perf tools: Add Intel PT support

Add support for Intel Processor Trace.

Intel PT support fits within the new auxtrace infrastructure.  Recording
is supporting by identifying the Intel PT PMU, parsing options and
setting up events.

Decoding is supported by queuing up trace data by cpu or thread and then
decoding synchronously delivering synthesized event samples into the
session processing for tools to consume.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1437150840-31811-7-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

1 files changed, 1911 insertions, 0 deletions

diff --git a/tools/perf/util/intel-pt.c b/tools/perf/util/intel-pt.c
new file mode 100644
index 000000000000..2a4a4120473b
--- /dev/null
+++ b/tools/perf/util/intel-pt.c
@@ -0,0 +1,1911 @@
+/*
+ * intel_pt.c: Intel Processor Trace support
+ * Copyright (c) 2013-2015, Intel Corporation.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ */
+
+#include <stdio.h>
+#include <stdbool.h>
+#include <errno.h>
+#include <linux/kernel.h>
+#include <linux/types.h>
+
+#include "../perf.h"
+#include "session.h"
+#include "machine.h"
+#include "tool.h"
+#include "event.h"
+#include "evlist.h"
+#include "evsel.h"
+#include "map.h"
+#include "color.h"
+#include "util.h"
+#include "thread.h"
+#include "thread-stack.h"
+#include "symbol.h"
+#include "callchain.h"
+#include "dso.h"
+#include "debug.h"
+#include "auxtrace.h"
+#include "tsc.h"
+#include "intel-pt.h"
+
+#include "intel-pt-decoder/intel-pt-log.h"
+#include "intel-pt-decoder/intel-pt-decoder.h"
+#include "intel-pt-decoder/intel-pt-insn-decoder.h"
+#include "intel-pt-decoder/intel-pt-pkt-decoder.h"
+
+#define MAX_TIMESTAMP (~0ULL)
+
+struct intel_pt {
+	struct auxtrace auxtrace;
+	struct auxtrace_queues queues;
+	struct auxtrace_heap heap;
+	u32 auxtrace_type;
+	struct perf_session *session;
+	struct machine *machine;
+	struct perf_evsel *switch_evsel;
+	struct thread *unknown_thread;
+	bool timeless_decoding;
+	bool sampling_mode;
+	bool snapshot_mode;
+	bool per_cpu_mmaps;
+	bool have_tsc;
+	bool data_queued;
+	bool est_tsc;
+	bool sync_switch;
+	int have_sched_switch;
+	u32 pmu_type;
+	u64 kernel_start;
+	u64 switch_ip;
+	u64 ptss_ip;
+
+	struct perf_tsc_conversion tc;
+	bool cap_user_time_zero;
+
+	struct itrace_synth_opts synth_opts;
+
+	bool sample_instructions;
+	u64 instructions_sample_type;
+	u64 instructions_sample_period;
+	u64 instructions_id;
+
+	bool sample_branches;
+	u32 branches_filter;
+	u64 branches_sample_type;
+	u64 branches_id;
+
+	bool sample_transactions;
+	u64 transactions_sample_type;
+	u64 transactions_id;
+
+	bool synth_needs_swap;
+
+	u64 tsc_bit;
+	u64 noretcomp_bit;
+	unsigned max_non_turbo_ratio;
+};
+
+enum switch_state {
+	INTEL_PT_SS_NOT_TRACING,
+	INTEL_PT_SS_UNKNOWN,
+	INTEL_PT_SS_TRACING,
+	INTEL_PT_SS_EXPECTING_SWITCH_EVENT,
+	INTEL_PT_SS_EXPECTING_SWITCH_IP,
+};
+
+struct intel_pt_queue {
+	struct intel_pt *pt;
+	unsigned int queue_nr;
+	struct auxtrace_buffer *buffer;
+	void *decoder;
+	const struct intel_pt_state *state;
+	struct ip_callchain *chain;
+	union perf_event *event_buf;
+	bool on_heap;
+	bool stop;
+	bool step_through_buffers;
+	bool use_buffer_pid_tid;
+	pid_t pid, tid;
+	int cpu;
+	int switch_state;
+	pid_t next_tid;
+	struct thread *thread;
+	bool exclude_kernel;
+	bool have_sample;
+	u64 time;
+	u64 timestamp;
+	u32 flags;
+	u16 insn_len;
+};
+
+static void intel_pt_dump(struct intel_pt *pt __maybe_unused,
+			  unsigned char *buf, size_t len)
+{
+	struct intel_pt_pkt packet;
+	size_t pos = 0;
+	int ret, pkt_len, i;
+	char desc[INTEL_PT_PKT_DESC_MAX];
+	const char *color = PERF_COLOR_BLUE;
+
+	color_fprintf(stdout, color,
+		      ". ... Intel Processor Trace data: size %zu bytes\n",
+		      len);
+
+	while (len) {
+		ret = intel_pt_get_packet(buf, len, &packet);
+		if (ret > 0)
+			pkt_len = ret;
+		else
+			pkt_len = 1;
+		printf(".");
+		color_fprintf(stdout, color, "  %08x: ", pos);
+		for (i = 0; i < pkt_len; i++)
+			color_fprintf(stdout, color, " %02x", buf[i]);
+		for (; i < 16; i++)
+			color_fprintf(stdout, color, "   ");
+		if (ret > 0) {
+			ret = intel_pt_pkt_desc(&packet, desc,
+						INTEL_PT_PKT_DESC_MAX);
+			if (ret > 0)
+				color_fprintf(stdout, color, " %s\n", desc);
+		} else {
+			color_fprintf(stdout, color, " Bad packet!\n");
+		}
+		pos += pkt_len;
+		buf += pkt_len;
+		len -= pkt_len;
+	}
+}
+
+static void intel_pt_dump_event(struct intel_pt *pt, unsigned char *buf,
+				size_t len)
+{
+	printf(".\n");
+	intel_pt_dump(pt, buf, len);
+}
+
+static int intel_pt_do_fix_overlap(struct intel_pt *pt, struct auxtrace_buffer *a,
+				   struct auxtrace_buffer *b)
+{
+	void *start;
+
+	start = intel_pt_find_overlap(a->data, a->size, b->data, b->size,
+				      pt->have_tsc);
+	if (!start)
+		return -EINVAL;
+	b->use_size = b->data + b->size - start;
+	b->use_data = start;
+	return 0;
+}
+
+static void intel_pt_use_buffer_pid_tid(struct intel_pt_queue *ptq,
+					struct auxtrace_queue *queue,
+					struct auxtrace_buffer *buffer)
+{
+	if (queue->cpu == -1 && buffer->cpu != -1)
+		ptq->cpu = buffer->cpu;
+
+	ptq->pid = buffer->pid;
+	ptq->tid = buffer->tid;
+
+	intel_pt_log("queue %u cpu %d pid %d tid %d\n",
+		     ptq->queue_nr, ptq->cpu, ptq->pid, ptq->tid);
+
+	thread__zput(ptq->thread);
+
+	if (ptq->tid != -1) {
+		if (ptq->pid != -1)
+			ptq->thread = machine__findnew_thread(ptq->pt->machine,
+							      ptq->pid,
+							      ptq->tid);
+		else
+			ptq->thread = machine__find_thread(ptq->pt->machine, -1,
+							   ptq->tid);
+	}
+}
+
+/* This function assumes data is processed sequentially only */
+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, *old_buffer = buffer;
+	struct auxtrace_queue *queue;
+
+	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_file__fd(ptq->pt->session->file);
+
+		buffer->data = auxtrace_buffer__get_data(buffer, fd);
+		if (!buffer->data)
+			return -ENOMEM;
+	}
+
+	if (ptq->pt->snapshot_mode && !buffer->consecutive && old_buffer &&
+	    intel_pt_do_fix_overlap(ptq->pt, old_buffer, buffer))
+		return -ENOMEM;
+
+	if (old_buffer)
+		auxtrace_buffer__drop_data(old_buffer);
+
+	if (buffer->use_data) {
+		b->len = buffer->use_size;
+		b->buf = buffer->use_data;
+	} else {
+		b->len = buffer->size;
+		b->buf = buffer->data;
+	}
+	b->ref_timestamp = buffer->reference;
+
+	if (!old_buffer || ptq->pt->sampling_mode || (ptq->pt->snapshot_mode &&
+						      !buffer->consecutive)) {
+		b->consecutive = false;
+		b->trace_nr = buffer->buffer_nr + 1;
+	} else {
+		b->consecutive = true;
+	}
+
+	if (ptq->use_buffer_pid_tid && (ptq->pid != buffer->pid ||
+					ptq->tid != buffer->tid))
+		intel_pt_use_buffer_pid_tid(ptq, queue, buffer);
+
+	if (ptq->step_through_buffers)
+		ptq->stop = true;
+
+	if (!b->len)
+		return intel_pt_get_trace(b, data);
+
+	return 0;
+}
+
+struct intel_pt_cache_entry {
+	struct auxtrace_cache_entry	entry;
+	u64				insn_cnt;
+	u64				byte_cnt;
+	enum intel_pt_insn_op		op;
+	enum intel_pt_insn_branch	branch;
+	int				length;
+	int32_t				rel;
+};
+
+static int intel_pt_config_div(const char *var, const char *value, void *data)
+{
+	int *d = data;
+	long val;
+
+	if (!strcmp(var, "intel-pt.cache-divisor")) {
+		val = strtol(value, NULL, 0);
+		if (val > 0 && val <= INT_MAX)
+			*d = val;
+	}
+
+	return 0;
+}
+
+static int intel_pt_cache_divisor(void)
+{
+	static int d;
+
+	if (d)
+		return d;
+
+	perf_config(intel_pt_config_div, &d);
+
+	if (!d)
+		d = 64;
+
+	return d;
+}
+
+static unsigned int intel_pt_cache_size(struct dso *dso,
+					struct machine *machine)
+{
+	off_t size;
+
+	size = dso__data_size(dso, machine);
+	size /= intel_pt_cache_divisor();
+	if (size < 1000)
+		return 10;
+	if (size > (1 << 21))
+		return 21;
+	return 32 - __builtin_clz(size);
+}
+
+static struct auxtrace_cache *intel_pt_cache(struct dso *dso,
+					     struct machine *machine)
+{
+	struct auxtrace_cache *c;
+	unsigned int bits;
+
+	if (dso->auxtrace_cache)
+		return dso->auxtrace_cache;
+
+	bits = intel_pt_cache_size(dso, machine);
+
+	/* Ignoring cache creation failure */
+	c = auxtrace_cache__new(bits, sizeof(struct intel_pt_cache_entry), 200);
+
+	dso->auxtrace_cache = c;
+
+	return c;
+}
+
+static int intel_pt_cache_add(struct dso *dso, struct machine *machine,
+			      u64 offset, u64 insn_cnt, u64 byte_cnt,
+			      struct intel_pt_insn *intel_pt_insn)
+{
+	struct auxtrace_cache *c = intel_pt_cache(dso, machine);
+	struct intel_pt_cache_entry *e;
+	int err;
+
+	if (!c)
+		return -ENOMEM;
+
+	e = auxtrace_cache__alloc_entry(c);
+	if (!e)
+		return -ENOMEM;
+
+	e->insn_cnt = insn_cnt;
+	e->byte_cnt = byte_cnt;
+	e->op = intel_pt_insn->op;
+	e->branch = intel_pt_insn->branch;
+	e->length = intel_pt_insn->length;
+	e->rel = intel_pt_insn->rel;
+
+	err = auxtrace_cache__add(c, offset, &e->entry);
+	if (err)
+		auxtrace_cache__free_entry(c, e);
+
+	return err;
+}
+
+static struct intel_pt_cache_entry *
+intel_pt_cache_lookup(struct dso *dso, struct machine *machine, u64 offset)
+{
+	struct auxtrace_cache *c = intel_pt_cache(dso, machine);
+
+	if (!c)
+		return NULL;
+
+	return auxtrace_cache__lookup(dso->auxtrace_cache, offset);
+}
+
+static int intel_pt_walk_next_insn(struct intel_pt_insn *intel_pt_insn,
+				   uint64_t *insn_cnt_ptr, uint64_t *ip,
+				   uint64_t to_ip, uint64_t max_insn_cnt,
+				   void *data)
+{
+	struct intel_pt_queue *ptq = data;
+	struct machine *machine = ptq->pt->machine;
+	struct thread *thread;
+	struct addr_location al;
+	unsigned char buf[1024];
+	size_t bufsz;
+	ssize_t len;
+	int x86_64;
+	u8 cpumode;
+	u64 offset, start_offset, start_ip;
+	u64 insn_cnt = 0;
+	bool one_map = true;
+
+	if (to_ip && *ip == to_ip)
+		goto out_no_cache;
+
+	bufsz = intel_pt_insn_max_size();
+
+	if (*ip >= ptq->pt->kernel_start)
+		cpumode = PERF_RECORD_MISC_KERNEL;
+	else
+		cpumode = PERF_RECORD_MISC_USER;
+
+	thread = ptq->thread;
+	if (!thread) {
+		if (cpumode != PERF_RECORD_MISC_KERNEL)
+			return -EINVAL;
+		thread = ptq->pt->unknown_thread;
+	}
+
+	while (1) {
+		thread__find_addr_map(thread, cpumode, MAP__FUNCTION, *ip, &al);
+		if (!al.map || !al.map->dso)
+			return -EINVAL;
+
+		if (al.map->dso->data.status == DSO_DATA_STATUS_ERROR &&
+		    dso__data_status_seen(al.map->dso,
+					  DSO_DATA_STATUS_SEEN_ITRACE))
+			return -ENOENT;
+
+		offset = al.map->map_ip(al.map, *ip);
+
+		if (!to_ip && one_map) {
+			struct intel_pt_cache_entry *e;
+
+			e = intel_pt_cache_lookup(al.map->dso, machine, offset);
+			if (e &&
+			    (!max_insn_cnt || e->insn_cnt <= max_insn_cnt)) {
+				*insn_cnt_ptr = e->insn_cnt;
+				*ip += e->byte_cnt;
+				intel_pt_insn->op = e->op;
+				intel_pt_insn->branch = e->branch;
+				intel_pt_insn->length = e->length;
+				intel_pt_insn->rel = e->rel;
+				intel_pt_log_insn_no_data(intel_pt_insn, *ip);
+				return 0;
+			}
+		}
+
+		start_offset = offset;
+		start_ip = *ip;
+
+		/* Load maps to ensure dso->is_64_bit has been updated */
+		map__load(al.map, machine->symbol_filter);
+
+		x86_64 = al.map->dso->is_64_bit;
+
+		while (1) {
+			len = dso__data_read_offset(al.map->dso, machine,
+						    offset, buf, bufsz);
+			if (len <= 0)
+				return -EINVAL;
+
+			if (intel_pt_get_insn(buf, len, x86_64, intel_pt_insn))
+				return -EINVAL;
+
+			intel_pt_log_insn(intel_pt_insn, *ip);
+
+			insn_cnt += 1;
+
+			if (intel_pt_insn->branch != INTEL_PT_BR_NO_BRANCH)
+				goto out;
+
+			if (max_insn_cnt && insn_cnt >= max_insn_cnt)
+				goto out_no_cache;
+
+			*ip += intel_pt_insn->length;
+
+			if (to_ip && *ip == to_ip)
+				goto out_no_cache;
+
+			if (*ip >= al.map->end)
+				break;
+
+			offset += intel_pt_insn->length;
+		}
+		one_map = false;
+	}
+out:
+	*insn_cnt_ptr = insn_cnt;
+
+	if (!one_map)
+		goto out_no_cache;
+
+	/*
+	 * Didn't lookup in the 'to_ip' case, so do it now to prevent duplicate
+	 * entries.
+	 */
+	if (to_ip) {
+		struct intel_pt_cache_entry *e;
+
+		e = intel_pt_cache_lookup(al.map->dso, machine, start_offset);
+		if (e)
+			return 0;
+	}
+
+	/* Ignore cache errors */
+	intel_pt_cache_add(al.map->dso, machine, start_offset, insn_cnt,
+			   *ip - start_ip, intel_pt_insn);
+
+	return 0;
+
+out_no_cache:
+	*insn_cnt_ptr = insn_cnt;
+	return 0;
+}
+
+static bool intel_pt_get_config(struct intel_pt *pt,
+				struct perf_event_attr *attr, u64 *config)
+{
+	if (attr->type == pt->pmu_type) {
+		if (config)
+			*config = attr->config;
+		return true;
+	}
+
+	return false;
+}
+
+static bool intel_pt_exclude_kernel(struct intel_pt *pt)
+{
+	struct perf_evsel *evsel;
+
+	evlist__for_each(pt->session->evlist, evsel) {
+		if (intel_pt_get_config(pt, &evsel->attr, NULL) &&
+		    !evsel->attr.exclude_kernel)
+			return false;
+	}
+	return true;
+}
+
+static bool intel_pt_return_compression(struct intel_pt *pt)
+{
+	struct perf_evsel *evsel;
+	u64 config;
+
+	if (!pt->noretcomp_bit)
+		return true;
+
+	evlist__for_each(pt->session->evlist, evsel) {
+		if (intel_pt_get_config(pt, &evsel->attr, &config) &&
+		    (config & pt->noretcomp_bit))
+			return false;
+	}
+	return true;
+}
+
+static bool intel_pt_timeless_decoding(struct intel_pt *pt)
+{
+	struct perf_evsel *evsel;
+	bool timeless_decoding = true;
+	u64 config;
+
+	if (!pt->tsc_bit || !pt->cap_user_time_zero)
+		return true;
+
+	evlist__for_each(pt->session->evlist, evsel) {
+		if (!(evsel->attr.sample_type & PERF_SAMPLE_TIME))
+			return true;
+		if (intel_pt_get_config(pt, &evsel->attr, &config)) {
+			if (config & pt->tsc_bit)
+				timeless_decoding = false;
+			else
+				return true;
+		}
+	}
+	return timeless_decoding;
+}
+
+static bool intel_pt_tracing_kernel(struct intel_pt *pt)
+{
+	struct perf_evsel *evsel;
+
+	evlist__for_each(pt->session->evlist, evsel) {
+		if (intel_pt_get_config(pt, &evsel->attr, NULL) &&
+		    !evsel->attr.exclude_kernel)
+			return true;
+	}
+	return false;
+}
+
+static bool intel_pt_have_tsc(struct intel_pt *pt)
+{
+	struct perf_evsel *evsel;
+	bool have_tsc = false;
+	u64 config;
+
+	if (!pt->tsc_bit)
+		return false;
+
+	evlist__for_each(pt->session->evlist, evsel) {
+		if (intel_pt_get_config(pt, &evsel->attr, &config)) {
+			if (config & pt->tsc_bit)
+				have_tsc = true;
+			else
+				return false;
+		}
+	}
+	return have_tsc;
+}
+
+static u64 intel_pt_ns_to_ticks(const struct intel_pt *pt, u64 ns)
+{
+	u64 quot, rem;
+
+	quot = ns / pt->tc.time_mult;
+	rem  = ns % pt->tc.time_mult;
+	return (quot << pt->tc.time_shift) + (rem << pt->tc.time_shift) /
+		pt->tc.time_mult;
+}
+
+static struct intel_pt_queue *intel_pt_alloc_queue(struct intel_pt *pt,
+						   unsigned int queue_nr)
+{
+	struct intel_pt_params params = { .get_trace = 0, };
+	struct intel_pt_queue *ptq;
+
+	ptq = zalloc(sizeof(struct intel_pt_queue));
+	if (!ptq)
+		return NULL;
+
+	if (pt->synth_opts.callchain) {
+		size_t sz = sizeof(struct ip_callchain);
+
+		sz += pt->synth_opts.callchain_sz * sizeof(u64);
+		ptq->chain = zalloc(sz);
+		if (!ptq->chain)
+			goto out_free;
+	}
+
+	ptq->event_buf = malloc(PERF_SAMPLE_MAX_SIZE);
+	if (!ptq->event_buf)
+		goto out_free;
+
+	ptq->pt = pt;
+	ptq->queue_nr = queue_nr;
+	ptq->exclude_kernel = intel_pt_exclude_kernel(pt);
+	ptq->pid = -1;
+	ptq->tid = -1;
+	ptq->cpu = -1;
+	ptq->next_tid = -1;
+
+	params.get_trace = intel_pt_get_trace;
+	params.walk_insn = intel_pt_walk_next_insn;
+	params.data = ptq;
+	params.return_compression = intel_pt_return_compression(pt);
+	params.max_non_turbo_ratio = pt->max_non_turbo_ratio;
+
+	if (pt->synth_opts.instructions) {
+		if (pt->synth_opts.period) {
+			switch (pt->synth_opts.period_type) {
+			case PERF_ITRACE_PERIOD_INSTRUCTIONS:
+				params.period_type =
+						INTEL_PT_PERIOD_INSTRUCTIONS;
+				params.period = pt->synth_opts.period;
+				break;
+			case PERF_ITRACE_PERIOD_TICKS:
+				params.period_type = INTEL_PT_PERIOD_TICKS;
+				params.period = pt->synth_opts.period;
+				break;
+			case PERF_ITRACE_PERIOD_NANOSECS:
+				params.period_type = INTEL_PT_PERIOD_TICKS;
+				params.period = intel_pt_ns_to_ticks(pt,
+							pt->synth_opts.period);
+				break;
+			default:
+				break;
+			}
+		}
+
+		if (!params.period) {
+			params.period_type = INTEL_PT_PERIOD_INSTRUCTIONS;
+			params.period = 1000;
+		}
+	}
+
+	ptq->decoder = intel_pt_decoder_new(&params);
+	if (!ptq->decoder)
+		goto out_free;
+
+	return ptq;
+
+out_free:
+	zfree(&ptq->event_buf);
+	zfree(&ptq->chain);
+	free(ptq);
+	return NULL;
+}
+
+static void intel_pt_free_queue(void *priv)
+{
+	struct intel_pt_queue *ptq = priv;
+
+	if (!ptq)
+		return;
+	thread__zput(ptq->thread);
+	intel_pt_decoder_free(ptq->decoder);
+	zfree(&ptq->event_buf);
+	zfree(&ptq->chain);
+	free(ptq);
+}
+
+static void intel_pt_set_pid_tid_cpu(struct intel_pt *pt,
+				     struct auxtrace_queue *queue)
+{
+	struct intel_pt_queue *ptq = queue->priv;
+
+	if (queue->tid == -1 || pt->have_sched_switch) {
+		ptq->tid = machine__get_current_tid(pt->machine, ptq->cpu);
+		thread__zput(ptq->thread);
+	}
+
+	if (!ptq->thread && ptq->tid != -1)
+		ptq->thread = machine__find_thread(pt->machine, -1, ptq->tid);
+
+	if (ptq->thread) {
+		ptq->pid = ptq->thread->pid_;
+		if (queue->cpu == -1)
+			ptq->cpu = ptq->thread->cpu;
+	}
+}
+
+static void intel_pt_sample_flags(struct intel_pt_queue *ptq)
+{
+	if (ptq->state->flags & INTEL_PT_ABORT_TX) {
+		ptq->flags = PERF_IP_FLAG_BRANCH | PERF_IP_FLAG_TX_ABORT;
+	} else if (ptq->state->flags & INTEL_PT_ASYNC) {
+		if (ptq->state->to_ip)
+			ptq->flags = PERF_IP_FLAG_BRANCH | PERF_IP_FLAG_CALL |
+				     PERF_IP_FLAG_ASYNC |
+				     PERF_IP_FLAG_INTERRUPT;
+		else
+			ptq->flags = PERF_IP_FLAG_BRANCH |
+				     PERF_IP_FLAG_TRACE_END;
+		ptq->insn_len = 0;
+	} else {
+		if (ptq->state->from_ip)
+			ptq->flags = intel_pt_insn_type(ptq->state->insn_op);
+		else
+			ptq->flags = PERF_IP_FLAG_BRANCH |
+				     PERF_IP_FLAG_TRACE_BEGIN;
+		if (ptq->state->flags & INTEL_PT_IN_TX)
+			ptq->flags |= PERF_IP_FLAG_IN_TX;
+		ptq->insn_len = ptq->state->insn_len;
+	}
+}
+
+static int intel_pt_setup_queue(struct intel_pt *pt,
+				struct auxtrace_queue *queue,
+				unsigned int queue_nr)
+{
+	struct intel_pt_queue *ptq = queue->priv;
+
+	if (list_empty(&queue->head))
+		return 0;
+
+	if (!ptq) {
+		ptq = intel_pt_alloc_queue(pt, queue_nr);
+		if (!ptq)
+			return -ENOMEM;
+		queue->priv = ptq;
+
+		if (queue->cpu != -1)
+			ptq->cpu = queue->cpu;
+		ptq->tid = queue->tid;
+
+		if (pt->sampling_mode) {
+			if (pt->timeless_decoding)
+				ptq->step_through_buffers = true;
+			if (pt->timeless_decoding || !pt->have_sched_switch)
+				ptq->use_buffer_pid_tid = true;
+		}
+	}
+
+	if (!ptq->on_heap &&
+	    (!pt->sync_switch ||
+	     ptq->switch_state != INTEL_PT_SS_EXPECTING_SWITCH_EVENT)) {
+		const struct intel_pt_state *state;
+		int ret;
+
+		if (pt->timeless_decoding)
+			return 0;
+
+		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);
+		while (1) {
+			state = intel_pt_decode(ptq->decoder);
+			if (state->err) {
+				if (state->err == INTEL_PT_ERR_NODATA) {
+					intel_pt_log("queue %u has no timestamp\n",
+						     queue_nr);
+					return 0;
+				}
+				continue;
+			}
+			if (state->timestamp)
+				break;
+		}
+
+		ptq->timestamp = state->timestamp;
+		intel_pt_log("queue %u timestamp 0x%" PRIx64 "\n",
+			     queue_nr, ptq->timestamp);
+		ptq->state = state;
+		ptq->have_sample = true;
+		intel_pt_sample_flags(ptq);
+		ret = auxtrace_heap__add(&pt->heap, queue_nr, ptq->timestamp);
+		if (ret)
+			return ret;
+		ptq->on_heap = true;
+	}
+
+	return 0;
+}
+
+static int intel_pt_setup_queues(struct intel_pt *pt)
+{
+	unsigned int i;
+	int ret;
+
+	for (i = 0; i < pt->queues.nr_queues; i++) {
+		ret = intel_pt_setup_queue(pt, &pt->queues.queue_array[i], i);
+		if (ret)
+			return ret;
+	}
+	return 0;
+}
+
+static int intel_pt_inject_event(union perf_event *event,
+				 struct perf_sample *sample, u64 type,
+				 bool swapped)
+{
+	event->header.size = perf_event__sample_event_size(sample, type, 0);
+	return perf_event__synthesize_sample(event, type, 0, sample, swapped);
+}
+
+static int intel_pt_synth_branch_sample(struct intel_pt_queue *ptq)
+{
+	int ret;
+	struct intel_pt *pt = ptq->pt;
+	union perf_event *event = ptq->event_buf;
+	struct perf_sample sample = { .ip = 0, };
+
+	event->sample.header.type = PERF_RECORD_SAMPLE;
+	event->sample.header.misc = PERF_RECORD_MISC_USER;
+	event->sample.header.size = sizeof(struct perf_event_header);
+
+	if (!pt->timeless_decoding)
+		sample.time = tsc_to_perf_time(ptq->timestamp, &pt->tc);
+
+	sample.ip = ptq->state->from_ip;
+	sample.pid = ptq->pid;
+	sample.tid = ptq->tid;
+	sample.addr = ptq->state->to_ip;
+	sample.id = ptq->pt->branches_id;
+	sample.stream_id = ptq->pt->branches_id;
+	sample.period = 1;
+	sample.cpu = ptq->cpu;
+	sample.flags = ptq->flags;
+	sample.insn_len = ptq->insn_len;
+
+	if (pt->branches_filter && !(pt->branches_filter & ptq->flags))
+		return 0;
+
+	if (pt->synth_opts.inject) {
+		ret = intel_pt_inject_event(event, &sample,
+					    pt->branches_sample_type,
+					    pt->synth_needs_swap);
+		if (ret)
+			return ret;
+	}
+
+	ret = perf_session__deliver_synth_event(pt->session, event, &sample);
+	if (ret)
+		pr_err("Intel Processor Trace: failed to deliver branch event, error %d\n",
+		       ret);
+
+	return ret;
+}
+
+static int intel_pt_synth_instruction_sample(struct intel_pt_queue *ptq)
+{
+	int ret;
+	struct intel_pt *pt = ptq->pt;
+	union perf_event *event = ptq->event_buf;
+	struct perf_sample sample = { .ip = 0, };
+
+	event->sample.header.type = PERF_RECORD_SAMPLE;
+	event->sample.header.misc = PERF_RECORD_MISC_USER;
+	event->sample.header.size = sizeof(struct perf_event_header);
+
+	if (!pt->timeless_decoding)
+		sample.time = tsc_to_perf_time(ptq->timestamp, &pt->tc);
+
+	sample.ip = ptq->state->from_ip;
+	sample.pid = ptq->pid;
+	sample.tid = ptq->tid;
+	sample.addr = ptq->state->to_ip;
+	sample.id = ptq->pt->instructions_id;
+	sample.stream_id = ptq->pt->instructions_id;
+	sample.period = ptq->pt->instructions_sample_period;
+	sample.cpu = ptq->cpu;
+	sample.flags = ptq->flags;
+	sample.insn_len = ptq->insn_len;
+
+	if (pt->synth_opts.callchain) {
+		thread_stack__sample(ptq->thread, ptq->chain,
+				     pt->synth_opts.callchain_sz, sample.ip);
+		sample.callchain = ptq->chain;
+	}
+
+	if (pt->synth_opts.inject) {
+		ret = intel_pt_inject_event(event, &sample,
+					    pt->instructions_sample_type,
+					    pt->synth_needs_swap);
+		if (ret)
+			return ret;
+	}
+
+	ret = perf_session__deliver_synth_event(pt->session, event, &sample);
+	if (ret)
+		pr_err("Intel Processor Trace: failed to deliver instruction event, error %d\n",
+		       ret);
+
+	return ret;
+}
+
+static int intel_pt_synth_transaction_sample(struct intel_pt_queue *ptq)
+{
+	int ret;
+	struct intel_pt *pt = ptq->pt;
+	union perf_event *event = ptq->event_buf;
+	struct perf_sample sample = { .ip = 0, };
+
+	event->sample.header.type = PERF_RECORD_SAMPLE;
+	event->sample.header.misc = PERF_RECORD_MISC_USER;
+	event->sample.header.size = sizeof(struct perf_event_header);
+
+	if (!pt->timeless_decoding)
+		sample.time = tsc_to_perf_time(ptq->timestamp, &pt->tc);
+
+	sample.ip = ptq->state->from_ip;
+	sample.pid = ptq->pid;
+	sample.tid = ptq->tid;
+	sample.addr = ptq->state->to_ip;
+	sample.id = ptq->pt->transactions_id;
+	sample.stream_id = ptq->pt->transactions_id;
+	sample.period = 1;
+	sample.cpu = ptq->cpu;
+	sample.flags = ptq->flags;
+	sample.insn_len = ptq->insn_len;
+
+	if (pt->synth_opts.callchain) {
+		thread_stack__sample(ptq->thread, ptq->chain,
+				     pt->synth_opts.callchain_sz, sample.ip);
+		sample.callchain = ptq->chain;
+	}
+
+	if (pt->synth_opts.inject) {
+		ret = intel_pt_inject_event(event, &sample,
+					    pt->transactions_sample_type,
+					    pt->synth_needs_swap);
+		if (ret)
+			return ret;
+	}
+
+	ret = perf_session__deliver_synth_event(pt->session, event, &sample);
+	if (ret)
+		pr_err("Intel Processor Trace: failed to deliver transaction event, error %d\n",
+		       ret);
+
+	return ret;
+}
+
+static int intel_pt_synth_error(struct intel_pt *pt, int code, int cpu,
+				pid_t pid, pid_t tid, u64 ip)
+{
+	union perf_event event;
+	char msg[MAX_AUXTRACE_ERROR_MSG];
+	int err;
+
+	intel_pt__strerror(code, msg, MAX_AUXTRACE_ERROR_MSG);
+
+	auxtrace_synth_error(&event.auxtrace_error, PERF_AUXTRACE_ERROR_ITRACE,
+			     code, cpu, pid, tid, ip, msg);
+
+	err = perf_session__deliver_synth_event(pt->session, &event, NULL);
+	if (err)
+		pr_err("Intel Processor Trace: failed to deliver error event, error %d\n",
+		       err);
+
+	return err;
+}
+
+static int intel_pt_next_tid(struct intel_pt *pt, struct intel_pt_queue *ptq)
+{
+	struct auxtrace_queue *queue;
+	pid_t tid = ptq->next_tid;
+	int err;
+
+	if (tid == -1)
+		return 0;
+
+	intel_pt_log("switch: cpu %d tid %d\n", ptq->cpu, tid);
+
+	err = machine__set_current_tid(pt->machine, ptq->cpu, -1, tid);
+
+	queue = &pt->queues.queue_array[ptq->queue_nr];
+	intel_pt_set_pid_tid_cpu(pt, queue);
+
+	ptq->next_tid = -1;
+
+	return err;
+}
+
+static inline bool intel_pt_is_switch_ip(struct intel_pt_queue *ptq, u64 ip)
+{
+	struct intel_pt *pt = ptq->pt;
+
+	return ip == pt->switch_ip &&
+	       (ptq->flags & PERF_IP_FLAG_BRANCH) &&
+	       !(ptq->flags & (PERF_IP_FLAG_CONDITIONAL | PERF_IP_FLAG_ASYNC |
+			       PERF_IP_FLAG_INTERRUPT | PERF_IP_FLAG_TX_ABORT));
+}
+
+static int intel_pt_sample(struct intel_pt_queue *ptq)
+{
+	const struct intel_pt_state *state = ptq->state;
+	struct intel_pt *pt = ptq->pt;
+	int err;
+
+	if (!ptq->have_sample)
+		return 0;
+
+	ptq->have_sample = false;
+
+	if (pt->sample_instructions &&
+	    (state->type & INTEL_PT_INSTRUCTION)) {
+		err = intel_pt_synth_instruction_sample(ptq);
+		if (err)
+			return err;
+	}
+
+	if (pt->sample_transactions &&
+	    (state->type & INTEL_PT_TRANSACTION)) {
+		err = intel_pt_synth_transaction_sample(ptq);
+		if (err)
+			return err;
+	}
+
+	if (!(state->type & INTEL_PT_BRANCH))
+		return 0;
+
+	if (pt->synth_opts.callchain)
+		thread_stack__event(ptq->thread, ptq->flags, state->from_ip,
+				    state->to_ip, ptq->insn_len,
+				    state->trace_nr);
+	else
+		thread_stack__set_trace_nr(ptq->thread, state->trace_nr);
+
+	if (pt->sample_branches) {
+		err = intel_pt_synth_branch_sample(ptq);
+		if (err)
+			return err;
+	}
+
+	if (!pt->sync_switch)
+		return 0;
+
+	if (intel_pt_is_switch_ip(ptq, state->to_ip)) {
+		switch (ptq->switch_state) {
+		case INTEL_PT_SS_UNKNOWN:
+		case INTEL_PT_SS_EXPECTING_SWITCH_IP:
+			err = intel_pt_next_tid(pt, ptq);
+			if (err)
+				return err;
+			ptq->switch_state = INTEL_PT_SS_TRACING;
+			break;
+		default:
+			ptq->switch_state = INTEL_PT_SS_EXPECTING_SWITCH_EVENT;
+			return 1;
+		}
+	} else if (!state->to_ip) {
+		ptq->switch_state = INTEL_PT_SS_NOT_TRACING;
+	} else if (ptq->switch_state == INTEL_PT_SS_NOT_TRACING) {
+		ptq->switch_state = INTEL_PT_SS_UNKNOWN;
+	} else if (ptq->switch_state == INTEL_PT_SS_UNKNOWN &&
+		   state->to_ip == pt->ptss_ip &&
+		   (ptq->flags & PERF_IP_FLAG_CALL)) {
+		ptq->switch_state = INTEL_PT_SS_TRACING;
+	}
+
+	return 0;
+}
+
+static u64 intel_pt_switch_ip(struct machine *machine, u64 *ptss_ip)
+{
+	struct map *map;
+	struct symbol *sym, *start;
+	u64 ip, switch_ip = 0;
+
+	if (ptss_ip)
+		*ptss_ip = 0;
+
+	map = machine__kernel_map(machine, MAP__FUNCTION);
+	if (!map)
+		return 0;
+
+	if (map__load(map, machine->symbol_filter))
+		return 0;
+
+	start = dso__first_symbol(map->dso, MAP__FUNCTION);
+
+	for (sym = start; sym; sym = dso__next_symbol(sym)) {
+		if (sym->binding == STB_GLOBAL &&
+		    !strcmp(sym->name, "__switch_to")) {
+			ip = map->unmap_ip(map, sym->start);
+			if (ip >= map->start && ip < map->end) {
+				switch_ip = ip;
+				break;
+			}
+		}
+	}
+
+	if (!switch_ip || !ptss_ip)
+		return 0;
+
+	for (sym = start; sym; sym = dso__next_symbol(sym)) {
+		if (!strcmp(sym->name, "perf_trace_sched_switch")) {
+			ip = map->unmap_ip(map, sym->start);
+			if (ip >= map->start && ip < map->end) {
+				*ptss_ip = ip;
+				break;
+			}
+		}
+	}
+
+	return switch_ip;
+}
+
+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;
+	int err;
+
+	if (!pt->kernel_start) {
+		pt->kernel_start = machine__kernel_start(pt->machine);
+		if (pt->per_cpu_mmaps && pt->have_sched_switch &&
+		    !pt->timeless_decoding && intel_pt_tracing_kernel(pt) &&
+		    !pt->sampling_mode) {
+			pt->switch_ip = intel_pt_switch_ip(pt->machine,
+							   &pt->ptss_ip);
+			if (pt->switch_ip) {
+				intel_pt_log("switch_ip: %"PRIx64" ptss_ip: %"PRIx64"\n",
+					     pt->switch_ip, pt->ptss_ip);
+				pt->sync_switch = true;
+			}
+		}
+	}
+
+	intel_pt_log("queue %u decoding cpu %d pid %d tid %d\n",
+		     ptq->queue_nr, ptq->cpu, ptq->pid, ptq->tid);
+	while (1) {
+		err = intel_pt_sample(ptq);
+		if (err)
+			return err;
+
+		state = intel_pt_decode(ptq->decoder);
+		if (state->err) {
+			if (state->err == INTEL_PT_ERR_NODATA)
+				return 1;
+			if (pt->sync_switch &&
+			    state->from_ip >= pt->kernel_start) {
+				pt->sync_switch = false;
+				intel_pt_next_tid(pt, ptq);
+			}
+			if (pt->synth_opts.errors) {
+				err = intel_pt_synth_error(pt, state->err,
+							   ptq->cpu, ptq->pid,
+							   ptq->tid,
+							   state->from_ip);