Use device service for retrieve gpu details

src/middlewared/middlewared/etc_files/docker.py

@@ -1,12 +1,8 @@
 import json
 import os
-import re
 import subprocess
 
 
-RE_LDCONFIG = re.compile(r'@/sbin/ldconfig')
-
-
 def nvidia_configuration(middleware):
     # this needs to happen for nvidia gpu to work properly for docker containers
     # https://github.com/NVIDIA/nvidia-docker/issues/854#issuecomment-572175484
@@ -15,10 +11,10 @@ def nvidia_configuration(middleware):
         return {}
 
     with open(nvidia_config_path, 'r') as f:
-        data = RE_LDCONFIG.sub('/sbin/ldconfig', f.read())
+        data = f.read()
 
     with open(nvidia_config_path, 'w') as f:
-        f.write(data)
+        f.write(data.replace('@/sbin/ldconfig', '/sbin/ldconfig'))
 
     return {
         'runtimes': {'nvidia': {'path': '/usr/bin/nvidia-container-runtime', 'runtimeArgs': []}},
@@ -27,7 +23,7 @@ def nvidia_configuration(middleware):
 
 
 def gpu_configuration(middleware):
-    available_gpu = middleware.call_sync('hardware.available_gpu')
+    available_gpu = middleware.call_sync('device.get_gpus')
 
     if available_gpu['vendor'] == 'NVIDIA':
         return nvidia_configuration(middleware)

src/middlewared/middlewared/plugins/device.py

@@ -4,9 +4,9 @@
 
 class DeviceService(Service):
 
-    @accepts(Str('type', enum=['SERIAL', 'DISK']))
+    @accepts(Str('type', enum=['SERIAL', 'DISK', 'GPU']))
     async def get_info(self, _type):
         """
-        Get info for SERIAL/DISK device types.
+        Get info for SERIAL/DISK/GPU device types.
         """
         return await self.middleware.call(f'device.get_{_type.lower()}s')

src/middlewared/middlewared/plugins/device_/device_info_base.py

@@ -33,6 +33,10 @@ class DeviceInfoBase(ServicePartBase):
         'blocks': None,
     }
 
+    @private
+    async def get_gpus(self):
+        raise NotImplementedError()
+
     @private
     async def get_serials(self):
         raise NotImplementedError()

src/middlewared/middlewared/plugins/device_/device_info_freebsd.py

@@ -120,3 +120,6 @@ async def get_serials(self):
 
     async def get_storage_devices_topology(self):
         return await camcontrol_list()
+
+    async def get_gpus(self):
+        raise NotImplementedError()

src/middlewared/middlewared/plugins/device_/device_info_linux.py

@@ -8,15 +8,19 @@
 
 from .device_info_base import DeviceInfoBase
 from middlewared.service import private, Service
+from middlewared.utils import run
 
 RE_DISK_SERIAL = re.compile(r'Unit serial number:\s*(.*)')
+RE_GPU_VENDOR = re.compile(r'description:\s*VGA compatible controller[\s\S]*vendor:\s*(.*)')
 RE_NVME_PRIVATE_NAMESPACE = re.compile(r'nvme[0-9]+c')
 RE_SERIAL = re.compile(r'state.*=\s*(\w*).*io (.*)-(\w*)\n.*', re.S | re.A)
 RE_UART_TYPE = re.compile(r'is a\s*(\w+)')
 
 
 class DeviceService(Service, DeviceInfoBase):
 
+    GPU = None
+
     def get_serials(self):
         devices = []
         for tty in map(lambda t: os.path.basename(t), glob.glob('/dev/ttyS*')):
@@ -225,3 +229,24 @@ def get_storage_devices_topology(self):
                     }
                 }
         return topology
+
+    async def get_gpus(self):
+        if self.GPU:
+            return self.GPU
+
+        not_available = {'available': False, 'vendor': None}
+        cp = await run(['lshw', '-numeric', '-C', 'display'], check=False)
+        if cp.returncode:
+            self.logger.error('Unable to retrieve GPU details: %s', cp.stderr.decode())
+            return not_available
+
+        vendor = RE_GPU_VENDOR.findall(cp.stdout.decode())
+        if not vendor:
+            self.GPU = not_available
+        else:
+            # We only support nvidia based GPU's right now based on equipment available
+            if 'nvidia' in vendor[0].lower():
+                self.GPU = {'available': True, 'vendor': 'NVIDIA'}
+            else:
+                self.GPU = not_available
+        return self.GPU

src/middlewared/middlewared/plugins/kubernetes_linux/gpu.py

@@ -49,7 +49,7 @@ async def setup(self):
             self.logger.error('Unable to configure GPU for node: %s', e)
 
     async def setup_internal(self):
-        gpu = await self.middleware.call('hardware.available_gpu')
+        gpu = await self.middleware.call('device.get_gpus')
         to_remove = list(GPU_CONFIG.keys())
         daemonsets = {
             f'{d["metadata"]["namespace"]}_{d["metadata"]["name"]}': d