Backtrace capture on SGX broken after (including) nightly-2023-11-20

[kostko]

It looks like that after #117895 (that seems to be the only SGX backtrace-related change between the two versions) enabling backtraces and attempting to capture one on the x86_64-fortanix-unknown-sgx target (sometimes) results in the enclave hanging.

Code

Unfortunately I don't currently have a minimal reproducer example, but roughly:

std::env::set_var("RUST_BACKTRACE", "1");
// ... unspecified logic, multiple nested frames ...
std::backtrace::Backtrace::capture();

I expected to see this happen: backtrace is captured without issues.

Instead, this happened: the Backtrace::capture() method appears to hang.

Version it worked on

It most recently worked on: nightly-2023-11-19

Version with regression

nightly-2023-11-20

[Nilstrieb]

cc @mzohreva

[Nilstrieb]

also @raoulstrackx @jethrogb as the other SGX maintainers

[apiraino]

WG-prioritization assigning priority (Zulip discussion).

@rustbot label -I-prioritize +P-medium

[BelalH]

Hello,
I am Belal Haikal an engineering manager at Fortanix, also colleague of Raoul and Jethro who were mentioned in a previous comment.
my team tried to reproduce the issue however we couldn't trigger the expected fault
is there a test service we can run to reproduce the issue?