Fixing compilation errors related to MUSL libc

Fix Musl libc NULL errors and unsupported pthread funcs for compatibility.
Also ensures cleanup and error handling irrespective of CPU affinity override.

Fix submitted by github dev - AngryLoki
https://github.com/ROCm/ROCR-Runtime/issues/181

Change-Id: Ia487315e504112be5d3370756f23f6e23b9ae4be

runtime/hsa-runtime/core/inc/checked.h

@@ -58,7 +58,7 @@ template <uint64_t code, bool multiProcess = false> class Check final {
   Check(const Check&) { object_ = uintptr_t(this) ^ uintptr_t(code); }
   Check(Check&&) { object_ = uintptr_t(this) ^ uintptr_t(code); }
 
-  ~Check() { object_ = NULL; }
+  ~Check() { object_ = uintptr_t(NULL); }
 
   const Check& operator=(Check&& rhs) { return *this; }
   const Check& operator=(const Check& rhs) { return *this; }

runtime/hsa-runtime/core/runtime/default_signal.cpp

@@ -57,7 +57,7 @@ int BusyWaitSignal::rtti_id_ = 0;
 BusyWaitSignal::BusyWaitSignal(SharedSignal* abi_block, bool enableIPC)
     : Signal(abi_block, enableIPC) {
   signal_.kind = AMD_SIGNAL_KIND_USER;
-  signal_.event_mailbox_ptr = NULL;
+  signal_.event_mailbox_ptr = uint64_t(NULL);
 }
 
 hsa_signal_value_t BusyWaitSignal::LoadRelaxed() {

runtime/hsa-runtime/core/util/lnx/os_linux.cpp

@@ -108,31 +108,44 @@ class os_thread {
       err = pthread_attr_setstacksize(&attrib, stackSize);
       if (err != 0) {
         fprintf(stderr, "pthread_attr_setstacksize failed: %s\n", strerror(err));
+        err = pthread_attr_destroy(&attrib);
+        if (err != 0) {
+          fprintf(stderr, "pthread_attr_destroy failed: %s\n", strerror(err));
+        }
         return;
       }
     }
-\
-    err = pthread_create(&thread, &attrib, ThreadTrampoline, args.get());
+
+    int create_err = pthread_create(&thread, &attrib, ThreadTrampoline, args.get());
 
     // Probably a stack size error since system limits can be different from PTHREAD_STACK_MIN
     // Attempt to grow the stack within reason.
-    if ((err == EINVAL) && stackSize != 0) {
+    if ((create_err == EINVAL) && stackSize != 0) {
       while (stackSize < 20 * 1024 * 1024) {
         stackSize *= 2;
         err = pthread_attr_setstacksize(&attrib, stackSize);
         if (err != 0) {
           fprintf(stderr, "pthread_attr_setstacksize failed: %s\n", strerror(err));
+          err = pthread_attr_destroy(&attrib);
+          if (err != 0) {
+            fprintf(stderr, "pthread_attr_destroy failed: %s\n", strerror(err));
+          }
+
           return;
         }
-        err = pthread_create(&thread, &attrib, ThreadTrampoline, args.get());
-        if (err != EINVAL) break;
+
+        create_err = pthread_create(&thread, &attrib, ThreadTrampoline, args.get());
+        if (create_err != EINVAL) break;
         debug_print("pthread_create returned EINVAL, doubling stack size\n");
       }
     }
 
+    int cores = 0;
+    cpu_set_t* cpuset = nullptr;
+
     if (core::Runtime::runtime_singleton_->flag().override_cpu_affinity()) {
-      int cores = get_nprocs_conf();
-      cpu_set_t* cpuset = CPU_ALLOC(cores);
+      cores = get_nprocs_conf();
+      cpuset = CPU_ALLOC(cores);
       if (cpuset == nullptr) {
         fprintf(stderr, "CPU_ALLOC failed: %s\n", strerror(errno));
         return;
@@ -144,12 +157,11 @@ class os_thread {
       err = pthread_setaffinity_np(thread, CPU_ALLOC_SIZE(cores), cpuset);
       CPU_FREE(cpuset);
       if (err != 0) {
-        fprintf(stderr, "pthread_attr_setaffinity_np failed: %s\n", strerror(err));
+        fprintf(stderr, "pthread_setaffinity_np failed: %s\n", strerror(err));
         return;
       }
     }
-
-    if (err == 0)
+    if (create_err == 0)
       args.release();
     else
       thread = 0;
@@ -642,11 +654,20 @@ SharedMutex CreateSharedMutex() {
     fprintf(stderr, "rw lock attribute init failed: %s\n", strerror(err));
     return nullptr;
   }
+
+#ifdef __GLIBC__
   err = pthread_rwlockattr_setkind_np(&attrib, PTHREAD_RWLOCK_PREFER_WRITER_NONRECURSIVE_NP);
   if (err != 0) {
     fprintf(stderr, "Set rw lock attribute failure: %s\n", strerror(err));
     return nullptr;
   }
+#else
+  err = pthread_rwlockattr_setkind(&attrib, PTHREAD_RWLOCK_PREFER_WRITER_NONRECURSIVE_NP);
+  if (err != 0) {
+    fprintf(stderr, "Set rw lock attribute failure: %s\n", strerror(err));
+    return nullptr;
+  }
+#endif
 
   pthread_rwlock_t* lock = new pthread_rwlock_t;
   err = pthread_rwlock_init(lock, &attrib);

runtime/hsa-runtime/core/util/utils.h

@@ -74,8 +74,7 @@ static __forceinline void* _aligned_malloc(size_t size, size_t alignment) {
   return aligned_alloc(alignment, size);
 #else
   void *mem = NULL;
-  if (NULL != posix_memalign(&mem, alignment, size))
-    return NULL;
+  if (0 != posix_memalign(&mem, alignment, size)) return NULL;
   return mem;
 #endif
 }

runtime/hsa-runtime/image/util.h

@@ -99,7 +99,7 @@ static __forceinline void* _aligned_malloc(size_t size, size_t alignment) {
   return aligned_alloc(alignment, size);
 #else
   void* mem = NULL;
-  if (NULL != posix_memalign(&mem, alignment, size)) return NULL;
+  if (0 != posix_memalign(&mem, alignment, size)) return NULL;
   return mem;
 #endif
 }