[SmartSwitch] Add tests for reboot of a smart switch

tests/common/devices/sonic.py

@@ -419,6 +419,42 @@ def is_supervisor_node(self):
         inv_files = im._sources
         return is_supervisor_node(inv_files, self.hostname)
 
+    def is_smartswitch(self):
+        """Check if the current node is a SmartSwitch
+
+        Returns:
+            True if the current node is a SmartSwitch, else False
+        """
+        config_facts = self.config_facts(host=self.hostname, source="running")['ansible_facts']
+        if (
+            "DEVICE_METADATA" in config_facts and
+            "localhost" in config_facts["DEVICE_METADATA"] and
+            "subtype" in config_facts["DEVICE_METADATA"]["localhost"] and
+            config_facts["DEVICE_METADATA"]["localhost"]["subtype"] == "SmartSwitch" and
+            "type" in config_facts["DEVICE_METADATA"]["localhost"] and
+            config_facts["DEVICE_METADATA"]["localhost"]["type"] != "SmartSwitchDPU"
+        ):
+            return True
+
+        return False
+
+    def is_dpu(self):
+        """Check if the current node is a DPU
+
+        Returns:
+            True if the current node is a DPU, else False
+        """
+        config_facts = self.config_facts(host=self.hostname, source="running")['ansible_facts']
+        if (
+            "DEVICE_METADATA" in config_facts and
+            "localhost" in config_facts["DEVICE_METADATA"] and
+            "type" in config_facts["DEVICE_METADATA"]["localhost"] and
+            config_facts["DEVICE_METADATA"]["localhost"]["type"] == "SmartSwitchDPU"
+        ):
+            return True
+
+        return False
+
     def is_frontend_node(self):
         """Check if the current node is a frontend node in case of multi-DUT.
 

tests/common/reboot.py

@@ -15,7 +15,6 @@
 from tests.common.helpers.dut_utils import ignore_t2_syslog_msgs, create_duthost_console, creds_on_dut
 from tests.common.fixtures.conn_graph_facts import get_graph_facts
 
-
 logger = logging.getLogger(__name__)
 
 # Create the waiting power on event
@@ -142,6 +141,17 @@
     }
 }
 
+'''
+command : command to reboot the smartswitch DUT
+'''
+reboot_ss_ctrl_dict = {
+    REBOOT_TYPE_COLD: {
+        "command": "reboot",
+        "timeout": 300,
+        "cause": r"'reboot'|Non-Hardware \(reboot|^reboot"
+    }
+}
+
 MAX_NUM_REBOOT_CAUSE_HISTORY = 10
 REBOOT_TYPE_HISTOYR_QUEUE = deque([], MAX_NUM_REBOOT_CAUSE_HISTORY)
 REBOOT_CAUSE_HISTORY_TITLE = ["name", "cause", "time", "user", "comment"]
@@ -224,6 +234,28 @@ def execute_reboot_helper():
     return [reboot_res, dut_datetime]
 
 
+@support_ignore_loganalyzer
+def reboot_smartswitch(duthost, reboot_type=REBOOT_TYPE_COLD):
+    """
+    reboots SmartSwitch or a DPU
+    :param duthost: DUT host object
+    :param reboot_type: reboot type (cold)
+    """
+
+    if reboot_type not in reboot_ss_ctrl_dict:
+        logger.info("Skipping the reboot test as the reboot type {} is not supported".format(reboot_type))
+        return
+
+    hostname = duthost.hostname
+    dut_datetime = duthost.get_now_time(utc_timezone=True)
+
+    logging.info("Rebooting the DUT {} with type {}".format(hostname, reboot_type))
+
+    reboot_res = duthost.command(reboot_ss_ctrl_dict[reboot_type]["command"])
+
+    return [reboot_res, dut_datetime]
+
+
 @support_ignore_loganalyzer
 def reboot(duthost, localhost, reboot_type='cold', delay=10,
            timeout=0, wait=0, wait_for_ssh=True, wait_warmboot_finalizer=False, warmboot_finalizer_timeout=0,
@@ -284,7 +316,12 @@ def reboot(duthost, localhost, reboot_type='cold', delay=10,
     console_thread_res = pool.apply_async(
         collect_console_log, args=(duthost, localhost, timeout + wait_conlsole_connection))
     time.sleep(wait_conlsole_connection)
-    reboot_res, dut_datetime = perform_reboot(duthost, pool, reboot_command, reboot_helper, reboot_kwargs, reboot_type)
+    # Perform reboot
+    if duthost.is_smartswitch():
+        reboot_res, dut_datetime = reboot_smartswitch(duthost, reboot_type)
+    else:
+        reboot_res, dut_datetime = perform_reboot(duthost, pool, reboot_command, reboot_helper,
+                                                  reboot_kwargs, reboot_type)
 
     wait_for_shutdown(duthost, localhost, delay, timeout, reboot_res)
 

tests/platform_tests/test_reboot.py

@@ -221,6 +221,9 @@ def test_fast_reboot(duthosts, enum_rand_one_per_hwsku_hostname,
     if duthost.is_multi_asic:
         pytest.skip("Multi-ASIC devices not supporting fast reboot")
 
+    if duthost.is_smartswitch():
+        pytest.skip("Smart Switch devices does not support fast reboot")
+
     reboot_and_check(localhost, duthost, conn_graph_facts.get("device_conn", {}).get(duthost.hostname, {}),
                      xcvr_skip_list, reboot_type=REBOOT_TYPE_FAST, duthosts=duthosts)
 
@@ -236,6 +239,9 @@ def test_warm_reboot(duthosts, enum_rand_one_per_hwsku_hostname,
     if duthost.is_multi_asic:
         pytest.skip("Multi-ASIC devices not supporting warm reboot")
 
+    if duthost.is_smartswitch():
+        pytest.skip("Smart Switch devices does not support warm reboot")
+
     asic_type = duthost.facts["asic_type"]
 
     if asic_type in ["mellanox"]:

tests/smartswitch/common/device_utils_dpu.py

@@ -433,7 +433,7 @@ def post_test_switch_check(duthost, localhost,
     return
 
 
-def post_test_dpu_check(duthost, dpuhosts, dpu_name):
+def post_test_dpu_check(duthost, dpuhosts, dpu_name, reboot_cause):
     """
     Runs all required checks for a given DPU
     Args:
@@ -464,14 +464,13 @@ def post_test_dpu_check(duthost, dpuhosts, dpu_name):
     logging.info(f"Checking reboot cause of {dpu_name}")
     pytest_assert(
         wait_until(REBOOT_CAUSE_TIMEOUT, REBOOT_CAUSE_INT, 0,
-                   check_dpu_reboot_cause, duthost, dpu_name, "Non-Hardware"),
+                   check_dpu_reboot_cause, duthost, dpu_name, reboot_cause),
         f"Reboot cause for DPU {dpu_name} is incorrect"
     )
 
 
-def post_test_dpus_check(duthost, dpuhosts,
-                         dpu_on_list, dpu_off_list,
-                         ip_address_list, num_dpu_modules):
+def post_test_dpus_check(duthost, dpuhosts, dpu_on_list, ip_address_list,
+                         num_dpu_modules, reboot_cause):
     """
     Checks DPU OFF/ON and reboot cause status Post Test
     Args:
@@ -489,7 +488,7 @@ def post_test_dpus_check(duthost, dpuhosts,
         logging.info("Post test DPUs check in parallel")
         for dpu in dpu_on_list:
             executor.submit(post_test_dpu_check, duthost,
-                            dpuhosts, dpu)
+                            dpuhosts, dpu, reboot_cause)
 
     logging.info("Checking all powered on DPUs connectivity")
     ping_status = check_dpu_ping_status(duthost, ip_address_list)

tests/smartswitch/common/reboot.py

@@ -0,0 +1,62 @@
+import logging
+import pytest
+from tests.common.reboot import reboot_ss_ctrl_dict as reboot_dict, REBOOT_TYPE_HISTOYR_QUEUE, \
+    sync_reboot_history_queue_with_dut
+
+logger = logging.getLogger(__name__)
+
+REBOOT_TYPE_COLD = "cold"
+REBOOT_TYPE_UNKNOWN = "unknown"
+REBOOT_TYPE_KERNEL_PANIC = "Kernel Panic"
+REBOOT_TYPE_WATCHDOG = "Watchdog"
+
+
+def log_and_perform_reboot(duthost, reboot_type, dpu_name):
+    """
+    Logs and initiates the reboot process based on the host type.
+    Skips the test if the host is a DPU.
+
+    @param duthost: DUT host object
+    @param reboot_type: Type of reboot to perform
+    @param dpu_name: Name of the DPU (optional)
+    """
+    hostname = duthost.hostname
+    logger.info("Rebooting the DUT {} with type {}".format(hostname, reboot_type))
+
+    if reboot_type == REBOOT_TYPE_COLD:
+        if duthost.is_smartswitch():
+            if dpu_name is None:
+                logger.info("Rebooting the switch {} with cold reboot".format(hostname))
+                return duthost.command("sudo reboot")
+            else:
+                logger.info("Rebooting the DUT {} with cold reboot".format(hostname))
+                return duthost.command("sudo reboot -d {}".format(dpu_name))
+        elif duthost.is_dpu():
+            pytest.skip("Skipping the reboot test as the DUT is a DPU")
+    else:
+        pytest.skip("Skipping the reboot test as the reboot type {} is not supported".format(reboot_type))
+
+
+def perform_reboot(duthost, reboot_type=REBOOT_TYPE_COLD, dpu_name=None):
+    """
+    Performs a reboot and validates the DPU status after reboot.
+
+    @param duthost: DUT host object
+    @param reboot_type: Reboot type
+    @param dpu_name: DPU name
+    """
+    if reboot_type not in reboot_dict:
+        pytest.skip("Skipping the reboot test as the reboot type {} is not supported".format(reboot_type))
+
+    logger.info("Sync reboot cause history queue with DUT reboot cause history queue")
+    sync_reboot_history_queue_with_dut(duthost)
+
+    res = log_and_perform_reboot(duthost, reboot_type, dpu_name)
+    if res['failed'] is True:
+        if dpu_name is None:
+            pytest.fail("Failed to reboot the {} with type {}".format(duthost.hostname, reboot_type))
+        else:
+            pytest.fail("Failed to reboot the DPU {}".format(dpu_name))
+
+    logger.info("Appending the last reboot type to the queue")
+    REBOOT_TYPE_HISTOYR_QUEUE.append(reboot_type)

tests/smartswitch/platform_tests/test_platform_dpu.py

@@ -113,9 +113,7 @@ def test_pcie_link(duthosts, dpuhosts,
         duthost.shell("sudo config chassis modules \
                        startup %s" % (dpu_on_list[index]))
 
-    post_test_dpus_check(duthost, dpuhosts,
-                         dpu_on_list, dpu_off_list,
-                         ip_address_list, num_dpu_modules)
+    post_test_dpus_check(duthost, dpuhosts, dpu_on_list, ip_address_list, num_dpu_modules, "Non-Hardware")
 
     logging.info("Verifying output of '{}' on '{}'..."
                  .format(CMD_PCIE_INFO, duthost.hostname))
@@ -297,9 +295,8 @@ def test_system_health_summary(duthosts, dpuhosts,
                                                  num_dpu_modules)
 
     logging.info("Checking DPU is completely UP")
-    post_test_dpus_check(duthost, dpuhosts,
-                         dpu_on_list, dpu_off_list,
-                         ip_address_list, num_dpu_modules)
+    post_test_dpus_check(duthost, dpuhosts, dpu_on_list,
+                         ip_address_list, num_dpu_modules, "Non-Hardware")
 
     logging.info("Checking show system-health summary on Switch")
     output_health_summary = duthost.command("show system-health summary")

tests/smartswitch/platform_tests/test_reload_dpu.py

@@ -7,11 +7,13 @@
 import re
 from tests.common.platform.processes_utils import wait_critical_processes
 from tests.common.reboot import reboot, REBOOT_TYPE_COLD
-from tests.smartswitch.common.device_utils_dpu import get_dpu_link_status,\
-    check_dpu_ping_status, check_dpu_link_and_status, check_dpu_module_status,\
+from tests.common.helpers.platform_api import module
+from tests.smartswitch.common.device_utils_dpu import check_dpu_link_and_status,\
     pre_test_check, post_test_switch_check, post_test_dpus_check,\
-    check_dpu_reboot_cause, num_dpu_modules  # noqa: F401
+    num_dpu_modules  # noqa: F401
 from tests.common.platform.device_utils import platform_api_conn, start_platform_api_service  # noqa: F401,F403
+from tests.smartswitch.common.reboot import perform_reboot
+from tests.common.helpers.multi_thread_utils import SafeThreadPoolExecutor
 
 pytestmark = [
     pytest.mark.topology('smartswitch')
@@ -153,9 +155,7 @@ def test_dpu_status_post_dpu_kernel_panic(duthosts, dpuhosts,
         dpuhosts[dpu_id].shell(kernel_panic_cmd, executable="/bin/bash")
 
     logging.info("Executing post test dpu check")
-    post_test_dpus_check(duthost, dpuhosts,
-                         dpu_on_list, dpu_off_list,
-                         ip_address_list, num_dpu_modules)
+    post_test_dpus_check(duthost, dpuhosts, dpu_on_list, ip_address_list, num_dpu_modules, "Non-Hardware")
 
 
 def test_dpu_check_post_dpu_mem_exhaustion(duthosts, dpuhosts,
@@ -182,6 +182,70 @@ def test_dpu_check_post_dpu_mem_exhaustion(duthosts, dpuhosts,
         dpuhosts[dpu_id].shell(memory_exhaustion_cmd, executable="/bin/bash")
 
     logging.info("Executing post test dpu check")
-    post_test_dpus_check(duthost, dpuhosts,
-                         dpu_on_list, dpu_off_list,
-                         ip_address_list, num_dpu_modules)
+    post_test_dpus_check(duthost, dpuhosts, dpu_on_list, ip_address_list,
+                         num_dpu_modules, "Non-Hardware")
+
+
+def test_cold_reboot_dpus(duthosts, dpuhosts, enum_rand_one_per_hwsku_hostname,
+                          platform_api_conn, num_dpu_modules):  # noqa: F811, E501
+    """
+    Test to cold reboot all DPUs in the DUT.
+    Steps:
+    1. Perform pre-test checks to gather DPU state.
+    2. Initiate cold reboot on all DPUs concurrently.
+    3. Perform post-test checks to verify the state after reboot.
+
+    Args:
+        duthosts: DUT hosts object
+        dpuhosts: DPU hosts object
+        enum_rand_one_per_hwsku_hostname: Randomized DUT hostname
+        platform_api_conn: Platform API connection object
+        num_dpu_modules: Number of DPU modules to reboot
+    """
+    duthost = duthosts[enum_rand_one_per_hwsku_hostname]
+
+    logging.info("Executing pre test check")
+    ip_address_list, dpu_on_list, dpu_off_list = pre_test_check(duthost, platform_api_conn, num_dpu_modules)
+
+    def reboot_dpu(duthost, platform_api_conn, index):
+        try:
+            dpu_name = module.get_name(platform_api_conn, index)
+            perform_reboot(duthost, REBOOT_TYPE_COLD, dpu_name)
+        except Exception as e:
+            logging.error(f"Failed to reboot DPU at index {index}: {e}")
+
+    with SafeThreadPoolExecutor(max_workers=num_dpu_modules) as executor:
+        logging.info("Rebooting all DPUs in parallel")
+        for index in range(num_dpu_modules):
+            executor.submit(reboot_dpu, duthost, platform_api_conn, index)
+
+    logging.info("Executing post test dpu check")
+    post_test_dpus_check(duthost, dpuhosts, dpu_on_list, ip_address_list, num_dpu_modules, "Non-Hardware")
+
+
+def test_cold_reboot_switch(duthosts, dpuhosts, enum_rand_one_per_hwsku_hostname,
+                            platform_api_conn, num_dpu_modules):  # noqa: F811, E501
+    """
+    Test to cold reboot the switch in the DUT.
+    Steps:
+    1. Perform pre-test checks to gather DPU state.
+    2. Initiate a cold reboot on the switch.
+    3. Perform post-test checks to verify the state of DPUs after the reboot.
+
+    Args:
+        duthosts: DUT hosts object
+        dpuhosts: DPU hosts object
+        enum_rand_one_per_hwsku_hostname: Randomized DUT hostname
+        platform_api_conn: Platform API connection object
+        num_dpu_modules: Number of DPU modules to verify
+    """
+    duthost = duthosts[enum_rand_one_per_hwsku_hostname]
+
+    logging.info("Executing pre test check")
+    ip_address_list, dpu_on_list, dpu_off_list = pre_test_check(duthost, platform_api_conn, num_dpu_modules)
+
+    logging.info("Starting switch reboot...")
+    perform_reboot(duthost, REBOOT_TYPE_COLD, None)
+
+    logging.info("Executing post switch reboot dpu check")
+    post_test_dpus_check(duthost, dpuhosts, dpu_on_list, ip_address_list, num_dpu_modules, "reboot")