1 files changed, 331 insertions, 23 deletions

diff --git a/tools/perf/util/intel-pt.c b/tools/perf/util/intel-pt.c
index 9276506d07f8e..8ed51f4e9e304 100644
--- a/tools/perf/util/intel-pt.c
+++ b/tools/perf/util/intel-pt.c
@@ -33,6 +33,7 @@
 #include "tsc.h"
 #include "intel-pt.h"
 #include "config.h"
+#include "time-utils.h"
 
 #include "intel-pt-decoder/intel-pt-log.h"
 #include "intel-pt-decoder/intel-pt-decoder.h"
@@ -41,6 +42,11 @@
 
 #define MAX_TIMESTAMP (~0ULL)
 
+struct range {
+	u64 start;
+	u64 end;
+};
+
 struct intel_pt {
 	struct auxtrace auxtrace;
 	struct auxtrace_queues queues;
@@ -109,6 +115,9 @@ struct intel_pt {
 
 	char *filter;
 	struct addr_filters filts;
+
+	struct range *time_ranges;
+	unsigned int range_cnt;
 };
 
 enum switch_state {
@@ -145,9 +154,18 @@ struct intel_pt_queue {
 	bool have_sample;
 	u64 time;
 	u64 timestamp;
+	u64 sel_timestamp;
+	bool sel_start;
+	unsigned int sel_idx;
 	u32 flags;
 	u16 insn_len;
 	u64 last_insn_cnt;
+	u64 ipc_insn_cnt;
+	u64 ipc_cyc_cnt;
+	u64 last_in_insn_cnt;
+	u64 last_in_cyc_cnt;
+	u64 last_br_insn_cnt;
+	u64 last_br_cyc_cnt;
 	char insn[INTEL_PT_INSN_BUF_SZ];
 };
 
@@ -224,32 +242,13 @@ static int intel_pt_do_fix_overlap(struct intel_pt *pt, struct auxtrace_buffer *
 	return 0;
 }
 
-/* This function assumes data is processed sequentially only */
-static int intel_pt_get_trace(struct intel_pt_buffer *b, void *data)
+static int intel_pt_get_buffer(struct intel_pt_queue *ptq,
+			       struct auxtrace_buffer *buffer,
+			       struct auxtrace_buffer *old_buffer,
+			       struct intel_pt_buffer *b)
 {
-	struct intel_pt_queue *ptq = data;
-	struct auxtrace_buffer *buffer = ptq->buffer;
-	struct auxtrace_buffer *old_buffer = ptq->old_buffer;
-	struct auxtrace_queue *queue;
 	bool might_overlap;
 
-	if (ptq->stop) {
-		b->len = 0;
-		return 0;
-	}
-
-	queue = &ptq->pt->queues.queue_array[ptq->queue_nr];
-
-	buffer = auxtrace_buffer__next(queue, buffer);
-	if (!buffer) {
-		if (old_buffer)
-			auxtrace_buffer__drop_data(old_buffer);
-		b->len = 0;
-		return 0;
-	}
-
-	ptq->buffer = buffer;
-
 	if (!buffer->data) {
 		int fd = perf_data__fd(ptq->pt->session->data);
 
@@ -279,6 +278,95 @@ static int intel_pt_get_trace(struct intel_pt_buffer *b, void *data)
 		b->consecutive = true;
 	}
 
+	return 0;
+}
+
+/* Do not drop buffers with references - refer intel_pt_get_trace() */
+static void intel_pt_lookahead_drop_buffer(struct intel_pt_queue *ptq,
+					   struct auxtrace_buffer *buffer)
+{
+	if (!buffer || buffer == ptq->buffer || buffer == ptq->old_buffer)
+		return;
+
+	auxtrace_buffer__drop_data(buffer);
+}
+
+/* Must be serialized with respect to intel_pt_get_trace() */
+static int intel_pt_lookahead(void *data, intel_pt_lookahead_cb_t cb,
+			      void *cb_data)
+{
+	struct intel_pt_queue *ptq = data;
+	struct auxtrace_buffer *buffer = ptq->buffer;
+	struct auxtrace_buffer *old_buffer = ptq->old_buffer;
+	struct auxtrace_queue *queue;
+	int err = 0;
+
+	queue = &ptq->pt->queues.queue_array[ptq->queue_nr];
+
+	while (1) {
+		struct intel_pt_buffer b = { .len = 0 };
+
+		buffer = auxtrace_buffer__next(queue, buffer);
+		if (!buffer)
+			break;
+
+		err = intel_pt_get_buffer(ptq, buffer, old_buffer, &b);
+		if (err)
+			break;
+
+		if (b.len) {
+			intel_pt_lookahead_drop_buffer(ptq, old_buffer);
+			old_buffer = buffer;
+		} else {
+			intel_pt_lookahead_drop_buffer(ptq, buffer);
+			continue;
+		}
+
+		err = cb(&b, cb_data);
+		if (err)
+			break;
+	}
+
+	if (buffer != old_buffer)
+		intel_pt_lookahead_drop_buffer(ptq, buffer);
+	intel_pt_lookahead_drop_buffer(ptq, old_buffer);
+
+	return err;
+}
+
+/*
+ * This function assumes data is processed sequentially only.
+ * Must be serialized with respect to intel_pt_lookahead()
+ */
+static int intel_pt_get_trace(struct intel_pt_buffer *b, void *data)
+{
+	struct intel_pt_queue *ptq = data;
+	struct auxtrace_buffer *buffer = ptq->buffer;
+	struct auxtrace_buffer *old_buffer = ptq->old_buffer;
+	struct auxtrace_queue *queue;
+	int err;
+
+	if (ptq->stop) {
+		b->len = 0;
+		return 0;
+	}
+
+	queue = &ptq->pt->queues.queue_array[ptq->queue_nr];
+
+	buffer = auxtrace_buffer__next(queue, buffer);
+	if (!buffer) {
+		if (old_buffer)
+			auxtrace_buffer__drop_data(old_buffer);
+		b->len = 0;
+		return 0;
+	}
+
+	ptq->buffer = buffer;
+
+	err = intel_pt_get_buffer(ptq, buffer, old_buffer, b);
+	if (err)
+		return err;
+
 	if (ptq->step_through_buffers)
 		ptq->stop = true;
 
@@ -798,6 +886,7 @@ static struct intel_pt_queue *intel_pt_alloc_queue(struct intel_pt *pt,
 
 	params.get_trace = intel_pt_get_trace;
 	params.walk_insn = intel_pt_walk_next_insn;
+	params.lookahead = intel_pt_lookahead;
 	params.data = ptq;
 	params.return_compression = intel_pt_return_compression(pt);
 	params.branch_enable = intel_pt_branch_enable(pt);
@@ -921,6 +1010,23 @@ static void intel_pt_sample_flags(struct intel_pt_queue *ptq)
 		ptq->flags |= PERF_IP_FLAG_TRACE_END;
 }
 
+static void intel_pt_setup_time_range(struct intel_pt *pt,
+				      struct intel_pt_queue *ptq)
+{
+	if (!pt->range_cnt)
+		return;
+
+	ptq->sel_timestamp = pt->time_ranges[0].start;
+	ptq->sel_idx = 0;
+
+	if (ptq->sel_timestamp) {
+		ptq->sel_start = true;
+	} else {
+		ptq->sel_timestamp = pt->time_ranges[0].end;
+		ptq->sel_start = false;
+	}
+}
+
 static int intel_pt_setup_queue(struct intel_pt *pt,
 				struct auxtrace_queue *queue,
 				unsigned int queue_nr)
@@ -945,6 +1051,8 @@ static int intel_pt_setup_queue(struct intel_pt *pt,
 			ptq->step_through_buffers = true;
 
 		ptq->sync_switch = pt->sync_switch;
+
+		intel_pt_setup_time_range(pt, ptq);
 	}
 
 	if (!ptq->on_heap &&
@@ -959,6 +1067,14 @@ static int intel_pt_setup_queue(struct intel_pt *pt,
 		intel_pt_log("queue %u getting timestamp\n", queue_nr);
 		intel_pt_log("queue %u decoding cpu %d pid %d tid %d\n",
 			     queue_nr, ptq->cpu, ptq->pid, ptq->tid);
+
+		if (ptq->sel_start && ptq->sel_timestamp) {
+			ret = intel_pt_fast_forward(ptq->decoder,
+						    ptq->sel_timestamp);
+			if (ret)
+				return ret;
+		}
+
 		while (1) {
 			state = intel_pt_decode(ptq->decoder);
 			if (state->err) {
@@ -978,6 +1094,9 @@ static int intel_pt_setup_queue(struct intel_pt *pt,
 			     queue_nr, ptq->timestamp);
 		ptq->state = state;
 		ptq->have_sample = true;
+		if (ptq->sel_start && ptq->sel_timestamp &&
+		    ptq->timestamp < ptq->sel_timestamp)
+			ptq->have_sample = false;
 		intel_pt_sample_flags(ptq);
 		ret = auxtrace_heap__add(&pt->heap, queue_nr, ptq->timestamp);
 		if (ret)
@@ -1153,6 +1272,13 @@ static int intel_pt_synth_branch_sample(struct intel_pt_queue *ptq)
 		sample.branch_stack = (struct branch_stack *)&dummy_bs;
 	}
 
+	sample.cyc_cnt = ptq->ipc_cyc_cnt - ptq->last_br_cyc_cnt;
+	if (sample.cyc_cnt) {
+		sample.insn_cnt = ptq->ipc_insn_cnt - ptq->last_br_insn_cnt;
+		ptq->last_br_insn_cnt = ptq->ipc_insn_cnt;
+		ptq->last_br_cyc_cnt = ptq->ipc_cyc_cnt;
+	}
+
 	return intel_pt_deliver_synth_b_event(pt, event, &sample,
 					      pt->branches_sample_type);
 }
@@ -1208,6 +1334,13 @@ static int intel_pt_synth_instruction_sample(struct intel_pt_queue *ptq)
 	sample.stream_id = ptq->pt->instructions_id;
 	sample.period = ptq->state->tot_insn_cnt - ptq->last_insn_cnt;
 
+	sample.cyc_cnt = ptq->ipc_cyc_cnt - ptq->last_in_cyc_cnt;
+	if (sample.cyc_cnt) {
+		sample.insn_cnt = ptq->ipc_insn_cnt - ptq->last_in_insn_cnt;
+		ptq->last_in_insn_cnt = ptq->ipc_insn_cnt;
+		ptq->last_in_cyc_cnt = ptq->ipc_cyc_cnt;
+	}
+
 	ptq->last_insn_cnt = ptq->state->tot_insn_cnt;
 
 	return intel_pt_deliver_synth_event(pt, ptq, event, &sample,
@@ -1479,6 +1612,15 @@ static int intel_pt_sample(struct intel_pt_queue *ptq)
 
 	ptq->have_sample = false;
 
+	if (ptq->state->tot_cyc_cnt > ptq->ipc_cyc_cnt) {
+		/*
+		 * Cycle count and instruction count only go together to create
+		 * a valid IPC ratio when the cycle count changes.
+		 */
+		ptq->ipc_insn_cnt = ptq->state->tot_insn_cnt;
+		ptq->ipc_cyc_cnt = ptq->state->tot_cyc_cnt;
+	}
+
 	if (pt->sample_pwr_events && (state->type & INTEL_PT_PWR_EVT)) {
 		if (state->type & INTEL_PT_CBR_CHG) {
 			err = intel_pt_synth_cbr_sample(ptq);
@@ -1641,10 +1783,83 @@ static void intel_pt_enable_sync_switch(struct intel_pt *pt)
 	}
 }
 
+/*
+ * To filter against time ranges, it is only necessary to look at the next start
+ * or end time.
+ */
+static bool intel_pt_next_time(struct intel_pt_queue *ptq)
+{
+	struct intel_pt *pt = ptq->pt;
+
+	if (ptq->sel_start) {
+		/* Next time is an end time */
+		ptq->sel_start = false;
+		ptq->sel_timestamp = pt->time_ranges[ptq->sel_idx].end;
+		return true;
+	} else if (ptq->sel_idx + 1 < pt->range_cnt) {
+		/* Next time is a start time */
+		ptq->sel_start = true;
+		ptq->sel_idx += 1;
+		ptq->sel_timestamp = pt->time_ranges[ptq->sel_idx].start;
+		return true;
+	}
+
+	/* No next time */
+	return false;
+}
+
+static int intel_pt_time_filter(struct intel_pt_queue *ptq, u64 *ff_timestamp)
+{
+	int err;
+
+	while (1) {
+		if (ptq->sel_start) {
+			if (ptq->timestamp >= ptq->sel_timestamp) {
+				/* After start time, so consider next time */
+				intel_pt_next_time(ptq);
+				if (!ptq->sel_timestamp) {
+					/* No end time */
+					return 0;
+				}
+				/* Check against end time */
+				continue;
+			}
+			/* Before start time, so fast forward */
+			ptq->have_sample = false;
+			if (ptq->sel_timestamp > *ff_timestamp) {
+				if (ptq->sync_switch) {
+					intel_pt_next_tid(ptq->pt, ptq);
+					ptq->switch_state = INTEL_PT_SS_UNKNOWN;
+				}
+				*ff_timestamp = ptq->sel_timestamp;
+				err = intel_pt_fast_forward(ptq->decoder,
+							    ptq->sel_timestamp);
+				if (err)
+					return err;
+			}
+			return 0;
+		} else if (ptq->timestamp > ptq->sel_timestamp) {
+			/* After end time, so consider next time */
+			if (!intel_pt_next_time(ptq)) {
+				/* No next time range, so stop decoding */
+				ptq->have_sample = false;
+				ptq->switch_state = INTEL_PT_SS_NOT_TRACING;
+				return 1;
+			}
+			/* Check against next start time */
+			continue;
+		} else {
+			/* Before end time */
+			return 0;
+		}
+	}
+}
+
 static int intel_pt_run_decoder(struct intel_pt_queue *ptq, u64 *timestamp)
 {
 	const struct intel_pt_state *state = ptq->state;
 	struct intel_pt *pt = ptq->pt;
+	u64 ff_timestamp = 0;
 	int err;
 
 	if (!pt->kernel_start) {
@@ -1709,6 +1924,12 @@ static int intel_pt_run_decoder(struct intel_pt_queue *ptq, u64 *timestamp)
 			ptq->timestamp = state->timestamp;
 		}
 
+		if (ptq->sel_timestamp) {
+			err = intel_pt_time_filter(ptq, &ff_timestamp);
+			if (err)
+				return err;
+		}
+
 		if (!pt->timeless_decoding && ptq->timestamp >= *timestamp) {
 			*timestamp = ptq->timestamp;
 			return 0;
@@ -2114,6 +2335,7 @@ static void intel_pt_free(struct perf_session *session)
 	thread__put(pt->unknown_thread);
 	addr_filters__exit(&pt->filts);
 	zfree(&pt->filter);
+	zfree(&pt->time_ranges);
 	free(pt);
 }
 
@@ -2411,6 +2633,85 @@ static int intel_pt_perf_config(const char *var, const char *value, void *data)
 	return 0;
 }
 
+/* Find least TSC which converts to ns or later */
+static u64 intel_pt_tsc_start(u64 ns, struct intel_pt *pt)
+{
+	u64 tsc, tm;
+
+	tsc = perf_time_to_tsc(ns, &pt->tc);
+
+	while (1) {
+		tm = tsc_to_perf_time(tsc, &pt->tc);
+		if (tm < ns)
+			break;
+		tsc -= 1;
+	}
+
+	while (tm < ns)
+		tm = tsc_to_perf_time(++tsc, &pt->tc);
+
+	return tsc;
+}
+
+/* Find greatest TSC which converts to ns or earlier */
+static u64 intel_pt_tsc_end(u64 ns, struct intel_pt *pt)
+{
+	u64 tsc, tm;
+
+	tsc = perf_time_to_tsc(ns, &pt->tc);
+
+	while (1) {
+		tm = tsc_to_perf_time(tsc, &pt->tc);
+		if (tm > ns)
+			break;
+		tsc += 1;
+	}
+
+	while (tm > ns)
+		tm = tsc_to_perf_time(--tsc, &pt->tc);
+
+	return tsc;
+}
+
+static int intel_pt_setup_time_ranges(struct intel_pt *pt,
+				      struct itrace_synth_opts *opts)
+{
+	struct perf_time_interval *p = opts->ptime_range;
+	int n = opts->range_num;
+	int i;
+
+	if (!n || !p || pt->timeless_decoding)
+		return 0;
+
+	pt->time_ranges = calloc(n, sizeof(struct range));
+	if (!pt->time_ranges)
+		return -ENOMEM;
+
+	pt->range_cnt = n;
+
+	intel_pt_log("%s: %u range(s)\n", __func__, n);
+
+	for (i = 0; i < n; i++) {
+		struct range *r = &pt->time_ranges[i];
+		u64 ts = p[i].start;
+		u64 te = p[i].end;
+
+		/*
+		 * Take care to ensure the TSC range matches the perf-time range
+		 * when converted back to perf-time.
+		 */
+		r->start = ts ? intel_pt_tsc_start(ts, pt) : 0;
+		r->end   = te ? intel_pt_tsc_end(te, pt) : 0;
+
+		intel_pt_log("range %d: perf time interval: %"PRIu64" to %"PRIu64"\n",
+			     i, ts, te);
+		intel_pt_log("range %d: TSC time interval: %#"PRIx64" to %#"PRIx64"\n",
+			     i, r->start, r->end);
+	}
+
+	return 0;
+}
+
 static const char * const intel_pt_info_fmts[] = {
 	[INTEL_PT_PMU_TYPE]		= "  PMU Type            %"PRId64"\n",
 	[INTEL_PT_TIME_SHIFT]		= "  Time Shift          %"PRIu64"\n",
@@ -2643,6 +2944,12 @@ int intel_pt_process_auxtrace_info(union perf_event *event,
 		pt->cbr2khz = tsc_freq / pt->max_non_turbo_ratio / 1000;
 	}
 
+	if (session->itrace_synth_opts) {
+		err = intel_pt_setup_time_ranges(pt, session->itrace_synth_opts);
+		if (err)
+			goto err_delete_thread;
+	}
+
 	if (pt->synth_opts.calls)
 		pt->branches_filter |= PERF_IP_FLAG_CALL | PERF_IP_FLAG_ASYNC |
 				       PERF_IP_FLAG_TRACE_END;
@@ -2683,6 +2990,7 @@ err_free_queues:
 err_free:
 	addr_filters__exit(&pt->filts);
 	zfree(&pt->filter);
+	zfree(&pt->time_ranges);
 	free(pt);
 	return err;
 }