github-mirror/linux
1
0
Fork 0
mirror of https://github.com/torvalds/linux.git synced 2026-05-12 16:18:45 +02:00
Code Issues Actions Packages Projects Releases Wiki Activity

docs: handling-regressions: add, trim, and sort quotes from Linus

Browse Source 
Add additional quotes from Linus while trimming the existing ones and
sorting them all into categories. That makes it easier for new
developers and maintainers to look up how Linus expects certain
situations wrt regressions to be handled. The earlier sections in the
document already explain this, but those parts are often questioned --
or not considered authoritative at all and plainly ignored. Having it
straight from the horse's mouth helps get everyone on the same page,
even if that makes the document quite a bit longer (the raw line count
of this section doubles, but the number of characters increases by
nearly 50%). In return, this covers a lot more aspects and, due to the
sub-headings, is easier to navigate.

In contrast to the more neutral description in the early sections of the
document, this also provides a better insight into how serious Linus is
about the "no regressions" rule and how he wants it to be interpreted in
practice; this makes it easier for new developers and maintainers to
understand things and prevent run-ins with higher-level maintainers.

Signed-off-by: Thorsten Leemhuis <linux@leemhuis.info>
Signed-off-by: Jonathan Corbet <corbet@lwn.net>
Message-ID: <c825d7981e1badb22d15f3f6fc9c95001a017f09.1771833924.git.linux@leemhuis.info>
This commit is contained in:
Thorsten Leemhuis 2026-02-23 09:20:47 +01:00 committed by Jonathan Corbet
parent 2b8060418d
commit d20e945e9f
1 changed files with 463 additions and 232 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

695
Documentation/process/handling-regressions.rst
View File

@ -461,325 +461,556 @@ which both cover more details than the above section.
Quotes from Linus about regression
----------------------------------
Find below a few real life examples of how Linus Torvalds expects regressions to
be handled:
The following statements from Linus Torvalds provide some insight into Linux
"no regressions" rule and how he expects regressions to be handled:
* From `2017-10-26 (1/2)
<https://lore.kernel.org/lkml/CA+55aFwiiQYJ+YoLKCXjN_beDVfu38mg=Ggg5LFOcqHE8Qi7Zw@mail.gmail.com/>`_::
On how quickly regressions should be fixed
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
If you break existing user space setups THAT IS A REGRESSION.
* From `2026-01-22 <https://lore.kernel.org/all/CAHk-=wheQNiW_WtHGO7bKkT7Uib-p+ai2JP9M+z+FYcZ6CAxYA@mail.gmail.com/>`_::
It's not ok to say "but we'll fix the user space setup".
But a user complaining should basically result in an immediate fix -
possibly a "revert and rethink".
Really. NOT OK.
With a later clarification on `2026-01-28 <https://lore.kernel.org/all/CAHk-%3Dwi86AosXs66-yi54%2BmpQjPu0upxB8ZAfG%2BLsMyJmcuMSA@mail.gmail.com/>`_::
[...]
It's also worth noting that "immediate" obviously doesn't mean "right
this *second* when the problem has been reported".
The first rule is:
But if it's a regression with a known commit that caused it, I think
the rule of thumb should generally be "within a week", preferably
before the next rc.
- we don't cause regressions
* From `2023-04-21 <https://lore.kernel.org/all/CAHk-=wgD98pmSK3ZyHk_d9kZ2bhgN6DuNZMAJaV0WTtbkf=RDw@mail.gmail.com/>`_::
and the corollary is that when regressions *do* occur, we admit to
them and fix them, instead of blaming user space.
Known-broken commits either
(a) get a timely fix that doesn't have other questions
or
(b) get reverted
The fact that you have apparently been denying the regression now for
three weeks means that I will revert, and I will stop pulling apparmor
requests until the people involved understand how kernel development
is done.
* From `2021-09-20(2) <https://lore.kernel.org/all/CAHk-=wgOvmtRw1TNbMC1rn5YqyTKyn0hz+sc4k0DGNn++u9aYw@mail.gmail.com/>`_::
* From `2017-10-26 (2/2)
<https://lore.kernel.org/lkml/CA+55aFxW7NMAMvYhkvz1UPbUTUJewRt6Yb51QAx5RtrWOwjebg@mail.gmail.com/>`_::
[...] review shouldn't hold up reported regressions of existing code. That's
just basic _testing_ - either the fix should be applied, or - if the fix is
too invasive or too ugly - the problematic source of the regression should
be reverted.
People should basically always feel like they can update their kernel
and simply not have to worry about it.
Review should be about new code, it shouldn't be holding up "there's a
bug report, here's the obvious fix".
I refuse to introduce "you can only update the kernel if you also
update that other program" kind of limitations. If the kernel used to
work for you, the rule is that it continues to work for you.
* From `2023-05-08 <https://lore.kernel.org/all/CAHk-=wgzU8_dGn0Yg+DyX7ammTkDUCyEJ4C=NvnHRhxKWC7Wpw@mail.gmail.com/>`_::
There have been exceptions, but they are few and far between, and they
generally have some major and fundamental reasons for having happened,
that were basically entirely unavoidable, and people _tried_hard_ to
avoid them. Maybe we can't practically support the hardware any more
after it is decades old and nobody uses it with modern kernels any
more. Maybe there's a serious security issue with how we did things,
and people actually depended on that fundamentally broken model. Maybe
there was some fundamental other breakage that just _had_ to have a
flag day for very core and fundamental reasons.
If something doesn't even build, it should damn well be fixed ASAP.
And notice that this is very much about *breaking* peoples environments.
On how fixing regressions with reverts can help prevent maintainer burnout
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Behavioral changes happen, and maybe we don't even support some
feature any more. There's a number of fields in /proc/<pid>/stat that
are printed out as zeroes, simply because they don't even *exist* in
the kernel any more, or because showing them was a mistake (typically
an information leak). But the numbers got replaced by zeroes, so that
the code that used to parse the fields still works. The user might not
see everything they used to see, and so behavior is clearly different,
but things still _work_, even if they might no longer show sensitive
(or no longer relevant) information.
* From `2026-01-28 <https://lore.kernel.org/all/CAHk-%3Dwi86AosXs66-yi54%2BmpQjPu0upxB8ZAfG%2BLsMyJmcuMSA@mail.gmail.com/>`_::
But if something actually breaks, then the change must get fixed or
reverted. And it gets fixed in the *kernel*. Not by saying "well, fix
your user space then". It was a kernel change that exposed the
problem, it needs to be the kernel that corrects for it, because we
have a "upgrade in place" model. We don't have a "upgrade with new
user space".
> So how can I/we make "immediate fixes" happen more often without
> contributing to maintainer burnout?
And I seriously will refuse to take code from people who do not
understand and honor this very simple rule.
[...] the "revert and rethink" model [...] often a good idea in general [...]
This rule is also not going to change.
Exactly so that maintainers don't get stressed out over having a pending
problem report that people keep pestering them about.
And yes, I realize that the kernel is "special" in this respect. I'm
proud of it.
I think people are sometimes a bit too bought into whatever changes
they made, and reverting is seen as "too drastic", but I think it's
often the quick and easy solution for when there isn't some obvious
response to a regression report.
I have seen, and can point to, lots of projects that go "We need to
break that use case in order to make progress" or "you relied on
undocumented behavior, it sucks to be you" or "there's a better way to
do what you want to do, and you have to change to that new better
way", and I simply don't think that's acceptable outside of very early
alpha releases that have experimental users that know what they signed
up for. The kernel hasn't been in that situation for the last two
decades.
On mainlining fixes when the last -rc or a new release is close
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
We do API breakage _inside_ the kernel all the time. We will fix
internal problems by saying "you now need to do XYZ", but then it's
about internal kernel API's, and the people who do that then also
obviously have to fix up all the in-kernel users of that API. Nobody
can say "I now broke the API you used, and now _you_ need to fix it
up". Whoever broke something gets to fix it too.
* From `2026-02-01 <https://lore.kernel.org/all/CAHk-%3DwhXTw1oPsa%2BTLuY1Rc9D1OAiPVOdR_-R2xG45kwDObKdA@mail.gmail.com/>`_::
And we simply do not break user space.
So I think I'd rather see them hit rc8 (later today) and have a week
of testing in my tree and be reverted if they cause problems, than
have them go in after rc8 and then cause problems in the 6.19 release
instead.
* From `2020-05-21
<https://lore.kernel.org/all/CAHk-=wiVi7mSrsMP=fLXQrXK_UimybW=ziLOwSzFTtoXUacWVQ@mail.gmail.com/>`_::
* From `2023-04-20 <https://lore.kernel.org/all/CAHk-=wis_qQy4oDNynNKi5b7Qhosmxtoj1jxo5wmB6SRUwQUBQ@mail.gmail.com/>`_::
The rules about regressions have never been about any kind of
documented behavior, or where the code lives.
But something like this, where the regression was in the previous release
and it's just a clear fix with no semantic subtlety, I consider to be just a
regular regression that should be expedited - partly to make it into stable,
and partly to avoid having to put the fix into _another_ stable kernel.
The rules about regressions are always about "breaks user workflow".
On sending merge requests with just one fix
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Users are literally the _only_ thing that matters.
* From `2024-04-24 <https://lore.kernel.org/all/CAHk-=wjy_ph9URuFt-pq+2AJ__p7gFDx=yzVSCsx16xAYvNw9g@mail.gmail.com/>`_::
No amount of "you shouldn't have used this" or "that behavior was
undefined, it's your own fault your app broke" or "that used to work
simply because of a kernel bug" is at all relevant.
If the issue is just that there's nothing else happening, I think people
should just point me to the patch and say "can you apply this single fix?"
Now, reality is never entirely black-and-white. So we've had things
like "serious security issue" etc that just forces us to make changes
that may break user space. But even then the rule is that we don't
really have other options that would allow things to continue.
* From `2023-04-20 <https://lore.kernel.org/all/CAHk-=wis_qQy4oDNynNKi5b7Qhosmxtoj1jxo5wmB6SRUwQUBQ@mail.gmail.com/>`_::
And obviously, if users take years to even notice that something
broke, or if we have sane ways to work around the breakage that
doesn't make for too much trouble for users (ie "ok, there are a
handful of users, and they can use a kernel command line to work
around it" kind of things) we've also been a bit less strict.
I'm always open to direct fixes when there is no controversy about the fix.
No problem. I still happily deal with individual patches.
But no, "that was documented to be broken" (whether it's because the
code was in staging or because the man-page said something else) is
irrelevant. If staging code is so useful that people end up using it,
that means that it's basically regular kernel code with a flag saying
"please clean this up".
On the importance of pointing to bug reports using Link:/Closes: tags
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The other side of the coin is that people who talk about "API
stability" are entirely wrong. API's don't matter either. You can make
any changes to an API you like - as long as nobody notices.
* From `2025-07-29(1) <https://lore.kernel.org/all/CAHk-=wj2kJRPWx8B09AAtzj+_g+T6UBX11TP0ebs1WJdTtv=WQ@mail.gmail.com/>`_::
Again, the regression rule is not about documentation, not about
API's, and not about the phase of the moon.
[...] revert like this, it really would be good to link to the problems, so
that when people try to re-enable it, they have the history for why it
didn't work the first time.
It's entirely about "we caused problems for user space that used to work".
* From `2022-05-08 <https://lore.kernel.org/all/CAHk-=wjMmSZzMJ3Xnskdg4+GGz=5p5p+GSYyFBTh0f-DgvdBWg@mail.gmail.com/>`_::
* From `2017-11-05
<https://lore.kernel.org/all/CA+55aFzUvbGjD8nQ-+3oiMBx14c_6zOj2n7KLN3UsJ-qsd4Dcw@mail.gmail.com/>`_::
So I have to once more complain [...]
And our regression rule has never been "behavior doesn't change".
That would mean that we could never make any changes at all.
[...] There's no link to the actual problem the patch fixes.
For example, we do things like add new error handling etc all the
time, which we then sometimes even add tests for in our kselftest
directory.
* From `2022-06-22 <https://lore.kernel.org/all/CAHk-=wjxzafG-=J8oT30s7upn4RhBs6TX-uVFZ5rME+L5_DoJA@mail.gmail.com/>`_::
So clearly behavior changes all the time and we don't consider that a
regression per se.
See, *that* link [to the report] would have been useful in the commit.
The rule for a regression for the kernel is that some real user
workflow breaks. Not some test. Not a "look, I used to be able to do
X, now I can't".
On why the "no regressions" rule exists
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
* From `2018-08-03
<https://lore.kernel.org/all/CA+55aFwWZX=CXmWDTkDGb36kf12XmTehmQjbiMPCqCRG2hi9kw@mail.gmail.com/>`_::
* From `2026-01-22 <https://lore.kernel.org/all/CAHk-=wheQNiW_WtHGO7bKkT7Uib-p+ai2JP9M+z+FYcZ6CAxYA@mail.gmail.com/>`_::
YOU ARE MISSING THE #1 KERNEL RULE.
But the basic rule is: be so good about backwards compatibility that
users never have to worry about upgrading. They should absolutely feel
confident that any kernel-reported problem will either be solved, or
have an easy solution that is appropriate for *them* (ie a
non-technical user shouldn't be expected to be able to do a lot).
We do not regress, and we do not regress exactly because your are 100% wrong.
Because the last thing we want is people holding back from trying new
kernels.
And the reason you state for your opinion is in fact exactly *WHY* you
are wrong.
* From `2024-05-28 <https://lore.kernel.org/all/CAHk-=wgtb7y-bEh7tPDvDWru7ZKQ8-KMjZ53Tsk37zsPPdwXbA@mail.gmail.com/>`_::
Your "good reasons" are pure and utter garbage.
I introduced that "no regressions" rule something like two decades
ago, because people need to be able to update their kernel without
fear of something they relied on suddenly stopping to work.
The whole point of "we do not regress" is so that people can upgrade
the kernel and never have to worry about it.
* From `2018-08-03 <https://lore.kernel.org/all/CA+55aFwWZX=CXmWDTkDGb36kf12XmTehmQjbiMPCqCRG2hi9kw@mail.gmail.com/>`_::
> Kernel had a bug which has been fixed
The whole point of "we do not regress" is so that people can upgrade
the kernel and never have to worry about it.
That is *ENTIRELY* immaterial.
[...]
Guys, whether something was buggy or not DOES NOT MATTER.
Because the only thing that matters IS THE USER.
Why?
* From `2017-10-26(1) <https://lore.kernel.org/lkml/CA+55aFxW7NMAMvYhkvz1UPbUTUJewRt6Yb51QAx5RtrWOwjebg@mail.gmail.com/>`_::
Bugs happen. That's a fact of life. Arguing that "we had to break
something because we were fixing a bug" is completely insane. We fix
tens of bugs every single day, thinking that "fixing a bug" means that
we can break something is simply NOT TRUE.
If the kernel used to work for you, the rule is that it continues to work
for you.
So bugs simply aren't even relevant to the discussion. They happen,
they get found, they get fixed, and it has nothing to do with "we
break users".
[...]
Because the only thing that matters IS THE USER.
People should basically always feel like they can update their kernel
and simply not have to worry about it.
How hard is that to understand?
I refuse to introduce "you can only update the kernel if you also
update that other program" kind of limitations. If the kernel used to
work for you, the rule is that it continues to work for you.
Anybody who uses "but it was buggy" as an argument is entirely missing
the point. As far as the USER was concerned, it wasn't buggy - it
worked for him/her.
On exceptions to the "no regressions" rule
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Maybe it worked *because* the user had taken the bug into account,
maybe it worked because the user didn't notice - again, it doesn't
matter. It worked for the user.
* From `2026-01-22 <https://lore.kernel.org/all/CAHk-=wheQNiW_WtHGO7bKkT7Uib-p+ai2JP9M+z+FYcZ6CAxYA@mail.gmail.com/>`_::
Breaking a user workflow for a "bug" is absolutely the WORST reason
for breakage you can imagine.
There are _very_ few exceptions to that rule, the main one being "the
problem was a fundamental huge and gaping security issue and we *had* to
make that change, and we couldn't even make your limited use-case just
continue to work".
It's basically saying "I took something that worked, and I broke it,
but now it's better". Do you not see how f*cking insane that statement
is?
The other exception is "the problem was reported years after it was
introduced, and now most people rely on the new behavior".
And without users, your program is not a program, it's a pointless
piece of code that you might as well throw away.
[...]
Seriously. This is *why* the #1 rule for kernel development is "we
don't break users". Because "I fixed a bug" is absolutely NOT AN
ARGUMENT if that bug fix broke a user setup. You actually introduced a
MUCH BIGGER bug by "fixing" something that the user clearly didn't
even care about.
Now, if it's one or two users and you can just get them to recompile,
that's one thing. Niche hardware and odd use-cases can sometimes be
solved that way, and regressions can sometimes be fixed by handholding
every single reporter if the reporter is willing and able to change
his or her workflow.
And dammit, we upgrade the kernel ALL THE TIME without upgrading any
other programs at all. It is absolutely required, because flag-days
and dependencies are horribly bad.
* From `2023-04-20 <https://lore.kernel.org/all/CAHk-=wis_qQy4oDNynNKi5b7Qhosmxtoj1jxo5wmB6SRUwQUBQ@mail.gmail.com/>`_::
And it is also required simply because I as a kernel developer do not
upgrade random other tools that I don't even care about as I develop
the kernel, and I want any of my users to feel safe doing the same
time.
And yes, I do consider "regression in an earlier release" to be a
regression that needs fixing.
So no. Your rule is COMPLETELY wrong. If you cannot upgrade a kernel
without upgrading some other random binary, then we have a problem.
There's obviously a time limit: if that "regression in an earlier
release" was a year or more ago, and just took forever for people to
notice, and it had semantic changes that now mean that fixing the
regression could cause a _new_ regression, then that can cause me to
go "Oh, now the new semantics are what we have to live with".
* From `2021-06-05
<https://lore.kernel.org/all/CAHk-=wiUVqHN76YUwhkjZzwTdjMMJf_zN4+u7vEJjmEGh3recw@mail.gmail.com/>`_::
* From `2017-10-26(2) <https://lore.kernel.org/lkml/CA+55aFxW7NMAMvYhkvz1UPbUTUJewRt6Yb51QAx5RtrWOwjebg@mail.gmail.com/>`_::
THERE ARE NO VALID ARGUMENTS FOR REGRESSIONS.
There have been exceptions, but they are few and far between, and they
generally have some major and fundamental reasons for having happened,
that were basically entirely unavoidable, and people _tried_hard_ to
avoid them. Maybe we can't practically support the hardware any more
after it is decades old and nobody uses it with modern kernels any
more. Maybe there's a serious security issue with how we did things,
and people actually depended on that fundamentally broken model. Maybe
there was some fundamental other breakage that just _had_ to have a
flag day for very core and fundamental reasons.
Honestly, security people need to understand that "not working" is not
a success case of security. It's a failure case.
On situations where updating something in userspace can resolve regressions
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Yes, "not working" may be secure. But security in that case is *pointless*.
* From `2018-08-03 <https://lore.kernel.org/all/CA+55aFwWZX=CXmWDTkDGb36kf12XmTehmQjbiMPCqCRG2hi9kw@mail.gmail.com/>`_::
* From `2011-05-06 (1/3)
<https://lore.kernel.org/all/BANLkTim9YvResB+PwRp7QTK-a5VNg2PvmQ@mail.gmail.com/>`_::
And dammit, we upgrade the kernel ALL THE TIME without upgrading any
other programs at all. It is absolutely required, because flag-days
and dependencies are horribly bad.
Binary compatibility is more important.
And it is also required simply because I as a kernel developer do not
upgrade random other tools that I don't even care about as I develop the
kernel, and I want any of my users to feel safe doing the same time.
And if binaries don't use the interface to parse the format (or just
parse it wrongly - see the fairly recent example of adding uuid's to
/proc/self/mountinfo), then it's a regression.
* From `2017-10-26(3) <https://lore.kernel.org/lkml/CA+55aFxW7NMAMvYhkvz1UPbUTUJewRt6Yb51QAx5RtrWOwjebg@mail.gmail.com/>`_::
And regressions get reverted, unless there are security issues or
similar that makes us go "Oh Gods, we really have to break things".
But if something actually breaks, then the change must get fixed or
reverted. And it gets fixed in the *kernel*. Not by saying "well, fix your
user space then". It was a kernel change that exposed the problem, it needs
to be the kernel that corrects for it, because we have a "upgrade in place"
model. We don't have a "upgrade with new user space".
I don't understand why this simple logic is so hard for some kernel
developers to understand. Reality matters. Your personal wishes matter
NOT AT ALL.
And I seriously will refuse to take code from people who do not understand
and honor this very simple rule.
If you made an interface that can be used without parsing the
interface description, then we're stuck with the interface. Theory
simply doesn't matter.
This rule is also not going to change.
You could help fix the tools, and try to avoid the compatibility
issues that way. There aren't that many of them.
And yes, I realize that the kernel is "special" in this respect. I'm proud
of it.
From `2011-05-06 (2/3)
<https://lore.kernel.org/all/BANLkTi=KVXjKR82sqsz4gwjr+E0vtqCmvA@mail.gmail.com/>`_::
* From `2017-10-26(4) <https://lore.kernel.org/all/CA+55aFwiiQYJ+YoLKCXjN_beDVfu38mg=Ggg5LFOcqHE8Qi7Zw@mail.gmail.com/>`_::
it's clearly NOT an internal tracepoint. By definition. It's being
used by powertop.
If you break existing user space setups THAT IS A REGRESSION.
From `2011-05-06 (3/3)
<https://lore.kernel.org/all/BANLkTinazaXRdGovYL7rRVp+j6HbJ7pzhg@mail.gmail.com/>`_::
It's not ok to say "but we'll fix the user space setup".
We have programs that use that ABI and thus it's a regression if they break.
Really. NOT OK.
* From `2012-07-06 <https://lore.kernel.org/all/CA+55aFwnLJ+0sjx92EGREGTWOx84wwKaraSzpTNJwPVV8edw8g@mail.gmail.com/>`_::
On what qualifies as userspace interface, ABI, API, documented interfaces, etc.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> Now this got me wondering if Debian _unstable_ actually qualifies as a
> standard distro userspace.
* From `2026-01-20 <https://lore.kernel.org/all/CAHk-=wga8Qu0-OSE9VZbviq9GuqwhPhLUXeAt-S7_9+fMCLkKg@mail.gmail.com/>`_::
Oh, if the kernel breaks some standard user space, that counts. Tons
of people run Debian unstable
So I absolutely detest the whole notion of "ABI changes". It's a
meaningless concept, and I hate it with a passion, [...]
* From `2019-09-15
<https://lore.kernel.org/lkml/CAHk-=wiP4K8DRJWsCo=20hn_6054xBamGKF2kPgUzpB5aMaofA@mail.gmail.com/>`_::
The Linux rule for regressions is basically based on the philosophical
question of "If a tree falls in the forest, and nobody is around to
hear it, does it make a sound?".
One _particularly_ last-minute revert is the top-most commit (ignoring
the version change itself) done just before the release, and while
it's very annoying, it's perhaps also instructive.
So the only thing that matters is if something breaks user-*conscious*
behavior.
What's instructive about it is that I reverted a commit that wasn't
actually buggy. In fact, it was doing exactly what it set out to do,
and did it very well. In fact it did it _so_ well that the much
improved IO patterns it caused then ended up revealing a user-visible
regression due to a real bug in a completely unrelated area.
And when that happens, the distinction between "bug fix" and "new
feature" and "ABI change" matters not one whit, and the change needs
to be done differently.
The actual details of that regression are not the reason I point that
revert out as instructive, though. It's more that it's an instructive
example of what counts as a regression, and what the whole "no
regressions" kernel rule means. The reverted commit didn't change any
API's, and it didn't introduce any new bugs. But it ended up exposing
another problem, and as such caused a kernel upgrade to fail for a
user. So it got reverted.
[...]
The point here being that we revert based on user-reported _behavior_,
not based on some "it changes the ABI" or "it caused a bug" concept.
The problem was really pre-existing, and it just didn't happen to
trigger before. The better IO patterns introduced by the change just
happened to expose an old bug, and people had grown to depend on the
previously benign behavior of that old issue.
I just wanted to point out that the argument about whether it's an ABI
change or not is irrelevant. If it turns out that some program - not a test
script, but something with relevance to conscious user expectations ~
depended on the old broken behavior, then it needs to be done some other
way.
And never fear, we'll re-introduce the fix that improved on the IO
patterns once we've decided just how to handle the fact that we had a
bad interaction with an interface that people had then just happened
to rely on incidental behavior for before. It's just that we'll have
to hash through how to do that (there are no less than three different
patches by three different developers being discussed, and there might
be more coming...). In the meantime, I reverted the thing that exposed
the problem to users for this release, even if I hope it will be
re-introduced (perhaps even backported as a stable patch) once we have
consensus about the issue it exposed.
* From `2026-02-13 <https://lore.kernel.org/all/CAHk-=whY-N8kjm8kiFUV5Ei-8AuYw--EPGD-AR3Pd+5GTx2sAQ@mail.gmail.com/>`_::
Take-away from the whole thing: it's not about whether you change the
kernel-userspace ABI, or fix a bug, or about whether the old code
"should never have worked in the first place". It's about whether
something breaks existing users' workflow.
> [...] this should not fall under the don't break user space rule [...]
Anyway, that was my little aside on the whole regression thing. Since
it's that "first rule of kernel programming", I felt it is perhaps
worth just bringing it up every once in a while
Note that the rule is about breaking *users*, not breaking user space per
se. [...]
If some user setup breaks, things need fixing.
[...] but I want to make it very clear that there are no excuses about "user
space applications".
* From `2021-09-20(4) <https://lore.kernel.org/all/CAHk-=wi7DB2SJ-wngVvsJ7Ak2cM556Q8437sOXo4EJt2BWPdEg@mail.gmail.com/>`_::
[...] a regression is a bit like Schrödinger's cat - if nobody is around
to notice it and it doesn't actually affect any real workload, then you
can treat the regression as if it doesn't exist.
* From `2020-05-21 <https://lore.kernel.org/all/CAHk-=wiVi7mSrsMP=fLXQrXK_UimybW=ziLOwSzFTtoXUacWVQ@mail.gmail.com/>`_::
The rules about regressions have never been about any kind of documented
behavior, or where the code lives.
The rules about regressions are always about "breaks user workflow".
Users are literally the _only_ thing that matters.
* From `2019-09-15 <https://lore.kernel.org/lkml/CAHk-=wiP4K8DRJWsCo=20hn_6054xBamGKF2kPgUzpB5aMaofA@mail.gmail.com/>`_::
One _particularly_ last-minute revert is the top-most commit (ignoring
the version change itself) done just before the release, and while
it's very annoying, it's perhaps also instructive.
What's instructive about it is that I reverted a commit that wasn't
actually buggy. In fact, it was doing exactly what it set out to do,
and did it very well. In fact it did it _so_ well that the much
improved IO patterns it caused then ended up revealing a user-visible
regression due to a real bug in a completely unrelated area.
The actual details of that regression are not the reason I point that
revert out as instructive, though. It's more that it's an instructive
example of what counts as a regression, and what the whole "no
regressions" kernel rule means.
[...] The reverted commit didn't change any API's, and it didn't introduce
any new bugs. But it ended up exposing another problem, and as such caused
a kernel upgrade to fail for a user. So it got reverted.
The point here being that we revert based on user-reported _behavior_, not
based on some "it changes the ABI" or "it caused a bug" concept. The problem
was really pre-existing, and it just didn't happen to trigger before. [...]
Take-away from the whole thing: it's not about whether you change the
kernel-userspace ABI, or fix a bug, or about whether the old code
"should never have worked in the first place". It's about whether
something breaks existing users' workflow.
* From `2017-11-05 <https://lore.kernel.org/all/CA+55aFzUvbGjD8nQ-+3oiMBx14c_6zOj2n7KLN3UsJ-qsd4Dcw@mail.gmail.com/>`_::
And our regression rule has never been "behavior doesn't change".
That would mean that we could never make any changes at all.
* From `2020-05-21 <https://lore.kernel.org/all/CAHk-=wiVi7mSrsMP=fLXQrXK_UimybW=ziLOwSzFTtoXUacWVQ@mail.gmail.com/>`_::
No amount of "you shouldn't have used this" or "that behavior was
undefined, it's your own fault your app broke" or "that used to work
simply because of a kernel bug" is at all relevant.
* From `2021-05-21 <https://lore.kernel.org/all/CAHk-=wiVi7mSrsMP=fLXQrXK_UimybW=ziLOwSzFTtoXUacWVQ@mail.gmail.com/>`_::
But no, "that was documented to be broken" (whether it's because the code
was in staging or because the man-page said something else) is irrelevant.
If staging code is so useful that people end up using it, that means that
it's basically regular kernel code with a flag saying "please clean this
up".
[...]
The other side of the coin is that people who talk about "API stability" are
entirely wrong. API's don't matter either. You can make any changes to an
API you like - as long as nobody notices.
Again, the regression rule is not about documentation, not about API's, and
not about the phase of the moon.
* From `2012-07-06 <https://lore.kernel.org/all/CA+55aFwnLJ+0sjx92EGREGTWOx84wwKaraSzpTNJwPVV8edw8g@mail.gmail.com/>`_::
> Now this got me wondering if Debian _unstable_ actually qualifies as a
> standard distro userspace.
Oh, if the kernel breaks some standard user space, that counts. Tons
of people run Debian unstable
* From `2011-05-06 <https://lore.kernel.org/all/BANLkTi=KVXjKR82sqsz4gwjr+E0vtqCmvA@mail.gmail.com/>`_::
It's clearly NOT an internal tracepoint. By definition. It's being
used by powertop.
On regressions noticed by users or test-suites/CIs
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
* From `2026-01-22 <https://lore.kernel.org/all/CAHk-=wheQNiW_WtHGO7bKkT7Uib-p+ai2JP9M+z+FYcZ6CAxYA@mail.gmail.com/>`_::
Users complaining is the only real line in the end.
[...] a test-suite complaining is then often a *very* good indication that
maybe users will hit some problem, and test suite issues should be taken
very seriously [...]
But a test-suite error isn't necessarily where you have to draw the
line - it's a big red flag [...]
* From `2024-29-01 <https://lore.kernel.org/all/CAHk-=wg8BrZEzjJ5kUyZzHPZmFqH6ooMN1gRBCofxxCfucgjaw@mail.gmail.com/>`_::
The "no regressions" rule is not about made-up "if I do this, behavior
changes".
The "no regressions" rule is about *users*.
If you have an actual user that has been doing insane things, and we
change something, and now the insane thing no longer works, at that
point it's a regression, and we'll sigh, and go "Users are insane" and
have to fix it.
But if you have some random test that now behaves differently, it's
not a regression. It's a *warning* sign, sure: tests are useful.
On accepting when a regression occurred
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
* From `2026-01-22 <https://lore.kernel.org/all/CAHk-=wheQNiW_WtHGO7bKkT7Uib-p+ai2JP9M+z+FYcZ6CAxYA@mail.gmail.com/>`_::
But starting to argue about users reporting breaking changes is
basically the final line for me. I have a couple of people that I have
in my spam block-list and refuse to have anything to do with, and they
have generally been about exactly that.
Note how it's not about making mistakes and _causing_ the regression.
That's normal. That's development. But then arguing about it is a
no-no.
* From `2024-06-23 <https://lore.kernel.org/all/CAHk-=wi_KMO_rJ6OCr8mAWBRg-irziM=T9wxGC+J1VVoQb39gw@mail.gmail.com/>`_::
We don't introduce regressions and then blame others.
There's a very clear rule in kernel development: things that break
other things ARE NOT FIXES.
EVER.
They get reverted, or the thing they broke gets fixed.
* From `2021-06-05 <https://lore.kernel.org/all/CAHk-=wiUVqHN76YUwhkjZzwTdjMMJf_zN4+u7vEJjmEGh3recw@mail.gmail.com/>`_::
THERE ARE NO VALID ARGUMENTS FOR REGRESSIONS.
Honestly, security people need to understand that "not working" is not
a success case of security. It's a failure case.
Yes, "not working" may be secure. But security in that case is *pointless*.
* From `2017-10-26(5) <https://lore.kernel.org/lkml/CA+55aFwiiQYJ+YoLKCXjN_beDVfu38mg=Ggg5LFOcqHE8Qi7Zw@mail.gmail.com/>`_::
[...] when regressions *do* occur, we admit to them and fix them, instead of
blaming user space.
The fact that you have apparently been denying the regression now for
three weeks means that I will revert, and I will stop pulling apparmor
requests until the people involved understand how kernel development
is done.
On back-and-forth
~~~~~~~~~~~~~~~~~
* From `2024-05-28 <https://lore.kernel.org/all/CAHk-=wgtb7y-bEh7tPDvDWru7ZKQ8-KMjZ53Tsk37zsPPdwXbA@mail.gmail.com/>`_::
The "no regressions" rule is that we do not introduce NEW bugs.
It *literally* came about because we had an endless dance of "fix two
bugs, introduce one new one", and that then resulted in a system that
you cannot TRUST.
* From `2021-09-20(1) <https://lore.kernel.org/all/CAHk-=wi7DB2SJ-wngVvsJ7Ak2cM556Q8437sOXo4EJt2BWPdEg@mail.gmail.com/>`_::
And the thing that makes regressions special is that back when I
wasn't so strict about these things, we'd end up in endless "seesaw
situations" where somebody would fix something, it would break
something else, then that something else would break, and it would
never actually converge on anything reliable at all.
* From `2015-08-13 <https://lore.kernel.org/all/CA+55aFxk8-BsiKwr_S-c+4G6wihKPQVMLE34H9wOZpeua6W9+Q@mail.gmail.com/>`_::
The strict policy of no regressions actually originally started mainly wrt
suspend/resume issues, where the "fix one machine, break another" kind of
back-and-forth caused endless problems, and meant that we didn't actually
necessarily make any forward progress, just moving a problem around.
On changes with a risk of causing regressions
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
* From `2023-06-02 <https://lore.kernel.org/all/CAHk-=wgyAGUMHmQM-5Eb556z5xiHZB7cF05qjrtUH4F7P-1rSA@mail.gmail.com/>`_::
So what I think you should do is to fix the bug right, with a clean
patch, and no crazy hacks. That is something we can then apply and
test. All the while knowing full well that "uhhuh, this is a visible
change, we may have to revert it".
If then some *real* load ends up showing a regression, we may just be
screwed. Our current behavior may be buggy, but we have the rule that
once user space depends on kernel bugs, they become features pretty
much by definition, however much we might dislike it.
On in-kernel workarounds to avoid regressions
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
* From `2017-10-26(6) <https://lore.kernel.org/lkml/CA+55aFxW7NMAMvYhkvz1UPbUTUJewRt6Yb51QAx5RtrWOwjebg@mail.gmail.com/>`_::
Behavioral changes happen, and maybe we don't even support some
feature any more. There's a number of fields in /proc/<pid>/stat that
are printed out as zeroes, simply because they don't even *exist* in
the kernel any more, or because showing them was a mistake (typically
an information leak). But the numbers got replaced by zeroes, so that
the code that used to parse the fields still works. The user might not
see everything they used to see, and so behavior is clearly different,
but things still _work_, even if they might no longer show sensitive
(or no longer relevant) information.
On regressions caused by bugfixes
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
* From `2018-08-03 <https://lore.kernel.org/all/CA+55aFwWZX=CXmWDTkDGb36kf12XmTehmQjbiMPCqCRG2hi9kw@mail.gmail.com/>`_::
> Kernel had a bug which has been fixed
That is *ENTIRELY* immaterial.
Guys, whether something was buggy or not DOES NOT MATTER.
[...]
It's basically saying "I took something that worked, and I broke it,
but now it's better". Do you not see how f*cking insane that statement
is?
On internal API changes
~~~~~~~~~~~~~~~~~~~~~~~
* From `2017-10-26(7) <https://lore.kernel.org/lkml/CA+55aFxW7NMAMvYhkvz1UPbUTUJewRt6Yb51QAx5RtrWOwjebg@mail.gmail.com/>`_::
We do API breakage _inside_ the kernel all the time. We will fix
internal problems by saying "you now need to do XYZ", but then it's
about internal kernel API's, and the people who do that then also
obviously have to fix up all the in-kernel users of that API. Nobody
can say "I now broke the API you used, and now _you_ need to fix it
up". Whoever broke something gets to fix it too.
On regressions only found after a long time
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
* From `2024-03-28 <https://lore.kernel.org/all/CAHk-=wgFuoHpMk_Z_R3qMXVDgq0N1592+bABkyGjwwSL4zBtHA@mail.gmail.com/>`_::
I'm definitely not reverting a patch from almost a decade ago as a
regression.
If it took that long to find, it can't be that critical of a regression.
So yes, let's treat it as a regular bug.
On testing regressions fixes in linux-next
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
* On `maintainers summit 2024 <https://lwn.net/Articles/990599/>`_::
So running fixes though linux-next is just a waste of time.
On a few other aspects related to regressions
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
* From `2025-07-29(2) <https://lore.kernel.org/all/CAHk-=wjj9DvOZtmTkoLtyfHmy5mNKy6q_96d9=4FUEDXre=cww@mail.gmail.com/>`_
[which `is not quite a regression, but a huge inconvenience <https://lore.kernel.org/all/CAHk-=wgO0Rx2LcYT4f75Xs46orbJ4JxO2jbAFQnVKDYAjV5HeQ@mail.gmail.com/>`_]::
I no longer have sound.
I also suspect that it's purely because "make oldconfig" doesn't work,
and probably turned off my old Intel HDA settings. Or something.
Renaming config parameters is *bad*. I've harped on the Kconfig phase
of the kernel build probably being our nastiest point, and a real pain
point to people getting involved with development simply because
building your own kernel can be so daunting with hundreds of fairly
esoteric questions.
..
end-of-content