Improve support for applications that rely on fork for parallelism

A forked process inherits the elfmap of the parent, thus we need to
copy that data when we encounter a fork event. To do so reliably,
we need to flush the event buffer, as otherwise the mmmap state may
not be accurate.

Furthermore, pass the ppid through the ThreadStart event, to allow
client applications to inherit the comm for the newly created process.

Fixes: KDAB/hotspot#241
Change-Id: I5de13644e12def6704c5f622428a815fd87d2af4

app/perfdata.h

@@ -576,6 +576,8 @@ class PerfRecordFork : public PerfRecord
                    bool sampleIdAll = false);
     qint32 childTid() const { return m_tid; }
     qint32 childPid() const { return m_pid; }
+    qint32 parentTid() const { return m_ptid; }
+    qint32 parentPid() const { return m_ppid; }
 private:
     qint32 m_pid, m_ppid;
     qint32 m_tid, m_ptid;

app/perfelfmap.h

@@ -106,6 +106,12 @@ class PerfElfMap : public QObject
 
     bool isAddressInRange(quint64 addr) const;
 
+    void copyDataFrom(const PerfElfMap *parent)
+    {
+        m_elfs = parent->m_elfs;
+        m_lastBase = parent->m_lastBase;
+    }
+
 signals:
     void aboutToInvalidate(const ElfInfo &elf);
 

app/perfsymboltable.cpp

@@ -1175,3 +1175,8 @@ PerfSymbolTable::ElfAndFile::ElfAndFile(PerfSymbolTable::ElfAndFile &&other)
     other.m_elf = nullptr;
     other.m_file = -1;
 }
+
+void PerfSymbolTable::initAfterFork(const PerfSymbolTable* parent)
+{
+    m_elfs.copyDataFrom(&parent->m_elfs);
+}

app/perfsymboltable.h

@@ -74,6 +74,8 @@ class PerfSymbolTable
     void clearCache();
     bool cacheIsDirty() const { return m_cacheIsDirty; }
 
+    void initAfterFork(const PerfSymbolTable *parent);
+
 private:
     // Report an mmap to dwfl and parse it for symbols and inlines, or simply return it if dwfl has
     // it already

app/perfunwind.cpp

@@ -689,8 +689,15 @@ void PerfUnwind::analyze(const PerfRecordSample &sample)
 void PerfUnwind::fork(const PerfRecordFork &sample)
 {
     bufferEvent(TaskEvent{sample.childPid(), sample.childTid(), sample.time(), sample.cpu(),
-                          0, ThreadStart},
+                          sample.parentPid(), ThreadStart},
                 &m_taskEventsBuffer, &m_stats.numTaskEventsInRound);
+
+    if (sample.childPid() != sample.parentPid()) {
+        // we need to flush to ensure the parent symbol table contains an accurate
+        // representation of the current elf mappings
+        flushEventBuffer(0);
+        symbolTable(sample.childPid())->initAfterFork(symbolTable(sample.parentPid()));
+    }
 }
 
 void PerfUnwind::exit(const PerfRecordExit &sample)
@@ -1034,7 +1041,7 @@ void PerfUnwind::sendTaskEvent(const TaskEvent& taskEvent)
 
     if (taskEvent.m_type == ContextSwitchDefinition)
         stream << static_cast<bool>(taskEvent.m_payload);
-    else if (taskEvent.m_type == Command)
+    else if (taskEvent.m_type == Command || taskEvent.m_type == ThreadStart)
         stream << taskEvent.m_payload;
 
     sendBuffer(buffer);