Skip to contentSkip to conversation

package/libnss: fix build failure with ARM without NEON extension

PATCH Superseded Giulio Benetti 3 participants 15 messages Oct 23, 2019 – Oct 29, 2019

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/4 package/libnss: fix build failure with ARM without NEON extension

At the moment libnss assumes that every ARM has NEON extension but it's not that way. So add a patch to make it aware of it and use native functions in place of NEON optimized ones.

---
Already sent upstream:
https://bugzilla.mozilla.org/show_bug.cgi?id=1590676

---
 ...ix-build-if-arm-doesn-t-support-NEON.patch | 47 +++++++++++++++++++
 1 file changed, 47 insertions(+)
 create mode 100644 package/libnss/0003-Bug-1590676-Fix-build-if-arm-doesn-t-support-NEON.patch
package/libnss/0003-Bug-1590676-Fix-build-if-arm-doesn-t-support-NEON.patch new file +47 -0
new file mode 100644
index 0000000000..e536282371
@@ -0,0 +1,47 @@
+From 946bc2864b0de7b63a4ec415e42e622d591a8753 Mon Sep 17 00:00:00 2001
+From: Giulio Benetti <[email protected]>
+Date: Wed, 23 Oct 2019 11:47:03 +0200
+Subject: [PATCH] Bug 1590676 - Fix build if arm doesn't support NEON
+
+At the moment NSS assumes that ARM supports NEON extension but this is
+not true and leads to build failure on ARM without NEON extension.
+Add check to assure USE_HW_AES is not defined if ARM without NEON
+extension is used.
+
+Signed-off-by: Giulio Benetti <[email protected]>
+---
+ nss/lib/freebl/aes-armv8.c | 3 ++-
+ nss/lib/freebl/rijndael.c  | 3 ++-
+ 2 files changed, 4 insertions(+), 2 deletions(-)
+
+diff --git a/nss/lib/freebl/aes-armv8.c b/nss/lib/freebl/aes-armv8.c
+index 40d5e2d34..a5df53891 100644
+--- a/nss/lib/freebl/aes-armv8.c
++++ b/nss/lib/freebl/aes-armv8.c
+@@ -7,7 +7,8 @@
+ 
+ #if (defined(__clang__) ||                            \
+      (defined(__GNUC__) && defined(__GNUC_MINOR__) && \
+-      (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ > 8))))
++      (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ > 8))) && \
++      (defined(__ARM_NEON) || defined(__ARM_NEON__)))
+ 
+ #ifndef __ARM_FEATURE_CRYPTO
+ #error "Compiler option is invalid"
+diff --git a/nss/lib/freebl/rijndael.c b/nss/lib/freebl/rijndael.c
+index 26bd58ee0..23893f419 100644
+--- a/nss/lib/freebl/rijndael.c
++++ b/nss/lib/freebl/rijndael.c
+@@ -20,7 +20,8 @@
+ #include "gcm.h"
+ #include "mpi.h"
+ 
+-#if !defined(IS_LITTLE_ENDIAN) && !defined(NSS_X86_OR_X64)
++#if (!defined(IS_LITTLE_ENDIAN) && !defined(NSS_X86_OR_X64)) || \
++    (!defined(__ARM_NEON) && !defined(__ARM_NEON__))
+ // not test yet on big endian platform of arm
+ #undef USE_HW_AES
+ #endif
+-- 
+2.20.1
+
Signature · 1 line
2.20.1
#1 lore raw

PATCH 2/4 package/libnss: fix build warning

Add patch to fix build warning due to uninitialized variable.

---
https://bugzilla.mozilla.org/show_bug.cgi?id=1590678

---
 ...ve-Wmaybe-uninitialized-warning-in-t.patch | 27 +++++++++++++++++++
 1 file changed, 27 insertions(+)
 create mode 100644 package/libnss/0004-Bug-1590678-Remove-Wmaybe-uninitialized-warning-in-t.patch
package/libnss/0004-Bug-1590678-Remove-Wmaybe-uninitialized-warning-in-t.patch new file +27 -0
new file mode 100644
index 0000000000..956abeb4a5
@@ -0,0 +1,27 @@
+From 24bcc8860310149da1524dbf25f3bc77f6476b11 Mon Sep 17 00:00:00 2001
+From: Giulio Benetti <[email protected]>
+Date: Wed, 23 Oct 2019 11:58:12 +0200
+Subject: [PATCH] Bug 1590678 - Remove -Wmaybe-uninitialized warning in
+ tls13esni.c
+
+Signed-off-by: Giulio Benetti <[email protected]>
+---
+ nss/lib/ssl/tls13esni.c | 2 +-
+ 1 file changed, 1 insertion(+), 1 deletion(-)
+
+diff --git a/nss/lib/ssl/tls13esni.c b/nss/lib/ssl/tls13esni.c
+index 4d2e12d62..a7ce6f568 100644
+--- a/nss/lib/ssl/tls13esni.c
++++ b/nss/lib/ssl/tls13esni.c
+@@ -728,7 +728,7 @@ tls13_ServerDecryptEsniXtn(const sslSocket *ss, const PRUint8 *in, unsigned int
+ {
+     sslReader rdr = SSL_READER(in, inLen);
+     PRUint64 suite;
+-    const ssl3CipherSuiteDef *suiteDef;
++    const ssl3CipherSuiteDef *suiteDef = NULL;
+     SSLAEADCipher aead = NULL;
+     TLSExtension *keyShareExtension;
+     TLS13KeyShareEntry *entry = NULL;
+-- 
+2.20.1
+
Signature · 1 line
2.20.1
#2 lore raw

PATCH 2/4 package/libnss: fix build warning

Quoted 27 lines · Giulio Benetti

Add patch to fix build warning due to uninitialized variable.

---
https://bugzilla.mozilla.org/show_bug.cgi?id=1590678

---
 ...ve-Wmaybe-uninitialized-warning-in-t.patch | 27 +++++++++++++++++++
 1 file changed, 27 insertions(+)
 create mode 100644 package/libnss/0004-Bug-1590678-Remove-Wmaybe-uninitialized-warning-in-t.patch
package/libnss/0004-Bug-1590678-Remove-Wmaybe-uninitialized-warning-in-t.patch new file +11 -0
new file mode 100644
index 0000000000..956abeb4a5
@@ -0,0 +1,27 @@
+From 24bcc8860310149da1524dbf25f3bc77f6476b11 Mon Sep 17 00:00:00 2001
+From: Giulio Benetti <[email protected]>
+Date: Wed, 23 Oct 2019 11:58:12 +0200
+Subject: [PATCH] Bug 1590678 - Remove -Wmaybe-uninitialized warning in
+ tls13esni.c
+
+Signed-off-by: Giulio Benetti <[email protected]>
+---
+ nss/lib/ssl/tls13esni.c | 2 +-
+ 1 file changed, 1 insertion(+), 1 deletion(-)
+

package/libnss/0004-Bug-1590678-Remove-Wmaybe-uninitialized-warning-in-t.patch line 21

Quoted 10 lines
package/libnss/0004-Bug-1590678-Remove-Wmaybe-uninitialized-warning-in-t.patch new file +10 -0
+diff --git a/nss/lib/ssl/tls13esni.c b/nss/lib/ssl/tls13esni.c
+index 4d2e12d62..a7ce6f568 100644
+--- a/nss/lib/ssl/tls13esni.c
++++ b/nss/lib/ssl/tls13esni.c
+@@ -728,7 +728,7 @@ tls13_ServerDecryptEsniXtn(const sslSocket *ss, const PRUint8 *in, unsigned int
+ {
+     sslReader rdr = SSL_READER(in, inLen);
+     PRUint64 suite;
+-    const ssl3CipherSuiteDef *suiteDef;
++    const ssl3CipherSuiteDef *suiteDef = NULL;

Although this looks OK at first sight, it's something that might be dangerous
(i.e. the wrong fix). Since this is a security package, I don't want to apply it until upstream gives feedback.

Regards,
Arnout

Quoted 7 lines
package/libnss/0004-Bug-1590678-Remove-Wmaybe-uninitialized-warning-in-t.patch new file +6 -0
+     SSLAEADCipher aead = NULL;
+     TLSExtension *keyShareExtension;
+     TLS13KeyShareEntry *entry = NULL;
+-- 
+2.20.1
+
#3 lore raw

PATCH 2/4 package/libnss: fix build warning

Quoted 36 lines · Arnout Vandecappelle
Quoted 33 lines · Giulio Benetti

Add patch to fix build warning due to uninitialized variable.

---
https://bugzilla.mozilla.org/show_bug.cgi?id=1590678

---
 ...ve-Wmaybe-uninitialized-warning-in-t.patch | 27 +++++++++++++++++++
 1 file changed, 27 insertions(+)
 create mode 100644 package/libnss/0004-Bug-1590678-Remove-Wmaybe-uninitialized-warning-in-t.patch
package/libnss/0004-Bug-1590678-Remove-Wmaybe-uninitialized-warning-in-t.patch new file +17 -0
new file mode 100644
index 0000000000..956abeb4a5
@@ -0,0 +1,27 @@
+From 24bcc8860310149da1524dbf25f3bc77f6476b11 Mon Sep 17 00:00:00 2001
+From: Giulio Benetti <[email protected]>
+Date: Wed, 23 Oct 2019 11:58:12 +0200
+Subject: [PATCH] Bug 1590678 - Remove -Wmaybe-uninitialized warning in
+ tls13esni.c
+
+Signed-off-by: Giulio Benetti <[email protected]>
+---
+ nss/lib/ssl/tls13esni.c | 2 +-
+ 1 file changed, 1 insertion(+), 1 deletion(-)
+
+diff --git a/nss/lib/ssl/tls13esni.c b/nss/lib/ssl/tls13esni.c
+index 4d2e12d62..a7ce6f568 100644
+--- a/nss/lib/ssl/tls13esni.c
++++ b/nss/lib/ssl/tls13esni.c
+@@ -728,7 +728,7 @@ tls13_ServerDecryptEsniXtn(const sslSocket *ss, const PRUint8 *in, unsigned int
+ {
Quoted 8 lines

package/libnss/0004-Bug-1590678-Remove-Wmaybe-uninitialized-warning-in-t.patch line 21

Quoted 4 lines
package/libnss/0004-Bug-1590678-Remove-Wmaybe-uninitialized-warning-in-t.patch new file +4 -0
+     sslReader rdr = SSL_READER(in, inLen);
+     PRUint64 suite;
+-    const ssl3CipherSuiteDef *suiteDef;
++    const ssl3CipherSuiteDef *suiteDef = NULL;

Although this looks OK at first sight, it's something that might be dangerous
(i.e. the wrong fix). Since this is a security package, I don't want to apply it until upstream gives feedback.

On second thought, since we don't enable -Werror in Buildroot, we don't need
this patch AFAICS. So it's good that you reported it upstream, but we're not going to apply this patch. I've marked it as Rejected.

Regards,
Arnout

#4 lore raw

PATCH 2/4 package/libnss: fix build warning

One more :-)

Quoted 48 lines · Arnout Vandecappelle
Quoted 36 lines · Arnout Vandecappelle
Quoted 33 lines · Giulio Benetti

Add patch to fix build warning due to uninitialized variable.

---
https://bugzilla.mozilla.org/show_bug.cgi?id=1590678

---
 ...ve-Wmaybe-uninitialized-warning-in-t.patch | 27 +++++++++++++++++++
 1 file changed, 27 insertions(+)
 create mode 100644 package/libnss/0004-Bug-1590678-Remove-Wmaybe-uninitialized-warning-in-t.patch
package/libnss/0004-Bug-1590678-Remove-Wmaybe-uninitialized-warning-in-t.patch new file +17 -0
new file mode 100644
index 0000000000..956abeb4a5
@@ -0,0 +1,27 @@
+From 24bcc8860310149da1524dbf25f3bc77f6476b11 Mon Sep 17 00:00:00 2001
+From: Giulio Benetti <[email protected]>
+Date: Wed, 23 Oct 2019 11:58:12 +0200
+Subject: [PATCH] Bug 1590678 - Remove -Wmaybe-uninitialized warning in
+ tls13esni.c
+
+Signed-off-by: Giulio Benetti <[email protected]>
+---
+ nss/lib/ssl/tls13esni.c | 2 +-
+ 1 file changed, 1 insertion(+), 1 deletion(-)
+
+diff --git a/nss/lib/ssl/tls13esni.c b/nss/lib/ssl/tls13esni.c
+index 4d2e12d62..a7ce6f568 100644
+--- a/nss/lib/ssl/tls13esni.c
++++ b/nss/lib/ssl/tls13esni.c
+@@ -728,7 +728,7 @@ tls13_ServerDecryptEsniXtn(const sslSocket *ss, const PRUint8 *in, unsigned int
+ {
Quoted 8 lines

package/libnss/0004-Bug-1590678-Remove-Wmaybe-uninitialized-warning-in-t.patch line 21

Quoted 4 lines
package/libnss/0004-Bug-1590678-Remove-Wmaybe-uninitialized-warning-in-t.patch new file +4 -0
+     sslReader rdr = SSL_READER(in, inLen);
+     PRUint64 suite;
+-    const ssl3CipherSuiteDef *suiteDef;
++    const ssl3CipherSuiteDef *suiteDef = NULL;

Although this looks OK at first sight, it's something that might be dangerous
(i.e. the wrong fix). Since this is a security package, I don't want to apply it until upstream gives feedback.

Quoted 3 lines

On second thought, since we don't enable -Werror in Buildroot, we don't need
this patch AFAICS. So it's good that you reported it upstream, but we're not going to apply this patch. I've marked it as Rejected.

If upstream decides that this really *is* an uninitialized variable reference
and not just a limitation of the compiler's analysis, it's a potential security issue, so in that case please resubmit.

Regards,
Arnout

#5 lore raw

PATCH 2/4 package/libnss: fix build warning

Hi Arnout,

Quoted 51 lines · Arnout Vandecappelle

One more :-)

Quoted 48 lines · Arnout Vandecappelle
Quoted 44 lines · Arnout Vandecappelle
Quoted 27 lines · Giulio Benetti

Add patch to fix build warning due to uninitialized variable.

---
https://bugzilla.mozilla.org/show_bug.cgi?id=1590678

---
  ...ve-Wmaybe-uninitialized-warning-in-t.patch | 27 +++++++++++++++++++
  1 file changed, 27 insertions(+)

create mode 100644 package/libnss/0004-Bug-1590678-Remove-Wmaybe-uninitialized-warning-in-t.patch

package/libnss/0004-Bug-1590678-Remove-Wmaybe-uninitialized-warning-in-t.patch new file +11 -0
new file mode 100644
index 0000000000..956abeb4a5
@@ -0,0 +1,27 @@
+From 24bcc8860310149da1524dbf25f3bc77f6476b11 Mon Sep 17 00:00:00 2001
+From: Giulio Benetti <[email protected]>
+Date: Wed, 23 Oct 2019 11:58:12 +0200
+Subject: [PATCH] Bug 1590678 - Remove -Wmaybe-uninitialized warning in
+ tls13esni.c
+
+Signed-off-by: Giulio Benetti <[email protected]>
+---
+ nss/lib/ssl/tls13esni.c | 2 +-
+ 1 file changed, 1 insertion(+), 1 deletion(-)
+

package/libnss/0004-Bug-1590678-Remove-Wmaybe-uninitialized-warning-in-t.patch line 21

Quoted 10 lines
package/libnss/0004-Bug-1590678-Remove-Wmaybe-uninitialized-warning-in-t.patch new file +10 -0
+diff --git a/nss/lib/ssl/tls13esni.c b/nss/lib/ssl/tls13esni.c
+index 4d2e12d62..a7ce6f568 100644
+--- a/nss/lib/ssl/tls13esni.c
++++ b/nss/lib/ssl/tls13esni.c
+@@ -728,7 +728,7 @@ tls13_ServerDecryptEsniXtn(const sslSocket *ss, const PRUint8 *in, unsigned int
+ {
+     sslReader rdr = SSL_READER(in, inLen);
+     PRUint64 suite;
+-    const ssl3CipherSuiteDef *suiteDef;
++    const ssl3CipherSuiteDef *suiteDef = NULL;

Although this looks OK at first sight, it's something that might be dangerous
(i.e. the wrong fix). Since this is a security package, I don't want to apply it until upstream gives feedback.

Quoted 7 lines
Quoted 3 lines

On second thought, since we don't enable -Werror in Buildroot, we don't need
this patch AFAICS. So it's good that you reported it upstream, but we're not going to apply this patch. I've marked it as Rejected.

If upstream decides that this really *is* an uninitialized variable reference
and not just a limitation of the compiler's analysis, it's a potential security issue, so in that case please resubmit.

Ok then,

thanks for reviewing!
Best regards

Signature · 2 lines
Giulio Benetti
Benetti Engineering sas
#6 lore raw

PATCH 3/4 package/libnss: remove useless NSS_ENABLE_ECC variable

In reply to Giulio Benetti

NSS_ENABLE_ECC is not supported anymore, then remove it from our libnss.mk file.

---
 package/libnss/libnss.mk | 2 --
 1 file changed, 2 deletions(-)
package/libnss/libnss.mk +0 -2
index 472936db89..ca1cde4c1c 100644
@@ -50,7 +50,6 @@ LIBNSS_BUILD_VARS = \
 	NS_USE_GCC=1 \
 	NSS_DISABLE_GTESTS=1 \
 	NSS_USE_SYSTEM_SQLITE=1 \
-	NSS_ENABLE_ECC=1 \
 	NATIVE_CC="$(HOSTCC)" \
 	OS_ARCH="Linux" \
 	OS_RELEASE="2.6" \
@@ -121,7 +120,6 @@ HOST_LIBNSS_BUILD_VARS = \
 	NSS_USE_SYSTEM_SQLITE=1 \
 	SQLITE_INCLUDE_DIR=$(HOST_DIR)/include \
 	ZLIB_INCLUDE_DIR=$(HOST_DIR)/include \
-	NSS_ENABLE_ECC=1 \
 	NSS_ENABLE_WERROR=0
 
 HOST_LIBNSS_DEPENDENCIES = host-libnspr host-sqlite host-zlib
Signature · 1 line
2.20.1
#7 lore raw

PATCH 3/4 package/libnss: remove useless NSS_ENABLE_ECC variable

Quoted 4 lines · Giulio Benetti

NSS_ENABLE_ECC is not supported anymore, then remove it from our libnss.mk file.

Applied to master, thanks.

Regards,
Arnout

Quoted 25 lines
---
 package/libnss/libnss.mk | 2 --
 1 file changed, 2 deletions(-)
package/libnss/libnss.mk +0 -2
index 472936db89..ca1cde4c1c 100644
@@ -50,7 +50,6 @@ LIBNSS_BUILD_VARS = \
 	NS_USE_GCC=1 \
 	NSS_DISABLE_GTESTS=1 \
 	NSS_USE_SYSTEM_SQLITE=1 \
-	NSS_ENABLE_ECC=1 \
 	NATIVE_CC="$(HOSTCC)" \
 	OS_ARCH="Linux" \
 	OS_RELEASE="2.6" \
@@ -121,7 +120,6 @@ HOST_LIBNSS_BUILD_VARS = \
 	NSS_USE_SYSTEM_SQLITE=1 \
 	SQLITE_INCLUDE_DIR=$(HOST_DIR)/include \
 	ZLIB_INCLUDE_DIR=$(HOST_DIR)/include \
-	NSS_ENABLE_ECC=1 \
 	NSS_ENABLE_WERROR=0
 
 HOST_LIBNSS_DEPENDENCIES = host-libnspr host-sqlite host-zlib
#8 lore raw

PATCH 4/4 package/libnss: remove NSS_ENABLE_WERROR=0

In reply to Giulio Benetti

All warnings have been fixed now, so let's remove NSS_ENABLE_WERROR and leave any warning to show up.

---
 package/libnss/libnss.mk | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)
package/libnss/libnss.mk +2 -4
index ca1cde4c1c..e1df7bbf33 100644
@@ -53,8 +53,7 @@ LIBNSS_BUILD_VARS = \
 	NATIVE_CC="$(HOSTCC)" \
 	OS_ARCH="Linux" \
 	OS_RELEASE="2.6" \
-	OS_TEST="$(LIBNSS_ARCH)" \
-	NSS_ENABLE_WERROR=0
+	OS_TEST="$(LIBNSS_ARCH)"
 
 # #pragma usage needs gcc >= 4.8
 # See https://bugzilla.mozilla.org/show_bug.cgi?id=1226179
@@ -119,8 +118,7 @@ HOST_LIBNSS_BUILD_VARS = \
 	NSS_DISABLE_GTESTS=1 \
 	NSS_USE_SYSTEM_SQLITE=1 \
 	SQLITE_INCLUDE_DIR=$(HOST_DIR)/include \
-	ZLIB_INCLUDE_DIR=$(HOST_DIR)/include \
-	NSS_ENABLE_WERROR=0
+	ZLIB_INCLUDE_DIR=$(HOST_DIR)/include
 
 HOST_LIBNSS_DEPENDENCIES = host-libnspr host-sqlite host-zlib
 
Signature · 1 line
2.20.1
#9 lore raw

PATCH 4/4 package/libnss: remove NSS_ENABLE_WERROR=0

Quoted 4 lines · On Wed, 23 Oct 2019 12:23:30 +0200 Giulio Benetti…

All warnings have been fixed now, so let's remove NSS_ENABLE_WERROR and leave any warning to show up.

No, we generally don't want to enable -Werror in Buildroot builds, as warnings keep popping up when upgrading gcc versions.

Thomas

Signature · 3 lines
Thomas Petazzoni, CTO, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com
#10 lore raw

PATCH 4/4 package/libnss: remove NSS_ENABLE_WERROR=0

Hello Thomas,

Quoted 10 lines · Thomas Petazzoni
Quoted 4 lines · On Wed, 23 Oct 2019 12:23:30 +0200 Giulio Benetti…

All warnings have been fixed now, so let's remove NSS_ENABLE_WERROR and leave any warning to show up.

No, we generally don't want to enable -Werror in Buildroot builds, as warnings keep popping up when upgrading gcc versions.

Ok, then I'll fix warnings upstream and keep -Werror disabled.

Thanks for reviewing
Best regards

Signature · 2 lines
Giulio Benetti
Benetti Engineering sas
#11 lore raw

PATCH 1/4 package/libnss: fix build failure with ARM without NEON extension

In reply to Giulio Benetti
Quoted 5 lines · Giulio Benetti

At the moment libnss assumes that every ARM has NEON extension but it's not that way. So add a patch to make it aware of it and use native functions in place of NEON optimized ones.

You forgot:

Fixes:
http://autobuild.buildroot.net/results/1342d305d1aeebef7af54a83afc094fda12421e2

Quoted 24 lines

---
Already sent upstream:
https://bugzilla.mozilla.org/show_bug.cgi?id=1590676

---
 ...ix-build-if-arm-doesn-t-support-NEON.patch | 47 +++++++++++++++++++
 1 file changed, 47 insertions(+)
 create mode 100644 package/libnss/0003-Bug-1590676-Fix-build-if-arm-doesn-t-support-NEON.patch
package/libnss/0003-Bug-1590676-Fix-build-if-arm-doesn-t-support-NEON.patch new file +10 -0
new file mode 100644
index 0000000000..e536282371
@@ -0,0 +1,47 @@
+From 946bc2864b0de7b63a4ec415e42e622d591a8753 Mon Sep 17 00:00:00 2001
+From: Giulio Benetti <[email protected]>
+Date: Wed, 23 Oct 2019 11:47:03 +0200
+Subject: [PATCH] Bug 1590676 - Fix build if arm doesn't support NEON
+
+At the moment NSS assumes that ARM supports NEON extension but this is
+not true and leads to build failure on ARM without NEON extension.
+Add check to assure USE_HW_AES is not defined if ARM without NEON
+extension is used.
+

package/libnss/0003-Bug-1590676-Fix-build-if-arm-doesn-t-support-NEON.patch line 20

Quoted 10 lines
package/libnss/0003-Bug-1590676-Fix-build-if-arm-doesn-t-support-NEON.patch new file +10 -0
+Signed-off-by: Giulio Benetti <[email protected]>
+---
+ nss/lib/freebl/aes-armv8.c | 3 ++-
+ nss/lib/freebl/rijndael.c  | 3 ++-
+ 2 files changed, 4 insertions(+), 2 deletions(-)
+
+diff --git a/nss/lib/freebl/aes-armv8.c b/nss/lib/freebl/aes-armv8.c
+index 40d5e2d34..a5df53891 100644
+--- a/nss/lib/freebl/aes-armv8.c
++++ b/nss/lib/freebl/aes-armv8.c

That looks as if it is something that should be built only on armv8... Since
the build failure is in fact for an armv4, don't we have a bigger issue here?

I believe 32-bit armv8 always has neon.

package/libnss/0003-Bug-1590676-Fix-build-if-arm-doesn-t-support-NEON.patch line 27

Quoted 7 lines
package/libnss/0003-Bug-1590676-Fix-build-if-arm-doesn-t-support-NEON.patch new file +7 -0
+@@ -7,7 +7,8 @@
+ 
+ #if (defined(__clang__) ||                            \
+      (defined(__GNUC__) && defined(__GNUC_MINOR__) && \
+-      (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ > 8))))
++      (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ > 8))) && \
++      (defined(__ARM_NEON) || defined(__ARM_NEON__)))

That parenthesis matching doesn't look right... I suppose with clang it doesn't
magically support neon, right? So there should be an additional set of parenthesis around the defined(__clang__) ... > 8 part.

Regards,
Arnout

Quoted 21 lines
package/libnss/0003-Bug-1590676-Fix-build-if-arm-doesn-t-support-NEON.patch new file +20 -0
+ 
+ #ifndef __ARM_FEATURE_CRYPTO
+ #error "Compiler option is invalid"
+diff --git a/nss/lib/freebl/rijndael.c b/nss/lib/freebl/rijndael.c
+index 26bd58ee0..23893f419 100644
+--- a/nss/lib/freebl/rijndael.c
++++ b/nss/lib/freebl/rijndael.c
+@@ -20,7 +20,8 @@
+ #include "gcm.h"
+ #include "mpi.h"
+ 
+-#if !defined(IS_LITTLE_ENDIAN) && !defined(NSS_X86_OR_X64)
++#if (!defined(IS_LITTLE_ENDIAN) && !defined(NSS_X86_OR_X64)) || \
++    (!defined(__ARM_NEON) && !defined(__ARM_NEON__))
+ // not test yet on big endian platform of arm
+ #undef USE_HW_AES
+ #endif
+-- 
+2.20.1
+
#12 lore raw

PATCH 1/4 package/libnss: fix build failure with ARM without NEON extension

Hi Arnout,

Quoted 7 lines · Arnout Vandecappelle
Quoted 4 lines · Giulio Benetti

At the moment libnss assumes that every ARM has NEON extension but it's not that way. So add a patch to make it aware of it and use native functions in place of NEON optimized ones.

Quoted 6 lines
Quoted 1 line

You forgot:

Fixes:
http://autobuild.buildroot.net/results/1342d305d1aeebef7af54a83afc094fda12421e2

Oops, yes.

Quoted 38 lines
Quoted 24 lines

---
Already sent upstream:
https://bugzilla.mozilla.org/show_bug.cgi?id=1590676

---
  ...ix-build-if-arm-doesn-t-support-NEON.patch | 47 +++++++++++++++++++
  1 file changed, 47 insertions(+)

create mode 100644 package/libnss/0003-Bug-1590676-Fix-build-if-arm-doesn-t-support-NEON.patch

package/libnss/0003-Bug-1590676-Fix-build-if-arm-doesn-t-support-NEON.patch new file +10 -0
new file mode 100644
index 0000000000..e536282371
@@ -0,0 +1,47 @@
+From 946bc2864b0de7b63a4ec415e42e622d591a8753 Mon Sep 17 00:00:00 2001
+From: Giulio Benetti <[email protected]>
+Date: Wed, 23 Oct 2019 11:47:03 +0200
+Subject: [PATCH] Bug 1590676 - Fix build if arm doesn't support NEON
+
+At the moment NSS assumes that ARM supports NEON extension but this is
+not true and leads to build failure on ARM without NEON extension.
+Add check to assure USE_HW_AES is not defined if ARM without NEON
+extension is used.
+

package/libnss/0003-Bug-1590676-Fix-build-if-arm-doesn-t-support-NEON.patch line 20

Quoted 10 lines
package/libnss/0003-Bug-1590676-Fix-build-if-arm-doesn-t-support-NEON.patch new file +10 -0
+Signed-off-by: Giulio Benetti <[email protected]>
+---
+ nss/lib/freebl/aes-armv8.c | 3 ++-
+ nss/lib/freebl/rijndael.c  | 3 ++-
+ 2 files changed, 4 insertions(+), 2 deletions(-)
+
+diff --git a/nss/lib/freebl/aes-armv8.c b/nss/lib/freebl/aes-armv8.c
+index 40d5e2d34..a5df53891 100644
+--- a/nss/lib/freebl/aes-armv8.c
++++ b/nss/lib/freebl/aes-armv8.c

That looks as if it is something that should be built only on armv8... Since
the build failure is in fact for an armv4, don't we have a bigger issue here?

Quoted 1 line

I believe 32-bit armv8 always has neon.

Would it be a good idea to provide a variable to enable or not building aes-armv8 and define or not USE_HW_AES? I mean, this is a Makefile and I don't know how to make libnss capable of understand if there is NEON support or not, do you have any suggestions?

Quoted 11 lines

package/libnss/0003-Bug-1590676-Fix-build-if-arm-doesn-t-support-NEON.patch line 27

Quoted 7 lines
package/libnss/0003-Bug-1590676-Fix-build-if-arm-doesn-t-support-NEON.patch new file +7 -0
+@@ -7,7 +7,8 @@
+
+ #if (defined(__clang__) ||                            \
+      (defined(__GNUC__) && defined(__GNUC_MINOR__) && \
+-      (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ > 8))))
++      (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ > 8))) && \
++      (defined(__ARM_NEON) || defined(__ARM_NEON__)))

That parenthesis matching doesn't look right... I suppose with clang it doesn't
magically support neon, right? So there should be an additional set of parenthesis around the defined(__clang__) ... > 8 part.

Oops again, you're right.

Thanks for reviewing
Best regards

Signature · 2 lines
Giulio Benetti
Benetti Engineering sas
#13 lore raw

PATCH v2 1/1 package/libnss: fix build failure with ARM without NEON extension

In reply to Arnout Vandecappelle

At the moment libnss assumes that every ARM has NEON extension but it's not that way. So add a patch to make it aware of it and use native functions in place of NEON optimized ones.

Fixes:
http://autobuild.buildroot.net/results/1342d305d1aeebef7af54a83afc094fda12421e2/

---
V1->V2:
* added Fixes to commit log
* fixed #ifdef statement in patch

---
 ...ix-build-if-arm-doesn-t-support-NEON.patch | 50 +++++++++++++++++++
 1 file changed, 50 insertions(+)
 create mode 100644 package/libnss/0003-Bug-1590676-Fix-build-if-arm-doesn-t-support-NEON.patch
package/libnss/0003-Bug-1590676-Fix-build-if-arm-doesn-t-support-NEON.patch new file +50 -0
new file mode 100644
index 0000000000..1d7f63611e
@@ -0,0 +1,50 @@
+From 76f35f9acd6c6b34318438bef30b0bdfdfd13f16 Mon Sep 17 00:00:00 2001
+From: Giulio Benetti <[email protected]>
+Date: Wed, 23 Oct 2019 11:47:03 +0200
+Subject: [PATCH] Bug 1590676 - Fix build if arm doesn't support NEON
+
+At the moment NSS assumes that ARM supports NEON extension but this is
+not true and leads to build failure on ARM without NEON extension.
+Add check to assure USE_HW_AES is not defined if ARM without NEON
+extension is used.
+
+Signed-off-by: Giulio Benetti <[email protected]>
+---
+ nss/lib/freebl/aes-armv8.c | 5 +++--
+ nss/lib/freebl/rijndael.c  | 3 ++-
+ 2 files changed, 5 insertions(+), 3 deletions(-)
+
+diff --git a/nss/lib/freebl/aes-armv8.c b/nss/lib/freebl/aes-armv8.c
+index 40d5e2d34..d1504e6fb 100644
+--- a/nss/lib/freebl/aes-armv8.c
++++ b/nss/lib/freebl/aes-armv8.c
+@@ -5,9 +5,10 @@
+ #include "secerr.h"
+ #include "rijndael.h"
+ 
+-#if (defined(__clang__) ||                            \
++#if ((defined(__clang__) ||                            \
+      (defined(__GNUC__) && defined(__GNUC_MINOR__) && \
+-      (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ > 8))))
++      (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ > 8)))) && \
++      (defined(__ARM_NEON) || defined(__ARM_NEON__)))
+ 
+ #ifndef __ARM_FEATURE_CRYPTO
+ #error "Compiler option is invalid"
+diff --git a/nss/lib/freebl/rijndael.c b/nss/lib/freebl/rijndael.c
+index 26bd58ee0..23893f419 100644
+--- a/nss/lib/freebl/rijndael.c
++++ b/nss/lib/freebl/rijndael.c
+@@ -20,7 +20,8 @@
+ #include "gcm.h"
+ #include "mpi.h"
+ 
+-#if !defined(IS_LITTLE_ENDIAN) && !defined(NSS_X86_OR_X64)
++#if (!defined(IS_LITTLE_ENDIAN) && !defined(NSS_X86_OR_X64)) || \
++    (!defined(__ARM_NEON) && !defined(__ARM_NEON__))
+ // not test yet on big endian platform of arm
+ #undef USE_HW_AES
+ #endif
+-- 
+2.20.1
+
Signature · 1 line
2.20.1
#14 lore raw

PATCH v2 1/1 package/libnss: fix build failure with ARM without NEON extension

Quoted 57 lines · Giulio Benetti

At the moment libnss assumes that every ARM has NEON extension but it's not that way. So add a patch to make it aware of it and use native functions in place of NEON optimized ones.

Fixes:
http://autobuild.buildroot.net/results/1342d305d1aeebef7af54a83afc094fda12421e2/

---
V1->V2:
* added Fixes to commit log
* fixed #ifdef statement in patch

---
  ...ix-build-if-arm-doesn-t-support-NEON.patch | 50 +++++++++++++++++++
  1 file changed, 50 insertions(+)

create mode 100644 package/libnss/0003-Bug-1590676-Fix-build-if-arm-doesn-t-support-NEON.patch

package/libnss/0003-Bug-1590676-Fix-build-if-arm-doesn-t-support-NEON.patch new file +34 -0
new file mode 100644
index 0000000000..1d7f63611e
@@ -0,0 +1,50 @@
+From 76f35f9acd6c6b34318438bef30b0bdfdfd13f16 Mon Sep 17 00:00:00 2001
+From: Giulio Benetti <[email protected]>
+Date: Wed, 23 Oct 2019 11:47:03 +0200
+Subject: [PATCH] Bug 1590676 - Fix build if arm doesn't support NEON
+
+At the moment NSS assumes that ARM supports NEON extension but this is
+not true and leads to build failure on ARM without NEON extension.
+Add check to assure USE_HW_AES is not defined if ARM without NEON
+extension is used.
+
+Signed-off-by: Giulio Benetti <[email protected]>
+---
+ nss/lib/freebl/aes-armv8.c | 5 +++--
+ nss/lib/freebl/rijndael.c  | 3 ++-
+ 2 files changed, 5 insertions(+), 3 deletions(-)
+
+diff --git a/nss/lib/freebl/aes-armv8.c b/nss/lib/freebl/aes-armv8.c
+index 40d5e2d34..d1504e6fb 100644
+--- a/nss/lib/freebl/aes-armv8.c
++++ b/nss/lib/freebl/aes-armv8.c
+@@ -5,9 +5,10 @@
+ #include "secerr.h"
+ #include "rijndael.h"
+
+-#if (defined(__clang__) ||                            \
++#if ((defined(__clang__) ||                            \
+      (defined(__GNUC__) && defined(__GNUC_MINOR__) && \
+-      (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ > 8))))
++      (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ > 8)))) && \
++      (defined(__ARM_NEON) || defined(__ARM_NEON__)))
+
+ #ifndef __ARM_FEATURE_CRYPTO
+ #error "Compiler option is invalid"
+diff --git a/nss/lib/freebl/rijndael.c b/nss/lib/freebl/rijndael.c

package/libnss/0003-Bug-1590676-Fix-build-if-arm-doesn-t-support-NEON.patch line 44

Quoted 10 lines
package/libnss/0003-Bug-1590676-Fix-build-if-arm-doesn-t-support-NEON.patch new file +10 -0
+index 26bd58ee0..23893f419 100644
+--- a/nss/lib/freebl/rijndael.c
++++ b/nss/lib/freebl/rijndael.c
+@@ -20,7 +20,8 @@
+ #include "gcm.h"
+ #include "mpi.h"
+
+-#if !defined(IS_LITTLE_ENDIAN) && !defined(NSS_X86_OR_X64)
++#if (!defined(IS_LITTLE_ENDIAN) && !defined(NSS_X86_OR_X64)) || \
++    (!defined(__ARM_NEON) && !defined(__ARM_NEON__))

This part is wrong, I've sent upstream corrected patch[1] and I'm going to send v3 patch for buildroot.

[1]: https://bugzilla.mozilla.org/show_bug.cgi?id=1590676

Quoted 7 lines
package/libnss/0003-Bug-1590676-Fix-build-if-arm-doesn-t-support-NEON.patch new file +6 -0
+ // not test yet on big endian platform of arm
+ #undef USE_HW_AES
+ #endif
+--
+2.20.1
+
Signature · 2 lines
Giulio Benetti
Benetti Engineering sas