Re: [PATCH 5.10 000/599] 5.10.110-rc1 review

[Steven Rostedt]

On Tue, 5 Apr 2022 22:52:12 -0400
Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:

> On Tue, 5 Apr 2022 19:30:25 -0700
> Guenter Roeck <linux@xxxxxxxxxxxx> wrote:
> 
> > > s390 tests crashed. Other failed qemu tests did not compile.    
> > 
> > Bisect points to commit 93fe2389e6fd ("tracing: Have TRACE_DEFINE_ENUM
> > affect trace event types as well"). Bisect log attached. Reverting the
> > offending patch fixes the problem. Copying Steven for comments/input.  
> 
> Do you have this commit?
> 
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=795301d3c2899
> 

OK, I see what happened. I wrote that patch because a new event appeared
that used enums in the field format that gets exposed to user space. This
patch is only needed for events that do that. Which until this release,
nothing did.

That patch was then flagged as something to backport (but it was never
tagged as stable, so this was one of the "AUTOSEL" backports). And it
should have only been backported if any of these new events were also
backported. Otherwise, the patch is useless.

When this patch was in linux-next, it caused s390 and powerpc to crash. I
found the issue and fixed it with the linked patch above, and even added a
"Fixes" tag to tie the two commits together.

Both of problem patch and the fix went into mainline in the same pull
request.

Here's a thought, if you decide to backport a patch to stable, and you see
that there's another commit with a "Fixes" tag to the automatically
selected commit. DO NOT BACKPORT IF THE FIXES PATCH FAILS TO GO BACK TOO!

This never should have been backported without the fix. Please revert that
patch for any stable release that the fixes patch fails to apply.

-- Steve