Register the default queue error handler for all internal queues.

Also make failure to handle queue errors fatal.

Motivation is to improve detection of queue error conditions
that currently appear as application hangs.

Change-Id: I655643616dc0bd303d7df3ce8aca2c099bec3d46


[ROCm/ROCR-Runtime commit: 907679c989]

projects/rocr-runtime/runtime/hsa-runtime/core/inc/amd_gpu_agent.h

@@ -332,8 +332,16 @@ class GpuAgent : public GpuAgentInt {
   static const uint32_t minAqlSize_ = 0x1000;   // 4KB min
   static const uint32_t maxAqlSize_ = 0x20000;  // 8MB max
 
-  // @brief Create a queue through HSA API to allow tools to intercept.
-  core::Queue* CreateInterceptibleQueue();
+  // @brief Create an internal queue allowing tools to be notified.
+  core::Queue* CreateInterceptibleQueue() {
+    return CreateInterceptibleQueue(core::Queue::DefaultErrorHandler, nullptr);
+  }
+
+  // @brief // @brief Create an internal queue, with a custom error handler, allowing tools to be
+  // notified.
+  core::Queue* CreateInterceptibleQueue(void (*callback)(hsa_status_t status, hsa_queue_t* source,
+                                                         void* data),
+                                        void* data);
 
   // @brief Create SDMA blit object.
   //

projects/rocr-runtime/runtime/hsa-runtime/core/runtime/amd_gpu_agent.cpp

@@ -533,10 +533,12 @@ hsa_status_t GpuAgent::VisitRegion(
   return HSA_STATUS_SUCCESS;
 }
 
-core::Queue* GpuAgent::CreateInterceptibleQueue() {
+core::Queue* GpuAgent::CreateInterceptibleQueue(void (*callback)(hsa_status_t status,
+                                                                 hsa_queue_t* source, void* data),
+                                                void* data) {
   // Disabled intercept of internal queues pending tools updates.
   core::Queue* queue = nullptr;
-  QueueCreate(minAqlSize_, HSA_QUEUE_TYPE_MULTI, NULL, NULL, 0, 0, &queue);
+  QueueCreate(minAqlSize_, HSA_QUEUE_TYPE_MULTI, callback, data, 0, 0, &queue);
   if (queue != nullptr)
     core::Runtime::runtime_singleton_->InternalQueueCreateNotify(core::Queue::Convert(queue),
                                                                  this->public_handle());

projects/rocr-runtime/runtime/hsa-runtime/core/runtime/queue.cpp

@@ -54,6 +54,7 @@ void Queue::DefaultErrorHandler(hsa_status_t status, hsa_queue_t* source, void*
     const char* msg = "UNKNOWN ERROR";
     HSA::hsa_status_string(status, &msg);
     fprintf(stderr, "Queue at %p inactivated due to async error:\n\t%s\n", source, msg);
+    abort();
   }
 }