Hi, It seems to me, that kernel-2.6.18-8.1.15 is maybe not ok. I have been trying to rebuild my patched kernel RPMS in virtual machine running i686 kernel-2.6.18-8.1.15. It had failed for the three times with segmentation fault. First run under kernel-2.6.18-8.1.14 went smoothly. So, maybe it's a kernel issue or it's VMWare (Server 1.04). Anyone to reproduce the issue? Thanks, David
On 10/27/07, David Hrbáč hrbac.conf@seznam.cz wrote:
Hi, It seems to me, that kernel-2.6.18-8.1.15 is maybe not ok. I have been trying to rebuild my patched kernel RPMS in virtual machine running i686 kernel-2.6.18-8.1.15. It had failed for the three times with segmentation fault. First run under kernel-2.6.18-8.1.14 went smoothly. So, maybe it's a kernel issue or it's VMWare (Server 1.04). Anyone to reproduce the issue? Thanks, David
I can certainly try it on my c5-i686 VM. But at which step does it segfault? Does it have to do with modsigning?
Akemi
On 10/27/07, David Hrbáč hrbac.conf@seznam.cz wrote:
Akemi Yagi napsal(a):
I can certainly try it on my c5-i686 VM. But at which step does it segfault? Does it have to do with modsigning?
Akemi
As far as I remember yes. D.
Change *all* signmodules to zero as a workround.
%define signmodules 0
It would go through then. Or else, build in mock. According to Johnny, the modsign problem does not occur in mock.
Akemi
David Hrbáč wrote:
Akemi Yagi napsal(a):
Change *all* signmodules to zero as a workround.
%define signmodules 0
It would go through then. Or else, build in mock. According to Johnny, the modsign problem does not occur in mock.
Akemi
OK, thanks. D.
I specifically do have that exact same problem when building the kernel in vmware. When I build the kernels inside mock I do not seem to have the problem, but my mock machine is not in VMware, so I am not sure whether it is a problem with VMware or not.
Thanks, Johnny Hughes
On 10/28/07, Johnny Hughes johnny@centos.org wrote:
David Hrbáč wrote:
Akemi Yagi napsal(a):
Change *all* signmodules to zero as a workround.
%define signmodules 0
It would go through then. Or else, build in mock. According to Johnny, the modsign problem does not occur in mock.
Akemi
OK, thanks. D.
I specifically do have that exact same problem when building the kernel in vmware. When I build the kernels inside mock I do not seem to have the problem, but my mock machine is not in VMware, so I am not sure whether it is a problem with VMware or not.
Thanks, Johnny Hughes
This particular problem has been known for a while, at least since FC6, and reported in:
https://bugzilla.redhat.com/show_bug.cgi?id=213974
As noted in that report (right after Johnny's comment), I suspect it shares the same root cause as another bug #246623 filed for Fedora. In the latter bug report, someone tracked it down to a bug in linux-2.6-execshield.patch (kernel) and provided a fix.
To find out if that fix cures the modsign problem as well, I first built a c5 i686 kernel in a vmware guest and made sure it ended up with a segfault. Then I applied the fix and tried again. This time the rebuild finished without any issue.
Testing it only once may not give us anything conclusive, but it is a good start. Anyway, I would like to add this observation to the above bug report. In the meantime if this is tested by more people, it will be helpful. David, could you try it? (I know you are already overloaded :-)
Akemi
Akemi Yagi napsal(a):
This particular problem has been known for a while, at least since FC6, and reported in:
https://bugzilla.redhat.com/show_bug.cgi?id=213974
As noted in that report (right after Johnny's comment), I suspect it shares the same root cause as another bug #246623 filed for Fedora. In the latter bug report, someone tracked it down to a bug in linux-2.6-execshield.patch (kernel) and provided a fix.
To find out if that fix cures the modsign problem as well, I first built a c5 i686 kernel in a vmware guest and made sure it ended up with a segfault. Then I applied the fix and tried again. This time the rebuild finished without any issue.
Testing it only once may not give us anything conclusive, but it is a good start. Anyway, I would like to add this observation to the above bug report. In the meantime if this is tested by more people, it will be helpful. David, could you try it? (I know you are already overloaded :-)
Akemi
Akemi, what revision did you test? http://cvs.fedora.redhat.com/viewcvs/devel/kernel/linux-2.6-execshield.patch... Tanks, David
On 10/29/07, David Hrbáč hrbac.conf@seznam.cz wrote:
Akemi Yagi napsal(a):
Testing it only once may not give us anything conclusive, but it is a good start. Anyway, I would like to add this observation to the above bug report. In the meantime if this is tested by more people, it will be helpful. David, could you try it? (I know you are already overloaded :-)
Akemi, what revision did you test? http://cvs.fedora.redhat.com/viewcvs/devel/kernel/linux-2.6-execshield.patch... Tanks, David
I did my testing with the latest CentOS kernel-2.6.18-8.1.15.el5.i686.
Akemi
Akemi Yagi napsal(a):
http://cvs.fedora.redhat.com/viewcvs/devel/kernel/linux-2.6-execshield.patch...
I did my testing with the latest CentOS kernel-2.6.18-8.1.15.el5.i686.
I mean linux-2.6-execshield.patch revision as on http://cvs.fedora.redhat.com/viewcvs/devel/kernel/linux-2.6-execshield.patch... Thanks, David
On 10/29/07, David Hrbáč hrbac.conf@seznam.cz wrote:
Akemi Yagi napsal(a):
http://cvs.fedora.redhat.com/viewcvs/devel/kernel/linux-2.6-execshield.patch...
I did my testing with the latest CentOS kernel-2.6.18-8.1.15.el5.i686.
I mean linux-2.6-execshield.patch revision as on http://cvs.fedora.redhat.com/viewcvs/devel/kernel/linux-2.6-execshield.patch... Thanks, David
Sorry, I misunderstood. I manually edited the file. :)
But wait for a while. I do not want to waste your time and your cputime. I realize I need to do more testing before asking others. As you know more than anyone else, these things take a lot of (cpu)time.
Akemi
On 10/29/07, Akemi Yagi amyagi@gmail.com wrote:
On 10/29/07, David Hrbáč hrbac.conf@seznam.cz wrote:
But wait for a while. I do not want to waste your time and your cputime. I realize I need to do more testing before asking others. As you know more than anyone else, these things take a lot of (cpu)time.
David,
After successful completion of a rebuild under the "fixed" kernel, I rolled back to the previous kernel and tried again. This time it did not segfault. And this was the exact same condition as before (same VMware guest, same kernel) and yet segfault did not occur. In a situation like this, it is very difficult to assess a bug fix unless test runs are repeated enough number of times -- which will take many days or even weeks.
I will wait for 5.1 because there is some possibility that this issue is resolved in the newer kernel. In the meantime, the best workaround is to disable signmodules or to use mock.
Akemi
On 10/30/07, David Hrbáč hrbac.conf@seznam.cz wrote:
Akemi Yagi napsal(a):
I will wait for 5.1 because there is some possibility that this issue is resolved in the newer kernel. In the meantime, the best workaround is to disable signmodules or to use mock.
Akemi
Or build under kernel-2.6.18-8.1.14 :o) David
You never had this problem under 2.6.18-8.1.14 ? That is curious. I do not know exactly how far back this started happening, but it must be quite sometime ago. Guess Johnny can tell us more.
Akemi
Akemi Yagi napsal(a):
You never had this problem under 2.6.18-8.1.14 ? That is curious. I do not know exactly how far back this started happening, but it must be quite sometime ago. Guess Johnny can tell us more.
Akemi
Yes, I have pointed it out in the very first post. A new ICH9 patched kernel builds are done in 2.6.18-8.1.14 again, smoothly. David
David Hrbáč wrote:
Akemi Yagi napsal(a):
You never had this problem under 2.6.18-8.1.14 ? That is curious. I do not know exactly how far back this started happening, but it must be quite sometime ago. Guess Johnny can tell us more.
Yes, I have pointed it out in the very first post. A new ICH9 patched kernel builds are done in 2.6.18-8.1.14 again, smoothly.
As far as this particular problem goes, it is hit and miss.
I might get 3 or 4 failures in ten kernel builds ... though it is always on a module signing.
I have not noticed if it happens more or less often on which kernels as I always just build my kernel via mock anyway, as that is how we build them for release.
Because of this problem, I have even started building all my kernels via mock (and everything else I build except kmod packages).
Thanks, Johnny Hughes
-
Akemi Yagi -
David Hrbáč -
Johnny Hughes