[4.0] Added missing calls to wasmifs on read-only threads

[linh2931]

Resolve #963

• Added missing calls to wasmifs on read-only threads
• Removed first_block_num_used from wasm_instantiation_cache.emplace
• Replaced thread_local for wasmifs with a local variable (unordered_map)

[spoonincode]

If I'm understanding right, I don't think this is the correct behavior. When a contract calls eosio_exit it's a "clean" exit of the executing contract. But this now calls

leap/libraries/chain/controller.cpp

Lines 2741 to 2747 in 1f46ff6

	void wasm_interface_exit() {
	// exit all running wasmifs
	wasmif.exit();
	for (auto& ele: threaded_wasmifs) {
	ele.second->exit();
	}
	}

which looks like it operates on all executing contracts

[linh2931]

Thanks Matt. You are correct. I was wrong to exit other executing contracts. Am reverting it.

[linh2931]

I am looking into #956 after this PR.

[heifner]

This static map causes instability:

leap/libraries/wasm-jit/Source/IR/Types.cpp

Line 20 in 9a10c80

static std::map<Key,FunctionType*> map;

Here is how I fixed it:

diff --git a/libraries/wasm-jit/Source/IR/Types.cpp b/libraries/wasm-jit/Source/IR/Types.cpp
index d1ade9c3b1..ff3704fd67 100644
--- a/libraries/wasm-jit/Source/IR/Types.cpp
+++ b/libraries/wasm-jit/Source/IR/Types.cpp
@@ -1,6 +1,7 @@
 #include "Types.h"

 #include <map>
+#include <mutex>

 namespace IR
 {
@@ -25,6 +26,8 @@ namespace IR
        template<typename Key,typename Value,typename CreateValueThunk>
        Value findExistingOrCreateNew(std::map<Key,Value>& map,Key&& key,CreateValueThunk createValueThunk)
        {
+      static std::mutex mtx;
+      std::lock_guard<std::mutex> g(mtx);
                auto mapIt = map.find(key);
                if(mapIt != map.end()) { return mapIt->second; }
                else

There is tabs in that file, so the indentation looks weird in the diff.

[heifner]

Fix use of static map

[spoonincode]

@heifner that maybe should be another issue; though it wouldn't surprise me if it's related to #969 since what is pointed to in there may be the root cause of WAVM code being touched from multiple threads.

[heifner]

Approved assuming that the static map will be fixed in a diff PR.

[linh2931]

I am doing a separate PR for #969 shortly.

[linh2931]

The PR is #975

[heifner]

This static map causes instability:

Verified that #975 fixes this issue.

[linh2931]

Thanks @heifner and @spoonincode!

[spoonincode]

Looks good

[spoonincode]

I'm not sure you need a unique_ptr here.. couldn't you just emplace() the wasm_interface in?