Re: kernel BUG at mm/swap_state.c:170!

[Mikhail Gavrilov <mikhail.v.gavrilov@gmail.com>]

On Mon, 17 Jun 2019 at 17:17, Vlastimil Babka <vbabka@suse.cz> wrote:
>
> That's commit "tcp: fix retrans timestamp on passive Fast Open" which is
> almost certainly not the culprit.

Yes, I seen also content of this commit.
And it looks like madness.
But I can proving that my bisect are properly created.
Here I saved all dmesg output from all bisecting steps:
https://mega.nz/#F!00wFHACA!nmaLgkkbrlt46DteERjl7Q
And only one of them ended without crash message "kernel BUG at
mm/swap_state.c:170!"
This is step5 with commit 3d21b6525cae.

I tried to cause kernel panic several times when kenel compiled from
commit 3d21b6525cae would be launched and all my attempts was be
unsuccessful.

So I can say that commit 3d21b6525cae is enough stable for me and I
now sitting on it.

> You told bisect that 5.2-rc1 is good, but it probably isn't.
> What you probably need to do is:
> git bisect good v5.1
> git bisect bad v5.2-rc2
>
> The presence of the other ext4 bug complicates the bisect, however.
> According to tytso in the thread you linked, it should be fixed by
> commit 0a944e8a6c66, while the bug was introduced by commit
> 345c0dbf3a30. So in each step of bisect, before building the kernel, you
> should cherry-pick the fix if the bug is there:
>
> git merge-base --is-ancestor 345c0dbf3a30 HEAD && git cherry-pick 0a944e8a6c66

Oh, thanks for advise. But I am used another solution.
(I applied the patch every time when bisect move to new step)

> Also in case you see a completely different problem in some bisect step, try
> 'git bisect skip' instead of guessing if it's good or bad.
> Hopefully that will lead to a better result.

If you take a look all my dmesg logs you can sure that all bad steps
ended with crash "kernel BUG at mm/swap_state.c:170!".

And yes, I look again at commit cd736d8b67fb still don't understand
how it can broke my system.

--
Best Regards,
Mike Gavrilov.