Fixing multithreaded tests (#265)

udesou · web-flow · commit 263e7121a3c8 · 2025-08-29T15:35:41.000+10:00
This PR aims to fix multithreaded tests as noted in #179. To fix the issue, we needed to: - Capture all tasks in the shadow stack of other tasks; - Scan/conservatively process `scope`s and references in `__jmpbuf` in `task.eh`.
diff --git a/.github/scripts/ci-test-other.sh b/.github/scripts/ci-test-other.sh
@@ -23,9 +23,7 @@ declare -a max_moving_tests_to_skip=(
     "cmdlineargs"
     "Downloads"
     "read"
-    "threads"
     "LibCURL"
-    "rounding"
     "loading"
     "misc"
 )
@@ -63,8 +61,24 @@ if [[ $CHOOSE_TESTS_JL_CONTENT =~ $REGEX_PATTERN ]]; then
                 continue
             fi
 
+            if [[ $test =~ "rounding" ]]; then
+                # Run rounding test with single thread and Julia's 
+                # heap resizing (it OOMs with a fixed heap)
+                echo "-> Run"
+                ci_run_jl_test $test 1 $moving_feature
+                continue
+            fi
+
+            if [[ $test =~ "ranges" ]]; then
+                # Run ranges test with single thread and Julia's 
+                # heap resizing (it OOMs with a fixed heap)
+                echo "-> Run"
+                ci_run_jl_test $test 1 $moving_feature
+                continue
+            fi
+
             echo "-> Run"
-            ci_run_jl_test $test 1 $moving_feature
+            ci_run_jl_test $test 2 $moving_feature
         fi
     done
 else
diff --git a/mmtk/src/julia_types.rs b/mmtk/src/julia_types.rs
@@ -418,23 +418,26 @@ const _: () = {
 pub struct jl_thread_heap_common_t {
     pub weak_refs: small_arraylist_t,
     pub live_tasks: small_arraylist_t,
+    pub all_tasks: small_arraylist_t,
     pub mallocarrays: small_arraylist_t,
     pub free_stacks: [small_arraylist_t; 16usize],
 }
 #[allow(clippy::unnecessary_operation, clippy::identity_op)]
 const _: () = {
     ["Size of jl_thread_heap_common_t"]
-        [::std::mem::size_of::<jl_thread_heap_common_t>() - 1216usize];
+        [::std::mem::size_of::<jl_thread_heap_common_t>() - 1280usize];
     ["Alignment of jl_thread_heap_common_t"]
         [::std::mem::align_of::<jl_thread_heap_common_t>() - 8usize];
     ["Offset of field: jl_thread_heap_common_t::weak_refs"]
         [::std::mem::offset_of!(jl_thread_heap_common_t, weak_refs) - 0usize];
     ["Offset of field: jl_thread_heap_common_t::live_tasks"]
         [::std::mem::offset_of!(jl_thread_heap_common_t, live_tasks) - 64usize];
+    ["Offset of field: jl_thread_heap_common_t::all_tasks"]
+        [::std::mem::offset_of!(jl_thread_heap_common_t, all_tasks) - 128usize];
     ["Offset of field: jl_thread_heap_common_t::mallocarrays"]
-        [::std::mem::offset_of!(jl_thread_heap_common_t, mallocarrays) - 128usize];
+        [::std::mem::offset_of!(jl_thread_heap_common_t, mallocarrays) - 192usize];
     ["Offset of field: jl_thread_heap_common_t::free_stacks"]
-        [::std::mem::offset_of!(jl_thread_heap_common_t, free_stacks) - 192usize];
+        [::std::mem::offset_of!(jl_thread_heap_common_t, free_stacks) - 256usize];
 };
 #[repr(C)]
 #[derive(Debug)]
@@ -480,13 +483,13 @@ pub struct jl_gc_tls_states_common_t {
 #[allow(clippy::unnecessary_operation, clippy::identity_op)]
 const _: () = {
     ["Size of jl_gc_tls_states_common_t"]
-        [::std::mem::size_of::<jl_gc_tls_states_common_t>() - 1280usize];
+        [::std::mem::size_of::<jl_gc_tls_states_common_t>() - 1344usize];
     ["Alignment of jl_gc_tls_states_common_t"]
         [::std::mem::align_of::<jl_gc_tls_states_common_t>() - 8usize];
     ["Offset of field: jl_gc_tls_states_common_t::heap"]
         [::std::mem::offset_of!(jl_gc_tls_states_common_t, heap) - 0usize];
     ["Offset of field: jl_gc_tls_states_common_t::gc_num"]
-        [::std::mem::offset_of!(jl_gc_tls_states_common_t, gc_num) - 1216usize];
+        [::std::mem::offset_of!(jl_gc_tls_states_common_t, gc_num) - 1280usize];
 };
 #[repr(C)]
 #[derive(Debug, Copy, Clone)]
@@ -675,7 +678,7 @@ pub struct _jl_tls_states_t {
 }
 #[allow(clippy::unnecessary_operation, clippy::identity_op)]
 const _: () = {
-    ["Size of _jl_tls_states_t"][::std::mem::size_of::<_jl_tls_states_t>() - 3616usize];
+    ["Size of _jl_tls_states_t"][::std::mem::size_of::<_jl_tls_states_t>() - 3680usize];
     ["Alignment of _jl_tls_states_t"][::std::mem::align_of::<_jl_tls_states_t>() - 8usize];
     ["Offset of field: _jl_tls_states_t::tid"]
         [::std::mem::offset_of!(_jl_tls_states_t, tid) - 0usize];
@@ -702,51 +705,51 @@ const _: () = {
     ["Offset of field: _jl_tls_states_t::gc_tls_common"]
         [::std::mem::offset_of!(_jl_tls_states_t, gc_tls_common) - 1800usize];
     ["Offset of field: _jl_tls_states_t::lazily_freed_mtarraylist_buffers"]
-        [::std::mem::offset_of!(_jl_tls_states_t, lazily_freed_mtarraylist_buffers) - 3080usize];
+        [::std::mem::offset_of!(_jl_tls_states_t, lazily_freed_mtarraylist_buffers) - 3144usize];
     ["Offset of field: _jl_tls_states_t::defer_signal"]
-        [::std::mem::offset_of!(_jl_tls_states_t, defer_signal) - 3144usize];
+        [::std::mem::offset_of!(_jl_tls_states_t, defer_signal) - 3208usize];
     ["Offset of field: _jl_tls_states_t::current_task"]
-        [::std::mem::offset_of!(_jl_tls_states_t, current_task) - 3152usize];
+        [::std::mem::offset_of!(_jl_tls_states_t, current_task) - 3216usize];
     ["Offset of field: _jl_tls_states_t::next_task"]
-        [::std::mem::offset_of!(_jl_tls_states_t, next_task) - 3160usize];
+        [::std::mem::offset_of!(_jl_tls_states_t, next_task) - 3224usize];
     ["Offset of field: _jl_tls_states_t::previous_task"]
-        [::std::mem::offset_of!(_jl_tls_states_t, previous_task) - 3168usize];
+        [::std::mem::offset_of!(_jl_tls_states_t, previous_task) - 3232usize];
     ["Offset of field: _jl_tls_states_t::root_task"]
-        [::std::mem::offset_of!(_jl_tls_states_t, root_task) - 3176usize];
+        [::std::mem::offset_of!(_jl_tls_states_t, root_task) - 3240usize];
     ["Offset of field: _jl_tls_states_t::timing_stack"]
-        [::std::mem::offset_of!(_jl_tls_states_t, timing_stack) - 3184usize];
+        [::std::mem::offset_of!(_jl_tls_states_t, timing_stack) - 3248usize];
     ["Offset of field: _jl_tls_states_t::stackbase"]
-        [::std::mem::offset_of!(_jl_tls_states_t, stackbase) - 3192usize];
+        [::std::mem::offset_of!(_jl_tls_states_t, stackbase) - 3256usize];
     ["Offset of field: _jl_tls_states_t::stacksize"]
-        [::std::mem::offset_of!(_jl_tls_states_t, stacksize) - 3200usize];
+        [::std::mem::offset_of!(_jl_tls_states_t, stacksize) - 3264usize];
     ["Offset of field: _jl_tls_states_t::sig_exception"]
-        [::std::mem::offset_of!(_jl_tls_states_t, sig_exception) - 3208usize];
+        [::std::mem::offset_of!(_jl_tls_states_t, sig_exception) - 3272usize];
     ["Offset of field: _jl_tls_states_t::bt_data"]
-        [::std::mem::offset_of!(_jl_tls_states_t, bt_data) - 3216usize];
+        [::std::mem::offset_of!(_jl_tls_states_t, bt_data) - 3280usize];
     ["Offset of field: _jl_tls_states_t::bt_size"]
-        [::std::mem::offset_of!(_jl_tls_states_t, bt_size) - 3224usize];
+        [::std::mem::offset_of!(_jl_tls_states_t, bt_size) - 3288usize];
     ["Offset of field: _jl_tls_states_t::profiling_bt_buffer"]
-        [::std::mem::offset_of!(_jl_tls_states_t, profiling_bt_buffer) - 3232usize];
+        [::std::mem::offset_of!(_jl_tls_states_t, profiling_bt_buffer) - 3296usize];
     ["Offset of field: _jl_tls_states_t::signal_request"]
-        [::std::mem::offset_of!(_jl_tls_states_t, signal_request) - 3240usize];
+        [::std::mem::offset_of!(_jl_tls_states_t, signal_request) - 3304usize];
     ["Offset of field: _jl_tls_states_t::io_wait"]
-        [::std::mem::offset_of!(_jl_tls_states_t, io_wait) - 3244usize];
+        [::std::mem::offset_of!(_jl_tls_states_t, io_wait) - 3308usize];
     ["Offset of field: _jl_tls_states_t::signal_stack"]
-        [::std::mem::offset_of!(_jl_tls_states_t, signal_stack) - 3248usize];
+        [::std::mem::offset_of!(_jl_tls_states_t, signal_stack) - 3312usize];
     ["Offset of field: _jl_tls_states_t::signal_stack_size"]
-        [::std::mem::offset_of!(_jl_tls_states_t, signal_stack_size) - 3256usize];
+        [::std::mem::offset_of!(_jl_tls_states_t, signal_stack_size) - 3320usize];
     ["Offset of field: _jl_tls_states_t::system_id"]
-        [::std::mem::offset_of!(_jl_tls_states_t, system_id) - 3264usize];
+        [::std::mem::offset_of!(_jl_tls_states_t, system_id) - 3328usize];
     ["Offset of field: _jl_tls_states_t::suspend_count"]
-        [::std::mem::offset_of!(_jl_tls_states_t, suspend_count) - 3272usize];
+        [::std::mem::offset_of!(_jl_tls_states_t, suspend_count) - 3336usize];
     ["Offset of field: _jl_tls_states_t::finalizers"]
-        [::std::mem::offset_of!(_jl_tls_states_t, finalizers) - 3280usize];
+        [::std::mem::offset_of!(_jl_tls_states_t, finalizers) - 3344usize];
     ["Offset of field: _jl_tls_states_t::previous_exception"]
-        [::std::mem::offset_of!(_jl_tls_states_t, previous_exception) - 3536usize];
+        [::std::mem::offset_of!(_jl_tls_states_t, previous_exception) - 3600usize];
     ["Offset of field: _jl_tls_states_t::locks"]
-        [::std::mem::offset_of!(_jl_tls_states_t, locks) - 3544usize];
+        [::std::mem::offset_of!(_jl_tls_states_t, locks) - 3608usize];
     ["Offset of field: _jl_tls_states_t::engine_nqueued"]
-        [::std::mem::offset_of!(_jl_tls_states_t, engine_nqueued) - 3608usize];
+        [::std::mem::offset_of!(_jl_tls_states_t, engine_nqueued) - 3672usize];
 };
 pub type jl_function_t = jl_value_t;
 pub type jl_timing_block_t = _jl_timing_block_t;
@@ -3520,10 +3523,3 @@ const _: () = {
     ["Align of template specialization: pinned_ref_open0_jl_value_t_close0"]
         [::std::mem::align_of::<pinned_ref<jl_value_t>>() - 8usize];
 };
-#[allow(clippy::unnecessary_operation, clippy::identity_op)]
-const _: () = {
-    ["Size of template specialization: pinned_ref_open0_jl_value_t_close0"]
-        [::std::mem::size_of::<pinned_ref<jl_value_t>>() - 8usize];
-    ["Align of template specialization: pinned_ref_open0_jl_value_t_close0"]
-        [::std::mem::align_of::<pinned_ref<jl_value_t>>() - 8usize];
-};
diff --git a/mmtk/src/scanning.rs b/mmtk/src/scanning.rs
@@ -62,7 +62,10 @@ impl Scanning<JuliaVM> for VMScanning {
         crate::conservative::mmtk_conservative_scan_ptls_registers(ptls);
 
         // Scan thread local from ptls: See gc_queue_thread_local in gc.c
-        let mut root_scan_task = |task: *const _jl_task_t, task_is_root: bool| {
+
+        // scan task object to conservatively pin references from shadow,
+        // gc_preserve and regular stack and registers
+        let mut scan_root_task = |task: *const _jl_task_t| {
             if !task.is_null() {
                 // Scan gc preserve and shadow stacks
                 unsafe {
@@ -92,40 +95,41 @@ impl Scanning<JuliaVM> for VMScanning {
                     crate::conservative::mmtk_conservative_scan_task_registers(task);
                 }
                 // }
+            };
+        };
 
-                if task_is_root {
-                    // captures wrong root nodes before creating the work
-                    debug_assert!(
-                        Address::from_ptr(task).as_usize() % 16 == 0
-                            || Address::from_ptr(task).as_usize() % 8 == 0,
-                        "root node {:?} is not aligned to 8 or 16",
-                        Address::from_ptr(task)
-                    );
+        // adding task object as root
+        let mut add_task_as_root = |task: *const _jl_task_t| {
+            if !task.is_null() {
+                // captures wrong root nodes before creating the work
+                debug_assert!(
+                    Address::from_ptr(task).as_usize() % 16 == 0
+                        || Address::from_ptr(task).as_usize() % 8 == 0,
+                    "root node {:?} is not aligned to 8 or 16",
+                    Address::from_ptr(task)
+                );
 
-                    // unsafe: We checked `!task.is_null()` before.
-                    let objref = unsafe {
-                        ObjectReference::from_raw_address_unchecked(Address::from_ptr(task))
-                    };
-                    node_buffer.push(objref);
-                }
+                let objref =
+                    unsafe { ObjectReference::from_raw_address_unchecked(Address::from_ptr(task)) };
+                node_buffer.push(objref);
             }
         };
-        root_scan_task(ptls.root_task, true);
 
-        // need to iterate over live tasks as well to process their shadow stacks
-        // we should not set the task themselves as roots as we will know which ones are still alive after GC
+        add_task_as_root(ptls.root_task);
+        add_task_as_root(ptls.current_task as *mut _jl_task_t);
+        add_task_as_root(ptls.next_task);
+        add_task_as_root(ptls.previous_task);
+
+        // need to iterate over all tasks as well to process their shadow stacks and scan their stacks
         let mut i = 0;
-        while i < ptls.gc_tls_common.heap.live_tasks.len {
-            let mut task_address = Address::from_ptr(ptls.gc_tls_common.heap.live_tasks.items);
+        while i < ptls.gc_tls_common.heap.all_tasks.len {
+            let mut task_address = Address::from_ptr(ptls.gc_tls_common.heap.all_tasks.items);
             task_address = task_address.shift::<Address>(i as isize);
             let task = unsafe { task_address.load::<*const jl_task_t>() };
-            root_scan_task(task, false);
+            scan_root_task(task);
             i += 1;
         }
 
-        root_scan_task(ptls.current_task as *mut _jl_task_t, true);
-        root_scan_task(ptls.next_task, true);
-        root_scan_task(ptls.previous_task, true);
         if !ptls.previous_exception.is_null() {
             node_buffer.push(unsafe {
                 // unsafe: We have just checked `ptls.previous_exception` is not null.
