[PATCH] vmcore: convert mmap_vmcore_fault() to use folios

[PATCH] vmcore: convert mmap_vmcore_fault() to use folios

[Tal Zussman]

Use a folio for the page cache page the s390 fault handler reads the
old kernel's memory into. This removes four compound_head() calls and
one of the last callers of find_or_create_page().

The vmcore mapping only has order-0 folios, so the logic is unchanged.

Compile tested for s390.

Signed-off-by: Tal Zussman <tz2294@columbia.edu>
---
This is based on the kexec-next branch of the liveupdate tree.
---
 fs/proc/vmcore.c | 27 ++++++++++++++-------------
 1 file changed, 14 insertions(+), 13 deletions(-)

diff --git a/fs/proc/vmcore.c b/fs/proc/vmcore.c
index 44d15436439f..2dc3803aa11d 100644
--- a/fs/proc/vmcore.c
+++ b/fs/proc/vmcore.c
@@ -474,29 +474,30 @@ static vm_fault_t mmap_vmcore_fault(struct vm_fault *vmf)
 	pgoff_t index = vmf->pgoff;
 	struct iov_iter iter;
 	struct kvec kvec;
-	struct page *page;
+	struct folio *folio;
 	loff_t offset;
 	int rc;
 
-	page = find_or_create_page(mapping, index, GFP_KERNEL);
-	if (!page)
+	folio = __filemap_get_folio(mapping, index,
+			FGP_LOCK | FGP_ACCESSED | FGP_CREAT, GFP_KERNEL);
+	if (IS_ERR(folio))
 		return VM_FAULT_OOM;
-	if (!PageUptodate(page)) {
-		offset = (loff_t) index << PAGE_SHIFT;
-		kvec.iov_base = page_address(page);
-		kvec.iov_len = PAGE_SIZE;
-		iov_iter_kvec(&iter, ITER_DEST, &kvec, 1, PAGE_SIZE);
+	if (!folio_test_uptodate(folio)) {
+		offset = folio_pos(folio);
+		kvec.iov_base = folio_address(folio);
+		kvec.iov_len = folio_size(folio);
+		iov_iter_kvec(&iter, ITER_DEST, &kvec, 1, folio_size(folio));
 
 		rc = __read_vmcore(&iter, &offset);
 		if (rc < 0) {
-			unlock_page(page);
-			put_page(page);
+			folio_unlock(folio);
+			folio_put(folio);
 			return vmf_error(rc);
 		}
-		SetPageUptodate(page);
+		folio_mark_uptodate(folio);
 	}
-	unlock_page(page);
-	vmf->page = page;
+	folio_unlock(folio);
+	vmf->page = folio_file_page(folio, index);
 	return 0;
 #else
 	return VM_FAULT_SIGBUS;

---
base-commit: 4d17a004cfd5489f88ad0a7a151dea11a533338b
change-id: 20260907-vmcore-fault-folio-4b382eeff262

Best regards,
--  
Tal Zussman <tz2294@columbia.edu>

Re: [PATCH] vmcore: convert mmap_vmcore_fault() to use folios

[sashiko-bot]

> Use a folio for the page cache page the s390 fault handler reads the
> old kernel's memory into. This removes four compound_head() calls and
> one of the last callers of find_or_create_page().
> 
> The vmcore mapping only has order-0 folios, so the logic is unchanged.
> 
> Compile tested for s390.
> 
> Signed-off-by: Tal Zussman <tz2294@columbia.edu>

Sashiko has reviewed this patch and found no issues. It looks great!

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260915-vmcore-fault-folio-v1-1-a0cb6278670f@columbia.edu?part=1

Re: [PATCH] vmcore: convert mmap_vmcore_fault() to use folios

[Pratyush Yadav]

On Tue, Sep 15 2026, Tal Zussman wrote:

> Use a folio for the page cache page the s390 fault handler reads the
> old kernel's memory into. This removes four compound_head() calls and
> one of the last callers of find_or_create_page().
>
> The vmcore mapping only has order-0 folios, so the logic is unchanged.
>
> Compile tested for s390.
>
> Signed-off-by: Tal Zussman <tz2294@columbia.edu>

Acked-by: Pratyush Yadav <pratyush@kernel.org>

[...]

-- 
Regards,
Pratyush Yadav

Re: [PATCH] vmcore: convert mmap_vmcore_fault() to use folios

[Mike Rapoport]

(cc s390 folks)

On Tue, Sep 15, 2026 at 07:42:15PM -0400, Tal Zussman wrote:
> Use a folio for the page cache page the s390 fault handler reads the

I tiny bit of context why is this only relevant for s390 would have been
nice :)

> old kernel's memory into. This removes four compound_head() calls and
> one of the last callers of find_or_create_page().
> 
> The vmcore mapping only has order-0 folios, so the logic is unchanged.
> 
> Compile tested for s390.
> 
> Signed-off-by: Tal Zussman <tz2294@columbia.edu>
> ---
> This is based on the kexec-next branch of the liveupdate tree.
> ---
>  fs/proc/vmcore.c | 27 ++++++++++++++-------------
>  1 file changed, 14 insertions(+), 13 deletions(-)
> 
> diff --git a/fs/proc/vmcore.c b/fs/proc/vmcore.c
> index 44d15436439f..2dc3803aa11d 100644
> --- a/fs/proc/vmcore.c
> +++ b/fs/proc/vmcore.c
> @@ -474,29 +474,30 @@ static vm_fault_t mmap_vmcore_fault(struct vm_fault *vmf)
>  	pgoff_t index = vmf->pgoff;
>  	struct iov_iter iter;
>  	struct kvec kvec;
> -	struct page *page;
> +	struct folio *folio;
>  	loff_t offset;
>  	int rc;
>  
> -	page = find_or_create_page(mapping, index, GFP_KERNEL);
> -	if (!page)
> +	folio = __filemap_get_folio(mapping, index,
> +			FGP_LOCK | FGP_ACCESSED | FGP_CREAT, GFP_KERNEL);
> +	if (IS_ERR(folio))
>  		return VM_FAULT_OOM;
> -	if (!PageUptodate(page)) {
> -		offset = (loff_t) index << PAGE_SHIFT;
> -		kvec.iov_base = page_address(page);
> -		kvec.iov_len = PAGE_SIZE;
> -		iov_iter_kvec(&iter, ITER_DEST, &kvec, 1, PAGE_SIZE);
> +	if (!folio_test_uptodate(folio)) {
> +		offset = folio_pos(folio);
> +		kvec.iov_base = folio_address(folio);
> +		kvec.iov_len = folio_size(folio);
> +		iov_iter_kvec(&iter, ITER_DEST, &kvec, 1, folio_size(folio));
>  
>  		rc = __read_vmcore(&iter, &offset);
>  		if (rc < 0) {
> -			unlock_page(page);
> -			put_page(page);
> +			folio_unlock(folio);
> +			folio_put(folio);
>  			return vmf_error(rc);
>  		}
> -		SetPageUptodate(page);
> +		folio_mark_uptodate(folio);
>  	}
> -	unlock_page(page);
> -	vmf->page = page;
> +	folio_unlock(folio);
> +	vmf->page = folio_file_page(folio, index);
>  	return 0;
>  #else
>  	return VM_FAULT_SIGBUS;
> 
> ---
> base-commit: 4d17a004cfd5489f88ad0a7a151dea11a533338b
> change-id: 20260907-vmcore-fault-folio-4b382eeff262
> 
> Best regards,
> --  
> Tal Zussman <tz2294@columbia.edu>
> 

-- 
Sincerely yours,
Mike.

Re: [PATCH] vmcore: convert mmap_vmcore_fault() to use folios

[Heiko Carstens]

On Wed, Sep 23, 2026 at 12:24:43PM +0300, Mike Rapoport wrote:
> (cc s390 folks)
> 
> On Tue, Sep 15, 2026 at 07:42:15PM -0400, Tal Zussman wrote:
> > Use a folio for the page cache page the s390 fault handler reads the
> 
> I tiny bit of context why is this only relevant for s390 would have been
> nice :)

It is the ifdef CONFIG_S390 just two lines above the context diff of
the patch; but I was wondering the same and had to apply the patch and
look at the result. :)

In any case:

Acked-by: Heiko Carstens <hca@linux.ibm.com>