[RFC PATCH 4/6] mm/mremap: Use mmu gather interface instead of flush_tlb_range

["Aneesh Kumar K.V" <aneesh.kumar@linux.ibm.com>]

Some architectures do have the concept of page walk cache and only mmu gather
interface supports flushing them. A fast mremap that involves moving page
table pages instead of copying pte entries should flush page walk cache since
the old translation cache is no more valid. Hence switch to mm gather to flush
TLB and mark tlb.freed_tables = 1. No page table pages need to be freed here.
With this the tlb flush is done outside page table lock (ptl).

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
---
 mm/mremap.c | 33 +++++++++++++++++++++++++++++----
 1 file changed, 29 insertions(+), 4 deletions(-)

diff --git a/mm/mremap.c b/mm/mremap.c
index 54fd2302b99d..14778d215011 100644
--- a/mm/mremap.c
+++ b/mm/mremap.c
@@ -217,6 +217,7 @@ static bool move_normal_pmd(struct vm_area_struct *vma, unsigned long old_addr,
 {
 	spinlock_t *old_ptl, *new_ptl;
 	struct mm_struct *mm = vma->vm_mm;
+	struct mmu_gather tlb;
 	pmd_t pmd;
 
 	/*
@@ -245,11 +246,12 @@ static bool move_normal_pmd(struct vm_area_struct *vma, unsigned long old_addr,
 	if (WARN_ON_ONCE(!pmd_none(*new_pmd)))
 		return false;
 
+	tlb_gather_mmu(&tlb, mm, old_addr, PMD_SIZE);
 	/*
 	 * We don't have to worry about the ordering of src and dst
 	 * ptlocks because exclusive mmap_lock prevents deadlock.
 	 */
-	old_ptl = pmd_lock(vma->vm_mm, old_pmd);
+	old_ptl = pmd_lock(mm, old_pmd);
 	new_ptl = pmd_lockptr(mm, new_pmd);
 	if (new_ptl != old_ptl)
 		spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING);
@@ -258,13 +260,23 @@ static bool move_normal_pmd(struct vm_area_struct *vma, unsigned long old_addr,
 	pmd = *old_pmd;
 	pmd_clear(old_pmd);
 
+	/*
+	 * Mark the range. We are not freeing page table pages nor
+	 * regular pages. Hence we don't need to call tlb_remove_table()
+	 * or tlb_remove_page().
+	 */
+	tlb_flush_pte_range(&tlb, old_addr, PMD_SIZE);
+	tlb.freed_tables = 1;
 	VM_BUG_ON(!pmd_none(*new_pmd));
 	pmd_populate(mm, new_pmd, (pgtable_t)pmd_page_vaddr(pmd));
 
-	flush_tlb_range(vma, old_addr, old_addr + PMD_SIZE);
 	if (new_ptl != old_ptl)
 		spin_unlock(new_ptl);
 	spin_unlock(old_ptl);
+	/*
+	 * This will invalidate both the old TLB and page table walk caches.
+	 */
+	tlb_finish_mmu(&tlb, old_addr, PMD_SIZE);
 
 	return true;
 }
@@ -283,6 +295,7 @@ static bool move_normal_pud(struct vm_area_struct *vma, unsigned long old_addr,
 {
 	spinlock_t *old_ptl, *new_ptl;
 	struct mm_struct *mm = vma->vm_mm;
+	struct mmu_gather tlb;
 	pud_t pud;
 
 	/*
@@ -292,11 +305,12 @@ static bool move_normal_pud(struct vm_area_struct *vma, unsigned long old_addr,
 	if (WARN_ON_ONCE(!pud_none(*new_pud)))
 		return false;
 
+	tlb_gather_mmu(&tlb, mm, old_addr, PUD_SIZE);
 	/*
 	 * We don't have to worry about the ordering of src and dst
 	 * ptlocks because exclusive mmap_lock prevents deadlock.
 	 */
-	old_ptl = pud_lock(vma->vm_mm, old_pud);
+	old_ptl = pud_lock(mm, old_pud);
 	new_ptl = pud_lockptr(mm, new_pud);
 	if (new_ptl != old_ptl)
 		spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING);
@@ -305,14 +319,25 @@ static bool move_normal_pud(struct vm_area_struct *vma, unsigned long old_addr,
 	pud = *old_pud;
 	pud_clear(old_pud);
 
+	/*
+	 * Mark the range. We are not freeing page table pages nor
+	 * regular pages. Hence we don't need to call tlb_remove_table()
+	 * or tlb_remove_page().
+	 */
+	tlb_flush_pte_range(&tlb, old_addr, PUD_SIZE);
+	tlb.freed_tables = 1;
 	VM_BUG_ON(!pud_none(*new_pud));
 
 	pud_populate(mm, new_pud, (pmd_t *)pud_page_vaddr(pud));
-	flush_tlb_range(vma, old_addr, old_addr + PUD_SIZE);
+
 	if (new_ptl != old_ptl)
 		spin_unlock(new_ptl);
 	spin_unlock(old_ptl);
 
+	/*
+	 * This will invalidate both the old TLB and page table walk caches.
+	 */
+	tlb_finish_mmu(&tlb, old_addr, PUD_SIZE);
 	return true;
 }
 #else
-- 
2.29.2