[AIX] Fix hangs during testing

[mustartt]

Fixes all current test hangs experienced during CI runs.

1. ipv6 link-local (the loopback device) gets assigned an automatic zone id of 1, causing the assert to fail and hang in library/std/src/net/udp/tests.rs
2. Const alloc does not fail gracefully
3. Debuginfo test has problem with gdb auto load safe path

[rustbot]

r? @ChrisDenton

rustbot has assigned @ChrisDenton.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[daltenty]

LGTM from the AIX perspective. These test cases hang the test run indefinitely at the moment, so this unblocks regular runs.

[workingjubilee]

If the system behaves badly on large allocations, then there is nothing to fix here.

Suggested change

	//@ ignore-aix: FIXME(#137966)
	//@ ignore-aix: alloc failure on AIX can result in SIGKILL instead of nullptr

[workingjubilee]

@daltenty Do you have any idea why it sometimes hangs and sometimes SIGKILLs?

[mustartt]

@daltenty Do you have any idea why it sometimes hangs and sometimes SIGKILLs?

It not exactly an "hang". mmap and zero initializing the mapped region takes quite a while on our dev machines which either times out our CI or get SIGKILL'd after a very long time.

[workingjubilee]

...Is the problem that you literally have 128TiB of RAM?

[workingjubilee]

Hm, wait... laziness in paging due to overcommit, resulting in the system accepting an allocation that can't possibly be respected if called but assuming that no one will actually call that bluff?

[workingjubilee]

Suggested change

	// AIX will allow allow the allocation to go through, and get SIGKILL when zero initializing
	// AIX will allow the allocation to go through, and get SIGKILL when zero initializing

[workingjubilee]

Suggested change

	// the overcommited page.
	// the overcommitted page.

[workingjubilee]

Suggested change

	// AIX will allow allow the allocation to go through, and get SIGKILL when zero initializing
	// the overcommited page.
	// AIX will allow the allocation to go through, and get SIGKILL when zero initializing
	// the overcommitted page.

[workingjubilee]

address nits, squash, and then r=me

[mustartt]

Yes the sigkills are from lazy paging and OS overcommitting.

Addressed nit and squashed.

[workingjubilee]

Hm, wait... what are the actual mismatches on the scope ID, exactly?

They have the same scope, but a different zone index? Because it's assigned the 0 ZoneID...? But we create the IP with a scope_id of 0...

[mustartt]

I think included the wrong scope id in the post. The peer we get back contains a scope id of 1 for the loopback.
(edited PR description to scope id to reflect it)

(gdb) list
41              });
42      
43              let server = t!(UdpSocket::bind(&server_ip));
44              tx1.send(()).unwrap();
45              let mut buf = [0];
46              let (nread, src) = t!(server.recv_from(&mut buf));
47              assert_eq!(nread, 1);
48              assert_eq!(buf[0], 99);
49              assert_eq!(compare_ignore_zoneid(&src, &client_ip), true);
50              rx2.recv().unwrap();
(gdb) p src
$2 = core::net::socket_addr::SocketAddr::V6(core::net::socket_addr::SocketAddrV6 {ip: core::net::ip_addr::Ipv6Addr {octets: [0 <repeats 15 times>, 1]}, port: 19603, flowinfo: 0, scope_id: 1})
(gdb) p client_ip
$3 = core::net::socket_addr::SocketAddr::V6(core::net::socket_addr::SocketAddrV6 {ip: core::net::ip_addr::Ipv6Addr {octets: [0 <repeats 15 times>, 1]}, port: 19603, flowinfo: 0, scope_id: 0})

bash-5.2$ cat /etc/hosts | grep ::1
::1                     loopback localhost      # IPv6 loopback (lo0) name/address
bash-5.2$ ibm-clang++_r test.cpp -o test
bash-5.2$ cat test.cpp
#include <net/if.h>
#include <iostream>
#include <sysexits.h>

int main(void)
{
        auto scope_id = if_nametoindex("lo0");
        std::cout << scope_id << std::endl;
        return EX_OK;
}
bash-5.2$ ./test 
1

[workingjubilee]

Aha.

We probably should be creating these with a scope ID of 1 for the loopback.

[workingjubilee]

For now this change is fine though.

[mustartt]

Fixed nits
@bors r=workingjubilee

[bors]

@mustartt: 🔑 Insufficient privileges: Not in reviewers

[workingjubilee]

Oh, sorry

@bors r+ rollup

[bors]

📌 Commit 2a7ad95 has been approved by workingjubilee

It is now in the queue for this repository.