Discussion of the improvement of libctest

[LIN-Matrix]

1. Implementing TLS for Cross-Architecture Compatibility

To enable compatibility with both x86_64 and aarch64, Thread-Local Storage (TLS) support is introduced and integrated into task context management. TLS is crucial for handling per-thread data and is frequently accessed by the C library during runtime.

Core Mechanism

The TLS area is managed through the fs_base field within the TaskContext structure. Two accessor methods are implemented:

pub fn tls(&self) -> VirtAddr {
    VirtAddr::from(self.fs_base)
}

pub fn set_tls(&mut self, tls_area: VirtAddr) {
    self.fs_base = tls_area.as_usize();
}

These functions respectively read from and write to the fs_base, which stores the base address of the current task’s TLS region. This mechanism is crucial for proper management of thread-local data.

Integration with Context Switching

During a context switch, TLS pointers must be saved and restored to maintain task-specific state:

#[cfg(any(feature = "tls", feature = "uspace"))]
unsafe {
    self.fs_base = super::read_thread_pointer();
    super::write_thread_pointer(next_ctx.fs_base);
}

This ensures that each task regains its thread-local data when it resumes execution. Moreover, clone_task() is modified to explicitly set the TLS base for new tasks by invoking:

new_task.ctx_mut().set_tls(axhal::arch::read_thread_pointer().into());

Without this step, cloned tasks would lose access to TLS, leading to unpredictable behavior, especially in libc-based environments.

---

2. Enhancing System Call Compatibility

Several system calls essential to libc startup routines were either implemented or stubbed to ensure smooth execution of dynamically linked applications.

Key Additions

• statfs: Returns dummy EXT4 values, allowing filesystem checks to pass.
• getppid, gettid, set_tid_address: Return reasonable defaults used by libc.
• prlimit64, rt_sigtimedwait: Stub implementations returning success to bypass failures during initialization.

These system calls are commonly invoked when programs initialize standard I/O, memory allocation, and environment parsing routines.

Dispatcher Updates

To ensure proper routing, all the newly supported system calls are registered in the syscall dispatcher. This eliminates the fallback to the ENOSYS path and avoids terminating tasks prematurely.

---

3. Adapting the Main Function for Structured Output

The structure of the main() function is modified to accommodate test output parsing during automated evaluation. The original approach merged all tests in a single loop, injecting markers based on case index to delineate test groups:

if i == 32 {
    println!("#### OS COMP TEST GROUP END basic-musl ####");
    println!("#### OS COMP TEST GROUP END basic-glibc ####");
    println!("#### OS COMP TEST GROUP START libctest-glibc ####");
    println!("#### OS COMP TEST GROUP START libctest-musl ####");
} else if i == 161 {
    println!("#### OS COMP TEST GROUP END libctest-musl ####");
    println!("#### OS COMP TEST GROUP END libctest-glibc ####");
}

This approach works without requiring additional environment variables, but it relies on knowing the precise number of test cases in each group, which can be fragile if the test set changes.

---

4. Alternative Grouping Strategy

Before adopting the above design, a different method was explored based on BattiestStone4’s approach. That strategy created separate environment variables and test case lists for each group (basic, libctest-musl, libctest-glibc), allowing the test runner to handle them in two distinct loops:

let testcases = option_env!("AX_TESTCASES_LIST")...;
let testcases = option_env!("AX_MUSL_LIBCTEST_TESTCASES_LIST")...;

This method is more robust when the number of test cases varies but introduces additional Makefile complexity and slight duplication in code structure.

---

5. Summary of Improvements

The following table outlines the major modifications and their impact:

Change Area	Description	Purpose
TLS Handling	Added tls() and set_tls() accessors and integrated into context switch	Enables thread-local storage for dynamic language runtimes
clone_task Update	Sets TLS pointer in new tasks	Ensures correct behavior of multi-threaded applications
System Call Extensions	Implemented or stubbed several libc-required syscalls	Prevents initialization failures during startup of dynamic binaries
Main Function Refactor	Adds group markers based on case index	Enables structured parsing of test outputs for evaluation
Grouped Test Strategy	Compared index-based and env-variable based grouping	Explored trade-offs between flexibility and simplicity

Together, these changes significantly improve the system’s support for POSIX-compliant applications and ensure compatibility with libctest, making ArceOS a more robust and versatile kernel for multi-architecture deployment.