Re: [PATCH 2/2] mm/frame-vec: use FOLL_LONGTERM

[Jason Gunthorpe]

On Wed, Oct 07, 2020 at 03:06:17PM +0200, Tomasz Figa wrote:

> Note that vb2_vmalloc is only used for in-kernel CPU usage, e.g. the
> contents being copied by the driver between vb2 buffers and some
> hardware FIFO or other dedicated buffers. The memory does not go to
> any hardware DMA.

That is even worse, the CPU can't just blindly touch VM_IO pages, that
isn't portable.

> Could you elaborate on what "the REQUIRED behavior is"? I can see that
> both follow the get_vaddr_frames() -> frame_vector_to_pages() flow, as
> you mentioned. Perhaps the only change needed is switching to
> pin_user_pages after all?

It is the comment right on top of get_vaddr_frames():

  if @start belongs to VM_IO | VM_PFNMAP vma, we don't
  touch page structures and the caller must make sure pfns aren't
  reused for anything else while he is using them.

Which means excluding every kind of VMA that is not something this
driver understands and then using special knowledge of the
driver-specific VMA to assure the above.

For instance if you could detect the VMA is from a carevout and do
something special like hold the fget() while knowning that the struct
file guarentees the carveout remains reserved - then you could use
follow_pfn.

But it would be faster and better to ask the carveout FD for the vaddr
range.

Jason