Re: reply: [RFC] pin_user_pages_fast failure count increased

[Hyesoo Yu <hyesoo.yu@samsung.com>]

[-- Attachment #1: Type: text/plain, Size: 2294 bytes --]

On Wed, Jun 04, 2025 at 11:48:54AM +0200, David Hildenbrand wrote:
> > > > > > 
> > > > > So, what's the status of that? We should fix it upstream (*not* caring
> > > > > about controversial out-of-tree pkvm issues).
> > > > Leaving aside the pkvm issue, we should also care about the CMA pages
> > > > mapping to VM by special driver which are intended to be long term
> > > > pinned (actually they are fetched by cma_alloc and then mapped to VM
> > > > instead of alloc_pages during normal page fault).
> > > 
> > > Is there any such "special driver" in the tree?
> > Not that I know of. However, pin_user_pages is exported symbol which
> > could be used for ko, should we make it be capable of dealing with
> > this scenario?
> 
> We have the tendency to not go above and beyond of adding features that we
> cannot even test without OOT drivers.
> 
> Note that we export symbols so other in-tree drivers that are built as a
> module can make use of them.
> 
> > > 
> > > > Could we distinguish
> > > > them by the patch below based on 1aaf8c122, that is, this kind of
> > > > pages is not on page cache and have equaled refcnt to mapcount
> > > 
> > > No, not like that. We'd need some proper indication that this page was
> > > allocated by the CMA area owner, and that owner agrees that the folio
> > > can be long-term pinned (maybe that agreement is by mapping it into user
> > > space, tbd).
> > I think the key point is to distinguish the cma pages which are
> > allocated from fallback of GFP_MOVABLE during common page faults from
> > the ones which got from cma_alloc within the special driver's
> > vm_ops->fault.
> 
> Yes. Using typed pages in the future might work. For now, this is not
> possible yet because the page type overlays page->mapcount.
> 
> Hm.
> 
> > > 
> > > Will you send the fix or should I do it? Discussing about broken use
> > > cases that do no apply upstream is not particularly helpful when we're
> > > dealing with a real upstream bug.
> > I would like to ask for your help on this since I have no further ideas. Thanks
> 
> Okay, let me send a fix for the original commit.
> 

I have prepared a patch and just sent it out.
I'd appreciate it if you could take a look and share your feedback.

Thanks,
Regards.

> -- 
> Cheers,
> 
> David / dhildenb
> 
> 

[-- Attachment #2: Type: text/plain, Size: 0 bytes --]