RFR(XXS): 8238180: RunThese30M failed "assert(t->jfr_thread_local()->shelved_buffer() == __null) failed: invariant"

[Markus Gronlund]

Greetings,

Please review this small adjustment to relax some assertions that can be triggered by the DeoptimizationEvent.

Bug: https://bugs.openjdk.java.net/browse/JDK-8238180
Testing: jdk_jfr
Description:
The newly introduced event for Deoptimization comes into flush from a previously unknown location. The flush path taken is the native path, but the thread has still its Java context, including its stored away regular buffer (its working with a large buffer lease on the Java side). There is nothing wrong in doing this, the native event can easily be accommodated after a regular flush, so the assertions should be relaxed.

Thanks
Markus

Change:

diff -r e804de2e6d4a -r 6c6e978c5f60 src/hotspot/share/jfr/recorder/storage/jfrStorage.cpp
--- a/src/hotspot/share/jfr/recorder/storage/jfrStorage.cpp    Thu Mar 05 16:46:24 2020 -0500
+++ b/src/hotspot/share/jfr/recorder/storage/jfrStorage.cpp    Thu Mar 05 23:41:31 2020 +0100
@@ -464,7 +464,6 @@

 static void assert_flush_regular_precondition(ConstBufferPtr cur, const u1* const cur_pos, size_t used, size_t req, const Thread* t) {
   assert(t != NULL, "invariant");
-  assert(t->jfr_thread_local()->shelved_buffer() == NULL, "invariant");
   assert(cur != NULL, "invariant");
   assert(!cur->lease(), "invariant");
   assert(cur_pos != NULL, "invariant");
@@ -513,7 +512,6 @@
       return cur;
     }
   }
-  assert(t->jfr_thread_local()->shelved_buffer() == NULL, "invariant");
   if (cur->free_size() >= req) {
     // simplest case, no switching of buffers
     if (used > 0) {
@@ -524,6 +522,7 @@
   }
   // Going for a "larger-than-regular" buffer.
   // Shelve the current buffer to make room for a temporary lease.
+  assert(t->jfr_thread_local()->shelved_buffer() == NULL, "invariant");
   t->jfr_thread_local()->shelve_buffer(cur);
   return provision_large(cur, cur_pos, used, req, native, t);
 }