| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Date: Mon, 22 Jan 2024 16:42:16 +0800
From: Changbin Du <changbin.du@...wei.com>
To: Adrian Hunter <adrian.hunter@...el.com>
CC: Changbin Du <changbin.du@...wei.com>, Mark Rutland <mark.rutland@....com>,
Alexander Shishkin <alexander.shishkin@...ux.intel.com>, Jiri Olsa
<jolsa@...nel.org>, Namhyung Kim <namhyung@...nel.org>, Ian Rogers
<irogers@...gle.com>, <linux-kernel@...r.kernel.org>,
<linux-perf-users@...r.kernel.org>, Andi Kleen <ak@...ux.intel.com>, Thomas
Richter <tmricht@...ux.ibm.com>, <changbin.du@...il.com>, Peter Zijlstra
<peterz@...radead.org>, Arnaldo Carvalho de Melo <acme@...nel.org>, Ingo
Molnar <mingo@...hat.com>
Subject: Re: [PATCH v4 2/5] perf: util: use capstone disasm engine to show
assembly instructions
On Mon, Jan 22, 2024 at 10:24:21AM +0200, Adrian Hunter wrote:
> On 20/01/24 11:13, Changbin Du wrote:
> > On Fri, Jan 19, 2024 at 08:39:19PM +0200, Adrian Hunter wrote:
> >> On 19/01/24 12:48, Changbin Du wrote:
> >>> Currently, the instructions of samples are shown as raw hex strings
> >>> which are hard to read. x86 has a special option '--xed' to disassemble
> >>> the hex string via intel XED tool.
> >>>
> >>> Here we use capstone as our disassembler engine to give more friendly
> >>> instructions. We select libcapstone because capstone can provide more
> >>> insn details. Perf will fallback to raw instructions if libcapstone is
> >>> not available.
> >>>
> >>> The advantages compared to XED tool:
> >>> * Support arm, arm64, x86-32, x86_64 (more could be supported),
> >>> xed only for x86_64.
> >>> * Immediate address operands are shown as symbol+offs.
> >>>
> >>> Signed-off-by: Changbin Du <changbin.du@...wei.com>
> >>> ---
> >>> tools/perf/builtin-script.c | 8 +--
> >>> tools/perf/util/Build | 1 +
> >>> tools/perf/util/print_insn.c | 122 +++++++++++++++++++++++++++++++++++
> >>> tools/perf/util/print_insn.h | 14 ++++
> >>> 4 files changed, 140 insertions(+), 5 deletions(-)
> >>> create mode 100644 tools/perf/util/print_insn.c
> >>> create mode 100644 tools/perf/util/print_insn.h
> >>>
> >>> diff --git a/tools/perf/builtin-script.c b/tools/perf/builtin-script.c
> >>> index b1f57401ff23..4817a37f16e2 100644
> >>> --- a/tools/perf/builtin-script.c
> >>> +++ b/tools/perf/builtin-script.c
> >>> @@ -34,6 +34,7 @@
> >>> #include "util/event.h"
> >>> #include "ui/ui.h"
> >>> #include "print_binary.h"
> >>> +#include "print_insn.h"
> >>> #include "archinsn.h"
> >>> #include <linux/bitmap.h>
> >>> #include <linux/kernel.h>
> >>> @@ -1511,11 +1512,8 @@ static int perf_sample__fprintf_insn(struct perf_sample *sample,
> >>> if (PRINT_FIELD(INSNLEN))
> >>> printed += fprintf(fp, " ilen: %d", sample->insn_len);
> >>> if (PRINT_FIELD(INSN) && sample->insn_len) {
> >>> - int i;
> >>> -
> >>> - printed += fprintf(fp, " insn:");
> >>> - for (i = 0; i < sample->insn_len; i++)
> >>> - printed += fprintf(fp, " %02x", (unsigned char)sample->insn[i]);
> >>> + printed += fprintf(fp, " insn: ");
> >>
> >> "insn:" seems unnecessary. Also xed prints 2 tabs, which
> >> helps line up the output. Perhaps 1 tab and 2 spaces is
> >> enough.
> >>
> > The "insn:" is used by xed. So it can not be removed if we preserve xed
> > function.
>
> I got mixed up - it is patch 3 that the comment is meant for.
>
> >
> > For 'insn' field, I keep the original output format.
> > For 'disasm' field, we can line up the output. I changed to 2 tabs and removed
> > 'insn:'.
>
> That is better, thanks!
>
> >
> >>> + printed += sample__fprintf_insn_raw(sample, fp);
> >>> }
> >>> if (PRINT_FIELD(BRSTACKINSN) || PRINT_FIELD(BRSTACKINSNLEN))
> >>> printed += perf_sample__fprintf_brstackinsn(sample, thread, attr, machine, fp);
> >>> diff --git a/tools/perf/util/Build b/tools/perf/util/Build
> >>> index 988473bf907a..c33aab53d8dd 100644
> >>> --- a/tools/perf/util/Build
> >>> +++ b/tools/perf/util/Build
> >>> @@ -32,6 +32,7 @@ perf-y += perf_regs.o
> >>> perf-y += perf-regs-arch/
> >>> perf-y += path.o
> >>> perf-y += print_binary.o
> >>> +perf-y += print_insn.o
> >>> perf-y += rlimit.o
> >>> perf-y += argv_split.o
> >>> perf-y += rbtree.o
> >>> diff --git a/tools/perf/util/print_insn.c b/tools/perf/util/print_insn.c
> >>> new file mode 100644
> >>> index 000000000000..162be4856f79
> >>> --- /dev/null
> >>> +++ b/tools/perf/util/print_insn.c
> >>> @@ -0,0 +1,122 @@
> >>> +// SPDX-License-Identifier: GPL-2.0
> >>> +/*
> >>> + * Instruction binary disassembler based on capstone.
> >>> + *
> >>> + * Author(s): Changbin Du <changbin.du@...wei.com>
> >>> + */
> >>> +#include "print_insn.h"
> >>
> >> Please put with the other non-system includes
> >>
> > done.
> >
> >>> +#include <stdlib.h>
> >>> +#include <string.h>
> >>> +#include <stdbool.h>
> >>> +#include "util/debug.h"
> >>
> >> util/ not needed
> >>
> > done.
> >
> >>> +#include "util/symbol.h"
> >>
> >> util/ not needed
> >>
> > done.
> >
> >>> +#include "machine.h"
> >>> +
> >>> +size_t sample__fprintf_insn_raw(struct perf_sample *sample, FILE *fp)
> >>> +{
> >>> + int printed = 0;
> >>> +
> >>> + for (int i = 0; i < sample->insn_len; i++)
> >>> + printed += fprintf(fp, "%02x ", (unsigned char)sample->insn[i]);
> >>
> >> Why change this to put a space on the end?
> >>
> > Removed the tailing space.
> >
> >>> + return printed;
> >>> +}
> >>> +
> >>> +#ifdef HAVE_LIBCAPSTONE_SUPPORT
> >>> +#include <capstone/capstone.h>
> >>> +
> >>> +static int capstone_init(struct machine *machine, csh *cs_handle)
> >>> +{
> >>> + cs_arch arch;
> >>> + cs_mode mode;
> >>> +
> >>> + if (machine__is(machine, "x86_64")) {
> >>> + arch = CS_ARCH_X86;
> >>> + mode = CS_MODE_64;
> >>> + } else if (machine__normalized_is(machine, "x86")) {
> >>> + arch = CS_ARCH_X86;
> >>> + mode = CS_MODE_32;
> >>> + } else if (machine__normalized_is(machine, "arm64")) {
> >>> + arch = CS_ARCH_ARM64;
> >>> + mode = CS_MODE_ARM;
> >>> + } else if (machine__normalized_is(machine, "arm")) {
> >>> + arch = CS_ARCH_ARM;
> >>> + mode = CS_MODE_ARM + CS_MODE_V8;
> >>> + } else if (machine__normalized_is(machine, "s390")) {
> >>> + arch = CS_ARCH_SYSZ;
> >>> + mode = CS_MODE_BIG_ENDIAN;
> >>> + } else {
> >>> + return -1;
> >>> + }
> >>> +
> >>> + if (cs_open(arch, mode, cs_handle) != CS_ERR_OK) {
> >>> + pr_warning_once("cs_open failed\n");
> >>> + return -1;
> >>> + }
> >>> +
> >>> + cs_option(*cs_handle, CS_OPT_SYNTAX, CS_OPT_SYNTAX_ATT);
> >>
> >> That's only needed for x86 isn't it
> >>
> > Moved into below branch.
> >
> >>> + if (machine__normalized_is(machine, "x86"))
> >>> + cs_option(*cs_handle, CS_OPT_DETAIL, CS_OPT_ON);
> >>
> >> Why? Could use a comment.
> >>
> > /*
> > * Resolving address oprands to symbols is implemented
>
> oprands -> operands
Fixed, thanks!
>
> > * on x86 by investigating instruction details.
> > */
> >
> >>> +
> >>> + return 0;
> >>> +}
> >>> +
> >>> +static size_t print_insn_x86(struct perf_sample *sample, struct thread *thread,
> >>> + cs_insn *insn, FILE *fp)
> >>> +{
> >>> + struct addr_location al;
> >>> + size_t printed = 0;
> >>> +
> >>> + if (insn->detail && insn->detail->x86.op_count == 1) {
> >>> + cs_x86_op *op = &insn->detail->x86.operands[0];
> >>> +
> >>> + addr_location__init(&al);
> >>
> >> Missing addr_location__exit()
> >>
> > Fixed.
> >
> >>> +
> >>> + if (op->type == X86_OP_IMM &&
> >>> + thread__find_symbol(thread, sample->cpumode, op->imm, &al)) {
> >>> + printed += fprintf(fp, "%s ", insn[0].mnemonic);
> >>> + printed += symbol__fprintf_symname_offs(al.sym, &al, fp);
> >>> + return printed;
> >>> + }
> >>> + }
> >>> +
> >>> + printed += fprintf(fp, "%s %s", insn[0].mnemonic, insn[0].op_str);
> >>> + return printed;
> >>> +}
> >>> +
> >>> +size_t sample__fprintf_insn(struct perf_sample *sample, struct thread *thread,
> >>> + struct machine *machine, FILE *fp)
> >>> +{
> >>> + static csh cs_handle;
> >>
> >> Why static?
> >>
> > Removed. See below.
> >
> >>> + cs_insn *insn;
> >>> + size_t count;
> >>> + size_t printed = 0;
> >>> + int ret;
> >>> +
> >>> + ret = capstone_init(machine, &cs_handle);
> >>
> >> Does this really need to be done every time?
> >>
> > Only need to init once exactly. The problem is I cannot find a appropriate
> > place to do the initiation.
> >
> > I tried to initiate on first call but we still need a global mutex to be
> > initiated.
> >
> > So finally I fallback to initiate every time. The redundant initiation is
> > acceptable per my test.
>
> OK, could add a FIXME
>
I added a TODO instead.
/* TODO: Try to initiate capstone only once but need a proper place. */
> >
> >>> + if (ret < 0) {
> >>> + /* fallback */
> >>> + return sample__fprintf_insn_raw(sample, fp);
> >>> + }
> >>> +
> >>> + count = cs_disasm(cs_handle, (uint8_t *)sample->insn, sample->insn_len,
> >>> + sample->ip, 1, &insn);
> >>> + if (count > 0) {
> >>> + if (machine__normalized_is(machine, "x86"))
> >>> + printed += print_insn_x86(sample, thread, &insn[0], fp);
> >>> + else
> >>> + printed += fprintf(fp, "%s %s", insn[0].mnemonic, insn[0].op_str);
> >>> + cs_free(insn, count);
> >>> + } else {
> >>> + printed += fprintf(fp, "illegal instruction");
> >>> + }
> >>> +
> >>> + cs_close(&cs_handle);
> >>> + return printed;
> >>> +}
> >>> +#else
> >>> +size_t sample__fprintf_insn(struct perf_sample *sample, struct thread *thread __maybe_unused,
> >>> + struct machine *machine __maybe_unused, FILE *fp)
> >>> +{
> >>> + return sample__fprintf_insn_raw(sample, fp);
> >>> +}
> >>> +#endif
> >>> diff --git a/tools/perf/util/print_insn.h b/tools/perf/util/print_insn.h
> >>> new file mode 100644
> >>> index 000000000000..af8fa5d01fb7
> >>> --- /dev/null
> >>> +++ b/tools/perf/util/print_insn.h
> >>> @@ -0,0 +1,14 @@
> >>> +/* SPDX-License-Identifier: GPL-2.0 */
> >>> +#ifndef PERF_PRINT_ISNS_H
> >>
> >> Here and elsewhere
> >>
> >> ISNS -> INSN
> >>
> > fixed.
> >
> >>> +#define PERF_PRINT_ISNS_H
> >>> +
> >>> +#include <stddef.h>
> >>> +#include <stdio.h>
> >>> +#include "event.h"
> >>> +#include "util/thread.h"
> >>
> >> Instead of including event.h and thread.h, just forward declare:
> >>
> >> struct perf_sample;
> >> struct thread;
> >> struct machine;
> >>
> > Why forward declaration?
>
> That is how it is done in perf tools. It avoids unnecessary dependency
> among headers.
ok, done.
>
> >
> >>> +
> >>> +size_t sample__fprintf_insn(struct perf_sample *sample, struct thread *thread,
> >>> + struct machine *machine, FILE *fp);
> >>> +size_t sample__fprintf_insn_raw(struct perf_sample *sample, FILE *fp);
> >>> +
> >>> +#endif /* PERF_PRINT_ISNS_H */
> >>
> >
>
--
Cheers,
Changbin Du
Powered by blists - more mailing lists