Re: [PATCH v2] mm/hugetlb: fix a addressing exception caused by huge_pte_offset()

在 2020/2/23 1:02, Matthew Wilcox 写道:

On Sat, Feb 22, 2020 at 02:33:10PM +0800, Longpeng (Mike) wrote:

quoted

在 2020/2/22 13:23, Qian Cai 写道:

quoted

quoted

On Feb 21, 2020, at 10:34 PM, Longpeng(Mike) [off-list ref] wrote:

diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index dd8737a..90daf37 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c

@@ -4910,28 +4910,30 @@ pte_t *huge_pte_offset(struct mm_struct *mm,

{
   pgd_t *pgd;
   p4d_t *p4d;
-    pud_t *pud;
-    pmd_t *pmd;
+    pud_t *pud, pud_entry;
+    pmd_t *pmd, pmd_entry;

   pgd = pgd_offset(mm, addr);
-    if (!pgd_present(*pgd))
+    if (!pgd_present(READ_ONCE(*pgd)))
       return NULL;
   p4d = p4d_offset(pgd, addr);
-    if (!p4d_present(*p4d))
+    if (!p4d_present(READ_ONCE(*p4d)))
       return NULL;

What’s the point of READ_ONCE() on those two places?

As explained in the commit messages, it's for safe(e.g. avoid the compilier
mischief). You can also find the same usage in the ARM64's huge_pte_offset() in
arch/arm64/mm/hugetlbpage.c

I rather agree with Qian; if we need something like READ_ONCE() here,
why don't we always need it as part of pgd_present()?  It seems like an
unnecessary burden for every user.

Hi Matthew & Qian,

Firstly, this is NOT a 'blindly copy', it's an unwise words. I don't know
whether you read the commit message (commit 20a004e7) of ARM64's huge_pte_offset
? If you read, I think worry about the safe is necessary.

Secondly, huge_pte_offset in mm/hugetlb.c is for ARCH_WANT_GENERAL_HUGETLB, many
architectures use it, can you make sure there is no issue on all the
architectures using it with all the version of gcc ?

Thirdly, there are several places use READ_ONCE to access the page table in mm/*
(e.g. gup_pmd_range), they're also generical for all architectures, and they're
much more like unnecessary than here, so why there can use but not here? What's
more, you can read this commit 688272809.

-- 
Regards,
Longpeng(Mike)