DPDK patches and discussions
 help / color / mirror / Atom feed
* [PATCH dpdk] mbuf: fix strict aliasing error in allocator
@ 2024-09-25 14:00 Robin Jarry
  2024-09-25 15:21 ` Morten Brørup
                   ` (2 more replies)
  0 siblings, 3 replies; 7+ messages in thread
From: Robin Jarry @ 2024-09-25 14:00 UTC (permalink / raw)
  To: dev

When building an application with -fstrict-aliasing -Wstrict-aliasing=2,
we get errors triggered by rte_mbuf_raw_alloc() which is called inline
from rte_pktmbuf_alloc().

 ../dpdk/lib/mbuf/rte_mbuf.h: In function ‘rte_mbuf_raw_alloc’:
 ../dpdk/lib/mbuf/rte_mbuf.h:600:42: error: dereferencing type-punned
 pointer might break strict-aliasing rules [-Werror=strict-aliasing]
   600 |         if (rte_mempool_get(mp, (void **)&m) < 0)
       |                                          ^~

Avoid incorrect casting by changing the type of the returned variable.

Signed-off-by: Robin Jarry <rjarry@redhat.com>
---
 lib/mbuf/rte_mbuf.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/lib/mbuf/rte_mbuf.h b/lib/mbuf/rte_mbuf.h
index babe16c72ccb..bab1fb94d41d 100644
--- a/lib/mbuf/rte_mbuf.h
+++ b/lib/mbuf/rte_mbuf.h
@@ -595,9 +595,9 @@ __rte_mbuf_raw_sanity_check(__rte_unused const struct rte_mbuf *m)
  */
 static inline struct rte_mbuf *rte_mbuf_raw_alloc(struct rte_mempool *mp)
 {
-	struct rte_mbuf *m;
+	void *m;
 
-	if (rte_mempool_get(mp, (void **)&m) < 0)
+	if (rte_mempool_get(mp, &m) < 0)
 		return NULL;
 	__rte_mbuf_raw_sanity_check(m);
 	return m;
-- 
2.46.1


^ permalink raw reply	[flat|nested] 7+ messages in thread

* RE: [PATCH dpdk] mbuf: fix strict aliasing error in allocator
  2024-09-25 14:00 [PATCH dpdk] mbuf: fix strict aliasing error in allocator Robin Jarry
@ 2024-09-25 15:21 ` Morten Brørup
  2024-09-25 15:23   ` Stephen Hemminger
  2024-09-25 15:40 ` [PATCH dpdk v2] " Robin Jarry
  2024-09-26 16:38 ` [PATCH dpdk] " Patrick Robb
  2 siblings, 1 reply; 7+ messages in thread
From: Morten Brørup @ 2024-09-25 15:21 UTC (permalink / raw)
  To: Robin Jarry, dev

> From: Robin Jarry [mailto:rjarry@redhat.com]
> Sent: Wednesday, 25 September 2024 10.00
> 
> When building an application with -fstrict-aliasing -Wstrict-
> aliasing=2,
> we get errors triggered by rte_mbuf_raw_alloc() which is called inline
> from rte_pktmbuf_alloc().
> 
>  ../dpdk/lib/mbuf/rte_mbuf.h: In function ‘rte_mbuf_raw_alloc’:
>  ../dpdk/lib/mbuf/rte_mbuf.h:600:42: error: dereferencing type-punned
>  pointer might break strict-aliasing rules [-Werror=strict-aliasing]
>    600 |         if (rte_mempool_get(mp, (void **)&m) < 0)
>        |                                          ^~
> 
> Avoid incorrect casting by changing the type of the returned variable.
> 
> Signed-off-by: Robin Jarry <rjarry@redhat.com>
> ---
>  lib/mbuf/rte_mbuf.h | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/lib/mbuf/rte_mbuf.h b/lib/mbuf/rte_mbuf.h
> index babe16c72ccb..bab1fb94d41d 100644
> --- a/lib/mbuf/rte_mbuf.h
> +++ b/lib/mbuf/rte_mbuf.h
> @@ -595,9 +595,9 @@ __rte_mbuf_raw_sanity_check(__rte_unused const
> struct rte_mbuf *m)
>   */
>  static inline struct rte_mbuf *rte_mbuf_raw_alloc(struct rte_mempool
> *mp)
>  {
> -	struct rte_mbuf *m;
> +	void *m;
> 
> -	if (rte_mempool_get(mp, (void **)&m) < 0)
> +	if (rte_mempool_get(mp, &m) < 0)
>  		return NULL;
>  	__rte_mbuf_raw_sanity_check(m);
>  	return m;

Suggest:
	__rte_mbuf_raw_sanity_check((struct rte_mbuf *)m);
	return (struct rte_mbuf *)m;

> --
> 2.46.1

Please confirm that this builds with RTE_LIBRTE_MBUF_DEBUG and RTE_ENABLE_ASSERT set in config/rte_config.h. When confirmed, you may add:
Acked-by: Morten Brørup <mb@smartsharesystems.com>


^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: [PATCH dpdk] mbuf: fix strict aliasing error in allocator
  2024-09-25 15:21 ` Morten Brørup
@ 2024-09-25 15:23   ` Stephen Hemminger
  2024-09-25 15:24     ` Robin Jarry
  0 siblings, 1 reply; 7+ messages in thread
From: Stephen Hemminger @ 2024-09-25 15:23 UTC (permalink / raw)
  To: Morten Brørup; +Cc: Robin Jarry, dev

On Wed, 25 Sep 2024 17:21:12 +0200
Morten Brørup <mb@smartsharesystems.com> wrote:

> From: Morten Brørup <mb@smartsharesystems.com>
> To: "Robin Jarry" <rjarry@redhat.com>,  <dev@dpdk.org>
> Subject: RE: [PATCH dpdk] mbuf: fix strict aliasing error in allocator
> Date: Wed, 25 Sep 2024 17:21:12 +0200
> 
> > From: Robin Jarry [mailto:rjarry@redhat.com]
> > Sent: Wednesday, 25 September 2024 10.00
> > 
> > When building an application with -fstrict-aliasing -Wstrict-
> > aliasing=2,
> > we get errors triggered by rte_mbuf_raw_alloc() which is called inline
> > from rte_pktmbuf_alloc().
> > 
> >  ../dpdk/lib/mbuf/rte_mbuf.h: In function ‘rte_mbuf_raw_alloc’:
> >  ../dpdk/lib/mbuf/rte_mbuf.h:600:42: error: dereferencing type-punned
> >  pointer might break strict-aliasing rules [-Werror=strict-aliasing]
> >    600 |         if (rte_mempool_get(mp, (void **)&m) < 0)
> >        |                                          ^~
> > 
> > Avoid incorrect casting by changing the type of the returned variable.
> > 
> > Signed-off-by: Robin Jarry <rjarry@redhat.com>
> > ---
> >  lib/mbuf/rte_mbuf.h | 4 ++--
> >  1 file changed, 2 insertions(+), 2 deletions(-)
> > 
> > diff --git a/lib/mbuf/rte_mbuf.h b/lib/mbuf/rte_mbuf.h
> > index babe16c72ccb..bab1fb94d41d 100644
> > --- a/lib/mbuf/rte_mbuf.h
> > +++ b/lib/mbuf/rte_mbuf.h
> > @@ -595,9 +595,9 @@ __rte_mbuf_raw_sanity_check(__rte_unused const
> > struct rte_mbuf *m)
> >   */
> >  static inline struct rte_mbuf *rte_mbuf_raw_alloc(struct rte_mempool
> > *mp)
> >  {
> > -	struct rte_mbuf *m;
> > +	void *m;
> > 
> > -	if (rte_mempool_get(mp, (void **)&m) < 0)
> > +	if (rte_mempool_get(mp, &m) < 0)
> >  		return NULL;
> >  	__rte_mbuf_raw_sanity_check(m);
> >  	return m;  
> 
> Suggest:
> 	__rte_mbuf_raw_sanity_check((struct rte_mbuf *)m);
> 	return (struct rte_mbuf *)m;

Another way to avoid the warning would be to use a union?

^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: [PATCH dpdk] mbuf: fix strict aliasing error in allocator
  2024-09-25 15:23   ` Stephen Hemminger
@ 2024-09-25 15:24     ` Robin Jarry
  0 siblings, 0 replies; 7+ messages in thread
From: Robin Jarry @ 2024-09-25 15:24 UTC (permalink / raw)
  To: Stephen Hemminger, Morten Brørup; +Cc: dev

Stephen Hemminger, Sep 25, 2024 at 11:23:
> On Wed, 25 Sep 2024 17:21:12 +0200
> Morten Brørup <mb@smartsharesystems.com> wrote:
>
> > From: Morten Brørup <mb@smartsharesystems.com>
> > To: "Robin Jarry" <rjarry@redhat.com>,  <dev@dpdk.org>
> > Subject: RE: [PATCH dpdk] mbuf: fix strict aliasing error in allocator
> > Date: Wed, 25 Sep 2024 17:21:12 +0200
> > 
> > > From: Robin Jarry [mailto:rjarry@redhat.com]
> > > Sent: Wednesday, 25 September 2024 10.00
> > > 
> > > When building an application with -fstrict-aliasing -Wstrict-
> > > aliasing=2,
> > > we get errors triggered by rte_mbuf_raw_alloc() which is called inline
> > > from rte_pktmbuf_alloc().
> > > 
> > >  ../dpdk/lib/mbuf/rte_mbuf.h: In function ‘rte_mbuf_raw_alloc’:
> > >  ../dpdk/lib/mbuf/rte_mbuf.h:600:42: error: dereferencing type-punned
> > >  pointer might break strict-aliasing rules [-Werror=strict-aliasing]
> > >    600 |         if (rte_mempool_get(mp, (void **)&m) < 0)
> > >        |                                          ^~
> > > 
> > > Avoid incorrect casting by changing the type of the returned variable.
> > > 
> > > Signed-off-by: Robin Jarry <rjarry@redhat.com>
> > > ---
> > >  lib/mbuf/rte_mbuf.h | 4 ++--
> > >  1 file changed, 2 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/lib/mbuf/rte_mbuf.h b/lib/mbuf/rte_mbuf.h
> > > index babe16c72ccb..bab1fb94d41d 100644
> > > --- a/lib/mbuf/rte_mbuf.h
> > > +++ b/lib/mbuf/rte_mbuf.h
> > > @@ -595,9 +595,9 @@ __rte_mbuf_raw_sanity_check(__rte_unused const
> > > struct rte_mbuf *m)
> > >   */
> > >  static inline struct rte_mbuf *rte_mbuf_raw_alloc(struct rte_mempool
> > > *mp)
> > >  {
> > > -	struct rte_mbuf *m;
> > > +	void *m;
> > > 
> > > -	if (rte_mempool_get(mp, (void **)&m) < 0)
> > > +	if (rte_mempool_get(mp, &m) < 0)
> > >  		return NULL;
> > >  	__rte_mbuf_raw_sanity_check(m);
> > >  	return m;  
> > 
> > Suggest:
> > 	__rte_mbuf_raw_sanity_check((struct rte_mbuf *)m);
> > 	return (struct rte_mbuf *)m;
>
> Another way to avoid the warning would be to use a union?

I can use an inline union, it will be less code.


^ permalink raw reply	[flat|nested] 7+ messages in thread

* [PATCH dpdk v2] mbuf: fix strict aliasing error in allocator
  2024-09-25 14:00 [PATCH dpdk] mbuf: fix strict aliasing error in allocator Robin Jarry
  2024-09-25 15:21 ` Morten Brørup
@ 2024-09-25 15:40 ` Robin Jarry
  2024-09-25 15:47   ` Stephen Hemminger
  2024-09-26 16:38 ` [PATCH dpdk] " Patrick Robb
  2 siblings, 1 reply; 7+ messages in thread
From: Robin Jarry @ 2024-09-25 15:40 UTC (permalink / raw)
  To: dev

When building an application with -fstrict-aliasing -Wstrict-aliasing=2,
we get errors triggered by rte_mbuf_raw_alloc() which is called inline
from rte_pktmbuf_alloc().

 ../dpdk/lib/mbuf/rte_mbuf.h: In function ‘rte_mbuf_raw_alloc’:
 ../dpdk/lib/mbuf/rte_mbuf.h:600:42: error: dereferencing type-punned
 pointer might break strict-aliasing rules [-Werror=strict-aliasing]
   600 |         if (rte_mempool_get(mp, (void **)&m) < 0)
       |                                          ^~

Avoid incorrect casting by using an inline union variable.

Signed-off-by: Robin Jarry <rjarry@redhat.com>
---

Notes:
    v2: use inline union to fix -Wincompatible-pointer-types

 lib/mbuf/rte_mbuf.h | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)

diff --git a/lib/mbuf/rte_mbuf.h b/lib/mbuf/rte_mbuf.h
index babe16c72ccb..0d2e0e64b3ce 100644
--- a/lib/mbuf/rte_mbuf.h
+++ b/lib/mbuf/rte_mbuf.h
@@ -595,12 +595,15 @@ __rte_mbuf_raw_sanity_check(__rte_unused const struct rte_mbuf *m)
  */
 static inline struct rte_mbuf *rte_mbuf_raw_alloc(struct rte_mempool *mp)
 {
-	struct rte_mbuf *m;
+	union {
+		void *ptr;
+		struct rte_mbuf *m;
+	} ret;
 
-	if (rte_mempool_get(mp, (void **)&m) < 0)
+	if (rte_mempool_get(mp, &ret.ptr) < 0)
 		return NULL;
-	__rte_mbuf_raw_sanity_check(m);
-	return m;
+	__rte_mbuf_raw_sanity_check(ret.m);
+	return ret.m;
 }
 
 /**
-- 
2.46.1


^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: [PATCH dpdk v2] mbuf: fix strict aliasing error in allocator
  2024-09-25 15:40 ` [PATCH dpdk v2] " Robin Jarry
@ 2024-09-25 15:47   ` Stephen Hemminger
  0 siblings, 0 replies; 7+ messages in thread
From: Stephen Hemminger @ 2024-09-25 15:47 UTC (permalink / raw)
  To: Robin Jarry; +Cc: dev

On Wed, 25 Sep 2024 11:40:54 -0400
Robin Jarry <rjarry@redhat.com> wrote:

> From: Robin Jarry <rjarry@redhat.com>
> To: dev@dpdk.org
> Subject: [PATCH dpdk v2] mbuf: fix strict aliasing error in allocator
> Date: Wed, 25 Sep 2024 11:40:54 -0400
> 
> When building an application with -fstrict-aliasing -Wstrict-aliasing=2,
> we get errors triggered by rte_mbuf_raw_alloc() which is called inline
> from rte_pktmbuf_alloc().
> 
>  ../dpdk/lib/mbuf/rte_mbuf.h: In function ‘rte_mbuf_raw_alloc’:
>  ../dpdk/lib/mbuf/rte_mbuf.h:600:42: error: dereferencing type-punned
>  pointer might break strict-aliasing rules [-Werror=strict-aliasing]
>    600 |         if (rte_mempool_get(mp, (void **)&m) < 0)
>        |                                          ^~
> 
> Avoid incorrect casting by using an inline union variable.
> 
> Signed-off-by: Robin Jarry <rjarry@redhat.com>

Thanks, union is safer than cast.

Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>

^ permalink raw reply	[flat|nested] 7+ messages in thread

* Re: [PATCH dpdk] mbuf: fix strict aliasing error in allocator
  2024-09-25 14:00 [PATCH dpdk] mbuf: fix strict aliasing error in allocator Robin Jarry
  2024-09-25 15:21 ` Morten Brørup
  2024-09-25 15:40 ` [PATCH dpdk v2] " Robin Jarry
@ 2024-09-26 16:38 ` Patrick Robb
  2 siblings, 0 replies; 7+ messages in thread
From: Patrick Robb @ 2024-09-26 16:38 UTC (permalink / raw)
  To: Robin Jarry; +Cc: dev

[-- Attachment #1: Type: text/plain, Size: 227 bytes --]

Recheck-request: iol-mellanox-Performance

When I enabled this test on Monday I accidentally set the delta threshold
to 1.5% instead of 5%. So the accepted results window was too small and led
to this false positive test fail.

[-- Attachment #2: Type: text/html, Size: 274 bytes --]

^ permalink raw reply	[flat|nested] 7+ messages in thread

end of thread, other threads:[~2024-09-26 16:39 UTC | newest]

Thread overview: 7+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2024-09-25 14:00 [PATCH dpdk] mbuf: fix strict aliasing error in allocator Robin Jarry
2024-09-25 15:21 ` Morten Brørup
2024-09-25 15:23   ` Stephen Hemminger
2024-09-25 15:24     ` Robin Jarry
2024-09-25 15:40 ` [PATCH dpdk v2] " Robin Jarry
2024-09-25 15:47   ` Stephen Hemminger
2024-09-26 16:38 ` [PATCH dpdk] " Patrick Robb

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).