[PATCH 10/23] perf script: Add "brstackasm" output for branch stacks

[Arnaldo Carvalho de Melo <acme@kernel.org>]

From: Andi Kleen <ak@linux.intel.com>

Implement printing full disassembled sequences for branch stacks in perf
script.

This allows to directly print hot paths for individual samples, together
with branch misprediction and cycle count / IPC information if available
(on Skylake systems).

This only works when no special branch filters are specified.

E.g.:

  % perf record -b ...
  % perf script -F brstackasm
  ...
        000055b55d1147d0        pushq  %rbp
        000055b55d1147d1        pushq  %r15
        000055b55d1147d3        pushq  %r14
        000055b55d1147d5        pushq  %r13
        000055b55d1147d7        pushq  %r12
        000055b55d1147d9        pushq  %rbx
        000055b55d1147da        sub $0x18, %rsp
        000055b55d1147de        mov %r8, %r13
        000055b55d1147e1        mov %rcx, %rbp
        000055b55d1147e4        mov %rdx, %r14
        000055b55d1147e7        mov %rsi, %r15
        000055b55d1147ea        mov %rdi, %rbx
        000055b55d1147ed        movl  $0x0, 0xc(%rsp)
        000055b55d1147f5        movq  (%rbp), %rax
        000055b55d1147f9        test $0x1, %al
        000055b55d1147fb        jnz 0x55b55d114890              # PRED 4 cycles 3.75 IPC
        000055b55d114890        mov %eax, %ecx
        000055b55d114892        and $0x3, %ecx
        000055b55d114895        cmp $0x1, %rcx
        000055b55d114899        jnz 0x55b55d1148f8
        000055b55d11489b        movq  -0x1(%rax), %rcx
        000055b55d11489f        cmpb  $0x81, 0xb(%rcx)
        000055b55d1148a3        jnz 0x55b55d1148fe              # PRED 1 cycles 6.00 IPC
...

Occasionally the path does not reach up to the sample IP, as the LBRs
may be frozen before executing a final jump. In this case we print a
special message.

v2:
Use low level abstracted disassembler interface.
Print symbols and source lines as labels.
Print first jump in LBR too.
Patch up blocks with filtered ring transfers.
Show IPC
Lots of cleanups and improvements.

v3:
Print special message for branches frozen early.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Link: http://lkml.kernel.org/r/20170119014150.19218-6-andi@firstfloor.org
[ Fix up u64 formatting on PRIu64, use {} on multiline if/else blocks ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
---
 tools/perf/Documentation/perf-script.txt |  13 +-
 tools/perf/builtin-script.c              | 268 +++++++++++++++++++++++++++++++
 2 files changed, 279 insertions(+), 2 deletions(-)

diff --git a/tools/perf/Documentation/perf-script.txt b/tools/perf/Documentation/perf-script.txt
index 497989ea9768..15a80815941e 100644
--- a/tools/perf/Documentation/perf-script.txt
+++ b/tools/perf/Documentation/perf-script.txt
@@ -116,7 +116,7 @@ OPTIONS
 --fields::
         Comma separated list of fields to print. Options are:
         comm, tid, pid, time, cpu, event, trace, ip, sym, dso, addr, symoff,
-        srcline, period, iregs, brstack, brstacksym, flags, bpf-output, asm.
+        srcline, period, iregs, brstack, brstacksym, flags, bpf-output, asm, brstackasm,
         callindent, insn, insnlen. Field list can be prepended with the type, trace, sw or hw,
         to indicate to which event type the field list applies.
         e.g., -F sw:comm,tid,time,ip,sym  and -F trace:time,cpu,trace
@@ -189,17 +189,22 @@ OPTIONS
 	i.e., -F "" is not allowed.
 
 	The brstack output includes branch related information with raw addresses using the
-	/v/v/v/v/ syntax in the following order:
+	/v/v/v/v/cycles syntax in the following order:
 	FROM: branch source instruction
 	TO  : branch target instruction
         M/P/-: M=branch target mispredicted or branch direction was mispredicted, P=target predicted or direction predicted, -=not supported
 	X/- : X=branch inside a transactional region, -=not in transaction region or not supported
 	A/- : A=TSX abort entry, -=not aborted region or not supported
+	cycles
 
 	The brstacksym is identical to brstack, except that the FROM and TO addresses are printed in a symbolic form if possible.
 
 	When asm is specified the assembler instruction of each sample is printed in disassembled form.
 
+	When brstackasm is specified the full assembler sequences of branch sequences for each sample
+	is printed. This is the full execution path leading to the sample. This is only supported when the
+	sample was recorded with perf record -b or -j any.
+
 -k::
 --vmlinux=<file>::
         vmlinux pathname
@@ -301,6 +306,10 @@ include::itrace.txt[]
 	stop time is not given (i.e, time string is 'x.y,') then analysis goes
 	to end of file.
 
+--max-blocks::
+	Set the maximum number of program blocks to print with brstackasm for
+	each sample.
+
 SEE ALSO
 --------
 linkperf:perf-record[1], linkperf:perf-script-perl[1],
diff --git a/tools/perf/builtin-script.c b/tools/perf/builtin-script.c
index 7a09c4f7df3f..512d298031b9 100644
--- a/tools/perf/builtin-script.c
+++ b/tools/perf/builtin-script.c
@@ -43,6 +43,7 @@ static bool			nanosecs;
 static const char		*cpu_list;
 static DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
 static struct perf_stat_config	stat_config;
+static int			max_blocks;
 
 unsigned int scripting_max_stack = PERF_MAX_STACK_DEPTH;
 
@@ -71,6 +72,7 @@ enum perf_output_field {
 	PERF_OUTPUT_INSN	    = 1U << 21,
 	PERF_OUTPUT_INSNLEN	    = 1U << 22,
 	PERF_OUTPUT_ASM		    = 1U << 23,
+	PERF_OUTPUT_BRSTACKASM	    = 1U << 24,
 };
 
 struct output_option {
@@ -101,6 +103,7 @@ struct output_option {
 	{.str = "insn", .field = PERF_OUTPUT_INSN},
 	{.str = "insnlen", .field = PERF_OUTPUT_INSNLEN},
 	{.str = "asm", .field = PERF_OUTPUT_ASM},
+	{.str = "brstackasm", .field = PERF_OUTPUT_BRSTACKASM},
 };
 
 /* default set to maintain compatibility with current format */
@@ -300,6 +303,13 @@ static int perf_evsel__check_attr(struct perf_evsel *evsel,
 		       "selected.\n");
 		return -EINVAL;
 	}
+	if (PRINT_FIELD(BRSTACKASM) &&
+	    !(perf_evlist__combined_branch_type(session->evlist) &
+	      PERF_SAMPLE_BRANCH_ANY)) {
+		pr_err("Display of branch stack assembler requested, but non all-branch filter set\n");
+		return -EINVAL;
+	}
+
 	if ((PRINT_FIELD(PID) || PRINT_FIELD(TID)) &&
 		perf_evsel__check_stype(evsel, PERF_SAMPLE_TID, "TID",
 					PERF_OUTPUT_TID|PERF_OUTPUT_PID))
@@ -586,6 +596,260 @@ static void print_sample_brstacksym(struct perf_sample *sample,
 	}
 }
 
+#define MAXBB 16384UL
+
+static int grab_bb(u8 *buffer, u64 start, u64 end,
+		    struct machine *machine, struct thread *thread,
+		    bool *is64bit, u8 *cpumode, bool last)
+{
+	int offset, len;
+	struct addr_location al;
+	bool kernel;
+
+	if (!start || !end)
+		return 0;
+
+	kernel = machine__kernel_ip(machine, start);
+	if (kernel)
+		*cpumode = PERF_RECORD_MISC_KERNEL;
+	else
+		*cpumode = PERF_RECORD_MISC_USER;
+
+	/*
+	 * Block overlaps between kernel and user.
+	 * This can happen due to ring filtering
+	 * On Intel CPUs the entry into the kernel is filtered,
+	 * but the exit is not. Let the caller patch it up.
+	 */
+	if (kernel != machine__kernel_ip(machine, end)) {
+		printf("\tblock %" PRIx64 "-%" PRIx64 " transfers between kernel and user\n",
+				start, end);
+		return -ENXIO;
+	}
+
+	memset(&al, 0, sizeof(al));
+	if (end - start > MAXBB - MAXINSN) {
+		if (last) {
+			printf("\tbrstack does not reach to final jump (%" PRIx64 "-%" PRIx64 ")\n",
+					start, end);
+		} else {
+			printf("\tblock %" PRIx64 "-%" PRIx64 " (%" PRIu64 ") too long to dump\n",
+					start, end, end - start);
+		}
+		return 0;
+	}
+
+	thread__find_addr_map(thread, *cpumode, MAP__FUNCTION, start, &al);
+	if (!al.map || !al.map->dso) {
+		printf("\tcannot resolve %" PRIx64 "-%" PRIx64 "\n",
+				start, end);
+		return 0;
+	}
+	if (al.map->dso->data.status == DSO_DATA_STATUS_ERROR) {
+		printf("\tcannot resolve %" PRIx64 "-%" PRIx64 "\n",
+				start, end);
+		return 0;
+	}
+
+	/* Load maps to ensure dso->is_64_bit has been updated */
+	map__load(al.map);
+
+	offset = al.map->map_ip(al.map, start);
+	len = dso__data_read_offset(al.map->dso, machine,
+				    offset, (u8 *)buffer,
+				    end - start + MAXINSN);
+
+	*is64bit = al.map->dso->is_64_bit;
+	if (len <= 0)
+		printf("\tcannot fetch code for block at %" PRIx64 "-%" PRIx64 "\n",
+			start, end);
+	return len;
+}
+
+static void print_jump(uint64_t ip, struct branch_entry *en,
+		       struct perf_dis *x, u8 *inbuf, int len,
+		       int insn)
+{
+	printf("\t%016" PRIx64 "\t%-30s\t#%s%s%s%s",
+	       ip,
+	       disas_inst(x, ip, inbuf, len, NULL),
+	       en->flags.predicted ? " PRED" : "",
+	       en->flags.mispred ? " MISPRED" : "",
+	       en->flags.in_tx ? " INTX" : "",
+	       en->flags.abort ? " ABORT" : "");
+	if (en->flags.cycles) {
+		printf(" %d cycles", en->flags.cycles);
+		if (insn)
+			printf(" %.2f IPC", (float)insn / en->flags.cycles);
+	}
+	putchar('\n');
+}
+
+static void print_ip_sym(struct thread *thread,
+			 u8 cpumode, int cpu,
+			 uint64_t addr,
+			 struct symbol **lastsym,
+			 struct perf_event_attr *attr)
+{
+	struct addr_location al;
+	int off;
+
+	memset(&al, 0, sizeof(struct addr_location));
+
+	thread__find_addr_map(thread, cpumode, MAP__FUNCTION, addr, &al);
+	if (!al.map)
+		thread__find_addr_map(thread, cpumode, MAP__VARIABLE,
+				      addr, &al);
+	if ((*lastsym) && al.addr >= (*lastsym)->start && al.addr < (*lastsym)->end)
+		return;
+
+	al.cpu = cpu;
+	al.sym = NULL;
+	if (al.map)
+		al.sym = map__find_symbol(al.map, al.addr);
+
+	if (!al.sym)
+		return;
+
+	if (al.addr < al.sym->end)
+		off = al.addr - al.sym->start;
+	else
+		off = al.addr - al.map->start - al.sym->start;
+	printf("\t%s", al.sym->name);
+	if (off)
+		printf("%+d", off);
+	putchar(':');
+	if (PRINT_FIELD(SRCLINE))
+		map__fprintf_srcline(al.map, al.addr, "\t", stdout);
+	putchar('\n');
+	*lastsym = al.sym;
+}
+
+static void print_sample_brstackasm(struct perf_sample *sample,
+				    struct thread *thread,
+				    struct perf_event_attr *attr,
+				    struct machine *machine)
+{
+	struct branch_stack *br = sample->branch_stack;
+	u64 start, end;
+	int i, insn;
+	struct perf_dis x;
+	u8 buffer[MAXBB];
+	int len;
+	int nr;
+	unsigned off;
+	int ilen;
+	struct symbol *lastsym = NULL;
+
+	if (!(br && br->nr))
+		return;
+	nr = br->nr;
+	if (max_blocks && nr > max_blocks + 1)
+		nr = max_blocks + 1;
+
+	x.thread = thread;
+	x.cpu = sample->cpu;
+
+	putchar('\n');
+
+	/* Handle first from jump, of which we don't know the entry. */
+	len = grab_bb(buffer, br->entries[nr-1].from,
+			br->entries[nr-1].from,
+			machine, thread, &x.is64bit, &x.cpumode, false);
+	if (len > 0) {
+		print_ip_sym(thread, x.cpumode, x.cpu,
+			     br->entries[nr - 1].from,
+			     &lastsym, attr);
+		print_jump(br->entries[nr - 1].from, &br->entries[nr - 1],
+			&x, buffer, len, 0);
+	}
+
+	/* Print all blocks */
+	for (i = nr - 2; i >= 0; i--) {
+		if (br->entries[i].from || br->entries[i].to)
+			pr_debug("%d: %" PRIx64 "-%" PRIx64 "\n", i,
+				 br->entries[i].from,
+				 br->entries[i].to);
+		start = br->entries[i + 1].to;
+		end = br->entries[i].from;
+
+		len = grab_bb(buffer, start, end,
+				machine, thread, &x.is64bit,
+				&x.cpumode, false);
+		/* Patch up missing kernel transfers due to ring filters */
+		if (len == -ENXIO && i > 0) {
+			end = br->entries[--i].from;
+			pr_debug("\tpatching up to %" PRIx64 "-%" PRIx64 "\n",
+					start, end);
+			len = grab_bb(buffer, start, end,
+				      machine, thread, &x.is64bit,
+				      &x.cpumode, false);
+		}
+		if (len <= 0)
+			continue;
+
+		insn = 0;
+		for (off = 0;; off += ilen) {
+			uint64_t ip = start + off;
+
+			print_ip_sym(thread, x.cpumode, x.cpu,
+				     ip,
+				     &lastsym, attr);
+			if (ip == end) {
+				print_jump(ip, &br->entries[i], &x,
+					   buffer + off,
+					   len - off, insn);
+				break;
+			} else {
+				printf("\t%016" PRIx64 "\t%s\n", ip,
+					disas_inst(&x, ip, buffer + off,
+						   len - off, &ilen));
+				if (ilen == 0)
+					break;
+				insn++;
+			}
+		}
+	}
+
+	/*
+	 * Hit the branch? In this case we are already done, and the target
+	 * has not been executed yet.
+	 */
+	if (br->entries[0].from == sample->ip)
+		return;
+	if (br->entries[0].flags.abort)
+		return;
+
+	/*
+	 * Print final block upto sample
+	 */
+	start = br->entries[0].to;
+	end = sample->ip;
+	len = grab_bb(buffer, start, end, machine, thread, &x.is64bit,
+			&x.cpumode, true);
+	print_ip_sym(thread, x.cpumode, x.cpu,
+		     start,
+		     &lastsym, attr);
+	if (len <= 0) {
+		/* Print at least last IP if basic block did not work */
+		len = grab_bb(buffer, sample->ip, sample->ip,
+				machine, thread, &x.is64bit, &x.cpumode,
+				false);
+		if (len <= 0)
+			return;
+
+		printf("\t%016" PRIx64 "\t%s\n", sample->ip,
+			disas_inst(&x, sample->ip, buffer, len, NULL));
+		return;
+	}
+	for (off = 0; off <= end - start; off += ilen) {
+		printf("\t%016" PRIx64 "\t%s\n", start + off,
+				disas_inst(&x, start + off, buffer + off,
+					   len - off, &ilen));
+		if (ilen == 0)
+			break;
+	}
+}
 
 static void print_sample_addr(struct perf_sample *sample,
 			  struct thread *thread,
@@ -689,6 +953,8 @@ static void print_insn(union perf_event *event,
 	}
 	if (PRINT_FIELD(ASM))
 		print_sample_asm(event, sample, thread, al, machine);
+	if (PRINT_FIELD(BRSTACKASM))
+		print_sample_brstackasm(sample, thread, attr, machine);
 }
 
 static void print_sample_bts(union perf_event *event,
@@ -2231,6 +2497,8 @@ int cmd_script(int argc, const char **argv, const char *prefix __maybe_unused)
 	OPT_BOOLEAN('\0', "show-switch-events", &script.show_switch_events,
 		    "Show context switch events (if recorded)"),
 	OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"),
+	OPT_INTEGER(0, "max-blocks", &max_blocks,
+		    "Maximum number of code blocks to dump with brstackasm"),
 	OPT_BOOLEAN(0, "ns", &nanosecs,
 		    "Use 9 decimal places when displaying time"),
 	OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts",
-- 
2.9.3