Re: [PATCH v3 6/7] mm: introduce mod_vm_flags_nolock and use it in untrack_pfn

[Suren Baghdasaryan]

On Thu, Jan 26, 2023 at 7:47 AM Mel Gorman <mgorman@xxxxxxxxxxxxxxxxxxx> wrote:
>
> On Wed, Jan 25, 2023 at 03:35:53PM -0800, Suren Baghdasaryan wrote:
> > In cases when VMA flags are modified after VMA was isolated and mmap_lock
> > was downgraded, flags modifications would result in an assertion because
> > mmap write lock is not held.
>
> Add note that it's also used during exit when the locking of the VMAs
> becomes irrelevant (mm users is 0, should be no VMA modifications taking
> place other than zap).

Ack.

>
> The typical naming pattern when a caller either knows it holds the necessary
> lock or knows it does not matter is __mod_vm_flags()

Ok. It sounds less explicit but plenty of examples, so I'm fine with
such rename. Will apply in the next version.

>
> > Introduce mod_vm_flags_nolock to be used in such situation, when VMA is
> > not part of VMA tree and locking it is not required.
>
> Instead of such situations, describe in as "used when the caller takes
> responsibility for the required locking".

Ack.

>
> > Pass a hint to untrack_pfn to conditionally use mod_vm_flags_nolock for
> > flags modification and to avoid assertion.
> >
> > Signed-off-by: Suren Baghdasaryan <surenb@xxxxxxxxxx>
>
> Patch itself looks ok. It strays close to being "conditional locking"
> though which might attract some complaints.

The description seems to accurately describe what's done here but I'm
open to better suggestions.
Thanks!

>
> --
> Mel Gorman
> SUSE Labs