Re: [PATCH v8 14/21] mm/mmap: Avoid zeroing vma tree in mmap_region()

[Lorenzo Stoakes]

On Thu, Oct 03, 2024 at 03:09:08PM +0200, Bert Karwatzki wrote:
> Here's the log procduced by this kernel:
>
> 2938bbdf68a7 (HEAD -> maple_tree_debug_2) hack: set of fix, info stuff v3
> 7e3bb072761a mm: correct error handling in mmap_region()
> 77df9e4bb222 (tag: next-20241001, origin/master, origin/HEAD, master) Add linux-next specific files for 20241001
>
> "maybe fix 3" did not work, either.

Thanks again for your continued help.

It's a stubborn problem...

At this point I am strongly suspecting an internal maple tree bug given we
are probably doing a rebalance or a split to overwrite 6 (!) VMAs we unmap
as part of this operation which is something I don't think we could have
hit before in practice.

Let's try and snapshot the maple tree state prior to the 'nuclear' event
then, and see if it's possible to reconstruct it in userland or such and
thus see if we can repro it that way.

Please apply the attached patch to the clean 1st oct next tree and let us
know the dmesg output, again hugely hugely appreciated!

Thanks, Lorenzo

----8<----