SuSe and issues with their kernel's.
I have copied this to both lists due to my disgust with SuSe's kernel modifications from the vanilla kernel and additional modules not compiling right off of them. Anyone compiling a vanilla kernel under SuSe 10? First here is my hardware configuration: Appro XtremeServer with 4 - 880 CPU's, 64GB Ram, 2 - 36GB 15k SCSI drives This server uses the NVidia nForce4 Pro 2200/2050 chipset I untar my kernel. Enter the directory, do a "zcat /proc/config.gz
.config" Then I do a "make menuconfig". Look through the settings not changing anything. Do my make, make modules_install, make install. Everthing looks good. Do a reboot, it starts the boot, gets to my 79xx SCSI driver, sees the first few drives, then starts dumping kernel error messages and finally freezes up. I am using a software RAID1 of 2 drives. But that should not matter. I have also downloaded the OpenSuse 10.1RC1 kernel which is 2.6.16-20 installed that. Everything booted fine. So I copied it's .config into the vanilla kernel directory, did my make routine, reboot and same issue.
Why do I need a vanilla kernel you ask? Well because there are a lot of applications that do not seem to compile when pointed to the SuSe kernel source tree. One being RedHat GFS which we are going to start using. There are other applications as well. And I am a bit upset that they change their kernel so much that some of these applications will not compile on their kernel. Or if anyone is using GFS on SuSe let me know what you did. Any help appreciated. Brad Dameron SeaTab Software www.seatab.com
On Wednesday 19 April 2006 17:49, Brad Dameron wrote:
I have copied this to both lists due to my disgust with SuSe's kernel modifications from the vanilla kernel and additional modules not compiling right off of them.
Anyone compiling a vanilla kernel under SuSe 10?
I have..... without problems. I didn't see you mention: make prepare or mk_initrd If you're not used to compiling kernels, I would suspect you left something out of the process. And just what part of what you did do you think you can blame on SuSE? You did say it was a vanilla kernel, right?
Brad Dameron wrote:
I have copied this to both lists due to my disgust with SuSe's kernel modifications from the vanilla kernel and additional modules not compiling right off of them.
Anyone compiling a vanilla kernel under SuSe 10?
Yep, and I have no problems doing so.
First here is my hardware configuration:
Appro XtremeServer with 4 - 880 CPU's, 64GB Ram, 2 - 36GB 15k SCSI drives This server uses the NVidia nForce4 Pro 2200/2050 chipset
What is an "880" CPU ? (just being curious)
I untar my kernel. Enter the directory, do a "zcat /proc/config.gz
.config" Then I do a "make menuconfig".
You might want to do a "make oldconfig" before "make menuconfig".
Look through the settings not changing anything. Do my make, make modules_install, make install. Everthing looks good.
You don't rebuild your initrd?? "mk_initrd"?
Do a reboot, it starts the boot, gets to my 79xx SCSI driver, sees the first few drives, then starts dumping kernel error messages and finally freezes up.
Check your /etc/sysconfig/kernel and see what it puts in the initrd. Unless you build the kernel with all boot-time modules built in, you will need to rebuild the initrd.
OpenSuse 10.1RC1 kernel which is 2.6.16-20 installed that. Everything booted fine. So I copied it's .config into the vanilla kernel directory, did my make routine, reboot and same issue.
And same reason.
Why do I need a vanilla kernel you ask? Well because there are a lot of applications that do not seem to compile when pointed to the SuSe kernel source tree. One being RedHat GFS which we are going to start using. There are other applications as well.
Have you got some more examples?
And I am a bit upset that they change their kernel so much that some of these applications will not compile on their kernel.
AFAIK, SUSE is working towards a mostly-vanilla kernel, and have already reduced the number of extra patches considerably.
Or if anyone is using GFS on SuSe let me know what you did.
I have only looked at it so far, but I find it very difficult to accept that SUSEs kernel modifications would prevent GFS from building. /Per Jessen, Zürich
And I am a bit upset that they change their kernel so much that some of these applications will not compile on their kernel.
AFAIK, SUSE is working towards a mostly-vanilla kernel, and have already reduced the number of extra patches considerably.
Yep, I have heard that SUSE kernels will be vanilla and for installing modules the "rpm" system would be used as for packages - which I believe is the right way to go. I hope, that Vanilla kernels will include more fnctionality built-in.
Or if anyone is using GFS on SuSe let me know what you did.
I have only looked at it so far, but I find it very difficult to accept that SUSEs kernel modifications would prevent GFS from building.
If the GFS is so great, why not include a GFS inside SUSE Linux 10.2 ? What OSes include GFS on DVD right now ? FC4/5 and RHEL4 ?
On Thu, 2006-04-20 at 10:37 +0200, Per Jessen wrote:
Brad Dameron wrote:
I have copied this to both lists due to my disgust with SuSe's kernel modifications from the vanilla kernel and additional modules not compiling right off of them.
Anyone compiling a vanilla kernel under SuSe 10?
Yep, and I have no problems doing so.
First here is my hardware configuration:
Appro XtremeServer with 4 - 880 CPU's, 64GB Ram, 2 - 36GB 15k SCSI drives This server uses the NVidia nForce4 Pro 2200/2050 chipset
What is an "880" CPU ? (just being curious)
880 is Opteron dual core 2.4Ghz.
I untar my kernel. Enter the directory, do a "zcat /proc/config.gz
.config" Then I do a "make menuconfig".
You might want to do a "make oldconfig" before "make menuconfig".
Look through the settings not changing anything. Do my make, make modules_install, make install. Everthing looks good.
You don't rebuild your initrd?? "mk_initrd"?
The make install actually does a mkinitrd for you.
Do a reboot, it starts the boot, gets to my 79xx SCSI driver, sees the first few drives, then starts dumping kernel error messages and finally freezes up.
Check your /etc/sysconfig/kernel and see what it puts in the initrd. Unless you build the kernel with all boot-time modules built in, you will need to rebuild the initrd.
OpenSuse 10.1RC1 kernel which is 2.6.16-20 installed that. Everything booted fine. So I copied it's .config into the vanilla kernel directory, did my make routine, reboot and same issue.
And same reason.
Why do I need a vanilla kernel you ask? Well because there are a lot of applications that do not seem to compile when pointed to the SuSe kernel source tree. One being RedHat GFS which we are going to start using. There are other applications as well.
Have you got some more examples?
And I am a bit upset that they change their kernel so much that some of these applications will not compile on their kernel.
AFAIK, SUSE is working towards a mostly-vanilla kernel, and have already reduced the number of extra patches considerably.
What is interesting is several apps we run don't like to compile against SuSe's kernel. However work perfectly against the vanilla one.
Or if anyone is using GFS on SuSe let me know what you did.
I have only looked at it so far, but I find it very difficult to accept that SUSEs kernel modifications would prevent GFS from building.
Actually they do. I tried many version's of GFS against several different SuSe kernel's. No go. They would error out in different locations each time. I finally got this working this morning. I compiled everything into the kernel so there are no modules and that seems to have worked for me. I would of preferred to use modules so I could start building my own kernel for my servers. But I won't be picky. Thanks for the comment. Brad Dameron SeaTab Software www.seatab.com
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 The Thursday 2006-04-20 at 12:21 -0700, Brad Dameron wrote:
The make install actually does a mkinitrd for you.
Yes, but... (*) ...
I finally got this working this morning. I compiled everything into the kernel so there are no modules and that seems to have worked for me.
(*) I would think then that you still need mkinitrd.
I would of preferred to use modules so I could start building my own kernel for my servers. But I won't be picky.
Did you remember to set up a different name for your kernel? If you don't, the make install_modules overwrites the default modules. Just two ideas. Also, Per Jessen said about certain modifications you might try. - -- Cheers, Carlos Robinson -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.0 (GNU/Linux) Comment: Made with pgp4pine 1.76 iD8DBQFER+FItTMYHG2NR9URAvDkAJ9DmmDfQZUTymi9gJIpkstM1b/vLgCfVZ30 aAkxEUg0u7r7Pe/213X2WMA= =KDMn -----END PGP SIGNATURE-----
Brad Dameron wrote:
You don't rebuild your initrd?? "mk_initrd"?
The make install actually does a mkinitrd for you.
Really? OK, I didn't know that. If your initrd is up-to-date, I cannot explain why you are having problems. I've sofar always rebuilt my initrd separately.
AFAIK, SUSE is working towards a mostly-vanilla kernel, and have already reduced the number of extra patches considerably.
What is interesting is several apps we run don't like to compile against SuSe's kernel. However work perfectly against the vanilla one.
Are these just applications or actually kernel extensions, such as GFS? For the latter, SUSEs extra "-Werror-implicit-function-declaration" could be causing a problem. And in this case, cluster-1.02.00 most certainly does not compile perfectly against the 2.6.16.9 vanilla kernel.
Or if anyone is using GFS on SuSe let me know what you did.
I have only looked at it so far, but I find it very difficult to accept that SUSEs kernel modifications would prevent GFS from building.
Actually they do. I tried many version's of GFS against several different SuSe kernel's. No go. They would error out in different locations each time.
Apart from the already mentioned "-Werror-implicit-function-declaration", _nothing_ in the SUSE 2.6.16-20 kernel prevented cluster-1.02.00 from building. But please do prove me wrong. Why haven't you reported this to SUSE/Novell?
But I won't be picky. Thanks for the comment.
One final comment then - to me, you were a little to quick in dishing out the blame here. You don't seem to have spent much time debugging/diagnosing the problem, yet you feel able to get up on your soapbox and point the finger. Maybe I'm a little sensitive, but it's not the really the right approach IMHO. /Per Jessen, Zürich
On Thu, 2006-04-20 at 21:50 +0200, Per Jessen wrote:
Brad Dameron wrote:
You don't rebuild your initrd?? "mk_initrd"?
The make install actually does a mkinitrd for you.
Really? OK, I didn't know that. If your initrd is up-to-date, I cannot explain why you are having problems. I've sofar always rebuilt my initrd separately.
AFAIK, SUSE is working towards a mostly-vanilla kernel, and have already reduced the number of extra patches considerably.
What is interesting is several apps we run don't like to compile against SuSe's kernel. However work perfectly against the vanilla one.
Are these just applications or actually kernel extensions, such as GFS? For the latter, SUSEs extra "-Werror-implicit-function-declaration" could be causing a problem. And in this case, cluster-1.02.00 most certainly does not compile perfectly against the 2.6.16.9 vanilla kernel.
Or if anyone is using GFS on SuSe let me know what you did.
I have only looked at it so far, but I find it very difficult to accept that SUSEs kernel modifications would prevent GFS from building.
Actually they do. I tried many version's of GFS against several different SuSe kernel's. No go. They would error out in different locations each time.
Apart from the already mentioned "-Werror-implicit-function-declaration", _nothing_ in the SUSE 2.6.16-20 kernel prevented cluster-1.02.00 from building. But please do prove me wrong. Why haven't you reported this to SUSE/Novell?
I tried to compile it against the 2.6.16-20 kernel and had problems as well. I installed the source rpm, did a make cloneconfig in the directory, did a make. Then went to the cluster package did a ./configure --kernel_src=/usr/src/linux-2.6.16-20, then a make install. It got down to dio.c and failed. I will post the exact error tomorrow. I would love to be able to do a simple SuSe 10 install and be able to compile GFS against the installed kernel.
But I won't be picky. Thanks for the comment.
One final comment then - to me, you were a little to quick in dishing out the blame here. You don't seem to have spent much time debugging/diagnosing the problem, yet you feel able to get up on your soapbox and point the finger. Maybe I'm a little sensitive, but it's not the really the right approach IMHO.
Actually I spent quite a lot of time trying to get this to work. And some of the things SuSe is doing had finally brought me to a boiling point. Like their lack of server side rpm updates. They seem to be focusing on mainly desktop updates. Yes I should go with enterprise version. But I checked that out as well and some server side packages are still back from early 2005. I am referring to things like LVM2, device-mapper for starters. Maybe I should not have posted it like this. But I was very flustered. Brad Dameron SeaTab Software www.seatab.com
On Friday 21 April 2006 02:44, Brad Dameron wrote:
Maybe I should not have posted it like this. But I was very flustered.
Maybe. You should have posted 1) where you got the third party software. Are you aware that there is a specific version of GFS for Red Hat's own kernel? Were you using that? How are we to know? 2) Specific error messages Just asking how to compile a vanilla kernel because the default is 'disgusting' is perhaps not a terribly professional approach, wouldn't you agree? I'm surprised you got as many replies as you did You also talk about compiling applications against the kernel, and vaguely saying it fails. How about a little detail? Like maybe a few names of those applications. Maybe then someone could see to it that whatever it is that's not working could be corrected. Not everything is through evil and malicious intent you know, there is such a thing as bugs in software. And with bug reports like yours they are likely to remain This is where I would offer something constructive, but since I have absolutely no information to go on, I can't
Brad Dameron wrote:
I tried to compile it against the 2.6.16-20 kernel and had problems as well. I installed the source rpm, did a make cloneconfig in the directory, did a make. Then went to the cluster package did a ./configure --kernel_src=/usr/src/linux-2.6.16-20, then a make install. It got down to dio.c and failed. I will post the exact error tomorrow.
Like I said yesterday, you will have to modify the kernel toplevel Makefile, and remove "-Werror-implicit-function-declaration" from the CFLAGS. Otherwise you will see the following: CC [M] /home/per/workspace/cluster-1.02.pj/gfs-kernel/src/gfs/dir.o /home/per/workspace/cluster-1.02.pj/gfs-kernel/src/gfs/dir.c: In function ?dir_split_leaf?: /home/per/workspace/cluster-1.02.pj/gfs-kernel/src/gfs/dir.c:761: error: implicit declaration of function ?vmalloc? /home/per/workspace/cluster-1.02.pj/gfs-kernel/src/gfs/dir.c:761: warning: assignment makes pointer from integer without a cast /home/per/workspace/cluster-1.02.pj/gfs-kernel/src/gfs/dir.c:792: error: implicit declaration of function ?vfree?
I would love to be able to do a simple SuSe 10 install and be able to compile GFS against the installed kernel.
1. cat /proc/config.gz >/usr/src/linux/.config 2. make -C /usr/src/linux/ oldconfig 3. make -C /usr/src/linux/ 4. correct the /usr/src/linux/Makefile wrt "-Werror-implicit-function-declaration" 5. apply my patch http://jessen.ch/files/cluster-1.02.00.patch 6. in cluster-1.02.00, run ./configure --kernel_src=/usr/src/linux 7. in cluster-1.02.00, run make Done.
Actually I spent quite a lot of time trying to get this to work. And some of the things SuSe is doing had finally brought me to a boiling point.
But they've only done ONE thing to the kernel source-tree that causes GFS not to compile cleanly. (due to an implicit declaration in dir.c)
Like their lack of server side rpm updates. They seem to be focusing on mainly desktop updates. Yes I should go with enterprise version. But I checked that out as well and some server side packages are still back from early 2005. I am referring to things like LVM2, device-mapper for starters.
It's a different discussion, one that it might be interesting to take to one of the opensuse lists instead. /Per Jessen, Zürich
Brad Dameron wrote:
Or if anyone is using GFS on SuSe let me know what you did.
The SUSE toplevel kernel Makefile is a little stricter than the vanilla. Looks for CFLAGS, then remove the line that reads: -Werror-implicit-function-declaration \ This will otherwise cause a problem in compiling cluster-1.02.00/gfs-kernel/src/gfs/dir.c But cluster-1.02.00 still fails to compile with both the SUSE 2.6.16-20 and the vanilla 2.6.16.9 kernel. I haven't investigated that any further. /Per Jessen, Zürich
Per Jessen wrote:
But cluster-1.02.00 still fails to compile with both the SUSE 2.6.16-20 and the vanilla 2.6.16.9 kernel. I haven't investigated that any further.
OK, I did look a little closer - here's a patch to cluster-1.02.00 that
will make it build with both 2.6.16.9 and SUSEs 2.6.16-20:
http://jessen.ch/files/cluster-1.02.00.patch
Contents:
diff -urN --exclude='*mk' --exclude='*symvers' --exclude='*d'
cluster-1.02.00/cman/lib/libcman.c cluster-1.02.pj/cman/lib/libcman.c
--- cluster-1.02.00/cman/lib/libcman.c 2006-03-13 12:45:11.000000000
+0100
+++ cluster-1.02.pj/cman/lib/libcman.c 2006-04-20 12:00:55.947309669
+0200
@@ -28,7 +28,7 @@
#include
On Wednesday 19 April 2006 22:49, Brad Dameron wrote:
Why do I need a vanilla kernel you ask? Well because there are a lot of applications that do not seem to compile when pointed to the SuSe kernel source tree. One being RedHat GFS which we are going to start using. There are other applications as well. And I am a bit upset that they change their kernel so much that some of these applications will not compile on their kernel.
Or if anyone is using GFS on SuSe let me know what you did.
Have you ever thought that the problem may be RED HATS PROBLEM and NOT SUSE's problem are Redhat not the ones that shy away from the file/tree standards and refuse to tow the line ?..from what i remember reading a while ago . And maybe a little bit MORE of a description as to the ERRORS would be usefull .!! we can all get hot under the collar but it dont get you nowhere but a hostile reception .. Supply some useable information and you will get what help can be given supply none you get none and BTW try giving the REDHAT list some GBH as well after all it is one of there packages is it not .. Pete . -- The Labour party has changed there emblem from a rose to a condom as it more accuratley reflects the governments political stance. A condom allows for inflation halts production destroys the next gereration, protects a bunch of pricks, and givesyou a sense of security while you are actually bieng fucked from GSM
participants (7)
-
Alexey Eremenko
-
Anders Johansson
-
Brad Dameron
-
Bruce Marshall
-
Carlos E. R.
-
Per Jessen
-
Peter Nikolic