Problems with late 3.8-rc5 and 3.8-rc6 on i686
I have been testing 3.8 kernels on Fedora. The last good kernel was 3.8.0-0.rc5.git1.1.fc19.i686.PAE. I tested 3.8.0-0.rc5.git2.1.fc19.i686.PAE, 3.8.0-0.rc5.git3.1.fc19.i686 and 3.8.0-0.rc6.git0.1.fc19.i686.PAE and found these all had two odd effects.
One is that the rtkit-daemon service fails to start up and the other is that there are issues with X. One one machine with an rv280 based video card the background image doesn't show up in gdm and I can't enter a password in gdm. C-A-F2 does get me back to a vt. On another machine with a nv28 based video card I get the Oh no something went wrong page, but again I can still use C-A-F2 to get to a vt.
On an x86_64 machine I have tested through 3.8.0-0.rc5.git2.1.fc19.x86_64 and am not seeing either of these issues.
Quoted 2 lines · Bruno Wolff III
One is that the rtkit-daemon service fails to start up and the other is that there are issues with X. One one machine with an rv280 based video card the background image doesn't show up in gdm and I can't enter a password in gdm. C-A-F2 does get me back to a vt. On another machine with a nv28 based video card I get the Oh no something went wrong page, but again I can still use C-A-F2 to get to a vt.
I am still seeing this with 3.8.0-0.rc6.git1.1.fc19.i686.PAE.
However, I have found the GUI issues are because gnome is waiting on the rtkit-daemon and once it times out, things are OK. So the issue is just that the recent kernels break something for the rtkit-daemon service on i686.
Quoted 11 lines · Bruno Wolff III
Quoted 8 lines · Bruno Wolff III
One is that the rtkit-daemon service fails to start up and the other is that there are issues with X. One one machine with an rv280 based video card the background image doesn't show up in gdm and I can't enter a password in gdm. C-A-F2 does get me back to a vt. On another machine with a nv28 based video card I get the Oh no something went wrong page, but again I can still use C-A-F2 to get to a vt.
Quoted 4 lines
I am still seeing this with 3.8.0-0.rc6.git1.1.fc19.i686.PAE.
However, I have found the GUI issues are because gnome is waiting on the rtkit-daemon and once it times out, things are OK.
I had also problems with this and other daemons...
If you are using systemd, check if rtkit wants to run with PrivateTmp=yes. If this is so, then check that your /tmp is _not_ a symlink, but a real dir...
Signature · 1 line
J.A. Magallon <jamagallon()ono!com> \ Winter is coming...
Quoted 5 lines · JA Magallón
I had also problems with this and other daemons...
If you are using systemd, check if rtkit wants to run with PrivateTmp=yes. If this is so, then check that your /tmp is _not_ a symlink, but a real dir...
In this case it was using privatenetwork=yes when namespaces are enabled on 32 bit kernels. Possible an error status changed and systemd is behaving differently now due to that. We have a bug open in Fedora for this against rtkit, but I am going to recommend the submitter move it to systemd.
Thanks.
Quoted 1 line · Bruno Wolff III
I have been testing 3.8 kernels on Fedora. The last good kernel was 3.8.0-0.rc5.git1.1.fc19.i686.PAE. I tested 3.8.0-0.rc5.git2.1.fc19.i686.PAE, 3.8.0-0.rc5.git3.1.fc19.i686 and 3.8.0-0.rc6.git0.1.fc19.i686.PAE and found these all had two odd effects.
This turned out to be due Fedora changing some config options to turn off namespaces for i686 when it was thought by some people that they had been off previously. (And so this wasn't thought to be a change.) This broke the private network feature of systemd, used by rtkit-daemon.
There is a new kernel build which fixes this.
Quoted 9 lines · Bruno Wolff III
Quoted 5 lines · Bruno Wolff III
I have been testing 3.8 kernels on Fedora. The last good kernel was 3.8.0-0.rc5.git1.1.fc19.i686.PAE. I tested 3.8.0-0.rc5.git2.1.fc19.i686.PAE, 3.8.0-0.rc5.git3.1.fc19.i686 and 3.8.0-0.rc6.git0.1.fc19.i686.PAE and found these all had two odd effects.
Quoted 4 lines
This turned out to be due Fedora changing some config options to turn off namespaces for i686 when it was thought by some people that they had been off previously. (And so this wasn't thought to be a change.) This broke the private network feature of systemd, used by rtkit-daemon.
Yeah, the Kconfig entry for NAMESPACES is... misleading. It seems it should depend on CONFIG_EXPERT, but if you have one of the _NS options set in the config it gets flipped on regardless. It's either that or the weird "default !EXPERT" causing it to be enabled even when the CONFIG_EXPERT option is not set.
Anyway, still seems to me that systemd/rtkit-daemon should be _checking_ to see if that feature is present instead of assuming it is and failing in weird ways when it isn't.
josh
Quoted 4 lines · Josh Boyer
Anyway, still seems to me that systemd/rtkit-daemon should be _checking_ to see if that feature is present instead of assuming it is and failing in weird ways when it isn't.
I agree. It would have been a lot quicker to figure this out if there had been a message that plainly stated that namespaces weren't available and either ignored the privatenamespace setting or failed, but suggested turning the setting off.