kernel version management
Hi,
So ... I guess I succeeded in killing off 2.6.24 support, but I think we should talk about when we drop which version?
http://en.wikipedia.org/wiki/Linux_kernel#Maintenance has a handy list of which version was supported until when, telling us that the oldest kernel still supported now is 2.6.32.
What do people think? What kernels do you use? How old is "really far too old to worry about"?
johannes
Quoted 11 lines · Johannes Berg
Hi,
So ... I guess I succeeded in killing off 2.6.24 support, but I think we should talk about when we drop which version?
http://en.wikipedia.org/wiki/Linux_kernel#Maintenance has a handy list of which version was supported until when, telling us that the oldest kernel still supported now is 2.6.32.
What do people think? What kernels do you use? How old is "really far too old to worry about"?
I'm ready to accept killing anything not listed as supported on kernel.org. That should help with speed and also pushing folks to ensure their ecosystem is upgraded appropriately.
Luis
Quoted 5 lines · Johannes Berg
Hi,
So ... I guess I succeeded in killing off 2.6.24 support, but I think we should talk about when we drop which version?
Quoted 8 lines
http://en.wikipedia.org/wiki/Linux_kernel#Maintenance has a handy list of which version was supported until when, telling us that the oldest kernel still supported now is 2.6.32.
What do people think? What kernels do you use? How old is "really far too old to worry about"?
johannes
When you look at the users of old backports no one cares about using a supported kernel version. One group are the desktop users which are using Ubuntu, Debian, or some RedHat kernel, these kernels are based on 2.6.32 or some 3.X version. The other big group are the embedded developers which are using any random kernel version with many patches for some specific SoC in it. They do not care about security updates.
I think we should drop support for some versions if there are some thinks that are harder to backport.
The bigger maintenance problem I currently see is the adding for more and more subsystems. The GPU drivers for example are taking about 30% of the time when adding support for a new linux next version.
Hauke
Quoted 6 lines · Hauke Mehrtens
When you look at the users of old backports no one cares about using a supported kernel version. One group are the desktop users which are using Ubuntu, Debian, or some RedHat kernel, these kernels are based on 2.6.32 or some 3.X version. The other big group are the embedded developers which are using any random kernel version with many patches for some specific SoC in it. They do not care about security updates.
For what it's worth, this meshes with my experience as well. I'd even go so far as to say that the latter group is the more significant.
Whatever kernel a random SoC vendor ships is what will be used until the end of time. Encouraging board vendors to upgrade their software stack to use to newer kernels is a battle that needs to be fought with the SoC folks (not even considering crap like binary-only drivers), and it'll subsequently trickle down to the folks craking out boards. The devicetree migration is a great help here, but it won't do anything to migrate the legacy platforms out there..
One particular battle I fought less than a year ago involved someone unwilling to move from a 2.6.13 (!) kernel, even though I told them in no uncertain terms that it would cost them an order of magnitude more money to backport a wifi driver (and mac80211 etc) to that 2.6.13 kernel than it would be to port their meagre handful of custom drivers to a modern long-term 3.0/3.2 kernel.
And this was a network-exposed system too. ${deity}-only knows how many security holes were in their even-more-ancient userspace.
Quoted 3 lines
The bigger maintenance problem I currently see is the adding for more and more subsystems. The GPU drivers for example are taking about 30% of the time when adding support for a new linux next version.
...Especially when you consider that the GPU/DRM stuff also requires updated userspace in order to do anything meaningful -- which is another big speedbump.
I personally think it's a mistake to drop support for older kernels because of the pain of porting GPU stuffs. I'm beginning to really miss the simple "compat-wireless" days...
- Solomon
Signature · 3 lines
Solomon Peachy pizza at shaftnet dot org Delray Beach, FL ^^ (email/xmpp) ^^ Quidquid latine dictum sit, altum viditur.