mm, hugepages: add mremap() support for hugepage backed vma

Support mremap() for hugepage backed vma segment by simply repositioning
page table entries.  The page table entries are repositioned to the new
virtual address on mremap().

Hugetlb mremap() support is of course generic; my motivating use case is
a library (hugepage_text), which reloads the ELF text of executables in
hugepages.  This significantly increases the execution performance of
said executables.

Restrict the mremap operation on hugepages to up to the size of the
original mapping as the underlying hugetlb reservation is not yet
capable of handling remapping to a larger size.

During the mremap() operation we detect pmd_share'd mappings and we
unshare those during the mremap().  On access and fault the sharing is
established again.

Link: https://lkml.kernel.org/r/20211013195825.3058275-1-almasrymina@google.com
Signed-off-by: Mina Almasry <almasrymina@google.com>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Ken Chen <kenchen@google.com>
Cc: Chris Kennelly <ckennelly@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Kirill Shutemov <kirill@shutemov.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

1 files changed, 105 insertions, 6 deletions

diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 1835d548fecc..8028fb7677eb 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -1014,6 +1014,35 @@ void reset_vma_resv_huge_pages(struct vm_area_struct *vma)
 		vma->vm_private_data = (void *)0;
 }
 
+/*
+ * Reset and decrement one ref on hugepage private reservation.
+ * Called with mm->mmap_sem writer semaphore held.
+ * This function should be only used by move_vma() and operate on
+ * same sized vma. It should never come here with last ref on the
+ * reservation.
+ */
+void clear_vma_resv_huge_pages(struct vm_area_struct *vma)
+{
+	/*
+	 * Clear the old hugetlb private page reservation.
+	 * It has already been transferred to new_vma.
+	 *
+	 * During a mremap() operation of a hugetlb vma we call move_vma()
+	 * which copies vma into new_vma and unmaps vma. After the copy
+	 * operation both new_vma and vma share a reference to the resv_map
+	 * struct, and at that point vma is about to be unmapped. We don't
+	 * want to return the reservation to the pool at unmap of vma because
+	 * the reservation still lives on in new_vma, so simply decrement the
+	 * ref here and remove the resv_map reference from this vma.
+	 */
+	struct resv_map *reservations = vma_resv_map(vma);
+
+	if (reservations && is_vma_resv_set(vma, HPAGE_RESV_OWNER))
+		kref_put(&reservations->refs, resv_map_release);
+
+	reset_vma_resv_huge_pages(vma);
+}
+
 /* Returns true if the VMA has associated reserve pages */
 static bool vma_has_reserves(struct vm_area_struct *vma, long chg)
 {
@@ -4718,6 +4747,82 @@ again:
 	return ret;
 }
 
+static void move_huge_pte(struct vm_area_struct *vma, unsigned long old_addr,
+			  unsigned long new_addr, pte_t *src_pte)
+{
+	struct hstate *h = hstate_vma(vma);
+	struct mm_struct *mm = vma->vm_mm;
+	pte_t *dst_pte, pte;
+	spinlock_t *src_ptl, *dst_ptl;
+
+	dst_pte = huge_pte_offset(mm, new_addr, huge_page_size(h));
+	dst_ptl = huge_pte_lock(h, mm, dst_pte);
+	src_ptl = huge_pte_lockptr(h, mm, src_pte);
+
+	/*
+	 * We don't have to worry about the ordering of src and dst ptlocks
+	 * because exclusive mmap_sem (or the i_mmap_lock) prevents deadlock.
+	 */
+	if (src_ptl != dst_ptl)
+		spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
+
+	pte = huge_ptep_get_and_clear(mm, old_addr, src_pte);
+	set_huge_pte_at(mm, new_addr, dst_pte, pte);
+
+	if (src_ptl != dst_ptl)
+		spin_unlock(src_ptl);
+	spin_unlock(dst_ptl);
+}
+
+int move_hugetlb_page_tables(struct vm_area_struct *vma,
+			     struct vm_area_struct *new_vma,
+			     unsigned long old_addr, unsigned long new_addr,
+			     unsigned long len)
+{
+	struct hstate *h = hstate_vma(vma);
+	struct address_space *mapping = vma->vm_file->f_mapping;
+	unsigned long sz = huge_page_size(h);
+	struct mm_struct *mm = vma->vm_mm;
+	unsigned long old_end = old_addr + len;
+	unsigned long old_addr_copy;
+	pte_t *src_pte, *dst_pte;
+	struct mmu_notifier_range range;
+
+	mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma, mm, old_addr,
+				old_end);
+	adjust_range_if_pmd_sharing_possible(vma, &range.start, &range.end);
+	mmu_notifier_invalidate_range_start(&range);
+	/* Prevent race with file truncation */
+	i_mmap_lock_write(mapping);
+	for (; old_addr < old_end; old_addr += sz, new_addr += sz) {
+		src_pte = huge_pte_offset(mm, old_addr, sz);
+		if (!src_pte)
+			continue;
+		if (huge_pte_none(huge_ptep_get(src_pte)))
+			continue;
+
+		/* old_addr arg to huge_pmd_unshare() is a pointer and so the
+		 * arg may be modified. Pass a copy instead to preserve the
+		 * value in old_addr.
+		 */
+		old_addr_copy = old_addr;
+
+		if (huge_pmd_unshare(mm, vma, &old_addr_copy, src_pte))
+			continue;
+
+		dst_pte = huge_pte_alloc(mm, new_vma, new_addr, sz);
+		if (!dst_pte)
+			break;
+
+		move_huge_pte(vma, old_addr, new_addr, src_pte);
+	}
+	i_mmap_unlock_write(mapping);
+	flush_tlb_range(vma, old_end - len, old_end);
+	mmu_notifier_invalidate_range_end(&range);
+
+	return len + old_addr - old_end;
+}
+
 static void __unmap_hugepage_range(struct mmu_gather *tlb, struct vm_area_struct *vma,
 				   unsigned long start, unsigned long end,
 				   struct page *ref_page)
@@ -6257,12 +6362,6 @@ void adjust_range_if_pmd_sharing_possible(struct vm_area_struct *vma,
  * sharing is possible.  For hugetlbfs, this prevents removal of any page
  * table entries associated with the address space.  This is important as we
  * are setting up sharing based on existing page table entries (mappings).
- *
- * NOTE: This routine is only called from huge_pte_alloc.  Some callers of
- * huge_pte_alloc know that sharing is not possible and do not take
- * i_mmap_rwsem as a performance optimization.  This is handled by the
- * if !vma_shareable check at the beginning of the routine. i_mmap_rwsem is
- * only required for subsequent processing.
  */
 pte_t *huge_pmd_share(struct mm_struct *mm, struct vm_area_struct *vma,
 		      unsigned long addr, pud_t *pud)