Donate to e Foundation | Murena handsets with /e/OS | Own a part of Murena! Learn more

Commit e040f218 authored by Hugh Dickins's avatar Hugh Dickins Committed by Linus Torvalds
Browse files

[PATCH] mm: copy_pte_range progress fix 



My latency breaking in copy_pte_range didn't work as intended: instead of
checking at regularish intervals, after the first interval it checked every
time around the loop, too impatient to be preempted.  Fix that.

Signed-off-by: default avatarHugh Dickins <hugh@veritas.com>
Signed-off-by: default avatarAndrew Morton <akpm@osdl.org>
Signed-off-by: default avatarLinus Torvalds <torvalds@osdl.org>
parent 09ad4bbc

mm/memory.c

+8 −6
Original line number Diff line number Diff line
@@ -410,7 +410,7 @@ static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 
{

	pte_t *src_pte, *dst_pte;

	unsigned long vm_flags = vma->vm_flags;

	int progress;

	int progress = 0;



again:

	dst_pte = pte_alloc_map(dst_mm, dst_pmd, addr);
@@ -418,17 +418,19 @@ again: 
		return -ENOMEM;

	src_pte = pte_offset_map_nested(src_pmd, addr);



	progress = 0;

	spin_lock(&src_mm->page_table_lock);

	do {

		/*

		 * We are holding two locks at this point - either of them

		 * could generate latencies in another task on another CPU.

		 */

		if (progress >= 32 && (need_resched() ||

		if (progress >= 32) {

			progress = 0;

			if (need_resched() ||

			    need_lockbreak(&src_mm->page_table_lock) ||

		    need_lockbreak(&dst_mm->page_table_lock)))

			    need_lockbreak(&dst_mm->page_table_lock))

				break;

		}

		if (pte_none(*src_pte)) {

			progress++;

			continue;