On 5/12/2022 7:07 PM, Catalin Marinas wrote:
On Thu, May 12, 2022 at 07:38:55PM +1000, Stephen Rothwell wrote:
After merging the mm tree, today's linux-next build (arm64 defconfig)
failed like this:
arch/arm64/mm/hugetlbpage.c: In function 'huge_ptep_clear_flush':
arch/arm64/mm/hugetlbpage.c:493:16: error: implicit declaration of function 'get_clear_flush'; did you mean 'ptep_clear_flush'? [-Werror=implicit-function-declaration]
493 | return get_clear_flush(vma->vm_mm, addr, ptep, pgsize, ncontig);
| ^~~~~~~~~~~~~~~
| ptep_clear_flush
arch/arm64/mm/hugetlbpage.c:493:16: error: incompatible types when returning type 'int' but 'pte_t' was expected
493 | return get_clear_flush(vma->vm_mm, addr, ptep, pgsize, ncontig);
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
arch/arm64/mm/hugetlbpage.c:494:1: error: control reaches end of non-void function [-Werror=return-type]
494 | }
| ^
Caused by commit
00df1f1a133b ("mm: change huge_ptep_clear_flush() to return the original pte")
interacting with commit
fb396bb459c1 ("arm64/hugetlb: Drop TLB flush from get_clear_flush()")
I have applied the following merg fix patch for today.
From: Stephen Rothwell <sfr@xxxxxxxxxxxxxxxx>
Date: Thu, 12 May 2022 19:33:11 +1000
Subject: [PATCH] fixup for "mm: change huge_ptep_clear_flush() to return the original pte"
It interacts with commit
fb396bb459c1 ("arm64/hugetlb: Drop TLB flush from get_clear_flush()")
from the arm64 tree
Signed-off-by: Stephen Rothwell <sfr@xxxxxxxxxxxxxxxx>
---
arch/arm64/mm/hugetlbpage.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/arch/arm64/mm/hugetlbpage.c b/arch/arm64/mm/hugetlbpage.c
index 5bdf913dedc7..30f5b76aabe9 100644
--- a/arch/arm64/mm/hugetlbpage.c
+++ b/arch/arm64/mm/hugetlbpage.c
@@ -490,7 +490,7 @@ pte_t huge_ptep_clear_flush(struct vm_area_struct *vma,
return ptep_clear_flush(vma, addr, ptep);
ncontig = find_num_contig(vma->vm_mm, addr, ptep, &pgsize);
- return get_clear_flush(vma->vm_mm, addr, ptep, pgsize, ncontig);
+ return get_clear_contig(vma->vm_mm, addr, ptep, pgsize, ncontig);
}
Note that after the arm64 commit, get_clear_contig() no longer flushes
the TLB. So maybe something like:
diff --git a/arch/arm64/mm/hugetlbpage.c b/arch/arm64/mm/hugetlbpage.c
index 30f5b76aabe9..9a999550df8e 100644
--- a/arch/arm64/mm/hugetlbpage.c
+++ b/arch/arm64/mm/hugetlbpage.c
@@ -485,12 +485,15 @@ pte_t huge_ptep_clear_flush(struct vm_area_struct *vma,
{
size_t pgsize;
int ncontig;
+ pte_t orig_pte;
if (!pte_cont(READ_ONCE(*ptep)))
return ptep_clear_flush(vma, addr, ptep);
ncontig = find_num_contig(vma->vm_mm, addr, ptep, &pgsize);
- return get_clear_contig(vma->vm_mm, addr, ptep, pgsize, ncontig);
+ orig_pte = get_clear_contig(vma->vm_mm, addr, ptep, pgsize, ncontig);
+ flush_tlb_range(vma, addr, addr + pgsize * ncontig);
+ return orig_pte;
}
Yes, after checking this fb396bb459c1 ("arm64/hugetlb: Drop TLB flush
from get_clear_flush()"), I also realized it will miss TLB flush.
So I am not sure I need send a incremental patch to fix this issue? Or
resend my patch set [1] with rebasing on the arm64 changes?
Catalin and Andrew, how do you think? Thanks.
[1]
https://lore.kernel.org/all/cover.1652270205.git.baolin.wang@xxxxxxxxxxxxxxxxx/
