Running current Zero with gtests:

$ CONF=linux-x86_64-zero-fastdebug make exploded-test TEST=jtreg:gtest/

# A fatal error has been detected by the Java Runtime Environment:
#
# Internal Error (/home/shade/trunks/jdk/src/hotspot/os/linux/os_linux.cpp:3227), pid=562477, tid=562477
# assert(imin <= imax) failed: Unexpected page size

The actual problem is underflow in os::pd_create_stack_guard_pages() here:

stack_extent = (uintptr_t) get_stack_commited_bottom(
os::Linux::initial_thread_stack_bottom(),
(size_t)addr - stack_extent);

For Zero x86_64, these are off by a page, so the (addr - stack_extent) underflows, then get_stack_commited_bottom gets called with super-large size, which then fails the assert above. The assert should check for underflow specifically.

[0.002s][warning][os] Addr: 0x00007fff35e73000, stack_extent: 0x00007fff35e74000

I believe the root cause is primordial thread (which is why it is only seen in gtests?) having a weird stack base, which is explicitly handled by the shared code commonned by JDK-8169373. I see little reason not to use the same code for Zero. There is the IA64 block in Zero version that handles the double stack on Itanium, but I think we can sacrifice it since Itanium is discontinued.