weird ppc64 out of memory on startup.

On a Book-E ppc64.
Same CPU running as a Book-E ppc32 seems fine?
There is no shortage of free mem.
Most of the times I try I can't get it to start. With a strace like below.
Redoing the exact same command back to back will eventually get it started.
Seems super early. It's just off the glibc initialization and into the process.
I'll probably stick a debugger into it tomorrow, but thought someone might have an idea what's going on here.

...
set_tid_address(0x3fff812bb0f0) = 922
set_robust_list(0x3fff812bb100, 24) = 0
rseq(0x3fff812bb740, 0x20, 0, 0xfe5000b) = 0
mprotect(0x3fff81116000, 106496, PROT_READ) = 0
mprotect(0x3fff8116e000, 8192, PROT_READ) = 0
mprotect(0x3fff8123b000, 20480, PROT_READ) = 0
mprotect(0x11deb9000, 28672, PROT_READ) = 0
mprotect(0x3fff812bf000, 16384, PROT_READ) = 0
prlimit64(0, RLIMIT_STACK, NULL, {rlim_cur=8192*1024, rlim_max=RLIM64_INFINITY}) = 0
getrandom("\xba\x56\xd9\x6a\x5c\x8a\x24\x00\xf1\xdf\xea\xf0\xc3\x48\x13\xe4\x38\x72\x84\x70\xca\x74\xb8\xb2\x5e\x48\x13\x3d\x29\x35\x80\xe1", 32, 0) = 32
mmap(NULL, 131072, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x3fff80ee0000
munmap(0x3fff80ee0000, 131072) = 0
mmap(0xc8867000, 131072, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0xc8867000
write(2, "/sbin/luajit", 12/sbin/luajit) = 12
write(2, ":", 1:) = 1
write(2, " ", 1 ) = 1
write(2, "cannot create state: not enough "..., 38cannot create state: not enough memory) = 38
write(2, "\n", 1
) = 1
exit_group(1) = ?
+++ exited with 1 +++

Start with the provided strace and reproduce the startup failure on the Book-E ppc64 system, then compare the same command on ppc32. Investigate the early startup path around the mmap and reported state-allocation failure. Done means the program starts reliably on ppc64 without reporting insufficient memory when free memory is available.