Re: [PATCH v5 05/15] mm/frame-vector: Use FOLL_LONGTERM

[Jason Gunthorpe]

On Fri, Nov 06, 2020 at 11:01:57AM +0100, Daniel Vetter wrote:

> gpu drivers also tend to use vmf_insert_pfn* directly, so we can do
> on-demand paging and move buffers around. From what I glanced for
> lowest level we to the pte_mkspecial correctly (I think I convinced
> myself that vm_insert_pfn does that), but for pud/pmd levels it seems
> just yolo.
> 
> remap_pfn_range seems to indeed split down to pte level always.

Thats what it looked like to me too.
 
> >  From my reading, yes. See ioremap_try_huge_pmd().
> 
> The ioremap here shouldn't matter, since this is for kernel-internal
> mappings. So that's all fine I think.

Right, sorry to be unclear, we are talking about io_remap_pfn_range()
which is for userspace mappings in VMAs

Jason