RFR rev1 (M): 8078556: Runtime: implement ranges (optionally constraints) for those flags that have them missing (original) (raw)

gerard ziemski gerard.ziemski at oracle.com
Fri Oct 9 18:47:41 UTC 2015

(resending, fixed webrev link)

hi all,

Please review this rev1 change that implements ranges and constraint for runtime flags.

The purpose of the ranges is to define valid ranges, that are guaranteed to be accepted and work when instantiating a VM. The VM is allowed to exit with an exception due to out of OS resources (like memory), but it may not assert or prematurely exit with an error that produces hs_err and/or dump core.

Summary of changes since rev0:

#1 Punted the following flags:

SharedBaseAddress SharedMiscCodeSize SharedMiscDataSize SharedReadOnlySize SharedReadWriteSize

to task JDK-8138983 since their ranges are nontrivial [Coleen, Jianghli]

#2. Removed unused "Arguments::get_default_heap_base_min_address" [Coleen]

#3. Users may need to use "TraceRedefineClasses" with their own custom events, so changed the max value to 0XFFFFFFFF [Dmitry, Dan]

#4. All other feedback from Dmitry implemented, see below

#5. PrintSafepointStatisticsTimeout and SafepointtimeoutDelat max value is LP64_ONLY(max_intx/MICROUNITS) NOT_LP64(max_intx) because intx is 4bytes on 32bits architectures, so dividing it by MICROUNITS would bring the value below default. I did, however, make sure to upcast the values to jlong, so on 32bit they become 8byte long long, and so no overflow will occur when multiplying by MICROUNITS.

#6. GuaranteedSafepointInterval range(0, max_intx) asserts (overflows) on Windows64 (max_intx on 64bit is 0x7FFFFFFFFFFFFFFF), so I changed the range max to max_jint (ie. 0x7FFFFFFF) which is roughly 25 days

The following flags had their ranges defined by the the domain experts:

AttachListenerTimeout: Staffan Larsen BiasedLockingBulkRebiasThreshold: Marcus Larsson BiasedLockingBulkRevokeThreshold: Marcus Larsson BiasedLockingDecayTime: Marcus Larsson BiasedLockingStartupDelay: Marcus Larsson SharedSymbolTableBucketSize: Yumin Qi StackShadowPages: Coleen Phillimore, Frederic Parain

For the rest of the flags, I looked at the existing code that was using them and also did local tests:

DeferPollingPageLoopCount DeferThrSuspendLoopCount GuaranteedSafepointInterval HeapBaseMinAddress InitialBootClassLoaderMetaspaceSize JavaPriority10_To_OSPriority JavaPriority1_To_OSPriority JavaPriority2_To_OSPriority JavaPriority3_To_OSPriority JavaPriority4_To_OSPriority JavaPriority5_To_OSPriority JavaPriority6_To_OSPriority JavaPriority7_To_OSPriority JavaPriority8_To_OSPriority JavaPriority9_To_OSPriority MallocMaxTestWords MallocVerifyInterval MallocVerifyStart MaxDirectMemorySize MaxJNILocalCapacity MaxJavaStackTraceDepth MaxRAM MinPassesBeforeFlush PerfDataMemorySize PerfDataSamplingInterval PerfMaxStringConstLength PrintSafepointStatisticsCount PrintSafepointStatisticsTimeout ProfileIntervalsTicks RTMRetryCount SafepointSpinBeforeYield SafepointTimeoutDelay SelfDestructTimer SuspendRetryCount SuspendRetryDelay ThreadSafetyMargin ThreadStackSize TraceRedefineClasses VMThreadPriority VMThreadStackSize

The change passes “JPRT -testset hotspot”, “RBT hotspot/test/runtime/CommandLine” and “RBT testlist,noncolo.testlist --add-tonga-keyword quick”

References:

bugid: https://bugs.openjdk.java.net/browse/JDK-8078556 webrev: http://cr.openjdk.java.net/~gziemski/8078556_rev1

Followup issues:

https://bugs.openjdk.java.net/browse/JDK-8133564 "Runtime - 2nd followup to JDK-8059557" https://bugs.openjdk.java.net/browse/JDK-8136766 "Enable ThreadStackSize range test” https://bugs.openjdk.java.net/browse/JDK-8134691 "CommandLineFlagConstraint::AtParse is not useful and not what we need" https://bugs.openjdk.java.net/browse/JDK-8138983 "Runtime: implement ranges for Shared*** flags"

Thank you.

More information about the hotspot-runtime-dev mailing list