Hi,
I came across commit 8b93d1d7dbd5 ("drm/shmem-helper: Switch to
vmf_insert_pfn") from 2021, which makes it very clear the PFNMAP is
strongly preferred over pages. I totally forgot about that change. The
next iteration of this series will therefore not contain this patch.
Best regards
Thomas
Am 04.02.26 um 17:03 schrieb Matthew Wilcox:
On Wed, Feb 04, 2026 at 12:39:30PM +0100, Thomas Zimmermann wrote:
+ ret = drm_gem_shmem_try_map_pmd(vmf, vmf->address, page);
+ if (ret != VM_FAULT_NOPAGE) {
+ struct folio *folio = page_folio(page);
+
+ get_page(page);
folio_get(folio);
- pfn = page_to_pfn(pages[page_offset]);
- ret = vmf_insert_pfn(vma, vmf->address, pfn);
+ folio_lock(folio);
+
+ vmf->page = page;
+ ret = VM_FAULT_LOCKED;
+ }
- out:
+out:
dma_resv_unlock(shmem->base.resv);
return ret;
@@ -689,7 +698,7 @@ int drm_gem_shmem_mmap(struct drm_gem_shmem_object *shmem,
struct vm_area_struct
if (ret)
return ret;
- vm_flags_set(vma, VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP);
+ vm_flags_mod(vma, VM_DONTEXPAND | VM_DONTDUMP, VM_PFNMAP);
Do you need to explicitly clear VM_PFNMAP here? I'm not familiar with
the DRM stack; maybe that's set for you higher in the stack.
--
--
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Frankenstr. 146, 90461 Nürnberg, Germany, www.suse.com
GF: Jochen Jaser, Andrew McDonald, Werner Knoblich, (HRB 36809, AG Nürnberg)
