From: Adam Stylinski <kungfujesus06@gmail.com>
Date: Thu, 17 Mar 2022 02:52:44 +0000 (-0400)
Subject: Fix UBSAN's cry afoul
X-Git-Tag: 2.1.0-beta1~327
X-Git-Url: http://git.ipfire.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=1f87fac9d3812a4932ccf3d73a613c172866889c;p=thirdparty%2Fzlib-ng.git

Fix UBSAN's cry afoul

Technically, we weren't actually doing this the way C wants us to,
legally.  The zmemcpy's turn into NOPs for pretty much all > 0
optimization levels and this gets us defined behavior with the
sanitizer, putting the optimized load by arbitrary alignment into the
compiler's hands instead of ours.
---

diff --git a/arch/x86/chunkset_avx.c b/arch/x86/chunkset_avx.c
index 358869bc..ff5d3c61 100644
--- a/arch/x86/chunkset_avx.c
+++ b/arch/x86/chunkset_avx.c
@@ -20,15 +20,21 @@ static inline void chunkmemset_1(uint8_t *from, chunk_t *chunk) {
 }
 
 static inline void chunkmemset_2(uint8_t *from, chunk_t *chunk) {
-    *chunk = _mm256_set1_epi16(*(int16_t *)from);
+    int16_t tmp;
+    zmemcpy_2(&tmp, from);
+    *chunk = _mm256_set1_epi16(tmp);
 }
 
 static inline void chunkmemset_4(uint8_t *from, chunk_t *chunk) {
-    *chunk = _mm256_set1_epi32(*(int32_t *)from);
+    int32_t tmp;
+    zmemcpy_4(&tmp, from);
+    *chunk = _mm256_set1_epi32(tmp);
 }
 
 static inline void chunkmemset_8(uint8_t *from, chunk_t *chunk) {
-    *chunk = _mm256_set1_epi64x(*(int64_t *)from);
+    int64_t tmp;
+    zmemcpy_8(&tmp, from);
+    *chunk = _mm256_set1_epi64x(tmp);
 }
 
 static inline void loadchunk(uint8_t const *s, chunk_t *chunk) {
diff --git a/arch/x86/chunkset_sse2.c b/arch/x86/chunkset_sse2.c
index 59b86f15..778e4ca1 100644
--- a/arch/x86/chunkset_sse2.c
+++ b/arch/x86/chunkset_sse2.c
@@ -21,15 +21,21 @@ static inline void chunkmemset_1(uint8_t *from, chunk_t *chunk) {
 }
 
 static inline void chunkmemset_2(uint8_t *from, chunk_t *chunk) {
-    *chunk = _mm_set1_epi16(*(int16_t *)from);
+    int16_t tmp;
+    zmemcpy_2(&tmp, from);
+    *chunk = _mm_set1_epi16(tmp);
 }
 
 static inline void chunkmemset_4(uint8_t *from, chunk_t *chunk) {
-    *chunk = _mm_set1_epi32(*(int32_t *)from);
+    int32_t tmp;
+    zmemcpy_4(&tmp, from);
+    *chunk = _mm_set1_epi32(tmp);
 }
 
 static inline void chunkmemset_8(uint8_t *from, chunk_t *chunk) {
-    *chunk = _mm_set1_epi64x(*(int64_t *)from);
+    int64_t tmp;
+    zmemcpy_8(&tmp, from);
+    *chunk = _mm_set1_epi64x(tmp);
 }
 
 static inline void loadchunk(uint8_t const *s, chunk_t *chunk) {