On Wed, Jun 25, 2025 at 5:30 AM Lorenzo Stoakes
<lorenzo.stoa...@oracle.com> wrote:
>
> This patch results in some spam :) there's a stray mmap_assert_locked() in
> anon_vma_name() that triggers constantly.
Ah, damn! This was triggered before and I completely forgot to fix it.
Thanks for reporting it!
>
> Andrew - I attach a fix-patch for this, could you apply as at least a
> temporary
> fix? As mm-new is broken at the moment with this patch.
Yes please.Andrew, if you would prefer me to respin the series please
let me know.
>
> Suren - could you check and obviously suggest something more sensible if you
> feel this isn't right.
This is exactly what I was planning to do but it slipped from my mind.
>
> I'm not actually sure if we'd always have the VMA read lock here, maybe we
> need
> an 'assert mmap lock or vma lock' predicate?
For now this is the only place that needs it. Maybe we can wait until
there is more demand for it?
>
> Worth auditing other mmap lock asserts that might have been missed with this
> change also.
I'll go over it once more but this was the only reported issue with
the previous version.
Thanks,
Suren.
>
> Cheers, Lorenzo
>
> ----8<----
> From 1ed3bd12d43be1f8303fd6b7b714f5ef7e60728a Mon Sep 17 00:00:00 2001
> From: Lorenzo Stoakes <lorenzo.stoa...@oracle.com>
> Date: Wed, 25 Jun 2025 13:28:36 +0100
> Subject: [PATCH] mm/madvise: fixup stray mmap lock assert in anon_vma_name()
>
> anon_vma_name() is being called under VMA lock, but is assert mmap lock which
> won't necessarily be held.
>
> This results in the kernel spamming warnings about this on startup.
>
> Replace this with an open-coded 'mmap or VMA lock' assert to resolve.
>
> Signed-off-by: Lorenzo Stoakes <lorenzo.stoa...@oracle.com>
Acked-by: Suren Baghdasaryan <sur...@google.com>
> ---
> mm/madvise.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/mm/madvise.c b/mm/madvise.c
> index c467ee42596f..0530d033b3dd 100644
> --- a/mm/madvise.c
> +++ b/mm/madvise.c
> @@ -108,7 +108,8 @@ void anon_vma_name_free(struct kref *kref)
>
> struct anon_vma_name *anon_vma_name(struct vm_area_struct *vma)
> {
> - mmap_assert_locked(vma->vm_mm);
> + if (!rwsem_is_locked(&vma->vm_mm->mmap_lock))
> + vma_assert_locked(vma);
>
> return vma->anon_name;
> }
> --
> 2.50.0
