Re: [PATCH v2] perf unwind: Fix map reference counts

[Dan Carpenter]

On Fri, Jun 23, 2023 at 10:49:36AM -0700, Ian Rogers wrote:
> >
> > How do you think about to add the tag “Fixes”?
> 
> In general we've not been adding Fixes as there is a danger a backport
> will introduce a use-after-free.

I feel like we have been discussing issues around Perf backports
recently.  Wasn't there some build breakage that wasn't detected?  Why
not just ask Sasha to leave perf out of the -stable tree?

Also Sasha has a tag to explain that patch AAA is included because
patch BBB depends on it.  I feel like maybe those tags are backwards,
it would be nicer to tag AAA as depending on BBB.  That way we could
add the dependency tags here.

I think at Linaro we have recently been testing taking the latest Perf
tools and using them on older kernels.  I don't know the details around
why we can't just use the perf that ships with the kernel...

To tell the truth, I also don't really understand the problem for this
patch specifically. From what I can see, the Fixes tag would have been:

Fixes: 0dd5041c9a0e ("perf addr_location: Add init/exit/copy functions")

1) Adding a Fixes tag would have automatically prevented any backports.
2) I don't see any possible use after frees.  That probably means I have
identified the wrong Fixes tag?

I'm not going to dig further than that because I don't care.  I'm just
looking at it because Markus added kernel-janitors to the CC list.  But
for subsystems where I'm more involved then I always look at how a bug
is introduced.  That information is essential to me as a reviewer.  So
if I'm writing a patch and even if it's not a bug fix but let's say it
deletes dead code then I often include include the information under the
--- cut off line.

---
This dead code was introduced by commit 23423423 ("blah blah blah").

regards,
dan carpenter