On Mon, Oct 12, 2020 at 12:47 PM Marek Szyprowski <m.szyprowski@xxxxxxxxxxx> wrote: > > Hi Jason, > > On 09.10.2020 14:48, Jason Gunthorpe wrote: > > On Fri, Oct 09, 2020 at 02:37:23PM +0200, Mauro Carvalho Chehab wrote: > > > >> I'm not a mm/ expert, but, from what I understood from Daniel's patch > >> description is that this is unsafe *only if* __GFP_MOVABLE is used. > > No, it is unconditionally unsafe. The CMA movable mappings are > > specific VMAs that will have bad issues here, but there are other > > types too. > > I'm trying to follow this thread, but I really wonder what do you mean > by CMA movable mappings? If a buffer has been allocated from CMA and > used for DMA, it won't be moved in the memory. It will stay at the same > physical memory address all the time until freed by the owner. It just a > matter of proper usage count tracking to delay freeing if it is still > used somewhere. Yup. The problem is that this usage count tracking doesn't exist. And drivers could at least in theory treat CMA like vram and swap buffers in&out of it, so just refcounting the userspace vma isn't enough. In practice, right now, it might be enough for CMA drivers though (but there's more that's possible here). -Daniel -- Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch
