Skip to contentSkip to conversation

eal: fix rte_memcpy build on ppc with gcc 9.3

PATCH Superseded David Christensen 3 participants 12 messages May 4, 2020 – May 6, 2020

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 eal: fix rte_memcpy build on ppc with gcc 9.3

Building DPDK on Ubuntu 20.04 with GCC 9.3.0 results in a "subscript is outside array bounds" message in rte_memcpy function. The build error is caused by an interaction between __builtin_constant_p and "-Werror=array-bounds" as described in this bugzilla:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90387

Modify the code to disable the array-bounds check for GCC versions 9.0 to 9.3.

---
 lib/librte_eal/ppc/include/rte_memcpy.h | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)
lib/librte_eal/ppc/include/rte_memcpy.h +10 -2
index 25311ba1d..dce173d4b 100644
@@ -8,8 +8,7 @@
 
 #include <stdint.h>
 #include <string.h>
-/*To include altivec.h, GCC version must  >= 4.8 */
-#include <altivec.h>
+#include "rte_altivec.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -17,6 +16,11 @@ extern "C" {
 
 #include "generic/rte_memcpy.h"
 
+#if (__GNUC__ == 9 && __GNUC_MINOR__ < 4)
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Warray-bounds"
+#endif
+
 static inline void
 rte_mov16(uint8_t *dst, const uint8_t *src)
 {
@@ -192,6 +196,10 @@ rte_memcpy_func(void *dst, const void *src, size_t n)
 	return ret;
 }
 
+#if (__GNUC__ == 9 && __GNUC_MINOR__ < 4)
+#pragma GCC diagnostic pop
+#endif
+
 #ifdef __cplusplus
 }
 #endif
Signature · 1 line
2.18.1
#1 lore raw

PATCH v2 1/1 eal: fix rte_memcpy build on ppc with gcc 9.3

Building DPDK on Ubuntu 20.04 with GCC 9.3.0 results in a "subscript is outside array bounds" message in rte_memcpy function. The build error is caused by an interaction between __builtin_constant_p and "-Werror=array-bounds" as described in this bugzilla:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90387

Modify the code to disable the array-bounds check for GCC versions 9.0 to 9.3.

---

v2:
* Replace __GNUC__ and __GNUC_MINOR__ with GCC_VERSION

---
 lib/librte_eal/ppc/include/rte_memcpy.h | 13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)
lib/librte_eal/ppc/include/rte_memcpy.h +11 -2
index 25311ba1d..de47a5d2e 100644
@@ -8,8 +8,8 @@
 
 #include <stdint.h>
 #include <string.h>
-/*To include altivec.h, GCC version must  >= 4.8 */
-#include <altivec.h>
+#include "rte_altivec.h"
+#include "rte_common.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -17,6 +17,11 @@ extern "C" {
 
 #include "generic/rte_memcpy.h"
 
+#if (GCC_VERSION >= 90000 && GCC_VERSION < 90400)
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Warray-bounds"
+#endif
+
 static inline void
 rte_mov16(uint8_t *dst, const uint8_t *src)
 {
@@ -192,6 +197,10 @@ rte_memcpy_func(void *dst, const void *src, size_t n)
 	return ret;
 }
 
+#if (GCC_VERSION >= 90000 && GCC_VERSION < 90400)
+#pragma GCC diagnostic pop
+#endif
+
 #ifdef __cplusplus
 }
 #endif
Signature · 1 line
2.18.1
#2 lore raw
Quoted 24 lines · David Christensen

Building DPDK on Ubuntu 20.04 with GCC 9.3.0 results in a "subscript is outside array bounds" message in rte_memcpy function. The build error is caused by an interaction between __builtin_constant_p and "-Werror=array-bounds" as described in this bugzilla:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90387

Modify the code to disable the array-bounds check for GCC versions 9.0 to 9.3.

---

v2:
* Replace __GNUC__ and __GNUC_MINOR__ with GCC_VERSION

---
 lib/librte_eal/ppc/include/rte_memcpy.h | 13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)
lib/librte_eal/ppc/include/rte_memcpy.h +0 -0
index 25311ba1d..de47a5d2e 100644

lib/librte_eal/ppc/include/rte_memcpy.h line 12

Quoted 7 lines
lib/librte_eal/ppc/include/rte_memcpy.h +2 -2
@@ -8,8 +8,8 @@
 
 #include <stdint.h>
 #include <string.h>
-/*To include altivec.h, GCC version must  >= 4.8 */
-#include <altivec.h>
+#include "rte_altivec.h"
+#include "rte_common.h"

I can't find "rte_altivec.h", am I missing something.

With just ignoring "-Warray-bounds" changes, I confirm ena build issue is fixed with gcc 9.1

Quoted 27 lines
lib/librte_eal/ppc/include/rte_memcpy.h +9 -0
 #ifdef __cplusplus
 extern "C" {
@@ -17,6 +17,11 @@ extern "C" {
 
 #include "generic/rte_memcpy.h"
 
+#if (GCC_VERSION >= 90000 && GCC_VERSION < 90400)
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Warray-bounds"
+#endif
+
 static inline void
 rte_mov16(uint8_t *dst, const uint8_t *src)
 {
@@ -192,6 +197,10 @@ rte_memcpy_func(void *dst, const void *src, size_t n)
 	return ret;
 }
 
+#if (GCC_VERSION >= 90000 && GCC_VERSION < 90400)
+#pragma GCC diagnostic pop
+#endif
+
 #ifdef __cplusplus
 }
 #endif
#3 lore raw
Quoted 17 lines
Quoted 7 lines
lib/librte_eal/ppc/include/rte_memcpy.h +0 -0
index 25311ba1d..de47a5d2e 100644

lib/librte_eal/ppc/include/rte_memcpy.h line 12

Quoted 7 lines
lib/librte_eal/ppc/include/rte_memcpy.h +2 -2
@@ -8,8 +8,8 @@
 
  #include <stdint.h>
  #include <string.h>
-/*To include altivec.h, GCC version must  >= 4.8 */
-#include <altivec.h>
+#include "rte_altivec.h"
+#include "rte_common.h"

I can't find "rte_altivec.h", am I missing something.

Quoted 2 lines

With just ignoring "-Warray-bounds" changes, I confirm ena build issue is fixed with gcc 9.1

The rte_altivec.h is related to another open patch required to build on POWER systems (http://patches.dpdk.org/patch/69605/) that's waiting to be accepted. You may not have encountered it if you're not building the MLX5 PMD which has additional library requirements.

Dave

#4 lore raw
Quoted 15 lines · David Christensen
Quoted 15 lines

lib/librte_eal/ppc/include/rte_memcpy.h line 12

Quoted 12 lines
lib/librte_eal/ppc/include/rte_memcpy.h +2 -2
index 25311ba1d..de47a5d2e 100644
@@ -8,8 +8,8 @@
 
  #include <stdint.h>
  #include <string.h>
-/*To include altivec.h, GCC version must  >= 4.8 */
-#include <altivec.h>
+#include "rte_altivec.h"
+#include "rte_common.h"

I can't find "rte_altivec.h", am I missing something.

Quoted 7 lines
Quoted 2 lines

With just ignoring "-Warray-bounds" changes, I confirm ena build issue is fixed with gcc 9.1

The rte_altivec.h is related to another open patch required to build on POWER systems (http://patches.dpdk.org/patch/69605/) that's waiting to be accepted. You may not have encountered it if you're not building the MLX5 PMD which has additional library requirements.

Is there a point in having two patches?

Signature · 1 line
David Marchand
#5 lore raw
Quoted 7 lines
Quoted 5 lines

The rte_altivec.h is related to another open patch required to build on POWER systems (http://patches.dpdk.org/patch/69605/) that's waiting to be accepted. You may not have encountered it if you're not building the MLX5 PMD which has additional library requirements.

Is there a point in having two patches?

The address different problems. This patch addresses a build issue with gcc in Ubuntu 20.04, the other patch is related to the new trace library that uses boolean values from stdbool.h (or the use of the --std=c11 build option) that causes a build error on POWER systems because of altivec definitions.

Dave

#6 lore raw
Quoted 7 lines · David Christensen
Quoted 7 lines
Quoted 6 lines

The rte_altivec.h is related to another open patch required to build on POWER systems (http://patches.dpdk.org/patch/69605/) that's waiting to be accepted. You may not have encountered it if you're not building the MLX5 PMD which has additional library requirements.

Quoted 7 lines
Quoted 1 line

Is there a point in having two patches?

The address different problems. This patch addresses a build issue with gcc in Ubuntu 20.04, the other patch is related to the new trace library that uses boolean values from stdbool.h (or the use of the --std=c11 build option) that causes a build error on POWER systems because of altivec definitions.

Ok, thanks.
WRT the conflict on the other ppc patch, no need for a new version.

Though it is probably worth backporting if it is a problem with gcc 9.x. WDYT?
I can add a Cc: [email protected] when applying.

Signature · 1 line
David Marchand
#7 lore raw
Quoted 3 lines

Though it is probably worth backporting if it is a problem with gcc 9.x. WDYT?
I can add a Cc: [email protected] when applying.

Agree, should be backported.

Dave

#8 lore raw
In reply to David Christensen
Quoted 17 lines · David Christensen
Quoted 17 lines
Quoted 7 lines
lib/librte_eal/ppc/include/rte_memcpy.h +0 -0
index 25311ba1d..de47a5d2e 100644

lib/librte_eal/ppc/include/rte_memcpy.h line 12

Quoted 7 lines
lib/librte_eal/ppc/include/rte_memcpy.h +2 -2
@@ -8,8 +8,8 @@
 
  #include <stdint.h>
  #include <string.h>
-/*To include altivec.h, GCC version must  >= 4.8 */
-#include <altivec.h>
+#include "rte_altivec.h"
+#include "rte_common.h"

I can't find "rte_altivec.h", am I missing something.

Quoted 8 lines
Quoted 2 lines

With just ignoring "-Warray-bounds" changes, I confirm ena build issue is fixed with gcc 9.1

The rte_altivec.h is related to another open patch required to build on POWER systems (http://patches.dpdk.org/patch/69605/) that's waiting to be accepted. You may not have encountered it if you're not building the MLX5 PMD which has additional library requirements.

I see, I missed it. Looks good on top of that patch, although it still doesn't apply cleanly.

It helps to put a comment to the patch about the dependent patch if it is not merged yet.

#9 lore raw
Quoted 11 lines
Quoted 10 lines
Quoted 4 lines

I can't find "rte_altivec.h", am I missing something.

With just ignoring "-Warray-bounds" changes, I confirm ena build issue is fixed with gcc 9.1

The rte_altivec.h is related to another open patch required to build on POWER systems (http://patches.dpdk.org/patch/69605/) that's waiting to be accepted. You may not have encountered it if you're not building the MLX5 PMD which has additional library requirements.

Quoted 5 lines

I see, I missed it. Looks good on top of that patch, although it still doesn't apply cleanly.

It helps to put a comment to the patch about the dependent patch if it is not merged yet.

It was an oversight on my part, forgetting that I had another patch installed. I can submit a corrected version if you think it's necessary but then that that patch will have to be adjusted when it's eventually merged. Sorry for the confusion.

Dave

#10 lore raw
Quoted 14 lines · David Christensen
Quoted 14 lines
Quoted 10 lines
Quoted 4 lines

I can't find "rte_altivec.h", am I missing something.

With just ignoring "-Warray-bounds" changes, I confirm ena build issue is fixed with gcc 9.1

The rte_altivec.h is related to another open patch required to build on POWER systems (http://patches.dpdk.org/patch/69605/) that's waiting to be accepted. You may not have encountered it if you're not building the MLX5 PMD which has additional library requirements.

I see, I missed it. Looks good on top of that patch, although it still doesn't apply cleanly.

Quoted 8 lines
Quoted 2 lines

It helps to put a comment to the patch about the dependent patch if it is not merged yet.

It was an oversight on my part, forgetting that I had another patch installed. I can submit a corrected version if you think it's necessary but then that that patch will have to be adjusted when it's eventually merged. Sorry for the confusion.

No worries, but I won't be merging the patches, it looks easy to manage the conflict but eventually it is up to David to have a new version or not.

#11 lore raw
In reply to David Christensen
Quoted 5 lines · David Christensen

Building DPDK on Ubuntu 20.04 with GCC 9.3.0 results in a "subscript is outside array bounds" message in rte_memcpy function. The build error is caused by an interaction between __builtin_constant_p and "-Werror=array-bounds" as described in this bugzilla:

Good to know, thanks.

Quoted 5 lines

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90387

Modify the code to disable the array-bounds check for GCC versions 9.0 to 9.3.

Quoted 1 line

Applied, thanks.

Signature · 1 line
David Marchand