mm/memory.c: add vm_insert_pages()

Add the ability to insert multiple pages at once to a user VM with lower
PTE spinlock operations.

The intention of this patch-set is to reduce atomic ops for tcp zerocopy
receives, which normally hits the same spinlock multiple times
consecutively.

[[email protected]: pte_alloc() no longer takes the `addr' argument]
[[email protected]: add missing page_count() check to vm_insert_pages()]
  Link: http://lkml.kernel.org/r/[email protected]
[[email protected]: vm_insert_pages() checks if pte_index defined]
  Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Arjun Roy <[email protected]>
Signed-off-by: Eric Dumazet <[email protected]>
Signed-off-by: Soheil Hassas Yeganeh <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Cc: David Miller <[email protected]>
Cc: Matthew Wilcox <[email protected]>
Cc: Jason Gunthorpe <[email protected]>
Cc: Stephen Rothwell <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Linus Torvalds <[email protected]>

Author: arjunroy

include/linux/mm.h

@@ -2689,6 +2689,8 @@ struct vm_area_struct *find_extend_vma(struct mm_struct *, unsigned long addr);
 int remap_pfn_range(struct vm_area_struct *, unsigned long addr,
 			unsigned long pfn, unsigned long size, pgprot_t);
 int vm_insert_page(struct vm_area_struct *, unsigned long addr, struct page *);
+int vm_insert_pages(struct vm_area_struct *vma, unsigned long addr,
+			struct page **pages, unsigned long *num);
 int vm_map_pages(struct vm_area_struct *vma, struct page **pages,
 				unsigned long num);
 int vm_map_pages_zero(struct vm_area_struct *vma, struct page **pages,

mm/memory.c

@@ -1419,8 +1419,7 @@ void zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
 }
 EXPORT_SYMBOL_GPL(zap_vma_ptes);
 
-pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr,
-			spinlock_t **ptl)
+static pmd_t *walk_to_pmd(struct mm_struct *mm, unsigned long addr)
 {
 	pgd_t *pgd;
 	p4d_t *p4d;
@@ -1439,6 +1438,16 @@ pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr,
 		return NULL;
 
 	VM_BUG_ON(pmd_trans_huge(*pmd));
+	return pmd;
+}
+
+pte_t *__get_locked_pte(struct mm_struct *mm, unsigned long addr,
+			spinlock_t **ptl)
+{
+	pmd_t *pmd = walk_to_pmd(mm, addr);
+
+	if (!pmd)
+		return NULL;
 	return pte_alloc_map_lock(mm, pmd, addr, ptl);
 }
 
@@ -1491,6 +1500,122 @@ static int insert_page(struct vm_area_struct *vma, unsigned long addr,
 	return retval;
 }
 
+#ifdef pte_index
+static int insert_page_in_batch_locked(struct mm_struct *mm, pmd_t *pmd,
+			unsigned long addr, struct page *page, pgprot_t prot)
+{
+	int err;
+
+	if (!page_count(page))
+		return -EINVAL;
+	err = validate_page_before_insert(page);
+	return err ? err : insert_page_into_pte_locked(
+		mm, pte_offset_map(pmd, addr), addr, page, prot);
+}
+
+/* insert_pages() amortizes the cost of spinlock operations
+ * when inserting pages in a loop. Arch *must* define pte_index.
+ */
+static int insert_pages(struct vm_area_struct *vma, unsigned long addr,
+			struct page **pages, unsigned long *num, pgprot_t prot)
+{
+	pmd_t *pmd = NULL;
+	spinlock_t *pte_lock = NULL;
+	struct mm_struct *const mm = vma->vm_mm;
+	unsigned long curr_page_idx = 0;
+	unsigned long remaining_pages_total = *num;
+	unsigned long pages_to_write_in_pmd;
+	int ret;
+more:
+	ret = -EFAULT;
+	pmd = walk_to_pmd(mm, addr);
+	if (!pmd)
+		goto out;
+
+	pages_to_write_in_pmd = min_t(unsigned long,
+		remaining_pages_total, PTRS_PER_PTE - pte_index(addr));
+
+	/* Allocate the PTE if necessary; takes PMD lock once only. */
+	ret = -ENOMEM;
+	if (pte_alloc(mm, pmd))
+		goto out;
+	pte_lock = pte_lockptr(mm, pmd);
+
+	while (pages_to_write_in_pmd) {
+		int pte_idx = 0;
+		const int batch_size = min_t(int, pages_to_write_in_pmd, 8);
+
+		spin_lock(pte_lock);
+		for (; pte_idx < batch_size; ++pte_idx) {
+			int err = insert_page_in_batch_locked(mm, pmd,
+				addr, pages[curr_page_idx], prot);
+			if (unlikely(err)) {
+				spin_unlock(pte_lock);
+				ret = err;
+				remaining_pages_total -= pte_idx;
+				goto out;
+			}
+			addr += PAGE_SIZE;
+			++curr_page_idx;
+		}
+		spin_unlock(pte_lock);
+		pages_to_write_in_pmd -= batch_size;
+		remaining_pages_total -= batch_size;
+	}
+	if (remaining_pages_total)
+		goto more;
+	ret = 0;
+out:
+	*num = remaining_pages_total;
+	return ret;
+}
+#endif  /* ifdef pte_index */
+
+/**
+ * vm_insert_pages - insert multiple pages into user vma, batching the pmd lock.
+ * @vma: user vma to map to
+ * @addr: target start user address of these pages
+ * @pages: source kernel pages
+ * @num: in: number of pages to map. out: number of pages that were *not*
+ * mapped. (0 means all pages were successfully mapped).
+ *
+ * Preferred over vm_insert_page() when inserting multiple pages.
+ *
+ * In case of error, we may have mapped a subset of the provided
+ * pages. It is the caller's responsibility to account for this case.
+ *
+ * The same restrictions apply as in vm_insert_page().
+ */
+int vm_insert_pages(struct vm_area_struct *vma, unsigned long addr,
+			struct page **pages, unsigned long *num)
+{
+#ifdef pte_index
+	const unsigned long end_addr = addr + (*num * PAGE_SIZE) - 1;
+
+	if (addr < vma->vm_start || end_addr >= vma->vm_end)
+		return -EFAULT;
+	if (!(vma->vm_flags & VM_MIXEDMAP)) {
+		BUG_ON(down_read_trylock(&vma->vm_mm->mmap_sem));
+		BUG_ON(vma->vm_flags & VM_PFNMAP);
+		vma->vm_flags |= VM_MIXEDMAP;
+	}
+	/* Defer page refcount checking till we're about to map that page. */
+	return insert_pages(vma, addr, pages, num, vma->vm_page_prot);
+#else
+	unsigned long idx = 0, pgcount = *num;
+	int err;
+
+	for (; idx < pgcount; ++idx) {
+		err = vm_insert_page(vma, addr + (PAGE_SIZE * idx), pages[idx]);
+		if (err)
+			break;
+	}
+	*num = pgcount - idx;
+	return err;
+#endif  /* ifdef pte_index */
+}
+EXPORT_SYMBOL(vm_insert_pages);
+
 /**
  * vm_insert_page - insert single page into user vma
  * @vma: user vma to map to