ppc64 has issues with multiple pagesizes in the same VMA. We currently
force the hugepages to be mapped for malloc at the next TB address, but
that is only necessary if the address is above the TB boundary. Fix that
up. Tested at run- and compile-time.
Signed-off-by: Nishanth Aravamudan <[EMAIL PR
Fixup our morecore implementation so that we deal with glibc trimming
correctly. A 0 increment is a request for the current heaptop and a
negative increment simply adjusts the heaptop (the return value is
actually ignored by glibc).
Signed-off-by: Nishanth Aravamudan <[EMAIL PROTECTED]>
diff --gi
During some benchmarking, it was noted that the current morecore
implementation will sometimes violate brk() semantics when the library
is PRELOAD'd. We always use hugetlbfs_vaddr_granularity() to bump to the
next appropriate mapping location for the heapbase. However, if any of
the linked-in libr
