Re: [PATCH v6 6/6] uprobe: collapse THP pmd after removing all uprobes

[Hillf Danton <hdanton@sina.com>]

Hello

On Sat, 22 Jun 2019 22:48:29 -0700 Song Liu wrote:
>After all uprobes are removed from the huge page (with PTE pgtable), it
>is possible to collapse the pmd and benefit from THP again. This patch
>does the collapse by setting AS_COLLAPSE_PMD. khugepage would retrace
>the page table.
>
>A check for vma->anon_vma is removed from retract_page_tables(). The
>check was initially marked as "probably overkill". The code works well
>without the check.
>
>An issue on earlier version was discovered by kbuild test robot.
>
>Reported-by: kbuild test robot <lkp@intel.com>
>Signed-off-by: Song Liu <songliubraving@fb.com>
>---
> kernel/events/uprobes.c | 6 +++++-
> mm/khugepaged.c         | 3 ---
> 2 files changed, 5 insertions(+), 4 deletions(-)
>
>diff --git a/kernel/events/uprobes.c b/kernel/events/uprobes.c
>index a20d7b43a056..418382259f61 100644
>--- a/kernel/events/uprobes.c
>+++ b/kernel/events/uprobes.c
>@@ -474,6 +474,7 @@ int uprobe_write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm,
> 	struct page *old_page, *new_page;
> 	struct vm_area_struct *vma;
> 	int ret, is_register, ref_ctr_updated = 0;
>+	struct page *orig_page = NULL;
> 
> 	is_register = is_swbp_insn(&opcode);
> 	uprobe = container_of(auprobe, struct uprobe, arch);
>@@ -512,7 +513,6 @@ int uprobe_write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm,
> 	copy_to_page(new_page, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE);
> 
> 	if (!is_register) {
>-		struct page *orig_page;
> 		pgoff_t index;
> 
> 		index = vaddr_to_offset(vma, vaddr & PAGE_MASK) >> PAGE_SHIFT;
>@@ -540,6 +540,10 @@ int uprobe_write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm,
> 	if (ret && is_register && ref_ctr_updated)
> 		update_ref_ctr(uprobe, mm, -1);
> 
>+	if (!ret && orig_page && PageTransCompound(orig_page))
>+		set_bit(AS_COLLAPSE_PMD,
>+			&compound_head(orig_page)->mapping->flags);
>+
orig_page may be invalid if it is not identical to new_page for instance.

> 	return ret;
> }
> 
>diff --git a/mm/khugepaged.c b/mm/khugepaged.c
>index 9b980327fd9b..2e277a2d731f 100644
>--- a/mm/khugepaged.c
>+++ b/mm/khugepaged.c
>@@ -1302,9 +1302,6 @@ static void retract_page_tables(struct address_space *mapping, pgoff_t pgoff,
> 
> 	i_mmap_lock_write(mapping);
> 	vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
>-		/* probably overkill */
>-		if (vma->anon_vma)
>-			continue;
> 		addr = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
> 		if (addr & ~HPAGE_PMD_MASK)
> 			continue;
>-- 
>2.17.1
>
>
Hillf