PUD and PMD entries do not have a special bit.
get_user_pages_fast() considers any page that passed pmd_huge() as
usable:
if (unlikely(pmd_trans_huge(pmd) || pmd_huge(pmd) ||
pmd_devmap(pmd))) {
And vmf_insert_pfn_pmd_prot() unconditionally sets
entry = pmd_mkhuge(pfn_t_pmd(pfn, prot));
eg on x86 the page will be _PAGE_PRESENT | PAGE_PSE.
As such gup_huge_pmd() will try to deref a struct page:
head = try_grab_compound_head(pmd_page(orig), refs, flags);
and thus crash.
Prevent this by never using IO memory with vmf_insert_pfn_pud/pmd_prot().
Fixes: 314b6580adc5 ("drm/ttm, drm/vmwgfx: Support huge TTM pagefaults")
Signed-off-by: Jason Gunthorpe <j...@nvidia.com>
---
drivers/gpu/drm/ttm/ttm_bo_vm.c | 9 ++++++---
1 file changed, 6 insertions(+), 3 deletions(-)
AFAICT only the vmwgfx driver makes use of this, and I can't tell which path
it is taking.
This function will also try to install a PUD - does vmwgfx have a case where
it has obtained a 1GB high order page - or is that dead code now?
diff --git a/drivers/gpu/drm/ttm/ttm_bo_vm.c b/drivers/gpu/drm/ttm/ttm_bo_vm.c
index f56be5bc0861ec..91d02c345fbba8 100644
--- a/drivers/gpu/drm/ttm/ttm_bo_vm.c
+++ b/drivers/gpu/drm/ttm/ttm_bo_vm.c
@@ -203,10 +203,13 @@ static vm_fault_t ttm_bo_vm_insert_huge(struct vm_fault
*vmf,
if (page_offset + fault_page_size > bo->resource->num_pages)
goto out_fallback;
+ /*
+ * vmf_insert_pfn_pud/pmd_prot() can only be called with struct page
+ * backed memory
+ */
if (bo->resource->bus.is_iomem)
- pfn = ttm_bo_io_mem_pfn(bo, page_offset);
- else
- pfn = page_to_pfn(ttm->pages[page_offset]);
+ goto out_fallback;
+ pfn = page_to_pfn(ttm->pages[page_offset]);
/* pfn must be fault_page_size aligned. */
if ((pfn & (fault_page_size - 1)) != 0)
base-commit: 519d81956ee277b4419c723adfb154603c2565ba
--
2.33.0
