Re: [PATCH 4/5] perf, tools, script: Add support for printing assembler

From: Arnaldo Carvalho de Melo
Date: Mon Jan 23 2017 - 14:49:18 EST

Em Wed, Jan 18, 2017 at 05:41:49PM -0800, Andi Kleen escreveu:
> From: Andi Kleen <ak@xxxxxxxxxxxxxxx>
>
> When dumping PT traces with perf script it is very useful to see the
> assembler for each sample, so that it is easily possible to follow
> the control flow.
>
> As using objdump is difficult and inefficient from perf script this
> patch uses the Intel xed library to implement assembler output.
> The library can be downloaded from http://github.com/intelxed/xed

What I have, and there are multiple changes to make this build, is
available at the tmp.perf/xed branch of my tree at:

git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux.git

https://git.kernel.org/cgit/linux/kernel/git/acme/linux.git/log/?h=tmp.perf/xed

So, while testing this I noticed some differences from what objdump -d produces
to what is produced with this patch, see below, where it is different I added
the bytes from the objdump output, have to lookup the tables manually to check
each case, and some seem harmless.

Do you know if there is any tool comparing the output of objdump -d to what is
produced by a similar xed based tool?

Disassembly of perf_evsel__enable() bits referenced in the samples
collected for this test:

'perf script' with this patch: DIFF objdump -d

4b8506 jz 0x4b84d0 <perf_evsel__enable+0x70> 74 c8 je 4b84d0 <perf_evsel__enable+0x70>
4b84d0 add $0x1, %r14 add $0x1,%r14
4b84d4 cmp %r14d, %ebx cmp %r14d,%ebx
4b84d7 jle 0x4b8530 <perf_evsel__enable+0xd0> jle 4b8530 <perf_evsel__enable+0xd0>
4b8530 add $0x1, %r12 add $0x1,%r12
4b8534 cmp %r12d, %r13d cmp %r12d,%r13d
4b8537 jnle 0x4b84c2 <perf_evsel__enable+0x62> 7f 89 jg 4b84c2 <perf_evsel__enable+0x62>
4b84c2 xor %r14d, %r14d xor %r14d,%r14d
4b84c5 test %ebx, %ebx test %ebx,%ebx
4b84c7 jnle 0x4b84d9 <perf_evsel__enable+0x79> 7f 10 jg 4b84d9 <perf_evsel__enable+0x79>
4b84d9 movq 0x90(%r15), %rax 49 8b 87 90 00 00 00 mov 0x90(%r15),%rax
4b84e0 mov %r12, %rdx mov %r12,%rdx
4b84e3 mov %r14, %rcx mov %r14,%rcx
4b84e6 mov $0x2400, %esi mov $0x2400,%esi
4b84eb imulq (%rax), %rdx 48 0f af 10 imul (%rax),%rdx
4b84ef imulq 0x8(%rax), %rcx 48 0f af 48 08 imul 0x8(%rax),%rcx
4b84f4 add %rdx, %rax add %rdx,%rax
4b84f7 xor %edx, %edx xor %edx,%edx
4b84f9 movl 0x18(%rcx,%rax,1), %edi 8b 7c 01 18 mov 0x18(%rcx,%rax,1),%edi
4b84fd xor %eax, %eax xor %eax,%eax
4b84ff callq 0x42d990 <ioctl@plt> callq 42d990 <ioctl@plt>
4b8504 test %eax, %eax test %eax,%eax
4b8506 jz 0x4b84d0 <perf_evsel__enable+0x70> 74 c8 je 4b84d0 <perf_evsel__enable+0x70>

Built today using xed from:

https://github.com/intelxed/xed

[acme@jouet xed]$ git log --oneline -5
4507b57ba629 rebase tests to account for new operand sorting
c7c1777216f5 generator: xed operand ordering sort was wrong
eb45a282de28 convert LOOPNE/E comments to XED COMMENT field
22427e1a4027 generator.py: duplicate iform check now checks isa-set conflicts
4a0a09a1542e Disambiguate iforms for VFPCLASS{PD,PS} mem forms, suffix with VL
[acme@jouet xed]$

> The previous version of this patch used udis86, but was
> rejected because udis86 was unmaintained and a runtime dependency.
> Using the recently released xed avoids both of these problems:
> - XED is well maintained and used by many Intel tools
> - XED is linked statically so there is no runtime dependency.
>
> The library is probed as an external dependency in the usual way. Then perf
> script calls into it when needed, and handles callbacks to resolve
> symbols.
>
> % perf record -e intel_pt//u true
> % perf script -F sym,symoff,ip,asm --itrace=i0ns | head
> 7fc7188b4190 _start+0x0 mov %rsp, %rdi
> 7fc7188b4193 _start+0x3 call _dl_start
> 7fc7188b7710 _dl_start+0x0 push %rbp
> 7fc7188b7711 _dl_start+0x1 mov %rsp, %rbp
> 7fc7188b7714 _dl_start+0x4 push %r15
> 7fc7188b7716 _dl_start+0x6 push %r14
> 7fc7188b7718 _dl_start+0x8 push %r13
> 7fc7188b771a _dl_start+0xa push %r12
> 7fc7188b771c _dl_start+0xc mov %rdi, %r12
> 7fc7188b771f _dl_start+0xf push %rbx
>
> v2:
> Converted to use XED instead of udis86.
> Separate disassembler interface into separate arch specific file.
> Lots of cleanups and improvements.
>
> Cc: adrian.hunter@xxxxxxxxx
> Signed-off-by: Andi Kleen <ak@xxxxxxxxxxxxxxx>
> ---
> tools/perf/Documentation/perf-script.txt | 4 +-
> tools/perf/builtin-script.c | 72 +++++++++++++++++++++++++++-----
> 2 files changed, 64 insertions(+), 12 deletions(-)
>
> diff --git a/tools/perf/Documentation/perf-script.txt b/tools/perf/Documentation/perf-script.txt
> index 4ed5f239ba7d..497989ea9768 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,
> + srcline, period, iregs, brstack, brstacksym, flags, bpf-output, asm.
> 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
> @@ -198,6 +198,8 @@ OPTIONS
>
> 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.
> +
> -k::
> --vmlinux=<file>::
> vmlinux pathname
> diff --git a/tools/perf/builtin-script.c b/tools/perf/builtin-script.c
> index c0783b4f7b6c..7a09c4f7df3f 100644
> --- a/tools/perf/builtin-script.c
> +++ b/tools/perf/builtin-script.c
> @@ -28,6 +28,7 @@
> #include <linux/time64.h>
> #include "asm/bug.h"
> #include "util/mem-events.h"
> +#include "util/dis.h"
>
> static char const *script_name;
> static char const *generate_script_lang;
> @@ -69,6 +70,7 @@ enum perf_output_field {
> PERF_OUTPUT_CALLINDENT = 1U << 20,
> PERF_OUTPUT_INSN = 1U << 21,
> PERF_OUTPUT_INSNLEN = 1U << 22,
> + PERF_OUTPUT_ASM = 1U << 23,
> };
>
> struct output_option {
> @@ -98,6 +100,7 @@ struct output_option {
> {.str = "callindent", .field = PERF_OUTPUT_CALLINDENT},
> {.str = "insn", .field = PERF_OUTPUT_INSN},
> {.str = "insnlen", .field = PERF_OUTPUT_INSNLEN},
> + {.str = "asm", .field = PERF_OUTPUT_ASM},
> };
>
> /* default set to maintain compatibility with current format */
> @@ -292,7 +295,11 @@ static int perf_evsel__check_attr(struct perf_evsel *evsel,
> "selected. Hence, no address to lookup the source line number.\n");
> return -EINVAL;
> }
> -
> + if (PRINT_FIELD(ASM) && !PRINT_FIELD(IP)) {
> + pr_err("Display of assembler requested but sample IP is not\n"
> + "selected.\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))
> @@ -436,6 +443,39 @@ static void print_sample_iregs(struct perf_sample *sample,
> }
> }
>
> +static void print_sample_asm(union perf_event *event,
> + struct perf_sample *sample,
> + struct thread *thread,
> + struct addr_location *al,
> + struct machine *machine)
> +{
> + struct perf_dis x;
> + u8 buffer[32];
> + int len;
> + u64 offset;
> +
> + x.thread = thread;
> + x.cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
> + x.cpu = sample->cpu;
> +
> + if (!al->map || !al->map->dso)
> + return;
> + if (al->map->dso->data.status == DSO_DATA_STATUS_ERROR)
> + return;
> +
> + /* Load maps to ensure dso->is_64_bit has been updated */
> + map__load(al->map);
> + x.is64bit = al->map->dso->is_64_bit;
> +
> + offset = al->map->map_ip(al->map, sample->ip);
> + len = dso__data_read_offset(al->map->dso, machine,
> + offset, buffer, MAXINSN);
> + if (len <= 0)
> + return;
> +
> + printf("\t%s", disas_inst(&x, sample->ip, buffer, len, NULL));
> +}
> +
> static void print_sample_start(struct perf_sample *sample,
> struct thread *thread,
> struct perf_evsel *evsel)
> @@ -631,8 +671,12 @@ static void print_sample_callindent(struct perf_sample *sample,
> printf("%*s", spacing - len, "");
> }
>
> -static void print_insn(struct perf_sample *sample,
> - struct perf_event_attr *attr)
> +static void print_insn(union perf_event *event,
> + struct perf_sample *sample,
> + struct perf_event_attr *attr,
> + struct thread *thread,
> + struct addr_location *al,
> + struct machine *machine)
> {
> if (PRINT_FIELD(INSNLEN))
> printf(" ilen: %d", sample->insn_len);
> @@ -643,12 +687,16 @@ static void print_insn(struct perf_sample *sample,
> for (i = 0; i < sample->insn_len; i++)
> printf(" %02x", (unsigned char)sample->insn[i]);
> }
> + if (PRINT_FIELD(ASM))
> + print_sample_asm(event, sample, thread, al, machine);
> }
>
> -static void print_sample_bts(struct perf_sample *sample,
> +static void print_sample_bts(union perf_event *event,
> + struct perf_sample *sample,
> struct perf_evsel *evsel,
> struct thread *thread,
> - struct addr_location *al)
> + struct addr_location *al,
> + struct machine *machine)
> {
> struct perf_event_attr *attr = &evsel->attr;
> bool print_srcline_last = false;
> @@ -689,7 +737,7 @@ static void print_sample_bts(struct perf_sample *sample,
> if (print_srcline_last)
> map__fprintf_srcline(al->map, al->addr, "\n ", stdout);
>
> - print_insn(sample, attr);
> + print_insn(event, sample, attr, thread, al, machine);
>
> printf("\n");
> }
> @@ -871,7 +919,9 @@ static size_t data_src__printf(u64 data_src)
>
> static void process_event(struct perf_script *script,
> struct perf_sample *sample, struct perf_evsel *evsel,
> - struct addr_location *al)
> + struct addr_location *al,
> + struct machine *machine,
> + union perf_event *event)
> {
> struct thread *thread = al->thread;
> struct perf_event_attr *attr = &evsel->attr;
> @@ -898,7 +948,7 @@ static void process_event(struct perf_script *script,
> print_sample_flags(sample->flags);
>
> if (is_bts_event(attr)) {
> - print_sample_bts(sample, evsel, thread, al);
> + print_sample_bts(event, sample, evsel, thread, al, machine);
> return;
> }
>
> @@ -936,7 +986,7 @@ static void process_event(struct perf_script *script,
>
> if (perf_evsel__is_bpf_output(evsel) && PRINT_FIELD(BPF_OUTPUT))
> print_sample_bpf_output(sample);
> - print_insn(sample, attr);
> + print_insn(event, sample, attr, thread, al, machine);
> printf("\n");
> }
>
> @@ -1046,7 +1096,7 @@ static int process_sample_event(struct perf_tool *tool,
> if (scripting_ops)
> scripting_ops->process_event(event, sample, evsel, &al);
> else
> - process_event(scr, sample, evsel, &al);
> + process_event(scr, sample, evsel, &al, machine, event);
>
> out_put:
> addr_location__put(&al);
> @@ -2152,7 +2202,7 @@ int cmd_script(int argc, const char **argv, const char *prefix __maybe_unused)
> "Valid types: hw,sw,trace,raw. "
> "Fields: comm,tid,pid,time,cpu,event,trace,ip,sym,dso,"
> "addr,symoff,period,iregs,brstack,brstacksym,flags,"
> - "bpf-output,callindent,insn,insnlen", parse_output_fields),
> + "bpf-output,callindent,insn,insnlen,asm", parse_output_fields),
> OPT_BOOLEAN('a', "all-cpus", &system_wide,
> "system-wide collection from all CPUs"),
> OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
> --
> 2.9.3