A piece of news here: JS engine initialization creates JIT code for atomic ops (including flushing the icache for this JITted code). That may happen on a non-main thread, so because of the current patch, we think the icache flushing needs to be broadcast to all the threads, while we never checked for the membarrier availability before, and in my case that happens to fail. The proper fix might be to add another parameter to `EnsureIAndDCacheCoherency` so it knows whether the icache invalidation must be broadcast or not. Then wasm compilation can set this parameter to "please broadcast the invalidation" since the code may be run on other threads, while other use cases don't have to do the heavyweight membarrier for their own usage.
Bug 1660944 Comment 4 Edit History
Note: The actual edited comment in the bug view page will always show the original commenter’s name and original timestamp.
A piece of news here: JS engine initialization creates JIT code for atomic ops (including flushing the icache for this JITted code). That may happen on a non-main thread, so because of the current patch, we think the icache flushing needs to be broadcast to all the threads, while we never checked for the membarrier availability before, and in my case that happens to fail (kernel too old). The proper fix might be to add another parameter to `EnsureIAndDCacheCoherency` so it knows whether the icache invalidation must be broadcast or not. Then wasm compilation can set this parameter to "please broadcast the invalidation" since the code may be run on other threads, while other use cases don't have to do the heavyweight membarrier for their own usage.