Implement redisAtomic to replace _Atomic C11 builtin (#7707)

Redis 6.0 introduces I/O threads, it is so cool and efficient, we use C11
_Atomic to establish inter-thread synchronization without mutex. But the
compiler that must supports C11 _Atomic can compile redis code, that brings a
lot of inconvenience since some common platforms can't support by default such
as CentOS7, so we want to implement redis atomic type to make it more portable.

We have implemented our atomic variable for redis that only has 'relaxed'
operations in src/atomicvar.h, so we implement some operations with
'sequentially-consistent', just like the default behavior of C11 _Atomic that
can establish inter-thread synchronization. And we replace all uses of C11
_Atomic with redis atomic variable.

Our implementation of redis atomic variable uses C11 _Atomic, __atomic or
__sync macros if available, it supports most common platforms, and we will
detect automatically which feature we use. In Makefile we use a dummy file to
detect if the compiler supports C11 _Atomic. Now for gcc, we can compile redis
code theoretically if your gcc version is not less than 4.1.2(starts to support
__sync_xxx operations). Otherwise, we remove use mutex fallback to implement
redis atomic variable for performance and test. You will get compiling errors
if your compiler doesn't support all features of above.

For cover redis atomic variable tests, we add other CI jobs that build redis on
CentOS6 and CentOS7 and workflow daily jobs that run the tests on them.
For them, we just install gcc by default in order to cover different compiler
versions, gcc is 4.4.7 by default installation on CentOS6 and 4.8.5 on CentOS7.

We restore the feature that we can test redis with Helgrind to find data race
errors. But you need install Valgrind in the default path configuration firstly
before running your tests, since we use macros in helgrind.h to tell Helgrind
inter-thread happens-before relationship explicitly for avoiding false positives.
Please open an issue on github if you find data race errors relate to this commit.

Unrelated:
- Fix redefinition of typedef 'RedisModuleUserChangedFunc'
  For some old version compilers, they will report errors or warnings, if we
  re-define function type.

1 files changed, 67 insertions, 42 deletions

diff --git a/src/atomicvar.h b/src/atomicvar.h
index ecd26ad70..6ac04c605 100644
--- a/src/atomicvar.h
+++ b/src/atomicvar.h
@@ -1,5 +1,5 @@
-/* This file implements atomic counters using __atomic or __sync macros if
- * available, otherwise synchronizing different threads using a mutex.
+/* This file implements atomic counters using c11 _Atomic, __atomic or __sync
+ * macros if available, otherwise we will throw an error when compile.
  *
  * The exported interface is composed of three macros:
  *
@@ -8,16 +8,8 @@
  * atomicDecr(var,count) -- Decrement the atomic counter
  * atomicGet(var,dstvar) -- Fetch the atomic counter value
  * atomicSet(var,value)  -- Set the atomic counter value
- *
- * The variable 'var' should also have a declared mutex with the same
- * name and the "_mutex" postfix, for instance:
- *
- *  long myvar;
- *  pthread_mutex_t myvar_mutex;
- *  atomicSet(myvar,12345);
- *
- * If atomic primitives are available (tested in config.h) the mutex
- * is not used.
+ * atomicGetWithSync(var,value)  -- 'atomicGet' with inter-thread synchronization
+ * atomicSetWithSync(var,value)  -- 'atomicSet' with inter-thread synchronization
  *
  * Never use return value from the macros, instead use the AtomicGetIncr()
  * if you need to get the current value and increment it atomically, like
@@ -58,17 +50,64 @@
  */
 
 #include <pthread.h>
+#include "config.h"
 
 #ifndef __ATOMIC_VAR_H
 #define __ATOMIC_VAR_H
 
+/* Define redisAtomic for atomic variable. */
+#define redisAtomic
+
 /* To test Redis with Helgrind (a Valgrind tool) it is useful to define
  * the following macro, so that __sync macros are used: those can be detected
  * by Helgrind (even if they are less efficient) so that no false positive
  * is reported. */
 // #define __ATOMIC_VAR_FORCE_SYNC_MACROS
 
-#if !defined(__ATOMIC_VAR_FORCE_SYNC_MACROS) && defined(__ATOMIC_RELAXED) && !defined(__sun) && (!defined(__clang__) || !defined(__APPLE__) || __apple_build_version__ > 4210057)
+/* There will be many false positives if we test Redis with Helgrind, since
+ * Helgrind can't understand we have imposed ordering on the program, so
+ * we use macros in helgrind.h to tell Helgrind inter-thread happens-before
+ * relationship explicitly for avoiding false positives.
+ *
+ * For more details, please see: valgrind/helgrind.h and
+ * https://www.valgrind.org/docs/manual/hg-manual.html#hg-manual.effective-use
+ *
+ * These macros take effect only when 'make helgrind', and you must first
+ * install Valgrind in the default path configuration. */
+#ifdef __ATOMIC_VAR_FORCE_SYNC_MACROS
+#include <valgrind/helgrind.h>
+#else
+#define ANNOTATE_HAPPENS_BEFORE(v) ((void) v)
+#define ANNOTATE_HAPPENS_AFTER(v)  ((void) v)
+#endif
+
+#if !defined(__ATOMIC_VAR_FORCE_SYNC_MACROS) && defined(__STDC_VERSION__) && \
+    (__STDC_VERSION__ >= 201112L) && !defined(__STDC_NO_ATOMICS__)
+/* Use '_Atomic' keyword if the compiler supports. */
+#undef  redisAtomic
+#define redisAtomic _Atomic
+/* Implementation using _Atomic in C11. */
+
+#include <stdatomic.h>
+#define atomicIncr(var,count) atomic_fetch_add_explicit(&var,(count),memory_order_relaxed)
+#define atomicGetIncr(var,oldvalue_var,count) do { \
+    oldvalue_var = atomic_fetch_add_explicit(&var,(count),memory_order_relaxed); \
+} while(0)
+#define atomicDecr(var,count) atomic_fetch_sub_explicit(&var,(count),memory_order_relaxed)
+#define atomicGet(var,dstvar) do { \
+    dstvar = atomic_load_explicit(&var,memory_order_relaxed); \
+} while(0)
+#define atomicSet(var,value) atomic_store_explicit(&var,value,memory_order_relaxed)
+#define atomicGetWithSync(var,dstvar) do { \
+    dstvar = atomic_load_explicit(&var,memory_order_seq_cst); \
+} while(0)
+#define atomicSetWithSync(var,value) \
+    atomic_store_explicit(&var,value,memory_order_seq_cst)
+#define REDIS_ATOMIC_API "c11-builtin"
+
+#elif !defined(__ATOMIC_VAR_FORCE_SYNC_MACROS) && !defined(__sun) && \
+    (!defined(__clang__) || !defined(__APPLE__) || __apple_build_version__ > 4210057) && \
+    defined(__ATOMIC_RELAXED) && defined(__ATOMIC_SEQ_CST)
 /* Implementation using __atomic macros. */
 
 #define atomicIncr(var,count) __atomic_add_fetch(&var,(count),__ATOMIC_RELAXED)
@@ -80,6 +119,11 @@
     dstvar = __atomic_load_n(&var,__ATOMIC_RELAXED); \
 } while(0)
 #define atomicSet(var,value) __atomic_store_n(&var,value,__ATOMIC_RELAXED)
+#define atomicGetWithSync(var,dstvar) do { \
+    dstvar = __atomic_load_n(&var,__ATOMIC_SEQ_CST); \
+} while(0)
+#define atomicSetWithSync(var,value) \
+    __atomic_store_n(&var,value,__ATOMIC_SEQ_CST)
 #define REDIS_ATOMIC_API "atomic-builtin"
 
 #elif defined(HAVE_ATOMIC)
@@ -96,38 +140,19 @@
 #define atomicSet(var,value) do { \
     while(!__sync_bool_compare_and_swap(&var,var,value)); \
 } while(0)
+/* Actually the builtin issues a full memory barrier by default. */
+#define atomicGetWithSync(var,dstvar) { \
+    dstvar = __sync_sub_and_fetch(&var,0,__sync_synchronize); \
+    ANNOTATE_HAPPENS_AFTER(&var); \
+} while(0)
+#define atomicSetWithSync(var,value) do { \
+    ANNOTATE_HAPPENS_BEFORE(&var);  \
+    while(!__sync_bool_compare_and_swap(&var,var,value,__sync_synchronize)); \
+} while(0)
 #define REDIS_ATOMIC_API "sync-builtin"
 
 #else
-/* Implementation using pthread mutex. */
-
-#define atomicIncr(var,count) do { \
-    pthread_mutex_lock(&var ## _mutex); \
-    var += (count); \
-    pthread_mutex_unlock(&var ## _mutex); \
-} while(0)
-#define atomicGetIncr(var,oldvalue_var,count) do { \
-    pthread_mutex_lock(&var ## _mutex); \
-    oldvalue_var = var; \
-    var += (count); \
-    pthread_mutex_unlock(&var ## _mutex); \
-} while(0)
-#define atomicDecr(var,count) do { \
-    pthread_mutex_lock(&var ## _mutex); \
-    var -= (count); \
-    pthread_mutex_unlock(&var ## _mutex); \
-} while(0)
-#define atomicGet(var,dstvar) do { \
-    pthread_mutex_lock(&var ## _mutex); \
-    dstvar = var; \
-    pthread_mutex_unlock(&var ## _mutex); \
-} while(0)
-#define atomicSet(var,value) do { \
-    pthread_mutex_lock(&var ## _mutex); \
-    var = value; \
-    pthread_mutex_unlock(&var ## _mutex); \
-} while(0)
-#define REDIS_ATOMIC_API "pthread-mutex"
+#error "Unable to determine atomic operations for your platform"
 
 #endif
 #endif /* __ATOMIC_VAR_H */