Why does teeos panic during exit?

@rustbot

I was reading the platform sources for the runtime and I found this peculiar sys::cleanup implementation on teeos (aarch64-unknown-teeos):

rust/library/std/src/sys/pal/teeos/mod.rs

Lines 33 to 38 in 28e8a8c

    
           pub unsafe fn cleanup() { 
        
               unimplemented!() 
        
               // We do NOT have stack overflow handler, because TEE OS will kill TA when it happens. 
        
               // So cleanup is commented 
        
               // stack_overflow::cleanup(); 
        
           }

This function is always invoked after main. The unsupported.rs implementation is to just do nothing, but this is panicking instead. Even if it is somehow intentional, an rtabort call would surely be much more appropriate in this context, given that unimplemented involves a call to the global allocator and panic hook.

It just looks wrong. Am I missing something?

@rustbot label T-libs O-teeos

@Sword-Destiny

Added in #116565.
Cc @Sword-Destiny @petrochenkov

The comment above that says that Teeos uses dylibs rather than executables for "trusted applications", so cleanup() should in fact be unreachable. Using unreachable!() in init and cleanup would make sense in that case.

Interesting. However, the function is also called in std::process::exit (although IIRC that can panic anyway on multiple platforms). So it is reachable. ~~Edit: nvm, it panics before that.~~ Edit: nvm my nvm, cleanup is called first

In any case, this should not panic IMO.

And teeos is apparently one of these ^^

rust/library/std/src/sys/exit.rs

Lines 94 to 111 in 28e8a8c

    
           pub fn exit(code: i32) -> ! { 
        
               cfg_select! { 
        
                   target_os = "hermit" => unsafe { hermit_abi::exit(code) }, 
        
                   target_os = "linux" => unsafe { 
        
                       unique_thread_exit(); 
        
                       libc::exit(code) 
        
                   }, 
        
                   target_os = "motor" => moto_rt::process::exit(code), 
        
                   all(target_vendor = "fortanix", target_env = "sgx") => { 
        
                       crate::sys::pal::abi::exit_with_code(code as _) 
        
                   } 
        
                   target_os = "solid_asp3" => { 
        
                       rtabort!("exit({}) called", code) 
        
                   } 
        
                   target_os = "teeos" => { 
        
                       let _ = code; 
        
                       panic!("TA should not call `exit`") 
        
                   }

IMO, an unconditional panic for std::process::exit is also inappropriate and should be replaced with rtabort.

Navigation Menu

Repository navigation

Why does teeos panic during exit? #162847

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

	pub unsafe fn cleanup() {
	unimplemented!()
	// We do NOT have stack overflow handler, because TEE OS will kill TA when it happens.
	// So cleanup is commented
	// stack_overflow::cleanup();
	}

	pub fn exit(code: i32) -> ! {
	cfg_select! {
	target_os = "hermit" => unsafe { hermit_abi::exit(code) },
	target_os = "linux" => unsafe {
	unique_thread_exit();
	libc::exit(code)
	},
	target_os = "motor" => moto_rt::process::exit(code),
	all(target_vendor = "fortanix", target_env = "sgx") => {
	crate::sys::pal::abi::exit_with_code(code as _)
	}
	target_os = "solid_asp3" => {
	rtabort!("exit({}) called", code)
	}
	target_os = "teeos" => {
	let _ = code;
	panic!("TA should not call `exit`")
	}

Why does teeos panic during exit? #162847

Description

Activity

RalfJung commented on Sep 16, 2026

bjorn3 commented on Sep 16, 2026

maxdexh commented on Sep 16, 2026

maxdexh commented on Sep 16, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions