bugfix dax fault page should be no cache

Commit Message

[e-mail] Jan. 12, 2016, 11:43 a.m. UTC

From: chenjie <[e-mail]>

We show the pte: 
	*pte=01600000159c8fd3
The page should be no_cache attr,it will be better. 

Cc: <[e-mail]>
Signed-off-by: chenjie <[e-mail]>
---
 fs/dax.c | 2 ++
 1 file changed, 2 insertions(+)

Comments

Matthew Wilcox Jan. 12, 2016, 12:51 p.m. UTC | #1

On Tue, Jan 12, 2016 at 07:43:45PM +0800, [e-mail] wrote:
> From: chenjie <[e-mail]>
> 
> We show the pte: 
> 	*pte=01600000159c8fd3
> The page should be no_cache attr,it will be better. 

I disagree.  What data do you have to show that the WC attribute would
be better than WB?

Even if we agree to that, this is the wrong place to change that
attribute, so the patch is NAKed.  But you need more than "it will
be better".
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[e-mail] Jan. 13, 2016, 3:47 a.m. UTC | #2

The DAX means Direct Access for files?if the pages have cache,
the file is not direct access. But in normal condition,
we can not find the problem directly.

  In our case, we use ext2 filesystem with dax on the reserved mem.
we mmap the file and write some things ,the file is not changed immediately.

this is the wrong place to change that attribute:
	Do you mean
add
	ext4_file_mmap/ext2_file_mmap

	if (IS_DAX(file_inode(file))) {
		vma->vm_ops = &ext4_dax_vm_ops;
		vma->vm_flags |= VM_MIXEDMAP;
+	vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
	} else {
like this ???

On 2016/1/12 20:51, Matthew Wilcox wrote:
> On Tue, Jan 12, 2016 at 07:43:45PM +0800, [e-mail] wrote:
>> From: chenjie <[e-mail]>
>>
>> We show the pte:
>> 	*pte=01600000159c8fd3
>> The page should be no_cache attr,it will be better.
>
> I disagree.  What data do you have to show that the WC attribute would
> be better than WB?
>
> Even if we agree to that, this is the wrong place to change that
> attribute, so the patch is NAKed.  But you need more than "it will
> be better".
>
> .
>

--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Matthew Wilcox Jan. 13, 2016, 4:32 p.m. UTC | #3

On Wed, Jan 13, 2016 at 11:47:12AM +0800, Chenjie (K) wrote:
> The DAX means Direct Access for files?if the pages have cache,
> the file is not direct access. But in normal condition,
> we can not find the problem directly.

The file absolutely is direct accessed.  The CPU cache is critical to
performance, and should not be bypassed lightly.  We have instructions to
flush the cache, and a library (see http://pmem.io/) so that applications
don't have to use them directly.

--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Patch

diff --git a/fs/dax.c b/fs/dax.c
index 43671b6..179ced6 100644
--- a/fs/dax.c
+++ b/fs/dax.c
@@ -328,6 +328,8 @@  static int dax_insert_mapping(struct inode *inode, struct buffer_head *bh,
 		wmb_pmem();
 	}
 
+	vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
+
 	error = vm_insert_mixed(vma, vaddr, pfn);
 
  out: