diff mbox series

[RFC,10/10] mm/hugetlb: Comment at rest huge_pte_offset() places

Message ID 20221030213045.335680-1-peterx@redhat.com (mailing list archive)
State New
Headers show
Series mm/hugetlb: Make huge_pte_offset() thread-safe for pmd unshare | expand

Commit Message

Peter Xu Oct. 30, 2022, 9:30 p.m. UTC 
This makes sure that we're covering all the existing huge_pte_offset()
callers and mention why they are safe regarding to pmd unsharing.

Signed-off-by: Peter Xu <peterx@redhat.com>
---
 mm/hugetlb.c | 14 ++++++++++++++
 1 file changed, 14 insertions(+)

Comments

Nadav Amit Nov. 1, 2022, 5:39 a.m. UTC | #1 
On Oct 30, 2022, at 2:30 PM, Peter Xu <peterx@redhat.com> wrote:

> +		/* With vma lock held, safe without RCU */
> 		src_pte = huge_pte_offset(src, addr, sz);

Just another option to consider: you can create an inline function
huge_pte_offset_locked_mm(), which would do the lockdep_assert_held_*().

I personally would prefer it, since it would clarify exactly the lock you
care about and "make the code document itself”.
Peter Xu Nov. 2, 2022, 9:21 p.m. UTC | #2 
On Mon, Oct 31, 2022 at 10:39:33PM -0700, Nadav Amit wrote:
> On Oct 30, 2022, at 2:30 PM, Peter Xu <peterx@redhat.com> wrote:
> 
> > +		/* With vma lock held, safe without RCU */
> > 		src_pte = huge_pte_offset(src, addr, sz);
> 
> Just another option to consider: you can create an inline function
> huge_pte_offset_locked_mm(), which would do the lockdep_assert_held_*().
> 
> I personally would prefer it, since it would clarify exactly the lock you
> care about and "make the code document itself”.

That's a great suggestion, I'll give it a shot in the next version.

Thanks!
diff mbox series

Patch

diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 6d336d286394..270bfc578115 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -4822,6 +4822,7 @@  int copy_hugetlb_page_range(struct mm_struct *dst, struct mm_struct *src,
 	last_addr_mask = hugetlb_mask_last_page(h);
 	for (addr = src_vma->vm_start; addr < src_vma->vm_end; addr += sz) {
 		spinlock_t *src_ptl, *dst_ptl;
+		/* With vma lock held, safe without RCU */
 		src_pte = huge_pte_offset(src, addr, sz);
 		if (!src_pte) {
 			addr |= last_addr_mask;
@@ -5026,6 +5027,7 @@  int move_hugetlb_page_tables(struct vm_area_struct *vma,
 	hugetlb_vma_lock_write(vma);
 	i_mmap_lock_write(mapping);
 	for (; old_addr < old_end; old_addr += sz, new_addr += sz) {
+		/* With vma lock held, safe without RCU */
 		src_pte = huge_pte_offset(mm, old_addr, sz);
 		if (!src_pte) {
 			old_addr |= last_addr_mask;
@@ -5097,6 +5099,7 @@  static void __unmap_hugepage_range(struct mmu_gather *tlb, struct vm_area_struct
 	last_addr_mask = hugetlb_mask_last_page(h);
 	address = start;
 	for (; address < end; address += sz) {
+		/* With vma lock held, safe without RCU */
 		ptep = huge_pte_offset(mm, address, sz);
 		if (!ptep) {
 			address |= last_addr_mask;
@@ -5402,6 +5405,7 @@  static vm_fault_t hugetlb_wp(struct mm_struct *mm, struct vm_area_struct *vma,
 			mutex_lock(&hugetlb_fault_mutex_table[hash]);
 			hugetlb_vma_lock_read(vma);
 			spin_lock(ptl);
+			/* With vma lock held, safe without RCU */
 			ptep = huge_pte_offset(mm, haddr, huge_page_size(h));
 			if (likely(ptep &&
 				   pte_same(huge_ptep_get(ptep), pte)))
@@ -5440,6 +5444,7 @@  static vm_fault_t hugetlb_wp(struct mm_struct *mm, struct vm_area_struct *vma,
 	 * before the page tables are altered
 	 */
 	spin_lock(ptl);
+	/* With vma lock (and even pgtable lock) held, safe without RCU */
 	ptep = huge_pte_offset(mm, haddr, huge_page_size(h));
 	if (likely(ptep && pte_same(huge_ptep_get(ptep), pte))) {
 		/* Break COW or unshare */
@@ -6511,6 +6516,7 @@  unsigned long hugetlb_change_protection(struct vm_area_struct *vma,
 	last_addr_mask = hugetlb_mask_last_page(h);
 	for (; address < end; address += psize) {
 		spinlock_t *ptl;
+		/* With vma lock held, safe without RCU */
 		ptep = huge_pte_offset(mm, address, psize);
 		if (!ptep) {
 			address |= last_addr_mask;
@@ -7060,7 +7066,14 @@  pte_t *huge_pmd_share(struct mm_struct *mm, struct vm_area_struct *vma,
 
 		saddr = page_table_shareable(svma, vma, addr, idx);
 		if (saddr) {
+			/*
+			 * huge_pmd_share() (or say its solo caller,
+			 * huge_pte_alloc()) always takes the hugetlb vma
+			 * lock, so it's always safe to walk the pgtable of
+			 * the process, even without RCU.
+			 */
 			spte = huge_pte_offset(svma->vm_mm, saddr,
+
 					       vma_mmu_pagesize(svma));
 			if (spte) {
 				get_page(virt_to_page(spte));
@@ -7420,6 +7433,7 @@  void hugetlb_unshare_all_pmds(struct vm_area_struct *vma)
 	hugetlb_vma_lock_write(vma);
 	i_mmap_lock_write(vma->vm_file->f_mapping);
 	for (address = start; address < end; address += PUD_SIZE) {
+		/* With vma lock held, safe without RCU */
 		ptep = huge_pte_offset(mm, address, sz);
 		if (!ptep)
 			continue;