[PATCH] memcontrol: use flexible-array member

[PATCH] memcontrol: use flexible-array member

[wenhuizhang]

Change depracated zero-length-and-one-element-arrays into flexible 
array member.
Zero-length and one-element arrays detected by Lukas's CodeChecker.
Zero/one element arrays causes undefined behaviours if sizeof() used.

https://www.kernel.org/doc/html/latest/processd/deprecated.html


Signed-off-by: wenhuizhang<wenhui@gwmail.gwu.edu>
---
 include/linux/memcontrol.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h
index 0ce97eff79e2..f879efb1bc6e 100644
--- a/include/linux/memcontrol.h
+++ b/include/linux/memcontrol.h
@@ -190,7 +190,7 @@ enum memcg_kmem_state {
 
 #if defined(CONFIG_SMP)
 struct memcg_padding {
-	char x[0];
+	char x[];
 } ____cacheline_internodealigned_in_smp;
 #define MEMCG_PADDING(name)      struct memcg_padding name;
 #else
@@ -349,7 +349,7 @@ struct mem_cgroup {
 	struct deferred_split deferred_split_queue;
 #endif
 
-	struct mem_cgroup_per_node *nodeinfo[0];
+	struct mem_cgroup_per_node *nodeinfo[];
 	/* WARNING: nodeinfo must be the last member here */
 };
 
-- 
2.17.1

Re: [External] [PATCH] memcontrol: use flexible-array member

[Muchun Song]

On Tue, May 18, 2021 at 1:20 PM wenhuizhang <wenhui@gwmail.gwu.edu> wrote:
>
>
> Change depracated zero-length-and-one-element-arrays into flexible
> array member.
> Zero-length and one-element arrays detected by Lukas's CodeChecker.
> Zero/one element arrays causes undefined behaviours if sizeof() used.
>
> https://www.kernel.org/doc/html/latest/processd/deprecated.html
>
>
> Signed-off-by: wenhuizhang<wenhui@gwmail.gwu.edu>
> ---
>  include/linux/memcontrol.h | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h
> index 0ce97eff79e2..f879efb1bc6e 100644
> --- a/include/linux/memcontrol.h
> +++ b/include/linux/memcontrol.h
> @@ -190,7 +190,7 @@ enum memcg_kmem_state {
>
>  #if defined(CONFIG_SMP)
>  struct memcg_padding {
> -       char x[0];
> +       char x[];

Have you tested this?

The compiler will complain about this. The output will be something like below.

    error: flexible array member in otherwise empty struct

Thanks.

>  } ____cacheline_internodealigned_in_smp;
>  #define MEMCG_PADDING(name)      struct memcg_padding name;
>  #else
> @@ -349,7 +349,7 @@ struct mem_cgroup {
>         struct deferred_split deferred_split_queue;
>  #endif
>
> -       struct mem_cgroup_per_node *nodeinfo[0];
> +       struct mem_cgroup_per_node *nodeinfo[];
>         /* WARNING: nodeinfo must be the last member here */
>  };
>
> --
> 2.17.1
>

Re: [PATCH] memcontrol: use flexible-array member

[Michal Hocko]

On Tue 18-05-21 01:20:36, wenhuizhang wrote:
> 
> Change depracated zero-length-and-one-element-arrays into flexible 
> array member.
> Zero-length and one-element arrays detected by Lukas's CodeChecker.
> Zero/one element arrays causes undefined behaviours if sizeof() used.
> 
> https://www.kernel.org/doc/html/latest/processd/deprecated.html

I do understand why zero-length structures are discouraged for trail
space but does the same concern apply to internal padding inside
structures to enforce alignment to different cache lines? AFAIK this is
a trick we use at other places and I do not remember having heard about
any problems.

nodeinfo is a different case where flexible arrays seems reasonable to
me.

> Signed-off-by: wenhuizhang<wenhui@gwmail.gwu.edu>
> ---
>  include/linux/memcontrol.h | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h
> index 0ce97eff79e2..f879efb1bc6e 100644
> --- a/include/linux/memcontrol.h
> +++ b/include/linux/memcontrol.h
> @@ -190,7 +190,7 @@ enum memcg_kmem_state {
>  
>  #if defined(CONFIG_SMP)
>  struct memcg_padding {
> -	char x[0];
> +	char x[];
>  } ____cacheline_internodealigned_in_smp;
>  #define MEMCG_PADDING(name)      struct memcg_padding name;
>  #else
> @@ -349,7 +349,7 @@ struct mem_cgroup {
>  	struct deferred_split deferred_split_queue;
>  #endif
>  
> -	struct mem_cgroup_per_node *nodeinfo[0];
> +	struct mem_cgroup_per_node *nodeinfo[];
>  	/* WARNING: nodeinfo must be the last member here */
>  };
>  
> -- 
> 2.17.1

-- 
Michal Hocko
SUSE Labs

Re: [External] Re: [PATCH] memcontrol: use flexible-array member

[Muchun Song]

On Tue, May 18, 2021 at 4:08 PM Michal Hocko <mhocko@suse.com> wrote:
>
> On Tue 18-05-21 01:20:36, wenhuizhang wrote:
> >
> > Change depracated zero-length-and-one-element-arrays into flexible
> > array member.
> > Zero-length and one-element arrays detected by Lukas's CodeChecker.
> > Zero/one element arrays causes undefined behaviours if sizeof() used.
> >
> > https://www.kernel.org/doc/html/latest/processd/deprecated.html
>
> I do understand why zero-length structures are discouraged for trail
> space but does the same concern apply to internal padding inside
> structures to enforce alignment to different cache lines? AFAIK this is
> a trick we use at other places and I do not remember having heard about
> any problems.
>
> nodeinfo is a different case where flexible arrays seems reasonable to
> me.

I do agree with Michal's suggestion. If you only change nodeinfo
to a flexible array in the next version. Please also remove the
comment below it.

>
> > Signed-off-by: wenhuizhang<wenhui@gwmail.gwu.edu>
> > ---
> >  include/linux/memcontrol.h | 4 ++--
> >  1 file changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h
> > index 0ce97eff79e2..f879efb1bc6e 100644
> > --- a/include/linux/memcontrol.h
> > +++ b/include/linux/memcontrol.h
> > @@ -190,7 +190,7 @@ enum memcg_kmem_state {
> >
> >  #if defined(CONFIG_SMP)
> >  struct memcg_padding {
> > -     char x[0];
> > +     char x[];
> >  } ____cacheline_internodealigned_in_smp;
> >  #define MEMCG_PADDING(name)      struct memcg_padding name;
> >  #else
> > @@ -349,7 +349,7 @@ struct mem_cgroup {
> >       struct deferred_split deferred_split_queue;
> >  #endif
> >
> > -     struct mem_cgroup_per_node *nodeinfo[0];
> > +     struct mem_cgroup_per_node *nodeinfo[];
> >       /* WARNING: nodeinfo must be the last member here */

I mean this comment. Because the compiler will help us to
check this if we convert it to a flexible array.

Thanks.

> >  };
> >
> > --
> > 2.17.1
>
> --
> Michal Hocko
> SUSE Labs