[SWDEV-474450] Removed DEVICE_MUTEX from gpu_reset

Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>
Change-Id: I706fb47288738bfbde94b56fee66bbf807b3c0cb


[ROCm/amdsmi commit: 8bc8307c60]

projects/amdsmi/rocm_smi/src/rocm_smi.cc

@@ -3435,7 +3435,8 @@ rsmi_dev_gpu_reset(uint32_t dv_ind) {
   ss << __PRETTY_FUNCTION__ << "| ======= start =======";
   LOG_TRACE(ss);
   REQUIRE_ROOT_ACCESS
-  DEVICE_MUTEX
+  // No longer using DEVICE_MUTEX as it blocks long running processes
+  // DEVICE_MUTEX
 
   rsmi_status_t ret;
   uint64_t status_code = 0;

projects/amdsmi/third_party/shared_mutex/shared_mutex.cc

@@ -255,10 +255,11 @@ shared_mutex_t shared_mutex_init(const char *name, mode_t mode, bool retried) {
     // When process crash before unlock the mutex, the mutex is in bad status.
     // reset the mutex if no process is using it, and then retry lock
     if (!retried) {
-      std::vector<std::string> ids = lsof(name);
+      std::string shared_mutex_filename = "/dev/shm" + std::string(name);
+      std::vector<std::string> ids = lsof(shared_mutex_filename.c_str());
       if (ids.size() == 0) {  // no process is using it
         fprintf(stderr, "%d re-init the mutex %s since no one use it. ret:%d ptr:%p\n",
-              cur_pid, name, ret, reinterpret_cast<shared_mutex_t *>(addr)->ptr);
+              cur_pid, shared_mutex_filename, ret, reinterpret_cast<shared_mutex_t *>(addr)->ptr);
         memset(mutex_ptr, 0, sizeof(pthread_mutex_t));
         // Set mutex.created == 1 so that it can be initialized latter.
         mutex.created = 1;