Don't turn on transparent huge pages (-XX:+UseTransparentHugePages) unless explicitly specified on the command line. This has the effect that large pages are never turned on Linux unless the user has explicitly enabled any of the large pages flags: -XX:+UseLargePages, -XX:+UseTransparentHugePages, -XX:+UseHugeTLBFS, and -XX:+UseSHM.
Reviewed-by: jwilhelm, tschatzl, brutisso
Use conservative assumptions of required alignment for the various garbage collector components into account when determining the maximum heap size that supports compressed oops. Using this conservative value avoids several circular dependencies in the calculation.
Reviewed-by: stefank, dholmes
When using NUMA and large pages we need to ease the requirement on which node the memory should be allocated on. To avoid the SIGBUS we now use the memory policy MPOL_PREFERRED, which prefers a certain node, instead of MPOL_BIND, which requires a certain node.
Reviewed-by: jmasa, pliden
Replace the broken large pages implementation on Linux. New flag: -XX:+UseTransparentHugePages - Linux specific flag to turn on transparent huge page hinting with madvise(..., MAP_HUGETLB). Changed behavior: -XX:+UseLargePages - tries to use -XX:+UseTransparentHugePages before trying other large pages implementations (on Linux). Changed behavior: -XX:+UseHugeTLBFS - Use upfront allocation of Large Pages instead of using the broken implementation to dynamically committing large pages. Changed behavior: -XX:LargePageSizeInBytes - Turned off the ability to use this flag on Linux and provides warning to user if set to a value different than the OS chosen large page size. Changed behavior: Setting no large page size - Now defaults to use -XX:UseTransparentHugePages if the OS supports it. Previously, -XX:+UseHugeTLBFS was chosen if the OS was configured to use large pages.
Reviewed-by: tschatzl, dcubed, brutisso
Intermittent java.io.IOException: Bad file number during HotSpotVirtualMachine.executeCommand
Reviewed-by: dcubed, dholmes, sspitsyn, mgerdin, ctornqvi, dsamersoff
Dl_info struct should only be used if dladdr() has returned non-zero (no errors) and always check the dladdr() return value; Dl_info.dli_sname and Dl_info.dli_saddr fields should only be used if non-NULL; update/improve runtime/6888954/vmerrors.sh test
Reviewed-by: dsamersoff, zgu, hseigel, coleenp
Detect mmap() commit failures in Linux and Solaris os::commit_memory() impls and call vm_exit_out_of_memory(). Add os::commit_memory_or_exit(). Also tidy up some NMT accounting and some mmap() return value checking.
Reviewed-by: zgu, stefank, dholmes, dsamersoff
Fixed the output of G1SummarizeRSetStats: too small datatype for the number of concurrently processed cards, added concurrent remembered set thread time retrieval for Linux and Windows (BSD uses os::elapsedTime() now), and other cleanup. The information presented during VM operation is now relative to the previous output, not always cumulative if G1SummarizeRSetStatsPeriod > 0. At VM exit, the code prints a cumulative summary.
Reviewed-by: johnc, jwilhelm
Ergonomics now also takes available virtual memory into account when deciding for a heap size. The helper method to determine the maximum allocatable memory block now uses the appropriate OS specific calls to retrieve available virtual memory for the java process. In 32 bit environments this method now also searches for the maximum actually reservable amount of memory. Merge previously separate implementations for Linux/BSD/Solaris into a single method.
Reviewed-by: jmasa, tamao
Fix the build without precompiled headers by either including the missing ".inline.hpp" files into the appropriate files or by turning inline-functions declared in header files into ordinary functions in ".cpp" files.
Reviewed-by: coleenp, stefank, dholmes
Enhanced virtual memory tracking to track committed regions as well as reserved regions, so NMT now can generate virtual memory map.
Reviewed-by: acorn, coleenp
