Push 2024 12 11

src/cpucounters.cpp

@@ -1444,29 +1444,19 @@ bool PCM::discoverSystemTopology()
     }
     if(num_sockets == 0) {
         num_sockets = (int32)(std::max)(socketIdMap.size(), (size_t)1);
+        // std::cerr << " num_sockets = " << num_sockets << "\n";
     }
 
     socketIdMap_type::iterator s = socketIdMap.begin();
     for (uint32 sid = 0; s != socketIdMap.end(); ++s)
     {
         s->second = sid++;
-        // first is apic id, second is logical socket id
-        systemTopology->addSocket( s->first, s->second );
     }
 
-    for (int32 cid = 0; cid < num_cores; ++cid)
-    {
-        //std::cerr << "Cid: " << cid << "\n";
-        systemTopology->addThread( cid, topology[cid] );
-    }
-
-    // All threads are here now so we can set the refCore for a socket
-    for ( auto& socket : systemTopology->sockets() )
-        socket->setRefCore();
-
     // use map to change apic socket id to the logical socket id
     for (int i = 0; (i < (int)num_cores) && (!socketIdMap.empty()); ++i)
     {
+        // std::cerr << "socket_id: " << topology[i].socket_id << ", socketIdMap tells me: " << socketIdMap[topology[i].socket_id] << "\n";
         if(isCoreOnline((int32)i))
           topology[i].socket_id = socketIdMap[topology[i].socket_id];
     }
@@ -1483,14 +1473,30 @@ bool PCM::discoverSystemTopology()
     {
         for (int i = 0; i < (int)num_cores; ++i)
         {
-            if (topology[i].socket_id == topology[0].socket_id && topology[i].core_id == topology[0].core_id)
+            if (topology[i].isSameCore( topology[0] ))
                 ++threads_per_core;
         }
         assert(threads_per_core != 0);
     }
     if(num_phys_cores_per_socket == 0 && num_cores == num_online_cores) num_phys_cores_per_socket = num_cores / num_sockets / threads_per_core;
     if(num_online_cores == 0) num_online_cores = num_cores;
 
+    s = socketIdMap.begin();
+    for (; s != socketIdMap.end(); ++s)
+    {
+        systemTopology->addSocket( s->second );
+    }
+
+    for (int32 cid = 0; cid < num_cores; ++cid)
+    {
+        //std::cerr << "Cid: " << cid << "\n";
+        systemTopology->addThread( cid, topology[cid] );
+    }
+
+    // All threads are here now so we can set the refCore for a socket
+    for ( auto& socket : systemTopology->sockets() )
+        socket->setRefCore();
+
     int32 i = 0;
 
     socketRefCore.resize(num_sockets, -1);
@@ -3421,11 +3427,11 @@ void PCM::destroyMSR()
 
 PCM::~PCM()
 {
+    deleteAndNullify(systemTopology);
     if (instance)
     {
         destroyMSR();
         instance = NULL;
-        deleteAndNullify(systemTopology);
     }
 }
 

src/pcm-sensor-server.cpp

@@ -645,7 +645,7 @@ class PrometheusPrinter : Visitor
     }
 
     virtual void dispatch( Core* c ) override {
-        addToHierarchy( std::string( "core=\"" ) + std::to_string( c->coreID() ) + "\"" );
+        addToHierarchy( std::string( "core=\"" ) + std::to_string( c->dieGroupID()*256 + c->dieID()*64 + c->tileID()*16 + c->moduleID()*4 + c->coreID() ) + "\"" );
         auto vec = c->threads();
         iterateVectorAndCallAccept( vec );
 
@@ -801,6 +801,7 @@ class PrometheusPrinter : Visitor
         }
         removeFromHierarchy();
     }
+
     void printSystemCounterState( SystemCounterState const& before, SystemCounterState const& after ) {
         addToHierarchy( "source=\"uncore\"" );
         PCM* pcm = PCM::getInstance();
@@ -3759,6 +3760,7 @@ int mainThrows(int argc, char * argv[]) {
     bool useRealtimePriority = false;
 #endif
     bool forceRTMAbortMode = false;
+    bool printTopology = false;
     unsigned short port = 0;
     unsigned short debug_level = 0;
     std::string certificateFile;
@@ -3774,7 +3776,12 @@ int mainThrows(int argc, char * argv[]) {
     MainLoop mainLoop;
     std::string ev_file_name;
 
-    if ( argc > 1 ) {
+    const char* PPTEnv = std::getenv( "PCMSENSORSERVER_PRINT_TOPOLOGY" );
+    if ( PPTEnv ) {
+        if ( *PPTEnv == '1' ) {
+            printTopology = true;
+        }
+    } else if ( argc > 1 ) {
         std::string arg_value;
 
         for ( int i=1; i < argc; ++i ) {
@@ -3901,7 +3908,7 @@ int mainThrows(int argc, char * argv[]) {
         }
     }
 
-    #ifdef __linux__
+#ifdef __linux__
     // check kernel version for driver dependency.
     if (accel != ACCEL_NOCONFIG)
     {
@@ -4035,6 +4042,17 @@ int mainThrows(int argc, char * argv[]) {
 
             accs_->programAccelCounters();
         }
+        if ( printTopology ) {
+            TopologyPrinter* tp = new TopologyPrinter();
+            tp->dispatch( PCM::getInstance()->getSystemTopology() );
+            std::vector<std::string> & tpData = tp->topologyDataStrings();
+            std::sort( tpData.begin(), tpData.end(), TopologyStringCompare );
+            for( auto& line: tpData ) {
+                std::cout << line << "\n";
+            }
+            deleteAndNullify( tp );
+            exit( 0 );
+        }
 #if defined (USE_SSL)
         if ( useSSL ) {
             if ( port == 0 )

src/topology.cpp

@@ -30,8 +30,8 @@ UncoreCounterState ClientUncore::uncoreCounterState( void ) const
     return ucs;
 }
 
-Socket::Socket( PCM* m, int32 apicID, int32 logicalID )
-    : pcm_(m), refCore_(nullptr), apicID_(apicID), logicalID_(logicalID)
+Socket::Socket( PCM* m, int32 logicalID )
+    : pcm_(m), refCore_(nullptr), logicalID_(logicalID)
 {
     if ( pcm_->isServerCPU() )
         uncore_ = new ServerUncore( pcm_, logicalID );
@@ -102,4 +102,16 @@ void Aggregator::dispatch( SystemRoot const& syp ) {
     readAccelCounters(sycs_);
 }
 
+bool TopologyStringCompare( const std::string& topology1, const std::string& topology2 ) {
+    if ( topology1.size() == 0 ) return true;
+    if ( topology2.size() == 0 ) return false;
+
+    int topo1asint, topo2asint;
+    std::stringstream ss1(topology1);
+    std::stringstream ss2(topology2);
+    ss1 >> topo1asint;
+    ss2 >> topo2asint;
+    return topo1asint < topo2asint;
+}
+
 }// namespace pcm