Re: [PATCH v7 4/4] userfaultfd: use per-vma locks in userfaultfd operations

[Matthew Wilcox <willy@infradead.org>]

On Thu, Jan 23, 2025 at 05:14:27PM +1300, Barry Song wrote:
> However, we’ve noticed that userfaultfd still remains one of the largest users
> of mmap_lock for write operations, with the other—binder—having been recently
> addressed by Carlos Llamas's "binder: faster page installations" series:
> 
> https://lore.kernel.org/lkml/20241203215452.2820071-1-cmllamas@google.com/
> 
> The HeapTaskDaemon(Java GC) might frequently perform userfaultfd_register()
> and userfaultfd_unregister() operations, both of which require the mmap_lock
> in write mode to either split or merge VMAs. Since HeapTaskDaemon is a

I don't think that's an innate necessity.  It does require quite some
work in order to fix it though.  Since it was so much work, nobody's
been motivated to fix it ... but now you are!  Congratulations ;-)

The reason we need the mmap_lock in write mode is that the VMA tree is
currently protected by that mmap_lock.  It shouldn't be.  Logically, it
should be protected by the ma_lock spinlock.  But today we use external
locking (ie the mmap_lock) because fixing that was a huge amount of work
that we didn't have time to do.

This MAP_VOLATILE idea is never going to fly.