[LSF/MM/BPF TOPIC] Enhancements to RCU-protected VMA walks

[LSF/MM/BPF TOPIC] Enhancements to RCU-protected VMA walks

[Liam R. Howlett]

With the introduction of the per-vma locking in the pagefault handler,
the VMA tree is beginning to be used without the mmap_lock taken at all.

I'd like to talk about where we are in the process of RCU-protected VMA
walks, and where we are headed with a discussion around problems which
may arise and some mitigation ideas.

AKA: The annual mmap_sem/mmap_lock talk - And this time we mean it!

Re: [LSF/MM/BPF TOPIC] Enhancements to RCU-protected VMA walks

[Liam R. Howlett]

Added Cc's

* Liam R. Howlett <Liam.Howlett@Oracle.com> [240119 11:42]:
> With the introduction of the per-vma locking in the pagefault handler,
> the VMA tree is beginning to be used without the mmap_lock taken at all.

Thanks to Suren for this [1] work, I should have included the reference
in the first email.  There are more patch sets that increases the impact
of the locks, but I believe v4 that I've linked is the start of the RCU
use of the maple tree.

> 
> I'd like to talk about where we are in the process of RCU-protected VMA
> walks, and where we are headed with a discussion around problems which
> may arise and some mitigation ideas.

There are some loose ends left in this work that we're still discussing
and working through and I think it would be beneficial to hash some of
it out in person, and have a wider audience exposure.

> 
> AKA: The annual mmap_sem/mmap_lock talk - And this time we mean it!
> 

[1] https://lore.kernel.org/linux-mm/20230227173632.3292573-1-surenb@google.com/