Merge branch 'raimo/ssl/tls-optimization/OTP-15529' into maint

* raimo/ssl/tls-optimization/OTP-15529:
  Inline local function
  Optimize binary matching
  Clean up module boundaries
  Remove redundant return of CipherState
  Use iovec() internally in send path
  Small binary handling optimizations
  Optimize read_application_data with Okasaki queue
  Try to optimize decode_cipher_text/3
  Optimize application data aggregation
  Optimize TLS record parsing with Okasaki queue
  Cache strong_random_bytes for IV
  Optimize padding
  Produce less garbage in encrypt loop
  Reorganize #data{}
  Tidy up state machine
  Add server GC info to bench results

2 files changed, 44 insertions, 23 deletions

diff --git a/lib/ssl/test/ssl_basic_SUITE.erl b/lib/ssl/test/ssl_basic_SUITE.erl
index d0fa0443f2..0a66dfa0b1 100644
--- a/lib/ssl/test/ssl_basic_SUITE.erl
+++ b/lib/ssl/test/ssl_basic_SUITE.erl
@@ -1,7 +1,7 @@
 %%
 %% %CopyrightBegin%
 %%
-%% Copyright Ericsson AB 2007-2018. All Rights Reserved.
+%% Copyright Ericsson AB 2007-2019. All Rights Reserved.
 %%
 %% Licensed under the Apache License, Version 2.0 (the "License");
 %% you may not use this file except in compliance with the License.
@@ -3555,7 +3555,7 @@ tls_dont_crash_on_handshake_garbage(Config) ->
                        <<22, 3,3, 5:16, 92,64,37,228,209>> % garbage
                       ]),
     % Send unexpected change_cipher_spec
-    ok = gen_tcp:send(Socket, <<20, 0,0,12, 111,40,244,7,137,224,16,109,197,110,249,152>>),
+    ok = gen_tcp:send(Socket, <<20, 3,3, 12:16, 111,40,244,7,137,224,16,109,197,110,249,152>>),
 
     % Ensure we receive an alert, not sudden disconnect
     {ok, <<21, _/binary>>} = drop_handshakes(Socket, 1000).
diff --git a/lib/ssl/test/ssl_dist_bench_SUITE.erl b/lib/ssl/test/ssl_dist_bench_SUITE.erl
index 7409b69639..618ad0789e 100644
--- a/lib/ssl/test/ssl_dist_bench_SUITE.erl
+++ b/lib/ssl/test/ssl_dist_bench_SUITE.erl
@@ -1,7 +1,7 @@
 %%%-------------------------------------------------------------------
 %% %CopyrightBegin%
 %%
-%% Copyright Ericsson AB 2017-2018. All Rights Reserved.
+%% Copyright Ericsson AB 2017-2019. All Rights Reserved.
 %%
 %% Licensed under the Apache License, Version 2.0 (the "License");
 %% you may not use this file except in compliance with the License.
@@ -42,7 +42,7 @@
     throughput_1048576/1]).
 
 %% Debug
--export([payload/1]).
+-export([payload/1, roundtrip_runner/3, setup_runner/3, throughput_runner/4]).
 
 %%%-------------------------------------------------------------------
 
@@ -407,17 +407,19 @@ throughput(A, B, Prefix, HA, HB, Packets, Size) ->
     [] = ssl_apply(HA, erlang, nodes, []),
     [] = ssl_apply(HB, erlang, nodes, []),
     #{time := Time,
-      dist_stats := DistStats,
+      client_dist_stats := ClientDistStats,
       client_msacc_stats := ClientMsaccStats,
       client_prof := ClientProf,
       server_msacc_stats := ServerMsaccStats,
-      server_prof := ServerProf} =
+      server_prof := ServerProf,
+      server_gc_before := Server_GC_Before,
+      server_gc_after := Server_GC_After} =
         ssl_apply(HA, fun () -> throughput_runner(A, B, Packets, Size) end),
     [B] = ssl_apply(HA, erlang, nodes, []),
     [A] = ssl_apply(HB, erlang, nodes, []),
     ClientMsaccStats =:= undefined orelse
         msacc:print(ClientMsaccStats),
-    io:format("DistStats: ~p~n", [DistStats]),
+    io:format("ClientDistStats: ~p~n", [ClientDistStats]),
     Overhead =
         50 % Distribution protocol headers (empirical) (TLS+=54)
         + byte_size(erlang:term_to_binary([0|<<>>])), % Benchmark overhead
@@ -436,6 +438,8 @@ throughput(A, B, Prefix, HA, HB, Packets, Size) ->
         end,
     io:format("******* ClientProf:~n", []), prof_print(ClientProf),
     io:format("******* ServerProf:~n", []), prof_print(ServerProf),
+    io:format("******* Server GC Before:~n~p~n", [Server_GC_Before]),
+    io:format("******* Server GC After:~n~p~n", [Server_GC_After]),
     Speed = round((Bytes * 1000000) / (1024 * Time)),
     report(Prefix++" Throughput_"++integer_to_list(Size), Speed, "kB/s").
 
@@ -457,10 +461,10 @@ throughput_runner(A, B, Rounds, Size) ->
             ok
         end,
     prof_start(),
-    {Time,ServerMsaccStats,ServerProf} =
+    #{time := Time} = Result =
         throughput_client(ServerPid, ServerMon, Payload, Rounds),
     prof_stop(),
-    ClientMsaccStats =
+    MsaccStats =
         case msacc:available() of
             true ->
                 MStats = msacc:stats(),
@@ -469,15 +473,13 @@ throughput_runner(A, B, Rounds, Size) ->
             false ->
                 undefined
         end,
-    ClientProf = prof_end(),
+    Prof = prof_end(),
     [{_Node,Socket}] = dig_dist_node_sockets(),
     DistStats = inet:getstat(Socket),
-    #{time => microseconds(Time),
-      dist_stats => DistStats,
-      client_msacc_stats => ClientMsaccStats,
-      client_prof => ClientProf,
-      server_msacc_stats => ServerMsaccStats,
-      server_prof => ServerProf}.
+    Result#{time := microseconds(Time),
+            client_dist_stats => DistStats,
+            client_msacc_stats => MsaccStats,
+            client_prof => Prof}.
 
 dig_dist_node_sockets() ->
     [case DistCtrl of
@@ -500,6 +502,9 @@ dig_dist_node_sockets() ->
 
 
 throughput_server(Pid, N) ->
+    GC_Before = get_server_gc_info(),
+    %% dbg:tracer(port, dbg:trace_port(file, "throughput_server_gc.log")),
+    %% dbg:p(TLSDistReceiver, garbage_collection),
     msacc:available() andalso
         begin
             msacc:stop(),
@@ -508,9 +513,9 @@ throughput_server(Pid, N) ->
             ok
         end,
     prof_start(),
-    throughput_server_loop(Pid, N).
+    throughput_server_loop(Pid, GC_Before, N).
 
-throughput_server_loop(_Pid, 0) ->
+throughput_server_loop(_Pid, GC_Before, 0) ->
     prof_stop(),
     MsaccStats =
         case msacc:available() of
@@ -523,11 +528,26 @@ throughput_server_loop(_Pid, 0) ->
                 undefined
         end,
     Prof = prof_end(),
-    exit({ok,MsaccStats,Prof});
-throughput_server_loop(Pid, N) ->
+    %% dbg:flush_trace_port(),
+    exit(#{server_msacc_stats => MsaccStats,
+           server_prof => Prof,
+           server_gc_before => GC_Before,
+           server_gc_after => get_server_gc_info()});
+throughput_server_loop(Pid, GC_Before, N) ->
     receive
         {Pid, N, _} ->
-            throughput_server_loop(Pid, N-1)
+            throughput_server_loop(Pid, GC_Before, N-1)
+    end.
+
+get_server_gc_info() ->
+    case whereis(ssl_connection_sup_dist) of
+        undefined ->
+            undefined;
+        SupPid ->
+            [{_Id,TLSDistReceiver,_Type,_Modules}|_] =
+                supervisor:which_children(SupPid),
+            erlang:process_info(
+              TLSDistReceiver, [garbage_collection,garbage_collection_info])
     end.
 
 throughput_client(Pid, Mon, Payload, N) ->
@@ -535,8 +555,8 @@ throughput_client(Pid, Mon, Payload, N) ->
 
 throughput_client_loop(_Pid, Mon, _Payload, 0, StartTime) ->
     receive
-        {'DOWN', Mon, _, _, {ok,MsaccStats,Prof}} ->
-            {elapsed_time(StartTime),MsaccStats,Prof};
+        {'DOWN', Mon, _, _, #{} = Result} ->
+            Result#{time => elapsed_time(StartTime)};
         {'DOWN', Mon, _, _, Other} ->
             exit(Other)
     end;
@@ -554,6 +574,7 @@ prof_start() ->
     ok.
 -elif(?prof =:= eprof).
 prof_start() ->
+    catch eprof:stop(),
     {ok,_} = eprof:start(),
     profiling = eprof:start_profiling(processes()),
     ok.