8024669: Native OOME when allocating after changes to maximum heap supporting Coops sizing on sparcv9

After changes in 8010722 the ergonomics for calculating the size of the heap that supports zero based compressed oops changed. This lead to the VM actually using zero based compressed oops. Due to low default HeapBaseMinAddress, the OS mapping in the application image at the same address, and limitations of the malloc implementation on Solaris this resulted in very little C heap available for the VM. So the VM immediately gives a native OOME when the machine has lots of physical memory (>=32G). The solution is to increase the HeapBaseMinAddress so that the VM has enough C heap.

Reviewed-by: kvn, brutisso
This commit is contained in:
Thomas Schatzl 2013-09-18 13:18:52 +02:00
parent 3e4a59f797
commit f6e98b8181

View file

@ -35,7 +35,9 @@ define_pd_global(intx, CompilerThreadStackSize, 0);
// Used on 64 bit platforms for UseCompressedOops base address
#ifdef _LP64
define_pd_global(uintx, HeapBaseMinAddress, CONST64(4)*G);
// use 6G as default base address because by default the OS maps the application
// to 4G on Solaris-Sparc. This leaves at least 2G for the native heap.
define_pd_global(uintx, HeapBaseMinAddress, CONST64(6)*G);
#else
define_pd_global(uintx, HeapBaseMinAddress, 2*G);
#endif