RFR: 8305819: LogConfigurationTest intermittently fails on AArch64

[gaogao-mem]

On Fri, 21 Apr 2023 08:24:08 GMT, Andrew Haley <aph at openjdk.org> wrote:

>> Thanks for your suggestion! The performance issue I mentioned earlier was due to adding OrderAccess::loadload while reading the list, but after researching some information for memory barrier, it was found to be unnecessary.
>
> Atomic access is unnecessary when reading the list? I'd like to know the reasoning behind that.
> I guess you could argue that dependency would save you on the read side, but I'd like to see and understand the code. It's very risky to depend on that,

The list is read in src/hotspot/share/logging/logTagSet.cpp by iterator in src/hotspot/share/logging/logOutputList.hpp.
For example, in void LogTagSet::log(LogLevelType level, const char* msg), the list is read as follows:
  ```
src/hotspot/share/logging/logTagSet.cpp

LogOutputList::Iterator it = _output_list.iterator(level);
  LogDecorations decorations(level, *this, _decorators);

  for (; it != _output_list.end(); it++) {
    (*it)->write(decorations, msg);
  }

And I intended to add loadload in Iterator as follows:
   ```
src/hotspot/share/logging/logOutputList.hpp

void operator++(int) {
          _current = _current->_next;
+      OrderAccess::loadload();
  }
  Iterator iterator(LogLevelType level = LogLevel::Last) {
        increase_readers();
-    return Iterator(this, _level_start[level]);
+    LogOutputNode* start=_level_start[level];
+    OrderAccess::loadload();
+    return Iterator(this, start);
   }

-------------

PR Review Comment: https://git.openjdk.org/jdk/pull/13421#discussion_r1173569182