[pmon] Update the platform_wait script to wait on asic init

.gitmodules

@@ -3,7 +3,7 @@
 	url = https://github.com/sonic-net/sonic-swss-common
 [submodule "sonic-linux-kernel"]
 	path = src/sonic-linux-kernel
-	url = https://github.com/sonic-net/sonic-linux-kernel
+	url = https://github.com/vivekrnv/sonic-linux-kernel
 [submodule "sonic-sairedis"]
 	path = src/sonic-sairedis
 	url = https://github.com/sonic-net/sonic-sairedis

build_debian.sh

@@ -398,6 +398,7 @@ sudo LANG=C DEBIAN_FRONTEND=noninteractive chroot $FILESYSTEM_ROOT apt-get -y in
     resolvconf              \
 	lsof                    \
 	sysstat                 \
+	xxd                     \
 	zstd
 
 # Have systemd create the auditd log directory

device/mellanox/x86_64-mlnx_msn2700-r0/platform_wait

@@ -8,8 +8,9 @@ declare -r SYSLOG_INFO="info"
 
 declare -r HW_MGMT_CONFIG="/var/run/hw-management/config"
 
-declare -r MODULE_COUNTER="${HW_MGMT_CONFIG}/module_counter"
-declare -r SFP_COUNTER="${HW_MGMT_CONFIG}/sfp_counter"
+declare -r ASIC_INIT_DONE="${HW_MGMT_CONFIG}/asics_init_done"
+declare -r NUM_ASICS="${HW_MGMT_CONFIG}/asic_num"
+declare -r ASIC_CHIPUP_COMPLETED="${HW_MGMT_CONFIG}/asic_chipup_completed"
 
 declare -r EXIT_SUCCESS="0"
 declare -r EXIT_TIMEOUT="1"
@@ -26,44 +27,42 @@ function log_info() {
     eval "${SYSLOG_LOGGER} -t ${SYSLOG_IDENTIFIER} -p ${SYSLOG_INFO} $@"
 }
 
-function wait_for_sfp() {
-    local -r _NUM_MATCH="^[0-9]+$"
-    local -r _NUM_ZERO="0"
+function wait_for_asic_chipup() {
 
-    local _MODULE_CNT="0"
-    local _SFP_CNT="0"
+    local _ASIC_INIT="0"
+    local _ASIC_COUNT="0"
+    local _ASICS_CHIPUP="0"
 
     local -i _WDOG_CNT="1"
     local -ir _WDOG_MAX="300"
 
     local -r _TIMEOUT="1s"
 
     while [[ "${_WDOG_CNT}" -le "${_WDOG_MAX}" ]]; do
-        _MODULE_CNT="$(cat ${MODULE_COUNTER} 2>&1)"
-        _SFP_CNT="$(cat ${SFP_COUNTER} 2>&1)"
+        _ASIC_INIT="$(cat ${ASIC_INIT_DONE} 2>&1)"
+        _ASIC_COUNT="$(cat ${NUM_ASICS} 2>&1)"
+        _ASICS_CHIPUP="$(cat ${ASIC_CHIPUP_COMPLETED} 2>&1)"
 
-        if [[ "${_MODULE_CNT}" =~ ${_NUM_MATCH} && "${_SFP_CNT}" =~ ${_NUM_MATCH} ]]; then
-            if [[ "${_SFP_CNT}" -gt "${_NUM_ZERO}" && "${_MODULE_CNT}" -eq "${_SFP_CNT}" ]]; then
-                return "${EXIT_SUCCESS}"
-            fi
+        if [[ "${_ASIC_INIT}" -eq 1 && "${_ASIC_COUNT}" -eq "${_ASICS_CHIPUP}" ]]; then
+            return "${EXIT_SUCCESS}"
         fi
 
         let "_WDOG_CNT++"
         sleep "${_TIMEOUT}"
     done
 
+    log_error "Mellanox ASIC is not ready: INIT: ${_ASIC_INIT}, NUM_ASIC: ${_ASIC_COUNT}, CHIPUP: ${_ASICS_CHIPUP} timeout...."
     return "${EXIT_TIMEOUT}"
 }
 
-log_info "Wait for SFP interfaces to be ready"
+log_info "Wait for Mellanox ASIC to be ready"
 
-wait_for_sfp
+wait_for_asic_chipup
 EXIT_CODE="$?"
 if [[ "${EXIT_CODE}" != "${EXIT_SUCCESS}" ]]; then
-    log_error "SFP interfaces are not ready: timeout"
     exit "${EXIT_CODE}"
 fi
 
-log_info "SFP interfaces are ready"
+log_notice "Mellanox ASIC is ready"
 
 exit "${EXIT_SUCCESS}"

files/scripts/syncd.sh

@@ -105,17 +105,6 @@ function waitplatform() {
             debug "Started pmon service"
         fi
     fi
-    if [[ x"$BOOT_TYPE" = @(x"fast"|x"warm"|x"fastfast") ]]; then
-        debug "LLDP service is delayed by a timer for better fast/warm boot performance"
-    else
-        lldp_state=$(systemctl is-enabled lldp.timer)
-        if [[ $lldp_state == "enabled" ]]
-        then
-            debug "Starting lldp service..."
-            /bin/systemctl start lldp
-            debug "Started lldp service"
-        fi
-    fi
 }
 
 function stopplatform1() {

platform/mellanox/integration-scripts/sdk_kernel_patches.py

@@ -82,7 +82,7 @@ def get_kernel_dir(self):
         major_kernel_path = os.path.join(KERNEL_BACKPORTS, "{}.{}".format(kernel, major))
 
         # if the k_dir with actual minor doesn't exit, use the closest minor version
-        for minor_i in range(minor_int, 0, -1):
+        for minor_i in range(minor_int, -1, -1):
             path = os.path.join(major_kernel_path, "{}.{}.{}".format(kernel, major, minor_i))
             if os.path.exists(os.path.join(self.args.patches, path)):
                 minor = str(minor_i)
@@ -137,6 +137,18 @@ def get_new_patches(self):
         Data.new_patches = FileHandler.read_dir(patches_path, "*.patch")
         Data.new_patches.sort()
 
+    def cleanup_old_patches(self):
+        patches_del = copy.deepcopy(Data.old_patches)
+        for patch in Data.new_patches:
+            if patch in Data.old_patches:
+                patches_del.remove(patch)
+        print(f" -> Patches to be removed are : {patches_del}")
+        for patch in patches_del:
+            file_n = os.path.join(self.args.build_root, os.path.join(SLK_PATCH_LOC, patch))
+            if os.path.exists(file_n):
+                print(f"{file_n} is deleted")
+                os.remove(file_n)
+
     def refresh_markers(self):
         print("-> INFO Refreshing Markers ")
         (Data.i_sdk_start, Data.i_sdk_end) = FileHandler.find_marker_indices(Data.old_series, SDK_MARKER)
@@ -187,6 +199,7 @@ def perform(self):
         self.refresh_markers()
         self.add_new_patch_series()
         self.process_update()
+        self.cleanup_old_patches()
         patch_table = self.fetch_patch_table(os.path.join(self.args.patches, Data.k_dir))
         slk_msg = self.create_commit_msg(patch_table)
         if self.args.slk_msg:

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

@@ -160,7 +160,7 @@ def is_psu_hotswapable(cls):
     @classmethod
     @utils.read_only_cache()
     def get_sfp_count(cls):
-        return utils.read_int_from_file('/run/hw-management/config/sfp_counter')
+        return utils.read_int_from_file('/run/hw-management/config/module_counter')
 
     @classmethod
     def get_linecard_sfp_count(cls, lc_index):

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

@@ -216,7 +216,10 @@ def get_target_speed(self):
         """
         try:
             # Get PSU fan target speed according to current system cooling level
-            cooling_level = utils.read_int_from_file('/run/hw-management/thermal/cooling_cur_state', log_func=None)
+            pwm = utils.read_int_from_file('/run/hw-management/thermal/pwm1', log_func=None)
+            if pwm >= PWM_MAX:
+                pwm = PWM_MAX - 1
+            cooling_level = int(pwm / PWM_MAX * 10)
             return int(self.PSU_FAN_SPEED[cooling_level], 16)
         except Exception:
             return self.get_speed()

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

@@ -242,8 +242,8 @@ def __init__(self, psu_index):
 
         self.psu_power_max_capacity = os.path.join(PSU_PATH, "config/psu{}_power_capacity".format(self.index))
 
-        self.psu_temp = os.path.join(PSU_PATH, 'thermal/psu{}_temp'.format(self.index))
-        self.psu_temp_threshold = os.path.join(PSU_PATH, 'thermal/psu{}_temp_max'.format(self.index))
+        self.psu_temp = os.path.join(PSU_PATH, 'thermal/psu{}_temp1'.format(self.index))
+        self.psu_temp_threshold = os.path.join(PSU_PATH, 'thermal/psu{}_temp1_max'.format(self.index))
 
         self.psu_power_slope = os.path.join(PSU_PATH, self.PSU_POWER_SLOPE.format(self.index))
 

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

@@ -65,16 +65,16 @@
     "psu thermals":
     {
         "name": "PSU-{} Temp",
-        "temperature": "psu{}_temp",
-        "high_threshold": "psu{}_temp_max",
+        "temperature": "psu{}_temp1",
+        "high_threshold": "psu{}_temp1_max",
         "type": "indexable"
     },
     "chassis thermals": [
         {
             "name": "ASIC",
             "temperature": "asic",
-            "high_threshold": "mlxsw/temp_trip_hot",
-            "high_critical_threshold": "mlxsw/temp_trip_crit"
+            "high_threshold": "asic_temp_emergency",
+            "high_critical_threshold": "asic_temp_trip_crit"
         },
         {
             "name": "Ambient Port Side Temp",
@@ -105,8 +105,8 @@
         {
             "name": "Gearbox {} Temp",
             "temperature": "gearbox{}_temp_input",
-            "high_threshold": "mlxsw-gearbox{}/temp_trip_hot",
-            "high_critical_threshold": "mlxsw-gearbox{}/temp_trip_crit",
+            "high_threshold": "gearbox{}_temp_emergency",
+            "high_critical_threshold": "gearbox{}_temp_trip_crit",
             "type": "indexable"
         },
         {
@@ -135,8 +135,8 @@
     'linecard thermals': {
         "name": "Gearbox {} Temp",
         "temperature": "gearbox{}_temp_input",
-        "high_threshold": "mlxsw-gearbox{}/temp_trip_hot",
-        "high_critical_threshold": "mlxsw-gearbox{}/temp_trip_crit",
+        "high_threshold": "gearbox{}_temp_emergency",
+        "high_critical_threshold": "gearbox{}_temp_trip_crit",
         "type": "indexable"
     }
 }
@@ -268,16 +268,6 @@ def _check_thermal_sysfs_existence(file_path):
 
 
 class Thermal(ThermalBase):
-    thermal_algorithm_status = False
-    # Expect cooling level, used for caching the cooling level value before commiting to hardware
-    expect_cooling_level = None
-    # Expect cooling state
-    expect_cooling_state = None
-    # Last committed cooling level
-    last_set_cooling_level = None
-    last_set_cooling_state = None
-    last_set_psu_cooling_level = None
-
     def __init__(self, name, temp_file, high_th_file, high_crit_th_file, position):
         """
         index should be a string for category ambient and int for other categories

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

@@ -1,5 +1,5 @@
 #
-# Copyright (c) 2021-2022 NVIDIA CORPORATION & AFFILIATES.
+# Copyright (c) 2021-2023 NVIDIA CORPORATION & AFFILIATES.
 # Apache-2.0
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -42,14 +42,3 @@ def auto_recover_mock():
     utils.read_str_from_file = origin_read_str_from_file
     utils.write_file = origin_write_file
     utils.read_float_from_file = origin_read_float_from_file
-
-
-@pytest.fixture(scope='function', autouse=True)
-def auto_reset_cooling_level():
-    from sonic_platform.thermal import Thermal
-    yield
-    Thermal.expect_cooling_level = None
-    Thermal.expect_cooling_state = None
-    Thermal.last_set_cooling_level = None
-    Thermal.last_set_cooling_state = None
-    Thermal.last_set_psu_cooling_level = None

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

@@ -124,8 +124,8 @@ def test_psu_fan_basic(self, mock_path_exists, mock_powergood, mock_presence, mo
         assert fan.get_presence() is False
         mock_path_exists.return_value = True
         assert fan.get_presence() is True
-        mock_read_int.return_value = 7
-        assert fan.get_target_speed() == 70
+        mock_read_int.return_value = int(255 / 10 * 7)
+        assert fan.get_target_speed() == 60
         mock_read_int.return_value = FAN_DIR_VALUE_INTAKE
         assert fan.get_direction() == Fan.FAN_DIRECTION_INTAKE
         mock_read_int.return_value = FAN_DIR_VALUE_EXHAUST

platform/mellanox/rules.mk

@@ -28,7 +28,7 @@ include $(PLATFORM_PATH)/docker-saiserver-mlnx.mk
 include $(PLATFORM_PATH)/one-image.mk
 include $(PLATFORM_PATH)/libsaithrift-dev.mk
 include $(PLATFORM_PATH)/mlnx-ffb.mk
-#include $(PLATFORM_PATH)/issu-version.mk
+include $(PLATFORM_PATH)/issu-version.mk
 include $(PLATFORM_PATH)/mlnx-onie-fw-update.mk
 include $(PLATFORM_PATH)/mlnx-ssd-fw-update.mk
 include $(PLATFORM_PATH)/iproute2.mk

platform/mellanox/sdk.dep

@@ -210,14 +210,13 @@ endif
 
 # SX_KERNEL
 
-# TODO: enable for bookworm
-#SPATH := $($(SX_KERNEL)_SRC_PATH)
-#SLINKS := $(shell find $(SPATH) -type l -exec echo {} \; | grep -Ev ' ')
-#SMDEP_PATHS := $(shell git submodule status --recursive -- $(SPATH) | awk '{print $$2}' | grep -Ev ' ')
-#SMDEP_FILES := $(foreach path,$(SMDEP_PATHS),$(filter-out $(SMDEP_PATHS),$(addprefix $(path)/,$(shell cd $(path) && git ls-files | grep -Ev ' '))))
-#DEP_FILES := $(SONIC_COMMON_FILES_LIST) $(PLATFORM_PATH)/sdk.mk $(PLATFORM_PATH)/sdk.dep
-#DEP_FILES += $(SONIC_COMMON_BASE_FILES_LIST)
-#DEP_FILES += $(filter-out $(SMDEP_PATHS),$(shell git ls-files -- $(SPATH) | grep -Ev ' '))
+SPATH := $($(SX_KERNEL)_SRC_PATH)
+SLINKS := $(shell find $(SPATH) -type l -exec echo {} \; | grep -Ev ' ')
+SMDEP_PATHS := $(shell git submodule status --recursive -- $(SPATH) | awk '{print $$2}' | grep -Ev ' ')
+SMDEP_FILES := $(foreach path,$(SMDEP_PATHS),$(filter-out $(SMDEP_PATHS),$(addprefix $(path)/,$(shell cd $(path) && git ls-files | grep -Ev ' '))))
+DEP_FILES := $(SONIC_COMMON_FILES_LIST) $(PLATFORM_PATH)/sdk.mk $(PLATFORM_PATH)/sdk.dep
+DEP_FILES += $(SONIC_COMMON_BASE_FILES_LIST)
+DEP_FILES += $(filter-out $(SMDEP_PATHS),$(shell git ls-files -- $(SPATH) | grep -Ev ' '))
 
 $(SX_KERNEL)_CACHE_MODE := GIT_CONTENT_SHA
 $(SX_KERNEL)_DEP_FLAGS := $(MLNX_SDK_COMMON_FLAGS_LIST)

platform/mellanox/sdk.mk

@@ -152,12 +152,11 @@ ifeq ($(SDK_FROM_SRC),y)
 $(eval $(call add_derived_package,$(SX_HASH_CALC),$(SX_HASH_CALC_DBGSYM)))
 endif
 
-# TODO: enable for bookworm
-#SX_KERNEL = sx-kernel_1.mlnx.$(MLNX_SDK_DEB_VERSION)_$(CONFIGURED_ARCH).deb
-#$(SX_KERNEL)_DEPENDS += $(LINUX_HEADERS) $(LINUX_HEADERS_COMMON)
-#$(SX_KERNEL)_SRC_PATH = $(PLATFORM_PATH)/sdk-src/sx-kernel
-#SX_KERNEL_DEV = sx-kernel-dev_1.mlnx.$(MLNX_SDK_DEB_VERSION)_$(CONFIGURED_ARCH).deb
-#$(eval $(call add_derived_package,$(SX_KERNEL),$(SX_KERNEL_DEV)))
+SX_KERNEL = sx-kernel_1.mlnx.$(MLNX_SDK_DEB_VERSION)_$(CONFIGURED_ARCH).deb
+$(SX_KERNEL)_DEPENDS += $(LINUX_HEADERS) $(LINUX_HEADERS_COMMON)
+$(SX_KERNEL)_SRC_PATH = $(PLATFORM_PATH)/sdk-src/sx-kernel
+SX_KERNEL_DEV = sx-kernel-dev_1.mlnx.$(MLNX_SDK_DEB_VERSION)_$(CONFIGURED_ARCH).deb
+$(eval $(call add_derived_package,$(SX_KERNEL),$(SX_KERNEL_DEV)))
 
 define make_url
 	$(1)_URL = $(MLNX_SDK_ASSETS_URL)/$(1)

rules/linux-kernel.dep

@@ -3,8 +3,8 @@ SPATH       := $($(LINUX_HEADERS_COMMON)_SRC_PATH)
 DEP_FILES   := rules/linux-kernel.mk rules/linux-kernel.dep
 SMDEP_FILES := $(addprefix $(SPATH)/,$(shell cd $(SPATH) && git ls-files))
 
-DEP_FLAGS := $(SONIC_COMMON_FLAGS_LIST) \
-	         $(KERNEL_PROCURE_METHOD) $(KERNEL_CACHE_PATH) $(SECURE_UPGRADE_MODE) $(SECURE_UPGRADE_SIGNING_CERT)
+DEP_FLAGS := $(SONIC_COMMON_FLAGS_LIST) $(INCLUDE_EXTERNAL_PATCHES) \
+	        $(KERNEL_PROCURE_METHOD) $(KERNEL_CACHE_PATH) $(SECURE_UPGRADE_MODE) $(SECURE_UPGRADE_SIGNING_CERT)
 
 $(LINUX_HEADERS_COMMON)_CACHE_MODE  := GIT_CONTENT_SHA
 $(LINUX_HEADERS_COMMON)_DEP_FLAGS   := $(DEP_FLAGS)

rules/linux-kernel.mk

@@ -12,7 +12,7 @@ endif
 # Place an URL here to .tar.gz file if you want to include those patches
 EXTERNAL_KERNEL_PATCH_URL =
 # Set y to include non upstream patches tarball provided by the corresponding platform
-INCLUDE_EXTERNAL_PATCHES = n
+INCLUDE_EXTERNAL_PATCHES ?= n
 # platforms should override this and provide an absolute location to the patches
 EXTERNAL_KERNEL_PATCH_LOC =
 

slave.mk

@@ -448,6 +448,7 @@ endif
 $(info "CROSS_BUILD_ENVIRON"             : "$(CROSS_BUILD_ENVIRON)")
 $(info "GZ_COMPRESS_PROGRAM"             : "$(GZ_COMPRESS_PROGRAM)")
 $(info "LEGACY_SONIC_MGMT_DOCKER"        : "$(LEGACY_SONIC_MGMT_DOCKER)")
+$(info "INCLUDE_EXTERNAL_PATCHES"        : "$(INCLUDE_EXTERNAL_PATCHES)")
 $(info )
 else
 $(info SONiC Build System for $(CONFIGURED_PLATFORM):$(CONFIGURED_ARCH))