RFR: 8305819: LogConfigurationTest intermittently fails on AArch64

[Andrew Haley]

On Tue, 11 Apr 2023 08:45:56 GMT, Dan Gao <duke at openjdk.org> wrote:

> LogConfigurationTest*reconfigure*MT* crash intermittently on AArch64.
> According to the crash log and coredump, we found it crash as follows:
> 
> void LogTagSet::log(LogLevelType level, const char* msg) {
>   LogOutputList::Iterator it = _output_list.iterator(level);
>   LogDecorations decorations(level, *this, _decorators);
> 
>   for (; it != _output_list.end(); it++) {
>     (*it)->write(decorations, msg);//crash 
>   }
> }
> 
> In the test, two threads write into the log while another thread dynamically changes the decorators and tags. During this time, the  _output_list will be modified. Because of the relax memory model of aarch64, while adding LogOutputNode to LogOutputList, adding node to list and setting the value of node may be reordered, therefore the read thread may not read the correct value of the node's content. Consequently, storestore memory barrier is needed to ensure the order of writing. 
> By the way, applying this patch may affect performance.
> 
> How to reproduce on Linux aarch64:
> test case
> 
> /* @test
>  * @library /test/lib
>  * @modules java.base/jdk.internal.misc
>  *          java.xml
>  * @run main/native GTestWrapper --gtest_filter=LogConfigurationTest*reconfigure*MT*
>  */
> 
> Crash may occasionally occur after running continuously for 5000 times.

I've had an idea. How about you use a load-acquire on the reader side, making the code correct, with a FIXME comment. Something like "// FIXME: memory_order_consume could be used here."
Then, I will add memory_order_consume to the shared atomic code, which we can define appropriately in the OS/CPU section. OK?

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

PR Comment: https://git.openjdk.org/jdk/pull/13421#issuecomment-1536167507