Move Event Configuration to EventConfig class

Designwise, the EventConfig class should handle all the ugly details of configuring the Events from the given Config(), such as setting specific bits. The sample::Reader for example asks the EventConfig for a sample event. The EventConfig then selects the correct sampling event using the EventProvider and the Config, configures all the options, such as sample_period and then gives it back to the sample::Reader. The sample::Reader does not need to change anything about the event. The EventProvider should only provide events, reading them from different sources, such as pre-defined, PFM, PMU events. The CounterProvider did not do anything special enough to warrant its continuing existence, merge it with EventConfig Smaller fixes - delete needless getter in CounterCollection - provide default values for bools in Config() - check exclude_kernel once in EventConfig() instead of for every event
tud-zih-energy · Jan 22, 2025 · baf36c2 · baf36c2
1 parent b82a776
commit baf36c2
Show file tree

Hide file tree

Showing 23 changed files with 630 additions and 739 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -188,7 +188,7 @@ set(SOURCE_FILES
     src/perf/event.cpp
 
     src/perf/bio/block_device.cpp
-    src/perf/counter/counter_provider.cpp
+    src/perf/event_config.cpp
     src/perf/counter/group/reader.cpp
     src/perf/counter/userspace/reader.cpp
 

diff --git a/include/lo2s/config.hpp b/include/lo2s/config.hpp
@@ -83,23 +83,26 @@ struct Config
     std::uint64_t metric_count;
     std::uint64_t metric_frequency;
 
+    std::string metric_leader;
+    std::vector<std::string> group_counters;
+    std::vector<std::string> userspace_counters;
     // time synchronization
     bool use_clockid;
     bool use_pebs;
     clockid_t clockid;
     // x86_energy
-    bool use_x86_energy;
+    bool use_x86_energy = false;
     // block I/O
-    bool use_block_io;
+    bool use_block_io = false;
     // syscalls
     bool use_syscalls = false;
     std::vector<int64_t> syscall_filter;
     // NEC SX-Aurora Tsubasa
-    bool use_nec;
+    bool use_nec = false;
     std::chrono::microseconds nec_read_interval;
     std::chrono::milliseconds nec_check_interval;
     // Nvidia CUPTI
-    bool use_nvidia;
+    bool use_nvidia = false;
     std::string cuda_injectionlib_path;
     uint64_t nvidia_ringbuf_size;
 };

diff --git a/include/lo2s/perf/bio/writer.hpp b/include/lo2s/perf/bio/writer.hpp
@@ -21,6 +21,7 @@
 
 #pragma once
 
+#include <lo2s/perf/event_config.hpp>
 #include <lo2s/perf/io_reader.hpp>
 #include <lo2s/perf/time/converter.hpp>
 #include <lo2s/perf/tracepoint/format.hpp>
@@ -157,12 +158,10 @@ class Writer
 
     std::vector<perf::tracepoint::TracepointEvent> get_tracepoints()
     {
-        bio_queue_ =
-            perf::EventProvider::instance().create_tracepoint_event("block:block_bio_queue");
-        bio_issue_ =
-            perf::EventProvider::instance().create_tracepoint_event("block:block_rq_issue");
+        bio_queue_ = perf::EventConfig::instance().create_tracepoint_event("block:block_bio_queue");
+        bio_issue_ = perf::EventConfig::instance().create_tracepoint_event("block:block_rq_issue");
         bio_complete_ =
-            perf::EventProvider::instance().create_tracepoint_event("block:block_rq_complete");
+            perf::EventConfig::instance().create_tracepoint_event("block:block_rq_complete");
 
         return { bio_queue_.value(), bio_issue_.value(), bio_complete_.value() };
     }

diff --git a/include/lo2s/perf/counter/counter_collection.hpp b/include/lo2s/perf/counter/counter_collection.hpp
@@ -34,29 +34,28 @@ namespace counter
 {
 struct CounterCollection
 {
-    std::optional<Event> leader_;
+    CounterCollection() : leader(std::nullopt)
+    {
+    }
+
+    std::optional<Event> leader = std::nullopt;
     std::vector<Event> counters;
 
     double get_scale(int index) const
     {
         if (index == 0)
         {
-            return leader_.value().scale();
+            return leader.value().scale();
         }
         else
         {
             return counters[index - 1].scale();
         }
     }
 
-    Event& leader() const
-    {
-        return const_cast<Event&>(leader_.value());
-    }
-
     friend bool operator==(const CounterCollection& lhs, const CounterCollection& rhs)
     {
-        if (lhs.leader_.value() == rhs.leader_.value())
+        if (lhs.leader.value() == rhs.leader.value())
         {
             return lhs.counters == rhs.counters;
         }
@@ -65,11 +64,11 @@ struct CounterCollection
 
     friend bool operator<(const CounterCollection& lhs, const CounterCollection& rhs)
     {
-        if (lhs.leader_.value() == rhs.leader_.value())
+        if (lhs.leader.value() == rhs.leader.value())
         {
             return lhs.counters < rhs.counters;
         }
-        return lhs.leader_.value() < rhs.leader_.value();
+        return lhs.leader.value() < rhs.leader.value();
     }
 };
 

diff --git a/include/lo2s/perf/counter/counter_provider.hpp b/include/lo2s/perf/counter/counter_provider.hpp
diff --git a/include/lo2s/perf/event_config.hpp b/include/lo2s/perf/event_config.hpp
@@ -0,0 +1,70 @@
+/*
+ * This file is part of the lo2s software.
+ * Linux OTF2 sampling
+ *
+ * Copyright (c) 2016,
+ *    Technische Universitaet Dresden, Germany
+ *
+ * lo2s is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * lo2s is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with lo2s.  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#pragma once
+
+#include <lo2s/config.hpp>
+#include <lo2s/measurement_scope.hpp>
+#include <lo2s/perf/counter/counter_collection.hpp>
+#include <lo2s/perf/event_provider.hpp>
+
+namespace lo2s
+{
+namespace perf
+{
+class EventConfig
+{
+
+public:
+    EventConfig();
+
+    static EventConfig& instance()
+    {
+        static EventConfig e;
+        return e;
+    }
+
+    counter::CounterCollection counters_for(MeasurementScope scope);
+
+    Event create_time_event(uint64_t local_time);
+    Event create_sampling_event();
+    perf::tracepoint::TracepointEvent create_tracepoint_event(std::string name);
+    std::vector<perf::tracepoint::TracepointEvent> get_tracepoints();
+
+private:
+    // When we poll on the fd given by perf_event_open, wakeup, when our buffer is 80% full
+    // Default behaviour is to wakeup on every event, which is horrible performance wise
+    void watermark(Event& ev)
+    {
+        ev.set_watermark(0.8 * config().mmap_pages * sysconf(_SC_PAGESIZE));
+    }
+
+    void read_userspace_counters();
+    void read_group_counters();
+
+    std::optional<Event> sampling_event_;
+    std::optional<counter::CounterCollection> group_counters_;
+    std::optional<counter::CounterCollection> userspace_counters_;
+    std::optional<std::vector<tracepoint::TracepointEvent>> tracepoint_events_;
+    bool exclude_kernel_;
+};
+} // namespace perf
+} // namespace lo2s
diff --git a/include/lo2s/perf/event_provider.hpp b/include/lo2s/perf/event_provider.hpp
@@ -36,33 +36,16 @@ namespace perf
 class EventProvider
 {
 public:
-    EventProvider();
-    EventProvider(const EventProvider&) = delete;
-    void operator=(const EventProvider&) = delete;
-
-    static const EventProvider& instance()
-    {
-        return instance_mutable();
-    }
-
-    static Event get_event_by_name(const std::string& name);
+    Event get_event_by_name(const std::string& name);
 
-    static bool has_event(const std::string& name);
+    bool has_event(const std::string& name);
 
-    static std::vector<Event> get_predefined_events();
-    static std::vector<SysfsEvent> get_pmu_events();
+    std::vector<Event> get_predefined_events();
+    std::vector<SysfsEvent> get_pmu_events();
 
-    static Event fallback_metric_leader_event();
-
-    static Event create_time_event(std::uint64_t local_time, bool enable_on_exec = false);
-    static Event create_event(const std::string& name, perf_type_id type, std::uint64_t config,
-                              std::uint64_t config1 = 0);
-    static SysfsEvent create_sampling_event(bool enable_on_exec);
-    static SysfsEvent create_sysfs_event(const std::string& name, bool use_config = true);
-    static tracepoint::TracepointEvent create_tracepoint_event(const std::string& name,
-                                                               bool use_config = true,
-                                                               bool enable_on_exec = false);
+    Event get_metric_leader(std::string metric_leader);
 
+    std::vector<std::string> get_tracepoint_event_names();
     class InvalidEvent : public std::runtime_error
     {
     public:
@@ -72,15 +55,17 @@ class EventProvider
         }
     };
 
-private:
-    static EventProvider& instance_mutable()
+    static EventProvider& instance()
     {
         static EventProvider e;
         return e;
     }
 
-    static void apply_config_attrs(Event& event);
-    static void apply_default_attrs(Event& event);
+private:
+    Event fallback_metric_leader_event();
+    EventProvider();
+    EventProvider(const EventProvider&) = delete;
+    void operator=(const EventProvider&) = delete;
 
     Event cache_event(const std::string& name);
 

diff --git a/include/lo2s/perf/io_reader.hpp b/include/lo2s/perf/io_reader.hpp
@@ -22,6 +22,7 @@
 #pragma once
 
 #include <lo2s/measurement_scope.hpp>
+#include <lo2s/perf/event_config.hpp>
 #include <lo2s/perf/event_provider.hpp>
 #include <lo2s/perf/event_reader.hpp>
 #include <lo2s/perf/tracepoint/event.hpp>

diff --git a/include/lo2s/perf/sample/reader.hpp b/include/lo2s/perf/sample/reader.hpp
@@ -21,6 +21,7 @@
 
 #pragma once
 
+#include <lo2s/perf/event_config.hpp>
 #include <lo2s/perf/event_provider.hpp>
 #include <lo2s/perf/event_reader.hpp>
 #include <lo2s/perf/util.hpp>
@@ -84,40 +85,18 @@ class Reader : public EventReader<T>
         Log::debug() << "initializing event_reader for:" << scope.name()
                      << ", enable_on_exec: " << enable_on_exec;
 
-        Event event = EventProvider::instance().create_sampling_event(enable_on_exec);
-
-        do
+        Event event = EventConfig::instance().create_sampling_event();
+        if (enable_on_exec)
         {
-            try
-            {
-                event_ = event.open(scope, config().cgroup_fd);
-            }
-            catch (const std::system_error& e)
-            {
-                if (e.code().value() == EACCES && !event.attr().exclude_kernel &&
-                    perf_event_paranoid() > 1)
-                {
-                    event.mut_attr().exclude_kernel = 1;
-                    perf_warn_paranoid();
-                    continue;
-                }
-
-                if (!event.degrade_precision())
-                {
-                    Log::error() << "perf_event_open for sampling failed: " << e.what();
-
-                    if (event.attr().use_clockid)
-                    {
-                        Log::error() << "maybe the specified clock is unavailable?";
-                    }
-                    throw_errno();
-                }
-            }
-        } while (!event_.value().is_valid());
+            event.set_flags({ EventFlag::ENABLE_ON_EXEC });
+        }
+        else
+        {
+            event.set_flags({ EventFlag::DISABLED });
+        }
 
-        Log::debug() << "Using precise_ip level: " << event.attr().precise_ip;
+        event_ = event.open(scope, config().cgroup_fd);
 
-        // Exception safe, so much wow!
         try
         {
             init_mmap(event_.value().get_fd());