Merge pull request #3184 from matrix/metal_opencl_gpu_utilization

Hardware Monitor: Add support for GPU device utilization readings using iokit on Apple Silicon (OpenCL and Metal)

docs/changes.txt

@@ -50,6 +50,7 @@
 - Backend Info: Added local memory size to output
 - Backend: with kernel build options, switch from -I to -D INCLUDE_PATH, in order to support Apple Metal runtime
 - CUDA Backend: moved functions to ext_cuda.c/ext_nvrtc.c and includes to ext_cuda.h/ext_nvrtc.h
+- Hardware Monitor: Add support for GPU device utilization readings using iokit on Apple Silicon (OpenCL and Metal)
 - Hash Info: show more information (Updated Hash-Format. Added Autodetect, Self-Test, Potfile and Plaintext encoding)
 - HIP Backend: moved functions to ext_hip.c/ext_hiprtc.c and includes to ext_hip.h/ext_hiprtc.h
 - Kernels: Refactored standard kernel declaration to use a structure holding u32/u64 attributes to reduce the number of attributes

include/ext_iokit.h

@@ -111,7 +111,7 @@ typedef struct hm_iokit_lib
 typedef hm_iokit_lib_t IOKIT_PTR;
 
 #if defined(__APPLE__)
-UInt32 hm_IOKIT_strtoul (char *str, int size, int base);
+UInt32 hm_IOKIT_strtoul (const char *str, int size, int base);
 void hm_IOKIT_ultostr (char *str, UInt32 val);
 kern_return_t hm_IOKIT_SMCOpen (void *hashcat_ctx, io_connect_t *conn);
 kern_return_t hm_IOKIT_SMCClose (io_connect_t conn);
@@ -121,6 +121,7 @@ int hm_IOKIT_SMCGetSensorGraphicHot (void *hashcat_ctx);
 int hm_IOKIT_SMCGetTemperature (void *hashcat_ctx, char *key, double *temp);
 bool hm_IOKIT_SMCGetFanRPM (char *key, io_connect_t conn, float *ret);
 int hm_IOKIT_get_fan_speed_current (void *hashcat_ctx, char *fan_speed_buf);
+int hm_IOKIT_get_utilization_current (void *hashcat_ctx, int *utilization);
 bool iokit_init (void *hashcat_ctx);
 bool iokit_close (void *hashcat_ctx);
 #endif // __APPLE__

src/backend.c

@@ -6746,10 +6746,7 @@ int backend_ctx_devices_init (hashcat_ctx_t *hashcat_ctx, const int comptime)
           #if defined (__APPLE__)
           if (device_param->opencl_platform_vendor_id == VENDOR_ID_APPLE)
           {
-            if (device_param->skipped == false)
-            {
-              need_iokit = true;
-            }
+            need_iokit = true;
           }
           #endif
 
@@ -6777,6 +6774,16 @@ int backend_ctx_devices_init (hashcat_ctx_t *hashcat_ctx, const int comptime)
             need_nvapi = true;
             #endif
           }
+
+          #if defined (__APPLE__)
+          if (strncmp (device_param->device_name, "Apple M", 7) == 0)
+          {
+            if (device_param->opencl_platform_vendor_id == VENDOR_ID_APPLE)
+            {
+              need_iokit = true;
+            }
+          }
+          #endif
         }
 
         if (device_param->opencl_device_type & CL_DEVICE_TYPE_CPU)

src/ext_iokit.c

@@ -13,7 +13,7 @@
 #if defined (__APPLE__)
 #include <IOKit/IOKitLib.h>
 
-UInt32 hm_IOKIT_strtoul (char *str, int size, int base)
+UInt32 hm_IOKIT_strtoul (const char *str, int size, int base)
 {
   int i;
 
@@ -223,6 +223,66 @@ bool hm_IOKIT_SMCGetFanRPM (char *key, io_connect_t conn, float *ret)
   return false;
 }
 
+int hm_IOKIT_get_utilization_current (void *hashcat_ctx, int *utilization)
+{
+  bool rc = false;
+
+  io_iterator_t iterator;
+
+  CFMutableDictionaryRef matching = IOServiceMatching ("IOAccelerator");
+
+  if (IOServiceGetMatchingServices (kIOMasterPortDefault, matching, &iterator) != kIOReturnSuccess)
+  {
+    event_log_error (hashcat_ctx, "IOServiceGetMatchingServices(): failure");
+
+    return rc;
+  }
+
+  io_registry_entry_t regEntry;
+
+  while ((regEntry = IOIteratorNext (iterator)))
+  {
+    // Put this services object into a dictionary object.
+    CFMutableDictionaryRef serviceDictionary;
+
+    if (IORegistryEntryCreateCFProperties (regEntry, &serviceDictionary, kCFAllocatorDefault, kNilOptions) != kIOReturnSuccess)
+    {
+      // Service dictionary creation failed.
+      IOObjectRelease (regEntry);
+
+      continue;
+    }
+
+    CFMutableDictionaryRef perf_properties = (CFMutableDictionaryRef) CFDictionaryGetValue (serviceDictionary, CFSTR ("PerformanceStatistics"));
+
+    if (perf_properties)
+    {
+      static ssize_t gpuCoreUtil = 0;
+
+      const void *gpuCoreUtilization = CFDictionaryGetValue (perf_properties, CFSTR ("Device Utilization %"));
+
+      if (gpuCoreUtilization != NULL)
+      {
+        CFNumberGetValue (gpuCoreUtilization, kCFNumberSInt64Type, &gpuCoreUtil);
+
+        *utilization = gpuCoreUtil;
+
+        rc = true;
+      }
+    }
+
+    CFRelease (serviceDictionary);
+
+    IOObjectRelease (regEntry);
+
+    if (rc == true) break;
+  }
+
+  IOObjectRelease (iterator);
+
+  return rc;
+}
+
 int hm_IOKIT_get_fan_speed_current (void *hashcat_ctx, char *fan_speed_buf)
 {
   hwmon_ctx_t *hwmon_ctx = ((hashcat_ctx_t *) hashcat_ctx)->hwmon_ctx;

src/hwmon.c

@@ -764,6 +764,31 @@ int hm_get_utilization_with_devices_idx (hashcat_ctx_t *hashcat_ctx, const int b
     }
   }
 
+  #if defined(__APPLE__)
+  if (backend_ctx->devices_param[backend_device_idx].is_metal == true || backend_ctx->devices_param[backend_device_idx].is_opencl == true)
+  {
+    if (backend_ctx->devices_param[backend_device_idx].opencl_platform_vendor_id == VENDOR_ID_APPLE)
+    {
+      if (backend_ctx->devices_param[backend_device_idx].opencl_device_type & CL_DEVICE_TYPE_GPU)
+      {
+        if (hwmon_ctx->hm_iokit)
+        {
+          int utilization = 0;
+
+          if (hm_IOKIT_get_utilization_current (hashcat_ctx, &utilization) == -1)
+          {
+            hwmon_ctx->hm_device[backend_device_idx].utilization_get_supported = false;
+
+            return -1;
+          }
+
+          return utilization;
+        }
+      }
+    }
+  }
+  #endif
+
   if ((backend_ctx->devices_param[backend_device_idx].is_opencl == true) || (backend_ctx->devices_param[backend_device_idx].is_hip == true))
   {
     if (backend_ctx->devices_param[backend_device_idx].opencl_device_type & CL_DEVICE_TYPE_GPU)
@@ -1565,6 +1590,24 @@ int hwmon_ctx_init (hashcat_ctx_t *hashcat_ctx)
           // nothing to do
         }
 
+        #if defined (__APPLE__)
+        if (device_param->is_metal == true)
+        {
+          const u32 device_id = device_param->device_id;
+
+          if ((device_param->opencl_platform_vendor_id == VENDOR_ID_APPLE) && (hwmon_ctx->hm_iokit))
+          {
+            hm_adapters_iokit[device_id].buslanes_get_supported    = false;
+            hm_adapters_iokit[device_id].corespeed_get_supported   = false;
+            hm_adapters_iokit[device_id].fanspeed_get_supported    = true;
+            hm_adapters_iokit[device_id].fanpolicy_get_supported   = false;
+            hm_adapters_iokit[device_id].memoryspeed_get_supported = false;
+            hm_adapters_iokit[device_id].temperature_get_supported = true;
+            hm_adapters_iokit[device_id].utilization_get_supported = true;
+          }
+        }
+        #endif
+
         if ((device_param->is_opencl == true) || (device_param->is_hip == true))
         {
           const u32 device_id = device_param->device_id;
@@ -1577,7 +1620,7 @@ int hwmon_ctx_init (hashcat_ctx_t *hashcat_ctx)
             hm_adapters_iokit[device_id].fanpolicy_get_supported   = false;
             hm_adapters_iokit[device_id].memoryspeed_get_supported = false;
             hm_adapters_iokit[device_id].temperature_get_supported = true;
-            hm_adapters_iokit[device_id].utilization_get_supported = false;
+            hm_adapters_iokit[device_id].utilization_get_supported = true;
           }
 
           if ((device_param->opencl_device_type & CL_DEVICE_TYPE_GPU) == 0) continue;
@@ -1647,7 +1690,6 @@ int hwmon_ctx_init (hashcat_ctx_t *hashcat_ctx)
   }
   #endif
 
-
   if (hwmon_ctx->hm_adl == NULL && hwmon_ctx->hm_nvml == NULL && hwmon_ctx->hm_sysfs_amdgpu == NULL && hwmon_ctx->hm_sysfs_cpu == NULL && hwmon_ctx->hm_iokit == NULL)
   {
     FREE_ADAPTERS;
@@ -1715,6 +1757,24 @@ int hwmon_ctx_init (hashcat_ctx_t *hashcat_ctx)
       }
     }
 
+    if (device_param->is_metal == true)
+    {
+      if (hwmon_ctx->hm_iokit)
+      {
+        hwmon_ctx->hm_device[backend_devices_idx].iokit                              = hm_adapters_iokit[device_id].iokit;
+        hwmon_ctx->hm_device[backend_devices_idx].buslanes_get_supported            |= hm_adapters_iokit[device_id].buslanes_get_supported;
+        hwmon_ctx->hm_device[backend_devices_idx].corespeed_get_supported           |= hm_adapters_iokit[device_id].corespeed_get_supported;
+        hwmon_ctx->hm_device[backend_devices_idx].fanspeed_get_supported            |= hm_adapters_iokit[device_id].fanspeed_get_supported;
+        hwmon_ctx->hm_device[backend_devices_idx].fanpolicy_get_supported           |= hm_adapters_iokit[device_id].fanpolicy_get_supported;
+        hwmon_ctx->hm_device[backend_devices_idx].memoryspeed_get_supported         |= hm_adapters_iokit[device_id].memoryspeed_get_supported;
+        hwmon_ctx->hm_device[backend_devices_idx].temperature_get_supported         |= hm_adapters_iokit[device_id].temperature_get_supported;
+        hwmon_ctx->hm_device[backend_devices_idx].threshold_shutdown_get_supported  |= hm_adapters_iokit[device_id].threshold_shutdown_get_supported;
+        hwmon_ctx->hm_device[backend_devices_idx].threshold_slowdown_get_supported  |= hm_adapters_iokit[device_id].threshold_slowdown_get_supported;
+        hwmon_ctx->hm_device[backend_devices_idx].throttle_get_supported            |= hm_adapters_iokit[device_id].throttle_get_supported;
+        hwmon_ctx->hm_device[backend_devices_idx].utilization_get_supported         |= hm_adapters_iokit[device_id].utilization_get_supported;
+      }
+    }
+
     if ((device_param->is_opencl == true) || (device_param->is_hip == true))
     {
       if (device_param->opencl_device_type & CL_DEVICE_TYPE_CPU)