from https://lists.busybox.net/pipermail/busybox/2024-September/090899.html

The ENABLE_SHA1_HWACCEL Kconfig symbol is meant to be archicture
agnostic, so can be enabled regardless of whether your build
architecture provides hardware acceleration or not.
 At the moment only
x86 implements this, so every piece of optimised code should be guarded
by both ENABLE_SHA1_HWACCEL and (__x86_64__ || __i386__).
 This is missing
at one place, so compiling for arm64 breaks when ENABLE_SHA1_HWACCEL is
enabled:
================================
libbb/hash_md5_sha.c: In function ‘sha1_end’:
libbb/hash_md5_sha.c:1316:28: error: ‘sha1_process_block64_shaNI’ undeclared (first use in this function); did you mean ‘sha1_process_block64’?

 1316 |   || ctx->process_block == sha1_process_block64_shaNI
      |                            ^~~~~~~~~~~~~~~~~~~~~~~~~~
      |                            sha1_process_block64
libbb/hash_md5_sha.c:1316:28: note: each undeclared identifier is reported only once for each function it appears in
make[1]: *** [scripts/Makefile.build:197: libbb/hash_md5_sha.o] Error 1
make: *** [Makefile:744: libbb] Error 2
================================

Add the missing guards around the call to sha1_process_block64_shaNI to
fix the build on other architectures with ENABLE_SHA1_HWACCEL enabled.

Change-Id: I40bba388422625f4230abf15a5de23e1fdc654fc
Signed-off-by: Andre Przywara <andre.przywara at arm.com>
---
 libbb/hash_md5_sha.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/libbb/hash_md5_sha.c b/libbb/hash_md5_sha.c
index 57a801459..75a61c32c 100644
--- a/libbb/hash_md5_sha.c
+++ b/libbb/hash_md5_sha.c
@@ -1313,7 +1313,9 @@ unsigned FAST_FUNC sha1_end(sha1_ctx_t *ctx, void *resbuf)
 	hash_size = 8;
 	if (ctx->process_block == sha1_process_block64
 #if ENABLE_SHA1_HWACCEL
+# if defined(__GNUC__) && (defined(__i386__) || defined(__x86_64__))
 	 || ctx->process_block == sha1_process_block64_shaNI
+# endif
 #endif
 	) {
 		hash_size = 5;
-- 
2.25.1