Re: [PATCH v8 22/23] mm: Enable PTE markers by default

[Peter Xu <peterx@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 2850 bytes --]

On Tue, Apr 19, 2022 at 11:13:48AM -0400, Johannes Weiner wrote:
> Hi Peter,

Hi, Johannes,

> 
> On Mon, Apr 04, 2022 at 09:49:29PM -0400, Peter Xu wrote:
> > Enable PTE markers by default.  On x86_64 it means it'll auto-enable
> > PTE_MARKER_UFFD_WP as well.
> > 
> > Signed-off-by: Peter Xu <peterx@redhat.com>
> > ---
> >  mm/Kconfig | 2 ++
> >  1 file changed, 2 insertions(+)
> > 
> > diff --git a/mm/Kconfig b/mm/Kconfig
> > index 6e7c2d59fa96..3eca34c864c5 100644
> > --- a/mm/Kconfig
> > +++ b/mm/Kconfig
> > @@ -911,12 +911,14 @@ config ANON_VMA_NAME
> >  
> >  config PTE_MARKER
> >  	bool "Marker PTEs support"
> > +	default y
> >  
> >  	help
> >  	  Allows to create marker PTEs for file-backed memory.
> 
> make oldconfig just prompted me on these:
> 
> ---
> Marker PTEs support (PTE_MARKER) [Y/n/?] (NEW) ?
> 
> CONFIG_PTE_MARKER:
> 
> Allows to create marker PTEs for file-backed memory.
> 
> Symbol: PTE_MARKER [=y]
> Type  : bool
> Defined at mm/Kconfig:1046
>   Prompt: Marker PTEs support
>   Location:
>     Main menu
>       -> Memory Management options
> ---
> 
> >  config PTE_MARKER_UFFD_WP
> >  	bool "Marker PTEs support for userfaultfd write protection"
> > +	default y
> >  	depends on PTE_MARKER && HAVE_ARCH_USERFAULTFD_WP
> 
> It's not possible to answer them without looking at the code.
> 
> But after looking at the code, I'm still not sure why it asks
> me. Isn't this infrastructure code?
> 
> Wouldn't it make more sense to remove the prompt string and have
> userfaultfd simply select those?
> 
> If this is too experimental to enable per default, a more reasonable
> question for the user would be a "userfaultfd file support" option or
> something, and have *that* select the marker code.

Thanks for raising this question.

Actually it's right now enabled by default, so I kept the options just to
make sure we can always explicitly disable those options when we want.
That's majorly why I kept this patch standalone one so if we want we can
even drop it.

Said that, I fully agree with you that having two options seem to be an
overkill, especially the PTE_MARKER option will be too challenging to be
correctly understood by anyone not familiar with it.

So after a 2nd thought I'm trying to refine what I want to achieve with the
kbuild system on this new feature:

- On supported systems (x86_64), should be by default y with "make
  olddefconfig", but able to turn it off using "make oldconfig" etc., so
  the user will have a choice when they want.

- On not-supported systems (non-x86_64), should be always n without
  any prompt asking, and user won't even see this entry.

- PTE_MARKER option should always be hidden for all archs

I plan to post a patch that is attached (I also reworded the entry to not
mention about pte markers).  Does that look acceptable on your side?

Thanks,

-- 
Peter Xu

[-- Attachment #2: 0001-mm-uffd-Hide-PTE_MARKER-option.patch --]
[-- Type: text/plain, Size: 1501 bytes --]

From 5d25e9d685bf129a1730caa61c1545fb16c094bb Mon Sep 17 00:00:00 2001
From: Peter Xu <peterx@redhat.com>
Date: Tue, 19 Apr 2022 15:31:12 -0400
Subject: [PATCH] mm/uffd: Hide PTE_MARKER option
Content-type: text/plain

The PTE_MARKER option should not need to be exposed to the kernel builder,
keep it internal and remove the prompt so it won't be seen.

Instead, make the PTE_MARKER_UFFD_WP option to explicitly choose PTE_MARKER
when necessary.

While PTE_MARKER_UFFD_WP will still prompt to user, change the wording so
that it'll not mention PTE_MARKER at all but renaming it to "Userfaultfd
write protection support for shmem/hugetlbfs".

Reported-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Peter Xu <peterx@redhat.com>
---
 mm/Kconfig | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/mm/Kconfig b/mm/Kconfig
index 3eca34c864c5..d740e1ff3b2f 100644
--- a/mm/Kconfig
+++ b/mm/Kconfig
@@ -910,16 +910,16 @@ config ANON_VMA_NAME
 	  difference in their name.
 
 config PTE_MARKER
-	bool "Marker PTEs support"
-	default y
+	bool
 
 	help
 	  Allows to create marker PTEs for file-backed memory.
 
 config PTE_MARKER_UFFD_WP
-	bool "Marker PTEs support for userfaultfd write protection"
+	bool "Userfaultfd write protection support for shmem/hugetlbfs"
 	default y
-	depends on PTE_MARKER && HAVE_ARCH_USERFAULTFD_WP
+	depends on HAVE_ARCH_USERFAULTFD_WP
+	select PTE_MARKER
 
 	help
 	  Allows to create marker PTEs for userfaultfd write protection
-- 
2.32.0