datadog 2.2.0 → 2.4.0

data/ext/datadog_profiling_native_extension/collectors_thread_context.h

@@ -3,13 +3,22 @@
 #include <ruby.h>
 #include <stdbool.h>
 
+#include "gvl_profiling_helper.h"
+
 void thread_context_collector_sample(
   VALUE self_instance,
   long current_monotonic_wall_time_ns,
   VALUE profiler_overhead_stack_thread
 );
 void thread_context_collector_sample_allocation(VALUE self_instance, unsigned int sample_weight, VALUE new_object);
+void thread_context_collector_sample_skipped_allocation_samples(VALUE self_instance, unsigned int skipped_samples);
 VALUE thread_context_collector_sample_after_gc(VALUE self_instance);
 void thread_context_collector_on_gc_start(VALUE self_instance);
-bool thread_context_collector_on_gc_finish(VALUE self_instance);
+__attribute__((warn_unused_result)) bool thread_context_collector_on_gc_finish(VALUE self_instance);
 VALUE enforce_thread_context_collector_instance(VALUE object);
+
+#ifndef NO_GVL_INSTRUMENTATION
+  void thread_context_collector_on_gvl_waiting(gvl_profiling_thread thread);
+  __attribute__((warn_unused_result)) bool thread_context_collector_on_gvl_running(gvl_profiling_thread thread);
+  VALUE thread_context_collector_sample_after_gvl_running(VALUE self_instance);
+#endif

data/ext/datadog_profiling_native_extension/datadog_ruby_common.c

@@ -0,0 +1,83 @@
+#include "datadog_ruby_common.h"
+
+// IMPORTANT: Currently this file is copy-pasted between extensions. Make sure to update all versions when doing any change!
+
+void raise_unexpected_type(VALUE value, const char *value_name, const char *type_name, const char *file, int line, const char* function_name) {
+  rb_exc_raise(
+    rb_exc_new_str(
+      rb_eTypeError,
+      rb_sprintf("wrong argument %"PRIsVALUE" for '%s' (expected a %s) at %s:%d:in `%s'",
+        rb_inspect(value),
+        value_name,
+        type_name,
+        file,
+        line,
+        function_name
+      )
+    )
+  );
+}
+
+VALUE datadog_gem_version(void) {
+  VALUE ddtrace_module = rb_const_get(rb_cObject, rb_intern("Datadog"));
+  ENFORCE_TYPE(ddtrace_module, T_MODULE);
+  VALUE version_module = rb_const_get(ddtrace_module, rb_intern("VERSION"));
+  ENFORCE_TYPE(version_module, T_MODULE);
+  VALUE version_string = rb_const_get(version_module, rb_intern("STRING"));
+  ENFORCE_TYPE(version_string, T_STRING);
+  return version_string;
+}
+
+static VALUE log_failure_to_process_tag(VALUE err_details) {
+  VALUE datadog_module = rb_const_get(rb_cObject, rb_intern("Datadog"));
+  VALUE logger = rb_funcall(datadog_module, rb_intern("logger"), 0);
+
+  return rb_funcall(logger, rb_intern("warn"), 1, rb_sprintf("Failed to convert tag: %"PRIsVALUE, err_details));
+}
+
+__attribute__((warn_unused_result))
+ddog_Vec_Tag convert_tags(VALUE tags_as_array) {
+  ENFORCE_TYPE(tags_as_array, T_ARRAY);
+
+  long tags_count = RARRAY_LEN(tags_as_array);
+  ddog_Vec_Tag tags = ddog_Vec_Tag_new();
+
+  for (long i = 0; i < tags_count; i++) {
+    VALUE name_value_pair = rb_ary_entry(tags_as_array, i);
+
+    if (!RB_TYPE_P(name_value_pair, T_ARRAY)) {
+      ddog_Vec_Tag_drop(tags);
+      ENFORCE_TYPE(name_value_pair, T_ARRAY);
+    }
+
+    // Note: We can index the array without checking its size first because rb_ary_entry returns Qnil if out of bounds
+    VALUE tag_name = rb_ary_entry(name_value_pair, 0);
+    VALUE tag_value = rb_ary_entry(name_value_pair, 1);
+
+    if (!(RB_TYPE_P(tag_name, T_STRING) && RB_TYPE_P(tag_value, T_STRING))) {
+      ddog_Vec_Tag_drop(tags);
+      ENFORCE_TYPE(tag_name, T_STRING);
+      ENFORCE_TYPE(tag_value, T_STRING);
+    }
+
+    ddog_Vec_Tag_PushResult push_result =
+      ddog_Vec_Tag_push(&tags, char_slice_from_ruby_string(tag_name), char_slice_from_ruby_string(tag_value));
+
+    if (push_result.tag == DDOG_VEC_TAG_PUSH_RESULT_ERR) {
+      // libdatadog validates tags and may catch invalid tags that ddtrace didn't actually catch.
+      // We warn users about such tags, and then just ignore them.
+
+      int exception_state;
+      rb_protect(log_failure_to_process_tag, get_error_details_and_drop(&push_result.err), &exception_state);
+
+      // Since we are calling into Ruby code, it may raise an exception. Ensure that dynamically-allocated tags
+      // get cleaned before propagating the exception.
+      if (exception_state) {
+        ddog_Vec_Tag_drop(tags);
+        rb_jump_tag(exception_state);  // "Re-raise" exception
+      }
+    }
+  }
+
+  return tags;
+}

data/ext/datadog_profiling_native_extension/datadog_ruby_common.h

@@ -0,0 +1,53 @@
+#pragma once
+
+// IMPORTANT: Currently this file is copy-pasted between extensions. Make sure to update all versions when doing any change!
+
+#include <ruby.h>
+#include <datadog/profiling.h>
+
+// Used to mark symbols to be exported to the outside of the extension.
+// Consider very carefully before tagging a function with this.
+#define DDTRACE_EXPORT __attribute__ ((visibility ("default")))
+
+// Used to mark function arguments that are deliberately left unused
+#ifdef __GNUC__
+  #define DDTRACE_UNUSED  __attribute__((unused))
+#else
+  #define DDTRACE_UNUSED
+#endif
+
+#define ADD_QUOTES_HELPER(x) #x
+#define ADD_QUOTES(x) ADD_QUOTES_HELPER(x)
+
+// Ruby has a Check_Type(value, type) that is roughly equivalent to this BUT Ruby's version is rather cryptic when it fails
+// e.g. "wrong argument type nil (expected String)". This is a replacement that prints more information to help debugging.
+#define ENFORCE_TYPE(value, type) \
+  { if (RB_UNLIKELY(!RB_TYPE_P(value, type))) raise_unexpected_type(value, ADD_QUOTES(value), ADD_QUOTES(type), __FILE__, __LINE__, __func__); }
+
+#define ENFORCE_BOOLEAN(value) \
+  { if (RB_UNLIKELY(value != Qtrue && value != Qfalse)) raise_unexpected_type(value, ADD_QUOTES(value), "true or false", __FILE__, __LINE__, __func__); }
+
+NORETURN(void raise_unexpected_type(VALUE value, const char *value_name, const char *type_name, const char *file, int line, const char* function_name));
+
+// Helper to retrieve Datadog::VERSION::STRING
+VALUE datadog_gem_version(void);
+
+static inline ddog_CharSlice char_slice_from_ruby_string(VALUE string) {
+  ENFORCE_TYPE(string, T_STRING);
+  ddog_CharSlice char_slice = {.ptr = RSTRING_PTR(string), .len = RSTRING_LEN(string)};
+  return char_slice;
+}
+
+__attribute__((warn_unused_result))
+ddog_Vec_Tag convert_tags(VALUE tags_as_array);
+
+static inline VALUE ruby_string_from_error(const ddog_Error *error) {
+  ddog_CharSlice char_slice = ddog_Error_message(error);
+  return rb_str_new(char_slice.ptr, char_slice.len);
+}
+
+static inline VALUE get_error_details_and_drop(ddog_Error *error) {
+  VALUE result = ruby_string_from_error(error);
+  ddog_Error_drop(error);
+  return result;
+}

data/ext/datadog_profiling_native_extension/extconf.rb

@@ -1,11 +1,16 @@
 # rubocop:disable Style/StderrPuts
 # rubocop:disable Style/GlobalVars
 
-require_relative 'native_extension_helpers'
+require_relative "native_extension_helpers"
+require_relative "../libdatadog_extconf_helpers"
 
 SKIPPED_REASON_FILE = "#{__dir__}/skipped_reason.txt".freeze
 # Not a problem if the file doesn't exist or we can't delete it
-File.delete(SKIPPED_REASON_FILE) rescue nil
+begin
+  File.delete(SKIPPED_REASON_FILE)
+rescue
+  nil
+end
 
 def skip_building_extension!(reason)
   fail_install_if_missing_extension =
@@ -24,13 +29,13 @@ def skip_building_extension!(reason)
   )
 
   if fail_install_if_missing_extension
-    require 'mkmf'
+    require "mkmf"
     Logging.message(
-      '[datadog] Failure cause: ' \
+      "[datadog] Failure cause: " \
       "#{Datadog::Profiling::NativeExtensionHelpers::Supported.render_skipped_reason_file(**reason)}\n"
     )
   else
-    File.write('Makefile', 'all install clean: # dummy makefile that does nothing')
+    File.write("Makefile", "all install clean: # dummy makefile that does nothing")
   end
 
   exit
@@ -63,41 +68,31 @@ $stderr.puts(
 
 # NOTE: we MUST NOT require 'mkmf' before we check the #skip_building_extension? because the require triggers checks
 # that may fail on an environment not properly setup for building Ruby extensions.
-require 'mkmf'
+require "mkmf"
 
 Logging.message("[datadog] Using compiler:\n")
-xsystem("#{CONFIG['CC']} -v")
+xsystem("#{CONFIG["CC"]} -v")
 Logging.message("[datadog] End of compiler information\n")
 
-# mkmf on modern Rubies actually has an append_cflags that does something similar
-# (see https://github.com/ruby/ruby/pull/5760), but as usual we need a bit more boilerplate to deal with legacy Rubies
-def add_compiler_flag(flag)
-  if try_cflags(flag)
-    $CFLAGS << ' ' << flag
-  else
-    $stderr.puts("WARNING: '#{flag}' not accepted by compiler, skipping it")
-  end
-end
-
 # Because we can't control what compiler versions our customers use, shipping with -Werror by default is a no-go.
 # But we can enable it in CI, so that we quickly spot any new warnings that just got introduced.
-add_compiler_flag '-Werror' if ENV['DATADOG_GEM_CI'] == 'true'
+append_cflags "-Werror" if ENV["DATADOG_GEM_CI"] == "true"
 
 # Older gcc releases may not default to C99 and we need to ask for this. This is also used:
 # * by upstream Ruby -- search for gnu99 in the codebase
 # * by msgpack, another datadog gem dependency
 #   (https://github.com/msgpack/msgpack-ruby/blob/18ce08f6d612fe973843c366ac9a0b74c4e50599/ext/msgpack/extconf.rb#L8)
-add_compiler_flag '-std=gnu99'
+append_cflags "-std=gnu99"
 
 # Gets really noisy when we include the MJIT header, let's omit it (TODO: Use #pragma GCC diagnostic instead?)
-add_compiler_flag '-Wno-unused-function'
+append_cflags "-Wno-unused-function"
 
 # Allow defining variables at any point in a function
-add_compiler_flag '-Wno-declaration-after-statement'
+append_cflags "-Wno-declaration-after-statement"
 
 # If we forget to include a Ruby header, the function call may still appear to work, but then
 # cause a segfault later. Let's ensure that never happens.
-add_compiler_flag '-Werror-implicit-function-declaration'
+append_cflags "-Werror-implicit-function-declaration"
 
 # The native extension is not intended to expose any symbols/functions for other native libraries to use;
 # the sole exception being `Init_datadog_profiling_native_extension` which needs to be visible for Ruby to call it when
@@ -105,22 +100,22 @@ add_compiler_flag '-Werror-implicit-function-declaration'
 #
 # By setting this compiler flag, we tell it to assume that everything is private unless explicitly stated.
 # For more details see https://gcc.gnu.org/wiki/Visibility
-add_compiler_flag '-fvisibility=hidden'
+append_cflags "-fvisibility=hidden"
 
 # Avoid legacy C definitions
-add_compiler_flag '-Wold-style-definition'
+append_cflags "-Wold-style-definition"
 
 # Enable all other compiler warnings
-add_compiler_flag '-Wall'
-add_compiler_flag '-Wextra'
+append_cflags "-Wall"
+append_cflags "-Wextra"
 
-if ENV['DDTRACE_DEBUG']
-  $defs << '-DDD_DEBUG'
-  CONFIG['optflags'] = '-O0'
-  CONFIG['debugflags'] = '-ggdb3'
+if ENV["DDTRACE_DEBUG"] == "true"
+  $defs << "-DDD_DEBUG"
+  CONFIG["optflags"] = "-O0"
+  CONFIG["debugflags"] = "-ggdb3"
 end
 
-if RUBY_PLATFORM.include?('linux')
+if RUBY_PLATFORM.include?("linux")
   # Supposedly, the correct way to do this is
   # ```
   # have_library 'pthread'
@@ -128,77 +123,83 @@ if RUBY_PLATFORM.include?('linux')
   # ```
   # but it's slower to build
   # so instead we just assume that we have the function we need on Linux, and nowhere else
-  $defs << '-DHAVE_PTHREAD_GETCPUCLOCKID'
+  $defs << "-DHAVE_PTHREAD_GETCPUCLOCKID"
+
+  # Not available on macOS
+  $defs << "-DHAVE_CLOCK_MONOTONIC_COARSE"
 end
 
-have_func 'malloc_stats'
+have_func "malloc_stats"
 
 # On older Rubies, rb_postponed_job_preregister/rb_postponed_job_trigger did not exist
-$defs << '-DNO_POSTPONED_TRIGGER' if RUBY_VERSION < '3.3'
+$defs << "-DNO_POSTPONED_TRIGGER" if RUBY_VERSION < "3.3"
 
 # On older Rubies, M:N threads were not available
-$defs << '-DNO_MN_THREADS_AVAILABLE' if RUBY_VERSION < '3.3'
+$defs << "-DNO_MN_THREADS_AVAILABLE" if RUBY_VERSION < "3.3"
 
 # On older Rubies, we did not need to include the ractor header (this was built into the MJIT header)
-$defs << '-DNO_RACTOR_HEADER_INCLUDE' if RUBY_VERSION < '3.3'
+$defs << "-DNO_RACTOR_HEADER_INCLUDE" if RUBY_VERSION < "3.3"
 
 # On older Rubies, some of the Ractor internal APIs were directly accessible
-$defs << '-DUSE_RACTOR_INTERNAL_APIS_DIRECTLY' if RUBY_VERSION < '3.3'
+$defs << "-DUSE_RACTOR_INTERNAL_APIS_DIRECTLY" if RUBY_VERSION < "3.3"
+
+# On older Rubies, there was no GVL instrumentation API and APIs created to support it
+$defs << "-DNO_GVL_INSTRUMENTATION" if RUBY_VERSION < "3.2"
+
+# Supporting GVL instrumentation on 3.2 needs some workarounds
+$defs << "-DUSE_GVL_PROFILING_3_2_WORKAROUNDS" if RUBY_VERSION.start_with?("3.2")
 
 # On older Rubies, there was no struct rb_native_thread. See private_vm_api_acccess.c for details.
-$defs << '-DNO_RB_NATIVE_THREAD' if RUBY_VERSION < '3.2'
+$defs << "-DNO_RB_NATIVE_THREAD" if RUBY_VERSION < "3.2"
 
 # On older Rubies, there was no struct rb_thread_sched (it was struct rb_global_vm_lock_struct)
-$defs << '-DNO_RB_THREAD_SCHED' if RUBY_VERSION < '3.2'
+$defs << "-DNO_RB_THREAD_SCHED" if RUBY_VERSION < "3.2"
 
 # On older Rubies, the first_lineno inside a location was a VALUE and not a int (https://github.com/ruby/ruby/pull/6430)
-$defs << '-DNO_INT_FIRST_LINENO' if RUBY_VERSION < '3.2'
+$defs << "-DNO_INT_FIRST_LINENO" if RUBY_VERSION < "3.2"
 
 # On older Rubies, "pop" was not a primitive operation
-$defs << '-DNO_PRIMITIVE_POP' if RUBY_VERSION < '3.2'
+$defs << "-DNO_PRIMITIVE_POP" if RUBY_VERSION < "3.2"
 
 # On older Rubies, there was no tid member in the internal thread structure
-$defs << '-DNO_THREAD_TID' if RUBY_VERSION < '3.1'
+$defs << "-DNO_THREAD_TID" if RUBY_VERSION < "3.1"
 
 # On older Rubies, there was no jit_return member on the rb_control_frame_t struct
-$defs << '-DNO_JIT_RETURN' if RUBY_VERSION < '3.1'
+$defs << "-DNO_JIT_RETURN" if RUBY_VERSION < "3.1"
 
 # On older Rubies, rb_gc_force_recycle allowed to free objects in a way that
 # would be invisible to free tracepoints, finalizers and without cleaning
 # obj_to_id_tbl mappings.
-$defs << '-DHAVE_WORKING_RB_GC_FORCE_RECYCLE' if RUBY_VERSION < '3.1'
-
-# On older Rubies, we need to use a backported version of this function. See private_vm_api_access.h for details.
-$defs << '-DUSE_BACKPORTED_RB_PROFILE_FRAME_METHOD_NAME' if RUBY_VERSION < '3'
+$defs << "-DHAVE_WORKING_RB_GC_FORCE_RECYCLE" if RUBY_VERSION < "3.1"
 
 # On older Rubies, there are no Ractors
-$defs << '-DNO_RACTORS' if RUBY_VERSION < '3'
+$defs << "-DNO_RACTORS" if RUBY_VERSION < "3"
 
 # On older Rubies, rb_imemo_name did not exist
-$defs << '-DNO_IMEMO_NAME' if RUBY_VERSION < '3'
+$defs << "-DNO_IMEMO_NAME" if RUBY_VERSION < "3"
 
 # On older Rubies, objects would not move
-$defs << '-DNO_T_MOVED' if RUBY_VERSION < '2.7'
+$defs << "-DNO_T_MOVED" if RUBY_VERSION < "2.7"
 
 # On older Rubies, there was no RUBY_SEEN_OBJ_ID flag
-$defs << '-DNO_SEEN_OBJ_ID_FLAG' if RUBY_VERSION < '2.7'
+$defs << "-DNO_SEEN_OBJ_ID_FLAG" if RUBY_VERSION < "2.7"
 
 # On older Rubies, rb_global_vm_lock_struct did not include the owner field
-$defs << '-DNO_GVL_OWNER' if RUBY_VERSION < '2.6'
+$defs << "-DNO_GVL_OWNER" if RUBY_VERSION < "2.6"
 
 # On older Rubies, there was no thread->invoke_arg
-$defs << '-DNO_THREAD_INVOKE_ARG' if RUBY_VERSION < '2.6'
+$defs << "-DNO_THREAD_INVOKE_ARG" if RUBY_VERSION < "2.6"
 
 # If we got here, libdatadog is available and loaded
-ENV['PKG_CONFIG_PATH'] = "#{ENV['PKG_CONFIG_PATH']}:#{Libdatadog.pkgconfig_folder}"
-Logging.message("[datadog] PKG_CONFIG_PATH set to #{ENV['PKG_CONFIG_PATH'].inspect}\n")
+ENV["PKG_CONFIG_PATH"] = "#{ENV["PKG_CONFIG_PATH"]}:#{Libdatadog.pkgconfig_folder}"
+Logging.message("[datadog] PKG_CONFIG_PATH set to #{ENV["PKG_CONFIG_PATH"].inspect}\n")
 $stderr.puts("Using libdatadog #{Libdatadog::VERSION} from #{Libdatadog.pkgconfig_folder}")
 
-unless pkg_config('datadog_profiling_with_rpath')
+unless pkg_config("datadog_profiling_with_rpath")
   Logging.message("[datadog] Ruby detected the pkg-config command is #{$PKGCONFIG.inspect}\n")
 
   skip_building_extension!(
-    if Datadog::Profiling::NativeExtensionHelpers::Supported.pkg_config_missing?
+    if Datadog::LibdatadogExtconfHelpers.pkg_config_missing?
       Datadog::Profiling::NativeExtensionHelpers::Supported::PKG_CONFIG_IS_MISSING
     else
       # Less specific error message
@@ -207,7 +208,7 @@ unless pkg_config('datadog_profiling_with_rpath')
   )
 end
 
-unless have_type('atomic_int', ['stdatomic.h'])
+unless have_type("atomic_int", ["stdatomic.h"])
   skip_building_extension!(Datadog::Profiling::NativeExtensionHelpers::Supported::COMPILER_ATOMIC_MISSING)
 end
 
@@ -215,8 +216,8 @@ end
 # The extremely excessive escaping around ORIGIN below seems to be correct and was determined after a lot of
 # experimentation. We need to get these special characters across a lot of tools untouched...
 extra_relative_rpaths = [
-  Datadog::Profiling::NativeExtensionHelpers.libdatadog_folder_relative_to_native_lib_folder,
-  *Datadog::Profiling::NativeExtensionHelpers.libdatadog_folder_relative_to_ruby_extensions_folders,
+  Datadog::LibdatadogExtconfHelpers.libdatadog_folder_relative_to_native_lib_folder(current_folder: __dir__),
+  *Datadog::LibdatadogExtconfHelpers.libdatadog_folder_relative_to_ruby_extensions_folders,
 ]
 extra_relative_rpaths.each { |folder| $LDFLAGS += " -Wl,-rpath,$$$\\\\{ORIGIN\\}/#{folder.to_str}" }
 Logging.message("[datadog] After pkg-config $LDFLAGS were set to: #{$LDFLAGS.inspect}\n")
@@ -237,8 +238,8 @@ if Datadog::Profiling::NativeExtensionHelpers::CAN_USE_MJIT_HEADER
   # use the MJIT header.
   # Finally, the `COMMON_HEADERS` conflict with the MJIT header so we need to temporarily disable them for this check.
   original_common_headers = MakeMakefile::COMMON_HEADERS
-  MakeMakefile::COMMON_HEADERS = ''.freeze
-  unless have_macro('RUBY_MJIT_H', mjit_header_file_name)
+  MakeMakefile::COMMON_HEADERS = "".freeze
+  unless have_macro("RUBY_MJIT_H", mjit_header_file_name)
     skip_building_extension!(Datadog::Profiling::NativeExtensionHelpers::Supported::COMPILATION_BROKEN)
   end
   MakeMakefile::COMMON_HEADERS = original_common_headers
@@ -250,7 +251,7 @@ if Datadog::Profiling::NativeExtensionHelpers::CAN_USE_MJIT_HEADER
 
   # Warn on unused parameters to functions. Use `DDTRACE_UNUSED` to mark things as known-to-not-be-used.
   # See the comment on the same flag below for why this is done last.
-  add_compiler_flag '-Wunused-parameter'
+  append_cflags "-Wunused-parameter"
 
   create_makefile EXTENSION_NAME
 else
@@ -261,23 +262,44 @@ else
 
   create_header
 
-  require 'debase/ruby_core_source'
-  dir_config('ruby') # allow user to pass in non-standard core include directory
+  require "debase/ruby_core_source"
+  dir_config("ruby") # allow user to pass in non-standard core include directory
+
+  # This is a workaround for a weird issue...
+  #
+  # The mkmf tool defines a `with_cppflags` helper that debase-ruby_core_source uses. This helper temporarily
+  # replaces `$CPPFLAGS` (aka the C pre-processor [not c++!] flags) with a different set when doing something.
+  #
+  # The debase-ruby_core_source gem uses `with_cppflags` during makefile generation to inject extra headers into the
+  # path. But because `with_cppflags` replaces `$CPPFLAGS`, well, the default `$CPPFLAGS` are not included in the
+  # makefile.
+  #
+  # This is a problem because the default `$CPPFLAGS` carries configuration that was set when Ruby was being built.
+  # Thus, if we ignore it, we don't compile the profiler with the exact same configuration as Ruby.
+  # In practice, this can generate crashes and weird bugs if the Ruby configuration is tweaked in a manner that
+  # changes some of the internal structures that the profiler relies on. Concretely, setting for instance
+  # `VM_CHECK_MODE=1` when building Ruby will trigger this issue (because somethings in structures the profiler reads
+  # are ifdef'd out using this setting).
+  #
+  # To workaround this issue, we override `with_cppflags` for debase-ruby_core_source to still include `$CPPFLAGS`.
+  Debase::RubyCoreSource.define_singleton_method(:with_cppflags) do |newflags, &block|
+    super("#{newflags} #{$CPPFLAGS}", &block)
+  end
 
   Debase::RubyCoreSource
     .create_makefile_with_core(
       proc do
         headers_available =
-          have_header('vm_core.h') &&
-          have_header('iseq.h') &&
-          (RUBY_VERSION < '3.3' || have_header('ractor_core.h'))
+          have_header("vm_core.h") &&
+          have_header("iseq.h") &&
+          (RUBY_VERSION < "3.3" || have_header("ractor_core.h"))
 
         if headers_available
           # Warn on unused parameters to functions. Use `DDTRACE_UNUSED` to mark things as known-to-not-be-used.
           # This is added as late as possible because in some Rubies we support (e.g. 3.3), adding this flag before
           # checking if internal VM headers are available causes those checks to fail because of this warning (and not
           # because the headers are not available.)
-          add_compiler_flag '-Wunused-parameter'
+          append_cflags "-Wunused-parameter"
         end
 
         headers_available

data/ext/datadog_profiling_native_extension/gvl_profiling_helper.c

@@ -0,0 +1,50 @@
+#include <ruby.h>
+#include <ruby/thread.h>
+#include "gvl_profiling_helper.h"
+
+#if !defined(NO_GVL_INSTRUMENTATION) && !defined(USE_GVL_PROFILING_3_2_WORKAROUNDS) // Ruby 3.3+
+  rb_internal_thread_specific_key_t gvl_waiting_tls_key;
+
+  void gvl_profiling_init(void) {
+    gvl_waiting_tls_key = rb_internal_thread_specific_key_create();
+  }
+
+#endif
+
+#ifdef USE_GVL_PROFILING_3_2_WORKAROUNDS // Ruby 3.2
+  __thread gvl_profiling_thread gvl_waiting_tls;
+  static bool gvl_profiling_state_thread_tracking_workaround_installed = false;
+
+  static void on_thread_start(
+    DDTRACE_UNUSED rb_event_flag_t _unused1,
+    DDTRACE_UNUSED const rb_internal_thread_event_data_t *_unused2,
+    DDTRACE_UNUSED void *_unused3
+  ) {
+    gvl_waiting_tls = (gvl_profiling_thread) {.thread = NULL};
+  }
+
+  // Hack: We're using the gvl_waiting_tls native thread-local to store per-thread information. Unfortunately, Ruby puts a big hole
+  // in our plan because it reuses native threads -- specifically, in Ruby 3.2, native threads are still 1:1 to Ruby
+  // threads (M:N wasn't a thing yet) BUT once a Ruby thread dies, the VM will keep the native thread around for a
+  // bit, and if another Ruby thread starts right after, Ruby will reuse the native thread, rather than create a new one.
+  //
+  // This will mean that the new Ruby thread will still have the same native thread-local data that we set on the
+  // old thread. For the purposes of our tracking, where we're keeping a pointer to the current thread object in
+  // thread-local storage **this is disastrous** since it means we'll be pointing at the wrong thread (and its
+  // memory may have been freed or reused since!)
+  //
+  // To work around this issue, once GVL profiling is enabled, we install an event hook on thread start
+  // events that clears the thread-local data. This guarantees that there will be no stale data -- any existing
+  // data will be cleared at thread start.
+  //
+  // Note that once installed, this event hook becomes permanent -- stopping the profiler does not stop this event
+  // hook, unlike all others. This is because we can't afford to miss any thread start events while the
+  // profiler is stopped (e.g. during reconfiguration) as that would mean stale data once the profiler starts again.
+  void gvl_profiling_state_thread_tracking_workaround(void) {
+    if (gvl_profiling_state_thread_tracking_workaround_installed) return;
+
+    rb_internal_thread_add_event_hook(on_thread_start, RUBY_INTERNAL_THREAD_EVENT_STARTED, NULL);
+
+    gvl_profiling_state_thread_tracking_workaround_installed = true;
+  }
+#endif

data/ext/datadog_profiling_native_extension/gvl_profiling_helper.h

@@ -0,0 +1,75 @@
+#pragma once
+
+// This helper is used by the Datadog::Profiling::Collectors::ThreadContext to store data used when profiling the GVL.
+// It's tested through that class' interfaces.
+// ---
+
+#include "extconf.h"
+
+#if !defined(NO_GVL_INSTRUMENTATION) && !defined(USE_GVL_PROFILING_3_2_WORKAROUNDS) // Ruby 3.3+
+  #include <ruby.h>
+  #include <ruby/thread.h>
+  #include "datadog_ruby_common.h"
+
+  typedef struct { VALUE thread; } gvl_profiling_thread;
+  extern rb_internal_thread_specific_key_t gvl_waiting_tls_key;
+
+  void gvl_profiling_init(void);
+
+  static inline gvl_profiling_thread thread_from_thread_object(VALUE thread) {
+    return (gvl_profiling_thread) {.thread = thread};
+  }
+
+  static inline gvl_profiling_thread thread_from_event(const rb_internal_thread_event_data_t *event_data) {
+    return thread_from_thread_object(event_data->thread);
+  }
+
+  static inline intptr_t gvl_profiling_state_get(gvl_profiling_thread thread) {
+    return (intptr_t) rb_internal_thread_specific_get(thread.thread, gvl_waiting_tls_key);
+  }
+
+  static inline void gvl_profiling_state_set(gvl_profiling_thread thread, intptr_t value) {
+    rb_internal_thread_specific_set(thread.thread, gvl_waiting_tls_key, (void *) value);
+  }
+#endif
+
+#ifdef USE_GVL_PROFILING_3_2_WORKAROUNDS // Ruby 3.2
+  typedef struct { void *thread; } gvl_profiling_thread;
+  extern __thread gvl_profiling_thread gvl_waiting_tls;
+
+  static inline void gvl_profiling_init(void) { }
+
+  // This header gets included in private_vm_access.c which can't include datadog_ruby_common.h so we replicate this
+  // helper here
+  #ifdef __GNUC__
+    #define DDTRACE_UNUSED  __attribute__((unused))
+  #else
+    #define DDTRACE_UNUSED
+  #endif
+
+  // NOTE: This is a hack that relies on the knowledge that on Ruby 3.2 the
+  // RUBY_INTERNAL_THREAD_EVENT_READY and RUBY_INTERNAL_THREAD_EVENT_RESUMED events always get called on the thread they
+  // are about. Thus, we can use our thread local storage hack to get this data, even though the event doesn't include it.
+  static inline gvl_profiling_thread thread_from_event(DDTRACE_UNUSED const void *event_data) {
+    return gvl_waiting_tls;
+  }
+
+  void gvl_profiling_state_thread_tracking_workaround(void);
+  gvl_profiling_thread gvl_profiling_state_maybe_initialize(void);
+
+  // Implementing these on Ruby 3.2 requires access to private VM things, so the following methods are
+  // implemented in `private_vm_api_access.c`
+  gvl_profiling_thread thread_from_thread_object(VALUE thread);
+  intptr_t gvl_profiling_state_get(gvl_profiling_thread thread);
+  void gvl_profiling_state_set(gvl_profiling_thread thread, intptr_t value);
+#endif
+
+#ifndef NO_GVL_INSTRUMENTATION // For all Rubies supporting GVL profiling (3.2+)
+  static inline intptr_t gvl_profiling_state_thread_object_get(VALUE thread) {
+    return gvl_profiling_state_get(thread_from_thread_object(thread));
+  }
+
+  static inline void gvl_profiling_state_thread_object_set(VALUE thread, intptr_t value) {
+    gvl_profiling_state_set(thread_from_thread_object(thread), value);
+  }
+#endif