[RFC,v1,4/4] mm/gup: flag to limit follow_page() to transhuge pages

Commit Message

Mircea CIRJALIU - MELIU Dec. 11, 2019, 9:29 a.m. UTC

Sometimes the user needs to look up a transhuge page mapped at a given address.
So instead of being given a normal page and having to test it, save some cycles
by filtering out PTE mappings.

Signed-off-by: Mircea Cirjaliu <[e-mail]>
---
 include/linux/mm.h |  1 +
 mm/gup.c           | 13 +++++++++++--
 2 files changed, 12 insertions(+), 2 deletions(-)

Comments

Kirill A . Shutemov Dec. 11, 2019, 3:56 p.m. UTC | #1

On Wed, Dec 11, 2019 at 09:29:19AM +0000, Mircea CIRJALIU - MELIU wrote:
> Sometimes the user needs to look up a transhuge page mapped at a given address.

There is no such examples in the current kernel, right?

Mircea CIRJALIU - MELIU Dec. 11, 2019, 5:13 p.m. UTC | #2

> On Wed, Dec 11, 2019 at 09:29:19AM +0000, Mircea CIRJALIU - MELIU wrote:
> > Sometimes the user needs to look up a transhuge page mapped at a given
> address.
> 
> There is no such examples in the current kernel, right?

This is a helper patch for remote mapping main patch.
It helps to establish an exact mirror of another process page table.
And exact also involves huge mappings.

> 
> --
>  Kirill A. Shutemov
> 
> ________________________
> This email was scanned by Bitdefender

Patch

diff --git a/include/linux/mm.h b/include/linux/mm.h
index c97ea3b..64bbf83 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -2579,6 +2579,7 @@  struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
 #define FOLL_ANON	0x8000	/* don't do file mappings */
 #define FOLL_LONGTERM	0x10000	/* mapping lifetime is indefinite: see below */
 #define FOLL_SPLIT_PMD	0x20000	/* split huge pmd before returning */
+#define FOLL_HUGE	0x40000 /* only return huge mappings */
 
 /*
  * NOTE on FOLL_LONGTERM:
diff --git a/mm/gup.c b/mm/gup.c
index 7646bf9..a776bdc 100644
--- a/mm/gup.c
+++ b/mm/gup.c
@@ -361,9 +361,11 @@  static struct page *follow_pmd_mask(struct vm_area_struct *vma,
 		if (page)
 			return page;
 	}
-	if (likely(!pmd_trans_huge(pmdval)))
+	if (likely(!pmd_trans_huge(pmdval))) {
+		if (flags & FOLL_HUGE)
+			return ERR_PTR(-EFAULT);
 		return follow_page_pte(vma, address, pmd, flags, &ctx->pgmap);
-
+	}
 	if ((flags & FOLL_NUMA) && pmd_protnone(pmdval))
 		return no_page_table(vma, flags);
 
@@ -382,6 +384,8 @@  static struct page *follow_pmd_mask(struct vm_area_struct *vma,
 	}
 	if (unlikely(!pmd_trans_huge(*pmd))) {
 		spin_unlock(ptl);
+		if (flags & FOLL_HUGE)
+			return ERR_PTR(-EFAULT);
 		return follow_page_pte(vma, address, pmd, flags, &ctx->pgmap);
 	}
 	if (flags & (FOLL_SPLIT | FOLL_SPLIT_PMD)) {
@@ -513,6 +517,8 @@  static struct page *follow_page_mask(struct vm_area_struct *vma,
 	struct page *page;
 	struct mm_struct *mm = vma->vm_mm;
 
+	VM_BUG_ON((flags & (FOLL_SPLIT | FOLL_HUGE)) == (FOLL_SPLIT | FOLL_HUGE));
+
 	ctx->page_mask = 0;
 
 	/* make this handle hugepd */
@@ -685,6 +691,9 @@  static int check_vma_flags(struct vm_area_struct *vma, unsigned long gup_flags)
 	if (gup_flags & FOLL_ANON && !vma_is_anonymous(vma))
 		return -EFAULT;
 
+	if (gup_flags & FOLL_HUGE && !transparent_hugepage_enabled(vma))
+		return -EFAULT;
+
 	if (write) {
 		if (!(vm_flags & VM_WRITE)) {
 			if (!(gup_flags & FOLL_FORCE))