[PATCH 4/4] rhashtable: fix vmalloc() deadlock in atomic rehash

[Linus Torvalds]

On Tue, 3 Nov 2026 at 19:58, Herbert Xu <herbert at gondor.apana.org.au> wrote:
>
> To recap, kvmalloc was modified in 2025 such that it can use vmalloc
> even when called with GFP_ATOMIC (commit c6307674ed82).  Unfortunately,
> it does so in a way where it has to wait for other CPUs to do certain
> work.  As a result, if you call kvmalloc with GFP_ATOMIC under an
> external lock, this easily leads to a dead-lock when other CPUs end
> up blocking on the same external lock.

Ok, so it sounds like that change was clearly complete garbage and needs fixing.

The only reason for GFP_ATOMIC is that you're in a context where you
can't block. That's the *point*. So if GFP_ATOMIC waits for something,
it's just fundamentally buggy.

Sadly, the commit you point to doesn't talk about *why* that
braindamage was done. Looking at the original thread it seems like it
was for kmsan.

I didn't actually find the kmsan use case, but I didn't search very
hard. I absolutely *DETEST* it when "debug infrastructure" causes
actual bugs and mis-designs. That's the opposite of debugging.

IOW, I guess we do have to split those allocation cases up, and we
damn well should remove GFP_ATOMIC support from vmalloc() again since
it's clearly buggered beyond belief.

              Linus