Mercurial > hg > truffle
changeset 5937:cf956638b844
7151089: PS NUMA: NUMA allocator should not attempt to free pages when using SHM large pages
Summary: Don't attempt to uncommit SHM-based large pages
Reviewed-by: kvn
author | iveresov |
---|---|
date | Mon, 12 Mar 2012 13:12:07 -0700 |
parents | fde683df4c27 |
children | 2c0751569716 |
files | src/os/linux/vm/os_linux.cpp |
diffstat | 1 files changed, 8 insertions(+), 1 deletions(-) [+] |
line wrap: on
line diff
--- a/src/os/linux/vm/os_linux.cpp Fri Mar 09 13:34:45 2012 -0800 +++ b/src/os/linux/vm/os_linux.cpp Mon Mar 12 13:12:07 2012 -0700 @@ -2547,7 +2547,14 @@ } void os::free_memory(char *addr, size_t bytes, size_t alignment_hint) { - commit_memory(addr, bytes, alignment_hint, false); + // This method works by doing an mmap over an existing mmaping and effectively discarding + // the existing pages. However it won't work for SHM-based large pages that cannot be + // uncommitted at all. We don't do anything in this case to avoid creating a segment with + // small pages on top of the SHM segment. This method always works for small pages, so we + // allow that in any case. + if (alignment_hint <= (size_t)os::vm_page_size() || !UseSHM) { + commit_memory(addr, bytes, alignment_hint, false); + } } void os::numa_make_global(char *addr, size_t bytes) {