24 files changed, 5260 insertions, 0 deletions

diff --git a/trace2.c b/trace2.c
new file mode 100644
index 0000000..279bddf
--- /dev/null
+++ b/trace2.c
@@ -0,0 +1,914 @@
+#include "cache.h"
+#include "config.h"
+#include "json-writer.h"
+#include "quote.h"
+#include "run-command.h"
+#include "sigchain.h"
+#include "thread-utils.h"
+#include "version.h"
+#include "trace2/tr2_cfg.h"
+#include "trace2/tr2_cmd_name.h"
+#include "trace2/tr2_ctr.h"
+#include "trace2/tr2_dst.h"
+#include "trace2/tr2_sid.h"
+#include "trace2/tr2_sysenv.h"
+#include "trace2/tr2_tgt.h"
+#include "trace2/tr2_tls.h"
+#include "trace2/tr2_tmr.h"
+
+static int trace2_enabled;
+
+static int tr2_next_child_id; /* modify under lock */
+static int tr2_next_exec_id; /* modify under lock */
+static int tr2_next_repo_id = 1; /* modify under lock. zero is reserved */
+
+/*
+ * A table of the builtin TRACE2 targets.  Each of these may be independently
+ * enabled or disabled.  Each TRACE2 API method will try to write an event to
+ * *each* of the enabled targets.
+ */
+/* clang-format off */
+static struct tr2_tgt *tr2_tgt_builtins[] =
+{
+	&tr2_tgt_normal,
+	&tr2_tgt_perf,
+	&tr2_tgt_event,
+	NULL
+};
+/* clang-format on */
+
+/* clang-format off */
+#define for_each_builtin(j, tgt_j)			\
+	for (j = 0, tgt_j = tr2_tgt_builtins[j];	\
+	     tgt_j;					\
+	     j++, tgt_j = tr2_tgt_builtins[j])
+/* clang-format on */
+
+/* clang-format off */
+#define for_each_wanted_builtin(j, tgt_j)            \
+	for_each_builtin(j, tgt_j)                   \
+		if (tr2_dst_trace_want(tgt_j->pdst))
+/* clang-format on */
+
+/*
+ * Force (rather than lazily) initialize any of the requested
+ * builtin TRACE2 targets at startup (and before we've seen an
+ * actual TRACE2 event call) so we can see if we need to setup
+ * private data structures and thread-local storage.
+ *
+ * Return the number of builtin targets enabled.
+ */
+static int tr2_tgt_want_builtins(void)
+{
+	struct tr2_tgt *tgt_j;
+	int j;
+	int sum = 0;
+
+	for_each_builtin (j, tgt_j)
+		if (tgt_j->pfn_init())
+			sum++;
+
+	return sum;
+}
+
+/*
+ * Properly terminate each builtin target.  Give each target
+ * a chance to write a summary event and/or flush if necessary
+ * and then close the fd.
+ */
+static void tr2_tgt_disable_builtins(void)
+{
+	struct tr2_tgt *tgt_j;
+	int j;
+
+	for_each_builtin (j, tgt_j)
+		tgt_j->pfn_term();
+}
+
+/*
+ * The signature of this function must match the pfn_timer
+ * method in the targets.  (Think of this is an apply operation
+ * across the set of active targets.)
+ */
+static void tr2_tgt_emit_a_timer(const struct tr2_timer_metadata *meta,
+				 const struct tr2_timer *timer,
+				 int is_final_data)
+{
+	struct tr2_tgt *tgt_j;
+	int j;
+
+	for_each_wanted_builtin (j, tgt_j)
+		if (tgt_j->pfn_timer)
+			tgt_j->pfn_timer(meta, timer, is_final_data);
+}
+
+/*
+ * The signature of this function must match the pfn_counter
+ * method in the targets.
+ */
+static void tr2_tgt_emit_a_counter(const struct tr2_counter_metadata *meta,
+				   const struct tr2_counter *counter,
+				   int is_final_data)
+{
+	struct tr2_tgt *tgt_j;
+	int j;
+
+	for_each_wanted_builtin (j, tgt_j)
+		if (tgt_j->pfn_counter)
+			tgt_j->pfn_counter(meta, counter, is_final_data);
+}
+
+static int tr2main_exit_code;
+
+/*
+ * Our atexit routine should run after everything has finished.
+ *
+ * Note that events generated here might not actually appear if
+ * we are writing to fd 1 or 2 and our atexit routine runs after
+ * the pager's atexit routine (since it closes them to shutdown
+ * the pipes).
+ */
+static void tr2main_atexit_handler(void)
+{
+	struct tr2_tgt *tgt_j;
+	int j;
+	uint64_t us_now;
+	uint64_t us_elapsed_absolute;
+
+	us_now = getnanotime() / 1000;
+	us_elapsed_absolute = tr2tls_absolute_elapsed(us_now);
+
+	/*
+	 * Clear any unbalanced regions so that our atexit message
+	 * does not appear nested.  This improves the appearance of
+	 * the trace output if someone calls die(), for example.
+	 */
+	tr2tls_pop_unwind_self();
+
+	/*
+	 * Some timers want per-thread details.  If the main thread
+	 * used one of those timers, emit the details now (before
+	 * we emit the aggregate timer values).
+	 *
+	 * Likewise for counters.
+	 */
+	tr2_emit_per_thread_timers(tr2_tgt_emit_a_timer);
+	tr2_emit_per_thread_counters(tr2_tgt_emit_a_counter);
+
+	/*
+	 * Add stopwatch timer and counter data for the main thread to
+	 * the final totals.  And then emit the final values.
+	 *
+	 * Technically, we shouldn't need to hold the lock to update
+	 * and output the final_timer_block and final_counter_block
+	 * (since all other threads should be dead by now), but it
+	 * doesn't hurt anything.
+	 */
+	tr2tls_lock();
+	tr2_update_final_timers();
+	tr2_update_final_counters();
+	tr2_emit_final_timers(tr2_tgt_emit_a_timer);
+	tr2_emit_final_counters(tr2_tgt_emit_a_counter);
+	tr2tls_unlock();
+
+	for_each_wanted_builtin (j, tgt_j)
+		if (tgt_j->pfn_atexit)
+			tgt_j->pfn_atexit(us_elapsed_absolute,
+					  tr2main_exit_code);
+
+	tr2_tgt_disable_builtins();
+
+	tr2tls_release();
+	tr2_sid_release();
+	tr2_cmd_name_release();
+	tr2_cfg_free_patterns();
+	tr2_cfg_free_env_vars();
+	tr2_sysenv_release();
+
+	trace2_enabled = 0;
+}
+
+static void tr2main_signal_handler(int signo)
+{
+	struct tr2_tgt *tgt_j;
+	int j;
+	uint64_t us_now;
+	uint64_t us_elapsed_absolute;
+
+	us_now = getnanotime() / 1000;
+	us_elapsed_absolute = tr2tls_absolute_elapsed(us_now);
+
+	for_each_wanted_builtin (j, tgt_j)
+		if (tgt_j->pfn_signal)
+			tgt_j->pfn_signal(us_elapsed_absolute, signo);
+
+	sigchain_pop(signo);
+	raise(signo);
+}
+
+void trace2_initialize_clock(void)
+{
+	tr2tls_start_process_clock();
+}
+
+void trace2_initialize_fl(const char *file, int line)
+{
+	struct tr2_tgt *tgt_j;
+	int j;
+
+	if (trace2_enabled)
+		return;
+
+	tr2_sysenv_load();
+
+	if (!tr2_tgt_want_builtins())
+		return;
+	trace2_enabled = 1;
+
+	tr2_sid_get();
+
+	atexit(tr2main_atexit_handler);
+	sigchain_push(SIGPIPE, tr2main_signal_handler);
+	tr2tls_init();
+
+	/*
+	 * Emit 'version' message on each active builtin target.
+	 */
+	for_each_wanted_builtin (j, tgt_j)
+		if (tgt_j->pfn_version_fl)
+			tgt_j->pfn_version_fl(file, line);
+}
+
+int trace2_is_enabled(void)
+{
+	return trace2_enabled;
+}
+
+void trace2_cmd_start_fl(const char *file, int line, const char **argv)
+{
+	struct tr2_tgt *tgt_j;
+	int j;
+	uint64_t us_now;
+	uint64_t us_elapsed_absolute;
+
+	if (!trace2_enabled)
+		return;
+
+	us_now = getnanotime() / 1000;
+	us_elapsed_absolute = tr2tls_absolute_elapsed(us_now);
+
+	for_each_wanted_builtin (j, tgt_j)
+		if (tgt_j->pfn_start_fl)
+			tgt_j->pfn_start_fl(file, line, us_elapsed_absolute,
+					    argv);
+}
+
+void trace2_cmd_exit_fl(const char *file, int line, int code)
+{
+	struct tr2_tgt *tgt_j;
+	int j;
+	uint64_t us_now;
+	uint64_t us_elapsed_absolute;
+
+	if (!trace2_enabled)
+		return;
+
+	trace_git_fsync_stats();
+	trace2_collect_process_info(TRACE2_PROCESS_INFO_EXIT);
+
+	tr2main_exit_code = code;
+
+	us_now = getnanotime() / 1000;
+	us_elapsed_absolute = tr2tls_absolute_elapsed(us_now);
+
+	for_each_wanted_builtin (j, tgt_j)
+		if (tgt_j->pfn_exit_fl)
+			tgt_j->pfn_exit_fl(file, line, us_elapsed_absolute,
+					   code);
+}
+
+void trace2_cmd_error_va_fl(const char *file, int line, const char *fmt,
+			    va_list ap)
+{
+	struct tr2_tgt *tgt_j;
+	int j;
+
+	if (!trace2_enabled)
+		return;
+
+	/*
+	 * We expect each target function to treat 'ap' as constant
+	 * and use va_copy (because an 'ap' can only be walked once).
+	 */
+	for_each_wanted_builtin (j, tgt_j)
+		if (tgt_j->pfn_error_va_fl)
+			tgt_j->pfn_error_va_fl(file, line, fmt, ap);
+}
+
+void trace2_cmd_path_fl(const char *file, int line, const char *pathname)
+{
+	struct tr2_tgt *tgt_j;
+	int j;
+
+	if (!trace2_enabled)
+		return;
+
+	for_each_wanted_builtin (j, tgt_j)
+		if (tgt_j->pfn_command_path_fl)
+			tgt_j->pfn_command_path_fl(file, line, pathname);
+}
+
+void trace2_cmd_ancestry_fl(const char *file, int line, const char **parent_names)
+{
+	struct tr2_tgt *tgt_j;
+	int j;
+
+	if (!trace2_enabled)
+		return;
+
+	for_each_wanted_builtin (j, tgt_j)
+		if (tgt_j->pfn_command_ancestry_fl)
+			tgt_j->pfn_command_ancestry_fl(file, line, parent_names);
+}
+
+void trace2_cmd_name_fl(const char *file, int line, const char *name)
+{
+	struct tr2_tgt *tgt_j;
+	const char *hierarchy;
+	int j;
+
+	if (!trace2_enabled)
+		return;
+
+	tr2_cmd_name_append_hierarchy(name);
+	hierarchy = tr2_cmd_name_get_hierarchy();
+
+	for_each_wanted_builtin (j, tgt_j)
+		if (tgt_j->pfn_command_name_fl)
+			tgt_j->pfn_command_name_fl(file, line, name, hierarchy);
+}
+
+void trace2_cmd_mode_fl(const char *file, int line, const char *mode)
+{
+	struct tr2_tgt *tgt_j;
+	int j;
+
+	if (!trace2_enabled)
+		return;
+
+	for_each_wanted_builtin (j, tgt_j)
+		if (tgt_j->pfn_command_mode_fl)
+			tgt_j->pfn_command_mode_fl(file, line, mode);
+}
+
+void trace2_cmd_alias_fl(const char *file, int line, const char *alias,
+			 const char **argv)
+{
+	struct tr2_tgt *tgt_j;
+	int j;
+
+	if (!trace2_enabled)
+		return;
+
+	for_each_wanted_builtin (j, tgt_j)
+		if (tgt_j->pfn_alias_fl)
+			tgt_j->pfn_alias_fl(file, line, alias, argv);
+}
+
+void trace2_cmd_list_config_fl(const char *file, int line)
+{
+	if (!trace2_enabled)
+		return;
+
+	tr2_cfg_list_config_fl(file, line);
+}
+
+void trace2_cmd_list_env_vars_fl(const char *file, int line)
+{
+	if (!trace2_enabled)
+		return;
+
+	tr2_list_env_vars_fl(file, line);
+}
+
+void trace2_cmd_set_config_fl(const char *file, int line, const char *key,
+			      const char *value)
+{
+	if (!trace2_enabled)
+		return;
+
+	tr2_cfg_set_fl(file, line, key, value);
+}
+
+void trace2_child_start_fl(const char *file, int line,
+			   struct child_process *cmd)
+{
+	struct tr2_tgt *tgt_j;
+	int j;
+	uint64_t us_now;
+	uint64_t us_elapsed_absolute;
+
+	if (!trace2_enabled)
+		return;
+
+	us_now = getnanotime() / 1000;
+	us_elapsed_absolute = tr2tls_absolute_elapsed(us_now);
+
+	cmd->trace2_child_id = tr2tls_locked_increment(&tr2_next_child_id);
+	cmd->trace2_child_us_start = us_now;
+
+	for_each_wanted_builtin (j, tgt_j)
+		if (tgt_j->pfn_child_start_fl)
+			tgt_j->pfn_child_start_fl(file, line,
+						  us_elapsed_absolute, cmd);
+}
+
+void trace2_child_exit_fl(const char *file, int line, struct child_process *cmd,
+			  int child_exit_code)
+{
+	struct tr2_tgt *tgt_j;
+	int j;
+	uint64_t us_now;
+	uint64_t us_elapsed_absolute;
+	uint64_t us_elapsed_child;
+
+	if (!trace2_enabled)
+		return;
+
+	us_now = getnanotime() / 1000;
+	us_elapsed_absolute = tr2tls_absolute_elapsed(us_now);
+
+	if (cmd->trace2_child_us_start)
+		us_elapsed_child = us_now - cmd->trace2_child_us_start;
+	else
+		us_elapsed_child = 0;
+
+	for_each_wanted_builtin (j, tgt_j)
+		if (tgt_j->pfn_child_exit_fl)
+			tgt_j->pfn_child_exit_fl(file, line,
+						 us_elapsed_absolute,
+						 cmd->trace2_child_id, cmd->pid,
+						 child_exit_code,
+						 us_elapsed_child);
+}
+
+void trace2_child_ready_fl(const char *file, int line,
+			   struct child_process *cmd,
+			   const char *ready)
+{
+	struct tr2_tgt *tgt_j;
+	int j;
+	uint64_t us_now;
+	uint64_t us_elapsed_absolute;
+	uint64_t us_elapsed_child;
+
+	if (!trace2_enabled)
+		return;
+
+	us_now = getnanotime() / 1000;
+	us_elapsed_absolute = tr2tls_absolute_elapsed(us_now);
+
+	if (cmd->trace2_child_us_start)
+		us_elapsed_child = us_now - cmd->trace2_child_us_start;
+	else
+		us_elapsed_child = 0;
+
+	for_each_wanted_builtin (j, tgt_j)
+		if (tgt_j->pfn_child_ready_fl)
+			tgt_j->pfn_child_ready_fl(file, line,
+						  us_elapsed_absolute,
+						  cmd->trace2_child_id,
+						  cmd->pid,
+						  ready,
+						  us_elapsed_child);
+}
+
+int trace2_exec_fl(const char *file, int line, const char *exe,
+		   const char **argv)
+{
+	struct tr2_tgt *tgt_j;
+	int j;
+	int exec_id;
+	uint64_t us_now;
+	uint64_t us_elapsed_absolute;
+
+	if (!trace2_enabled)
+		return -1;
+
+	us_now = getnanotime() / 1000;
+	us_elapsed_absolute = tr2tls_absolute_elapsed(us_now);
+
+	exec_id = tr2tls_locked_increment(&tr2_next_exec_id);
+
+	for_each_wanted_builtin (j, tgt_j)
+		if (tgt_j->pfn_exec_fl)
+			tgt_j->pfn_exec_fl(file, line, us_elapsed_absolute,
+					   exec_id, exe, argv);
+
+	return exec_id;
+}
+
+void trace2_exec_result_fl(const char *file, int line, int exec_id, int code)
+{
+	struct tr2_tgt *tgt_j;
+	int j;
+	uint64_t us_now;
+	uint64_t us_elapsed_absolute;
+
+	if (!trace2_enabled)
+		return;
+
+	us_now = getnanotime() / 1000;
+	us_elapsed_absolute = tr2tls_absolute_elapsed(us_now);
+
+	for_each_wanted_builtin (j, tgt_j)
+		if (tgt_j->pfn_exec_result_fl)
+			tgt_j->pfn_exec_result_fl(
+				file, line, us_elapsed_absolute, exec_id, code);
+}
+
+void trace2_thread_start_fl(const char *file, int line, const char *thread_base_name)
+{
+	struct tr2_tgt *tgt_j;
+	int j;
+	uint64_t us_now;
+	uint64_t us_elapsed_absolute;
+
+	if (!trace2_enabled)
+		return;
+
+	if (tr2tls_is_main_thread()) {
+		/*
+		 * We should only be called from the new thread's thread-proc,
+		 * so this is technically a bug.  But in those cases where the
+		 * main thread also runs the thread-proc function (or when we
+		 * are built with threading disabled), we need to allow it.
+		 *
+		 * Convert this call to a region-enter so the nesting looks
+		 * correct.
+		 */
+		trace2_region_enter_printf_fl(file, line, NULL, NULL, NULL,
+					      "thread-proc on main: %s",
+					      thread_base_name);
+		return;
+	}
+
+	us_now = getnanotime() / 1000;
+	us_elapsed_absolute = tr2tls_absolute_elapsed(us_now);
+
+	tr2tls_create_self(thread_base_name, us_now);
+
+	for_each_wanted_builtin (j, tgt_j)
+		if (tgt_j->pfn_thread_start_fl)
+			tgt_j->pfn_thread_start_fl(file, line,
+						   us_elapsed_absolute);
+}
+
+void trace2_thread_exit_fl(const char *file, int line)
+{
+	struct tr2_tgt *tgt_j;
+	int j;
+	uint64_t us_now;
+	uint64_t us_elapsed_absolute;
+	uint64_t us_elapsed_thread;
+
+	if (!trace2_enabled)
+		return;
+
+	if (tr2tls_is_main_thread()) {
+		/*
+		 * We should only be called from the exiting thread's
+		 * thread-proc, so this is technically a bug.  But in
+		 * those cases where the main thread also runs the
+		 * thread-proc function (or when we are built with
+		 * threading disabled), we need to allow it.
+		 *
+		 * Convert this call to a region-leave so the nesting
+		 * looks correct.
+		 */
+		trace2_region_leave_printf_fl(file, line, NULL, NULL, NULL,
+					      "thread-proc on main");
+		return;
+	}
+
+	us_now = getnanotime() / 1000;
+	us_elapsed_absolute = tr2tls_absolute_elapsed(us_now);
+
+	/*
+	 * Clear any unbalanced regions and then get the relative time
+	 * for the outer-most region (which we pushed when the thread
+	 * started).  This gives us the run time of the thread.
+	 */
+	tr2tls_pop_unwind_self();
+	us_elapsed_thread = tr2tls_region_elasped_self(us_now);
+
+	/*
+	 * Some timers want per-thread details.  If this thread used
+	 * one of those timers, emit the details now.
+	 *
+	 * Likewise for counters.
+	 */
+	tr2_emit_per_thread_timers(tr2_tgt_emit_a_timer);
+	tr2_emit_per_thread_counters(tr2_tgt_emit_a_counter);
+
+	/*
+	 * Add stopwatch timer and counter data from the current
+	 * (non-main) thread to the final totals.  (We'll accumulate
+	 * data for the main thread later during "atexit".)
+	 */
+	tr2tls_lock();
+	tr2_update_final_timers();
+	tr2_update_final_counters();
+	tr2tls_unlock();
+
+	for_each_wanted_builtin (j, tgt_j)
+		if (tgt_j->pfn_thread_exit_fl)
+			tgt_j->pfn_thread_exit_fl(file, line,
+						  us_elapsed_absolute,
+						  us_elapsed_thread);
+
+	tr2tls_unset_self();
+}
+
+void trace2_def_param_fl(const char *file, int line, const char *param,
+			 const char *value)
+{
+	struct tr2_tgt *tgt_j;
+	int j;
+
+	if (!trace2_enabled)
+		return;
+
+	for_each_wanted_builtin (j, tgt_j)
+		if (tgt_j->pfn_param_fl)
+			tgt_j->pfn_param_fl(file, line, param, value);
+}
+
+void trace2_def_repo_fl(const char *file, int line, struct repository *repo)
+{
+	struct tr2_tgt *tgt_j;
+	int j;
+
+	if (!trace2_enabled)
+		return;
+
+	if (repo->trace2_repo_id)
+		return;
+
+	repo->trace2_repo_id = tr2tls_locked_increment(&tr2_next_repo_id);
+
+	for_each_wanted_builtin (j, tgt_j)
+		if (tgt_j->pfn_repo_fl)
+			tgt_j->pfn_repo_fl(file, line, repo);
+}
+
+void trace2_region_enter_printf_va_fl(const char *file, int line,
+				      const char *category, const char *label,
+				      const struct repository *repo,
+				      const char *fmt, va_list ap)
+{
+	struct tr2_tgt *tgt_j;
+	int j;
+	uint64_t us_now;
+	uint64_t us_elapsed_absolute;
+
+	if (!trace2_enabled)
+		return;
+
+	us_now = getnanotime() / 1000;
+	us_elapsed_absolute = tr2tls_absolute_elapsed(us_now);
+
+	/*
+	 * Print the region-enter message at the current nesting
+	 * (indentation) level and then push a new level.
+	 *
+	 * We expect each target function to treat 'ap' as constant
+	 * and use va_copy.
+	 */
+	for_each_wanted_builtin (j, tgt_j)
+		if (tgt_j->pfn_region_enter_printf_va_fl)
+			tgt_j->pfn_region_enter_printf_va_fl(
+				file, line, us_elapsed_absolute, category,
+				label, repo, fmt, ap);
+
+	tr2tls_push_self(us_now);
+}
+
+void trace2_region_enter_fl(const char *file, int line, const char *category,
+			    const char *label, const struct repository *repo, ...)
+{
+	va_list ap;
+	va_start(ap, repo);
+	trace2_region_enter_printf_va_fl(file, line, category, label, repo,
+					 NULL, ap);
+	va_end(ap);
+
+}
+
+void trace2_region_enter_printf_fl(const char *file, int line,
+				   const char *category, const char *label,
+				   const struct repository *repo,
+				   const char *fmt, ...)
+{
+	va_list ap;
+
+	va_start(ap, fmt);
+	trace2_region_enter_printf_va_fl(file, line, category, label, repo, fmt,
+					 ap);
+	va_end(ap);
+}
+
+void trace2_region_leave_printf_va_fl(const char *file, int line,
+				      const char *category, const char *label,
+				      const struct repository *repo,
+				      const char *fmt, va_list ap)
+{
+	struct tr2_tgt *tgt_j;
+	int j;
+	uint64_t us_now;
+	uint64_t us_elapsed_absolute;
+	uint64_t us_elapsed_region;
+
+	if (!trace2_enabled)
+		return;
+
+	us_now = getnanotime() / 1000;
+	us_elapsed_absolute = tr2tls_absolute_elapsed(us_now);
+
+	/*
+	 * Get the elapsed time in the current region before we
+	 * pop it off the stack.  Pop the stack.  And then print
+	 * the perf message at the new (shallower) level so that
+	 * it lines up with the corresponding push/enter.
+	 */
+	us_elapsed_region = tr2tls_region_elasped_self(us_now);
+
+	tr2tls_pop_self();
+
+	/*
+	 * We expect each target function to treat 'ap' as constant
+	 * and use va_copy.
+	 */
+	for_each_wanted_builtin (j, tgt_j)
+		if (tgt_j->pfn_region_leave_printf_va_fl)
+			tgt_j->pfn_region_leave_printf_va_fl(
+				file, line, us_elapsed_absolute,
+				us_elapsed_region, category, label, repo, fmt,
+				ap);
+}
+
+void trace2_region_leave_fl(const char *file, int line, const char *category,
+			    const char *label, const struct repository *repo, ...)
+{
+	va_list ap;
+	va_start(ap, repo);
+	trace2_region_leave_printf_va_fl(file, line, category, label, repo,
+					 NULL, ap);
+	va_end(ap);
+}
+
+void trace2_region_leave_printf_fl(const char *file, int line,
+				   const char *category, const char *label,
+				   const struct repository *repo,
+				   const char *fmt, ...)
+{
+	va_list ap;
+
+	va_start(ap, fmt);
+	trace2_region_leave_printf_va_fl(file, line, category, label, repo, fmt,
+					 ap);
+	va_end(ap);
+}
+
+void trace2_data_string_fl(const char *file, int line, const char *category,
+			   const struct repository *repo, const char *key,
+			   const char *value)
+{
+	struct tr2_tgt *tgt_j;
+	int j;
+	uint64_t us_now;
+	uint64_t us_elapsed_absolute;
+	uint64_t us_elapsed_region;
+
+	if (!trace2_enabled)
+		return;
+
+	us_now = getnanotime() / 1000;
+	us_elapsed_absolute = tr2tls_absolute_elapsed(us_now);
+	us_elapsed_region = tr2tls_region_elasped_self(us_now);
+
+	for_each_wanted_builtin (j, tgt_j)
+		if (tgt_j->pfn_data_fl)
+			tgt_j->pfn_data_fl(file, line, us_elapsed_absolute,
+					   us_elapsed_region, category, repo,
+					   key, value);
+}
+
+void trace2_data_intmax_fl(const char *file, int line, const char *category,
+			   const struct repository *repo, const char *key,
+			   intmax_t value)
+{
+	struct strbuf buf_string = STRBUF_INIT;
+
+	if (!trace2_enabled)
+		return;
+
+	strbuf_addf(&buf_string, "%" PRIdMAX, value);
+	trace2_data_string_fl(file, line, category, repo, key, buf_string.buf);
+	strbuf_release(&buf_string);
+}
+
+void trace2_data_json_fl(const char *file, int line, const char *category,
+			 const struct repository *repo, const char *key,
+			 const struct json_writer *value)
+{
+	struct tr2_tgt *tgt_j;
+	int j;
+	uint64_t us_now;
+	uint64_t us_elapsed_absolute;
+	uint64_t us_elapsed_region;
+
+	if (!trace2_enabled)
+		return;
+
+	us_now = getnanotime() / 1000;
+	us_elapsed_absolute = tr2tls_absolute_elapsed(us_now);
+	us_elapsed_region = tr2tls_region_elasped_self(us_now);
+
+	for_each_wanted_builtin (j, tgt_j)
+		if (tgt_j->pfn_data_json_fl)
+			tgt_j->pfn_data_json_fl(file, line, us_elapsed_absolute,
+						us_elapsed_region, category,
+						repo, key, value);
+}
+
+void trace2_printf_va_fl(const char *file, int line, const char *fmt,
+			 va_list ap)
+{
+	struct tr2_tgt *tgt_j;
+	int j;
+	uint64_t us_now;
+	uint64_t us_elapsed_absolute;
+
+	if (!trace2_enabled)
+		return;
+
+	us_now = getnanotime() / 1000;
+	us_elapsed_absolute = tr2tls_absolute_elapsed(us_now);
+
+	/*
+	 * We expect each target function to treat 'ap' as constant
+	 * and use va_copy.
+	 */
+	for_each_wanted_builtin (j, tgt_j)
+		if (tgt_j->pfn_printf_va_fl)
+			tgt_j->pfn_printf_va_fl(file, line, us_elapsed_absolute,
+						fmt, ap);
+}
+
+void trace2_printf_fl(const char *file, int line, const char *fmt, ...)
+{
+	va_list ap;
+
+	va_start(ap, fmt);
+	trace2_printf_va_fl(file, line, fmt, ap);
+	va_end(ap);
+}
+
+void trace2_timer_start(enum trace2_timer_id tid)
+{
+	if (!trace2_enabled)
+		return;
+
+	if (tid < 0 || tid >= TRACE2_NUMBER_OF_TIMERS)
+		BUG("trace2_timer_start: invalid timer id: %d", tid);
+
+	tr2_start_timer(tid);
+}
+
+void trace2_timer_stop(enum trace2_timer_id tid)
+{
+	if (!trace2_enabled)
+		return;
+
+	if (tid < 0 || tid >= TRACE2_NUMBER_OF_TIMERS)
+		BUG("trace2_timer_stop: invalid timer id: %d", tid);
+
+	tr2_stop_timer(tid);
+}
+
+void trace2_counter_add(enum trace2_counter_id cid, uint64_t value)
+{
+	if (!trace2_enabled)
+		return;
+
+	if (cid < 0 || cid >= TRACE2_NUMBER_OF_COUNTERS)
+		BUG("trace2_counter_add: invalid counter id: %d", cid);
+
+	tr2_counter_increment(cid, value);
+}
+
+const char *trace2_session_id(void)
+{
+	return tr2_sid_get();
+}
diff --git a/trace2.h b/trace2.h
new file mode 100644
index 0000000..4ced30c
--- /dev/null
+++ b/trace2.h
@@ -0,0 +1,584 @@
+#ifndef TRACE2_H
+#define TRACE2_H
+
+/**
+ * The Trace2 API can be used to print debug, performance, and telemetry
+ * information to stderr or a file.  The Trace2 feature is inactive unless
+ * explicitly enabled by enabling one or more Trace2 Targets.
+ *
+ * The Trace2 API is intended to replace the existing (Trace1)
+ * printf-style tracing provided by the existing `GIT_TRACE` and
+ * `GIT_TRACE_PERFORMANCE` facilities.  During initial implementation,
+ * Trace2 and Trace1 may operate in parallel.
+ *
+ * The Trace2 API defines a set of high-level messages with known fields,
+ * such as (`start`: `argv`) and (`exit`: {`exit-code`, `elapsed-time`}).
+ *
+ * Trace2 instrumentation throughout the Git code base sends Trace2
+ * messages to the enabled Trace2 Targets.  Targets transform these
+ * messages content into purpose-specific formats and write events to
+ * their data streams.  In this manner, the Trace2 API can drive
+ * many different types of analysis.
+ *
+ * Targets are defined using a VTable allowing easy extension to other
+ * formats in the future.  This might be used to define a binary format,
+ * for example.
+ *
+ * Trace2 is controlled using `trace2.*` config values in the system and
+ * global config files and `GIT_TRACE2*` environment variables.  Trace2 does
+ * not read from repo local or worktree config files or respect `-c`
+ * command line config settings.
+ *
+ * For more info about: trace2 targets, conventions for public functions and
+ * macros, trace2 target formats and examples on trace2 API usage refer to
+ * Documentation/technical/api-trace2.txt
+ *
+ */
+
+struct child_process;
+struct repository;
+struct json_writer;
+
+/*
+ * The public TRACE2 routines are grouped into the following groups:
+ *
+ * [] trace2_initialize -- initialization.
+ * [] trace2_cmd_*      -- emit command/control messages.
+ * [] trace2_child*     -- emit child start/stop messages.
+ * [] trace2_exec*      -- emit exec start/stop messages.
+ * [] trace2_thread*    -- emit thread start/stop messages.
+ * [] trace2_def*       -- emit definition/parameter mesasges.
+ * [] trace2_region*    -- emit region nesting messages.
+ * [] trace2_data*      -- emit region/thread/repo data messages.
+ * [] trace2_printf*    -- legacy trace[1] messages.
+ * [] trace2_timer*     -- stopwatch timers (messages are deferred).
+ * [] trace2_counter*   -- global counters (messages are deferred).
+ */
+
+/*
+ * Initialize the TRACE2 clock and do nothing else, in particular
+ * no mallocs, no system inspection, and no environment inspection.
+ *
+ * This should be called at the very top of main() to capture the
+ * process start time.  This is intended to reduce chicken-n-egg
+ * bootstrap pressure.
+ *
+ * It is safe to call this more than once.  This allows capturing
+ * absolute startup costs on Windows which uses a little trickery
+ * to do setup work before common-main.c:main() is called.
+ *
+ * The main trace2_initialize_fl() may be called a little later
+ * after more infrastructure is established.
+ */
+void trace2_initialize_clock(void);
+
+/*
+ * Initialize TRACE2 tracing facility if any of the builtin TRACE2
+ * targets are enabled in the system config or the environment.
+ * This emits a 'version' message containing the version of git
+ * and the Trace2 protocol.
+ *
+ * This function should be called from `main()` as early as possible in
+ * the life of the process after essential process initialization.
+ *
+ * Cleanup/Termination is handled automatically by a registered
+ * atexit() routine.
+ */
+void trace2_initialize_fl(const char *file, int line);
+
+#define trace2_initialize() trace2_initialize_fl(__FILE__, __LINE__)
+
+/*
+ * Return 1 if trace2 is enabled (at least one target is active).
+ */
+int trace2_is_enabled(void);
+
+/*
+ * Emit a 'start' event with the original (unmodified) argv.
+ */
+void trace2_cmd_start_fl(const char *file, int line, const char **argv);
+
+#define trace2_cmd_start(argv) trace2_cmd_start_fl(__FILE__, __LINE__, (argv))
+
+/*
+ * Emit an 'exit' event.
+ */
+void trace2_cmd_exit_fl(const char *file, int line, int code);
+
+#define trace2_cmd_exit(code) (trace2_cmd_exit_fl(__FILE__, __LINE__, (code)))
+
+/*
+ * Emit an 'error' event.
+ *
+ * Write an error message to the TRACE2 targets.
+ */
+void trace2_cmd_error_va_fl(const char *file, int line, const char *fmt,
+			    va_list ap);
+
+#define trace2_cmd_error_va(fmt, ap) \
+	trace2_cmd_error_va_fl(__FILE__, __LINE__, (fmt), (ap))
+
+/*
+ * Emit a 'pathname' event with the canonical pathname of the current process
+ * This gives post-processors a simple field to identify the command without
+ * having to parse the argv.  For example, to distinguish invocations from
+ * installed versus debug executables.
+ */
+void trace2_cmd_path_fl(const char *file, int line, const char *pathname);
+
+#define trace2_cmd_path(p) trace2_cmd_path_fl(__FILE__, __LINE__, (p))
+
+/*
+ * Emit an 'ancestry' event with the process name of the current process's
+ * parent process.
+ * This gives post-processors a way to determine what invoked the command and
+ * learn more about usage patterns.
+ */
+void trace2_cmd_ancestry_fl(const char *file, int line, const char **parent_names);
+
+#define trace2_cmd_ancestry(v) trace2_cmd_ancestry_fl(__FILE__, __LINE__, (v))
+
+/*
+ * Emit a 'cmd_name' event with the canonical name of the command.
+ * This gives post-processors a simple field to identify the command
+ * without having to parse the argv.
+ */
+void trace2_cmd_name_fl(const char *file, int line, const char *name);
+
+#define trace2_cmd_name(v) trace2_cmd_name_fl(__FILE__, __LINE__, (v))
+
+/*
+ * Emit a 'cmd_mode' event to further describe the command being run.
+ * For example, "checkout" can checkout a single file or can checkout a
+ * different branch.  This gives post-processors a simple field to compare
+ * equivalent commands without having to parse the argv.
+ */
+void trace2_cmd_mode_fl(const char *file, int line, const char *mode);
+
+#define trace2_cmd_mode(sv) trace2_cmd_mode_fl(__FILE__, __LINE__, (sv))
+
+/*
+ * Emits an "alias" message containing the alias used and the argument
+ * expansion.
+ */
+void trace2_cmd_alias_fl(const char *file, int line, const char *alias,
+			 const char **argv);
+
+#define trace2_cmd_alias(alias, argv) \
+	trace2_cmd_alias_fl(__FILE__, __LINE__, (alias), (argv))
+
+/*
+ * Emit one or more 'def_param' events for "important" configuration
+ * settings.
+ *
+ * Use the TR2_SYSENV_CFG_PARAM setting to register a comma-separated
+ * list of patterns configured important.  For example:
+ *     git config --system trace2.configParams 'core.*,remote.*.url'
+ * or:
+ *     GIT_TRACE2_CONFIG_PARAMS=core.*,remote.*.url"
+ *
+ * Note: this routine does a read-only iteration on the config data
+ * (using read_early_config()), so it must not be called until enough
+ * of the process environment has been established.  This includes the
+ * location of the git and worktree directories, expansion of any "-c"
+ * and "-C" command line options, and etc.
+ */
+void trace2_cmd_list_config_fl(const char *file, int line);
+
+#define trace2_cmd_list_config() trace2_cmd_list_config_fl(__FILE__, __LINE__)
+
+/*
+ * Emit one or more 'def_param' events for "important" environment variables.
+ *
+ * Use the TR2_SYSENV_ENV_VARS setting to register a comma-separated list of
+ * environment variables considered important.  For example:
+ *     git config --system trace2.envVars 'GIT_HTTP_USER_AGENT,GIT_CONFIG'
+ * or:
+ *     GIT_TRACE2_ENV_VARS="GIT_HTTP_USER_AGENT,GIT_CONFIG"
+ */
+void trace2_cmd_list_env_vars_fl(const char *file, int line);
+
+#define trace2_cmd_list_env_vars() trace2_cmd_list_env_vars_fl(__FILE__, __LINE__)
+
+/*
+ * Emit a "def_param" event for the given config key/value pair IF
+ * we consider the key to be "important".
+ *
+ * Use this for new/updated config settings created/updated after
+ * trace2_cmd_list_config() is called.
+ */
+void trace2_cmd_set_config_fl(const char *file, int line, const char *key,
+			      const char *value);
+
+#define trace2_cmd_set_config(k, v) \
+	trace2_cmd_set_config_fl(__FILE__, __LINE__, (k), (v))
+
+/**
+ * Emits a "child_start" message containing the "child-id",
+ * "child-argv", and "child-classification".
+ *
+ * Before calling optionally set "cmd->trace2_child_class" to a string
+ * describing the type of the child process.  For example, "editor" or
+ * "pager".
+ *
+ * This function assigns a unique "child-id" to `cmd->trace2_child_id`.
+ * This field is used later during the "child_exit" message to associate
+ * it with the "child_start" message.
+ *
+ * This function should be called before spawning the child process.
+ */
+void trace2_child_start_fl(const char *file, int line,
+			   struct child_process *cmd);
+
+#define trace2_child_start(cmd) trace2_child_start_fl(__FILE__, __LINE__, (cmd))
+
+/**
+ * Emits a "child_exit" message containing the "child-id",
+ * the child's elapsed time and exit-code.
+ *
+ * The reported elapsed time includes the process creation overhead and
+ * time spend waiting for it to exit, so it may be slightly longer than
+ * the time reported by the child itself.
+ *
+ * This function should be called after reaping the child process.
+ */
+void trace2_child_exit_fl(const char *file, int line, struct child_process *cmd,
+			  int child_exit_code);
+
+#define trace2_child_exit(cmd, code) \
+	trace2_child_exit_fl(__FILE__, __LINE__, (cmd), (code))
+
+/**
+ * Emits a "child_ready" message containing the "child-id" and a flag
+ * indicating whether the child was considered "ready" when we
+ * released it.
+ *
+ * This function should be called after starting a daemon process in
+ * the background (and after giving it sufficient time to boot
+ * up) to indicate that we no longer control or own it.
+ *
+ * The "ready" argument should contain one of { "ready", "timeout",
+ * "error" } to indicate the state of the running daemon when we
+ * released it.
+ *
+ * If the daemon process fails to start or it exits or is terminated
+ * while we are still waiting for it, the caller should emit a
+ * regular "child_exit" to report the normal process exit information.
+ *
+ */
+void trace2_child_ready_fl(const char *file, int line,
+			   struct child_process *cmd,
+			   const char *ready);
+
+#define trace2_child_ready(cmd, ready) \
+	trace2_child_ready_fl(__FILE__, __LINE__, (cmd), (ready))
+
+/**
+ * Emit an 'exec' event prior to calling one of exec(), execv(),
+ * execvp(), and etc.  On Unix-derived systems, this will be the
+ * last event emitted for the current process, unless the exec
+ * fails.  On Windows, exec() behaves like 'child_start' and a
+ * waitpid(), so additional events may be emitted.
+ *
+ * Returns a unique "exec-id".  This value is used later
+ * if the exec() fails and a "exec-result" message is necessary.
+ */
+int trace2_exec_fl(const char *file, int line, const char *exe,
+		   const char **argv);
+
+#define trace2_exec(exe, argv) trace2_exec_fl(__FILE__, __LINE__, (exe), (argv))
+
+/**
+ * Emit an 'exec_result' when possible.  On Unix-derived systems,
+ * this should be called after exec() returns (which only happens
+ * when there is an error starting the new process).  On Windows,
+ * this should be called after the waitpid().
+ *
+ * The "exec_id" should be the value returned from trace2_exec().
+ */
+void trace2_exec_result_fl(const char *file, int line, int exec_id, int code);
+
+#define trace2_exec_result(id, code) \
+	trace2_exec_result_fl(__FILE__, __LINE__, (id), (code))
+
+/*
+ * Emit a 'thread_start' event.  This must be called from inside the
+ * thread-proc to allow the thread to create its own thread-local
+ * storage.
+ *
+ * The thread base name should be descriptive, like "preload_index" or
+ * taken from the thread-proc function.  A unique thread name will be
+ * created from the given base name and the thread id automatically.
+ */
+void trace2_thread_start_fl(const char *file, int line,
+			    const char *thread_base_name);
+
+#define trace2_thread_start(thread_base_name) \
+	trace2_thread_start_fl(__FILE__, __LINE__, (thread_base_name))
+
+/*
+ * Emit a 'thread_exit' event.  This must be called from inside the
+ * thread-proc so that the thread can access and clean up its
+ * thread-local storage.
+ */
+void trace2_thread_exit_fl(const char *file, int line);
+
+#define trace2_thread_exit() trace2_thread_exit_fl(__FILE__, __LINE__)
+
+/*
+ * Emits a "def_param" message containing a key/value pair.
+ *
+ * This message is intended to report some global aspect of the current
+ * command, such as a configuration setting or command line switch that
+ * significantly affects program performance or behavior, such as
+ * `core.abbrev`, `status.showUntrackedFiles`, or `--no-ahead-behind`.
+ */
+void trace2_def_param_fl(const char *file, int line, const char *param,
+			 const char *value);
+
+#define trace2_def_param(param, value) \
+	trace2_def_param_fl(__FILE__, __LINE__, (param), (value))
+
+/*
+ * Tell trace2 about a newly instantiated repo object and assign
+ * a trace2-repo-id to be used in subsequent activity events.
+ *
+ * Emits a 'worktree' event for this repo instance.
+ *
+ * Region and data messages may refer to this repo-id.
+ *
+ * The main/top-level repository will have repo-id value 1 (aka "r1").
+ *
+ * The repo-id field is in anticipation of future in-proc submodule
+ * repositories.
+ */
+void trace2_def_repo_fl(const char *file, int line, struct repository *repo);
+
+#define trace2_def_repo(repo) trace2_def_repo_fl(__FILE__, __LINE__, repo)
+
+/**
+ * Emit a 'region_enter' event for <category>.<label> with optional
+ * repo-id and printf message.
+ *
+ * This function pushes a new region nesting stack level on the current
+ * thread and starts a clock for the new stack frame.
+ *
+ * The `category` field is an arbitrary category name used to classify
+ * regions by feature area, such as "status" or "index".  At this time
+ * it is only just printed along with the rest of the message.  It may
+ * be used in the future to filter messages.
+ *
+ * The `label` field is an arbitrary label used to describe the activity
+ * being started, such as "read_recursive" or "do_read_index".
+ *
+ * The `repo` field, if set, will be used to get the "repo-id", so that
+ * recursive operations can be attributed to the correct repository.
+ */
+void trace2_region_enter_fl(const char *file, int line, const char *category,
+			    const char *label, const struct repository *repo, ...);
+
+#define trace2_region_enter(category, label, repo) \
+	trace2_region_enter_fl(__FILE__, __LINE__, (category), (label), (repo))
+
+void trace2_region_enter_printf_va_fl(const char *file, int line,
+				      const char *category, const char *label,
+				      const struct repository *repo,
+				      const char *fmt, va_list ap);
+
+#define trace2_region_enter_printf_va(category, label, repo, fmt, ap)    \
+	trace2_region_enter_printf_va_fl(__FILE__, __LINE__, (category), \
+					 (label), (repo), (fmt), (ap))
+
+void trace2_region_enter_printf_fl(const char *file, int line,
+				   const char *category, const char *label,
+				   const struct repository *repo,
+				   const char *fmt, ...);
+
+#define trace2_region_enter_printf(category, label, repo, ...)                 \
+	trace2_region_enter_printf_fl(__FILE__, __LINE__, (category), (label), \
+				      (repo), __VA_ARGS__)
+
+/**
+ * Emit a 'region_leave' event for <category>.<label> with optional
+ * repo-id and printf message.
+ *
+ * Leave current nesting level and report the elapsed time spent
+ * in this nesting level.
+ *
+ * The `category`, `label`, and `repo` fields are the same as
+ * trace2_region_enter_fl. The `category` and `label` do not
+ * need to match the corresponding "region_enter" message,
+ * but it makes the data stream easier to understand.
+ */
+void trace2_region_leave_fl(const char *file, int line, const char *category,
+			    const char *label, const struct repository *repo, ...);
+
+#define trace2_region_leave(category, label, repo) \
+	trace2_region_leave_fl(__FILE__, __LINE__, (category), (label), (repo))
+
+void trace2_region_leave_printf_va_fl(const char *file, int line,
+				      const char *category, const char *label,
+				      const struct repository *repo,
+				      const char *fmt, va_list ap);
+
+#define trace2_region_leave_printf_va(category, label, repo, fmt, ap)    \
+	trace2_region_leave_printf_va_fl(__FILE__, __LINE__, (category), \
+					 (label), (repo), (fmt), (ap))
+
+void trace2_region_leave_printf_fl(const char *file, int line,
+				   const char *category, const char *label,
+				   const struct repository *repo,
+				   const char *fmt, ...);
+
+#define trace2_region_leave_printf(category, label, repo, ...)                 \
+	trace2_region_leave_printf_fl(__FILE__, __LINE__, (category), (label), \
+				      (repo), __VA_ARGS__)
+
+/**
+ * Emit a key-value pair 'data' event of the form <category>.<key> = <value>.
+ * This event implicitly contains information about thread, nesting region,
+ * and optional repo-id.
+ * This could be used to print the number of files in a directory during
+ * a multi-threaded recursive tree walk.
+ *
+ * On event-based TRACE2 targets, this generates a 'data' event suitable
+ * for post-processing.  On printf-based TRACE2 targets, this is converted
+ * into a fixed-format printf message.
+ */
+void trace2_data_string_fl(const char *file, int line, const char *category,
+			   const struct repository *repo, const char *key,
+			   const char *value);
+
+#define trace2_data_string(category, repo, key, value)                       \
+	trace2_data_string_fl(__FILE__, __LINE__, (category), (repo), (key), \
+			      (value))
+
+void trace2_data_intmax_fl(const char *file, int line, const char *category,
+			   const struct repository *repo, const char *key,
+			   intmax_t value);
+
+#define trace2_data_intmax(category, repo, key, value)                       \
+	trace2_data_intmax_fl(__FILE__, __LINE__, (category), (repo), (key), \
+			      (value))
+
+void trace2_data_json_fl(const char *file, int line, const char *category,
+			 const struct repository *repo, const char *key,
+			 const struct json_writer *jw);
+
+#define trace2_data_json(category, repo, key, value)                       \
+	trace2_data_json_fl(__FILE__, __LINE__, (category), (repo), (key), \
+			    (value))
+
+/*
+ * Emit a 'printf' event.
+ *
+ * Write an arbitrary formatted message to the TRACE2 targets.  These
+ * text messages should be considered as human-readable strings without
+ * any formatting guidelines.  Post-processors may choose to ignore
+ * them.
+ */
+void trace2_printf_va_fl(const char *file, int line, const char *fmt,
+			 va_list ap);
+
+#define trace2_printf_va(fmt, ap) \
+	trace2_printf_va_fl(__FILE__, __LINE__, (fmt), (ap))
+
+void trace2_printf_fl(const char *file, int line, const char *fmt, ...);
+
+#define trace2_printf(...) trace2_printf_fl(__FILE__, __LINE__, __VA_ARGS__)
+
+/*
+ * Define the set of stopwatch timers.
+ *
+ * We can add more at any time, but they must be defined at compile
+ * time (to avoid the need to dynamically allocate and synchronize
+ * them between different threads).
+ *
+ * These must start at 0 and be contiguous (because we use them
+ * elsewhere as array indexes).
+ *
+ * Any values added to this enum must also be added to the
+ * `tr2_timer_metadata[]` in `trace2/tr2_tmr.c`.
+ */
+enum trace2_timer_id {
+	/*
+	 * Define two timers for testing.  See `t/helper/test-trace2.c`.
+	 * These can be used for ad hoc testing, but should not be used
+	 * for permanent analysis code.
+	 */
+	TRACE2_TIMER_ID_TEST1 = 0, /* emits summary event only */
+	TRACE2_TIMER_ID_TEST2,     /* emits summary and thread events */
+
+	/* Add additional timer definitions before here. */
+	TRACE2_NUMBER_OF_TIMERS
+};
+
+/*
+ * Start/Stop the indicated stopwatch timer in the current thread.
+ *
+ * The time spent by the current thread between the _start and _stop
+ * calls will be added to the thread's partial sum for this timer.
+ *
+ * Timer events are emitted at thread and program exit.
+ *
+ * Note: Since the stopwatch API routines do not generate individual
+ * events, they do not take (file, line) arguments.  Similarly, the
+ * category and timer name values are defined at compile-time in the
+ * timer definitions array, so they are not needed here in the API.
+ */
+void trace2_timer_start(enum trace2_timer_id tid);
+void trace2_timer_stop(enum trace2_timer_id tid);
+
+/*
+ * Define the set of global counters.
+ *
+ * We can add more at any time, but they must be defined at compile
+ * time (to avoid the need to dynamically allocate and synchronize
+ * them between different threads).
+ *
+ * These must start at 0 and be contiguous (because we use them
+ * elsewhere as array indexes).
+ *
+ * Any values added to this enum be also be added to the
+ * `tr2_counter_metadata[]` in `trace2/tr2_tr2_ctr.c`.
+ */
+enum trace2_counter_id {
+	/*
+	 * Define two counters for testing.  See `t/helper/test-trace2.c`.
+	 * These can be used for ad hoc testing, but should not be used
+	 * for permanent analysis code.
+	 */
+	TRACE2_COUNTER_ID_TEST1 = 0, /* emits summary event only */
+	TRACE2_COUNTER_ID_TEST2,     /* emits summary and thread events */
+
+	/* Add additional counter definitions before here. */
+	TRACE2_NUMBER_OF_COUNTERS
+};
+
+/*
+ * Increase the named global counter by value.
+ *
+ * Note that this adds `value` to the current thread's partial sum for
+ * this counter (without locking) and that the complete sum is not
+ * available until all threads have exited, so it does not return the
+ * new value of the counter.
+ */
+void trace2_counter_add(enum trace2_counter_id cid, uint64_t value);
+
+/*
+ * Optional platform-specific code to dump information about the
+ * current and any parent process(es).  This is intended to allow
+ * post-processors to know who spawned this git instance and anything
+ * else that the platform may be able to tell us about the current process.
+ */
+
+enum trace2_process_info_reason {
+	TRACE2_PROCESS_INFO_STARTUP,
+	TRACE2_PROCESS_INFO_EXIT,
+};
+
+void trace2_collect_process_info(enum trace2_process_info_reason reason);
+
+const char *trace2_session_id(void);
+
+#endif /* TRACE2_H */
diff --git a/trace2/tr2_cfg.c b/trace2/tr2_cfg.c
new file mode 100644
index 0000000..ec9ac1a
--- /dev/null
+++ b/trace2/tr2_cfg.c
@@ -0,0 +1,147 @@
+#include "cache.h"
+#include "config.h"
+#include "trace2/tr2_cfg.h"
+#include "trace2/tr2_sysenv.h"
+
+static struct strbuf **tr2_cfg_patterns;
+static int tr2_cfg_count_patterns;
+static int tr2_cfg_loaded;
+
+static struct strbuf **tr2_cfg_env_vars;
+static int tr2_cfg_env_vars_count;
+static int tr2_cfg_env_vars_loaded;
+
+/*
+ * Parse a string containing a comma-delimited list of config keys
+ * or wildcard patterns into a list of strbufs.
+ */
+static int tr2_cfg_load_patterns(void)
+{
+	struct strbuf **s;
+	const char *envvar;
+
+	if (tr2_cfg_loaded)
+		return tr2_cfg_count_patterns;
+	tr2_cfg_loaded = 1;
+
+	envvar = tr2_sysenv_get(TR2_SYSENV_CFG_PARAM);
+	if (!envvar || !*envvar)
+		return tr2_cfg_count_patterns;
+
+	tr2_cfg_patterns = strbuf_split_buf(envvar, strlen(envvar), ',', -1);
+	for (s = tr2_cfg_patterns; *s; s++) {
+		struct strbuf *buf = *s;
+
+		if (buf->len && buf->buf[buf->len - 1] == ',')
+			strbuf_setlen(buf, buf->len - 1);
+		strbuf_trim_trailing_newline(*s);
+		strbuf_trim(*s);
+	}
+
+	tr2_cfg_count_patterns = s - tr2_cfg_patterns;
+	return tr2_cfg_count_patterns;
+}
+
+void tr2_cfg_free_patterns(void)
+{
+	if (tr2_cfg_patterns)
+		strbuf_list_free(tr2_cfg_patterns);
+	tr2_cfg_count_patterns = 0;
+	tr2_cfg_loaded = 0;
+}
+
+/*
+ * Parse a string containing a comma-delimited list of environment variable
+ * names into a list of strbufs.
+ */
+static int tr2_load_env_vars(void)
+{
+	struct strbuf **s;
+	const char *varlist;
+
+	if (tr2_cfg_env_vars_loaded)
+		return tr2_cfg_env_vars_count;
+	tr2_cfg_env_vars_loaded = 1;
+
+	varlist = tr2_sysenv_get(TR2_SYSENV_ENV_VARS);
+	if (!varlist || !*varlist)
+		return tr2_cfg_env_vars_count;
+
+	tr2_cfg_env_vars = strbuf_split_buf(varlist, strlen(varlist), ',', -1);
+	for (s = tr2_cfg_env_vars; *s; s++) {
+		struct strbuf *buf = *s;
+
+		if (buf->len && buf->buf[buf->len - 1] == ',')
+			strbuf_setlen(buf, buf->len - 1);
+		strbuf_trim_trailing_newline(*s);
+		strbuf_trim(*s);
+	}
+
+	tr2_cfg_env_vars_count = s - tr2_cfg_env_vars;
+	return tr2_cfg_env_vars_count;
+}
+
+void tr2_cfg_free_env_vars(void)
+{
+	if (tr2_cfg_env_vars)
+		strbuf_list_free(tr2_cfg_env_vars);
+	tr2_cfg_env_vars_count = 0;
+	tr2_cfg_env_vars_loaded = 0;
+}
+
+struct tr2_cfg_data {
+	const char *file;
+	int line;
+};
+
+/*
+ * See if the given config key matches any of our patterns of interest.
+ */
+static int tr2_cfg_cb(const char *key, const char *value, void *d)
+{
+	struct strbuf **s;
+	struct tr2_cfg_data *data = (struct tr2_cfg_data *)d;
+
+	for (s = tr2_cfg_patterns; *s; s++) {
+		struct strbuf *buf = *s;
+		int wm = wildmatch(buf->buf, key, WM_CASEFOLD);
+		if (wm == WM_MATCH) {
+			trace2_def_param_fl(data->file, data->line, key, value);
+			return 0;
+		}
+	}
+
+	return 0;
+}
+
+void tr2_cfg_list_config_fl(const char *file, int line)
+{
+	struct tr2_cfg_data data = { file, line };
+
+	if (tr2_cfg_load_patterns() > 0)
+		read_early_config(tr2_cfg_cb, &data);
+}
+
+void tr2_list_env_vars_fl(const char *file, int line)
+{
+	struct strbuf **s;
+
+	if (tr2_load_env_vars() <= 0)
+		return;
+
+	for (s = tr2_cfg_env_vars; *s; s++) {
+		struct strbuf *buf = *s;
+		const char *val = getenv(buf->buf);
+		if (val && *val)
+			trace2_def_param_fl(file, line, buf->buf, val);
+	}
+}
+
+void tr2_cfg_set_fl(const char *file, int line, const char *key,
+		    const char *value)
+{
+	struct tr2_cfg_data data = { file, line };
+
+	if (tr2_cfg_load_patterns() > 0)
+		tr2_cfg_cb(key, value, &data);
+}
diff --git a/trace2/tr2_cfg.h b/trace2/tr2_cfg.h
new file mode 100644
index 0000000..a11d71f
--- /dev/null
+++ b/trace2/tr2_cfg.h
@@ -0,0 +1,27 @@
+#ifndef TR2_CFG_H
+#define TR2_CFG_H
+
+/*
+ * Iterate over all config settings and emit 'def_param' events for the
+ * "interesting" ones to TRACE2.
+ */
+void tr2_cfg_list_config_fl(const char *file, int line);
+
+/*
+ * Iterate over all "interesting" environment variables and emit 'def_param'
+ * events for them to TRACE2.
+ */
+void tr2_list_env_vars_fl(const char *file, int line);
+
+/*
+ * Emit a "def_param" event for the given key/value pair IF we consider
+ * the key to be "interesting".
+ */
+void tr2_cfg_set_fl(const char *file, int line, const char *key,
+		    const char *value);
+
+void tr2_cfg_free_patterns(void);
+
+void tr2_cfg_free_env_vars(void);
+
+#endif /* TR2_CFG_H */
diff --git a/trace2/tr2_cmd_name.c b/trace2/tr2_cmd_name.c
new file mode 100644
index 0000000..dd31320
--- /dev/null
+++ b/trace2/tr2_cmd_name.c
@@ -0,0 +1,30 @@
+#include "cache.h"
+#include "trace2/tr2_cmd_name.h"
+
+#define TR2_ENVVAR_PARENT_NAME "GIT_TRACE2_PARENT_NAME"
+
+static struct strbuf tr2cmdname_hierarchy = STRBUF_INIT;
+
+void tr2_cmd_name_append_hierarchy(const char *name)
+{
+	const char *parent_name = getenv(TR2_ENVVAR_PARENT_NAME);
+
+	strbuf_reset(&tr2cmdname_hierarchy);
+	if (parent_name && *parent_name) {
+		strbuf_addstr(&tr2cmdname_hierarchy, parent_name);
+		strbuf_addch(&tr2cmdname_hierarchy, '/');
+	}
+	strbuf_addstr(&tr2cmdname_hierarchy, name);
+
+	setenv(TR2_ENVVAR_PARENT_NAME, tr2cmdname_hierarchy.buf, 1);
+}
+
+const char *tr2_cmd_name_get_hierarchy(void)
+{
+	return tr2cmdname_hierarchy.buf;
+}
+
+void tr2_cmd_name_release(void)
+{
+	strbuf_release(&tr2cmdname_hierarchy);
+}
diff --git a/trace2/tr2_cmd_name.h b/trace2/tr2_cmd_name.h
new file mode 100644
index 0000000..ab70b67
--- /dev/null
+++ b/trace2/tr2_cmd_name.h
@@ -0,0 +1,24 @@
+#ifndef TR2_CMD_NAME_H
+#define TR2_CMD_NAME_H
+
+/*
+ * Append the current command name to the list being maintained
+ * in the environment.
+ *
+ * The hierarchy for a top-level git command is just the current
+ * command name.  For a child git process, the hierarchy includes the
+ * names of the parent processes.
+ *
+ * The hierarchy for the current process will be exported to the
+ * environment and inherited by child processes.
+ */
+void tr2_cmd_name_append_hierarchy(const char *name);
+
+/*
+ * Get the command name hierarchy for the current process.
+ */
+const char *tr2_cmd_name_get_hierarchy(void);
+
+void tr2_cmd_name_release(void);
+
+#endif /* TR2_CMD_NAME_H */
diff --git a/trace2/tr2_ctr.c b/trace2/tr2_ctr.c
new file mode 100644
index 0000000..483ca7c
--- /dev/null
+++ b/trace2/tr2_ctr.c
@@ -0,0 +1,101 @@
+#include "cache.h"
+#include "thread-utils.h"
+#include "trace2/tr2_tgt.h"
+#include "trace2/tr2_tls.h"
+#include "trace2/tr2_ctr.h"
+
+/*
+ * A global counter block to aggregrate values from the partial sums
+ * from each thread.
+ */
+static struct tr2_counter_block final_counter_block; /* access under tr2tls_mutex */
+
+/*
+ * Define metadata for each global counter.
+ *
+ * This array must match the "enum trace2_counter_id" and the values
+ * in "struct tr2_counter_block.counter[*]".
+ */
+static struct tr2_counter_metadata tr2_counter_metadata[TRACE2_NUMBER_OF_COUNTERS] = {
+	[TRACE2_COUNTER_ID_TEST1] = {
+		.category = "test",
+		.name = "test1",
+		.want_per_thread_events = 0,
+	},
+	[TRACE2_COUNTER_ID_TEST2] = {
+		.category = "test",
+		.name = "test2",
+		.want_per_thread_events = 1,
+	},
+
+	/* Add additional metadata before here. */
+};
+
+void tr2_counter_increment(enum trace2_counter_id cid, uint64_t value)
+{
+	struct tr2tls_thread_ctx *ctx = tr2tls_get_self();
+	struct tr2_counter *c = &ctx->counter_block.counter[cid];
+
+	c->value += value;
+
+	ctx->used_any_counter = 1;
+	if (tr2_counter_metadata[cid].want_per_thread_events)
+		ctx->used_any_per_thread_counter = 1;
+}
+
+void tr2_update_final_counters(void)
+{
+	struct tr2tls_thread_ctx *ctx = tr2tls_get_self();
+	enum trace2_counter_id cid;
+
+	if (!ctx->used_any_counter)
+		return;
+
+	/*
+	 * Access `final_counter_block` requires holding `tr2tls_mutex`.
+	 * We assume that our caller is holding the lock.
+	 */
+
+	for (cid = 0; cid < TRACE2_NUMBER_OF_COUNTERS; cid++) {
+		struct tr2_counter *c_final = &final_counter_block.counter[cid];
+		const struct tr2_counter *c = &ctx->counter_block.counter[cid];
+
+		c_final->value += c->value;
+	}
+}
+
+void tr2_emit_per_thread_counters(tr2_tgt_evt_counter_t *fn_apply)
+{
+	struct tr2tls_thread_ctx *ctx = tr2tls_get_self();
+	enum trace2_counter_id cid;
+
+	if (!ctx->used_any_per_thread_counter)
+		return;
+
+	/*
+	 * For each counter, if the counter wants per-thread events
+	 * and this thread used it (the value is non-zero), emit it.
+	 */
+	for (cid = 0; cid < TRACE2_NUMBER_OF_COUNTERS; cid++)
+		if (tr2_counter_metadata[cid].want_per_thread_events &&
+		    ctx->counter_block.counter[cid].value)
+			fn_apply(&tr2_counter_metadata[cid],
+				 &ctx->counter_block.counter[cid],
+				 0);
+}
+
+void tr2_emit_final_counters(tr2_tgt_evt_counter_t *fn_apply)
+{
+	enum trace2_counter_id cid;
+
+	/*
+	 * Access `final_counter_block` requires holding `tr2tls_mutex`.
+	 * We assume that our caller is holding the lock.
+	 */
+
+	for (cid = 0; cid < TRACE2_NUMBER_OF_COUNTERS; cid++)
+		if (final_counter_block.counter[cid].value)
+			fn_apply(&tr2_counter_metadata[cid],
+				 &final_counter_block.counter[cid],
+				 1);
+}
diff --git a/trace2/tr2_ctr.h b/trace2/tr2_ctr.h
new file mode 100644
index 0000000..a2267ee
--- /dev/null
+++ b/trace2/tr2_ctr.h
@@ -0,0 +1,104 @@
+#ifndef TR2_CTR_H
+#define TR2_CTR_H
+
+#include "trace2.h"
+#include "trace2/tr2_tgt.h"
+
+/*
+ * Define a mechanism to allow global "counters".
+ *
+ * Counters can be used count interesting activity that does not fit
+ * the "region and data" model, such as code called from many
+ * different regions and/or where you want to count a number of items,
+ * but don't have control of when the last item will be processed,
+ * such as counter the number of calls to `lstat()`.
+ *
+ * Counters differ from Trace2 "data" events.  Data events are emitted
+ * immediately and are appropriate for documenting loop counters at
+ * the end of a region, for example.  Counter values are accumulated
+ * during the program and final counter values are emitted at program
+ * exit.
+ *
+ * To make this model efficient, we define a compile-time fixed set of
+ * counters and counter ids using a fixed size "counter block" array
+ * in thread-local storage.  This gives us constant time, lock-free
+ * access to each counter within each thread.  This lets us avoid the
+ * complexities of dynamically allocating a counter and sharing that
+ * definition with other threads.
+ *
+ * Each thread uses the counter block in its thread-local storage to
+ * increment partial sums for each counter (without locking).  When a
+ * thread exits, those partial sums are (under lock) added to the
+ * global final sum.
+ *
+ * Partial sums for each counter are optionally emitted when a thread
+ * exits.
+ *
+ * Final sums for each counter are emitted between the "exit" and
+ * "atexit" events.
+ *
+ * A parallel "counter metadata" table contains the "category" and
+ * "name" fields for each counter.  This eliminates the need to
+ * include those args in the various counter APIs.
+ */
+
+/*
+ * The definition of an individual counter as used by an individual
+ * thread (and later in aggregation).
+ */
+struct tr2_counter {
+	uint64_t value;
+};
+
+/*
+ * Metadata for a counter.
+ */
+struct tr2_counter_metadata {
+	const char *category;
+	const char *name;
+
+	/*
+	 * True if we should emit per-thread events for this counter
+	 * when individual threads exit.
+	 */
+	unsigned int want_per_thread_events:1;
+};
+
+/*
+ * A compile-time fixed block of counters to insert into thread-local
+ * storage.  This wrapper is used to avoid quirks of C and the usual
+ * need to pass an array size argument.
+ */
+struct tr2_counter_block {
+	struct tr2_counter counter[TRACE2_NUMBER_OF_COUNTERS];
+};
+
+/*
+ * Private routines used by trace2.c to increment a counter for the
+ * current thread.
+ */
+void tr2_counter_increment(enum trace2_counter_id cid, uint64_t value);
+
+/*
+ * Add the current thread's counter data to the global totals.
+ * This is called during thread-exit.
+ *
+ * Caller must be holding the tr2tls_mutex.
+ */
+void tr2_update_final_counters(void);
+
+/*
+ * Emit per-thread counter data for the current thread.
+ * This is called during thread-exit.
+ */
+void tr2_emit_per_thread_counters(tr2_tgt_evt_counter_t *fn_apply);
+
+/*
+ * Emit global counter values.
+ * This is called during atexit handling.
+ *
+ * Caller must be holding the tr2tls_mutex.
+ */
+void tr2_emit_final_counters(tr2_tgt_evt_counter_t *fn_apply);
+
+#endif /* TR2_CTR_H */
diff --git a/trace2/tr2_dst.c b/trace2/tr2_dst.c
new file mode 100644
index 0000000..8a21dd2
--- /dev/null
+++ b/trace2/tr2_dst.c
@@ -0,0 +1,394 @@
+#include "cache.h"
+#include "sigchain.h"
+#include "trace2/tr2_dst.h"
+#include "trace2/tr2_sid.h"
+#include "trace2/tr2_sysenv.h"
+
+/*
+ * How many attempts we will make at creating an automatically-named trace file.
+ */
+#define MAX_AUTO_ATTEMPTS 10
+
+/*
+ * Sentinel file used to detect when we should discard new traces to avoid
+ * writing too many trace files to a directory.
+ */
+#define DISCARD_SENTINEL_NAME "git-trace2-discard"
+
+/*
+ * When set to zero, disables directory file count checks. Otherwise, controls
+ * how many files we can write to a directory before entering discard mode.
+ * This can be overridden via the TR2_SYSENV_MAX_FILES setting.
+ */
+static int tr2env_max_files = 0;
+
+static int tr2_dst_want_warning(void)
+{
+	static int tr2env_dst_debug = -1;
+
+	if (tr2env_dst_debug == -1) {
+		const char *env_value = tr2_sysenv_get(TR2_SYSENV_DST_DEBUG);
+		if (!env_value || !*env_value)
+			tr2env_dst_debug = 0;
+		else
+			tr2env_dst_debug = atoi(env_value) > 0;
+	}
+
+	return tr2env_dst_debug;
+}
+
+void tr2_dst_trace_disable(struct tr2_dst *dst)
+{
+	if (dst->need_close)
+		close(dst->fd);
+	dst->fd = 0;
+	dst->initialized = 1;
+	dst->need_close = 0;
+}
+
+/*
+ * Check to make sure we're not overloading the target directory with too many
+ * files. First get the threshold (if present) from the config or envvar. If
+ * it's zero or unset, disable this check. Next check for the presence of a
+ * sentinel file, then check file count.
+ *
+ * Returns 0 if tracing should proceed as normal. Returns 1 if the sentinel file
+ * already exists, which means tracing should be disabled. Returns -1 if there
+ * are too many files but there was no sentinel file, which means we have
+ * created and should write traces to the sentinel file.
+ *
+ * We expect that some trace processing system is gradually collecting files
+ * from the target directory; after it removes the sentinel file we'll start
+ * writing traces again.
+ */
+static int tr2_dst_too_many_files(struct tr2_dst *dst, const char *tgt_prefix)
+{
+	int file_count = 0, max_files = 0, ret = 0;
+	const char *max_files_var;
+	DIR *dirp;
+	struct strbuf path = STRBUF_INIT, sentinel_path = STRBUF_INIT;
+	struct stat statbuf;
+
+	/* Get the config or envvar and decide if we should continue this check */
+	max_files_var = tr2_sysenv_get(TR2_SYSENV_MAX_FILES);
+	if (max_files_var && *max_files_var && ((max_files = atoi(max_files_var)) >= 0))
+		tr2env_max_files = max_files;
+
+	if (!tr2env_max_files) {
+		ret = 0;
+		goto cleanup;
+	}
+
+	strbuf_addstr(&path, tgt_prefix);
+	if (!is_dir_sep(path.buf[path.len - 1])) {
+		strbuf_addch(&path, '/');
+	}
+
+	/* check sentinel */
+	strbuf_addbuf(&sentinel_path, &path);
+	strbuf_addstr(&sentinel_path, DISCARD_SENTINEL_NAME);
+	if (!stat(sentinel_path.buf, &statbuf)) {
+		ret = 1;
+		goto cleanup;
+	}
+
+	/* check file count */
+	dirp = opendir(path.buf);
+	while (file_count < tr2env_max_files && dirp && readdir(dirp))
+		file_count++;
+	if (dirp)
+		closedir(dirp);
+
+	if (file_count >= tr2env_max_files) {
+		dst->too_many_files = 1;
+		dst->fd = open(sentinel_path.buf, O_WRONLY | O_CREAT | O_EXCL, 0666);
+		ret = -1;
+		goto cleanup;
+	}
+
+cleanup:
+	strbuf_release(&path);
+	strbuf_release(&sentinel_path);
+	return ret;
+}
+
+static int tr2_dst_try_auto_path(struct tr2_dst *dst, const char *tgt_prefix)
+{
+	int too_many_files;
+	const char *last_slash, *sid = tr2_sid_get();
+	struct strbuf path = STRBUF_INIT;
+	size_t base_path_len;
+	unsigned attempt_count;
+
+	last_slash = strrchr(sid, '/');
+	if (last_slash)
+		sid = last_slash + 1;
+
+	strbuf_addstr(&path, tgt_prefix);
+	if (!is_dir_sep(path.buf[path.len - 1]))
+		strbuf_addch(&path, '/');
+	strbuf_addstr(&path, sid);
+	base_path_len = path.len;
+
+	too_many_files = tr2_dst_too_many_files(dst, tgt_prefix);
+	if (!too_many_files) {
+		for (attempt_count = 0; attempt_count < MAX_AUTO_ATTEMPTS; attempt_count++) {
+			if (attempt_count > 0) {
+				strbuf_setlen(&path, base_path_len);
+				strbuf_addf(&path, ".%d", attempt_count);
+			}
+
+			dst->fd = open(path.buf, O_WRONLY | O_CREAT | O_EXCL, 0666);
+			if (dst->fd != -1)
+				break;
+		}
+	} else if (too_many_files == 1) {
+		strbuf_release(&path);
+		if (tr2_dst_want_warning())
+			warning("trace2: not opening %s trace file due to too "
+				"many files in target directory %s",
+				tr2_sysenv_display_name(dst->sysenv_var),
+				tgt_prefix);
+		return 0;
+	}
+
+	if (dst->fd == -1) {
+		if (tr2_dst_want_warning())
+			warning("trace2: could not open '%.*s' for '%s' tracing: %s",
+				(int) base_path_len, path.buf,
+				tr2_sysenv_display_name(dst->sysenv_var),
+				strerror(errno));
+
+		tr2_dst_trace_disable(dst);
+		strbuf_release(&path);
+		return 0;
+	}
+
+	strbuf_release(&path);
+
+	dst->need_close = 1;
+	dst->initialized = 1;
+
+	return dst->fd;
+}
+
+static int tr2_dst_try_path(struct tr2_dst *dst, const char *tgt_value)
+{
+	int fd = open(tgt_value, O_WRONLY | O_APPEND | O_CREAT, 0666);
+	if (fd == -1) {
+		if (tr2_dst_want_warning())
+			warning("trace2: could not open '%s' for '%s' tracing: %s",
+				tgt_value,
+				tr2_sysenv_display_name(dst->sysenv_var),
+				strerror(errno));
+
+		tr2_dst_trace_disable(dst);
+		return 0;
+	}
+
+	dst->fd = fd;
+	dst->need_close = 1;
+	dst->initialized = 1;
+
+	return dst->fd;
+}
+
+#ifndef NO_UNIX_SOCKETS
+#define PREFIX_AF_UNIX "af_unix:"
+#define PREFIX_AF_UNIX_STREAM "af_unix:stream:"
+#define PREFIX_AF_UNIX_DGRAM "af_unix:dgram:"
+
+static int tr2_dst_try_uds_connect(const char *path, int sock_type, int *out_fd)
+{
+	int fd;
+	struct sockaddr_un sa;
+
+	fd = socket(AF_UNIX, sock_type, 0);
+	if (fd == -1)
+		return -1;
+
+	sa.sun_family = AF_UNIX;
+	strlcpy(sa.sun_path, path, sizeof(sa.sun_path));
+
+	if (connect(fd, (struct sockaddr *)&sa, sizeof(sa)) == -1) {
+		int saved_errno = errno;
+		close(fd);
+		errno = saved_errno;
+		return -1;
+	}
+
+	*out_fd = fd;
+	return 0;
+}
+
+#define TR2_DST_UDS_TRY_STREAM (1 << 0)
+#define TR2_DST_UDS_TRY_DGRAM  (1 << 1)
+
+static int tr2_dst_try_unix_domain_socket(struct tr2_dst *dst,
+					  const char *tgt_value)
+{
+	unsigned int uds_try = 0;
+	int fd;
+	const char *path = NULL;
+
+	/*
+	 * Allow "af_unix:[<type>:]<absolute_path>"
+	 *
+	 * Trace2 always writes complete individual messages (without
+	 * chunking), so we can talk to either DGRAM or STREAM type sockets.
+	 *
+	 * Allow the user to explicitly request the socket type.
+	 *
+	 * If they omit the socket type, try one and then the other.
+	 */
+
+	if (skip_prefix(tgt_value, PREFIX_AF_UNIX_STREAM, &path))
+		uds_try |= TR2_DST_UDS_TRY_STREAM;
+
+	else if (skip_prefix(tgt_value, PREFIX_AF_UNIX_DGRAM, &path))
+		uds_try |= TR2_DST_UDS_TRY_DGRAM;
+
+	else if (skip_prefix(tgt_value, PREFIX_AF_UNIX, &path))
+		uds_try |= TR2_DST_UDS_TRY_STREAM | TR2_DST_UDS_TRY_DGRAM;
+
+	if (!path || !*path) {
+		if (tr2_dst_want_warning())
+			warning("trace2: invalid AF_UNIX value '%s' for '%s' tracing",
+				tgt_value,
+				tr2_sysenv_display_name(dst->sysenv_var));
+
+		tr2_dst_trace_disable(dst);
+		return 0;
+	}
+
+	if (!is_absolute_path(path) ||
+	    strlen(path) >= sizeof(((struct sockaddr_un *)0)->sun_path)) {
+		if (tr2_dst_want_warning())
+			warning("trace2: invalid AF_UNIX path '%s' for '%s' tracing",
+				path, tr2_sysenv_display_name(dst->sysenv_var));
+
+		tr2_dst_trace_disable(dst);
+		return 0;
+	}
+
+	if (uds_try & TR2_DST_UDS_TRY_STREAM) {
+		if (!tr2_dst_try_uds_connect(path, SOCK_STREAM, &fd))
+			goto connected;
+		if (errno != EPROTOTYPE)
+			goto error;
+	}
+	if (uds_try & TR2_DST_UDS_TRY_DGRAM) {
+		if (!tr2_dst_try_uds_connect(path, SOCK_DGRAM, &fd))
+			goto connected;
+	}
+
+error:
+	if (tr2_dst_want_warning())
+		warning("trace2: could not connect to socket '%s' for '%s' tracing: %s",
+			path, tr2_sysenv_display_name(dst->sysenv_var),
+			strerror(errno));
+
+	tr2_dst_trace_disable(dst);
+	return 0;
+
+connected:
+	dst->fd = fd;
+	dst->need_close = 1;
+	dst->initialized = 1;
+
+	return dst->fd;
+}
+#endif
+
+static void tr2_dst_malformed_warning(struct tr2_dst *dst,
+				      const char *tgt_value)
+{
+	warning("trace2: unknown value for '%s': '%s'",
+		tr2_sysenv_display_name(dst->sysenv_var), tgt_value);
+}
+
+int tr2_dst_get_trace_fd(struct tr2_dst *dst)
+{
+	const char *tgt_value;
+
+	/* don't open twice */
+	if (dst->initialized)
+		return dst->fd;
+
+	dst->initialized = 1;
+
+	tgt_value = tr2_sysenv_get(dst->sysenv_var);
+
+	if (!tgt_value || !strcmp(tgt_value, "") || !strcmp(tgt_value, "0") ||
+	    !strcasecmp(tgt_value, "false")) {
+		dst->fd = 0;
+		return dst->fd;
+	}
+
+	if (!strcmp(tgt_value, "1") || !strcasecmp(tgt_value, "true")) {
+		dst->fd = STDERR_FILENO;
+		return dst->fd;
+	}
+
+	if (strlen(tgt_value) == 1 && isdigit(*tgt_value)) {
+		dst->fd = atoi(tgt_value);
+		return dst->fd;
+	}
+
+	if (is_absolute_path(tgt_value)) {
+		if (is_directory(tgt_value))
+			return tr2_dst_try_auto_path(dst, tgt_value);
+		else
+			return tr2_dst_try_path(dst, tgt_value);
+	}
+
+#ifndef NO_UNIX_SOCKETS
+	if (starts_with(tgt_value, PREFIX_AF_UNIX))
+		return tr2_dst_try_unix_domain_socket(dst, tgt_value);
+#endif
+
+	/* Always warn about malformed values. */
+	tr2_dst_malformed_warning(dst, tgt_value);
+	tr2_dst_trace_disable(dst);
+	return 0;
+}
+
+int tr2_dst_trace_want(struct tr2_dst *dst)
+{
+	return !!tr2_dst_get_trace_fd(dst);
+}
+
+void tr2_dst_write_line(struct tr2_dst *dst, struct strbuf *buf_line)
+{
+	int fd = tr2_dst_get_trace_fd(dst);
+	ssize_t bytes;
+
+	strbuf_complete_line(buf_line); /* ensure final NL on buffer */
+
+	/*
+	 * We do not use write_in_full() because we do not want
+	 * a short-write to try again.  We are using O_APPEND mode
+	 * files and the kernel handles the atomic seek+write. If
+	 * another thread or git process is concurrently writing to
+	 * this fd or file, our remainder-write may not be contiguous
+	 * with our initial write of this message.  And that will
+	 * confuse readers.  So just don't bother.
+	 *
+	 * It is assumed that TRACE2 messages are short enough that
+	 * the system can write them in 1 attempt and we won't see
+	 * a short-write.
+	 *
+	 * If we get an IO error, just close the trace dst.
+	 */
+	sigchain_push(SIGPIPE, SIG_IGN);
+	bytes = write(fd, buf_line->buf, buf_line->len);
+	sigchain_pop(SIGPIPE);
+	if (bytes >= 0)
+		return;
+
+	tr2_dst_trace_disable(dst);
+	if (tr2_dst_want_warning())
+		warning("unable to write trace to '%s': %s",
+			tr2_sysenv_display_name(dst->sysenv_var),
+			strerror(errno));
+}
diff --git a/trace2/tr2_dst.h b/trace2/tr2_dst.h
new file mode 100644
index 0000000..b1a8c14
--- /dev/null
+++ b/trace2/tr2_dst.h
@@ -0,0 +1,38 @@
+#ifndef TR2_DST_H
+#define TR2_DST_H
+
+struct strbuf;
+#include "trace2/tr2_sysenv.h"
+
+struct tr2_dst {
+	enum tr2_sysenv_variable sysenv_var;
+	int fd;
+	unsigned int initialized : 1;
+	unsigned int need_close : 1;
+	unsigned int too_many_files : 1;
+};
+
+/*
+ * Disable TRACE2 on the destination.  In TRACE2 a destination (DST)
+ * wraps a file descriptor; it is associated with a TARGET which
+ * defines the formatting.
+ */
+void tr2_dst_trace_disable(struct tr2_dst *dst);
+
+/*
+ * Return the file descriptor for the DST.
+ * If 0, the dst is closed or disabled.
+ */
+int tr2_dst_get_trace_fd(struct tr2_dst *dst);
+
+/*
+ * Return true if the DST is opened for writing.
+ */
+int tr2_dst_trace_want(struct tr2_dst *dst);
+
+/*
+ * Write a single line/message to the trace file.
+ */
+void tr2_dst_write_line(struct tr2_dst *dst, struct strbuf *buf_line);
+
+#endif /* TR2_DST_H */
diff --git a/trace2/tr2_sid.c b/trace2/tr2_sid.c
new file mode 100644
index 0000000..dc6e75e
--- /dev/null
+++ b/trace2/tr2_sid.c
@@ -0,0 +1,112 @@
+#include "cache.h"
+#include "trace2/tr2_tbuf.h"
+#include "trace2/tr2_sid.h"
+
+#define TR2_ENVVAR_PARENT_SID "GIT_TRACE2_PARENT_SID"
+
+static struct strbuf tr2sid_buf = STRBUF_INIT;
+static int tr2sid_nr_git_parents;
+
+/*
+ * Compute the final component of the SID representing the current process.
+ * This should uniquely identify the process and be a valid filename (to
+ * allow writing trace2 data to per-process files).  It should also be fixed
+ * length for possible use as a database key.
+ *
+ * "<yyyymmdd>T<hhmmss>.<fraction>Z-<host>-<process>"
+ *
+ * where <host> is a 9 character string:
+ *    "H<first_8_chars_of_sha1_of_hostname>"
+ *    "Localhost" when no hostname.
+ *
+ * where <process> is a 9 character string containing the least significant
+ * 32 bits in the process-id.
+ *    "P<pid>"
+ * (This is an abribrary choice.  On most systems pid_t is a 32 bit value,
+ * so limit doesn't matter.  On larger systems, a truncated value is fine
+ * for our purposes here.)
+ */
+static void tr2_sid_append_my_sid_component(void)
+{
+	const struct git_hash_algo *algo = &hash_algos[GIT_HASH_SHA1];
+	struct tr2_tbuf tb_now;
+	git_hash_ctx ctx;
+	pid_t pid = getpid();
+	unsigned char hash[GIT_MAX_RAWSZ + 1];
+	char hex[GIT_MAX_HEXSZ + 1];
+	char hostname[HOST_NAME_MAX + 1];
+
+	tr2_tbuf_utc_datetime(&tb_now);
+	strbuf_addstr(&tr2sid_buf, tb_now.buf);
+
+	strbuf_addch(&tr2sid_buf, '-');
+	if (xgethostname(hostname, sizeof(hostname)))
+		strbuf_add(&tr2sid_buf, "Localhost", 9);
+	else {
+		algo->init_fn(&ctx);
+		algo->update_fn(&ctx, hostname, strlen(hostname));
+		algo->final_fn(hash, &ctx);
+		hash_to_hex_algop_r(hex, hash, algo);
+		strbuf_addch(&tr2sid_buf, 'H');
+		strbuf_add(&tr2sid_buf, hex, 8);
+	}
+
+	strbuf_addf(&tr2sid_buf, "-P%08"PRIx32, (uint32_t)pid);
+}
+
+/*
+ * Compute a "unique" session id (SID) for the current process.  This allows
+ * all events from this process to have a single label (much like a PID).
+ *
+ * Export this into our environment so that all child processes inherit it.
+ *
+ * If we were started by another git instance, use our parent's SID as a
+ * prefix.  (This lets us track parent/child relationships even if there
+ * is an intermediate shell process.)
+ *
+ * Additionally, count the number of nested git processes.
+ */
+static void tr2_sid_compute(void)
+{
+	const char *parent_sid;
+
+	if (tr2sid_buf.len)
+		return;
+
+	parent_sid = getenv(TR2_ENVVAR_PARENT_SID);
+	if (parent_sid && *parent_sid) {
+		const char *p;
+		for (p = parent_sid; *p; p++)
+			if (*p == '/')
+				tr2sid_nr_git_parents++;
+
+		strbuf_addstr(&tr2sid_buf, parent_sid);
+		strbuf_addch(&tr2sid_buf, '/');
+		tr2sid_nr_git_parents++;
+	}
+
+	tr2_sid_append_my_sid_component();
+
+	setenv(TR2_ENVVAR_PARENT_SID, tr2sid_buf.buf, 1);
+}
+
+const char *tr2_sid_get(void)
+{
+	if (!tr2sid_buf.len)
+		tr2_sid_compute();
+
+	return tr2sid_buf.buf;
+}
+
+int tr2_sid_depth(void)
+{
+	if (!tr2sid_buf.len)
+		tr2_sid_compute();
+
+	return tr2sid_nr_git_parents;
+}
+
+void tr2_sid_release(void)
+{
+	strbuf_release(&tr2sid_buf);
+}
diff --git a/trace2/tr2_sid.h b/trace2/tr2_sid.h
new file mode 100644
index 0000000..9bef321
--- /dev/null
+++ b/trace2/tr2_sid.h
@@ -0,0 +1,18 @@
+#ifndef TR2_SID_H
+#define TR2_SID_H
+
+/*
+ * Get our session id. Compute if necessary.
+ */
+const char *tr2_sid_get(void);
+
+/*
+ * Get our process depth.  A top-level git process invoked from the
+ * command line will have depth=0.  A child git process will have
+ * depth=1 and so on.
+ */
+int tr2_sid_depth(void);
+
+void tr2_sid_release(void);
+
+#endif /* TR2_SID_H */
diff --git a/trace2/tr2_sysenv.c b/trace2/tr2_sysenv.c
new file mode 100644
index 0000000..a380dcf
--- /dev/null
+++ b/trace2/tr2_sysenv.c
@@ -0,0 +1,132 @@
+#include "cache.h"
+#include "config.h"
+#include "dir.h"
+#include "tr2_sysenv.h"
+
+/*
+ * Each entry represents a trace2 setting.
+ * See Documentation/technical/api-trace2.txt
+ */
+struct tr2_sysenv_entry {
+	const char *env_var_name;
+	const char *git_config_name;
+
+	char *value;
+	unsigned int getenv_called : 1;
+};
+
+/*
+ * This table must match "enum tr2_sysenv_variable" in tr2_sysenv.h.
+ *
+ * The strings in this table are constant and must match the published
+ * config and environment variable names as described in the documentation.
+ *
+ * We do not define entries for the GIT_TRACE2_PARENT_* environment
+ * variables because they are transient and used to pass information
+ * from parent to child git processes, rather than settings.
+ */
+/* clang-format off */
+static struct tr2_sysenv_entry tr2_sysenv_settings[] = {
+	[TR2_SYSENV_CFG_PARAM]     = { "GIT_TRACE2_CONFIG_PARAMS",
+				       "trace2.configparams" },
+	[TR2_SYSENV_ENV_VARS]      = { "GIT_TRACE2_ENV_VARS",
+				       "trace2.envvars" },
+
+	[TR2_SYSENV_DST_DEBUG]     = { "GIT_TRACE2_DST_DEBUG",
+				       "trace2.destinationdebug" },
+
+	[TR2_SYSENV_NORMAL]        = { "GIT_TRACE2",
+				       "trace2.normaltarget" },
+	[TR2_SYSENV_NORMAL_BRIEF]  = { "GIT_TRACE2_BRIEF",
+				       "trace2.normalbrief" },
+
+	[TR2_SYSENV_EVENT]         = { "GIT_TRACE2_EVENT",
+				       "trace2.eventtarget" },
+	[TR2_SYSENV_EVENT_BRIEF]   = { "GIT_TRACE2_EVENT_BRIEF",
+				       "trace2.eventbrief" },
+	[TR2_SYSENV_EVENT_NESTING] = { "GIT_TRACE2_EVENT_NESTING",
+				       "trace2.eventnesting" },
+
+	[TR2_SYSENV_PERF]          = { "GIT_TRACE2_PERF",
+				       "trace2.perftarget" },
+	[TR2_SYSENV_PERF_BRIEF]    = { "GIT_TRACE2_PERF_BRIEF",
+				       "trace2.perfbrief" },
+
+	[TR2_SYSENV_MAX_FILES]     = { "GIT_TRACE2_MAX_FILES",
+				       "trace2.maxfiles" },
+};
+/* clang-format on */
+
+static int tr2_sysenv_cb(const char *key, const char *value, void *d)
+{
+	int k;
+
+	if (!starts_with(key, "trace2."))
+		return 0;
+
+	for (k = 0; k < ARRAY_SIZE(tr2_sysenv_settings); k++) {
+		if (!strcmp(key, tr2_sysenv_settings[k].git_config_name)) {
+			free(tr2_sysenv_settings[k].value);
+			tr2_sysenv_settings[k].value = xstrdup(value);
+			return 0;
+		}
+	}
+
+	return 0;
+}
+
+/*
+ * Load Trace2 settings from the system config (usually "/etc/gitconfig"
+ * unless we were built with a runtime-prefix).  These are intended to
+ * define the default values for Trace2 as requested by the administrator.
+ *
+ * Then override with the Trace2 settings from the global config.
+ */
+void tr2_sysenv_load(void)
+{
+	if (ARRAY_SIZE(tr2_sysenv_settings) != TR2_SYSENV_MUST_BE_LAST)
+		BUG("tr2_sysenv_settings size is wrong");
+
+	read_very_early_config(tr2_sysenv_cb, NULL);
+}
+
+/*
+ * Return the value for the requested Trace2 setting from these sources:
+ * the system config, the global config, and the environment.
+ */
+const char *tr2_sysenv_get(enum tr2_sysenv_variable var)
+{
+	if (var >= TR2_SYSENV_MUST_BE_LAST)
+		BUG("tr2_sysenv_get invalid var '%d'", var);
+
+	if (!tr2_sysenv_settings[var].getenv_called) {
+		const char *v = getenv(tr2_sysenv_settings[var].env_var_name);
+		if (v && *v) {
+			free(tr2_sysenv_settings[var].value);
+			tr2_sysenv_settings[var].value = xstrdup(v);
+		}
+		tr2_sysenv_settings[var].getenv_called = 1;
+	}
+
+	return tr2_sysenv_settings[var].value;
+}
+
+/*
+ * Return a friendly name for this setting that is suitable for printing
+ * in an error messages.
+ */
+const char *tr2_sysenv_display_name(enum tr2_sysenv_variable var)
+{
+	if (var >= TR2_SYSENV_MUST_BE_LAST)
+		BUG("tr2_sysenv_get invalid var '%d'", var);
+
+	return tr2_sysenv_settings[var].env_var_name;
+}
+
+void tr2_sysenv_release(void)
+{
+	int k;
+
+	for (k = 0; k < ARRAY_SIZE(tr2_sysenv_settings); k++)
+		free(tr2_sysenv_settings[k].value);
+}
diff --git a/trace2/tr2_sysenv.h b/trace2/tr2_sysenv.h
new file mode 100644
index 0000000..3292ee1
--- /dev/null
+++ b/trace2/tr2_sysenv.h
@@ -0,0 +1,39 @@
+#ifndef TR2_SYSENV_H
+#define TR2_SYSENV_H
+
+/*
+ * The Trace2 settings that can be loaded from /etc/gitconfig
+ * and/or user environment variables.
+ *
+ * Note that this set does not contain any of the transient
+ * environment variables used to pass information from parent
+ * to child git processes, such "GIT_TRACE2_PARENT_SID".
+ */
+enum tr2_sysenv_variable {
+	TR2_SYSENV_CFG_PARAM = 0,
+	TR2_SYSENV_ENV_VARS,
+
+	TR2_SYSENV_DST_DEBUG,
+
+	TR2_SYSENV_NORMAL,
+	TR2_SYSENV_NORMAL_BRIEF,
+
+	TR2_SYSENV_EVENT,
+	TR2_SYSENV_EVENT_BRIEF,
+	TR2_SYSENV_EVENT_NESTING,
+
+	TR2_SYSENV_PERF,
+	TR2_SYSENV_PERF_BRIEF,
+
+	TR2_SYSENV_MAX_FILES,
+
+	TR2_SYSENV_MUST_BE_LAST
+};
+
+void tr2_sysenv_load(void);
+
+const char *tr2_sysenv_get(enum tr2_sysenv_variable);
+const char *tr2_sysenv_display_name(enum tr2_sysenv_variable var);
+void tr2_sysenv_release(void);
+
+#endif /* TR2_SYSENV_H */
diff --git a/trace2/tr2_tbuf.c b/trace2/tr2_tbuf.c
new file mode 100644
index 0000000..2498482
--- /dev/null
+++ b/trace2/tr2_tbuf.c
@@ -0,0 +1,47 @@
+#include "cache.h"
+#include "tr2_tbuf.h"
+
+void tr2_tbuf_local_time(struct tr2_tbuf *tb)
+{
+	struct timeval tv;
+	struct tm tm;
+	time_t secs;
+
+	gettimeofday(&tv, NULL);
+	secs = tv.tv_sec;
+	localtime_r(&secs, &tm);
+
+	xsnprintf(tb->buf, sizeof(tb->buf), "%02d:%02d:%02d.%06ld", tm.tm_hour,
+		  tm.tm_min, tm.tm_sec, (long)tv.tv_usec);
+}
+
+void tr2_tbuf_utc_datetime_extended(struct tr2_tbuf *tb)
+{
+	struct timeval tv;
+	struct tm tm;
+	time_t secs;
+
+	gettimeofday(&tv, NULL);
+	secs = tv.tv_sec;
+	gmtime_r(&secs, &tm);
+
+	xsnprintf(tb->buf, sizeof(tb->buf),
+		  "%4d-%02d-%02dT%02d:%02d:%02d.%06ldZ", tm.tm_year + 1900,
+		  tm.tm_mon + 1, tm.tm_mday, tm.tm_hour, tm.tm_min, tm.tm_sec,
+		  (long)tv.tv_usec);
+}
+
+void tr2_tbuf_utc_datetime(struct tr2_tbuf *tb)
+{
+	struct timeval tv;
+	struct tm tm;
+	time_t secs;
+
+	gettimeofday(&tv, NULL);
+	secs = tv.tv_sec;
+	gmtime_r(&secs, &tm);
+
+	xsnprintf(tb->buf, sizeof(tb->buf), "%4d%02d%02dT%02d%02d%02d.%06ldZ",
+		  tm.tm_year + 1900, tm.tm_mon + 1, tm.tm_mday, tm.tm_hour,
+		  tm.tm_min, tm.tm_sec, (long)tv.tv_usec);
+}
diff --git a/trace2/tr2_tbuf.h b/trace2/tr2_tbuf.h
new file mode 100644
index 0000000..fa853d8
--- /dev/null
+++ b/trace2/tr2_tbuf.h
@@ -0,0 +1,24 @@
+#ifndef TR2_TBUF_H
+#define TR2_TBUF_H
+
+/*
+ * A simple wrapper around a fixed buffer to avoid C syntax
+ * quirks and the need to pass around an additional size_t
+ * argument.
+ */
+struct tr2_tbuf {
+	char buf[32];
+};
+
+/*
+ * Fill buffer with formatted local time string.
+ */
+void tr2_tbuf_local_time(struct tr2_tbuf *tb);
+
+/*
+ * Fill buffer with formatted UTC datatime string.
+ */
+void tr2_tbuf_utc_datetime_extended(struct tr2_tbuf *tb);
+void tr2_tbuf_utc_datetime(struct tr2_tbuf *tb);
+
+#endif /* TR2_TBUF_H */
diff --git a/trace2/tr2_tgt.h b/trace2/tr2_tgt.h
new file mode 100644
index 0000000..bf8745c
--- /dev/null
+++ b/trace2/tr2_tgt.h
@@ -0,0 +1,158 @@
+#ifndef TR2_TGT_H
+#define TR2_TGT_H
+
+struct child_process;
+struct repository;
+struct json_writer;
+struct tr2_timer_metadata;
+struct tr2_timer;
+struct tr2_counter_metadata;
+struct tr2_counter;
+
+#define NS_TO_SEC(ns) ((double)(ns) / 1.0e9)
+
+/*
+ * Function prototypes for a TRACE2 "target" vtable.
+ */
+
+typedef int(tr2_tgt_init_t)(void);
+typedef void(tr2_tgt_term_t)(void);
+
+typedef void(tr2_tgt_evt_version_fl_t)(const char *file, int line);
+
+typedef void(tr2_tgt_evt_start_fl_t)(const char *file, int line,
+				     uint64_t us_elapsed_absolute,
+				     const char **argv);
+typedef void(tr2_tgt_evt_exit_fl_t)(const char *file, int line,
+				    uint64_t us_elapsed_absolute, int code);
+typedef void(tr2_tgt_evt_signal_t)(uint64_t us_elapsed_absolute, int signo);
+typedef void(tr2_tgt_evt_atexit_t)(uint64_t us_elapsed_absolute, int code);
+
+typedef void(tr2_tgt_evt_error_va_fl_t)(const char *file, int line,
+					const char *fmt, va_list ap);
+
+typedef void(tr2_tgt_evt_command_path_fl_t)(const char *file, int line,
+					    const char *command_path);
+typedef void(tr2_tgt_evt_command_ancestry_fl_t)(const char *file, int line,
+						const char **parent_names);
+typedef void(tr2_tgt_evt_command_name_fl_t)(const char *file, int line,
+					    const char *name,
+					    const char *hierarchy);
+typedef void(tr2_tgt_evt_command_mode_fl_t)(const char *file, int line,
+					    const char *mode);
+
+typedef void(tr2_tgt_evt_alias_fl_t)(const char *file, int line,
+				     const char *alias, const char **argv);
+
+typedef void(tr2_tgt_evt_child_start_fl_t)(const char *file, int line,
+					   uint64_t us_elapsed_absolute,
+					   const struct child_process *cmd);
+typedef void(tr2_tgt_evt_child_exit_fl_t)(const char *file, int line,
+					  uint64_t us_elapsed_absolute, int cid,
+					  int pid, int code,
+					  uint64_t us_elapsed_child);
+typedef void(tr2_tgt_evt_child_ready_fl_t)(const char *file, int line,
+					   uint64_t us_elapsed_absolute,
+					   int cid, int pid, const char *ready,
+					   uint64_t us_elapsed_child);
+
+typedef void(tr2_tgt_evt_thread_start_fl_t)(const char *file, int line,
+					    uint64_t us_elapsed_absolute);
+typedef void(tr2_tgt_evt_thread_exit_fl_t)(const char *file, int line,
+					   uint64_t us_elapsed_absolute,
+					   uint64_t us_elapsed_thread);
+
+typedef void(tr2_tgt_evt_exec_fl_t)(const char *file, int line,
+				    uint64_t us_elapsed_absolute, int exec_id,
+				    const char *exe, const char **argv);
+typedef void(tr2_tgt_evt_exec_result_fl_t)(const char *file, int line,
+					   uint64_t us_elapsed_absolute,
+					   int exec_id, int code);
+
+typedef void(tr2_tgt_evt_param_fl_t)(const char *file, int line,
+				     const char *param, const char *value);
+
+typedef void(tr2_tgt_evt_repo_fl_t)(const char *file, int line,
+				    const struct repository *repo);
+
+typedef void(tr2_tgt_evt_region_enter_printf_va_fl_t)(
+	const char *file, int line, uint64_t us_elapsed_absolute,
+	const char *category, const char *label, const struct repository *repo,
+	const char *fmt, va_list ap);
+typedef void(tr2_tgt_evt_region_leave_printf_va_fl_t)(
+	const char *file, int line, uint64_t us_elapsed_absolute,
+	uint64_t us_elapsed_region, const char *category, const char *label,
+	const struct repository *repo, const char *fmt, va_list ap);
+
+typedef void(tr2_tgt_evt_data_fl_t)(const char *file, int line,
+				    uint64_t us_elapsed_absolute,
+				    uint64_t us_elapsed_region,
+				    const char *category,
+				    const struct repository *repo,
+				    const char *key, const char *value);
+typedef void(tr2_tgt_evt_data_json_fl_t)(const char *file, int line,
+					 uint64_t us_elapsed_absolute,
+					 uint64_t us_elapsed_region,
+					 const char *category,
+					 const struct repository *repo,
+					 const char *key,
+					 const struct json_writer *value);
+
+typedef void(tr2_tgt_evt_printf_va_fl_t)(const char *file, int line,
+					 uint64_t us_elapsed_absolute,
+					 const char *fmt, va_list ap);
+
+typedef void(tr2_tgt_evt_timer_t)(const struct tr2_timer_metadata *meta,
+				  const struct tr2_timer *timer,
+				  int is_final_data);
+
+typedef void(tr2_tgt_evt_counter_t)(const struct tr2_counter_metadata *meta,
+				    const struct tr2_counter *counter,
+				    int is_final_data);
+
+/*
+ * "vtable" for a TRACE2 target.  Use NULL if a target does not want
+ * to emit that message.
+ */
+/* clang-format off */
+struct tr2_tgt {
+	struct tr2_dst                          *pdst;
+
+	tr2_tgt_init_t                          *pfn_init;
+	tr2_tgt_term_t                          *pfn_term;
+
+	tr2_tgt_evt_version_fl_t                *pfn_version_fl;
+	tr2_tgt_evt_start_fl_t                  *pfn_start_fl;
+	tr2_tgt_evt_exit_fl_t                   *pfn_exit_fl;
+	tr2_tgt_evt_signal_t                    *pfn_signal;
+	tr2_tgt_evt_atexit_t                    *pfn_atexit;
+	tr2_tgt_evt_error_va_fl_t               *pfn_error_va_fl;
+	tr2_tgt_evt_command_path_fl_t           *pfn_command_path_fl;
+	tr2_tgt_evt_command_ancestry_fl_t	*pfn_command_ancestry_fl;
+	tr2_tgt_evt_command_name_fl_t           *pfn_command_name_fl;
+	tr2_tgt_evt_command_mode_fl_t           *pfn_command_mode_fl;
+	tr2_tgt_evt_alias_fl_t                  *pfn_alias_fl;
+	tr2_tgt_evt_child_start_fl_t            *pfn_child_start_fl;
+	tr2_tgt_evt_child_exit_fl_t             *pfn_child_exit_fl;
+	tr2_tgt_evt_child_ready_fl_t            *pfn_child_ready_fl;
+	tr2_tgt_evt_thread_start_fl_t           *pfn_thread_start_fl;
+	tr2_tgt_evt_thread_exit_fl_t            *pfn_thread_exit_fl;
+	tr2_tgt_evt_exec_fl_t                   *pfn_exec_fl;
+	tr2_tgt_evt_exec_result_fl_t            *pfn_exec_result_fl;
+	tr2_tgt_evt_param_fl_t                  *pfn_param_fl;
+	tr2_tgt_evt_repo_fl_t                   *pfn_repo_fl;
+	tr2_tgt_evt_region_enter_printf_va_fl_t *pfn_region_enter_printf_va_fl;
+	tr2_tgt_evt_region_leave_printf_va_fl_t *pfn_region_leave_printf_va_fl;
+	tr2_tgt_evt_data_fl_t                   *pfn_data_fl;
+	tr2_tgt_evt_data_json_fl_t              *pfn_data_json_fl;
+	tr2_tgt_evt_printf_va_fl_t              *pfn_printf_va_fl;
+	tr2_tgt_evt_timer_t                     *pfn_timer;
+	tr2_tgt_evt_counter_t                   *pfn_counter;
+};
+/* clang-format on */
+
+extern struct tr2_tgt tr2_tgt_event;
+extern struct tr2_tgt tr2_tgt_normal;
+extern struct tr2_tgt tr2_tgt_perf;
+
+#endif /* TR2_TGT_H */
diff --git a/trace2/tr2_tgt_event.c b/trace2/tr2_tgt_event.c
new file mode 100644
index 0000000..16f6332
--- /dev/null
+++ b/trace2/tr2_tgt_event.c
@@ -0,0 +1,696 @@
+#include "cache.h"
+#include "config.h"
+#include "json-writer.h"
+#include "run-command.h"
+#include "version.h"
+#include "trace2/tr2_dst.h"
+#include "trace2/tr2_tbuf.h"
+#include "trace2/tr2_sid.h"
+#include "trace2/tr2_sysenv.h"
+#include "trace2/tr2_tgt.h"
+#include "trace2/tr2_tls.h"
+#include "trace2/tr2_tmr.h"
+
+static struct tr2_dst tr2dst_event = {
+	.sysenv_var = TR2_SYSENV_EVENT,
+};
+
+/*
+ * The version number of the JSON data generated by the EVENT target in this
+ * source file. The version should be incremented if new event types are added,
+ * if existing fields are removed, or if there are significant changes in
+ * interpretation of existing events or fields. Smaller changes, such as adding
+ * a new field to an existing event, do not require an increment to the EVENT
+ * format version.
+ */
+#define TR2_EVENT_VERSION "3"
+
+/*
+ * Region nesting limit for messages written to the event target.
+ *
+ * The "region_enter" and "region_leave" messages (especially recursive
+ * messages such as those produced while diving the worktree or index)
+ * are primarily intended for the performance target during debugging.
+ *
+ * Some of the outer-most messages, however, may be of interest to the
+ * event target.  Use the TR2_SYSENV_EVENT_NESTING setting to increase
+ * region details in the event target.
+ */
+static int tr2env_event_max_nesting_levels = 2;
+
+/*
+ * Use the TR2_SYSENV_EVENT_BRIEF to omit the <time>, <file>, and
+ * <line> fields from most events.
+ */
+static int tr2env_event_be_brief;
+
+static int fn_init(void)
+{
+	int want = tr2_dst_trace_want(&tr2dst_event);
+	int max_nesting;
+	int want_brief;
+	const char *nesting;
+	const char *brief;
+
+	if (!want)
+		return want;
+
+	nesting = tr2_sysenv_get(TR2_SYSENV_EVENT_NESTING);
+	if (nesting && *nesting && ((max_nesting = atoi(nesting)) > 0))
+		tr2env_event_max_nesting_levels = max_nesting;
+
+	brief = tr2_sysenv_get(TR2_SYSENV_EVENT_BRIEF);
+	if (brief && *brief &&
+	    ((want_brief = git_parse_maybe_bool(brief)) != -1))
+		tr2env_event_be_brief = want_brief;
+
+	return want;
+}
+
+static void fn_term(void)
+{
+	tr2_dst_trace_disable(&tr2dst_event);
+}
+
+/*
+ * Append common key-value pairs to the currently open JSON object.
+ *     "event:"<event_name>"
+ *      "sid":"<sid>"
+ *   "thread":"<thread_name>"
+ *     "time":"<time>"
+ *     "file":"<filename>"
+ *     "line":<line_number>
+ *     "repo":<repo_id>
+ */
+static void event_fmt_prepare(const char *event_name, const char *file,
+			      int line, const struct repository *repo,
+			      struct json_writer *jw)
+{
+	struct tr2tls_thread_ctx *ctx = tr2tls_get_self();
+	struct tr2_tbuf tb_now;
+
+	jw_object_string(jw, "event", event_name);
+	jw_object_string(jw, "sid", tr2_sid_get());
+	jw_object_string(jw, "thread", ctx->thread_name);
+
+	/*
+	 * In brief mode, only emit <time> on these 2 event types.
+	 */
+	if (!tr2env_event_be_brief || !strcmp(event_name, "version") ||
+	    !strcmp(event_name, "atexit")) {
+		tr2_tbuf_utc_datetime_extended(&tb_now);
+		jw_object_string(jw, "time", tb_now.buf);
+	}
+
+	if (!tr2env_event_be_brief && file && *file) {
+		jw_object_string(jw, "file", file);
+		jw_object_intmax(jw, "line", line);
+	}
+
+	if (repo)
+		jw_object_intmax(jw, "repo", repo->trace2_repo_id);
+}
+
+static void fn_too_many_files_fl(const char *file, int line)
+{
+	const char *event_name = "too_many_files";
+	struct json_writer jw = JSON_WRITER_INIT;
+
+	jw_object_begin(&jw, 0);
+	event_fmt_prepare(event_name, file, line, NULL, &jw);
+	jw_end(&jw);
+
+	tr2_dst_write_line(&tr2dst_event, &jw.json);
+	jw_release(&jw);
+}
+
+static void fn_version_fl(const char *file, int line)
+{
+	const char *event_name = "version";
+	struct json_writer jw = JSON_WRITER_INIT;
+
+	jw_object_begin(&jw, 0);
+	event_fmt_prepare(event_name, file, line, NULL, &jw);
+	jw_object_string(&jw, "evt", TR2_EVENT_VERSION);
+	jw_object_string(&jw, "exe", git_version_string);
+	jw_end(&jw);
+
+	tr2_dst_write_line(&tr2dst_event, &jw.json);
+	jw_release(&jw);
+
+	if (tr2dst_event.too_many_files)
+		fn_too_many_files_fl(file, line);
+}
+
+static void fn_start_fl(const char *file, int line,
+			uint64_t us_elapsed_absolute, const char **argv)
+{
+	const char *event_name = "start";
+	struct json_writer jw = JSON_WRITER_INIT;
+	double t_abs = (double)us_elapsed_absolute / 1000000.0;
+
+	jw_object_begin(&jw, 0);
+	event_fmt_prepare(event_name, file, line, NULL, &jw);
+	jw_object_double(&jw, "t_abs", 6, t_abs);
+	jw_object_inline_begin_array(&jw, "argv");
+	jw_array_argv(&jw, argv);
+	jw_end(&jw);
+	jw_end(&jw);
+
+	tr2_dst_write_line(&tr2dst_event, &jw.json);
+	jw_release(&jw);
+}
+
+static void fn_exit_fl(const char *file, int line, uint64_t us_elapsed_absolute,
+		       int code)
+{
+	const char *event_name = "exit";
+	struct json_writer jw = JSON_WRITER_INIT;
+	double t_abs = (double)us_elapsed_absolute / 1000000.0;
+
+	jw_object_begin(&jw, 0);
+	event_fmt_prepare(event_name, file, line, NULL, &jw);
+	jw_object_double(&jw, "t_abs", 6, t_abs);
+	jw_object_intmax(&jw, "code", code);
+	jw_end(&jw);
+
+	tr2_dst_write_line(&tr2dst_event, &jw.json);
+	jw_release(&jw);
+}
+
+static void fn_signal(uint64_t us_elapsed_absolute, int signo)
+{
+	const char *event_name = "signal";
+	struct json_writer jw = JSON_WRITER_INIT;
+	double t_abs = (double)us_elapsed_absolute / 1000000.0;
+
+	jw_object_begin(&jw, 0);
+	event_fmt_prepare(event_name, __FILE__, __LINE__, NULL, &jw);
+	jw_object_double(&jw, "t_abs", 6, t_abs);
+	jw_object_intmax(&jw, "signo", signo);
+	jw_end(&jw);
+
+	tr2_dst_write_line(&tr2dst_event, &jw.json);
+	jw_release(&jw);
+}
+
+static void fn_atexit(uint64_t us_elapsed_absolute, int code)
+{
+	const char *event_name = "atexit";
+	struct json_writer jw = JSON_WRITER_INIT;
+	double t_abs = (double)us_elapsed_absolute / 1000000.0;
+
+	jw_object_begin(&jw, 0);
+	event_fmt_prepare(event_name, __FILE__, __LINE__, NULL, &jw);
+	jw_object_double(&jw, "t_abs", 6, t_abs);
+	jw_object_intmax(&jw, "code", code);
+	jw_end(&jw);
+
+	tr2_dst_write_line(&tr2dst_event, &jw.json);
+	jw_release(&jw);
+}
+
+static void maybe_add_string_va(struct json_writer *jw, const char *field_name,
+				const char *fmt, va_list ap)
+{
+	if (fmt && *fmt) {
+		va_list copy_ap;
+		struct strbuf buf = STRBUF_INIT;
+
+		va_copy(copy_ap, ap);
+		strbuf_vaddf(&buf, fmt, copy_ap);
+		va_end(copy_ap);
+
+		jw_object_string(jw, field_name, buf.buf);
+		strbuf_release(&buf);
+		return;
+	}
+}
+
+static void fn_error_va_fl(const char *file, int line, const char *fmt,
+			   va_list ap)
+{
+	const char *event_name = "error";
+	struct json_writer jw = JSON_WRITER_INIT;
+
+	jw_object_begin(&jw, 0);
+	event_fmt_prepare(event_name, file, line, NULL, &jw);
+	maybe_add_string_va(&jw, "msg", fmt, ap);
+	/*
+	 * Also emit the format string as a field in case
+	 * post-processors want to aggregate common error
+	 * messages by type without argument fields (such
+	 * as pathnames or branch names) cluttering it up.
+	 */
+	if (fmt && *fmt)
+		jw_object_string(&jw, "fmt", fmt);
+	jw_end(&jw);
+
+	tr2_dst_write_line(&tr2dst_event, &jw.json);
+	jw_release(&jw);
+}
+
+static void fn_command_path_fl(const char *file, int line, const char *pathname)
+{
+	const char *event_name = "cmd_path";
+	struct json_writer jw = JSON_WRITER_INIT;
+
+	jw_object_begin(&jw, 0);
+	event_fmt_prepare(event_name, file, line, NULL, &jw);
+	jw_object_string(&jw, "path", pathname);
+	jw_end(&jw);
+
+	tr2_dst_write_line(&tr2dst_event, &jw.json);
+	jw_release(&jw);
+}
+
+static void fn_command_ancestry_fl(const char *file, int line, const char **parent_names)
+{
+	const char *event_name = "cmd_ancestry";
+	const char *parent_name = NULL;
+	struct json_writer jw = JSON_WRITER_INIT;
+
+	jw_object_begin(&jw, 0);
+	event_fmt_prepare(event_name, file, line, NULL, &jw);
+	jw_object_inline_begin_array(&jw, "ancestry");
+
+	while ((parent_name = *parent_names++))
+		jw_array_string(&jw, parent_name);
+
+	jw_end(&jw); /* 'ancestry' array */
+	jw_end(&jw); /* event object */
+
+	tr2_dst_write_line(&tr2dst_event, &jw.json);
+	jw_release(&jw);
+}
+
+static void fn_command_name_fl(const char *file, int line, const char *name,
+			       const char *hierarchy)
+{
+	const char *event_name = "cmd_name";
+	struct json_writer jw = JSON_WRITER_INIT;
+
+	jw_object_begin(&jw, 0);
+	event_fmt_prepare(event_name, file, line, NULL, &jw);
+	jw_object_string(&jw, "name", name);
+	if (hierarchy && *hierarchy)
+		jw_object_string(&jw, "hierarchy", hierarchy);
+	jw_end(&jw);
+
+	tr2_dst_write_line(&tr2dst_event, &jw.json);
+	jw_release(&jw);
+}
+
+static void fn_command_mode_fl(const char *file, int line, const char *mode)
+{
+	const char *event_name = "cmd_mode";
+	struct json_writer jw = JSON_WRITER_INIT;
+
+	jw_object_begin(&jw, 0);
+	event_fmt_prepare(event_name, file, line, NULL, &jw);
+	jw_object_string(&jw, "name", mode);
+	jw_end(&jw);
+
+	tr2_dst_write_line(&tr2dst_event, &jw.json);
+	jw_release(&jw);
+}
+
+static void fn_alias_fl(const char *file, int line, const char *alias,
+			const char **argv)
+{
+	const char *event_name = "alias";
+	struct json_writer jw = JSON_WRITER_INIT;
+
+	jw_object_begin(&jw, 0);
+	event_fmt_prepare(event_name, file, line, NULL, &jw);
+	jw_object_string(&jw, "alias", alias);
+	jw_object_inline_begin_array(&jw, "argv");
+	jw_array_argv(&jw, argv);
+	jw_end(&jw);
+	jw_end(&jw);
+
+	tr2_dst_write_line(&tr2dst_event, &jw.json);
+	jw_release(&jw);
+}
+
+static void fn_child_start_fl(const char *file, int line,
+			      uint64_t us_elapsed_absolute,
+			      const struct child_process *cmd)
+{
+	const char *event_name = "child_start";
+	struct json_writer jw = JSON_WRITER_INIT;
+
+	jw_object_begin(&jw, 0);
+	event_fmt_prepare(event_name, file, line, NULL, &jw);
+	jw_object_intmax(&jw, "child_id", cmd->trace2_child_id);
+	if (cmd->trace2_hook_name) {
+		jw_object_string(&jw, "child_class", "hook");
+		jw_object_string(&jw, "hook_name", cmd->trace2_hook_name);
+	} else {
+		const char *child_class =
+			cmd->trace2_child_class ? cmd->trace2_child_class : "?";
+		jw_object_string(&jw, "child_class", child_class);
+	}
+	if (cmd->dir)
+		jw_object_string(&jw, "cd", cmd->dir);
+	jw_object_bool(&jw, "use_shell", cmd->use_shell);
+	jw_object_inline_begin_array(&jw, "argv");
+	if (cmd->git_cmd)
+		jw_array_string(&jw, "git");
+	jw_array_argv(&jw, cmd->args.v);
+	jw_end(&jw);
+	jw_end(&jw);
+
+	tr2_dst_write_line(&tr2dst_event, &jw.json);
+	jw_release(&jw);
+}
+
+static void fn_child_exit_fl(const char *file, int line,
+			     uint64_t us_elapsed_absolute, int cid, int pid,
+			     int code, uint64_t us_elapsed_child)
+{
+	const char *event_name = "child_exit";
+	struct json_writer jw = JSON_WRITER_INIT;
+	double t_rel = (double)us_elapsed_child / 1000000.0;
+
+	jw_object_begin(&jw, 0);
+	event_fmt_prepare(event_name, file, line, NULL, &jw);
+	jw_object_intmax(&jw, "child_id", cid);
+	jw_object_intmax(&jw, "pid", pid);
+	jw_object_intmax(&jw, "code", code);
+	jw_object_double(&jw, "t_rel", 6, t_rel);
+	jw_end(&jw);
+
+	tr2_dst_write_line(&tr2dst_event, &jw.json);
+
+	jw_release(&jw);
+}
+
+static void fn_child_ready_fl(const char *file, int line,
+			      uint64_t us_elapsed_absolute, int cid, int pid,
+			      const char *ready, uint64_t us_elapsed_child)
+{
+	const char *event_name = "child_ready";
+	struct json_writer jw = JSON_WRITER_INIT;
+	double t_rel = (double)us_elapsed_child / 1000000.0;
+
+	jw_object_begin(&jw, 0);
+	event_fmt_prepare(event_name, file, line, NULL, &jw);
+	jw_object_intmax(&jw, "child_id", cid);
+	jw_object_intmax(&jw, "pid", pid);
+	jw_object_string(&jw, "ready", ready);
+	jw_object_double(&jw, "t_rel", 6, t_rel);
+	jw_end(&jw);
+
+	tr2_dst_write_line(&tr2dst_event, &jw.json);
+
+	jw_release(&jw);
+}
+
+static void fn_thread_start_fl(const char *file, int line,
+			       uint64_t us_elapsed_absolute)
+{
+	const char *event_name = "thread_start";
+	struct json_writer jw = JSON_WRITER_INIT;
+
+	jw_object_begin(&jw, 0);
+	event_fmt_prepare(event_name, file, line, NULL, &jw);
+	jw_end(&jw);
+
+	tr2_dst_write_line(&tr2dst_event, &jw.json);
+	jw_release(&jw);
+}
+
+static void fn_thread_exit_fl(const char *file, int line,
+			      uint64_t us_elapsed_absolute,
+			      uint64_t us_elapsed_thread)
+{
+	const char *event_name = "thread_exit";
+	struct json_writer jw = JSON_WRITER_INIT;
+	double t_rel = (double)us_elapsed_thread / 1000000.0;
+
+	jw_object_begin(&jw, 0);
+	event_fmt_prepare(event_name, file, line, NULL, &jw);
+	jw_object_double(&jw, "t_rel", 6, t_rel);
+	jw_end(&jw);
+
+	tr2_dst_write_line(&tr2dst_event, &jw.json);
+	jw_release(&jw);
+}
+
+static void fn_exec_fl(const char *file, int line, uint64_t us_elapsed_absolute,
+		       int exec_id, const char *exe, const char **argv)
+{
+	const char *event_name = "exec";
+	struct json_writer jw = JSON_WRITER_INIT;
+
+	jw_object_begin(&jw, 0);
+	event_fmt_prepare(event_name, file, line, NULL, &jw);
+	jw_object_intmax(&jw, "exec_id", exec_id);
+	if (exe)
+		jw_object_string(&jw, "exe", exe);
+	jw_object_inline_begin_array(&jw, "argv");
+	jw_array_argv(&jw, argv);
+	jw_end(&jw);
+	jw_end(&jw);
+
+	tr2_dst_write_line(&tr2dst_event, &jw.json);
+	jw_release(&jw);
+}
+
+static void fn_exec_result_fl(const char *file, int line,
+			      uint64_t us_elapsed_absolute, int exec_id,
+			      int code)
+{
+	const char *event_name = "exec_result";
+	struct json_writer jw = JSON_WRITER_INIT;
+
+	jw_object_begin(&jw, 0);
+	event_fmt_prepare(event_name, file, line, NULL, &jw);
+	jw_object_intmax(&jw, "exec_id", exec_id);
+	jw_object_intmax(&jw, "code", code);
+	jw_end(&jw);
+
+	tr2_dst_write_line(&tr2dst_event, &jw.json);
+	jw_release(&jw);
+}
+
+static void fn_param_fl(const char *file, int line, const char *param,
+			const char *value)
+{
+	const char *event_name = "def_param";
+	struct json_writer jw = JSON_WRITER_INIT;
+	enum config_scope scope = current_config_scope();
+	const char *scope_name = config_scope_name(scope);
+
+	jw_object_begin(&jw, 0);
+	event_fmt_prepare(event_name, file, line, NULL, &jw);
+	jw_object_string(&jw, "scope", scope_name);
+	jw_object_string(&jw, "param", param);
+	jw_object_string(&jw, "value", value);
+	jw_end(&jw);
+
+	tr2_dst_write_line(&tr2dst_event, &jw.json);
+	jw_release(&jw);
+}
+
+static void fn_repo_fl(const char *file, int line,
+		       const struct repository *repo)
+{
+	const char *event_name = "def_repo";
+	struct json_writer jw = JSON_WRITER_INIT;
+
+	jw_object_begin(&jw, 0);
+	event_fmt_prepare(event_name, file, line, repo, &jw);
+	jw_object_string(&jw, "worktree", repo->worktree);
+	jw_end(&jw);
+
+	tr2_dst_write_line(&tr2dst_event, &jw.json);
+	jw_release(&jw);
+}
+
+static void fn_region_enter_printf_va_fl(const char *file, int line,
+					 uint64_t us_elapsed_absolute,
+					 const char *category,
+					 const char *label,
+					 const struct repository *repo,
+					 const char *fmt, va_list ap)
+{
+	const char *event_name = "region_enter";
+	struct tr2tls_thread_ctx *ctx = tr2tls_get_self();
+	if (ctx->nr_open_regions <= tr2env_event_max_nesting_levels) {
+		struct json_writer jw = JSON_WRITER_INIT;
+
+		jw_object_begin(&jw, 0);
+		event_fmt_prepare(event_name, file, line, repo, &jw);
+		jw_object_intmax(&jw, "nesting", ctx->nr_open_regions);
+		if (category)
+			jw_object_string(&jw, "category", category);
+		if (label)
+			jw_object_string(&jw, "label", label);
+		maybe_add_string_va(&jw, "msg", fmt, ap);
+		jw_end(&jw);
+
+		tr2_dst_write_line(&tr2dst_event, &jw.json);
+		jw_release(&jw);
+	}
+}
+
+static void fn_region_leave_printf_va_fl(
+	const char *file, int line, uint64_t us_elapsed_absolute,
+	uint64_t us_elapsed_region, const char *category, const char *label,
+	const struct repository *repo, const char *fmt, va_list ap)
+{
+	const char *event_name = "region_leave";
+	struct tr2tls_thread_ctx *ctx = tr2tls_get_self();
+	if (ctx->nr_open_regions <= tr2env_event_max_nesting_levels) {
+		struct json_writer jw = JSON_WRITER_INIT;
+		double t_rel = (double)us_elapsed_region / 1000000.0;
+
+		jw_object_begin(&jw, 0);
+		event_fmt_prepare(event_name, file, line, repo, &jw);
+		jw_object_double(&jw, "t_rel", 6, t_rel);
+		jw_object_intmax(&jw, "nesting", ctx->nr_open_regions);
+		if (category)
+			jw_object_string(&jw, "category", category);
+		if (label)
+			jw_object_string(&jw, "label", label);
+		maybe_add_string_va(&jw, "msg", fmt, ap);
+		jw_end(&jw);
+
+		tr2_dst_write_line(&tr2dst_event, &jw.json);
+		jw_release(&jw);
+	}
+}
+
+static void fn_data_fl(const char *file, int line, uint64_t us_elapsed_absolute,
+		       uint64_t us_elapsed_region, const char *category,
+		       const struct repository *repo, const char *key,
+		       const char *value)
+{
+	const char *event_name = "data";
+	struct tr2tls_thread_ctx *ctx = tr2tls_get_self();
+	if (ctx->nr_open_regions <= tr2env_event_max_nesting_levels) {
+		struct json_writer jw = JSON_WRITER_INIT;
+		double t_abs = (double)us_elapsed_absolute / 1000000.0;
+		double t_rel = (double)us_elapsed_region / 1000000.0;
+
+		jw_object_begin(&jw, 0);
+		event_fmt_prepare(event_name, file, line, repo, &jw);
+		jw_object_double(&jw, "t_abs", 6, t_abs);
+		jw_object_double(&jw, "t_rel", 6, t_rel);
+		jw_object_intmax(&jw, "nesting", ctx->nr_open_regions);
+		jw_object_string(&jw, "category", category);
+		jw_object_string(&jw, "key", key);
+		jw_object_string(&jw, "value", value);
+		jw_end(&jw);
+
+		tr2_dst_write_line(&tr2dst_event, &jw.json);
+		jw_release(&jw);
+	}
+}
+
+static void fn_data_json_fl(const char *file, int line,
+			    uint64_t us_elapsed_absolute,
+			    uint64_t us_elapsed_region, const char *category,
+			    const struct repository *repo, const char *key,
+			    const struct json_writer *value)
+{
+	const char *event_name = "data_json";
+	struct tr2tls_thread_ctx *ctx = tr2tls_get_self();
+	if (ctx->nr_open_regions <= tr2env_event_max_nesting_levels) {
+		struct json_writer jw = JSON_WRITER_INIT;
+		double t_abs = (double)us_elapsed_absolute / 1000000.0;
+		double t_rel = (double)us_elapsed_region / 1000000.0;
+
+		jw_object_begin(&jw, 0);
+		event_fmt_prepare(event_name, file, line, repo, &jw);
+		jw_object_double(&jw, "t_abs", 6, t_abs);
+		jw_object_double(&jw, "t_rel", 6, t_rel);
+		jw_object_intmax(&jw, "nesting", ctx->nr_open_regions);
+		jw_object_string(&jw, "category", category);
+		jw_object_string(&jw, "key", key);
+		jw_object_sub_jw(&jw, "value", value);
+		jw_end(&jw);
+
+		tr2_dst_write_line(&tr2dst_event, &jw.json);
+		jw_release(&jw);
+	}
+}
+
+static void fn_timer(const struct tr2_timer_metadata *meta,
+		     const struct tr2_timer *timer,
+		     int is_final_data)
+{
+	const char *event_name = is_final_data ? "timer" : "th_timer";
+	struct json_writer jw = JSON_WRITER_INIT;
+	double t_total = NS_TO_SEC(timer->total_ns);
+	double t_min = NS_TO_SEC(timer->min_ns);
+	double t_max = NS_TO_SEC(timer->max_ns);
+
+	jw_object_begin(&jw, 0);
+	event_fmt_prepare(event_name, __FILE__, __LINE__, NULL, &jw);
+	jw_object_string(&jw, "category", meta->category);
+	jw_object_string(&jw, "name", meta->name);
+	jw_object_intmax(&jw, "intervals", timer->interval_count);
+	jw_object_double(&jw, "t_total", 6, t_total);
+	jw_object_double(&jw, "t_min", 6, t_min);
+	jw_object_double(&jw, "t_max", 6, t_max);
+	jw_end(&jw);
+
+	tr2_dst_write_line(&tr2dst_event, &jw.json);
+	jw_release(&jw);
+}
+
+static void fn_counter(const struct tr2_counter_metadata *meta,
+		       const struct tr2_counter *counter,
+		       int is_final_data)
+{
+	const char *event_name = is_final_data ? "counter" : "th_counter";
+	struct json_writer jw = JSON_WRITER_INIT;
+
+	jw_object_begin(&jw, 0);
+	event_fmt_prepare(event_name, __FILE__, __LINE__, NULL, &jw);
+	jw_object_string(&jw, "category", meta->category);
+	jw_object_string(&jw, "name", meta->name);
+	jw_object_intmax(&jw, "count", counter->value);
+	jw_end(&jw);
+
+	tr2_dst_write_line(&tr2dst_event, &jw.json);
+	jw_release(&jw);
+}
+
+struct tr2_tgt tr2_tgt_event = {
+	.pdst = &tr2dst_event,
+
+	.pfn_init = fn_init,
+	.pfn_term = fn_term,
+
+	.pfn_version_fl = fn_version_fl,
+	.pfn_start_fl = fn_start_fl,
+	.pfn_exit_fl = fn_exit_fl,
+	.pfn_signal = fn_signal,
+	.pfn_atexit = fn_atexit,
+	.pfn_error_va_fl = fn_error_va_fl,
+	.pfn_command_path_fl = fn_command_path_fl,
+	.pfn_command_ancestry_fl = fn_command_ancestry_fl,
+	.pfn_command_name_fl = fn_command_name_fl,
+	.pfn_command_mode_fl = fn_command_mode_fl,
+	.pfn_alias_fl = fn_alias_fl,
+	.pfn_child_start_fl = fn_child_start_fl,
+	.pfn_child_exit_fl = fn_child_exit_fl,
+	.pfn_child_ready_fl = fn_child_ready_fl,
+	.pfn_thread_start_fl = fn_thread_start_fl,
+	.pfn_thread_exit_fl = fn_thread_exit_fl,
+	.pfn_exec_fl = fn_exec_fl,
+	.pfn_exec_result_fl = fn_exec_result_fl,
+	.pfn_param_fl = fn_param_fl,
+	.pfn_repo_fl = fn_repo_fl,
+	.pfn_region_enter_printf_va_fl = fn_region_enter_printf_va_fl,
+	.pfn_region_leave_printf_va_fl = fn_region_leave_printf_va_fl,
+	.pfn_data_fl = fn_data_fl,
+	.pfn_data_json_fl = fn_data_json_fl,
+	.pfn_printf_va_fl = NULL,
+	.pfn_timer = fn_timer,
+	.pfn_counter = fn_counter,
+};
diff --git a/trace2/tr2_tgt_normal.c b/trace2/tr2_tgt_normal.c
new file mode 100644
index 0000000..fbbef68
--- /dev/null
+++ b/trace2/tr2_tgt_normal.c
@@ -0,0 +1,402 @@
+#include "cache.h"
+#include "config.h"
+#include "run-command.h"
+#include "quote.h"
+#include "version.h"
+#include "trace2/tr2_dst.h"
+#include "trace2/tr2_sysenv.h"
+#include "trace2/tr2_tbuf.h"
+#include "trace2/tr2_tgt.h"
+#include "trace2/tr2_tls.h"
+#include "trace2/tr2_tmr.h"
+
+static struct tr2_dst tr2dst_normal = {
+	.sysenv_var = TR2_SYSENV_NORMAL,
+};
+
+/*
+ * Use the TR2_SYSENV_NORMAL_BRIEF setting to omit the "<time> <file>:<line>"
+ * fields from each line written to the builtin normal target.
+ *
+ * Unit tests may want to use this to help with testing.
+ */
+static int tr2env_normal_be_brief;
+
+#define TR2FMT_NORMAL_FL_WIDTH (50)
+
+static int fn_init(void)
+{
+	int want = tr2_dst_trace_want(&tr2dst_normal);
+	int want_brief;
+	const char *brief;
+
+	if (!want)
+		return want;
+
+	brief = tr2_sysenv_get(TR2_SYSENV_NORMAL_BRIEF);
+	if (brief && *brief &&
+	    ((want_brief = git_parse_maybe_bool(brief)) != -1))
+		tr2env_normal_be_brief = want_brief;
+
+	return want;
+}
+
+static void fn_term(void)
+{
+	tr2_dst_trace_disable(&tr2dst_normal);
+}
+
+static void normal_fmt_prepare(const char *file, int line, struct strbuf *buf)
+{
+	strbuf_setlen(buf, 0);
+
+	if (!tr2env_normal_be_brief) {
+		struct tr2_tbuf tb_now;
+
+		tr2_tbuf_local_time(&tb_now);
+		strbuf_addstr(buf, tb_now.buf);
+		strbuf_addch(buf, ' ');
+
+		if (file && *file)
+			strbuf_addf(buf, "%s:%d ", file, line);
+		while (buf->len < TR2FMT_NORMAL_FL_WIDTH)
+			strbuf_addch(buf, ' ');
+	}
+}
+
+static void normal_io_write_fl(const char *file, int line,
+			       const struct strbuf *buf_payload)
+{
+	struct strbuf buf_line = STRBUF_INIT;
+
+	normal_fmt_prepare(file, line, &buf_line);
+	strbuf_addbuf(&buf_line, buf_payload);
+	tr2_dst_write_line(&tr2dst_normal, &buf_line);
+	strbuf_release(&buf_line);
+}
+
+static void fn_version_fl(const char *file, int line)
+{
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	strbuf_addf(&buf_payload, "version %s", git_version_string);
+	normal_io_write_fl(file, line, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_start_fl(const char *file, int line,
+			uint64_t us_elapsed_absolute, const char **argv)
+{
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	strbuf_addstr(&buf_payload, "start ");
+	sq_append_quote_argv_pretty(&buf_payload, argv);
+	normal_io_write_fl(file, line, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_exit_fl(const char *file, int line, uint64_t us_elapsed_absolute,
+		       int code)
+{
+	struct strbuf buf_payload = STRBUF_INIT;
+	double elapsed = (double)us_elapsed_absolute / 1000000.0;
+
+	strbuf_addf(&buf_payload, "exit elapsed:%.6f code:%d", elapsed, code);
+	normal_io_write_fl(file, line, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_signal(uint64_t us_elapsed_absolute, int signo)
+{
+	struct strbuf buf_payload = STRBUF_INIT;
+	double elapsed = (double)us_elapsed_absolute / 1000000.0;
+
+	strbuf_addf(&buf_payload, "signal elapsed:%.6f code:%d", elapsed,
+		    signo);
+	normal_io_write_fl(__FILE__, __LINE__, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_atexit(uint64_t us_elapsed_absolute, int code)
+{
+	struct strbuf buf_payload = STRBUF_INIT;
+	double elapsed = (double)us_elapsed_absolute / 1000000.0;
+
+	strbuf_addf(&buf_payload, "atexit elapsed:%.6f code:%d", elapsed, code);
+	normal_io_write_fl(__FILE__, __LINE__, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void maybe_append_string_va(struct strbuf *buf, const char *fmt,
+				   va_list ap)
+{
+	if (fmt && *fmt) {
+		va_list copy_ap;
+
+		va_copy(copy_ap, ap);
+		strbuf_vaddf(buf, fmt, copy_ap);
+		va_end(copy_ap);
+		return;
+	}
+}
+
+static void fn_error_va_fl(const char *file, int line, const char *fmt,
+			   va_list ap)
+{
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	strbuf_addstr(&buf_payload, "error");
+	if (fmt && *fmt) {
+		strbuf_addch(&buf_payload, ' ');
+		maybe_append_string_va(&buf_payload, fmt, ap);
+	}
+	normal_io_write_fl(file, line, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_command_path_fl(const char *file, int line, const char *pathname)
+{
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	strbuf_addf(&buf_payload, "cmd_path %s", pathname);
+	normal_io_write_fl(file, line, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_command_ancestry_fl(const char *file, int line, const char **parent_names)
+{
+	const char *parent_name = NULL;
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	/* cmd_ancestry parent <- grandparent <- great-grandparent */
+	strbuf_addstr(&buf_payload, "cmd_ancestry ");
+	while ((parent_name = *parent_names++)) {
+		strbuf_addstr(&buf_payload, parent_name);
+		/* if we'll write another one after this, add a delimiter */
+		if (parent_names && *parent_names)
+			strbuf_addstr(&buf_payload, " <- ");
+	}
+
+	normal_io_write_fl(file, line, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_command_name_fl(const char *file, int line, const char *name,
+			       const char *hierarchy)
+{
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	strbuf_addf(&buf_payload, "cmd_name %s", name);
+	if (hierarchy && *hierarchy)
+		strbuf_addf(&buf_payload, " (%s)", hierarchy);
+	normal_io_write_fl(file, line, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_command_mode_fl(const char *file, int line, const char *mode)
+{
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	strbuf_addf(&buf_payload, "cmd_mode %s", mode);
+	normal_io_write_fl(file, line, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_alias_fl(const char *file, int line, const char *alias,
+			const char **argv)
+{
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	strbuf_addf(&buf_payload, "alias %s -> ", alias);
+	sq_append_quote_argv_pretty(&buf_payload, argv);
+	normal_io_write_fl(file, line, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_child_start_fl(const char *file, int line,
+			      uint64_t us_elapsed_absolute,
+			      const struct child_process *cmd)
+{
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	strbuf_addf(&buf_payload, "child_start[%d]", cmd->trace2_child_id);
+
+	if (cmd->dir) {
+		strbuf_addstr(&buf_payload, " cd ");
+		sq_quote_buf_pretty(&buf_payload, cmd->dir);
+		strbuf_addstr(&buf_payload, ";");
+	}
+
+	/*
+	 * TODO if (cmd->env) { Consider dumping changes to environment. }
+	 * See trace_add_env() in run-command.c as used by original trace.c
+	 */
+
+	strbuf_addch(&buf_payload, ' ');
+	if (cmd->git_cmd)
+		strbuf_addstr(&buf_payload, "git ");
+	sq_append_quote_argv_pretty(&buf_payload, cmd->args.v);
+
+	normal_io_write_fl(file, line, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_child_exit_fl(const char *file, int line,
+			     uint64_t us_elapsed_absolute, int cid, int pid,
+			     int code, uint64_t us_elapsed_child)
+{
+	struct strbuf buf_payload = STRBUF_INIT;
+	double elapsed = (double)us_elapsed_child / 1000000.0;
+
+	strbuf_addf(&buf_payload, "child_exit[%d] pid:%d code:%d elapsed:%.6f",
+		    cid, pid, code, elapsed);
+	normal_io_write_fl(file, line, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_child_ready_fl(const char *file, int line,
+			      uint64_t us_elapsed_absolute, int cid, int pid,
+			      const char *ready, uint64_t us_elapsed_child)
+{
+	struct strbuf buf_payload = STRBUF_INIT;
+	double elapsed = (double)us_elapsed_child / 1000000.0;
+
+	strbuf_addf(&buf_payload, "child_ready[%d] pid:%d ready:%s elapsed:%.6f",
+		    cid, pid, ready, elapsed);
+	normal_io_write_fl(file, line, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_exec_fl(const char *file, int line, uint64_t us_elapsed_absolute,
+		       int exec_id, const char *exe, const char **argv)
+{
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	strbuf_addf(&buf_payload, "exec[%d] ", exec_id);
+	if (exe) {
+		strbuf_addstr(&buf_payload, exe);
+		strbuf_addch(&buf_payload, ' ');
+	}
+	sq_append_quote_argv_pretty(&buf_payload, argv);
+	normal_io_write_fl(file, line, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_exec_result_fl(const char *file, int line,
+			      uint64_t us_elapsed_absolute, int exec_id,
+			      int code)
+{
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	strbuf_addf(&buf_payload, "exec_result[%d] code:%d", exec_id, code);
+	if (code > 0)
+		strbuf_addf(&buf_payload, " err:%s", strerror(code));
+	normal_io_write_fl(file, line, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_param_fl(const char *file, int line, const char *param,
+			const char *value)
+{
+	struct strbuf buf_payload = STRBUF_INIT;
+	enum config_scope scope = current_config_scope();
+	const char *scope_name = config_scope_name(scope);
+
+	strbuf_addf(&buf_payload, "def_param scope:%s %s=%s", scope_name, param,
+		    value);
+	normal_io_write_fl(file, line, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_repo_fl(const char *file, int line,
+		       const struct repository *repo)
+{
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	strbuf_addstr(&buf_payload, "worktree ");
+	sq_quote_buf_pretty(&buf_payload, repo->worktree);
+	normal_io_write_fl(file, line, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_printf_va_fl(const char *file, int line,
+			    uint64_t us_elapsed_absolute, const char *fmt,
+			    va_list ap)
+{
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	maybe_append_string_va(&buf_payload, fmt, ap);
+	normal_io_write_fl(file, line, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_timer(const struct tr2_timer_metadata *meta,
+		     const struct tr2_timer *timer,
+		     int is_final_data)
+{
+	const char *event_name = is_final_data ? "timer" : "th_timer";
+	struct strbuf buf_payload = STRBUF_INIT;
+	double t_total = NS_TO_SEC(timer->total_ns);
+	double t_min = NS_TO_SEC(timer->min_ns);
+	double t_max = NS_TO_SEC(timer->max_ns);
+
+	strbuf_addf(&buf_payload, ("%s %s/%s"
+				   " intervals:%"PRIu64
+				   " total:%8.6f min:%8.6f max:%8.6f"),
+		    event_name, meta->category, meta->name,
+		    timer->interval_count,
+		    t_total, t_min, t_max);
+
+	normal_io_write_fl(__FILE__, __LINE__, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_counter(const struct tr2_counter_metadata *meta,
+		       const struct tr2_counter *counter,
+		       int is_final_data)
+{
+	const char *event_name = is_final_data ? "counter" : "th_counter";
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	strbuf_addf(&buf_payload, "%s %s/%s value:%"PRIu64,
+		    event_name, meta->category, meta->name,
+		    counter->value);
+
+	normal_io_write_fl(__FILE__, __LINE__, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+struct tr2_tgt tr2_tgt_normal = {
+	.pdst = &tr2dst_normal,
+
+	.pfn_init = fn_init,
+	.pfn_term = fn_term,
+
+	.pfn_version_fl = fn_version_fl,
+	.pfn_start_fl = fn_start_fl,
+	.pfn_exit_fl = fn_exit_fl,
+	.pfn_signal = fn_signal,
+	.pfn_atexit = fn_atexit,
+	.pfn_error_va_fl = fn_error_va_fl,
+	.pfn_command_path_fl = fn_command_path_fl,
+	.pfn_command_ancestry_fl = fn_command_ancestry_fl,
+	.pfn_command_name_fl = fn_command_name_fl,
+	.pfn_command_mode_fl = fn_command_mode_fl,
+	.pfn_alias_fl = fn_alias_fl,
+	.pfn_child_start_fl = fn_child_start_fl,
+	.pfn_child_exit_fl = fn_child_exit_fl,
+	.pfn_child_ready_fl = fn_child_ready_fl,
+	.pfn_thread_start_fl = NULL,
+	.pfn_thread_exit_fl = NULL,
+	.pfn_exec_fl = fn_exec_fl,
+	.pfn_exec_result_fl = fn_exec_result_fl,
+	.pfn_param_fl = fn_param_fl,
+	.pfn_repo_fl = fn_repo_fl,
+	.pfn_region_enter_printf_va_fl = NULL,
+	.pfn_region_leave_printf_va_fl = NULL,
+	.pfn_data_fl = NULL,
+	.pfn_data_json_fl = NULL,
+	.pfn_printf_va_fl = fn_printf_va_fl,
+	.pfn_timer = fn_timer,
+	.pfn_counter = fn_counter,
+};
diff --git a/trace2/tr2_tgt_perf.c b/trace2/tr2_tgt_perf.c
new file mode 100644
index 0000000..adae803
--- /dev/null
+++ b/trace2/tr2_tgt_perf.c
@@ -0,0 +1,630 @@
+#include "cache.h"
+#include "config.h"
+#include "run-command.h"
+#include "quote.h"
+#include "version.h"
+#include "json-writer.h"
+#include "trace2/tr2_dst.h"
+#include "trace2/tr2_sid.h"
+#include "trace2/tr2_sysenv.h"
+#include "trace2/tr2_tbuf.h"
+#include "trace2/tr2_tgt.h"
+#include "trace2/tr2_tls.h"
+#include "trace2/tr2_tmr.h"
+
+static struct tr2_dst tr2dst_perf = {
+	.sysenv_var = TR2_SYSENV_PERF,
+};
+
+/*
+ * Use TR2_SYSENV_PERF_BRIEF to omit the "<time> <file>:<line>"
+ * fields from each line written to the builtin performance target.
+ *
+ * Unit tests may want to use this to help with testing.
+ */
+static int tr2env_perf_be_brief;
+
+#define TR2FMT_PERF_FL_WIDTH (28)
+#define TR2FMT_PERF_MAX_EVENT_NAME (12)
+#define TR2FMT_PERF_REPO_WIDTH (3)
+#define TR2FMT_PERF_CATEGORY_WIDTH (12)
+
+#define TR2_INDENT (2)
+#define TR2_INDENT_LENGTH(ctx) (((ctx)->nr_open_regions - 1) * TR2_INDENT)
+
+static int fn_init(void)
+{
+	int want = tr2_dst_trace_want(&tr2dst_perf);
+	int want_brief;
+	const char *brief;
+
+	if (!want)
+		return want;
+
+	brief = tr2_sysenv_get(TR2_SYSENV_PERF_BRIEF);
+	if (brief && *brief &&
+	    ((want_brief = git_parse_maybe_bool(brief)) != -1))
+		tr2env_perf_be_brief = want_brief;
+
+	return want;
+}
+
+static void fn_term(void)
+{
+	tr2_dst_trace_disable(&tr2dst_perf);
+}
+
+/*
+ * Format trace line prefix in human-readable classic format for
+ * the performance target:
+ *     "[<time> [<file>:<line>] <bar>] <nr_parents> <bar>
+ *         <thread_name> <bar> <event_name> <bar> [<repo>] <bar>
+ *         [<elapsed_absolute>] [<elapsed_relative>] <bar>
+ *         [<category>] <bar> [<dots>] "
+ */
+static void perf_fmt_prepare(const char *event_name,
+			     struct tr2tls_thread_ctx *ctx, const char *file,
+			     int line, const struct repository *repo,
+			     uint64_t *p_us_elapsed_absolute,
+			     uint64_t *p_us_elapsed_relative,
+			     const char *category, struct strbuf *buf)
+{
+	int len;
+
+	strbuf_setlen(buf, 0);
+
+	if (!tr2env_perf_be_brief) {
+		struct tr2_tbuf tb_now;
+		size_t fl_end_col;
+
+		tr2_tbuf_local_time(&tb_now);
+		strbuf_addstr(buf, tb_now.buf);
+		strbuf_addch(buf, ' ');
+
+		fl_end_col = buf->len + TR2FMT_PERF_FL_WIDTH;
+
+		if (file && *file) {
+			struct strbuf buf_fl = STRBUF_INIT;
+
+			strbuf_addf(&buf_fl, "%s:%d", file, line);
+
+			if (buf_fl.len <= TR2FMT_PERF_FL_WIDTH)
+				strbuf_addbuf(buf, &buf_fl);
+			else {
+				size_t avail = TR2FMT_PERF_FL_WIDTH - 3;
+				strbuf_addstr(buf, "...");
+				strbuf_add(buf,
+					   &buf_fl.buf[buf_fl.len - avail],
+					   avail);
+			}
+
+			strbuf_release(&buf_fl);
+		}
+
+		while (buf->len < fl_end_col)
+			strbuf_addch(buf, ' ');
+
+		strbuf_addstr(buf, " | ");
+	}
+
+	strbuf_addf(buf, "d%d | ", tr2_sid_depth());
+	strbuf_addf(buf, "%-*s | %-*s | ", TR2_MAX_THREAD_NAME,
+		    ctx->thread_name, TR2FMT_PERF_MAX_EVENT_NAME,
+		    event_name);
+
+	len = buf->len + TR2FMT_PERF_REPO_WIDTH;
+	if (repo)
+		strbuf_addf(buf, "r%d ", repo->trace2_repo_id);
+	while (buf->len < len)
+		strbuf_addch(buf, ' ');
+	strbuf_addstr(buf, " | ");
+
+	if (p_us_elapsed_absolute)
+		strbuf_addf(buf, "%9.6f | ",
+			    ((double)(*p_us_elapsed_absolute)) / 1000000.0);
+	else
+		strbuf_addf(buf, "%9s | ", " ");
+
+	if (p_us_elapsed_relative)
+		strbuf_addf(buf, "%9.6f | ",
+			    ((double)(*p_us_elapsed_relative)) / 1000000.0);
+	else
+		strbuf_addf(buf, "%9s | ", " ");
+
+	strbuf_addf(buf, "%-*.*s | ", TR2FMT_PERF_CATEGORY_WIDTH,
+		    TR2FMT_PERF_CATEGORY_WIDTH, (category ? category : ""));
+
+	if (ctx->nr_open_regions > 0)
+		strbuf_addchars(buf, '.', TR2_INDENT_LENGTH(ctx));
+}
+
+static void perf_io_write_fl(const char *file, int line, const char *event_name,
+			     const struct repository *repo,
+			     uint64_t *p_us_elapsed_absolute,
+			     uint64_t *p_us_elapsed_relative,
+			     const char *category,
+			     const struct strbuf *buf_payload)
+{
+	struct tr2tls_thread_ctx *ctx = tr2tls_get_self();
+	struct strbuf buf_line = STRBUF_INIT;
+
+	perf_fmt_prepare(event_name, ctx, file, line, repo,
+			 p_us_elapsed_absolute, p_us_elapsed_relative, category,
+			 &buf_line);
+	strbuf_addbuf(&buf_line, buf_payload);
+	tr2_dst_write_line(&tr2dst_perf, &buf_line);
+	strbuf_release(&buf_line);
+}
+
+static void fn_version_fl(const char *file, int line)
+{
+	const char *event_name = "version";
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	strbuf_addstr(&buf_payload, git_version_string);
+
+	perf_io_write_fl(file, line, event_name, NULL, NULL, NULL, NULL,
+			 &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_start_fl(const char *file, int line,
+			uint64_t us_elapsed_absolute, const char **argv)
+{
+	const char *event_name = "start";
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	sq_append_quote_argv_pretty(&buf_payload, argv);
+
+	perf_io_write_fl(file, line, event_name, NULL, &us_elapsed_absolute,
+			 NULL, NULL, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_exit_fl(const char *file, int line, uint64_t us_elapsed_absolute,
+		       int code)
+{
+	const char *event_name = "exit";
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	strbuf_addf(&buf_payload, "code:%d", code);
+
+	perf_io_write_fl(file, line, event_name, NULL, &us_elapsed_absolute,
+			 NULL, NULL, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_signal(uint64_t us_elapsed_absolute, int signo)
+{
+	const char *event_name = "signal";
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	strbuf_addf(&buf_payload, "signo:%d", signo);
+
+	perf_io_write_fl(__FILE__, __LINE__, event_name, NULL,
+			 &us_elapsed_absolute, NULL, NULL, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_atexit(uint64_t us_elapsed_absolute, int code)
+{
+	const char *event_name = "atexit";
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	strbuf_addf(&buf_payload, "code:%d", code);
+
+	perf_io_write_fl(__FILE__, __LINE__, event_name, NULL,
+			 &us_elapsed_absolute, NULL, NULL, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void maybe_append_string_va(struct strbuf *buf, const char *fmt,
+				   va_list ap)
+{
+	if (fmt && *fmt) {
+		va_list copy_ap;
+
+		va_copy(copy_ap, ap);
+		strbuf_vaddf(buf, fmt, copy_ap);
+		va_end(copy_ap);
+		return;
+	}
+}
+
+static void fn_error_va_fl(const char *file, int line, const char *fmt,
+			   va_list ap)
+{
+	const char *event_name = "error";
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	maybe_append_string_va(&buf_payload, fmt, ap);
+
+	perf_io_write_fl(file, line, event_name, NULL, NULL, NULL, NULL,
+			 &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_command_path_fl(const char *file, int line, const char *pathname)
+{
+	const char *event_name = "cmd_path";
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	strbuf_addstr(&buf_payload, pathname);
+
+	perf_io_write_fl(file, line, event_name, NULL, NULL, NULL, NULL,
+			 &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_command_ancestry_fl(const char *file, int line, const char **parent_names)
+{
+	const char *event_name = "cmd_ancestry";
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	strbuf_addstr(&buf_payload, "ancestry:[");
+	/* It's not an argv but the rules are basically the same. */
+	sq_append_quote_argv_pretty(&buf_payload, parent_names);
+	strbuf_addch(&buf_payload, ']');
+
+	perf_io_write_fl(file, line, event_name, NULL, NULL, NULL, NULL,
+			 &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_command_name_fl(const char *file, int line, const char *name,
+			       const char *hierarchy)
+{
+	const char *event_name = "cmd_name";
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	strbuf_addstr(&buf_payload, name);
+	if (hierarchy && *hierarchy)
+		strbuf_addf(&buf_payload, " (%s)", hierarchy);
+
+	perf_io_write_fl(file, line, event_name, NULL, NULL, NULL, NULL,
+			 &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_command_mode_fl(const char *file, int line, const char *mode)
+{
+	const char *event_name = "cmd_mode";
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	strbuf_addstr(&buf_payload, mode);
+
+	perf_io_write_fl(file, line, event_name, NULL, NULL, NULL, NULL,
+			 &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_alias_fl(const char *file, int line, const char *alias,
+			const char **argv)
+{
+	const char *event_name = "alias";
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	strbuf_addf(&buf_payload, "alias:%s argv:[", alias);
+	sq_append_quote_argv_pretty(&buf_payload, argv);
+	strbuf_addch(&buf_payload, ']');
+
+	perf_io_write_fl(file, line, event_name, NULL, NULL, NULL, NULL,
+			 &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_child_start_fl(const char *file, int line,
+			      uint64_t us_elapsed_absolute,
+			      const struct child_process *cmd)
+{
+	const char *event_name = "child_start";
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	if (cmd->trace2_hook_name) {
+		strbuf_addf(&buf_payload, "[ch%d] class:hook hook:%s",
+			    cmd->trace2_child_id, cmd->trace2_hook_name);
+	} else {
+		const char *child_class =
+			cmd->trace2_child_class ? cmd->trace2_child_class : "?";
+		strbuf_addf(&buf_payload, "[ch%d] class:%s",
+			    cmd->trace2_child_id, child_class);
+	}
+
+	if (cmd->dir) {
+		strbuf_addstr(&buf_payload, " cd:");
+		sq_quote_buf_pretty(&buf_payload, cmd->dir);
+	}
+
+	strbuf_addstr(&buf_payload, " argv:[");
+	if (cmd->git_cmd) {
+		strbuf_addstr(&buf_payload, "git");
+		if (cmd->args.nr)
+			strbuf_addch(&buf_payload, ' ');
+	}
+	sq_append_quote_argv_pretty(&buf_payload, cmd->args.v);
+	strbuf_addch(&buf_payload, ']');
+
+	perf_io_write_fl(file, line, event_name, NULL, &us_elapsed_absolute,
+			 NULL, NULL, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_child_exit_fl(const char *file, int line,
+			     uint64_t us_elapsed_absolute, int cid, int pid,
+			     int code, uint64_t us_elapsed_child)
+{
+	const char *event_name = "child_exit";
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	strbuf_addf(&buf_payload, "[ch%d] pid:%d code:%d", cid, pid, code);
+
+	perf_io_write_fl(file, line, event_name, NULL, &us_elapsed_absolute,
+			 &us_elapsed_child, NULL, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_child_ready_fl(const char *file, int line,
+			      uint64_t us_elapsed_absolute, int cid, int pid,
+			      const char *ready, uint64_t us_elapsed_child)
+{
+	const char *event_name = "child_ready";
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	strbuf_addf(&buf_payload, "[ch%d] pid:%d ready:%s", cid, pid, ready);
+
+	perf_io_write_fl(file, line, event_name, NULL, &us_elapsed_absolute,
+			 &us_elapsed_child, NULL, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_thread_start_fl(const char *file, int line,
+			       uint64_t us_elapsed_absolute)
+{
+	const char *event_name = "thread_start";
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	perf_io_write_fl(file, line, event_name, NULL, &us_elapsed_absolute,
+			 NULL, NULL, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_thread_exit_fl(const char *file, int line,
+			      uint64_t us_elapsed_absolute,
+			      uint64_t us_elapsed_thread)
+{
+	const char *event_name = "thread_exit";
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	perf_io_write_fl(file, line, event_name, NULL, &us_elapsed_absolute,
+			 &us_elapsed_thread, NULL, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_exec_fl(const char *file, int line, uint64_t us_elapsed_absolute,
+		       int exec_id, const char *exe, const char **argv)
+{
+	const char *event_name = "exec";
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	strbuf_addf(&buf_payload, "id:%d ", exec_id);
+	strbuf_addstr(&buf_payload, "argv:[");
+	if (exe) {
+		strbuf_addstr(&buf_payload, exe);
+		if (argv[0])
+			strbuf_addch(&buf_payload, ' ');
+	}
+	sq_append_quote_argv_pretty(&buf_payload, argv);
+	strbuf_addch(&buf_payload, ']');
+
+	perf_io_write_fl(file, line, event_name, NULL, &us_elapsed_absolute,
+			 NULL, NULL, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_exec_result_fl(const char *file, int line,
+			      uint64_t us_elapsed_absolute, int exec_id,
+			      int code)
+{
+	const char *event_name = "exec_result";
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	strbuf_addf(&buf_payload, "id:%d code:%d", exec_id, code);
+	if (code > 0)
+		strbuf_addf(&buf_payload, " err:%s", strerror(code));
+
+	perf_io_write_fl(file, line, event_name, NULL, &us_elapsed_absolute,
+			 NULL, NULL, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_param_fl(const char *file, int line, const char *param,
+			const char *value)
+{
+	const char *event_name = "def_param";
+	struct strbuf buf_payload = STRBUF_INIT;
+	struct strbuf scope_payload = STRBUF_INIT;
+	enum config_scope scope = current_config_scope();
+	const char *scope_name = config_scope_name(scope);
+
+	strbuf_addf(&buf_payload, "%s:%s", param, value);
+	strbuf_addf(&scope_payload, "%s:%s", "scope", scope_name);
+
+	perf_io_write_fl(file, line, event_name, NULL, NULL, NULL,
+			 scope_payload.buf, &buf_payload);
+	strbuf_release(&buf_payload);
+	strbuf_release(&scope_payload);
+}
+
+static void fn_repo_fl(const char *file, int line,
+		       const struct repository *repo)
+{
+	const char *event_name = "def_repo";
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	strbuf_addstr(&buf_payload, "worktree:");
+	sq_quote_buf_pretty(&buf_payload, repo->worktree);
+
+	perf_io_write_fl(file, line, event_name, repo, NULL, NULL, NULL,
+			 &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_region_enter_printf_va_fl(const char *file, int line,
+					 uint64_t us_elapsed_absolute,
+					 const char *category,
+					 const char *label,
+					 const struct repository *repo,
+					 const char *fmt, va_list ap)
+{
+	const char *event_name = "region_enter";
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	if (label)
+		strbuf_addf(&buf_payload, "label:%s", label);
+	if (fmt && *fmt) {
+		strbuf_addch(&buf_payload, ' ');
+		maybe_append_string_va(&buf_payload, fmt, ap);
+	}
+
+	perf_io_write_fl(file, line, event_name, repo, &us_elapsed_absolute,
+			 NULL, category, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_region_leave_printf_va_fl(
+	const char *file, int line, uint64_t us_elapsed_absolute,
+	uint64_t us_elapsed_region, const char *category, const char *label,
+	const struct repository *repo, const char *fmt, va_list ap)
+{
+	const char *event_name = "region_leave";
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	if (label)
+		strbuf_addf(&buf_payload, "label:%s", label);
+	if (fmt && *fmt) {
+		strbuf_addch(&buf_payload, ' ' );
+		maybe_append_string_va(&buf_payload, fmt, ap);
+	}
+
+	perf_io_write_fl(file, line, event_name, repo, &us_elapsed_absolute,
+			 &us_elapsed_region, category, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_data_fl(const char *file, int line, uint64_t us_elapsed_absolute,
+		       uint64_t us_elapsed_region, const char *category,
+		       const struct repository *repo, const char *key,
+		       const char *value)
+{
+	const char *event_name = "data";
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	strbuf_addf(&buf_payload, "%s:%s", key, value);
+
+	perf_io_write_fl(file, line, event_name, repo, &us_elapsed_absolute,
+			 &us_elapsed_region, category, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_data_json_fl(const char *file, int line,
+			    uint64_t us_elapsed_absolute,
+			    uint64_t us_elapsed_region, const char *category,
+			    const struct repository *repo, const char *key,
+			    const struct json_writer *value)
+{
+	const char *event_name = "data_json";
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	strbuf_addf(&buf_payload, "%s:%s", key, value->json.buf);
+
+	perf_io_write_fl(file, line, event_name, repo, &us_elapsed_absolute,
+			 &us_elapsed_region, category, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_printf_va_fl(const char *file, int line,
+			    uint64_t us_elapsed_absolute, const char *fmt,
+			    va_list ap)
+{
+	const char *event_name = "printf";
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	maybe_append_string_va(&buf_payload, fmt, ap);
+
+	perf_io_write_fl(file, line, event_name, NULL, &us_elapsed_absolute,
+			 NULL, NULL, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_timer(const struct tr2_timer_metadata *meta,
+		     const struct tr2_timer *timer,
+		     int is_final_data)
+{
+	const char *event_name = is_final_data ? "timer" : "th_timer";
+	struct strbuf buf_payload = STRBUF_INIT;
+	double t_total = NS_TO_SEC(timer->total_ns);
+	double t_min = NS_TO_SEC(timer->min_ns);
+	double t_max = NS_TO_SEC(timer->max_ns);
+
+	strbuf_addf(&buf_payload, ("name:%s"
+				   " intervals:%"PRIu64
+				   " total:%8.6f min:%8.6f max:%8.6f"),
+		    meta->name,
+		    timer->interval_count,
+		    t_total, t_min, t_max);
+
+	perf_io_write_fl(__FILE__, __LINE__, event_name, NULL, NULL, NULL,
+			 meta->category, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+static void fn_counter(const struct tr2_counter_metadata *meta,
+		       const struct tr2_counter *counter,
+		       int is_final_data)
+{
+	const char *event_name = is_final_data ? "counter" : "th_counter";
+	struct strbuf buf_payload = STRBUF_INIT;
+
+	strbuf_addf(&buf_payload, "name:%s value:%"PRIu64,
+		    meta->name,
+		    counter->value);
+
+	perf_io_write_fl(__FILE__, __LINE__, event_name, NULL, NULL, NULL,
+			 meta->category, &buf_payload);
+	strbuf_release(&buf_payload);
+}
+
+struct tr2_tgt tr2_tgt_perf = {
+	.pdst = &tr2dst_perf,
+
+	.pfn_init = fn_init,
+	.pfn_term = fn_term,
+
+	.pfn_version_fl = fn_version_fl,
+	.pfn_start_fl = fn_start_fl,
+	.pfn_exit_fl = fn_exit_fl,
+	.pfn_signal = fn_signal,
+	.pfn_atexit = fn_atexit,
+	.pfn_error_va_fl = fn_error_va_fl,
+	.pfn_command_path_fl = fn_command_path_fl,
+	.pfn_command_ancestry_fl = fn_command_ancestry_fl,
+	.pfn_command_name_fl = fn_command_name_fl,
+	.pfn_command_mode_fl = fn_command_mode_fl,
+	.pfn_alias_fl = fn_alias_fl,
+	.pfn_child_start_fl = fn_child_start_fl,
+	.pfn_child_exit_fl = fn_child_exit_fl,
+	.pfn_child_ready_fl = fn_child_ready_fl,
+	.pfn_thread_start_fl = fn_thread_start_fl,
+	.pfn_thread_exit_fl = fn_thread_exit_fl,
+	.pfn_exec_fl = fn_exec_fl,
+	.pfn_exec_result_fl = fn_exec_result_fl,
+	.pfn_param_fl = fn_param_fl,
+	.pfn_repo_fl = fn_repo_fl,
+	.pfn_region_enter_printf_va_fl = fn_region_enter_printf_va_fl,
+	.pfn_region_leave_printf_va_fl = fn_region_leave_printf_va_fl,
+	.pfn_data_fl = fn_data_fl,
+	.pfn_data_json_fl = fn_data_json_fl,
+	.pfn_printf_va_fl = fn_printf_va_fl,
+	.pfn_timer = fn_timer,
+	.pfn_counter = fn_counter,
+};
diff --git a/trace2/tr2_tls.c b/trace2/tr2_tls.c
new file mode 100644
index 0000000..04900bb
--- /dev/null
+++ b/trace2/tr2_tls.c
@@ -0,0 +1,193 @@
+#include "cache.h"
+#include "thread-utils.h"
+#include "trace2/tr2_tls.h"
+
+/*
+ * Initialize size of the thread stack for nested regions.
+ * This is used to store nested region start times.  Note that
+ * this stack is per-thread and not per-trace-key.
+ */
+#define TR2_REGION_NESTING_INITIAL_SIZE (100)
+
+static struct tr2tls_thread_ctx *tr2tls_thread_main;
+static uint64_t tr2tls_us_start_process;
+
+static pthread_mutex_t tr2tls_mutex;
+static pthread_key_t tr2tls_key;
+
+static int tr2_next_thread_id; /* modify under lock */
+
+void tr2tls_start_process_clock(void)
+{
+	if (tr2tls_us_start_process)
+		return;
+
+	/*
+	 * Keep the absolute start time of the process (i.e. the main
+	 * process) in a fixed variable since other threads need to
+	 * access it.  This allows them to do that without a lock on
+	 * main thread's array data (because of reallocs).
+	 */
+	tr2tls_us_start_process = getnanotime() / 1000;
+}
+
+struct tr2tls_thread_ctx *tr2tls_create_self(const char *thread_base_name,
+					     uint64_t us_thread_start)
+{
+	struct tr2tls_thread_ctx *ctx = xcalloc(1, sizeof(*ctx));
+	struct strbuf buf = STRBUF_INIT;
+
+	/*
+	 * Implicitly "tr2tls_push_self()" to capture the thread's start
+	 * time in array_us_start[0].  For the main thread this gives us the
+	 * application run time.
+	 */
+	ctx->alloc = TR2_REGION_NESTING_INITIAL_SIZE;
+	ctx->array_us_start = (uint64_t *)xcalloc(ctx->alloc, sizeof(uint64_t));
+	ctx->array_us_start[ctx->nr_open_regions++] = us_thread_start;
+
+	ctx->thread_id = tr2tls_locked_increment(&tr2_next_thread_id);
+
+	strbuf_init(&buf, 0);
+	if (ctx->thread_id)
+		strbuf_addf(&buf, "th%02d:", ctx->thread_id);
+	strbuf_addstr(&buf, thread_base_name);
+	if (buf.len > TR2_MAX_THREAD_NAME)
+		strbuf_setlen(&buf, TR2_MAX_THREAD_NAME);
+	ctx->thread_name = strbuf_detach(&buf, NULL);
+
+	pthread_setspecific(tr2tls_key, ctx);
+
+	return ctx;
+}
+
+struct tr2tls_thread_ctx *tr2tls_get_self(void)
+{
+	struct tr2tls_thread_ctx *ctx;
+
+	if (!HAVE_THREADS)
+		return tr2tls_thread_main;
+
+	ctx = pthread_getspecific(tr2tls_key);
+
+	/*
+	 * If the current thread's thread-proc did not call
+	 * trace2_thread_start(), then the thread will not have any
+	 * thread-local storage.  Create it now and silently continue.
+	 */
+	if (!ctx)
+		ctx = tr2tls_create_self("unknown", getnanotime() / 1000);
+
+	return ctx;
+}
+
+int tr2tls_is_main_thread(void)
+{
+	if (!HAVE_THREADS)
+		return 1;
+
+	return pthread_getspecific(tr2tls_key) == tr2tls_thread_main;
+}
+
+void tr2tls_unset_self(void)
+{
+	struct tr2tls_thread_ctx *ctx;
+
+	ctx = tr2tls_get_self();
+
+	pthread_setspecific(tr2tls_key, NULL);
+
+	free((char *)ctx->thread_name);
+	free(ctx->array_us_start);
+	free(ctx);
+}
+
+void tr2tls_push_self(uint64_t us_now)
+{
+	struct tr2tls_thread_ctx *ctx = tr2tls_get_self();
+
+	ALLOC_GROW(ctx->array_us_start, ctx->nr_open_regions + 1, ctx->alloc);
+	ctx->array_us_start[ctx->nr_open_regions++] = us_now;
+}
+
+void tr2tls_pop_self(void)
+{
+	struct tr2tls_thread_ctx *ctx = tr2tls_get_self();
+
+	if (!ctx->nr_open_regions)
+		BUG("no open regions in thread '%s'", ctx->thread_name);
+
+	ctx->nr_open_regions--;
+}
+
+void tr2tls_pop_unwind_self(void)
+{
+	struct tr2tls_thread_ctx *ctx = tr2tls_get_self();
+
+	while (ctx->nr_open_regions > 1)
+		tr2tls_pop_self();
+}
+
+uint64_t tr2tls_region_elasped_self(uint64_t us)
+{
+	struct tr2tls_thread_ctx *ctx;
+	uint64_t us_start;
+
+	ctx = tr2tls_get_self();
+	if (!ctx->nr_open_regions)
+		return 0;
+
+	us_start = ctx->array_us_start[ctx->nr_open_regions - 1];
+
+	return us - us_start;
+}
+
+uint64_t tr2tls_absolute_elapsed(uint64_t us)
+{
+	if (!tr2tls_thread_main)
+		return 0;
+
+	return us - tr2tls_us_start_process;
+}
+
+void tr2tls_init(void)
+{
+	tr2tls_start_process_clock();
+
+	pthread_key_create(&tr2tls_key, NULL);
+	init_recursive_mutex(&tr2tls_mutex);
+
+	tr2tls_thread_main =
+		tr2tls_create_self("main", tr2tls_us_start_process);
+}
+
+void tr2tls_release(void)
+{
+	tr2tls_unset_self();
+	tr2tls_thread_main = NULL;
+
+	pthread_mutex_destroy(&tr2tls_mutex);
+	pthread_key_delete(tr2tls_key);
+}
+
+int tr2tls_locked_increment(int *p)
+{
+	int current_value;
+
+	pthread_mutex_lock(&tr2tls_mutex);
+	current_value = *p;
+	*p = current_value + 1;
+	pthread_mutex_unlock(&tr2tls_mutex);
+
+	return current_value;
+}
+
+void tr2tls_lock(void)
+{
+	pthread_mutex_lock(&tr2tls_mutex);
+}
+
+void tr2tls_unlock(void)
+{
+	pthread_mutex_unlock(&tr2tls_mutex);
+}
diff --git a/trace2/tr2_tls.h b/trace2/tr2_tls.h
new file mode 100644
index 0000000..f904980
--- /dev/null
+++ b/trace2/tr2_tls.h
@@ -0,0 +1,124 @@
+#ifndef TR2_TLS_H
+#define TR2_TLS_H
+
+#include "strbuf.h"
+#include "trace2/tr2_ctr.h"
+#include "trace2/tr2_tmr.h"
+
+/*
+ * Notice: the term "TLS" refers to "thread-local storage" in the
+ * Trace2 source files.  This usage is borrowed from GCC and Windows.
+ * There is NO relation to "transport layer security".
+ */
+
+/*
+ * Arbitry limit for thread names for column alignment.
+ */
+#define TR2_MAX_THREAD_NAME (24)
+
+struct tr2tls_thread_ctx {
+	const char *thread_name;
+	uint64_t *array_us_start;
+	size_t alloc;
+	size_t nr_open_regions; /* plays role of "nr" in ALLOC_GROW */
+	int thread_id;
+	struct tr2_timer_block timer_block;
+	struct tr2_counter_block counter_block;
+	unsigned int used_any_timer:1;
+	unsigned int used_any_per_thread_timer:1;
+	unsigned int used_any_counter:1;
+	unsigned int used_any_per_thread_counter:1;
+};
+
+/*
+ * Create thread-local storage for the current thread.
+ *
+ * The first thread in the process will have:
+ *     { .thread_id=0, .thread_name="main" }
+ * Subsequent threads are given a non-zero thread_id and a thread_name
+ * constructed from the id and a thread base name (which is usually just
+ * the name of the thread-proc function).  For example:
+ *     { .thread_id=10, .thread_name="th10:fsm-listen" }
+ * This helps to identify and distinguish messages from concurrent threads.
+ * The ctx.thread_name field is truncated if necessary to help with column
+ * alignment in printf-style messages.
+ *
+ * In this and all following functions the term "self" refers to the
+ * current thread.
+ */
+struct tr2tls_thread_ctx *tr2tls_create_self(const char *thread_base_name,
+					     uint64_t us_thread_start);
+
+/*
+ * Get the thread-local storage pointer of the current thread.
+ */
+struct tr2tls_thread_ctx *tr2tls_get_self(void);
+
+/*
+ * return true if the current thread is the main thread.
+ */
+int tr2tls_is_main_thread(void);
+
+/*
+ * Free the current thread's thread-local storage.
+ */
+void tr2tls_unset_self(void);
+
+/*
+ * Begin a new nested region and remember the start time.
+ */
+void tr2tls_push_self(uint64_t us_now);
+
+/*
+ * End the innermost nested region.
+ */
+void tr2tls_pop_self(void);
+
+/*
+ * Pop any extra (above the first) open regions on the current
+ * thread and discard.  During a thread-exit, we should only
+ * have region[0] that was pushed in trace2_thread_start() if
+ * the thread exits normally.
+ */
+void tr2tls_pop_unwind_self(void);
+
+/*
+ * Compute the elapsed time since the innermost region in the
+ * current thread started and the given time (usually now).
+ */
+uint64_t tr2tls_region_elasped_self(uint64_t us);
+
+/*
+ * Compute the elapsed time since the main thread started
+ * and the given time (usually now).  This is assumed to
+ * be the absolute run time of the process.
+ */
+uint64_t tr2tls_absolute_elapsed(uint64_t us);
+
+/*
+ * Initialize thread-local storage for Trace2.
+ */
+void tr2tls_init(void);
+
+/*
+ * Free all Trace2 thread-local storage resources.
+ */
+void tr2tls_release(void);
+
+/*
+ * Protected increment of an integer.
+ */
+int tr2tls_locked_increment(int *p);
+
+/*
+ * Capture the process start time and do nothing else.
+ */
+void tr2tls_start_process_clock(void);
+
+/*
+ * Explicitly lock/unlock our mutex.
+ */
+void tr2tls_lock(void);
+void tr2tls_unlock(void);
+
+#endif /* TR2_TLS_H */
diff --git a/trace2/tr2_tmr.c b/trace2/tr2_tmr.c
new file mode 100644
index 0000000..786762d
--- /dev/null
+++ b/trace2/tr2_tmr.c
@@ -0,0 +1,182 @@
+#include "cache.h"
+#include "thread-utils.h"
+#include "trace2/tr2_tgt.h"
+#include "trace2/tr2_tls.h"
+#include "trace2/tr2_tmr.h"
+
+#define MY_MAX(a, b) ((a) > (b) ? (a) : (b))
+#define MY_MIN(a, b) ((a) < (b) ? (a) : (b))
+
+/*
+ * A global timer block to aggregate values from the partial sums from
+ * each thread.
+ */
+static struct tr2_timer_block final_timer_block; /* access under tr2tls_mutex */
+
+/*
+ * Define metadata for each stopwatch timer.
+ *
+ * This array must match "enum trace2_timer_id" and the values
+ * in "struct tr2_timer_block.timer[*]".
+ */
+static struct tr2_timer_metadata tr2_timer_metadata[TRACE2_NUMBER_OF_TIMERS] = {
+	[TRACE2_TIMER_ID_TEST1] = {
+		.category = "test",
+		.name = "test1",
+		.want_per_thread_events = 0,
+	},
+	[TRACE2_TIMER_ID_TEST2] = {
+		.category = "test",
+		.name = "test2",
+		.want_per_thread_events = 1,
+	},
+
+	/* Add additional metadata before here. */
+};
+
+void tr2_start_timer(enum trace2_timer_id tid)
+{
+	struct tr2tls_thread_ctx *ctx = tr2tls_get_self();
+	struct tr2_timer *t = &ctx->timer_block.timer[tid];
+
+	t->recursion_count++;
+	if (t->recursion_count > 1)
+		return; /* ignore recursive starts */
+
+	t->start_ns = getnanotime();
+}
+
+void tr2_stop_timer(enum trace2_timer_id tid)
+{
+	struct tr2tls_thread_ctx *ctx = tr2tls_get_self();
+	struct tr2_timer *t = &ctx->timer_block.timer[tid];
+	uint64_t ns_now;
+	uint64_t ns_interval;
+
+	assert(t->recursion_count > 0);
+
+	t->recursion_count--;
+	if (t->recursion_count)
+		return; /* still in recursive call(s) */
+
+	ns_now = getnanotime();
+	ns_interval = ns_now - t->start_ns;
+
+	t->total_ns += ns_interval;
+
+	/*
+	 * min_ns was initialized to zero (in the xcalloc()) rather
+	 * than UINT_MAX when the block of timers was allocated,
+	 * so we should always set both the min_ns and max_ns values
+	 * the first time that the timer is used.
+	 */
+	if (!t->interval_count) {
+		t->min_ns = ns_interval;
+		t->max_ns = ns_interval;
+	} else {
+		t->min_ns = MY_MIN(ns_interval, t->min_ns);
+		t->max_ns = MY_MAX(ns_interval, t->max_ns);
+	}
+
+	t->interval_count++;
+
+	ctx->used_any_timer = 1;
+	if (tr2_timer_metadata[tid].want_per_thread_events)
+		ctx->used_any_per_thread_timer = 1;
+}
+
+void tr2_update_final_timers(void)
+{
+	struct tr2tls_thread_ctx *ctx = tr2tls_get_self();
+	enum trace2_timer_id tid;
+
+	if (!ctx->used_any_timer)
+		return;
+
+	/*
+	 * Accessing `final_timer_block` requires holding `tr2tls_mutex`.
+	 * We assume that our caller is holding the lock.
+	 */
+
+	for (tid = 0; tid < TRACE2_NUMBER_OF_TIMERS; tid++) {
+		struct tr2_timer *t_final = &final_timer_block.timer[tid];
+		struct tr2_timer *t = &ctx->timer_block.timer[tid];
+
+		if (t->recursion_count) {
+			/*
+			 * The current thread is exiting with
+			 * timer[tid] still running.
+			 *
+			 * Technically, this is a bug, but I'm going
+			 * to ignore it.
+			 *
+			 * I don't think it is worth calling die()
+			 * for.  I don't think it is worth killing the
+			 * process for this bookkeeping error.  We
+			 * might want to call warning(), but I'm going
+			 * to wait on that.
+			 *
+			 * The downside here is that total_ns won't
+			 * include the current open interval (now -
+			 * start_ns).  I can live with that.
+			 */
+		}
+
+		if (!t->interval_count)
+			continue; /* this timer was not used by this thread */
+
+		t_final->total_ns += t->total_ns;
+
+		/*
+		 * final_timer_block.timer[tid].min_ns was initialized to
+		 * was initialized to zero rather than UINT_MAX, so we should
+		 * always set both the min_ns and max_ns values the first time
+		 * that we add a partial sum into it.
+		 */
+		if (!t_final->interval_count) {
+			t_final->min_ns = t->min_ns;
+			t_final->max_ns = t->max_ns;
+		} else {
+			t_final->min_ns = MY_MIN(t_final->min_ns, t->min_ns);
+			t_final->max_ns = MY_MAX(t_final->max_ns, t->max_ns);
+		}
+
+		t_final->interval_count += t->interval_count;
+	}
+}
+
+void tr2_emit_per_thread_timers(tr2_tgt_evt_timer_t *fn_apply)
+{
+	struct tr2tls_thread_ctx *ctx = tr2tls_get_self();
+	enum trace2_timer_id tid;
+
+	if (!ctx->used_any_per_thread_timer)
+		return;
+
+	/*
+	 * For each timer, if the timer wants per-thread events and
+	 * this thread used it, emit it.
+	 */
+	for (tid = 0; tid < TRACE2_NUMBER_OF_TIMERS; tid++)
+		if (tr2_timer_metadata[tid].want_per_thread_events &&
+		    ctx->timer_block.timer[tid].interval_count)
+			fn_apply(&tr2_timer_metadata[tid],
+				 &ctx->timer_block.timer[tid],
+				 0);
+}
+
+void tr2_emit_final_timers(tr2_tgt_evt_timer_t *fn_apply)
+{
+	enum trace2_timer_id tid;
+
+	/*
+	 * Accessing `final_timer_block` requires holding `tr2tls_mutex`.
+	 * We assume that our caller is holding the lock.
+	 */
+
+	for (tid = 0; tid < TRACE2_NUMBER_OF_TIMERS; tid++)
+		if (final_timer_block.timer[tid].interval_count)
+			fn_apply(&tr2_timer_metadata[tid],
+				 &final_timer_block.timer[tid],
+				 1);
+}
diff --git a/trace2/tr2_tmr.h b/trace2/tr2_tmr.h
new file mode 100644
index 0000000..d575357
--- /dev/null
+++ b/trace2/tr2_tmr.h
@@ -0,0 +1,140 @@
+#ifndef TR2_TMR_H
+#define TR2_TMR_H
+
+#include "trace2.h"
+#include "trace2/tr2_tgt.h"
+
+/*
+ * Define a mechanism to allow "stopwatch" timers.
+ *
+ * Timers can be used to measure "interesting" activity that does not
+ * fit the "region" model, such as code called from many different
+ * regions (like zlib) and/or where data for individual calls are not
+ * interesting or are too numerous to be efficiently logged.
+ *
+ * Timer values are accumulated during program execution and emitted
+ * to the Trace2 logs at program exit.
+ *
+ * To make this model efficient, we define a compile-time fixed set of
+ * timers and timer ids using a "timer block" array in thread-local
+ * storage.  This gives us constant time access to each timer within
+ * each thread, since we want start/stop operations to be as fast as
+ * possible.  This lets us avoid the complexities of dynamically
+ * allocating a timer on the first use by a thread and/or possibly
+ * sharing that timer definition with other concurrent threads.
+ * However, this does require that we define time the set of timers at
+ * compile time.
+ *
+ * Each thread uses the timer block in its thread-local storage to
+ * compute partial sums for each timer (without locking).  When a
+ * thread exits, those partial sums are (under lock) added to the
+ * global final sum.
+ *
+ * Using this "timer block" model costs ~48 bytes per timer per thread
+ * (we have about six uint64 fields per timer).  This does increase
+ * the size of the thread-local storage block, but it is allocated (at
+ * thread create time) and not on the thread stack, so I'm not worried
+ * about the size.
+ *
+ * Partial sums for each timer are optionally emitted when a thread
+ * exits.
+ *
+ * Final sums for each timer are emitted between the "exit" and
+ * "atexit" events.
+ *
+ * A parallel "timer metadata" table contains the "category" and "name"
+ * fields for each timer.  This eliminates the need to include those
+ * args in the various timer APIs.
+ */
+
+/*
+ * The definition of an individual timer and used by an individual
+ * thread.
+ */
+struct tr2_timer {
+	/*
+	 * Total elapsed time for this timer in this thread in nanoseconds.
+	 */
+	uint64_t total_ns;
+
+	/*
+	 * The maximum and minimum interval values observed for this
+	 * timer in this thread.
+	 */
+	uint64_t min_ns;
+	uint64_t max_ns;
+
+	/*
+	 * The value of the clock when this timer was started in this
+	 * thread.  (Undefined when the timer is not active in this
+	 * thread.)
+	 */
+	uint64_t start_ns;
+
+	/*
+	 * Number of times that this timer has been started and stopped
+	 * in this thread.  (Recursive starts are ignored.)
+	 */
+	uint64_t interval_count;
+
+	/*
+	 * Number of nested starts on the stack in this thread.  (We
+	 * ignore recursive starts and use this to track the recursive
+	 * calls.)
+	 */
+	unsigned int recursion_count;
+};
+
+/*
+ * Metadata for a timer.
+ */
+struct tr2_timer_metadata {
+	const char *category;
+	const char *name;
+
+	/*
+	 * True if we should emit per-thread events for this timer
+	 * when individual threads exit.
+	 */
+	unsigned int want_per_thread_events:1;
+};
+
+/*
+ * A compile-time fixed-size block of timers to insert into
+ * thread-local storage.  This wrapper is used to avoid quirks
+ * of C and the usual need to pass an array size argument.
+ */
+struct tr2_timer_block {
+	struct tr2_timer timer[TRACE2_NUMBER_OF_TIMERS];
+};
+
+/*
+ * Private routines used by trace2.c to actually start/stop an
+ * individual timer in the current thread.
+ */
+void tr2_start_timer(enum trace2_timer_id tid);
+void tr2_stop_timer(enum trace2_timer_id tid);
+
+/*
+ * Add the current thread's timer data to the global totals.
+ * This is called during thread-exit.
+ *
+ * Caller must be holding the tr2tls_mutex.
+ */
+void tr2_update_final_timers(void);
+
+/*
+ * Emit per-thread timer data for the current thread.
+ * This is called during thread-exit.
+ */
+void tr2_emit_per_thread_timers(tr2_tgt_evt_timer_t *fn_apply);
+
+/*
+ * Emit global total timer values.
+ * This is called during atexit handling.
+ *
+ * Caller must be holding the tr2tls_mutex.
+ */
+void tr2_emit_final_timers(tr2_tgt_evt_timer_t *fn_apply);
+
+#endif /* TR2_TMR_H */