Re: Does GUP page unpinning have to be done in the pinning context?

[David Hildenbrand <david@redhat.com>]

On 04.04.25 12:20, David Howells wrote:
> Hi Kirill, John, David,

Hi,

> 
> I don't know if you're the experts on GUP, but can you tell me if unpinning,
> e.g. unpin_user_page(), needs to be done in the same MM context and/or the
> same user context as the initial pinning?  And, if so, is there some way to
> break that link?

gup_put_folio() seems to only rely on per-folio information (esp. 
node_stat_mod_folio).

So there should not be such a context requirement.

> 
> I'm looking at how I might extend page pinning into the socket layer and
> sendmsg(MSG_ZEROCOPY).  The problem is that sendmsg() merely queues the
> buffers with no guarantee that it's finished with them by the time it's
> returned.  You get a SO_EE_ORIGIN_ZEROCOPY message to tell you that.
> 
> I have to deal with the issue that the process that did the sendmsg(), the MM
> context and the user context may all have ceased to exist by the time the
> transmission completes if I don't pin them.

AFAIKS that should work.

-- 
Cheers,

David / dhildenb