG_UNUSED_SYMBOLS=y. Without this option, the symbols needed for nvidia / frglx won't be exported to modules.
i will try that. what i found out today is, that the package "kernel-syms" contains the file /usr/src/linux-2.6.27.xxx-obj/x86_64/default/Module.symvers in this file i found the following lines: 0xf16b27a5 pv_mmu_ops arch/x86/kernel/built-in EXPORT_SYMBOL 0xf97a76a6 pv_cpu_ops arch/x86/kernel/built-in EXPORT_SYMBOL 0x522f5573 pv_irq_ops arch/x86/kernel/built-in EXPORT_SYMBOL 0x60400b7f pv_lock_ops arch/x86/kernel/built-in EXPORT_SYMBOL
maybe this file is broken in version 2.6.27.13-2.1 and the above lines are not there. what would happen, if i uninstall the package "kernel-syms"? will it be possible to make nvidia-driver without it?
kernel-syms is needed only when you build modules with all flavors. If you need to build one flavor, it's not needed at all. The file Module.symvers is contained in kernel-$FLAVOR-base.rpm, too. Takashi -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
Am Tuesday 27 January 2009 10:40:27 schrieb Takashi Iwai:
At Tue, 27 Jan 2009 09:30:04 +0100, Rainer Klier wrote:
Am Tuesday 27 January 2009 09:13:38 schrieb Takashi Iwai:
At Tue, 27 Jan 2009 08:55:58 +0100, Rainer Klier wrote:
Am Monday 26 January 2009 18:22:51 schrieb Markus Koßmann:
Am Montag, 26. Januar 2009 schrieb Rainer Klier:
i had to downgrade kernel to latest available 2.6.27.7-9 from oss-repo. with this version building the nvidia driver also was no problem. right now i am writing this mail with this setup.
there must have been some changes between 2.6.27.12-1 and 2.6.27.13-2 which have something to do with these missing symbols. Hmm , I'am running 2.6.27.13-HEAD_20090125010735_8d0b1ab5-default x86_64 from http://ftp.suse.com/pub/projects/kernel/kotd/HEAD/x86_64/ and I had no problems to build 180.22. However I didn't clone the configuration from the previous version , but used the SUSE default configuration from linux-2.6.27.13- HEAD_20090125010735_8d0b1ab5-obj/x86_64/default/.config
interesting approach! i will try that. normally i always do "make cloneconfig". what is the difference between doing "make cloneconfig" and using a downloaded .config? i always thought "make cloneconfig" gives me the config from the kernel i am running right now, and this is the best possible option.
Basically it works so. I wonder what got broken on your setup, too.
yes, me too. i always do it this way. it never failed in the past. and i did nothing special this time. installed new kernel with the following packages: kernel-default-base-2.6.27.13-2.1 kernel-default-extra-2.6.27.13-2.1 kernel-source-2.6.27.13-2.1 kernel-default-2.6.27.13-2.1 kernel-syms-2.6.27.13-2.1
Anyway, check whether you have CONFIG_UNUSED_SYMBOLS=y. Without this option, the symbols needed for nvidia / frglx won't be exported to modules.
i found that in alle tested .config files.
i will try that. what i found out today is, that the package "kernel-syms" contains the file /usr/src/linux-2.6.27.xxx-obj/x86_64/default/Module.symvers in this file i found the following lines: 0xf16b27a5 pv_mmu_ops arch/x86/kernel/built-in EXPORT_SYMBOL 0xf97a76a6 pv_cpu_ops arch/x86/kernel/built-in EXPORT_SYMBOL 0x522f5573 pv_irq_ops arch/x86/kernel/built-in EXPORT_SYMBOL 0x60400b7f pv_lock_ops arch/x86/kernel/built-in EXPORT_SYMBOL
maybe this file is broken in version 2.6.27.13-2.1 and the above lines are not there. what would happen, if i uninstall the package "kernel-syms"? will it be possible to make nvidia-driver without it?
kernel-syms is needed only when you build modules with all flavors. If you need to build one flavor, it's not needed at all. The file Module.symvers is contained in kernel-$FLAVOR-base.rpm, too.
thank you. ok, i noticed that just a few minutes ago. i found factory-kernel 2.6.27.11-2.1 on http://ftp.iasi.roedu.net/mirrors/opensuse.org/repositories/Kernel:/HEAD/ope... this kernel is a little older than latest 2.6.27.13-3.1. like with 2.6.27.12-1 from last week (not available any more) it was no problem building the nvidia-kernel module. i noticed one thing about .config file when i had installed 2.6.27.13-3: in /usr/src/linux-2.6.27.13-3 there was the file .config.running. when i did "make cloneconfig" i got the error/warning about setting some XFS related config-variable to "m" and the resulting .config was NOT the same as the .config.running. it looked this way: linux-2.6.27.13-3: -rw-r--r-- 1 root root 89700 Jan 27 10:40 .config -rw-r--r-- 1 root root 90134 Jan 27 10:36 .config.running and in the directory /usr/src/linux-2.6.27.13-3-obj/x86_64/default there was this .config: -rw-r--r-- 1 root root 89782 Jan 27 10:45 .config so THREE different .config-files. right now, with 2.6.27.11-2 installed, it looks like this in /usr/src/linux-2.6.27.11-2: -rw-r--r-- 1 root root 90123 Jan 27 11:01 .config -rw-r--r-- 1 root root 90123 Jan 27 11:01 .config.running and like this in /usr/src/linux-2.6.27.11-2-obj/x86_64/default: -rw-r--r-- 1 root root 90123 Jan 16 04:13 .config so, as you can see, ALL .config files are the same with kernel 2.6.27.11-2 which is working great with nvidia-kernel-module. and with 2.6.27.13-3 there are 3 different .config files... maybe this has something to do with the problem... -- einen schönen Tag noch DI Rainer Klier Und bitte keine TOFU-Mails (http://de.wikipedia.org/wiki/TOFU) -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
At Tue, 27 Jan 2009 09:13:38 +0100, I wrote:
At Tue, 27 Jan 2009 08:55:58 +0100, Rainer Klier wrote:
Am Monday 26 January 2009 18:22:51 schrieb Markus Koßmann:
Am Montag, 26. Januar 2009 schrieb Rainer Klier:
i had to downgrade kernel to latest available 2.6.27.7-9 from oss-repo. with this version building the nvidia driver also was no problem. right now i am writing this mail with this setup.
there must have been some changes between 2.6.27.12-1 and 2.6.27.13-2 which have something to do with these missing symbols. Hmm , I'am running 2.6.27.13-HEAD_20090125010735_8d0b1ab5-default x86_64 from http://ftp.suse.com/pub/projects/kernel/kotd/HEAD/x86_64/ and I had no problems to build 180.22. However I didn't clone the configuration from the previous version , but used the SUSE default configuration from linux-2.6.27.13- HEAD_20090125010735_8d0b1ab5-obj/x86_64/default/.config
interesting approach! i will try that. normally i always do "make cloneconfig". what is the difference between doing "make cloneconfig" and using a downloaded .config? i always thought "make cloneconfig" gives me the config from the kernel i am running right now, and this is the best possible option.
Basically it works so. I wonder what got broken on your setup, too.
Anyway, check whether you have CONFIG_UNUSED_SYMBOLS=y. Without this option, the symbols needed for nvidia / frglx won't be exported to modules.
It seems that we recently disabled paravirt on default config. So pv_* don't exist any more. I guess it's a problem nvidia side, e.g. you are building with the old object refering to non-existing pv_*. Takashi -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
Am Tuesday 27 January 2009 11:34:53 schrieb Takashi Iwai:
At Tue, 27 Jan 2009 09:13:38 +0100, I wrote:
At Tue, 27 Jan 2009 08:55:58 +0100, Rainer Klier wrote:
Am Monday 26 January 2009 18:22:51 schrieb Markus Koßmann:
Am Montag, 26. Januar 2009 schrieb Rainer Klier:
It seems that we recently disabled paravirt on default config. So pv_* don't exist any more.
that would explain the errors nvidia: Unknown symbol pv_mmu_ops nvidia: Unknown symbol pv_cpu_ops nvidia: Unknown symbol pv_irq_ops nvidia: Unknown symbol pv_lock_ops last version which worked for me was 2.6.27.12-1.
I guess it's a problem nvidia side, e.g. you are building with the old object refering to non-existing pv_*.
ok, understood. so either nvidia changes the pre-built module-part (AFAIK the nvidia-kernel-module consists of 2 parts, a pre-built part and a part that has to be made at install-time) or i have a kernel, with paravirt enabled, so that latest pre-built nvidia-kernel-module fits. right? this issue has to be fixed in any way, BEFORE this factory-kernel becomes some day the official next opensuse-kernel or the official update-kernel for opensuse11.1. if not, alle users working with proprietary nvidia-driver will habe black-screens...... :-( -- einen schönen Tag noch DI Rainer Klier Und bitte keine TOFU-Mails (http://de.wikipedia.org/wiki/TOFU) -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
On 2009-01-27T12:35:03, Rainer Klier <r.klier@eta.co.at> wrote:
this issue has to be fixed in any way, BEFORE this factory-kernel becomes some day the official next opensuse-kernel or the official update-kernel for opensuse11.1.
if not, alle users working with proprietary nvidia-driver will habe black-screens...... :-(
This must be fixed by nvidia. Their driver is faulty. Regards, Lars -- Teamlead Kernel, SuSE Labs, Research and Development SUSE LINUX Products GmbH, GF: Markus Rex, HRB 16746 (AG Nürnberg) "Experience is the name everyone gives to their mistakes." -- Oscar Wilde -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
At Tue, 27 Jan 2009 12:35:03 +0100, Rainer Klier wrote:
Am Tuesday 27 January 2009 11:34:53 schrieb Takashi Iwai:
At Tue, 27 Jan 2009 09:13:38 +0100, I wrote:
At Tue, 27 Jan 2009 08:55:58 +0100, Rainer Klier wrote:
Am Monday 26 January 2009 18:22:51 schrieb Markus Koßmann:
Am Montag, 26. Januar 2009 schrieb Rainer Klier:
It seems that we recently disabled paravirt on default config. So pv_* don't exist any more.
that would explain the errors nvidia: Unknown symbol pv_mmu_ops nvidia: Unknown symbol pv_cpu_ops nvidia: Unknown symbol pv_irq_ops nvidia: Unknown symbol pv_lock_ops
last version which worked for me was 2.6.27.12-1.
I guess it's a problem nvidia side, e.g. you are building with the old object refering to non-existing pv_*.
ok, understood. so either nvidia changes the pre-built module-part (AFAIK the nvidia-kernel-module consists of 2 parts, a pre-built part and a part that has to be made at install-time) or i have a kernel, with paravirt enabled, so that latest pre-built nvidia-kernel-module fits.
You don't need pre-built stuff for nvidia module. The binary-only part of nvidia driver has nothing to do with paravirt, and the rest can be compiled from the source code.
this issue has to be fixed in any way, BEFORE this factory-kernel becomes some day the official next opensuse-kernel or the official update-kernel for opensuse11.1.
It works on my machine. I just built nvidia.ko manually with the latest KOTD kernel. It's built and loaded fine. So, it's likely your nvidia build-setup problem. Takashi -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
Am Tuesday 27 January 2009 15:51:01 schrieb Takashi Iwai:
At Tue, 27 Jan 2009 12:35:03 +0100, Rainer Klier wrote:
Am Tuesday 27 January 2009 11:34:53 schrieb Takashi Iwai:
At Tue, 27 Jan 2009 09:13:38 +0100, I wrote:
At Tue, 27 Jan 2009 08:55:58 +0100, Rainer Klier wrote:
Am Monday 26 January 2009 18:22:51 schrieb Markus Koßmann:
Am Montag, 26. Januar 2009 schrieb Rainer Klier:
ok, understood. so either nvidia changes the pre-built module-part (AFAIK the nvidia-kernel-module consists of 2 parts, a pre-built part and a part that has to be made at install-time) or i have a kernel, with paravirt enabled, so that latest pre-built nvidia-kernel-module fits.
You don't need pre-built stuff for nvidia module. The binary-only part of nvidia driver has nothing to do with paravirt, and the rest can be compiled from the source code.
ok, then it should work i think.
this issue has to be fixed in any way, BEFORE this factory-kernel becomes some day the official next opensuse-kernel or the official update-kernel for opensuse11.1.
It works on my machine. I just built nvidia.ko manually with the latest KOTD kernel. It's built and loaded fine.
whow! how did you do that? from where did you get your KOTD? which nvidia driver did you install? i did the following: 1. downloaded nvidia driver from http://www.nvidia.com/object/linux_display_amd64_180.22.html 2. extracted install-package with "NVIDIA-Linux-x86_64-180.22-pkg2.run --extract-only" 3. updated kernel from within yast from repo at http://download.opensuse.org/repositories/Kernel:/HEAD/openSUSE_Factory/ 4. rebooted to runlevel 3 5. cd /usr/src/linux 6. make cloneconfig 7. make prepare 8. make modules_prepare 9. cd /tmp/NVIDIA-Linux-x86_64-180.22-pkg2.run 10. ./nvidia-installer -N then the installer stopped and in dmesg i could see the messages about missing symbols. -- einen schönen Tag noch DI Rainer Klier Und bitte keine TOFU-Mails (http://de.wikipedia.org/wiki/TOFU) -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
At Wed, 28 Jan 2009 09:42:55 +0100, Rainer Klier wrote:
Am Tuesday 27 January 2009 15:51:01 schrieb Takashi Iwai:
At Tue, 27 Jan 2009 12:35:03 +0100, Rainer Klier wrote:
Am Tuesday 27 January 2009 11:34:53 schrieb Takashi Iwai:
At Tue, 27 Jan 2009 09:13:38 +0100, I wrote:
At Tue, 27 Jan 2009 08:55:58 +0100, Rainer Klier wrote:
Am Monday 26 January 2009 18:22:51 schrieb Markus Koßmann: > Am Montag, 26. Januar 2009 schrieb Rainer Klier:
ok, understood. so either nvidia changes the pre-built module-part (AFAIK the nvidia-kernel-module consists of 2 parts, a pre-built part and a part that has to be made at install-time) or i have a kernel, with paravirt enabled, so that latest pre-built nvidia-kernel-module fits.
You don't need pre-built stuff for nvidia module. The binary-only part of nvidia driver has nothing to do with paravirt, and the rest can be compiled from the source code.
ok, then it should work i think.
this issue has to be fixed in any way, BEFORE this factory-kernel becomes some day the official next opensuse-kernel or the official update-kernel for opensuse11.1.
It works on my machine. I just built nvidia.ko manually with the latest KOTD kernel. It's built and loaded fine.
whow! how did you do that? from where did you get your KOTD?
ftp://ftp.suse.com/pub/projects/kernel/kotd/
which nvidia driver did you install?
nvidia-gfxG01-173.14.15.
i did the following: 1. downloaded nvidia driver from http://www.nvidia.com/object/linux_display_amd64_180.22.html 2. extracted install-package with "NVIDIA-Linux-x86_64-180.22-pkg2.run --extract-only" 3. updated kernel from within yast from repo at http://download.opensuse.org/repositories/Kernel:/HEAD/openSUSE_Factory/ 4. rebooted to runlevel 3 5. cd /usr/src/linux 6. make cloneconfig
Why do you need this at all? Installing kernel-source.rpm suffices.
7. make prepare 8. make modules_prepare 9. cd /tmp/NVIDIA-Linux-x86_64-180.22-pkg2.run 10. ./nvidia-installer -N
I'd build nvidia.ko directly by going to $NVIDIA/usr/src/nv and run "make module". Better to remove precompiled stuff, though. Takashi -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
Am Wednesday 28 January 2009 12:26:35 schrieb Takashi Iwai:
At Wed, 28 Jan 2009 09:42:55 +0100,
Rainer Klier wrote:
Am Tuesday 27 January 2009 15:51:01 schrieb Takashi Iwai:
At Tue, 27 Jan 2009 12:35:03 +0100,
Rainer Klier wrote:
Am Tuesday 27 January 2009 11:34:53 schrieb Takashi Iwai:
At Tue, 27 Jan 2009 09:13:38 +0100,
I wrote:
At Tue, 27 Jan 2009 08:55:58 +0100,
Rainer Klier wrote: > Am Monday 26 January 2009 18:22:51 schrieb Markus Koßmann: > > Am Montag, 26. Januar 2009 schrieb Rainer Klier:
It works on my machine. I just built nvidia.ko manually with the latest KOTD kernel. It's built and loaded fine.
whow! how did you do that? from where did you get your KOTD?
ftp://ftp.suse.com/pub/projects/kernel/kotd/
ok, i tried http://download.opensuse.org/repositories/Kernel:/HEAD/openSUSE_Factory
which nvidia driver did you install?
nvidia-gfxG01-173.14.15.
ok, i tried http://www.nvidia.com/object/linux_display_amd64_180.22.html
6. make cloneconfig
Why do you need this at all? Installing kernel-source.rpm suffices.
i learned in the past, for installing vmware-server, and cisco-vpn-client (which also compile kernel-modules) you have to do "make cloneconfig; make prepare; make modules_prepare"
I'd build nvidia.ko directly by going to $NVIDIA/usr/src/nv and run "make module". Better to remove precompiled stuff, though.
ok, this way i didn't know. documentation on nvidia-homepage says you should run nvidia-installer. -- einen schönen Tag noch DI Rainer Klier Und bitte keine TOFU-Mails (http://de.wikipedia.org/wiki/TOFU) -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
At Wed, 28 Jan 2009 12:37:14 +0100, Rainer Klier wrote:
6. make cloneconfig
Why do you need this at all? Installing kernel-source.rpm suffices.
i learned in the past, for installing vmware-server, and cisco-vpn-client (which also compile kernel-modules) you have to do "make cloneconfig; make prepare; make modules_prepare"
In general you don't have to do it. Sometimes you need to fix the other side, though -- it's basically a problem of external driver trees. The include files and symbols needed to build kernel modules are already stored in /usr/src/linux*-obj/$ARCH/$FLAVOR. This directory is symlinked from /lib/modules/$VERSION/build, but some external drivers don't refer this correctly (or you may have some options to specify manually). But, nvidia driver tree seems to work as is. Takashi -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
Am Tuesday 27 January 2009 15:51:01 schrieb Takashi Iwai:
At Tue, 27 Jan 2009 12:35:03 +0100,
Rainer Klier wrote:
Am Tuesday 27 January 2009 11:34:53 schrieb Takashi Iwai:
At Tue, 27 Jan 2009 09:13:38 +0100,
I wrote:
At Tue, 27 Jan 2009 08:55:58 +0100,
Rainer Klier wrote:
Am Monday 26 January 2009 18:22:51 schrieb Markus Koßmann:
Am Montag, 26. Januar 2009 schrieb Rainer Klier:
It works on my machine. I just built nvidia.ko manually with the latest KOTD kernel. It's built and loaded fine.
So, it's likely your nvidia build-setup problem.
VERY VERY STRANGE! it also works for me now! today i installed latest 2.6.27.13-5.1 from http://download.opensuse.org/repositories/Kernel:/HEAD/openSUSE_Factory and then i copied the files ".config" and "Module.symvers" from /usr/src/linux-obj/x86_64/default/ to /usr/src/linux/ then i went to /usr/src/linux/ and made "make prepare" and "make modules_prepare". then i did the usual "./nvidia-installer" and that's it. it worked this time! WHITHOUT CHANGING ANYTHING AT THE NVIDIA-DRIVER! so the problem MUST come from something inside the 2.6.27.13-2.1 from yesterday! or it was the "make cloneconfig" i did yesterday and the missing file Module.symvers in /usr/src/linux/. or the guys at suse/novell changed something in latest 2.6.27.13-5.1. -- einen schönen Tag noch DI Rainer Klier Und bitte keine TOFU-Mails (http://de.wikipedia.org/wiki/TOFU) -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
On Tue, Jan 27, 2009 at 12:35:03PM +0100, Rainer Klier wrote:
this issue has to be fixed in any way, BEFORE this factory-kernel becomes some day the official next opensuse-kernel or the official update-kernel for opensuse11.1.
Step back and think about this statement for a moment. Please realize that the nvidia driver is closed source and we have absolutely _NO_ control over what they do in their code. For their driver to be controlling the direction of the opensuse kernel seems to be exactly backwards, right? I understand that people get upset when their closed-source, GPL violating driver breaks, but why not talk to the people that create that driver instead of asking people who can't do anything about it? Remember, the tail doesn't wag the dog here, greg k-h -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
Am Tuesday 27 January 2009 17:32:52 schrieb Greg KH:
On Tue, Jan 27, 2009 at 12:35:03PM +0100, Rainer Klier wrote:
this issue has to be fixed in any way, BEFORE this factory-kernel becomes some day the official next opensuse-kernel or the official update-kernel for opensuse11.1.
Step back and think about this statement for a moment. Please realize that the nvidia driver is closed source and we have absolutely _NO_ control over what they do in their code. For their driver to be
yes, of course i know that.
controlling the direction of the opensuse kernel seems to be exactly backwards, right?
no, of course not. you are absolutely right!
I understand that people get upset when their closed-source, GPL violating driver breaks, but why not talk to the people that create that driver instead of asking people who can't do anything about it?
at the time the problem occured, i didn't know, from where it came from. i updated kermel like i alwas do. i installed nvidia-driver like i always do. there was no problem in the past with that. kernel 2.6.27.12 worked perfectly with nvidia 180.22. and changing the kernel to 2.6.27.13 and installing the same nvidia 180.22 driver failed. so, of course, i thought of a problem in the latest kernel. ok, there have been made changes in 2.6.27.13 which are incompatible with nvidia driver. and the nvidia guys have to fix this. but as i said, it will not be good for us all, if this will not happen before this new kernel becomes some kind of "official". because in this case many many people will update to the latest official kernel through online-update or base-installation and will have black screens when using the proprietary nvidia driver. i hope this will not happen. -- einen schönen Tag noch DI Rainer Klier Und bitte keine TOFU-Mails (http://de.wikipedia.org/wiki/TOFU) -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
On 2009-01-28T09:26:28, Rainer Klier <r.klier@eta.co.at> wrote:
because in this case many many people will update to the latest official kernel through online-update or base-installation and will have black screens when using the proprietary nvidia driver.
Best of course would be if they stopped relying on proprietary & license-violating kernel modules. ;-) Regards, Lars -- Teamlead Kernel, SuSE Labs, Research and Development SUSE LINUX Products GmbH, GF: Markus Rex, HRB 16746 (AG Nürnberg) "Experience is the name everyone gives to their mistakes." -- Oscar Wilde -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
Am Wednesday 28 January 2009 10:29:21 schrieb Lars Marowsky-Bree:
On 2009-01-28T09:26:28, Rainer Klier <r.klier@eta.co.at> wrote:
because in this case many many people will update to the latest official kernel through online-update or base-installation and will have black screens when using the proprietary nvidia driver.
Best of course would be if they stopped relying on proprietary & license-violating kernel modules. ;-)
yes, of course it would be. but i own a nvidia-gfx-card, and the open nv-driver ist far behind the proprietary driver from nvidia! and i want to use my hardware with all its features and at full speed. -- einen schönen Tag noch DI Rainer Klier Und bitte keine TOFU-Mails (http://de.wikipedia.org/wiki/TOFU) -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
Am Tuesday 27 January 2009 11:34:53 schrieb Takashi Iwai:
At Tue, 27 Jan 2009 09:13:38 +0100, I wrote:
At Tue, 27 Jan 2009 08:55:58 +0100, Rainer Klier wrote:
Am Monday 26 January 2009 18:22:51 schrieb Markus Koßmann:
Am Montag, 26. Januar 2009 schrieb Rainer Klier:
It seems that we recently disabled paravirt on default config. So pv_* don't exist any more.
can you re-enable paravirt on default config? -- einen schönen Tag noch DI Rainer Klier Und bitte keine TOFU-Mails (http://de.wikipedia.org/wiki/TOFU) -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
On Tue, 27 Jan 2009, Rainer Klier wrote:
It seems that we recently disabled paravirt on default config. So pv_* don't exist any more. can you re-enable paravirt on default config?
Currently not. Paravirt introduced measurable overhead into pagetable handling, and until this is sorted out, we keep it disabled for -default. -- Jiri Kosina SUSE Labs -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
Am Monday 26 January 2009 18:22:51 schrieb Markus Koßmann:
Am Montag, 26. Januar 2009 schrieb Rainer Klier:
there must have been some changes between 2.6.27.12-1 and 2.6.27.13-2 which have something to do with these missing symbols. Hmm , I'am running 2.6.27.13-HEAD_20090125010735_8d0b1ab5-default x86_64 from http://ftp.suse.com/pub/projects/kernel/kotd/HEAD/x86_64/ and I had no problems to build 180.22. However I didn't clone the configuration from the previous version , but used the SUSE default configuration from linux-2.6.27.13- HEAD_20090125010735_8d0b1ab5-obj/x86_64/default/.config
i tried using .config from /usr/src/linux-2.6.27.13-3-obj/x86_64/default/.config but it also failed. -- einen schönen Tag noch DI Rainer Klier Und bitte keine TOFU-Mails (http://de.wikipedia.org/wiki/TOFU) -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
participants (8)
-
"Markus Koßmann" -
Bernhard Walle -
Greg KH -
Jiri Kosina -
Lars Marowsky-Bree -
Markus Koßmann
-
Rainer Klier -
Takashi Iwai