On Fri, Oct 31, 2025 at 04:12:51PM -0700, Suren Baghdasaryan wrote:

Overall, I think it should be possible to set this flag atomically
under VMA read-lock. However, if you introduce new vm_flags
manipulation functions, please make sure they can't be used for other
vm_flags. In Android I've seen several "interesting" attempts to
update vm_flags under a read-lock (specifically in the page-fault
path) and had to explain why that's a bad idea.

Yeah agreed, so the idea would be to absolutely ring-fence any flag we do
this with entirely. Probably a VM_WARN_ON_ONCE() for anybody trying it with
other flags so bots can catch anybody being naughty.

That kind of 'interesting' stuff is another reason I prefer to really limit
what drivers can do btw ;)

Will have a respin with this idea applied relatively soon hopefully.

Cheers, Lorenzo