Merge pull request #300 from insertinterestingnamehere/msan

Memory Sanitizer Fixes

.github/workflows/CI.yml

@@ -240,6 +240,11 @@ jobs:
         scheduler: [nemesis, sherwood, distrib]
         topology: [hwloc, binders, no]
         use_libcxx: [false] # disable testing on libcxx since its effect seems very limited for now.
+        exclude:
+          - sanitizer: memory
+            topology: hwloc
+          - sanitizer: memory
+            topology: binders
     env:
       CC: clang-19
       CXX: clang++-19

include/qt_macros.h

@@ -22,6 +22,12 @@
 #define Q_UNUSED(x) x
 #endif
 
+#if defined(__has_feature)
+#if __has_feature(memory_sanitizer)
+#define QTHREAD_MSAN
+#endif
+#endif
+
 #define TLS_DECL(type, name) thread_local type name
 #define TLS_DECL_INIT(type, name) thread_local type name = 0
 #define TLS_GET(name) name

src/fastcontext/context.c

@@ -16,6 +16,7 @@
 #include "qthread/common.h"
 
 #include "qt_asserts.h"
+#include "qt_macros.h"
 #include "qt_prefetch.h"
 #include "qt_visibility.h"
 
@@ -149,7 +150,9 @@ QT_SKIP_THREAD_SANITIZER int INTERNAL qt_swapctxt(uctxt_t *oucp, uctxt_t *ucp) {
   if (getcontext(oucp) == 0) {
 #if ((QTHREAD_ASSEMBLY_ARCH == QTHREAD_IA32) ||                                \
      (QTHREAD_ASSEMBLY_ARCH == QTHREAD_AMD64))
+#ifndef QTHREAD_MSAN
     Q_PREFETCH((void *)ucp->mc.mc_esp, 1, 3);
+#endif
 #endif
     setcontext(ucp);
   }

src/threadqueues/distrib_threadqueues.c

@@ -88,6 +88,7 @@ static qt_threadqueue_t *alloc_threadqueue(void) {
   t->t = qt_malloc(sizeof(qt_threadqueue_internal) * t->num_queues);
   t->w_inds =
     qt_calloc(qlib->nshepherds * qlib->nworkerspershep, sizeof(w_ind));
+  atomic_store_explicit(&t->numwaiters, 0ull, memory_order_relaxed);
   return t;
 }
 

test/basics/tasklocal_data.c

@@ -4,6 +4,8 @@
 #include <stdlib.h>
 #include <unistd.h>
 
+int *main_local = NULL;
+
 static aligned_t use_default_space(void *arg) {
   unsigned int size = 64 * sizeof(int);
 
@@ -15,10 +17,9 @@ static aligned_t use_default_space(void *arg) {
           qthread_size_tasklocal());
   iprintf("use_default_space(): local_int_arr at %p\n", local_int_arr);
   local_int_arr = (int *)qthread_get_tasklocal(size);
-  test_check(42 != local_int_arr[42]);
-  iprintf(
-    "use_default_space(): retrieved tasklocal data successfully (42 != %d)\n",
-    local_int_arr[42]);
+  test_check(main_local != local_int_arr);
+  iprintf("use_default_space(): retrieved tasklocal data successfully "
+          "(distinct from main thread's local)\n");
   local_int_arr = NULL;
 
   return 0;
@@ -35,10 +36,9 @@ static aligned_t use_allocated_space(void *arg) {
           qthread_size_tasklocal());
   iprintf("use_allocated_space(): local_int_arr at %p\n", local_int_arr);
   local_int_arr = (int *)qthread_get_tasklocal(size);
-  test_check(42 != local_int_arr[42]);
-  iprintf(
-    "use_allocated_space(): retrieved tasklocal data successfully (42 != %d)\n",
-    local_int_arr[42]);
+  test_check(main_local != local_int_arr);
+  iprintf("use_allocated_space(): retrieved tasklocal data successfully "
+          "(distinct from main thread's local)\n");
   local_int_arr = NULL;
 
   return 0;
@@ -121,6 +121,7 @@ int main(int argc, char *argv[]) {
     iprintf("main(): size_tasklocal() is %d\n", qthread_size_tasklocal());
     iprintf("main(): local_int_arr at %p\n", local_int_arr);
     local_int_arr = (int *)qthread_get_tasklocal(size);
+    main_local = local_int_arr;
     test_check(42 == local_int_arr[42]);
     iprintf("main(): retrieved tasklocal data successfully (42 == %d)\n",
             local_int_arr[42]);

test/basics/tasklocal_data_no_argcopy.c

@@ -5,6 +5,8 @@
 #include <stdlib.h>
 #include <unistd.h>
 
+int *main_local = NULL;
+
 static aligned_t use_default_space(void *arg) {
   unsigned int size = 64 * sizeof(int);
 
@@ -16,10 +18,9 @@ static aligned_t use_default_space(void *arg) {
           qthread_size_tasklocal());
   iprintf("use_default_space(): local_int_arr at %p\n", local_int_arr);
   local_int_arr = (int *)qthread_get_tasklocal(size);
-  test_check(42 != local_int_arr[42]);
-  iprintf(
-    "use_default_space(): retrieved tasklocal data successfully (42 != %d)\n",
-    local_int_arr[42]);
+  test_check(main_local != local_int_arr);
+  iprintf("use_default_space(): retrieved tasklocal data successfully "
+          "(distinct from main thread's local)\n");
   local_int_arr = NULL;
 
   return 0;
@@ -36,10 +37,9 @@ static aligned_t use_allocated_space(void *arg) {
           qthread_size_tasklocal());
   iprintf("use_allocated_space(): local_int_arr at %p\n", local_int_arr);
   local_int_arr = (int *)qthread_get_tasklocal(size);
-  test_check(42 != local_int_arr[42]);
-  iprintf(
-    "use_allocated_space(): retrieved tasklocal data successfully (42 != %d)\n",
-    local_int_arr[42]);
+  test_check(main_local != local_int_arr);
+  iprintf("use_allocated_space(): retrieved tasklocal data successfully "
+          "(distinct from main thread's local)\n");
   local_int_arr = NULL;
 
   return 0;
@@ -123,6 +123,7 @@ int main(int argc, char *argv[]) {
     iprintf("main(): size_tasklocal() is %d\n", qthread_size_tasklocal());
     iprintf("main(): local_int_arr at %p\n", local_int_arr);
     local_int_arr = (int *)qthread_get_tasklocal(size);
+    main_local = local_int_arr;
     test_check(42 == local_int_arr[42]);
     iprintf("main(): retrieved tasklocal data successfully (42 == %d)\n",
             local_int_arr[42]);

test/features/qpool.c

@@ -44,7 +44,7 @@ int main(int argc, char *argv[]) {
     fprintf(stderr, "qpool_alloc() failed!\n");
     exit(-1);
   }
-  iprintf("allocated: %p (%lu)\n", (void *)rets, (unsigned long)*rets);
+  iprintf("allocated: %p\n", (void *)rets);
   *rets = 1;
   if (*rets != 1) {
     fprintf(stderr, "assigning a value to the allocated memory failed!\n");