Backporting Selftests to Stable Kernels?

Backporting Selftests to Stable Kernels?

[Brett Sheffield]

Dear Stable Maintainers,

When a bugfix is backported to stable kernels, should we be backporting the
associated selftest(s)?

eg.

9e30ecf23b1b ("net: ipv4: fix incorrect MTU in broadcast routes")
was backported to stable, but the selftest was not:
5777d1871bf6 ("selftests: net: add test for variable PMTU in broadcast routes")

Does stable policy say whether it should be?

It does not fix a bug, per se, but it does enable those of us running stable
kernel tests to more thoroughly test stable RCs.

Should mainline authors be encouraged to mark related tests for backporting?

Cheers,


Brett

Re: Backporting Selftests to Stable Kernels?

[Greg KH]

On Sat, Aug 23, 2025 at 03:05:46PM +0200, Brett Sheffield wrote:
> Dear Stable Maintainers,
> 
> When a bugfix is backported to stable kernels, should we be backporting the
> associated selftest(s)?

If you want to, sure!

> eg.
> 
> 9e30ecf23b1b ("net: ipv4: fix incorrect MTU in broadcast routes")
> was backported to stable, but the selftest was not:
> 5777d1871bf6 ("selftests: net: add test for variable PMTU in broadcast routes")
> 
> Does stable policy say whether it should be?

It's up to the subsystem/maintainer/developer/whomever if they wish to
do that or not.  Some subsystems want to do this, others don't, others
don't care.

> It does not fix a bug, per se, but it does enable those of us running stable
> kernel tests to more thoroughly test stable RCs.

Note that you should always be running the latest selftests for older
kernels, and I think that's what many of the CI systems are already
doing, so maybe that's why you don't notice stuff like this?

That's the only "rule" we have, all new selftests need to work properly
for older kernels.

> Should mainline authors be encouraged to mark related tests for backporting?

Again, it's up to them if they want to or not.  I will point out mptcp
as one subsystem that does mark selftests to backport, and provides
backports for when they do not apply correctly.

hope this helps,

greg k-h

Re: Backporting Selftests to Stable Kernels?

[Brett A C Sheffield]

On 2025-08-23 15:51, Greg KH wrote:
> On Sat, Aug 23, 2025 at 03:05:46PM +0200, Brett Sheffield wrote:
> > Dear Stable Maintainers,
> > 
> > When a bugfix is backported to stable kernels, should we be backporting the
> > associated selftest(s)?
> 
> If you want to, sure!
> 
> > eg.
> > 
> > 9e30ecf23b1b ("net: ipv4: fix incorrect MTU in broadcast routes")
> > was backported to stable, but the selftest was not:
> > 5777d1871bf6 ("selftests: net: add test for variable PMTU in broadcast routes")
> > 
> > Does stable policy say whether it should be?
> 
> It's up to the subsystem/maintainer/developer/whomever if they wish to
> do that or not.  Some subsystems want to do this, others don't, others
> don't care.
> 
> > It does not fix a bug, per se, but it does enable those of us running stable
> > kernel tests to more thoroughly test stable RCs.
> 
> Note that you should always be running the latest selftests for older
> kernels, and I think that's what many of the CI systems are already
> doing, so maybe that's why you don't notice stuff like this?
> 
> That's the only "rule" we have, all new selftests need to work properly
> for older kernels.
> 
> > Should mainline authors be encouraged to mark related tests for backporting?
> 
> Again, it's up to them if they want to or not.  I will point out mptcp
> as one subsystem that does mark selftests to backport, and provides
> backports for when they do not apply correctly.
> 
> hope this helps,

Thanks Greg. It does.


Brett