diff --git a/snappy-stubs-internal.h b/snappy-stubs-internal.h
index f834bdb..22407ef 100644
--- a/snappy-stubs-internal.h
+++ b/snappy-stubs-internal.h
@@ -123,7 +123,7 @@ static const int64 kint64max = static_cast<int64>(0x7FFFFFFFFFFFFFFFLL);
 // x86, PowerPC, and ARM64 can simply do these loads and stores native.
 
 #if defined(__i386__) || defined(__x86_64__) || defined(__powerpc__) || \
-    defined(__aarch64__)
+    defined(__aarch64__) || defined(__s390x__)
 
 #define UNALIGNED_LOAD16(_p) (*reinterpret_cast<const uint16 *>(_p))
 #define UNALIGNED_LOAD32(_p) (*reinterpret_cast<const uint32 *>(_p))
@@ -150,6 +150,8 @@ static const int64 kint64max = static_cast<int64>(0x7FFFFFFFFFFFFFFFLL);
 // (it ignores __attribute__((packed)) on individual variables). However,
 // we can tell it that a _struct_ is unaligned, which has the same effect,
 // so we do that.
+//
+// On pre-R6 MIPS just let the compiler use LWL/LWR, SWL/SWR etc.
 
 #elif defined(__arm__) && \
       !defined(__ARM_ARCH_4__) && \
@@ -163,7 +165,8 @@ static const int64 kint64max = static_cast<int64>(0x7FFFFFFFFFFFFFFFLL);
       !defined(__ARM_ARCH_6K__) && \
       !defined(__ARM_ARCH_6Z__) && \
       !defined(__ARM_ARCH_6ZK__) && \
-      !defined(__ARM_ARCH_6T2__)
+      !defined(__ARM_ARCH_6T2__) || \
+      (defined(__mips__) && (!defined(__mips_isa_rev) || __mips_isa_rev < 6))
 
 #if __GNUC__
 #define ATTRIBUTE_PACKED __attribute__((__packed__))
@@ -184,6 +187,11 @@ struct Unaligned32Struct {
   uint8 dummy;  // To make the size non-power-of-two.
 } ATTRIBUTE_PACKED;
 
+struct Unaligned64Struct {
+  uint64 value;
+  uint8 dummy;  // To make the size non-power-of-two.
+} ATTRIBUTE_PACKED;
+
 }  // namespace internal
 }  // namespace base
 
@@ -203,6 +211,7 @@ struct Unaligned32Struct {
 // See if that would be more efficient on platforms supporting it,
 // at least for copies.
 
+#ifndef __mips__
 inline uint64 UNALIGNED_LOAD64(const void *p) {
   uint64 t;
   memcpy(&t, p, sizeof t);
@@ -212,6 +221,13 @@ inline uint64 UNALIGNED_LOAD64(const void *p) {
 inline void UNALIGNED_STORE64(void *p, uint64 v) {
   memcpy(p, &v, sizeof v);
 }
+#else
+#define UNALIGNED_LOAD64(_p) \
+    ((reinterpret_cast<const ::snappy::base::internal::Unaligned64Struct *>(_p))->value)
+#define UNALIGNED_STORE64(_p, _val) \
+    ((reinterpret_cast< ::snappy::base::internal::Unaligned64Struct *>(_p))->value = \
+         (_val))
+#endif
 
 #else
 
@@ -274,6 +290,13 @@ inline void UNALIGNED_STORE64(void *p, uint64 v) {
 #define bswap_32(x) OSSwapInt32(x)
 #define bswap_64(x) OSSwapInt64(x)
 
+#elif defined(__GNUC__)
+/* musl provides suboptimal (generic) bswap_xx implementations in <byteswap.h> */
+
+#define bswap_16	__builtin_bswap16
+#define bswap_32	__builtin_bswap32
+#define bswap_64	__builtin_bswap64
+
 #elif defined(HAVE_BYTESWAP_H)
 #include <byteswap.h>