[mellanox_firmware] PCC collector (opt in)

sos-nvdebug.conf

@@ -8,7 +8,7 @@
 batch = yes
 log-size = 10
 journal_size = 50
-plugin_timeout = 2000
+plugin_timeout = 2160
 cmd_timeout = 300
 low_priority = True
 

sos-nvidia.conf

@@ -9,7 +9,7 @@ batch = yes
 log-size = 10
 # 50 MB per journalctl capture, 7 days of journal text rarely exceeds this
 journal_size = 50
-plugin_timeout = 2000
+plugin_timeout = 2160
 cmd_timeout = 300
 low_priority = True
 

sos/report/mellanox_firmware_suite/collectors/collector_manager.py

@@ -1,6 +1,8 @@
 from .system_collector import SystemCollector
 from .firmware_collector import FirmwareCollector
 from .cable_collector import CableCollector
+from .pcc_collector import PccCollector
+from .doca_pcc_collector import DocaPccCollector
 
 
 class CollectorManager(object):
@@ -12,6 +14,7 @@ def collect_all(self):
         self.collect_system_info()
         self.collect_firmware_info()
         self.collect_cable_info()
+        self.collect_pcc_info()
 
     def collect_system_info(self):
         for ctx in self.device_contexts:
@@ -26,3 +29,13 @@ def collect_firmware_info(self):
     def collect_cable_info(self):
         for ctx in self.device_contexts:
             CableCollector().run(self.plugin, ctx)
+
+    def collect_pcc_info(self):
+        if not self.plugin.get_option("pcc", default=False):
+            return
+
+        for ctx in self.device_contexts:
+            PccCollector().run(self.plugin, ctx)
+
+        for ctx in self.device_contexts:
+            DocaPccCollector().run(self.plugin, ctx)

sos/report/mellanox_firmware_suite/collectors/doca_pcc_collector.py

@@ -0,0 +1,65 @@
+import os
+import re
+
+
+class DocaPccCollector(object):
+    _TOOL_PATH = "/opt/mellanox/doca/tools/pcc_counters.sh"
+    _OUTPUT_SUBDIR = "pcc_info/doca_pcc_counters"
+
+    # sw_dev_id values embedded in MST device names for supported devices:
+    #   ConnectX6DX: 4125  (/dev/mst/mt4125_*)
+    #   ConnectX7:   4129  (/dev/mst/mt4129_*)
+    #   BlueField3:  41692 (/dev/mst/mt41692_*)
+    _SUPPORTED_DEV_IDS = frozenset({"4125", "4129", "41692"})
+    _MST_DEV_ID_RE = re.compile(r"/dev/mst/mt(\d+)_")
+
+    def _is_supported_device(self, mst_device):
+        match = self._MST_DEV_ID_RE.search(mst_device)
+        return match is not None and match.group(1) in self._SUPPORTED_DEV_IDS
+
+    def run(self, plugin, ctx):
+        if not ctx.primary:
+            return
+
+        if ctx.mst_device is None:
+            plugin._log_info(
+                f"doca_pcc: skipping {ctx.pci} — no MST device path available"
+            )
+            return
+
+        if not self._is_supported_device(ctx.mst_device):
+            plugin._log_info(
+                f"doca_pcc: skipping {ctx.pci} — "
+                f"{ctx.mst_device!r} is not a supported device"
+            )
+            return
+
+        if not os.path.isfile(self._TOOL_PATH):
+            plugin._log_info(
+                f"doca_pcc: {self._TOOL_PATH} not found; skipping collection"
+            )
+            return
+
+        self._collect(plugin, ctx)
+
+    def _collect(self, plugin, ctx):
+        mst_dev = ctx.mst_device
+
+        result = plugin.exec_cmd(cmd=f"{self._TOOL_PATH} set {mst_dev}")
+        rc = result.get("status", 1)
+
+        if rc != 0:
+            plugin._log_info(
+                f"doca_pcc: setting counters failed for {mst_dev} (rc={rc}); "
+                "skipping query"
+            )
+            return
+
+        filename = f"pcc_counters_{ctx.bdf}_query"
+
+        plugin._collect_cmd_output(
+            cmd=f"{self._TOOL_PATH} query {mst_dev}",
+            suggest_filename=filename,
+            subdir=self._OUTPUT_SUBDIR,
+            stderr=True,
+        )