diff mbox series

[v2,3/4] mm/memory: convert do_shared_fault() to folios

Message ID 20230705194335.273790-3-sidhartha.kumar@oracle.com
State New
Headers
Series [v2,1/4] mm/memory: convert do_page_mkwrite() to use folios |

Commit Message

Sidhartha Kumar July 5, 2023, 7:43 p.m. UTC 
  Saves three implicit calls to compound_head().

Signed-off-by: Sidhartha Kumar <sidhartha.kumar@oracle.com>
---

v2:
	 - move folio initialization after __do_fault()

 mm/memory.c | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)

Comments

Matthew Wilcox July 5, 2023, 8:13 p.m. UTC | #1 
On Wed, Jul 05, 2023 at 12:43:34PM -0700, Sidhartha Kumar wrote:
>  	/*
>  	 * Check if the backing address space wants to know that the page is
>  	 * about to become writable
>  	 */
>  	if (vma->vm_ops->page_mkwrite) {
> -		unlock_page(vmf->page);
> +		folio_unlock(folio);
>  		tmp = do_page_mkwrite(vmf);
>  		if (unlikely(!tmp ||
>  				(tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) {
> -			put_page(vmf->page);
> +			folio_put(folio);

This is _probably_ OK.  However, do_page_mkwrite() calls
vm_ops->page_mkwrite(), and I think it's theoretically possible for the
driver to replace vmf->page with a different one.  The chance of them
actually doing that is pretty low (particularly if they return error or
nopage!), but I'm going to flag it just in case it comes up.

Also, should we pass a folio to do_page_mkwrite() instead of having it
extract the folio from vmf->page?

Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Sidhartha Kumar July 5, 2023, 8:16 p.m. UTC | #2 
On 7/5/23 1:13 PM, Matthew Wilcox wrote:
> On Wed, Jul 05, 2023 at 12:43:34PM -0700, Sidhartha Kumar wrote:
>>   	/*
>>   	 * Check if the backing address space wants to know that the page is
>>   	 * about to become writable
>>   	 */
>>   	if (vma->vm_ops->page_mkwrite) {
>> -		unlock_page(vmf->page);
>> +		folio_unlock(folio);
>>   		tmp = do_page_mkwrite(vmf);
>>   		if (unlikely(!tmp ||
>>   				(tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) {
>> -			put_page(vmf->page);
>> +			folio_put(folio);
> 
> This is _probably_ OK.  However, do_page_mkwrite() calls
> vm_ops->page_mkwrite(), and I think it's theoretically possible for the
> driver to replace vmf->page with a different one.  The chance of them
> actually doing that is pretty low (particularly if they return error or
> nopage!), but I'm going to flag it just in case it comes up.
> 
> Also, should we pass a folio to do_page_mkwrite() instead of having it
> extract the folio from vmf->page?

I can take a look at doing this in a follow-up patch.
> 
> Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>

Did you mean for this to be reviewed-by?

Thanks,
Sidhartha Kumar
Matthew Wilcox July 5, 2023, 8:19 p.m. UTC | #3 
On Wed, Jul 05, 2023 at 01:16:25PM -0700, Sidhartha Kumar wrote:
> On 7/5/23 1:13 PM, Matthew Wilcox wrote:
> > On Wed, Jul 05, 2023 at 12:43:34PM -0700, Sidhartha Kumar wrote:
> > >   	/*
> > >   	 * Check if the backing address space wants to know that the page is
> > >   	 * about to become writable
> > >   	 */
> > >   	if (vma->vm_ops->page_mkwrite) {
> > > -		unlock_page(vmf->page);
> > > +		folio_unlock(folio);
> > >   		tmp = do_page_mkwrite(vmf);
> > >   		if (unlikely(!tmp ||
> > >   				(tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) {
> > > -			put_page(vmf->page);
> > > +			folio_put(folio);
> > 
> > This is _probably_ OK.  However, do_page_mkwrite() calls
> > vm_ops->page_mkwrite(), and I think it's theoretically possible for the
> > driver to replace vmf->page with a different one.  The chance of them
> > actually doing that is pretty low (particularly if they return error or
> > nopage!), but I'm going to flag it just in case it comes up.
> > 
> > Also, should we pass a folio to do_page_mkwrite() instead of having it
> > extract the folio from vmf->page?
> 
> I can take a look at doing this in a follow-up patch.
> > 
> > Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
> 
> Did you mean for this to be reviewed-by?

Uh, yes.  Maybe I need to get more rest ...
diff mbox series

Patch

diff --git a/mm/memory.c b/mm/memory.c
index fba33fb55a010..ce7826d3f6200 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -4593,21 +4593,24 @@  static vm_fault_t do_shared_fault(struct vm_fault *vmf)
 {
 	struct vm_area_struct *vma = vmf->vma;
 	vm_fault_t ret, tmp;
+	struct folio *folio;
 
 	ret = __do_fault(vmf);
 	if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE | VM_FAULT_RETRY)))
 		return ret;
 
+	folio = page_folio(vmf->page);
+
 	/*
 	 * Check if the backing address space wants to know that the page is
 	 * about to become writable
 	 */
 	if (vma->vm_ops->page_mkwrite) {
-		unlock_page(vmf->page);
+		folio_unlock(folio);
 		tmp = do_page_mkwrite(vmf);
 		if (unlikely(!tmp ||
 				(tmp & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))) {
-			put_page(vmf->page);
+			folio_put(folio);
 			return tmp;
 		}
 	}
@@ -4615,8 +4618,8 @@  static vm_fault_t do_shared_fault(struct vm_fault *vmf)
 	ret |= finish_fault(vmf);
 	if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE |
 					VM_FAULT_RETRY))) {
-		unlock_page(vmf->page);
-		put_page(vmf->page);
+		folio_unlock(folio);
+		folio_put(folio);
 		return ret;
 	}