[Mellanox] wait reset cause ready (#16722)

Why I did it
SONiC service determine-reboot-cause might run before driver creating reset cause files. In that case, the reset cause will be "Unknown". This PR introduces a wait mechanism to wait for reset cause sysfs files ready.

How I did it
/run/hw-management/config/reset_attr_ready is the file to indicate all reset cause files are ready. In chassis.get_reboot_cause function, it waits /run/hw-management/config/reset_attr_ready for up to 45 seconds.

How to verify it
Manual test on master/202211/202205

platform/mellanox/mlnx-platform-api/sonic_platform/chassis.py

@@ -59,8 +59,9 @@ HWMGMT_SYSTEM_ROOT = '/var/run/hw-management/system/'
 
 #reboot cause related definitions
 REBOOT_CAUSE_ROOT = HWMGMT_SYSTEM_ROOT
-
-REBOOT_CAUSE_FILE_LENGTH = 1
+REBOOT_CAUSE_MAX_WAIT_TIME = 45
+REBOOT_CAUSE_CHECK_INTERVAL = 5
+REBOOT_CAUSE_READY_FILE = '/run/hw-management/config/reset_attr_ready'
 
 REBOOT_TYPE_KEXEC_FILE = "/proc/cmdline"
 REBOOT_TYPE_KEXEC_PATTERN_WARM = ".*SONIC_BOOT_TYPE=(warm|fastfast).*"
@@ -756,6 +757,16 @@ class Chassis(ChassisBase):
                 return 'fast-reboot'
         return None
 
+    def _wait_reboot_cause_ready(self):
+        max_wait_time = REBOOT_CAUSE_MAX_WAIT_TIME
+        while max_wait_time > 0:
+            if utils.read_int_from_file(REBOOT_CAUSE_READY_FILE, log_func=None) == 1:
+                return True
+            time.sleep(REBOOT_CAUSE_CHECK_INTERVAL)
+            max_wait_time -= REBOOT_CAUSE_CHECK_INTERVAL
+
+        return False
+
     def get_reboot_cause(self):
         """
         Retrieves the cause of the previous reboot
@@ -776,6 +787,10 @@ class Chassis(ChassisBase):
             if reboot_cause:
                 return self.REBOOT_CAUSE_NON_HARDWARE, ''
 
+        if not self._wait_reboot_cause_ready():
+            logger.log_error("Hardware reboot cause is not ready")
+            return self.REBOOT_CAUSE_NON_HARDWARE, ''
+
         if not self.reboot_cause_initialized:
             self.initialize_reboot_cause()
 

platform/mellanox/mlnx-platform-api/tests/test_chassis.py

@@ -194,6 +194,7 @@ class TestChassis:
         assert status is True
         assert 'sfp' in event_dict and not event_dict['sfp']
 
+    @mock.patch('sonic_platform.chassis.Chassis._wait_reboot_cause_ready', MagicMock(return_value=True))
     def test_reboot_cause(self):
         from sonic_platform import utils
         from sonic_platform.chassis import REBOOT_CAUSE_ROOT
@@ -242,6 +243,22 @@ class TestChassis:
             assert minor == value
             mock_file_content[file_path] = 0
 
+    @mock.patch('sonic_platform.chassis.Chassis._wait_reboot_cause_ready', MagicMock(return_value=False))
+    def test_reboot_cause_timeout(self):
+        chassis = Chassis()
+        major, minor = chassis.get_reboot_cause()
+        assert major == chassis.REBOOT_CAUSE_NON_HARDWARE
+        assert minor == ''
+
+    @mock.patch('sonic_platform.utils.read_int_from_file')
+    @mock.patch('sonic_platform.chassis.time.sleep', mock.MagicMock())
+    def test_wait_reboot_cause_ready(self, mock_read_int):
+        mock_read_int.return_value = 1
+        chassis = Chassis()
+        assert chassis._wait_reboot_cause_ready()
+        mock_read_int.return_value = 0
+        assert not chassis._wait_reboot_cause_ready()
+
     def test_parse_warmfast_reboot_from_proc_cmdline(self):
         chassis = Chassis()
         with mock.patch("builtins.open", mock.mock_open(read_data="SONIC_BOOT_TYPE=warm")):