@@ -800,6 +800,12 @@ static inline bool pud_user_accessible_page(pud_t pud)
#endif
#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+#define pte_trans_huge pte_trans_huge
+static inline int pte_trans_huge(pte_t pte)
+{
+ return pte_huge(pte) && pte_napot(pte);
+}
+
static inline int pmd_trans_huge(pmd_t pmd)
{
return pmd_leaf(pmd);
@@ -1584,6 +1584,14 @@ static inline unsigned long my_zero_pfn(unsigned long addr)
#ifdef CONFIG_MMU
+#if (defined(CONFIG_TRANSPARENT_HUGEPAGE) && !defined(pte_trans_huge)) || \
+ (!defined(CONFIG_TRANSPARENT_HUGEPAGE))
+static inline int pte_trans_huge(pte_t pte)
+{
+ return 0;
+}
+#endif
+
#ifndef CONFIG_TRANSPARENT_HUGEPAGE
static inline int pmd_trans_huge(pmd_t pmd)
{
@@ -838,7 +838,7 @@ static inline bool can_follow_write_pte(pte_t pte, struct page *page,
static struct page *follow_page_pte(struct vm_area_struct *vma,
unsigned long address, pmd_t *pmd, unsigned int flags,
- struct dev_pagemap **pgmap)
+ struct follow_page_context *ctx)
{
struct mm_struct *mm = vma->vm_mm;
struct folio *folio;
@@ -879,8 +879,8 @@ static struct page *follow_page_pte(struct vm_area_struct *vma,
* case since they are only valid while holding the pgmap
* reference.
*/
- *pgmap = get_dev_pagemap(pte_pfn(pte), *pgmap);
- if (*pgmap)
+ ctx->pgmap = get_dev_pagemap(pte_pfn(pte), ctx->pgmap);
+ if (ctx->pgmap)
page = pte_page(pte);
else
goto no_page;
@@ -940,6 +940,11 @@ static struct page *follow_page_pte(struct vm_area_struct *vma,
*/
folio_mark_accessed(folio);
}
+ if (is_vm_hugetlb_page(vma) || pte_trans_huge(pte)) {
+ ctx->page_mask = (1 << folio_order(folio)) - 1;
+ page = folio_page(folio, 0) +
+ ((address & (folio_size(folio) - 1)) >> PAGE_SHIFT);
+ }
out:
pte_unmap_unlock(ptep, ptl);
return page;
@@ -975,7 +980,7 @@ static struct page *follow_pmd_mask(struct vm_area_struct *vma,
return no_page_table(vma, flags, address);
}
if (likely(!pmd_leaf(pmdval)))
- return follow_page_pte(vma, address, pmd, flags, &ctx->pgmap);
+ return follow_page_pte(vma, address, pmd, flags, ctx);
if (pmd_protnone(pmdval) && !gup_can_follow_protnone(vma, flags))
return no_page_table(vma, flags, address);
@@ -988,14 +993,14 @@ static struct page *follow_pmd_mask(struct vm_area_struct *vma,
}
if (unlikely(!pmd_leaf(pmdval))) {
spin_unlock(ptl);
- return follow_page_pte(vma, address, pmd, flags, &ctx->pgmap);
+ return follow_page_pte(vma, address, pmd, flags, ctx);
}
if (pmd_trans_huge(pmdval) && (flags & FOLL_SPLIT_PMD)) {
spin_unlock(ptl);
split_huge_pmd(vma, pmd, address);
/* If pmd was left empty, stuff a page table in there quickly */
return pte_alloc(mm, pmd) ? ERR_PTR(-ENOMEM) :
- follow_page_pte(vma, address, pmd, flags, &ctx->pgmap);
+ follow_page_pte(vma, address, pmd, flags, ctx);
}
page = follow_huge_pmd(vma, address, pmd, flags, ctx);
spin_unlock(ptl);
Page mapped at pte level can also be huge page when ARM CONT_PTE or RISC-V SVNAPOT is applied. Lack of huge pte handling logic in follow_page_pte() may lead to both performance and correctness issues. For example, on RISC-V platform, pages in the same 64K huge page have the same pte value, which means follow_page_pte() will get the same page for all of them using pte_pfn(). Then __get_user_pages() will return an array of pages with the same pfn. Mapping these pages causes memory confusion. This error can be triggered by the following code: void *addr = mmap(NULL, 0x10000, PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_PRIVATE | MAP_HUGETLB | MAP_HUGE_64KB, -1, 0); struct vfio_iommu_type1_dma_map dmap_map = { .argsz = sizeof(dma_map), .flags = VFIO_DMA_MAP_FLAG_READ | VFIO_DMA_MAP_FLAG_WRITE, .vaddr = (uint64_t)addr, .size = 0x10000, }; ioctl(vfio_container_fd, VFIO_IOMMU_MAP_DMA, &dma_map); This commit supplies huge pte handling logic in follow_page_pte() to avoid such problems. Signed-off-by: Xu Lu <luxu.kernel@bytedance.com> --- arch/riscv/include/asm/pgtable.h | 6 ++++++ include/linux/pgtable.h | 8 ++++++++ mm/gup.c | 17 +++++++++++------ 3 files changed, 25 insertions(+), 6 deletions(-)