kdbus for 4.2
Hi all,
Given the email threads about kdbus this week from various community members, I figured it would be good to let people know what was going on with things at the moment.
We (David, Daniel, Djalal, and myself) do not need to rush anything and will be delaying the kdbus merge request for another cycle. There still seems to be some disagreement and we are confident that with more time to review this can be sorted out. We have asked distributions to start testing kdbus and have already gotten back valuable feedback from real world testing. We are incorporating that feedback, fixing bugs, and optimizing some internal details which will all be part of the pull request for the next release cycle.
Eric, I hope you can relax now and get some sleep :)
thanks,
greg k-h
Quoted 6 lines · Greg Kroah-Hartman
Hi all,
Given the email threads about kdbus this week from various community members, I figured it would be good to let people know what was going on with things at the moment.
Quoted 8 lines
We (David, Daniel, Djalal, and myself) do not need to rush anything and will be delaying the kdbus merge request for another cycle. There still seems to be some disagreement and we are confident that with more time to review this can be sorted out. We have asked distributions to start testing kdbus and have already gotten back valuable feedback from real world testing. We are incorporating that feedback, fixing bugs, and optimizing some internal details which will all be part of the pull request for the next release cycle.
What's a good distro on which to poke at a full running system? Fedora Rawhide seems to still build systemd with --disable-kdbus, although I suppose that means that I could boot it with kdbus=1?
--Andy
Quoted 1 line · Andy Lutomirski
What's a good distro on which to poke at a full running system?
Fedora Rawhide is probably currently your best bet.
Quoted 2 lines
Fedora Rawhide seems to still build systemd with --disable-kdbus, although I suppose that means that I could boot it with kdbus=1?
Correct.
Cheers,
Tom
Quoted 9 lines · Tom Gundersen
Quoted 1 line · Andy Lutomirski
What's a good distro on which to poke at a full running system?
Fedora Rawhide is probably currently your best bet.
Quoted 2 lines
Fedora Rawhide seems to still build systemd with --disable-kdbus, although I suppose that means that I could boot it with kdbus=1?
Correct.
Does Fedora offer a RPM with systemd from git?
I wonder how to install a bleeding edge systemd in a sane way. ./configure && make && make install it a bit archaic.
Signature · 2 lines
Thanks, //richard
Quoted 4 lines · Tom Gundersen
Quoted 1 line · Andy Lutomirski
What's a good distro on which to poke at a full running system?
Fedora Rawhide is probably currently your best bet.
I won't swear that all of the problems I'm seeing are 100% kdbus' fault, but Rawhide, updated today, has serious problems with the kdbus branch in Greg's char-misc tree.
Problem 1: Booting a kdbus-enabled kernel (CONFIG_KDBUS=y) causes gdm to bail saying "oops, something went wrong" or whatever the useless standard error message is.
I can work around problem 1 by booting with kdbus=1, but that's not okay. Unless this is limited to just some narrow range of Rawhide versions, I don't think the kernel gets to make changes that break userspace like that. Maybe this is a kernel issue, not a user issue, in which case it's not a big deal as long as it gets fixed.
Problem 2: Running 'sudo mount /mnt/share' from a terminal hangs the whole graphical session hard. This is repeatable. /mnt/share is virtfs, but I doubt that matters.
If I build exactly the same sources but set CONFIG_KDBUS=n, all is well.
I suspect that there are multiple issues here:
1. The current state of systemd's --disable-kdbus could be buggy. kdbusfs still gets mounted but isn't set up and then things break.
2. mount generates a bunch of dbus traffic. Apparently on the latest char-misc kdbus tree with the latest Rawhide, it blows up somehow.
That's all from me until next week.
--Andy
Hi
Quoted 9 lines · Andy Lutomirski
Problem 1: Booting a kdbus-enabled kernel (
CONFIG_KDBUS=y) causes gdm to bail saying "oops, something went wrong" or whatever the useless standard error message is.I can work around problem 1 by booting with kdbus=1, but that's not okay. Unless this is limited to just some narrow range of Rawhide versions, I don't think the kernel gets to make changes that break userspace like that. Maybe this is a kernel issue, not a user issue, in which case it's not a big deal as long as it gets fixed.
You're saying booting with the same kernel but kdbus not compiled in works?
Quoted 3 lines
Problem 2: Running 'sudo mount
/mnt/share' from a terminal hangs the whole graphical session hard. This is repeatable./mnt/shareis virtfs, but I doubt that matters.
This is triggered by running through pam from outside the gfx-session but on a shared VT. It's not directly related to kdbus, though. It's fixed in systemd-git. As a workaround, you can remove pam_systemd from the sudo/su pam config.
Thanks
David
Quoted 15 lines · David Herrmann
Hi
Quoted 9 lines · Andy Lutomirski
Problem 1: Booting a kdbus-enabled kernel (
CONFIG_KDBUS=y) causes gdm to bail saying "oops, something went wrong" or whatever the useless standard error message is.I can work around problem 1 by booting with kdbus=1, but that's not okay. Unless this is limited to just some narrow range of Rawhide versions, I don't think the kernel gets to make changes that break userspace like that. Maybe this is a kernel issue, not a user issue, in which case it's not a big deal as long as it gets fixed.
You're saying booting with the same kernel but kdbus not compiled in works?
Quoted 8 lines
Quoted 3 lines
Problem 2: Running 'sudo mount
/mnt/share' from a terminal hangs the whole graphical session hard. This is repeatable./mnt/shareis virtfs, but I doubt that matters.This is triggered by running through pam from outside the gfx-session but on a shared VT. It's not directly related to kdbus, though. It's fixed in systemd-git. As a workaround, you can remove
pam_systemdfrom the sudo/su pam config.
Just because I'm starring right now at similar issues, how can I run systemd-git?
I fear to use a plain "make && make install" as I don't want to break my distro's package management and I want to make sure that nothing from the currently installed systemd influences the system.
Signature · 2 lines
Thanks, //richard
Quoted 25 lines · Richard Weinberger
Quoted 15 lines · David Herrmann
Hi
Quoted 9 lines · Andy Lutomirski
Problem 1: Booting a kdbus-enabled kernel (
CONFIG_KDBUS=y) causes gdm to bail saying "oops, something went wrong" or whatever the useless standard error message is.I can work around problem 1 by booting with kdbus=1, but that's not okay. Unless this is limited to just some narrow range of Rawhide versions, I don't think the kernel gets to make changes that break userspace like that. Maybe this is a kernel issue, not a user issue, in which case it's not a big deal as long as it gets fixed.
You're saying booting with the same kernel but kdbus not compiled in works?
Quoted 8 lines
Quoted 3 lines
Problem 2: Running 'sudo mount
/mnt/share' from a terminal hangs the whole graphical session hard. This is repeatable./mnt/shareis virtfs, but I doubt that matters.This is triggered by running through pam from outside the gfx-session but on a shared VT. It's not directly related to kdbus, though. It's fixed in systemd-git. As a workaround, you can remove
pam_systemdfrom the sudo/su pam config.
Quoted 5 lines
Just because I'm starring right now at similar issues, how can I run systemd-git?
I fear to use a plain "make && make install" as I don't want to break my distro's package management and I want to make sure that nothing from the currently installed systemd influences the system.
Use your distro's package management system to build a new systemd package based on just adding the single big patch diff from the last release to the git head, or pick individual ones. But really, this isn't a kernel issue, you know how to build userspace distro packages...
greg k-h
Am 03.07.2015 um 17:27 schrieb Greg Kroah-Hartman:
Quoted 32 lines · Richard Weinberger
Quoted 25 lines · David Herrmann
Quoted 15 lines
Hi
Quoted 9 lines · Andy Lutomirski
Problem 1: Booting a kdbus-enabled kernel (
CONFIG_KDBUS=y) causes gdm to bail saying "oops, something went wrong" or whatever the useless standard error message is.I can work around problem 1 by booting with kdbus=1, but that's not okay. Unless this is limited to just some narrow range of Rawhide versions, I don't think the kernel gets to make changes that break userspace like that. Maybe this is a kernel issue, not a user issue, in which case it's not a big deal as long as it gets fixed.
You're saying booting with the same kernel but kdbus not compiled in works?
Quoted 8 lines
Quoted 3 lines
Problem 2: Running 'sudo mount
/mnt/share' from a terminal hangs the whole graphical session hard. This is repeatable./mnt/shareis virtfs, but I doubt that matters.This is triggered by running through pam from outside the gfx-session but on a shared VT. It's not directly related to kdbus, though. It's fixed in systemd-git. As a workaround, you can remove
pam_systemdfrom the sudo/su pam config.Quoted 5 lines
Just because I'm starring right now at similar issues, how can I run systemd-git?
I fear to use a plain "make && make install" as I don't want to break my distro's package management and I want to make sure that nothing from the currently installed systemd influences the system.
Quoted 4 lines
Use your distro's package management system to build a new systemd package based on just adding the single big patch diff from the last release to the git head, or pick individual ones. But really, this isn't a kernel issue, you know how to build userspace distro packages...
I hoped some distros already offer such a package. Maybe Fedora? Building packages is no fun nor trivial.
And IMHO this belongs here as systemd is the only user of kdbus and kernel folks may want to give it a try.
Thanks,
//richard
Quoted 8 lines · David Herrmann
Hi
Quoted 4 lines · Andy Lutomirski
Problem 1: Booting a kdbus-enabled kernel (
CONFIG_KDBUS=y) causes gdm to bail saying "oops, something went wrong" or whatever the useless standard error message is.
Quoted 8 lines
Quoted 5 lines
I can work around problem 1 by booting with kdbus=1, but that's not okay. Unless this is limited to just some narrow range of Rawhide versions, I don't think the kernel gets to make changes that break userspace like that. Maybe this is a kernel issue, not a user issue, in which case it's not a big deal as long as it gets fixed.
You're saying booting with the same kernel but kdbus not compiled in works?
Yes, exactly.
The bad boot gets a couple warnings about dbus names getting lost or something like that.
Quoted 9 lines
Quoted 3 lines
Problem 2: Running 'sudo mount
/mnt/share' from a terminal hangs the whole graphical session hard. This is repeatable./mnt/shareis virtfs, but I doubt that matters.This is triggered by running through pam from outside the gfx-session but on a shared VT. It's not directly related to kdbus, though. It's fixed in systemd-git. As a workaround, you can remove
pam_systemdfrom the sudo/su pam config.
I'll try on Monday. Or maybe someone will build a new systemd on Rawhide :)
--Andy
Richard Weinberger
Acked-by Daniel MackGreg,
Quoted 6 lines · Greg Kroah-Hartman
Hi all,
Given the email threads about kdbus this week from various community members, I figured it would be good to let people know what was going on with things at the moment.
Quoted 8 lines
We (David, Daniel, Djalal, and myself) do not need to rush anything and will be delaying the kdbus merge request for another cycle. There still seems to be some disagreement and we are confident that with more time to review this can be sorted out. We have asked distributions to start testing kdbus and have already gotten back valuable feedback from real world testing. We are incorporating that feedback, fixing bugs, and optimizing some internal details which will all be part of the pull request for the next release cycle.
I've started digging into current kdbus to understand the changes you made and have a question.
Where does all the development happen?
For example your current tree contains the following commit:
commit f3adf84302fb4fdb6698cf129b96546b47e27d33
Author: David Herrmann <[email protected]> Date: Thu May 21 20:03:29 2015 +0200
kdbus: skip mandatory items on negotiation
The kdbus negotiation is used to figure out what items and flags an ioctl supports. It is highly impractical to pass in mandatory items when all we do is negotiation. Therefore, allow user-space to skip mandatory items if KDBUS_FLAG_NEGOTIATE is passed.
Where was this patch posted to? Do you have mailinglist where one can follow the discussions? As it carries an ACK there must be some review process.
Signature · 2 lines
Thanks, //richard
Quoted 36 lines · Richard Weinberger
Greg,
Quoted 6 lines · Greg Kroah-Hartman
Hi all,
Given the email threads about kdbus this week from various community members, I figured it would be good to let people know what was going on with things at the moment.
Quoted 8 lines
We (David, Daniel, Djalal, and myself) do not need to rush anything and will be delaying the kdbus merge request for another cycle. There still seems to be some disagreement and we are confident that with more time to review this can be sorted out. We have asked distributions to start testing kdbus and have already gotten back valuable feedback from real world testing. We are incorporating that feedback, fixing bugs, and optimizing some internal details which will all be part of the pull request for the next release cycle.
I've started digging into current kdbus to understand the changes you made and have a question.
Where does all the development happen?For example your current tree contains the following commit:
commit
f3adf84302fb4fdb6698cf129b96546b47e27d33Author: David Herrmann <[email protected]> Date: Thu May 21 20:03:29 2015 +0200kdbus: skip mandatory items on negotiation
The kdbus negotiation is used to figure out what items and flags an ioctl supports. It is highly impractical to pass in mandatory items when all we do is negotiation. Therefore, allow user-space to skip mandatory items if
KDBUS_FLAG_NEGOTIATEis passed.
Quoted 4 lines
Signed-off-by David Herrmann <[email protected]>Acked-by Daniel Mack <[email protected]>Where was this patch posted to?
There was a pull request that happened on the linux-kernel mailing list with this patch in it, along with others.
thanks,
greg k-h
Am 29.06.2015 um 17:45 schrieb Greg Kroah-Hartman:
Quoted 37 lines · Richard Weinberger
Quoted 36 lines
Greg,
Quoted 6 lines · Greg Kroah-Hartman
Hi all,
Given the email threads about kdbus this week from various community members, I figured it would be good to let people know what was going on with things at the moment.
Quoted 8 lines
We (David, Daniel, Djalal, and myself) do not need to rush anything and will be delaying the kdbus merge request for another cycle. There still seems to be some disagreement and we are confident that with more time to review this can be sorted out. We have asked distributions to start testing kdbus and have already gotten back valuable feedback from real world testing. We are incorporating that feedback, fixing bugs, and optimizing some internal details which will all be part of the pull request for the next release cycle.
I've started digging into current kdbus to understand the changes you made and have a question.
Where does all the development happen?For example your current tree contains the following commit:
commit
f3adf84302fb4fdb6698cf129b96546b47e27d33Author: David Herrmann <[email protected]> Date: Thu May 21 20:03:29 2015 +0200kdbus: skip mandatory items on negotiation
The kdbus negotiation is used to figure out what items and flags an ioctl supports. It is highly impractical to pass in mandatory items when all we do is negotiation. Therefore, allow user-space to skip mandatory items if
KDBUS_FLAG_NEGOTIATEis passed.
Quoted 7 lines
Quoted 4 lines
Signed-off-by David Herrmann <[email protected]>Acked-by Daniel Mack <[email protected]>Where was this patch posted to?
There was a pull request that happened on the linux-kernel mailing list with this patch in it, along with others.
I saw the pull request on LKML but not the patch. That's why I'm asking. :)
Thanks,
//richard
David's out tree modules kdbus tree
https://github.com/systemd/kdbus
Quoted 52 lines · Richard Weinberger
Am 29.06.2015 um 17:45 schrieb Greg Kroah-Hartman:
Quoted 36 lines · Richard Weinberger
Quoted 35 lines
Greg,
Quoted 6 lines · Greg Kroah-Hartman
Hi all,
Given the email threads about kdbus this week from various community members, I figured it would be good to let people know what was going on with things at the moment.
Quoted 8 lines
We (David, Daniel, Djalal, and myself) do not need to rush anything and will be delaying the kdbus merge request for another cycle. There still seems to be some disagreement and we are confident that with more time to review this can be sorted out. We have asked distributions to start testing kdbus and have already gotten back valuable feedback from real world testing. We are incorporating that feedback, fixing bugs, and optimizing some internal details which will all be part of the pull request for the next release cycle.
I've started digging into current kdbus to understand the changes you made and have a question.
Where does all the development happen?For example your current tree contains the following commit:
commit
f3adf84302fb4fdb6698cf129b96546b47e27d33Author: David Herrmann <[email protected]> Date: Thu May 21 20:03:29 2015 +0200kdbus: skip mandatory items on negotiation
The kdbus negotiation is used to figure out what items and flags an ioctl supports. It is highly impractical to pass in mandatory items when all we do is negotiation. Therefore, allow user-space to skip mandatory items if
KDBUS_FLAG_NEGOTIATEis passed.Quoted 6 lines
Quoted 4 lines
Signed-off-by David Herrmann <[email protected]>Acked-by Daniel Mack <[email protected]>Where was this patch posted to?
There was a pull request that happened on the linux-kernel mailing list with this patch in it, along with others.
I saw the pull request on LKML but not the patch. That's why I'm asking. :)
Thanks,
//richard
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Am 30.06.2015 um 03:06 schrieb chunshan.zhu:
Quoted 4 lines
David's out tree modules kdbus tree
Thank you Chunshan, but this does not an answer to my questions.
Let's try again. ;)
Where does (or did?) the development happen?
Will it happen on LKML in future?
Is there a mailinglist?
The reason I'm asking is, reading the discussion behind patches is very useful to understand them.
If a reviewer can see the path to a patch with all patch iterations it can help to understand it better and faster.
Of course this is not a must, but *very* useful.
Thanks,
//richard
Hi
Quoted 1 line · Richard Weinberger
Where does (or did?) the development happen?
So far development happened at conferences, on Freenode #kdbus, and in personal meetings.
Thanks
David
Quoted 8 lines · David Herrmann
Hi
Quoted 1 line · Richard Weinberger
Where does (or did?) the development happen?
So far development happened at conferences, on Freenode #kdbus, and in personal meetings.
So the code is done there too?
For me, I discuss designs and such at conferences, IRC and in personal meetings. But when I get down and code, I post my results to a mailing list and ask for feedback. The nice thing about doing it this way is that it leaves a historical archive and nice history of the development.
Again, as Richard stated. This isn't a must have, but a really nice to have.
-- Steve