diff --git a/benchmarks/property_benchmark.cpp b/benchmarks/property_benchmark.cpp
index 77814bf..ba54ed1 100644
--- a/benchmarks/property_benchmark.cpp
+++ b/benchmarks/property_benchmark.cpp
@@ -184,7 +184,7 @@
 
   size_t i = 0;
   while (state.KeepRunning()) {
-    pa.system_properties().Serial(pinfo[i]);
+    __system_property_serial(pinfo[i]);
     i = (i + 1) % nprops;
   }
 
diff --git a/libc/Android.bp b/libc/Android.bp
index c5ea4c5..cdbdf63 100644
--- a/libc/Android.bp
+++ b/libc/Android.bp
@@ -42,7 +42,6 @@
     "-Wno-deprecated-declarations",
     "-Wno-gcc-compat",
     "-Wframe-larger-than=2048",
-    "-Wimplicit-fallthrough",
 
     // Try to catch typical 32-bit assumptions that break with 64-bit pointers.
     "-Werror=pointer-to-int-cast",
@@ -1711,6 +1710,8 @@
             strip: {
                 keep_symbols_and_debug_frame: true,
             },
+
+            whole_static_libs: [ "libgcc_stripped" ],
         },
         arm64: {
             version_script: ":libc.arm64.map",
diff --git a/libc/arch-x86_64/string/cache.h b/libc/arch-x86_64/string/cache.h
index 38acc6e..4131509 100644
--- a/libc/arch-x86_64/string/cache.h
+++ b/libc/arch-x86_64/string/cache.h
@@ -28,9 +28,9 @@
 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
 
-/* Values are optimized for Silvermont */
-#define SHARED_CACHE_SIZE	(1024*1024)			/* Silvermont L2 Cache */
-#define DATA_CACHE_SIZE		(24*1024)			/* Silvermont L1 Data Cache */
+/* Values are optimized for Core Architecture */
+#define SHARED_CACHE_SIZE (4096*1024)  /* Core Architecture L2 Cache */
+#define DATA_CACHE_SIZE   (24*1024)    /* Core Architecture L1 Data Cache */
 
-#define SHARED_CACHE_SIZE_HALF	(SHARED_CACHE_SIZE / 2)
-#define DATA_CACHE_SIZE_HALF	(DATA_CACHE_SIZE / 2)
+#define SHARED_CACHE_SIZE_HALF (SHARED_CACHE_SIZE / 2)
+#define DATA_CACHE_SIZE_HALF   (DATA_CACHE_SIZE / 2)
diff --git a/libc/bionic/jemalloc.h b/libc/bionic/jemalloc.h
index ef77c9c..4ce51c0 100644
--- a/libc/bionic/jemalloc.h
+++ b/libc/bionic/jemalloc.h
@@ -29,7 +29,7 @@
 __BEGIN_DECLS
 
 void* je_aligned_alloc_wrapper(size_t, size_t);
-int je_iterate(uintptr_t, size_t, void (*)(uintptr_t, size_t, void*), void*);
+int je_malloc_iterate(uintptr_t, size_t, void (*)(uintptr_t, size_t, void*), void*);
 int je_mallctl(const char *name, void *oldp, size_t *oldlenp, void *newp, size_t newlen) __attribute__((nothrow));
 struct mallinfo je_mallinfo();
 void je_malloc_disable();
diff --git a/libc/bionic/malloc_common.cpp b/libc/bionic/malloc_common.cpp
index 96e6140..a0da3db 100644
--- a/libc/bionic/malloc_common.cpp
+++ b/libc/bionic/malloc_common.cpp
@@ -215,9 +215,9 @@
     void (*callback)(uintptr_t base, size_t size, void* arg), void* arg) {
   auto dispatch_table = GetDispatchTable();
   if (__predict_false(dispatch_table != nullptr)) {
-    return dispatch_table->iterate(base, size, callback, arg);
+    return dispatch_table->malloc_iterate(base, size, callback, arg);
   }
-  return Malloc(iterate)(base, size, callback, arg);
+  return Malloc(malloc_iterate)(base, size, callback, arg);
 }
 
 // Disable calls to malloc so malloc_iterate gets a consistent view of
@@ -247,9 +247,10 @@
 
 #if __has_feature(hwaddress_sanitizer)
 // FIXME: implement these in HWASan allocator.
-extern "C" int __sanitizer_iterate(uintptr_t base __unused, size_t size __unused,
-                                   void (*callback)(uintptr_t base, size_t size, void* arg) __unused,
-                                   void* arg __unused) {
+extern "C" int __sanitizer_malloc_iterate(uintptr_t base __unused, size_t size __unused,
+                                          void (*callback)(uintptr_t base, size_t size, void* arg)
+                                              __unused,
+                                          void* arg __unused) {
   return 0;
 }
 
diff --git a/libc/bionic/malloc_common.h b/libc/bionic/malloc_common.h
index 2176e63..89dccc3 100644
--- a/libc/bionic/malloc_common.h
+++ b/libc/bionic/malloc_common.h
@@ -42,9 +42,9 @@
 __BEGIN_DECLS
 
 // FIXME: implement these in HWASan allocator.
-int __sanitizer_iterate(uintptr_t base, size_t size,
-                        void (*callback)(uintptr_t base, size_t size, void* arg),
-                        void* arg);
+int __sanitizer_malloc_iterate(uintptr_t base, size_t size,
+                               void (*callback)(uintptr_t base, size_t size, void* arg),
+                               void* arg);
 void __sanitizer_malloc_disable();
 void __sanitizer_malloc_enable();
 int __sanitizer_malloc_info(int options, FILE* fp);
diff --git a/libc/bionic/malloc_common_dynamic.cpp b/libc/bionic/malloc_common_dynamic.cpp
index 9ad79a0..0ac3f62 100644
--- a/libc/bionic/malloc_common_dynamic.cpp
+++ b/libc/bionic/malloc_common_dynamic.cpp
@@ -95,7 +95,7 @@
 #if defined(HAVE_DEPRECATED_MALLOC_FUNCS)
     Malloc(valloc),
 #endif
-    Malloc(iterate),
+    Malloc(malloc_iterate),
     Malloc(malloc_disable),
     Malloc(malloc_enable),
     Malloc(mallopt),
@@ -184,7 +184,8 @@
   if (!InitMallocFunction<MallocRealloc>(impl_handler, &table->realloc, prefix, "realloc")) {
     return false;
   }
-  if (!InitMallocFunction<MallocIterate>(impl_handler, &table->iterate, prefix, "iterate")) {
+  if (!InitMallocFunction<MallocIterate>(impl_handler, &table->malloc_iterate, prefix,
+                                         "malloc_iterate")) {
     return false;
   }
   if (!InitMallocFunction<MallocMallocDisable>(impl_handler, &table->malloc_disable, prefix,
diff --git a/libc/bionic/malloc_heapprofd.cpp b/libc/bionic/malloc_heapprofd.cpp
index 62249fb..5860222 100644
--- a/libc/bionic/malloc_heapprofd.cpp
+++ b/libc/bionic/malloc_heapprofd.cpp
@@ -104,7 +104,7 @@
 #if defined(HAVE_DEPRECATED_MALLOC_FUNCS)
     Malloc(valloc),
 #endif
-    Malloc(iterate),
+    Malloc(malloc_iterate),
     Malloc(malloc_disable),
     Malloc(malloc_enable),
     Malloc(mallopt),
diff --git a/libc/bionic/malloc_limit.cpp b/libc/bionic/malloc_limit.cpp
index 6a67cae..d8c0ebe 100644
--- a/libc/bionic/malloc_limit.cpp
+++ b/libc/bionic/malloc_limit.cpp
@@ -350,9 +350,9 @@
 static int LimitIterate(uintptr_t base, size_t size, void (*callback)(uintptr_t, size_t, void*), void* arg) {
   auto dispatch_table = GetDefaultDispatchTable();
   if (__predict_false(dispatch_table != nullptr)) {
-    return dispatch_table->iterate(base, size, callback, arg);
+    return dispatch_table->malloc_iterate(base, size, callback, arg);
   }
-  return Malloc(iterate)(base, size, callback, arg);
+  return Malloc(malloc_iterate)(base, size, callback, arg);
 }
 
 static void LimitMallocDisable() {
diff --git a/libc/bionic/scudo.h b/libc/bionic/scudo.h
index d9933c4..a80d754 100644
--- a/libc/bionic/scudo.h
+++ b/libc/bionic/scudo.h
@@ -52,7 +52,7 @@
 void* scudo_valloc(size_t);
 #endif
 
-int scudo_iterate(uintptr_t, size_t, void (*)(uintptr_t, size_t, void*), void*);
+int scudo_malloc_iterate(uintptr_t, size_t, void (*)(uintptr_t, size_t, void*), void*);
 void scudo_malloc_disable();
 void scudo_malloc_enable();
 
diff --git a/libc/bionic/scudo_wrapper.cpp b/libc/bionic/scudo_wrapper.cpp
index e17f20b..debd8d9 100644
--- a/libc/bionic/scudo_wrapper.cpp
+++ b/libc/bionic/scudo_wrapper.cpp
@@ -48,7 +48,7 @@
   return -1;
 }
 
-int scudo_iterate(uintptr_t, size_t, void (*)(uintptr_t, size_t, void*), void*) {
+int scudo_malloc_iterate(uintptr_t, size_t, void (*)(uintptr_t, size_t, void*), void*) {
   return 0;
 }
 
diff --git a/libc/bionic/system_property_api.cpp b/libc/bionic/system_property_api.cpp
index 051bc4c..a641f12 100644
--- a/libc/bionic/system_property_api.cpp
+++ b/libc/bionic/system_property_api.cpp
@@ -100,7 +100,13 @@
 
 __BIONIC_WEAK_FOR_NATIVE_BRIDGE
 uint32_t __system_property_serial(const prop_info* pi) {
-  return system_properties.Serial(pi);
+  // N.B. a previous version of this function was much heavier-weight
+  // and enforced acquire semantics, so give our load here acquire
+  // semantics just in case somebody depends on
+  // __system_property_serial enforcing memory order, e.g., in case
+  // someone spins on the result of this function changing before
+  // loading some value.
+  return atomic_load_explicit(&pi->serial, memory_order_acquire);
 }
 
 __BIONIC_WEAK_FOR_NATIVE_BRIDGE
diff --git a/libc/include/bits/sys_statvfs_inlines.h b/libc/include/bits/sys_statvfs_inlines.h
index fd4578c..991fac7 100644
--- a/libc/include/bits/sys_statvfs_inlines.h
+++ b/libc/include/bits/sys_statvfs_inlines.h
@@ -38,20 +38,20 @@
 
 #if defined(__BIONIC_NEED_STATVFS_INLINES)
 
-static __inline void __bionic_statfs_to_statvfs(const struct statfs* __in,
-                                                struct statvfs* __out) {
-  __out->f_bsize = __in->f_bsize;
-  __out->f_frsize = __in->f_frsize;
-  __out->f_blocks = __in->f_blocks;
-  __out->f_bfree = __in->f_bfree;
-  __out->f_bavail = __in->f_bavail;
-  __out->f_files = __in->f_files;
-  __out->f_ffree = __in->f_ffree;
-  __out->f_favail = __in->f_ffree;
-  __out->f_fsid = __in->f_fsid.__val[0] |
-      __BIONIC_CAST(static_cast, uint64_t, __in->f_fsid.__val[1]) << 32;
-  __out->f_flag = __in->f_flags;
-  __out->f_namemax = __in->f_namelen;
+static __inline void __bionic_statfs_to_statvfs(const struct statfs* __src,
+                                                struct statvfs* __dst) {
+  __dst->f_bsize = __src->f_bsize;
+  __dst->f_frsize = __src->f_frsize;
+  __dst->f_blocks = __src->f_blocks;
+  __dst->f_bfree = __src->f_bfree;
+  __dst->f_bavail = __src->f_bavail;
+  __dst->f_files = __src->f_files;
+  __dst->f_ffree = __src->f_ffree;
+  __dst->f_favail = __src->f_ffree;
+  __dst->f_fsid = __src->f_fsid.__val[0] |
+      __BIONIC_CAST(static_cast, uint64_t, __src->f_fsid.__val[1]) << 32;
+  __dst->f_flag = __src->f_flags;
+  __dst->f_namemax = __src->f_namelen;
 }
 
 __BIONIC_SYS_STATVFS_INLINE int statvfs(const char* __path,
diff --git a/libc/libc.map.txt b/libc/libc.map.txt
index a4ab600..8e843af 100644
--- a/libc/libc.map.txt
+++ b/libc/libc.map.txt
@@ -1468,10 +1468,10 @@
     __system_properties_init; # apex
 
     # Used by libmemunreachable
-    malloc_backtrace; # apex vndk
-    malloc_disable; # apex vndk
-    malloc_enable; # apex vndk
-    malloc_iterate; # apex vndk
+    malloc_backtrace; # apex llndk
+    malloc_disable; # apex llndk
+    malloc_enable; # apex llndk
+    malloc_iterate; # apex llndk
 
     # Used by libandroid_net
     android_getaddrinfofornet; # apex
diff --git a/libc/malloc_debug/exported32.map b/libc/malloc_debug/exported32.map
index 8ed37fa..f75d173 100644
--- a/libc/malloc_debug/exported32.map
+++ b/libc/malloc_debug/exported32.map
@@ -8,13 +8,13 @@
     debug_free_malloc_leak_info;
     debug_get_malloc_leak_info;
     debug_initialize;
-    debug_iterate;
     debug_mallinfo;
     debug_malloc;
     debug_malloc_backtrace;
     debug_malloc_disable;
     debug_malloc_enable;
     debug_malloc_info;
+    debug_malloc_iterate;
     debug_malloc_usable_size;
     debug_mallopt;
     debug_memalign;
diff --git a/libc/malloc_debug/exported64.map b/libc/malloc_debug/exported64.map
index cdff88b..6dea58c 100644
--- a/libc/malloc_debug/exported64.map
+++ b/libc/malloc_debug/exported64.map
@@ -8,13 +8,13 @@
     debug_free_malloc_leak_info;
     debug_get_malloc_leak_info;
     debug_initialize;
-    debug_iterate;
     debug_mallinfo;
     debug_malloc;
     debug_malloc_backtrace;
     debug_malloc_disable;
     debug_malloc_enable;
     debug_malloc_info;
+    debug_malloc_iterate;
     debug_malloc_usable_size;
     debug_mallopt;
     debug_memalign;
diff --git a/libc/malloc_debug/malloc_debug.cpp b/libc/malloc_debug/malloc_debug.cpp
index 3c0e630..c0765a9 100644
--- a/libc/malloc_debug/malloc_debug.cpp
+++ b/libc/malloc_debug/malloc_debug.cpp
@@ -91,8 +91,8 @@
 int debug_mallopt(int param, int value);
 int debug_malloc_info(int options, FILE* fp);
 int debug_posix_memalign(void** memptr, size_t alignment, size_t size);
-int debug_iterate(uintptr_t base, size_t size,
-                  void (*callback)(uintptr_t base, size_t size, void* arg), void* arg);
+int debug_malloc_iterate(uintptr_t base, size_t size,
+                         void (*callback)(uintptr_t base, size_t size, void* arg), void* arg);
 void debug_malloc_disable();
 void debug_malloc_enable();
 
@@ -841,7 +841,7 @@
   return (*memptr != nullptr) ? 0 : ENOMEM;
 }
 
-int debug_iterate(uintptr_t base, size_t size, void (*callback)(uintptr_t, size_t, void*),
+int debug_malloc_iterate(uintptr_t base, size_t size, void (*callback)(uintptr_t, size_t, void*),
                   void* arg) {
   ScopedConcurrentLock lock;
   if (g_debug->TrackPointers()) {
@@ -854,7 +854,7 @@
 
   // An option that adds a header will add pointer tracking, so no need to
   // check if headers are enabled.
-  return g_dispatch->iterate(base, size, callback, arg);
+  return g_dispatch->malloc_iterate(base, size, callback, arg);
 }
 
 void debug_malloc_disable() {
diff --git a/libc/malloc_hooks/exported32.map b/libc/malloc_hooks/exported32.map
index 293d9ac..2b02806 100644
--- a/libc/malloc_hooks/exported32.map
+++ b/libc/malloc_hooks/exported32.map
@@ -7,13 +7,13 @@
     hooks_free_malloc_leak_info;
     hooks_get_malloc_leak_info;
     hooks_initialize;
-    hooks_iterate;
     hooks_mallinfo;
     hooks_malloc;
     hooks_malloc_backtrace;
     hooks_malloc_disable;
     hooks_malloc_enable;
     hooks_malloc_info;
+    hooks_malloc_iterate;
     hooks_malloc_usable_size;
     hooks_mallopt;
     hooks_memalign;
diff --git a/libc/malloc_hooks/exported64.map b/libc/malloc_hooks/exported64.map
index 340106b..59ec1f0 100644
--- a/libc/malloc_hooks/exported64.map
+++ b/libc/malloc_hooks/exported64.map
@@ -7,13 +7,13 @@
     hooks_free_malloc_leak_info;
     hooks_get_malloc_leak_info;
     hooks_initialize;
-    hooks_iterate;
     hooks_mallinfo;
     hooks_malloc;
     hooks_malloc_backtrace;
     hooks_malloc_disable;
     hooks_malloc_enable;
     hooks_malloc_info;
+    hooks_malloc_iterate;
     hooks_malloc_usable_size;
     hooks_mallopt;
     hooks_memalign;
diff --git a/libc/malloc_hooks/malloc_hooks.cpp b/libc/malloc_hooks/malloc_hooks.cpp
index b1c1d50..1ba8696 100644
--- a/libc/malloc_hooks/malloc_hooks.cpp
+++ b/libc/malloc_hooks/malloc_hooks.cpp
@@ -67,7 +67,7 @@
 struct mallinfo hooks_mallinfo();
 int hooks_mallopt(int param, int value);
 int hooks_posix_memalign(void** memptr, size_t alignment, size_t size);
-int hooks_iterate(uintptr_t base, size_t size,
+int hooks_malloc_iterate(uintptr_t base, size_t size,
     void (*callback)(uintptr_t base, size_t size, void* arg), void* arg);
 void hooks_malloc_disable();
 void hooks_malloc_enable();
@@ -209,7 +209,7 @@
   return g_dispatch->posix_memalign(memptr, alignment, size);
 }
 
-int hooks_iterate(uintptr_t, size_t, void (*)(uintptr_t, size_t, void*), void*) {
+int hooks_malloc_iterate(uintptr_t, size_t, void (*)(uintptr_t, size_t, void*), void*) {
   return 0;
 }
 
diff --git a/libc/private/bionic_malloc_dispatch.h b/libc/private/bionic_malloc_dispatch.h
index aea3a1c..52d8573 100644
--- a/libc/private/bionic_malloc_dispatch.h
+++ b/libc/private/bionic_malloc_dispatch.h
@@ -70,7 +70,7 @@
 #if defined(HAVE_DEPRECATED_MALLOC_FUNCS)
   MallocValloc valloc;
 #endif
-  MallocIterate iterate;
+  MallocIterate malloc_iterate;
   MallocMallocDisable malloc_disable;
   MallocMallocEnable malloc_enable;
   MallocMallopt mallopt;
diff --git a/libc/system_properties/include/system_properties/prop_area.h b/libc/system_properties/include/system_properties/prop_area.h
index a69f90e..53b2745 100644
--- a/libc/system_properties/include/system_properties/prop_area.h
+++ b/libc/system_properties/include/system_properties/prop_area.h
@@ -106,6 +106,20 @@
     memset(reserved_, 0, sizeof(reserved_));
     // Allocate enough space for the root node.
     bytes_used_ = sizeof(prop_bt);
+    // To make property reads wait-free, we reserve a
+    // PROP_VALUE_MAX-sized block of memory, the "dirty backup area",
+    // just after the root node. When we're about to modify a
+    // property, we copy the old value into the dirty backup area and
+    // copy the new value into the prop_info structure. Before
+    // starting the latter copy, we mark the property's serial as
+    // being dirty. If a reader comes along while we're doing the
+    // property update and sees a dirty serial, the reader copies from
+    // the dirty backup area instead of the property value
+    // proper. After the copy, the reader checks whether the property
+    // serial is the same: if it is, the dirty backup area hasn't been
+    // reused for something else and we can complete the
+    // read immediately.
+    bytes_used_ +=  __BIONIC_ALIGN(PROP_VALUE_MAX, sizeof(uint_least32_t));
   }
 
   const prop_info* find(const char* name);
@@ -122,6 +136,9 @@
   uint32_t version() const {
     return version_;
   }
+  char* dirty_backup_area() {
+    return data_ + sizeof (prop_bt);
+  }
 
  private:
   static prop_area* map_fd_ro(const int fd);
diff --git a/libc/system_properties/include/system_properties/system_properties.h b/libc/system_properties/include/system_properties/system_properties.h
index cad29cc..0666e28 100644
--- a/libc/system_properties/include/system_properties/system_properties.h
+++ b/libc/system_properties/include/system_properties/system_properties.h
@@ -66,7 +66,6 @@
   int Get(const char* name, char* value);
   int Update(prop_info* pi, const char* value, unsigned int len);
   int Add(const char* name, unsigned int namelen, const char* value, unsigned int valuelen);
-  uint32_t Serial(const prop_info* pi);
   uint32_t WaitAny(uint32_t old_serial);
   bool Wait(const prop_info* pi, uint32_t old_serial, uint32_t* new_serial_ptr,
             const timespec* relative_timeout);
@@ -74,6 +73,8 @@
   int Foreach(void (*propfn)(const prop_info* pi, void* cookie), void* cookie);
 
  private:
+  uint32_t ReadMutablePropertyValue(const prop_info* pi, char* value);
+
   // We don't want to use new or malloc in properties (b/31659220), and we don't want to waste a
   // full page by using mmap(), so we set aside enough space to create any context of the three
   // contexts.
diff --git a/libc/system_properties/system_properties.cpp b/libc/system_properties/system_properties.cpp
index d7c441b..8404778 100644
--- a/libc/system_properties/system_properties.cpp
+++ b/libc/system_properties/system_properties.cpp
@@ -140,42 +140,58 @@
   return strncmp(name, "ro.", 3) == 0;
 }
 
-int SystemProperties::Read(const prop_info* pi, char* name, char* value) {
-  while (true) {
-    uint32_t serial = Serial(pi);  // acquire semantics
-    size_t len = SERIAL_VALUE_LEN(serial);
-    memcpy(value, pi->value, len + 1);
-    // TODO: Fix the synchronization scheme here.
-    // There is no fully supported way to implement this kind
-    // of synchronization in C++11, since the memcpy races with
-    // updates to pi, and the data being accessed is not atomic.
-    // The following fence is unintuitive, but would be the
-    // correct one if memcpy used memory_order_relaxed atomic accesses.
-    // In practice it seems unlikely that the generated code would
-    // would be any different, so this should be OK.
+uint32_t SystemProperties::ReadMutablePropertyValue(const prop_info* pi, char* value) {
+  // We assume the memcpy below gets serialized by the acquire fence.
+  uint32_t new_serial = load_const_atomic(&pi->serial, memory_order_acquire);
+  uint32_t serial;
+  unsigned int len;
+  for (;;) {
+    serial = new_serial;
+    len = SERIAL_VALUE_LEN(serial);
+    if (__predict_false(SERIAL_DIRTY(serial))) {
+      // See the comment in the prop_area constructor.
+      prop_area* pa = contexts_->GetPropAreaForName(pi->name);
+      memcpy(value, pa->dirty_backup_area(), len + 1);
+    } else {
+      memcpy(value, pi->value, len + 1);
+    }
     atomic_thread_fence(memory_order_acquire);
-    if (serial == load_const_atomic(&(pi->serial), memory_order_relaxed)) {
-      if (name != nullptr) {
-        size_t namelen = strlcpy(name, pi->name, PROP_NAME_MAX);
-        if (namelen >= PROP_NAME_MAX) {
-          async_safe_format_log(ANDROID_LOG_ERROR, "libc",
-                                "The property name length for \"%s\" is >= %d;"
-                                " please use __system_property_read_callback"
-                                " to read this property. (the name is truncated to \"%s\")",
-                                pi->name, PROP_NAME_MAX - 1, name);
-        }
-      }
-      if (is_read_only(pi->name) && pi->is_long()) {
-        async_safe_format_log(
-            ANDROID_LOG_ERROR, "libc",
-            "The property \"%s\" has a value with length %zu that is too large for"
-            " __system_property_get()/__system_property_read(); use"
-            " __system_property_read_callback() instead.",
-            pi->name, strlen(pi->long_value()));
-      }
-      return len;
+    new_serial = load_const_atomic(&pi->serial, memory_order_relaxed);
+    if (__predict_true(serial == new_serial)) {
+      break;
+    }
+    // We need another fence here because we want to ensure that the memcpy in the
+    // next iteration of the loop occurs after the load of new_serial above. We could
+    // get this guarantee by making the load_const_atomic of new_serial
+    // memory_order_acquire instead of memory_order_relaxed, but then we'd pay the
+    // penalty of the memory_order_acquire even in the overwhelmingly common case
+    // that the serial number didn't change.
+    atomic_thread_fence(memory_order_acquire);
+  }
+  return serial;
+}
+
+int SystemProperties::Read(const prop_info* pi, char* name, char* value) {
+  uint32_t serial = ReadMutablePropertyValue(pi, value);
+  if (name != nullptr) {
+    size_t namelen = strlcpy(name, pi->name, PROP_NAME_MAX);
+    if (namelen >= PROP_NAME_MAX) {
+      async_safe_format_log(ANDROID_LOG_ERROR, "libc",
+                            "The property name length for \"%s\" is >= %d;"
+                            " please use __system_property_read_callback"
+                            " to read this property. (the name is truncated to \"%s\")",
+                            pi->name, PROP_NAME_MAX - 1, name);
     }
   }
+  if (is_read_only(pi->name) && pi->is_long()) {
+    async_safe_format_log(
+        ANDROID_LOG_ERROR, "libc",
+        "The property \"%s\" has a value with length %zu that is too large for"
+        " __system_property_get()/__system_property_read(); use"
+        " __system_property_read_callback() instead.",
+        pi->name, strlen(pi->long_value()));
+  }
+  return SERIAL_VALUE_LEN(serial);
 }
 
 void SystemProperties::ReadCallback(const prop_info* pi,
@@ -183,9 +199,9 @@
                                                      const char* value, uint32_t serial),
                                     void* cookie) {
   // Read only properties don't need to copy the value to a temporary buffer, since it can never
-  // change.
+  // change.  We use relaxed memory order on the serial load for the same reason.
   if (is_read_only(pi->name)) {
-    uint32_t serial = Serial(pi);
+    uint32_t serial = load_const_atomic(&pi->serial, memory_order_relaxed);
     if (pi->is_long()) {
       callback(cookie, pi->name, pi->long_value(), serial);
     } else {
@@ -194,21 +210,9 @@
     return;
   }
 
-  while (true) {
-    uint32_t serial = Serial(pi);  // acquire semantics
-    size_t len = SERIAL_VALUE_LEN(serial);
-    char value_buf[len + 1];
-
-    memcpy(value_buf, pi->value, len);
-    value_buf[len] = '\0';
-
-    // TODO: see todo in Read function
-    atomic_thread_fence(memory_order_acquire);
-    if (serial == load_const_atomic(&(pi->serial), memory_order_relaxed)) {
-      callback(cookie, pi->name, value_buf, serial);
-      return;
-    }
-  }
+  char value_buf[PROP_VALUE_MAX];
+  uint32_t serial = ReadMutablePropertyValue(pi, value_buf);
+  callback(cookie, pi->name, value_buf, serial);
 }
 
 int SystemProperties::Get(const char* name, char* value) {
@@ -231,26 +235,37 @@
     return -1;
   }
 
-  prop_area* pa = contexts_->GetSerialPropArea();
-  if (!pa) {
+  prop_area* serial_pa = contexts_->GetSerialPropArea();
+  if (!serial_pa) {
+    return -1;
+  }
+  prop_area* pa = contexts_->GetPropAreaForName(pi->name);
+  if (__predict_false(!pa)) {
+    async_safe_format_log(ANDROID_LOG_ERROR, "libc", "Could not find area for \"%s\"", pi->name);
     return -1;
   }
 
   uint32_t serial = atomic_load_explicit(&pi->serial, memory_order_relaxed);
+  unsigned int old_len = SERIAL_VALUE_LEN(serial);
+
+  // The contract with readers is that whenever the dirty bit is set, an undamaged copy
+  // of the pre-dirty value is available in the dirty backup area. The fence ensures
+  // that we publish our dirty area update before allowing readers to see a
+  // dirty serial.
+  memcpy(pa->dirty_backup_area(), pi->value, old_len + 1);
+  atomic_thread_fence(memory_order_release);
   serial |= 1;
   atomic_store_explicit(&pi->serial, serial, memory_order_relaxed);
-  // The memcpy call here also races.  Again pretend it
-  // used memory_order_relaxed atomics, and use the analogous
-  // counterintuitive fence.
-  atomic_thread_fence(memory_order_release);
   strlcpy(pi->value, value, len + 1);
-
-  atomic_store_explicit(&pi->serial, (len << 24) | ((serial + 1) & 0xffffff), memory_order_release);
-  __futex_wake(&pi->serial, INT32_MAX);
-
-  atomic_store_explicit(pa->serial(), atomic_load_explicit(pa->serial(), memory_order_relaxed) + 1,
+  // Now the primary value property area is up-to-date. Let readers know that they should
+  // look at the property value instead of the backup area.
+  atomic_thread_fence(memory_order_release);
+  atomic_store_explicit(&pi->serial, (len << 24) | ((serial + 1) & 0xffffff), memory_order_relaxed);
+  __futex_wake(&pi->serial, INT32_MAX);  // Fence by side effect
+  atomic_store_explicit(serial_pa->serial(),
+                        atomic_load_explicit(serial_pa->serial(), memory_order_relaxed) + 1,
                         memory_order_release);
-  __futex_wake(pa->serial(), INT32_MAX);
+  __futex_wake(serial_pa->serial(), INT32_MAX);
 
   return 0;
 }
@@ -294,16 +309,6 @@
   return 0;
 }
 
-// Wait for non-locked serial, and retrieve it with acquire semantics.
-uint32_t SystemProperties::Serial(const prop_info* pi) {
-  uint32_t serial = load_const_atomic(&pi->serial, memory_order_acquire);
-  while (SERIAL_DIRTY(serial)) {
-    __futex_wait(const_cast<_Atomic(uint_least32_t)*>(&pi->serial), serial, nullptr);
-    serial = load_const_atomic(&pi->serial, memory_order_acquire);
-  }
-  return serial;
-}
-
 uint32_t SystemProperties::WaitAny(uint32_t old_serial) {
   uint32_t new_serial;
   Wait(nullptr, old_serial, &new_serial, nullptr);
diff --git a/libm/Android.bp b/libm/Android.bp
index bf05b17..de32185 100644
--- a/libm/Android.bp
+++ b/libm/Android.bp
@@ -286,6 +286,7 @@
             pack_relocations: false,
             ldflags: ["-Wl,--hash-style=both"],
             version_script: ":libm.arm.map",
+            whole_static_libs: [ "libgcc_stripped" ],
         },
 
         arm64: {
diff --git a/linker/linker.cpp b/linker/linker.cpp
index 1393eb5..e09b2a4 100644
--- a/linker/linker.cpp
+++ b/linker/linker.cpp
@@ -1503,7 +1503,16 @@
   // Open the file.
   int fd = open_library(ns, zip_archive_cache, name, needed_by, &file_offset, &realpath);
   if (fd == -1) {
-    DL_OPEN_ERR("library \"%s\" not found", name);
+    if (task->is_dt_needed()) {
+      if (needed_by->is_main_executable()) {
+        DL_OPEN_ERR("library \"%s\" not found: needed by main executable", name);
+      } else {
+        DL_OPEN_ERR("library \"%s\" not found: needed by %s in namespace %s", name,
+                    needed_by->get_realpath(), task->get_start_from()->get_name());
+      }
+    } else {
+      DL_OPEN_ERR("library \"%s\" not found", name);
+    }
     return false;
   }
 
diff --git a/tests/dl_test.cpp b/tests/dl_test.cpp
index 9e46394..b69da97 100644
--- a/tests/dl_test.cpp
+++ b/tests/dl_test.cpp
@@ -204,7 +204,10 @@
 // The two libs are in ns2/ subdir.
 TEST(dl, exec_without_ld_config_file) {
 #if defined(__BIONIC__)
-  std::string error_message = "CANNOT LINK EXECUTABLE \"" + GetTestlibRoot() + "/ld_config_test_helper/ld_config_test_helper\": library \"ld_config_test_helper_lib1.so\" not found\n";
+  std::string error_message =
+      "CANNOT LINK EXECUTABLE \"" + GetTestlibRoot() +
+      "/ld_config_test_helper/ld_config_test_helper\": library \"ld_config_test_helper_lib1.so\" "
+      "not found: needed by main executable\n";
   std::string helper = GetTestlibRoot() +
       "/ld_config_test_helper/ld_config_test_helper";
   chmod(helper.c_str(), 0755);
diff --git a/tests/dlext_test.cpp b/tests/dlext_test.cpp
index e98d66f..7e772b8 100644
--- a/tests/dlext_test.cpp
+++ b/tests/dlext_test.cpp
@@ -28,6 +28,7 @@
 #include <android/dlext.h>
 #include <android-base/file.h>
 #include <android-base/strings.h>
+#include <android-base/test_utils.h>
 
 #include <sys/mman.h>
 #include <sys/types.h>
@@ -1374,7 +1375,10 @@
 
   void* handle2 = android_dlopen_ext(root_lib, RTLD_NOW, &extinfo);
   ASSERT_TRUE(handle2 == nullptr);
-  ASSERT_STREQ("dlopen failed: library \"libnstest_private_external.so\" not found", dlerror());
+  const char* error = dlerror();
+  ASSERT_MATCH(error,
+               R"(dlopen failed: library "libnstest_private_external.so" not found: needed by )"
+               R"(\S+libnstest_root_not_isolated.so in namespace private_isolated1)");
 
   // Check dlopen by absolute path
   handle2 = android_dlopen_ext(lib_private_external_path.c_str(), RTLD_NOW, &extinfo);
@@ -1502,7 +1506,9 @@
 
   void* handle2 = android_dlopen_ext(root_lib, RTLD_NOW, &extinfo);
   ASSERT_TRUE(handle2 == nullptr);
-  ASSERT_STREQ("dlopen failed: library \"libnstest_private_external.so\" not found", dlerror());
+  ASSERT_MATCH(dlerror(),
+               R"(dlopen failed: library "libnstest_private_external.so" not found: needed by )"
+               R"(\S+libnstest_root_not_isolated.so in namespace private_isolated_shared)");
 
   // Check dlopen by absolute path
   handle2 = android_dlopen_ext(lib_private_external_path.c_str(), RTLD_NOW, &extinfo);
@@ -1762,7 +1768,10 @@
 
   handle1 = android_dlopen_ext(root_lib, RTLD_NOW, &extinfo);
   ASSERT_TRUE(handle1 == nullptr);
-  ASSERT_STREQ("dlopen failed: library \"libnstest_public.so\" not found", dlerror());
+  ASSERT_MATCH(
+      dlerror(),
+      R"(dlopen failed: library "libnstest_public.so" not found: needed by \S+libnstest_root.so)"
+      R"( in namespace isolated2)");
 }
 
 TEST(dlext, ns_inaccessible_error_message) {
diff --git a/tests/system_properties_test.cpp b/tests/system_properties_test.cpp
index 245e42f..6d696c7 100644
--- a/tests/system_properties_test.cpp
+++ b/tests/system_properties_test.cpp
@@ -340,9 +340,9 @@
     ASSERT_EQ(0, system_properties.Add("property", 8, "value1", 6));
     const prop_info* pi = system_properties.Find("property");
     ASSERT_TRUE(pi != nullptr);
-    unsigned serial = system_properties.Serial(pi);
+    unsigned serial = __system_property_serial(pi);
     ASSERT_EQ(0, system_properties.Update(const_cast<prop_info*>(pi), "value2", 6));
-    ASSERT_NE(serial, system_properties.Serial(pi));
+    ASSERT_NE(serial, __system_property_serial(pi));
 #else // __BIONIC__
     GTEST_SKIP() << "bionic-only test";
 #endif // __BIONIC__
@@ -389,7 +389,7 @@
     prop_info* pi = const_cast<prop_info*>(system_properties.Find("property"));
     ASSERT_TRUE(pi != nullptr);
 
-    unsigned serial = system_properties.Serial(pi);
+    unsigned serial = __system_property_serial(pi);
 
     std::thread thread([&system_properties]() {
         prop_info* pi = const_cast<prop_info*>(system_properties.Find("property"));