Kernel: Start work on full system profiling :^)

The superuser can now call sys$profiling_enable() with PID -1 to enable
profiling of all running threads in the system. The perf events are
collected in a global PerformanceEventBuffer (currently 32 MiB in size.)

The events can be accessed via /proc/profile

Kernel/FileSystem/ProcFS.cpp

@@ -474,6 +474,15 @@ static bool procfs$modules(InodeIdentifier, KBufferBuilder& builder)
     return true;
 }
 
+static bool procfs$profile(InodeIdentifier, KBufferBuilder& builder)
+{
+    extern PerformanceEventBuffer* g_global_perf_events;
+    if (!g_global_perf_events)
+        return false;
+
+    return g_global_perf_events->to_json(builder);
+}
+
 static bool procfs$pid_perf_events(InodeIdentifier identifier, KBufferBuilder& builder)
 {
     auto process = Process::from_pid(to_pid(identifier));
@@ -1703,6 +1712,7 @@ ProcFS::ProcFS()
     m_entries[FI_Root_uptime] = { "uptime", FI_Root_uptime, false, procfs$uptime };
     m_entries[FI_Root_cmdline] = { "cmdline", FI_Root_cmdline, true, procfs$cmdline };
     m_entries[FI_Root_modules] = { "modules", FI_Root_modules, true, procfs$modules };
+    m_entries[FI_Root_profile] = { "profile", FI_Root_profile, true, procfs$profile };
     m_entries[FI_Root_sys] = { "sys", FI_Root_sys, true };
     m_entries[FI_Root_net] = { "net", FI_Root_net, false };
 

Kernel/PerformanceEventBuffer.cpp

@@ -98,6 +98,7 @@ KResult PerformanceEventBuffer::append_with_eip_and_ebp(u32 eip, u32 ebp, int ty
     event.stack_size = min(sizeof(event.stack) / sizeof(FlatPtr), static_cast<size_t>(backtrace.size()));
     memcpy(event.stack, backtrace.data(), event.stack_size * sizeof(FlatPtr));
 
+    event.tid = Thread::current()->tid().value();
     event.timestamp = TimeManagement::the().uptime_ms();
     at(m_count++) = event;
     return KSuccess;
@@ -118,27 +119,9 @@ OwnPtr<KBuffer> PerformanceEventBuffer::to_json(ProcessID pid, const String& exe
     return builder.build();
 }
 
-bool PerformanceEventBuffer::to_json(KBufferBuilder& builder, ProcessID pid, const String& executable_path) const
+template<typename Serializer>
+bool PerformanceEventBuffer::to_json_impl(Serializer& object) const
 {
-    auto process = Process::from_pid(pid);
-    VERIFY(process);
-    ScopedSpinLock locker(process->space().get_lock());
-
-    JsonObjectSerializer object(builder);
-    object.add("pid", pid.value());
-    object.add("executable", executable_path);
-
-    {
-        auto region_array = object.add_array("regions");
-        for (const auto& region : process->space().regions()) {
-            auto region_object = region_array.add_object();
-            region_object.add("base", region.vaddr().get());
-            region_object.add("size", region.size());
-            region_object.add("name", region.name());
-        }
-        region_array.finish();
-    }
-
     auto array = object.add_array("events");
     for (size_t i = 0; i < m_count; ++i) {
         auto& event = at(i);
@@ -171,6 +154,36 @@ bool PerformanceEventBuffer::to_json(KBufferBuilder& builder, ProcessID pid, con
     return true;
 }
 
+bool PerformanceEventBuffer::to_json(KBufferBuilder& builder)
+{
+    JsonObjectSerializer object(builder);
+    return to_json_impl(object);
+}
+
+bool PerformanceEventBuffer::to_json(KBufferBuilder& builder, ProcessID pid, const String& executable_path) const
+{
+    auto process = Process::from_pid(pid);
+    VERIFY(process);
+    ScopedSpinLock locker(process->space().get_lock());
+
+    JsonObjectSerializer object(builder);
+    object.add("pid", pid.value());
+    object.add("executable", executable_path);
+
+    {
+        auto region_array = object.add_array("regions");
+        for (const auto& region : process->space().regions()) {
+            auto region_object = region_array.add_object();
+            region_object.add("base", region.vaddr().get());
+            region_object.add("size", region.size());
+            region_object.add("name", region.name());
+        }
+        region_array.finish();
+    }
+
+    return to_json_impl(object);
+}
+
 OwnPtr<PerformanceEventBuffer> PerformanceEventBuffer::try_create_with_size(size_t buffer_size)
 {
     auto buffer = KBuffer::try_create_with_size(buffer_size, Region::Access::Read | Region::Access::Write, "Performance events", AllocationStrategy::AllocateNow);

Kernel/PerformanceEventBuffer.h

@@ -78,9 +78,15 @@ public:
     OwnPtr<KBuffer> to_json(ProcessID, const String& executable_path) const;
     bool to_json(KBufferBuilder&, ProcessID, const String& executable_path) const;
 
+    // Used by full-system profile (/proc/profile)
+    bool to_json(KBufferBuilder&);
+
 private:
     explicit PerformanceEventBuffer(NonnullOwnPtr<KBuffer>);
 
+    template<typename Serializer>
+    bool to_json_impl(Serializer&) const;
+
     PerformanceEvent& at(size_t index);
 
     size_t m_count { 0 };

Kernel/Scheduler.cpp

@@ -42,6 +42,9 @@
 
 namespace Kernel {
 
+extern bool g_profiling_all_threads;
+extern PerformanceEventBuffer* g_global_perf_events;
+
 class SchedulerPerProcessorData {
     AK_MAKE_NONCOPYABLE(SchedulerPerProcessorData);
     AK_MAKE_NONMOVABLE(SchedulerPerProcessorData);
@@ -542,10 +545,22 @@ void Scheduler::timer_tick(const RegisterState& regs)
     if (!is_bsp)
         return; // TODO: This prevents scheduling on other CPUs!
 #endif
-    if (current_thread->process().is_profiling()) {
+
+    PerformanceEventBuffer* perf_events = nullptr;
+
+    if (g_profiling_all_threads) {
+        VERIFY(g_global_perf_events);
+        // FIXME: We currently don't collect samples while idle.
+        //        That will be an interesting mode to add in the future. :^)
+        if (current_thread != Processor::current().idle_thread())
+            perf_events = g_global_perf_events;
+    } else if (current_thread->process().is_profiling()) {
         VERIFY(current_thread->process().perf_events());
-        auto& perf_events = *current_thread->process().perf_events();
+        perf_events = current_thread->process().perf_events();
-        [[maybe_unused]] auto rc = perf_events.append_with_eip_and_ebp(regs.eip, regs.ebp, PERF_EVENT_SAMPLE, 0, 0);
+    }
+
+    if (perf_events) {
+        [[maybe_unused]] auto rc = perf_events->append_with_eip_and_ebp(regs.eip, regs.ebp, PERF_EVENT_SAMPLE, 0, 0);
     }
 
     if (current_thread->tick())

Kernel/Syscalls/profiling.cpp

@@ -32,9 +32,25 @@
 
 namespace Kernel {
 
+PerformanceEventBuffer* g_global_perf_events;
+bool g_profiling_all_threads;
+
 KResultOr<int> Process::sys$profiling_enable(pid_t pid)
 {
     REQUIRE_NO_PROMISES;
+
+    if (pid == -1) {
+        if (!is_superuser())
+            return EPERM;
+        ScopedCritical critical;
+        if (g_global_perf_events)
+            g_global_perf_events->clear();
+        else
+            g_global_perf_events = PerformanceEventBuffer::try_create_with_size(32 * MiB).leak_ptr();
+        g_profiling_all_threads = true;
+        return 0;
+    }
+
     ScopedSpinLock lock(g_processes_lock);
     auto process = Process::from_pid(pid);
     if (!process)
@@ -51,6 +67,14 @@ KResultOr<int> Process::sys$profiling_enable(pid_t pid)
 
 KResultOr<int> Process::sys$profiling_disable(pid_t pid)
 {
+    if (pid == -1) {
+        if (!is_superuser())
+            return EPERM;
+        ScopedCritical critical;
+        g_profiling_all_threads = false;
+        return 0;
+    }
+
     ScopedSpinLock lock(g_processes_lock);
     auto process = Process::from_pid(pid);
     if (!process)