Skip to contentSkip to conversation

package/wolfssl: enable ARMv8 hardware acceleration

PATCH New Sergio Prado 2 participants 2 messages Nov 25, 2018 – Nov 29, 2018

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 package/wolfssl: enable ARMv8 hardware acceleration

Enable hardware acceleration for ARMv8 targets.

When ARMv8 hardware acceleration is enabled without any additional flags, the build fails with the following messages:

/tmp/cciv7Oei.s: Assembler messages:
/tmp/cciv7Oei.s:580: Error: invalid addressing mode at operand 2 -- ld1 {v0.2d},[x0,256]' /tmp/cciv7Oei.s:616: Error: invalid addressing mode at operand 2 -- st1 {v0.2d},[x0,256]' /tmp/cciv7Oei.s:629: Error: invalid addressing mode at operand 2 -- ld1 {v0.2d},[x0,256]' /tmp/cciv7Oei.s:669: Error: invalid addressing mode at operand 2 -- st1 {v0.2d},[x0,256]' /tmp/cciv7Oei.s:1211: Error: invalid addressing mode at operand 2 -- ld1 {v16.2d},[x0,304]' /tmp/cciv7Oei.s:1368: Error: invalid addressing mode at operand 2 -- ld1 {v17.16b},[x19,304]' /tmp/cciv7Oei.s:1554: Error: invalid addressing mode at operand 2 -- ld1 {v16.2d},[x0,304]' /tmp/cciv7Oei.s:1719: Error: invalid addressing mode at operand 2 -- ld1 {v17.16b},[x19,304]' /tmp/cciv7Oei.s:1870: Error: invalid addressing mode at operand 2 -- ld1 {v16.2d},[x0,304]' /tmp/cciv7Oei.s:2043: Error: invalid addressing mode at operand 2 -- ld1 {v17.16b},[x19,304]' make[3]: *** [Makefile:3801: wolfcrypt/src/port/arm/src_libwolfssl_la-armv8-aes.lo] Error 1

This is because of some inline assembly in parts of the AES structure using the "m" constraint.

So lets use the flag -mstrict-align to prevent this error.

---
 package/wolfssl/wolfssl.mk | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)
package/wolfssl/wolfssl.mk +8 -1
index 68b638c3e2a8..ab5f5ea4550c 100644
@@ -30,7 +30,14 @@ else
 WOLFSSL_CONF_OPTS += --disable-sslv3
 endif
 
-# build fails when ARMv8 hardware acceleration is enabled
+# enable ARMv8 hardware acceleration (the flag -mstrict-align is
+# needed to prevent build errors caused by some inline assembly
+# in parts of the AES structure using the "m" constraint)
+ifeq ($(BR2_ARM_CPU_ARMV8A),y)
+WOLFSSL_CONF_OPTS += --enable-armasm
+WOLFSSL_CONF_ENV += CPPFLAGS="$(TARGET_CPPFLAGS) -mstrict-align"
+else
 WOLFSSL_CONF_OPTS += --disable-armasm
+endif
 
 $(eval $(autotools-package))
Signature · 1 line
2.17.1
#1 lore raw

PATCH 1/1 package/wolfssl: enable ARMv8 hardware acceleration

Hello,

package/wolfssl/wolfssl.mk else · line 38

Quoted 7 lines · Sergio Prado
package/wolfssl/wolfssl.mk +6 -1
-# build fails when ARMv8 hardware acceleration is enabled
+# enable ARMv8 hardware acceleration (the flag -mstrict-align is
+# needed to prevent build errors caused by some inline assembly
+# in parts of the AES structure using the "m" constraint)
+ifeq ($(BR2_ARM_CPU_ARMV8A),y)
+WOLFSSL_CONF_OPTS += --enable-armasm
+WOLFSSL_CONF_ENV += CPPFLAGS="$(TARGET_CPPFLAGS) -mstrict-align"

Actually, this doesn't work well when doing an ARM 32 bit configuration targeting Cortex-A53. Cortex-A53 is an ARMv8-A core, so BR2_ARM_CPU_ARMV8A=y, but the gcc for ARM 32 bits doesn't support -mstrict-align, causing a build failure.

However, the ARM optimization of wolfssl applies to both ARMv8 in 32 and 64 bits mode. So I've added another condition on BR2_aarch64=y around the -mstrict-align workaround.

Applied with this change. Thanks!

Thomas

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