armv4cpuid.S: fix compilation error in pre-ARMv7 build.

PR: 3474
Reviewed-by: Kurt Roeckx <kurt@openssl.org>
(cherry picked from commit 6696203963ba209fbbef12098e6eb6f364379c65)

diff --git a/crypto/armv4cpuid.S b/crypto/armv4cpuid.S
index add11d405ef994497dc49b4e90f76f54a6e5cb9c..0059311303ac9f1dedebc5e3a7c920822eb1b8c7 100644 (file)
--- a/crypto/armv4cpuid.S
+++ b/crypto/armv4cpuid.S
@@ -3,6 +3,25 @@
 .text
 .code  32
 
+@ Special note about using .byte directives to encode instructions.
+@ Initial reason for hand-coding instructions was to allow module to
+@ be compilable by legacy tool-chains. At later point it was pointed
+@ out that since ARMv7, instructions are always encoded in little-endian
+@ order, therefore one has to opt for endian-neutral presentation.
+@ Contemporary tool-chains offer .inst directive for this purpose,
+@ but not legacy ones. Therefore .byte. But there is an exception,
+@ namely ARMv7-R profile still allows for big-endian encoding even for
+@ instructions. This raises the question what if probe instructions
+@ appear executable to such processor operating in big-endian order?
+@ They have to be chosen in a way that avoids this problem. As failed
+@ NEON probe disables a number of other probes we have to ensure that
+@ only NEON probe instruction doesn't appear executable in big-endian
+@ order, therefore 'vorr q8,q8,q8', and not some other register. The
+@ only probe that is not bypassed on failed NEON probe is _armv7_tick,
+@ where you'll spot 'mov r0,r6' that serves this purpose. Basic idea is
+@ that if fetched in alternative byte oder instruction should crash to
+@ denote lack of probed capability...
+
 .align 5
 .global        _armv7_neon_probe
 .type  _armv7_neon_probe,%function
@@ -14,12 +33,10 @@ _armv7_neon_probe:
 .global        _armv7_tick
 .type  _armv7_tick,%function
 _armv7_tick:
-       mrrc    p15,1,r0,r1,c14         @ CNTVCT
-#if __ARM_ARCH__>=5
-       bx      lr
-#else
-       .word   0xe12fff1e              @ bx    lr
-#endif
+       .byte   0x06,0x00,0xa0,0xe1     @ mov   r0,r6
+       .byte   0x1e,0x0f,0x51,0xec     @ mrrc  p15,1,r0,r1,c14 @ CNTVCT
+       .byte   0x1e,0xff,0x2f,0xe1     @ bx    lr
+       nop
 .size  _armv7_tick,.-_armv7_tick
 
 .global        _armv8_aes_probe