Looks like JDK-8186027 is causing a few performance regressions:

• ~ 1.5% in SPECjbb2005-G1 on Solaris-SPARC

• 2-5% in SHA1PRNG crypto benchmarks on Linux-x64, Mac-x64, and Solaris-SPARC:
   • micro.benchmarks.crypto.small.SecureRandomBench.nextBytes-algorithm:SHA1PRNG-dataSize:64-provider:-shared:false
   • micro.benchmarks.crypto.small.SecureRandomBench.nextBytes-algorithm:SHA1PRNG-dataSize:64-provider:-shared:true

Regressions hold at least through b39.

Reproduces in triage (reproduction run on same machine).
Reproduces in re-triage (reproduction run on different machine).
VMswap results indicate the cause is in the JVM.

Measurements of b36 using -XX:-UseCountedLoopSafepoints show the regressions go away. So the cause appears to be JDK-8186027