systemd requiring Linux >= 3.7

classic Classic list List threaded Threaded
12 messages Options
Reply | Threaded
Open this post in threaded view
|

systemd requiring Linux >= 3.7

csirac2
With our dear universal operating system set to switch over to systemd, I am just wondering if anybody has communicated that this breaks many ARM platforms with "typical" vendors who only care to ship a kernel they once hacked at product launch, and/or the one provided by CPU vendor who barely does much more than fork and abandon stuff on linaro.org.

Okay, Linaro isn't that bad, the expensive ARM chips are better supported than that and the sky isn't really falling. I actually really do like systemd features (though I think complaints about its monolithic approach are valid) and I currently maintain a systemd  build of my work for a candidate ARM target which mostly works well.

Except that critical out-of-tree kernel modules written for 3.0 need to be ported to a newer kernel, and undergo expensive re-validation.

Eg. Congatec still actively maintains its fork of Freescale's fork of Linaro kernel <a href="tel:3015">3.0.15: https://git.congatec.com/yocto/meta-fsl-arm-extra

Count how many of gumstix' offerings officially run Linux kernels >= 3.7 (hint: zero) http://www.gumstix.org/access-source-code.html

These vendor's products easily run Debian today but won't boot a Jessie image with systemd.

Not because the CPUs are unable but just the sheer fork-happy, hack&slash insanity of  software practice in the embedded space. Has this been communicated to the vote participants?

Or am I completely off-base here? Most of my career has been x86-only until recently.

Cheers
--
Paul Harvey

Sent from my android device.
Reply | Threaded
Open this post in threaded view
|

Re: systemd requiring Linux >= 3.7

Jeremiah Foster-10
Hi Paul,

On Fri, Oct 24, 2014 at 7:49 AM, <[hidden email]> wrote:
With our dear universal operating system set to switch over to systemd, I am just wondering if anybody has communicated that this breaks many ARM platforms

I'm not sure what you mean here by "breaks many ARM platforms." The examples you give below, like Congatec, are already broken from a Debian perspective since they're not "based" on Debian like the Raspberry Pi and the Beagle Bone Black are. I think instead the vendors have chosen another approach, Yocto in Congatech's case which matches Freescale's approach. But even there you can still run Debian on those machines no? And its not just systemd that won't work in Jessie, its anything that relies on control groups from the kernel. 

You can of course swap out systemd for another init system in Jessie, nothing prevents you from doing that. That shouldn't be too hard since most of those embedded devices can boot from a SD card on which you can craft an image with debootstrap without systemd, but of course that may depend on your build process and might get fiddly with cross compilation. 
 
with "typical" vendors who only care to ship a kernel they once hacked at product launch, and/or the one provided by CPU vendor who barely does much more than fork and abandon stuff on linaro.org.

Okay, Linaro isn't that bad, the expensive ARM chips are better supported than that and the sky isn't really falling. I actually really do like systemd features (though I think complaints about its monolithic approach are valid) and I currently maintain a systemd  build of my work for a candidate ARM target which mostly works well.

Except that critical out-of-tree kernel modules written for 3.0 need to be ported to a newer kernel, and undergo expensive re-validation.

Eg. Congatec still actively maintains its fork of Freescale's fork of Linaro kernel <a href="tel:3015" target="_blank">3.0.15: https://git.congatec.com/yocto/meta-fsl-arm-extra

Count how many of gumstix' offerings officially run Linux kernels >= 3.7 (hint: zero) http://www.gumstix.org/access-source-code.html

These vendor's products easily run Debian today but won't boot a Jessie image with systemd.

Not because the CPUs are unable but just the sheer fork-happy, hack&slash insanity of  software practice in the embedded space. Has this been communicated to the vote participants?

How would this be communicated? Would you say that "some companies have old devices that use old kernels and they won't update." Wouldn't the reply be "backports"?
 

Or am I completely off-base here? Most of my career has been x86-only until recently.

x86 is embedded, for $value of embedded. ;)

Cheers,

Jeremiah
 
Cheers
--
Paul Harvey
Reply | Threaded
Open this post in threaded view
|

Re: systemd requiring Linux >= 3.7

csirac2
In reply to this post by csirac2
Thanks for your response.

If you read the proposal https://lists.debian.org/debian-vote/2014/10/msg00001.html that's exactly what is being voted on: to preserve the option to boot with sysvinit.

Presumably if the vote comes out in the negative, then at best, we will slowly lose this option as packages remove their init.d scripts during Jessie's life cycle.

Sent from my android device.

-----Original Message-----
From: Johannes Schauer <[hidden email]>
To: [hidden email], [hidden email]
Sent: Fri, 24 Oct 2014 18:00
Subject: Re: systemd requiring Linux >= 3.7

Hi,

Quoting [hidden email] (2014-10-24 07:49:46)
> These vendor's products easily run Debian today but won't boot a Jessie image
> with systemd.

how about booting a Jessie image without systemd then?

cheers, josch
Reply | Threaded
Open this post in threaded view
|

Re: systemd requiring Linux >= 3.7

Jean-Christian de Rivaz
In reply to this post by csirac2
Le 24. 10. 14 07:49, [hidden email] a écrit :

> With our dear universal operating system set to switch over to
> systemd, I am just wondering if anybody has communicated that this
> breaks many ARM platforms with "typical" vendors who only care to ship
> a kernel they once hacked at product launch, and/or the one provided
> by CPU vendor who barely does much more than fork and abandon stuff on
> linaro <http://linaro.org>. <http://linaro.org>org <http://linaro.org>.
>
> Okay, Linaro isn't that bad, the expensive ARM chips are better
> supported than that and the sky isn't really falling. I actually
> really do like systemd features (though I think complaints about its
> monolithic approach are valid) and I currently maintain a systemd  
> build of my work for a candidate ARM target which mostly works well.
>
> Except that critical out-of-tree kernel modules written for 3.0 need
> to be ported to a newer kernel, and undergo expensive re-validation.
>
> Eg. Congatec still actively maintains its fork of Freescale's fork of
> Linaro kernel 3.0.15 <tel:3015>: https
> <https://git.congatec.com/yocto/meta-fsl-arm-extra>://
> <https://git.congatec.com/yocto/meta-fsl-arm-extra>git.congatec.com
> <https://git.congatec.com/yocto/meta-fsl-arm-extra>/
> <https://git.congatec.com/yocto/meta-fsl-arm-extra>yocto
> <https://git.congatec.com/yocto/meta-fsl-arm-extra>/
> <https://git.congatec.com/yocto/meta-fsl-arm-extra>meta-
> <https://git.congatec.com/yocto/meta-fsl-arm-extra>fsl-arm-extra
> <https://git.congatec.com/yocto/meta-fsl-arm-extra>
>
> Count how many of gumstix' offerings officially run Linux kernels >=
> 3.7 (hint: zero) http://www.gumstix.org/access-source-code.html
>
> These vendor's products easily run Debian today but won't boot a
> Jessie image with systemd.
>
> Not because the CPUs are unable but just the sheer fork-happy,
> hack&slash insanity of  software practice in the embedded space. Has
> this been communicated to the vote participants?
>
> Or am I completely off-base here? Most of my career has been x86-only
> until recently.

Hi Paul,

I few days ago I completely switched a SAMA5D35 ARM Cortex-A5 custom
board from EmDebian Whezzy Grip with system V inito a pure Debian Jessie
with systemd. I simply configured the apt sources files and do a normal
dist-upgrade (aside as having to force the version of a few packages to
completely avoid gripped version). The process was so smooth that all
the realtime processing applications running on that board didn't even
notice the unusual activities (well until the postgresql database
restarted with both 9.1 and 9.4 revisions running alongside). At the
reboot, systemd was in charge of the base of the system and everything
was good. Even the old custom /etc/init.d/* custom scripts specific on
that board was executed the right way. It was a complete success, and I
enjoy the systemd-analyze command.

Debian Jessy with systemd is just already incredibly perfect. I predict
that Jessis will be one of the most successful Debian release to date
and will play a major role in the embedded market. For the first time
ever, the armhf port is so complete that you can do on armhf everything
you can do on a amd64 port. I used to work on custom build, on
scratchbox build, on buildroot build, on openembedded build, but now I
do everything natively on armhf directly on the target board and really
enjoy doing so. Really, Debian Jessie is a major wonderful advance, try it.

Best Regards,
Jean-Christian


--
To UNSUBSCRIBE, email to [hidden email]
with a subject of "unsubscribe". Trouble? Contact [hidden email]
Archive: https://lists.debian.org/544A419E.6090308@...

Reply | Threaded
Open this post in threaded view
|

Re: systemd requiring Linux >= 3.7

Paul Boddie
On Friday 24. October 2014 14.10.06 Jean-Christian de Rivaz wrote:
>
> Debian Jessy with systemd is just already incredibly perfect. I predict
> that Jessis will be one of the most successful Debian release to date
> and will play a major role in the embedded market. For the first time
> ever, the armhf port is so complete that you can do on armhf everything
> you can do on a amd64 port. I used to work on custom build, on
> scratchbox build, on buildroot build, on openembedded build, but now I
> do everything natively on armhf directly on the target board and really
> enjoy doing so. Really, Debian Jessie is a major wonderful advance, try it.

It's worth noting that the cross-toolchains are now in Debian proper, at least
if the Debian Wiki can be believed:

"Packages which implement some of what is described here (without the gcc-for-
host/build multiarch magic - just cross toolchains) are available in unstable,
and (soon) in jessie"

https://wiki.debian.org/CrossToolchains

Congratulations and thanks to those responsible!

Paul


--
To UNSUBSCRIBE, email to [hidden email]
with a subject of "unsubscribe". Trouble? Contact [hidden email]
Archive: https://lists.debian.org/201410241505.00128.paul@...

Reply | Threaded
Open this post in threaded view
|

Re: systemd requiring Linux >= 3.7

W. Martin Borgert
In reply to this post by csirac2
Quoting [hidden email]:
> With our dear universal operating system set to switch over to  
> systemd, I am just wondering if anybody has communicated that this  
> breaks many ARM platforms with "typical" vendors who only care to  
> ship a kernel they once hacked at product launch, and/or the one  
> provided by CPU vendor who barely does much more than fork and  
> abandon stuff on linaro.org.

This is not a problem, because Debian will continue to support
sysvinit. It is just not the default. Maybe you will not be able
to run certain programs, such as GDM, but how relevant is this?

I'm maintaining an embedded product myself. It is based on
Debian with kernel 3.2 and upstart! We will probably update the
kernel and move to systemd. But if we could not update the kernel,
we would just stay with upstart or downgrade to sysvinit.

HTH


--
To UNSUBSCRIBE, email to [hidden email]
with a subject of "unsubscribe". Trouble? Contact [hidden email]
Archive: https://lists.debian.org/20141024152414.Horde.sJE5FAq1ZMCQgT67YKezsw1@...

Reply | Threaded
Open this post in threaded view
|

Re: systemd requiring Linux >= 3.7

csirac2
In reply to this post by Jean-Christian de Rivaz
Hi Jean, I'm not here to complain about systemd. Did you read my Emails? I said that Jessie with systemd is good. I already maintain a Jessie+systemd  build of my work, though it's not useful because platform-specific kernel modules were written for 3.0 even though 3.0 was already quite old at the time.

I'm here wondering if adding this circumstance to consideration for retaining sysvinit capability in Jessir is worth throwing into the discussion. Given that so far I seem to be the only one on the planet in this predicament stuck supporting new, current hardware that's locked into kernel 3.0, perhaps not.

Cheers

Sent from my android device.

-----Original Message-----
From: Jean-Christian de Rivaz <[hidden email]>
To: [hidden email]
Sent: Fri, 24 Oct 2014 23:27
Subject: Re: systemd requiring Linux >= 3.7

Le <a href="tel:24101407">24. 10. 14 07:49, [hidden email] a écrit :
> With our dear universal operating system set to switch over to
> systemd, I am just wondering if anybody has communicated that this
> breaks many ARM platforms with "typical" vendors who only care to ship
> a kernel they once hacked at product launch, and/or the one provided
> by CPU vendor who barely does much more than fork and abandon stuff on
> linaro <http://linaro.org>. <http://linaro.org>org <http://linaro.org>.
>
> Okay, Linaro isn't that bad, the expensive ARM chips are better
> supported than that and the sky isn't really falling. I actually
> really do like systemd features (though I think complaints about its
> monolithic approach are valid) and I currently maintain a systemd 
> build of my work for a candidate ARM target which mostly works well.
>
> Except that critical out-of-tree kernel modules written for 3.0 need
> to be ported to a newer kernel, and undergo expensive re-validation.
>
> Eg. Congatec still actively maintains its fork of Freescale's fork of
> Linaro kernel <a href="tel:3015">3.0.15 <tel:<a href="tel:3015">3015>: https
> <https://git.congatec.com/yocto/meta-fsl-arm-extra>://
> <https://git.congatec.com/yocto/meta-fsl-arm-extra>git.congatec.com
> <https://git.congatec.com/yocto/meta-fsl-arm-extra>/
> <https://git.congatec.com/yocto/meta-fsl-arm-extra>yocto
> <https://git.congatec.com/yocto/meta-fsl-arm-extra>/
> <https://git.congatec.com/yocto/meta-fsl-arm-extra>meta-
> <https://git.congatec.com/yocto/meta-fsl-arm-extra>fsl-arm-extra
> <https://git.congatec.com/yocto/meta-fsl-arm-extra>
>
> Count how many of gumstix' offerings officially run Linux kernels >=
> 3.7 (hint: zero) http://www.gumstix.org/access-source-code.html
>
> These vendor's products easily run Debian today but won't boot a
> Jessie image with systemd.
>
> Not because the CPUs are unable but just the sheer fork-happy,
> hack&slash insanity of  software practice in the embedded space. Has
> this been communicated to the vote participants?
>
> Or am I completely off-base here? Most of my career has been x86-only
> until recently.

Hi Paul,

I few days ago I completely switched a SAMA5D35 ARM Cortex-A5 custom
board from EmDebian Whezzy Grip with system V inito a pure Debian Jessie
with systemd. I simply configured the apt sources files and do a normal
dist-upgrade (aside as having to force the version of a few packages to
completely avoid gripped version). The process was so smooth that all
the realtime processing applications running on that board didn't even
notice the unusual activities (well until the postgresql database
restarted with both 9.1 and 9.4 revisions running alongside). At the
reboot, systemd was in charge of the base of the system and everything
was good. Even the old custom /etc/init.d/* custom scripts specific on
that board was executed the right way. It was a complete success, and I
enjoy the systemd-analyze command.

Debian Jessy with systemd is just already incredibly perfect. I predict
that Jessis will be one of the most successful Debian release to date
and will play a major role in the embedded market. For the first time
ever, the armhf port is so complete that you can do on armhf everything
you can do on a amd64 port. I used to work on custom build, on
scratchbox build, on buildroot build, on openembedded build, but now I
do everything natively on armhf directly on the target board and really
enjoy doing so. Really, Debian Jessie is a major wonderful advance, try it.

Best Regards,
Jean-Christian


--
To UNSUBSCRIBE, email to [hidden email]
with a subject of "unsubscribe". Trouble? Contact [hidden email]
Archive: https://lists.debian.org/544A419E.6090308@...

Reply | Threaded
Open this post in threaded view
|

Re: systemd requiring Linux >= 3.7

csirac2
In reply to this post by Paul Boddie
Yep, they are fantastic. I still have some qemu builds for cmake stuff where xapt doesn't quite get package naming right (somehow arch: all packages don't translate well), the emdebian toolchain has been fantastic and is exactly why I'm using Debian.

Sent from my android device.

-----Original Message-----
From: Paul Boddie <[hidden email]>
To: [hidden email]
Sent: Sat, 25 Oct 2014 0:21
Subject: Re: systemd requiring Linux >= 3.7

On Friday 24. October <a href="tel:2014141006">2014 14.10.06 Jean-Christian de Rivaz wrote:
>
> Debian Jessy with systemd is just already incredibly perfect. I predict
> that Jessis will be one of the most successful Debian release to date
> and will play a major role in the embedded market. For the first time
> ever, the armhf port is so complete that you can do on armhf everything
> you can do on a amd64 port. I used to work on custom build, on
> scratchbox build, on buildroot build, on openembedded build, but now I
> do everything natively on armhf directly on the target board and really
> enjoy doing so. Really, Debian Jessie is a major wonderful advance, try it.

It's worth noting that the cross-toolchains are now in Debian proper, at least
if the Debian Wiki can be believed:

"Packages which implement some of what is described here (without the gcc-for-
host/build multiarch magic - just cross toolchains) are available in unstable,
and (soon) in jessie"

https://wiki.debian.org/CrossToolchains

Congratulations and thanks to those responsible!

Paul


--
To UNSUBSCRIBE, email to [hidden email]
with a subject of "unsubscribe". Trouble? Contact [hidden email]
Archive: https://lists.debian.org/201410241505.00128.paul@...

Reply | Threaded
Open this post in threaded view
|

Re: systemd requiring Linux >= 3.7

csirac2
In reply to this post by W. Martin Borgert
Thanks Martin. Could you give me your thoughts on what a negative vote on the current proposal would mean in practice?

Sent from my android device.

-----Original Message-----
From: "W. Martin Borgert" <[hidden email]>
To: [hidden email]
Sent: Sat, 25 Oct 2014 0:42
Subject: Re: systemd requiring Linux >= 3.7

Quoting [hidden email]:
> With our dear universal operating system set to switch over to 
> systemd, I am just wondering if anybody has communicated that this 
> breaks many ARM platforms with "typical" vendors who only care to 
> ship a kernel they once hacked at product launch, and/or the one 
> provided by CPU vendor who barely does much more than fork and 
> abandon stuff on linaro.org.

This is not a problem, because Debian will continue to support
sysvinit. It is just not the default. Maybe you will not be able
to run certain programs, such as GDM, but how relevant is this?

I'm maintaining an embedded product myself. It is based on
Debian with kernel 3.2 and upstart! We will probably update the
kernel and move to systemd. But if we could not update the kernel,
we would just stay with upstart or downgrade to sysvinit.

HTH


--
To UNSUBSCRIBE, email to [hidden email]
with a subject of "unsubscribe". Trouble? Contact [hidden email]
Archive: https://lists.debian.org/20141024152414.Horde.sJE5FAq1ZMCQgT67YKezsw1@...

Reply | Threaded
Open this post in threaded view
|

Re: systemd requiring Linux >= 3.7

W. Martin Borgert
Quoting [hidden email]:
> Could you give me your thoughts on what a negative vote on the  
> current proposal would mean in practice?

The impact of different results has been analysed by Lucas Nussbaum:
"Tentative summary of the amendments of the init system coupling GR"
(http://www.lucas-nussbaum.net/blog/?p=845)

For Debian 8 (Jessie), there is no need to fear anything. For later
versions it is hard to tell, because it does not depend much on the
vote, but mainly of what both upstream and Debian developers will do.

Just take a look in the crystal ball:

If systemd works well for most people, less effort will be put in
alternatives. If many people have problems with systemd, e.g. in
the embedded community, alternatives will continue to be relevant.

Maybe there will be sufficient pressure on hardware companies to
support newer kernels on their hardware?

Maybe in two, three years from now, there are good reasons, such as
security aspects, not to run a <= 3.7 kernel anyway?

If security is not a concern, maybe just keep Debian 6/7/8 forever?

Cheers


--
To UNSUBSCRIBE, email to [hidden email]
with a subject of "unsubscribe". Trouble? Contact [hidden email]
Archive: https://lists.debian.org/20141024173103.Horde.uSIwd0ilQTie-EZMLyFThg3@...

Reply | Threaded
Open this post in threaded view
|

Re: systemd requiring Linux >= 3.7

Jean-Christian de Rivaz
In reply to this post by csirac2
Le 24. 10. 14 15:49, [hidden email] a écrit :
> Hi Jean, I'm not here to complain about systemd. Did you read my
> Emails? I said that Jessie with systemd is good. I already maintain a
> Jessie+systemd  build of my work, though it's not useful because
> platform-specific kernel modules were written for 3.0 even though 3.0
> was already quite old at the time.

Nothing wrong, I don't complain either :-) I have taken the opportunity
to share how easy it is today to migrate to systemd with Jessie in a
hope to motivate more peoples to try it. It was in response to the
'These vendor's products easily run Debian today but won't boot a Jessie
image with systemd.' part of your email.

> I'm here wondering if adding this circumstance to consideration for
> retaining sysvinit capability in Jessir is worth throwing into the
> discussion. Given that so far I seem to be the only one on the planet
> in this predicament stuck supporting new, current hardware that's
> locked into kernel 3.0, perhaps not.'

Maybe can you explain to this vendors that the cost of the migration
there modules to a newer kernel will be less than the cost of supporting
them self an old, officially unsupported system. Ok, it's hard to say it
it's true today, but this will be certainly true in the future if there
expect to have up to date system. My point of view is that system V init
are counted not only since systemd adoption, but already since upstart
proved his goal.

Note: While mandated for a new board architecture, I strongly push the
hardware to only use standard interfaces available from userspace
applications, to avoids out-of-tree kernel module.

Regards,

Jean-Christian


--
To UNSUBSCRIBE, email to [hidden email]
with a subject of "unsubscribe". Trouble? Contact [hidden email]
Archive: https://lists.debian.org/544A71A3.2080102@...

Reply | Threaded
Open this post in threaded view
|

Re: systemd requiring Linux >= 3.7

csirac2
In reply to this post by W. Martin Borgert
Thanks for the link to Lucas Nussbaum's blog entry, that was very useful.

Given that debsecan typically seems to show fewer, less serious CVEs in
sid/jessie than stable/wheezy for the workloads I have, I will do what I
can to run Jessie for now.

There are certainly a bunch of important privilege escalation and bypass
issues fixed leading up to 3.7 kernels (and since!), only some of which
can be mitigated - but with a tight enough .config, I'd say it's the
userland application vulnerabilities which are going to be the less
complex attack vectors for a given system.

On 25/10/14 02:31, W. Martin Borgert wrote:

> Quoting [hidden email]:
>> Could you give me your thoughts on what a negative vote on the  
>> current proposal would mean in practice?
> The impact of different results has been analysed by Lucas Nussbaum:
> "Tentative summary of the amendments of the init system coupling GR"
> (http://www.lucas-nussbaum.net/blog/?p=845)
>
> For Debian 8 (Jessie), there is no need to fear anything. For later
> versions it is hard to tell, because it does not depend much on the
> vote, but mainly of what both upstream and Debian developers will do.
>
> Just take a look in the crystal ball:
>
> If systemd works well for most people, less effort will be put in
> alternatives. If many people have problems with systemd, e.g. in
> the embedded community, alternatives will continue to be relevant.
>
> Maybe there will be sufficient pressure on hardware companies to
> support newer kernels on their hardware?
>
> Maybe in two, three years from now, there are good reasons, such as
> security aspects, not to run a <= 3.7 kernel anyway?
>
> If security is not a concern, maybe just keep Debian 6/7/8 forever?
>
> Cheers
>
>


--
To UNSUBSCRIBE, email to [hidden email]
with a subject of "unsubscribe". Trouble? Contact [hidden email]
Archive: https://lists.debian.org/544B5462.7020800@...